blob: 1ddb77ba3995c2d8f486baf32c1409b3daf8093f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Christoph Lameter930fc452005-10-29 18:15:41 -07008 * Numa awareness, Christoph Lameter, SGI, June 2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Nick Piggindb64fe02008-10-18 20:27:03 -070011#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#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>
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +040018#include <linux/proc_fs.h>
Christoph Lametera10aa572008-04-28 02:12:40 -070019#include <linux/seq_file.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070020#include <linux/debugobjects.h>
Christoph Lameter23016962008-04-28 02:12:42 -070021#include <linux/kallsyms.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070022#include <linux/list.h>
23#include <linux/rbtree.h>
24#include <linux/radix-tree.h>
25#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Nick Piggindb64fe02008-10-18 20:27:03 -070027#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/uaccess.h>
29#include <asm/tlbflush.h>
30
31
Nick Piggindb64fe02008-10-18 20:27:03 -070032/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070033
Linus Torvalds1da177e2005-04-16 15:20:36 -070034static 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
Nick Piggindb64fe02008-10-18 20:27:03 -070045static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070046{
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
Nick Piggindb64fe02008-10-18 20:27:03 -070059static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
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
Nick Piggindb64fe02008-10-18 20:27:03 -070073static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 pgd_t *pgd;
76 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 BUG_ON(addr >= end);
79 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086}
87
88static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -070089 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
91 pte_t *pte;
92
Nick Piggindb64fe02008-10-18 20:27:03 -070093 /*
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
Hugh Dickins872fec12005-10-29 18:16:21 -070098 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 if (!pte)
100 return -ENOMEM;
101 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700102 struct page *page = pages[*nr];
103
104 if (WARN_ON(!pte_none(*pte)))
105 return -EBUSY;
106 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 return -ENOMEM;
108 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700109 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 } while (pte++, addr += PAGE_SIZE, addr != end);
111 return 0;
112}
113
Nick Piggindb64fe02008-10-18 20:27:03 -0700114static int vmap_pmd_range(pud_t *pud, unsigned long addr,
115 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
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);
Nick Piggindb64fe02008-10-18 20:27:03 -0700125 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 return -ENOMEM;
127 } while (pmd++, addr = next, addr != end);
128 return 0;
129}
130
Nick Piggindb64fe02008-10-18 20:27:03 -0700131static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
132 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
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);
Nick Piggindb64fe02008-10-18 20:27:03 -0700142 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 return -ENOMEM;
144 } while (pud++, addr = next, addr != end);
145 return 0;
146}
147
Nick Piggindb64fe02008-10-18 20:27:03 -0700148/*
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 */
154static int vmap_page_range(unsigned long addr, unsigned long end,
155 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156{
157 pgd_t *pgd;
158 unsigned long next;
Nick Piggindb64fe02008-10-18 20:27:03 -0700159 int err = 0;
160 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 BUG_ON(addr >= end);
163 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 do {
165 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700166 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 if (err)
168 break;
169 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700170 flush_cache_vmap(addr, end);
171
172 if (unlikely(err))
173 return err;
174 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700177static inline int is_vmalloc_or_module_addr(const void *x)
178{
179 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000180 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700181 * 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
Christoph Lameter48667e72008-02-04 22:28:31 -0800192/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700193 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800194 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800195struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800196{
197 unsigned long addr = (unsigned long) vmalloc_addr;
198 struct page *page = NULL;
199 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800200
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200201 /*
202 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
203 * architectures that do not vmalloc module space
204 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700205 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200206
Christoph Lameter48667e72008-02-04 22:28:31 -0800207 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700208 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800209 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700210 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800211 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700212 pte_t *ptep, pte;
213
Christoph Lameter48667e72008-02-04 22:28:31 -0800214 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}
224EXPORT_SYMBOL(vmalloc_to_page);
225
226/*
227 * Map a vmalloc()-space virtual address to the physical page frame number.
228 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800229unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800230{
231 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
232}
233EXPORT_SYMBOL(vmalloc_to_pfn);
234
Nick Piggindb64fe02008-10-18 20:27:03 -0700235
236/*** Global kva allocator ***/
237
238#define VM_LAZY_FREE 0x01
239#define VM_LAZY_FREEING 0x02
240#define VM_VM_AREA 0x04
241
242struct 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
253static DEFINE_SPINLOCK(vmap_area_lock);
254static struct rb_root vmap_area_root = RB_ROOT;
255static LIST_HEAD(vmap_area_list);
256
257static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Nick Piggindb64fe02008-10-18 20:27:03 -0700259 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
276static 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
308static 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 */
314static 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700322 int purged = 0;
323
324 BUG_ON(size & ~PAGE_MASK);
325
Nick Piggindb64fe02008-10-18 20:27:03 -0700326 va = kmalloc_node(sizeof(struct vmap_area),
327 gfp_mask & GFP_RECLAIM_MASK, node);
328 if (unlikely(!va))
329 return ERR_PTR(-ENOMEM);
330
331retry:
Glauber Costa0ae15132008-11-19 15:36:33 -0800332 addr = ALIGN(vstart, align);
333
Nick Piggindb64fe02008-10-18 20:27:03 -0700334 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
Nick Pigginf011c2d2008-11-19 15:36:32 -0800364 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700365 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 }
374found:
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
399static 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
406static 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 */
419static 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 */
429static 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 */
450static 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
459static 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 */
471static 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/*
Nick Piggin496850e2008-11-19 15:36:33 -0800524 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
525 * is already purging.
526 */
527static void try_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/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700535 * Kick off a purge of the outstanding lazy areas.
536 */
537static void purge_vmap_area_lazy(void)
538{
539 unsigned long start = ULONG_MAX, end = 0;
540
Nick Piggin496850e2008-11-19 15:36:33 -0800541 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700542}
543
544/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800545 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
546 * called for the correct range previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700547 */
Nick Pigginb29acbd2008-12-01 13:13:47 -0800548static void free_unmap_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700549{
550 va->flags |= VM_LAZY_FREE;
551 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
552 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800553 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700554}
555
Nick Pigginb29acbd2008-12-01 13:13:47 -0800556/*
557 * Free and unmap a vmap area
558 */
559static void free_unmap_vmap_area(struct vmap_area *va)
560{
561 flush_cache_vunmap(va->va_start, va->va_end);
562 free_unmap_vmap_area_noflush(va);
563}
564
Nick Piggindb64fe02008-10-18 20:27:03 -0700565static struct vmap_area *find_vmap_area(unsigned long addr)
566{
567 struct vmap_area *va;
568
569 spin_lock(&vmap_area_lock);
570 va = __find_vmap_area(addr);
571 spin_unlock(&vmap_area_lock);
572
573 return va;
574}
575
576static void free_unmap_vmap_area_addr(unsigned long addr)
577{
578 struct vmap_area *va;
579
580 va = find_vmap_area(addr);
581 BUG_ON(!va);
582 free_unmap_vmap_area(va);
583}
584
585
586/*** Per cpu kva allocator ***/
587
588/*
589 * vmap space is limited especially on 32 bit architectures. Ensure there is
590 * room for at least 16 percpu vmap blocks per CPU.
591 */
592/*
593 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
594 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
595 * instead (we just need a rough idea)
596 */
597#if BITS_PER_LONG == 32
598#define VMALLOC_SPACE (128UL*1024*1024)
599#else
600#define VMALLOC_SPACE (128UL*1024*1024*1024)
601#endif
602
603#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
604#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
605#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
606#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
607#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
608#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
609#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
610 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
611 VMALLOC_PAGES / NR_CPUS / 16))
612
613#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
614
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100615static bool vmap_initialized __read_mostly = false;
616
Nick Piggindb64fe02008-10-18 20:27:03 -0700617struct vmap_block_queue {
618 spinlock_t lock;
619 struct list_head free;
620 struct list_head dirty;
621 unsigned int nr_dirty;
622};
623
624struct vmap_block {
625 spinlock_t lock;
626 struct vmap_area *va;
627 struct vmap_block_queue *vbq;
628 unsigned long free, dirty;
629 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
630 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
631 union {
632 struct {
633 struct list_head free_list;
634 struct list_head dirty_list;
635 };
636 struct rcu_head rcu_head;
637 };
638};
639
640/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
641static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
642
643/*
644 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
645 * in the free path. Could get rid of this if we change the API to return a
646 * "cookie" from alloc, to be passed to free. But no big deal yet.
647 */
648static DEFINE_SPINLOCK(vmap_block_tree_lock);
649static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
650
651/*
652 * We should probably have a fallback mechanism to allocate virtual memory
653 * out of partially filled vmap blocks. However vmap block sizing should be
654 * fairly reasonable according to the vmalloc size, so it shouldn't be a
655 * big problem.
656 */
657
658static unsigned long addr_to_vb_idx(unsigned long addr)
659{
660 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
661 addr /= VMAP_BLOCK_SIZE;
662 return addr;
663}
664
665static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
666{
667 struct vmap_block_queue *vbq;
668 struct vmap_block *vb;
669 struct vmap_area *va;
670 unsigned long vb_idx;
671 int node, err;
672
673 node = numa_node_id();
674
675 vb = kmalloc_node(sizeof(struct vmap_block),
676 gfp_mask & GFP_RECLAIM_MASK, node);
677 if (unlikely(!vb))
678 return ERR_PTR(-ENOMEM);
679
680 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
681 VMALLOC_START, VMALLOC_END,
682 node, gfp_mask);
683 if (unlikely(IS_ERR(va))) {
684 kfree(vb);
685 return ERR_PTR(PTR_ERR(va));
686 }
687
688 err = radix_tree_preload(gfp_mask);
689 if (unlikely(err)) {
690 kfree(vb);
691 free_vmap_area(va);
692 return ERR_PTR(err);
693 }
694
695 spin_lock_init(&vb->lock);
696 vb->va = va;
697 vb->free = VMAP_BBMAP_BITS;
698 vb->dirty = 0;
699 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
700 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
701 INIT_LIST_HEAD(&vb->free_list);
702 INIT_LIST_HEAD(&vb->dirty_list);
703
704 vb_idx = addr_to_vb_idx(va->va_start);
705 spin_lock(&vmap_block_tree_lock);
706 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
707 spin_unlock(&vmap_block_tree_lock);
708 BUG_ON(err);
709 radix_tree_preload_end();
710
711 vbq = &get_cpu_var(vmap_block_queue);
712 vb->vbq = vbq;
713 spin_lock(&vbq->lock);
714 list_add(&vb->free_list, &vbq->free);
715 spin_unlock(&vbq->lock);
716 put_cpu_var(vmap_cpu_blocks);
717
718 return vb;
719}
720
721static void rcu_free_vb(struct rcu_head *head)
722{
723 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
724
725 kfree(vb);
726}
727
728static void free_vmap_block(struct vmap_block *vb)
729{
730 struct vmap_block *tmp;
731 unsigned long vb_idx;
732
733 spin_lock(&vb->vbq->lock);
734 if (!list_empty(&vb->free_list))
735 list_del(&vb->free_list);
736 if (!list_empty(&vb->dirty_list))
737 list_del(&vb->dirty_list);
738 spin_unlock(&vb->vbq->lock);
739
740 vb_idx = addr_to_vb_idx(vb->va->va_start);
741 spin_lock(&vmap_block_tree_lock);
742 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
743 spin_unlock(&vmap_block_tree_lock);
744 BUG_ON(tmp != vb);
745
Nick Pigginb29acbd2008-12-01 13:13:47 -0800746 free_unmap_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700747 call_rcu(&vb->rcu_head, rcu_free_vb);
748}
749
750static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
751{
752 struct vmap_block_queue *vbq;
753 struct vmap_block *vb;
754 unsigned long addr = 0;
755 unsigned int order;
756
757 BUG_ON(size & ~PAGE_MASK);
758 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
759 order = get_order(size);
760
761again:
762 rcu_read_lock();
763 vbq = &get_cpu_var(vmap_block_queue);
764 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
765 int i;
766
767 spin_lock(&vb->lock);
768 i = bitmap_find_free_region(vb->alloc_map,
769 VMAP_BBMAP_BITS, order);
770
771 if (i >= 0) {
772 addr = vb->va->va_start + (i << PAGE_SHIFT);
773 BUG_ON(addr_to_vb_idx(addr) !=
774 addr_to_vb_idx(vb->va->va_start));
775 vb->free -= 1UL << order;
776 if (vb->free == 0) {
777 spin_lock(&vbq->lock);
778 list_del_init(&vb->free_list);
779 spin_unlock(&vbq->lock);
780 }
781 spin_unlock(&vb->lock);
782 break;
783 }
784 spin_unlock(&vb->lock);
785 }
786 put_cpu_var(vmap_cpu_blocks);
787 rcu_read_unlock();
788
789 if (!addr) {
790 vb = new_vmap_block(gfp_mask);
791 if (IS_ERR(vb))
792 return vb;
793 goto again;
794 }
795
796 return (void *)addr;
797}
798
799static void vb_free(const void *addr, unsigned long size)
800{
801 unsigned long offset;
802 unsigned long vb_idx;
803 unsigned int order;
804 struct vmap_block *vb;
805
806 BUG_ON(size & ~PAGE_MASK);
807 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800808
809 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
810
Nick Piggindb64fe02008-10-18 20:27:03 -0700811 order = get_order(size);
812
813 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
814
815 vb_idx = addr_to_vb_idx((unsigned long)addr);
816 rcu_read_lock();
817 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
818 rcu_read_unlock();
819 BUG_ON(!vb);
820
821 spin_lock(&vb->lock);
822 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
823 if (!vb->dirty) {
824 spin_lock(&vb->vbq->lock);
825 list_add(&vb->dirty_list, &vb->vbq->dirty);
826 spin_unlock(&vb->vbq->lock);
827 }
828 vb->dirty += 1UL << order;
829 if (vb->dirty == VMAP_BBMAP_BITS) {
830 BUG_ON(vb->free || !list_empty(&vb->free_list));
831 spin_unlock(&vb->lock);
832 free_vmap_block(vb);
833 } else
834 spin_unlock(&vb->lock);
835}
836
837/**
838 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
839 *
840 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
841 * to amortize TLB flushing overheads. What this means is that any page you
842 * have now, may, in a former life, have been mapped into kernel virtual
843 * address by the vmap layer and so there might be some CPUs with TLB entries
844 * still referencing that page (additional to the regular 1:1 kernel mapping).
845 *
846 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
847 * be sure that none of the pages we have control over will have any aliases
848 * from the vmap layer.
849 */
850void vm_unmap_aliases(void)
851{
852 unsigned long start = ULONG_MAX, end = 0;
853 int cpu;
854 int flush = 0;
855
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100856 if (unlikely(!vmap_initialized))
857 return;
858
Nick Piggindb64fe02008-10-18 20:27:03 -0700859 for_each_possible_cpu(cpu) {
860 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
861 struct vmap_block *vb;
862
863 rcu_read_lock();
864 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
865 int i;
866
867 spin_lock(&vb->lock);
868 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
869 while (i < VMAP_BBMAP_BITS) {
870 unsigned long s, e;
871 int j;
872 j = find_next_zero_bit(vb->dirty_map,
873 VMAP_BBMAP_BITS, i);
874
875 s = vb->va->va_start + (i << PAGE_SHIFT);
876 e = vb->va->va_start + (j << PAGE_SHIFT);
877 vunmap_page_range(s, e);
878 flush = 1;
879
880 if (s < start)
881 start = s;
882 if (e > end)
883 end = e;
884
885 i = j;
886 i = find_next_bit(vb->dirty_map,
887 VMAP_BBMAP_BITS, i);
888 }
889 spin_unlock(&vb->lock);
890 }
891 rcu_read_unlock();
892 }
893
894 __purge_vmap_area_lazy(&start, &end, 1, flush);
895}
896EXPORT_SYMBOL_GPL(vm_unmap_aliases);
897
898/**
899 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
900 * @mem: the pointer returned by vm_map_ram
901 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
902 */
903void vm_unmap_ram(const void *mem, unsigned int count)
904{
905 unsigned long size = count << PAGE_SHIFT;
906 unsigned long addr = (unsigned long)mem;
907
908 BUG_ON(!addr);
909 BUG_ON(addr < VMALLOC_START);
910 BUG_ON(addr > VMALLOC_END);
911 BUG_ON(addr & (PAGE_SIZE-1));
912
913 debug_check_no_locks_freed(mem, size);
914
915 if (likely(count <= VMAP_MAX_ALLOC))
916 vb_free(mem, size);
917 else
918 free_unmap_vmap_area_addr(addr);
919}
920EXPORT_SYMBOL(vm_unmap_ram);
921
922/**
923 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
924 * @pages: an array of pointers to the pages to be mapped
925 * @count: number of pages
926 * @node: prefer to allocate data structures on this node
927 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -0700928 *
929 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -0700930 */
931void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
932{
933 unsigned long size = count << PAGE_SHIFT;
934 unsigned long addr;
935 void *mem;
936
937 if (likely(count <= VMAP_MAX_ALLOC)) {
938 mem = vb_alloc(size, GFP_KERNEL);
939 if (IS_ERR(mem))
940 return NULL;
941 addr = (unsigned long)mem;
942 } else {
943 struct vmap_area *va;
944 va = alloc_vmap_area(size, PAGE_SIZE,
945 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
946 if (IS_ERR(va))
947 return NULL;
948
949 addr = va->va_start;
950 mem = (void *)addr;
951 }
952 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
953 vm_unmap_ram(mem, count);
954 return NULL;
955 }
956 return mem;
957}
958EXPORT_SYMBOL(vm_map_ram);
959
960void __init vmalloc_init(void)
961{
962 int i;
963
964 for_each_possible_cpu(i) {
965 struct vmap_block_queue *vbq;
966
967 vbq = &per_cpu(vmap_block_queue, i);
968 spin_lock_init(&vbq->lock);
969 INIT_LIST_HEAD(&vbq->free);
970 INIT_LIST_HEAD(&vbq->dirty);
971 vbq->nr_dirty = 0;
972 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100973
974 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700975}
976
977void unmap_kernel_range(unsigned long addr, unsigned long size)
978{
979 unsigned long end = addr + size;
980 vunmap_page_range(addr, end);
981 flush_tlb_kernel_range(addr, end);
982}
983
984int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
985{
986 unsigned long addr = (unsigned long)area->addr;
987 unsigned long end = addr + area->size - PAGE_SIZE;
988 int err;
989
990 err = vmap_page_range(addr, end, prot, *pages);
991 if (err > 0) {
992 *pages += err;
993 err = 0;
994 }
995
996 return err;
997}
998EXPORT_SYMBOL_GPL(map_vm_area);
999
1000/*** Old vmalloc interfaces ***/
1001DEFINE_RWLOCK(vmlist_lock);
1002struct vm_struct *vmlist;
1003
1004static struct vm_struct *__get_vm_area_node(unsigned long size,
1005 unsigned long flags, unsigned long start, unsigned long end,
1006 int node, gfp_t gfp_mask, void *caller)
1007{
1008 static struct vmap_area *va;
1009 struct vm_struct *area;
1010 struct vm_struct *tmp, **p;
1011 unsigned long align = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001013 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (flags & VM_IOREMAP) {
1015 int bit = fls(size);
1016
1017 if (bit > IOREMAP_MAX_ORDER)
1018 bit = IOREMAP_MAX_ORDER;
1019 else if (bit < PAGE_SHIFT)
1020 bit = PAGE_SHIFT;
1021
1022 align = 1ul << bit;
1023 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001026 if (unlikely(!size))
1027 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Christoph Lameter6cb06222007-10-16 01:25:41 -07001029 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (unlikely(!area))
1031 return NULL;
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 /*
1034 * We always allocate a guard page.
1035 */
1036 size += PAGE_SIZE;
1037
Nick Piggindb64fe02008-10-18 20:27:03 -07001038 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1039 if (IS_ERR(va)) {
1040 kfree(area);
1041 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 area->flags = flags;
Nick Piggindb64fe02008-10-18 20:27:03 -07001045 area->addr = (void *)va->va_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 area->size = size;
1047 area->pages = NULL;
1048 area->nr_pages = 0;
1049 area->phys_addr = 0;
Christoph Lameter23016962008-04-28 02:12:42 -07001050 area->caller = caller;
Nick Piggindb64fe02008-10-18 20:27:03 -07001051 va->private = area;
1052 va->flags |= VM_VM_AREA;
1053
1054 write_lock(&vmlist_lock);
1055 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1056 if (tmp->addr >= area->addr)
1057 break;
1058 }
1059 area->next = *p;
1060 *p = area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 write_unlock(&vmlist_lock);
1062
1063 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064}
1065
Christoph Lameter930fc452005-10-29 18:15:41 -07001066struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1067 unsigned long start, unsigned long end)
1068{
Christoph Lameter23016962008-04-28 02:12:42 -07001069 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1070 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001071}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001072EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001073
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074/**
Simon Arlott183ff222007-10-20 01:27:18 +02001075 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 * @size: size of the area
1077 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1078 *
1079 * Search an area of @size in the kernel virtual mapping area,
1080 * and reserved it for out purposes. Returns the area descriptor
1081 * on success or %NULL on failure.
1082 */
1083struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1084{
Christoph Lameter23016962008-04-28 02:12:42 -07001085 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1086 -1, GFP_KERNEL, __builtin_return_address(0));
1087}
1088
1089struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1090 void *caller)
1091{
1092 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1093 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094}
1095
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001096struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1097 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07001098{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001099 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
Christoph Lameter23016962008-04-28 02:12:42 -07001100 gfp_mask, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001101}
1102
Nick Piggindb64fe02008-10-18 20:27:03 -07001103static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001104{
Nick Piggindb64fe02008-10-18 20:27:03 -07001105 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001106
Nick Piggindb64fe02008-10-18 20:27:03 -07001107 va = find_vmap_area((unsigned long)addr);
1108 if (va && va->flags & VM_VM_AREA)
1109 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001110
Andi Kleen7856dfe2005-05-20 14:27:57 -07001111 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001112}
1113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114/**
Simon Arlott183ff222007-10-20 01:27:18 +02001115 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 * @addr: base address
1117 *
1118 * Search for the kernel VM area starting at @addr, and remove it.
1119 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001120 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001122struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123{
Nick Piggindb64fe02008-10-18 20:27:03 -07001124 struct vmap_area *va;
1125
1126 va = find_vmap_area((unsigned long)addr);
1127 if (va && va->flags & VM_VM_AREA) {
1128 struct vm_struct *vm = va->private;
1129 struct vm_struct *tmp, **p;
1130 free_unmap_vmap_area(va);
1131 vm->size -= PAGE_SIZE;
1132
1133 write_lock(&vmlist_lock);
1134 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1135 ;
1136 *p = tmp->next;
1137 write_unlock(&vmlist_lock);
1138
1139 return vm;
1140 }
1141 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142}
1143
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001144static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145{
1146 struct vm_struct *area;
1147
1148 if (!addr)
1149 return;
1150
1151 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001152 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 return;
1154 }
1155
1156 area = remove_vm_area(addr);
1157 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001158 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 return;
1161 }
1162
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001163 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001164 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 if (deallocate_pages) {
1167 int i;
1168
1169 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001170 struct page *page = area->pages[i];
1171
1172 BUG_ON(!page);
1173 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
1175
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001176 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 vfree(area->pages);
1178 else
1179 kfree(area->pages);
1180 }
1181
1182 kfree(area);
1183 return;
1184}
1185
1186/**
1187 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 * @addr: memory base address
1189 *
Simon Arlott183ff222007-10-20 01:27:18 +02001190 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001191 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1192 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001194 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001196void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
1198 BUG_ON(in_interrupt());
1199 __vunmap(addr, 1);
1200}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201EXPORT_SYMBOL(vfree);
1202
1203/**
1204 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 * @addr: memory base address
1206 *
1207 * Free the virtually contiguous memory area starting at @addr,
1208 * which was created from the page array passed to vmap().
1209 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001210 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001212void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
1214 BUG_ON(in_interrupt());
1215 __vunmap(addr, 0);
1216}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217EXPORT_SYMBOL(vunmap);
1218
1219/**
1220 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 * @pages: array of page pointers
1222 * @count: number of pages to map
1223 * @flags: vm_area->flags
1224 * @prot: page protection for the mapping
1225 *
1226 * Maps @count pages from @pages into contiguous kernel virtual
1227 * space.
1228 */
1229void *vmap(struct page **pages, unsigned int count,
1230 unsigned long flags, pgprot_t prot)
1231{
1232 struct vm_struct *area;
1233
1234 if (count > num_physpages)
1235 return NULL;
1236
Christoph Lameter23016962008-04-28 02:12:42 -07001237 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1238 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (!area)
1240 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (map_vm_area(area, prot, &pages)) {
1243 vunmap(area->addr);
1244 return NULL;
1245 }
1246
1247 return area->addr;
1248}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249EXPORT_SYMBOL(vmap);
1250
Nick Piggindb64fe02008-10-18 20:27:03 -07001251static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1252 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001253static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001254 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255{
1256 struct page **pages;
1257 unsigned int nr_pages, array_size, i;
1258
1259 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1260 array_size = (nr_pages * sizeof(struct page *));
1261
1262 area->nr_pages = nr_pages;
1263 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001264 if (array_size > PAGE_SIZE) {
Christoph Lameter94f60302007-07-17 04:03:29 -07001265 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
Christoph Lameter23016962008-04-28 02:12:42 -07001266 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001267 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001268 } else {
1269 pages = kmalloc_node(array_size,
Christoph Lameter6cb06222007-10-16 01:25:41 -07001270 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
Andrew Morton286e1ea2006-10-17 00:09:57 -07001271 node);
1272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001274 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 if (!area->pages) {
1276 remove_vm_area(area->addr);
1277 kfree(area);
1278 return NULL;
1279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
1281 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001282 struct page *page;
1283
Christoph Lameter930fc452005-10-29 18:15:41 -07001284 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001285 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001286 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001287 page = alloc_pages_node(node, gfp_mask, 0);
1288
1289 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 /* Successfully allocated i pages, free them in __vunmap() */
1291 area->nr_pages = i;
1292 goto fail;
1293 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001294 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 }
1296
1297 if (map_vm_area(area, prot, &pages))
1298 goto fail;
1299 return area->addr;
1300
1301fail:
1302 vfree(area->addr);
1303 return NULL;
1304}
1305
Christoph Lameter930fc452005-10-29 18:15:41 -07001306void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1307{
Christoph Lameter23016962008-04-28 02:12:42 -07001308 return __vmalloc_area_node(area, gfp_mask, prot, -1,
1309 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001310}
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001313 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 * @size: allocation size
1315 * @gfp_mask: flags for the page level allocator
1316 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001317 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001318 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 *
1320 * Allocate enough pages to cover @size from the page level
1321 * allocator with @gfp_mask flags. Map them into contiguous
1322 * kernel virtual space, using a pagetable protection of @prot.
1323 */
Adrian Bunkb2213852006-09-25 23:31:02 -07001324static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
Christoph Lameter23016962008-04-28 02:12:42 -07001325 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
1327 struct vm_struct *area;
1328
1329 size = PAGE_ALIGN(size);
1330 if (!size || (size >> PAGE_SHIFT) > num_physpages)
1331 return NULL;
1332
Christoph Lameter23016962008-04-28 02:12:42 -07001333 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1334 node, gfp_mask, caller);
1335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 if (!area)
1337 return NULL;
1338
Christoph Lameter23016962008-04-28 02:12:42 -07001339 return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340}
1341
Christoph Lameter930fc452005-10-29 18:15:41 -07001342void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1343{
Christoph Lameter23016962008-04-28 02:12:42 -07001344 return __vmalloc_node(size, gfp_mask, prot, -1,
1345 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001346}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347EXPORT_SYMBOL(__vmalloc);
1348
1349/**
1350 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 * Allocate enough pages to cover @size from the page level
1353 * allocator and map them into contiguous kernel virtual space.
1354 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001355 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 * use __vmalloc() instead.
1357 */
1358void *vmalloc(unsigned long size)
1359{
Christoph Lameter23016962008-04-28 02:12:42 -07001360 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1361 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363EXPORT_SYMBOL(vmalloc);
1364
Christoph Lameter930fc452005-10-29 18:15:41 -07001365/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001366 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1367 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001368 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001369 * The resulting memory area is zeroed so it can be mapped to userspace
1370 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001371 */
1372void *vmalloc_user(unsigned long size)
1373{
1374 struct vm_struct *area;
1375 void *ret;
1376
1377 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001378 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001379 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001380 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001381 }
Nick Piggin83342312006-06-23 02:03:20 -07001382 return ret;
1383}
1384EXPORT_SYMBOL(vmalloc_user);
1385
1386/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001387 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001388 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001389 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001390 *
1391 * Allocate enough pages to cover @size from the page level
1392 * allocator and map them into contiguous kernel virtual space.
1393 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001394 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001395 * use __vmalloc() instead.
1396 */
1397void *vmalloc_node(unsigned long size, int node)
1398{
Christoph Lameter23016962008-04-28 02:12:42 -07001399 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1400 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001401}
1402EXPORT_SYMBOL(vmalloc_node);
1403
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001404#ifndef PAGE_KERNEL_EXEC
1405# define PAGE_KERNEL_EXEC PAGE_KERNEL
1406#endif
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408/**
1409 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 * @size: allocation size
1411 *
1412 * Kernel-internal function to allocate enough pages to cover @size
1413 * the page level allocator and map them into contiguous and
1414 * executable kernel virtual space.
1415 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001416 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 * use __vmalloc() instead.
1418 */
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420void *vmalloc_exec(unsigned long size)
1421{
1422 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
1423}
1424
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001425#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001426#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001427#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001428#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001429#else
1430#define GFP_VMALLOC32 GFP_KERNEL
1431#endif
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433/**
1434 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 * @size: allocation size
1436 *
1437 * Allocate enough 32bit PA addressable pages to cover @size from the
1438 * page level allocator and map them into contiguous kernel virtual space.
1439 */
1440void *vmalloc_32(unsigned long size)
1441{
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001442 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444EXPORT_SYMBOL(vmalloc_32);
1445
Nick Piggin83342312006-06-23 02:03:20 -07001446/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001447 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001448 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001449 *
1450 * The resulting memory area is 32bit addressable and zeroed so it can be
1451 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001452 */
1453void *vmalloc_32_user(unsigned long size)
1454{
1455 struct vm_struct *area;
1456 void *ret;
1457
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001458 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001459 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001460 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001461 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001462 }
Nick Piggin83342312006-06-23 02:03:20 -07001463 return ret;
1464}
1465EXPORT_SYMBOL(vmalloc_32_user);
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467long vread(char *buf, char *addr, unsigned long count)
1468{
1469 struct vm_struct *tmp;
1470 char *vaddr, *buf_start = buf;
1471 unsigned long n;
1472
1473 /* Don't allow overflow */
1474 if ((unsigned long) addr + count < count)
1475 count = -(unsigned long) addr;
1476
1477 read_lock(&vmlist_lock);
1478 for (tmp = vmlist; tmp; tmp = tmp->next) {
1479 vaddr = (char *) tmp->addr;
1480 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1481 continue;
1482 while (addr < vaddr) {
1483 if (count == 0)
1484 goto finished;
1485 *buf = '\0';
1486 buf++;
1487 addr++;
1488 count--;
1489 }
1490 n = vaddr + tmp->size - PAGE_SIZE - addr;
1491 do {
1492 if (count == 0)
1493 goto finished;
1494 *buf = *addr;
1495 buf++;
1496 addr++;
1497 count--;
1498 } while (--n > 0);
1499 }
1500finished:
1501 read_unlock(&vmlist_lock);
1502 return buf - buf_start;
1503}
1504
1505long vwrite(char *buf, char *addr, unsigned long count)
1506{
1507 struct vm_struct *tmp;
1508 char *vaddr, *buf_start = buf;
1509 unsigned long n;
1510
1511 /* Don't allow overflow */
1512 if ((unsigned long) addr + count < count)
1513 count = -(unsigned long) addr;
1514
1515 read_lock(&vmlist_lock);
1516 for (tmp = vmlist; tmp; tmp = tmp->next) {
1517 vaddr = (char *) tmp->addr;
1518 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1519 continue;
1520 while (addr < vaddr) {
1521 if (count == 0)
1522 goto finished;
1523 buf++;
1524 addr++;
1525 count--;
1526 }
1527 n = vaddr + tmp->size - PAGE_SIZE - addr;
1528 do {
1529 if (count == 0)
1530 goto finished;
1531 *addr = *buf;
1532 buf++;
1533 addr++;
1534 count--;
1535 } while (--n > 0);
1536 }
1537finished:
1538 read_unlock(&vmlist_lock);
1539 return buf - buf_start;
1540}
Nick Piggin83342312006-06-23 02:03:20 -07001541
1542/**
1543 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07001544 * @vma: vma to cover (map full range of vma)
1545 * @addr: vmalloc memory
1546 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07001547 *
1548 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07001549 *
1550 * This function checks that addr is a valid vmalloc'ed area, and
1551 * that it is big enough to cover the vma. Will return failure if
1552 * that criteria isn't met.
1553 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001554 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07001555 */
1556int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1557 unsigned long pgoff)
1558{
1559 struct vm_struct *area;
1560 unsigned long uaddr = vma->vm_start;
1561 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07001562
1563 if ((PAGE_SIZE-1) & (unsigned long)addr)
1564 return -EINVAL;
1565
Nick Piggindb64fe02008-10-18 20:27:03 -07001566 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07001567 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07001568 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001569
1570 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07001571 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001572
1573 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07001574 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001575
1576 addr += pgoff << PAGE_SHIFT;
1577 do {
1578 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07001579 int ret;
1580
Nick Piggin83342312006-06-23 02:03:20 -07001581 ret = vm_insert_page(vma, uaddr, page);
1582 if (ret)
1583 return ret;
1584
1585 uaddr += PAGE_SIZE;
1586 addr += PAGE_SIZE;
1587 usize -= PAGE_SIZE;
1588 } while (usize > 0);
1589
1590 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1591 vma->vm_flags |= VM_RESERVED;
1592
Nick Piggindb64fe02008-10-18 20:27:03 -07001593 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07001594}
1595EXPORT_SYMBOL(remap_vmalloc_range);
1596
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07001597/*
1598 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1599 * have one.
1600 */
1601void __attribute__((weak)) vmalloc_sync_all(void)
1602{
1603}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001604
1605
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001606static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001607{
1608 /* apply_to_page_range() does all the hard work. */
1609 return 0;
1610}
1611
1612/**
1613 * alloc_vm_area - allocate a range of kernel address space
1614 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07001615 *
1616 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001617 *
1618 * This function reserves a range of kernel address space, and
1619 * allocates pagetables to map that range. No actual mappings
1620 * are created. If the kernel address space is not shared
1621 * between processes, it syncs the pagetable across all
1622 * processes.
1623 */
1624struct vm_struct *alloc_vm_area(size_t size)
1625{
1626 struct vm_struct *area;
1627
Christoph Lameter23016962008-04-28 02:12:42 -07001628 area = get_vm_area_caller(size, VM_IOREMAP,
1629 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001630 if (area == NULL)
1631 return NULL;
1632
1633 /*
1634 * This ensures that page tables are constructed for this region
1635 * of kernel virtual address space and mapped into init_mm.
1636 */
1637 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1638 area->size, f, NULL)) {
1639 free_vm_area(area);
1640 return NULL;
1641 }
1642
1643 /* Make sure the pagetables are constructed in process kernel
1644 mappings */
1645 vmalloc_sync_all();
1646
1647 return area;
1648}
1649EXPORT_SYMBOL_GPL(alloc_vm_area);
1650
1651void free_vm_area(struct vm_struct *area)
1652{
1653 struct vm_struct *ret;
1654 ret = remove_vm_area(area->addr);
1655 BUG_ON(ret != area);
1656 kfree(area);
1657}
1658EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07001659
1660
1661#ifdef CONFIG_PROC_FS
1662static void *s_start(struct seq_file *m, loff_t *pos)
1663{
1664 loff_t n = *pos;
1665 struct vm_struct *v;
1666
1667 read_lock(&vmlist_lock);
1668 v = vmlist;
1669 while (n > 0 && v) {
1670 n--;
1671 v = v->next;
1672 }
1673 if (!n)
1674 return v;
1675
1676 return NULL;
1677
1678}
1679
1680static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1681{
1682 struct vm_struct *v = p;
1683
1684 ++*pos;
1685 return v->next;
1686}
1687
1688static void s_stop(struct seq_file *m, void *p)
1689{
1690 read_unlock(&vmlist_lock);
1691}
1692
Eric Dumazeta47a1262008-07-23 21:27:38 -07001693static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1694{
1695 if (NUMA_BUILD) {
1696 unsigned int nr, *counters = m->private;
1697
1698 if (!counters)
1699 return;
1700
1701 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1702
1703 for (nr = 0; nr < v->nr_pages; nr++)
1704 counters[page_to_nid(v->pages[nr])]++;
1705
1706 for_each_node_state(nr, N_HIGH_MEMORY)
1707 if (counters[nr])
1708 seq_printf(m, " N%u=%u", nr, counters[nr]);
1709 }
1710}
1711
Christoph Lametera10aa572008-04-28 02:12:40 -07001712static int s_show(struct seq_file *m, void *p)
1713{
1714 struct vm_struct *v = p;
1715
1716 seq_printf(m, "0x%p-0x%p %7ld",
1717 v->addr, v->addr + v->size, v->size);
1718
Christoph Lameter23016962008-04-28 02:12:42 -07001719 if (v->caller) {
Hugh Dickins9c246242008-12-09 13:14:27 -08001720 char buff[KSYM_SYMBOL_LEN];
Christoph Lameter23016962008-04-28 02:12:42 -07001721
1722 seq_putc(m, ' ');
1723 sprint_symbol(buff, (unsigned long)v->caller);
1724 seq_puts(m, buff);
1725 }
1726
Christoph Lametera10aa572008-04-28 02:12:40 -07001727 if (v->nr_pages)
1728 seq_printf(m, " pages=%d", v->nr_pages);
1729
1730 if (v->phys_addr)
1731 seq_printf(m, " phys=%lx", v->phys_addr);
1732
1733 if (v->flags & VM_IOREMAP)
1734 seq_printf(m, " ioremap");
1735
1736 if (v->flags & VM_ALLOC)
1737 seq_printf(m, " vmalloc");
1738
1739 if (v->flags & VM_MAP)
1740 seq_printf(m, " vmap");
1741
1742 if (v->flags & VM_USERMAP)
1743 seq_printf(m, " user");
1744
1745 if (v->flags & VM_VPAGES)
1746 seq_printf(m, " vpages");
1747
Eric Dumazeta47a1262008-07-23 21:27:38 -07001748 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07001749 seq_putc(m, '\n');
1750 return 0;
1751}
1752
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001753static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07001754 .start = s_start,
1755 .next = s_next,
1756 .stop = s_stop,
1757 .show = s_show,
1758};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001759
1760static int vmalloc_open(struct inode *inode, struct file *file)
1761{
1762 unsigned int *ptr = NULL;
1763 int ret;
1764
1765 if (NUMA_BUILD)
1766 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1767 ret = seq_open(file, &vmalloc_op);
1768 if (!ret) {
1769 struct seq_file *m = file->private_data;
1770 m->private = ptr;
1771 } else
1772 kfree(ptr);
1773 return ret;
1774}
1775
1776static const struct file_operations proc_vmalloc_operations = {
1777 .open = vmalloc_open,
1778 .read = seq_read,
1779 .llseek = seq_lseek,
1780 .release = seq_release_private,
1781};
1782
1783static int __init proc_vmalloc_init(void)
1784{
1785 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1786 return 0;
1787}
1788module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07001789#endif
1790