Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Generic hugetlb support. |
| 3 | * (C) William Irwin, April 2004 |
| 4 | */ |
| 5 | #include <linux/gfp.h> |
| 6 | #include <linux/list.h> |
| 7 | #include <linux/init.h> |
| 8 | #include <linux/module.h> |
| 9 | #include <linux/mm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | #include <linux/sysctl.h> |
| 11 | #include <linux/highmem.h> |
| 12 | #include <linux/nodemask.h> |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 13 | #include <linux/pagemap.h> |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 14 | #include <linux/mempolicy.h> |
Christoph Lameter | aea47ff | 2006-01-08 01:00:57 -0800 | [diff] [blame] | 15 | #include <linux/cpuset.h> |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 16 | #include <linux/mutex.h> |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 17 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 18 | #include <asm/page.h> |
| 19 | #include <asm/pgtable.h> |
| 20 | |
| 21 | #include <linux/hugetlb.h> |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 22 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | |
| 24 | const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL; |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 25 | static unsigned long nr_huge_pages, free_huge_pages, resv_huge_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | unsigned long max_huge_pages; |
| 27 | static struct list_head hugepage_freelists[MAX_NUMNODES]; |
| 28 | static unsigned int nr_huge_pages_node[MAX_NUMNODES]; |
| 29 | static unsigned int free_huge_pages_node[MAX_NUMNODES]; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 30 | /* |
| 31 | * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages |
| 32 | */ |
| 33 | static DEFINE_SPINLOCK(hugetlb_lock); |
Eric Paris | 0bd0f9f | 2005-11-21 21:32:28 -0800 | [diff] [blame] | 34 | |
David Gibson | 79ac6ba | 2006-03-22 00:08:51 -0800 | [diff] [blame] | 35 | static void clear_huge_page(struct page *page, unsigned long addr) |
| 36 | { |
| 37 | int i; |
| 38 | |
| 39 | might_sleep(); |
| 40 | for (i = 0; i < (HPAGE_SIZE/PAGE_SIZE); i++) { |
| 41 | cond_resched(); |
| 42 | clear_user_highpage(page + i, addr); |
| 43 | } |
| 44 | } |
| 45 | |
| 46 | static void copy_huge_page(struct page *dst, struct page *src, |
| 47 | unsigned long addr) |
| 48 | { |
| 49 | int i; |
| 50 | |
| 51 | might_sleep(); |
| 52 | for (i = 0; i < HPAGE_SIZE/PAGE_SIZE; i++) { |
| 53 | cond_resched(); |
| 54 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE); |
| 55 | } |
| 56 | } |
| 57 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | static void enqueue_huge_page(struct page *page) |
| 59 | { |
| 60 | int nid = page_to_nid(page); |
| 61 | list_add(&page->lru, &hugepage_freelists[nid]); |
| 62 | free_huge_pages++; |
| 63 | free_huge_pages_node[nid]++; |
| 64 | } |
| 65 | |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 66 | static struct page *dequeue_huge_page(struct vm_area_struct *vma, |
| 67 | unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | { |
| 69 | int nid = numa_node_id(); |
| 70 | struct page *page = NULL; |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 71 | struct zonelist *zonelist = huge_zonelist(vma, address); |
Christoph Lameter | 96df933 | 2006-01-06 00:10:45 -0800 | [diff] [blame] | 72 | struct zone **z; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | |
Christoph Lameter | 96df933 | 2006-01-06 00:10:45 -0800 | [diff] [blame] | 74 | for (z = zonelist->zones; *z; z++) { |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 75 | nid = zone_to_nid(*z); |
Christoph Lameter | aea47ff | 2006-01-08 01:00:57 -0800 | [diff] [blame] | 76 | if (cpuset_zone_allowed(*z, GFP_HIGHUSER) && |
| 77 | !list_empty(&hugepage_freelists[nid])) |
Christoph Lameter | 96df933 | 2006-01-06 00:10:45 -0800 | [diff] [blame] | 78 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | } |
Christoph Lameter | 96df933 | 2006-01-06 00:10:45 -0800 | [diff] [blame] | 80 | |
| 81 | if (*z) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | page = list_entry(hugepage_freelists[nid].next, |
| 83 | struct page, lru); |
| 84 | list_del(&page->lru); |
| 85 | free_huge_pages--; |
| 86 | free_huge_pages_node[nid]--; |
| 87 | } |
| 88 | return page; |
| 89 | } |
| 90 | |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 91 | static void free_huge_page(struct page *page) |
| 92 | { |
| 93 | BUG_ON(page_count(page)); |
| 94 | |
| 95 | INIT_LIST_HEAD(&page->lru); |
| 96 | |
| 97 | spin_lock(&hugetlb_lock); |
| 98 | enqueue_huge_page(page); |
| 99 | spin_unlock(&hugetlb_lock); |
| 100 | } |
| 101 | |
Nick Piggin | a482289 | 2006-03-22 00:08:08 -0800 | [diff] [blame] | 102 | static int alloc_fresh_huge_page(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | { |
| 104 | static int nid = 0; |
| 105 | struct page *page; |
| 106 | page = alloc_pages_node(nid, GFP_HIGHUSER|__GFP_COMP|__GFP_NOWARN, |
| 107 | HUGETLB_PAGE_ORDER); |
Paul Jackson | fdb7cc5 | 2006-03-22 00:09:10 -0800 | [diff] [blame] | 108 | nid = next_node(nid, node_online_map); |
| 109 | if (nid == MAX_NUMNODES) |
| 110 | nid = first_node(node_online_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | if (page) { |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame^] | 112 | set_compound_page_dtor(page, free_huge_page); |
Eric Paris | 0bd0f9f | 2005-11-21 21:32:28 -0800 | [diff] [blame] | 113 | spin_lock(&hugetlb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | nr_huge_pages++; |
| 115 | nr_huge_pages_node[page_to_nid(page)]++; |
Eric Paris | 0bd0f9f | 2005-11-21 21:32:28 -0800 | [diff] [blame] | 116 | spin_unlock(&hugetlb_lock); |
Nick Piggin | a482289 | 2006-03-22 00:08:08 -0800 | [diff] [blame] | 117 | put_page(page); /* free it into the hugepage allocator */ |
| 118 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | } |
Nick Piggin | a482289 | 2006-03-22 00:08:08 -0800 | [diff] [blame] | 120 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | } |
| 122 | |
David Gibson | 27a85ef | 2006-03-22 00:08:56 -0800 | [diff] [blame] | 123 | static struct page *alloc_huge_page(struct vm_area_struct *vma, |
| 124 | unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | { |
| 126 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | |
| 128 | spin_lock(&hugetlb_lock); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 129 | if (vma->vm_flags & VM_MAYSHARE) |
| 130 | resv_huge_pages--; |
| 131 | else if (free_huge_pages <= resv_huge_pages) |
| 132 | goto fail; |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 133 | |
| 134 | page = dequeue_huge_page(vma, addr); |
| 135 | if (!page) |
| 136 | goto fail; |
| 137 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | spin_unlock(&hugetlb_lock); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 139 | set_page_refcounted(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | return page; |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 141 | |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 142 | fail: |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 143 | spin_unlock(&hugetlb_lock); |
| 144 | return NULL; |
| 145 | } |
| 146 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | static int __init hugetlb_init(void) |
| 148 | { |
| 149 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | |
Benjamin Herrenschmidt | 3c726f8 | 2005-11-07 11:06:55 +1100 | [diff] [blame] | 151 | if (HPAGE_SHIFT == 0) |
| 152 | return 0; |
| 153 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | for (i = 0; i < MAX_NUMNODES; ++i) |
| 155 | INIT_LIST_HEAD(&hugepage_freelists[i]); |
| 156 | |
| 157 | for (i = 0; i < max_huge_pages; ++i) { |
Nick Piggin | a482289 | 2006-03-22 00:08:08 -0800 | [diff] [blame] | 158 | if (!alloc_fresh_huge_page()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | } |
| 161 | max_huge_pages = free_huge_pages = nr_huge_pages = i; |
| 162 | printk("Total HugeTLB memory allocated, %ld\n", free_huge_pages); |
| 163 | return 0; |
| 164 | } |
| 165 | module_init(hugetlb_init); |
| 166 | |
| 167 | static int __init hugetlb_setup(char *s) |
| 168 | { |
| 169 | if (sscanf(s, "%lu", &max_huge_pages) <= 0) |
| 170 | max_huge_pages = 0; |
| 171 | return 1; |
| 172 | } |
| 173 | __setup("hugepages=", hugetlb_setup); |
| 174 | |
| 175 | #ifdef CONFIG_SYSCTL |
| 176 | static void update_and_free_page(struct page *page) |
| 177 | { |
| 178 | int i; |
| 179 | nr_huge_pages--; |
Christoph Lameter | 4415cc8 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 180 | nr_huge_pages_node[page_to_nid(page)]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | for (i = 0; i < (HPAGE_SIZE / PAGE_SIZE); i++) { |
| 182 | page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 1 << PG_referenced | |
| 183 | 1 << PG_dirty | 1 << PG_active | 1 << PG_reserved | |
| 184 | 1 << PG_private | 1<< PG_writeback); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | } |
Nick Piggin | a482289 | 2006-03-22 00:08:08 -0800 | [diff] [blame] | 186 | page[1].lru.next = NULL; |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 187 | set_page_refcounted(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | __free_pages(page, HUGETLB_PAGE_ORDER); |
| 189 | } |
| 190 | |
| 191 | #ifdef CONFIG_HIGHMEM |
| 192 | static void try_to_free_low(unsigned long count) |
| 193 | { |
Christoph Lameter | 4415cc8 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 194 | int i; |
| 195 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | for (i = 0; i < MAX_NUMNODES; ++i) { |
| 197 | struct page *page, *next; |
| 198 | list_for_each_entry_safe(page, next, &hugepage_freelists[i], lru) { |
| 199 | if (PageHighMem(page)) |
| 200 | continue; |
| 201 | list_del(&page->lru); |
| 202 | update_and_free_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | free_huge_pages--; |
Christoph Lameter | 4415cc8 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 204 | free_huge_pages_node[page_to_nid(page)]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | if (count >= nr_huge_pages) |
| 206 | return; |
| 207 | } |
| 208 | } |
| 209 | } |
| 210 | #else |
| 211 | static inline void try_to_free_low(unsigned long count) |
| 212 | { |
| 213 | } |
| 214 | #endif |
| 215 | |
| 216 | static unsigned long set_max_huge_pages(unsigned long count) |
| 217 | { |
| 218 | while (count > nr_huge_pages) { |
Nick Piggin | a482289 | 2006-03-22 00:08:08 -0800 | [diff] [blame] | 219 | if (!alloc_fresh_huge_page()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | return nr_huge_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | } |
| 222 | if (count >= nr_huge_pages) |
| 223 | return nr_huge_pages; |
| 224 | |
| 225 | spin_lock(&hugetlb_lock); |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 226 | count = max(count, resv_huge_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | try_to_free_low(count); |
| 228 | while (count < nr_huge_pages) { |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 229 | struct page *page = dequeue_huge_page(NULL, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | if (!page) |
| 231 | break; |
| 232 | update_and_free_page(page); |
| 233 | } |
| 234 | spin_unlock(&hugetlb_lock); |
| 235 | return nr_huge_pages; |
| 236 | } |
| 237 | |
| 238 | int hugetlb_sysctl_handler(struct ctl_table *table, int write, |
| 239 | struct file *file, void __user *buffer, |
| 240 | size_t *length, loff_t *ppos) |
| 241 | { |
| 242 | proc_doulongvec_minmax(table, write, file, buffer, length, ppos); |
| 243 | max_huge_pages = set_max_huge_pages(max_huge_pages); |
| 244 | return 0; |
| 245 | } |
| 246 | #endif /* CONFIG_SYSCTL */ |
| 247 | |
| 248 | int hugetlb_report_meminfo(char *buf) |
| 249 | { |
| 250 | return sprintf(buf, |
| 251 | "HugePages_Total: %5lu\n" |
| 252 | "HugePages_Free: %5lu\n" |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 253 | "HugePages_Rsvd: %5lu\n" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | "Hugepagesize: %5lu kB\n", |
| 255 | nr_huge_pages, |
| 256 | free_huge_pages, |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 257 | resv_huge_pages, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | HPAGE_SIZE/1024); |
| 259 | } |
| 260 | |
| 261 | int hugetlb_report_node_meminfo(int nid, char *buf) |
| 262 | { |
| 263 | return sprintf(buf, |
| 264 | "Node %d HugePages_Total: %5u\n" |
| 265 | "Node %d HugePages_Free: %5u\n", |
| 266 | nid, nr_huge_pages_node[nid], |
| 267 | nid, free_huge_pages_node[nid]); |
| 268 | } |
| 269 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ |
| 271 | unsigned long hugetlb_total_pages(void) |
| 272 | { |
| 273 | return nr_huge_pages * (HPAGE_SIZE / PAGE_SIZE); |
| 274 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | |
| 276 | /* |
| 277 | * We cannot handle pagefaults against hugetlb pages at all. They cause |
| 278 | * handle_mm_fault() to try to instantiate regular-sized pages in the |
| 279 | * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get |
| 280 | * this far. |
| 281 | */ |
| 282 | static struct page *hugetlb_nopage(struct vm_area_struct *vma, |
| 283 | unsigned long address, int *unused) |
| 284 | { |
| 285 | BUG(); |
| 286 | return NULL; |
| 287 | } |
| 288 | |
| 289 | struct vm_operations_struct hugetlb_vm_ops = { |
| 290 | .nopage = hugetlb_nopage, |
| 291 | }; |
| 292 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 293 | static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, |
| 294 | int writable) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 295 | { |
| 296 | pte_t entry; |
| 297 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 298 | if (writable) { |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 299 | entry = |
| 300 | pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))); |
| 301 | } else { |
| 302 | entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot)); |
| 303 | } |
| 304 | entry = pte_mkyoung(entry); |
| 305 | entry = pte_mkhuge(entry); |
| 306 | |
| 307 | return entry; |
| 308 | } |
| 309 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 310 | static void set_huge_ptep_writable(struct vm_area_struct *vma, |
| 311 | unsigned long address, pte_t *ptep) |
| 312 | { |
| 313 | pte_t entry; |
| 314 | |
| 315 | entry = pte_mkwrite(pte_mkdirty(*ptep)); |
| 316 | ptep_set_access_flags(vma, address, ptep, entry, 1); |
| 317 | update_mmu_cache(vma, address, entry); |
| 318 | lazy_mmu_prot_update(entry); |
| 319 | } |
| 320 | |
| 321 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 322 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, |
| 323 | struct vm_area_struct *vma) |
| 324 | { |
| 325 | pte_t *src_pte, *dst_pte, entry; |
| 326 | struct page *ptepage; |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 327 | unsigned long addr; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 328 | int cow; |
| 329 | |
| 330 | cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 331 | |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 332 | for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) { |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 333 | src_pte = huge_pte_offset(src, addr); |
| 334 | if (!src_pte) |
| 335 | continue; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 336 | dst_pte = huge_pte_alloc(dst, addr); |
| 337 | if (!dst_pte) |
| 338 | goto nomem; |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 339 | spin_lock(&dst->page_table_lock); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 340 | spin_lock(&src->page_table_lock); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 341 | if (!pte_none(*src_pte)) { |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 342 | if (cow) |
| 343 | ptep_set_wrprotect(src, addr, src_pte); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 344 | entry = *src_pte; |
| 345 | ptepage = pte_page(entry); |
| 346 | get_page(ptepage); |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 347 | set_huge_pte_at(dst, addr, dst_pte, entry); |
| 348 | } |
| 349 | spin_unlock(&src->page_table_lock); |
Hugh Dickins | c74df32 | 2005-10-29 18:16:23 -0700 | [diff] [blame] | 350 | spin_unlock(&dst->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 351 | } |
| 352 | return 0; |
| 353 | |
| 354 | nomem: |
| 355 | return -ENOMEM; |
| 356 | } |
| 357 | |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 358 | void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
| 359 | unsigned long end) |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 360 | { |
| 361 | struct mm_struct *mm = vma->vm_mm; |
| 362 | unsigned long address; |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 363 | pte_t *ptep; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 364 | pte_t pte; |
| 365 | struct page *page; |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 366 | struct page *tmp; |
Chen, Kenneth W | c0a499c | 2006-12-06 20:31:39 -0800 | [diff] [blame] | 367 | /* |
| 368 | * A page gathering list, protected by per file i_mmap_lock. The |
| 369 | * lock is used to avoid list corruption from multiple unmapping |
| 370 | * of the same page since we are using page->lru. |
| 371 | */ |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 372 | LIST_HEAD(page_list); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 373 | |
| 374 | WARN_ON(!is_vm_hugetlb_page(vma)); |
| 375 | BUG_ON(start & ~HPAGE_MASK); |
| 376 | BUG_ON(end & ~HPAGE_MASK); |
| 377 | |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 378 | spin_lock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 379 | for (address = start; address < end; address += HPAGE_SIZE) { |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 380 | ptep = huge_pte_offset(mm, address); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 381 | if (!ptep) |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 382 | continue; |
| 383 | |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 384 | if (huge_pmd_unshare(mm, &address, ptep)) |
| 385 | continue; |
| 386 | |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 387 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 388 | if (pte_none(pte)) |
| 389 | continue; |
David Gibson | c7546f8 | 2005-08-05 11:59:35 -0700 | [diff] [blame] | 390 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 391 | page = pte_page(pte); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 392 | list_add(&page->lru, &page_list); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 393 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | spin_unlock(&mm->page_table_lock); |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 395 | flush_tlb_range(vma, start, end); |
Chen, Kenneth W | fe1668a | 2006-10-04 02:15:24 -0700 | [diff] [blame] | 396 | list_for_each_entry_safe(page, tmp, &page_list, lru) { |
| 397 | list_del(&page->lru); |
| 398 | put_page(page); |
| 399 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 401 | |
Chen, Kenneth W | 502717f | 2006-10-11 01:20:46 -0700 | [diff] [blame] | 402 | void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
| 403 | unsigned long end) |
| 404 | { |
| 405 | /* |
| 406 | * It is undesirable to test vma->vm_file as it should be non-null |
| 407 | * for valid hugetlb area. However, vm_file will be NULL in the error |
| 408 | * cleanup path of do_mmap_pgoff. When hugetlbfs ->mmap method fails, |
| 409 | * do_mmap_pgoff() nullifies vma->vm_file before calling this function |
| 410 | * to clean up. Since no pte has actually been setup, it is safe to |
| 411 | * do nothing in this case. |
| 412 | */ |
| 413 | if (vma->vm_file) { |
| 414 | spin_lock(&vma->vm_file->f_mapping->i_mmap_lock); |
| 415 | __unmap_hugepage_range(vma, start, end); |
| 416 | spin_unlock(&vma->vm_file->f_mapping->i_mmap_lock); |
| 417 | } |
| 418 | } |
| 419 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 420 | static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma, |
| 421 | unsigned long address, pte_t *ptep, pte_t pte) |
| 422 | { |
| 423 | struct page *old_page, *new_page; |
David Gibson | 79ac6ba | 2006-03-22 00:08:51 -0800 | [diff] [blame] | 424 | int avoidcopy; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 425 | |
| 426 | old_page = pte_page(pte); |
| 427 | |
| 428 | /* If no-one else is actually using this page, avoid the copy |
| 429 | * and just make the page writable */ |
| 430 | avoidcopy = (page_count(old_page) == 1); |
| 431 | if (avoidcopy) { |
| 432 | set_huge_ptep_writable(vma, address, ptep); |
| 433 | return VM_FAULT_MINOR; |
| 434 | } |
| 435 | |
| 436 | page_cache_get(old_page); |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 437 | new_page = alloc_huge_page(vma, address); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 438 | |
| 439 | if (!new_page) { |
| 440 | page_cache_release(old_page); |
Christoph Lameter | 0df420d | 2006-02-07 12:58:30 -0800 | [diff] [blame] | 441 | return VM_FAULT_OOM; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 442 | } |
| 443 | |
| 444 | spin_unlock(&mm->page_table_lock); |
David Gibson | 79ac6ba | 2006-03-22 00:08:51 -0800 | [diff] [blame] | 445 | copy_huge_page(new_page, old_page, address); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 446 | spin_lock(&mm->page_table_lock); |
| 447 | |
| 448 | ptep = huge_pte_offset(mm, address & HPAGE_MASK); |
| 449 | if (likely(pte_same(*ptep, pte))) { |
| 450 | /* Break COW */ |
| 451 | set_huge_pte_at(mm, address, ptep, |
| 452 | make_huge_pte(vma, new_page, 1)); |
| 453 | /* Make the old page be freed below */ |
| 454 | new_page = old_page; |
| 455 | } |
| 456 | page_cache_release(new_page); |
| 457 | page_cache_release(old_page); |
| 458 | return VM_FAULT_MINOR; |
| 459 | } |
| 460 | |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 461 | int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma, |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 462 | unsigned long address, pte_t *ptep, int write_access) |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 463 | { |
| 464 | int ret = VM_FAULT_SIGBUS; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 465 | unsigned long idx; |
| 466 | unsigned long size; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 467 | struct page *page; |
| 468 | struct address_space *mapping; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 469 | pte_t new_pte; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 470 | |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 471 | mapping = vma->vm_file->f_mapping; |
| 472 | idx = ((address - vma->vm_start) >> HPAGE_SHIFT) |
| 473 | + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT)); |
| 474 | |
| 475 | /* |
| 476 | * Use page lock to guard against racing truncation |
| 477 | * before we get page_table_lock. |
| 478 | */ |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 479 | retry: |
| 480 | page = find_lock_page(mapping, idx); |
| 481 | if (!page) { |
Hugh Dickins | ebed4bf | 2006-10-28 10:38:43 -0700 | [diff] [blame] | 482 | size = i_size_read(mapping->host) >> HPAGE_SHIFT; |
| 483 | if (idx >= size) |
| 484 | goto out; |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 485 | if (hugetlb_get_quota(mapping)) |
| 486 | goto out; |
| 487 | page = alloc_huge_page(vma, address); |
| 488 | if (!page) { |
| 489 | hugetlb_put_quota(mapping); |
Christoph Lameter | 0df420d | 2006-02-07 12:58:30 -0800 | [diff] [blame] | 490 | ret = VM_FAULT_OOM; |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 491 | goto out; |
| 492 | } |
David Gibson | 79ac6ba | 2006-03-22 00:08:51 -0800 | [diff] [blame] | 493 | clear_huge_page(page, address); |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 494 | |
Christoph Lameter | 6bda666 | 2006-01-06 00:10:49 -0800 | [diff] [blame] | 495 | if (vma->vm_flags & VM_SHARED) { |
| 496 | int err; |
| 497 | |
| 498 | err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); |
| 499 | if (err) { |
| 500 | put_page(page); |
| 501 | hugetlb_put_quota(mapping); |
| 502 | if (err == -EEXIST) |
| 503 | goto retry; |
| 504 | goto out; |
| 505 | } |
| 506 | } else |
| 507 | lock_page(page); |
| 508 | } |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 509 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 510 | spin_lock(&mm->page_table_lock); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 511 | size = i_size_read(mapping->host) >> HPAGE_SHIFT; |
| 512 | if (idx >= size) |
| 513 | goto backout; |
| 514 | |
| 515 | ret = VM_FAULT_MINOR; |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 516 | if (!pte_none(*ptep)) |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 517 | goto backout; |
| 518 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 519 | new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) |
| 520 | && (vma->vm_flags & VM_SHARED))); |
| 521 | set_huge_pte_at(mm, address, ptep, new_pte); |
| 522 | |
| 523 | if (write_access && !(vma->vm_flags & VM_SHARED)) { |
| 524 | /* Optimization, do the COW without a second fault */ |
| 525 | ret = hugetlb_cow(mm, vma, address, ptep, new_pte); |
| 526 | } |
| 527 | |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 528 | spin_unlock(&mm->page_table_lock); |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 529 | unlock_page(page); |
| 530 | out: |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 531 | return ret; |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 532 | |
| 533 | backout: |
| 534 | spin_unlock(&mm->page_table_lock); |
| 535 | hugetlb_put_quota(mapping); |
| 536 | unlock_page(page); |
| 537 | put_page(page); |
| 538 | goto out; |
Hugh Dickins | ac9b9c6 | 2005-10-20 16:24:28 +0100 | [diff] [blame] | 539 | } |
| 540 | |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 541 | int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
| 542 | unsigned long address, int write_access) |
| 543 | { |
| 544 | pte_t *ptep; |
| 545 | pte_t entry; |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 546 | int ret; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 547 | static DEFINE_MUTEX(hugetlb_instantiation_mutex); |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 548 | |
| 549 | ptep = huge_pte_alloc(mm, address); |
| 550 | if (!ptep) |
| 551 | return VM_FAULT_OOM; |
| 552 | |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 553 | /* |
| 554 | * Serialize hugepage allocation and instantiation, so that we don't |
| 555 | * get spurious allocation failures if two CPUs race to instantiate |
| 556 | * the same page in the page cache. |
| 557 | */ |
| 558 | mutex_lock(&hugetlb_instantiation_mutex); |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 559 | entry = *ptep; |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 560 | if (pte_none(entry)) { |
| 561 | ret = hugetlb_no_page(mm, vma, address, ptep, write_access); |
| 562 | mutex_unlock(&hugetlb_instantiation_mutex); |
| 563 | return ret; |
| 564 | } |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 565 | |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 566 | ret = VM_FAULT_MINOR; |
| 567 | |
| 568 | spin_lock(&mm->page_table_lock); |
| 569 | /* Check for a racing update before calling hugetlb_cow */ |
| 570 | if (likely(pte_same(entry, *ptep))) |
| 571 | if (write_access && !pte_write(entry)) |
| 572 | ret = hugetlb_cow(mm, vma, address, ptep, entry); |
| 573 | spin_unlock(&mm->page_table_lock); |
David Gibson | 3935baa | 2006-03-22 00:08:53 -0800 | [diff] [blame] | 574 | mutex_unlock(&hugetlb_instantiation_mutex); |
David Gibson | 1e8f889 | 2006-01-06 00:10:44 -0800 | [diff] [blame] | 575 | |
| 576 | return ret; |
Adam Litke | 86e5216 | 2006-01-06 00:10:43 -0800 | [diff] [blame] | 577 | } |
| 578 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 579 | int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, |
| 580 | struct page **pages, struct vm_area_struct **vmas, |
| 581 | unsigned long *position, int *length, int i) |
| 582 | { |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 583 | unsigned long pfn_offset; |
| 584 | unsigned long vaddr = *position; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 585 | int remainder = *length; |
| 586 | |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 587 | spin_lock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 588 | while (vaddr < vma->vm_end && remainder) { |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 589 | pte_t *pte; |
| 590 | struct page *page; |
| 591 | |
| 592 | /* |
| 593 | * Some archs (sparc64, sh*) have multiple pte_ts to |
| 594 | * each hugepage. We have to make * sure we get the |
| 595 | * first, for the page indexing below to work. |
| 596 | */ |
| 597 | pte = huge_pte_offset(mm, vaddr & HPAGE_MASK); |
| 598 | |
| 599 | if (!pte || pte_none(*pte)) { |
| 600 | int ret; |
| 601 | |
| 602 | spin_unlock(&mm->page_table_lock); |
| 603 | ret = hugetlb_fault(mm, vma, vaddr, 0); |
| 604 | spin_lock(&mm->page_table_lock); |
| 605 | if (ret == VM_FAULT_MINOR) |
| 606 | continue; |
| 607 | |
| 608 | remainder = 0; |
| 609 | if (!i) |
| 610 | i = -EFAULT; |
| 611 | break; |
| 612 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 613 | |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 614 | pfn_offset = (vaddr & ~HPAGE_MASK) >> PAGE_SHIFT; |
| 615 | page = pte_page(*pte); |
| 616 | same_page: |
Chen, Kenneth W | d669218 | 2006-03-31 02:29:57 -0800 | [diff] [blame] | 617 | if (pages) { |
| 618 | get_page(page); |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 619 | pages[i] = page + pfn_offset; |
Chen, Kenneth W | d669218 | 2006-03-31 02:29:57 -0800 | [diff] [blame] | 620 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 621 | |
| 622 | if (vmas) |
| 623 | vmas[i] = vma; |
| 624 | |
| 625 | vaddr += PAGE_SIZE; |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 626 | ++pfn_offset; |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 627 | --remainder; |
| 628 | ++i; |
Chen, Kenneth W | d5d4b0a | 2006-03-22 00:09:03 -0800 | [diff] [blame] | 629 | if (vaddr < vma->vm_end && remainder && |
| 630 | pfn_offset < HPAGE_SIZE/PAGE_SIZE) { |
| 631 | /* |
| 632 | * We use pfn_offset to avoid touching the pageframes |
| 633 | * of this compound page. |
| 634 | */ |
| 635 | goto same_page; |
| 636 | } |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 637 | } |
Hugh Dickins | 1c59827 | 2005-10-19 21:23:43 -0700 | [diff] [blame] | 638 | spin_unlock(&mm->page_table_lock); |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 639 | *length = remainder; |
| 640 | *position = vaddr; |
| 641 | |
| 642 | return i; |
| 643 | } |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 644 | |
| 645 | void hugetlb_change_protection(struct vm_area_struct *vma, |
| 646 | unsigned long address, unsigned long end, pgprot_t newprot) |
| 647 | { |
| 648 | struct mm_struct *mm = vma->vm_mm; |
| 649 | unsigned long start = address; |
| 650 | pte_t *ptep; |
| 651 | pte_t pte; |
| 652 | |
| 653 | BUG_ON(address >= end); |
| 654 | flush_cache_range(vma, address, end); |
| 655 | |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 656 | spin_lock(&vma->vm_file->f_mapping->i_mmap_lock); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 657 | spin_lock(&mm->page_table_lock); |
| 658 | for (; address < end; address += HPAGE_SIZE) { |
| 659 | ptep = huge_pte_offset(mm, address); |
| 660 | if (!ptep) |
| 661 | continue; |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 662 | if (huge_pmd_unshare(mm, &address, ptep)) |
| 663 | continue; |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 664 | if (!pte_none(*ptep)) { |
| 665 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
| 666 | pte = pte_mkhuge(pte_modify(pte, newprot)); |
| 667 | set_huge_pte_at(mm, address, ptep, pte); |
| 668 | lazy_mmu_prot_update(pte); |
| 669 | } |
| 670 | } |
| 671 | spin_unlock(&mm->page_table_lock); |
Chen, Kenneth W | 39dde65 | 2006-12-06 20:32:03 -0800 | [diff] [blame] | 672 | spin_unlock(&vma->vm_file->f_mapping->i_mmap_lock); |
Zhang, Yanmin | 8f86059 | 2006-03-22 00:08:50 -0800 | [diff] [blame] | 673 | |
| 674 | flush_tlb_range(vma, start, end); |
| 675 | } |
| 676 | |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 677 | struct file_region { |
| 678 | struct list_head link; |
| 679 | long from; |
| 680 | long to; |
| 681 | }; |
| 682 | |
| 683 | static long region_add(struct list_head *head, long f, long t) |
| 684 | { |
| 685 | struct file_region *rg, *nrg, *trg; |
| 686 | |
| 687 | /* Locate the region we are either in or before. */ |
| 688 | list_for_each_entry(rg, head, link) |
| 689 | if (f <= rg->to) |
| 690 | break; |
| 691 | |
| 692 | /* Round our left edge to the current segment if it encloses us. */ |
| 693 | if (f > rg->from) |
| 694 | f = rg->from; |
| 695 | |
| 696 | /* Check for and consume any regions we now overlap with. */ |
| 697 | nrg = rg; |
| 698 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { |
| 699 | if (&rg->link == head) |
| 700 | break; |
| 701 | if (rg->from > t) |
| 702 | break; |
| 703 | |
| 704 | /* If this area reaches higher then extend our area to |
| 705 | * include it completely. If this is not the first area |
| 706 | * which we intend to reuse, free it. */ |
| 707 | if (rg->to > t) |
| 708 | t = rg->to; |
| 709 | if (rg != nrg) { |
| 710 | list_del(&rg->link); |
| 711 | kfree(rg); |
| 712 | } |
| 713 | } |
| 714 | nrg->from = f; |
| 715 | nrg->to = t; |
| 716 | return 0; |
| 717 | } |
| 718 | |
| 719 | static long region_chg(struct list_head *head, long f, long t) |
| 720 | { |
| 721 | struct file_region *rg, *nrg; |
| 722 | long chg = 0; |
| 723 | |
| 724 | /* Locate the region we are before or in. */ |
| 725 | list_for_each_entry(rg, head, link) |
| 726 | if (f <= rg->to) |
| 727 | break; |
| 728 | |
| 729 | /* If we are below the current region then a new region is required. |
| 730 | * Subtle, allocate a new region at the position but make it zero |
| 731 | * size such that we can guarentee to record the reservation. */ |
| 732 | if (&rg->link == head || t < rg->from) { |
| 733 | nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); |
| 734 | if (nrg == 0) |
| 735 | return -ENOMEM; |
| 736 | nrg->from = f; |
| 737 | nrg->to = f; |
| 738 | INIT_LIST_HEAD(&nrg->link); |
| 739 | list_add(&nrg->link, rg->link.prev); |
| 740 | |
| 741 | return t - f; |
| 742 | } |
| 743 | |
| 744 | /* Round our left edge to the current segment if it encloses us. */ |
| 745 | if (f > rg->from) |
| 746 | f = rg->from; |
| 747 | chg = t - f; |
| 748 | |
| 749 | /* Check for and consume any regions we now overlap with. */ |
| 750 | list_for_each_entry(rg, rg->link.prev, link) { |
| 751 | if (&rg->link == head) |
| 752 | break; |
| 753 | if (rg->from > t) |
| 754 | return chg; |
| 755 | |
| 756 | /* We overlap with this area, if it extends futher than |
| 757 | * us then we must extend ourselves. Account for its |
| 758 | * existing reservation. */ |
| 759 | if (rg->to > t) { |
| 760 | chg += rg->to - t; |
| 761 | t = rg->to; |
| 762 | } |
| 763 | chg -= rg->to - rg->from; |
| 764 | } |
| 765 | return chg; |
| 766 | } |
| 767 | |
| 768 | static long region_truncate(struct list_head *head, long end) |
| 769 | { |
| 770 | struct file_region *rg, *trg; |
| 771 | long chg = 0; |
| 772 | |
| 773 | /* Locate the region we are either in or before. */ |
| 774 | list_for_each_entry(rg, head, link) |
| 775 | if (end <= rg->to) |
| 776 | break; |
| 777 | if (&rg->link == head) |
| 778 | return 0; |
| 779 | |
| 780 | /* If we are in the middle of a region then adjust it. */ |
| 781 | if (end > rg->from) { |
| 782 | chg = rg->to - end; |
| 783 | rg->to = end; |
| 784 | rg = list_entry(rg->link.next, typeof(*rg), link); |
| 785 | } |
| 786 | |
| 787 | /* Drop any remaining regions. */ |
| 788 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { |
| 789 | if (&rg->link == head) |
| 790 | break; |
| 791 | chg += rg->to - rg->from; |
| 792 | list_del(&rg->link); |
| 793 | kfree(rg); |
| 794 | } |
| 795 | return chg; |
| 796 | } |
| 797 | |
| 798 | static int hugetlb_acct_memory(long delta) |
| 799 | { |
| 800 | int ret = -ENOMEM; |
| 801 | |
| 802 | spin_lock(&hugetlb_lock); |
| 803 | if ((delta + resv_huge_pages) <= free_huge_pages) { |
| 804 | resv_huge_pages += delta; |
| 805 | ret = 0; |
| 806 | } |
| 807 | spin_unlock(&hugetlb_lock); |
| 808 | return ret; |
| 809 | } |
| 810 | |
| 811 | int hugetlb_reserve_pages(struct inode *inode, long from, long to) |
| 812 | { |
| 813 | long ret, chg; |
| 814 | |
| 815 | chg = region_chg(&inode->i_mapping->private_list, from, to); |
| 816 | if (chg < 0) |
| 817 | return chg; |
| 818 | ret = hugetlb_acct_memory(chg); |
| 819 | if (ret < 0) |
| 820 | return ret; |
| 821 | region_add(&inode->i_mapping->private_list, from, to); |
| 822 | return 0; |
| 823 | } |
| 824 | |
| 825 | void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) |
| 826 | { |
| 827 | long chg = region_truncate(&inode->i_mapping->private_list, offset); |
| 828 | hugetlb_acct_memory(freed - chg); |
| 829 | } |