Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Generic hugetlb support. |
| 3 | * (C) William Irwin, April 2004 |
| 4 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | #include <linux/list.h> |
| 6 | #include <linux/init.h> |
| 7 | #include <linux/module.h> |
| 8 | #include <linux/mm.h> |
Alexey Dobriyan | e1759c2 | 2008-10-15 23:50:22 +0400 | [diff] [blame] | 9 | #include <linux/seq_file.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | #include <linux/sysctl.h> |
| 11 | #include <linux/highmem.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 12 | #include <linux/mmu_notifier.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | #include <linux/nodemask.h> |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 14 | #include <linux/pagemap.h> |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 15 | #include <linux/mempolicy.h> |
Christoph Lameter | aea47ff | 2006-01-08 01:00:57 -0800 | [diff] [blame] | 16 | #include <linux/cpuset.h> |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 17 | #include <linux/mutex.h> |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 18 | #include <linux/bootmem.h> |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 19 | #include <linux/sysfs.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 20 | #include <linux/slab.h> |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 21 | #include <linux/rmap.h> |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 22 | #include <linux/swap.h> |
| 23 | #include <linux/swapops.h> |
Linus Torvalds | d660668 | 2008-08-06 12:04:54 -0700 | [diff] [blame] | 24 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 25 | #include <asm/page.h> |
| 26 | #include <asm/pgtable.h> |
Adrian Bunk | 78a34ae | 2008-07-28 15:46:30 -0700 | [diff] [blame] | 27 | #include <asm/io.h> |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 28 | |
| 29 | #include <linux/hugetlb.h> |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 30 | #include <linux/node.h> |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 31 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | |
| 33 | const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL; |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 34 | static gfp_t htlb_alloc_mask = GFP_HIGHUSER; |
| 35 | unsigned long hugepages_treat_as_movable; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 36 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 37 | static int max_hstate; |
| 38 | unsigned int default_hstate_idx; |
| 39 | struct hstate hstates[HUGE_MAX_HSTATE]; |
| 40 | |
Jon Tollefson | 53ba51d | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 41 | __initdata LIST_HEAD(huge_boot_pages); |
| 42 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 43 | /* for command line parsing */ |
| 44 | static struct hstate * __initdata parsed_hstate; |
| 45 | static unsigned long __initdata default_hstate_max_huge_pages; |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 46 | static unsigned long __initdata default_hstate_size; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 47 | |
| 48 | #define for_each_hstate(h) \ |
| 49 | for ((h) = hstates; (h) < &hstates[max_hstate]; (h)++) |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 50 | |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 51 | /* |
| 52 | * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages |
| 53 | */ |
| 54 | static DEFINE_SPINLOCK(hugetlb_lock); |
Eric Paris | 0bd0f9f | 2005-11-21 21:32:28 -0800 | [diff] [blame] | 55 | |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 56 | /* |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 57 | * Region tracking -- allows tracking of reservations and instantiated pages |
| 58 | * across the pages in a mapping. |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 59 | * |
| 60 | * The region data structures are protected by a combination of the mmap_sem |
| 61 | * and the hugetlb_instantion_mutex. To access or modify a region the caller |
| 62 | * must either hold the mmap_sem for write, or the mmap_sem for read and |
| 63 | * the hugetlb_instantiation mutex: |
| 64 | * |
| 65 | * down_write(&mm->mmap_sem); |
| 66 | * or |
| 67 | * down_read(&mm->mmap_sem); |
| 68 | * mutex_lock(&hugetlb_instantiation_mutex); |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 69 | */ |
| 70 | struct file_region { |
| 71 | struct list_head link; |
| 72 | long from; |
| 73 | long to; |
| 74 | }; |
| 75 | |
| 76 | static long region_add(struct list_head *head, long f, long t) |
| 77 | { |
| 78 | struct file_region *rg, *nrg, *trg; |
| 79 | |
| 80 | /* Locate the region we are either in or before. */ |
| 81 | list_for_each_entry(rg, head, link) |
| 82 | if (f <= rg->to) |
| 83 | break; |
| 84 | |
| 85 | /* Round our left edge to the current segment if it encloses us. */ |
| 86 | if (f > rg->from) |
| 87 | f = rg->from; |
| 88 | |
| 89 | /* Check for and consume any regions we now overlap with. */ |
| 90 | nrg = rg; |
| 91 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { |
| 92 | if (&rg->link == head) |
| 93 | break; |
| 94 | if (rg->from > t) |
| 95 | break; |
| 96 | |
| 97 | /* If this area reaches higher then extend our area to |
| 98 | * include it completely. If this is not the first area |
| 99 | * which we intend to reuse, free it. */ |
| 100 | if (rg->to > t) |
| 101 | t = rg->to; |
| 102 | if (rg != nrg) { |
| 103 | list_del(&rg->link); |
| 104 | kfree(rg); |
| 105 | } |
| 106 | } |
| 107 | nrg->from = f; |
| 108 | nrg->to = t; |
| 109 | return 0; |
| 110 | } |
| 111 | |
| 112 | static long region_chg(struct list_head *head, long f, long t) |
| 113 | { |
| 114 | struct file_region *rg, *nrg; |
| 115 | long chg = 0; |
| 116 | |
| 117 | /* Locate the region we are before or in. */ |
| 118 | list_for_each_entry(rg, head, link) |
| 119 | if (f <= rg->to) |
| 120 | break; |
| 121 | |
| 122 | /* If we are below the current region then a new region is required. |
| 123 | * Subtle, allocate a new region at the position but make it zero |
| 124 | * size such that we can guarantee to record the reservation. */ |
| 125 | if (&rg->link == head || t < rg->from) { |
| 126 | nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); |
| 127 | if (!nrg) |
| 128 | return -ENOMEM; |
| 129 | nrg->from = f; |
| 130 | nrg->to = f; |
| 131 | INIT_LIST_HEAD(&nrg->link); |
| 132 | list_add(&nrg->link, rg->link.prev); |
| 133 | |
| 134 | return t - f; |
| 135 | } |
| 136 | |
| 137 | /* Round our left edge to the current segment if it encloses us. */ |
| 138 | if (f > rg->from) |
| 139 | f = rg->from; |
| 140 | chg = t - f; |
| 141 | |
| 142 | /* Check for and consume any regions we now overlap with. */ |
| 143 | list_for_each_entry(rg, rg->link.prev, link) { |
| 144 | if (&rg->link == head) |
| 145 | break; |
| 146 | if (rg->from > t) |
| 147 | return chg; |
| 148 | |
| 149 | /* We overlap with this area, if it extends futher than |
| 150 | * us then we must extend ourselves. Account for its |
| 151 | * existing reservation. */ |
| 152 | if (rg->to > t) { |
| 153 | chg += rg->to - t; |
| 154 | t = rg->to; |
| 155 | } |
| 156 | chg -= rg->to - rg->from; |
| 157 | } |
| 158 | return chg; |
| 159 | } |
| 160 | |
| 161 | static long region_truncate(struct list_head *head, long end) |
| 162 | { |
| 163 | struct file_region *rg, *trg; |
| 164 | long chg = 0; |
| 165 | |
| 166 | /* Locate the region we are either in or before. */ |
| 167 | list_for_each_entry(rg, head, link) |
| 168 | if (end <= rg->to) |
| 169 | break; |
| 170 | if (&rg->link == head) |
| 171 | return 0; |
| 172 | |
| 173 | /* If we are in the middle of a region then adjust it. */ |
| 174 | if (end > rg->from) { |
| 175 | chg = rg->to - end; |
| 176 | rg->to = end; |
| 177 | rg = list_entry(rg->link.next, typeof(*rg), link); |
| 178 | } |
| 179 | |
| 180 | /* Drop any remaining regions. */ |
| 181 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { |
| 182 | if (&rg->link == head) |
| 183 | break; |
| 184 | chg += rg->to - rg->from; |
| 185 | list_del(&rg->link); |
| 186 | kfree(rg); |
| 187 | } |
| 188 | return chg; |
| 189 | } |
| 190 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 191 | static long region_count(struct list_head *head, long f, long t) |
| 192 | { |
| 193 | struct file_region *rg; |
| 194 | long chg = 0; |
| 195 | |
| 196 | /* Locate each segment we overlap with, and count that overlap. */ |
| 197 | list_for_each_entry(rg, head, link) { |
| 198 | int seg_from; |
| 199 | int seg_to; |
| 200 | |
| 201 | if (rg->to <= f) |
| 202 | continue; |
| 203 | if (rg->from >= t) |
| 204 | break; |
| 205 | |
| 206 | seg_from = max(rg->from, f); |
| 207 | seg_to = min(rg->to, t); |
| 208 | |
| 209 | chg += seg_to - seg_from; |
| 210 | } |
| 211 | |
| 212 | return chg; |
| 213 | } |
| 214 | |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 215 | /* |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 216 | * Convert the address within this vma to the page offset within |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 217 | * the mapping, in pagecache page units; huge pages here. |
| 218 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 219 | static pgoff_t vma_hugecache_offset(struct hstate *h, |
| 220 | struct vm_area_struct *vma, unsigned long address) |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 221 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 222 | return ((address - vma->vm_start) >> huge_page_shift(h)) + |
| 223 | (vma->vm_pgoff >> huge_page_order(h)); |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 224 | } |
| 225 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 226 | pgoff_t linear_hugepage_index(struct vm_area_struct *vma, |
| 227 | unsigned long address) |
| 228 | { |
| 229 | return vma_hugecache_offset(hstate_vma(vma), vma, address); |
| 230 | } |
| 231 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 232 | /* |
Mel Gorman | 08fba69 | 2009-01-06 14:38:53 -0800 | [diff] [blame] | 233 | * Return the size of the pages allocated when backing a VMA. In the majority |
| 234 | * cases this will be same size as used by the page table entries. |
| 235 | */ |
| 236 | unsigned long vma_kernel_pagesize(struct vm_area_struct *vma) |
| 237 | { |
| 238 | struct hstate *hstate; |
| 239 | |
| 240 | if (!is_vm_hugetlb_page(vma)) |
| 241 | return PAGE_SIZE; |
| 242 | |
| 243 | hstate = hstate_vma(vma); |
| 244 | |
| 245 | return 1UL << (hstate->order + PAGE_SHIFT); |
| 246 | } |
Joerg Roedel | f340ca0 | 2009-06-19 15:16:22 +0200 | [diff] [blame] | 247 | EXPORT_SYMBOL_GPL(vma_kernel_pagesize); |
Mel Gorman | 08fba69 | 2009-01-06 14:38:53 -0800 | [diff] [blame] | 248 | |
| 249 | /* |
Mel Gorman | 3340289 | 2009-01-06 14:38:54 -0800 | [diff] [blame] | 250 | * Return the page size being used by the MMU to back a VMA. In the majority |
| 251 | * of cases, the page size used by the kernel matches the MMU size. On |
| 252 | * architectures where it differs, an architecture-specific version of this |
| 253 | * function is required. |
| 254 | */ |
| 255 | #ifndef vma_mmu_pagesize |
| 256 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) |
| 257 | { |
| 258 | return vma_kernel_pagesize(vma); |
| 259 | } |
| 260 | #endif |
| 261 | |
| 262 | /* |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 263 | * Flags for MAP_PRIVATE reservations. These are stored in the bottom |
| 264 | * bits of the reservation map pointer, which are always clear due to |
| 265 | * alignment. |
| 266 | */ |
| 267 | #define HPAGE_RESV_OWNER (1UL << 0) |
| 268 | #define HPAGE_RESV_UNMAPPED (1UL << 1) |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 269 | #define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 270 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 271 | /* |
| 272 | * These helpers are used to track how many pages are reserved for |
| 273 | * faults in a MAP_PRIVATE mapping. Only the process that called mmap() |
| 274 | * is guaranteed to have their future faults succeed. |
| 275 | * |
| 276 | * With the exception of reset_vma_resv_huge_pages() which is called at fork(), |
| 277 | * the reserve counters are updated with the hugetlb_lock held. It is safe |
| 278 | * to reset the VMA at fork() time as it is not in use yet and there is no |
| 279 | * chance of the global counters getting corrupted as a result of the values. |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 280 | * |
| 281 | * The private mapping reservation is represented in a subtly different |
| 282 | * manner to a shared mapping. A shared mapping has a region map associated |
| 283 | * with the underlying file, this region map represents the backing file |
| 284 | * pages which have ever had a reservation assigned which this persists even |
| 285 | * after the page is instantiated. A private mapping has a region map |
| 286 | * associated with the original mmap which is attached to all VMAs which |
| 287 | * reference it, this region map represents those offsets which have consumed |
| 288 | * reservation ie. where pages have been instantiated. |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 289 | */ |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 290 | static unsigned long get_vma_private_data(struct vm_area_struct *vma) |
| 291 | { |
| 292 | return (unsigned long)vma->vm_private_data; |
| 293 | } |
| 294 | |
| 295 | static void set_vma_private_data(struct vm_area_struct *vma, |
| 296 | unsigned long value) |
| 297 | { |
| 298 | vma->vm_private_data = (void *)value; |
| 299 | } |
| 300 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 301 | struct resv_map { |
| 302 | struct kref refs; |
| 303 | struct list_head regions; |
| 304 | }; |
| 305 | |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 306 | static struct resv_map *resv_map_alloc(void) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 307 | { |
| 308 | struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); |
| 309 | if (!resv_map) |
| 310 | return NULL; |
| 311 | |
| 312 | kref_init(&resv_map->refs); |
| 313 | INIT_LIST_HEAD(&resv_map->regions); |
| 314 | |
| 315 | return resv_map; |
| 316 | } |
| 317 | |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 318 | static void resv_map_release(struct kref *ref) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 319 | { |
| 320 | struct resv_map *resv_map = container_of(ref, struct resv_map, refs); |
| 321 | |
| 322 | /* Clear out any active regions before we release the map. */ |
| 323 | region_truncate(&resv_map->regions, 0); |
| 324 | kfree(resv_map); |
| 325 | } |
| 326 | |
| 327 | static struct resv_map *vma_resv_map(struct vm_area_struct *vma) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 328 | { |
| 329 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 330 | if (!(vma->vm_flags & VM_MAYSHARE)) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 331 | return (struct resv_map *)(get_vma_private_data(vma) & |
| 332 | ~HPAGE_RESV_MASK); |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 333 | return NULL; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 334 | } |
| 335 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 336 | static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 337 | { |
| 338 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 339 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 340 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 341 | set_vma_private_data(vma, (get_vma_private_data(vma) & |
| 342 | HPAGE_RESV_MASK) | (unsigned long)map); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 343 | } |
| 344 | |
| 345 | static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags) |
| 346 | { |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 347 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 348 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 349 | |
| 350 | set_vma_private_data(vma, get_vma_private_data(vma) | flags); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 351 | } |
| 352 | |
| 353 | static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag) |
| 354 | { |
| 355 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 356 | |
| 357 | return (get_vma_private_data(vma) & flag) != 0; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 358 | } |
| 359 | |
| 360 | /* Decrement the reserved pages in the hugepage pool by one */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 361 | static void decrement_hugepage_resv_vma(struct hstate *h, |
| 362 | struct vm_area_struct *vma) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 363 | { |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 364 | if (vma->vm_flags & VM_NORESERVE) |
| 365 | return; |
| 366 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 367 | if (vma->vm_flags & VM_MAYSHARE) { |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 368 | /* Shared mappings always use reserves */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 369 | h->resv_huge_pages--; |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 370 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 371 | /* |
| 372 | * Only the process that called mmap() has reserves for |
| 373 | * private mappings. |
| 374 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 375 | h->resv_huge_pages--; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 376 | } |
| 377 | } |
| 378 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 379 | /* Reset counters to 0 and clear all HPAGE_RESV_* flags */ |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 380 | void reset_vma_resv_huge_pages(struct vm_area_struct *vma) |
| 381 | { |
| 382 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 383 | if (!(vma->vm_flags & VM_MAYSHARE)) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 384 | vma->vm_private_data = (void *)0; |
| 385 | } |
| 386 | |
| 387 | /* Returns true if the VMA has associated reserve pages */ |
Mel Gorman | 7f09ca5 | 2008-07-23 21:27:58 -0700 | [diff] [blame] | 388 | static int vma_has_reserves(struct vm_area_struct *vma) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 389 | { |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 390 | if (vma->vm_flags & VM_MAYSHARE) |
Mel Gorman | 7f09ca5 | 2008-07-23 21:27:58 -0700 | [diff] [blame] | 391 | return 1; |
| 392 | if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) |
| 393 | return 1; |
| 394 | return 0; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 395 | } |
| 396 | |
Naoya Horiguchi | 0ebabb4 | 2010-09-08 10:19:34 +0900 | [diff] [blame] | 397 | static void copy_gigantic_page(struct page *dst, struct page *src) |
| 398 | { |
| 399 | int i; |
| 400 | struct hstate *h = page_hstate(src); |
| 401 | struct page *dst_base = dst; |
| 402 | struct page *src_base = src; |
| 403 | |
| 404 | for (i = 0; i < pages_per_huge_page(h); ) { |
| 405 | cond_resched(); |
| 406 | copy_highpage(dst, src); |
| 407 | |
| 408 | i++; |
| 409 | dst = mem_map_next(dst, dst_base, i); |
| 410 | src = mem_map_next(src, src_base, i); |
| 411 | } |
| 412 | } |
| 413 | |
| 414 | void copy_huge_page(struct page *dst, struct page *src) |
| 415 | { |
| 416 | int i; |
| 417 | struct hstate *h = page_hstate(src); |
| 418 | |
| 419 | if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) { |
| 420 | copy_gigantic_page(dst, src); |
| 421 | return; |
| 422 | } |
| 423 | |
| 424 | might_sleep(); |
| 425 | for (i = 0; i < pages_per_huge_page(h); i++) { |
| 426 | cond_resched(); |
| 427 | copy_highpage(dst + i, src + i); |
| 428 | } |
| 429 | } |
| 430 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 431 | static void enqueue_huge_page(struct hstate *h, struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | { |
| 433 | int nid = page_to_nid(page); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 434 | list_add(&page->lru, &h->hugepage_freelists[nid]); |
| 435 | h->free_huge_pages++; |
| 436 | h->free_huge_pages_node[nid]++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | } |
| 438 | |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 439 | static struct page *dequeue_huge_page_node(struct hstate *h, int nid) |
| 440 | { |
| 441 | struct page *page; |
| 442 | |
| 443 | if (list_empty(&h->hugepage_freelists[nid])) |
| 444 | return NULL; |
| 445 | page = list_entry(h->hugepage_freelists[nid].next, struct page, lru); |
| 446 | list_del(&page->lru); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 447 | set_page_refcounted(page); |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 448 | h->free_huge_pages--; |
| 449 | h->free_huge_pages_node[nid]--; |
| 450 | return page; |
| 451 | } |
| 452 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 453 | static struct page *dequeue_huge_page_vma(struct hstate *h, |
| 454 | struct vm_area_struct *vma, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 455 | unsigned long address, int avoid_reserve) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | struct page *page = NULL; |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 458 | struct mempolicy *mpol; |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 459 | nodemask_t *nodemask; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 460 | struct zonelist *zonelist; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 461 | struct zone *zone; |
| 462 | struct zoneref *z; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 463 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 464 | get_mems_allowed(); |
| 465 | zonelist = huge_zonelist(vma, address, |
| 466 | htlb_alloc_mask, &mpol, &nodemask); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 467 | /* |
| 468 | * A child process with MAP_PRIVATE mappings created by their parent |
| 469 | * have no page reserves. This check ensures that reservations are |
| 470 | * not "stolen". The child may still get SIGKILLed |
| 471 | */ |
Mel Gorman | 7f09ca5 | 2008-07-23 21:27:58 -0700 | [diff] [blame] | 472 | if (!vma_has_reserves(vma) && |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 473 | h->free_huge_pages - h->resv_huge_pages == 0) |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 474 | goto err; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 475 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 476 | /* If reserves cannot be used, ensure enough pages are in the pool */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 477 | if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 478 | goto err;; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 479 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 480 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 481 | MAX_NR_ZONES - 1, nodemask) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 482 | if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) { |
| 483 | page = dequeue_huge_page_node(h, zone_to_nid(zone)); |
| 484 | if (page) { |
| 485 | if (!avoid_reserve) |
| 486 | decrement_hugepage_resv_vma(h, vma); |
| 487 | break; |
| 488 | } |
Andrew Morton | 3abf7af | 2007-07-19 01:49:08 -0700 | [diff] [blame] | 489 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | } |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 491 | err: |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 492 | mpol_cond_put(mpol); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 493 | put_mems_allowed(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | return page; |
| 495 | } |
| 496 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 497 | static void update_and_free_page(struct hstate *h, struct page *page) |
Adam Litke | 6af2acb | 2007-10-16 01:26:16 -0700 | [diff] [blame] | 498 | { |
| 499 | int i; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 500 | |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 501 | VM_BUG_ON(h->order >= MAX_ORDER); |
| 502 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 503 | h->nr_huge_pages--; |
| 504 | h->nr_huge_pages_node[page_to_nid(page)]--; |
| 505 | for (i = 0; i < pages_per_huge_page(h); i++) { |
Adam Litke | 6af2acb | 2007-10-16 01:26:16 -0700 | [diff] [blame] | 506 | page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 1 << PG_referenced | |
| 507 | 1 << PG_dirty | 1 << PG_active | 1 << PG_reserved | |
| 508 | 1 << PG_private | 1<< PG_writeback); |
| 509 | } |
| 510 | set_compound_page_dtor(page, NULL); |
| 511 | set_page_refcounted(page); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 512 | arch_release_hugepage(page); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 513 | __free_pages(page, huge_page_order(h)); |
Adam Litke | 6af2acb | 2007-10-16 01:26:16 -0700 | [diff] [blame] | 514 | } |
| 515 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 516 | struct hstate *size_to_hstate(unsigned long size) |
| 517 | { |
| 518 | struct hstate *h; |
| 519 | |
| 520 | for_each_hstate(h) { |
| 521 | if (huge_page_size(h) == size) |
| 522 | return h; |
| 523 | } |
| 524 | return NULL; |
| 525 | } |
| 526 | |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 527 | static void free_huge_page(struct page *page) |
| 528 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 529 | /* |
| 530 | * Can't pass hstate in here because it is called from the |
| 531 | * compound page destructor. |
| 532 | */ |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 533 | struct hstate *h = page_hstate(page); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 534 | int nid = page_to_nid(page); |
Adam Litke | c79fb75 | 2007-11-14 16:59:38 -0800 | [diff] [blame] | 535 | struct address_space *mapping; |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 536 | |
Adam Litke | c79fb75 | 2007-11-14 16:59:38 -0800 | [diff] [blame] | 537 | mapping = (struct address_space *) page_private(page); |
Andy Whitcroft | e5df70a | 2008-02-23 15:23:32 -0800 | [diff] [blame] | 538 | set_page_private(page, 0); |
Mel Gorman | 23be746 | 2010-04-23 13:17:56 -0400 | [diff] [blame] | 539 | page->mapping = NULL; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 540 | BUG_ON(page_count(page)); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 541 | BUG_ON(page_mapcount(page)); |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 542 | INIT_LIST_HEAD(&page->lru); |
| 543 | |
| 544 | spin_lock(&hugetlb_lock); |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 545 | if (h->surplus_huge_pages_node[nid] && huge_page_order(h) < MAX_ORDER) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 546 | update_and_free_page(h, page); |
| 547 | h->surplus_huge_pages--; |
| 548 | h->surplus_huge_pages_node[nid]--; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 549 | } else { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 550 | enqueue_huge_page(h, page); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 551 | } |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 552 | spin_unlock(&hugetlb_lock); |
Adam Litke | c79fb75 | 2007-11-14 16:59:38 -0800 | [diff] [blame] | 553 | if (mapping) |
Adam Litke | 9a119c0 | 2007-11-14 16:59:41 -0800 | [diff] [blame] | 554 | hugetlb_put_quota(mapping, 1); |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 555 | } |
| 556 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 557 | static void prep_new_huge_page(struct hstate *h, struct page *page, int nid) |
Andi Kleen | b7ba30c | 2008-07-23 21:27:40 -0700 | [diff] [blame] | 558 | { |
| 559 | set_compound_page_dtor(page, free_huge_page); |
| 560 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 561 | h->nr_huge_pages++; |
| 562 | h->nr_huge_pages_node[nid]++; |
Andi Kleen | b7ba30c | 2008-07-23 21:27:40 -0700 | [diff] [blame] | 563 | spin_unlock(&hugetlb_lock); |
| 564 | put_page(page); /* free it into the hugepage allocator */ |
| 565 | } |
| 566 | |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 567 | static void prep_compound_gigantic_page(struct page *page, unsigned long order) |
| 568 | { |
| 569 | int i; |
| 570 | int nr_pages = 1 << order; |
| 571 | struct page *p = page + 1; |
| 572 | |
| 573 | /* we rely on prep_new_huge_page to set the destructor */ |
| 574 | set_compound_order(page, order); |
| 575 | __SetPageHead(page); |
| 576 | for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { |
| 577 | __SetPageTail(p); |
| 578 | p->first_page = page; |
| 579 | } |
| 580 | } |
| 581 | |
| 582 | int PageHuge(struct page *page) |
| 583 | { |
| 584 | compound_page_dtor *dtor; |
| 585 | |
| 586 | if (!PageCompound(page)) |
| 587 | return 0; |
| 588 | |
| 589 | page = compound_head(page); |
| 590 | dtor = get_compound_page_dtor(page); |
| 591 | |
| 592 | return dtor == free_huge_page; |
| 593 | } |
| 594 | |
Naoya Horiguchi | 43131e1 | 2010-05-28 09:29:22 +0900 | [diff] [blame] | 595 | EXPORT_SYMBOL_GPL(PageHuge); |
| 596 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 597 | static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | struct page *page; |
Joe Jin | f96efd5 | 2007-07-15 23:38:12 -0700 | [diff] [blame] | 600 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 601 | if (h->order >= MAX_ORDER) |
| 602 | return NULL; |
| 603 | |
Mel Gorman | 6484eb3 | 2009-06-16 15:31:54 -0700 | [diff] [blame] | 604 | page = alloc_pages_exact_node(nid, |
Nishanth Aravamudan | 551883a | 2008-04-29 00:58:26 -0700 | [diff] [blame] | 605 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
| 606 | __GFP_REPEAT|__GFP_NOWARN, |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 607 | huge_page_order(h)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | if (page) { |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 609 | if (arch_prepare_hugepage(page)) { |
Gerald Schaefer | caff3a2 | 2008-08-12 15:08:38 -0700 | [diff] [blame] | 610 | __free_pages(page, huge_page_order(h)); |
Harvey Harrison | 7b8ee84 | 2008-04-28 14:13:19 -0700 | [diff] [blame] | 611 | return NULL; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 612 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 613 | prep_new_huge_page(h, page, nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | } |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 615 | |
| 616 | return page; |
| 617 | } |
| 618 | |
Andi Kleen | 5ced66c | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 619 | /* |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 620 | * common helper functions for hstate_next_node_to_{alloc|free}. |
| 621 | * We may have allocated or freed a huge page based on a different |
| 622 | * nodes_allowed previously, so h->next_node_to_{alloc|free} might |
| 623 | * be outside of *nodes_allowed. Ensure that we use an allowed |
| 624 | * node for alloc or free. |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 625 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 626 | static int next_node_allowed(int nid, nodemask_t *nodes_allowed) |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 627 | { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 628 | nid = next_node(nid, *nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 629 | if (nid == MAX_NUMNODES) |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 630 | nid = first_node(*nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 631 | VM_BUG_ON(nid >= MAX_NUMNODES); |
| 632 | |
| 633 | return nid; |
| 634 | } |
| 635 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 636 | static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed) |
Andi Kleen | 5ced66c | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 637 | { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 638 | if (!node_isset(nid, *nodes_allowed)) |
| 639 | nid = next_node_allowed(nid, nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 640 | return nid; |
Andi Kleen | 5ced66c | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 641 | } |
| 642 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 643 | /* |
| 644 | * returns the previously saved node ["this node"] from which to |
| 645 | * allocate a persistent huge page for the pool and advance the |
| 646 | * next node from which to allocate, handling wrap at end of node |
| 647 | * mask. |
| 648 | */ |
| 649 | static int hstate_next_node_to_alloc(struct hstate *h, |
| 650 | nodemask_t *nodes_allowed) |
| 651 | { |
| 652 | int nid; |
| 653 | |
| 654 | VM_BUG_ON(!nodes_allowed); |
| 655 | |
| 656 | nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); |
| 657 | h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); |
| 658 | |
| 659 | return nid; |
| 660 | } |
| 661 | |
| 662 | static int alloc_fresh_huge_page(struct hstate *h, nodemask_t *nodes_allowed) |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 663 | { |
| 664 | struct page *page; |
| 665 | int start_nid; |
| 666 | int next_nid; |
| 667 | int ret = 0; |
| 668 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 669 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 670 | next_nid = start_nid; |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 671 | |
| 672 | do { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 673 | page = alloc_fresh_huge_page_node(h, next_nid); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 674 | if (page) { |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 675 | ret = 1; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 676 | break; |
| 677 | } |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 678 | next_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 679 | } while (next_nid != start_nid); |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 680 | |
Adam Litke | 3b11630 | 2008-04-28 02:13:06 -0700 | [diff] [blame] | 681 | if (ret) |
| 682 | count_vm_event(HTLB_BUDDY_PGALLOC); |
| 683 | else |
| 684 | count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
| 685 | |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 686 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | } |
| 688 | |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 689 | /* |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 690 | * helper for free_pool_huge_page() - return the previously saved |
| 691 | * node ["this node"] from which to free a huge page. Advance the |
| 692 | * next node id whether or not we find a free huge page to free so |
| 693 | * that the next attempt to free addresses the next node. |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 694 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 695 | static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed) |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 696 | { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 697 | int nid; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 698 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 699 | VM_BUG_ON(!nodes_allowed); |
| 700 | |
| 701 | nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); |
| 702 | h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); |
| 703 | |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 704 | return nid; |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 705 | } |
| 706 | |
| 707 | /* |
| 708 | * Free huge page from pool from next node to free. |
| 709 | * Attempt to keep persistent huge pages more or less |
| 710 | * balanced over allowed nodes. |
| 711 | * Called with hugetlb_lock locked. |
| 712 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 713 | static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, |
| 714 | bool acct_surplus) |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 715 | { |
| 716 | int start_nid; |
| 717 | int next_nid; |
| 718 | int ret = 0; |
| 719 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 720 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 721 | next_nid = start_nid; |
| 722 | |
| 723 | do { |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 724 | /* |
| 725 | * If we're returning unused surplus pages, only examine |
| 726 | * nodes with surplus pages. |
| 727 | */ |
| 728 | if ((!acct_surplus || h->surplus_huge_pages_node[next_nid]) && |
| 729 | !list_empty(&h->hugepage_freelists[next_nid])) { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 730 | struct page *page = |
| 731 | list_entry(h->hugepage_freelists[next_nid].next, |
| 732 | struct page, lru); |
| 733 | list_del(&page->lru); |
| 734 | h->free_huge_pages--; |
| 735 | h->free_huge_pages_node[next_nid]--; |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 736 | if (acct_surplus) { |
| 737 | h->surplus_huge_pages--; |
| 738 | h->surplus_huge_pages_node[next_nid]--; |
| 739 | } |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 740 | update_and_free_page(h, page); |
| 741 | ret = 1; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 742 | break; |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 743 | } |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 744 | next_nid = hstate_next_node_to_free(h, nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 745 | } while (next_nid != start_nid); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 746 | |
| 747 | return ret; |
| 748 | } |
| 749 | |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 750 | static struct page *alloc_buddy_huge_page(struct hstate *h, int nid) |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 751 | { |
| 752 | struct page *page; |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 753 | unsigned int r_nid; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 754 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 755 | if (h->order >= MAX_ORDER) |
| 756 | return NULL; |
| 757 | |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 758 | /* |
| 759 | * Assume we will successfully allocate the surplus page to |
| 760 | * prevent racing processes from causing the surplus to exceed |
| 761 | * overcommit |
| 762 | * |
| 763 | * This however introduces a different race, where a process B |
| 764 | * tries to grow the static hugepage pool while alloc_pages() is |
| 765 | * called by process A. B will only examine the per-node |
| 766 | * counters in determining if surplus huge pages can be |
| 767 | * converted to normal huge pages in adjust_pool_surplus(). A |
| 768 | * won't be able to increment the per-node counter, until the |
| 769 | * lock is dropped by B, but B doesn't drop hugetlb_lock until |
| 770 | * no more huge pages can be converted from surplus to normal |
| 771 | * state (and doesn't try to convert again). Thus, we have a |
| 772 | * case where a surplus huge page exists, the pool is grown, and |
| 773 | * the surplus huge page still exists after, even though it |
| 774 | * should just have been converted to a normal huge page. This |
| 775 | * does not leak memory, though, as the hugepage will be freed |
| 776 | * once it is out of use. It also does not allow the counters to |
| 777 | * go out of whack in adjust_pool_surplus() as we don't modify |
| 778 | * the node values until we've gotten the hugepage and only the |
| 779 | * per-node value is checked there. |
| 780 | */ |
| 781 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 782 | if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 783 | spin_unlock(&hugetlb_lock); |
| 784 | return NULL; |
| 785 | } else { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 786 | h->nr_huge_pages++; |
| 787 | h->surplus_huge_pages++; |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 788 | } |
| 789 | spin_unlock(&hugetlb_lock); |
| 790 | |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 791 | if (nid == NUMA_NO_NODE) |
| 792 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP| |
| 793 | __GFP_REPEAT|__GFP_NOWARN, |
| 794 | huge_page_order(h)); |
| 795 | else |
| 796 | page = alloc_pages_exact_node(nid, |
| 797 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
| 798 | __GFP_REPEAT|__GFP_NOWARN, huge_page_order(h)); |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 799 | |
Gerald Schaefer | caff3a2 | 2008-08-12 15:08:38 -0700 | [diff] [blame] | 800 | if (page && arch_prepare_hugepage(page)) { |
| 801 | __free_pages(page, huge_page_order(h)); |
| 802 | return NULL; |
| 803 | } |
| 804 | |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 805 | spin_lock(&hugetlb_lock); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 806 | if (page) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 807 | r_nid = page_to_nid(page); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 808 | set_compound_page_dtor(page, free_huge_page); |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 809 | /* |
| 810 | * We incremented the global counters already |
| 811 | */ |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 812 | h->nr_huge_pages_node[r_nid]++; |
| 813 | h->surplus_huge_pages_node[r_nid]++; |
Adam Litke | 3b11630 | 2008-04-28 02:13:06 -0700 | [diff] [blame] | 814 | __count_vm_event(HTLB_BUDDY_PGALLOC); |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 815 | } else { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 816 | h->nr_huge_pages--; |
| 817 | h->surplus_huge_pages--; |
Adam Litke | 3b11630 | 2008-04-28 02:13:06 -0700 | [diff] [blame] | 818 | __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 819 | } |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 820 | spin_unlock(&hugetlb_lock); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 821 | |
| 822 | return page; |
| 823 | } |
| 824 | |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 825 | /* |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 826 | * This allocation function is useful in the context where vma is irrelevant. |
| 827 | * E.g. soft-offlining uses this function because it only cares physical |
| 828 | * address of error page. |
| 829 | */ |
| 830 | struct page *alloc_huge_page_node(struct hstate *h, int nid) |
| 831 | { |
| 832 | struct page *page; |
| 833 | |
| 834 | spin_lock(&hugetlb_lock); |
| 835 | page = dequeue_huge_page_node(h, nid); |
| 836 | spin_unlock(&hugetlb_lock); |
| 837 | |
| 838 | if (!page) |
| 839 | page = alloc_buddy_huge_page(h, nid); |
| 840 | |
| 841 | return page; |
| 842 | } |
| 843 | |
| 844 | /* |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 845 | * Increase the hugetlb pool such that it can accomodate a reservation |
| 846 | * of size 'delta'. |
| 847 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 848 | static int gather_surplus_pages(struct hstate *h, int delta) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 849 | { |
| 850 | struct list_head surplus_list; |
| 851 | struct page *page, *tmp; |
| 852 | int ret, i; |
| 853 | int needed, allocated; |
| 854 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 855 | needed = (h->resv_huge_pages + delta) - h->free_huge_pages; |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 856 | if (needed <= 0) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 857 | h->resv_huge_pages += delta; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 858 | return 0; |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 859 | } |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 860 | |
| 861 | allocated = 0; |
| 862 | INIT_LIST_HEAD(&surplus_list); |
| 863 | |
| 864 | ret = -ENOMEM; |
| 865 | retry: |
| 866 | spin_unlock(&hugetlb_lock); |
| 867 | for (i = 0; i < needed; i++) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 868 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 869 | if (!page) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 870 | /* |
| 871 | * We were not able to allocate enough pages to |
| 872 | * satisfy the entire reservation so we free what |
| 873 | * we've allocated so far. |
| 874 | */ |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 875 | goto free; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 876 | |
| 877 | list_add(&page->lru, &surplus_list); |
| 878 | } |
| 879 | allocated += needed; |
| 880 | |
| 881 | /* |
| 882 | * After retaking hugetlb_lock, we need to recalculate 'needed' |
| 883 | * because either resv_huge_pages or free_huge_pages may have changed. |
| 884 | */ |
| 885 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 886 | needed = (h->resv_huge_pages + delta) - |
| 887 | (h->free_huge_pages + allocated); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 888 | if (needed > 0) |
| 889 | goto retry; |
| 890 | |
| 891 | /* |
| 892 | * The surplus_list now contains _at_least_ the number of extra pages |
| 893 | * needed to accomodate the reservation. Add the appropriate number |
| 894 | * of pages to the hugetlb pool and free the extras back to the buddy |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 895 | * allocator. Commit the entire reservation here to prevent another |
| 896 | * process from stealing the pages as they are added to the pool but |
| 897 | * before they are reserved. |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 898 | */ |
| 899 | needed += allocated; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 900 | h->resv_huge_pages += delta; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 901 | ret = 0; |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 902 | |
| 903 | spin_unlock(&hugetlb_lock); |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 904 | /* Free the needed pages to the hugetlb pool */ |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 905 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 906 | if ((--needed) < 0) |
| 907 | break; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 908 | list_del(&page->lru); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 909 | /* |
| 910 | * This page is now managed by the hugetlb allocator and has |
| 911 | * no users -- drop the buddy allocator's reference. |
| 912 | */ |
| 913 | put_page_testzero(page); |
| 914 | VM_BUG_ON(page_count(page)); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 915 | enqueue_huge_page(h, page); |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 916 | } |
| 917 | |
| 918 | /* Free unnecessary surplus pages to the buddy allocator */ |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 919 | free: |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 920 | if (!list_empty(&surplus_list)) { |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 921 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
| 922 | list_del(&page->lru); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 923 | put_page(page); |
Adam Litke | af767cb | 2007-10-16 01:26:25 -0700 | [diff] [blame] | 924 | } |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 925 | } |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 926 | spin_lock(&hugetlb_lock); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 927 | |
| 928 | return ret; |
| 929 | } |
| 930 | |
| 931 | /* |
| 932 | * When releasing a hugetlb pool reservation, any surplus pages that were |
| 933 | * allocated to satisfy the reservation must be explicitly freed if they were |
| 934 | * never used. |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 935 | * Called with hugetlb_lock held. |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 936 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 937 | static void return_unused_surplus_pages(struct hstate *h, |
| 938 | unsigned long unused_resv_pages) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 939 | { |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 940 | unsigned long nr_pages; |
| 941 | |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 942 | /* Uncommit the reservation */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 943 | h->resv_huge_pages -= unused_resv_pages; |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 944 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 945 | /* Cannot return gigantic pages currently */ |
| 946 | if (h->order >= MAX_ORDER) |
| 947 | return; |
| 948 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 949 | nr_pages = min(unused_resv_pages, h->surplus_huge_pages); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 950 | |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 951 | /* |
| 952 | * We want to release as many surplus pages as possible, spread |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 953 | * evenly across all nodes with memory. Iterate across these nodes |
| 954 | * until we can no longer free unreserved surplus pages. This occurs |
| 955 | * when the nodes with surplus pages have no free pages. |
| 956 | * free_pool_huge_page() will balance the the freed pages across the |
| 957 | * on-line nodes with memory and will handle the hstate accounting. |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 958 | */ |
| 959 | while (nr_pages--) { |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 960 | if (!free_pool_huge_page(h, &node_states[N_HIGH_MEMORY], 1)) |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 961 | break; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 962 | } |
| 963 | } |
| 964 | |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 965 | /* |
| 966 | * Determine if the huge page at addr within the vma has an associated |
| 967 | * reservation. Where it does not we will need to logically increase |
| 968 | * reservation and actually increase quota before an allocation can occur. |
| 969 | * Where any new reservation would be required the reservation change is |
| 970 | * prepared, but not committed. Once the page has been quota'd allocated |
| 971 | * an instantiated the change should be committed via vma_commit_reservation. |
| 972 | * No action is required on failure. |
| 973 | */ |
Roel Kluin | e2f17d9 | 2009-03-31 15:23:15 -0700 | [diff] [blame] | 974 | static long vma_needs_reservation(struct hstate *h, |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 975 | struct vm_area_struct *vma, unsigned long addr) |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 976 | { |
| 977 | struct address_space *mapping = vma->vm_file->f_mapping; |
| 978 | struct inode *inode = mapping->host; |
| 979 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 980 | if (vma->vm_flags & VM_MAYSHARE) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 981 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 982 | return region_chg(&inode->i_mapping->private_list, |
| 983 | idx, idx + 1); |
| 984 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 985 | } else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
| 986 | return 1; |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 987 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 988 | } else { |
Roel Kluin | e2f17d9 | 2009-03-31 15:23:15 -0700 | [diff] [blame] | 989 | long err; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 990 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 991 | struct resv_map *reservations = vma_resv_map(vma); |
| 992 | |
| 993 | err = region_chg(&reservations->regions, idx, idx + 1); |
| 994 | if (err < 0) |
| 995 | return err; |
| 996 | return 0; |
| 997 | } |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 998 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 999 | static void vma_commit_reservation(struct hstate *h, |
| 1000 | struct vm_area_struct *vma, unsigned long addr) |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1001 | { |
| 1002 | struct address_space *mapping = vma->vm_file->f_mapping; |
| 1003 | struct inode *inode = mapping->host; |
| 1004 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 1005 | if (vma->vm_flags & VM_MAYSHARE) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1006 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1007 | region_add(&inode->i_mapping->private_list, idx, idx + 1); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1008 | |
| 1009 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1010 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1011 | struct resv_map *reservations = vma_resv_map(vma); |
| 1012 | |
| 1013 | /* Mark this page used in the map. */ |
| 1014 | region_add(&reservations->regions, idx, idx + 1); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1015 | } |
| 1016 | } |
| 1017 | |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 1018 | static struct page *alloc_huge_page(struct vm_area_struct *vma, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 1019 | unsigned long addr, int avoid_reserve) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1021 | struct hstate *h = hstate_vma(vma); |
Adam Litke | 348ea20 | 2007-11-14 16:59:37 -0800 | [diff] [blame] | 1022 | struct page *page; |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 1023 | struct address_space *mapping = vma->vm_file->f_mapping; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1024 | struct inode *inode = mapping->host; |
Roel Kluin | e2f17d9 | 2009-03-31 15:23:15 -0700 | [diff] [blame] | 1025 | long chg; |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 1026 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1027 | /* |
| 1028 | * Processes that did not create the mapping will have no reserves and |
| 1029 | * will not have accounted against quota. Check that the quota can be |
| 1030 | * made before satisfying the allocation |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1031 | * MAP_NORESERVE mappings may also need pages and quota allocated |
| 1032 | * if no reserve mapping overlaps. |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1033 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1034 | chg = vma_needs_reservation(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1035 | if (chg < 0) |
| 1036 | return ERR_PTR(chg); |
| 1037 | if (chg) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1038 | if (hugetlb_get_quota(inode->i_mapping, chg)) |
| 1039 | return ERR_PTR(-ENOSPC); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1040 | |
| 1041 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1042 | page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1043 | spin_unlock(&hugetlb_lock); |
| 1044 | |
| 1045 | if (!page) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 1046 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1047 | if (!page) { |
| 1048 | hugetlb_put_quota(inode->i_mapping, chg); |
Mel Gorman | 4a6018f | 2010-05-11 14:06:53 -0700 | [diff] [blame] | 1049 | return ERR_PTR(-VM_FAULT_SIGBUS); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1050 | } |
| 1051 | } |
| 1052 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1053 | set_page_private(page, (unsigned long) mapping); |
| 1054 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1055 | vma_commit_reservation(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1056 | |
Adam Litke | 90d8b7e | 2007-11-14 16:59:42 -0800 | [diff] [blame] | 1057 | return page; |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 1058 | } |
| 1059 | |
Cyrill Gorcunov | 91f4766 | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 1060 | int __weak alloc_bootmem_huge_page(struct hstate *h) |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1061 | { |
| 1062 | struct huge_bootmem_page *m; |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1063 | int nr_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1064 | |
| 1065 | while (nr_nodes) { |
| 1066 | void *addr; |
| 1067 | |
| 1068 | addr = __alloc_bootmem_node_nopanic( |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1069 | NODE_DATA(hstate_next_node_to_alloc(h, |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1070 | &node_states[N_HIGH_MEMORY])), |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1071 | huge_page_size(h), huge_page_size(h), 0); |
| 1072 | |
| 1073 | if (addr) { |
| 1074 | /* |
| 1075 | * Use the beginning of the huge page to store the |
| 1076 | * huge_bootmem_page struct (until gather_bootmem |
| 1077 | * puts them into the mem_map). |
| 1078 | */ |
| 1079 | m = addr; |
Cyrill Gorcunov | 91f4766 | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 1080 | goto found; |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1081 | } |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1082 | nr_nodes--; |
| 1083 | } |
| 1084 | return 0; |
| 1085 | |
| 1086 | found: |
| 1087 | BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1)); |
| 1088 | /* Put them into a private list first because mem_map is not up yet */ |
| 1089 | list_add(&m->list, &huge_boot_pages); |
| 1090 | m->hstate = h; |
| 1091 | return 1; |
| 1092 | } |
| 1093 | |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 1094 | static void prep_compound_huge_page(struct page *page, int order) |
| 1095 | { |
| 1096 | if (unlikely(order > (MAX_ORDER - 1))) |
| 1097 | prep_compound_gigantic_page(page, order); |
| 1098 | else |
| 1099 | prep_compound_page(page, order); |
| 1100 | } |
| 1101 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1102 | /* Put bootmem huge pages into the standard lists after mem_map is up */ |
| 1103 | static void __init gather_bootmem_prealloc(void) |
| 1104 | { |
| 1105 | struct huge_bootmem_page *m; |
| 1106 | |
| 1107 | list_for_each_entry(m, &huge_boot_pages, list) { |
| 1108 | struct page *page = virt_to_page(m); |
| 1109 | struct hstate *h = m->hstate; |
| 1110 | __ClearPageReserved(page); |
| 1111 | WARN_ON(page_count(page) != 1); |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 1112 | prep_compound_huge_page(page, h->order); |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1113 | prep_new_huge_page(h, page, page_to_nid(page)); |
| 1114 | } |
| 1115 | } |
| 1116 | |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1117 | static void __init hugetlb_hstate_alloc_pages(struct hstate *h) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | { |
| 1119 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1121 | for (i = 0; i < h->max_huge_pages; ++i) { |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1122 | if (h->order >= MAX_ORDER) { |
| 1123 | if (!alloc_bootmem_huge_page(h)) |
| 1124 | break; |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1125 | } else if (!alloc_fresh_huge_page(h, |
| 1126 | &node_states[N_HIGH_MEMORY])) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1128 | } |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1129 | h->max_huge_pages = i; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1130 | } |
| 1131 | |
| 1132 | static void __init hugetlb_init_hstates(void) |
| 1133 | { |
| 1134 | struct hstate *h; |
| 1135 | |
| 1136 | for_each_hstate(h) { |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1137 | /* oversize hugepages were init'ed in early boot */ |
| 1138 | if (h->order < MAX_ORDER) |
| 1139 | hugetlb_hstate_alloc_pages(h); |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1140 | } |
| 1141 | } |
| 1142 | |
Andi Kleen | 4abd32d | 2008-07-23 21:27:49 -0700 | [diff] [blame] | 1143 | static char * __init memfmt(char *buf, unsigned long n) |
| 1144 | { |
| 1145 | if (n >= (1UL << 30)) |
| 1146 | sprintf(buf, "%lu GB", n >> 30); |
| 1147 | else if (n >= (1UL << 20)) |
| 1148 | sprintf(buf, "%lu MB", n >> 20); |
| 1149 | else |
| 1150 | sprintf(buf, "%lu KB", n >> 10); |
| 1151 | return buf; |
| 1152 | } |
| 1153 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1154 | static void __init report_hugepages(void) |
| 1155 | { |
| 1156 | struct hstate *h; |
| 1157 | |
| 1158 | for_each_hstate(h) { |
Andi Kleen | 4abd32d | 2008-07-23 21:27:49 -0700 | [diff] [blame] | 1159 | char buf[32]; |
| 1160 | printk(KERN_INFO "HugeTLB registered %s page size, " |
| 1161 | "pre-allocated %ld pages\n", |
| 1162 | memfmt(buf, huge_page_size(h)), |
| 1163 | h->free_huge_pages); |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1164 | } |
| 1165 | } |
| 1166 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1167 | #ifdef CONFIG_HIGHMEM |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1168 | static void try_to_free_low(struct hstate *h, unsigned long count, |
| 1169 | nodemask_t *nodes_allowed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | { |
Christoph Lameter | 4415cc8 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 1171 | int i; |
| 1172 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1173 | if (h->order >= MAX_ORDER) |
| 1174 | return; |
| 1175 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1176 | for_each_node_mask(i, *nodes_allowed) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | struct page *page, *next; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1178 | struct list_head *freel = &h->hugepage_freelists[i]; |
| 1179 | list_for_each_entry_safe(page, next, freel, lru) { |
| 1180 | if (count >= h->nr_huge_pages) |
Adam Litke | 6b0c880 | 2007-10-16 01:26:23 -0700 | [diff] [blame] | 1181 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1182 | if (PageHighMem(page)) |
| 1183 | continue; |
| 1184 | list_del(&page->lru); |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1185 | update_and_free_page(h, page); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1186 | h->free_huge_pages--; |
| 1187 | h->free_huge_pages_node[page_to_nid(page)]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | } |
| 1189 | } |
| 1190 | } |
| 1191 | #else |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1192 | static inline void try_to_free_low(struct hstate *h, unsigned long count, |
| 1193 | nodemask_t *nodes_allowed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1194 | { |
| 1195 | } |
| 1196 | #endif |
| 1197 | |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1198 | /* |
| 1199 | * Increment or decrement surplus_huge_pages. Keep node-specific counters |
| 1200 | * balanced by operating on them in a round-robin fashion. |
| 1201 | * Returns 1 if an adjustment was made. |
| 1202 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1203 | static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed, |
| 1204 | int delta) |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1205 | { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1206 | int start_nid, next_nid; |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1207 | int ret = 0; |
| 1208 | |
| 1209 | VM_BUG_ON(delta != -1 && delta != 1); |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1210 | |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1211 | if (delta < 0) |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1212 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1213 | else |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1214 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1215 | next_nid = start_nid; |
| 1216 | |
| 1217 | do { |
| 1218 | int nid = next_nid; |
| 1219 | if (delta < 0) { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1220 | /* |
| 1221 | * To shrink on this node, there must be a surplus page |
| 1222 | */ |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1223 | if (!h->surplus_huge_pages_node[nid]) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1224 | next_nid = hstate_next_node_to_alloc(h, |
| 1225 | nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1226 | continue; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1227 | } |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1228 | } |
| 1229 | if (delta > 0) { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1230 | /* |
| 1231 | * Surplus cannot exceed the total number of pages |
| 1232 | */ |
| 1233 | if (h->surplus_huge_pages_node[nid] >= |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1234 | h->nr_huge_pages_node[nid]) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1235 | next_nid = hstate_next_node_to_free(h, |
| 1236 | nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1237 | continue; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1238 | } |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1239 | } |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1240 | |
| 1241 | h->surplus_huge_pages += delta; |
| 1242 | h->surplus_huge_pages_node[nid] += delta; |
| 1243 | ret = 1; |
| 1244 | break; |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1245 | } while (next_nid != start_nid); |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1246 | |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1247 | return ret; |
| 1248 | } |
| 1249 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1250 | #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages) |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1251 | static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count, |
| 1252 | nodemask_t *nodes_allowed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1253 | { |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1254 | unsigned long min_count, ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1256 | if (h->order >= MAX_ORDER) |
| 1257 | return h->max_huge_pages; |
| 1258 | |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1259 | /* |
| 1260 | * Increase the pool size |
| 1261 | * First take pages out of surplus state. Then make up the |
| 1262 | * remaining difference by allocating fresh huge pages. |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 1263 | * |
| 1264 | * We might race with alloc_buddy_huge_page() here and be unable |
| 1265 | * to convert a surplus huge page to a normal huge page. That is |
| 1266 | * not critical, though, it just means the overall size of the |
| 1267 | * pool might be one hugepage larger than it needs to be, but |
| 1268 | * within all the constraints specified by the sysctls. |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1269 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1271 | while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1272 | if (!adjust_pool_surplus(h, nodes_allowed, -1)) |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1273 | break; |
| 1274 | } |
| 1275 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1276 | while (count > persistent_huge_pages(h)) { |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1277 | /* |
| 1278 | * If this allocation races such that we no longer need the |
| 1279 | * page, free_huge_page will handle it by freeing the page |
| 1280 | * and reducing the surplus. |
| 1281 | */ |
| 1282 | spin_unlock(&hugetlb_lock); |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1283 | ret = alloc_fresh_huge_page(h, nodes_allowed); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1284 | spin_lock(&hugetlb_lock); |
| 1285 | if (!ret) |
| 1286 | goto out; |
| 1287 | |
Mel Gorman | 536240f2 | 2009-12-14 17:59:56 -0800 | [diff] [blame] | 1288 | /* Bail for signals. Probably ctrl-c from user */ |
| 1289 | if (signal_pending(current)) |
| 1290 | goto out; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1291 | } |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1292 | |
| 1293 | /* |
| 1294 | * Decrease the pool size |
| 1295 | * First return free pages to the buddy allocator (being careful |
| 1296 | * to keep enough around to satisfy reservations). Then place |
| 1297 | * pages into surplus state as needed so the pool will shrink |
| 1298 | * to the desired size as pages become free. |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 1299 | * |
| 1300 | * By placing pages into the surplus state independent of the |
| 1301 | * overcommit value, we are allowing the surplus pool size to |
| 1302 | * exceed overcommit. There are few sane options here. Since |
| 1303 | * alloc_buddy_huge_page() is checking the global counter, |
| 1304 | * though, we'll note that we're not allowed to exceed surplus |
| 1305 | * and won't grow the pool anywhere else. Not until one of the |
| 1306 | * sysctls are changed, or the surplus pages go out of use. |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1307 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1308 | min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; |
Adam Litke | 6b0c880 | 2007-10-16 01:26:23 -0700 | [diff] [blame] | 1309 | min_count = max(count, min_count); |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1310 | try_to_free_low(h, min_count, nodes_allowed); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1311 | while (min_count < persistent_huge_pages(h)) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1312 | if (!free_pool_huge_page(h, nodes_allowed, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1313 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1314 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1315 | while (count < persistent_huge_pages(h)) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1316 | if (!adjust_pool_surplus(h, nodes_allowed, 1)) |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1317 | break; |
| 1318 | } |
| 1319 | out: |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1320 | ret = persistent_huge_pages(h); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | spin_unlock(&hugetlb_lock); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1322 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1323 | } |
| 1324 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1325 | #define HSTATE_ATTR_RO(_name) \ |
| 1326 | static struct kobj_attribute _name##_attr = __ATTR_RO(_name) |
| 1327 | |
| 1328 | #define HSTATE_ATTR(_name) \ |
| 1329 | static struct kobj_attribute _name##_attr = \ |
| 1330 | __ATTR(_name, 0644, _name##_show, _name##_store) |
| 1331 | |
| 1332 | static struct kobject *hugepages_kobj; |
| 1333 | static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; |
| 1334 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1335 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp); |
| 1336 | |
| 1337 | static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1338 | { |
| 1339 | int i; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1340 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1341 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1342 | if (hstate_kobjs[i] == kobj) { |
| 1343 | if (nidp) |
| 1344 | *nidp = NUMA_NO_NODE; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1345 | return &hstates[i]; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1346 | } |
| 1347 | |
| 1348 | return kobj_to_node_hstate(kobj, nidp); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1349 | } |
| 1350 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1351 | static ssize_t nr_hugepages_show_common(struct kobject *kobj, |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1352 | struct kobj_attribute *attr, char *buf) |
| 1353 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1354 | struct hstate *h; |
| 1355 | unsigned long nr_huge_pages; |
| 1356 | int nid; |
| 1357 | |
| 1358 | h = kobj_to_hstate(kobj, &nid); |
| 1359 | if (nid == NUMA_NO_NODE) |
| 1360 | nr_huge_pages = h->nr_huge_pages; |
| 1361 | else |
| 1362 | nr_huge_pages = h->nr_huge_pages_node[nid]; |
| 1363 | |
| 1364 | return sprintf(buf, "%lu\n", nr_huge_pages); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1365 | } |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1366 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1367 | static ssize_t nr_hugepages_store_common(bool obey_mempolicy, |
| 1368 | struct kobject *kobj, struct kobj_attribute *attr, |
| 1369 | const char *buf, size_t len) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1370 | { |
| 1371 | int err; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1372 | int nid; |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1373 | unsigned long count; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1374 | struct hstate *h; |
David Rientjes | bad44b5 | 2009-12-14 17:58:38 -0800 | [diff] [blame] | 1375 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1376 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1377 | err = strict_strtoul(buf, 10, &count); |
Eric B Munson | 73ae31e | 2011-01-13 15:47:28 -0800 | [diff] [blame] | 1378 | if (err) |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1379 | goto out; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1380 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1381 | h = kobj_to_hstate(kobj, &nid); |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1382 | if (h->order >= MAX_ORDER) { |
| 1383 | err = -EINVAL; |
| 1384 | goto out; |
| 1385 | } |
| 1386 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1387 | if (nid == NUMA_NO_NODE) { |
| 1388 | /* |
| 1389 | * global hstate attribute |
| 1390 | */ |
| 1391 | if (!(obey_mempolicy && |
| 1392 | init_nodemask_of_mempolicy(nodes_allowed))) { |
| 1393 | NODEMASK_FREE(nodes_allowed); |
| 1394 | nodes_allowed = &node_states[N_HIGH_MEMORY]; |
| 1395 | } |
| 1396 | } else if (nodes_allowed) { |
| 1397 | /* |
| 1398 | * per node hstate attribute: adjust count to global, |
| 1399 | * but restrict alloc/free to the specified node. |
| 1400 | */ |
| 1401 | count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; |
| 1402 | init_nodemask_of_node(nodes_allowed, nid); |
| 1403 | } else |
| 1404 | nodes_allowed = &node_states[N_HIGH_MEMORY]; |
| 1405 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1406 | h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1407 | |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1408 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1409 | NODEMASK_FREE(nodes_allowed); |
| 1410 | |
| 1411 | return len; |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1412 | out: |
| 1413 | NODEMASK_FREE(nodes_allowed); |
| 1414 | return err; |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1415 | } |
| 1416 | |
| 1417 | static ssize_t nr_hugepages_show(struct kobject *kobj, |
| 1418 | struct kobj_attribute *attr, char *buf) |
| 1419 | { |
| 1420 | return nr_hugepages_show_common(kobj, attr, buf); |
| 1421 | } |
| 1422 | |
| 1423 | static ssize_t nr_hugepages_store(struct kobject *kobj, |
| 1424 | struct kobj_attribute *attr, const char *buf, size_t len) |
| 1425 | { |
| 1426 | return nr_hugepages_store_common(false, kobj, attr, buf, len); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1427 | } |
| 1428 | HSTATE_ATTR(nr_hugepages); |
| 1429 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1430 | #ifdef CONFIG_NUMA |
| 1431 | |
| 1432 | /* |
| 1433 | * hstate attribute for optionally mempolicy-based constraint on persistent |
| 1434 | * huge page alloc/free. |
| 1435 | */ |
| 1436 | static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj, |
| 1437 | struct kobj_attribute *attr, char *buf) |
| 1438 | { |
| 1439 | return nr_hugepages_show_common(kobj, attr, buf); |
| 1440 | } |
| 1441 | |
| 1442 | static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj, |
| 1443 | struct kobj_attribute *attr, const char *buf, size_t len) |
| 1444 | { |
| 1445 | return nr_hugepages_store_common(true, kobj, attr, buf, len); |
| 1446 | } |
| 1447 | HSTATE_ATTR(nr_hugepages_mempolicy); |
| 1448 | #endif |
| 1449 | |
| 1450 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1451 | static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj, |
| 1452 | struct kobj_attribute *attr, char *buf) |
| 1453 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1454 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1455 | return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); |
| 1456 | } |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1457 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1458 | static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj, |
| 1459 | struct kobj_attribute *attr, const char *buf, size_t count) |
| 1460 | { |
| 1461 | int err; |
| 1462 | unsigned long input; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1463 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1464 | |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1465 | if (h->order >= MAX_ORDER) |
| 1466 | return -EINVAL; |
| 1467 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1468 | err = strict_strtoul(buf, 10, &input); |
| 1469 | if (err) |
Eric B Munson | 73ae31e | 2011-01-13 15:47:28 -0800 | [diff] [blame] | 1470 | return err; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1471 | |
| 1472 | spin_lock(&hugetlb_lock); |
| 1473 | h->nr_overcommit_huge_pages = input; |
| 1474 | spin_unlock(&hugetlb_lock); |
| 1475 | |
| 1476 | return count; |
| 1477 | } |
| 1478 | HSTATE_ATTR(nr_overcommit_hugepages); |
| 1479 | |
| 1480 | static ssize_t free_hugepages_show(struct kobject *kobj, |
| 1481 | struct kobj_attribute *attr, char *buf) |
| 1482 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1483 | struct hstate *h; |
| 1484 | unsigned long free_huge_pages; |
| 1485 | int nid; |
| 1486 | |
| 1487 | h = kobj_to_hstate(kobj, &nid); |
| 1488 | if (nid == NUMA_NO_NODE) |
| 1489 | free_huge_pages = h->free_huge_pages; |
| 1490 | else |
| 1491 | free_huge_pages = h->free_huge_pages_node[nid]; |
| 1492 | |
| 1493 | return sprintf(buf, "%lu\n", free_huge_pages); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1494 | } |
| 1495 | HSTATE_ATTR_RO(free_hugepages); |
| 1496 | |
| 1497 | static ssize_t resv_hugepages_show(struct kobject *kobj, |
| 1498 | struct kobj_attribute *attr, char *buf) |
| 1499 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1500 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1501 | return sprintf(buf, "%lu\n", h->resv_huge_pages); |
| 1502 | } |
| 1503 | HSTATE_ATTR_RO(resv_hugepages); |
| 1504 | |
| 1505 | static ssize_t surplus_hugepages_show(struct kobject *kobj, |
| 1506 | struct kobj_attribute *attr, char *buf) |
| 1507 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1508 | struct hstate *h; |
| 1509 | unsigned long surplus_huge_pages; |
| 1510 | int nid; |
| 1511 | |
| 1512 | h = kobj_to_hstate(kobj, &nid); |
| 1513 | if (nid == NUMA_NO_NODE) |
| 1514 | surplus_huge_pages = h->surplus_huge_pages; |
| 1515 | else |
| 1516 | surplus_huge_pages = h->surplus_huge_pages_node[nid]; |
| 1517 | |
| 1518 | return sprintf(buf, "%lu\n", surplus_huge_pages); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1519 | } |
| 1520 | HSTATE_ATTR_RO(surplus_hugepages); |
| 1521 | |
| 1522 | static struct attribute *hstate_attrs[] = { |
| 1523 | &nr_hugepages_attr.attr, |
| 1524 | &nr_overcommit_hugepages_attr.attr, |
| 1525 | &free_hugepages_attr.attr, |
| 1526 | &resv_hugepages_attr.attr, |
| 1527 | &surplus_hugepages_attr.attr, |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1528 | #ifdef CONFIG_NUMA |
| 1529 | &nr_hugepages_mempolicy_attr.attr, |
| 1530 | #endif |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1531 | NULL, |
| 1532 | }; |
| 1533 | |
| 1534 | static struct attribute_group hstate_attr_group = { |
| 1535 | .attrs = hstate_attrs, |
| 1536 | }; |
| 1537 | |
Jeff Mahoney | 094e953 | 2010-02-02 13:44:14 -0800 | [diff] [blame] | 1538 | static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent, |
| 1539 | struct kobject **hstate_kobjs, |
| 1540 | struct attribute_group *hstate_attr_group) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1541 | { |
| 1542 | int retval; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1543 | int hi = h - hstates; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1544 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1545 | hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); |
| 1546 | if (!hstate_kobjs[hi]) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1547 | return -ENOMEM; |
| 1548 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1549 | retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1550 | if (retval) |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1551 | kobject_put(hstate_kobjs[hi]); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1552 | |
| 1553 | return retval; |
| 1554 | } |
| 1555 | |
| 1556 | static void __init hugetlb_sysfs_init(void) |
| 1557 | { |
| 1558 | struct hstate *h; |
| 1559 | int err; |
| 1560 | |
| 1561 | hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj); |
| 1562 | if (!hugepages_kobj) |
| 1563 | return; |
| 1564 | |
| 1565 | for_each_hstate(h) { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1566 | err = hugetlb_sysfs_add_hstate(h, hugepages_kobj, |
| 1567 | hstate_kobjs, &hstate_attr_group); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1568 | if (err) |
| 1569 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s", |
| 1570 | h->name); |
| 1571 | } |
| 1572 | } |
| 1573 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1574 | #ifdef CONFIG_NUMA |
| 1575 | |
| 1576 | /* |
| 1577 | * node_hstate/s - associate per node hstate attributes, via their kobjects, |
| 1578 | * with node sysdevs in node_devices[] using a parallel array. The array |
| 1579 | * index of a node sysdev or _hstate == node id. |
| 1580 | * This is here to avoid any static dependency of the node sysdev driver, in |
| 1581 | * the base kernel, on the hugetlb module. |
| 1582 | */ |
| 1583 | struct node_hstate { |
| 1584 | struct kobject *hugepages_kobj; |
| 1585 | struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; |
| 1586 | }; |
| 1587 | struct node_hstate node_hstates[MAX_NUMNODES]; |
| 1588 | |
| 1589 | /* |
| 1590 | * A subset of global hstate attributes for node sysdevs |
| 1591 | */ |
| 1592 | static struct attribute *per_node_hstate_attrs[] = { |
| 1593 | &nr_hugepages_attr.attr, |
| 1594 | &free_hugepages_attr.attr, |
| 1595 | &surplus_hugepages_attr.attr, |
| 1596 | NULL, |
| 1597 | }; |
| 1598 | |
| 1599 | static struct attribute_group per_node_hstate_attr_group = { |
| 1600 | .attrs = per_node_hstate_attrs, |
| 1601 | }; |
| 1602 | |
| 1603 | /* |
| 1604 | * kobj_to_node_hstate - lookup global hstate for node sysdev hstate attr kobj. |
| 1605 | * Returns node id via non-NULL nidp. |
| 1606 | */ |
| 1607 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) |
| 1608 | { |
| 1609 | int nid; |
| 1610 | |
| 1611 | for (nid = 0; nid < nr_node_ids; nid++) { |
| 1612 | struct node_hstate *nhs = &node_hstates[nid]; |
| 1613 | int i; |
| 1614 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
| 1615 | if (nhs->hstate_kobjs[i] == kobj) { |
| 1616 | if (nidp) |
| 1617 | *nidp = nid; |
| 1618 | return &hstates[i]; |
| 1619 | } |
| 1620 | } |
| 1621 | |
| 1622 | BUG(); |
| 1623 | return NULL; |
| 1624 | } |
| 1625 | |
| 1626 | /* |
| 1627 | * Unregister hstate attributes from a single node sysdev. |
| 1628 | * No-op if no hstate attributes attached. |
| 1629 | */ |
| 1630 | void hugetlb_unregister_node(struct node *node) |
| 1631 | { |
| 1632 | struct hstate *h; |
| 1633 | struct node_hstate *nhs = &node_hstates[node->sysdev.id]; |
| 1634 | |
| 1635 | if (!nhs->hugepages_kobj) |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1636 | return; /* no hstate attributes */ |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1637 | |
| 1638 | for_each_hstate(h) |
| 1639 | if (nhs->hstate_kobjs[h - hstates]) { |
| 1640 | kobject_put(nhs->hstate_kobjs[h - hstates]); |
| 1641 | nhs->hstate_kobjs[h - hstates] = NULL; |
| 1642 | } |
| 1643 | |
| 1644 | kobject_put(nhs->hugepages_kobj); |
| 1645 | nhs->hugepages_kobj = NULL; |
| 1646 | } |
| 1647 | |
| 1648 | /* |
| 1649 | * hugetlb module exit: unregister hstate attributes from node sysdevs |
| 1650 | * that have them. |
| 1651 | */ |
| 1652 | static void hugetlb_unregister_all_nodes(void) |
| 1653 | { |
| 1654 | int nid; |
| 1655 | |
| 1656 | /* |
| 1657 | * disable node sysdev registrations. |
| 1658 | */ |
| 1659 | register_hugetlbfs_with_node(NULL, NULL); |
| 1660 | |
| 1661 | /* |
| 1662 | * remove hstate attributes from any nodes that have them. |
| 1663 | */ |
| 1664 | for (nid = 0; nid < nr_node_ids; nid++) |
| 1665 | hugetlb_unregister_node(&node_devices[nid]); |
| 1666 | } |
| 1667 | |
| 1668 | /* |
| 1669 | * Register hstate attributes for a single node sysdev. |
| 1670 | * No-op if attributes already registered. |
| 1671 | */ |
| 1672 | void hugetlb_register_node(struct node *node) |
| 1673 | { |
| 1674 | struct hstate *h; |
| 1675 | struct node_hstate *nhs = &node_hstates[node->sysdev.id]; |
| 1676 | int err; |
| 1677 | |
| 1678 | if (nhs->hugepages_kobj) |
| 1679 | return; /* already allocated */ |
| 1680 | |
| 1681 | nhs->hugepages_kobj = kobject_create_and_add("hugepages", |
| 1682 | &node->sysdev.kobj); |
| 1683 | if (!nhs->hugepages_kobj) |
| 1684 | return; |
| 1685 | |
| 1686 | for_each_hstate(h) { |
| 1687 | err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, |
| 1688 | nhs->hstate_kobjs, |
| 1689 | &per_node_hstate_attr_group); |
| 1690 | if (err) { |
| 1691 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s" |
| 1692 | " for node %d\n", |
| 1693 | h->name, node->sysdev.id); |
| 1694 | hugetlb_unregister_node(node); |
| 1695 | break; |
| 1696 | } |
| 1697 | } |
| 1698 | } |
| 1699 | |
| 1700 | /* |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1701 | * hugetlb init time: register hstate attributes for all registered node |
| 1702 | * sysdevs of nodes that have memory. All on-line nodes should have |
| 1703 | * registered their associated sysdev by this time. |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1704 | */ |
| 1705 | static void hugetlb_register_all_nodes(void) |
| 1706 | { |
| 1707 | int nid; |
| 1708 | |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1709 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1710 | struct node *node = &node_devices[nid]; |
| 1711 | if (node->sysdev.id == nid) |
| 1712 | hugetlb_register_node(node); |
| 1713 | } |
| 1714 | |
| 1715 | /* |
| 1716 | * Let the node sysdev driver know we're here so it can |
| 1717 | * [un]register hstate attributes on node hotplug. |
| 1718 | */ |
| 1719 | register_hugetlbfs_with_node(hugetlb_register_node, |
| 1720 | hugetlb_unregister_node); |
| 1721 | } |
| 1722 | #else /* !CONFIG_NUMA */ |
| 1723 | |
| 1724 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) |
| 1725 | { |
| 1726 | BUG(); |
| 1727 | if (nidp) |
| 1728 | *nidp = -1; |
| 1729 | return NULL; |
| 1730 | } |
| 1731 | |
| 1732 | static void hugetlb_unregister_all_nodes(void) { } |
| 1733 | |
| 1734 | static void hugetlb_register_all_nodes(void) { } |
| 1735 | |
| 1736 | #endif |
| 1737 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1738 | static void __exit hugetlb_exit(void) |
| 1739 | { |
| 1740 | struct hstate *h; |
| 1741 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1742 | hugetlb_unregister_all_nodes(); |
| 1743 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1744 | for_each_hstate(h) { |
| 1745 | kobject_put(hstate_kobjs[h - hstates]); |
| 1746 | } |
| 1747 | |
| 1748 | kobject_put(hugepages_kobj); |
| 1749 | } |
| 1750 | module_exit(hugetlb_exit); |
| 1751 | |
| 1752 | static int __init hugetlb_init(void) |
| 1753 | { |
Benjamin Herrenschmidt | 0ef89d2 | 2008-07-31 00:07:30 -0700 | [diff] [blame] | 1754 | /* Some platform decide whether they support huge pages at boot |
| 1755 | * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when |
| 1756 | * there is no such support |
| 1757 | */ |
| 1758 | if (HPAGE_SHIFT == 0) |
| 1759 | return 0; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1760 | |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1761 | if (!size_to_hstate(default_hstate_size)) { |
| 1762 | default_hstate_size = HPAGE_SIZE; |
| 1763 | if (!size_to_hstate(default_hstate_size)) |
| 1764 | hugetlb_add_hstate(HUGETLB_PAGE_ORDER); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1765 | } |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1766 | default_hstate_idx = size_to_hstate(default_hstate_size) - hstates; |
| 1767 | if (default_hstate_max_huge_pages) |
| 1768 | default_hstate.max_huge_pages = default_hstate_max_huge_pages; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1769 | |
| 1770 | hugetlb_init_hstates(); |
| 1771 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1772 | gather_bootmem_prealloc(); |
| 1773 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1774 | report_hugepages(); |
| 1775 | |
| 1776 | hugetlb_sysfs_init(); |
| 1777 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1778 | hugetlb_register_all_nodes(); |
| 1779 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1780 | return 0; |
| 1781 | } |
| 1782 | module_init(hugetlb_init); |
| 1783 | |
| 1784 | /* Should be called on processing a hugepagesz=... option */ |
| 1785 | void __init hugetlb_add_hstate(unsigned order) |
| 1786 | { |
| 1787 | struct hstate *h; |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1788 | unsigned long i; |
| 1789 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1790 | if (size_to_hstate(PAGE_SIZE << order)) { |
| 1791 | printk(KERN_WARNING "hugepagesz= specified twice, ignoring\n"); |
| 1792 | return; |
| 1793 | } |
| 1794 | BUG_ON(max_hstate >= HUGE_MAX_HSTATE); |
| 1795 | BUG_ON(order == 0); |
| 1796 | h = &hstates[max_hstate++]; |
| 1797 | h->order = order; |
| 1798 | h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1799 | h->nr_huge_pages = 0; |
| 1800 | h->free_huge_pages = 0; |
| 1801 | for (i = 0; i < MAX_NUMNODES; ++i) |
| 1802 | INIT_LIST_HEAD(&h->hugepage_freelists[i]); |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1803 | h->next_nid_to_alloc = first_node(node_states[N_HIGH_MEMORY]); |
| 1804 | h->next_nid_to_free = first_node(node_states[N_HIGH_MEMORY]); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1805 | snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", |
| 1806 | huge_page_size(h)/1024); |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1807 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1808 | parsed_hstate = h; |
| 1809 | } |
| 1810 | |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1811 | static int __init hugetlb_nrpages_setup(char *s) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1812 | { |
| 1813 | unsigned long *mhp; |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1814 | static unsigned long *last_mhp; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1815 | |
| 1816 | /* |
| 1817 | * !max_hstate means we haven't parsed a hugepagesz= parameter yet, |
| 1818 | * so this hugepages= parameter goes to the "default hstate". |
| 1819 | */ |
| 1820 | if (!max_hstate) |
| 1821 | mhp = &default_hstate_max_huge_pages; |
| 1822 | else |
| 1823 | mhp = &parsed_hstate->max_huge_pages; |
| 1824 | |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1825 | if (mhp == last_mhp) { |
| 1826 | printk(KERN_WARNING "hugepages= specified twice without " |
| 1827 | "interleaving hugepagesz=, ignoring\n"); |
| 1828 | return 1; |
| 1829 | } |
| 1830 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1831 | if (sscanf(s, "%lu", mhp) <= 0) |
| 1832 | *mhp = 0; |
| 1833 | |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1834 | /* |
| 1835 | * Global state is always initialized later in hugetlb_init. |
| 1836 | * But we need to allocate >= MAX_ORDER hstates here early to still |
| 1837 | * use the bootmem allocator. |
| 1838 | */ |
| 1839 | if (max_hstate && parsed_hstate->order >= MAX_ORDER) |
| 1840 | hugetlb_hstate_alloc_pages(parsed_hstate); |
| 1841 | |
| 1842 | last_mhp = mhp; |
| 1843 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1844 | return 1; |
| 1845 | } |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1846 | __setup("hugepages=", hugetlb_nrpages_setup); |
| 1847 | |
| 1848 | static int __init hugetlb_default_setup(char *s) |
| 1849 | { |
| 1850 | default_hstate_size = memparse(s, &s); |
| 1851 | return 1; |
| 1852 | } |
| 1853 | __setup("default_hugepagesz=", hugetlb_default_setup); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1854 | |
Nishanth Aravamudan | 8a21346 | 2008-07-25 19:44:37 -0700 | [diff] [blame] | 1855 | static unsigned int cpuset_mems_nr(unsigned int *array) |
| 1856 | { |
| 1857 | int node; |
| 1858 | unsigned int nr = 0; |
| 1859 | |
| 1860 | for_each_node_mask(node, cpuset_current_mems_allowed) |
| 1861 | nr += array[node]; |
| 1862 | |
| 1863 | return nr; |
| 1864 | } |
| 1865 | |
| 1866 | #ifdef CONFIG_SYSCTL |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1867 | static int hugetlb_sysctl_handler_common(bool obey_mempolicy, |
| 1868 | struct ctl_table *table, int write, |
| 1869 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1870 | { |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1871 | struct hstate *h = &default_hstate; |
| 1872 | unsigned long tmp; |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1873 | int ret; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1874 | |
| 1875 | if (!write) |
| 1876 | tmp = h->max_huge_pages; |
| 1877 | |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1878 | if (write && h->order >= MAX_ORDER) |
| 1879 | return -EINVAL; |
| 1880 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1881 | table->data = &tmp; |
| 1882 | table->maxlen = sizeof(unsigned long); |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1883 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
| 1884 | if (ret) |
| 1885 | goto out; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1886 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1887 | if (write) { |
David Rientjes | bad44b5 | 2009-12-14 17:58:38 -0800 | [diff] [blame] | 1888 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, |
| 1889 | GFP_KERNEL | __GFP_NORETRY); |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1890 | if (!(obey_mempolicy && |
| 1891 | init_nodemask_of_mempolicy(nodes_allowed))) { |
| 1892 | NODEMASK_FREE(nodes_allowed); |
| 1893 | nodes_allowed = &node_states[N_HIGH_MEMORY]; |
| 1894 | } |
| 1895 | h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed); |
| 1896 | |
| 1897 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) |
| 1898 | NODEMASK_FREE(nodes_allowed); |
| 1899 | } |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1900 | out: |
| 1901 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1902 | } |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 1903 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1904 | int hugetlb_sysctl_handler(struct ctl_table *table, int write, |
| 1905 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1906 | { |
| 1907 | |
| 1908 | return hugetlb_sysctl_handler_common(false, table, write, |
| 1909 | buffer, length, ppos); |
| 1910 | } |
| 1911 | |
| 1912 | #ifdef CONFIG_NUMA |
| 1913 | int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write, |
| 1914 | void __user *buffer, size_t *length, loff_t *ppos) |
| 1915 | { |
| 1916 | return hugetlb_sysctl_handler_common(true, table, write, |
| 1917 | buffer, length, ppos); |
| 1918 | } |
| 1919 | #endif /* CONFIG_NUMA */ |
| 1920 | |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 1921 | int hugetlb_treat_movable_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 1922 | void __user *buffer, |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 1923 | size_t *length, loff_t *ppos) |
| 1924 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 1925 | proc_dointvec(table, write, buffer, length, ppos); |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 1926 | if (hugepages_treat_as_movable) |
| 1927 | htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; |
| 1928 | else |
| 1929 | htlb_alloc_mask = GFP_HIGHUSER; |
| 1930 | return 0; |
| 1931 | } |
| 1932 | |
Nishanth Aravamudan | a3d0c6a | 2008-02-08 04:18:18 -0800 | [diff] [blame] | 1933 | int hugetlb_overcommit_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 1934 | void __user *buffer, |
Nishanth Aravamudan | a3d0c6a | 2008-02-08 04:18:18 -0800 | [diff] [blame] | 1935 | size_t *length, loff_t *ppos) |
| 1936 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1937 | struct hstate *h = &default_hstate; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1938 | unsigned long tmp; |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1939 | int ret; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1940 | |
| 1941 | if (!write) |
| 1942 | tmp = h->nr_overcommit_huge_pages; |
| 1943 | |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1944 | if (write && h->order >= MAX_ORDER) |
| 1945 | return -EINVAL; |
| 1946 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1947 | table->data = &tmp; |
| 1948 | table->maxlen = sizeof(unsigned long); |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1949 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
| 1950 | if (ret) |
| 1951 | goto out; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1952 | |
| 1953 | if (write) { |
| 1954 | spin_lock(&hugetlb_lock); |
| 1955 | h->nr_overcommit_huge_pages = tmp; |
| 1956 | spin_unlock(&hugetlb_lock); |
| 1957 | } |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1958 | out: |
| 1959 | return ret; |
Nishanth Aravamudan | a3d0c6a | 2008-02-08 04:18:18 -0800 | [diff] [blame] | 1960 | } |
| 1961 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1962 | #endif /* CONFIG_SYSCTL */ |
| 1963 | |
Alexey Dobriyan | e1759c2 | 2008-10-15 23:50:22 +0400 | [diff] [blame] | 1964 | void hugetlb_report_meminfo(struct seq_file *m) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1965 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1966 | struct hstate *h = &default_hstate; |
Alexey Dobriyan | e1759c2 | 2008-10-15 23:50:22 +0400 | [diff] [blame] | 1967 | seq_printf(m, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1968 | "HugePages_Total: %5lu\n" |
| 1969 | "HugePages_Free: %5lu\n" |
| 1970 | "HugePages_Rsvd: %5lu\n" |
| 1971 | "HugePages_Surp: %5lu\n" |
| 1972 | "Hugepagesize: %8lu kB\n", |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1973 | h->nr_huge_pages, |
| 1974 | h->free_huge_pages, |
| 1975 | h->resv_huge_pages, |
| 1976 | h->surplus_huge_pages, |
| 1977 | 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1978 | } |
| 1979 | |
| 1980 | int hugetlb_report_node_meminfo(int nid, char *buf) |
| 1981 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1982 | struct hstate *h = &default_hstate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1983 | return sprintf(buf, |
| 1984 | "Node %d HugePages_Total: %5u\n" |
Nishanth Aravamudan | a1de091 | 2008-03-26 14:37:53 -0700 | [diff] [blame] | 1985 | "Node %d HugePages_Free: %5u\n" |
| 1986 | "Node %d HugePages_Surp: %5u\n", |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1987 | nid, h->nr_huge_pages_node[nid], |
| 1988 | nid, h->free_huge_pages_node[nid], |
| 1989 | nid, h->surplus_huge_pages_node[nid]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1990 | } |
| 1991 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1992 | /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ |
| 1993 | unsigned long hugetlb_total_pages(void) |
| 1994 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1995 | struct hstate *h = &default_hstate; |
| 1996 | return h->nr_huge_pages * pages_per_huge_page(h); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1997 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1998 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1999 | static int hugetlb_acct_memory(struct hstate *h, long delta) |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2000 | { |
| 2001 | int ret = -ENOMEM; |
| 2002 | |
| 2003 | spin_lock(&hugetlb_lock); |
| 2004 | /* |
| 2005 | * When cpuset is configured, it breaks the strict hugetlb page |
| 2006 | * reservation as the accounting is done on a global variable. Such |
| 2007 | * reservation is completely rubbish in the presence of cpuset because |
| 2008 | * the reservation is not checked against page availability for the |
| 2009 | * current cpuset. Application can still potentially OOM'ed by kernel |
| 2010 | * with lack of free htlb page in cpuset that the task is in. |
| 2011 | * Attempt to enforce strict accounting with cpuset is almost |
| 2012 | * impossible (or too ugly) because cpuset is too fluid that |
| 2013 | * task or memory node can be dynamically moved between cpusets. |
| 2014 | * |
| 2015 | * The change of semantics for shared hugetlb mapping with cpuset is |
| 2016 | * undesirable. However, in order to preserve some of the semantics, |
| 2017 | * we fall back to check against current free page availability as |
| 2018 | * a best attempt and hopefully to minimize the impact of changing |
| 2019 | * semantics that cpuset has. |
| 2020 | */ |
| 2021 | if (delta > 0) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2022 | if (gather_surplus_pages(h, delta) < 0) |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2023 | goto out; |
| 2024 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2025 | if (delta > cpuset_mems_nr(h->free_huge_pages_node)) { |
| 2026 | return_unused_surplus_pages(h, delta); |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2027 | goto out; |
| 2028 | } |
| 2029 | } |
| 2030 | |
| 2031 | ret = 0; |
| 2032 | if (delta < 0) |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2033 | return_unused_surplus_pages(h, (unsigned long) -delta); |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2034 | |
| 2035 | out: |
| 2036 | spin_unlock(&hugetlb_lock); |
| 2037 | return ret; |
| 2038 | } |
| 2039 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2040 | static void hugetlb_vm_op_open(struct vm_area_struct *vma) |
| 2041 | { |
| 2042 | struct resv_map *reservations = vma_resv_map(vma); |
| 2043 | |
| 2044 | /* |
| 2045 | * This new VMA should share its siblings reservation map if present. |
| 2046 | * The VMA will only ever have a valid reservation map pointer where |
| 2047 | * it is being copied for another still existing VMA. As that VMA |
| 2048 | * has a reference to the reservation map it cannot dissappear until |
| 2049 | * after this open call completes. It is therefore safe to take a |
| 2050 | * new reference here without additional locking. |
| 2051 | */ |
| 2052 | if (reservations) |
| 2053 | kref_get(&reservations->refs); |
| 2054 | } |
| 2055 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2056 | static void hugetlb_vm_op_close(struct vm_area_struct *vma) |
| 2057 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2058 | struct hstate *h = hstate_vma(vma); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2059 | struct resv_map *reservations = vma_resv_map(vma); |
| 2060 | unsigned long reserve; |
| 2061 | unsigned long start; |
| 2062 | unsigned long end; |
| 2063 | |
| 2064 | if (reservations) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2065 | start = vma_hugecache_offset(h, vma, vma->vm_start); |
| 2066 | end = vma_hugecache_offset(h, vma, vma->vm_end); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2067 | |
| 2068 | reserve = (end - start) - |
| 2069 | region_count(&reservations->regions, start, end); |
| 2070 | |
| 2071 | kref_put(&reservations->refs, resv_map_release); |
| 2072 | |
Adam Litke | 7251ff7 | 2008-07-23 21:27:59 -0700 | [diff] [blame] | 2073 | if (reserve) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2074 | hugetlb_acct_memory(h, -reserve); |
Adam Litke | 7251ff7 | 2008-07-23 21:27:59 -0700 | [diff] [blame] | 2075 | hugetlb_put_quota(vma->vm_file->f_mapping, reserve); |
| 2076 | } |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2077 | } |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2078 | } |
| 2079 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2080 | /* |
| 2081 | * We cannot handle pagefaults against hugetlb pages at all. They cause |
| 2082 | * handle_mm_fault() to try to instantiate regular-sized pages in the |
| 2083 | * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get |
| 2084 | * this far. |
| 2085 | */ |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2086 | static int hugetlb_vm_op_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2087 | { |
| 2088 | BUG(); |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2089 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | } |
| 2091 | |
Alexey Dobriyan | f0f37e2 | 2009-09-27 22:29:37 +0400 | [diff] [blame] | 2092 | const struct vm_operations_struct hugetlb_vm_ops = { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2093 | .fault = hugetlb_vm_op_fault, |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2094 | .open = hugetlb_vm_op_open, |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2095 | .close = hugetlb_vm_op_close, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2096 | }; |
| 2097 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2098 | static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, |
| 2099 | int writable) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2100 | { |
| 2101 | pte_t entry; |
| 2102 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2103 | if (writable) { |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2104 | entry = |
| 2105 | pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))); |
| 2106 | } else { |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2107 | entry = huge_pte_wrprotect(mk_pte(page, vma->vm_page_prot)); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2108 | } |
| 2109 | entry = pte_mkyoung(entry); |
| 2110 | entry = pte_mkhuge(entry); |
| 2111 | |
| 2112 | return entry; |
| 2113 | } |
| 2114 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2115 | static void set_huge_ptep_writable(struct vm_area_struct *vma, |
| 2116 | unsigned long address, pte_t *ptep) |
| 2117 | { |
| 2118 | pte_t entry; |
| 2119 | |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2120 | entry = pte_mkwrite(pte_mkdirty(huge_ptep_get(ptep))); |
| 2121 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1)) { |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 2122 | update_mmu_cache(vma, address, ptep); |
Benjamin Herrenschmidt | 8dab524 | 2007-06-16 10:16:12 -0700 | [diff] [blame] | 2123 | } |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2124 | } |
| 2125 | |
| 2126 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2127 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, |
| 2128 | struct vm_area_struct *vma) |
| 2129 | { |
| 2130 | pte_t *src_pte, *dst_pte, entry; |
| 2131 | struct page *ptepage; |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2132 | unsigned long addr; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2133 | int cow; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2134 | struct hstate *h = hstate_vma(vma); |
| 2135 | unsigned long sz = huge_page_size(h); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2136 | |
| 2137 | cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2138 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2139 | for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2140 | src_pte = huge_pte_offset(src, addr); |
| 2141 | if (!src_pte) |
| 2142 | continue; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2143 | dst_pte = huge_pte_alloc(dst, addr, sz); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2144 | if (!dst_pte) |
| 2145 | goto nomem; |
Larry Woodman | c5c9942 | 2008-01-24 05:49:25 -0800 | [diff] [blame] | 2146 | |
| 2147 | /* If the pagetables are shared don't copy or take references */ |
| 2148 | if (dst_pte == src_pte) |
| 2149 | continue; |
| 2150 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2151 | spin_lock(&dst->page_table_lock); |
Nick Piggin | 4647875 | 2008-06-05 22:45:57 -0700 | [diff] [blame] | 2152 | spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2153 | if (!huge_pte_none(huge_ptep_get(src_pte))) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2154 | if (cow) |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2155 | huge_ptep_set_wrprotect(src, addr, src_pte); |
| 2156 | entry = huge_ptep_get(src_pte); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2157 | ptepage = pte_page(entry); |
| 2158 | get_page(ptepage); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2159 | page_dup_rmap(ptepage); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2160 | set_huge_pte_at(dst, addr, dst_pte, entry); |
| 2161 | } |
| 2162 | spin_unlock(&src->page_table_lock); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2163 | spin_unlock(&dst->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2164 | } |
| 2165 | return 0; |
| 2166 | |
| 2167 | nomem: |
| 2168 | return -ENOMEM; |
| 2169 | } |
| 2170 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 2171 | static int is_hugetlb_entry_migration(pte_t pte) |
| 2172 | { |
| 2173 | swp_entry_t swp; |
| 2174 | |
| 2175 | if (huge_pte_none(pte) || pte_present(pte)) |
| 2176 | return 0; |
| 2177 | swp = pte_to_swp_entry(pte); |
| 2178 | if (non_swap_entry(swp) && is_migration_entry(swp)) { |
| 2179 | return 1; |
| 2180 | } else |
| 2181 | return 0; |
| 2182 | } |
| 2183 | |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2184 | static int is_hugetlb_entry_hwpoisoned(pte_t pte) |
| 2185 | { |
| 2186 | swp_entry_t swp; |
| 2187 | |
| 2188 | if (huge_pte_none(pte) || pte_present(pte)) |
| 2189 | return 0; |
| 2190 | swp = pte_to_swp_entry(pte); |
| 2191 | if (non_swap_entry(swp) && is_hwpoison_entry(swp)) { |
| 2192 | return 1; |
| 2193 | } else |
| 2194 | return 0; |
| 2195 | } |
| 2196 | |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2197 | void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2198 | unsigned long end, struct page *ref_page) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2199 | { |
| 2200 | struct mm_struct *mm = vma->vm_mm; |
| 2201 | unsigned long address; |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2202 | pte_t *ptep; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2203 | pte_t pte; |
| 2204 | struct page *page; |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2205 | struct page *tmp; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2206 | struct hstate *h = hstate_vma(vma); |
| 2207 | unsigned long sz = huge_page_size(h); |
| 2208 | |
Chen, Kenneth W | c0a499c | 2006-12-06 20:31:39 -0800 | [diff] [blame] | 2209 | /* |
| 2210 | * A page gathering list, protected by per file i_mmap_lock. The |
| 2211 | * lock is used to avoid list corruption from multiple unmapping |
| 2212 | * of the same page since we are using page->lru. |
| 2213 | */ |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2214 | LIST_HEAD(page_list); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2215 | |
| 2216 | WARN_ON(!is_vm_hugetlb_page(vma)); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2217 | BUG_ON(start & ~huge_page_mask(h)); |
| 2218 | BUG_ON(end & ~huge_page_mask(h)); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2219 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 2220 | mmu_notifier_invalidate_range_start(mm, start, end); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 2221 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2222 | for (address = start; address < end; address += sz) { |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2223 | ptep = huge_pte_offset(mm, address); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2224 | if (!ptep) |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2225 | continue; |
| 2226 | |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 2227 | if (huge_pmd_unshare(mm, &address, ptep)) |
| 2228 | continue; |
| 2229 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2230 | /* |
| 2231 | * If a reference page is supplied, it is because a specific |
| 2232 | * page is being unmapped, not a range. Ensure the page we |
| 2233 | * are about to unmap is the actual page of interest. |
| 2234 | */ |
| 2235 | if (ref_page) { |
| 2236 | pte = huge_ptep_get(ptep); |
| 2237 | if (huge_pte_none(pte)) |
| 2238 | continue; |
| 2239 | page = pte_page(pte); |
| 2240 | if (page != ref_page) |
| 2241 | continue; |
| 2242 | |
| 2243 | /* |
| 2244 | * Mark the VMA as having unmapped its page so that |
| 2245 | * future faults in this VMA will fail rather than |
| 2246 | * looking like data was lost |
| 2247 | */ |
| 2248 | set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED); |
| 2249 | } |
| 2250 | |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2251 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2252 | if (huge_pte_none(pte)) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2253 | continue; |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2254 | |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2255 | /* |
| 2256 | * HWPoisoned hugepage is already unmapped and dropped reference |
| 2257 | */ |
| 2258 | if (unlikely(is_hugetlb_entry_hwpoisoned(pte))) |
| 2259 | continue; |
| 2260 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2261 | page = pte_page(pte); |
Ken Chen | 6649a38 | 2007-02-08 14:20:27 -0800 | [diff] [blame] | 2262 | if (pte_dirty(pte)) |
| 2263 | set_page_dirty(page); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2264 | list_add(&page->lru, &page_list); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2265 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2266 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 2267 | flush_tlb_range(vma, start, end); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 2268 | mmu_notifier_invalidate_range_end(mm, start, end); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2269 | list_for_each_entry_safe(page, tmp, &page_list, lru) { |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2270 | page_remove_rmap(page); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2271 | list_del(&page->lru); |
| 2272 | put_page(page); |
| 2273 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2275 | |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2276 | void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2277 | unsigned long end, struct page *ref_page) |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2278 | { |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 2279 | spin_lock(&vma->vm_file->f_mapping->i_mmap_lock); |
| 2280 | __unmap_hugepage_range(vma, start, end, ref_page); |
| 2281 | spin_unlock(&vma->vm_file->f_mapping->i_mmap_lock); |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2282 | } |
| 2283 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2284 | /* |
| 2285 | * This is called when the original mapper is failing to COW a MAP_PRIVATE |
| 2286 | * mappping it owns the reserve page for. The intention is to unmap the page |
| 2287 | * from other VMAs and let the children be SIGKILLed if they are faulting the |
| 2288 | * same region. |
| 2289 | */ |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 2290 | static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2291 | struct page *page, unsigned long address) |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2292 | { |
Adam Litke | 7526674 | 2008-11-12 13:24:56 -0800 | [diff] [blame] | 2293 | struct hstate *h = hstate_vma(vma); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2294 | struct vm_area_struct *iter_vma; |
| 2295 | struct address_space *mapping; |
| 2296 | struct prio_tree_iter iter; |
| 2297 | pgoff_t pgoff; |
| 2298 | |
| 2299 | /* |
| 2300 | * vm_pgoff is in PAGE_SIZE units, hence the different calculation |
| 2301 | * from page cache lookup which is in HPAGE_SIZE units. |
| 2302 | */ |
Adam Litke | 7526674 | 2008-11-12 13:24:56 -0800 | [diff] [blame] | 2303 | address = address & huge_page_mask(h); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2304 | pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) |
| 2305 | + (vma->vm_pgoff >> PAGE_SHIFT); |
| 2306 | mapping = (struct address_space *)page_private(page); |
| 2307 | |
Mel Gorman | 4eb2b1d | 2009-12-14 17:59:53 -0800 | [diff] [blame] | 2308 | /* |
| 2309 | * Take the mapping lock for the duration of the table walk. As |
| 2310 | * this mapping should be shared between all the VMAs, |
| 2311 | * __unmap_hugepage_range() is called as the lock is already held |
| 2312 | */ |
| 2313 | spin_lock(&mapping->i_mmap_lock); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2314 | vma_prio_tree_foreach(iter_vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
| 2315 | /* Do not unmap the current VMA */ |
| 2316 | if (iter_vma == vma) |
| 2317 | continue; |
| 2318 | |
| 2319 | /* |
| 2320 | * Unmap the page from other VMAs without their own reserves. |
| 2321 | * They get marked to be SIGKILLed if they fault in these |
| 2322 | * areas. This is because a future no-page fault on this VMA |
| 2323 | * could insert a zeroed page instead of the data existing |
| 2324 | * from the time of fork. This would look like data corruption |
| 2325 | */ |
| 2326 | if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER)) |
Mel Gorman | 4eb2b1d | 2009-12-14 17:59:53 -0800 | [diff] [blame] | 2327 | __unmap_hugepage_range(iter_vma, |
Adam Litke | 7526674 | 2008-11-12 13:24:56 -0800 | [diff] [blame] | 2328 | address, address + huge_page_size(h), |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2329 | page); |
| 2330 | } |
Mel Gorman | 4eb2b1d | 2009-12-14 17:59:53 -0800 | [diff] [blame] | 2331 | spin_unlock(&mapping->i_mmap_lock); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2332 | |
| 2333 | return 1; |
| 2334 | } |
| 2335 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2336 | /* |
| 2337 | * Hugetlb_cow() should be called with page lock of the original hugepage held. |
| 2338 | */ |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2339 | static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2340 | unsigned long address, pte_t *ptep, pte_t pte, |
| 2341 | struct page *pagecache_page) |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2342 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2343 | struct hstate *h = hstate_vma(vma); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2344 | struct page *old_page, *new_page; |
David Gibson | 79ac6ba | 2006-03-22 00:08:51 -0800 | [diff] [blame] | 2345 | int avoidcopy; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2346 | int outside_reserve = 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2347 | |
| 2348 | old_page = pte_page(pte); |
| 2349 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2350 | retry_avoidcopy: |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2351 | /* If no-one else is actually using this page, avoid the copy |
| 2352 | * and just make the page writable */ |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2353 | avoidcopy = (page_mapcount(old_page) == 1); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2354 | if (avoidcopy) { |
Naoya Horiguchi | 56c9cfb | 2010-09-10 13:23:04 +0900 | [diff] [blame] | 2355 | if (PageAnon(old_page)) |
| 2356 | page_move_anon_rmap(old_page, vma, address); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2357 | set_huge_ptep_writable(vma, address, ptep); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2358 | return 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2359 | } |
| 2360 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2361 | /* |
| 2362 | * If the process that created a MAP_PRIVATE mapping is about to |
| 2363 | * perform a COW due to a shared page count, attempt to satisfy |
| 2364 | * the allocation without using the existing reserves. The pagecache |
| 2365 | * page is used to determine if the reserve at this address was |
| 2366 | * consumed or not. If reserves were used, a partial faulted mapping |
| 2367 | * at the time of fork() could consume its reserves on COW instead |
| 2368 | * of the full address range. |
| 2369 | */ |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2370 | if (!(vma->vm_flags & VM_MAYSHARE) && |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2371 | is_vma_resv_set(vma, HPAGE_RESV_OWNER) && |
| 2372 | old_page != pagecache_page) |
| 2373 | outside_reserve = 1; |
| 2374 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2375 | page_cache_get(old_page); |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2376 | |
| 2377 | /* Drop page_table_lock as buddy allocator may be called */ |
| 2378 | spin_unlock(&mm->page_table_lock); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2379 | new_page = alloc_huge_page(vma, address, outside_reserve); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2380 | |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 2381 | if (IS_ERR(new_page)) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2382 | page_cache_release(old_page); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2383 | |
| 2384 | /* |
| 2385 | * If a process owning a MAP_PRIVATE mapping fails to COW, |
| 2386 | * it is due to references held by a child and an insufficient |
| 2387 | * huge page pool. To guarantee the original mappers |
| 2388 | * reliability, unmap the page from child processes. The child |
| 2389 | * may get SIGKILLed if it later faults. |
| 2390 | */ |
| 2391 | if (outside_reserve) { |
| 2392 | BUG_ON(huge_pte_none(pte)); |
| 2393 | if (unmap_ref_private(mm, vma, old_page, address)) { |
| 2394 | BUG_ON(page_count(old_page) != 1); |
| 2395 | BUG_ON(huge_pte_none(pte)); |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2396 | spin_lock(&mm->page_table_lock); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2397 | goto retry_avoidcopy; |
| 2398 | } |
| 2399 | WARN_ON_ONCE(1); |
| 2400 | } |
| 2401 | |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2402 | /* Caller expects lock to be held */ |
| 2403 | spin_lock(&mm->page_table_lock); |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 2404 | return -PTR_ERR(new_page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2405 | } |
| 2406 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2407 | /* |
| 2408 | * When the original hugepage is shared one, it does not have |
| 2409 | * anon_vma prepared. |
| 2410 | */ |
Dean Nelson | 44e2aa9 | 2010-10-26 14:22:08 -0700 | [diff] [blame] | 2411 | if (unlikely(anon_vma_prepare(vma))) { |
| 2412 | /* Caller expects lock to be held */ |
| 2413 | spin_lock(&mm->page_table_lock); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2414 | return VM_FAULT_OOM; |
Dean Nelson | 44e2aa9 | 2010-10-26 14:22:08 -0700 | [diff] [blame] | 2415 | } |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2416 | |
Andrea Arcangeli | 47ad847 | 2011-01-13 15:46:47 -0800 | [diff] [blame] | 2417 | copy_user_huge_page(new_page, old_page, address, vma, |
| 2418 | pages_per_huge_page(h)); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2419 | __SetPageUptodate(new_page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2420 | |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2421 | /* |
| 2422 | * Retake the page_table_lock to check for racing updates |
| 2423 | * before the page tables are altered |
| 2424 | */ |
| 2425 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2426 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2427 | if (likely(pte_same(huge_ptep_get(ptep), pte))) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2428 | /* Break COW */ |
Doug Doan | 3edd4fc | 2010-08-09 17:18:30 -0700 | [diff] [blame] | 2429 | mmu_notifier_invalidate_range_start(mm, |
| 2430 | address & huge_page_mask(h), |
| 2431 | (address & huge_page_mask(h)) + huge_page_size(h)); |
Gerald Schaefer | 8fe627e | 2008-04-28 02:13:28 -0700 | [diff] [blame] | 2432 | huge_ptep_clear_flush(vma, address, ptep); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2433 | set_huge_pte_at(mm, address, ptep, |
| 2434 | make_huge_pte(vma, new_page, 1)); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2435 | page_remove_rmap(old_page); |
Naoya Horiguchi | cd67f0d | 2010-09-10 13:23:04 +0900 | [diff] [blame] | 2436 | hugepage_add_new_anon_rmap(new_page, vma, address); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2437 | /* Make the old page be freed below */ |
| 2438 | new_page = old_page; |
Doug Doan | 3edd4fc | 2010-08-09 17:18:30 -0700 | [diff] [blame] | 2439 | mmu_notifier_invalidate_range_end(mm, |
| 2440 | address & huge_page_mask(h), |
| 2441 | (address & huge_page_mask(h)) + huge_page_size(h)); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2442 | } |
| 2443 | page_cache_release(new_page); |
| 2444 | page_cache_release(old_page); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2445 | return 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2446 | } |
| 2447 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2448 | /* Return the pagecache page at a given address within a VMA */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2449 | static struct page *hugetlbfs_pagecache_page(struct hstate *h, |
| 2450 | struct vm_area_struct *vma, unsigned long address) |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2451 | { |
| 2452 | struct address_space *mapping; |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 2453 | pgoff_t idx; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2454 | |
| 2455 | mapping = vma->vm_file->f_mapping; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2456 | idx = vma_hugecache_offset(h, vma, address); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2457 | |
| 2458 | return find_lock_page(mapping, idx); |
| 2459 | } |
| 2460 | |
Hugh Dickins | 3ae77f4 | 2009-09-21 17:03:33 -0700 | [diff] [blame] | 2461 | /* |
| 2462 | * Return whether there is a pagecache page to back given address within VMA. |
| 2463 | * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page. |
| 2464 | */ |
| 2465 | static bool hugetlbfs_pagecache_present(struct hstate *h, |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2466 | struct vm_area_struct *vma, unsigned long address) |
| 2467 | { |
| 2468 | struct address_space *mapping; |
| 2469 | pgoff_t idx; |
| 2470 | struct page *page; |
| 2471 | |
| 2472 | mapping = vma->vm_file->f_mapping; |
| 2473 | idx = vma_hugecache_offset(h, vma, address); |
| 2474 | |
| 2475 | page = find_get_page(mapping, idx); |
| 2476 | if (page) |
| 2477 | put_page(page); |
| 2478 | return page != NULL; |
| 2479 | } |
| 2480 | |
Robert P. J. Day | a1ed3dd | 2007-07-17 04:03:33 -0700 | [diff] [blame] | 2481 | static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma, |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2482 | unsigned long address, pte_t *ptep, unsigned int flags) |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2483 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2484 | struct hstate *h = hstate_vma(vma); |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2485 | int ret = VM_FAULT_SIGBUS; |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 2486 | pgoff_t idx; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2487 | unsigned long size; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2488 | struct page *page; |
| 2489 | struct address_space *mapping; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2490 | pte_t new_pte; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2491 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2492 | /* |
| 2493 | * Currently, we are forced to kill the process in the event the |
| 2494 | * original mapper has unmapped pages from the child due to a failed |
| 2495 | * COW. Warn that such a situation has occured as it may not be obvious |
| 2496 | */ |
| 2497 | if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) { |
| 2498 | printk(KERN_WARNING |
| 2499 | "PID %d killed due to inadequate hugepage pool\n", |
| 2500 | current->pid); |
| 2501 | return ret; |
| 2502 | } |
| 2503 | |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2504 | mapping = vma->vm_file->f_mapping; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2505 | idx = vma_hugecache_offset(h, vma, address); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2506 | |
| 2507 | /* |
| 2508 | * Use page lock to guard against racing truncation |
| 2509 | * before we get page_table_lock. |
| 2510 | */ |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2511 | retry: |
| 2512 | page = find_lock_page(mapping, idx); |
| 2513 | if (!page) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2514 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
Hugh Dickins | ebed4bf | 2006-10-28 10:38:43 -0700 | [diff] [blame] | 2515 | if (idx >= size) |
| 2516 | goto out; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2517 | page = alloc_huge_page(vma, address, 0); |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 2518 | if (IS_ERR(page)) { |
| 2519 | ret = -PTR_ERR(page); |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2520 | goto out; |
| 2521 | } |
Andrea Arcangeli | 47ad847 | 2011-01-13 15:46:47 -0800 | [diff] [blame] | 2522 | clear_huge_page(page, address, pages_per_huge_page(h)); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2523 | __SetPageUptodate(page); |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2524 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2525 | if (vma->vm_flags & VM_MAYSHARE) { |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2526 | int err; |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2527 | struct inode *inode = mapping->host; |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2528 | |
| 2529 | err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); |
| 2530 | if (err) { |
| 2531 | put_page(page); |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2532 | if (err == -EEXIST) |
| 2533 | goto retry; |
| 2534 | goto out; |
| 2535 | } |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2536 | |
| 2537 | spin_lock(&inode->i_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2538 | inode->i_blocks += blocks_per_huge_page(h); |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2539 | spin_unlock(&inode->i_lock); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2540 | page_dup_rmap(page); |
Mel Gorman | 23be746 | 2010-04-23 13:17:56 -0400 | [diff] [blame] | 2541 | } else { |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2542 | lock_page(page); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2543 | if (unlikely(anon_vma_prepare(vma))) { |
| 2544 | ret = VM_FAULT_OOM; |
| 2545 | goto backout_unlocked; |
| 2546 | } |
| 2547 | hugepage_add_new_anon_rmap(page, vma, address); |
Mel Gorman | 23be746 | 2010-04-23 13:17:56 -0400 | [diff] [blame] | 2548 | } |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2549 | } else { |
Naoya Horiguchi | 998b438 | 2010-09-08 10:19:32 +0900 | [diff] [blame] | 2550 | /* |
| 2551 | * If memory error occurs between mmap() and fault, some process |
| 2552 | * don't have hwpoisoned swap entry for errored virtual address. |
| 2553 | * So we need to block hugepage fault by PG_hwpoison bit check. |
| 2554 | */ |
| 2555 | if (unlikely(PageHWPoison(page))) { |
Andi Kleen | aa50d3a | 2010-10-06 21:45:00 +0200 | [diff] [blame] | 2556 | ret = VM_FAULT_HWPOISON | |
| 2557 | VM_FAULT_SET_HINDEX(h - hstates); |
Naoya Horiguchi | 998b438 | 2010-09-08 10:19:32 +0900 | [diff] [blame] | 2558 | goto backout_unlocked; |
| 2559 | } |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2560 | page_dup_rmap(page); |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2561 | } |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2562 | |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2563 | /* |
| 2564 | * If we are going to COW a private mapping later, we examine the |
| 2565 | * pending reservations for this page now. This will ensure that |
| 2566 | * any allocations necessary to record that reservation occur outside |
| 2567 | * the spinlock. |
| 2568 | */ |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2569 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2570 | if (vma_needs_reservation(h, vma, address) < 0) { |
| 2571 | ret = VM_FAULT_OOM; |
| 2572 | goto backout_unlocked; |
| 2573 | } |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2574 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2575 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2576 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2577 | if (idx >= size) |
| 2578 | goto backout; |
| 2579 | |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2580 | ret = 0; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2581 | if (!huge_pte_none(huge_ptep_get(ptep))) |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2582 | goto backout; |
| 2583 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2584 | new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) |
| 2585 | && (vma->vm_flags & VM_SHARED))); |
| 2586 | set_huge_pte_at(mm, address, ptep, new_pte); |
| 2587 | |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2588 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2589 | /* Optimization, do the COW without a second fault */ |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2590 | ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2591 | } |
| 2592 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2593 | spin_unlock(&mm->page_table_lock); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2594 | unlock_page(page); |
| 2595 | out: |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2596 | return ret; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2597 | |
| 2598 | backout: |
| 2599 | spin_unlock(&mm->page_table_lock); |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2600 | backout_unlocked: |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2601 | unlock_page(page); |
| 2602 | put_page(page); |
| 2603 | goto out; |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2604 | } |
| 2605 | |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2606 | int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2607 | unsigned long address, unsigned int flags) |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2608 | { |
| 2609 | pte_t *ptep; |
| 2610 | pte_t entry; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2611 | int ret; |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2612 | struct page *page = NULL; |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2613 | struct page *pagecache_page = NULL; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2614 | static DEFINE_MUTEX(hugetlb_instantiation_mutex); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2615 | struct hstate *h = hstate_vma(vma); |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2616 | |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2617 | ptep = huge_pte_offset(mm, address); |
| 2618 | if (ptep) { |
| 2619 | entry = huge_ptep_get(ptep); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 2620 | if (unlikely(is_hugetlb_entry_migration(entry))) { |
| 2621 | migration_entry_wait(mm, (pmd_t *)ptep, address); |
| 2622 | return 0; |
| 2623 | } else if (unlikely(is_hugetlb_entry_hwpoisoned(entry))) |
Andi Kleen | aa50d3a | 2010-10-06 21:45:00 +0200 | [diff] [blame] | 2624 | return VM_FAULT_HWPOISON_LARGE | |
| 2625 | VM_FAULT_SET_HINDEX(h - hstates); |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2626 | } |
| 2627 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2628 | ptep = huge_pte_alloc(mm, address, huge_page_size(h)); |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2629 | if (!ptep) |
| 2630 | return VM_FAULT_OOM; |
| 2631 | |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2632 | /* |
| 2633 | * Serialize hugepage allocation and instantiation, so that we don't |
| 2634 | * get spurious allocation failures if two CPUs race to instantiate |
| 2635 | * the same page in the page cache. |
| 2636 | */ |
| 2637 | mutex_lock(&hugetlb_instantiation_mutex); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2638 | entry = huge_ptep_get(ptep); |
| 2639 | if (huge_pte_none(entry)) { |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2640 | ret = hugetlb_no_page(mm, vma, address, ptep, flags); |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2641 | goto out_mutex; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2642 | } |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2643 | |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2644 | ret = 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2645 | |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2646 | /* |
| 2647 | * If we are going to COW the mapping later, we examine the pending |
| 2648 | * reservations for this page now. This will ensure that any |
| 2649 | * allocations necessary to record that reservation occur outside the |
| 2650 | * spinlock. For private mappings, we also lookup the pagecache |
| 2651 | * page now as it is used to determine if a reservation has been |
| 2652 | * consumed. |
| 2653 | */ |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2654 | if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) { |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2655 | if (vma_needs_reservation(h, vma, address) < 0) { |
| 2656 | ret = VM_FAULT_OOM; |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2657 | goto out_mutex; |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2658 | } |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2659 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2660 | if (!(vma->vm_flags & VM_MAYSHARE)) |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2661 | pagecache_page = hugetlbfs_pagecache_page(h, |
| 2662 | vma, address); |
| 2663 | } |
| 2664 | |
Naoya Horiguchi | 56c9cfb | 2010-09-10 13:23:04 +0900 | [diff] [blame] | 2665 | /* |
| 2666 | * hugetlb_cow() requires page locks of pte_page(entry) and |
| 2667 | * pagecache_page, so here we need take the former one |
| 2668 | * when page != pagecache_page or !pagecache_page. |
| 2669 | * Note that locking order is always pagecache_page -> page, |
| 2670 | * so no worry about deadlock. |
| 2671 | */ |
| 2672 | page = pte_page(entry); |
| 2673 | if (page != pagecache_page) |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2674 | lock_page(page); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2675 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2676 | spin_lock(&mm->page_table_lock); |
| 2677 | /* Check for a racing update before calling hugetlb_cow */ |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2678 | if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) |
| 2679 | goto out_page_table_lock; |
| 2680 | |
| 2681 | |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2682 | if (flags & FAULT_FLAG_WRITE) { |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2683 | if (!pte_write(entry)) { |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2684 | ret = hugetlb_cow(mm, vma, address, ptep, entry, |
| 2685 | pagecache_page); |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2686 | goto out_page_table_lock; |
| 2687 | } |
| 2688 | entry = pte_mkdirty(entry); |
| 2689 | } |
| 2690 | entry = pte_mkyoung(entry); |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2691 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, |
| 2692 | flags & FAULT_FLAG_WRITE)) |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 2693 | update_mmu_cache(vma, address, ptep); |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2694 | |
| 2695 | out_page_table_lock: |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2696 | spin_unlock(&mm->page_table_lock); |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2697 | |
| 2698 | if (pagecache_page) { |
| 2699 | unlock_page(pagecache_page); |
| 2700 | put_page(pagecache_page); |
| 2701 | } |
Dean Nelson | 1f64d69 | 2010-12-02 14:31:12 -0800 | [diff] [blame] | 2702 | if (page != pagecache_page) |
| 2703 | unlock_page(page); |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2704 | |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2705 | out_mutex: |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2706 | mutex_unlock(&hugetlb_instantiation_mutex); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2707 | |
| 2708 | return ret; |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2709 | } |
| 2710 | |
Andi Kleen | ceb8687 | 2008-07-23 21:27:50 -0700 | [diff] [blame] | 2711 | /* Can be overriden by architectures */ |
| 2712 | __attribute__((weak)) struct page * |
| 2713 | follow_huge_pud(struct mm_struct *mm, unsigned long address, |
| 2714 | pud_t *pud, int write) |
| 2715 | { |
| 2716 | BUG(); |
| 2717 | return NULL; |
| 2718 | } |
| 2719 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2720 | int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2721 | struct page **pages, struct vm_area_struct **vmas, |
Adam Litke | 5b23dbe | 2007-11-14 16:59:33 -0800 | [diff] [blame] | 2722 | unsigned long *position, int *length, int i, |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2723 | unsigned int flags) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2724 | { |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2725 | unsigned long pfn_offset; |
| 2726 | unsigned long vaddr = *position; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2727 | int remainder = *length; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2728 | struct hstate *h = hstate_vma(vma); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2729 | |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2730 | spin_lock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2731 | while (vaddr < vma->vm_end && remainder) { |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2732 | pte_t *pte; |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2733 | int absent; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2734 | struct page *page; |
| 2735 | |
| 2736 | /* |
| 2737 | * Some archs (sparc64, sh*) have multiple pte_ts to |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2738 | * each hugepage. We have to make sure we get the |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2739 | * first, for the page indexing below to work. |
| 2740 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2741 | pte = huge_pte_offset(mm, vaddr & huge_page_mask(h)); |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2742 | absent = !pte || huge_pte_none(huge_ptep_get(pte)); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2743 | |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2744 | /* |
| 2745 | * When coredumping, it suits get_dump_page if we just return |
Hugh Dickins | 3ae77f4 | 2009-09-21 17:03:33 -0700 | [diff] [blame] | 2746 | * an error where there's an empty slot with no huge pagecache |
| 2747 | * to back it. This way, we avoid allocating a hugepage, and |
| 2748 | * the sparse dumpfile avoids allocating disk blocks, but its |
| 2749 | * huge holes still show up with zeroes where they need to be. |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2750 | */ |
Hugh Dickins | 3ae77f4 | 2009-09-21 17:03:33 -0700 | [diff] [blame] | 2751 | if (absent && (flags & FOLL_DUMP) && |
| 2752 | !hugetlbfs_pagecache_present(h, vma, vaddr)) { |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2753 | remainder = 0; |
| 2754 | break; |
| 2755 | } |
| 2756 | |
| 2757 | if (absent || |
| 2758 | ((flags & FOLL_WRITE) && !pte_write(huge_ptep_get(pte)))) { |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2759 | int ret; |
| 2760 | |
| 2761 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2762 | ret = hugetlb_fault(mm, vma, vaddr, |
| 2763 | (flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2764 | spin_lock(&mm->page_table_lock); |
Adam Litke | a89182c | 2007-08-22 14:01:51 -0700 | [diff] [blame] | 2765 | if (!(ret & VM_FAULT_ERROR)) |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2766 | continue; |
| 2767 | |
| 2768 | remainder = 0; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2769 | break; |
| 2770 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2771 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2772 | pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2773 | page = pte_page(huge_ptep_get(pte)); |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2774 | same_page: |
Chen, Kenneth W | d669218 | 2006-03-31 02:29:57 -0800 | [diff] [blame] | 2775 | if (pages) { |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2776 | pages[i] = mem_map_offset(page, pfn_offset); |
KOSAKI Motohiro | 4b2e38a | 2008-10-18 20:27:10 -0700 | [diff] [blame] | 2777 | get_page(pages[i]); |
Chen, Kenneth W | d669218 | 2006-03-31 02:29:57 -0800 | [diff] [blame] | 2778 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2779 | |
| 2780 | if (vmas) |
| 2781 | vmas[i] = vma; |
| 2782 | |
| 2783 | vaddr += PAGE_SIZE; |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2784 | ++pfn_offset; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2785 | --remainder; |
| 2786 | ++i; |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2787 | if (vaddr < vma->vm_end && remainder && |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2788 | pfn_offset < pages_per_huge_page(h)) { |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2789 | /* |
| 2790 | * We use pfn_offset to avoid touching the pageframes |
| 2791 | * of this compound page. |
| 2792 | */ |
| 2793 | goto same_page; |
| 2794 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2795 | } |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2796 | spin_unlock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2797 | *length = remainder; |
| 2798 | *position = vaddr; |
| 2799 | |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2800 | return i ? i : -EFAULT; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2801 | } |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2802 | |
| 2803 | void hugetlb_change_protection(struct vm_area_struct *vma, |
| 2804 | unsigned long address, unsigned long end, pgprot_t newprot) |
| 2805 | { |
| 2806 | struct mm_struct *mm = vma->vm_mm; |
| 2807 | unsigned long start = address; |
| 2808 | pte_t *ptep; |
| 2809 | pte_t pte; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2810 | struct hstate *h = hstate_vma(vma); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2811 | |
| 2812 | BUG_ON(address >= end); |
| 2813 | flush_cache_range(vma, address, end); |
| 2814 | |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 2815 | spin_lock(&vma->vm_file->f_mapping->i_mmap_lock); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2816 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2817 | for (; address < end; address += huge_page_size(h)) { |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2818 | ptep = huge_pte_offset(mm, address); |
| 2819 | if (!ptep) |
| 2820 | continue; |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 2821 | if (huge_pmd_unshare(mm, &address, ptep)) |
| 2822 | continue; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2823 | if (!huge_pte_none(huge_ptep_get(ptep))) { |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2824 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
| 2825 | pte = pte_mkhuge(pte_modify(pte, newprot)); |
| 2826 | set_huge_pte_at(mm, address, ptep, pte); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2827 | } |
| 2828 | } |
| 2829 | spin_unlock(&mm->page_table_lock); |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 2830 | spin_unlock(&vma->vm_file->f_mapping->i_mmap_lock); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2831 | |
| 2832 | flush_tlb_range(vma, start, end); |
| 2833 | } |
| 2834 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2835 | int hugetlb_reserve_pages(struct inode *inode, |
| 2836 | long from, long to, |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2837 | struct vm_area_struct *vma, |
| 2838 | int acctflag) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 2839 | { |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2840 | long ret, chg; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2841 | struct hstate *h = hstate_inode(inode); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 2842 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2843 | /* |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2844 | * Only apply hugepage reservation if asked. At fault time, an |
| 2845 | * attempt will be made for VM_NORESERVE to allocate a page |
| 2846 | * and filesystem quota without using reserves |
| 2847 | */ |
| 2848 | if (acctflag & VM_NORESERVE) |
| 2849 | return 0; |
| 2850 | |
| 2851 | /* |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2852 | * Shared mappings base their reservation on the number of pages that |
| 2853 | * are already allocated on behalf of the file. Private mappings need |
| 2854 | * to reserve the full area even if read-only as mprotect() may be |
| 2855 | * called to make the mapping read-write. Assume !vma is a shm mapping |
| 2856 | */ |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2857 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2858 | chg = region_chg(&inode->i_mapping->private_list, from, to); |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2859 | else { |
| 2860 | struct resv_map *resv_map = resv_map_alloc(); |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2861 | if (!resv_map) |
| 2862 | return -ENOMEM; |
| 2863 | |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2864 | chg = to - from; |
| 2865 | |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2866 | set_vma_resv_map(vma, resv_map); |
| 2867 | set_vma_resv_flags(vma, HPAGE_RESV_OWNER); |
| 2868 | } |
| 2869 | |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2870 | if (chg < 0) |
| 2871 | return chg; |
| 2872 | |
| 2873 | /* There must be enough filesystem quota for the mapping */ |
| 2874 | if (hugetlb_get_quota(inode->i_mapping, chg)) |
| 2875 | return -ENOSPC; |
| 2876 | |
| 2877 | /* |
| 2878 | * Check enough hugepages are available for the reservation. |
| 2879 | * Hand back the quota if there are not |
| 2880 | */ |
| 2881 | ret = hugetlb_acct_memory(h, chg); |
| 2882 | if (ret < 0) { |
| 2883 | hugetlb_put_quota(inode->i_mapping, chg); |
| 2884 | return ret; |
| 2885 | } |
| 2886 | |
| 2887 | /* |
| 2888 | * Account for the reservations made. Shared mappings record regions |
| 2889 | * that have reservations as they are shared by multiple VMAs. |
| 2890 | * When the last VMA disappears, the region map says how much |
| 2891 | * the reservation was and the page cache tells how much of |
| 2892 | * the reservation was consumed. Private mappings are per-VMA and |
| 2893 | * only the consumed reservations are tracked. When the VMA |
| 2894 | * disappears, the original reservation is the VMA size and the |
| 2895 | * consumed reservations are stored in the map. Hence, nothing |
| 2896 | * else has to be done for private mappings here |
| 2897 | */ |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2898 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2899 | region_add(&inode->i_mapping->private_list, from, to); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 2900 | return 0; |
| 2901 | } |
| 2902 | |
| 2903 | void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) |
| 2904 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2905 | struct hstate *h = hstate_inode(inode); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 2906 | long chg = region_truncate(&inode->i_mapping->private_list, offset); |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2907 | |
| 2908 | spin_lock(&inode->i_lock); |
Eric Sandeen | e4c6f8b | 2009-07-29 15:02:16 -0700 | [diff] [blame] | 2909 | inode->i_blocks -= (blocks_per_huge_page(h) * freed); |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2910 | spin_unlock(&inode->i_lock); |
| 2911 | |
Adam Litke | 90d8b7e | 2007-11-14 16:59:42 -0800 | [diff] [blame] | 2912 | hugetlb_put_quota(inode->i_mapping, (chg - freed)); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2913 | hugetlb_acct_memory(h, -(chg - freed)); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 2914 | } |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 2915 | |
Andi Kleen | d5bd910 | 2010-09-27 09:00:12 +0200 | [diff] [blame] | 2916 | #ifdef CONFIG_MEMORY_FAILURE |
| 2917 | |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2918 | /* Should be called in hugetlb_lock */ |
| 2919 | static int is_hugepage_on_freelist(struct page *hpage) |
| 2920 | { |
| 2921 | struct page *page; |
| 2922 | struct page *tmp; |
| 2923 | struct hstate *h = page_hstate(hpage); |
| 2924 | int nid = page_to_nid(hpage); |
| 2925 | |
| 2926 | list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru) |
| 2927 | if (page == hpage) |
| 2928 | return 1; |
| 2929 | return 0; |
| 2930 | } |
| 2931 | |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 2932 | /* |
| 2933 | * This function is called from memory failure code. |
| 2934 | * Assume the caller holds page lock of the head page. |
| 2935 | */ |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2936 | int dequeue_hwpoisoned_huge_page(struct page *hpage) |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 2937 | { |
| 2938 | struct hstate *h = page_hstate(hpage); |
| 2939 | int nid = page_to_nid(hpage); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2940 | int ret = -EBUSY; |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 2941 | |
| 2942 | spin_lock(&hugetlb_lock); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2943 | if (is_hugepage_on_freelist(hpage)) { |
| 2944 | list_del(&hpage->lru); |
Naoya Horiguchi | 8c6c2ec | 2010-09-08 10:19:38 +0900 | [diff] [blame] | 2945 | set_page_refcounted(hpage); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2946 | h->free_huge_pages--; |
| 2947 | h->free_huge_pages_node[nid]--; |
| 2948 | ret = 0; |
| 2949 | } |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 2950 | spin_unlock(&hugetlb_lock); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2951 | return ret; |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 2952 | } |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 2953 | #endif |