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> | |
a10aa579 | 18 | #include <linux/seq_file.h> |
3ac7fe5a | 19 | #include <linux/debugobjects.h> |
23016969 | 20 | #include <linux/kallsyms.h> |
db64fe02 NP |
21 | #include <linux/list.h> |
22 | #include <linux/rbtree.h> | |
23 | #include <linux/radix-tree.h> | |
24 | #include <linux/rcupdate.h> | |
1da177e4 | 25 | |
db64fe02 | 26 | #include <asm/atomic.h> |
1da177e4 LT |
27 | #include <asm/uaccess.h> |
28 | #include <asm/tlbflush.h> | |
29 | ||
30 | ||
db64fe02 | 31 | /*** Page table manipulation functions ***/ |
b221385b | 32 | |
1da177e4 LT |
33 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
34 | { | |
35 | pte_t *pte; | |
36 | ||
37 | pte = pte_offset_kernel(pmd, addr); | |
38 | do { | |
39 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
40 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
41 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
42 | } | |
43 | ||
db64fe02 | 44 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
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 | ||
db64fe02 | 58 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
1da177e4 LT |
59 | { |
60 | pud_t *pud; | |
61 | unsigned long next; | |
62 | ||
63 | pud = pud_offset(pgd, addr); | |
64 | do { | |
65 | next = pud_addr_end(addr, end); | |
66 | if (pud_none_or_clear_bad(pud)) | |
67 | continue; | |
68 | vunmap_pmd_range(pud, addr, next); | |
69 | } while (pud++, addr = next, addr != end); | |
70 | } | |
71 | ||
db64fe02 | 72 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
73 | { |
74 | pgd_t *pgd; | |
75 | unsigned long next; | |
1da177e4 LT |
76 | |
77 | BUG_ON(addr >= end); | |
78 | pgd = pgd_offset_k(addr); | |
79 | flush_cache_vunmap(addr, end); | |
80 | do { | |
81 | next = pgd_addr_end(addr, end); | |
82 | if (pgd_none_or_clear_bad(pgd)) | |
83 | continue; | |
84 | vunmap_pud_range(pgd, addr, next); | |
85 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
86 | } |
87 | ||
88 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 89 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
90 | { |
91 | pte_t *pte; | |
92 | ||
db64fe02 NP |
93 | /* |
94 | * nr is a running index into the array which helps higher level | |
95 | * callers keep track of where we're up to. | |
96 | */ | |
97 | ||
872fec16 | 98 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
99 | if (!pte) |
100 | return -ENOMEM; | |
101 | do { | |
db64fe02 NP |
102 | struct page *page = pages[*nr]; |
103 | ||
104 | if (WARN_ON(!pte_none(*pte))) | |
105 | return -EBUSY; | |
106 | if (WARN_ON(!page)) | |
1da177e4 LT |
107 | return -ENOMEM; |
108 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 109 | (*nr)++; |
1da177e4 LT |
110 | } while (pte++, addr += PAGE_SIZE, addr != end); |
111 | return 0; | |
112 | } | |
113 | ||
db64fe02 NP |
114 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
115 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
116 | { |
117 | pmd_t *pmd; | |
118 | unsigned long next; | |
119 | ||
120 | pmd = pmd_alloc(&init_mm, pud, addr); | |
121 | if (!pmd) | |
122 | return -ENOMEM; | |
123 | do { | |
124 | next = pmd_addr_end(addr, end); | |
db64fe02 | 125 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
126 | return -ENOMEM; |
127 | } while (pmd++, addr = next, addr != end); | |
128 | return 0; | |
129 | } | |
130 | ||
db64fe02 NP |
131 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
132 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
133 | { |
134 | pud_t *pud; | |
135 | unsigned long next; | |
136 | ||
137 | pud = pud_alloc(&init_mm, pgd, addr); | |
138 | if (!pud) | |
139 | return -ENOMEM; | |
140 | do { | |
141 | next = pud_addr_end(addr, end); | |
db64fe02 | 142 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
143 | return -ENOMEM; |
144 | } while (pud++, addr = next, addr != end); | |
145 | return 0; | |
146 | } | |
147 | ||
db64fe02 NP |
148 | /* |
149 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
150 | * will have pfns corresponding to the "pages" array. | |
151 | * | |
152 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
153 | */ | |
154 | static int vmap_page_range(unsigned long addr, unsigned long end, | |
155 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
156 | { |
157 | pgd_t *pgd; | |
158 | unsigned long next; | |
db64fe02 NP |
159 | int err = 0; |
160 | int nr = 0; | |
1da177e4 LT |
161 | |
162 | BUG_ON(addr >= end); | |
163 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
164 | do { |
165 | next = pgd_addr_end(addr, end); | |
db64fe02 | 166 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 LT |
167 | if (err) |
168 | break; | |
169 | } while (pgd++, addr = next, addr != end); | |
db64fe02 NP |
170 | flush_cache_vmap(addr, end); |
171 | ||
172 | if (unlikely(err)) | |
173 | return err; | |
174 | return nr; | |
1da177e4 LT |
175 | } |
176 | ||
73bdf0a6 LT |
177 | static inline int is_vmalloc_or_module_addr(const void *x) |
178 | { | |
179 | /* | |
180 | * x86-64 and sparc64 put modules in a special place, | |
181 | * and fall back on vmalloc() if that fails. Others | |
182 | * just put it in the vmalloc space. | |
183 | */ | |
184 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
185 | unsigned long addr = (unsigned long)x; | |
186 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
187 | return 1; | |
188 | #endif | |
189 | return is_vmalloc_addr(x); | |
190 | } | |
191 | ||
48667e7a | 192 | /* |
db64fe02 | 193 | * Walk a vmap address to the struct page it maps. |
48667e7a | 194 | */ |
b3bdda02 | 195 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
196 | { |
197 | unsigned long addr = (unsigned long) vmalloc_addr; | |
198 | struct page *page = NULL; | |
199 | pgd_t *pgd = pgd_offset_k(addr); | |
48667e7a | 200 | |
7aa413de IM |
201 | /* |
202 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
203 | * architectures that do not vmalloc module space | |
204 | */ | |
73bdf0a6 | 205 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 206 | |
48667e7a | 207 | if (!pgd_none(*pgd)) { |
db64fe02 | 208 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 209 | if (!pud_none(*pud)) { |
db64fe02 | 210 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 211 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
212 | pte_t *ptep, pte; |
213 | ||
48667e7a CL |
214 | ptep = pte_offset_map(pmd, addr); |
215 | pte = *ptep; | |
216 | if (pte_present(pte)) | |
217 | page = pte_page(pte); | |
218 | pte_unmap(ptep); | |
219 | } | |
220 | } | |
221 | } | |
222 | return page; | |
223 | } | |
224 | EXPORT_SYMBOL(vmalloc_to_page); | |
225 | ||
226 | /* | |
227 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
228 | */ | |
b3bdda02 | 229 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
230 | { |
231 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
232 | } | |
233 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
234 | ||
db64fe02 NP |
235 | |
236 | /*** Global kva allocator ***/ | |
237 | ||
238 | #define VM_LAZY_FREE 0x01 | |
239 | #define VM_LAZY_FREEING 0x02 | |
240 | #define VM_VM_AREA 0x04 | |
241 | ||
242 | struct vmap_area { | |
243 | unsigned long va_start; | |
244 | unsigned long va_end; | |
245 | unsigned long flags; | |
246 | struct rb_node rb_node; /* address sorted rbtree */ | |
247 | struct list_head list; /* address sorted list */ | |
248 | struct list_head purge_list; /* "lazy purge" list */ | |
249 | void *private; | |
250 | struct rcu_head rcu_head; | |
251 | }; | |
252 | ||
253 | static DEFINE_SPINLOCK(vmap_area_lock); | |
254 | static struct rb_root vmap_area_root = RB_ROOT; | |
255 | static LIST_HEAD(vmap_area_list); | |
256 | ||
257 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 258 | { |
db64fe02 NP |
259 | struct rb_node *n = vmap_area_root.rb_node; |
260 | ||
261 | while (n) { | |
262 | struct vmap_area *va; | |
263 | ||
264 | va = rb_entry(n, struct vmap_area, rb_node); | |
265 | if (addr < va->va_start) | |
266 | n = n->rb_left; | |
267 | else if (addr > va->va_start) | |
268 | n = n->rb_right; | |
269 | else | |
270 | return va; | |
271 | } | |
272 | ||
273 | return NULL; | |
274 | } | |
275 | ||
276 | static void __insert_vmap_area(struct vmap_area *va) | |
277 | { | |
278 | struct rb_node **p = &vmap_area_root.rb_node; | |
279 | struct rb_node *parent = NULL; | |
280 | struct rb_node *tmp; | |
281 | ||
282 | while (*p) { | |
283 | struct vmap_area *tmp; | |
284 | ||
285 | parent = *p; | |
286 | tmp = rb_entry(parent, struct vmap_area, rb_node); | |
287 | if (va->va_start < tmp->va_end) | |
288 | p = &(*p)->rb_left; | |
289 | else if (va->va_end > tmp->va_start) | |
290 | p = &(*p)->rb_right; | |
291 | else | |
292 | BUG(); | |
293 | } | |
294 | ||
295 | rb_link_node(&va->rb_node, parent, p); | |
296 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
297 | ||
298 | /* address-sort this list so it is usable like the vmlist */ | |
299 | tmp = rb_prev(&va->rb_node); | |
300 | if (tmp) { | |
301 | struct vmap_area *prev; | |
302 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
303 | list_add_rcu(&va->list, &prev->list); | |
304 | } else | |
305 | list_add_rcu(&va->list, &vmap_area_list); | |
306 | } | |
307 | ||
308 | static void purge_vmap_area_lazy(void); | |
309 | ||
310 | /* | |
311 | * Allocate a region of KVA of the specified size and alignment, within the | |
312 | * vstart and vend. | |
313 | */ | |
314 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
315 | unsigned long align, | |
316 | unsigned long vstart, unsigned long vend, | |
317 | int node, gfp_t gfp_mask) | |
318 | { | |
319 | struct vmap_area *va; | |
320 | struct rb_node *n; | |
1da177e4 | 321 | unsigned long addr; |
db64fe02 NP |
322 | int purged = 0; |
323 | ||
324 | BUG_ON(size & ~PAGE_MASK); | |
325 | ||
326 | addr = ALIGN(vstart, align); | |
327 | ||
328 | va = kmalloc_node(sizeof(struct vmap_area), | |
329 | gfp_mask & GFP_RECLAIM_MASK, node); | |
330 | if (unlikely(!va)) | |
331 | return ERR_PTR(-ENOMEM); | |
332 | ||
333 | retry: | |
334 | spin_lock(&vmap_area_lock); | |
335 | /* XXX: could have a last_hole cache */ | |
336 | n = vmap_area_root.rb_node; | |
337 | if (n) { | |
338 | struct vmap_area *first = NULL; | |
339 | ||
340 | do { | |
341 | struct vmap_area *tmp; | |
342 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
343 | if (tmp->va_end >= addr) { | |
344 | if (!first && tmp->va_start < addr + size) | |
345 | first = tmp; | |
346 | n = n->rb_left; | |
347 | } else { | |
348 | first = tmp; | |
349 | n = n->rb_right; | |
350 | } | |
351 | } while (n); | |
352 | ||
353 | if (!first) | |
354 | goto found; | |
355 | ||
356 | if (first->va_end < addr) { | |
357 | n = rb_next(&first->rb_node); | |
358 | if (n) | |
359 | first = rb_entry(n, struct vmap_area, rb_node); | |
360 | else | |
361 | goto found; | |
362 | } | |
363 | ||
364 | while (addr + size >= first->va_start && addr + size <= vend) { | |
365 | addr = ALIGN(first->va_end + PAGE_SIZE, align); | |
366 | ||
367 | n = rb_next(&first->rb_node); | |
368 | if (n) | |
369 | first = rb_entry(n, struct vmap_area, rb_node); | |
370 | else | |
371 | goto found; | |
372 | } | |
373 | } | |
374 | found: | |
375 | if (addr + size > vend) { | |
376 | spin_unlock(&vmap_area_lock); | |
377 | if (!purged) { | |
378 | purge_vmap_area_lazy(); | |
379 | purged = 1; | |
380 | goto retry; | |
381 | } | |
382 | if (printk_ratelimit()) | |
383 | printk(KERN_WARNING "vmap allocation failed: " | |
384 | "use vmalloc=<size> to increase size.\n"); | |
385 | return ERR_PTR(-EBUSY); | |
386 | } | |
387 | ||
388 | BUG_ON(addr & (align-1)); | |
389 | ||
390 | va->va_start = addr; | |
391 | va->va_end = addr + size; | |
392 | va->flags = 0; | |
393 | __insert_vmap_area(va); | |
394 | spin_unlock(&vmap_area_lock); | |
395 | ||
396 | return va; | |
397 | } | |
398 | ||
399 | static void rcu_free_va(struct rcu_head *head) | |
400 | { | |
401 | struct vmap_area *va = container_of(head, struct vmap_area, rcu_head); | |
402 | ||
403 | kfree(va); | |
404 | } | |
405 | ||
406 | static void __free_vmap_area(struct vmap_area *va) | |
407 | { | |
408 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
409 | rb_erase(&va->rb_node, &vmap_area_root); | |
410 | RB_CLEAR_NODE(&va->rb_node); | |
411 | list_del_rcu(&va->list); | |
412 | ||
413 | call_rcu(&va->rcu_head, rcu_free_va); | |
414 | } | |
415 | ||
416 | /* | |
417 | * Free a region of KVA allocated by alloc_vmap_area | |
418 | */ | |
419 | static void free_vmap_area(struct vmap_area *va) | |
420 | { | |
421 | spin_lock(&vmap_area_lock); | |
422 | __free_vmap_area(va); | |
423 | spin_unlock(&vmap_area_lock); | |
424 | } | |
425 | ||
426 | /* | |
427 | * Clear the pagetable entries of a given vmap_area | |
428 | */ | |
429 | static void unmap_vmap_area(struct vmap_area *va) | |
430 | { | |
431 | vunmap_page_range(va->va_start, va->va_end); | |
432 | } | |
433 | ||
434 | /* | |
435 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
436 | * before attempting to purge with a TLB flush. | |
437 | * | |
438 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
439 | * and take slightly longer to purge, but it will linearly reduce the number of | |
440 | * global TLB flushes that must be performed. It would seem natural to scale | |
441 | * this number up linearly with the number of CPUs (because vmapping activity | |
442 | * could also scale linearly with the number of CPUs), however it is likely | |
443 | * that in practice, workloads might be constrained in other ways that mean | |
444 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
445 | * conservative and not introduce a big latency on huge systems, so go with | |
446 | * a less aggressive log scale. It will still be an improvement over the old | |
447 | * code, and it will be simple to change the scale factor if we find that it | |
448 | * becomes a problem on bigger systems. | |
449 | */ | |
450 | static unsigned long lazy_max_pages(void) | |
451 | { | |
452 | unsigned int log; | |
453 | ||
454 | log = fls(num_online_cpus()); | |
455 | ||
456 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
457 | } | |
458 | ||
459 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
460 | ||
461 | /* | |
462 | * Purges all lazily-freed vmap areas. | |
463 | * | |
464 | * If sync is 0 then don't purge if there is already a purge in progress. | |
465 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
466 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
467 | * their own TLB flushing). | |
468 | * Returns with *start = min(*start, lowest purged address) | |
469 | * *end = max(*end, highest purged address) | |
470 | */ | |
471 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
472 | int sync, int force_flush) | |
473 | { | |
474 | static DEFINE_SPINLOCK(purge_lock); | |
475 | LIST_HEAD(valist); | |
476 | struct vmap_area *va; | |
477 | int nr = 0; | |
478 | ||
479 | /* | |
480 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
481 | * should not expect such behaviour. This just simplifies locking for | |
482 | * the case that isn't actually used at the moment anyway. | |
483 | */ | |
484 | if (!sync && !force_flush) { | |
485 | if (!spin_trylock(&purge_lock)) | |
486 | return; | |
487 | } else | |
488 | spin_lock(&purge_lock); | |
489 | ||
490 | rcu_read_lock(); | |
491 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
492 | if (va->flags & VM_LAZY_FREE) { | |
493 | if (va->va_start < *start) | |
494 | *start = va->va_start; | |
495 | if (va->va_end > *end) | |
496 | *end = va->va_end; | |
497 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
498 | unmap_vmap_area(va); | |
499 | list_add_tail(&va->purge_list, &valist); | |
500 | va->flags |= VM_LAZY_FREEING; | |
501 | va->flags &= ~VM_LAZY_FREE; | |
502 | } | |
503 | } | |
504 | rcu_read_unlock(); | |
505 | ||
506 | if (nr) { | |
507 | BUG_ON(nr > atomic_read(&vmap_lazy_nr)); | |
508 | atomic_sub(nr, &vmap_lazy_nr); | |
509 | } | |
510 | ||
511 | if (nr || force_flush) | |
512 | flush_tlb_kernel_range(*start, *end); | |
513 | ||
514 | if (nr) { | |
515 | spin_lock(&vmap_area_lock); | |
516 | list_for_each_entry(va, &valist, purge_list) | |
517 | __free_vmap_area(va); | |
518 | spin_unlock(&vmap_area_lock); | |
519 | } | |
520 | spin_unlock(&purge_lock); | |
521 | } | |
522 | ||
523 | /* | |
524 | * Kick off a purge of the outstanding lazy areas. | |
525 | */ | |
526 | static void purge_vmap_area_lazy(void) | |
527 | { | |
528 | unsigned long start = ULONG_MAX, end = 0; | |
529 | ||
530 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
531 | } | |
532 | ||
533 | /* | |
534 | * Free and unmap a vmap area | |
535 | */ | |
536 | static void free_unmap_vmap_area(struct vmap_area *va) | |
537 | { | |
538 | va->flags |= VM_LAZY_FREE; | |
539 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
540 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
541 | purge_vmap_area_lazy(); | |
542 | } | |
543 | ||
544 | static struct vmap_area *find_vmap_area(unsigned long addr) | |
545 | { | |
546 | struct vmap_area *va; | |
547 | ||
548 | spin_lock(&vmap_area_lock); | |
549 | va = __find_vmap_area(addr); | |
550 | spin_unlock(&vmap_area_lock); | |
551 | ||
552 | return va; | |
553 | } | |
554 | ||
555 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
556 | { | |
557 | struct vmap_area *va; | |
558 | ||
559 | va = find_vmap_area(addr); | |
560 | BUG_ON(!va); | |
561 | free_unmap_vmap_area(va); | |
562 | } | |
563 | ||
564 | ||
565 | /*** Per cpu kva allocator ***/ | |
566 | ||
567 | /* | |
568 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
569 | * room for at least 16 percpu vmap blocks per CPU. | |
570 | */ | |
571 | /* | |
572 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
573 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
574 | * instead (we just need a rough idea) | |
575 | */ | |
576 | #if BITS_PER_LONG == 32 | |
577 | #define VMALLOC_SPACE (128UL*1024*1024) | |
578 | #else | |
579 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
580 | #endif | |
581 | ||
582 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
583 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
584 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
585 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
586 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
587 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
588 | #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
589 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
590 | VMALLOC_PAGES / NR_CPUS / 16)) | |
591 | ||
592 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
593 | ||
594 | struct vmap_block_queue { | |
595 | spinlock_t lock; | |
596 | struct list_head free; | |
597 | struct list_head dirty; | |
598 | unsigned int nr_dirty; | |
599 | }; | |
600 | ||
601 | struct vmap_block { | |
602 | spinlock_t lock; | |
603 | struct vmap_area *va; | |
604 | struct vmap_block_queue *vbq; | |
605 | unsigned long free, dirty; | |
606 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
607 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
608 | union { | |
609 | struct { | |
610 | struct list_head free_list; | |
611 | struct list_head dirty_list; | |
612 | }; | |
613 | struct rcu_head rcu_head; | |
614 | }; | |
615 | }; | |
616 | ||
617 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
618 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
619 | ||
620 | /* | |
621 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
622 | * in the free path. Could get rid of this if we change the API to return a | |
623 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
624 | */ | |
625 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
626 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
627 | ||
628 | /* | |
629 | * We should probably have a fallback mechanism to allocate virtual memory | |
630 | * out of partially filled vmap blocks. However vmap block sizing should be | |
631 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
632 | * big problem. | |
633 | */ | |
634 | ||
635 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
636 | { | |
637 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
638 | addr /= VMAP_BLOCK_SIZE; | |
639 | return addr; | |
640 | } | |
641 | ||
642 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
643 | { | |
644 | struct vmap_block_queue *vbq; | |
645 | struct vmap_block *vb; | |
646 | struct vmap_area *va; | |
647 | unsigned long vb_idx; | |
648 | int node, err; | |
649 | ||
650 | node = numa_node_id(); | |
651 | ||
652 | vb = kmalloc_node(sizeof(struct vmap_block), | |
653 | gfp_mask & GFP_RECLAIM_MASK, node); | |
654 | if (unlikely(!vb)) | |
655 | return ERR_PTR(-ENOMEM); | |
656 | ||
657 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
658 | VMALLOC_START, VMALLOC_END, | |
659 | node, gfp_mask); | |
660 | if (unlikely(IS_ERR(va))) { | |
661 | kfree(vb); | |
662 | return ERR_PTR(PTR_ERR(va)); | |
663 | } | |
664 | ||
665 | err = radix_tree_preload(gfp_mask); | |
666 | if (unlikely(err)) { | |
667 | kfree(vb); | |
668 | free_vmap_area(va); | |
669 | return ERR_PTR(err); | |
670 | } | |
671 | ||
672 | spin_lock_init(&vb->lock); | |
673 | vb->va = va; | |
674 | vb->free = VMAP_BBMAP_BITS; | |
675 | vb->dirty = 0; | |
676 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
677 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
678 | INIT_LIST_HEAD(&vb->free_list); | |
679 | INIT_LIST_HEAD(&vb->dirty_list); | |
680 | ||
681 | vb_idx = addr_to_vb_idx(va->va_start); | |
682 | spin_lock(&vmap_block_tree_lock); | |
683 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
684 | spin_unlock(&vmap_block_tree_lock); | |
685 | BUG_ON(err); | |
686 | radix_tree_preload_end(); | |
687 | ||
688 | vbq = &get_cpu_var(vmap_block_queue); | |
689 | vb->vbq = vbq; | |
690 | spin_lock(&vbq->lock); | |
691 | list_add(&vb->free_list, &vbq->free); | |
692 | spin_unlock(&vbq->lock); | |
693 | put_cpu_var(vmap_cpu_blocks); | |
694 | ||
695 | return vb; | |
696 | } | |
697 | ||
698 | static void rcu_free_vb(struct rcu_head *head) | |
699 | { | |
700 | struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head); | |
701 | ||
702 | kfree(vb); | |
703 | } | |
704 | ||
705 | static void free_vmap_block(struct vmap_block *vb) | |
706 | { | |
707 | struct vmap_block *tmp; | |
708 | unsigned long vb_idx; | |
709 | ||
710 | spin_lock(&vb->vbq->lock); | |
711 | if (!list_empty(&vb->free_list)) | |
712 | list_del(&vb->free_list); | |
713 | if (!list_empty(&vb->dirty_list)) | |
714 | list_del(&vb->dirty_list); | |
715 | spin_unlock(&vb->vbq->lock); | |
716 | ||
717 | vb_idx = addr_to_vb_idx(vb->va->va_start); | |
718 | spin_lock(&vmap_block_tree_lock); | |
719 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
720 | spin_unlock(&vmap_block_tree_lock); | |
721 | BUG_ON(tmp != vb); | |
722 | ||
723 | free_unmap_vmap_area(vb->va); | |
724 | call_rcu(&vb->rcu_head, rcu_free_vb); | |
725 | } | |
726 | ||
727 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) | |
728 | { | |
729 | struct vmap_block_queue *vbq; | |
730 | struct vmap_block *vb; | |
731 | unsigned long addr = 0; | |
732 | unsigned int order; | |
733 | ||
734 | BUG_ON(size & ~PAGE_MASK); | |
735 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
736 | order = get_order(size); | |
737 | ||
738 | again: | |
739 | rcu_read_lock(); | |
740 | vbq = &get_cpu_var(vmap_block_queue); | |
741 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
742 | int i; | |
743 | ||
744 | spin_lock(&vb->lock); | |
745 | i = bitmap_find_free_region(vb->alloc_map, | |
746 | VMAP_BBMAP_BITS, order); | |
747 | ||
748 | if (i >= 0) { | |
749 | addr = vb->va->va_start + (i << PAGE_SHIFT); | |
750 | BUG_ON(addr_to_vb_idx(addr) != | |
751 | addr_to_vb_idx(vb->va->va_start)); | |
752 | vb->free -= 1UL << order; | |
753 | if (vb->free == 0) { | |
754 | spin_lock(&vbq->lock); | |
755 | list_del_init(&vb->free_list); | |
756 | spin_unlock(&vbq->lock); | |
757 | } | |
758 | spin_unlock(&vb->lock); | |
759 | break; | |
760 | } | |
761 | spin_unlock(&vb->lock); | |
762 | } | |
763 | put_cpu_var(vmap_cpu_blocks); | |
764 | rcu_read_unlock(); | |
765 | ||
766 | if (!addr) { | |
767 | vb = new_vmap_block(gfp_mask); | |
768 | if (IS_ERR(vb)) | |
769 | return vb; | |
770 | goto again; | |
771 | } | |
772 | ||
773 | return (void *)addr; | |
774 | } | |
775 | ||
776 | static void vb_free(const void *addr, unsigned long size) | |
777 | { | |
778 | unsigned long offset; | |
779 | unsigned long vb_idx; | |
780 | unsigned int order; | |
781 | struct vmap_block *vb; | |
782 | ||
783 | BUG_ON(size & ~PAGE_MASK); | |
784 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
785 | order = get_order(size); | |
786 | ||
787 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
788 | ||
789 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
790 | rcu_read_lock(); | |
791 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
792 | rcu_read_unlock(); | |
793 | BUG_ON(!vb); | |
794 | ||
795 | spin_lock(&vb->lock); | |
796 | bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order); | |
797 | if (!vb->dirty) { | |
798 | spin_lock(&vb->vbq->lock); | |
799 | list_add(&vb->dirty_list, &vb->vbq->dirty); | |
800 | spin_unlock(&vb->vbq->lock); | |
801 | } | |
802 | vb->dirty += 1UL << order; | |
803 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
804 | BUG_ON(vb->free || !list_empty(&vb->free_list)); | |
805 | spin_unlock(&vb->lock); | |
806 | free_vmap_block(vb); | |
807 | } else | |
808 | spin_unlock(&vb->lock); | |
809 | } | |
810 | ||
811 | /** | |
812 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
813 | * | |
814 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
815 | * to amortize TLB flushing overheads. What this means is that any page you | |
816 | * have now, may, in a former life, have been mapped into kernel virtual | |
817 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
818 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
819 | * | |
820 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
821 | * be sure that none of the pages we have control over will have any aliases | |
822 | * from the vmap layer. | |
823 | */ | |
824 | void vm_unmap_aliases(void) | |
825 | { | |
826 | unsigned long start = ULONG_MAX, end = 0; | |
827 | int cpu; | |
828 | int flush = 0; | |
829 | ||
830 | for_each_possible_cpu(cpu) { | |
831 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
832 | struct vmap_block *vb; | |
833 | ||
834 | rcu_read_lock(); | |
835 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
836 | int i; | |
837 | ||
838 | spin_lock(&vb->lock); | |
839 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
840 | while (i < VMAP_BBMAP_BITS) { | |
841 | unsigned long s, e; | |
842 | int j; | |
843 | j = find_next_zero_bit(vb->dirty_map, | |
844 | VMAP_BBMAP_BITS, i); | |
845 | ||
846 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
847 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
848 | vunmap_page_range(s, e); | |
849 | flush = 1; | |
850 | ||
851 | if (s < start) | |
852 | start = s; | |
853 | if (e > end) | |
854 | end = e; | |
855 | ||
856 | i = j; | |
857 | i = find_next_bit(vb->dirty_map, | |
858 | VMAP_BBMAP_BITS, i); | |
859 | } | |
860 | spin_unlock(&vb->lock); | |
861 | } | |
862 | rcu_read_unlock(); | |
863 | } | |
864 | ||
865 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
866 | } | |
867 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
868 | ||
869 | /** | |
870 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
871 | * @mem: the pointer returned by vm_map_ram | |
872 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
873 | */ | |
874 | void vm_unmap_ram(const void *mem, unsigned int count) | |
875 | { | |
876 | unsigned long size = count << PAGE_SHIFT; | |
877 | unsigned long addr = (unsigned long)mem; | |
878 | ||
879 | BUG_ON(!addr); | |
880 | BUG_ON(addr < VMALLOC_START); | |
881 | BUG_ON(addr > VMALLOC_END); | |
882 | BUG_ON(addr & (PAGE_SIZE-1)); | |
883 | ||
884 | debug_check_no_locks_freed(mem, size); | |
885 | ||
886 | if (likely(count <= VMAP_MAX_ALLOC)) | |
887 | vb_free(mem, size); | |
888 | else | |
889 | free_unmap_vmap_area_addr(addr); | |
890 | } | |
891 | EXPORT_SYMBOL(vm_unmap_ram); | |
892 | ||
893 | /** | |
894 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
895 | * @pages: an array of pointers to the pages to be mapped | |
896 | * @count: number of pages | |
897 | * @node: prefer to allocate data structures on this node | |
898 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
899 | * @returns: a pointer to the address that has been mapped, or NULL on failure | |
900 | */ | |
901 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
902 | { | |
903 | unsigned long size = count << PAGE_SHIFT; | |
904 | unsigned long addr; | |
905 | void *mem; | |
906 | ||
907 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
908 | mem = vb_alloc(size, GFP_KERNEL); | |
909 | if (IS_ERR(mem)) | |
910 | return NULL; | |
911 | addr = (unsigned long)mem; | |
912 | } else { | |
913 | struct vmap_area *va; | |
914 | va = alloc_vmap_area(size, PAGE_SIZE, | |
915 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
916 | if (IS_ERR(va)) | |
917 | return NULL; | |
918 | ||
919 | addr = va->va_start; | |
920 | mem = (void *)addr; | |
921 | } | |
922 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
923 | vm_unmap_ram(mem, count); | |
924 | return NULL; | |
925 | } | |
926 | return mem; | |
927 | } | |
928 | EXPORT_SYMBOL(vm_map_ram); | |
929 | ||
930 | void __init vmalloc_init(void) | |
931 | { | |
932 | int i; | |
933 | ||
934 | for_each_possible_cpu(i) { | |
935 | struct vmap_block_queue *vbq; | |
936 | ||
937 | vbq = &per_cpu(vmap_block_queue, i); | |
938 | spin_lock_init(&vbq->lock); | |
939 | INIT_LIST_HEAD(&vbq->free); | |
940 | INIT_LIST_HEAD(&vbq->dirty); | |
941 | vbq->nr_dirty = 0; | |
942 | } | |
943 | } | |
944 | ||
945 | void unmap_kernel_range(unsigned long addr, unsigned long size) | |
946 | { | |
947 | unsigned long end = addr + size; | |
948 | vunmap_page_range(addr, end); | |
949 | flush_tlb_kernel_range(addr, end); | |
950 | } | |
951 | ||
952 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
953 | { | |
954 | unsigned long addr = (unsigned long)area->addr; | |
955 | unsigned long end = addr + area->size - PAGE_SIZE; | |
956 | int err; | |
957 | ||
958 | err = vmap_page_range(addr, end, prot, *pages); | |
959 | if (err > 0) { | |
960 | *pages += err; | |
961 | err = 0; | |
962 | } | |
963 | ||
964 | return err; | |
965 | } | |
966 | EXPORT_SYMBOL_GPL(map_vm_area); | |
967 | ||
968 | /*** Old vmalloc interfaces ***/ | |
969 | DEFINE_RWLOCK(vmlist_lock); | |
970 | struct vm_struct *vmlist; | |
971 | ||
972 | static struct vm_struct *__get_vm_area_node(unsigned long size, | |
973 | unsigned long flags, unsigned long start, unsigned long end, | |
974 | int node, gfp_t gfp_mask, void *caller) | |
975 | { | |
976 | static struct vmap_area *va; | |
977 | struct vm_struct *area; | |
978 | struct vm_struct *tmp, **p; | |
979 | unsigned long align = 1; | |
1da177e4 | 980 | |
52fd24ca | 981 | BUG_ON(in_interrupt()); |
1da177e4 LT |
982 | if (flags & VM_IOREMAP) { |
983 | int bit = fls(size); | |
984 | ||
985 | if (bit > IOREMAP_MAX_ORDER) | |
986 | bit = IOREMAP_MAX_ORDER; | |
987 | else if (bit < PAGE_SHIFT) | |
988 | bit = PAGE_SHIFT; | |
989 | ||
990 | align = 1ul << bit; | |
991 | } | |
db64fe02 | 992 | |
1da177e4 | 993 | size = PAGE_ALIGN(size); |
31be8309 OH |
994 | if (unlikely(!size)) |
995 | return NULL; | |
1da177e4 | 996 | |
6cb06229 | 997 | area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
998 | if (unlikely(!area)) |
999 | return NULL; | |
1000 | ||
1da177e4 LT |
1001 | /* |
1002 | * We always allocate a guard page. | |
1003 | */ | |
1004 | size += PAGE_SIZE; | |
1005 | ||
db64fe02 NP |
1006 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1007 | if (IS_ERR(va)) { | |
1008 | kfree(area); | |
1009 | return NULL; | |
1da177e4 | 1010 | } |
1da177e4 LT |
1011 | |
1012 | area->flags = flags; | |
db64fe02 | 1013 | area->addr = (void *)va->va_start; |
1da177e4 LT |
1014 | area->size = size; |
1015 | area->pages = NULL; | |
1016 | area->nr_pages = 0; | |
1017 | area->phys_addr = 0; | |
23016969 | 1018 | area->caller = caller; |
db64fe02 NP |
1019 | va->private = area; |
1020 | va->flags |= VM_VM_AREA; | |
1021 | ||
1022 | write_lock(&vmlist_lock); | |
1023 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1024 | if (tmp->addr >= area->addr) | |
1025 | break; | |
1026 | } | |
1027 | area->next = *p; | |
1028 | *p = area; | |
1da177e4 LT |
1029 | write_unlock(&vmlist_lock); |
1030 | ||
1031 | return area; | |
1da177e4 LT |
1032 | } |
1033 | ||
930fc45a CL |
1034 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1035 | unsigned long start, unsigned long end) | |
1036 | { | |
23016969 CL |
1037 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, |
1038 | __builtin_return_address(0)); | |
930fc45a | 1039 | } |
5992b6da | 1040 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1041 | |
1da177e4 | 1042 | /** |
183ff22b | 1043 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1044 | * @size: size of the area |
1045 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1046 | * | |
1047 | * Search an area of @size in the kernel virtual mapping area, | |
1048 | * and reserved it for out purposes. Returns the area descriptor | |
1049 | * on success or %NULL on failure. | |
1050 | */ | |
1051 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1052 | { | |
23016969 CL |
1053 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, |
1054 | -1, GFP_KERNEL, __builtin_return_address(0)); | |
1055 | } | |
1056 | ||
1057 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
1058 | void *caller) | |
1059 | { | |
1060 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, | |
1061 | -1, GFP_KERNEL, caller); | |
1da177e4 LT |
1062 | } |
1063 | ||
52fd24ca GP |
1064 | struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, |
1065 | int node, gfp_t gfp_mask) | |
930fc45a | 1066 | { |
52fd24ca | 1067 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node, |
23016969 | 1068 | gfp_mask, __builtin_return_address(0)); |
930fc45a CL |
1069 | } |
1070 | ||
db64fe02 | 1071 | static struct vm_struct *find_vm_area(const void *addr) |
83342314 | 1072 | { |
db64fe02 | 1073 | struct vmap_area *va; |
83342314 | 1074 | |
db64fe02 NP |
1075 | va = find_vmap_area((unsigned long)addr); |
1076 | if (va && va->flags & VM_VM_AREA) | |
1077 | return va->private; | |
1da177e4 | 1078 | |
1da177e4 | 1079 | return NULL; |
1da177e4 LT |
1080 | } |
1081 | ||
7856dfeb | 1082 | /** |
183ff22b | 1083 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1084 | * @addr: base address |
1085 | * | |
1086 | * Search for the kernel VM area starting at @addr, and remove it. | |
1087 | * This function returns the found VM area, but using it is NOT safe | |
1088 | * on SMP machines, except for its size or flags. | |
1089 | */ | |
b3bdda02 | 1090 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1091 | { |
db64fe02 NP |
1092 | struct vmap_area *va; |
1093 | ||
1094 | va = find_vmap_area((unsigned long)addr); | |
1095 | if (va && va->flags & VM_VM_AREA) { | |
1096 | struct vm_struct *vm = va->private; | |
1097 | struct vm_struct *tmp, **p; | |
1098 | free_unmap_vmap_area(va); | |
1099 | vm->size -= PAGE_SIZE; | |
1100 | ||
1101 | write_lock(&vmlist_lock); | |
1102 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) | |
1103 | ; | |
1104 | *p = tmp->next; | |
1105 | write_unlock(&vmlist_lock); | |
1106 | ||
1107 | return vm; | |
1108 | } | |
1109 | return NULL; | |
7856dfeb AK |
1110 | } |
1111 | ||
b3bdda02 | 1112 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1113 | { |
1114 | struct vm_struct *area; | |
1115 | ||
1116 | if (!addr) | |
1117 | return; | |
1118 | ||
1119 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1120 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1121 | return; |
1122 | } | |
1123 | ||
1124 | area = remove_vm_area(addr); | |
1125 | if (unlikely(!area)) { | |
4c8573e2 | 1126 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1127 | addr); |
1da177e4 LT |
1128 | return; |
1129 | } | |
1130 | ||
9a11b49a | 1131 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1132 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1133 | |
1da177e4 LT |
1134 | if (deallocate_pages) { |
1135 | int i; | |
1136 | ||
1137 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1138 | struct page *page = area->pages[i]; |
1139 | ||
1140 | BUG_ON(!page); | |
1141 | __free_page(page); | |
1da177e4 LT |
1142 | } |
1143 | ||
8757d5fa | 1144 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1145 | vfree(area->pages); |
1146 | else | |
1147 | kfree(area->pages); | |
1148 | } | |
1149 | ||
1150 | kfree(area); | |
1151 | return; | |
1152 | } | |
1153 | ||
1154 | /** | |
1155 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1156 | * @addr: memory base address |
1157 | * | |
183ff22b | 1158 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1159 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1160 | * NULL, no operation is performed. | |
1da177e4 | 1161 | * |
80e93eff | 1162 | * Must not be called in interrupt context. |
1da177e4 | 1163 | */ |
b3bdda02 | 1164 | void vfree(const void *addr) |
1da177e4 LT |
1165 | { |
1166 | BUG_ON(in_interrupt()); | |
1167 | __vunmap(addr, 1); | |
1168 | } | |
1da177e4 LT |
1169 | EXPORT_SYMBOL(vfree); |
1170 | ||
1171 | /** | |
1172 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1173 | * @addr: memory base address |
1174 | * | |
1175 | * Free the virtually contiguous memory area starting at @addr, | |
1176 | * which was created from the page array passed to vmap(). | |
1177 | * | |
80e93eff | 1178 | * Must not be called in interrupt context. |
1da177e4 | 1179 | */ |
b3bdda02 | 1180 | void vunmap(const void *addr) |
1da177e4 LT |
1181 | { |
1182 | BUG_ON(in_interrupt()); | |
1183 | __vunmap(addr, 0); | |
1184 | } | |
1da177e4 LT |
1185 | EXPORT_SYMBOL(vunmap); |
1186 | ||
1187 | /** | |
1188 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1189 | * @pages: array of page pointers |
1190 | * @count: number of pages to map | |
1191 | * @flags: vm_area->flags | |
1192 | * @prot: page protection for the mapping | |
1193 | * | |
1194 | * Maps @count pages from @pages into contiguous kernel virtual | |
1195 | * space. | |
1196 | */ | |
1197 | void *vmap(struct page **pages, unsigned int count, | |
1198 | unsigned long flags, pgprot_t prot) | |
1199 | { | |
1200 | struct vm_struct *area; | |
1201 | ||
1202 | if (count > num_physpages) | |
1203 | return NULL; | |
1204 | ||
23016969 CL |
1205 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1206 | __builtin_return_address(0)); | |
1da177e4 LT |
1207 | if (!area) |
1208 | return NULL; | |
23016969 | 1209 | |
1da177e4 LT |
1210 | if (map_vm_area(area, prot, &pages)) { |
1211 | vunmap(area->addr); | |
1212 | return NULL; | |
1213 | } | |
1214 | ||
1215 | return area->addr; | |
1216 | } | |
1da177e4 LT |
1217 | EXPORT_SYMBOL(vmap); |
1218 | ||
db64fe02 NP |
1219 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
1220 | int node, void *caller); | |
e31d9eb5 | 1221 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 1222 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
1223 | { |
1224 | struct page **pages; | |
1225 | unsigned int nr_pages, array_size, i; | |
1226 | ||
1227 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1228 | array_size = (nr_pages * sizeof(struct page *)); | |
1229 | ||
1230 | area->nr_pages = nr_pages; | |
1231 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1232 | if (array_size > PAGE_SIZE) { |
94f6030c | 1233 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
23016969 | 1234 | PAGE_KERNEL, node, caller); |
8757d5fa | 1235 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
1236 | } else { |
1237 | pages = kmalloc_node(array_size, | |
6cb06229 | 1238 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
1239 | node); |
1240 | } | |
1da177e4 | 1241 | area->pages = pages; |
23016969 | 1242 | area->caller = caller; |
1da177e4 LT |
1243 | if (!area->pages) { |
1244 | remove_vm_area(area->addr); | |
1245 | kfree(area); | |
1246 | return NULL; | |
1247 | } | |
1da177e4 LT |
1248 | |
1249 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1250 | struct page *page; |
1251 | ||
930fc45a | 1252 | if (node < 0) |
bf53d6f8 | 1253 | page = alloc_page(gfp_mask); |
930fc45a | 1254 | else |
bf53d6f8 CL |
1255 | page = alloc_pages_node(node, gfp_mask, 0); |
1256 | ||
1257 | if (unlikely(!page)) { | |
1da177e4 LT |
1258 | /* Successfully allocated i pages, free them in __vunmap() */ |
1259 | area->nr_pages = i; | |
1260 | goto fail; | |
1261 | } | |
bf53d6f8 | 1262 | area->pages[i] = page; |
1da177e4 LT |
1263 | } |
1264 | ||
1265 | if (map_vm_area(area, prot, &pages)) | |
1266 | goto fail; | |
1267 | return area->addr; | |
1268 | ||
1269 | fail: | |
1270 | vfree(area->addr); | |
1271 | return NULL; | |
1272 | } | |
1273 | ||
930fc45a CL |
1274 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
1275 | { | |
23016969 CL |
1276 | return __vmalloc_area_node(area, gfp_mask, prot, -1, |
1277 | __builtin_return_address(0)); | |
930fc45a CL |
1278 | } |
1279 | ||
1da177e4 | 1280 | /** |
930fc45a | 1281 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
1282 | * @size: allocation size |
1283 | * @gfp_mask: flags for the page level allocator | |
1284 | * @prot: protection mask for the allocated pages | |
d44e0780 | 1285 | * @node: node to use for allocation or -1 |
c85d194b | 1286 | * @caller: caller's return address |
1da177e4 LT |
1287 | * |
1288 | * Allocate enough pages to cover @size from the page level | |
1289 | * allocator with @gfp_mask flags. Map them into contiguous | |
1290 | * kernel virtual space, using a pagetable protection of @prot. | |
1291 | */ | |
b221385b | 1292 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 1293 | int node, void *caller) |
1da177e4 LT |
1294 | { |
1295 | struct vm_struct *area; | |
1296 | ||
1297 | size = PAGE_ALIGN(size); | |
1298 | if (!size || (size >> PAGE_SHIFT) > num_physpages) | |
1299 | return NULL; | |
1300 | ||
23016969 CL |
1301 | area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END, |
1302 | node, gfp_mask, caller); | |
1303 | ||
1da177e4 LT |
1304 | if (!area) |
1305 | return NULL; | |
1306 | ||
23016969 | 1307 | return __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1da177e4 LT |
1308 | } |
1309 | ||
930fc45a CL |
1310 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1311 | { | |
23016969 CL |
1312 | return __vmalloc_node(size, gfp_mask, prot, -1, |
1313 | __builtin_return_address(0)); | |
930fc45a | 1314 | } |
1da177e4 LT |
1315 | EXPORT_SYMBOL(__vmalloc); |
1316 | ||
1317 | /** | |
1318 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1319 | * @size: allocation size |
1da177e4 LT |
1320 | * Allocate enough pages to cover @size from the page level |
1321 | * allocator and map them into contiguous kernel virtual space. | |
1322 | * | |
c1c8897f | 1323 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1324 | * use __vmalloc() instead. |
1325 | */ | |
1326 | void *vmalloc(unsigned long size) | |
1327 | { | |
23016969 CL |
1328 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1329 | -1, __builtin_return_address(0)); | |
1da177e4 | 1330 | } |
1da177e4 LT |
1331 | EXPORT_SYMBOL(vmalloc); |
1332 | ||
83342314 | 1333 | /** |
ead04089 REB |
1334 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1335 | * @size: allocation size | |
83342314 | 1336 | * |
ead04089 REB |
1337 | * The resulting memory area is zeroed so it can be mapped to userspace |
1338 | * without leaking data. | |
83342314 NP |
1339 | */ |
1340 | void *vmalloc_user(unsigned long size) | |
1341 | { | |
1342 | struct vm_struct *area; | |
1343 | void *ret; | |
1344 | ||
1345 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); | |
2b4ac44e | 1346 | if (ret) { |
db64fe02 | 1347 | area = find_vm_area(ret); |
2b4ac44e | 1348 | area->flags |= VM_USERMAP; |
2b4ac44e | 1349 | } |
83342314 NP |
1350 | return ret; |
1351 | } | |
1352 | EXPORT_SYMBOL(vmalloc_user); | |
1353 | ||
930fc45a CL |
1354 | /** |
1355 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1356 | * @size: allocation size |
d44e0780 | 1357 | * @node: numa node |
930fc45a CL |
1358 | * |
1359 | * Allocate enough pages to cover @size from the page level | |
1360 | * allocator and map them into contiguous kernel virtual space. | |
1361 | * | |
c1c8897f | 1362 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1363 | * use __vmalloc() instead. |
1364 | */ | |
1365 | void *vmalloc_node(unsigned long size, int node) | |
1366 | { | |
23016969 CL |
1367 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1368 | node, __builtin_return_address(0)); | |
930fc45a CL |
1369 | } |
1370 | EXPORT_SYMBOL(vmalloc_node); | |
1371 | ||
4dc3b16b PP |
1372 | #ifndef PAGE_KERNEL_EXEC |
1373 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1374 | #endif | |
1375 | ||
1da177e4 LT |
1376 | /** |
1377 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1378 | * @size: allocation size |
1379 | * | |
1380 | * Kernel-internal function to allocate enough pages to cover @size | |
1381 | * the page level allocator and map them into contiguous and | |
1382 | * executable kernel virtual space. | |
1383 | * | |
c1c8897f | 1384 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1385 | * use __vmalloc() instead. |
1386 | */ | |
1387 | ||
1da177e4 LT |
1388 | void *vmalloc_exec(unsigned long size) |
1389 | { | |
1390 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
1391 | } | |
1392 | ||
0d08e0d3 | 1393 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1394 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1395 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1396 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1397 | #else |
1398 | #define GFP_VMALLOC32 GFP_KERNEL | |
1399 | #endif | |
1400 | ||
1da177e4 LT |
1401 | /** |
1402 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1403 | * @size: allocation size |
1404 | * | |
1405 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1406 | * page level allocator and map them into contiguous kernel virtual space. | |
1407 | */ | |
1408 | void *vmalloc_32(unsigned long size) | |
1409 | { | |
0d08e0d3 | 1410 | return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL); |
1da177e4 | 1411 | } |
1da177e4 LT |
1412 | EXPORT_SYMBOL(vmalloc_32); |
1413 | ||
83342314 | 1414 | /** |
ead04089 | 1415 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1416 | * @size: allocation size |
ead04089 REB |
1417 | * |
1418 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1419 | * mapped to userspace without leaking data. | |
83342314 NP |
1420 | */ |
1421 | void *vmalloc_32_user(unsigned long size) | |
1422 | { | |
1423 | struct vm_struct *area; | |
1424 | void *ret; | |
1425 | ||
0d08e0d3 | 1426 | ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL); |
2b4ac44e | 1427 | if (ret) { |
db64fe02 | 1428 | area = find_vm_area(ret); |
2b4ac44e | 1429 | area->flags |= VM_USERMAP; |
2b4ac44e | 1430 | } |
83342314 NP |
1431 | return ret; |
1432 | } | |
1433 | EXPORT_SYMBOL(vmalloc_32_user); | |
1434 | ||
1da177e4 LT |
1435 | long vread(char *buf, char *addr, unsigned long count) |
1436 | { | |
1437 | struct vm_struct *tmp; | |
1438 | char *vaddr, *buf_start = buf; | |
1439 | unsigned long n; | |
1440 | ||
1441 | /* Don't allow overflow */ | |
1442 | if ((unsigned long) addr + count < count) | |
1443 | count = -(unsigned long) addr; | |
1444 | ||
1445 | read_lock(&vmlist_lock); | |
1446 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
1447 | vaddr = (char *) tmp->addr; | |
1448 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1449 | continue; | |
1450 | while (addr < vaddr) { | |
1451 | if (count == 0) | |
1452 | goto finished; | |
1453 | *buf = '\0'; | |
1454 | buf++; | |
1455 | addr++; | |
1456 | count--; | |
1457 | } | |
1458 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
1459 | do { | |
1460 | if (count == 0) | |
1461 | goto finished; | |
1462 | *buf = *addr; | |
1463 | buf++; | |
1464 | addr++; | |
1465 | count--; | |
1466 | } while (--n > 0); | |
1467 | } | |
1468 | finished: | |
1469 | read_unlock(&vmlist_lock); | |
1470 | return buf - buf_start; | |
1471 | } | |
1472 | ||
1473 | long vwrite(char *buf, char *addr, unsigned long count) | |
1474 | { | |
1475 | struct vm_struct *tmp; | |
1476 | char *vaddr, *buf_start = buf; | |
1477 | unsigned long n; | |
1478 | ||
1479 | /* Don't allow overflow */ | |
1480 | if ((unsigned long) addr + count < count) | |
1481 | count = -(unsigned long) addr; | |
1482 | ||
1483 | read_lock(&vmlist_lock); | |
1484 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
1485 | vaddr = (char *) tmp->addr; | |
1486 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1487 | continue; | |
1488 | while (addr < vaddr) { | |
1489 | if (count == 0) | |
1490 | goto finished; | |
1491 | buf++; | |
1492 | addr++; | |
1493 | count--; | |
1494 | } | |
1495 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
1496 | do { | |
1497 | if (count == 0) | |
1498 | goto finished; | |
1499 | *addr = *buf; | |
1500 | buf++; | |
1501 | addr++; | |
1502 | count--; | |
1503 | } while (--n > 0); | |
1504 | } | |
1505 | finished: | |
1506 | read_unlock(&vmlist_lock); | |
1507 | return buf - buf_start; | |
1508 | } | |
83342314 NP |
1509 | |
1510 | /** | |
1511 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
1512 | * @vma: vma to cover (map full range of vma) |
1513 | * @addr: vmalloc memory | |
1514 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
1515 | * |
1516 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
1517 | * |
1518 | * This function checks that addr is a valid vmalloc'ed area, and | |
1519 | * that it is big enough to cover the vma. Will return failure if | |
1520 | * that criteria isn't met. | |
1521 | * | |
72fd4a35 | 1522 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
1523 | */ |
1524 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
1525 | unsigned long pgoff) | |
1526 | { | |
1527 | struct vm_struct *area; | |
1528 | unsigned long uaddr = vma->vm_start; | |
1529 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
1530 | |
1531 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
1532 | return -EINVAL; | |
1533 | ||
db64fe02 | 1534 | area = find_vm_area(addr); |
83342314 | 1535 | if (!area) |
db64fe02 | 1536 | return -EINVAL; |
83342314 NP |
1537 | |
1538 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 1539 | return -EINVAL; |
83342314 NP |
1540 | |
1541 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 1542 | return -EINVAL; |
83342314 NP |
1543 | |
1544 | addr += pgoff << PAGE_SHIFT; | |
1545 | do { | |
1546 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
1547 | int ret; |
1548 | ||
83342314 NP |
1549 | ret = vm_insert_page(vma, uaddr, page); |
1550 | if (ret) | |
1551 | return ret; | |
1552 | ||
1553 | uaddr += PAGE_SIZE; | |
1554 | addr += PAGE_SIZE; | |
1555 | usize -= PAGE_SIZE; | |
1556 | } while (usize > 0); | |
1557 | ||
1558 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
1559 | vma->vm_flags |= VM_RESERVED; | |
1560 | ||
db64fe02 | 1561 | return 0; |
83342314 NP |
1562 | } |
1563 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1564 | ||
1eeb66a1 CH |
1565 | /* |
1566 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
1567 | * have one. | |
1568 | */ | |
1569 | void __attribute__((weak)) vmalloc_sync_all(void) | |
1570 | { | |
1571 | } | |
5f4352fb JF |
1572 | |
1573 | ||
2f569afd | 1574 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
1575 | { |
1576 | /* apply_to_page_range() does all the hard work. */ | |
1577 | return 0; | |
1578 | } | |
1579 | ||
1580 | /** | |
1581 | * alloc_vm_area - allocate a range of kernel address space | |
1582 | * @size: size of the area | |
7682486b RD |
1583 | * |
1584 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
1585 | * |
1586 | * This function reserves a range of kernel address space, and | |
1587 | * allocates pagetables to map that range. No actual mappings | |
1588 | * are created. If the kernel address space is not shared | |
1589 | * between processes, it syncs the pagetable across all | |
1590 | * processes. | |
1591 | */ | |
1592 | struct vm_struct *alloc_vm_area(size_t size) | |
1593 | { | |
1594 | struct vm_struct *area; | |
1595 | ||
23016969 CL |
1596 | area = get_vm_area_caller(size, VM_IOREMAP, |
1597 | __builtin_return_address(0)); | |
5f4352fb JF |
1598 | if (area == NULL) |
1599 | return NULL; | |
1600 | ||
1601 | /* | |
1602 | * This ensures that page tables are constructed for this region | |
1603 | * of kernel virtual address space and mapped into init_mm. | |
1604 | */ | |
1605 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
1606 | area->size, f, NULL)) { | |
1607 | free_vm_area(area); | |
1608 | return NULL; | |
1609 | } | |
1610 | ||
1611 | /* Make sure the pagetables are constructed in process kernel | |
1612 | mappings */ | |
1613 | vmalloc_sync_all(); | |
1614 | ||
1615 | return area; | |
1616 | } | |
1617 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
1618 | ||
1619 | void free_vm_area(struct vm_struct *area) | |
1620 | { | |
1621 | struct vm_struct *ret; | |
1622 | ret = remove_vm_area(area->addr); | |
1623 | BUG_ON(ret != area); | |
1624 | kfree(area); | |
1625 | } | |
1626 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 CL |
1627 | |
1628 | ||
1629 | #ifdef CONFIG_PROC_FS | |
1630 | static void *s_start(struct seq_file *m, loff_t *pos) | |
1631 | { | |
1632 | loff_t n = *pos; | |
1633 | struct vm_struct *v; | |
1634 | ||
1635 | read_lock(&vmlist_lock); | |
1636 | v = vmlist; | |
1637 | while (n > 0 && v) { | |
1638 | n--; | |
1639 | v = v->next; | |
1640 | } | |
1641 | if (!n) | |
1642 | return v; | |
1643 | ||
1644 | return NULL; | |
1645 | ||
1646 | } | |
1647 | ||
1648 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
1649 | { | |
1650 | struct vm_struct *v = p; | |
1651 | ||
1652 | ++*pos; | |
1653 | return v->next; | |
1654 | } | |
1655 | ||
1656 | static void s_stop(struct seq_file *m, void *p) | |
1657 | { | |
1658 | read_unlock(&vmlist_lock); | |
1659 | } | |
1660 | ||
a47a126a ED |
1661 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
1662 | { | |
1663 | if (NUMA_BUILD) { | |
1664 | unsigned int nr, *counters = m->private; | |
1665 | ||
1666 | if (!counters) | |
1667 | return; | |
1668 | ||
1669 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); | |
1670 | ||
1671 | for (nr = 0; nr < v->nr_pages; nr++) | |
1672 | counters[page_to_nid(v->pages[nr])]++; | |
1673 | ||
1674 | for_each_node_state(nr, N_HIGH_MEMORY) | |
1675 | if (counters[nr]) | |
1676 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
1677 | } | |
1678 | } | |
1679 | ||
a10aa579 CL |
1680 | static int s_show(struct seq_file *m, void *p) |
1681 | { | |
1682 | struct vm_struct *v = p; | |
1683 | ||
1684 | seq_printf(m, "0x%p-0x%p %7ld", | |
1685 | v->addr, v->addr + v->size, v->size); | |
1686 | ||
23016969 CL |
1687 | if (v->caller) { |
1688 | char buff[2 * KSYM_NAME_LEN]; | |
1689 | ||
1690 | seq_putc(m, ' '); | |
1691 | sprint_symbol(buff, (unsigned long)v->caller); | |
1692 | seq_puts(m, buff); | |
1693 | } | |
1694 | ||
a10aa579 CL |
1695 | if (v->nr_pages) |
1696 | seq_printf(m, " pages=%d", v->nr_pages); | |
1697 | ||
1698 | if (v->phys_addr) | |
1699 | seq_printf(m, " phys=%lx", v->phys_addr); | |
1700 | ||
1701 | if (v->flags & VM_IOREMAP) | |
1702 | seq_printf(m, " ioremap"); | |
1703 | ||
1704 | if (v->flags & VM_ALLOC) | |
1705 | seq_printf(m, " vmalloc"); | |
1706 | ||
1707 | if (v->flags & VM_MAP) | |
1708 | seq_printf(m, " vmap"); | |
1709 | ||
1710 | if (v->flags & VM_USERMAP) | |
1711 | seq_printf(m, " user"); | |
1712 | ||
1713 | if (v->flags & VM_VPAGES) | |
1714 | seq_printf(m, " vpages"); | |
1715 | ||
a47a126a | 1716 | show_numa_info(m, v); |
a10aa579 CL |
1717 | seq_putc(m, '\n'); |
1718 | return 0; | |
1719 | } | |
1720 | ||
1721 | const struct seq_operations vmalloc_op = { | |
1722 | .start = s_start, | |
1723 | .next = s_next, | |
1724 | .stop = s_stop, | |
1725 | .show = s_show, | |
1726 | }; | |
1727 | #endif | |
1728 |