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> | |
a10aa579 | 17 | #include <linux/seq_file.h> |
3ac7fe5a | 18 | #include <linux/debugobjects.h> |
1da177e4 | 19 | #include <linux/vmalloc.h> |
23016969 | 20 | #include <linux/kallsyms.h> |
1da177e4 LT |
21 | |
22 | #include <asm/uaccess.h> | |
23 | #include <asm/tlbflush.h> | |
24 | ||
25 | ||
26 | DEFINE_RWLOCK(vmlist_lock); | |
27 | struct vm_struct *vmlist; | |
28 | ||
b221385b | 29 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 30 | int node, void *caller); |
b221385b | 31 | |
1da177e4 LT |
32 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
33 | { | |
34 | pte_t *pte; | |
35 | ||
36 | pte = pte_offset_kernel(pmd, addr); | |
37 | do { | |
38 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
39 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
40 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
41 | } | |
42 | ||
43 | static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr, | |
44 | unsigned long end) | |
45 | { | |
46 | pmd_t *pmd; | |
47 | unsigned long next; | |
48 | ||
49 | pmd = pmd_offset(pud, addr); | |
50 | do { | |
51 | next = pmd_addr_end(addr, end); | |
52 | if (pmd_none_or_clear_bad(pmd)) | |
53 | continue; | |
54 | vunmap_pte_range(pmd, addr, next); | |
55 | } while (pmd++, addr = next, addr != end); | |
56 | } | |
57 | ||
58 | static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr, | |
59 | unsigned long end) | |
60 | { | |
61 | pud_t *pud; | |
62 | unsigned long next; | |
63 | ||
64 | pud = pud_offset(pgd, addr); | |
65 | do { | |
66 | next = pud_addr_end(addr, end); | |
67 | if (pud_none_or_clear_bad(pud)) | |
68 | continue; | |
69 | vunmap_pmd_range(pud, addr, next); | |
70 | } while (pud++, addr = next, addr != end); | |
71 | } | |
72 | ||
c19c03fc | 73 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1da177e4 LT |
74 | { |
75 | pgd_t *pgd; | |
76 | unsigned long next; | |
c19c03fc BH |
77 | unsigned long start = addr; |
78 | unsigned long end = addr + size; | |
1da177e4 LT |
79 | |
80 | BUG_ON(addr >= end); | |
81 | pgd = pgd_offset_k(addr); | |
82 | flush_cache_vunmap(addr, end); | |
83 | do { | |
84 | next = pgd_addr_end(addr, end); | |
85 | if (pgd_none_or_clear_bad(pgd)) | |
86 | continue; | |
87 | vunmap_pud_range(pgd, addr, next); | |
88 | } while (pgd++, addr = next, addr != end); | |
c19c03fc BH |
89 | flush_tlb_kernel_range(start, end); |
90 | } | |
91 | ||
92 | static void unmap_vm_area(struct vm_struct *area) | |
93 | { | |
94 | unmap_kernel_range((unsigned long)area->addr, area->size); | |
1da177e4 LT |
95 | } |
96 | ||
97 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
98 | unsigned long end, pgprot_t prot, struct page ***pages) | |
99 | { | |
100 | pte_t *pte; | |
101 | ||
872fec16 | 102 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
103 | if (!pte) |
104 | return -ENOMEM; | |
105 | do { | |
106 | struct page *page = **pages; | |
107 | WARN_ON(!pte_none(*pte)); | |
108 | if (!page) | |
109 | return -ENOMEM; | |
110 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
111 | (*pages)++; | |
112 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
113 | return 0; | |
114 | } | |
115 | ||
116 | static inline int vmap_pmd_range(pud_t *pud, unsigned long addr, | |
117 | unsigned long end, pgprot_t prot, struct page ***pages) | |
118 | { | |
119 | pmd_t *pmd; | |
120 | unsigned long next; | |
121 | ||
122 | pmd = pmd_alloc(&init_mm, pud, addr); | |
123 | if (!pmd) | |
124 | return -ENOMEM; | |
125 | do { | |
126 | next = pmd_addr_end(addr, end); | |
127 | if (vmap_pte_range(pmd, addr, next, prot, pages)) | |
128 | return -ENOMEM; | |
129 | } while (pmd++, addr = next, addr != end); | |
130 | return 0; | |
131 | } | |
132 | ||
133 | static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr, | |
134 | unsigned long end, pgprot_t prot, struct page ***pages) | |
135 | { | |
136 | pud_t *pud; | |
137 | unsigned long next; | |
138 | ||
139 | pud = pud_alloc(&init_mm, pgd, addr); | |
140 | if (!pud) | |
141 | return -ENOMEM; | |
142 | do { | |
143 | next = pud_addr_end(addr, end); | |
144 | if (vmap_pmd_range(pud, addr, next, prot, pages)) | |
145 | return -ENOMEM; | |
146 | } while (pud++, addr = next, addr != end); | |
147 | return 0; | |
148 | } | |
149 | ||
150 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
151 | { | |
152 | pgd_t *pgd; | |
153 | unsigned long next; | |
154 | unsigned long addr = (unsigned long) area->addr; | |
155 | unsigned long end = addr + area->size - PAGE_SIZE; | |
156 | int err; | |
157 | ||
158 | BUG_ON(addr >= end); | |
159 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
160 | do { |
161 | next = pgd_addr_end(addr, end); | |
162 | err = vmap_pud_range(pgd, addr, next, prot, pages); | |
163 | if (err) | |
164 | break; | |
165 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
166 | flush_cache_vmap((unsigned long) area->addr, end); |
167 | return err; | |
168 | } | |
5992b6da | 169 | EXPORT_SYMBOL_GPL(map_vm_area); |
1da177e4 | 170 | |
48667e7a CL |
171 | /* |
172 | * Map a vmalloc()-space virtual address to the physical page. | |
173 | */ | |
b3bdda02 | 174 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
175 | { |
176 | unsigned long addr = (unsigned long) vmalloc_addr; | |
177 | struct page *page = NULL; | |
178 | pgd_t *pgd = pgd_offset_k(addr); | |
179 | pud_t *pud; | |
180 | pmd_t *pmd; | |
181 | pte_t *ptep, pte; | |
182 | ||
183 | if (!pgd_none(*pgd)) { | |
184 | pud = pud_offset(pgd, addr); | |
185 | if (!pud_none(*pud)) { | |
186 | pmd = pmd_offset(pud, addr); | |
187 | if (!pmd_none(*pmd)) { | |
188 | ptep = pte_offset_map(pmd, addr); | |
189 | pte = *ptep; | |
190 | if (pte_present(pte)) | |
191 | page = pte_page(pte); | |
192 | pte_unmap(ptep); | |
193 | } | |
194 | } | |
195 | } | |
196 | return page; | |
197 | } | |
198 | EXPORT_SYMBOL(vmalloc_to_page); | |
199 | ||
200 | /* | |
201 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
202 | */ | |
b3bdda02 | 203 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
204 | { |
205 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
206 | } | |
207 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
208 | ||
23016969 CL |
209 | static struct vm_struct * |
210 | __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start, | |
211 | unsigned long end, int node, gfp_t gfp_mask, void *caller) | |
1da177e4 LT |
212 | { |
213 | struct vm_struct **p, *tmp, *area; | |
214 | unsigned long align = 1; | |
215 | unsigned long addr; | |
216 | ||
52fd24ca | 217 | BUG_ON(in_interrupt()); |
1da177e4 LT |
218 | if (flags & VM_IOREMAP) { |
219 | int bit = fls(size); | |
220 | ||
221 | if (bit > IOREMAP_MAX_ORDER) | |
222 | bit = IOREMAP_MAX_ORDER; | |
223 | else if (bit < PAGE_SHIFT) | |
224 | bit = PAGE_SHIFT; | |
225 | ||
226 | align = 1ul << bit; | |
227 | } | |
228 | addr = ALIGN(start, align); | |
229 | size = PAGE_ALIGN(size); | |
31be8309 OH |
230 | if (unlikely(!size)) |
231 | return NULL; | |
1da177e4 | 232 | |
6cb06229 CL |
233 | area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
234 | ||
1da177e4 LT |
235 | if (unlikely(!area)) |
236 | return NULL; | |
237 | ||
1da177e4 LT |
238 | /* |
239 | * We always allocate a guard page. | |
240 | */ | |
241 | size += PAGE_SIZE; | |
242 | ||
243 | write_lock(&vmlist_lock); | |
244 | for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) { | |
245 | if ((unsigned long)tmp->addr < addr) { | |
246 | if((unsigned long)tmp->addr + tmp->size >= addr) | |
247 | addr = ALIGN(tmp->size + | |
248 | (unsigned long)tmp->addr, align); | |
249 | continue; | |
250 | } | |
251 | if ((size + addr) < addr) | |
252 | goto out; | |
253 | if (size + addr <= (unsigned long)tmp->addr) | |
254 | goto found; | |
255 | addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align); | |
256 | if (addr > end - size) | |
257 | goto out; | |
258 | } | |
5dc33185 RB |
259 | if ((size + addr) < addr) |
260 | goto out; | |
261 | if (addr > end - size) | |
262 | goto out; | |
1da177e4 LT |
263 | |
264 | found: | |
265 | area->next = *p; | |
266 | *p = area; | |
267 | ||
268 | area->flags = flags; | |
269 | area->addr = (void *)addr; | |
270 | area->size = size; | |
271 | area->pages = NULL; | |
272 | area->nr_pages = 0; | |
273 | area->phys_addr = 0; | |
23016969 | 274 | area->caller = caller; |
1da177e4 LT |
275 | write_unlock(&vmlist_lock); |
276 | ||
277 | return area; | |
278 | ||
279 | out: | |
280 | write_unlock(&vmlist_lock); | |
281 | kfree(area); | |
282 | if (printk_ratelimit()) | |
283 | printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n"); | |
284 | return NULL; | |
285 | } | |
286 | ||
930fc45a CL |
287 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
288 | unsigned long start, unsigned long end) | |
289 | { | |
23016969 CL |
290 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, |
291 | __builtin_return_address(0)); | |
930fc45a | 292 | } |
5992b6da | 293 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 294 | |
1da177e4 | 295 | /** |
183ff22b | 296 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
297 | * @size: size of the area |
298 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
299 | * | |
300 | * Search an area of @size in the kernel virtual mapping area, | |
301 | * and reserved it for out purposes. Returns the area descriptor | |
302 | * on success or %NULL on failure. | |
303 | */ | |
304 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
305 | { | |
23016969 CL |
306 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, |
307 | -1, GFP_KERNEL, __builtin_return_address(0)); | |
308 | } | |
309 | ||
310 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
311 | void *caller) | |
312 | { | |
313 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, | |
314 | -1, GFP_KERNEL, caller); | |
1da177e4 LT |
315 | } |
316 | ||
52fd24ca GP |
317 | struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, |
318 | int node, gfp_t gfp_mask) | |
930fc45a | 319 | { |
52fd24ca | 320 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node, |
23016969 | 321 | gfp_mask, __builtin_return_address(0)); |
930fc45a CL |
322 | } |
323 | ||
83342314 | 324 | /* Caller must hold vmlist_lock */ |
b3bdda02 | 325 | static struct vm_struct *__find_vm_area(const void *addr) |
83342314 NP |
326 | { |
327 | struct vm_struct *tmp; | |
328 | ||
329 | for (tmp = vmlist; tmp != NULL; tmp = tmp->next) { | |
330 | if (tmp->addr == addr) | |
331 | break; | |
332 | } | |
333 | ||
334 | return tmp; | |
335 | } | |
336 | ||
7856dfeb | 337 | /* Caller must hold vmlist_lock */ |
b3bdda02 | 338 | static struct vm_struct *__remove_vm_area(const void *addr) |
1da177e4 LT |
339 | { |
340 | struct vm_struct **p, *tmp; | |
341 | ||
1da177e4 LT |
342 | for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) { |
343 | if (tmp->addr == addr) | |
344 | goto found; | |
345 | } | |
1da177e4 LT |
346 | return NULL; |
347 | ||
348 | found: | |
349 | unmap_vm_area(tmp); | |
350 | *p = tmp->next; | |
1da177e4 LT |
351 | |
352 | /* | |
353 | * Remove the guard page. | |
354 | */ | |
355 | tmp->size -= PAGE_SIZE; | |
356 | return tmp; | |
357 | } | |
358 | ||
7856dfeb | 359 | /** |
183ff22b | 360 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
361 | * @addr: base address |
362 | * | |
363 | * Search for the kernel VM area starting at @addr, and remove it. | |
364 | * This function returns the found VM area, but using it is NOT safe | |
365 | * on SMP machines, except for its size or flags. | |
366 | */ | |
b3bdda02 | 367 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb AK |
368 | { |
369 | struct vm_struct *v; | |
370 | write_lock(&vmlist_lock); | |
371 | v = __remove_vm_area(addr); | |
372 | write_unlock(&vmlist_lock); | |
373 | return v; | |
374 | } | |
375 | ||
b3bdda02 | 376 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
377 | { |
378 | struct vm_struct *area; | |
379 | ||
380 | if (!addr) | |
381 | return; | |
382 | ||
383 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
384 | printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr); | |
385 | WARN_ON(1); | |
386 | return; | |
387 | } | |
388 | ||
389 | area = remove_vm_area(addr); | |
390 | if (unlikely(!area)) { | |
391 | printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", | |
392 | addr); | |
393 | WARN_ON(1); | |
394 | return; | |
395 | } | |
396 | ||
9a11b49a | 397 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 398 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 399 | |
1da177e4 LT |
400 | if (deallocate_pages) { |
401 | int i; | |
402 | ||
403 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
404 | struct page *page = area->pages[i]; |
405 | ||
406 | BUG_ON(!page); | |
407 | __free_page(page); | |
1da177e4 LT |
408 | } |
409 | ||
8757d5fa | 410 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
411 | vfree(area->pages); |
412 | else | |
413 | kfree(area->pages); | |
414 | } | |
415 | ||
416 | kfree(area); | |
417 | return; | |
418 | } | |
419 | ||
420 | /** | |
421 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
422 | * @addr: memory base address |
423 | * | |
183ff22b | 424 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
425 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
426 | * NULL, no operation is performed. | |
1da177e4 | 427 | * |
80e93eff | 428 | * Must not be called in interrupt context. |
1da177e4 | 429 | */ |
b3bdda02 | 430 | void vfree(const void *addr) |
1da177e4 LT |
431 | { |
432 | BUG_ON(in_interrupt()); | |
433 | __vunmap(addr, 1); | |
434 | } | |
1da177e4 LT |
435 | EXPORT_SYMBOL(vfree); |
436 | ||
437 | /** | |
438 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
439 | * @addr: memory base address |
440 | * | |
441 | * Free the virtually contiguous memory area starting at @addr, | |
442 | * which was created from the page array passed to vmap(). | |
443 | * | |
80e93eff | 444 | * Must not be called in interrupt context. |
1da177e4 | 445 | */ |
b3bdda02 | 446 | void vunmap(const void *addr) |
1da177e4 LT |
447 | { |
448 | BUG_ON(in_interrupt()); | |
449 | __vunmap(addr, 0); | |
450 | } | |
1da177e4 LT |
451 | EXPORT_SYMBOL(vunmap); |
452 | ||
453 | /** | |
454 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
455 | * @pages: array of page pointers |
456 | * @count: number of pages to map | |
457 | * @flags: vm_area->flags | |
458 | * @prot: page protection for the mapping | |
459 | * | |
460 | * Maps @count pages from @pages into contiguous kernel virtual | |
461 | * space. | |
462 | */ | |
463 | void *vmap(struct page **pages, unsigned int count, | |
464 | unsigned long flags, pgprot_t prot) | |
465 | { | |
466 | struct vm_struct *area; | |
467 | ||
468 | if (count > num_physpages) | |
469 | return NULL; | |
470 | ||
23016969 CL |
471 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
472 | __builtin_return_address(0)); | |
1da177e4 LT |
473 | if (!area) |
474 | return NULL; | |
23016969 | 475 | |
1da177e4 LT |
476 | if (map_vm_area(area, prot, &pages)) { |
477 | vunmap(area->addr); | |
478 | return NULL; | |
479 | } | |
480 | ||
481 | return area->addr; | |
482 | } | |
1da177e4 LT |
483 | EXPORT_SYMBOL(vmap); |
484 | ||
e31d9eb5 | 485 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 486 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
487 | { |
488 | struct page **pages; | |
489 | unsigned int nr_pages, array_size, i; | |
490 | ||
491 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
492 | array_size = (nr_pages * sizeof(struct page *)); | |
493 | ||
494 | area->nr_pages = nr_pages; | |
495 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 496 | if (array_size > PAGE_SIZE) { |
94f6030c | 497 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
23016969 | 498 | PAGE_KERNEL, node, caller); |
8757d5fa | 499 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
500 | } else { |
501 | pages = kmalloc_node(array_size, | |
6cb06229 | 502 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
503 | node); |
504 | } | |
1da177e4 | 505 | area->pages = pages; |
23016969 | 506 | area->caller = caller; |
1da177e4 LT |
507 | if (!area->pages) { |
508 | remove_vm_area(area->addr); | |
509 | kfree(area); | |
510 | return NULL; | |
511 | } | |
1da177e4 LT |
512 | |
513 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
514 | struct page *page; |
515 | ||
930fc45a | 516 | if (node < 0) |
bf53d6f8 | 517 | page = alloc_page(gfp_mask); |
930fc45a | 518 | else |
bf53d6f8 CL |
519 | page = alloc_pages_node(node, gfp_mask, 0); |
520 | ||
521 | if (unlikely(!page)) { | |
1da177e4 LT |
522 | /* Successfully allocated i pages, free them in __vunmap() */ |
523 | area->nr_pages = i; | |
524 | goto fail; | |
525 | } | |
bf53d6f8 | 526 | area->pages[i] = page; |
1da177e4 LT |
527 | } |
528 | ||
529 | if (map_vm_area(area, prot, &pages)) | |
530 | goto fail; | |
531 | return area->addr; | |
532 | ||
533 | fail: | |
534 | vfree(area->addr); | |
535 | return NULL; | |
536 | } | |
537 | ||
930fc45a CL |
538 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
539 | { | |
23016969 CL |
540 | return __vmalloc_area_node(area, gfp_mask, prot, -1, |
541 | __builtin_return_address(0)); | |
930fc45a CL |
542 | } |
543 | ||
1da177e4 | 544 | /** |
930fc45a | 545 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
546 | * @size: allocation size |
547 | * @gfp_mask: flags for the page level allocator | |
548 | * @prot: protection mask for the allocated pages | |
d44e0780 | 549 | * @node: node to use for allocation or -1 |
1da177e4 LT |
550 | * |
551 | * Allocate enough pages to cover @size from the page level | |
552 | * allocator with @gfp_mask flags. Map them into contiguous | |
553 | * kernel virtual space, using a pagetable protection of @prot. | |
554 | */ | |
b221385b | 555 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 556 | int node, void *caller) |
1da177e4 LT |
557 | { |
558 | struct vm_struct *area; | |
559 | ||
560 | size = PAGE_ALIGN(size); | |
561 | if (!size || (size >> PAGE_SHIFT) > num_physpages) | |
562 | return NULL; | |
563 | ||
23016969 CL |
564 | area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END, |
565 | node, gfp_mask, caller); | |
566 | ||
1da177e4 LT |
567 | if (!area) |
568 | return NULL; | |
569 | ||
23016969 | 570 | return __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1da177e4 LT |
571 | } |
572 | ||
930fc45a CL |
573 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
574 | { | |
23016969 CL |
575 | return __vmalloc_node(size, gfp_mask, prot, -1, |
576 | __builtin_return_address(0)); | |
930fc45a | 577 | } |
1da177e4 LT |
578 | EXPORT_SYMBOL(__vmalloc); |
579 | ||
580 | /** | |
581 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 582 | * @size: allocation size |
1da177e4 LT |
583 | * Allocate enough pages to cover @size from the page level |
584 | * allocator and map them into contiguous kernel virtual space. | |
585 | * | |
c1c8897f | 586 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
587 | * use __vmalloc() instead. |
588 | */ | |
589 | void *vmalloc(unsigned long size) | |
590 | { | |
23016969 CL |
591 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
592 | -1, __builtin_return_address(0)); | |
1da177e4 | 593 | } |
1da177e4 LT |
594 | EXPORT_SYMBOL(vmalloc); |
595 | ||
83342314 | 596 | /** |
ead04089 REB |
597 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
598 | * @size: allocation size | |
83342314 | 599 | * |
ead04089 REB |
600 | * The resulting memory area is zeroed so it can be mapped to userspace |
601 | * without leaking data. | |
83342314 NP |
602 | */ |
603 | void *vmalloc_user(unsigned long size) | |
604 | { | |
605 | struct vm_struct *area; | |
606 | void *ret; | |
607 | ||
608 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); | |
2b4ac44e ED |
609 | if (ret) { |
610 | write_lock(&vmlist_lock); | |
611 | area = __find_vm_area(ret); | |
612 | area->flags |= VM_USERMAP; | |
613 | write_unlock(&vmlist_lock); | |
614 | } | |
83342314 NP |
615 | return ret; |
616 | } | |
617 | EXPORT_SYMBOL(vmalloc_user); | |
618 | ||
930fc45a CL |
619 | /** |
620 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 621 | * @size: allocation size |
d44e0780 | 622 | * @node: numa node |
930fc45a CL |
623 | * |
624 | * Allocate enough pages to cover @size from the page level | |
625 | * allocator and map them into contiguous kernel virtual space. | |
626 | * | |
c1c8897f | 627 | * For tight control over page level allocator and protection flags |
930fc45a CL |
628 | * use __vmalloc() instead. |
629 | */ | |
630 | void *vmalloc_node(unsigned long size, int node) | |
631 | { | |
23016969 CL |
632 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
633 | node, __builtin_return_address(0)); | |
930fc45a CL |
634 | } |
635 | EXPORT_SYMBOL(vmalloc_node); | |
636 | ||
4dc3b16b PP |
637 | #ifndef PAGE_KERNEL_EXEC |
638 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
639 | #endif | |
640 | ||
1da177e4 LT |
641 | /** |
642 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
643 | * @size: allocation size |
644 | * | |
645 | * Kernel-internal function to allocate enough pages to cover @size | |
646 | * the page level allocator and map them into contiguous and | |
647 | * executable kernel virtual space. | |
648 | * | |
c1c8897f | 649 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
650 | * use __vmalloc() instead. |
651 | */ | |
652 | ||
1da177e4 LT |
653 | void *vmalloc_exec(unsigned long size) |
654 | { | |
655 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
656 | } | |
657 | ||
0d08e0d3 | 658 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 659 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 660 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 661 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
662 | #else |
663 | #define GFP_VMALLOC32 GFP_KERNEL | |
664 | #endif | |
665 | ||
1da177e4 LT |
666 | /** |
667 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
668 | * @size: allocation size |
669 | * | |
670 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
671 | * page level allocator and map them into contiguous kernel virtual space. | |
672 | */ | |
673 | void *vmalloc_32(unsigned long size) | |
674 | { | |
0d08e0d3 | 675 | return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL); |
1da177e4 | 676 | } |
1da177e4 LT |
677 | EXPORT_SYMBOL(vmalloc_32); |
678 | ||
83342314 | 679 | /** |
ead04089 | 680 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 681 | * @size: allocation size |
ead04089 REB |
682 | * |
683 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
684 | * mapped to userspace without leaking data. | |
83342314 NP |
685 | */ |
686 | void *vmalloc_32_user(unsigned long size) | |
687 | { | |
688 | struct vm_struct *area; | |
689 | void *ret; | |
690 | ||
0d08e0d3 | 691 | ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL); |
2b4ac44e ED |
692 | if (ret) { |
693 | write_lock(&vmlist_lock); | |
694 | area = __find_vm_area(ret); | |
695 | area->flags |= VM_USERMAP; | |
696 | write_unlock(&vmlist_lock); | |
697 | } | |
83342314 NP |
698 | return ret; |
699 | } | |
700 | EXPORT_SYMBOL(vmalloc_32_user); | |
701 | ||
1da177e4 LT |
702 | long vread(char *buf, char *addr, unsigned long count) |
703 | { | |
704 | struct vm_struct *tmp; | |
705 | char *vaddr, *buf_start = buf; | |
706 | unsigned long n; | |
707 | ||
708 | /* Don't allow overflow */ | |
709 | if ((unsigned long) addr + count < count) | |
710 | count = -(unsigned long) addr; | |
711 | ||
712 | read_lock(&vmlist_lock); | |
713 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
714 | vaddr = (char *) tmp->addr; | |
715 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
716 | continue; | |
717 | while (addr < vaddr) { | |
718 | if (count == 0) | |
719 | goto finished; | |
720 | *buf = '\0'; | |
721 | buf++; | |
722 | addr++; | |
723 | count--; | |
724 | } | |
725 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
726 | do { | |
727 | if (count == 0) | |
728 | goto finished; | |
729 | *buf = *addr; | |
730 | buf++; | |
731 | addr++; | |
732 | count--; | |
733 | } while (--n > 0); | |
734 | } | |
735 | finished: | |
736 | read_unlock(&vmlist_lock); | |
737 | return buf - buf_start; | |
738 | } | |
739 | ||
740 | long vwrite(char *buf, char *addr, unsigned long count) | |
741 | { | |
742 | struct vm_struct *tmp; | |
743 | char *vaddr, *buf_start = buf; | |
744 | unsigned long n; | |
745 | ||
746 | /* Don't allow overflow */ | |
747 | if ((unsigned long) addr + count < count) | |
748 | count = -(unsigned long) addr; | |
749 | ||
750 | read_lock(&vmlist_lock); | |
751 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
752 | vaddr = (char *) tmp->addr; | |
753 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
754 | continue; | |
755 | while (addr < vaddr) { | |
756 | if (count == 0) | |
757 | goto finished; | |
758 | buf++; | |
759 | addr++; | |
760 | count--; | |
761 | } | |
762 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
763 | do { | |
764 | if (count == 0) | |
765 | goto finished; | |
766 | *addr = *buf; | |
767 | buf++; | |
768 | addr++; | |
769 | count--; | |
770 | } while (--n > 0); | |
771 | } | |
772 | finished: | |
773 | read_unlock(&vmlist_lock); | |
774 | return buf - buf_start; | |
775 | } | |
83342314 NP |
776 | |
777 | /** | |
778 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
779 | * @vma: vma to cover (map full range of vma) |
780 | * @addr: vmalloc memory | |
781 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
782 | * |
783 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
784 | * |
785 | * This function checks that addr is a valid vmalloc'ed area, and | |
786 | * that it is big enough to cover the vma. Will return failure if | |
787 | * that criteria isn't met. | |
788 | * | |
72fd4a35 | 789 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
790 | */ |
791 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
792 | unsigned long pgoff) | |
793 | { | |
794 | struct vm_struct *area; | |
795 | unsigned long uaddr = vma->vm_start; | |
796 | unsigned long usize = vma->vm_end - vma->vm_start; | |
797 | int ret; | |
798 | ||
799 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
800 | return -EINVAL; | |
801 | ||
802 | read_lock(&vmlist_lock); | |
803 | area = __find_vm_area(addr); | |
804 | if (!area) | |
805 | goto out_einval_locked; | |
806 | ||
807 | if (!(area->flags & VM_USERMAP)) | |
808 | goto out_einval_locked; | |
809 | ||
810 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
811 | goto out_einval_locked; | |
812 | read_unlock(&vmlist_lock); | |
813 | ||
814 | addr += pgoff << PAGE_SHIFT; | |
815 | do { | |
816 | struct page *page = vmalloc_to_page(addr); | |
817 | ret = vm_insert_page(vma, uaddr, page); | |
818 | if (ret) | |
819 | return ret; | |
820 | ||
821 | uaddr += PAGE_SIZE; | |
822 | addr += PAGE_SIZE; | |
823 | usize -= PAGE_SIZE; | |
824 | } while (usize > 0); | |
825 | ||
826 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
827 | vma->vm_flags |= VM_RESERVED; | |
828 | ||
829 | return ret; | |
830 | ||
831 | out_einval_locked: | |
832 | read_unlock(&vmlist_lock); | |
833 | return -EINVAL; | |
834 | } | |
835 | EXPORT_SYMBOL(remap_vmalloc_range); | |
836 | ||
1eeb66a1 CH |
837 | /* |
838 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
839 | * have one. | |
840 | */ | |
841 | void __attribute__((weak)) vmalloc_sync_all(void) | |
842 | { | |
843 | } | |
5f4352fb JF |
844 | |
845 | ||
2f569afd | 846 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
847 | { |
848 | /* apply_to_page_range() does all the hard work. */ | |
849 | return 0; | |
850 | } | |
851 | ||
852 | /** | |
853 | * alloc_vm_area - allocate a range of kernel address space | |
854 | * @size: size of the area | |
7682486b RD |
855 | * |
856 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
857 | * |
858 | * This function reserves a range of kernel address space, and | |
859 | * allocates pagetables to map that range. No actual mappings | |
860 | * are created. If the kernel address space is not shared | |
861 | * between processes, it syncs the pagetable across all | |
862 | * processes. | |
863 | */ | |
864 | struct vm_struct *alloc_vm_area(size_t size) | |
865 | { | |
866 | struct vm_struct *area; | |
867 | ||
23016969 CL |
868 | area = get_vm_area_caller(size, VM_IOREMAP, |
869 | __builtin_return_address(0)); | |
5f4352fb JF |
870 | if (area == NULL) |
871 | return NULL; | |
872 | ||
873 | /* | |
874 | * This ensures that page tables are constructed for this region | |
875 | * of kernel virtual address space and mapped into init_mm. | |
876 | */ | |
877 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
878 | area->size, f, NULL)) { | |
879 | free_vm_area(area); | |
880 | return NULL; | |
881 | } | |
882 | ||
883 | /* Make sure the pagetables are constructed in process kernel | |
884 | mappings */ | |
885 | vmalloc_sync_all(); | |
886 | ||
887 | return area; | |
888 | } | |
889 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
890 | ||
891 | void free_vm_area(struct vm_struct *area) | |
892 | { | |
893 | struct vm_struct *ret; | |
894 | ret = remove_vm_area(area->addr); | |
895 | BUG_ON(ret != area); | |
896 | kfree(area); | |
897 | } | |
898 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 CL |
899 | |
900 | ||
901 | #ifdef CONFIG_PROC_FS | |
902 | static void *s_start(struct seq_file *m, loff_t *pos) | |
903 | { | |
904 | loff_t n = *pos; | |
905 | struct vm_struct *v; | |
906 | ||
907 | read_lock(&vmlist_lock); | |
908 | v = vmlist; | |
909 | while (n > 0 && v) { | |
910 | n--; | |
911 | v = v->next; | |
912 | } | |
913 | if (!n) | |
914 | return v; | |
915 | ||
916 | return NULL; | |
917 | ||
918 | } | |
919 | ||
920 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
921 | { | |
922 | struct vm_struct *v = p; | |
923 | ||
924 | ++*pos; | |
925 | return v->next; | |
926 | } | |
927 | ||
928 | static void s_stop(struct seq_file *m, void *p) | |
929 | { | |
930 | read_unlock(&vmlist_lock); | |
931 | } | |
932 | ||
933 | static int s_show(struct seq_file *m, void *p) | |
934 | { | |
935 | struct vm_struct *v = p; | |
936 | ||
937 | seq_printf(m, "0x%p-0x%p %7ld", | |
938 | v->addr, v->addr + v->size, v->size); | |
939 | ||
23016969 CL |
940 | if (v->caller) { |
941 | char buff[2 * KSYM_NAME_LEN]; | |
942 | ||
943 | seq_putc(m, ' '); | |
944 | sprint_symbol(buff, (unsigned long)v->caller); | |
945 | seq_puts(m, buff); | |
946 | } | |
947 | ||
a10aa579 CL |
948 | if (v->nr_pages) |
949 | seq_printf(m, " pages=%d", v->nr_pages); | |
950 | ||
951 | if (v->phys_addr) | |
952 | seq_printf(m, " phys=%lx", v->phys_addr); | |
953 | ||
954 | if (v->flags & VM_IOREMAP) | |
955 | seq_printf(m, " ioremap"); | |
956 | ||
957 | if (v->flags & VM_ALLOC) | |
958 | seq_printf(m, " vmalloc"); | |
959 | ||
960 | if (v->flags & VM_MAP) | |
961 | seq_printf(m, " vmap"); | |
962 | ||
963 | if (v->flags & VM_USERMAP) | |
964 | seq_printf(m, " user"); | |
965 | ||
966 | if (v->flags & VM_VPAGES) | |
967 | seq_printf(m, " vpages"); | |
968 | ||
969 | seq_putc(m, '\n'); | |
970 | return 0; | |
971 | } | |
972 | ||
973 | const struct seq_operations vmalloc_op = { | |
974 | .start = s_start, | |
975 | .next = s_next, | |
976 | .stop = s_stop, | |
977 | .show = s_show, | |
978 | }; | |
979 | #endif | |
980 |