kernel/memremap.c

   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  */
  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>
  18 #include <linux/io.h>
  19 #include <linux/mm.h>
  20 #include <linux/memory_hotplug.h>
  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
  30 static void *try_ram_remap(resource_size_t offset, size_t size)
  31 {
  32         unsigned long pfn = PHYS_PFN(offset);
  33
  34         /* In the simple case just return the existing linear address */
  35         if (pfn_valid(pfn) && !PageHighMem(pfn_to_page(pfn)))
  36                 return __va(offset);
  37         return NULL; /* fallback to ioremap_cache */
  38 }
  39
  40 /**
  41  * memremap() - remap an iomem_resource as cacheable memory
  42  * @offset: iomem resource start address
  43  * @size: size of remap
  44  * @flags: any of MEMREMAP_WB, MEMREMAP_WT and MEMREMAP_WC
  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. In the case of multiple flags, the different
  49  * mapping types will be attempted in the order listed below until one of
  50  * them succeeds.
  51  *
  52  * MEMREMAP_WB - matches the default mapping for System RAM on
  53  * the architecture.  This is usually a read-allocate write-back cache.
  54  * Morever, if MEMREMAP_WB is specified and the requested remap region is RAM
  55  * memremap() will bypass establishing a new mapping and instead return
  56  * a pointer into the direct map.
  57  *
  58  * MEMREMAP_WT - establish a mapping whereby writes either bypass the
  59  * cache or are written through to memory and never exist in a
  60  * cache-dirty state with respect to program visibility.  Attempts to
  61  * map System RAM with this mapping type will fail.
  62  *
  63  * MEMREMAP_WC - establish a writecombine mapping, whereby writes may
  64  * be coalesced together (e.g. in the CPU's write buffers), but is otherwise
  65  * uncached. Attempts to map System RAM with this mapping type will fail.
  66  */
  67 void *memremap(resource_size_t offset, size_t size, unsigned long flags)
  68 {
  69         int is_ram = region_intersects(offset, size,
  70                                        IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
  71         void *addr = NULL;
  72
  73         if (!flags)
  74                 return NULL;
  75
  76         if (is_ram == REGION_MIXED) {
  77                 WARN_ONCE(1, "memremap attempted on mixed range %pa size: %#lx\n",
  78                                 &offset, (unsigned long) size);
  79                 return NULL;
  80         }
  81
  82         /* Try all mapping types requested until one returns non-NULL */
  83         if (flags & MEMREMAP_WB) {
  84                 /*
  85                  * MEMREMAP_WB is special in that it can be satisifed
  86                  * from the direct map.  Some archs depend on the
  87                  * capability of memremap() to autodetect cases where
  88                  * the requested range is potentially in System RAM.
  89                  */
  90                 if (is_ram == REGION_INTERSECTS)
  91                         addr = try_ram_remap(offset, size);
  92                 if (!addr)
  93                         addr = ioremap_cache(offset, size);
  94         }
  95
  96         /*
  97          * If we don't have a mapping yet and other request flags are
  98          * present then we will be attempting to establish a new virtual
  99          * address mapping.  Enforce that this mapping is not aliasing
 100          * System RAM.
 101          */
 102         if (!addr && is_ram == REGION_INTERSECTS && flags != MEMREMAP_WB) {
 103                 WARN_ONCE(1, "memremap attempted on ram %pa size: %#lx\n",
 104                                 &offset, (unsigned long) size);
 105                 return NULL;
 106         }
 107
 108         if (!addr && (flags & MEMREMAP_WT))
 109                 addr = ioremap_wt(offset, size);
 110
 111         if (!addr && (flags & MEMREMAP_WC))
 112                 addr = ioremap_wc(offset, size);
 113
 114         return addr;
 115 }
 116 EXPORT_SYMBOL(memremap);
 117
 118 void memunmap(void *addr)
 119 {
 120         if (is_vmalloc_addr(addr))
 121                 iounmap((void __iomem *) addr);
 122 }
 123 EXPORT_SYMBOL(memunmap);
 124
 125 static void devm_memremap_release(struct device *dev, void *res)
 126 {
 127         memunmap(*(void **)res);
 128 }
 129
 130 static int devm_memremap_match(struct device *dev, void *res, void *match_data)
 131 {
 132         return *(void **)res == match_data;
 133 }
 134
 135 void *devm_memremap(struct device *dev, resource_size_t offset,
 136                 size_t size, unsigned long flags)
 137 {
 138         void **ptr, *addr;
 139
 140         ptr = devres_alloc_node(devm_memremap_release, sizeof(*ptr), GFP_KERNEL,
 141                         dev_to_node(dev));
 142         if (!ptr)
 143                 return ERR_PTR(-ENOMEM);
 144
 145         addr = memremap(offset, size, flags);
 146         if (addr) {
 147                 *ptr = addr;
 148                 devres_add(dev, ptr);
 149         } else {
 150                 devres_free(ptr);
 151                 return ERR_PTR(-ENXIO);
 152         }
 153
 154         return addr;
 155 }
 156 EXPORT_SYMBOL(devm_memremap);
 157
 158 void devm_memunmap(struct device *dev, void *addr)
 159 {
 160         WARN_ON(devres_release(dev, devm_memremap_release,
 161                                 devm_memremap_match, addr));
 162 }
 163 EXPORT_SYMBOL(devm_memunmap);
 164
 165 pfn_t phys_to_pfn_t(phys_addr_t addr, u64 flags)
 166 {
 167         return __pfn_to_pfn_t(addr >> PAGE_SHIFT, flags);
 168 }
 169 EXPORT_SYMBOL(phys_to_pfn_t);
 170
 171 #ifdef CONFIG_ZONE_DEVICE
 172 static DEFINE_MUTEX(pgmap_lock);
 173 static RADIX_TREE(pgmap_radix, GFP_KERNEL);
 174 #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
 175 #define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
 176
 177 struct page_map {
 178         struct resource res;
 179         struct percpu_ref *ref;
 180         struct dev_pagemap pgmap;
 181         struct vmem_altmap altmap;
 182 };
 183
 184 void get_zone_device_page(struct page *page)
 185 {
 186         percpu_ref_get(page->pgmap->ref);
 187 }
 188 EXPORT_SYMBOL(get_zone_device_page);
 189
 190 void put_zone_device_page(struct page *page)
 191 {
 192         put_dev_pagemap(page->pgmap);
 193 }
 194 EXPORT_SYMBOL(put_zone_device_page);
 195
 196 static void pgmap_radix_release(struct resource *res)
 197 {
 198         resource_size_t key, align_start, align_size, align_end;
 199
 200         align_start = res->start & ~(SECTION_SIZE - 1);
 201         align_size = ALIGN(resource_size(res), SECTION_SIZE);
 202         align_end = align_start + align_size - 1;
 203
 204         mutex_lock(&pgmap_lock);
 205         for (key = res->start; key <= res->end; key += SECTION_SIZE)
 206                 radix_tree_delete(&pgmap_radix, key >> PA_SECTION_SHIFT);
 207         mutex_unlock(&pgmap_lock);
 208 }
 209
 210 static unsigned long pfn_first(struct page_map *page_map)
 211 {
 212         struct dev_pagemap *pgmap = &page_map->pgmap;
 213         const struct resource *res = &page_map->res;
 214         struct vmem_altmap *altmap = pgmap->altmap;
 215         unsigned long pfn;
 216
 217         pfn = res->start >> PAGE_SHIFT;
 218         if (altmap)
 219                 pfn += vmem_altmap_offset(altmap);
 220         return pfn;
 221 }
 222
 223 static unsigned long pfn_end(struct page_map *page_map)
 224 {
 225         const struct resource *res = &page_map->res;
 226
 227         return (res->start + resource_size(res)) >> PAGE_SHIFT;
 228 }
 229
 230 #define for_each_device_pfn(pfn, map) \
 231         for (pfn = pfn_first(map); pfn < pfn_end(map); pfn++)
 232
 233 static void devm_memremap_pages_release(struct device *dev, void *data)
 234 {
 235         struct page_map *page_map = data;
 236         struct resource *res = &page_map->res;
 237         resource_size_t align_start, align_size;
 238         struct dev_pagemap *pgmap = &page_map->pgmap;
 239
 240         if (percpu_ref_tryget_live(pgmap->ref)) {
 241                 dev_WARN(dev, "%s: page mapping is still live!\n", __func__);
 242                 percpu_ref_put(pgmap->ref);
 243         }
 244
 245         /* pages are dead and unused, undo the arch mapping */
 246         align_start = res->start & ~(SECTION_SIZE - 1);
 247         align_size = ALIGN(resource_size(res), SECTION_SIZE);
 248         arch_remove_memory(align_start, align_size);
 249         pgmap_radix_release(res);
 250         dev_WARN_ONCE(dev, pgmap->altmap && pgmap->altmap->alloc,
 251                         "%s: failed to free all reserved pages\n", __func__);
 252 }
 253
 254 /* assumes rcu_read_lock() held at entry */
 255 struct dev_pagemap *find_dev_pagemap(resource_size_t phys)
 256 {
 257         struct page_map *page_map;
 258
 259         WARN_ON_ONCE(!rcu_read_lock_held());
 260
 261         page_map = radix_tree_lookup(&pgmap_radix, phys >> PA_SECTION_SHIFT);
 262         return page_map ? &page_map->pgmap : NULL;
 263 }
 264
 265 /**
 266  * devm_memremap_pages - remap and provide memmap backing for the given resource
 267  * @dev: hosting device for @res
 268  * @res: "host memory" address range
 269  * @ref: a live per-cpu reference count
 270  * @altmap: optional descriptor for allocating the memmap from @res
 271  *
 272  * Notes:
 273  * 1/ @ref must be 'live' on entry and 'dead' before devm_memunmap_pages() time
 274  *    (or devm release event).
 275  *
 276  * 2/ @res is expected to be a host memory range that could feasibly be
 277  *    treated as a "System RAM" range, i.e. not a device mmio range, but
 278  *    this is not enforced.
 279  */
 280 void *devm_memremap_pages(struct device *dev, struct resource *res,
 281                 struct percpu_ref *ref, struct vmem_altmap *altmap)
 282 {
 283         resource_size_t key, align_start, align_size, align_end;
 284         struct dev_pagemap *pgmap;
 285         struct page_map *page_map;
 286         int error, nid, is_ram;
 287         unsigned long pfn;
 288
 289         align_start = res->start & ~(SECTION_SIZE - 1);
 290         align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE)
 291                 - align_start;
 292         is_ram = region_intersects(align_start, align_size,
 293                 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
 294
 295         if (is_ram == REGION_MIXED) {
 296                 WARN_ONCE(1, "%s attempted on mixed region %pr\n",
 297                                 __func__, res);
 298                 return ERR_PTR(-ENXIO);
 299         }
 300
 301         if (is_ram == REGION_INTERSECTS)
 302                 return __va(res->start);
 303
 304         if (altmap && !IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) {
 305                 dev_err(dev, "%s: altmap requires CONFIG_SPARSEMEM_VMEMMAP=y\n",
 306                                 __func__);
 307                 return ERR_PTR(-ENXIO);
 308         }
 309
 310         if (!ref)
 311                 return ERR_PTR(-EINVAL);
 312
 313         page_map = devres_alloc_node(devm_memremap_pages_release,
 314                         sizeof(*page_map), GFP_KERNEL, dev_to_node(dev));
 315         if (!page_map)
 316                 return ERR_PTR(-ENOMEM);
 317         pgmap = &page_map->pgmap;
 318
 319         memcpy(&page_map->res, res, sizeof(*res));
 320
 321         pgmap->dev = dev;
 322         if (altmap) {
 323                 memcpy(&page_map->altmap, altmap, sizeof(*altmap));
 324                 pgmap->altmap = &page_map->altmap;
 325         }
 326         pgmap->ref = ref;
 327         pgmap->res = &page_map->res;
 328
 329         mutex_lock(&pgmap_lock);
 330         error = 0;
 331         align_end = align_start + align_size - 1;
 332         for (key = align_start; key <= align_end; key += SECTION_SIZE) {
 333                 struct dev_pagemap *dup;
 334
 335                 rcu_read_lock();
 336                 dup = find_dev_pagemap(key);
 337                 rcu_read_unlock();
 338                 if (dup) {
 339                         dev_err(dev, "%s: %pr collides with mapping for %s\n",
 340                                         __func__, res, dev_name(dup->dev));
 341                         error = -EBUSY;
 342                         break;
 343                 }
 344                 error = radix_tree_insert(&pgmap_radix, key >> PA_SECTION_SHIFT,
 345                                 page_map);
 346                 if (error) {
 347                         dev_err(dev, "%s: failed: %d\n", __func__, error);
 348                         break;
 349                 }
 350         }
 351         mutex_unlock(&pgmap_lock);
 352         if (error)
 353                 goto err_radix;
 354
 355         nid = dev_to_node(dev);
 356         if (nid < 0)
 357                 nid = numa_mem_id();
 358
 359         error = arch_add_memory(nid, align_start, align_size, true);
 360         if (error)
 361                 goto err_add_memory;
 362
 363         for_each_device_pfn(pfn, page_map) {
 364                 struct page *page = pfn_to_page(pfn);
 365
 366                 /*
 367                  * ZONE_DEVICE pages union ->lru with a ->pgmap back
 368                  * pointer.  It is a bug if a ZONE_DEVICE page is ever
 369                  * freed or placed on a driver-private list.  Seed the
 370                  * storage with LIST_POISON* values.
 371                  */
 372                 list_del(&page->lru);
 373                 page->pgmap = pgmap;
 374         }
 375         devres_add(dev, page_map);
 376         return __va(res->start);
 377
 378  err_add_memory:
 379  err_radix:
 380         pgmap_radix_release(res);
 381         devres_free(page_map);
 382         return ERR_PTR(error);
 383 }
 384 EXPORT_SYMBOL(devm_memremap_pages);
 385
 386 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
 387 {
 388         /* number of pfns from base where pfn_to_page() is valid */
 389         return altmap->reserve + altmap->free;
 390 }
 391
 392 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
 393 {
 394         altmap->alloc -= nr_pfns;
 395 }
 396
 397 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 398 struct vmem_altmap *to_vmem_altmap(unsigned long memmap_start)
 399 {
 400         /*
 401          * 'memmap_start' is the virtual address for the first "struct
 402          * page" in this range of the vmemmap array.  In the case of
 403          * CONFIG_SPARSEMEM_VMEMMAP a page_to_pfn conversion is simple
 404          * pointer arithmetic, so we can perform this to_vmem_altmap()
 405          * conversion without concern for the initialization state of
 406          * the struct page fields.
 407          */
 408         struct page *page = (struct page *) memmap_start;
 409         struct dev_pagemap *pgmap;
 410
 411         /*
 412          * Unconditionally retrieve a dev_pagemap associated with the
 413          * given physical address, this is only for use in the
 414          * arch_{add|remove}_memory() for setting up and tearing down
 415          * the memmap.
 416          */
 417         rcu_read_lock();
 418         pgmap = find_dev_pagemap(__pfn_to_phys(page_to_pfn(page)));
 419         rcu_read_unlock();
 420
 421         return pgmap ? pgmap->altmap : NULL;
 422 }
 423 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
 424 #endif /* CONFIG_ZONE_DEVICE */