]>
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 | ||
db64fe02 | 11 | #include <linux/vmalloc.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> | |
14 | #include <linux/highmem.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/spinlock.h> | |
17 | #include <linux/interrupt.h> | |
5f6a6a9c | 18 | #include <linux/proc_fs.h> |
a10aa579 | 19 | #include <linux/seq_file.h> |
3ac7fe5a | 20 | #include <linux/debugobjects.h> |
23016969 | 21 | #include <linux/kallsyms.h> |
db64fe02 NP |
22 | #include <linux/list.h> |
23 | #include <linux/rbtree.h> | |
24 | #include <linux/radix-tree.h> | |
25 | #include <linux/rcupdate.h> | |
1da177e4 | 26 | |
db64fe02 | 27 | #include <asm/atomic.h> |
1da177e4 LT |
28 | #include <asm/uaccess.h> |
29 | #include <asm/tlbflush.h> | |
30 | ||
31 | ||
db64fe02 | 32 | /*** Page table manipulation functions ***/ |
b221385b | 33 | |
1da177e4 LT |
34 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
35 | { | |
36 | pte_t *pte; | |
37 | ||
38 | pte = pte_offset_kernel(pmd, addr); | |
39 | do { | |
40 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
41 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
42 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
43 | } | |
44 | ||
db64fe02 | 45 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
46 | { |
47 | pmd_t *pmd; | |
48 | unsigned long next; | |
49 | ||
50 | pmd = pmd_offset(pud, addr); | |
51 | do { | |
52 | next = pmd_addr_end(addr, end); | |
53 | if (pmd_none_or_clear_bad(pmd)) | |
54 | continue; | |
55 | vunmap_pte_range(pmd, addr, next); | |
56 | } while (pmd++, addr = next, addr != end); | |
57 | } | |
58 | ||
db64fe02 | 59 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
1da177e4 LT |
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 | ||
db64fe02 | 73 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
74 | { |
75 | pgd_t *pgd; | |
76 | unsigned long next; | |
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); | |
1da177e4 LT |
87 | } |
88 | ||
89 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 90 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
91 | { |
92 | pte_t *pte; | |
93 | ||
db64fe02 NP |
94 | /* |
95 | * nr is a running index into the array which helps higher level | |
96 | * callers keep track of where we're up to. | |
97 | */ | |
98 | ||
872fec16 | 99 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
100 | if (!pte) |
101 | return -ENOMEM; | |
102 | do { | |
db64fe02 NP |
103 | struct page *page = pages[*nr]; |
104 | ||
105 | if (WARN_ON(!pte_none(*pte))) | |
106 | return -EBUSY; | |
107 | if (WARN_ON(!page)) | |
1da177e4 LT |
108 | return -ENOMEM; |
109 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 110 | (*nr)++; |
1da177e4 LT |
111 | } while (pte++, addr += PAGE_SIZE, addr != end); |
112 | return 0; | |
113 | } | |
114 | ||
db64fe02 NP |
115 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
116 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
117 | { |
118 | pmd_t *pmd; | |
119 | unsigned long next; | |
120 | ||
121 | pmd = pmd_alloc(&init_mm, pud, addr); | |
122 | if (!pmd) | |
123 | return -ENOMEM; | |
124 | do { | |
125 | next = pmd_addr_end(addr, end); | |
db64fe02 | 126 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
127 | return -ENOMEM; |
128 | } while (pmd++, addr = next, addr != end); | |
129 | return 0; | |
130 | } | |
131 | ||
db64fe02 NP |
132 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
133 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
134 | { |
135 | pud_t *pud; | |
136 | unsigned long next; | |
137 | ||
138 | pud = pud_alloc(&init_mm, pgd, addr); | |
139 | if (!pud) | |
140 | return -ENOMEM; | |
141 | do { | |
142 | next = pud_addr_end(addr, end); | |
db64fe02 | 143 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
144 | return -ENOMEM; |
145 | } while (pud++, addr = next, addr != end); | |
146 | return 0; | |
147 | } | |
148 | ||
db64fe02 NP |
149 | /* |
150 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
151 | * will have pfns corresponding to the "pages" array. | |
152 | * | |
153 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
154 | */ | |
155 | static int vmap_page_range(unsigned long addr, unsigned long end, | |
156 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
157 | { |
158 | pgd_t *pgd; | |
159 | unsigned long next; | |
db64fe02 NP |
160 | int err = 0; |
161 | int nr = 0; | |
1da177e4 LT |
162 | |
163 | BUG_ON(addr >= end); | |
164 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
165 | do { |
166 | next = pgd_addr_end(addr, end); | |
db64fe02 | 167 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 LT |
168 | if (err) |
169 | break; | |
170 | } while (pgd++, addr = next, addr != end); | |
db64fe02 NP |
171 | flush_cache_vmap(addr, end); |
172 | ||
173 | if (unlikely(err)) | |
174 | return err; | |
175 | return nr; | |
1da177e4 LT |
176 | } |
177 | ||
73bdf0a6 LT |
178 | static inline int is_vmalloc_or_module_addr(const void *x) |
179 | { | |
180 | /* | |
ab4f2ee1 | 181 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
182 | * and fall back on vmalloc() if that fails. Others |
183 | * just put it in the vmalloc space. | |
184 | */ | |
185 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
186 | unsigned long addr = (unsigned long)x; | |
187 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
188 | return 1; | |
189 | #endif | |
190 | return is_vmalloc_addr(x); | |
191 | } | |
192 | ||
48667e7a | 193 | /* |
db64fe02 | 194 | * Walk a vmap address to the struct page it maps. |
48667e7a | 195 | */ |
b3bdda02 | 196 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
197 | { |
198 | unsigned long addr = (unsigned long) vmalloc_addr; | |
199 | struct page *page = NULL; | |
200 | pgd_t *pgd = pgd_offset_k(addr); | |
48667e7a | 201 | |
7aa413de IM |
202 | /* |
203 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
204 | * architectures that do not vmalloc module space | |
205 | */ | |
73bdf0a6 | 206 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 207 | |
48667e7a | 208 | if (!pgd_none(*pgd)) { |
db64fe02 | 209 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 210 | if (!pud_none(*pud)) { |
db64fe02 | 211 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 212 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
213 | pte_t *ptep, pte; |
214 | ||
48667e7a CL |
215 | ptep = pte_offset_map(pmd, addr); |
216 | pte = *ptep; | |
217 | if (pte_present(pte)) | |
218 | page = pte_page(pte); | |
219 | pte_unmap(ptep); | |
220 | } | |
221 | } | |
222 | } | |
223 | return page; | |
224 | } | |
225 | EXPORT_SYMBOL(vmalloc_to_page); | |
226 | ||
227 | /* | |
228 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
229 | */ | |
b3bdda02 | 230 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
231 | { |
232 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
233 | } | |
234 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
235 | ||
db64fe02 NP |
236 | |
237 | /*** Global kva allocator ***/ | |
238 | ||
239 | #define VM_LAZY_FREE 0x01 | |
240 | #define VM_LAZY_FREEING 0x02 | |
241 | #define VM_VM_AREA 0x04 | |
242 | ||
243 | struct vmap_area { | |
244 | unsigned long va_start; | |
245 | unsigned long va_end; | |
246 | unsigned long flags; | |
247 | struct rb_node rb_node; /* address sorted rbtree */ | |
248 | struct list_head list; /* address sorted list */ | |
249 | struct list_head purge_list; /* "lazy purge" list */ | |
250 | void *private; | |
251 | struct rcu_head rcu_head; | |
252 | }; | |
253 | ||
254 | static DEFINE_SPINLOCK(vmap_area_lock); | |
255 | static struct rb_root vmap_area_root = RB_ROOT; | |
256 | static LIST_HEAD(vmap_area_list); | |
257 | ||
258 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 259 | { |
db64fe02 NP |
260 | struct rb_node *n = vmap_area_root.rb_node; |
261 | ||
262 | while (n) { | |
263 | struct vmap_area *va; | |
264 | ||
265 | va = rb_entry(n, struct vmap_area, rb_node); | |
266 | if (addr < va->va_start) | |
267 | n = n->rb_left; | |
268 | else if (addr > va->va_start) | |
269 | n = n->rb_right; | |
270 | else | |
271 | return va; | |
272 | } | |
273 | ||
274 | return NULL; | |
275 | } | |
276 | ||
277 | static void __insert_vmap_area(struct vmap_area *va) | |
278 | { | |
279 | struct rb_node **p = &vmap_area_root.rb_node; | |
280 | struct rb_node *parent = NULL; | |
281 | struct rb_node *tmp; | |
282 | ||
283 | while (*p) { | |
284 | struct vmap_area *tmp; | |
285 | ||
286 | parent = *p; | |
287 | tmp = rb_entry(parent, struct vmap_area, rb_node); | |
288 | if (va->va_start < tmp->va_end) | |
289 | p = &(*p)->rb_left; | |
290 | else if (va->va_end > tmp->va_start) | |
291 | p = &(*p)->rb_right; | |
292 | else | |
293 | BUG(); | |
294 | } | |
295 | ||
296 | rb_link_node(&va->rb_node, parent, p); | |
297 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
298 | ||
299 | /* address-sort this list so it is usable like the vmlist */ | |
300 | tmp = rb_prev(&va->rb_node); | |
301 | if (tmp) { | |
302 | struct vmap_area *prev; | |
303 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
304 | list_add_rcu(&va->list, &prev->list); | |
305 | } else | |
306 | list_add_rcu(&va->list, &vmap_area_list); | |
307 | } | |
308 | ||
309 | static void purge_vmap_area_lazy(void); | |
310 | ||
311 | /* | |
312 | * Allocate a region of KVA of the specified size and alignment, within the | |
313 | * vstart and vend. | |
314 | */ | |
315 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
316 | unsigned long align, | |
317 | unsigned long vstart, unsigned long vend, | |
318 | int node, gfp_t gfp_mask) | |
319 | { | |
320 | struct vmap_area *va; | |
321 | struct rb_node *n; | |
1da177e4 | 322 | unsigned long addr; |
db64fe02 NP |
323 | int purged = 0; |
324 | ||
325 | BUG_ON(size & ~PAGE_MASK); | |
326 | ||
327 | addr = ALIGN(vstart, align); | |
328 | ||
329 | va = kmalloc_node(sizeof(struct vmap_area), | |
330 | gfp_mask & GFP_RECLAIM_MASK, node); | |
331 | if (unlikely(!va)) | |
332 | return ERR_PTR(-ENOMEM); | |
333 | ||
334 | retry: | |
335 | spin_lock(&vmap_area_lock); | |
336 | /* XXX: could have a last_hole cache */ | |
337 | n = vmap_area_root.rb_node; | |
338 | if (n) { | |
339 | struct vmap_area *first = NULL; | |
340 | ||
341 | do { | |
342 | struct vmap_area *tmp; | |
343 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
344 | if (tmp->va_end >= addr) { | |
345 | if (!first && tmp->va_start < addr + size) | |
346 | first = tmp; | |
347 | n = n->rb_left; | |
348 | } else { | |
349 | first = tmp; | |
350 | n = n->rb_right; | |
351 | } | |
352 | } while (n); | |
353 | ||
354 | if (!first) | |
355 | goto found; | |
356 | ||
357 | if (first->va_end < addr) { | |
358 | n = rb_next(&first->rb_node); | |
359 | if (n) | |
360 | first = rb_entry(n, struct vmap_area, rb_node); | |
361 | else | |
362 | goto found; | |
363 | } | |
364 | ||
365 | while (addr + size >= first->va_start && addr + size <= vend) { | |
366 | addr = ALIGN(first->va_end + PAGE_SIZE, align); | |
367 | ||
368 | n = rb_next(&first->rb_node); | |
369 | if (n) | |
370 | first = rb_entry(n, struct vmap_area, rb_node); | |
371 | else | |
372 | goto found; | |
373 | } | |
374 | } | |
375 | found: | |
376 | if (addr + size > vend) { | |
377 | spin_unlock(&vmap_area_lock); | |
378 | if (!purged) { | |
379 | purge_vmap_area_lazy(); | |
380 | purged = 1; | |
381 | goto retry; | |
382 | } | |
383 | if (printk_ratelimit()) | |
384 | printk(KERN_WARNING "vmap allocation failed: " | |
385 | "use vmalloc=<size> to increase size.\n"); | |
386 | return ERR_PTR(-EBUSY); | |
387 | } | |
388 | ||
389 | BUG_ON(addr & (align-1)); | |
390 | ||
391 | va->va_start = addr; | |
392 | va->va_end = addr + size; | |
393 | va->flags = 0; | |
394 | __insert_vmap_area(va); | |
395 | spin_unlock(&vmap_area_lock); | |
396 | ||
397 | return va; | |
398 | } | |
399 | ||
400 | static void rcu_free_va(struct rcu_head *head) | |
401 | { | |
402 | struct vmap_area *va = container_of(head, struct vmap_area, rcu_head); | |
403 | ||
404 | kfree(va); | |
405 | } | |
406 | ||
407 | static void __free_vmap_area(struct vmap_area *va) | |
408 | { | |
409 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
410 | rb_erase(&va->rb_node, &vmap_area_root); | |
411 | RB_CLEAR_NODE(&va->rb_node); | |
412 | list_del_rcu(&va->list); | |
413 | ||
414 | call_rcu(&va->rcu_head, rcu_free_va); | |
415 | } | |
416 | ||
417 | /* | |
418 | * Free a region of KVA allocated by alloc_vmap_area | |
419 | */ | |
420 | static void free_vmap_area(struct vmap_area *va) | |
421 | { | |
422 | spin_lock(&vmap_area_lock); | |
423 | __free_vmap_area(va); | |
424 | spin_unlock(&vmap_area_lock); | |
425 | } | |
426 | ||
427 | /* | |
428 | * Clear the pagetable entries of a given vmap_area | |
429 | */ | |
430 | static void unmap_vmap_area(struct vmap_area *va) | |
431 | { | |
432 | vunmap_page_range(va->va_start, va->va_end); | |
433 | } | |
434 | ||
435 | /* | |
436 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
437 | * before attempting to purge with a TLB flush. | |
438 | * | |
439 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
440 | * and take slightly longer to purge, but it will linearly reduce the number of | |
441 | * global TLB flushes that must be performed. It would seem natural to scale | |
442 | * this number up linearly with the number of CPUs (because vmapping activity | |
443 | * could also scale linearly with the number of CPUs), however it is likely | |
444 | * that in practice, workloads might be constrained in other ways that mean | |
445 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
446 | * conservative and not introduce a big latency on huge systems, so go with | |
447 | * a less aggressive log scale. It will still be an improvement over the old | |
448 | * code, and it will be simple to change the scale factor if we find that it | |
449 | * becomes a problem on bigger systems. | |
450 | */ | |
451 | static unsigned long lazy_max_pages(void) | |
452 | { | |
453 | unsigned int log; | |
454 | ||
455 | log = fls(num_online_cpus()); | |
456 | ||
457 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
458 | } | |
459 | ||
460 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
461 | ||
462 | /* | |
463 | * Purges all lazily-freed vmap areas. | |
464 | * | |
465 | * If sync is 0 then don't purge if there is already a purge in progress. | |
466 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
467 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
468 | * their own TLB flushing). | |
469 | * Returns with *start = min(*start, lowest purged address) | |
470 | * *end = max(*end, highest purged address) | |
471 | */ | |
472 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
473 | int sync, int force_flush) | |
474 | { | |
475 | static DEFINE_SPINLOCK(purge_lock); | |
476 | LIST_HEAD(valist); | |
477 | struct vmap_area *va; | |
478 | int nr = 0; | |
479 | ||
480 | /* | |
481 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
482 | * should not expect such behaviour. This just simplifies locking for | |
483 | * the case that isn't actually used at the moment anyway. | |
484 | */ | |
485 | if (!sync && !force_flush) { | |
486 | if (!spin_trylock(&purge_lock)) | |
487 | return; | |
488 | } else | |
489 | spin_lock(&purge_lock); | |
490 | ||
491 | rcu_read_lock(); | |
492 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
493 | if (va->flags & VM_LAZY_FREE) { | |
494 | if (va->va_start < *start) | |
495 | *start = va->va_start; | |
496 | if (va->va_end > *end) | |
497 | *end = va->va_end; | |
498 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
499 | unmap_vmap_area(va); | |
500 | list_add_tail(&va->purge_list, &valist); | |
501 | va->flags |= VM_LAZY_FREEING; | |
502 | va->flags &= ~VM_LAZY_FREE; | |
503 | } | |
504 | } | |
505 | rcu_read_unlock(); | |
506 | ||
507 | if (nr) { | |
508 | BUG_ON(nr > atomic_read(&vmap_lazy_nr)); | |
509 | atomic_sub(nr, &vmap_lazy_nr); | |
510 | } | |
511 | ||
512 | if (nr || force_flush) | |
513 | flush_tlb_kernel_range(*start, *end); | |
514 | ||
515 | if (nr) { | |
516 | spin_lock(&vmap_area_lock); | |
517 | list_for_each_entry(va, &valist, purge_list) | |
518 | __free_vmap_area(va); | |
519 | spin_unlock(&vmap_area_lock); | |
520 | } | |
521 | spin_unlock(&purge_lock); | |
522 | } | |
523 | ||
524 | /* | |
525 | * Kick off a purge of the outstanding lazy areas. | |
526 | */ | |
527 | static void purge_vmap_area_lazy(void) | |
528 | { | |
529 | unsigned long start = ULONG_MAX, end = 0; | |
530 | ||
531 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
532 | } | |
533 | ||
534 | /* | |
535 | * Free and unmap a vmap area | |
536 | */ | |
537 | static void free_unmap_vmap_area(struct vmap_area *va) | |
538 | { | |
539 | va->flags |= VM_LAZY_FREE; | |
540 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
541 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
542 | purge_vmap_area_lazy(); | |
543 | } | |
544 | ||
545 | static struct vmap_area *find_vmap_area(unsigned long addr) | |
546 | { | |
547 | struct vmap_area *va; | |
548 | ||
549 | spin_lock(&vmap_area_lock); | |
550 | va = __find_vmap_area(addr); | |
551 | spin_unlock(&vmap_area_lock); | |
552 | ||
553 | return va; | |
554 | } | |
555 | ||
556 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
557 | { | |
558 | struct vmap_area *va; | |
559 | ||
560 | va = find_vmap_area(addr); | |
561 | BUG_ON(!va); | |
562 | free_unmap_vmap_area(va); | |
563 | } | |
564 | ||
565 | ||
566 | /*** Per cpu kva allocator ***/ | |
567 | ||
568 | /* | |
569 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
570 | * room for at least 16 percpu vmap blocks per CPU. | |
571 | */ | |
572 | /* | |
573 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
574 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
575 | * instead (we just need a rough idea) | |
576 | */ | |
577 | #if BITS_PER_LONG == 32 | |
578 | #define VMALLOC_SPACE (128UL*1024*1024) | |
579 | #else | |
580 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
581 | #endif | |
582 | ||
583 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
584 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
585 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
586 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
587 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
588 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
589 | #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
590 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
591 | VMALLOC_PAGES / NR_CPUS / 16)) | |
592 | ||
593 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
594 | ||
9b463334 JF |
595 | static bool vmap_initialized __read_mostly = false; |
596 | ||
db64fe02 NP |
597 | struct vmap_block_queue { |
598 | spinlock_t lock; | |
599 | struct list_head free; | |
600 | struct list_head dirty; | |
601 | unsigned int nr_dirty; | |
602 | }; | |
603 | ||
604 | struct vmap_block { | |
605 | spinlock_t lock; | |
606 | struct vmap_area *va; | |
607 | struct vmap_block_queue *vbq; | |
608 | unsigned long free, dirty; | |
609 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
610 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
611 | union { | |
612 | struct { | |
613 | struct list_head free_list; | |
614 | struct list_head dirty_list; | |
615 | }; | |
616 | struct rcu_head rcu_head; | |
617 | }; | |
618 | }; | |
619 | ||
620 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
621 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
622 | ||
623 | /* | |
624 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
625 | * in the free path. Could get rid of this if we change the API to return a | |
626 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
627 | */ | |
628 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
629 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
630 | ||
631 | /* | |
632 | * We should probably have a fallback mechanism to allocate virtual memory | |
633 | * out of partially filled vmap blocks. However vmap block sizing should be | |
634 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
635 | * big problem. | |
636 | */ | |
637 | ||
638 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
639 | { | |
640 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
641 | addr /= VMAP_BLOCK_SIZE; | |
642 | return addr; | |
643 | } | |
644 | ||
645 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
646 | { | |
647 | struct vmap_block_queue *vbq; | |
648 | struct vmap_block *vb; | |
649 | struct vmap_area *va; | |
650 | unsigned long vb_idx; | |
651 | int node, err; | |
652 | ||
653 | node = numa_node_id(); | |
654 | ||
655 | vb = kmalloc_node(sizeof(struct vmap_block), | |
656 | gfp_mask & GFP_RECLAIM_MASK, node); | |
657 | if (unlikely(!vb)) | |
658 | return ERR_PTR(-ENOMEM); | |
659 | ||
660 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
661 | VMALLOC_START, VMALLOC_END, | |
662 | node, gfp_mask); | |
663 | if (unlikely(IS_ERR(va))) { | |
664 | kfree(vb); | |
665 | return ERR_PTR(PTR_ERR(va)); | |
666 | } | |
667 | ||
668 | err = radix_tree_preload(gfp_mask); | |
669 | if (unlikely(err)) { | |
670 | kfree(vb); | |
671 | free_vmap_area(va); | |
672 | return ERR_PTR(err); | |
673 | } | |
674 | ||
675 | spin_lock_init(&vb->lock); | |
676 | vb->va = va; | |
677 | vb->free = VMAP_BBMAP_BITS; | |
678 | vb->dirty = 0; | |
679 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
680 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
681 | INIT_LIST_HEAD(&vb->free_list); | |
682 | INIT_LIST_HEAD(&vb->dirty_list); | |
683 | ||
684 | vb_idx = addr_to_vb_idx(va->va_start); | |
685 | spin_lock(&vmap_block_tree_lock); | |
686 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
687 | spin_unlock(&vmap_block_tree_lock); | |
688 | BUG_ON(err); | |
689 | radix_tree_preload_end(); | |
690 | ||
691 | vbq = &get_cpu_var(vmap_block_queue); | |
692 | vb->vbq = vbq; | |
693 | spin_lock(&vbq->lock); | |
694 | list_add(&vb->free_list, &vbq->free); | |
695 | spin_unlock(&vbq->lock); | |
696 | put_cpu_var(vmap_cpu_blocks); | |
697 | ||
698 | return vb; | |
699 | } | |
700 | ||
701 | static void rcu_free_vb(struct rcu_head *head) | |
702 | { | |
703 | struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head); | |
704 | ||
705 | kfree(vb); | |
706 | } | |
707 | ||
708 | static void free_vmap_block(struct vmap_block *vb) | |
709 | { | |
710 | struct vmap_block *tmp; | |
711 | unsigned long vb_idx; | |
712 | ||
713 | spin_lock(&vb->vbq->lock); | |
714 | if (!list_empty(&vb->free_list)) | |
715 | list_del(&vb->free_list); | |
716 | if (!list_empty(&vb->dirty_list)) | |
717 | list_del(&vb->dirty_list); | |
718 | spin_unlock(&vb->vbq->lock); | |
719 | ||
720 | vb_idx = addr_to_vb_idx(vb->va->va_start); | |
721 | spin_lock(&vmap_block_tree_lock); | |
722 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
723 | spin_unlock(&vmap_block_tree_lock); | |
724 | BUG_ON(tmp != vb); | |
725 | ||
726 | free_unmap_vmap_area(vb->va); | |
727 | call_rcu(&vb->rcu_head, rcu_free_vb); | |
728 | } | |
729 | ||
730 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) | |
731 | { | |
732 | struct vmap_block_queue *vbq; | |
733 | struct vmap_block *vb; | |
734 | unsigned long addr = 0; | |
735 | unsigned int order; | |
736 | ||
737 | BUG_ON(size & ~PAGE_MASK); | |
738 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
739 | order = get_order(size); | |
740 | ||
741 | again: | |
742 | rcu_read_lock(); | |
743 | vbq = &get_cpu_var(vmap_block_queue); | |
744 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
745 | int i; | |
746 | ||
747 | spin_lock(&vb->lock); | |
748 | i = bitmap_find_free_region(vb->alloc_map, | |
749 | VMAP_BBMAP_BITS, order); | |
750 | ||
751 | if (i >= 0) { | |
752 | addr = vb->va->va_start + (i << PAGE_SHIFT); | |
753 | BUG_ON(addr_to_vb_idx(addr) != | |
754 | addr_to_vb_idx(vb->va->va_start)); | |
755 | vb->free -= 1UL << order; | |
756 | if (vb->free == 0) { | |
757 | spin_lock(&vbq->lock); | |
758 | list_del_init(&vb->free_list); | |
759 | spin_unlock(&vbq->lock); | |
760 | } | |
761 | spin_unlock(&vb->lock); | |
762 | break; | |
763 | } | |
764 | spin_unlock(&vb->lock); | |
765 | } | |
766 | put_cpu_var(vmap_cpu_blocks); | |
767 | rcu_read_unlock(); | |
768 | ||
769 | if (!addr) { | |
770 | vb = new_vmap_block(gfp_mask); | |
771 | if (IS_ERR(vb)) | |
772 | return vb; | |
773 | goto again; | |
774 | } | |
775 | ||
776 | return (void *)addr; | |
777 | } | |
778 | ||
779 | static void vb_free(const void *addr, unsigned long size) | |
780 | { | |
781 | unsigned long offset; | |
782 | unsigned long vb_idx; | |
783 | unsigned int order; | |
784 | struct vmap_block *vb; | |
785 | ||
786 | BUG_ON(size & ~PAGE_MASK); | |
787 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
788 | order = get_order(size); | |
789 | ||
790 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
791 | ||
792 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
793 | rcu_read_lock(); | |
794 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
795 | rcu_read_unlock(); | |
796 | BUG_ON(!vb); | |
797 | ||
798 | spin_lock(&vb->lock); | |
799 | bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order); | |
800 | if (!vb->dirty) { | |
801 | spin_lock(&vb->vbq->lock); | |
802 | list_add(&vb->dirty_list, &vb->vbq->dirty); | |
803 | spin_unlock(&vb->vbq->lock); | |
804 | } | |
805 | vb->dirty += 1UL << order; | |
806 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
807 | BUG_ON(vb->free || !list_empty(&vb->free_list)); | |
808 | spin_unlock(&vb->lock); | |
809 | free_vmap_block(vb); | |
810 | } else | |
811 | spin_unlock(&vb->lock); | |
812 | } | |
813 | ||
814 | /** | |
815 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
816 | * | |
817 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
818 | * to amortize TLB flushing overheads. What this means is that any page you | |
819 | * have now, may, in a former life, have been mapped into kernel virtual | |
820 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
821 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
822 | * | |
823 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
824 | * be sure that none of the pages we have control over will have any aliases | |
825 | * from the vmap layer. | |
826 | */ | |
827 | void vm_unmap_aliases(void) | |
828 | { | |
829 | unsigned long start = ULONG_MAX, end = 0; | |
830 | int cpu; | |
831 | int flush = 0; | |
832 | ||
9b463334 JF |
833 | if (unlikely(!vmap_initialized)) |
834 | return; | |
835 | ||
db64fe02 NP |
836 | for_each_possible_cpu(cpu) { |
837 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
838 | struct vmap_block *vb; | |
839 | ||
840 | rcu_read_lock(); | |
841 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
842 | int i; | |
843 | ||
844 | spin_lock(&vb->lock); | |
845 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
846 | while (i < VMAP_BBMAP_BITS) { | |
847 | unsigned long s, e; | |
848 | int j; | |
849 | j = find_next_zero_bit(vb->dirty_map, | |
850 | VMAP_BBMAP_BITS, i); | |
851 | ||
852 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
853 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
854 | vunmap_page_range(s, e); | |
855 | flush = 1; | |
856 | ||
857 | if (s < start) | |
858 | start = s; | |
859 | if (e > end) | |
860 | end = e; | |
861 | ||
862 | i = j; | |
863 | i = find_next_bit(vb->dirty_map, | |
864 | VMAP_BBMAP_BITS, i); | |
865 | } | |
866 | spin_unlock(&vb->lock); | |
867 | } | |
868 | rcu_read_unlock(); | |
869 | } | |
870 | ||
871 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
872 | } | |
873 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
874 | ||
875 | /** | |
876 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
877 | * @mem: the pointer returned by vm_map_ram | |
878 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
879 | */ | |
880 | void vm_unmap_ram(const void *mem, unsigned int count) | |
881 | { | |
882 | unsigned long size = count << PAGE_SHIFT; | |
883 | unsigned long addr = (unsigned long)mem; | |
884 | ||
885 | BUG_ON(!addr); | |
886 | BUG_ON(addr < VMALLOC_START); | |
887 | BUG_ON(addr > VMALLOC_END); | |
888 | BUG_ON(addr & (PAGE_SIZE-1)); | |
889 | ||
890 | debug_check_no_locks_freed(mem, size); | |
891 | ||
892 | if (likely(count <= VMAP_MAX_ALLOC)) | |
893 | vb_free(mem, size); | |
894 | else | |
895 | free_unmap_vmap_area_addr(addr); | |
896 | } | |
897 | EXPORT_SYMBOL(vm_unmap_ram); | |
898 | ||
899 | /** | |
900 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
901 | * @pages: an array of pointers to the pages to be mapped | |
902 | * @count: number of pages | |
903 | * @node: prefer to allocate data structures on this node | |
904 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad RD |
905 | * |
906 | * Returns: a pointer to the address that has been mapped, or %NULL on failure | |
db64fe02 NP |
907 | */ |
908 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
909 | { | |
910 | unsigned long size = count << PAGE_SHIFT; | |
911 | unsigned long addr; | |
912 | void *mem; | |
913 | ||
914 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
915 | mem = vb_alloc(size, GFP_KERNEL); | |
916 | if (IS_ERR(mem)) | |
917 | return NULL; | |
918 | addr = (unsigned long)mem; | |
919 | } else { | |
920 | struct vmap_area *va; | |
921 | va = alloc_vmap_area(size, PAGE_SIZE, | |
922 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
923 | if (IS_ERR(va)) | |
924 | return NULL; | |
925 | ||
926 | addr = va->va_start; | |
927 | mem = (void *)addr; | |
928 | } | |
929 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
930 | vm_unmap_ram(mem, count); | |
931 | return NULL; | |
932 | } | |
933 | return mem; | |
934 | } | |
935 | EXPORT_SYMBOL(vm_map_ram); | |
936 | ||
937 | void __init vmalloc_init(void) | |
938 | { | |
939 | int i; | |
940 | ||
941 | for_each_possible_cpu(i) { | |
942 | struct vmap_block_queue *vbq; | |
943 | ||
944 | vbq = &per_cpu(vmap_block_queue, i); | |
945 | spin_lock_init(&vbq->lock); | |
946 | INIT_LIST_HEAD(&vbq->free); | |
947 | INIT_LIST_HEAD(&vbq->dirty); | |
948 | vbq->nr_dirty = 0; | |
949 | } | |
9b463334 JF |
950 | |
951 | vmap_initialized = true; | |
db64fe02 NP |
952 | } |
953 | ||
954 | void unmap_kernel_range(unsigned long addr, unsigned long size) | |
955 | { | |
956 | unsigned long end = addr + size; | |
957 | vunmap_page_range(addr, end); | |
958 | flush_tlb_kernel_range(addr, end); | |
959 | } | |
960 | ||
961 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
962 | { | |
963 | unsigned long addr = (unsigned long)area->addr; | |
964 | unsigned long end = addr + area->size - PAGE_SIZE; | |
965 | int err; | |
966 | ||
967 | err = vmap_page_range(addr, end, prot, *pages); | |
968 | if (err > 0) { | |
969 | *pages += err; | |
970 | err = 0; | |
971 | } | |
972 | ||
973 | return err; | |
974 | } | |
975 | EXPORT_SYMBOL_GPL(map_vm_area); | |
976 | ||
977 | /*** Old vmalloc interfaces ***/ | |
978 | DEFINE_RWLOCK(vmlist_lock); | |
979 | struct vm_struct *vmlist; | |
980 | ||
981 | static struct vm_struct *__get_vm_area_node(unsigned long size, | |
982 | unsigned long flags, unsigned long start, unsigned long end, | |
983 | int node, gfp_t gfp_mask, void *caller) | |
984 | { | |
985 | static struct vmap_area *va; | |
986 | struct vm_struct *area; | |
987 | struct vm_struct *tmp, **p; | |
988 | unsigned long align = 1; | |
1da177e4 | 989 | |
52fd24ca | 990 | BUG_ON(in_interrupt()); |
1da177e4 LT |
991 | if (flags & VM_IOREMAP) { |
992 | int bit = fls(size); | |
993 | ||
994 | if (bit > IOREMAP_MAX_ORDER) | |
995 | bit = IOREMAP_MAX_ORDER; | |
996 | else if (bit < PAGE_SHIFT) | |
997 | bit = PAGE_SHIFT; | |
998 | ||
999 | align = 1ul << bit; | |
1000 | } | |
db64fe02 | 1001 | |
1da177e4 | 1002 | size = PAGE_ALIGN(size); |
31be8309 OH |
1003 | if (unlikely(!size)) |
1004 | return NULL; | |
1da177e4 | 1005 | |
6cb06229 | 1006 | area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1007 | if (unlikely(!area)) |
1008 | return NULL; | |
1009 | ||
1da177e4 LT |
1010 | /* |
1011 | * We always allocate a guard page. | |
1012 | */ | |
1013 | size += PAGE_SIZE; | |
1014 | ||
db64fe02 NP |
1015 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1016 | if (IS_ERR(va)) { | |
1017 | kfree(area); | |
1018 | return NULL; | |
1da177e4 | 1019 | } |
1da177e4 LT |
1020 | |
1021 | area->flags = flags; | |
db64fe02 | 1022 | area->addr = (void *)va->va_start; |
1da177e4 LT |
1023 | area->size = size; |
1024 | area->pages = NULL; | |
1025 | area->nr_pages = 0; | |
1026 | area->phys_addr = 0; | |
23016969 | 1027 | area->caller = caller; |
db64fe02 NP |
1028 | va->private = area; |
1029 | va->flags |= VM_VM_AREA; | |
1030 | ||
1031 | write_lock(&vmlist_lock); | |
1032 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1033 | if (tmp->addr >= area->addr) | |
1034 | break; | |
1035 | } | |
1036 | area->next = *p; | |
1037 | *p = area; | |
1da177e4 LT |
1038 | write_unlock(&vmlist_lock); |
1039 | ||
1040 | return area; | |
1da177e4 LT |
1041 | } |
1042 | ||
930fc45a CL |
1043 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1044 | unsigned long start, unsigned long end) | |
1045 | { | |
23016969 CL |
1046 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, |
1047 | __builtin_return_address(0)); | |
930fc45a | 1048 | } |
5992b6da | 1049 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1050 | |
1da177e4 | 1051 | /** |
183ff22b | 1052 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1053 | * @size: size of the area |
1054 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1055 | * | |
1056 | * Search an area of @size in the kernel virtual mapping area, | |
1057 | * and reserved it for out purposes. Returns the area descriptor | |
1058 | * on success or %NULL on failure. | |
1059 | */ | |
1060 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1061 | { | |
23016969 CL |
1062 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, |
1063 | -1, GFP_KERNEL, __builtin_return_address(0)); | |
1064 | } | |
1065 | ||
1066 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
1067 | void *caller) | |
1068 | { | |
1069 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, | |
1070 | -1, GFP_KERNEL, caller); | |
1da177e4 LT |
1071 | } |
1072 | ||
52fd24ca GP |
1073 | struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, |
1074 | int node, gfp_t gfp_mask) | |
930fc45a | 1075 | { |
52fd24ca | 1076 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node, |
23016969 | 1077 | gfp_mask, __builtin_return_address(0)); |
930fc45a CL |
1078 | } |
1079 | ||
db64fe02 | 1080 | static struct vm_struct *find_vm_area(const void *addr) |
83342314 | 1081 | { |
db64fe02 | 1082 | struct vmap_area *va; |
83342314 | 1083 | |
db64fe02 NP |
1084 | va = find_vmap_area((unsigned long)addr); |
1085 | if (va && va->flags & VM_VM_AREA) | |
1086 | return va->private; | |
1da177e4 | 1087 | |
1da177e4 | 1088 | return NULL; |
1da177e4 LT |
1089 | } |
1090 | ||
7856dfeb | 1091 | /** |
183ff22b | 1092 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1093 | * @addr: base address |
1094 | * | |
1095 | * Search for the kernel VM area starting at @addr, and remove it. | |
1096 | * This function returns the found VM area, but using it is NOT safe | |
1097 | * on SMP machines, except for its size or flags. | |
1098 | */ | |
b3bdda02 | 1099 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1100 | { |
db64fe02 NP |
1101 | struct vmap_area *va; |
1102 | ||
1103 | va = find_vmap_area((unsigned long)addr); | |
1104 | if (va && va->flags & VM_VM_AREA) { | |
1105 | struct vm_struct *vm = va->private; | |
1106 | struct vm_struct *tmp, **p; | |
1107 | free_unmap_vmap_area(va); | |
1108 | vm->size -= PAGE_SIZE; | |
1109 | ||
1110 | write_lock(&vmlist_lock); | |
1111 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) | |
1112 | ; | |
1113 | *p = tmp->next; | |
1114 | write_unlock(&vmlist_lock); | |
1115 | ||
1116 | return vm; | |
1117 | } | |
1118 | return NULL; | |
7856dfeb AK |
1119 | } |
1120 | ||
b3bdda02 | 1121 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1122 | { |
1123 | struct vm_struct *area; | |
1124 | ||
1125 | if (!addr) | |
1126 | return; | |
1127 | ||
1128 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1129 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1130 | return; |
1131 | } | |
1132 | ||
1133 | area = remove_vm_area(addr); | |
1134 | if (unlikely(!area)) { | |
4c8573e2 | 1135 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1136 | addr); |
1da177e4 LT |
1137 | return; |
1138 | } | |
1139 | ||
9a11b49a | 1140 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1141 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1142 | |
1da177e4 LT |
1143 | if (deallocate_pages) { |
1144 | int i; | |
1145 | ||
1146 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1147 | struct page *page = area->pages[i]; |
1148 | ||
1149 | BUG_ON(!page); | |
1150 | __free_page(page); | |
1da177e4 LT |
1151 | } |
1152 | ||
8757d5fa | 1153 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1154 | vfree(area->pages); |
1155 | else | |
1156 | kfree(area->pages); | |
1157 | } | |
1158 | ||
1159 | kfree(area); | |
1160 | return; | |
1161 | } | |
1162 | ||
1163 | /** | |
1164 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1165 | * @addr: memory base address |
1166 | * | |
183ff22b | 1167 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1168 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1169 | * NULL, no operation is performed. | |
1da177e4 | 1170 | * |
80e93eff | 1171 | * Must not be called in interrupt context. |
1da177e4 | 1172 | */ |
b3bdda02 | 1173 | void vfree(const void *addr) |
1da177e4 LT |
1174 | { |
1175 | BUG_ON(in_interrupt()); | |
1176 | __vunmap(addr, 1); | |
1177 | } | |
1da177e4 LT |
1178 | EXPORT_SYMBOL(vfree); |
1179 | ||
1180 | /** | |
1181 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1182 | * @addr: memory base address |
1183 | * | |
1184 | * Free the virtually contiguous memory area starting at @addr, | |
1185 | * which was created from the page array passed to vmap(). | |
1186 | * | |
80e93eff | 1187 | * Must not be called in interrupt context. |
1da177e4 | 1188 | */ |
b3bdda02 | 1189 | void vunmap(const void *addr) |
1da177e4 LT |
1190 | { |
1191 | BUG_ON(in_interrupt()); | |
1192 | __vunmap(addr, 0); | |
1193 | } | |
1da177e4 LT |
1194 | EXPORT_SYMBOL(vunmap); |
1195 | ||
1196 | /** | |
1197 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1198 | * @pages: array of page pointers |
1199 | * @count: number of pages to map | |
1200 | * @flags: vm_area->flags | |
1201 | * @prot: page protection for the mapping | |
1202 | * | |
1203 | * Maps @count pages from @pages into contiguous kernel virtual | |
1204 | * space. | |
1205 | */ | |
1206 | void *vmap(struct page **pages, unsigned int count, | |
1207 | unsigned long flags, pgprot_t prot) | |
1208 | { | |
1209 | struct vm_struct *area; | |
1210 | ||
1211 | if (count > num_physpages) | |
1212 | return NULL; | |
1213 | ||
23016969 CL |
1214 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1215 | __builtin_return_address(0)); | |
1da177e4 LT |
1216 | if (!area) |
1217 | return NULL; | |
23016969 | 1218 | |
1da177e4 LT |
1219 | if (map_vm_area(area, prot, &pages)) { |
1220 | vunmap(area->addr); | |
1221 | return NULL; | |
1222 | } | |
1223 | ||
1224 | return area->addr; | |
1225 | } | |
1da177e4 LT |
1226 | EXPORT_SYMBOL(vmap); |
1227 | ||
db64fe02 NP |
1228 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
1229 | int node, void *caller); | |
e31d9eb5 | 1230 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 1231 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
1232 | { |
1233 | struct page **pages; | |
1234 | unsigned int nr_pages, array_size, i; | |
1235 | ||
1236 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1237 | array_size = (nr_pages * sizeof(struct page *)); | |
1238 | ||
1239 | area->nr_pages = nr_pages; | |
1240 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1241 | if (array_size > PAGE_SIZE) { |
94f6030c | 1242 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
23016969 | 1243 | PAGE_KERNEL, node, caller); |
8757d5fa | 1244 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
1245 | } else { |
1246 | pages = kmalloc_node(array_size, | |
6cb06229 | 1247 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
1248 | node); |
1249 | } | |
1da177e4 | 1250 | area->pages = pages; |
23016969 | 1251 | area->caller = caller; |
1da177e4 LT |
1252 | if (!area->pages) { |
1253 | remove_vm_area(area->addr); | |
1254 | kfree(area); | |
1255 | return NULL; | |
1256 | } | |
1da177e4 LT |
1257 | |
1258 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1259 | struct page *page; |
1260 | ||
930fc45a | 1261 | if (node < 0) |
bf53d6f8 | 1262 | page = alloc_page(gfp_mask); |
930fc45a | 1263 | else |
bf53d6f8 CL |
1264 | page = alloc_pages_node(node, gfp_mask, 0); |
1265 | ||
1266 | if (unlikely(!page)) { | |
1da177e4 LT |
1267 | /* Successfully allocated i pages, free them in __vunmap() */ |
1268 | area->nr_pages = i; | |
1269 | goto fail; | |
1270 | } | |
bf53d6f8 | 1271 | area->pages[i] = page; |
1da177e4 LT |
1272 | } |
1273 | ||
1274 | if (map_vm_area(area, prot, &pages)) | |
1275 | goto fail; | |
1276 | return area->addr; | |
1277 | ||
1278 | fail: | |
1279 | vfree(area->addr); | |
1280 | return NULL; | |
1281 | } | |
1282 | ||
930fc45a CL |
1283 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
1284 | { | |
23016969 CL |
1285 | return __vmalloc_area_node(area, gfp_mask, prot, -1, |
1286 | __builtin_return_address(0)); | |
930fc45a CL |
1287 | } |
1288 | ||
1da177e4 | 1289 | /** |
930fc45a | 1290 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
1291 | * @size: allocation size |
1292 | * @gfp_mask: flags for the page level allocator | |
1293 | * @prot: protection mask for the allocated pages | |
d44e0780 | 1294 | * @node: node to use for allocation or -1 |
c85d194b | 1295 | * @caller: caller's return address |
1da177e4 LT |
1296 | * |
1297 | * Allocate enough pages to cover @size from the page level | |
1298 | * allocator with @gfp_mask flags. Map them into contiguous | |
1299 | * kernel virtual space, using a pagetable protection of @prot. | |
1300 | */ | |
b221385b | 1301 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 1302 | int node, void *caller) |
1da177e4 LT |
1303 | { |
1304 | struct vm_struct *area; | |
1305 | ||
1306 | size = PAGE_ALIGN(size); | |
1307 | if (!size || (size >> PAGE_SHIFT) > num_physpages) | |
1308 | return NULL; | |
1309 | ||
23016969 CL |
1310 | area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END, |
1311 | node, gfp_mask, caller); | |
1312 | ||
1da177e4 LT |
1313 | if (!area) |
1314 | return NULL; | |
1315 | ||
23016969 | 1316 | return __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1da177e4 LT |
1317 | } |
1318 | ||
930fc45a CL |
1319 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1320 | { | |
23016969 CL |
1321 | return __vmalloc_node(size, gfp_mask, prot, -1, |
1322 | __builtin_return_address(0)); | |
930fc45a | 1323 | } |
1da177e4 LT |
1324 | EXPORT_SYMBOL(__vmalloc); |
1325 | ||
1326 | /** | |
1327 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1328 | * @size: allocation size |
1da177e4 LT |
1329 | * Allocate enough pages to cover @size from the page level |
1330 | * allocator and map them into contiguous kernel virtual space. | |
1331 | * | |
c1c8897f | 1332 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1333 | * use __vmalloc() instead. |
1334 | */ | |
1335 | void *vmalloc(unsigned long size) | |
1336 | { | |
23016969 CL |
1337 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1338 | -1, __builtin_return_address(0)); | |
1da177e4 | 1339 | } |
1da177e4 LT |
1340 | EXPORT_SYMBOL(vmalloc); |
1341 | ||
83342314 | 1342 | /** |
ead04089 REB |
1343 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1344 | * @size: allocation size | |
83342314 | 1345 | * |
ead04089 REB |
1346 | * The resulting memory area is zeroed so it can be mapped to userspace |
1347 | * without leaking data. | |
83342314 NP |
1348 | */ |
1349 | void *vmalloc_user(unsigned long size) | |
1350 | { | |
1351 | struct vm_struct *area; | |
1352 | void *ret; | |
1353 | ||
1354 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); | |
2b4ac44e | 1355 | if (ret) { |
db64fe02 | 1356 | area = find_vm_area(ret); |
2b4ac44e | 1357 | area->flags |= VM_USERMAP; |
2b4ac44e | 1358 | } |
83342314 NP |
1359 | return ret; |
1360 | } | |
1361 | EXPORT_SYMBOL(vmalloc_user); | |
1362 | ||
930fc45a CL |
1363 | /** |
1364 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1365 | * @size: allocation size |
d44e0780 | 1366 | * @node: numa node |
930fc45a CL |
1367 | * |
1368 | * Allocate enough pages to cover @size from the page level | |
1369 | * allocator and map them into contiguous kernel virtual space. | |
1370 | * | |
c1c8897f | 1371 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1372 | * use __vmalloc() instead. |
1373 | */ | |
1374 | void *vmalloc_node(unsigned long size, int node) | |
1375 | { | |
23016969 CL |
1376 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1377 | node, __builtin_return_address(0)); | |
930fc45a CL |
1378 | } |
1379 | EXPORT_SYMBOL(vmalloc_node); | |
1380 | ||
4dc3b16b PP |
1381 | #ifndef PAGE_KERNEL_EXEC |
1382 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1383 | #endif | |
1384 | ||
1da177e4 LT |
1385 | /** |
1386 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1387 | * @size: allocation size |
1388 | * | |
1389 | * Kernel-internal function to allocate enough pages to cover @size | |
1390 | * the page level allocator and map them into contiguous and | |
1391 | * executable kernel virtual space. | |
1392 | * | |
c1c8897f | 1393 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1394 | * use __vmalloc() instead. |
1395 | */ | |
1396 | ||
1da177e4 LT |
1397 | void *vmalloc_exec(unsigned long size) |
1398 | { | |
1399 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
1400 | } | |
1401 | ||
0d08e0d3 | 1402 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1403 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1404 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1405 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1406 | #else |
1407 | #define GFP_VMALLOC32 GFP_KERNEL | |
1408 | #endif | |
1409 | ||
1da177e4 LT |
1410 | /** |
1411 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1412 | * @size: allocation size |
1413 | * | |
1414 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1415 | * page level allocator and map them into contiguous kernel virtual space. | |
1416 | */ | |
1417 | void *vmalloc_32(unsigned long size) | |
1418 | { | |
0d08e0d3 | 1419 | return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL); |
1da177e4 | 1420 | } |
1da177e4 LT |
1421 | EXPORT_SYMBOL(vmalloc_32); |
1422 | ||
83342314 | 1423 | /** |
ead04089 | 1424 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1425 | * @size: allocation size |
ead04089 REB |
1426 | * |
1427 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1428 | * mapped to userspace without leaking data. | |
83342314 NP |
1429 | */ |
1430 | void *vmalloc_32_user(unsigned long size) | |
1431 | { | |
1432 | struct vm_struct *area; | |
1433 | void *ret; | |
1434 | ||
0d08e0d3 | 1435 | ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL); |
2b4ac44e | 1436 | if (ret) { |
db64fe02 | 1437 | area = find_vm_area(ret); |
2b4ac44e | 1438 | area->flags |= VM_USERMAP; |
2b4ac44e | 1439 | } |
83342314 NP |
1440 | return ret; |
1441 | } | |
1442 | EXPORT_SYMBOL(vmalloc_32_user); | |
1443 | ||
1da177e4 LT |
1444 | long vread(char *buf, char *addr, unsigned long count) |
1445 | { | |
1446 | struct vm_struct *tmp; | |
1447 | char *vaddr, *buf_start = buf; | |
1448 | unsigned long n; | |
1449 | ||
1450 | /* Don't allow overflow */ | |
1451 | if ((unsigned long) addr + count < count) | |
1452 | count = -(unsigned long) addr; | |
1453 | ||
1454 | read_lock(&vmlist_lock); | |
1455 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
1456 | vaddr = (char *) tmp->addr; | |
1457 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1458 | continue; | |
1459 | while (addr < vaddr) { | |
1460 | if (count == 0) | |
1461 | goto finished; | |
1462 | *buf = '\0'; | |
1463 | buf++; | |
1464 | addr++; | |
1465 | count--; | |
1466 | } | |
1467 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
1468 | do { | |
1469 | if (count == 0) | |
1470 | goto finished; | |
1471 | *buf = *addr; | |
1472 | buf++; | |
1473 | addr++; | |
1474 | count--; | |
1475 | } while (--n > 0); | |
1476 | } | |
1477 | finished: | |
1478 | read_unlock(&vmlist_lock); | |
1479 | return buf - buf_start; | |
1480 | } | |
1481 | ||
1482 | long vwrite(char *buf, char *addr, unsigned long count) | |
1483 | { | |
1484 | struct vm_struct *tmp; | |
1485 | char *vaddr, *buf_start = buf; | |
1486 | unsigned long n; | |
1487 | ||
1488 | /* Don't allow overflow */ | |
1489 | if ((unsigned long) addr + count < count) | |
1490 | count = -(unsigned long) addr; | |
1491 | ||
1492 | read_lock(&vmlist_lock); | |
1493 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
1494 | vaddr = (char *) tmp->addr; | |
1495 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1496 | continue; | |
1497 | while (addr < vaddr) { | |
1498 | if (count == 0) | |
1499 | goto finished; | |
1500 | buf++; | |
1501 | addr++; | |
1502 | count--; | |
1503 | } | |
1504 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
1505 | do { | |
1506 | if (count == 0) | |
1507 | goto finished; | |
1508 | *addr = *buf; | |
1509 | buf++; | |
1510 | addr++; | |
1511 | count--; | |
1512 | } while (--n > 0); | |
1513 | } | |
1514 | finished: | |
1515 | read_unlock(&vmlist_lock); | |
1516 | return buf - buf_start; | |
1517 | } | |
83342314 NP |
1518 | |
1519 | /** | |
1520 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
1521 | * @vma: vma to cover (map full range of vma) |
1522 | * @addr: vmalloc memory | |
1523 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
1524 | * |
1525 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
1526 | * |
1527 | * This function checks that addr is a valid vmalloc'ed area, and | |
1528 | * that it is big enough to cover the vma. Will return failure if | |
1529 | * that criteria isn't met. | |
1530 | * | |
72fd4a35 | 1531 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
1532 | */ |
1533 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
1534 | unsigned long pgoff) | |
1535 | { | |
1536 | struct vm_struct *area; | |
1537 | unsigned long uaddr = vma->vm_start; | |
1538 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
1539 | |
1540 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
1541 | return -EINVAL; | |
1542 | ||
db64fe02 | 1543 | area = find_vm_area(addr); |
83342314 | 1544 | if (!area) |
db64fe02 | 1545 | return -EINVAL; |
83342314 NP |
1546 | |
1547 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 1548 | return -EINVAL; |
83342314 NP |
1549 | |
1550 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 1551 | return -EINVAL; |
83342314 NP |
1552 | |
1553 | addr += pgoff << PAGE_SHIFT; | |
1554 | do { | |
1555 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
1556 | int ret; |
1557 | ||
83342314 NP |
1558 | ret = vm_insert_page(vma, uaddr, page); |
1559 | if (ret) | |
1560 | return ret; | |
1561 | ||
1562 | uaddr += PAGE_SIZE; | |
1563 | addr += PAGE_SIZE; | |
1564 | usize -= PAGE_SIZE; | |
1565 | } while (usize > 0); | |
1566 | ||
1567 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
1568 | vma->vm_flags |= VM_RESERVED; | |
1569 | ||
db64fe02 | 1570 | return 0; |
83342314 NP |
1571 | } |
1572 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1573 | ||
1eeb66a1 CH |
1574 | /* |
1575 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
1576 | * have one. | |
1577 | */ | |
1578 | void __attribute__((weak)) vmalloc_sync_all(void) | |
1579 | { | |
1580 | } | |
5f4352fb JF |
1581 | |
1582 | ||
2f569afd | 1583 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
1584 | { |
1585 | /* apply_to_page_range() does all the hard work. */ | |
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | /** | |
1590 | * alloc_vm_area - allocate a range of kernel address space | |
1591 | * @size: size of the area | |
7682486b RD |
1592 | * |
1593 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
1594 | * |
1595 | * This function reserves a range of kernel address space, and | |
1596 | * allocates pagetables to map that range. No actual mappings | |
1597 | * are created. If the kernel address space is not shared | |
1598 | * between processes, it syncs the pagetable across all | |
1599 | * processes. | |
1600 | */ | |
1601 | struct vm_struct *alloc_vm_area(size_t size) | |
1602 | { | |
1603 | struct vm_struct *area; | |
1604 | ||
23016969 CL |
1605 | area = get_vm_area_caller(size, VM_IOREMAP, |
1606 | __builtin_return_address(0)); | |
5f4352fb JF |
1607 | if (area == NULL) |
1608 | return NULL; | |
1609 | ||
1610 | /* | |
1611 | * This ensures that page tables are constructed for this region | |
1612 | * of kernel virtual address space and mapped into init_mm. | |
1613 | */ | |
1614 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
1615 | area->size, f, NULL)) { | |
1616 | free_vm_area(area); | |
1617 | return NULL; | |
1618 | } | |
1619 | ||
1620 | /* Make sure the pagetables are constructed in process kernel | |
1621 | mappings */ | |
1622 | vmalloc_sync_all(); | |
1623 | ||
1624 | return area; | |
1625 | } | |
1626 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
1627 | ||
1628 | void free_vm_area(struct vm_struct *area) | |
1629 | { | |
1630 | struct vm_struct *ret; | |
1631 | ret = remove_vm_area(area->addr); | |
1632 | BUG_ON(ret != area); | |
1633 | kfree(area); | |
1634 | } | |
1635 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 CL |
1636 | |
1637 | ||
1638 | #ifdef CONFIG_PROC_FS | |
1639 | static void *s_start(struct seq_file *m, loff_t *pos) | |
1640 | { | |
1641 | loff_t n = *pos; | |
1642 | struct vm_struct *v; | |
1643 | ||
1644 | read_lock(&vmlist_lock); | |
1645 | v = vmlist; | |
1646 | while (n > 0 && v) { | |
1647 | n--; | |
1648 | v = v->next; | |
1649 | } | |
1650 | if (!n) | |
1651 | return v; | |
1652 | ||
1653 | return NULL; | |
1654 | ||
1655 | } | |
1656 | ||
1657 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
1658 | { | |
1659 | struct vm_struct *v = p; | |
1660 | ||
1661 | ++*pos; | |
1662 | return v->next; | |
1663 | } | |
1664 | ||
1665 | static void s_stop(struct seq_file *m, void *p) | |
1666 | { | |
1667 | read_unlock(&vmlist_lock); | |
1668 | } | |
1669 | ||
a47a126a ED |
1670 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
1671 | { | |
1672 | if (NUMA_BUILD) { | |
1673 | unsigned int nr, *counters = m->private; | |
1674 | ||
1675 | if (!counters) | |
1676 | return; | |
1677 | ||
1678 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); | |
1679 | ||
1680 | for (nr = 0; nr < v->nr_pages; nr++) | |
1681 | counters[page_to_nid(v->pages[nr])]++; | |
1682 | ||
1683 | for_each_node_state(nr, N_HIGH_MEMORY) | |
1684 | if (counters[nr]) | |
1685 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
1686 | } | |
1687 | } | |
1688 | ||
a10aa579 CL |
1689 | static int s_show(struct seq_file *m, void *p) |
1690 | { | |
1691 | struct vm_struct *v = p; | |
1692 | ||
1693 | seq_printf(m, "0x%p-0x%p %7ld", | |
1694 | v->addr, v->addr + v->size, v->size); | |
1695 | ||
23016969 CL |
1696 | if (v->caller) { |
1697 | char buff[2 * KSYM_NAME_LEN]; | |
1698 | ||
1699 | seq_putc(m, ' '); | |
1700 | sprint_symbol(buff, (unsigned long)v->caller); | |
1701 | seq_puts(m, buff); | |
1702 | } | |
1703 | ||
a10aa579 CL |
1704 | if (v->nr_pages) |
1705 | seq_printf(m, " pages=%d", v->nr_pages); | |
1706 | ||
1707 | if (v->phys_addr) | |
1708 | seq_printf(m, " phys=%lx", v->phys_addr); | |
1709 | ||
1710 | if (v->flags & VM_IOREMAP) | |
1711 | seq_printf(m, " ioremap"); | |
1712 | ||
1713 | if (v->flags & VM_ALLOC) | |
1714 | seq_printf(m, " vmalloc"); | |
1715 | ||
1716 | if (v->flags & VM_MAP) | |
1717 | seq_printf(m, " vmap"); | |
1718 | ||
1719 | if (v->flags & VM_USERMAP) | |
1720 | seq_printf(m, " user"); | |
1721 | ||
1722 | if (v->flags & VM_VPAGES) | |
1723 | seq_printf(m, " vpages"); | |
1724 | ||
a47a126a | 1725 | show_numa_info(m, v); |
a10aa579 CL |
1726 | seq_putc(m, '\n'); |
1727 | return 0; | |
1728 | } | |
1729 | ||
5f6a6a9c | 1730 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
1731 | .start = s_start, |
1732 | .next = s_next, | |
1733 | .stop = s_stop, | |
1734 | .show = s_show, | |
1735 | }; | |
5f6a6a9c AD |
1736 | |
1737 | static int vmalloc_open(struct inode *inode, struct file *file) | |
1738 | { | |
1739 | unsigned int *ptr = NULL; | |
1740 | int ret; | |
1741 | ||
1742 | if (NUMA_BUILD) | |
1743 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); | |
1744 | ret = seq_open(file, &vmalloc_op); | |
1745 | if (!ret) { | |
1746 | struct seq_file *m = file->private_data; | |
1747 | m->private = ptr; | |
1748 | } else | |
1749 | kfree(ptr); | |
1750 | return ret; | |
1751 | } | |
1752 | ||
1753 | static const struct file_operations proc_vmalloc_operations = { | |
1754 | .open = vmalloc_open, | |
1755 | .read = seq_read, | |
1756 | .llseek = seq_lseek, | |
1757 | .release = seq_release_private, | |
1758 | }; | |
1759 | ||
1760 | static int __init proc_vmalloc_init(void) | |
1761 | { | |
1762 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
1763 | return 0; | |
1764 | } | |
1765 | module_init(proc_vmalloc_init); | |
a10aa579 CL |
1766 | #endif |
1767 |