Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | |
930fc45a | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
9 | */ |
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 | ||
24 | DEFINE_RWLOCK(vmlist_lock); | |
25 | struct vm_struct *vmlist; | |
26 | ||
b221385b AB |
27 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
28 | int node); | |
29 | ||
1da177e4 LT |
30 | static 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 | ||
41 | static 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 | ||
56 | static 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 | ||
c19c03fc | 71 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1da177e4 LT |
72 | { |
73 | pgd_t *pgd; | |
74 | unsigned long next; | |
c19c03fc BH |
75 | unsigned long start = addr; |
76 | unsigned long end = addr + size; | |
1da177e4 LT |
77 | |
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); | |
c19c03fc BH |
87 | flush_tlb_kernel_range(start, end); |
88 | } | |
89 | ||
90 | static void unmap_vm_area(struct vm_struct *area) | |
91 | { | |
92 | unmap_kernel_range((unsigned long)area->addr, area->size); | |
1da177e4 LT |
93 | } |
94 | ||
95 | static 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 | ||
872fec16 | 100 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
101 | 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 | ||
114 | static 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 | ||
131 | static 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 | ||
148 | int 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); | |
1da177e4 LT |
158 | 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); | |
1da177e4 LT |
164 | flush_cache_vmap((unsigned long) area->addr, end); |
165 | return err; | |
166 | } | |
5992b6da | 167 | EXPORT_SYMBOL_GPL(map_vm_area); |
1da177e4 | 168 | |
48667e7a CL |
169 | /* |
170 | * Map a vmalloc()-space virtual address to the physical page. | |
171 | */ | |
172 | struct page *vmalloc_to_page(void *vmalloc_addr) | |
173 | { | |
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 | } | |
196 | EXPORT_SYMBOL(vmalloc_to_page); | |
197 | ||
198 | /* | |
199 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
200 | */ | |
201 | unsigned long vmalloc_to_pfn(void *vmalloc_addr) | |
202 | { | |
203 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
204 | } | |
205 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
206 | ||
52fd24ca GP |
207 | static 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) | |
1da177e4 LT |
210 | { |
211 | struct vm_struct **p, *tmp, *area; | |
212 | unsigned long align = 1; | |
213 | unsigned long addr; | |
214 | ||
52fd24ca | 215 | BUG_ON(in_interrupt()); |
1da177e4 LT |
216 | 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); | |
31be8309 OH |
228 | if (unlikely(!size)) |
229 | return NULL; | |
1da177e4 | 230 | |
6cb06229 CL |
231 | area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
232 | ||
1da177e4 LT |
233 | if (unlikely(!area)) |
234 | return NULL; | |
235 | ||
1da177e4 LT |
236 | /* |
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 | } | |
257 | ||
258 | found: | |
259 | area->next = *p; | |
260 | *p = area; | |
261 | ||
262 | area->flags = flags; | |
263 | area->addr = (void *)addr; | |
264 | area->size = size; | |
265 | area->pages = NULL; | |
266 | area->nr_pages = 0; | |
267 | area->phys_addr = 0; | |
268 | write_unlock(&vmlist_lock); | |
269 | ||
270 | return area; | |
271 | ||
272 | out: | |
273 | write_unlock(&vmlist_lock); | |
274 | kfree(area); | |
275 | if (printk_ratelimit()) | |
276 | printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n"); | |
277 | return NULL; | |
278 | } | |
279 | ||
930fc45a CL |
280 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
281 | unsigned long start, unsigned long end) | |
282 | { | |
52fd24ca | 283 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL); |
930fc45a | 284 | } |
5992b6da | 285 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 286 | |
1da177e4 | 287 | /** |
183ff22b | 288 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
289 | * @size: size of the area |
290 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
291 | * | |
292 | * Search an area of @size in the kernel virtual mapping area, | |
293 | * and reserved it for out purposes. Returns the area descriptor | |
294 | * on success or %NULL on failure. | |
295 | */ | |
296 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
297 | { | |
298 | return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END); | |
299 | } | |
300 | ||
52fd24ca GP |
301 | struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, |
302 | int node, gfp_t gfp_mask) | |
930fc45a | 303 | { |
52fd24ca GP |
304 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node, |
305 | gfp_mask); | |
930fc45a CL |
306 | } |
307 | ||
83342314 NP |
308 | /* Caller must hold vmlist_lock */ |
309 | static struct vm_struct *__find_vm_area(void *addr) | |
310 | { | |
311 | struct vm_struct *tmp; | |
312 | ||
313 | for (tmp = vmlist; tmp != NULL; tmp = tmp->next) { | |
314 | if (tmp->addr == addr) | |
315 | break; | |
316 | } | |
317 | ||
318 | return tmp; | |
319 | } | |
320 | ||
7856dfeb | 321 | /* Caller must hold vmlist_lock */ |
d24afc57 | 322 | static struct vm_struct *__remove_vm_area(void *addr) |
1da177e4 LT |
323 | { |
324 | struct vm_struct **p, *tmp; | |
325 | ||
1da177e4 LT |
326 | for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) { |
327 | if (tmp->addr == addr) | |
328 | goto found; | |
329 | } | |
1da177e4 LT |
330 | return NULL; |
331 | ||
332 | found: | |
333 | unmap_vm_area(tmp); | |
334 | *p = tmp->next; | |
1da177e4 LT |
335 | |
336 | /* | |
337 | * Remove the guard page. | |
338 | */ | |
339 | tmp->size -= PAGE_SIZE; | |
340 | return tmp; | |
341 | } | |
342 | ||
7856dfeb | 343 | /** |
183ff22b | 344 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
345 | * @addr: base address |
346 | * | |
347 | * Search for the kernel VM area starting at @addr, and remove it. | |
348 | * This function returns the found VM area, but using it is NOT safe | |
349 | * on SMP machines, except for its size or flags. | |
350 | */ | |
351 | struct vm_struct *remove_vm_area(void *addr) | |
352 | { | |
353 | struct vm_struct *v; | |
354 | write_lock(&vmlist_lock); | |
355 | v = __remove_vm_area(addr); | |
356 | write_unlock(&vmlist_lock); | |
357 | return v; | |
358 | } | |
359 | ||
d55e2ca8 | 360 | static void __vunmap(void *addr, int deallocate_pages) |
1da177e4 LT |
361 | { |
362 | struct vm_struct *area; | |
363 | ||
364 | if (!addr) | |
365 | return; | |
366 | ||
367 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
368 | printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr); | |
369 | WARN_ON(1); | |
370 | return; | |
371 | } | |
372 | ||
373 | area = remove_vm_area(addr); | |
374 | if (unlikely(!area)) { | |
375 | printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", | |
376 | addr); | |
377 | WARN_ON(1); | |
378 | return; | |
379 | } | |
380 | ||
9a11b49a IM |
381 | debug_check_no_locks_freed(addr, area->size); |
382 | ||
1da177e4 LT |
383 | if (deallocate_pages) { |
384 | int i; | |
385 | ||
386 | for (i = 0; i < area->nr_pages; i++) { | |
5aae277e | 387 | BUG_ON(!area->pages[i]); |
1da177e4 LT |
388 | __free_page(area->pages[i]); |
389 | } | |
390 | ||
8757d5fa | 391 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
392 | vfree(area->pages); |
393 | else | |
394 | kfree(area->pages); | |
395 | } | |
396 | ||
397 | kfree(area); | |
398 | return; | |
399 | } | |
400 | ||
401 | /** | |
402 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
403 | * @addr: memory base address |
404 | * | |
183ff22b | 405 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
406 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
407 | * NULL, no operation is performed. | |
1da177e4 | 408 | * |
80e93eff | 409 | * Must not be called in interrupt context. |
1da177e4 LT |
410 | */ |
411 | void vfree(void *addr) | |
412 | { | |
413 | BUG_ON(in_interrupt()); | |
414 | __vunmap(addr, 1); | |
415 | } | |
1da177e4 LT |
416 | EXPORT_SYMBOL(vfree); |
417 | ||
418 | /** | |
419 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
420 | * @addr: memory base address |
421 | * | |
422 | * Free the virtually contiguous memory area starting at @addr, | |
423 | * which was created from the page array passed to vmap(). | |
424 | * | |
80e93eff | 425 | * Must not be called in interrupt context. |
1da177e4 LT |
426 | */ |
427 | void vunmap(void *addr) | |
428 | { | |
429 | BUG_ON(in_interrupt()); | |
430 | __vunmap(addr, 0); | |
431 | } | |
1da177e4 LT |
432 | EXPORT_SYMBOL(vunmap); |
433 | ||
434 | /** | |
435 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
436 | * @pages: array of page pointers |
437 | * @count: number of pages to map | |
438 | * @flags: vm_area->flags | |
439 | * @prot: page protection for the mapping | |
440 | * | |
441 | * Maps @count pages from @pages into contiguous kernel virtual | |
442 | * space. | |
443 | */ | |
444 | void *vmap(struct page **pages, unsigned int count, | |
445 | unsigned long flags, pgprot_t prot) | |
446 | { | |
447 | struct vm_struct *area; | |
448 | ||
449 | if (count > num_physpages) | |
450 | return NULL; | |
451 | ||
452 | area = get_vm_area((count << PAGE_SHIFT), flags); | |
453 | if (!area) | |
454 | return NULL; | |
455 | if (map_vm_area(area, prot, &pages)) { | |
456 | vunmap(area->addr); | |
457 | return NULL; | |
458 | } | |
459 | ||
460 | return area->addr; | |
461 | } | |
1da177e4 LT |
462 | EXPORT_SYMBOL(vmap); |
463 | ||
930fc45a CL |
464 | void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
465 | pgprot_t prot, int node) | |
1da177e4 LT |
466 | { |
467 | struct page **pages; | |
468 | unsigned int nr_pages, array_size, i; | |
469 | ||
470 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
471 | array_size = (nr_pages * sizeof(struct page *)); | |
472 | ||
473 | area->nr_pages = nr_pages; | |
474 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 475 | if (array_size > PAGE_SIZE) { |
94f6030c CL |
476 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
477 | PAGE_KERNEL, node); | |
8757d5fa | 478 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
479 | } else { |
480 | pages = kmalloc_node(array_size, | |
6cb06229 | 481 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
482 | node); |
483 | } | |
1da177e4 LT |
484 | area->pages = pages; |
485 | if (!area->pages) { | |
486 | remove_vm_area(area->addr); | |
487 | kfree(area); | |
488 | return NULL; | |
489 | } | |
1da177e4 LT |
490 | |
491 | for (i = 0; i < area->nr_pages; i++) { | |
930fc45a CL |
492 | if (node < 0) |
493 | area->pages[i] = alloc_page(gfp_mask); | |
494 | else | |
495 | area->pages[i] = alloc_pages_node(node, gfp_mask, 0); | |
1da177e4 LT |
496 | if (unlikely(!area->pages[i])) { |
497 | /* Successfully allocated i pages, free them in __vunmap() */ | |
498 | area->nr_pages = i; | |
499 | goto fail; | |
500 | } | |
501 | } | |
502 | ||
503 | if (map_vm_area(area, prot, &pages)) | |
504 | goto fail; | |
505 | return area->addr; | |
506 | ||
507 | fail: | |
508 | vfree(area->addr); | |
509 | return NULL; | |
510 | } | |
511 | ||
930fc45a CL |
512 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
513 | { | |
514 | return __vmalloc_area_node(area, gfp_mask, prot, -1); | |
515 | } | |
516 | ||
1da177e4 | 517 | /** |
930fc45a | 518 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
519 | * @size: allocation size |
520 | * @gfp_mask: flags for the page level allocator | |
521 | * @prot: protection mask for the allocated pages | |
d44e0780 | 522 | * @node: node to use for allocation or -1 |
1da177e4 LT |
523 | * |
524 | * Allocate enough pages to cover @size from the page level | |
525 | * allocator with @gfp_mask flags. Map them into contiguous | |
526 | * kernel virtual space, using a pagetable protection of @prot. | |
527 | */ | |
b221385b AB |
528 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
529 | int node) | |
1da177e4 LT |
530 | { |
531 | struct vm_struct *area; | |
532 | ||
533 | size = PAGE_ALIGN(size); | |
534 | if (!size || (size >> PAGE_SHIFT) > num_physpages) | |
535 | return NULL; | |
536 | ||
52fd24ca | 537 | area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask); |
1da177e4 LT |
538 | if (!area) |
539 | return NULL; | |
540 | ||
930fc45a | 541 | return __vmalloc_area_node(area, gfp_mask, prot, node); |
1da177e4 LT |
542 | } |
543 | ||
930fc45a CL |
544 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
545 | { | |
546 | return __vmalloc_node(size, gfp_mask, prot, -1); | |
547 | } | |
1da177e4 LT |
548 | EXPORT_SYMBOL(__vmalloc); |
549 | ||
550 | /** | |
551 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 552 | * @size: allocation size |
1da177e4 LT |
553 | * Allocate enough pages to cover @size from the page level |
554 | * allocator and map them into contiguous kernel virtual space. | |
555 | * | |
c1c8897f | 556 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
557 | * use __vmalloc() instead. |
558 | */ | |
559 | void *vmalloc(unsigned long size) | |
560 | { | |
83342314 | 561 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); |
1da177e4 | 562 | } |
1da177e4 LT |
563 | EXPORT_SYMBOL(vmalloc); |
564 | ||
83342314 | 565 | /** |
ead04089 REB |
566 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
567 | * @size: allocation size | |
83342314 | 568 | * |
ead04089 REB |
569 | * The resulting memory area is zeroed so it can be mapped to userspace |
570 | * without leaking data. | |
83342314 NP |
571 | */ |
572 | void *vmalloc_user(unsigned long size) | |
573 | { | |
574 | struct vm_struct *area; | |
575 | void *ret; | |
576 | ||
577 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); | |
2b4ac44e ED |
578 | if (ret) { |
579 | write_lock(&vmlist_lock); | |
580 | area = __find_vm_area(ret); | |
581 | area->flags |= VM_USERMAP; | |
582 | write_unlock(&vmlist_lock); | |
583 | } | |
83342314 NP |
584 | return ret; |
585 | } | |
586 | EXPORT_SYMBOL(vmalloc_user); | |
587 | ||
930fc45a CL |
588 | /** |
589 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 590 | * @size: allocation size |
d44e0780 | 591 | * @node: numa node |
930fc45a CL |
592 | * |
593 | * Allocate enough pages to cover @size from the page level | |
594 | * allocator and map them into contiguous kernel virtual space. | |
595 | * | |
c1c8897f | 596 | * For tight control over page level allocator and protection flags |
930fc45a CL |
597 | * use __vmalloc() instead. |
598 | */ | |
599 | void *vmalloc_node(unsigned long size, int node) | |
600 | { | |
83342314 | 601 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node); |
930fc45a CL |
602 | } |
603 | EXPORT_SYMBOL(vmalloc_node); | |
604 | ||
4dc3b16b PP |
605 | #ifndef PAGE_KERNEL_EXEC |
606 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
607 | #endif | |
608 | ||
1da177e4 LT |
609 | /** |
610 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
611 | * @size: allocation size |
612 | * | |
613 | * Kernel-internal function to allocate enough pages to cover @size | |
614 | * the page level allocator and map them into contiguous and | |
615 | * executable kernel virtual space. | |
616 | * | |
c1c8897f | 617 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
618 | * use __vmalloc() instead. |
619 | */ | |
620 | ||
1da177e4 LT |
621 | void *vmalloc_exec(unsigned long size) |
622 | { | |
623 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
624 | } | |
625 | ||
0d08e0d3 | 626 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 627 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 628 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 629 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
630 | #else |
631 | #define GFP_VMALLOC32 GFP_KERNEL | |
632 | #endif | |
633 | ||
1da177e4 LT |
634 | /** |
635 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
636 | * @size: allocation size |
637 | * | |
638 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
639 | * page level allocator and map them into contiguous kernel virtual space. | |
640 | */ | |
641 | void *vmalloc_32(unsigned long size) | |
642 | { | |
0d08e0d3 | 643 | return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL); |
1da177e4 | 644 | } |
1da177e4 LT |
645 | EXPORT_SYMBOL(vmalloc_32); |
646 | ||
83342314 | 647 | /** |
ead04089 | 648 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 649 | * @size: allocation size |
ead04089 REB |
650 | * |
651 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
652 | * mapped to userspace without leaking data. | |
83342314 NP |
653 | */ |
654 | void *vmalloc_32_user(unsigned long size) | |
655 | { | |
656 | struct vm_struct *area; | |
657 | void *ret; | |
658 | ||
0d08e0d3 | 659 | ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL); |
2b4ac44e ED |
660 | if (ret) { |
661 | write_lock(&vmlist_lock); | |
662 | area = __find_vm_area(ret); | |
663 | area->flags |= VM_USERMAP; | |
664 | write_unlock(&vmlist_lock); | |
665 | } | |
83342314 NP |
666 | return ret; |
667 | } | |
668 | EXPORT_SYMBOL(vmalloc_32_user); | |
669 | ||
1da177e4 LT |
670 | long vread(char *buf, char *addr, unsigned long count) |
671 | { | |
672 | struct vm_struct *tmp; | |
673 | char *vaddr, *buf_start = buf; | |
674 | unsigned long n; | |
675 | ||
676 | /* Don't allow overflow */ | |
677 | if ((unsigned long) addr + count < count) | |
678 | count = -(unsigned long) addr; | |
679 | ||
680 | read_lock(&vmlist_lock); | |
681 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
682 | vaddr = (char *) tmp->addr; | |
683 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
684 | continue; | |
685 | while (addr < vaddr) { | |
686 | if (count == 0) | |
687 | goto finished; | |
688 | *buf = '\0'; | |
689 | buf++; | |
690 | addr++; | |
691 | count--; | |
692 | } | |
693 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
694 | do { | |
695 | if (count == 0) | |
696 | goto finished; | |
697 | *buf = *addr; | |
698 | buf++; | |
699 | addr++; | |
700 | count--; | |
701 | } while (--n > 0); | |
702 | } | |
703 | finished: | |
704 | read_unlock(&vmlist_lock); | |
705 | return buf - buf_start; | |
706 | } | |
707 | ||
708 | long vwrite(char *buf, char *addr, unsigned long count) | |
709 | { | |
710 | struct vm_struct *tmp; | |
711 | char *vaddr, *buf_start = buf; | |
712 | unsigned long n; | |
713 | ||
714 | /* Don't allow overflow */ | |
715 | if ((unsigned long) addr + count < count) | |
716 | count = -(unsigned long) addr; | |
717 | ||
718 | read_lock(&vmlist_lock); | |
719 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
720 | vaddr = (char *) tmp->addr; | |
721 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
722 | continue; | |
723 | while (addr < vaddr) { | |
724 | if (count == 0) | |
725 | goto finished; | |
726 | buf++; | |
727 | addr++; | |
728 | count--; | |
729 | } | |
730 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
731 | do { | |
732 | if (count == 0) | |
733 | goto finished; | |
734 | *addr = *buf; | |
735 | buf++; | |
736 | addr++; | |
737 | count--; | |
738 | } while (--n > 0); | |
739 | } | |
740 | finished: | |
741 | read_unlock(&vmlist_lock); | |
742 | return buf - buf_start; | |
743 | } | |
83342314 NP |
744 | |
745 | /** | |
746 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
747 | * @vma: vma to cover (map full range of vma) |
748 | * @addr: vmalloc memory | |
749 | * @pgoff: number of pages into addr before first page to map | |
750 | * @returns: 0 for success, -Exxx on failure | |
751 | * | |
752 | * This function checks that addr is a valid vmalloc'ed area, and | |
753 | * that it is big enough to cover the vma. Will return failure if | |
754 | * that criteria isn't met. | |
755 | * | |
72fd4a35 | 756 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
757 | */ |
758 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
759 | unsigned long pgoff) | |
760 | { | |
761 | struct vm_struct *area; | |
762 | unsigned long uaddr = vma->vm_start; | |
763 | unsigned long usize = vma->vm_end - vma->vm_start; | |
764 | int ret; | |
765 | ||
766 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
767 | return -EINVAL; | |
768 | ||
769 | read_lock(&vmlist_lock); | |
770 | area = __find_vm_area(addr); | |
771 | if (!area) | |
772 | goto out_einval_locked; | |
773 | ||
774 | if (!(area->flags & VM_USERMAP)) | |
775 | goto out_einval_locked; | |
776 | ||
777 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
778 | goto out_einval_locked; | |
779 | read_unlock(&vmlist_lock); | |
780 | ||
781 | addr += pgoff << PAGE_SHIFT; | |
782 | do { | |
783 | struct page *page = vmalloc_to_page(addr); | |
784 | ret = vm_insert_page(vma, uaddr, page); | |
785 | if (ret) | |
786 | return ret; | |
787 | ||
788 | uaddr += PAGE_SIZE; | |
789 | addr += PAGE_SIZE; | |
790 | usize -= PAGE_SIZE; | |
791 | } while (usize > 0); | |
792 | ||
793 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
794 | vma->vm_flags |= VM_RESERVED; | |
795 | ||
796 | return ret; | |
797 | ||
798 | out_einval_locked: | |
799 | read_unlock(&vmlist_lock); | |
800 | return -EINVAL; | |
801 | } | |
802 | EXPORT_SYMBOL(remap_vmalloc_range); | |
803 | ||
1eeb66a1 CH |
804 | /* |
805 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
806 | * have one. | |
807 | */ | |
808 | void __attribute__((weak)) vmalloc_sync_all(void) | |
809 | { | |
810 | } | |
5f4352fb JF |
811 | |
812 | ||
813 | static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data) | |
814 | { | |
815 | /* apply_to_page_range() does all the hard work. */ | |
816 | return 0; | |
817 | } | |
818 | ||
819 | /** | |
820 | * alloc_vm_area - allocate a range of kernel address space | |
821 | * @size: size of the area | |
822 | * @returns: NULL on failure, vm_struct on success | |
823 | * | |
824 | * This function reserves a range of kernel address space, and | |
825 | * allocates pagetables to map that range. No actual mappings | |
826 | * are created. If the kernel address space is not shared | |
827 | * between processes, it syncs the pagetable across all | |
828 | * processes. | |
829 | */ | |
830 | struct vm_struct *alloc_vm_area(size_t size) | |
831 | { | |
832 | struct vm_struct *area; | |
833 | ||
834 | area = get_vm_area(size, VM_IOREMAP); | |
835 | if (area == NULL) | |
836 | return NULL; | |
837 | ||
838 | /* | |
839 | * This ensures that page tables are constructed for this region | |
840 | * of kernel virtual address space and mapped into init_mm. | |
841 | */ | |
842 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
843 | area->size, f, NULL)) { | |
844 | free_vm_area(area); | |
845 | return NULL; | |
846 | } | |
847 | ||
848 | /* Make sure the pagetables are constructed in process kernel | |
849 | mappings */ | |
850 | vmalloc_sync_all(); | |
851 | ||
852 | return area; | |
853 | } | |
854 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
855 | ||
856 | void free_vm_area(struct vm_struct *area) | |
857 | { | |
858 | struct vm_struct *ret; | |
859 | ret = remove_vm_area(area->addr); | |
860 | BUG_ON(ret != area); | |
861 | kfree(area); | |
862 | } | |
863 | EXPORT_SYMBOL_GPL(free_vm_area); |