blob: 41b1038f76dafa88d6f14c43319522656680e4e7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/sysctl.h>
11#include <linux/highmem.h>
12#include <linux/nodemask.h>
David Gibson63551ae2005-06-21 17:14:44 -070013#include <linux/pagemap.h>
Christoph Lameter5da7ca82006-01-06 00:10:46 -080014#include <linux/mempolicy.h>
Christoph Lameteraea47ff2006-01-08 01:00:57 -080015#include <linux/cpuset.h>
Christoph Lameter5da7ca82006-01-06 00:10:46 -080016
David Gibson63551ae2005-06-21 17:14:44 -070017#include <asm/page.h>
18#include <asm/pgtable.h>
19
20#include <linux/hugetlb.h>
Nick Piggin7835e982006-03-22 00:08:40 -080021#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL;
24static unsigned long nr_huge_pages, free_huge_pages;
25unsigned long max_huge_pages;
26static struct list_head hugepage_freelists[MAX_NUMNODES];
27static unsigned int nr_huge_pages_node[MAX_NUMNODES];
28static unsigned int free_huge_pages_node[MAX_NUMNODES];
Eric Paris0bd0f9f2005-11-21 21:32:28 -080029
David Gibson79ac6ba2006-03-22 00:08:51 -080030static void clear_huge_page(struct page *page, unsigned long addr)
31{
32 int i;
33
34 might_sleep();
35 for (i = 0; i < (HPAGE_SIZE/PAGE_SIZE); i++) {
36 cond_resched();
37 clear_user_highpage(page + i, addr);
38 }
39}
40
41static void copy_huge_page(struct page *dst, struct page *src,
42 unsigned long addr)
43{
44 int i;
45
46 might_sleep();
47 for (i = 0; i < HPAGE_SIZE/PAGE_SIZE; i++) {
48 cond_resched();
49 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE);
50 }
51}
52
Eric Paris0bd0f9f2005-11-21 21:32:28 -080053/*
54 * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages
55 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static DEFINE_SPINLOCK(hugetlb_lock);
57
58static 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 Lameter5da7ca82006-01-06 00:10:46 -080066static struct page *dequeue_huge_page(struct vm_area_struct *vma,
67 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
69 int nid = numa_node_id();
70 struct page *page = NULL;
Christoph Lameter5da7ca82006-01-06 00:10:46 -080071 struct zonelist *zonelist = huge_zonelist(vma, address);
Christoph Lameter96df9332006-01-06 00:10:45 -080072 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
Christoph Lameter96df9332006-01-06 00:10:45 -080074 for (z = zonelist->zones; *z; z++) {
75 nid = (*z)->zone_pgdat->node_id;
Christoph Lameteraea47ff2006-01-08 01:00:57 -080076 if (cpuset_zone_allowed(*z, GFP_HIGHUSER) &&
77 !list_empty(&hugepage_freelists[nid]))
Christoph Lameter96df9332006-01-06 00:10:45 -080078 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 }
Christoph Lameter96df9332006-01-06 00:10:45 -080080
81 if (*z) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 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
Nick Piggina4822892006-03-22 00:08:08 -080091static int alloc_fresh_huge_page(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 static int nid = 0;
94 struct page *page;
95 page = alloc_pages_node(nid, GFP_HIGHUSER|__GFP_COMP|__GFP_NOWARN,
96 HUGETLB_PAGE_ORDER);
97 nid = (nid + 1) % num_online_nodes();
98 if (page) {
Nick Piggina4822892006-03-22 00:08:08 -080099 page[1].lru.next = (void *)free_huge_page; /* dtor */
Eric Paris0bd0f9f2005-11-21 21:32:28 -0800100 spin_lock(&hugetlb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 nr_huge_pages++;
102 nr_huge_pages_node[page_to_nid(page)]++;
Eric Paris0bd0f9f2005-11-21 21:32:28 -0800103 spin_unlock(&hugetlb_lock);
Nick Piggina4822892006-03-22 00:08:08 -0800104 put_page(page); /* free it into the hugepage allocator */
105 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 }
Nick Piggina4822892006-03-22 00:08:08 -0800107 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109
110void free_huge_page(struct page *page)
111{
112 BUG_ON(page_count(page));
113
114 INIT_LIST_HEAD(&page->lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 spin_lock(&hugetlb_lock);
117 enqueue_huge_page(page);
118 spin_unlock(&hugetlb_lock);
119}
120
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800121struct page *alloc_huge_page(struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125 spin_lock(&hugetlb_lock);
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800126 page = dequeue_huge_page(vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 if (!page) {
128 spin_unlock(&hugetlb_lock);
129 return NULL;
130 }
131 spin_unlock(&hugetlb_lock);
Nick Piggin7835e982006-03-22 00:08:40 -0800132 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 return page;
134}
135
136static int __init hugetlb_init(void)
137{
138 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100140 if (HPAGE_SHIFT == 0)
141 return 0;
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 for (i = 0; i < MAX_NUMNODES; ++i)
144 INIT_LIST_HEAD(&hugepage_freelists[i]);
145
146 for (i = 0; i < max_huge_pages; ++i) {
Nick Piggina4822892006-03-22 00:08:08 -0800147 if (!alloc_fresh_huge_page())
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 }
150 max_huge_pages = free_huge_pages = nr_huge_pages = i;
151 printk("Total HugeTLB memory allocated, %ld\n", free_huge_pages);
152 return 0;
153}
154module_init(hugetlb_init);
155
156static int __init hugetlb_setup(char *s)
157{
158 if (sscanf(s, "%lu", &max_huge_pages) <= 0)
159 max_huge_pages = 0;
160 return 1;
161}
162__setup("hugepages=", hugetlb_setup);
163
164#ifdef CONFIG_SYSCTL
165static void update_and_free_page(struct page *page)
166{
167 int i;
168 nr_huge_pages--;
169 nr_huge_pages_node[page_zone(page)->zone_pgdat->node_id]--;
170 for (i = 0; i < (HPAGE_SIZE / PAGE_SIZE); i++) {
171 page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 1 << PG_referenced |
172 1 << PG_dirty | 1 << PG_active | 1 << PG_reserved |
173 1 << PG_private | 1<< PG_writeback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 }
Nick Piggina4822892006-03-22 00:08:08 -0800175 page[1].lru.next = NULL;
Nick Piggin7835e982006-03-22 00:08:40 -0800176 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 __free_pages(page, HUGETLB_PAGE_ORDER);
178}
179
180#ifdef CONFIG_HIGHMEM
181static void try_to_free_low(unsigned long count)
182{
183 int i, nid;
184 for (i = 0; i < MAX_NUMNODES; ++i) {
185 struct page *page, *next;
186 list_for_each_entry_safe(page, next, &hugepage_freelists[i], lru) {
187 if (PageHighMem(page))
188 continue;
189 list_del(&page->lru);
190 update_and_free_page(page);
191 nid = page_zone(page)->zone_pgdat->node_id;
192 free_huge_pages--;
193 free_huge_pages_node[nid]--;
194 if (count >= nr_huge_pages)
195 return;
196 }
197 }
198}
199#else
200static inline void try_to_free_low(unsigned long count)
201{
202}
203#endif
204
205static unsigned long set_max_huge_pages(unsigned long count)
206{
207 while (count > nr_huge_pages) {
Nick Piggina4822892006-03-22 00:08:08 -0800208 if (!alloc_fresh_huge_page())
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 return nr_huge_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211 if (count >= nr_huge_pages)
212 return nr_huge_pages;
213
214 spin_lock(&hugetlb_lock);
215 try_to_free_low(count);
216 while (count < nr_huge_pages) {
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800217 struct page *page = dequeue_huge_page(NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 if (!page)
219 break;
220 update_and_free_page(page);
221 }
222 spin_unlock(&hugetlb_lock);
223 return nr_huge_pages;
224}
225
226int hugetlb_sysctl_handler(struct ctl_table *table, int write,
227 struct file *file, void __user *buffer,
228 size_t *length, loff_t *ppos)
229{
230 proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
231 max_huge_pages = set_max_huge_pages(max_huge_pages);
232 return 0;
233}
234#endif /* CONFIG_SYSCTL */
235
236int hugetlb_report_meminfo(char *buf)
237{
238 return sprintf(buf,
239 "HugePages_Total: %5lu\n"
240 "HugePages_Free: %5lu\n"
241 "Hugepagesize: %5lu kB\n",
242 nr_huge_pages,
243 free_huge_pages,
244 HPAGE_SIZE/1024);
245}
246
247int hugetlb_report_node_meminfo(int nid, char *buf)
248{
249 return sprintf(buf,
250 "Node %d HugePages_Total: %5u\n"
251 "Node %d HugePages_Free: %5u\n",
252 nid, nr_huge_pages_node[nid],
253 nid, free_huge_pages_node[nid]);
254}
255
256int is_hugepage_mem_enough(size_t size)
257{
258 return (size + ~HPAGE_MASK)/HPAGE_SIZE <= free_huge_pages;
259}
260
261/* Return the number pages of memory we physically have, in PAGE_SIZE units. */
262unsigned long hugetlb_total_pages(void)
263{
264 return nr_huge_pages * (HPAGE_SIZE / PAGE_SIZE);
265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
267/*
268 * We cannot handle pagefaults against hugetlb pages at all. They cause
269 * handle_mm_fault() to try to instantiate regular-sized pages in the
270 * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get
271 * this far.
272 */
273static struct page *hugetlb_nopage(struct vm_area_struct *vma,
274 unsigned long address, int *unused)
275{
276 BUG();
277 return NULL;
278}
279
280struct vm_operations_struct hugetlb_vm_ops = {
281 .nopage = hugetlb_nopage,
282};
283
David Gibson1e8f8892006-01-06 00:10:44 -0800284static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page,
285 int writable)
David Gibson63551ae2005-06-21 17:14:44 -0700286{
287 pte_t entry;
288
David Gibson1e8f8892006-01-06 00:10:44 -0800289 if (writable) {
David Gibson63551ae2005-06-21 17:14:44 -0700290 entry =
291 pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
292 } else {
293 entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot));
294 }
295 entry = pte_mkyoung(entry);
296 entry = pte_mkhuge(entry);
297
298 return entry;
299}
300
David Gibson1e8f8892006-01-06 00:10:44 -0800301static void set_huge_ptep_writable(struct vm_area_struct *vma,
302 unsigned long address, pte_t *ptep)
303{
304 pte_t entry;
305
306 entry = pte_mkwrite(pte_mkdirty(*ptep));
307 ptep_set_access_flags(vma, address, ptep, entry, 1);
308 update_mmu_cache(vma, address, entry);
309 lazy_mmu_prot_update(entry);
310}
311
312
David Gibson63551ae2005-06-21 17:14:44 -0700313int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
314 struct vm_area_struct *vma)
315{
316 pte_t *src_pte, *dst_pte, entry;
317 struct page *ptepage;
Hugh Dickins1c598272005-10-19 21:23:43 -0700318 unsigned long addr;
David Gibson1e8f8892006-01-06 00:10:44 -0800319 int cow;
320
321 cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
David Gibson63551ae2005-06-21 17:14:44 -0700322
Hugh Dickins1c598272005-10-19 21:23:43 -0700323 for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) {
Hugh Dickinsc74df322005-10-29 18:16:23 -0700324 src_pte = huge_pte_offset(src, addr);
325 if (!src_pte)
326 continue;
David Gibson63551ae2005-06-21 17:14:44 -0700327 dst_pte = huge_pte_alloc(dst, addr);
328 if (!dst_pte)
329 goto nomem;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700330 spin_lock(&dst->page_table_lock);
Hugh Dickins1c598272005-10-19 21:23:43 -0700331 spin_lock(&src->page_table_lock);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700332 if (!pte_none(*src_pte)) {
David Gibson1e8f8892006-01-06 00:10:44 -0800333 if (cow)
334 ptep_set_wrprotect(src, addr, src_pte);
Hugh Dickins1c598272005-10-19 21:23:43 -0700335 entry = *src_pte;
336 ptepage = pte_page(entry);
337 get_page(ptepage);
Hugh Dickins42946212005-10-29 18:16:05 -0700338 add_mm_counter(dst, file_rss, HPAGE_SIZE / PAGE_SIZE);
Hugh Dickins1c598272005-10-19 21:23:43 -0700339 set_huge_pte_at(dst, addr, dst_pte, entry);
340 }
341 spin_unlock(&src->page_table_lock);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700342 spin_unlock(&dst->page_table_lock);
David Gibson63551ae2005-06-21 17:14:44 -0700343 }
344 return 0;
345
346nomem:
347 return -ENOMEM;
348}
349
350void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start,
351 unsigned long end)
352{
353 struct mm_struct *mm = vma->vm_mm;
354 unsigned long address;
David Gibsonc7546f82005-08-05 11:59:35 -0700355 pte_t *ptep;
David Gibson63551ae2005-06-21 17:14:44 -0700356 pte_t pte;
357 struct page *page;
358
359 WARN_ON(!is_vm_hugetlb_page(vma));
360 BUG_ON(start & ~HPAGE_MASK);
361 BUG_ON(end & ~HPAGE_MASK);
362
Hugh Dickins508034a2005-10-29 18:16:30 -0700363 spin_lock(&mm->page_table_lock);
364
Hugh Dickins365e9c872005-10-29 18:16:18 -0700365 /* Update high watermark before we lower rss */
366 update_hiwater_rss(mm);
367
David Gibson63551ae2005-06-21 17:14:44 -0700368 for (address = start; address < end; address += HPAGE_SIZE) {
David Gibsonc7546f82005-08-05 11:59:35 -0700369 ptep = huge_pte_offset(mm, address);
Adam Litke4c887262005-10-29 18:16:46 -0700370 if (!ptep)
David Gibsonc7546f82005-08-05 11:59:35 -0700371 continue;
372
373 pte = huge_ptep_get_and_clear(mm, address, ptep);
David Gibson63551ae2005-06-21 17:14:44 -0700374 if (pte_none(pte))
375 continue;
David Gibsonc7546f82005-08-05 11:59:35 -0700376
David Gibson63551ae2005-06-21 17:14:44 -0700377 page = pte_page(pte);
378 put_page(page);
Hugh Dickins42946212005-10-29 18:16:05 -0700379 add_mm_counter(mm, file_rss, (int) -(HPAGE_SIZE / PAGE_SIZE));
David Gibson63551ae2005-06-21 17:14:44 -0700380 }
David Gibson63551ae2005-06-21 17:14:44 -0700381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 spin_unlock(&mm->page_table_lock);
Hugh Dickins508034a2005-10-29 18:16:30 -0700383 flush_tlb_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
David Gibson63551ae2005-06-21 17:14:44 -0700385
David Gibson1e8f8892006-01-06 00:10:44 -0800386static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma,
387 unsigned long address, pte_t *ptep, pte_t pte)
388{
389 struct page *old_page, *new_page;
David Gibson79ac6ba2006-03-22 00:08:51 -0800390 int avoidcopy;
David Gibson1e8f8892006-01-06 00:10:44 -0800391
392 old_page = pte_page(pte);
393
394 /* If no-one else is actually using this page, avoid the copy
395 * and just make the page writable */
396 avoidcopy = (page_count(old_page) == 1);
397 if (avoidcopy) {
398 set_huge_ptep_writable(vma, address, ptep);
399 return VM_FAULT_MINOR;
400 }
401
402 page_cache_get(old_page);
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800403 new_page = alloc_huge_page(vma, address);
David Gibson1e8f8892006-01-06 00:10:44 -0800404
405 if (!new_page) {
406 page_cache_release(old_page);
Christoph Lameter0df420d2006-02-07 12:58:30 -0800407 return VM_FAULT_OOM;
David Gibson1e8f8892006-01-06 00:10:44 -0800408 }
409
410 spin_unlock(&mm->page_table_lock);
David Gibson79ac6ba2006-03-22 00:08:51 -0800411 copy_huge_page(new_page, old_page, address);
David Gibson1e8f8892006-01-06 00:10:44 -0800412 spin_lock(&mm->page_table_lock);
413
414 ptep = huge_pte_offset(mm, address & HPAGE_MASK);
415 if (likely(pte_same(*ptep, pte))) {
416 /* Break COW */
417 set_huge_pte_at(mm, address, ptep,
418 make_huge_pte(vma, new_page, 1));
419 /* Make the old page be freed below */
420 new_page = old_page;
421 }
422 page_cache_release(new_page);
423 page_cache_release(old_page);
424 return VM_FAULT_MINOR;
425}
426
Adam Litke86e52162006-01-06 00:10:43 -0800427int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
David Gibson1e8f8892006-01-06 00:10:44 -0800428 unsigned long address, pte_t *ptep, int write_access)
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100429{
430 int ret = VM_FAULT_SIGBUS;
Adam Litke4c887262005-10-29 18:16:46 -0700431 unsigned long idx;
432 unsigned long size;
Adam Litke4c887262005-10-29 18:16:46 -0700433 struct page *page;
434 struct address_space *mapping;
David Gibson1e8f8892006-01-06 00:10:44 -0800435 pte_t new_pte;
Adam Litke4c887262005-10-29 18:16:46 -0700436
Adam Litke4c887262005-10-29 18:16:46 -0700437 mapping = vma->vm_file->f_mapping;
438 idx = ((address - vma->vm_start) >> HPAGE_SHIFT)
439 + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT));
440
441 /*
442 * Use page lock to guard against racing truncation
443 * before we get page_table_lock.
444 */
Christoph Lameter6bda6662006-01-06 00:10:49 -0800445retry:
446 page = find_lock_page(mapping, idx);
447 if (!page) {
448 if (hugetlb_get_quota(mapping))
449 goto out;
450 page = alloc_huge_page(vma, address);
451 if (!page) {
452 hugetlb_put_quota(mapping);
Christoph Lameter0df420d2006-02-07 12:58:30 -0800453 ret = VM_FAULT_OOM;
Christoph Lameter6bda6662006-01-06 00:10:49 -0800454 goto out;
455 }
David Gibson79ac6ba2006-03-22 00:08:51 -0800456 clear_huge_page(page, address);
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100457
Christoph Lameter6bda6662006-01-06 00:10:49 -0800458 if (vma->vm_flags & VM_SHARED) {
459 int err;
460
461 err = add_to_page_cache(page, mapping, idx, GFP_KERNEL);
462 if (err) {
463 put_page(page);
464 hugetlb_put_quota(mapping);
465 if (err == -EEXIST)
466 goto retry;
467 goto out;
468 }
469 } else
470 lock_page(page);
471 }
David Gibson1e8f8892006-01-06 00:10:44 -0800472
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100473 spin_lock(&mm->page_table_lock);
Adam Litke4c887262005-10-29 18:16:46 -0700474 size = i_size_read(mapping->host) >> HPAGE_SHIFT;
475 if (idx >= size)
476 goto backout;
477
478 ret = VM_FAULT_MINOR;
Adam Litke86e52162006-01-06 00:10:43 -0800479 if (!pte_none(*ptep))
Adam Litke4c887262005-10-29 18:16:46 -0700480 goto backout;
481
482 add_mm_counter(mm, file_rss, HPAGE_SIZE / PAGE_SIZE);
David Gibson1e8f8892006-01-06 00:10:44 -0800483 new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE)
484 && (vma->vm_flags & VM_SHARED)));
485 set_huge_pte_at(mm, address, ptep, new_pte);
486
487 if (write_access && !(vma->vm_flags & VM_SHARED)) {
488 /* Optimization, do the COW without a second fault */
489 ret = hugetlb_cow(mm, vma, address, ptep, new_pte);
490 }
491
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100492 spin_unlock(&mm->page_table_lock);
Adam Litke4c887262005-10-29 18:16:46 -0700493 unlock_page(page);
494out:
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100495 return ret;
Adam Litke4c887262005-10-29 18:16:46 -0700496
497backout:
498 spin_unlock(&mm->page_table_lock);
499 hugetlb_put_quota(mapping);
500 unlock_page(page);
501 put_page(page);
502 goto out;
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100503}
504
Adam Litke86e52162006-01-06 00:10:43 -0800505int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
506 unsigned long address, int write_access)
507{
508 pte_t *ptep;
509 pte_t entry;
David Gibson1e8f8892006-01-06 00:10:44 -0800510 int ret;
Adam Litke86e52162006-01-06 00:10:43 -0800511
512 ptep = huge_pte_alloc(mm, address);
513 if (!ptep)
514 return VM_FAULT_OOM;
515
516 entry = *ptep;
517 if (pte_none(entry))
David Gibson1e8f8892006-01-06 00:10:44 -0800518 return hugetlb_no_page(mm, vma, address, ptep, write_access);
Adam Litke86e52162006-01-06 00:10:43 -0800519
David Gibson1e8f8892006-01-06 00:10:44 -0800520 ret = VM_FAULT_MINOR;
521
522 spin_lock(&mm->page_table_lock);
523 /* Check for a racing update before calling hugetlb_cow */
524 if (likely(pte_same(entry, *ptep)))
525 if (write_access && !pte_write(entry))
526 ret = hugetlb_cow(mm, vma, address, ptep, entry);
527 spin_unlock(&mm->page_table_lock);
528
529 return ret;
Adam Litke86e52162006-01-06 00:10:43 -0800530}
531
David Gibson63551ae2005-06-21 17:14:44 -0700532int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
533 struct page **pages, struct vm_area_struct **vmas,
534 unsigned long *position, int *length, int i)
535{
536 unsigned long vpfn, vaddr = *position;
537 int remainder = *length;
538
David Gibson63551ae2005-06-21 17:14:44 -0700539 vpfn = vaddr/PAGE_SIZE;
Hugh Dickins1c598272005-10-19 21:23:43 -0700540 spin_lock(&mm->page_table_lock);
David Gibson63551ae2005-06-21 17:14:44 -0700541 while (vaddr < vma->vm_end && remainder) {
Adam Litke4c887262005-10-29 18:16:46 -0700542 pte_t *pte;
543 struct page *page;
544
545 /*
546 * Some archs (sparc64, sh*) have multiple pte_ts to
547 * each hugepage. We have to make * sure we get the
548 * first, for the page indexing below to work.
549 */
550 pte = huge_pte_offset(mm, vaddr & HPAGE_MASK);
551
552 if (!pte || pte_none(*pte)) {
553 int ret;
554
555 spin_unlock(&mm->page_table_lock);
556 ret = hugetlb_fault(mm, vma, vaddr, 0);
557 spin_lock(&mm->page_table_lock);
558 if (ret == VM_FAULT_MINOR)
559 continue;
560
561 remainder = 0;
562 if (!i)
563 i = -EFAULT;
564 break;
565 }
David Gibson63551ae2005-06-21 17:14:44 -0700566
567 if (pages) {
David Gibson63551ae2005-06-21 17:14:44 -0700568 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
David Gibson63551ae2005-06-21 17:14:44 -0700569 get_page(page);
570 pages[i] = page;
571 }
572
573 if (vmas)
574 vmas[i] = vma;
575
576 vaddr += PAGE_SIZE;
577 ++vpfn;
578 --remainder;
579 ++i;
580 }
Hugh Dickins1c598272005-10-19 21:23:43 -0700581 spin_unlock(&mm->page_table_lock);
David Gibson63551ae2005-06-21 17:14:44 -0700582 *length = remainder;
583 *position = vaddr;
584
585 return i;
586}
Zhang, Yanmin8f860592006-03-22 00:08:50 -0800587
588void hugetlb_change_protection(struct vm_area_struct *vma,
589 unsigned long address, unsigned long end, pgprot_t newprot)
590{
591 struct mm_struct *mm = vma->vm_mm;
592 unsigned long start = address;
593 pte_t *ptep;
594 pte_t pte;
595
596 BUG_ON(address >= end);
597 flush_cache_range(vma, address, end);
598
599 spin_lock(&mm->page_table_lock);
600 for (; address < end; address += HPAGE_SIZE) {
601 ptep = huge_pte_offset(mm, address);
602 if (!ptep)
603 continue;
604 if (!pte_none(*ptep)) {
605 pte = huge_ptep_get_and_clear(mm, address, ptep);
606 pte = pte_mkhuge(pte_modify(pte, newprot));
607 set_huge_pte_at(mm, address, ptep, pte);
608 lazy_mmu_prot_update(pte);
609 }
610 }
611 spin_unlock(&mm->page_table_lock);
612
613 flush_tlb_range(vma, start, end);
614}
615