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> |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 27 | #include <linux/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 | |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 56 | static inline void unlock_or_release_subpool(struct hugepage_subpool *spool) |
| 57 | { |
| 58 | bool free = (spool->count == 0) && (spool->used_hpages == 0); |
| 59 | |
| 60 | spin_unlock(&spool->lock); |
| 61 | |
| 62 | /* If no pages are used, and no other handles to the subpool |
| 63 | * remain, free the subpool the subpool remain */ |
| 64 | if (free) |
| 65 | kfree(spool); |
| 66 | } |
| 67 | |
| 68 | struct hugepage_subpool *hugepage_new_subpool(long nr_blocks) |
| 69 | { |
| 70 | struct hugepage_subpool *spool; |
| 71 | |
| 72 | spool = kmalloc(sizeof(*spool), GFP_KERNEL); |
| 73 | if (!spool) |
| 74 | return NULL; |
| 75 | |
| 76 | spin_lock_init(&spool->lock); |
| 77 | spool->count = 1; |
| 78 | spool->max_hpages = nr_blocks; |
| 79 | spool->used_hpages = 0; |
| 80 | |
| 81 | return spool; |
| 82 | } |
| 83 | |
| 84 | void hugepage_put_subpool(struct hugepage_subpool *spool) |
| 85 | { |
| 86 | spin_lock(&spool->lock); |
| 87 | BUG_ON(!spool->count); |
| 88 | spool->count--; |
| 89 | unlock_or_release_subpool(spool); |
| 90 | } |
| 91 | |
| 92 | static int hugepage_subpool_get_pages(struct hugepage_subpool *spool, |
| 93 | long delta) |
| 94 | { |
| 95 | int ret = 0; |
| 96 | |
| 97 | if (!spool) |
| 98 | return 0; |
| 99 | |
| 100 | spin_lock(&spool->lock); |
| 101 | if ((spool->used_hpages + delta) <= spool->max_hpages) { |
| 102 | spool->used_hpages += delta; |
| 103 | } else { |
| 104 | ret = -ENOMEM; |
| 105 | } |
| 106 | spin_unlock(&spool->lock); |
| 107 | |
| 108 | return ret; |
| 109 | } |
| 110 | |
| 111 | static void hugepage_subpool_put_pages(struct hugepage_subpool *spool, |
| 112 | long delta) |
| 113 | { |
| 114 | if (!spool) |
| 115 | return; |
| 116 | |
| 117 | spin_lock(&spool->lock); |
| 118 | spool->used_hpages -= delta; |
| 119 | /* If hugetlbfs_put_super couldn't free spool due to |
| 120 | * an outstanding quota reference, free it now. */ |
| 121 | unlock_or_release_subpool(spool); |
| 122 | } |
| 123 | |
| 124 | static inline struct hugepage_subpool *subpool_inode(struct inode *inode) |
| 125 | { |
| 126 | return HUGETLBFS_SB(inode->i_sb)->spool; |
| 127 | } |
| 128 | |
| 129 | static inline struct hugepage_subpool *subpool_vma(struct vm_area_struct *vma) |
| 130 | { |
| 131 | return subpool_inode(vma->vm_file->f_dentry->d_inode); |
| 132 | } |
| 133 | |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 134 | /* |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 135 | * Region tracking -- allows tracking of reservations and instantiated pages |
| 136 | * across the pages in a mapping. |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 137 | * |
| 138 | * The region data structures are protected by a combination of the mmap_sem |
| 139 | * and the hugetlb_instantion_mutex. To access or modify a region the caller |
| 140 | * must either hold the mmap_sem for write, or the mmap_sem for read and |
| 141 | * the hugetlb_instantiation mutex: |
| 142 | * |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 143 | * down_write(&mm->mmap_sem); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 144 | * or |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 145 | * down_read(&mm->mmap_sem); |
| 146 | * mutex_lock(&hugetlb_instantiation_mutex); |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 147 | */ |
| 148 | struct file_region { |
| 149 | struct list_head link; |
| 150 | long from; |
| 151 | long to; |
| 152 | }; |
| 153 | |
| 154 | static long region_add(struct list_head *head, long f, long t) |
| 155 | { |
| 156 | struct file_region *rg, *nrg, *trg; |
| 157 | |
| 158 | /* Locate the region we are either in or before. */ |
| 159 | list_for_each_entry(rg, head, link) |
| 160 | if (f <= rg->to) |
| 161 | break; |
| 162 | |
| 163 | /* Round our left edge to the current segment if it encloses us. */ |
| 164 | if (f > rg->from) |
| 165 | f = rg->from; |
| 166 | |
| 167 | /* Check for and consume any regions we now overlap with. */ |
| 168 | nrg = rg; |
| 169 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { |
| 170 | if (&rg->link == head) |
| 171 | break; |
| 172 | if (rg->from > t) |
| 173 | break; |
| 174 | |
| 175 | /* If this area reaches higher then extend our area to |
| 176 | * include it completely. If this is not the first area |
| 177 | * which we intend to reuse, free it. */ |
| 178 | if (rg->to > t) |
| 179 | t = rg->to; |
| 180 | if (rg != nrg) { |
| 181 | list_del(&rg->link); |
| 182 | kfree(rg); |
| 183 | } |
| 184 | } |
| 185 | nrg->from = f; |
| 186 | nrg->to = t; |
| 187 | return 0; |
| 188 | } |
| 189 | |
| 190 | static long region_chg(struct list_head *head, long f, long t) |
| 191 | { |
| 192 | struct file_region *rg, *nrg; |
| 193 | long chg = 0; |
| 194 | |
| 195 | /* Locate the region we are before or in. */ |
| 196 | list_for_each_entry(rg, head, link) |
| 197 | if (f <= rg->to) |
| 198 | break; |
| 199 | |
| 200 | /* If we are below the current region then a new region is required. |
| 201 | * Subtle, allocate a new region at the position but make it zero |
| 202 | * size such that we can guarantee to record the reservation. */ |
| 203 | if (&rg->link == head || t < rg->from) { |
| 204 | nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); |
| 205 | if (!nrg) |
| 206 | return -ENOMEM; |
| 207 | nrg->from = f; |
| 208 | nrg->to = f; |
| 209 | INIT_LIST_HEAD(&nrg->link); |
| 210 | list_add(&nrg->link, rg->link.prev); |
| 211 | |
| 212 | return t - f; |
| 213 | } |
| 214 | |
| 215 | /* Round our left edge to the current segment if it encloses us. */ |
| 216 | if (f > rg->from) |
| 217 | f = rg->from; |
| 218 | chg = t - f; |
| 219 | |
| 220 | /* Check for and consume any regions we now overlap with. */ |
| 221 | list_for_each_entry(rg, rg->link.prev, link) { |
| 222 | if (&rg->link == head) |
| 223 | break; |
| 224 | if (rg->from > t) |
| 225 | return chg; |
| 226 | |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 227 | /* We overlap with this area, if it extends further than |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 228 | * us then we must extend ourselves. Account for its |
| 229 | * existing reservation. */ |
| 230 | if (rg->to > t) { |
| 231 | chg += rg->to - t; |
| 232 | t = rg->to; |
| 233 | } |
| 234 | chg -= rg->to - rg->from; |
| 235 | } |
| 236 | return chg; |
| 237 | } |
| 238 | |
| 239 | static long region_truncate(struct list_head *head, long end) |
| 240 | { |
| 241 | struct file_region *rg, *trg; |
| 242 | long chg = 0; |
| 243 | |
| 244 | /* Locate the region we are either in or before. */ |
| 245 | list_for_each_entry(rg, head, link) |
| 246 | if (end <= rg->to) |
| 247 | break; |
| 248 | if (&rg->link == head) |
| 249 | return 0; |
| 250 | |
| 251 | /* If we are in the middle of a region then adjust it. */ |
| 252 | if (end > rg->from) { |
| 253 | chg = rg->to - end; |
| 254 | rg->to = end; |
| 255 | rg = list_entry(rg->link.next, typeof(*rg), link); |
| 256 | } |
| 257 | |
| 258 | /* Drop any remaining regions. */ |
| 259 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { |
| 260 | if (&rg->link == head) |
| 261 | break; |
| 262 | chg += rg->to - rg->from; |
| 263 | list_del(&rg->link); |
| 264 | kfree(rg); |
| 265 | } |
| 266 | return chg; |
| 267 | } |
| 268 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 269 | static long region_count(struct list_head *head, long f, long t) |
| 270 | { |
| 271 | struct file_region *rg; |
| 272 | long chg = 0; |
| 273 | |
| 274 | /* Locate each segment we overlap with, and count that overlap. */ |
| 275 | list_for_each_entry(rg, head, link) { |
| 276 | int seg_from; |
| 277 | int seg_to; |
| 278 | |
| 279 | if (rg->to <= f) |
| 280 | continue; |
| 281 | if (rg->from >= t) |
| 282 | break; |
| 283 | |
| 284 | seg_from = max(rg->from, f); |
| 285 | seg_to = min(rg->to, t); |
| 286 | |
| 287 | chg += seg_to - seg_from; |
| 288 | } |
| 289 | |
| 290 | return chg; |
| 291 | } |
| 292 | |
Andy Whitcroft | 9682290 | 2008-07-23 21:27:29 -0700 | [diff] [blame] | 293 | /* |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 294 | * Convert the address within this vma to the page offset within |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 295 | * the mapping, in pagecache page units; huge pages here. |
| 296 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 297 | static pgoff_t vma_hugecache_offset(struct hstate *h, |
| 298 | struct vm_area_struct *vma, unsigned long address) |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 299 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 300 | return ((address - vma->vm_start) >> huge_page_shift(h)) + |
| 301 | (vma->vm_pgoff >> huge_page_order(h)); |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 302 | } |
| 303 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 304 | pgoff_t linear_hugepage_index(struct vm_area_struct *vma, |
| 305 | unsigned long address) |
| 306 | { |
| 307 | return vma_hugecache_offset(hstate_vma(vma), vma, address); |
| 308 | } |
| 309 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 310 | /* |
Mel Gorman | 08fba69 | 2009-01-06 14:38:53 -0800 | [diff] [blame] | 311 | * Return the size of the pages allocated when backing a VMA. In the majority |
| 312 | * cases this will be same size as used by the page table entries. |
| 313 | */ |
| 314 | unsigned long vma_kernel_pagesize(struct vm_area_struct *vma) |
| 315 | { |
| 316 | struct hstate *hstate; |
| 317 | |
| 318 | if (!is_vm_hugetlb_page(vma)) |
| 319 | return PAGE_SIZE; |
| 320 | |
| 321 | hstate = hstate_vma(vma); |
| 322 | |
| 323 | return 1UL << (hstate->order + PAGE_SHIFT); |
| 324 | } |
Joerg Roedel | f340ca0 | 2009-06-19 15:16:22 +0200 | [diff] [blame] | 325 | EXPORT_SYMBOL_GPL(vma_kernel_pagesize); |
Mel Gorman | 08fba69 | 2009-01-06 14:38:53 -0800 | [diff] [blame] | 326 | |
| 327 | /* |
Mel Gorman | 3340289 | 2009-01-06 14:38:54 -0800 | [diff] [blame] | 328 | * Return the page size being used by the MMU to back a VMA. In the majority |
| 329 | * of cases, the page size used by the kernel matches the MMU size. On |
| 330 | * architectures where it differs, an architecture-specific version of this |
| 331 | * function is required. |
| 332 | */ |
| 333 | #ifndef vma_mmu_pagesize |
| 334 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) |
| 335 | { |
| 336 | return vma_kernel_pagesize(vma); |
| 337 | } |
| 338 | #endif |
| 339 | |
| 340 | /* |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 341 | * Flags for MAP_PRIVATE reservations. These are stored in the bottom |
| 342 | * bits of the reservation map pointer, which are always clear due to |
| 343 | * alignment. |
| 344 | */ |
| 345 | #define HPAGE_RESV_OWNER (1UL << 0) |
| 346 | #define HPAGE_RESV_UNMAPPED (1UL << 1) |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 347 | #define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 348 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 349 | /* |
| 350 | * These helpers are used to track how many pages are reserved for |
| 351 | * faults in a MAP_PRIVATE mapping. Only the process that called mmap() |
| 352 | * is guaranteed to have their future faults succeed. |
| 353 | * |
| 354 | * With the exception of reset_vma_resv_huge_pages() which is called at fork(), |
| 355 | * the reserve counters are updated with the hugetlb_lock held. It is safe |
| 356 | * to reset the VMA at fork() time as it is not in use yet and there is no |
| 357 | * 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] | 358 | * |
| 359 | * The private mapping reservation is represented in a subtly different |
| 360 | * manner to a shared mapping. A shared mapping has a region map associated |
| 361 | * with the underlying file, this region map represents the backing file |
| 362 | * pages which have ever had a reservation assigned which this persists even |
| 363 | * after the page is instantiated. A private mapping has a region map |
| 364 | * associated with the original mmap which is attached to all VMAs which |
| 365 | * reference it, this region map represents those offsets which have consumed |
| 366 | * reservation ie. where pages have been instantiated. |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 367 | */ |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 368 | static unsigned long get_vma_private_data(struct vm_area_struct *vma) |
| 369 | { |
| 370 | return (unsigned long)vma->vm_private_data; |
| 371 | } |
| 372 | |
| 373 | static void set_vma_private_data(struct vm_area_struct *vma, |
| 374 | unsigned long value) |
| 375 | { |
| 376 | vma->vm_private_data = (void *)value; |
| 377 | } |
| 378 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 379 | struct resv_map { |
| 380 | struct kref refs; |
| 381 | struct list_head regions; |
| 382 | }; |
| 383 | |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 384 | static struct resv_map *resv_map_alloc(void) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 385 | { |
| 386 | struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); |
| 387 | if (!resv_map) |
| 388 | return NULL; |
| 389 | |
| 390 | kref_init(&resv_map->refs); |
| 391 | INIT_LIST_HEAD(&resv_map->regions); |
| 392 | |
| 393 | return resv_map; |
| 394 | } |
| 395 | |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 396 | static void resv_map_release(struct kref *ref) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 397 | { |
| 398 | struct resv_map *resv_map = container_of(ref, struct resv_map, refs); |
| 399 | |
| 400 | /* Clear out any active regions before we release the map. */ |
| 401 | region_truncate(&resv_map->regions, 0); |
| 402 | kfree(resv_map); |
| 403 | } |
| 404 | |
| 405 | static struct resv_map *vma_resv_map(struct vm_area_struct *vma) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 406 | { |
| 407 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 408 | if (!(vma->vm_flags & VM_MAYSHARE)) |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 409 | return (struct resv_map *)(get_vma_private_data(vma) & |
| 410 | ~HPAGE_RESV_MASK); |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 411 | return NULL; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 412 | } |
| 413 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 414 | 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] | 415 | { |
| 416 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 417 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 418 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 419 | set_vma_private_data(vma, (get_vma_private_data(vma) & |
| 420 | HPAGE_RESV_MASK) | (unsigned long)map); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 421 | } |
| 422 | |
| 423 | static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags) |
| 424 | { |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 425 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 426 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 427 | |
| 428 | set_vma_private_data(vma, get_vma_private_data(vma) | flags); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 429 | } |
| 430 | |
| 431 | static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag) |
| 432 | { |
| 433 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 434 | |
| 435 | return (get_vma_private_data(vma) & flag) != 0; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 436 | } |
| 437 | |
| 438 | /* Decrement the reserved pages in the hugepage pool by one */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 439 | static void decrement_hugepage_resv_vma(struct hstate *h, |
| 440 | struct vm_area_struct *vma) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 441 | { |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 442 | if (vma->vm_flags & VM_NORESERVE) |
| 443 | return; |
| 444 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 445 | if (vma->vm_flags & VM_MAYSHARE) { |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 446 | /* Shared mappings always use reserves */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 447 | h->resv_huge_pages--; |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 448 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 449 | /* |
| 450 | * Only the process that called mmap() has reserves for |
| 451 | * private mappings. |
| 452 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 453 | h->resv_huge_pages--; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 454 | } |
| 455 | } |
| 456 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 457 | /* Reset counters to 0 and clear all HPAGE_RESV_* flags */ |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 458 | void reset_vma_resv_huge_pages(struct vm_area_struct *vma) |
| 459 | { |
| 460 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 461 | if (!(vma->vm_flags & VM_MAYSHARE)) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 462 | vma->vm_private_data = (void *)0; |
| 463 | } |
| 464 | |
| 465 | /* Returns true if the VMA has associated reserve pages */ |
Mel Gorman | 7f09ca5 | 2008-07-23 21:27:58 -0700 | [diff] [blame] | 466 | static int vma_has_reserves(struct vm_area_struct *vma) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 467 | { |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 468 | if (vma->vm_flags & VM_MAYSHARE) |
Mel Gorman | 7f09ca5 | 2008-07-23 21:27:58 -0700 | [diff] [blame] | 469 | return 1; |
| 470 | if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) |
| 471 | return 1; |
| 472 | return 0; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 473 | } |
| 474 | |
Naoya Horiguchi | 0ebabb4 | 2010-09-08 10:19:34 +0900 | [diff] [blame] | 475 | static void copy_gigantic_page(struct page *dst, struct page *src) |
| 476 | { |
| 477 | int i; |
| 478 | struct hstate *h = page_hstate(src); |
| 479 | struct page *dst_base = dst; |
| 480 | struct page *src_base = src; |
| 481 | |
| 482 | for (i = 0; i < pages_per_huge_page(h); ) { |
| 483 | cond_resched(); |
| 484 | copy_highpage(dst, src); |
| 485 | |
| 486 | i++; |
| 487 | dst = mem_map_next(dst, dst_base, i); |
| 488 | src = mem_map_next(src, src_base, i); |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | void copy_huge_page(struct page *dst, struct page *src) |
| 493 | { |
| 494 | int i; |
| 495 | struct hstate *h = page_hstate(src); |
| 496 | |
| 497 | if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) { |
| 498 | copy_gigantic_page(dst, src); |
| 499 | return; |
| 500 | } |
| 501 | |
| 502 | might_sleep(); |
| 503 | for (i = 0; i < pages_per_huge_page(h); i++) { |
| 504 | cond_resched(); |
| 505 | copy_highpage(dst + i, src + i); |
| 506 | } |
| 507 | } |
| 508 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 509 | static void enqueue_huge_page(struct hstate *h, struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | { |
| 511 | int nid = page_to_nid(page); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 512 | list_add(&page->lru, &h->hugepage_freelists[nid]); |
| 513 | h->free_huge_pages++; |
| 514 | h->free_huge_pages_node[nid]++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | } |
| 516 | |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 517 | static struct page *dequeue_huge_page_node(struct hstate *h, int nid) |
| 518 | { |
| 519 | struct page *page; |
| 520 | |
| 521 | if (list_empty(&h->hugepage_freelists[nid])) |
| 522 | return NULL; |
| 523 | page = list_entry(h->hugepage_freelists[nid].next, struct page, lru); |
| 524 | list_del(&page->lru); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 525 | set_page_refcounted(page); |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 526 | h->free_huge_pages--; |
| 527 | h->free_huge_pages_node[nid]--; |
| 528 | return page; |
| 529 | } |
| 530 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 531 | static struct page *dequeue_huge_page_vma(struct hstate *h, |
| 532 | struct vm_area_struct *vma, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 533 | unsigned long address, int avoid_reserve) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | { |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 535 | struct page *page; |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 536 | struct mempolicy *mpol; |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 537 | nodemask_t *nodemask; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 538 | struct zonelist *zonelist; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 539 | struct zone *zone; |
| 540 | struct zoneref *z; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 541 | unsigned int cpuset_mems_cookie; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 543 | retry_cpuset: |
| 544 | cpuset_mems_cookie = get_mems_allowed(); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 545 | zonelist = huge_zonelist(vma, address, |
| 546 | htlb_alloc_mask, &mpol, &nodemask); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 547 | /* |
| 548 | * A child process with MAP_PRIVATE mappings created by their parent |
| 549 | * have no page reserves. This check ensures that reservations are |
| 550 | * not "stolen". The child may still get SIGKILLed |
| 551 | */ |
Mel Gorman | 7f09ca5 | 2008-07-23 21:27:58 -0700 | [diff] [blame] | 552 | if (!vma_has_reserves(vma) && |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 553 | h->free_huge_pages - h->resv_huge_pages == 0) |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 554 | goto err; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 555 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 556 | /* If reserves cannot be used, ensure enough pages are in the pool */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 557 | if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) |
Justin P. Mattock | 6eab04a | 2011-04-08 19:49:08 -0700 | [diff] [blame] | 558 | goto err; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 559 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 560 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 561 | MAX_NR_ZONES - 1, nodemask) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 562 | if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) { |
| 563 | page = dequeue_huge_page_node(h, zone_to_nid(zone)); |
| 564 | if (page) { |
| 565 | if (!avoid_reserve) |
| 566 | decrement_hugepage_resv_vma(h, vma); |
| 567 | break; |
| 568 | } |
Andrew Morton | 3abf7af | 2007-07-19 01:49:08 -0700 | [diff] [blame] | 569 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | } |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 571 | |
| 572 | mpol_cond_put(mpol); |
| 573 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
| 574 | goto retry_cpuset; |
| 575 | return page; |
| 576 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 577 | err: |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 578 | mpol_cond_put(mpol); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 579 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 580 | } |
| 581 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 582 | static void update_and_free_page(struct hstate *h, struct page *page) |
Adam Litke | 6af2acb | 2007-10-16 01:26:16 -0700 | [diff] [blame] | 583 | { |
| 584 | int i; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 585 | |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 586 | VM_BUG_ON(h->order >= MAX_ORDER); |
| 587 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 588 | h->nr_huge_pages--; |
| 589 | h->nr_huge_pages_node[page_to_nid(page)]--; |
| 590 | for (i = 0; i < pages_per_huge_page(h); i++) { |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 591 | page[i].flags &= ~(1 << PG_locked | 1 << PG_error | |
| 592 | 1 << PG_referenced | 1 << PG_dirty | |
| 593 | 1 << PG_active | 1 << PG_reserved | |
| 594 | 1 << PG_private | 1 << PG_writeback); |
Adam Litke | 6af2acb | 2007-10-16 01:26:16 -0700 | [diff] [blame] | 595 | } |
| 596 | set_compound_page_dtor(page, NULL); |
| 597 | set_page_refcounted(page); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 598 | arch_release_hugepage(page); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 599 | __free_pages(page, huge_page_order(h)); |
Adam Litke | 6af2acb | 2007-10-16 01:26:16 -0700 | [diff] [blame] | 600 | } |
| 601 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 602 | struct hstate *size_to_hstate(unsigned long size) |
| 603 | { |
| 604 | struct hstate *h; |
| 605 | |
| 606 | for_each_hstate(h) { |
| 607 | if (huge_page_size(h) == size) |
| 608 | return h; |
| 609 | } |
| 610 | return NULL; |
| 611 | } |
| 612 | |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 613 | static void free_huge_page(struct page *page) |
| 614 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 615 | /* |
| 616 | * Can't pass hstate in here because it is called from the |
| 617 | * compound page destructor. |
| 618 | */ |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 619 | struct hstate *h = page_hstate(page); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 620 | int nid = page_to_nid(page); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 621 | struct hugepage_subpool *spool = |
| 622 | (struct hugepage_subpool *)page_private(page); |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 623 | |
Andy Whitcroft | e5df70a | 2008-02-23 15:23:32 -0800 | [diff] [blame] | 624 | set_page_private(page, 0); |
Mel Gorman | 23be746 | 2010-04-23 13:17:56 -0400 | [diff] [blame] | 625 | page->mapping = NULL; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 626 | BUG_ON(page_count(page)); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 627 | BUG_ON(page_mapcount(page)); |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 628 | INIT_LIST_HEAD(&page->lru); |
| 629 | |
| 630 | spin_lock(&hugetlb_lock); |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 631 | 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] | 632 | update_and_free_page(h, page); |
| 633 | h->surplus_huge_pages--; |
| 634 | h->surplus_huge_pages_node[nid]--; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 635 | } else { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 636 | enqueue_huge_page(h, page); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 637 | } |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 638 | spin_unlock(&hugetlb_lock); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 639 | hugepage_subpool_put_pages(spool, 1); |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 640 | } |
| 641 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 642 | 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] | 643 | { |
| 644 | set_compound_page_dtor(page, free_huge_page); |
| 645 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 646 | h->nr_huge_pages++; |
| 647 | h->nr_huge_pages_node[nid]++; |
Andi Kleen | b7ba30c | 2008-07-23 21:27:40 -0700 | [diff] [blame] | 648 | spin_unlock(&hugetlb_lock); |
| 649 | put_page(page); /* free it into the hugepage allocator */ |
| 650 | } |
| 651 | |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 652 | static void prep_compound_gigantic_page(struct page *page, unsigned long order) |
| 653 | { |
| 654 | int i; |
| 655 | int nr_pages = 1 << order; |
| 656 | struct page *p = page + 1; |
| 657 | |
| 658 | /* we rely on prep_new_huge_page to set the destructor */ |
| 659 | set_compound_order(page, order); |
| 660 | __SetPageHead(page); |
| 661 | for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { |
| 662 | __SetPageTail(p); |
Youquan Song | 58a84aa | 2011-12-08 14:34:18 -0800 | [diff] [blame] | 663 | set_page_count(p, 0); |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 664 | p->first_page = page; |
| 665 | } |
| 666 | } |
| 667 | |
| 668 | int PageHuge(struct page *page) |
| 669 | { |
| 670 | compound_page_dtor *dtor; |
| 671 | |
| 672 | if (!PageCompound(page)) |
| 673 | return 0; |
| 674 | |
| 675 | page = compound_head(page); |
| 676 | dtor = get_compound_page_dtor(page); |
| 677 | |
| 678 | return dtor == free_huge_page; |
| 679 | } |
Naoya Horiguchi | 43131e1 | 2010-05-28 09:29:22 +0900 | [diff] [blame] | 680 | EXPORT_SYMBOL_GPL(PageHuge); |
| 681 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 682 | 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] | 683 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | struct page *page; |
Joe Jin | f96efd5 | 2007-07-15 23:38:12 -0700 | [diff] [blame] | 685 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 686 | if (h->order >= MAX_ORDER) |
| 687 | return NULL; |
| 688 | |
Mel Gorman | 6484eb3 | 2009-06-16 15:31:54 -0700 | [diff] [blame] | 689 | page = alloc_pages_exact_node(nid, |
Nishanth Aravamudan | 551883a | 2008-04-29 00:58:26 -0700 | [diff] [blame] | 690 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
| 691 | __GFP_REPEAT|__GFP_NOWARN, |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 692 | huge_page_order(h)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | if (page) { |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 694 | if (arch_prepare_hugepage(page)) { |
Gerald Schaefer | caff3a2 | 2008-08-12 15:08:38 -0700 | [diff] [blame] | 695 | __free_pages(page, huge_page_order(h)); |
Harvey Harrison | 7b8ee84 | 2008-04-28 14:13:19 -0700 | [diff] [blame] | 696 | return NULL; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 697 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 698 | prep_new_huge_page(h, page, nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | } |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 700 | |
| 701 | return page; |
| 702 | } |
| 703 | |
Andi Kleen | 5ced66c | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 704 | /* |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 705 | * common helper functions for hstate_next_node_to_{alloc|free}. |
| 706 | * We may have allocated or freed a huge page based on a different |
| 707 | * nodes_allowed previously, so h->next_node_to_{alloc|free} might |
| 708 | * be outside of *nodes_allowed. Ensure that we use an allowed |
| 709 | * node for alloc or free. |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 710 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 711 | static int next_node_allowed(int nid, nodemask_t *nodes_allowed) |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 712 | { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 713 | nid = next_node(nid, *nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 714 | if (nid == MAX_NUMNODES) |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 715 | nid = first_node(*nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 716 | VM_BUG_ON(nid >= MAX_NUMNODES); |
| 717 | |
| 718 | return nid; |
| 719 | } |
| 720 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 721 | static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed) |
Andi Kleen | 5ced66c | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 722 | { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 723 | if (!node_isset(nid, *nodes_allowed)) |
| 724 | nid = next_node_allowed(nid, nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 725 | return nid; |
Andi Kleen | 5ced66c | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 726 | } |
| 727 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 728 | /* |
| 729 | * returns the previously saved node ["this node"] from which to |
| 730 | * allocate a persistent huge page for the pool and advance the |
| 731 | * next node from which to allocate, handling wrap at end of node |
| 732 | * mask. |
| 733 | */ |
| 734 | static int hstate_next_node_to_alloc(struct hstate *h, |
| 735 | nodemask_t *nodes_allowed) |
| 736 | { |
| 737 | int nid; |
| 738 | |
| 739 | VM_BUG_ON(!nodes_allowed); |
| 740 | |
| 741 | nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); |
| 742 | h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); |
| 743 | |
| 744 | return nid; |
| 745 | } |
| 746 | |
| 747 | 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] | 748 | { |
| 749 | struct page *page; |
| 750 | int start_nid; |
| 751 | int next_nid; |
| 752 | int ret = 0; |
| 753 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 754 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 755 | next_nid = start_nid; |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 756 | |
| 757 | do { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 758 | page = alloc_fresh_huge_page_node(h, next_nid); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 759 | if (page) { |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 760 | ret = 1; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 761 | break; |
| 762 | } |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 763 | next_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 764 | } while (next_nid != start_nid); |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 765 | |
Adam Litke | 3b11630 | 2008-04-28 02:13:06 -0700 | [diff] [blame] | 766 | if (ret) |
| 767 | count_vm_event(HTLB_BUDDY_PGALLOC); |
| 768 | else |
| 769 | count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
| 770 | |
Nishanth Aravamudan | 63b4613 | 2007-10-16 01:26:24 -0700 | [diff] [blame] | 771 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | } |
| 773 | |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 774 | /* |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 775 | * helper for free_pool_huge_page() - return the previously saved |
| 776 | * node ["this node"] from which to free a huge page. Advance the |
| 777 | * next node id whether or not we find a free huge page to free so |
| 778 | * that the next attempt to free addresses the next node. |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 779 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 780 | 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] | 781 | { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 782 | int nid; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 783 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 784 | VM_BUG_ON(!nodes_allowed); |
| 785 | |
| 786 | nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); |
| 787 | h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); |
| 788 | |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 789 | return nid; |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 790 | } |
| 791 | |
| 792 | /* |
| 793 | * Free huge page from pool from next node to free. |
| 794 | * Attempt to keep persistent huge pages more or less |
| 795 | * balanced over allowed nodes. |
| 796 | * Called with hugetlb_lock locked. |
| 797 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 798 | static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, |
| 799 | bool acct_surplus) |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 800 | { |
| 801 | int start_nid; |
| 802 | int next_nid; |
| 803 | int ret = 0; |
| 804 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 805 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 806 | next_nid = start_nid; |
| 807 | |
| 808 | do { |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 809 | /* |
| 810 | * If we're returning unused surplus pages, only examine |
| 811 | * nodes with surplus pages. |
| 812 | */ |
| 813 | if ((!acct_surplus || h->surplus_huge_pages_node[next_nid]) && |
| 814 | !list_empty(&h->hugepage_freelists[next_nid])) { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 815 | struct page *page = |
| 816 | list_entry(h->hugepage_freelists[next_nid].next, |
| 817 | struct page, lru); |
| 818 | list_del(&page->lru); |
| 819 | h->free_huge_pages--; |
| 820 | h->free_huge_pages_node[next_nid]--; |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 821 | if (acct_surplus) { |
| 822 | h->surplus_huge_pages--; |
| 823 | h->surplus_huge_pages_node[next_nid]--; |
| 824 | } |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 825 | update_and_free_page(h, page); |
| 826 | ret = 1; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 827 | break; |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 828 | } |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 829 | next_nid = hstate_next_node_to_free(h, nodes_allowed); |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 830 | } while (next_nid != start_nid); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 831 | |
| 832 | return ret; |
| 833 | } |
| 834 | |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 835 | static struct page *alloc_buddy_huge_page(struct hstate *h, int nid) |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 836 | { |
| 837 | struct page *page; |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 838 | unsigned int r_nid; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 839 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 840 | if (h->order >= MAX_ORDER) |
| 841 | return NULL; |
| 842 | |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 843 | /* |
| 844 | * Assume we will successfully allocate the surplus page to |
| 845 | * prevent racing processes from causing the surplus to exceed |
| 846 | * overcommit |
| 847 | * |
| 848 | * This however introduces a different race, where a process B |
| 849 | * tries to grow the static hugepage pool while alloc_pages() is |
| 850 | * called by process A. B will only examine the per-node |
| 851 | * counters in determining if surplus huge pages can be |
| 852 | * converted to normal huge pages in adjust_pool_surplus(). A |
| 853 | * won't be able to increment the per-node counter, until the |
| 854 | * lock is dropped by B, but B doesn't drop hugetlb_lock until |
| 855 | * no more huge pages can be converted from surplus to normal |
| 856 | * state (and doesn't try to convert again). Thus, we have a |
| 857 | * case where a surplus huge page exists, the pool is grown, and |
| 858 | * the surplus huge page still exists after, even though it |
| 859 | * should just have been converted to a normal huge page. This |
| 860 | * does not leak memory, though, as the hugepage will be freed |
| 861 | * once it is out of use. It also does not allow the counters to |
| 862 | * go out of whack in adjust_pool_surplus() as we don't modify |
| 863 | * the node values until we've gotten the hugepage and only the |
| 864 | * per-node value is checked there. |
| 865 | */ |
| 866 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 867 | if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 868 | spin_unlock(&hugetlb_lock); |
| 869 | return NULL; |
| 870 | } else { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 871 | h->nr_huge_pages++; |
| 872 | h->surplus_huge_pages++; |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 873 | } |
| 874 | spin_unlock(&hugetlb_lock); |
| 875 | |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 876 | if (nid == NUMA_NO_NODE) |
| 877 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP| |
| 878 | __GFP_REPEAT|__GFP_NOWARN, |
| 879 | huge_page_order(h)); |
| 880 | else |
| 881 | page = alloc_pages_exact_node(nid, |
| 882 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
| 883 | __GFP_REPEAT|__GFP_NOWARN, huge_page_order(h)); |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 884 | |
Gerald Schaefer | caff3a2 | 2008-08-12 15:08:38 -0700 | [diff] [blame] | 885 | if (page && arch_prepare_hugepage(page)) { |
| 886 | __free_pages(page, huge_page_order(h)); |
Hillf Danton | ea5768c | 2012-01-10 15:08:30 -0800 | [diff] [blame] | 887 | page = NULL; |
Gerald Schaefer | caff3a2 | 2008-08-12 15:08:38 -0700 | [diff] [blame] | 888 | } |
| 889 | |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 890 | spin_lock(&hugetlb_lock); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 891 | if (page) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 892 | r_nid = page_to_nid(page); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 893 | set_compound_page_dtor(page, free_huge_page); |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 894 | /* |
| 895 | * We incremented the global counters already |
| 896 | */ |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 897 | h->nr_huge_pages_node[r_nid]++; |
| 898 | h->surplus_huge_pages_node[r_nid]++; |
Adam Litke | 3b11630 | 2008-04-28 02:13:06 -0700 | [diff] [blame] | 899 | __count_vm_event(HTLB_BUDDY_PGALLOC); |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 900 | } else { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 901 | h->nr_huge_pages--; |
| 902 | h->surplus_huge_pages--; |
Adam Litke | 3b11630 | 2008-04-28 02:13:06 -0700 | [diff] [blame] | 903 | __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 904 | } |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 905 | spin_unlock(&hugetlb_lock); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 906 | |
| 907 | return page; |
| 908 | } |
| 909 | |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 910 | /* |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 911 | * This allocation function is useful in the context where vma is irrelevant. |
| 912 | * E.g. soft-offlining uses this function because it only cares physical |
| 913 | * address of error page. |
| 914 | */ |
| 915 | struct page *alloc_huge_page_node(struct hstate *h, int nid) |
| 916 | { |
| 917 | struct page *page; |
| 918 | |
| 919 | spin_lock(&hugetlb_lock); |
| 920 | page = dequeue_huge_page_node(h, nid); |
| 921 | spin_unlock(&hugetlb_lock); |
| 922 | |
| 923 | if (!page) |
| 924 | page = alloc_buddy_huge_page(h, nid); |
| 925 | |
| 926 | return page; |
| 927 | } |
| 928 | |
| 929 | /* |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 930 | * Increase the hugetlb pool such that it can accommodate a reservation |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 931 | * of size 'delta'. |
| 932 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 933 | static int gather_surplus_pages(struct hstate *h, int delta) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 934 | { |
| 935 | struct list_head surplus_list; |
| 936 | struct page *page, *tmp; |
| 937 | int ret, i; |
| 938 | int needed, allocated; |
Hillf Danton | 28073b0 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 939 | bool alloc_ok = true; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 940 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 941 | needed = (h->resv_huge_pages + delta) - h->free_huge_pages; |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 942 | if (needed <= 0) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 943 | h->resv_huge_pages += delta; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 944 | return 0; |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 945 | } |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 946 | |
| 947 | allocated = 0; |
| 948 | INIT_LIST_HEAD(&surplus_list); |
| 949 | |
| 950 | ret = -ENOMEM; |
| 951 | retry: |
| 952 | spin_unlock(&hugetlb_lock); |
| 953 | for (i = 0; i < needed; i++) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 954 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
Hillf Danton | 28073b0 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 955 | if (!page) { |
| 956 | alloc_ok = false; |
| 957 | break; |
| 958 | } |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 959 | list_add(&page->lru, &surplus_list); |
| 960 | } |
Hillf Danton | 28073b0 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 961 | allocated += i; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 962 | |
| 963 | /* |
| 964 | * After retaking hugetlb_lock, we need to recalculate 'needed' |
| 965 | * because either resv_huge_pages or free_huge_pages may have changed. |
| 966 | */ |
| 967 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 968 | needed = (h->resv_huge_pages + delta) - |
| 969 | (h->free_huge_pages + allocated); |
Hillf Danton | 28073b0 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 970 | if (needed > 0) { |
| 971 | if (alloc_ok) |
| 972 | goto retry; |
| 973 | /* |
| 974 | * We were not able to allocate enough pages to |
| 975 | * satisfy the entire reservation so we free what |
| 976 | * we've allocated so far. |
| 977 | */ |
| 978 | goto free; |
| 979 | } |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 980 | /* |
| 981 | * The surplus_list now contains _at_least_ the number of extra pages |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 982 | * needed to accommodate the reservation. Add the appropriate number |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 983 | * 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] | 984 | * allocator. Commit the entire reservation here to prevent another |
| 985 | * process from stealing the pages as they are added to the pool but |
| 986 | * before they are reserved. |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 987 | */ |
| 988 | needed += allocated; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 989 | h->resv_huge_pages += delta; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 990 | ret = 0; |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 991 | |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 992 | /* Free the needed pages to the hugetlb pool */ |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 993 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 994 | if ((--needed) < 0) |
| 995 | break; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 996 | list_del(&page->lru); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 997 | /* |
| 998 | * This page is now managed by the hugetlb allocator and has |
| 999 | * no users -- drop the buddy allocator's reference. |
| 1000 | */ |
| 1001 | put_page_testzero(page); |
| 1002 | VM_BUG_ON(page_count(page)); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1003 | enqueue_huge_page(h, page); |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 1004 | } |
Hillf Danton | 28073b0 | 2012-03-21 16:34:00 -0700 | [diff] [blame] | 1005 | free: |
Hillf Danton | b0365c8 | 2011-12-28 15:57:16 -0800 | [diff] [blame] | 1006 | spin_unlock(&hugetlb_lock); |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 1007 | |
| 1008 | /* Free unnecessary surplus pages to the buddy allocator */ |
| 1009 | if (!list_empty(&surplus_list)) { |
Adam Litke | 19fc3f0 | 2008-04-28 02:12:20 -0700 | [diff] [blame] | 1010 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
| 1011 | list_del(&page->lru); |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 1012 | put_page(page); |
Adam Litke | af767cb | 2007-10-16 01:26:25 -0700 | [diff] [blame] | 1013 | } |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1014 | } |
Naoya Horiguchi | a9869b8 | 2010-09-08 10:19:37 +0900 | [diff] [blame] | 1015 | spin_lock(&hugetlb_lock); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1016 | |
| 1017 | return ret; |
| 1018 | } |
| 1019 | |
| 1020 | /* |
| 1021 | * When releasing a hugetlb pool reservation, any surplus pages that were |
| 1022 | * allocated to satisfy the reservation must be explicitly freed if they were |
| 1023 | * never used. |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 1024 | * Called with hugetlb_lock held. |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1025 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1026 | static void return_unused_surplus_pages(struct hstate *h, |
| 1027 | unsigned long unused_resv_pages) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1028 | { |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1029 | unsigned long nr_pages; |
| 1030 | |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 1031 | /* Uncommit the reservation */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1032 | h->resv_huge_pages -= unused_resv_pages; |
Adam Litke | ac09b3a | 2008-03-04 14:29:38 -0800 | [diff] [blame] | 1033 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1034 | /* Cannot return gigantic pages currently */ |
| 1035 | if (h->order >= MAX_ORDER) |
| 1036 | return; |
| 1037 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1038 | nr_pages = min(unused_resv_pages, h->surplus_huge_pages); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1039 | |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 1040 | /* |
| 1041 | * We want to release as many surplus pages as possible, spread |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1042 | * evenly across all nodes with memory. Iterate across these nodes |
| 1043 | * until we can no longer free unreserved surplus pages. This occurs |
| 1044 | * when the nodes with surplus pages have no free pages. |
| 1045 | * free_pool_huge_page() will balance the the freed pages across the |
| 1046 | * on-line nodes with memory and will handle the hstate accounting. |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 1047 | */ |
| 1048 | while (nr_pages--) { |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1049 | if (!free_pool_huge_page(h, &node_states[N_HIGH_MEMORY], 1)) |
Lee Schermerhorn | 685f345 | 2009-09-21 17:01:23 -0700 | [diff] [blame] | 1050 | break; |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 1051 | } |
| 1052 | } |
| 1053 | |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1054 | /* |
| 1055 | * Determine if the huge page at addr within the vma has an associated |
| 1056 | * reservation. Where it does not we will need to logically increase |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1057 | * reservation and actually increase subpool usage before an allocation |
| 1058 | * can occur. Where any new reservation would be required the |
| 1059 | * reservation change is prepared, but not committed. Once the page |
| 1060 | * has been allocated from the subpool and instantiated the change should |
| 1061 | * be committed via vma_commit_reservation. No action is required on |
| 1062 | * failure. |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1063 | */ |
Roel Kluin | e2f17d9 | 2009-03-31 15:23:15 -0700 | [diff] [blame] | 1064 | static long vma_needs_reservation(struct hstate *h, |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1065 | struct vm_area_struct *vma, unsigned long addr) |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1066 | { |
| 1067 | struct address_space *mapping = vma->vm_file->f_mapping; |
| 1068 | struct inode *inode = mapping->host; |
| 1069 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 1070 | if (vma->vm_flags & VM_MAYSHARE) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1071 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1072 | return region_chg(&inode->i_mapping->private_list, |
| 1073 | idx, idx + 1); |
| 1074 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1075 | } else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
| 1076 | return 1; |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1077 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1078 | } else { |
Roel Kluin | e2f17d9 | 2009-03-31 15:23:15 -0700 | [diff] [blame] | 1079 | long err; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1080 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1081 | struct resv_map *reservations = vma_resv_map(vma); |
| 1082 | |
| 1083 | err = region_chg(&reservations->regions, idx, idx + 1); |
| 1084 | if (err < 0) |
| 1085 | return err; |
| 1086 | return 0; |
| 1087 | } |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1088 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1089 | static void vma_commit_reservation(struct hstate *h, |
| 1090 | struct vm_area_struct *vma, unsigned long addr) |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1091 | { |
| 1092 | struct address_space *mapping = vma->vm_file->f_mapping; |
| 1093 | struct inode *inode = mapping->host; |
| 1094 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 1095 | if (vma->vm_flags & VM_MAYSHARE) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1096 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1097 | region_add(&inode->i_mapping->private_list, idx, idx + 1); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1098 | |
| 1099 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1100 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 1101 | struct resv_map *reservations = vma_resv_map(vma); |
| 1102 | |
| 1103 | /* Mark this page used in the map. */ |
| 1104 | region_add(&reservations->regions, idx, idx + 1); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1105 | } |
| 1106 | } |
| 1107 | |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 1108 | static struct page *alloc_huge_page(struct vm_area_struct *vma, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 1109 | unsigned long addr, int avoid_reserve) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | { |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1111 | struct hugepage_subpool *spool = subpool_vma(vma); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1112 | struct hstate *h = hstate_vma(vma); |
Adam Litke | 348ea20 | 2007-11-14 16:59:37 -0800 | [diff] [blame] | 1113 | struct page *page; |
Roel Kluin | e2f17d9 | 2009-03-31 15:23:15 -0700 | [diff] [blame] | 1114 | long chg; |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 1115 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1116 | /* |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1117 | * Processes that did not create the mapping will have no |
| 1118 | * reserves and will not have accounted against subpool |
| 1119 | * limit. Check that the subpool limit can be made before |
| 1120 | * satisfying the allocation MAP_NORESERVE mappings may also |
| 1121 | * need pages and subpool limit allocated allocated if no reserve |
| 1122 | * mapping overlaps. |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1123 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1124 | chg = vma_needs_reservation(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1125 | if (chg < 0) |
Hugh Dickins | e0dcd8a | 2011-06-05 22:03:13 -0700 | [diff] [blame] | 1126 | return ERR_PTR(-VM_FAULT_OOM); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1127 | if (chg) |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1128 | if (hugepage_subpool_get_pages(spool, chg)) |
Hugh Dickins | e0dcd8a | 2011-06-05 22:03:13 -0700 | [diff] [blame] | 1129 | return ERR_PTR(-VM_FAULT_SIGBUS); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1130 | |
| 1131 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1132 | page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1133 | spin_unlock(&hugetlb_lock); |
| 1134 | |
| 1135 | if (!page) { |
Naoya Horiguchi | bf50bab | 2010-09-08 10:19:33 +0900 | [diff] [blame] | 1136 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1137 | if (!page) { |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1138 | hugepage_subpool_put_pages(spool, chg); |
Mel Gorman | 4a6018f | 2010-05-11 14:06:53 -0700 | [diff] [blame] | 1139 | return ERR_PTR(-VM_FAULT_SIGBUS); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1140 | } |
| 1141 | } |
| 1142 | |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1143 | set_page_private(page, (unsigned long)spool); |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 1144 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1145 | vma_commit_reservation(h, vma, addr); |
Andy Whitcroft | c37f9fb | 2008-07-23 21:27:30 -0700 | [diff] [blame] | 1146 | |
Adam Litke | 90d8b7e | 2007-11-14 16:59:42 -0800 | [diff] [blame] | 1147 | return page; |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 1148 | } |
| 1149 | |
Cyrill Gorcunov | 91f4766 | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 1150 | int __weak alloc_bootmem_huge_page(struct hstate *h) |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1151 | { |
| 1152 | struct huge_bootmem_page *m; |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1153 | int nr_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1154 | |
| 1155 | while (nr_nodes) { |
| 1156 | void *addr; |
| 1157 | |
| 1158 | addr = __alloc_bootmem_node_nopanic( |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1159 | NODE_DATA(hstate_next_node_to_alloc(h, |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1160 | &node_states[N_HIGH_MEMORY])), |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1161 | huge_page_size(h), huge_page_size(h), 0); |
| 1162 | |
| 1163 | if (addr) { |
| 1164 | /* |
| 1165 | * Use the beginning of the huge page to store the |
| 1166 | * huge_bootmem_page struct (until gather_bootmem |
| 1167 | * puts them into the mem_map). |
| 1168 | */ |
| 1169 | m = addr; |
Cyrill Gorcunov | 91f4766 | 2009-01-06 14:40:33 -0800 | [diff] [blame] | 1170 | goto found; |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1171 | } |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1172 | nr_nodes--; |
| 1173 | } |
| 1174 | return 0; |
| 1175 | |
| 1176 | found: |
| 1177 | BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1)); |
| 1178 | /* Put them into a private list first because mem_map is not up yet */ |
| 1179 | list_add(&m->list, &huge_boot_pages); |
| 1180 | m->hstate = h; |
| 1181 | return 1; |
| 1182 | } |
| 1183 | |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 1184 | static void prep_compound_huge_page(struct page *page, int order) |
| 1185 | { |
| 1186 | if (unlikely(order > (MAX_ORDER - 1))) |
| 1187 | prep_compound_gigantic_page(page, order); |
| 1188 | else |
| 1189 | prep_compound_page(page, order); |
| 1190 | } |
| 1191 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1192 | /* Put bootmem huge pages into the standard lists after mem_map is up */ |
| 1193 | static void __init gather_bootmem_prealloc(void) |
| 1194 | { |
| 1195 | struct huge_bootmem_page *m; |
| 1196 | |
| 1197 | list_for_each_entry(m, &huge_boot_pages, list) { |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1198 | struct hstate *h = m->hstate; |
Becky Bruce | ee8f248 | 2011-07-25 17:11:50 -0700 | [diff] [blame] | 1199 | struct page *page; |
| 1200 | |
| 1201 | #ifdef CONFIG_HIGHMEM |
| 1202 | page = pfn_to_page(m->phys >> PAGE_SHIFT); |
| 1203 | free_bootmem_late((unsigned long)m, |
| 1204 | sizeof(struct huge_bootmem_page)); |
| 1205 | #else |
| 1206 | page = virt_to_page(m); |
| 1207 | #endif |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1208 | __ClearPageReserved(page); |
| 1209 | WARN_ON(page_count(page) != 1); |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 1210 | prep_compound_huge_page(page, h->order); |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1211 | prep_new_huge_page(h, page, page_to_nid(page)); |
Rafael Aquini | b0320c7 | 2011-06-15 15:08:39 -0700 | [diff] [blame] | 1212 | /* |
| 1213 | * If we had gigantic hugepages allocated at boot time, we need |
| 1214 | * to restore the 'stolen' pages to totalram_pages in order to |
| 1215 | * fix confusing memory reports from free(1) and another |
| 1216 | * side-effects, like CommitLimit going negative. |
| 1217 | */ |
| 1218 | if (h->order > (MAX_ORDER - 1)) |
| 1219 | totalram_pages += 1 << h->order; |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1220 | } |
| 1221 | } |
| 1222 | |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1223 | static void __init hugetlb_hstate_alloc_pages(struct hstate *h) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | { |
| 1225 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1226 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1227 | for (i = 0; i < h->max_huge_pages; ++i) { |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1228 | if (h->order >= MAX_ORDER) { |
| 1229 | if (!alloc_bootmem_huge_page(h)) |
| 1230 | break; |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1231 | } else if (!alloc_fresh_huge_page(h, |
| 1232 | &node_states[N_HIGH_MEMORY])) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1234 | } |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1235 | h->max_huge_pages = i; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1236 | } |
| 1237 | |
| 1238 | static void __init hugetlb_init_hstates(void) |
| 1239 | { |
| 1240 | struct hstate *h; |
| 1241 | |
| 1242 | for_each_hstate(h) { |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1243 | /* oversize hugepages were init'ed in early boot */ |
| 1244 | if (h->order < MAX_ORDER) |
| 1245 | hugetlb_hstate_alloc_pages(h); |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1246 | } |
| 1247 | } |
| 1248 | |
Andi Kleen | 4abd32d | 2008-07-23 21:27:49 -0700 | [diff] [blame] | 1249 | static char * __init memfmt(char *buf, unsigned long n) |
| 1250 | { |
| 1251 | if (n >= (1UL << 30)) |
| 1252 | sprintf(buf, "%lu GB", n >> 30); |
| 1253 | else if (n >= (1UL << 20)) |
| 1254 | sprintf(buf, "%lu MB", n >> 20); |
| 1255 | else |
| 1256 | sprintf(buf, "%lu KB", n >> 10); |
| 1257 | return buf; |
| 1258 | } |
| 1259 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1260 | static void __init report_hugepages(void) |
| 1261 | { |
| 1262 | struct hstate *h; |
| 1263 | |
| 1264 | for_each_hstate(h) { |
Andi Kleen | 4abd32d | 2008-07-23 21:27:49 -0700 | [diff] [blame] | 1265 | char buf[32]; |
| 1266 | printk(KERN_INFO "HugeTLB registered %s page size, " |
| 1267 | "pre-allocated %ld pages\n", |
| 1268 | memfmt(buf, huge_page_size(h)), |
| 1269 | h->free_huge_pages); |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1270 | } |
| 1271 | } |
| 1272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | #ifdef CONFIG_HIGHMEM |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1274 | static void try_to_free_low(struct hstate *h, unsigned long count, |
| 1275 | nodemask_t *nodes_allowed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | { |
Christoph Lameter | 4415cc8 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 1277 | int i; |
| 1278 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1279 | if (h->order >= MAX_ORDER) |
| 1280 | return; |
| 1281 | |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1282 | for_each_node_mask(i, *nodes_allowed) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | struct page *page, *next; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1284 | struct list_head *freel = &h->hugepage_freelists[i]; |
| 1285 | list_for_each_entry_safe(page, next, freel, lru) { |
| 1286 | if (count >= h->nr_huge_pages) |
Adam Litke | 6b0c880 | 2007-10-16 01:26:23 -0700 | [diff] [blame] | 1287 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | if (PageHighMem(page)) |
| 1289 | continue; |
| 1290 | list_del(&page->lru); |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1291 | update_and_free_page(h, page); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1292 | h->free_huge_pages--; |
| 1293 | h->free_huge_pages_node[page_to_nid(page)]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | } |
| 1295 | } |
| 1296 | } |
| 1297 | #else |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1298 | static inline void try_to_free_low(struct hstate *h, unsigned long count, |
| 1299 | nodemask_t *nodes_allowed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1300 | { |
| 1301 | } |
| 1302 | #endif |
| 1303 | |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1304 | /* |
| 1305 | * Increment or decrement surplus_huge_pages. Keep node-specific counters |
| 1306 | * balanced by operating on them in a round-robin fashion. |
| 1307 | * Returns 1 if an adjustment was made. |
| 1308 | */ |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1309 | static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed, |
| 1310 | int delta) |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1311 | { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1312 | int start_nid, next_nid; |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1313 | int ret = 0; |
| 1314 | |
| 1315 | VM_BUG_ON(delta != -1 && delta != 1); |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1316 | |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1317 | if (delta < 0) |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1318 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1319 | else |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1320 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1321 | next_nid = start_nid; |
| 1322 | |
| 1323 | do { |
| 1324 | int nid = next_nid; |
| 1325 | if (delta < 0) { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1326 | /* |
| 1327 | * To shrink on this node, there must be a surplus page |
| 1328 | */ |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1329 | if (!h->surplus_huge_pages_node[nid]) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1330 | next_nid = hstate_next_node_to_alloc(h, |
| 1331 | nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1332 | continue; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1333 | } |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1334 | } |
| 1335 | if (delta > 0) { |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1336 | /* |
| 1337 | * Surplus cannot exceed the total number of pages |
| 1338 | */ |
| 1339 | if (h->surplus_huge_pages_node[nid] >= |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1340 | h->nr_huge_pages_node[nid]) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1341 | next_nid = hstate_next_node_to_free(h, |
| 1342 | nodes_allowed); |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1343 | continue; |
Lee Schermerhorn | 9a76db0 | 2009-12-14 17:58:15 -0800 | [diff] [blame] | 1344 | } |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1345 | } |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1346 | |
| 1347 | h->surplus_huge_pages += delta; |
| 1348 | h->surplus_huge_pages_node[nid] += delta; |
| 1349 | ret = 1; |
| 1350 | break; |
Lee Schermerhorn | e8c5c82 | 2009-09-21 17:01:22 -0700 | [diff] [blame] | 1351 | } while (next_nid != start_nid); |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1352 | |
Wu Fengguang | 20a0307 | 2009-06-16 15:32:22 -0700 | [diff] [blame] | 1353 | return ret; |
| 1354 | } |
| 1355 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1356 | #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] | 1357 | static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count, |
| 1358 | nodemask_t *nodes_allowed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1359 | { |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1360 | unsigned long min_count, ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1362 | if (h->order >= MAX_ORDER) |
| 1363 | return h->max_huge_pages; |
| 1364 | |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1365 | /* |
| 1366 | * Increase the pool size |
| 1367 | * First take pages out of surplus state. Then make up the |
| 1368 | * remaining difference by allocating fresh huge pages. |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 1369 | * |
| 1370 | * We might race with alloc_buddy_huge_page() here and be unable |
| 1371 | * to convert a surplus huge page to a normal huge page. That is |
| 1372 | * not critical, though, it just means the overall size of the |
| 1373 | * pool might be one hugepage larger than it needs to be, but |
| 1374 | * within all the constraints specified by the sysctls. |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1375 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | spin_lock(&hugetlb_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1377 | while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1378 | if (!adjust_pool_surplus(h, nodes_allowed, -1)) |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1379 | break; |
| 1380 | } |
| 1381 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1382 | while (count > persistent_huge_pages(h)) { |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1383 | /* |
| 1384 | * If this allocation races such that we no longer need the |
| 1385 | * page, free_huge_page will handle it by freeing the page |
| 1386 | * and reducing the surplus. |
| 1387 | */ |
| 1388 | spin_unlock(&hugetlb_lock); |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1389 | ret = alloc_fresh_huge_page(h, nodes_allowed); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1390 | spin_lock(&hugetlb_lock); |
| 1391 | if (!ret) |
| 1392 | goto out; |
| 1393 | |
Mel Gorman | 536240f | 2009-12-14 17:59:56 -0800 | [diff] [blame] | 1394 | /* Bail for signals. Probably ctrl-c from user */ |
| 1395 | if (signal_pending(current)) |
| 1396 | goto out; |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1397 | } |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1398 | |
| 1399 | /* |
| 1400 | * Decrease the pool size |
| 1401 | * First return free pages to the buddy allocator (being careful |
| 1402 | * to keep enough around to satisfy reservations). Then place |
| 1403 | * pages into surplus state as needed so the pool will shrink |
| 1404 | * to the desired size as pages become free. |
Nishanth Aravamudan | d1c3fb1 | 2007-12-17 16:20:12 -0800 | [diff] [blame] | 1405 | * |
| 1406 | * By placing pages into the surplus state independent of the |
| 1407 | * overcommit value, we are allowing the surplus pool size to |
| 1408 | * exceed overcommit. There are few sane options here. Since |
| 1409 | * alloc_buddy_huge_page() is checking the global counter, |
| 1410 | * though, we'll note that we're not allowed to exceed surplus |
| 1411 | * and won't grow the pool anywhere else. Not until one of the |
| 1412 | * sysctls are changed, or the surplus pages go out of use. |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1413 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1414 | 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] | 1415 | min_count = max(count, min_count); |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1416 | try_to_free_low(h, min_count, nodes_allowed); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1417 | while (min_count < persistent_huge_pages(h)) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1418 | if (!free_pool_huge_page(h, nodes_allowed, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1420 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1421 | while (count < persistent_huge_pages(h)) { |
Lee Schermerhorn | 6ae11b2 | 2009-12-14 17:58:16 -0800 | [diff] [blame] | 1422 | if (!adjust_pool_surplus(h, nodes_allowed, 1)) |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1423 | break; |
| 1424 | } |
| 1425 | out: |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1426 | ret = persistent_huge_pages(h); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1427 | spin_unlock(&hugetlb_lock); |
Adam Litke | 7893d1d | 2007-10-16 01:26:18 -0700 | [diff] [blame] | 1428 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1429 | } |
| 1430 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1431 | #define HSTATE_ATTR_RO(_name) \ |
| 1432 | static struct kobj_attribute _name##_attr = __ATTR_RO(_name) |
| 1433 | |
| 1434 | #define HSTATE_ATTR(_name) \ |
| 1435 | static struct kobj_attribute _name##_attr = \ |
| 1436 | __ATTR(_name, 0644, _name##_show, _name##_store) |
| 1437 | |
| 1438 | static struct kobject *hugepages_kobj; |
| 1439 | static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; |
| 1440 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1441 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp); |
| 1442 | |
| 1443 | static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1444 | { |
| 1445 | int i; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1446 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1447 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1448 | if (hstate_kobjs[i] == kobj) { |
| 1449 | if (nidp) |
| 1450 | *nidp = NUMA_NO_NODE; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1451 | return &hstates[i]; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1452 | } |
| 1453 | |
| 1454 | return kobj_to_node_hstate(kobj, nidp); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1455 | } |
| 1456 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1457 | static ssize_t nr_hugepages_show_common(struct kobject *kobj, |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1458 | struct kobj_attribute *attr, char *buf) |
| 1459 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1460 | struct hstate *h; |
| 1461 | unsigned long nr_huge_pages; |
| 1462 | int nid; |
| 1463 | |
| 1464 | h = kobj_to_hstate(kobj, &nid); |
| 1465 | if (nid == NUMA_NO_NODE) |
| 1466 | nr_huge_pages = h->nr_huge_pages; |
| 1467 | else |
| 1468 | nr_huge_pages = h->nr_huge_pages_node[nid]; |
| 1469 | |
| 1470 | return sprintf(buf, "%lu\n", nr_huge_pages); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1471 | } |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1472 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1473 | static ssize_t nr_hugepages_store_common(bool obey_mempolicy, |
| 1474 | struct kobject *kobj, struct kobj_attribute *attr, |
| 1475 | const char *buf, size_t len) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1476 | { |
| 1477 | int err; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1478 | int nid; |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1479 | unsigned long count; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1480 | struct hstate *h; |
David Rientjes | bad44b5 | 2009-12-14 17:58:38 -0800 | [diff] [blame] | 1481 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1482 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1483 | err = strict_strtoul(buf, 10, &count); |
Eric B Munson | 73ae31e | 2011-01-13 15:47:28 -0800 | [diff] [blame] | 1484 | if (err) |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1485 | goto out; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1486 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1487 | h = kobj_to_hstate(kobj, &nid); |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1488 | if (h->order >= MAX_ORDER) { |
| 1489 | err = -EINVAL; |
| 1490 | goto out; |
| 1491 | } |
| 1492 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1493 | if (nid == NUMA_NO_NODE) { |
| 1494 | /* |
| 1495 | * global hstate attribute |
| 1496 | */ |
| 1497 | if (!(obey_mempolicy && |
| 1498 | init_nodemask_of_mempolicy(nodes_allowed))) { |
| 1499 | NODEMASK_FREE(nodes_allowed); |
| 1500 | nodes_allowed = &node_states[N_HIGH_MEMORY]; |
| 1501 | } |
| 1502 | } else if (nodes_allowed) { |
| 1503 | /* |
| 1504 | * per node hstate attribute: adjust count to global, |
| 1505 | * but restrict alloc/free to the specified node. |
| 1506 | */ |
| 1507 | count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; |
| 1508 | init_nodemask_of_node(nodes_allowed, nid); |
| 1509 | } else |
| 1510 | nodes_allowed = &node_states[N_HIGH_MEMORY]; |
| 1511 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1512 | h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1513 | |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1514 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1515 | NODEMASK_FREE(nodes_allowed); |
| 1516 | |
| 1517 | return len; |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1518 | out: |
| 1519 | NODEMASK_FREE(nodes_allowed); |
| 1520 | return err; |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1521 | } |
| 1522 | |
| 1523 | static ssize_t nr_hugepages_show(struct kobject *kobj, |
| 1524 | struct kobj_attribute *attr, char *buf) |
| 1525 | { |
| 1526 | return nr_hugepages_show_common(kobj, attr, buf); |
| 1527 | } |
| 1528 | |
| 1529 | static ssize_t nr_hugepages_store(struct kobject *kobj, |
| 1530 | struct kobj_attribute *attr, const char *buf, size_t len) |
| 1531 | { |
| 1532 | return nr_hugepages_store_common(false, kobj, attr, buf, len); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1533 | } |
| 1534 | HSTATE_ATTR(nr_hugepages); |
| 1535 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1536 | #ifdef CONFIG_NUMA |
| 1537 | |
| 1538 | /* |
| 1539 | * hstate attribute for optionally mempolicy-based constraint on persistent |
| 1540 | * huge page alloc/free. |
| 1541 | */ |
| 1542 | static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj, |
| 1543 | struct kobj_attribute *attr, char *buf) |
| 1544 | { |
| 1545 | return nr_hugepages_show_common(kobj, attr, buf); |
| 1546 | } |
| 1547 | |
| 1548 | static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj, |
| 1549 | struct kobj_attribute *attr, const char *buf, size_t len) |
| 1550 | { |
| 1551 | return nr_hugepages_store_common(true, kobj, attr, buf, len); |
| 1552 | } |
| 1553 | HSTATE_ATTR(nr_hugepages_mempolicy); |
| 1554 | #endif |
| 1555 | |
| 1556 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1557 | static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj, |
| 1558 | struct kobj_attribute *attr, char *buf) |
| 1559 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1560 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1561 | return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); |
| 1562 | } |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1563 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1564 | static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj, |
| 1565 | struct kobj_attribute *attr, const char *buf, size_t count) |
| 1566 | { |
| 1567 | int err; |
| 1568 | unsigned long input; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1569 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1570 | |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1571 | if (h->order >= MAX_ORDER) |
| 1572 | return -EINVAL; |
| 1573 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1574 | err = strict_strtoul(buf, 10, &input); |
| 1575 | if (err) |
Eric B Munson | 73ae31e | 2011-01-13 15:47:28 -0800 | [diff] [blame] | 1576 | return err; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1577 | |
| 1578 | spin_lock(&hugetlb_lock); |
| 1579 | h->nr_overcommit_huge_pages = input; |
| 1580 | spin_unlock(&hugetlb_lock); |
| 1581 | |
| 1582 | return count; |
| 1583 | } |
| 1584 | HSTATE_ATTR(nr_overcommit_hugepages); |
| 1585 | |
| 1586 | static ssize_t free_hugepages_show(struct kobject *kobj, |
| 1587 | struct kobj_attribute *attr, char *buf) |
| 1588 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1589 | struct hstate *h; |
| 1590 | unsigned long free_huge_pages; |
| 1591 | int nid; |
| 1592 | |
| 1593 | h = kobj_to_hstate(kobj, &nid); |
| 1594 | if (nid == NUMA_NO_NODE) |
| 1595 | free_huge_pages = h->free_huge_pages; |
| 1596 | else |
| 1597 | free_huge_pages = h->free_huge_pages_node[nid]; |
| 1598 | |
| 1599 | return sprintf(buf, "%lu\n", free_huge_pages); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1600 | } |
| 1601 | HSTATE_ATTR_RO(free_hugepages); |
| 1602 | |
| 1603 | static ssize_t resv_hugepages_show(struct kobject *kobj, |
| 1604 | struct kobj_attribute *attr, char *buf) |
| 1605 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1606 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1607 | return sprintf(buf, "%lu\n", h->resv_huge_pages); |
| 1608 | } |
| 1609 | HSTATE_ATTR_RO(resv_hugepages); |
| 1610 | |
| 1611 | static ssize_t surplus_hugepages_show(struct kobject *kobj, |
| 1612 | struct kobj_attribute *attr, char *buf) |
| 1613 | { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1614 | struct hstate *h; |
| 1615 | unsigned long surplus_huge_pages; |
| 1616 | int nid; |
| 1617 | |
| 1618 | h = kobj_to_hstate(kobj, &nid); |
| 1619 | if (nid == NUMA_NO_NODE) |
| 1620 | surplus_huge_pages = h->surplus_huge_pages; |
| 1621 | else |
| 1622 | surplus_huge_pages = h->surplus_huge_pages_node[nid]; |
| 1623 | |
| 1624 | return sprintf(buf, "%lu\n", surplus_huge_pages); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1625 | } |
| 1626 | HSTATE_ATTR_RO(surplus_hugepages); |
| 1627 | |
| 1628 | static struct attribute *hstate_attrs[] = { |
| 1629 | &nr_hugepages_attr.attr, |
| 1630 | &nr_overcommit_hugepages_attr.attr, |
| 1631 | &free_hugepages_attr.attr, |
| 1632 | &resv_hugepages_attr.attr, |
| 1633 | &surplus_hugepages_attr.attr, |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1634 | #ifdef CONFIG_NUMA |
| 1635 | &nr_hugepages_mempolicy_attr.attr, |
| 1636 | #endif |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1637 | NULL, |
| 1638 | }; |
| 1639 | |
| 1640 | static struct attribute_group hstate_attr_group = { |
| 1641 | .attrs = hstate_attrs, |
| 1642 | }; |
| 1643 | |
Jeff Mahoney | 094e953 | 2010-02-02 13:44:14 -0800 | [diff] [blame] | 1644 | static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent, |
| 1645 | struct kobject **hstate_kobjs, |
| 1646 | struct attribute_group *hstate_attr_group) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1647 | { |
| 1648 | int retval; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1649 | int hi = h - hstates; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1650 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1651 | hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); |
| 1652 | if (!hstate_kobjs[hi]) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1653 | return -ENOMEM; |
| 1654 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1655 | retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1656 | if (retval) |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1657 | kobject_put(hstate_kobjs[hi]); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1658 | |
| 1659 | return retval; |
| 1660 | } |
| 1661 | |
| 1662 | static void __init hugetlb_sysfs_init(void) |
| 1663 | { |
| 1664 | struct hstate *h; |
| 1665 | int err; |
| 1666 | |
| 1667 | hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj); |
| 1668 | if (!hugepages_kobj) |
| 1669 | return; |
| 1670 | |
| 1671 | for_each_hstate(h) { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1672 | err = hugetlb_sysfs_add_hstate(h, hugepages_kobj, |
| 1673 | hstate_kobjs, &hstate_attr_group); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1674 | if (err) |
| 1675 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s", |
| 1676 | h->name); |
| 1677 | } |
| 1678 | } |
| 1679 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1680 | #ifdef CONFIG_NUMA |
| 1681 | |
| 1682 | /* |
| 1683 | * node_hstate/s - associate per node hstate attributes, via their kobjects, |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1684 | * with node devices in node_devices[] using a parallel array. The array |
| 1685 | * index of a node device or _hstate == node id. |
| 1686 | * This is here to avoid any static dependency of the node device driver, in |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1687 | * the base kernel, on the hugetlb module. |
| 1688 | */ |
| 1689 | struct node_hstate { |
| 1690 | struct kobject *hugepages_kobj; |
| 1691 | struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; |
| 1692 | }; |
| 1693 | struct node_hstate node_hstates[MAX_NUMNODES]; |
| 1694 | |
| 1695 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1696 | * A subset of global hstate attributes for node devices |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1697 | */ |
| 1698 | static struct attribute *per_node_hstate_attrs[] = { |
| 1699 | &nr_hugepages_attr.attr, |
| 1700 | &free_hugepages_attr.attr, |
| 1701 | &surplus_hugepages_attr.attr, |
| 1702 | NULL, |
| 1703 | }; |
| 1704 | |
| 1705 | static struct attribute_group per_node_hstate_attr_group = { |
| 1706 | .attrs = per_node_hstate_attrs, |
| 1707 | }; |
| 1708 | |
| 1709 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1710 | * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj. |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1711 | * Returns node id via non-NULL nidp. |
| 1712 | */ |
| 1713 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) |
| 1714 | { |
| 1715 | int nid; |
| 1716 | |
| 1717 | for (nid = 0; nid < nr_node_ids; nid++) { |
| 1718 | struct node_hstate *nhs = &node_hstates[nid]; |
| 1719 | int i; |
| 1720 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
| 1721 | if (nhs->hstate_kobjs[i] == kobj) { |
| 1722 | if (nidp) |
| 1723 | *nidp = nid; |
| 1724 | return &hstates[i]; |
| 1725 | } |
| 1726 | } |
| 1727 | |
| 1728 | BUG(); |
| 1729 | return NULL; |
| 1730 | } |
| 1731 | |
| 1732 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1733 | * Unregister hstate attributes from a single node device. |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1734 | * No-op if no hstate attributes attached. |
| 1735 | */ |
| 1736 | void hugetlb_unregister_node(struct node *node) |
| 1737 | { |
| 1738 | struct hstate *h; |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1739 | struct node_hstate *nhs = &node_hstates[node->dev.id]; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1740 | |
| 1741 | if (!nhs->hugepages_kobj) |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1742 | return; /* no hstate attributes */ |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1743 | |
| 1744 | for_each_hstate(h) |
| 1745 | if (nhs->hstate_kobjs[h - hstates]) { |
| 1746 | kobject_put(nhs->hstate_kobjs[h - hstates]); |
| 1747 | nhs->hstate_kobjs[h - hstates] = NULL; |
| 1748 | } |
| 1749 | |
| 1750 | kobject_put(nhs->hugepages_kobj); |
| 1751 | nhs->hugepages_kobj = NULL; |
| 1752 | } |
| 1753 | |
| 1754 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1755 | * hugetlb module exit: unregister hstate attributes from node devices |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1756 | * that have them. |
| 1757 | */ |
| 1758 | static void hugetlb_unregister_all_nodes(void) |
| 1759 | { |
| 1760 | int nid; |
| 1761 | |
| 1762 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1763 | * disable node device registrations. |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1764 | */ |
| 1765 | register_hugetlbfs_with_node(NULL, NULL); |
| 1766 | |
| 1767 | /* |
| 1768 | * remove hstate attributes from any nodes that have them. |
| 1769 | */ |
| 1770 | for (nid = 0; nid < nr_node_ids; nid++) |
| 1771 | hugetlb_unregister_node(&node_devices[nid]); |
| 1772 | } |
| 1773 | |
| 1774 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1775 | * Register hstate attributes for a single node device. |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1776 | * No-op if attributes already registered. |
| 1777 | */ |
| 1778 | void hugetlb_register_node(struct node *node) |
| 1779 | { |
| 1780 | struct hstate *h; |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1781 | struct node_hstate *nhs = &node_hstates[node->dev.id]; |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1782 | int err; |
| 1783 | |
| 1784 | if (nhs->hugepages_kobj) |
| 1785 | return; /* already allocated */ |
| 1786 | |
| 1787 | nhs->hugepages_kobj = kobject_create_and_add("hugepages", |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1788 | &node->dev.kobj); |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1789 | if (!nhs->hugepages_kobj) |
| 1790 | return; |
| 1791 | |
| 1792 | for_each_hstate(h) { |
| 1793 | err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, |
| 1794 | nhs->hstate_kobjs, |
| 1795 | &per_node_hstate_attr_group); |
| 1796 | if (err) { |
| 1797 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s" |
| 1798 | " for node %d\n", |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1799 | h->name, node->dev.id); |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1800 | hugetlb_unregister_node(node); |
| 1801 | break; |
| 1802 | } |
| 1803 | } |
| 1804 | } |
| 1805 | |
| 1806 | /* |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1807 | * hugetlb init time: register hstate attributes for all registered node |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1808 | * devices of nodes that have memory. All on-line nodes should have |
| 1809 | * registered their associated device by this time. |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1810 | */ |
| 1811 | static void hugetlb_register_all_nodes(void) |
| 1812 | { |
| 1813 | int nid; |
| 1814 | |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1815 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1816 | struct node *node = &node_devices[nid]; |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1817 | if (node->dev.id == nid) |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1818 | hugetlb_register_node(node); |
| 1819 | } |
| 1820 | |
| 1821 | /* |
Kay Sievers | 10fbcf4 | 2011-12-21 14:48:43 -0800 | [diff] [blame] | 1822 | * Let the node device driver know we're here so it can |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1823 | * [un]register hstate attributes on node hotplug. |
| 1824 | */ |
| 1825 | register_hugetlbfs_with_node(hugetlb_register_node, |
| 1826 | hugetlb_unregister_node); |
| 1827 | } |
| 1828 | #else /* !CONFIG_NUMA */ |
| 1829 | |
| 1830 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) |
| 1831 | { |
| 1832 | BUG(); |
| 1833 | if (nidp) |
| 1834 | *nidp = -1; |
| 1835 | return NULL; |
| 1836 | } |
| 1837 | |
| 1838 | static void hugetlb_unregister_all_nodes(void) { } |
| 1839 | |
| 1840 | static void hugetlb_register_all_nodes(void) { } |
| 1841 | |
| 1842 | #endif |
| 1843 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1844 | static void __exit hugetlb_exit(void) |
| 1845 | { |
| 1846 | struct hstate *h; |
| 1847 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1848 | hugetlb_unregister_all_nodes(); |
| 1849 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1850 | for_each_hstate(h) { |
| 1851 | kobject_put(hstate_kobjs[h - hstates]); |
| 1852 | } |
| 1853 | |
| 1854 | kobject_put(hugepages_kobj); |
| 1855 | } |
| 1856 | module_exit(hugetlb_exit); |
| 1857 | |
| 1858 | static int __init hugetlb_init(void) |
| 1859 | { |
Benjamin Herrenschmidt | 0ef89d2 | 2008-07-31 00:07:30 -0700 | [diff] [blame] | 1860 | /* Some platform decide whether they support huge pages at boot |
| 1861 | * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when |
| 1862 | * there is no such support |
| 1863 | */ |
| 1864 | if (HPAGE_SHIFT == 0) |
| 1865 | return 0; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1866 | |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1867 | if (!size_to_hstate(default_hstate_size)) { |
| 1868 | default_hstate_size = HPAGE_SIZE; |
| 1869 | if (!size_to_hstate(default_hstate_size)) |
| 1870 | hugetlb_add_hstate(HUGETLB_PAGE_ORDER); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1871 | } |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1872 | default_hstate_idx = size_to_hstate(default_hstate_size) - hstates; |
| 1873 | if (default_hstate_max_huge_pages) |
| 1874 | default_hstate.max_huge_pages = default_hstate_max_huge_pages; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1875 | |
| 1876 | hugetlb_init_hstates(); |
| 1877 | |
Andi Kleen | aa888a7 | 2008-07-23 21:27:47 -0700 | [diff] [blame] | 1878 | gather_bootmem_prealloc(); |
| 1879 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1880 | report_hugepages(); |
| 1881 | |
| 1882 | hugetlb_sysfs_init(); |
| 1883 | |
Lee Schermerhorn | 9a30523 | 2009-12-14 17:58:25 -0800 | [diff] [blame] | 1884 | hugetlb_register_all_nodes(); |
| 1885 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1886 | return 0; |
| 1887 | } |
| 1888 | module_init(hugetlb_init); |
| 1889 | |
| 1890 | /* Should be called on processing a hugepagesz=... option */ |
| 1891 | void __init hugetlb_add_hstate(unsigned order) |
| 1892 | { |
| 1893 | struct hstate *h; |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1894 | unsigned long i; |
| 1895 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1896 | if (size_to_hstate(PAGE_SIZE << order)) { |
| 1897 | printk(KERN_WARNING "hugepagesz= specified twice, ignoring\n"); |
| 1898 | return; |
| 1899 | } |
| 1900 | BUG_ON(max_hstate >= HUGE_MAX_HSTATE); |
| 1901 | BUG_ON(order == 0); |
| 1902 | h = &hstates[max_hstate++]; |
| 1903 | h->order = order; |
| 1904 | h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1905 | h->nr_huge_pages = 0; |
| 1906 | h->free_huge_pages = 0; |
| 1907 | for (i = 0; i < MAX_NUMNODES; ++i) |
| 1908 | INIT_LIST_HEAD(&h->hugepage_freelists[i]); |
Lee Schermerhorn | 9b5e5d0 | 2009-12-14 17:58:32 -0800 | [diff] [blame] | 1909 | h->next_nid_to_alloc = first_node(node_states[N_HIGH_MEMORY]); |
| 1910 | h->next_nid_to_free = first_node(node_states[N_HIGH_MEMORY]); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1911 | snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", |
| 1912 | huge_page_size(h)/1024); |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1913 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1914 | parsed_hstate = h; |
| 1915 | } |
| 1916 | |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1917 | static int __init hugetlb_nrpages_setup(char *s) |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1918 | { |
| 1919 | unsigned long *mhp; |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1920 | static unsigned long *last_mhp; |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1921 | |
| 1922 | /* |
| 1923 | * !max_hstate means we haven't parsed a hugepagesz= parameter yet, |
| 1924 | * so this hugepages= parameter goes to the "default hstate". |
| 1925 | */ |
| 1926 | if (!max_hstate) |
| 1927 | mhp = &default_hstate_max_huge_pages; |
| 1928 | else |
| 1929 | mhp = &parsed_hstate->max_huge_pages; |
| 1930 | |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1931 | if (mhp == last_mhp) { |
| 1932 | printk(KERN_WARNING "hugepages= specified twice without " |
| 1933 | "interleaving hugepagesz=, ignoring\n"); |
| 1934 | return 1; |
| 1935 | } |
| 1936 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1937 | if (sscanf(s, "%lu", mhp) <= 0) |
| 1938 | *mhp = 0; |
| 1939 | |
Andi Kleen | 8faa8b0 | 2008-07-23 21:27:48 -0700 | [diff] [blame] | 1940 | /* |
| 1941 | * Global state is always initialized later in hugetlb_init. |
| 1942 | * But we need to allocate >= MAX_ORDER hstates here early to still |
| 1943 | * use the bootmem allocator. |
| 1944 | */ |
| 1945 | if (max_hstate && parsed_hstate->order >= MAX_ORDER) |
| 1946 | hugetlb_hstate_alloc_pages(parsed_hstate); |
| 1947 | |
| 1948 | last_mhp = mhp; |
| 1949 | |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1950 | return 1; |
| 1951 | } |
Nick Piggin | e11bfbf | 2008-07-23 21:27:52 -0700 | [diff] [blame] | 1952 | __setup("hugepages=", hugetlb_nrpages_setup); |
| 1953 | |
| 1954 | static int __init hugetlb_default_setup(char *s) |
| 1955 | { |
| 1956 | default_hstate_size = memparse(s, &s); |
| 1957 | return 1; |
| 1958 | } |
| 1959 | __setup("default_hugepagesz=", hugetlb_default_setup); |
Nishanth Aravamudan | a343787 | 2008-07-23 21:27:44 -0700 | [diff] [blame] | 1960 | |
Nishanth Aravamudan | 8a21346 | 2008-07-25 19:44:37 -0700 | [diff] [blame] | 1961 | static unsigned int cpuset_mems_nr(unsigned int *array) |
| 1962 | { |
| 1963 | int node; |
| 1964 | unsigned int nr = 0; |
| 1965 | |
| 1966 | for_each_node_mask(node, cpuset_current_mems_allowed) |
| 1967 | nr += array[node]; |
| 1968 | |
| 1969 | return nr; |
| 1970 | } |
| 1971 | |
| 1972 | #ifdef CONFIG_SYSCTL |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1973 | static int hugetlb_sysctl_handler_common(bool obey_mempolicy, |
| 1974 | struct ctl_table *table, int write, |
| 1975 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1976 | { |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1977 | struct hstate *h = &default_hstate; |
| 1978 | unsigned long tmp; |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1979 | int ret; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1980 | |
Petr Holasek | c033a93 | 2011-03-22 16:33:05 -0700 | [diff] [blame] | 1981 | tmp = h->max_huge_pages; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1982 | |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 1983 | if (write && h->order >= MAX_ORDER) |
| 1984 | return -EINVAL; |
| 1985 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1986 | table->data = &tmp; |
| 1987 | table->maxlen = sizeof(unsigned long); |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1988 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
| 1989 | if (ret) |
| 1990 | goto out; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 1991 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1992 | if (write) { |
David Rientjes | bad44b5 | 2009-12-14 17:58:38 -0800 | [diff] [blame] | 1993 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, |
| 1994 | GFP_KERNEL | __GFP_NORETRY); |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1995 | if (!(obey_mempolicy && |
| 1996 | init_nodemask_of_mempolicy(nodes_allowed))) { |
| 1997 | NODEMASK_FREE(nodes_allowed); |
| 1998 | nodes_allowed = &node_states[N_HIGH_MEMORY]; |
| 1999 | } |
| 2000 | h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed); |
| 2001 | |
| 2002 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) |
| 2003 | NODEMASK_FREE(nodes_allowed); |
| 2004 | } |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 2005 | out: |
| 2006 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2007 | } |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 2008 | |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 2009 | int hugetlb_sysctl_handler(struct ctl_table *table, int write, |
| 2010 | void __user *buffer, size_t *length, loff_t *ppos) |
| 2011 | { |
| 2012 | |
| 2013 | return hugetlb_sysctl_handler_common(false, table, write, |
| 2014 | buffer, length, ppos); |
| 2015 | } |
| 2016 | |
| 2017 | #ifdef CONFIG_NUMA |
| 2018 | int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write, |
| 2019 | void __user *buffer, size_t *length, loff_t *ppos) |
| 2020 | { |
| 2021 | return hugetlb_sysctl_handler_common(true, table, write, |
| 2022 | buffer, length, ppos); |
| 2023 | } |
| 2024 | #endif /* CONFIG_NUMA */ |
| 2025 | |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 2026 | int hugetlb_treat_movable_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2027 | void __user *buffer, |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 2028 | size_t *length, loff_t *ppos) |
| 2029 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2030 | proc_dointvec(table, write, buffer, length, ppos); |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 2031 | if (hugepages_treat_as_movable) |
| 2032 | htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; |
| 2033 | else |
| 2034 | htlb_alloc_mask = GFP_HIGHUSER; |
| 2035 | return 0; |
| 2036 | } |
| 2037 | |
Nishanth Aravamudan | a3d0c6a | 2008-02-08 04:18:18 -0800 | [diff] [blame] | 2038 | int hugetlb_overcommit_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2039 | void __user *buffer, |
Nishanth Aravamudan | a3d0c6a | 2008-02-08 04:18:18 -0800 | [diff] [blame] | 2040 | size_t *length, loff_t *ppos) |
| 2041 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2042 | struct hstate *h = &default_hstate; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 2043 | unsigned long tmp; |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 2044 | int ret; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 2045 | |
Petr Holasek | c033a93 | 2011-03-22 16:33:05 -0700 | [diff] [blame] | 2046 | tmp = h->nr_overcommit_huge_pages; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 2047 | |
Eric B Munson | adbe872 | 2011-01-13 15:47:27 -0800 | [diff] [blame] | 2048 | if (write && h->order >= MAX_ORDER) |
| 2049 | return -EINVAL; |
| 2050 | |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 2051 | table->data = &tmp; |
| 2052 | table->maxlen = sizeof(unsigned long); |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 2053 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
| 2054 | if (ret) |
| 2055 | goto out; |
Andi Kleen | e5ff215 | 2008-07-23 21:27:42 -0700 | [diff] [blame] | 2056 | |
| 2057 | if (write) { |
| 2058 | spin_lock(&hugetlb_lock); |
| 2059 | h->nr_overcommit_huge_pages = tmp; |
| 2060 | spin_unlock(&hugetlb_lock); |
| 2061 | } |
Michal Hocko | 08d4a24 | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 2062 | out: |
| 2063 | return ret; |
Nishanth Aravamudan | a3d0c6a | 2008-02-08 04:18:18 -0800 | [diff] [blame] | 2064 | } |
| 2065 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2066 | #endif /* CONFIG_SYSCTL */ |
| 2067 | |
Alexey Dobriyan | e1759c2 | 2008-10-15 23:50:22 +0400 | [diff] [blame] | 2068 | void hugetlb_report_meminfo(struct seq_file *m) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2069 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2070 | struct hstate *h = &default_hstate; |
Alexey Dobriyan | e1759c2 | 2008-10-15 23:50:22 +0400 | [diff] [blame] | 2071 | seq_printf(m, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2072 | "HugePages_Total: %5lu\n" |
| 2073 | "HugePages_Free: %5lu\n" |
| 2074 | "HugePages_Rsvd: %5lu\n" |
| 2075 | "HugePages_Surp: %5lu\n" |
| 2076 | "Hugepagesize: %8lu kB\n", |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2077 | h->nr_huge_pages, |
| 2078 | h->free_huge_pages, |
| 2079 | h->resv_huge_pages, |
| 2080 | h->surplus_huge_pages, |
| 2081 | 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2082 | } |
| 2083 | |
| 2084 | int hugetlb_report_node_meminfo(int nid, char *buf) |
| 2085 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2086 | struct hstate *h = &default_hstate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2087 | return sprintf(buf, |
| 2088 | "Node %d HugePages_Total: %5u\n" |
Nishanth Aravamudan | a1de091 | 2008-03-26 14:37:53 -0700 | [diff] [blame] | 2089 | "Node %d HugePages_Free: %5u\n" |
| 2090 | "Node %d HugePages_Surp: %5u\n", |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2091 | nid, h->nr_huge_pages_node[nid], |
| 2092 | nid, h->free_huge_pages_node[nid], |
| 2093 | nid, h->surplus_huge_pages_node[nid]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | } |
| 2095 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2096 | /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ |
| 2097 | unsigned long hugetlb_total_pages(void) |
| 2098 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2099 | struct hstate *h = &default_hstate; |
| 2100 | return h->nr_huge_pages * pages_per_huge_page(h); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2101 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2103 | static int hugetlb_acct_memory(struct hstate *h, long delta) |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2104 | { |
| 2105 | int ret = -ENOMEM; |
| 2106 | |
| 2107 | spin_lock(&hugetlb_lock); |
| 2108 | /* |
| 2109 | * When cpuset is configured, it breaks the strict hugetlb page |
| 2110 | * reservation as the accounting is done on a global variable. Such |
| 2111 | * reservation is completely rubbish in the presence of cpuset because |
| 2112 | * the reservation is not checked against page availability for the |
| 2113 | * current cpuset. Application can still potentially OOM'ed by kernel |
| 2114 | * with lack of free htlb page in cpuset that the task is in. |
| 2115 | * Attempt to enforce strict accounting with cpuset is almost |
| 2116 | * impossible (or too ugly) because cpuset is too fluid that |
| 2117 | * task or memory node can be dynamically moved between cpusets. |
| 2118 | * |
| 2119 | * The change of semantics for shared hugetlb mapping with cpuset is |
| 2120 | * undesirable. However, in order to preserve some of the semantics, |
| 2121 | * we fall back to check against current free page availability as |
| 2122 | * a best attempt and hopefully to minimize the impact of changing |
| 2123 | * semantics that cpuset has. |
| 2124 | */ |
| 2125 | if (delta > 0) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2126 | if (gather_surplus_pages(h, delta) < 0) |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2127 | goto out; |
| 2128 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2129 | if (delta > cpuset_mems_nr(h->free_huge_pages_node)) { |
| 2130 | return_unused_surplus_pages(h, delta); |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2131 | goto out; |
| 2132 | } |
| 2133 | } |
| 2134 | |
| 2135 | ret = 0; |
| 2136 | if (delta < 0) |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2137 | return_unused_surplus_pages(h, (unsigned long) -delta); |
Mel Gorman | fc1b8a7 | 2008-07-23 21:27:22 -0700 | [diff] [blame] | 2138 | |
| 2139 | out: |
| 2140 | spin_unlock(&hugetlb_lock); |
| 2141 | return ret; |
| 2142 | } |
| 2143 | |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2144 | static void hugetlb_vm_op_open(struct vm_area_struct *vma) |
| 2145 | { |
| 2146 | struct resv_map *reservations = vma_resv_map(vma); |
| 2147 | |
| 2148 | /* |
| 2149 | * This new VMA should share its siblings reservation map if present. |
| 2150 | * The VMA will only ever have a valid reservation map pointer where |
| 2151 | * it is being copied for another still existing VMA. As that VMA |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2152 | * has a reference to the reservation map it cannot disappear until |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2153 | * after this open call completes. It is therefore safe to take a |
| 2154 | * new reference here without additional locking. |
| 2155 | */ |
| 2156 | if (reservations) |
| 2157 | kref_get(&reservations->refs); |
| 2158 | } |
| 2159 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2160 | static void hugetlb_vm_op_close(struct vm_area_struct *vma) |
| 2161 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2162 | struct hstate *h = hstate_vma(vma); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2163 | struct resv_map *reservations = vma_resv_map(vma); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 2164 | struct hugepage_subpool *spool = subpool_vma(vma); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2165 | unsigned long reserve; |
| 2166 | unsigned long start; |
| 2167 | unsigned long end; |
| 2168 | |
| 2169 | if (reservations) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2170 | start = vma_hugecache_offset(h, vma, vma->vm_start); |
| 2171 | end = vma_hugecache_offset(h, vma, vma->vm_end); |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2172 | |
| 2173 | reserve = (end - start) - |
| 2174 | region_count(&reservations->regions, start, end); |
| 2175 | |
| 2176 | kref_put(&reservations->refs, resv_map_release); |
| 2177 | |
Adam Litke | 7251ff78 | 2008-07-23 21:27:59 -0700 | [diff] [blame] | 2178 | if (reserve) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2179 | hugetlb_acct_memory(h, -reserve); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 2180 | hugepage_subpool_put_pages(spool, reserve); |
Adam Litke | 7251ff78 | 2008-07-23 21:27:59 -0700 | [diff] [blame] | 2181 | } |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2182 | } |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2183 | } |
| 2184 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2185 | /* |
| 2186 | * We cannot handle pagefaults against hugetlb pages at all. They cause |
| 2187 | * handle_mm_fault() to try to instantiate regular-sized pages in the |
| 2188 | * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get |
| 2189 | * this far. |
| 2190 | */ |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2191 | 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] | 2192 | { |
| 2193 | BUG(); |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2194 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2195 | } |
| 2196 | |
Alexey Dobriyan | f0f37e2 | 2009-09-27 22:29:37 +0400 | [diff] [blame] | 2197 | const struct vm_operations_struct hugetlb_vm_ops = { |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 2198 | .fault = hugetlb_vm_op_fault, |
Andy Whitcroft | 84afd99 | 2008-07-23 21:27:32 -0700 | [diff] [blame] | 2199 | .open = hugetlb_vm_op_open, |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2200 | .close = hugetlb_vm_op_close, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2201 | }; |
| 2202 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2203 | static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, |
| 2204 | int writable) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2205 | { |
| 2206 | pte_t entry; |
| 2207 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2208 | if (writable) { |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2209 | entry = |
| 2210 | pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))); |
| 2211 | } else { |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2212 | entry = huge_pte_wrprotect(mk_pte(page, vma->vm_page_prot)); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2213 | } |
| 2214 | entry = pte_mkyoung(entry); |
| 2215 | entry = pte_mkhuge(entry); |
| 2216 | |
| 2217 | return entry; |
| 2218 | } |
| 2219 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2220 | static void set_huge_ptep_writable(struct vm_area_struct *vma, |
| 2221 | unsigned long address, pte_t *ptep) |
| 2222 | { |
| 2223 | pte_t entry; |
| 2224 | |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2225 | entry = pte_mkwrite(pte_mkdirty(huge_ptep_get(ptep))); |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2226 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1)) |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 2227 | update_mmu_cache(vma, address, ptep); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2228 | } |
| 2229 | |
| 2230 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2231 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, |
| 2232 | struct vm_area_struct *vma) |
| 2233 | { |
| 2234 | pte_t *src_pte, *dst_pte, entry; |
| 2235 | struct page *ptepage; |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2236 | unsigned long addr; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2237 | int cow; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2238 | struct hstate *h = hstate_vma(vma); |
| 2239 | unsigned long sz = huge_page_size(h); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2240 | |
| 2241 | cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2242 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2243 | for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2244 | src_pte = huge_pte_offset(src, addr); |
| 2245 | if (!src_pte) |
| 2246 | continue; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2247 | dst_pte = huge_pte_alloc(dst, addr, sz); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2248 | if (!dst_pte) |
| 2249 | goto nomem; |
Larry Woodman | c5c9942 | 2008-01-24 05:49:25 -0800 | [diff] [blame] | 2250 | |
| 2251 | /* If the pagetables are shared don't copy or take references */ |
| 2252 | if (dst_pte == src_pte) |
| 2253 | continue; |
| 2254 | |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2255 | spin_lock(&dst->page_table_lock); |
Nick Piggin | 4647875 | 2008-06-05 22:45:57 -0700 | [diff] [blame] | 2256 | spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2257 | if (!huge_pte_none(huge_ptep_get(src_pte))) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2258 | if (cow) |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2259 | huge_ptep_set_wrprotect(src, addr, src_pte); |
| 2260 | entry = huge_ptep_get(src_pte); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2261 | ptepage = pte_page(entry); |
| 2262 | get_page(ptepage); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2263 | page_dup_rmap(ptepage); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2264 | set_huge_pte_at(dst, addr, dst_pte, entry); |
| 2265 | } |
| 2266 | spin_unlock(&src->page_table_lock); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 2267 | spin_unlock(&dst->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2268 | } |
| 2269 | return 0; |
| 2270 | |
| 2271 | nomem: |
| 2272 | return -ENOMEM; |
| 2273 | } |
| 2274 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 2275 | static int is_hugetlb_entry_migration(pte_t pte) |
| 2276 | { |
| 2277 | swp_entry_t swp; |
| 2278 | |
| 2279 | if (huge_pte_none(pte) || pte_present(pte)) |
| 2280 | return 0; |
| 2281 | swp = pte_to_swp_entry(pte); |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2282 | if (non_swap_entry(swp) && is_migration_entry(swp)) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 2283 | return 1; |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2284 | else |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 2285 | return 0; |
| 2286 | } |
| 2287 | |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2288 | static int is_hugetlb_entry_hwpoisoned(pte_t pte) |
| 2289 | { |
| 2290 | swp_entry_t swp; |
| 2291 | |
| 2292 | if (huge_pte_none(pte) || pte_present(pte)) |
| 2293 | return 0; |
| 2294 | swp = pte_to_swp_entry(pte); |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2295 | if (non_swap_entry(swp) && is_hwpoison_entry(swp)) |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2296 | return 1; |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2297 | else |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2298 | return 0; |
| 2299 | } |
| 2300 | |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2301 | void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2302 | unsigned long end, struct page *ref_page) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2303 | { |
| 2304 | struct mm_struct *mm = vma->vm_mm; |
| 2305 | unsigned long address; |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2306 | pte_t *ptep; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2307 | pte_t pte; |
| 2308 | struct page *page; |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2309 | struct page *tmp; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2310 | struct hstate *h = hstate_vma(vma); |
| 2311 | unsigned long sz = huge_page_size(h); |
| 2312 | |
Chen, Kenneth W | c0a499c | 2006-12-06 20:31:39 -0800 | [diff] [blame] | 2313 | /* |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2314 | * A page gathering list, protected by per file i_mmap_mutex. The |
Chen, Kenneth W | c0a499c | 2006-12-06 20:31:39 -0800 | [diff] [blame] | 2315 | * lock is used to avoid list corruption from multiple unmapping |
| 2316 | * of the same page since we are using page->lru. |
| 2317 | */ |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2318 | LIST_HEAD(page_list); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2319 | |
| 2320 | WARN_ON(!is_vm_hugetlb_page(vma)); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2321 | BUG_ON(start & ~huge_page_mask(h)); |
| 2322 | BUG_ON(end & ~huge_page_mask(h)); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2323 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 2324 | mmu_notifier_invalidate_range_start(mm, start, end); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 2325 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2326 | for (address = start; address < end; address += sz) { |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2327 | ptep = huge_pte_offset(mm, address); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2328 | if (!ptep) |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2329 | continue; |
| 2330 | |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 2331 | if (huge_pmd_unshare(mm, &address, ptep)) |
| 2332 | continue; |
| 2333 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2334 | /* |
| 2335 | * If a reference page is supplied, it is because a specific |
| 2336 | * page is being unmapped, not a range. Ensure the page we |
| 2337 | * are about to unmap is the actual page of interest. |
| 2338 | */ |
| 2339 | if (ref_page) { |
| 2340 | pte = huge_ptep_get(ptep); |
| 2341 | if (huge_pte_none(pte)) |
| 2342 | continue; |
| 2343 | page = pte_page(pte); |
| 2344 | if (page != ref_page) |
| 2345 | continue; |
| 2346 | |
| 2347 | /* |
| 2348 | * Mark the VMA as having unmapped its page so that |
| 2349 | * future faults in this VMA will fail rather than |
| 2350 | * looking like data was lost |
| 2351 | */ |
| 2352 | set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED); |
| 2353 | } |
| 2354 | |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2355 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2356 | if (huge_pte_none(pte)) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2357 | continue; |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 2358 | |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2359 | /* |
| 2360 | * HWPoisoned hugepage is already unmapped and dropped reference |
| 2361 | */ |
| 2362 | if (unlikely(is_hugetlb_entry_hwpoisoned(pte))) |
| 2363 | continue; |
| 2364 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2365 | page = pte_page(pte); |
Ken Chen | 6649a38 | 2007-02-08 14:20:27 -0800 | [diff] [blame] | 2366 | if (pte_dirty(pte)) |
| 2367 | set_page_dirty(page); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2368 | list_add(&page->lru, &page_list); |
Hillf Danton | 9e81130 | 2012-03-21 16:34:03 -0700 | [diff] [blame] | 2369 | |
| 2370 | /* Bail out after unmapping reference page if supplied */ |
| 2371 | if (ref_page) |
| 2372 | break; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2373 | } |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 2374 | flush_tlb_range(vma, start, end); |
Al Viro | cd2934a | 2012-03-05 06:40:29 +0000 | [diff] [blame] | 2375 | spin_unlock(&mm->page_table_lock); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 2376 | mmu_notifier_invalidate_range_end(mm, start, end); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2377 | list_for_each_entry_safe(page, tmp, &page_list, lru) { |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2378 | page_remove_rmap(page); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 2379 | list_del(&page->lru); |
| 2380 | put_page(page); |
| 2381 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2382 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2383 | |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2384 | void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2385 | unsigned long end, struct page *ref_page) |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2386 | { |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2387 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 2388 | __unmap_hugepage_range(vma, start, end, ref_page); |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2389 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 2390 | } |
| 2391 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2392 | /* |
| 2393 | * This is called when the original mapper is failing to COW a MAP_PRIVATE |
| 2394 | * mappping it owns the reserve page for. The intention is to unmap the page |
| 2395 | * from other VMAs and let the children be SIGKILLed if they are faulting the |
| 2396 | * same region. |
| 2397 | */ |
Harvey Harrison | 2a4b3de | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 2398 | static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2399 | struct page *page, unsigned long address) |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2400 | { |
Adam Litke | 7526674 | 2008-11-12 13:24:56 -0800 | [diff] [blame] | 2401 | struct hstate *h = hstate_vma(vma); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2402 | struct vm_area_struct *iter_vma; |
| 2403 | struct address_space *mapping; |
| 2404 | struct prio_tree_iter iter; |
| 2405 | pgoff_t pgoff; |
| 2406 | |
| 2407 | /* |
| 2408 | * vm_pgoff is in PAGE_SIZE units, hence the different calculation |
| 2409 | * from page cache lookup which is in HPAGE_SIZE units. |
| 2410 | */ |
Adam Litke | 7526674 | 2008-11-12 13:24:56 -0800 | [diff] [blame] | 2411 | address = address & huge_page_mask(h); |
Hillf Danton | 0c176d5 | 2012-01-10 15:08:19 -0800 | [diff] [blame] | 2412 | pgoff = vma_hugecache_offset(h, vma, address); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 2413 | mapping = vma->vm_file->f_dentry->d_inode->i_mapping; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2414 | |
Mel Gorman | 4eb2b1d | 2009-12-14 17:59:53 -0800 | [diff] [blame] | 2415 | /* |
| 2416 | * Take the mapping lock for the duration of the table walk. As |
| 2417 | * this mapping should be shared between all the VMAs, |
| 2418 | * __unmap_hugepage_range() is called as the lock is already held |
| 2419 | */ |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2420 | mutex_lock(&mapping->i_mmap_mutex); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2421 | vma_prio_tree_foreach(iter_vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
| 2422 | /* Do not unmap the current VMA */ |
| 2423 | if (iter_vma == vma) |
| 2424 | continue; |
| 2425 | |
| 2426 | /* |
| 2427 | * Unmap the page from other VMAs without their own reserves. |
| 2428 | * They get marked to be SIGKILLed if they fault in these |
| 2429 | * areas. This is because a future no-page fault on this VMA |
| 2430 | * could insert a zeroed page instead of the data existing |
| 2431 | * from the time of fork. This would look like data corruption |
| 2432 | */ |
| 2433 | if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER)) |
Mel Gorman | 4eb2b1d | 2009-12-14 17:59:53 -0800 | [diff] [blame] | 2434 | __unmap_hugepage_range(iter_vma, |
Adam Litke | 7526674 | 2008-11-12 13:24:56 -0800 | [diff] [blame] | 2435 | address, address + huge_page_size(h), |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2436 | page); |
| 2437 | } |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2438 | mutex_unlock(&mapping->i_mmap_mutex); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2439 | |
| 2440 | return 1; |
| 2441 | } |
| 2442 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2443 | /* |
| 2444 | * Hugetlb_cow() should be called with page lock of the original hugepage held. |
Michal Hocko | ef009b2 | 2012-01-10 15:07:21 -0800 | [diff] [blame] | 2445 | * Called with hugetlb_instantiation_mutex held and pte_page locked so we |
| 2446 | * cannot race with other handlers or page migration. |
| 2447 | * Keep the pte_same checks anyway to make transition from the mutex easier. |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2448 | */ |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2449 | 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] | 2450 | unsigned long address, pte_t *ptep, pte_t pte, |
| 2451 | struct page *pagecache_page) |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2452 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2453 | struct hstate *h = hstate_vma(vma); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2454 | struct page *old_page, *new_page; |
David Gibson | 79ac6ba | 2006-03-22 00:08:51 -0800 | [diff] [blame] | 2455 | int avoidcopy; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2456 | int outside_reserve = 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2457 | |
| 2458 | old_page = pte_page(pte); |
| 2459 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2460 | retry_avoidcopy: |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2461 | /* If no-one else is actually using this page, avoid the copy |
| 2462 | * and just make the page writable */ |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2463 | avoidcopy = (page_mapcount(old_page) == 1); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2464 | if (avoidcopy) { |
Naoya Horiguchi | 56c9cfb | 2010-09-10 13:23:04 +0900 | [diff] [blame] | 2465 | if (PageAnon(old_page)) |
| 2466 | page_move_anon_rmap(old_page, vma, address); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2467 | set_huge_ptep_writable(vma, address, ptep); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2468 | return 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2469 | } |
| 2470 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2471 | /* |
| 2472 | * If the process that created a MAP_PRIVATE mapping is about to |
| 2473 | * perform a COW due to a shared page count, attempt to satisfy |
| 2474 | * the allocation without using the existing reserves. The pagecache |
| 2475 | * page is used to determine if the reserve at this address was |
| 2476 | * consumed or not. If reserves were used, a partial faulted mapping |
| 2477 | * at the time of fork() could consume its reserves on COW instead |
| 2478 | * of the full address range. |
| 2479 | */ |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2480 | if (!(vma->vm_flags & VM_MAYSHARE) && |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2481 | is_vma_resv_set(vma, HPAGE_RESV_OWNER) && |
| 2482 | old_page != pagecache_page) |
| 2483 | outside_reserve = 1; |
| 2484 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2485 | page_cache_get(old_page); |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2486 | |
| 2487 | /* Drop page_table_lock as buddy allocator may be called */ |
| 2488 | spin_unlock(&mm->page_table_lock); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2489 | new_page = alloc_huge_page(vma, address, outside_reserve); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2490 | |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 2491 | if (IS_ERR(new_page)) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2492 | page_cache_release(old_page); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2493 | |
| 2494 | /* |
| 2495 | * If a process owning a MAP_PRIVATE mapping fails to COW, |
| 2496 | * it is due to references held by a child and an insufficient |
| 2497 | * huge page pool. To guarantee the original mappers |
| 2498 | * reliability, unmap the page from child processes. The child |
| 2499 | * may get SIGKILLed if it later faults. |
| 2500 | */ |
| 2501 | if (outside_reserve) { |
| 2502 | BUG_ON(huge_pte_none(pte)); |
| 2503 | if (unmap_ref_private(mm, vma, old_page, address)) { |
| 2504 | BUG_ON(page_count(old_page) != 1); |
| 2505 | BUG_ON(huge_pte_none(pte)); |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2506 | spin_lock(&mm->page_table_lock); |
Hillf Danton | a734bcc | 2012-01-10 15:07:20 -0800 | [diff] [blame] | 2507 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
| 2508 | if (likely(pte_same(huge_ptep_get(ptep), pte))) |
| 2509 | goto retry_avoidcopy; |
| 2510 | /* |
| 2511 | * race occurs while re-acquiring page_table_lock, and |
| 2512 | * our job is done. |
| 2513 | */ |
| 2514 | return 0; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2515 | } |
| 2516 | WARN_ON_ONCE(1); |
| 2517 | } |
| 2518 | |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2519 | /* Caller expects lock to be held */ |
| 2520 | spin_lock(&mm->page_table_lock); |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 2521 | return -PTR_ERR(new_page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2522 | } |
| 2523 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2524 | /* |
| 2525 | * When the original hugepage is shared one, it does not have |
| 2526 | * anon_vma prepared. |
| 2527 | */ |
Dean Nelson | 44e2aa9 | 2010-10-26 14:22:08 -0700 | [diff] [blame] | 2528 | if (unlikely(anon_vma_prepare(vma))) { |
Hillf Danton | ea4039a | 2011-11-15 14:36:12 -0800 | [diff] [blame] | 2529 | page_cache_release(new_page); |
| 2530 | page_cache_release(old_page); |
Dean Nelson | 44e2aa9 | 2010-10-26 14:22:08 -0700 | [diff] [blame] | 2531 | /* Caller expects lock to be held */ |
| 2532 | spin_lock(&mm->page_table_lock); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2533 | return VM_FAULT_OOM; |
Dean Nelson | 44e2aa9 | 2010-10-26 14:22:08 -0700 | [diff] [blame] | 2534 | } |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2535 | |
Andrea Arcangeli | 47ad847 | 2011-01-13 15:46:47 -0800 | [diff] [blame] | 2536 | copy_user_huge_page(new_page, old_page, address, vma, |
| 2537 | pages_per_huge_page(h)); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2538 | __SetPageUptodate(new_page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2539 | |
Larry Woodman | b76c8cf | 2009-12-14 17:59:37 -0800 | [diff] [blame] | 2540 | /* |
| 2541 | * Retake the page_table_lock to check for racing updates |
| 2542 | * before the page tables are altered |
| 2543 | */ |
| 2544 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2545 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2546 | if (likely(pte_same(huge_ptep_get(ptep), pte))) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2547 | /* Break COW */ |
Doug Doan | 3edd4fc | 2010-08-09 17:18:30 -0700 | [diff] [blame] | 2548 | mmu_notifier_invalidate_range_start(mm, |
| 2549 | address & huge_page_mask(h), |
| 2550 | (address & huge_page_mask(h)) + huge_page_size(h)); |
Gerald Schaefer | 8fe627e | 2008-04-28 02:13:28 -0700 | [diff] [blame] | 2551 | huge_ptep_clear_flush(vma, address, ptep); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2552 | set_huge_pte_at(mm, address, ptep, |
| 2553 | make_huge_pte(vma, new_page, 1)); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2554 | page_remove_rmap(old_page); |
Naoya Horiguchi | cd67f0d | 2010-09-10 13:23:04 +0900 | [diff] [blame] | 2555 | hugepage_add_new_anon_rmap(new_page, vma, address); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2556 | /* Make the old page be freed below */ |
| 2557 | new_page = old_page; |
Doug Doan | 3edd4fc | 2010-08-09 17:18:30 -0700 | [diff] [blame] | 2558 | mmu_notifier_invalidate_range_end(mm, |
| 2559 | address & huge_page_mask(h), |
| 2560 | (address & huge_page_mask(h)) + huge_page_size(h)); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2561 | } |
| 2562 | page_cache_release(new_page); |
| 2563 | page_cache_release(old_page); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2564 | return 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2565 | } |
| 2566 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2567 | /* Return the pagecache page at a given address within a VMA */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2568 | static struct page *hugetlbfs_pagecache_page(struct hstate *h, |
| 2569 | struct vm_area_struct *vma, unsigned long address) |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2570 | { |
| 2571 | struct address_space *mapping; |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 2572 | pgoff_t idx; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2573 | |
| 2574 | mapping = vma->vm_file->f_mapping; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2575 | idx = vma_hugecache_offset(h, vma, address); |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2576 | |
| 2577 | return find_lock_page(mapping, idx); |
| 2578 | } |
| 2579 | |
Hugh Dickins | 3ae77f4 | 2009-09-21 17:03:33 -0700 | [diff] [blame] | 2580 | /* |
| 2581 | * Return whether there is a pagecache page to back given address within VMA. |
| 2582 | * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page. |
| 2583 | */ |
| 2584 | static bool hugetlbfs_pagecache_present(struct hstate *h, |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2585 | struct vm_area_struct *vma, unsigned long address) |
| 2586 | { |
| 2587 | struct address_space *mapping; |
| 2588 | pgoff_t idx; |
| 2589 | struct page *page; |
| 2590 | |
| 2591 | mapping = vma->vm_file->f_mapping; |
| 2592 | idx = vma_hugecache_offset(h, vma, address); |
| 2593 | |
| 2594 | page = find_get_page(mapping, idx); |
| 2595 | if (page) |
| 2596 | put_page(page); |
| 2597 | return page != NULL; |
| 2598 | } |
| 2599 | |
Robert P. J. Day | a1ed3dd | 2007-07-17 04:03:33 -0700 | [diff] [blame] | 2600 | 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] | 2601 | unsigned long address, pte_t *ptep, unsigned int flags) |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2602 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2603 | struct hstate *h = hstate_vma(vma); |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2604 | int ret = VM_FAULT_SIGBUS; |
Hillf Danton | 409eb8c | 2012-01-20 14:34:13 -0800 | [diff] [blame] | 2605 | int anon_rmap = 0; |
Andy Whitcroft | e7c4b0b | 2008-07-23 21:27:26 -0700 | [diff] [blame] | 2606 | pgoff_t idx; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2607 | unsigned long size; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2608 | struct page *page; |
| 2609 | struct address_space *mapping; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2610 | pte_t new_pte; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2611 | |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2612 | /* |
| 2613 | * Currently, we are forced to kill the process in the event the |
| 2614 | * original mapper has unmapped pages from the child due to a failed |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2615 | * COW. Warn that such a situation has occurred as it may not be obvious |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2616 | */ |
| 2617 | if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) { |
| 2618 | printk(KERN_WARNING |
| 2619 | "PID %d killed due to inadequate hugepage pool\n", |
| 2620 | current->pid); |
| 2621 | return ret; |
| 2622 | } |
| 2623 | |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2624 | mapping = vma->vm_file->f_mapping; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2625 | idx = vma_hugecache_offset(h, vma, address); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2626 | |
| 2627 | /* |
| 2628 | * Use page lock to guard against racing truncation |
| 2629 | * before we get page_table_lock. |
| 2630 | */ |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2631 | retry: |
| 2632 | page = find_lock_page(mapping, idx); |
| 2633 | if (!page) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2634 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
Hugh Dickins | ebed4bf | 2006-10-28 10:38:43 -0700 | [diff] [blame] | 2635 | if (idx >= size) |
| 2636 | goto out; |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2637 | page = alloc_huge_page(vma, address, 0); |
Adam Litke | 2fc39ce | 2007-11-14 16:59:39 -0800 | [diff] [blame] | 2638 | if (IS_ERR(page)) { |
| 2639 | ret = -PTR_ERR(page); |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2640 | goto out; |
| 2641 | } |
Andrea Arcangeli | 47ad847 | 2011-01-13 15:46:47 -0800 | [diff] [blame] | 2642 | clear_huge_page(page, address, pages_per_huge_page(h)); |
Nick Piggin | 0ed361d | 2008-02-04 22:29:34 -0800 | [diff] [blame] | 2643 | __SetPageUptodate(page); |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2644 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2645 | if (vma->vm_flags & VM_MAYSHARE) { |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2646 | int err; |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2647 | struct inode *inode = mapping->host; |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2648 | |
| 2649 | err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); |
| 2650 | if (err) { |
| 2651 | put_page(page); |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2652 | if (err == -EEXIST) |
| 2653 | goto retry; |
| 2654 | goto out; |
| 2655 | } |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2656 | |
| 2657 | spin_lock(&inode->i_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2658 | inode->i_blocks += blocks_per_huge_page(h); |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 2659 | spin_unlock(&inode->i_lock); |
Mel Gorman | 23be746 | 2010-04-23 13:17:56 -0400 | [diff] [blame] | 2660 | } else { |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2661 | lock_page(page); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2662 | if (unlikely(anon_vma_prepare(vma))) { |
| 2663 | ret = VM_FAULT_OOM; |
| 2664 | goto backout_unlocked; |
| 2665 | } |
Hillf Danton | 409eb8c | 2012-01-20 14:34:13 -0800 | [diff] [blame] | 2666 | anon_rmap = 1; |
Mel Gorman | 23be746 | 2010-04-23 13:17:56 -0400 | [diff] [blame] | 2667 | } |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2668 | } else { |
Naoya Horiguchi | 998b438 | 2010-09-08 10:19:32 +0900 | [diff] [blame] | 2669 | /* |
| 2670 | * If memory error occurs between mmap() and fault, some process |
| 2671 | * don't have hwpoisoned swap entry for errored virtual address. |
| 2672 | * So we need to block hugepage fault by PG_hwpoison bit check. |
| 2673 | */ |
| 2674 | if (unlikely(PageHWPoison(page))) { |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2675 | ret = VM_FAULT_HWPOISON | |
Andi Kleen | aa50d3a | 2010-10-06 21:45:00 +0200 | [diff] [blame] | 2676 | VM_FAULT_SET_HINDEX(h - hstates); |
Naoya Horiguchi | 998b438 | 2010-09-08 10:19:32 +0900 | [diff] [blame] | 2677 | goto backout_unlocked; |
| 2678 | } |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 2679 | } |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2680 | |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2681 | /* |
| 2682 | * If we are going to COW a private mapping later, we examine the |
| 2683 | * pending reservations for this page now. This will ensure that |
| 2684 | * any allocations necessary to record that reservation occur outside |
| 2685 | * the spinlock. |
| 2686 | */ |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2687 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2688 | if (vma_needs_reservation(h, vma, address) < 0) { |
| 2689 | ret = VM_FAULT_OOM; |
| 2690 | goto backout_unlocked; |
| 2691 | } |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2692 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2693 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2694 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2695 | if (idx >= size) |
| 2696 | goto backout; |
| 2697 | |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2698 | ret = 0; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2699 | if (!huge_pte_none(huge_ptep_get(ptep))) |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2700 | goto backout; |
| 2701 | |
Hillf Danton | 409eb8c | 2012-01-20 14:34:13 -0800 | [diff] [blame] | 2702 | if (anon_rmap) |
| 2703 | hugepage_add_new_anon_rmap(page, vma, address); |
| 2704 | else |
| 2705 | page_dup_rmap(page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2706 | new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) |
| 2707 | && (vma->vm_flags & VM_SHARED))); |
| 2708 | set_huge_pte_at(mm, address, ptep, new_pte); |
| 2709 | |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2710 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2711 | /* Optimization, do the COW without a second fault */ |
Mel Gorman | 04f2cbe | 2008-07-23 21:27:25 -0700 | [diff] [blame] | 2712 | ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2713 | } |
| 2714 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2715 | spin_unlock(&mm->page_table_lock); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2716 | unlock_page(page); |
| 2717 | out: |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2718 | return ret; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2719 | |
| 2720 | backout: |
| 2721 | spin_unlock(&mm->page_table_lock); |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2722 | backout_unlocked: |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2723 | unlock_page(page); |
| 2724 | put_page(page); |
| 2725 | goto out; |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 2726 | } |
| 2727 | |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2728 | int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2729 | unsigned long address, unsigned int flags) |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2730 | { |
| 2731 | pte_t *ptep; |
| 2732 | pte_t entry; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2733 | int ret; |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2734 | struct page *page = NULL; |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2735 | struct page *pagecache_page = NULL; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2736 | static DEFINE_MUTEX(hugetlb_instantiation_mutex); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2737 | struct hstate *h = hstate_vma(vma); |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2738 | |
KAMEZAWA Hiroyuki | 1e16a53 | 2012-01-10 15:07:22 -0800 | [diff] [blame] | 2739 | address &= huge_page_mask(h); |
| 2740 | |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2741 | ptep = huge_pte_offset(mm, address); |
| 2742 | if (ptep) { |
| 2743 | entry = huge_ptep_get(ptep); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 2744 | if (unlikely(is_hugetlb_entry_migration(entry))) { |
| 2745 | migration_entry_wait(mm, (pmd_t *)ptep, address); |
| 2746 | return 0; |
| 2747 | } else if (unlikely(is_hugetlb_entry_hwpoisoned(entry))) |
Chris Forbes | 32f8452 | 2011-07-25 17:12:14 -0700 | [diff] [blame] | 2748 | return VM_FAULT_HWPOISON_LARGE | |
Andi Kleen | aa50d3a | 2010-10-06 21:45:00 +0200 | [diff] [blame] | 2749 | VM_FAULT_SET_HINDEX(h - hstates); |
Naoya Horiguchi | fd6a03e | 2010-05-28 09:29:21 +0900 | [diff] [blame] | 2750 | } |
| 2751 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2752 | ptep = huge_pte_alloc(mm, address, huge_page_size(h)); |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2753 | if (!ptep) |
| 2754 | return VM_FAULT_OOM; |
| 2755 | |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2756 | /* |
| 2757 | * Serialize hugepage allocation and instantiation, so that we don't |
| 2758 | * get spurious allocation failures if two CPUs race to instantiate |
| 2759 | * the same page in the page cache. |
| 2760 | */ |
| 2761 | mutex_lock(&hugetlb_instantiation_mutex); |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2762 | entry = huge_ptep_get(ptep); |
| 2763 | if (huge_pte_none(entry)) { |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2764 | ret = hugetlb_no_page(mm, vma, address, ptep, flags); |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2765 | goto out_mutex; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2766 | } |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2767 | |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 2768 | ret = 0; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2769 | |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2770 | /* |
| 2771 | * If we are going to COW the mapping later, we examine the pending |
| 2772 | * reservations for this page now. This will ensure that any |
| 2773 | * allocations necessary to record that reservation occur outside the |
| 2774 | * spinlock. For private mappings, we also lookup the pagecache |
| 2775 | * page now as it is used to determine if a reservation has been |
| 2776 | * consumed. |
| 2777 | */ |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2778 | if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) { |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2779 | if (vma_needs_reservation(h, vma, address) < 0) { |
| 2780 | ret = VM_FAULT_OOM; |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2781 | goto out_mutex; |
Andy Whitcroft | 2b26736 | 2008-08-12 15:08:49 -0700 | [diff] [blame] | 2782 | } |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2783 | |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2784 | if (!(vma->vm_flags & VM_MAYSHARE)) |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2785 | pagecache_page = hugetlbfs_pagecache_page(h, |
| 2786 | vma, address); |
| 2787 | } |
| 2788 | |
Naoya Horiguchi | 56c9cfb | 2010-09-10 13:23:04 +0900 | [diff] [blame] | 2789 | /* |
| 2790 | * hugetlb_cow() requires page locks of pte_page(entry) and |
| 2791 | * pagecache_page, so here we need take the former one |
| 2792 | * when page != pagecache_page or !pagecache_page. |
| 2793 | * Note that locking order is always pagecache_page -> page, |
| 2794 | * so no worry about deadlock. |
| 2795 | */ |
| 2796 | page = pte_page(entry); |
| 2797 | if (page != pagecache_page) |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2798 | lock_page(page); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 2799 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2800 | spin_lock(&mm->page_table_lock); |
| 2801 | /* Check for a racing update before calling hugetlb_cow */ |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2802 | if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) |
| 2803 | goto out_page_table_lock; |
| 2804 | |
| 2805 | |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2806 | if (flags & FAULT_FLAG_WRITE) { |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2807 | if (!pte_write(entry)) { |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2808 | ret = hugetlb_cow(mm, vma, address, ptep, entry, |
| 2809 | pagecache_page); |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2810 | goto out_page_table_lock; |
| 2811 | } |
| 2812 | entry = pte_mkdirty(entry); |
| 2813 | } |
| 2814 | entry = pte_mkyoung(entry); |
Hugh Dickins | 788c7df | 2009-06-23 13:49:05 +0100 | [diff] [blame] | 2815 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, |
| 2816 | flags & FAULT_FLAG_WRITE)) |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 2817 | update_mmu_cache(vma, address, ptep); |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2818 | |
| 2819 | out_page_table_lock: |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2820 | spin_unlock(&mm->page_table_lock); |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2821 | |
| 2822 | if (pagecache_page) { |
| 2823 | unlock_page(pagecache_page); |
| 2824 | put_page(pagecache_page); |
| 2825 | } |
Dean Nelson | 1f64d69 | 2010-12-02 14:31:12 -0800 | [diff] [blame] | 2826 | if (page != pagecache_page) |
| 2827 | unlock_page(page); |
Andy Whitcroft | 57303d8 | 2008-08-12 15:08:47 -0700 | [diff] [blame] | 2828 | |
David Gibson | b4d1d99 | 2008-10-15 22:01:11 -0700 | [diff] [blame] | 2829 | out_mutex: |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 2830 | mutex_unlock(&hugetlb_instantiation_mutex); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 2831 | |
| 2832 | return ret; |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 2833 | } |
| 2834 | |
Andi Kleen | ceb8687 | 2008-07-23 21:27:50 -0700 | [diff] [blame] | 2835 | /* Can be overriden by architectures */ |
| 2836 | __attribute__((weak)) struct page * |
| 2837 | follow_huge_pud(struct mm_struct *mm, unsigned long address, |
| 2838 | pud_t *pud, int write) |
| 2839 | { |
| 2840 | BUG(); |
| 2841 | return NULL; |
| 2842 | } |
| 2843 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2844 | int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2845 | struct page **pages, struct vm_area_struct **vmas, |
Adam Litke | 5b23dbe | 2007-11-14 16:59:33 -0800 | [diff] [blame] | 2846 | unsigned long *position, int *length, int i, |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2847 | unsigned int flags) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2848 | { |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2849 | unsigned long pfn_offset; |
| 2850 | unsigned long vaddr = *position; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2851 | int remainder = *length; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2852 | struct hstate *h = hstate_vma(vma); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2853 | |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2854 | spin_lock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2855 | while (vaddr < vma->vm_end && remainder) { |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2856 | pte_t *pte; |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2857 | int absent; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2858 | struct page *page; |
| 2859 | |
| 2860 | /* |
| 2861 | * Some archs (sparc64, sh*) have multiple pte_ts to |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2862 | * each hugepage. We have to make sure we get the |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2863 | * first, for the page indexing below to work. |
| 2864 | */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2865 | pte = huge_pte_offset(mm, vaddr & huge_page_mask(h)); |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2866 | absent = !pte || huge_pte_none(huge_ptep_get(pte)); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2867 | |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2868 | /* |
| 2869 | * When coredumping, it suits get_dump_page if we just return |
Hugh Dickins | 3ae77f4 | 2009-09-21 17:03:33 -0700 | [diff] [blame] | 2870 | * an error where there's an empty slot with no huge pagecache |
| 2871 | * to back it. This way, we avoid allocating a hugepage, and |
| 2872 | * the sparse dumpfile avoids allocating disk blocks, but its |
| 2873 | * huge holes still show up with zeroes where they need to be. |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2874 | */ |
Hugh Dickins | 3ae77f4 | 2009-09-21 17:03:33 -0700 | [diff] [blame] | 2875 | if (absent && (flags & FOLL_DUMP) && |
| 2876 | !hugetlbfs_pagecache_present(h, vma, vaddr)) { |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2877 | remainder = 0; |
| 2878 | break; |
| 2879 | } |
| 2880 | |
| 2881 | if (absent || |
| 2882 | ((flags & FOLL_WRITE) && !pte_write(huge_ptep_get(pte)))) { |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2883 | int ret; |
| 2884 | |
| 2885 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2886 | ret = hugetlb_fault(mm, vma, vaddr, |
| 2887 | (flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2888 | spin_lock(&mm->page_table_lock); |
Adam Litke | a89182c | 2007-08-22 14:01:51 -0700 | [diff] [blame] | 2889 | if (!(ret & VM_FAULT_ERROR)) |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2890 | continue; |
| 2891 | |
| 2892 | remainder = 0; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 2893 | break; |
| 2894 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2895 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2896 | pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2897 | page = pte_page(huge_ptep_get(pte)); |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2898 | same_page: |
Chen, Kenneth W | d669218 | 2006-03-31 02:29:57 -0800 | [diff] [blame] | 2899 | if (pages) { |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2900 | pages[i] = mem_map_offset(page, pfn_offset); |
KOSAKI Motohiro | 4b2e38a | 2008-10-18 20:27:10 -0700 | [diff] [blame] | 2901 | get_page(pages[i]); |
Chen, Kenneth W | d669218 | 2006-03-31 02:29:57 -0800 | [diff] [blame] | 2902 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2903 | |
| 2904 | if (vmas) |
| 2905 | vmas[i] = vma; |
| 2906 | |
| 2907 | vaddr += PAGE_SIZE; |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2908 | ++pfn_offset; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2909 | --remainder; |
| 2910 | ++i; |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2911 | if (vaddr < vma->vm_end && remainder && |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2912 | pfn_offset < pages_per_huge_page(h)) { |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 2913 | /* |
| 2914 | * We use pfn_offset to avoid touching the pageframes |
| 2915 | * of this compound page. |
| 2916 | */ |
| 2917 | goto same_page; |
| 2918 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2919 | } |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 2920 | spin_unlock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2921 | *length = remainder; |
| 2922 | *position = vaddr; |
| 2923 | |
Hugh Dickins | 2a15efc | 2009-09-21 17:03:27 -0700 | [diff] [blame] | 2924 | return i ? i : -EFAULT; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 2925 | } |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2926 | |
| 2927 | void hugetlb_change_protection(struct vm_area_struct *vma, |
| 2928 | unsigned long address, unsigned long end, pgprot_t newprot) |
| 2929 | { |
| 2930 | struct mm_struct *mm = vma->vm_mm; |
| 2931 | unsigned long start = address; |
| 2932 | pte_t *ptep; |
| 2933 | pte_t pte; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2934 | struct hstate *h = hstate_vma(vma); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2935 | |
| 2936 | BUG_ON(address >= end); |
| 2937 | flush_cache_range(vma, address, end); |
| 2938 | |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2939 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2940 | spin_lock(&mm->page_table_lock); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2941 | for (; address < end; address += huge_page_size(h)) { |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2942 | ptep = huge_pte_offset(mm, address); |
| 2943 | if (!ptep) |
| 2944 | continue; |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 2945 | if (huge_pmd_unshare(mm, &address, ptep)) |
| 2946 | continue; |
Gerald Schaefer | 7f2e952 | 2008-04-28 02:13:29 -0700 | [diff] [blame] | 2947 | if (!huge_pte_none(huge_ptep_get(ptep))) { |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2948 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
| 2949 | pte = pte_mkhuge(pte_modify(pte, newprot)); |
| 2950 | set_huge_pte_at(mm, address, ptep, pte); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2951 | } |
| 2952 | } |
| 2953 | spin_unlock(&mm->page_table_lock); |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 2954 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 2955 | |
| 2956 | flush_tlb_range(vma, start, end); |
| 2957 | } |
| 2958 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2959 | int hugetlb_reserve_pages(struct inode *inode, |
| 2960 | long from, long to, |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2961 | struct vm_area_struct *vma, |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 2962 | vm_flags_t vm_flags) |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 2963 | { |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2964 | long ret, chg; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2965 | struct hstate *h = hstate_inode(inode); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 2966 | struct hugepage_subpool *spool = subpool_inode(inode); |
Adam Litke | e4e574b | 2007-10-16 01:26:19 -0700 | [diff] [blame] | 2967 | |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2968 | /* |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2969 | * Only apply hugepage reservation if asked. At fault time, an |
| 2970 | * attempt will be made for VM_NORESERVE to allocate a page |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 2971 | * without using reserves |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2972 | */ |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 2973 | if (vm_flags & VM_NORESERVE) |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2974 | return 0; |
| 2975 | |
| 2976 | /* |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2977 | * Shared mappings base their reservation on the number of pages that |
| 2978 | * are already allocated on behalf of the file. Private mappings need |
| 2979 | * to reserve the full area even if read-only as mprotect() may be |
| 2980 | * called to make the mapping read-write. Assume !vma is a shm mapping |
| 2981 | */ |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 2982 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 2983 | chg = region_chg(&inode->i_mapping->private_list, from, to); |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2984 | else { |
| 2985 | struct resv_map *resv_map = resv_map_alloc(); |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2986 | if (!resv_map) |
| 2987 | return -ENOMEM; |
| 2988 | |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2989 | chg = to - from; |
| 2990 | |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 2991 | set_vma_resv_map(vma, resv_map); |
| 2992 | set_vma_resv_flags(vma, HPAGE_RESV_OWNER); |
| 2993 | } |
| 2994 | |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 2995 | if (chg < 0) |
| 2996 | return chg; |
| 2997 | |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 2998 | /* There must be enough pages in the subpool for the mapping */ |
| 2999 | if (hugepage_subpool_get_pages(spool, chg)) |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 3000 | return -ENOSPC; |
| 3001 | |
| 3002 | /* |
| 3003 | * Check enough hugepages are available for the reservation. |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 3004 | * Hand the pages back to the subpool if there are not |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 3005 | */ |
| 3006 | ret = hugetlb_acct_memory(h, chg); |
| 3007 | if (ret < 0) { |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 3008 | hugepage_subpool_put_pages(spool, chg); |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 3009 | return ret; |
| 3010 | } |
| 3011 | |
| 3012 | /* |
| 3013 | * Account for the reservations made. Shared mappings record regions |
| 3014 | * that have reservations as they are shared by multiple VMAs. |
| 3015 | * When the last VMA disappears, the region map says how much |
| 3016 | * the reservation was and the page cache tells how much of |
| 3017 | * the reservation was consumed. Private mappings are per-VMA and |
| 3018 | * only the consumed reservations are tracked. When the VMA |
| 3019 | * disappears, the original reservation is the VMA size and the |
| 3020 | * consumed reservations are stored in the map. Hence, nothing |
| 3021 | * else has to be done for private mappings here |
| 3022 | */ |
Mel Gorman | f83a275 | 2009-05-28 14:34:40 -0700 | [diff] [blame] | 3023 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
Mel Gorman | 17c9d12 | 2009-02-11 16:34:16 +0000 | [diff] [blame] | 3024 | region_add(&inode->i_mapping->private_list, from, to); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 3025 | return 0; |
| 3026 | } |
| 3027 | |
| 3028 | void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) |
| 3029 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 3030 | struct hstate *h = hstate_inode(inode); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 3031 | long chg = region_truncate(&inode->i_mapping->private_list, offset); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 3032 | struct hugepage_subpool *spool = subpool_inode(inode); |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 3033 | |
| 3034 | spin_lock(&inode->i_lock); |
Eric Sandeen | e4c6f8b | 2009-07-29 15:02:16 -0700 | [diff] [blame] | 3035 | inode->i_blocks -= (blocks_per_huge_page(h) * freed); |
Ken Chen | 45c682a | 2007-11-14 16:59:44 -0800 | [diff] [blame] | 3036 | spin_unlock(&inode->i_lock); |
| 3037 | |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 3038 | hugepage_subpool_put_pages(spool, (chg - freed)); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 3039 | hugetlb_acct_memory(h, -(chg - freed)); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 3040 | } |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 3041 | |
Andi Kleen | d5bd910 | 2010-09-27 09:00:12 +0200 | [diff] [blame] | 3042 | #ifdef CONFIG_MEMORY_FAILURE |
| 3043 | |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3044 | /* Should be called in hugetlb_lock */ |
| 3045 | static int is_hugepage_on_freelist(struct page *hpage) |
| 3046 | { |
| 3047 | struct page *page; |
| 3048 | struct page *tmp; |
| 3049 | struct hstate *h = page_hstate(hpage); |
| 3050 | int nid = page_to_nid(hpage); |
| 3051 | |
| 3052 | list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru) |
| 3053 | if (page == hpage) |
| 3054 | return 1; |
| 3055 | return 0; |
| 3056 | } |
| 3057 | |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 3058 | /* |
| 3059 | * This function is called from memory failure code. |
| 3060 | * Assume the caller holds page lock of the head page. |
| 3061 | */ |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3062 | int dequeue_hwpoisoned_huge_page(struct page *hpage) |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 3063 | { |
| 3064 | struct hstate *h = page_hstate(hpage); |
| 3065 | int nid = page_to_nid(hpage); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3066 | int ret = -EBUSY; |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 3067 | |
| 3068 | spin_lock(&hugetlb_lock); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3069 | if (is_hugepage_on_freelist(hpage)) { |
| 3070 | list_del(&hpage->lru); |
Naoya Horiguchi | 8c6c2ec | 2010-09-08 10:19:38 +0900 | [diff] [blame] | 3071 | set_page_refcounted(hpage); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3072 | h->free_huge_pages--; |
| 3073 | h->free_huge_pages_node[nid]--; |
| 3074 | ret = 0; |
| 3075 | } |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 3076 | spin_unlock(&hugetlb_lock); |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3077 | return ret; |
Naoya Horiguchi | 93f70f9 | 2010-05-28 09:29:20 +0900 | [diff] [blame] | 3078 | } |
Naoya Horiguchi | 6de2b1a | 2010-09-08 10:19:36 +0900 | [diff] [blame] | 3079 | #endif |