blob: 508707704d2cb714968a779e128084d640afc439 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL;
23static unsigned long nr_huge_pages, free_huge_pages;
24unsigned long max_huge_pages;
25static struct list_head hugepage_freelists[MAX_NUMNODES];
26static unsigned int nr_huge_pages_node[MAX_NUMNODES];
27static unsigned int free_huge_pages_node[MAX_NUMNODES];
Eric Paris0bd0f9f2005-11-21 21:32:28 -080028
29/*
30 * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages
31 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070032static DEFINE_SPINLOCK(hugetlb_lock);
33
34static void enqueue_huge_page(struct page *page)
35{
36 int nid = page_to_nid(page);
37 list_add(&page->lru, &hugepage_freelists[nid]);
38 free_huge_pages++;
39 free_huge_pages_node[nid]++;
40}
41
Christoph Lameter5da7ca82006-01-06 00:10:46 -080042static struct page *dequeue_huge_page(struct vm_area_struct *vma,
43 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
45 int nid = numa_node_id();
46 struct page *page = NULL;
Christoph Lameter5da7ca82006-01-06 00:10:46 -080047 struct zonelist *zonelist = huge_zonelist(vma, address);
Christoph Lameter96df9332006-01-06 00:10:45 -080048 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Christoph Lameter96df9332006-01-06 00:10:45 -080050 for (z = zonelist->zones; *z; z++) {
51 nid = (*z)->zone_pgdat->node_id;
Christoph Lameteraea47ff2006-01-08 01:00:57 -080052 if (cpuset_zone_allowed(*z, GFP_HIGHUSER) &&
53 !list_empty(&hugepage_freelists[nid]))
Christoph Lameter96df9332006-01-06 00:10:45 -080054 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 }
Christoph Lameter96df9332006-01-06 00:10:45 -080056
57 if (*z) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 page = list_entry(hugepage_freelists[nid].next,
59 struct page, lru);
60 list_del(&page->lru);
61 free_huge_pages--;
62 free_huge_pages_node[nid]--;
63 }
64 return page;
65}
66
67static struct page *alloc_fresh_huge_page(void)
68{
69 static int nid = 0;
70 struct page *page;
71 page = alloc_pages_node(nid, GFP_HIGHUSER|__GFP_COMP|__GFP_NOWARN,
72 HUGETLB_PAGE_ORDER);
73 nid = (nid + 1) % num_online_nodes();
74 if (page) {
Eric Paris0bd0f9f2005-11-21 21:32:28 -080075 spin_lock(&hugetlb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 nr_huge_pages++;
77 nr_huge_pages_node[page_to_nid(page)]++;
Eric Paris0bd0f9f2005-11-21 21:32:28 -080078 spin_unlock(&hugetlb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 }
80 return page;
81}
82
83void free_huge_page(struct page *page)
84{
85 BUG_ON(page_count(page));
86
87 INIT_LIST_HEAD(&page->lru);
Hugh Dickins41d78ba2006-02-14 13:52:58 -080088 page[1].lru.next = NULL; /* reset dtor */
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90 spin_lock(&hugetlb_lock);
91 enqueue_huge_page(page);
92 spin_unlock(&hugetlb_lock);
93}
94
Christoph Lameter5da7ca82006-01-06 00:10:46 -080095struct page *alloc_huge_page(struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
97 struct page *page;
98 int i;
99
100 spin_lock(&hugetlb_lock);
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800101 page = dequeue_huge_page(vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 if (!page) {
103 spin_unlock(&hugetlb_lock);
104 return NULL;
105 }
106 spin_unlock(&hugetlb_lock);
107 set_page_count(page, 1);
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800108 page[1].lru.next = (void *)free_huge_page; /* set dtor */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 for (i = 0; i < (HPAGE_SIZE/PAGE_SIZE); ++i)
David Gibsona2dfef62006-02-07 12:58:25 -0800110 clear_user_highpage(&page[i], addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 return page;
112}
113
114static int __init hugetlb_init(void)
115{
116 unsigned long i;
117 struct page *page;
118
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100119 if (HPAGE_SHIFT == 0)
120 return 0;
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 for (i = 0; i < MAX_NUMNODES; ++i)
123 INIT_LIST_HEAD(&hugepage_freelists[i]);
124
125 for (i = 0; i < max_huge_pages; ++i) {
126 page = alloc_fresh_huge_page();
127 if (!page)
128 break;
129 spin_lock(&hugetlb_lock);
130 enqueue_huge_page(page);
131 spin_unlock(&hugetlb_lock);
132 }
133 max_huge_pages = free_huge_pages = nr_huge_pages = i;
134 printk("Total HugeTLB memory allocated, %ld\n", free_huge_pages);
135 return 0;
136}
137module_init(hugetlb_init);
138
139static int __init hugetlb_setup(char *s)
140{
141 if (sscanf(s, "%lu", &max_huge_pages) <= 0)
142 max_huge_pages = 0;
143 return 1;
144}
145__setup("hugepages=", hugetlb_setup);
146
147#ifdef CONFIG_SYSCTL
148static void update_and_free_page(struct page *page)
149{
150 int i;
151 nr_huge_pages--;
152 nr_huge_pages_node[page_zone(page)->zone_pgdat->node_id]--;
153 for (i = 0; i < (HPAGE_SIZE / PAGE_SIZE); i++) {
154 page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 1 << PG_referenced |
155 1 << PG_dirty | 1 << PG_active | 1 << PG_reserved |
156 1 << PG_private | 1<< PG_writeback);
157 set_page_count(&page[i], 0);
158 }
159 set_page_count(page, 1);
160 __free_pages(page, HUGETLB_PAGE_ORDER);
161}
162
163#ifdef CONFIG_HIGHMEM
164static void try_to_free_low(unsigned long count)
165{
166 int i, nid;
167 for (i = 0; i < MAX_NUMNODES; ++i) {
168 struct page *page, *next;
169 list_for_each_entry_safe(page, next, &hugepage_freelists[i], lru) {
170 if (PageHighMem(page))
171 continue;
172 list_del(&page->lru);
173 update_and_free_page(page);
174 nid = page_zone(page)->zone_pgdat->node_id;
175 free_huge_pages--;
176 free_huge_pages_node[nid]--;
177 if (count >= nr_huge_pages)
178 return;
179 }
180 }
181}
182#else
183static inline void try_to_free_low(unsigned long count)
184{
185}
186#endif
187
188static unsigned long set_max_huge_pages(unsigned long count)
189{
190 while (count > nr_huge_pages) {
191 struct page *page = alloc_fresh_huge_page();
192 if (!page)
193 return nr_huge_pages;
194 spin_lock(&hugetlb_lock);
195 enqueue_huge_page(page);
196 spin_unlock(&hugetlb_lock);
197 }
198 if (count >= nr_huge_pages)
199 return nr_huge_pages;
200
201 spin_lock(&hugetlb_lock);
202 try_to_free_low(count);
203 while (count < nr_huge_pages) {
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800204 struct page *page = dequeue_huge_page(NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if (!page)
206 break;
207 update_and_free_page(page);
208 }
209 spin_unlock(&hugetlb_lock);
210 return nr_huge_pages;
211}
212
213int hugetlb_sysctl_handler(struct ctl_table *table, int write,
214 struct file *file, void __user *buffer,
215 size_t *length, loff_t *ppos)
216{
217 proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
218 max_huge_pages = set_max_huge_pages(max_huge_pages);
219 return 0;
220}
221#endif /* CONFIG_SYSCTL */
222
223int hugetlb_report_meminfo(char *buf)
224{
225 return sprintf(buf,
226 "HugePages_Total: %5lu\n"
227 "HugePages_Free: %5lu\n"
228 "Hugepagesize: %5lu kB\n",
229 nr_huge_pages,
230 free_huge_pages,
231 HPAGE_SIZE/1024);
232}
233
234int hugetlb_report_node_meminfo(int nid, char *buf)
235{
236 return sprintf(buf,
237 "Node %d HugePages_Total: %5u\n"
238 "Node %d HugePages_Free: %5u\n",
239 nid, nr_huge_pages_node[nid],
240 nid, free_huge_pages_node[nid]);
241}
242
243int is_hugepage_mem_enough(size_t size)
244{
245 return (size + ~HPAGE_MASK)/HPAGE_SIZE <= free_huge_pages;
246}
247
248/* Return the number pages of memory we physically have, in PAGE_SIZE units. */
249unsigned long hugetlb_total_pages(void)
250{
251 return nr_huge_pages * (HPAGE_SIZE / PAGE_SIZE);
252}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
254/*
255 * We cannot handle pagefaults against hugetlb pages at all. They cause
256 * handle_mm_fault() to try to instantiate regular-sized pages in the
257 * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get
258 * this far.
259 */
260static struct page *hugetlb_nopage(struct vm_area_struct *vma,
261 unsigned long address, int *unused)
262{
263 BUG();
264 return NULL;
265}
266
267struct vm_operations_struct hugetlb_vm_ops = {
268 .nopage = hugetlb_nopage,
269};
270
David Gibson1e8f8892006-01-06 00:10:44 -0800271static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page,
272 int writable)
David Gibson63551ae2005-06-21 17:14:44 -0700273{
274 pte_t entry;
275
David Gibson1e8f8892006-01-06 00:10:44 -0800276 if (writable) {
David Gibson63551ae2005-06-21 17:14:44 -0700277 entry =
278 pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
279 } else {
280 entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot));
281 }
282 entry = pte_mkyoung(entry);
283 entry = pte_mkhuge(entry);
284
285 return entry;
286}
287
David Gibson1e8f8892006-01-06 00:10:44 -0800288static void set_huge_ptep_writable(struct vm_area_struct *vma,
289 unsigned long address, pte_t *ptep)
290{
291 pte_t entry;
292
293 entry = pte_mkwrite(pte_mkdirty(*ptep));
294 ptep_set_access_flags(vma, address, ptep, entry, 1);
295 update_mmu_cache(vma, address, entry);
296 lazy_mmu_prot_update(entry);
297}
298
299
David Gibson63551ae2005-06-21 17:14:44 -0700300int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
301 struct vm_area_struct *vma)
302{
303 pte_t *src_pte, *dst_pte, entry;
304 struct page *ptepage;
Hugh Dickins1c598272005-10-19 21:23:43 -0700305 unsigned long addr;
David Gibson1e8f8892006-01-06 00:10:44 -0800306 int cow;
307
308 cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
David Gibson63551ae2005-06-21 17:14:44 -0700309
Hugh Dickins1c598272005-10-19 21:23:43 -0700310 for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) {
Hugh Dickinsc74df322005-10-29 18:16:23 -0700311 src_pte = huge_pte_offset(src, addr);
312 if (!src_pte)
313 continue;
David Gibson63551ae2005-06-21 17:14:44 -0700314 dst_pte = huge_pte_alloc(dst, addr);
315 if (!dst_pte)
316 goto nomem;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700317 spin_lock(&dst->page_table_lock);
Hugh Dickins1c598272005-10-19 21:23:43 -0700318 spin_lock(&src->page_table_lock);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700319 if (!pte_none(*src_pte)) {
David Gibson1e8f8892006-01-06 00:10:44 -0800320 if (cow)
321 ptep_set_wrprotect(src, addr, src_pte);
Hugh Dickins1c598272005-10-19 21:23:43 -0700322 entry = *src_pte;
323 ptepage = pte_page(entry);
324 get_page(ptepage);
Hugh Dickins42946212005-10-29 18:16:05 -0700325 add_mm_counter(dst, file_rss, HPAGE_SIZE / PAGE_SIZE);
Hugh Dickins1c598272005-10-19 21:23:43 -0700326 set_huge_pte_at(dst, addr, dst_pte, entry);
327 }
328 spin_unlock(&src->page_table_lock);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700329 spin_unlock(&dst->page_table_lock);
David Gibson63551ae2005-06-21 17:14:44 -0700330 }
331 return 0;
332
333nomem:
334 return -ENOMEM;
335}
336
337void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start,
338 unsigned long end)
339{
340 struct mm_struct *mm = vma->vm_mm;
341 unsigned long address;
David Gibsonc7546f82005-08-05 11:59:35 -0700342 pte_t *ptep;
David Gibson63551ae2005-06-21 17:14:44 -0700343 pte_t pte;
344 struct page *page;
345
346 WARN_ON(!is_vm_hugetlb_page(vma));
347 BUG_ON(start & ~HPAGE_MASK);
348 BUG_ON(end & ~HPAGE_MASK);
349
Hugh Dickins508034a2005-10-29 18:16:30 -0700350 spin_lock(&mm->page_table_lock);
351
Hugh Dickins365e9c872005-10-29 18:16:18 -0700352 /* Update high watermark before we lower rss */
353 update_hiwater_rss(mm);
354
David Gibson63551ae2005-06-21 17:14:44 -0700355 for (address = start; address < end; address += HPAGE_SIZE) {
David Gibsonc7546f82005-08-05 11:59:35 -0700356 ptep = huge_pte_offset(mm, address);
Adam Litke4c887262005-10-29 18:16:46 -0700357 if (!ptep)
David Gibsonc7546f82005-08-05 11:59:35 -0700358 continue;
359
360 pte = huge_ptep_get_and_clear(mm, address, ptep);
David Gibson63551ae2005-06-21 17:14:44 -0700361 if (pte_none(pte))
362 continue;
David Gibsonc7546f82005-08-05 11:59:35 -0700363
David Gibson63551ae2005-06-21 17:14:44 -0700364 page = pte_page(pte);
365 put_page(page);
Hugh Dickins42946212005-10-29 18:16:05 -0700366 add_mm_counter(mm, file_rss, (int) -(HPAGE_SIZE / PAGE_SIZE));
David Gibson63551ae2005-06-21 17:14:44 -0700367 }
David Gibson63551ae2005-06-21 17:14:44 -0700368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 spin_unlock(&mm->page_table_lock);
Hugh Dickins508034a2005-10-29 18:16:30 -0700370 flush_tlb_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
David Gibson63551ae2005-06-21 17:14:44 -0700372
David Gibson1e8f8892006-01-06 00:10:44 -0800373static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma,
374 unsigned long address, pte_t *ptep, pte_t pte)
375{
376 struct page *old_page, *new_page;
377 int i, avoidcopy;
378
379 old_page = pte_page(pte);
380
381 /* If no-one else is actually using this page, avoid the copy
382 * and just make the page writable */
383 avoidcopy = (page_count(old_page) == 1);
384 if (avoidcopy) {
385 set_huge_ptep_writable(vma, address, ptep);
386 return VM_FAULT_MINOR;
387 }
388
389 page_cache_get(old_page);
Christoph Lameter5da7ca82006-01-06 00:10:46 -0800390 new_page = alloc_huge_page(vma, address);
David Gibson1e8f8892006-01-06 00:10:44 -0800391
392 if (!new_page) {
393 page_cache_release(old_page);
Christoph Lameter0df420d2006-02-07 12:58:30 -0800394 return VM_FAULT_OOM;
David Gibson1e8f8892006-01-06 00:10:44 -0800395 }
396
397 spin_unlock(&mm->page_table_lock);
398 for (i = 0; i < HPAGE_SIZE/PAGE_SIZE; i++)
399 copy_user_highpage(new_page + i, old_page + i,
400 address + i*PAGE_SIZE);
401 spin_lock(&mm->page_table_lock);
402
403 ptep = huge_pte_offset(mm, address & HPAGE_MASK);
404 if (likely(pte_same(*ptep, pte))) {
405 /* Break COW */
406 set_huge_pte_at(mm, address, ptep,
407 make_huge_pte(vma, new_page, 1));
408 /* Make the old page be freed below */
409 new_page = old_page;
410 }
411 page_cache_release(new_page);
412 page_cache_release(old_page);
413 return VM_FAULT_MINOR;
414}
415
Adam Litke86e52162006-01-06 00:10:43 -0800416int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
David Gibson1e8f8892006-01-06 00:10:44 -0800417 unsigned long address, pte_t *ptep, int write_access)
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100418{
419 int ret = VM_FAULT_SIGBUS;
Adam Litke4c887262005-10-29 18:16:46 -0700420 unsigned long idx;
421 unsigned long size;
Adam Litke4c887262005-10-29 18:16:46 -0700422 struct page *page;
423 struct address_space *mapping;
David Gibson1e8f8892006-01-06 00:10:44 -0800424 pte_t new_pte;
Adam Litke4c887262005-10-29 18:16:46 -0700425
Adam Litke4c887262005-10-29 18:16:46 -0700426 mapping = vma->vm_file->f_mapping;
427 idx = ((address - vma->vm_start) >> HPAGE_SHIFT)
428 + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT));
429
430 /*
431 * Use page lock to guard against racing truncation
432 * before we get page_table_lock.
433 */
Christoph Lameter6bda6662006-01-06 00:10:49 -0800434retry:
435 page = find_lock_page(mapping, idx);
436 if (!page) {
437 if (hugetlb_get_quota(mapping))
438 goto out;
439 page = alloc_huge_page(vma, address);
440 if (!page) {
441 hugetlb_put_quota(mapping);
Christoph Lameter0df420d2006-02-07 12:58:30 -0800442 ret = VM_FAULT_OOM;
Christoph Lameter6bda6662006-01-06 00:10:49 -0800443 goto out;
444 }
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100445
Christoph Lameter6bda6662006-01-06 00:10:49 -0800446 if (vma->vm_flags & VM_SHARED) {
447 int err;
448
449 err = add_to_page_cache(page, mapping, idx, GFP_KERNEL);
450 if (err) {
451 put_page(page);
452 hugetlb_put_quota(mapping);
453 if (err == -EEXIST)
454 goto retry;
455 goto out;
456 }
457 } else
458 lock_page(page);
459 }
David Gibson1e8f8892006-01-06 00:10:44 -0800460
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100461 spin_lock(&mm->page_table_lock);
Adam Litke4c887262005-10-29 18:16:46 -0700462 size = i_size_read(mapping->host) >> HPAGE_SHIFT;
463 if (idx >= size)
464 goto backout;
465
466 ret = VM_FAULT_MINOR;
Adam Litke86e52162006-01-06 00:10:43 -0800467 if (!pte_none(*ptep))
Adam Litke4c887262005-10-29 18:16:46 -0700468 goto backout;
469
470 add_mm_counter(mm, file_rss, HPAGE_SIZE / PAGE_SIZE);
David Gibson1e8f8892006-01-06 00:10:44 -0800471 new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE)
472 && (vma->vm_flags & VM_SHARED)));
473 set_huge_pte_at(mm, address, ptep, new_pte);
474
475 if (write_access && !(vma->vm_flags & VM_SHARED)) {
476 /* Optimization, do the COW without a second fault */
477 ret = hugetlb_cow(mm, vma, address, ptep, new_pte);
478 }
479
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100480 spin_unlock(&mm->page_table_lock);
Adam Litke4c887262005-10-29 18:16:46 -0700481 unlock_page(page);
482out:
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100483 return ret;
Adam Litke4c887262005-10-29 18:16:46 -0700484
485backout:
486 spin_unlock(&mm->page_table_lock);
487 hugetlb_put_quota(mapping);
488 unlock_page(page);
489 put_page(page);
490 goto out;
Hugh Dickinsac9b9c62005-10-20 16:24:28 +0100491}
492
Adam Litke86e52162006-01-06 00:10:43 -0800493int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
494 unsigned long address, int write_access)
495{
496 pte_t *ptep;
497 pte_t entry;
David Gibson1e8f8892006-01-06 00:10:44 -0800498 int ret;
Adam Litke86e52162006-01-06 00:10:43 -0800499
500 ptep = huge_pte_alloc(mm, address);
501 if (!ptep)
502 return VM_FAULT_OOM;
503
504 entry = *ptep;
505 if (pte_none(entry))
David Gibson1e8f8892006-01-06 00:10:44 -0800506 return hugetlb_no_page(mm, vma, address, ptep, write_access);
Adam Litke86e52162006-01-06 00:10:43 -0800507
David Gibson1e8f8892006-01-06 00:10:44 -0800508 ret = VM_FAULT_MINOR;
509
510 spin_lock(&mm->page_table_lock);
511 /* Check for a racing update before calling hugetlb_cow */
512 if (likely(pte_same(entry, *ptep)))
513 if (write_access && !pte_write(entry))
514 ret = hugetlb_cow(mm, vma, address, ptep, entry);
515 spin_unlock(&mm->page_table_lock);
516
517 return ret;
Adam Litke86e52162006-01-06 00:10:43 -0800518}
519
David Gibson63551ae2005-06-21 17:14:44 -0700520int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
521 struct page **pages, struct vm_area_struct **vmas,
522 unsigned long *position, int *length, int i)
523{
524 unsigned long vpfn, vaddr = *position;
525 int remainder = *length;
526
David Gibson63551ae2005-06-21 17:14:44 -0700527 vpfn = vaddr/PAGE_SIZE;
Hugh Dickins1c598272005-10-19 21:23:43 -0700528 spin_lock(&mm->page_table_lock);
David Gibson63551ae2005-06-21 17:14:44 -0700529 while (vaddr < vma->vm_end && remainder) {
Adam Litke4c887262005-10-29 18:16:46 -0700530 pte_t *pte;
531 struct page *page;
532
533 /*
534 * Some archs (sparc64, sh*) have multiple pte_ts to
535 * each hugepage. We have to make * sure we get the
536 * first, for the page indexing below to work.
537 */
538 pte = huge_pte_offset(mm, vaddr & HPAGE_MASK);
539
540 if (!pte || pte_none(*pte)) {
541 int ret;
542
543 spin_unlock(&mm->page_table_lock);
544 ret = hugetlb_fault(mm, vma, vaddr, 0);
545 spin_lock(&mm->page_table_lock);
546 if (ret == VM_FAULT_MINOR)
547 continue;
548
549 remainder = 0;
550 if (!i)
551 i = -EFAULT;
552 break;
553 }
David Gibson63551ae2005-06-21 17:14:44 -0700554
555 if (pages) {
David Gibson63551ae2005-06-21 17:14:44 -0700556 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
David Gibson63551ae2005-06-21 17:14:44 -0700557 get_page(page);
558 pages[i] = page;
559 }
560
561 if (vmas)
562 vmas[i] = vma;
563
564 vaddr += PAGE_SIZE;
565 ++vpfn;
566 --remainder;
567 ++i;
568 }
Hugh Dickins1c598272005-10-19 21:23:43 -0700569 spin_unlock(&mm->page_table_lock);
David Gibson63551ae2005-06-21 17:14:44 -0700570 *length = remainder;
571 *position = vaddr;
572
573 return i;
574}