blob: 0536dde139d1e1f6f28fc93ccd05a9fc337e2c7e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/vmalloc.c
3 *
4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
Christoph Lameter930fc452005-10-29 18:15:41 -07008 * Numa awareness, Christoph Lameter, SGI, June 2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
11#include <linux/mm.h>
12#include <linux/module.h>
13#include <linux/highmem.h>
14#include <linux/slab.h>
15#include <linux/spinlock.h>
16#include <linux/interrupt.h>
17
18#include <linux/vmalloc.h>
19
20#include <asm/uaccess.h>
21#include <asm/tlbflush.h>
22
23
24DEFINE_RWLOCK(vmlist_lock);
25struct vm_struct *vmlist;
26
Adrian Bunkb2213852006-09-25 23:31:02 -070027static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
28 int node);
29
Linus Torvalds1da177e2005-04-16 15:20:36 -070030static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
31{
32 pte_t *pte;
33
34 pte = pte_offset_kernel(pmd, addr);
35 do {
36 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
37 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
38 } while (pte++, addr += PAGE_SIZE, addr != end);
39}
40
41static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
42 unsigned long end)
43{
44 pmd_t *pmd;
45 unsigned long next;
46
47 pmd = pmd_offset(pud, addr);
48 do {
49 next = pmd_addr_end(addr, end);
50 if (pmd_none_or_clear_bad(pmd))
51 continue;
52 vunmap_pte_range(pmd, addr, next);
53 } while (pmd++, addr = next, addr != end);
54}
55
56static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
57 unsigned long end)
58{
59 pud_t *pud;
60 unsigned long next;
61
62 pud = pud_offset(pgd, addr);
63 do {
64 next = pud_addr_end(addr, end);
65 if (pud_none_or_clear_bad(pud))
66 continue;
67 vunmap_pmd_range(pud, addr, next);
68 } while (pud++, addr = next, addr != end);
69}
70
Benjamin Herrenschmidtc19c03f2007-06-04 15:15:35 +100071void unmap_kernel_range(unsigned long addr, unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 pgd_t *pgd;
74 unsigned long next;
Benjamin Herrenschmidtc19c03f2007-06-04 15:15:35 +100075 unsigned long start = addr;
76 unsigned long end = addr + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 BUG_ON(addr >= end);
79 pgd = pgd_offset_k(addr);
80 flush_cache_vunmap(addr, end);
81 do {
82 next = pgd_addr_end(addr, end);
83 if (pgd_none_or_clear_bad(pgd))
84 continue;
85 vunmap_pud_range(pgd, addr, next);
86 } while (pgd++, addr = next, addr != end);
Benjamin Herrenschmidtc19c03f2007-06-04 15:15:35 +100087 flush_tlb_kernel_range(start, end);
88}
89
90static void unmap_vm_area(struct vm_struct *area)
91{
92 unmap_kernel_range((unsigned long)area->addr, area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093}
94
95static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
96 unsigned long end, pgprot_t prot, struct page ***pages)
97{
98 pte_t *pte;
99
Hugh Dickins872fec12005-10-29 18:16:21 -0700100 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 if (!pte)
102 return -ENOMEM;
103 do {
104 struct page *page = **pages;
105 WARN_ON(!pte_none(*pte));
106 if (!page)
107 return -ENOMEM;
108 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
109 (*pages)++;
110 } while (pte++, addr += PAGE_SIZE, addr != end);
111 return 0;
112}
113
114static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
115 unsigned long end, pgprot_t prot, struct page ***pages)
116{
117 pmd_t *pmd;
118 unsigned long next;
119
120 pmd = pmd_alloc(&init_mm, pud, addr);
121 if (!pmd)
122 return -ENOMEM;
123 do {
124 next = pmd_addr_end(addr, end);
125 if (vmap_pte_range(pmd, addr, next, prot, pages))
126 return -ENOMEM;
127 } while (pmd++, addr = next, addr != end);
128 return 0;
129}
130
131static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
132 unsigned long end, pgprot_t prot, struct page ***pages)
133{
134 pud_t *pud;
135 unsigned long next;
136
137 pud = pud_alloc(&init_mm, pgd, addr);
138 if (!pud)
139 return -ENOMEM;
140 do {
141 next = pud_addr_end(addr, end);
142 if (vmap_pmd_range(pud, addr, next, prot, pages))
143 return -ENOMEM;
144 } while (pud++, addr = next, addr != end);
145 return 0;
146}
147
148int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
149{
150 pgd_t *pgd;
151 unsigned long next;
152 unsigned long addr = (unsigned long) area->addr;
153 unsigned long end = addr + area->size - PAGE_SIZE;
154 int err;
155
156 BUG_ON(addr >= end);
157 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 do {
159 next = pgd_addr_end(addr, end);
160 err = vmap_pud_range(pgd, addr, next, prot, pages);
161 if (err)
162 break;
163 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 flush_cache_vmap((unsigned long) area->addr, end);
165 return err;
166}
Rusty Russell5992b6d2007-07-19 01:49:21 -0700167EXPORT_SYMBOL_GPL(map_vm_area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Christoph Lameter48667e72008-02-04 22:28:31 -0800169/*
170 * Map a vmalloc()-space virtual address to the physical page.
171 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800172struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800173{
174 unsigned long addr = (unsigned long) vmalloc_addr;
175 struct page *page = NULL;
176 pgd_t *pgd = pgd_offset_k(addr);
177 pud_t *pud;
178 pmd_t *pmd;
179 pte_t *ptep, pte;
180
181 if (!pgd_none(*pgd)) {
182 pud = pud_offset(pgd, addr);
183 if (!pud_none(*pud)) {
184 pmd = pmd_offset(pud, addr);
185 if (!pmd_none(*pmd)) {
186 ptep = pte_offset_map(pmd, addr);
187 pte = *ptep;
188 if (pte_present(pte))
189 page = pte_page(pte);
190 pte_unmap(ptep);
191 }
192 }
193 }
194 return page;
195}
196EXPORT_SYMBOL(vmalloc_to_page);
197
198/*
199 * Map a vmalloc()-space virtual address to the physical page frame number.
200 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800201unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800202{
203 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
204}
205EXPORT_SYMBOL(vmalloc_to_pfn);
206
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -0700207static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
208 unsigned long start, unsigned long end,
209 int node, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210{
211 struct vm_struct **p, *tmp, *area;
212 unsigned long align = 1;
213 unsigned long addr;
214
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -0700215 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (flags & VM_IOREMAP) {
217 int bit = fls(size);
218
219 if (bit > IOREMAP_MAX_ORDER)
220 bit = IOREMAP_MAX_ORDER;
221 else if (bit < PAGE_SHIFT)
222 bit = PAGE_SHIFT;
223
224 align = 1ul << bit;
225 }
226 addr = ALIGN(start, align);
227 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -0800228 if (unlikely(!size))
229 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Christoph Lameter6cb06222007-10-16 01:25:41 -0700231 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (unlikely(!area))
234 return NULL;
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 /*
237 * We always allocate a guard page.
238 */
239 size += PAGE_SIZE;
240
241 write_lock(&vmlist_lock);
242 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
243 if ((unsigned long)tmp->addr < addr) {
244 if((unsigned long)tmp->addr + tmp->size >= addr)
245 addr = ALIGN(tmp->size +
246 (unsigned long)tmp->addr, align);
247 continue;
248 }
249 if ((size + addr) < addr)
250 goto out;
251 if (size + addr <= (unsigned long)tmp->addr)
252 goto found;
253 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
254 if (addr > end - size)
255 goto out;
256 }
Robert Bragg5dc33182008-02-04 22:29:18 -0800257 if ((size + addr) < addr)
258 goto out;
259 if (addr > end - size)
260 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262found:
263 area->next = *p;
264 *p = area;
265
266 area->flags = flags;
267 area->addr = (void *)addr;
268 area->size = size;
269 area->pages = NULL;
270 area->nr_pages = 0;
271 area->phys_addr = 0;
272 write_unlock(&vmlist_lock);
273
274 return area;
275
276out:
277 write_unlock(&vmlist_lock);
278 kfree(area);
279 if (printk_ratelimit())
280 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
281 return NULL;
282}
283
Christoph Lameter930fc452005-10-29 18:15:41 -0700284struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
285 unsigned long start, unsigned long end)
286{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -0700287 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
Christoph Lameter930fc452005-10-29 18:15:41 -0700288}
Rusty Russell5992b6d2007-07-19 01:49:21 -0700289EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -0700290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291/**
Simon Arlott183ff222007-10-20 01:27:18 +0200292 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 * @size: size of the area
294 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
295 *
296 * Search an area of @size in the kernel virtual mapping area,
297 * and reserved it for out purposes. Returns the area descriptor
298 * on success or %NULL on failure.
299 */
300struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
301{
302 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
303}
304
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -0700305struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
306 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -0700307{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -0700308 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
309 gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -0700310}
311
Andi Kleen7856dfe2005-05-20 14:27:57 -0700312/* Caller must hold vmlist_lock */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800313static struct vm_struct *__find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -0700314{
315 struct vm_struct *tmp;
316
317 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
318 if (tmp->addr == addr)
319 break;
320 }
321
322 return tmp;
323}
324
325/* Caller must hold vmlist_lock */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800326static struct vm_struct *__remove_vm_area(const void *addr)
Andi Kleen7856dfe2005-05-20 14:27:57 -0700327{
328 struct vm_struct **p, *tmp;
329
330 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
331 if (tmp->addr == addr)
332 goto found;
333 }
334 return NULL;
335
336found:
337 unmap_vm_area(tmp);
338 *p = tmp->next;
339
340 /*
341 * Remove the guard page.
342 */
343 tmp->size -= PAGE_SIZE;
344 return tmp;
345}
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347/**
Simon Arlott183ff222007-10-20 01:27:18 +0200348 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 * @addr: base address
350 *
351 * Search for the kernel VM area starting at @addr, and remove it.
352 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -0700353 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800355struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
Andi Kleen7856dfe2005-05-20 14:27:57 -0700357 struct vm_struct *v;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 write_lock(&vmlist_lock);
Andi Kleen7856dfe2005-05-20 14:27:57 -0700359 v = __remove_vm_area(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 write_unlock(&vmlist_lock);
Andi Kleen7856dfe2005-05-20 14:27:57 -0700361 return v;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800364static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 struct vm_struct *area;
367
368 if (!addr)
369 return;
370
371 if ((PAGE_SIZE-1) & (unsigned long)addr) {
372 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
373 WARN_ON(1);
374 return;
375 }
376
377 area = remove_vm_area(addr);
378 if (unlikely(!area)) {
379 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
380 addr);
381 WARN_ON(1);
382 return;
383 }
384
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700385 debug_check_no_locks_freed(addr, area->size);
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 if (deallocate_pages) {
388 int i;
389
390 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -0800391 struct page *page = area->pages[i];
392
393 BUG_ON(!page);
394 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
396
Jan Kiszka8757d5f2006-07-14 00:23:56 -0700397 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 vfree(area->pages);
399 else
400 kfree(area->pages);
401 }
402
403 kfree(area);
404 return;
405}
406
407/**
408 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 * @addr: memory base address
410 *
Simon Arlott183ff222007-10-20 01:27:18 +0200411 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -0700412 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
413 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -0700415 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800417void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
419 BUG_ON(in_interrupt());
420 __vunmap(addr, 1);
421}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422EXPORT_SYMBOL(vfree);
423
424/**
425 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 * @addr: memory base address
427 *
428 * Free the virtually contiguous memory area starting at @addr,
429 * which was created from the page array passed to vmap().
430 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -0700431 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800433void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 BUG_ON(in_interrupt());
436 __vunmap(addr, 0);
437}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438EXPORT_SYMBOL(vunmap);
439
440/**
441 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * @pages: array of page pointers
443 * @count: number of pages to map
444 * @flags: vm_area->flags
445 * @prot: page protection for the mapping
446 *
447 * Maps @count pages from @pages into contiguous kernel virtual
448 * space.
449 */
450void *vmap(struct page **pages, unsigned int count,
451 unsigned long flags, pgprot_t prot)
452{
453 struct vm_struct *area;
454
455 if (count > num_physpages)
456 return NULL;
457
458 area = get_vm_area((count << PAGE_SHIFT), flags);
459 if (!area)
460 return NULL;
461 if (map_vm_area(area, prot, &pages)) {
462 vunmap(area->addr);
463 return NULL;
464 }
465
466 return area->addr;
467}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468EXPORT_SYMBOL(vmap);
469
Adrian Bunke31d9eb2008-02-04 22:29:09 -0800470static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
471 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
473 struct page **pages;
474 unsigned int nr_pages, array_size, i;
475
476 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
477 array_size = (nr_pages * sizeof(struct page *));
478
479 area->nr_pages = nr_pages;
480 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -0700481 if (array_size > PAGE_SIZE) {
Christoph Lameter94f60302007-07-17 04:03:29 -0700482 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
483 PAGE_KERNEL, node);
Jan Kiszka8757d5f2006-07-14 00:23:56 -0700484 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -0700485 } else {
486 pages = kmalloc_node(array_size,
Christoph Lameter6cb06222007-10-16 01:25:41 -0700487 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
Andrew Morton286e1ea2006-10-17 00:09:57 -0700488 node);
489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 area->pages = pages;
491 if (!area->pages) {
492 remove_vm_area(area->addr);
493 kfree(area);
494 return NULL;
495 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -0800498 struct page *page;
499
Christoph Lameter930fc452005-10-29 18:15:41 -0700500 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -0800501 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -0700502 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -0800503 page = alloc_pages_node(node, gfp_mask, 0);
504
505 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 /* Successfully allocated i pages, free them in __vunmap() */
507 area->nr_pages = i;
508 goto fail;
509 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -0800510 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512
513 if (map_vm_area(area, prot, &pages))
514 goto fail;
515 return area->addr;
516
517fail:
518 vfree(area->addr);
519 return NULL;
520}
521
Christoph Lameter930fc452005-10-29 18:15:41 -0700522void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
523{
524 return __vmalloc_area_node(area, gfp_mask, prot, -1);
525}
526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527/**
Christoph Lameter930fc452005-10-29 18:15:41 -0700528 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 * @size: allocation size
530 * @gfp_mask: flags for the page level allocator
531 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -0800532 * @node: node to use for allocation or -1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 *
534 * Allocate enough pages to cover @size from the page level
535 * allocator with @gfp_mask flags. Map them into contiguous
536 * kernel virtual space, using a pagetable protection of @prot.
537 */
Adrian Bunkb2213852006-09-25 23:31:02 -0700538static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
539 int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
541 struct vm_struct *area;
542
543 size = PAGE_ALIGN(size);
544 if (!size || (size >> PAGE_SHIFT) > num_physpages)
545 return NULL;
546
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -0700547 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 if (!area)
549 return NULL;
550
Christoph Lameter930fc452005-10-29 18:15:41 -0700551 return __vmalloc_area_node(area, gfp_mask, prot, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
Christoph Lameter930fc452005-10-29 18:15:41 -0700554void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
555{
556 return __vmalloc_node(size, gfp_mask, prot, -1);
557}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558EXPORT_SYMBOL(__vmalloc);
559
560/**
561 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 * Allocate enough pages to cover @size from the page level
564 * allocator and map them into contiguous kernel virtual space.
565 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200566 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 * use __vmalloc() instead.
568 */
569void *vmalloc(unsigned long size)
570{
Nick Piggin83342312006-06-23 02:03:20 -0700571 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573EXPORT_SYMBOL(vmalloc);
574
Christoph Lameter930fc452005-10-29 18:15:41 -0700575/**
Rolf Eike Beeread04082006-09-27 01:50:13 -0700576 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
577 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -0700578 *
Rolf Eike Beeread04082006-09-27 01:50:13 -0700579 * The resulting memory area is zeroed so it can be mapped to userspace
580 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -0700581 */
582void *vmalloc_user(unsigned long size)
583{
584 struct vm_struct *area;
585 void *ret;
586
587 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
Eric Dumazet2b4ac442006-11-10 12:27:48 -0800588 if (ret) {
589 write_lock(&vmlist_lock);
590 area = __find_vm_area(ret);
591 area->flags |= VM_USERMAP;
592 write_unlock(&vmlist_lock);
593 }
Nick Piggin83342312006-06-23 02:03:20 -0700594 return ret;
595}
596EXPORT_SYMBOL(vmalloc_user);
597
598/**
Christoph Lameter930fc452005-10-29 18:15:41 -0700599 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -0700600 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -0800601 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -0700602 *
603 * Allocate enough pages to cover @size from the page level
604 * allocator and map them into contiguous kernel virtual space.
605 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200606 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -0700607 * use __vmalloc() instead.
608 */
609void *vmalloc_node(unsigned long size, int node)
610{
Nick Piggin83342312006-06-23 02:03:20 -0700611 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
Christoph Lameter930fc452005-10-29 18:15:41 -0700612}
613EXPORT_SYMBOL(vmalloc_node);
614
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700615#ifndef PAGE_KERNEL_EXEC
616# define PAGE_KERNEL_EXEC PAGE_KERNEL
617#endif
618
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619/**
620 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 * @size: allocation size
622 *
623 * Kernel-internal function to allocate enough pages to cover @size
624 * the page level allocator and map them into contiguous and
625 * executable kernel virtual space.
626 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200627 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 * use __vmalloc() instead.
629 */
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631void *vmalloc_exec(unsigned long size)
632{
633 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
634}
635
Andi Kleen0d08e0d2007-05-02 19:27:12 +0200636#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -0700637#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +0200638#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -0700639#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +0200640#else
641#define GFP_VMALLOC32 GFP_KERNEL
642#endif
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644/**
645 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 * @size: allocation size
647 *
648 * Allocate enough 32bit PA addressable pages to cover @size from the
649 * page level allocator and map them into contiguous kernel virtual space.
650 */
651void *vmalloc_32(unsigned long size)
652{
Andi Kleen0d08e0d2007-05-02 19:27:12 +0200653 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655EXPORT_SYMBOL(vmalloc_32);
656
Nick Piggin83342312006-06-23 02:03:20 -0700657/**
Rolf Eike Beeread04082006-09-27 01:50:13 -0700658 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -0700659 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -0700660 *
661 * The resulting memory area is 32bit addressable and zeroed so it can be
662 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -0700663 */
664void *vmalloc_32_user(unsigned long size)
665{
666 struct vm_struct *area;
667 void *ret;
668
Andi Kleen0d08e0d2007-05-02 19:27:12 +0200669 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
Eric Dumazet2b4ac442006-11-10 12:27:48 -0800670 if (ret) {
671 write_lock(&vmlist_lock);
672 area = __find_vm_area(ret);
673 area->flags |= VM_USERMAP;
674 write_unlock(&vmlist_lock);
675 }
Nick Piggin83342312006-06-23 02:03:20 -0700676 return ret;
677}
678EXPORT_SYMBOL(vmalloc_32_user);
679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680long vread(char *buf, char *addr, unsigned long count)
681{
682 struct vm_struct *tmp;
683 char *vaddr, *buf_start = buf;
684 unsigned long n;
685
686 /* Don't allow overflow */
687 if ((unsigned long) addr + count < count)
688 count = -(unsigned long) addr;
689
690 read_lock(&vmlist_lock);
691 for (tmp = vmlist; tmp; tmp = tmp->next) {
692 vaddr = (char *) tmp->addr;
693 if (addr >= vaddr + tmp->size - PAGE_SIZE)
694 continue;
695 while (addr < vaddr) {
696 if (count == 0)
697 goto finished;
698 *buf = '\0';
699 buf++;
700 addr++;
701 count--;
702 }
703 n = vaddr + tmp->size - PAGE_SIZE - addr;
704 do {
705 if (count == 0)
706 goto finished;
707 *buf = *addr;
708 buf++;
709 addr++;
710 count--;
711 } while (--n > 0);
712 }
713finished:
714 read_unlock(&vmlist_lock);
715 return buf - buf_start;
716}
717
718long vwrite(char *buf, char *addr, unsigned long count)
719{
720 struct vm_struct *tmp;
721 char *vaddr, *buf_start = buf;
722 unsigned long n;
723
724 /* Don't allow overflow */
725 if ((unsigned long) addr + count < count)
726 count = -(unsigned long) addr;
727
728 read_lock(&vmlist_lock);
729 for (tmp = vmlist; tmp; tmp = tmp->next) {
730 vaddr = (char *) tmp->addr;
731 if (addr >= vaddr + tmp->size - PAGE_SIZE)
732 continue;
733 while (addr < vaddr) {
734 if (count == 0)
735 goto finished;
736 buf++;
737 addr++;
738 count--;
739 }
740 n = vaddr + tmp->size - PAGE_SIZE - addr;
741 do {
742 if (count == 0)
743 goto finished;
744 *addr = *buf;
745 buf++;
746 addr++;
747 count--;
748 } while (--n > 0);
749 }
750finished:
751 read_unlock(&vmlist_lock);
752 return buf - buf_start;
753}
Nick Piggin83342312006-06-23 02:03:20 -0700754
755/**
756 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -0700757 * @vma: vma to cover (map full range of vma)
758 * @addr: vmalloc memory
759 * @pgoff: number of pages into addr before first page to map
760 * @returns: 0 for success, -Exxx on failure
761 *
762 * This function checks that addr is a valid vmalloc'ed area, and
763 * that it is big enough to cover the vma. Will return failure if
764 * that criteria isn't met.
765 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800766 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -0700767 */
768int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
769 unsigned long pgoff)
770{
771 struct vm_struct *area;
772 unsigned long uaddr = vma->vm_start;
773 unsigned long usize = vma->vm_end - vma->vm_start;
774 int ret;
775
776 if ((PAGE_SIZE-1) & (unsigned long)addr)
777 return -EINVAL;
778
779 read_lock(&vmlist_lock);
780 area = __find_vm_area(addr);
781 if (!area)
782 goto out_einval_locked;
783
784 if (!(area->flags & VM_USERMAP))
785 goto out_einval_locked;
786
787 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
788 goto out_einval_locked;
789 read_unlock(&vmlist_lock);
790
791 addr += pgoff << PAGE_SHIFT;
792 do {
793 struct page *page = vmalloc_to_page(addr);
794 ret = vm_insert_page(vma, uaddr, page);
795 if (ret)
796 return ret;
797
798 uaddr += PAGE_SIZE;
799 addr += PAGE_SIZE;
800 usize -= PAGE_SIZE;
801 } while (usize > 0);
802
803 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
804 vma->vm_flags |= VM_RESERVED;
805
806 return ret;
807
808out_einval_locked:
809 read_unlock(&vmlist_lock);
810 return -EINVAL;
811}
812EXPORT_SYMBOL(remap_vmalloc_range);
813
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700814/*
815 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
816 * have one.
817 */
818void __attribute__((weak)) vmalloc_sync_all(void)
819{
820}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -0700821
822
823static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
824{
825 /* apply_to_page_range() does all the hard work. */
826 return 0;
827}
828
829/**
830 * alloc_vm_area - allocate a range of kernel address space
831 * @size: size of the area
832 * @returns: NULL on failure, vm_struct on success
833 *
834 * This function reserves a range of kernel address space, and
835 * allocates pagetables to map that range. No actual mappings
836 * are created. If the kernel address space is not shared
837 * between processes, it syncs the pagetable across all
838 * processes.
839 */
840struct vm_struct *alloc_vm_area(size_t size)
841{
842 struct vm_struct *area;
843
844 area = get_vm_area(size, VM_IOREMAP);
845 if (area == NULL)
846 return NULL;
847
848 /*
849 * This ensures that page tables are constructed for this region
850 * of kernel virtual address space and mapped into init_mm.
851 */
852 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
853 area->size, f, NULL)) {
854 free_vm_area(area);
855 return NULL;
856 }
857
858 /* Make sure the pagetables are constructed in process kernel
859 mappings */
860 vmalloc_sync_all();
861
862 return area;
863}
864EXPORT_SYMBOL_GPL(alloc_vm_area);
865
866void free_vm_area(struct vm_struct *area)
867{
868 struct vm_struct *ret;
869 ret = remove_vm_area(area->addr);
870 BUG_ON(ret != area);
871 kfree(area);
872}
873EXPORT_SYMBOL_GPL(free_vm_area);