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) { | |
4c8573e2 | 384 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
385 | return; |
386 | } | |
387 | ||
388 | area = remove_vm_area(addr); | |
389 | if (unlikely(!area)) { | |
4c8573e2 | 390 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 391 | addr); |
1da177e4 LT |
392 | return; |
393 | } | |
394 | ||
9a11b49a | 395 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 396 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 397 | |
1da177e4 LT |
398 | if (deallocate_pages) { |
399 | int i; | |
400 | ||
401 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
402 | struct page *page = area->pages[i]; |
403 | ||
404 | BUG_ON(!page); | |
405 | __free_page(page); | |
1da177e4 LT |
406 | } |
407 | ||
8757d5fa | 408 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
409 | vfree(area->pages); |
410 | else | |
411 | kfree(area->pages); | |
412 | } | |
413 | ||
414 | kfree(area); | |
415 | return; | |
416 | } | |
417 | ||
418 | /** | |
419 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
420 | * @addr: memory base address |
421 | * | |
183ff22b | 422 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
423 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
424 | * NULL, no operation is performed. | |
1da177e4 | 425 | * |
80e93eff | 426 | * Must not be called in interrupt context. |
1da177e4 | 427 | */ |
b3bdda02 | 428 | void vfree(const void *addr) |
1da177e4 LT |
429 | { |
430 | BUG_ON(in_interrupt()); | |
431 | __vunmap(addr, 1); | |
432 | } | |
1da177e4 LT |
433 | EXPORT_SYMBOL(vfree); |
434 | ||
435 | /** | |
436 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
437 | * @addr: memory base address |
438 | * | |
439 | * Free the virtually contiguous memory area starting at @addr, | |
440 | * which was created from the page array passed to vmap(). | |
441 | * | |
80e93eff | 442 | * Must not be called in interrupt context. |
1da177e4 | 443 | */ |
b3bdda02 | 444 | void vunmap(const void *addr) |
1da177e4 LT |
445 | { |
446 | BUG_ON(in_interrupt()); | |
447 | __vunmap(addr, 0); | |
448 | } | |
1da177e4 LT |
449 | EXPORT_SYMBOL(vunmap); |
450 | ||
451 | /** | |
452 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
453 | * @pages: array of page pointers |
454 | * @count: number of pages to map | |
455 | * @flags: vm_area->flags | |
456 | * @prot: page protection for the mapping | |
457 | * | |
458 | * Maps @count pages from @pages into contiguous kernel virtual | |
459 | * space. | |
460 | */ | |
461 | void *vmap(struct page **pages, unsigned int count, | |
462 | unsigned long flags, pgprot_t prot) | |
463 | { | |
464 | struct vm_struct *area; | |
465 | ||
466 | if (count > num_physpages) | |
467 | return NULL; | |
468 | ||
23016969 CL |
469 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
470 | __builtin_return_address(0)); | |
1da177e4 LT |
471 | if (!area) |
472 | return NULL; | |
23016969 | 473 | |
1da177e4 LT |
474 | if (map_vm_area(area, prot, &pages)) { |
475 | vunmap(area->addr); | |
476 | return NULL; | |
477 | } | |
478 | ||
479 | return area->addr; | |
480 | } | |
1da177e4 LT |
481 | EXPORT_SYMBOL(vmap); |
482 | ||
e31d9eb5 | 483 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 484 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
485 | { |
486 | struct page **pages; | |
487 | unsigned int nr_pages, array_size, i; | |
488 | ||
489 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
490 | array_size = (nr_pages * sizeof(struct page *)); | |
491 | ||
492 | area->nr_pages = nr_pages; | |
493 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 494 | if (array_size > PAGE_SIZE) { |
94f6030c | 495 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
23016969 | 496 | PAGE_KERNEL, node, caller); |
8757d5fa | 497 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
498 | } else { |
499 | pages = kmalloc_node(array_size, | |
6cb06229 | 500 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
501 | node); |
502 | } | |
1da177e4 | 503 | area->pages = pages; |
23016969 | 504 | area->caller = caller; |
1da177e4 LT |
505 | if (!area->pages) { |
506 | remove_vm_area(area->addr); | |
507 | kfree(area); | |
508 | return NULL; | |
509 | } | |
1da177e4 LT |
510 | |
511 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
512 | struct page *page; |
513 | ||
930fc45a | 514 | if (node < 0) |
bf53d6f8 | 515 | page = alloc_page(gfp_mask); |
930fc45a | 516 | else |
bf53d6f8 CL |
517 | page = alloc_pages_node(node, gfp_mask, 0); |
518 | ||
519 | if (unlikely(!page)) { | |
1da177e4 LT |
520 | /* Successfully allocated i pages, free them in __vunmap() */ |
521 | area->nr_pages = i; | |
522 | goto fail; | |
523 | } | |
bf53d6f8 | 524 | area->pages[i] = page; |
1da177e4 LT |
525 | } |
526 | ||
527 | if (map_vm_area(area, prot, &pages)) | |
528 | goto fail; | |
529 | return area->addr; | |
530 | ||
531 | fail: | |
532 | vfree(area->addr); | |
533 | return NULL; | |
534 | } | |
535 | ||
930fc45a CL |
536 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
537 | { | |
23016969 CL |
538 | return __vmalloc_area_node(area, gfp_mask, prot, -1, |
539 | __builtin_return_address(0)); | |
930fc45a CL |
540 | } |
541 | ||
1da177e4 | 542 | /** |
930fc45a | 543 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
544 | * @size: allocation size |
545 | * @gfp_mask: flags for the page level allocator | |
546 | * @prot: protection mask for the allocated pages | |
d44e0780 | 547 | * @node: node to use for allocation or -1 |
c85d194b | 548 | * @caller: caller's return address |
1da177e4 LT |
549 | * |
550 | * Allocate enough pages to cover @size from the page level | |
551 | * allocator with @gfp_mask flags. Map them into contiguous | |
552 | * kernel virtual space, using a pagetable protection of @prot. | |
553 | */ | |
b221385b | 554 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 555 | int node, void *caller) |
1da177e4 LT |
556 | { |
557 | struct vm_struct *area; | |
558 | ||
559 | size = PAGE_ALIGN(size); | |
560 | if (!size || (size >> PAGE_SHIFT) > num_physpages) | |
561 | return NULL; | |
562 | ||
23016969 CL |
563 | area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END, |
564 | node, gfp_mask, caller); | |
565 | ||
1da177e4 LT |
566 | if (!area) |
567 | return NULL; | |
568 | ||
23016969 | 569 | return __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1da177e4 LT |
570 | } |
571 | ||
930fc45a CL |
572 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
573 | { | |
23016969 CL |
574 | return __vmalloc_node(size, gfp_mask, prot, -1, |
575 | __builtin_return_address(0)); | |
930fc45a | 576 | } |
1da177e4 LT |
577 | EXPORT_SYMBOL(__vmalloc); |
578 | ||
579 | /** | |
580 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 581 | * @size: allocation size |
1da177e4 LT |
582 | * Allocate enough pages to cover @size from the page level |
583 | * allocator and map them into contiguous kernel virtual space. | |
584 | * | |
c1c8897f | 585 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
586 | * use __vmalloc() instead. |
587 | */ | |
588 | void *vmalloc(unsigned long size) | |
589 | { | |
23016969 CL |
590 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
591 | -1, __builtin_return_address(0)); | |
1da177e4 | 592 | } |
1da177e4 LT |
593 | EXPORT_SYMBOL(vmalloc); |
594 | ||
83342314 | 595 | /** |
ead04089 REB |
596 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
597 | * @size: allocation size | |
83342314 | 598 | * |
ead04089 REB |
599 | * The resulting memory area is zeroed so it can be mapped to userspace |
600 | * without leaking data. | |
83342314 NP |
601 | */ |
602 | void *vmalloc_user(unsigned long size) | |
603 | { | |
604 | struct vm_struct *area; | |
605 | void *ret; | |
606 | ||
607 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); | |
2b4ac44e ED |
608 | if (ret) { |
609 | write_lock(&vmlist_lock); | |
610 | area = __find_vm_area(ret); | |
611 | area->flags |= VM_USERMAP; | |
612 | write_unlock(&vmlist_lock); | |
613 | } | |
83342314 NP |
614 | return ret; |
615 | } | |
616 | EXPORT_SYMBOL(vmalloc_user); | |
617 | ||
930fc45a CL |
618 | /** |
619 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 620 | * @size: allocation size |
d44e0780 | 621 | * @node: numa node |
930fc45a CL |
622 | * |
623 | * Allocate enough pages to cover @size from the page level | |
624 | * allocator and map them into contiguous kernel virtual space. | |
625 | * | |
c1c8897f | 626 | * For tight control over page level allocator and protection flags |
930fc45a CL |
627 | * use __vmalloc() instead. |
628 | */ | |
629 | void *vmalloc_node(unsigned long size, int node) | |
630 | { | |
23016969 CL |
631 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
632 | node, __builtin_return_address(0)); | |
930fc45a CL |
633 | } |
634 | EXPORT_SYMBOL(vmalloc_node); | |
635 | ||
4dc3b16b PP |
636 | #ifndef PAGE_KERNEL_EXEC |
637 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
638 | #endif | |
639 | ||
1da177e4 LT |
640 | /** |
641 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
642 | * @size: allocation size |
643 | * | |
644 | * Kernel-internal function to allocate enough pages to cover @size | |
645 | * the page level allocator and map them into contiguous and | |
646 | * executable kernel virtual space. | |
647 | * | |
c1c8897f | 648 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
649 | * use __vmalloc() instead. |
650 | */ | |
651 | ||
1da177e4 LT |
652 | void *vmalloc_exec(unsigned long size) |
653 | { | |
654 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
655 | } | |
656 | ||
0d08e0d3 | 657 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 658 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 659 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 660 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
661 | #else |
662 | #define GFP_VMALLOC32 GFP_KERNEL | |
663 | #endif | |
664 | ||
1da177e4 LT |
665 | /** |
666 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
667 | * @size: allocation size |
668 | * | |
669 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
670 | * page level allocator and map them into contiguous kernel virtual space. | |
671 | */ | |
672 | void *vmalloc_32(unsigned long size) | |
673 | { | |
0d08e0d3 | 674 | return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL); |
1da177e4 | 675 | } |
1da177e4 LT |
676 | EXPORT_SYMBOL(vmalloc_32); |
677 | ||
83342314 | 678 | /** |
ead04089 | 679 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 680 | * @size: allocation size |
ead04089 REB |
681 | * |
682 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
683 | * mapped to userspace without leaking data. | |
83342314 NP |
684 | */ |
685 | void *vmalloc_32_user(unsigned long size) | |
686 | { | |
687 | struct vm_struct *area; | |
688 | void *ret; | |
689 | ||
0d08e0d3 | 690 | ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL); |
2b4ac44e ED |
691 | if (ret) { |
692 | write_lock(&vmlist_lock); | |
693 | area = __find_vm_area(ret); | |
694 | area->flags |= VM_USERMAP; | |
695 | write_unlock(&vmlist_lock); | |
696 | } | |
83342314 NP |
697 | return ret; |
698 | } | |
699 | EXPORT_SYMBOL(vmalloc_32_user); | |
700 | ||
1da177e4 LT |
701 | long vread(char *buf, char *addr, unsigned long count) |
702 | { | |
703 | struct vm_struct *tmp; | |
704 | char *vaddr, *buf_start = buf; | |
705 | unsigned long n; | |
706 | ||
707 | /* Don't allow overflow */ | |
708 | if ((unsigned long) addr + count < count) | |
709 | count = -(unsigned long) addr; | |
710 | ||
711 | read_lock(&vmlist_lock); | |
712 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
713 | vaddr = (char *) tmp->addr; | |
714 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
715 | continue; | |
716 | while (addr < vaddr) { | |
717 | if (count == 0) | |
718 | goto finished; | |
719 | *buf = '\0'; | |
720 | buf++; | |
721 | addr++; | |
722 | count--; | |
723 | } | |
724 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
725 | do { | |
726 | if (count == 0) | |
727 | goto finished; | |
728 | *buf = *addr; | |
729 | buf++; | |
730 | addr++; | |
731 | count--; | |
732 | } while (--n > 0); | |
733 | } | |
734 | finished: | |
735 | read_unlock(&vmlist_lock); | |
736 | return buf - buf_start; | |
737 | } | |
738 | ||
739 | long vwrite(char *buf, char *addr, unsigned long count) | |
740 | { | |
741 | struct vm_struct *tmp; | |
742 | char *vaddr, *buf_start = buf; | |
743 | unsigned long n; | |
744 | ||
745 | /* Don't allow overflow */ | |
746 | if ((unsigned long) addr + count < count) | |
747 | count = -(unsigned long) addr; | |
748 | ||
749 | read_lock(&vmlist_lock); | |
750 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
751 | vaddr = (char *) tmp->addr; | |
752 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
753 | continue; | |
754 | while (addr < vaddr) { | |
755 | if (count == 0) | |
756 | goto finished; | |
757 | buf++; | |
758 | addr++; | |
759 | count--; | |
760 | } | |
761 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
762 | do { | |
763 | if (count == 0) | |
764 | goto finished; | |
765 | *addr = *buf; | |
766 | buf++; | |
767 | addr++; | |
768 | count--; | |
769 | } while (--n > 0); | |
770 | } | |
771 | finished: | |
772 | read_unlock(&vmlist_lock); | |
773 | return buf - buf_start; | |
774 | } | |
83342314 NP |
775 | |
776 | /** | |
777 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
778 | * @vma: vma to cover (map full range of vma) |
779 | * @addr: vmalloc memory | |
780 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
781 | * |
782 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
783 | * |
784 | * This function checks that addr is a valid vmalloc'ed area, and | |
785 | * that it is big enough to cover the vma. Will return failure if | |
786 | * that criteria isn't met. | |
787 | * | |
72fd4a35 | 788 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
789 | */ |
790 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
791 | unsigned long pgoff) | |
792 | { | |
793 | struct vm_struct *area; | |
794 | unsigned long uaddr = vma->vm_start; | |
795 | unsigned long usize = vma->vm_end - vma->vm_start; | |
796 | int ret; | |
797 | ||
798 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
799 | return -EINVAL; | |
800 | ||
801 | read_lock(&vmlist_lock); | |
802 | area = __find_vm_area(addr); | |
803 | if (!area) | |
804 | goto out_einval_locked; | |
805 | ||
806 | if (!(area->flags & VM_USERMAP)) | |
807 | goto out_einval_locked; | |
808 | ||
809 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
810 | goto out_einval_locked; | |
811 | read_unlock(&vmlist_lock); | |
812 | ||
813 | addr += pgoff << PAGE_SHIFT; | |
814 | do { | |
815 | struct page *page = vmalloc_to_page(addr); | |
816 | ret = vm_insert_page(vma, uaddr, page); | |
817 | if (ret) | |
818 | return ret; | |
819 | ||
820 | uaddr += PAGE_SIZE; | |
821 | addr += PAGE_SIZE; | |
822 | usize -= PAGE_SIZE; | |
823 | } while (usize > 0); | |
824 | ||
825 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
826 | vma->vm_flags |= VM_RESERVED; | |
827 | ||
828 | return ret; | |
829 | ||
830 | out_einval_locked: | |
831 | read_unlock(&vmlist_lock); | |
832 | return -EINVAL; | |
833 | } | |
834 | EXPORT_SYMBOL(remap_vmalloc_range); | |
835 | ||
1eeb66a1 CH |
836 | /* |
837 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
838 | * have one. | |
839 | */ | |
840 | void __attribute__((weak)) vmalloc_sync_all(void) | |
841 | { | |
842 | } | |
5f4352fb JF |
843 | |
844 | ||
2f569afd | 845 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
846 | { |
847 | /* apply_to_page_range() does all the hard work. */ | |
848 | return 0; | |
849 | } | |
850 | ||
851 | /** | |
852 | * alloc_vm_area - allocate a range of kernel address space | |
853 | * @size: size of the area | |
7682486b RD |
854 | * |
855 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
856 | * |
857 | * This function reserves a range of kernel address space, and | |
858 | * allocates pagetables to map that range. No actual mappings | |
859 | * are created. If the kernel address space is not shared | |
860 | * between processes, it syncs the pagetable across all | |
861 | * processes. | |
862 | */ | |
863 | struct vm_struct *alloc_vm_area(size_t size) | |
864 | { | |
865 | struct vm_struct *area; | |
866 | ||
23016969 CL |
867 | area = get_vm_area_caller(size, VM_IOREMAP, |
868 | __builtin_return_address(0)); | |
5f4352fb JF |
869 | if (area == NULL) |
870 | return NULL; | |
871 | ||
872 | /* | |
873 | * This ensures that page tables are constructed for this region | |
874 | * of kernel virtual address space and mapped into init_mm. | |
875 | */ | |
876 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
877 | area->size, f, NULL)) { | |
878 | free_vm_area(area); | |
879 | return NULL; | |
880 | } | |
881 | ||
882 | /* Make sure the pagetables are constructed in process kernel | |
883 | mappings */ | |
884 | vmalloc_sync_all(); | |
885 | ||
886 | return area; | |
887 | } | |
888 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
889 | ||
890 | void free_vm_area(struct vm_struct *area) | |
891 | { | |
892 | struct vm_struct *ret; | |
893 | ret = remove_vm_area(area->addr); | |
894 | BUG_ON(ret != area); | |
895 | kfree(area); | |
896 | } | |
897 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 CL |
898 | |
899 | ||
900 | #ifdef CONFIG_PROC_FS | |
901 | static void *s_start(struct seq_file *m, loff_t *pos) | |
902 | { | |
903 | loff_t n = *pos; | |
904 | struct vm_struct *v; | |
905 | ||
906 | read_lock(&vmlist_lock); | |
907 | v = vmlist; | |
908 | while (n > 0 && v) { | |
909 | n--; | |
910 | v = v->next; | |
911 | } | |
912 | if (!n) | |
913 | return v; | |
914 | ||
915 | return NULL; | |
916 | ||
917 | } | |
918 | ||
919 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
920 | { | |
921 | struct vm_struct *v = p; | |
922 | ||
923 | ++*pos; | |
924 | return v->next; | |
925 | } | |
926 | ||
927 | static void s_stop(struct seq_file *m, void *p) | |
928 | { | |
929 | read_unlock(&vmlist_lock); | |
930 | } | |
931 | ||
a47a126a ED |
932 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
933 | { | |
934 | if (NUMA_BUILD) { | |
935 | unsigned int nr, *counters = m->private; | |
936 | ||
937 | if (!counters) | |
938 | return; | |
939 | ||
940 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); | |
941 | ||
942 | for (nr = 0; nr < v->nr_pages; nr++) | |
943 | counters[page_to_nid(v->pages[nr])]++; | |
944 | ||
945 | for_each_node_state(nr, N_HIGH_MEMORY) | |
946 | if (counters[nr]) | |
947 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
948 | } | |
949 | } | |
950 | ||
a10aa579 CL |
951 | static int s_show(struct seq_file *m, void *p) |
952 | { | |
953 | struct vm_struct *v = p; | |
954 | ||
955 | seq_printf(m, "0x%p-0x%p %7ld", | |
956 | v->addr, v->addr + v->size, v->size); | |
957 | ||
23016969 CL |
958 | if (v->caller) { |
959 | char buff[2 * KSYM_NAME_LEN]; | |
960 | ||
961 | seq_putc(m, ' '); | |
962 | sprint_symbol(buff, (unsigned long)v->caller); | |
963 | seq_puts(m, buff); | |
964 | } | |
965 | ||
a10aa579 CL |
966 | if (v->nr_pages) |
967 | seq_printf(m, " pages=%d", v->nr_pages); | |
968 | ||
969 | if (v->phys_addr) | |
970 | seq_printf(m, " phys=%lx", v->phys_addr); | |
971 | ||
972 | if (v->flags & VM_IOREMAP) | |
973 | seq_printf(m, " ioremap"); | |
974 | ||
975 | if (v->flags & VM_ALLOC) | |
976 | seq_printf(m, " vmalloc"); | |
977 | ||
978 | if (v->flags & VM_MAP) | |
979 | seq_printf(m, " vmap"); | |
980 | ||
981 | if (v->flags & VM_USERMAP) | |
982 | seq_printf(m, " user"); | |
983 | ||
984 | if (v->flags & VM_VPAGES) | |
985 | seq_printf(m, " vpages"); | |
986 | ||
a47a126a | 987 | show_numa_info(m, v); |
a10aa579 CL |
988 | seq_putc(m, '\n'); |
989 | return 0; | |
990 | } | |
991 | ||
992 | const struct seq_operations vmalloc_op = { | |
993 | .start = s_start, | |
994 | .next = s_next, | |
995 | .stop = s_stop, | |
996 | .show = s_show, | |
997 | }; | |
998 | #endif | |
999 |