]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/memremap.c
bridge: add new BR_NEIGH_SUPPRESS port flag to suppress arp and nd flood
[mirror_ubuntu-bionic-kernel.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 13#include <linux/radix-tree.h>
7d3dcf26 14#include <linux/device.h>
92281dee 15#include <linux/types.h>
34c0fd54 16#include <linux/pfn_t.h>
92281dee
DW
17#include <linux/io.h>
18#include <linux/mm.h>
41e94a85 19#include <linux/memory_hotplug.h>
5042db43
JG
20#include <linux/swap.h>
21#include <linux/swapops.h>
92281dee
DW
22
23#ifndef ioremap_cache
24/* temporary while we convert existing ioremap_cache users to memremap */
25__weak void __iomem *ioremap_cache(resource_size_t offset, unsigned long size)
26{
27 return ioremap(offset, size);
28}
29#endif
30
c269cba3
AB
31#ifndef arch_memremap_wb
32static void *arch_memremap_wb(resource_size_t offset, unsigned long size)
33{
34 return (__force void *)ioremap_cache(offset, size);
35}
36#endif
37
8f716c9b
TL
38#ifndef arch_memremap_can_ram_remap
39static bool arch_memremap_can_ram_remap(resource_size_t offset, size_t size,
40 unsigned long flags)
41{
42 return true;
43}
44#endif
45
46static void *try_ram_remap(resource_size_t offset, size_t size,
47 unsigned long flags)
182475b7 48{
ac343e88 49 unsigned long pfn = PHYS_PFN(offset);
182475b7
DW
50
51 /* In the simple case just return the existing linear address */
8f716c9b
TL
52 if (pfn_valid(pfn) && !PageHighMem(pfn_to_page(pfn)) &&
53 arch_memremap_can_ram_remap(offset, size, flags))
182475b7 54 return __va(offset);
8f716c9b 55
c269cba3 56 return NULL; /* fallback to arch_memremap_wb */
182475b7
DW
57}
58
92281dee
DW
59/**
60 * memremap() - remap an iomem_resource as cacheable memory
61 * @offset: iomem resource start address
62 * @size: size of remap
8f716c9b
TL
63 * @flags: any of MEMREMAP_WB, MEMREMAP_WT, MEMREMAP_WC,
64 * MEMREMAP_ENC, MEMREMAP_DEC
92281dee
DW
65 *
66 * memremap() is "ioremap" for cases where it is known that the resource
67 * being mapped does not have i/o side effects and the __iomem
c907e0eb
BS
68 * annotation is not applicable. In the case of multiple flags, the different
69 * mapping types will be attempted in the order listed below until one of
70 * them succeeds.
92281dee 71 *
1c29f25b 72 * MEMREMAP_WB - matches the default mapping for System RAM on
92281dee
DW
73 * the architecture. This is usually a read-allocate write-back cache.
74 * Morever, if MEMREMAP_WB is specified and the requested remap region is RAM
75 * memremap() will bypass establishing a new mapping and instead return
76 * a pointer into the direct map.
77 *
78 * MEMREMAP_WT - establish a mapping whereby writes either bypass the
79 * cache or are written through to memory and never exist in a
80 * cache-dirty state with respect to program visibility. Attempts to
1c29f25b 81 * map System RAM with this mapping type will fail.
c907e0eb
BS
82 *
83 * MEMREMAP_WC - establish a writecombine mapping, whereby writes may
84 * be coalesced together (e.g. in the CPU's write buffers), but is otherwise
85 * uncached. Attempts to map System RAM with this mapping type will fail.
92281dee
DW
86 */
87void *memremap(resource_size_t offset, size_t size, unsigned long flags)
88{
1c29f25b
TK
89 int is_ram = region_intersects(offset, size,
90 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
92281dee
DW
91 void *addr = NULL;
92
cf61e2a1
BS
93 if (!flags)
94 return NULL;
95
92281dee
DW
96 if (is_ram == REGION_MIXED) {
97 WARN_ONCE(1, "memremap attempted on mixed range %pa size: %#lx\n",
98 &offset, (unsigned long) size);
99 return NULL;
100 }
101
102 /* Try all mapping types requested until one returns non-NULL */
103 if (flags & MEMREMAP_WB) {
92281dee
DW
104 /*
105 * MEMREMAP_WB is special in that it can be satisifed
106 * from the direct map. Some archs depend on the
107 * capability of memremap() to autodetect cases where
1c29f25b 108 * the requested range is potentially in System RAM.
92281dee
DW
109 */
110 if (is_ram == REGION_INTERSECTS)
8f716c9b 111 addr = try_ram_remap(offset, size, flags);
182475b7 112 if (!addr)
c269cba3 113 addr = arch_memremap_wb(offset, size);
92281dee
DW
114 }
115
116 /*
cf61e2a1
BS
117 * If we don't have a mapping yet and other request flags are
118 * present then we will be attempting to establish a new virtual
92281dee 119 * address mapping. Enforce that this mapping is not aliasing
1c29f25b 120 * System RAM.
92281dee 121 */
cf61e2a1 122 if (!addr && is_ram == REGION_INTERSECTS && flags != MEMREMAP_WB) {
92281dee
DW
123 WARN_ONCE(1, "memremap attempted on ram %pa size: %#lx\n",
124 &offset, (unsigned long) size);
125 return NULL;
126 }
127
cf61e2a1 128 if (!addr && (flags & MEMREMAP_WT))
92281dee 129 addr = ioremap_wt(offset, size);
c907e0eb
BS
130
131 if (!addr && (flags & MEMREMAP_WC))
132 addr = ioremap_wc(offset, size);
92281dee
DW
133
134 return addr;
135}
136EXPORT_SYMBOL(memremap);
137
138void memunmap(void *addr)
139{
140 if (is_vmalloc_addr(addr))
141 iounmap((void __iomem *) addr);
142}
143EXPORT_SYMBOL(memunmap);
7d3dcf26
CH
144
145static void devm_memremap_release(struct device *dev, void *res)
146{
9273a8bb 147 memunmap(*(void **)res);
7d3dcf26
CH
148}
149
150static int devm_memremap_match(struct device *dev, void *res, void *match_data)
151{
152 return *(void **)res == match_data;
153}
154
155void *devm_memremap(struct device *dev, resource_size_t offset,
156 size_t size, unsigned long flags)
157{
158 void **ptr, *addr;
159
538ea4aa
DW
160 ptr = devres_alloc_node(devm_memremap_release, sizeof(*ptr), GFP_KERNEL,
161 dev_to_node(dev));
7d3dcf26 162 if (!ptr)
b36f4761 163 return ERR_PTR(-ENOMEM);
7d3dcf26
CH
164
165 addr = memremap(offset, size, flags);
166 if (addr) {
167 *ptr = addr;
168 devres_add(dev, ptr);
93f834df 169 } else {
7d3dcf26 170 devres_free(ptr);
93f834df
TK
171 return ERR_PTR(-ENXIO);
172 }
7d3dcf26
CH
173
174 return addr;
175}
176EXPORT_SYMBOL(devm_memremap);
177
178void devm_memunmap(struct device *dev, void *addr)
179{
d741314f
DW
180 WARN_ON(devres_release(dev, devm_memremap_release,
181 devm_memremap_match, addr));
7d3dcf26
CH
182}
183EXPORT_SYMBOL(devm_memunmap);
41e94a85
CH
184
185#ifdef CONFIG_ZONE_DEVICE
9476df7d
DW
186static DEFINE_MUTEX(pgmap_lock);
187static RADIX_TREE(pgmap_radix, GFP_KERNEL);
188#define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
189#define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
190
41e94a85
CH
191struct page_map {
192 struct resource res;
9476df7d
DW
193 struct percpu_ref *ref;
194 struct dev_pagemap pgmap;
4b94ffdc 195 struct vmem_altmap altmap;
41e94a85
CH
196};
197
ab1b597e 198static unsigned long order_at(struct resource *res, unsigned long pgoff)
9476df7d 199{
ab1b597e
DW
200 unsigned long phys_pgoff = PHYS_PFN(res->start) + pgoff;
201 unsigned long nr_pages, mask;
eb7d78c9 202
ab1b597e
DW
203 nr_pages = PHYS_PFN(resource_size(res));
204 if (nr_pages == pgoff)
205 return ULONG_MAX;
206
207 /*
208 * What is the largest aligned power-of-2 range available from
209 * this resource pgoff to the end of the resource range,
210 * considering the alignment of the current pgoff?
211 */
212 mask = phys_pgoff | rounddown_pow_of_two(nr_pages - pgoff);
213 if (!mask)
214 return ULONG_MAX;
215
216 return find_first_bit(&mask, BITS_PER_LONG);
217}
218
219#define foreach_order_pgoff(res, order, pgoff) \
220 for (pgoff = 0, order = order_at((res), pgoff); order < ULONG_MAX; \
221 pgoff += 1UL << order, order = order_at((res), pgoff))
222
5042db43
JG
223#if IS_ENABLED(CONFIG_DEVICE_PRIVATE)
224int device_private_entry_fault(struct vm_area_struct *vma,
225 unsigned long addr,
226 swp_entry_t entry,
227 unsigned int flags,
228 pmd_t *pmdp)
229{
230 struct page *page = device_private_entry_to_page(entry);
231
232 /*
233 * The page_fault() callback must migrate page back to system memory
234 * so that CPU can access it. This might fail for various reasons
235 * (device issue, device was unsafely unplugged, ...). When such
236 * error conditions happen, the callback must return VM_FAULT_SIGBUS.
237 *
238 * Note that because memory cgroup charges are accounted to the device
239 * memory, this should never fail because of memory restrictions (but
240 * allocation of regular system page might still fail because we are
241 * out of memory).
242 *
243 * There is a more in-depth description of what that callback can and
244 * cannot do, in include/linux/memremap.h
245 */
246 return page->pgmap->page_fault(vma, addr, page, flags, pmdp);
247}
248EXPORT_SYMBOL(device_private_entry_fault);
249#endif /* CONFIG_DEVICE_PRIVATE */
250
ab1b597e
DW
251static void pgmap_radix_release(struct resource *res)
252{
253 unsigned long pgoff, order;
9476df7d
DW
254
255 mutex_lock(&pgmap_lock);
ab1b597e
DW
256 foreach_order_pgoff(res, order, pgoff)
257 radix_tree_delete(&pgmap_radix, PHYS_PFN(res->start) + pgoff);
9476df7d 258 mutex_unlock(&pgmap_lock);
ab1b597e
DW
259
260 synchronize_rcu();
9476df7d
DW
261}
262
5c2c2587
DW
263static unsigned long pfn_first(struct page_map *page_map)
264{
265 struct dev_pagemap *pgmap = &page_map->pgmap;
266 const struct resource *res = &page_map->res;
267 struct vmem_altmap *altmap = pgmap->altmap;
268 unsigned long pfn;
269
270 pfn = res->start >> PAGE_SHIFT;
271 if (altmap)
272 pfn += vmem_altmap_offset(altmap);
273 return pfn;
274}
275
276static unsigned long pfn_end(struct page_map *page_map)
277{
278 const struct resource *res = &page_map->res;
279
280 return (res->start + resource_size(res)) >> PAGE_SHIFT;
281}
282
283#define for_each_device_pfn(pfn, map) \
284 for (pfn = pfn_first(map); pfn < pfn_end(map); pfn++)
285
9476df7d 286static void devm_memremap_pages_release(struct device *dev, void *data)
41e94a85 287{
9476df7d
DW
288 struct page_map *page_map = data;
289 struct resource *res = &page_map->res;
290 resource_size_t align_start, align_size;
4b94ffdc 291 struct dev_pagemap *pgmap = &page_map->pgmap;
71389703
DW
292 unsigned long pfn;
293
294 for_each_device_pfn(pfn, page_map)
295 put_page(pfn_to_page(pfn));
9476df7d 296
5c2c2587
DW
297 if (percpu_ref_tryget_live(pgmap->ref)) {
298 dev_WARN(dev, "%s: page mapping is still live!\n", __func__);
299 percpu_ref_put(pgmap->ref);
300 }
301
41e94a85 302 /* pages are dead and unused, undo the arch mapping */
9476df7d
DW
303 align_start = res->start & ~(SECTION_SIZE - 1);
304 align_size = ALIGN(resource_size(res), SECTION_SIZE);
b5d24fda 305
f931ab47 306 mem_hotplug_begin();
9476df7d 307 arch_remove_memory(align_start, align_size);
f931ab47 308 mem_hotplug_done();
b5d24fda 309
9049771f 310 untrack_pfn(NULL, PHYS_PFN(align_start), align_size);
eb7d78c9 311 pgmap_radix_release(res);
4b94ffdc
DW
312 dev_WARN_ONCE(dev, pgmap->altmap && pgmap->altmap->alloc,
313 "%s: failed to free all reserved pages\n", __func__);
9476df7d
DW
314}
315
316/* assumes rcu_read_lock() held at entry */
317struct dev_pagemap *find_dev_pagemap(resource_size_t phys)
318{
319 struct page_map *page_map;
320
321 WARN_ON_ONCE(!rcu_read_lock_held());
322
ab1b597e 323 page_map = radix_tree_lookup(&pgmap_radix, PHYS_PFN(phys));
9476df7d 324 return page_map ? &page_map->pgmap : NULL;
41e94a85
CH
325}
326
4b94ffdc
DW
327/**
328 * devm_memremap_pages - remap and provide memmap backing for the given resource
329 * @dev: hosting device for @res
330 * @res: "host memory" address range
5c2c2587 331 * @ref: a live per-cpu reference count
4b94ffdc
DW
332 * @altmap: optional descriptor for allocating the memmap from @res
333 *
5c2c2587
DW
334 * Notes:
335 * 1/ @ref must be 'live' on entry and 'dead' before devm_memunmap_pages() time
71389703
DW
336 * (or devm release event). The expected order of events is that @ref has
337 * been through percpu_ref_kill() before devm_memremap_pages_release(). The
338 * wait for the completion of all references being dropped and
339 * percpu_ref_exit() must occur after devm_memremap_pages_release().
5c2c2587
DW
340 *
341 * 2/ @res is expected to be a host memory range that could feasibly be
342 * treated as a "System RAM" range, i.e. not a device mmio range, but
343 * this is not enforced.
4b94ffdc
DW
344 */
345void *devm_memremap_pages(struct device *dev, struct resource *res,
5c2c2587 346 struct percpu_ref *ref, struct vmem_altmap *altmap)
41e94a85 347{
ab1b597e
DW
348 resource_size_t align_start, align_size, align_end;
349 unsigned long pfn, pgoff, order;
9049771f 350 pgprot_t pgprot = PAGE_KERNEL;
4b94ffdc 351 struct dev_pagemap *pgmap;
41e94a85 352 struct page_map *page_map;
1fdcce6e 353 int error, nid, is_ram, i = 0;
5f29a77c
DW
354
355 align_start = res->start & ~(SECTION_SIZE - 1);
356 align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE)
357 - align_start;
d37a14bb
LT
358 is_ram = region_intersects(align_start, align_size,
359 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
41e94a85
CH
360
361 if (is_ram == REGION_MIXED) {
362 WARN_ONCE(1, "%s attempted on mixed region %pr\n",
363 __func__, res);
364 return ERR_PTR(-ENXIO);
365 }
366
367 if (is_ram == REGION_INTERSECTS)
368 return __va(res->start);
369
5c2c2587
DW
370 if (!ref)
371 return ERR_PTR(-EINVAL);
372
538ea4aa
DW
373 page_map = devres_alloc_node(devm_memremap_pages_release,
374 sizeof(*page_map), GFP_KERNEL, dev_to_node(dev));
41e94a85
CH
375 if (!page_map)
376 return ERR_PTR(-ENOMEM);
4b94ffdc 377 pgmap = &page_map->pgmap;
41e94a85
CH
378
379 memcpy(&page_map->res, res, sizeof(*res));
380
4b94ffdc
DW
381 pgmap->dev = dev;
382 if (altmap) {
383 memcpy(&page_map->altmap, altmap, sizeof(*altmap));
384 pgmap->altmap = &page_map->altmap;
385 }
5c2c2587 386 pgmap->ref = ref;
4b94ffdc 387 pgmap->res = &page_map->res;
5042db43
JG
388 pgmap->type = MEMORY_DEVICE_HOST;
389 pgmap->page_fault = NULL;
390 pgmap->page_free = NULL;
391 pgmap->data = NULL;
4b94ffdc 392
9476df7d
DW
393 mutex_lock(&pgmap_lock);
394 error = 0;
eb7d78c9 395 align_end = align_start + align_size - 1;
ab1b597e
DW
396
397 foreach_order_pgoff(res, order, pgoff) {
9476df7d
DW
398 struct dev_pagemap *dup;
399
400 rcu_read_lock();
ab1b597e 401 dup = find_dev_pagemap(res->start + PFN_PHYS(pgoff));
9476df7d
DW
402 rcu_read_unlock();
403 if (dup) {
404 dev_err(dev, "%s: %pr collides with mapping for %s\n",
405 __func__, res, dev_name(dup->dev));
406 error = -EBUSY;
407 break;
408 }
ab1b597e
DW
409 error = __radix_tree_insert(&pgmap_radix,
410 PHYS_PFN(res->start) + pgoff, order, page_map);
9476df7d
DW
411 if (error) {
412 dev_err(dev, "%s: failed: %d\n", __func__, error);
413 break;
414 }
415 }
416 mutex_unlock(&pgmap_lock);
417 if (error)
418 goto err_radix;
419
41e94a85
CH
420 nid = dev_to_node(dev);
421 if (nid < 0)
7eff93b7 422 nid = numa_mem_id();
41e94a85 423
9049771f
DW
424 error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(align_start), 0,
425 align_size);
426 if (error)
427 goto err_pfn_remap;
428
f931ab47 429 mem_hotplug_begin();
3d79a728 430 error = arch_add_memory(nid, align_start, align_size, false);
f1dd2cd1
MH
431 if (!error)
432 move_pfn_range_to_zone(&NODE_DATA(nid)->node_zones[ZONE_DEVICE],
433 align_start >> PAGE_SHIFT,
434 align_size >> PAGE_SHIFT);
f931ab47 435 mem_hotplug_done();
9476df7d
DW
436 if (error)
437 goto err_add_memory;
41e94a85 438
5c2c2587
DW
439 for_each_device_pfn(pfn, page_map) {
440 struct page *page = pfn_to_page(pfn);
441
d77a117e
DW
442 /*
443 * ZONE_DEVICE pages union ->lru with a ->pgmap back
444 * pointer. It is a bug if a ZONE_DEVICE page is ever
445 * freed or placed on a driver-private list. Seed the
446 * storage with LIST_POISON* values.
447 */
448 list_del(&page->lru);
5c2c2587 449 page->pgmap = pgmap;
71389703 450 percpu_ref_get(ref);
1fdcce6e
MH
451 if (!(++i % 1024))
452 cond_resched();
5c2c2587 453 }
41e94a85
CH
454 devres_add(dev, page_map);
455 return __va(res->start);
9476df7d
DW
456
457 err_add_memory:
9049771f
DW
458 untrack_pfn(NULL, PHYS_PFN(align_start), align_size);
459 err_pfn_remap:
9476df7d
DW
460 err_radix:
461 pgmap_radix_release(res);
462 devres_free(page_map);
463 return ERR_PTR(error);
41e94a85
CH
464}
465EXPORT_SYMBOL(devm_memremap_pages);
4b94ffdc
DW
466
467unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
468{
469 /* number of pfns from base where pfn_to_page() is valid */
470 return altmap->reserve + altmap->free;
471}
472
473void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
474{
475 altmap->alloc -= nr_pfns;
476}
477
4b94ffdc
DW
478struct vmem_altmap *to_vmem_altmap(unsigned long memmap_start)
479{
480 /*
481 * 'memmap_start' is the virtual address for the first "struct
482 * page" in this range of the vmemmap array. In the case of
07061aab 483 * CONFIG_SPARSEMEM_VMEMMAP a page_to_pfn conversion is simple
4b94ffdc
DW
484 * pointer arithmetic, so we can perform this to_vmem_altmap()
485 * conversion without concern for the initialization state of
486 * the struct page fields.
487 */
488 struct page *page = (struct page *) memmap_start;
489 struct dev_pagemap *pgmap;
490
491 /*
07061aab 492 * Unconditionally retrieve a dev_pagemap associated with the
4b94ffdc
DW
493 * given physical address, this is only for use in the
494 * arch_{add|remove}_memory() for setting up and tearing down
495 * the memmap.
496 */
497 rcu_read_lock();
498 pgmap = find_dev_pagemap(__pfn_to_phys(page_to_pfn(page)));
499 rcu_read_unlock();
500
501 return pgmap ? pgmap->altmap : NULL;
502}
41e94a85 503#endif /* CONFIG_ZONE_DEVICE */
7b2d55d2
JG
504
505
df6ad698
JG
506#if IS_ENABLED(CONFIG_DEVICE_PRIVATE) || IS_ENABLED(CONFIG_DEVICE_PUBLIC)
507void put_zone_device_private_or_public_page(struct page *page)
7b2d55d2
JG
508{
509 int count = page_ref_dec_return(page);
510
511 /*
512 * If refcount is 1 then page is freed and refcount is stable as nobody
513 * holds a reference on the page.
514 */
515 if (count == 1) {
516 /* Clear Active bit in case of parallel mark_page_accessed */
517 __ClearPageActive(page);
518 __ClearPageWaiters(page);
519
520 page->mapping = NULL;
c733a828 521 mem_cgroup_uncharge(page);
7b2d55d2
JG
522
523 page->pgmap->page_free(page, page->pgmap->data);
524 } else if (!count)
525 __put_page(page);
526}
df6ad698
JG
527EXPORT_SYMBOL(put_zone_device_private_or_public_page);
528#endif /* CONFIG_DEVICE_PRIVATE || CONFIG_DEVICE_PUBLIC */