blob: 75f49d312e8c1d47648f3e96b8a1eb6d14076405 [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>
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -080026#include <linux/bootmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Nick Piggindb64fe02008-10-18 20:27:03 -070028#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <asm/uaccess.h>
30#include <asm/tlbflush.h>
31
32
Nick Piggindb64fe02008-10-18 20:27:03 -070033/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
36{
37 pte_t *pte;
38
39 pte = pte_offset_kernel(pmd, addr);
40 do {
41 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
42 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
43 } while (pte++, addr += PAGE_SIZE, addr != end);
44}
45
Nick Piggindb64fe02008-10-18 20:27:03 -070046static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
48 pmd_t *pmd;
49 unsigned long next;
50
51 pmd = pmd_offset(pud, addr);
52 do {
53 next = pmd_addr_end(addr, end);
54 if (pmd_none_or_clear_bad(pmd))
55 continue;
56 vunmap_pte_range(pmd, addr, next);
57 } while (pmd++, addr = next, addr != end);
58}
59
Nick Piggindb64fe02008-10-18 20:27:03 -070060static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061{
62 pud_t *pud;
63 unsigned long next;
64
65 pud = pud_offset(pgd, addr);
66 do {
67 next = pud_addr_end(addr, end);
68 if (pud_none_or_clear_bad(pud))
69 continue;
70 vunmap_pmd_range(pud, addr, next);
71 } while (pud++, addr = next, addr != end);
72}
73
Nick Piggindb64fe02008-10-18 20:27:03 -070074static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
76 pgd_t *pgd;
77 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79 BUG_ON(addr >= end);
80 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 do {
82 next = pgd_addr_end(addr, end);
83 if (pgd_none_or_clear_bad(pgd))
84 continue;
85 vunmap_pud_range(pgd, addr, next);
86 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087}
88
89static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -070090 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
92 pte_t *pte;
93
Nick Piggindb64fe02008-10-18 20:27:03 -070094 /*
95 * nr is a running index into the array which helps higher level
96 * callers keep track of where we're up to.
97 */
98
Hugh Dickins872fec12005-10-29 18:16:21 -070099 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 if (!pte)
101 return -ENOMEM;
102 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700103 struct page *page = pages[*nr];
104
105 if (WARN_ON(!pte_none(*pte)))
106 return -EBUSY;
107 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 return -ENOMEM;
109 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700110 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 } while (pte++, addr += PAGE_SIZE, addr != end);
112 return 0;
113}
114
Nick Piggindb64fe02008-10-18 20:27:03 -0700115static int vmap_pmd_range(pud_t *pud, unsigned long addr,
116 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
118 pmd_t *pmd;
119 unsigned long next;
120
121 pmd = pmd_alloc(&init_mm, pud, addr);
122 if (!pmd)
123 return -ENOMEM;
124 do {
125 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700126 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 return -ENOMEM;
128 } while (pmd++, addr = next, addr != end);
129 return 0;
130}
131
Nick Piggindb64fe02008-10-18 20:27:03 -0700132static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
133 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134{
135 pud_t *pud;
136 unsigned long next;
137
138 pud = pud_alloc(&init_mm, pgd, addr);
139 if (!pud)
140 return -ENOMEM;
141 do {
142 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700143 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 return -ENOMEM;
145 } while (pud++, addr = next, addr != end);
146 return 0;
147}
148
Nick Piggindb64fe02008-10-18 20:27:03 -0700149/*
150 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
151 * will have pfns corresponding to the "pages" array.
152 *
153 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
154 */
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800155static int vmap_page_range(unsigned long start, unsigned long end,
Nick Piggindb64fe02008-10-18 20:27:03 -0700156 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
158 pgd_t *pgd;
159 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800160 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700161 int err = 0;
162 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164 BUG_ON(addr >= end);
165 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 do {
167 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700168 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 if (err)
170 break;
171 } while (pgd++, addr = next, addr != end);
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800172 flush_cache_vmap(start, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700173
174 if (unlikely(err))
175 return err;
176 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700179static inline int is_vmalloc_or_module_addr(const void *x)
180{
181 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000182 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700183 * and fall back on vmalloc() if that fails. Others
184 * just put it in the vmalloc space.
185 */
186#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
187 unsigned long addr = (unsigned long)x;
188 if (addr >= MODULES_VADDR && addr < MODULES_END)
189 return 1;
190#endif
191 return is_vmalloc_addr(x);
192}
193
Christoph Lameter48667e72008-02-04 22:28:31 -0800194/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700195 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800196 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800197struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800198{
199 unsigned long addr = (unsigned long) vmalloc_addr;
200 struct page *page = NULL;
201 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800202
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200203 /*
204 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
205 * architectures that do not vmalloc module space
206 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700207 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200208
Christoph Lameter48667e72008-02-04 22:28:31 -0800209 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700210 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800211 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700212 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800213 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700214 pte_t *ptep, pte;
215
Christoph Lameter48667e72008-02-04 22:28:31 -0800216 ptep = pte_offset_map(pmd, addr);
217 pte = *ptep;
218 if (pte_present(pte))
219 page = pte_page(pte);
220 pte_unmap(ptep);
221 }
222 }
223 }
224 return page;
225}
226EXPORT_SYMBOL(vmalloc_to_page);
227
228/*
229 * Map a vmalloc()-space virtual address to the physical page frame number.
230 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800231unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800232{
233 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
234}
235EXPORT_SYMBOL(vmalloc_to_pfn);
236
Nick Piggindb64fe02008-10-18 20:27:03 -0700237
238/*** Global kva allocator ***/
239
240#define VM_LAZY_FREE 0x01
241#define VM_LAZY_FREEING 0x02
242#define VM_VM_AREA 0x04
243
244struct vmap_area {
245 unsigned long va_start;
246 unsigned long va_end;
247 unsigned long flags;
248 struct rb_node rb_node; /* address sorted rbtree */
249 struct list_head list; /* address sorted list */
250 struct list_head purge_list; /* "lazy purge" list */
251 void *private;
252 struct rcu_head rcu_head;
253};
254
255static DEFINE_SPINLOCK(vmap_area_lock);
256static struct rb_root vmap_area_root = RB_ROOT;
257static LIST_HEAD(vmap_area_list);
258
259static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Nick Piggindb64fe02008-10-18 20:27:03 -0700261 struct rb_node *n = vmap_area_root.rb_node;
262
263 while (n) {
264 struct vmap_area *va;
265
266 va = rb_entry(n, struct vmap_area, rb_node);
267 if (addr < va->va_start)
268 n = n->rb_left;
269 else if (addr > va->va_start)
270 n = n->rb_right;
271 else
272 return va;
273 }
274
275 return NULL;
276}
277
278static void __insert_vmap_area(struct vmap_area *va)
279{
280 struct rb_node **p = &vmap_area_root.rb_node;
281 struct rb_node *parent = NULL;
282 struct rb_node *tmp;
283
284 while (*p) {
285 struct vmap_area *tmp;
286
287 parent = *p;
288 tmp = rb_entry(parent, struct vmap_area, rb_node);
289 if (va->va_start < tmp->va_end)
290 p = &(*p)->rb_left;
291 else if (va->va_end > tmp->va_start)
292 p = &(*p)->rb_right;
293 else
294 BUG();
295 }
296
297 rb_link_node(&va->rb_node, parent, p);
298 rb_insert_color(&va->rb_node, &vmap_area_root);
299
300 /* address-sort this list so it is usable like the vmlist */
301 tmp = rb_prev(&va->rb_node);
302 if (tmp) {
303 struct vmap_area *prev;
304 prev = rb_entry(tmp, struct vmap_area, rb_node);
305 list_add_rcu(&va->list, &prev->list);
306 } else
307 list_add_rcu(&va->list, &vmap_area_list);
308}
309
310static void purge_vmap_area_lazy(void);
311
312/*
313 * Allocate a region of KVA of the specified size and alignment, within the
314 * vstart and vend.
315 */
316static struct vmap_area *alloc_vmap_area(unsigned long size,
317 unsigned long align,
318 unsigned long vstart, unsigned long vend,
319 int node, gfp_t gfp_mask)
320{
321 struct vmap_area *va;
322 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700324 int purged = 0;
325
326 BUG_ON(size & ~PAGE_MASK);
327
Nick Piggindb64fe02008-10-18 20:27:03 -0700328 va = kmalloc_node(sizeof(struct vmap_area),
329 gfp_mask & GFP_RECLAIM_MASK, node);
330 if (unlikely(!va))
331 return ERR_PTR(-ENOMEM);
332
333retry:
Glauber Costa0ae15132008-11-19 15:36:33 -0800334 addr = ALIGN(vstart, align);
335
Nick Piggindb64fe02008-10-18 20:27:03 -0700336 spin_lock(&vmap_area_lock);
337 /* XXX: could have a last_hole cache */
338 n = vmap_area_root.rb_node;
339 if (n) {
340 struct vmap_area *first = NULL;
341
342 do {
343 struct vmap_area *tmp;
344 tmp = rb_entry(n, struct vmap_area, rb_node);
345 if (tmp->va_end >= addr) {
346 if (!first && tmp->va_start < addr + size)
347 first = tmp;
348 n = n->rb_left;
349 } else {
350 first = tmp;
351 n = n->rb_right;
352 }
353 } while (n);
354
355 if (!first)
356 goto found;
357
358 if (first->va_end < addr) {
359 n = rb_next(&first->rb_node);
360 if (n)
361 first = rb_entry(n, struct vmap_area, rb_node);
362 else
363 goto found;
364 }
365
Nick Pigginf011c2d2008-11-19 15:36:32 -0800366 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700367 addr = ALIGN(first->va_end + PAGE_SIZE, align);
368
369 n = rb_next(&first->rb_node);
370 if (n)
371 first = rb_entry(n, struct vmap_area, rb_node);
372 else
373 goto found;
374 }
375 }
376found:
377 if (addr + size > vend) {
378 spin_unlock(&vmap_area_lock);
379 if (!purged) {
380 purge_vmap_area_lazy();
381 purged = 1;
382 goto retry;
383 }
384 if (printk_ratelimit())
Glauber Costac1279c42009-01-06 14:39:18 -0800385 printk(KERN_WARNING
386 "vmap allocation for size %lu failed: "
387 "use vmalloc=<size> to increase size.\n", size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700388 return ERR_PTR(-EBUSY);
389 }
390
391 BUG_ON(addr & (align-1));
392
393 va->va_start = addr;
394 va->va_end = addr + size;
395 va->flags = 0;
396 __insert_vmap_area(va);
397 spin_unlock(&vmap_area_lock);
398
399 return va;
400}
401
402static void rcu_free_va(struct rcu_head *head)
403{
404 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
405
406 kfree(va);
407}
408
409static void __free_vmap_area(struct vmap_area *va)
410{
411 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
412 rb_erase(&va->rb_node, &vmap_area_root);
413 RB_CLEAR_NODE(&va->rb_node);
414 list_del_rcu(&va->list);
415
416 call_rcu(&va->rcu_head, rcu_free_va);
417}
418
419/*
420 * Free a region of KVA allocated by alloc_vmap_area
421 */
422static void free_vmap_area(struct vmap_area *va)
423{
424 spin_lock(&vmap_area_lock);
425 __free_vmap_area(va);
426 spin_unlock(&vmap_area_lock);
427}
428
429/*
430 * Clear the pagetable entries of a given vmap_area
431 */
432static void unmap_vmap_area(struct vmap_area *va)
433{
434 vunmap_page_range(va->va_start, va->va_end);
435}
436
Nick Piggincd528582009-01-06 14:39:20 -0800437static void vmap_debug_free_range(unsigned long start, unsigned long end)
438{
439 /*
440 * Unmap page tables and force a TLB flush immediately if
441 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
442 * bugs similarly to those in linear kernel virtual address
443 * space after a page has been freed.
444 *
445 * All the lazy freeing logic is still retained, in order to
446 * minimise intrusiveness of this debugging feature.
447 *
448 * This is going to be *slow* (linear kernel virtual address
449 * debugging doesn't do a broadcast TLB flush so it is a lot
450 * faster).
451 */
452#ifdef CONFIG_DEBUG_PAGEALLOC
453 vunmap_page_range(start, end);
454 flush_tlb_kernel_range(start, end);
455#endif
456}
457
Nick Piggindb64fe02008-10-18 20:27:03 -0700458/*
459 * lazy_max_pages is the maximum amount of virtual address space we gather up
460 * before attempting to purge with a TLB flush.
461 *
462 * There is a tradeoff here: a larger number will cover more kernel page tables
463 * and take slightly longer to purge, but it will linearly reduce the number of
464 * global TLB flushes that must be performed. It would seem natural to scale
465 * this number up linearly with the number of CPUs (because vmapping activity
466 * could also scale linearly with the number of CPUs), however it is likely
467 * that in practice, workloads might be constrained in other ways that mean
468 * vmap activity will not scale linearly with CPUs. Also, I want to be
469 * conservative and not introduce a big latency on huge systems, so go with
470 * a less aggressive log scale. It will still be an improvement over the old
471 * code, and it will be simple to change the scale factor if we find that it
472 * becomes a problem on bigger systems.
473 */
474static unsigned long lazy_max_pages(void)
475{
476 unsigned int log;
477
478 log = fls(num_online_cpus());
479
480 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
481}
482
483static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
484
485/*
486 * Purges all lazily-freed vmap areas.
487 *
488 * If sync is 0 then don't purge if there is already a purge in progress.
489 * If force_flush is 1, then flush kernel TLBs between *start and *end even
490 * if we found no lazy vmap areas to unmap (callers can use this to optimise
491 * their own TLB flushing).
492 * Returns with *start = min(*start, lowest purged address)
493 * *end = max(*end, highest purged address)
494 */
495static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
496 int sync, int force_flush)
497{
Andrew Morton46666d82009-01-15 13:51:15 -0800498 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700499 LIST_HEAD(valist);
500 struct vmap_area *va;
501 int nr = 0;
502
503 /*
504 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
505 * should not expect such behaviour. This just simplifies locking for
506 * the case that isn't actually used at the moment anyway.
507 */
508 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800509 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700510 return;
511 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800512 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700513
514 rcu_read_lock();
515 list_for_each_entry_rcu(va, &vmap_area_list, list) {
516 if (va->flags & VM_LAZY_FREE) {
517 if (va->va_start < *start)
518 *start = va->va_start;
519 if (va->va_end > *end)
520 *end = va->va_end;
521 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
522 unmap_vmap_area(va);
523 list_add_tail(&va->purge_list, &valist);
524 va->flags |= VM_LAZY_FREEING;
525 va->flags &= ~VM_LAZY_FREE;
526 }
527 }
528 rcu_read_unlock();
529
530 if (nr) {
531 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
532 atomic_sub(nr, &vmap_lazy_nr);
533 }
534
535 if (nr || force_flush)
536 flush_tlb_kernel_range(*start, *end);
537
538 if (nr) {
539 spin_lock(&vmap_area_lock);
540 list_for_each_entry(va, &valist, purge_list)
541 __free_vmap_area(va);
542 spin_unlock(&vmap_area_lock);
543 }
Andrew Morton46666d82009-01-15 13:51:15 -0800544 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700545}
546
547/*
Nick Piggin496850e2008-11-19 15:36:33 -0800548 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
549 * is already purging.
550 */
551static void try_purge_vmap_area_lazy(void)
552{
553 unsigned long start = ULONG_MAX, end = 0;
554
555 __purge_vmap_area_lazy(&start, &end, 0, 0);
556}
557
558/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700559 * Kick off a purge of the outstanding lazy areas.
560 */
561static void purge_vmap_area_lazy(void)
562{
563 unsigned long start = ULONG_MAX, end = 0;
564
Nick Piggin496850e2008-11-19 15:36:33 -0800565 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700566}
567
568/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800569 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
570 * called for the correct range previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700571 */
Nick Pigginb29acbd2008-12-01 13:13:47 -0800572static void free_unmap_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700573{
574 va->flags |= VM_LAZY_FREE;
575 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
576 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800577 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700578}
579
Nick Pigginb29acbd2008-12-01 13:13:47 -0800580/*
581 * Free and unmap a vmap area
582 */
583static void free_unmap_vmap_area(struct vmap_area *va)
584{
585 flush_cache_vunmap(va->va_start, va->va_end);
586 free_unmap_vmap_area_noflush(va);
587}
588
Nick Piggindb64fe02008-10-18 20:27:03 -0700589static struct vmap_area *find_vmap_area(unsigned long addr)
590{
591 struct vmap_area *va;
592
593 spin_lock(&vmap_area_lock);
594 va = __find_vmap_area(addr);
595 spin_unlock(&vmap_area_lock);
596
597 return va;
598}
599
600static void free_unmap_vmap_area_addr(unsigned long addr)
601{
602 struct vmap_area *va;
603
604 va = find_vmap_area(addr);
605 BUG_ON(!va);
606 free_unmap_vmap_area(va);
607}
608
609
610/*** Per cpu kva allocator ***/
611
612/*
613 * vmap space is limited especially on 32 bit architectures. Ensure there is
614 * room for at least 16 percpu vmap blocks per CPU.
615 */
616/*
617 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
618 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
619 * instead (we just need a rough idea)
620 */
621#if BITS_PER_LONG == 32
622#define VMALLOC_SPACE (128UL*1024*1024)
623#else
624#define VMALLOC_SPACE (128UL*1024*1024*1024)
625#endif
626
627#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
628#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
629#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
630#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
631#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
632#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
633#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
634 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
635 VMALLOC_PAGES / NR_CPUS / 16))
636
637#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
638
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100639static bool vmap_initialized __read_mostly = false;
640
Nick Piggindb64fe02008-10-18 20:27:03 -0700641struct vmap_block_queue {
642 spinlock_t lock;
643 struct list_head free;
644 struct list_head dirty;
645 unsigned int nr_dirty;
646};
647
648struct vmap_block {
649 spinlock_t lock;
650 struct vmap_area *va;
651 struct vmap_block_queue *vbq;
652 unsigned long free, dirty;
653 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
654 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
655 union {
656 struct {
657 struct list_head free_list;
658 struct list_head dirty_list;
659 };
660 struct rcu_head rcu_head;
661 };
662};
663
664/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
665static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
666
667/*
668 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
669 * in the free path. Could get rid of this if we change the API to return a
670 * "cookie" from alloc, to be passed to free. But no big deal yet.
671 */
672static DEFINE_SPINLOCK(vmap_block_tree_lock);
673static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
674
675/*
676 * We should probably have a fallback mechanism to allocate virtual memory
677 * out of partially filled vmap blocks. However vmap block sizing should be
678 * fairly reasonable according to the vmalloc size, so it shouldn't be a
679 * big problem.
680 */
681
682static unsigned long addr_to_vb_idx(unsigned long addr)
683{
684 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
685 addr /= VMAP_BLOCK_SIZE;
686 return addr;
687}
688
689static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
690{
691 struct vmap_block_queue *vbq;
692 struct vmap_block *vb;
693 struct vmap_area *va;
694 unsigned long vb_idx;
695 int node, err;
696
697 node = numa_node_id();
698
699 vb = kmalloc_node(sizeof(struct vmap_block),
700 gfp_mask & GFP_RECLAIM_MASK, node);
701 if (unlikely(!vb))
702 return ERR_PTR(-ENOMEM);
703
704 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
705 VMALLOC_START, VMALLOC_END,
706 node, gfp_mask);
707 if (unlikely(IS_ERR(va))) {
708 kfree(vb);
709 return ERR_PTR(PTR_ERR(va));
710 }
711
712 err = radix_tree_preload(gfp_mask);
713 if (unlikely(err)) {
714 kfree(vb);
715 free_vmap_area(va);
716 return ERR_PTR(err);
717 }
718
719 spin_lock_init(&vb->lock);
720 vb->va = va;
721 vb->free = VMAP_BBMAP_BITS;
722 vb->dirty = 0;
723 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
724 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
725 INIT_LIST_HEAD(&vb->free_list);
726 INIT_LIST_HEAD(&vb->dirty_list);
727
728 vb_idx = addr_to_vb_idx(va->va_start);
729 spin_lock(&vmap_block_tree_lock);
730 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
731 spin_unlock(&vmap_block_tree_lock);
732 BUG_ON(err);
733 radix_tree_preload_end();
734
735 vbq = &get_cpu_var(vmap_block_queue);
736 vb->vbq = vbq;
737 spin_lock(&vbq->lock);
738 list_add(&vb->free_list, &vbq->free);
739 spin_unlock(&vbq->lock);
740 put_cpu_var(vmap_cpu_blocks);
741
742 return vb;
743}
744
745static void rcu_free_vb(struct rcu_head *head)
746{
747 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
748
749 kfree(vb);
750}
751
752static void free_vmap_block(struct vmap_block *vb)
753{
754 struct vmap_block *tmp;
755 unsigned long vb_idx;
756
757 spin_lock(&vb->vbq->lock);
758 if (!list_empty(&vb->free_list))
759 list_del(&vb->free_list);
760 if (!list_empty(&vb->dirty_list))
761 list_del(&vb->dirty_list);
762 spin_unlock(&vb->vbq->lock);
763
764 vb_idx = addr_to_vb_idx(vb->va->va_start);
765 spin_lock(&vmap_block_tree_lock);
766 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
767 spin_unlock(&vmap_block_tree_lock);
768 BUG_ON(tmp != vb);
769
Nick Pigginb29acbd2008-12-01 13:13:47 -0800770 free_unmap_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700771 call_rcu(&vb->rcu_head, rcu_free_vb);
772}
773
774static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
775{
776 struct vmap_block_queue *vbq;
777 struct vmap_block *vb;
778 unsigned long addr = 0;
779 unsigned int order;
780
781 BUG_ON(size & ~PAGE_MASK);
782 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
783 order = get_order(size);
784
785again:
786 rcu_read_lock();
787 vbq = &get_cpu_var(vmap_block_queue);
788 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
789 int i;
790
791 spin_lock(&vb->lock);
792 i = bitmap_find_free_region(vb->alloc_map,
793 VMAP_BBMAP_BITS, order);
794
795 if (i >= 0) {
796 addr = vb->va->va_start + (i << PAGE_SHIFT);
797 BUG_ON(addr_to_vb_idx(addr) !=
798 addr_to_vb_idx(vb->va->va_start));
799 vb->free -= 1UL << order;
800 if (vb->free == 0) {
801 spin_lock(&vbq->lock);
802 list_del_init(&vb->free_list);
803 spin_unlock(&vbq->lock);
804 }
805 spin_unlock(&vb->lock);
806 break;
807 }
808 spin_unlock(&vb->lock);
809 }
810 put_cpu_var(vmap_cpu_blocks);
811 rcu_read_unlock();
812
813 if (!addr) {
814 vb = new_vmap_block(gfp_mask);
815 if (IS_ERR(vb))
816 return vb;
817 goto again;
818 }
819
820 return (void *)addr;
821}
822
823static void vb_free(const void *addr, unsigned long size)
824{
825 unsigned long offset;
826 unsigned long vb_idx;
827 unsigned int order;
828 struct vmap_block *vb;
829
830 BUG_ON(size & ~PAGE_MASK);
831 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800832
833 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
834
Nick Piggindb64fe02008-10-18 20:27:03 -0700835 order = get_order(size);
836
837 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
838
839 vb_idx = addr_to_vb_idx((unsigned long)addr);
840 rcu_read_lock();
841 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
842 rcu_read_unlock();
843 BUG_ON(!vb);
844
845 spin_lock(&vb->lock);
846 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
847 if (!vb->dirty) {
848 spin_lock(&vb->vbq->lock);
849 list_add(&vb->dirty_list, &vb->vbq->dirty);
850 spin_unlock(&vb->vbq->lock);
851 }
852 vb->dirty += 1UL << order;
853 if (vb->dirty == VMAP_BBMAP_BITS) {
854 BUG_ON(vb->free || !list_empty(&vb->free_list));
855 spin_unlock(&vb->lock);
856 free_vmap_block(vb);
857 } else
858 spin_unlock(&vb->lock);
859}
860
861/**
862 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
863 *
864 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
865 * to amortize TLB flushing overheads. What this means is that any page you
866 * have now, may, in a former life, have been mapped into kernel virtual
867 * address by the vmap layer and so there might be some CPUs with TLB entries
868 * still referencing that page (additional to the regular 1:1 kernel mapping).
869 *
870 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
871 * be sure that none of the pages we have control over will have any aliases
872 * from the vmap layer.
873 */
874void vm_unmap_aliases(void)
875{
876 unsigned long start = ULONG_MAX, end = 0;
877 int cpu;
878 int flush = 0;
879
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100880 if (unlikely(!vmap_initialized))
881 return;
882
Nick Piggindb64fe02008-10-18 20:27:03 -0700883 for_each_possible_cpu(cpu) {
884 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
885 struct vmap_block *vb;
886
887 rcu_read_lock();
888 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
889 int i;
890
891 spin_lock(&vb->lock);
892 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
893 while (i < VMAP_BBMAP_BITS) {
894 unsigned long s, e;
895 int j;
896 j = find_next_zero_bit(vb->dirty_map,
897 VMAP_BBMAP_BITS, i);
898
899 s = vb->va->va_start + (i << PAGE_SHIFT);
900 e = vb->va->va_start + (j << PAGE_SHIFT);
901 vunmap_page_range(s, e);
902 flush = 1;
903
904 if (s < start)
905 start = s;
906 if (e > end)
907 end = e;
908
909 i = j;
910 i = find_next_bit(vb->dirty_map,
911 VMAP_BBMAP_BITS, i);
912 }
913 spin_unlock(&vb->lock);
914 }
915 rcu_read_unlock();
916 }
917
918 __purge_vmap_area_lazy(&start, &end, 1, flush);
919}
920EXPORT_SYMBOL_GPL(vm_unmap_aliases);
921
922/**
923 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
924 * @mem: the pointer returned by vm_map_ram
925 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
926 */
927void vm_unmap_ram(const void *mem, unsigned int count)
928{
929 unsigned long size = count << PAGE_SHIFT;
930 unsigned long addr = (unsigned long)mem;
931
932 BUG_ON(!addr);
933 BUG_ON(addr < VMALLOC_START);
934 BUG_ON(addr > VMALLOC_END);
935 BUG_ON(addr & (PAGE_SIZE-1));
936
937 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -0800938 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700939
940 if (likely(count <= VMAP_MAX_ALLOC))
941 vb_free(mem, size);
942 else
943 free_unmap_vmap_area_addr(addr);
944}
945EXPORT_SYMBOL(vm_unmap_ram);
946
947/**
948 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
949 * @pages: an array of pointers to the pages to be mapped
950 * @count: number of pages
951 * @node: prefer to allocate data structures on this node
952 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -0700953 *
954 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -0700955 */
956void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
957{
958 unsigned long size = count << PAGE_SHIFT;
959 unsigned long addr;
960 void *mem;
961
962 if (likely(count <= VMAP_MAX_ALLOC)) {
963 mem = vb_alloc(size, GFP_KERNEL);
964 if (IS_ERR(mem))
965 return NULL;
966 addr = (unsigned long)mem;
967 } else {
968 struct vmap_area *va;
969 va = alloc_vmap_area(size, PAGE_SIZE,
970 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
971 if (IS_ERR(va))
972 return NULL;
973
974 addr = va->va_start;
975 mem = (void *)addr;
976 }
977 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
978 vm_unmap_ram(mem, count);
979 return NULL;
980 }
981 return mem;
982}
983EXPORT_SYMBOL(vm_map_ram);
984
985void __init vmalloc_init(void)
986{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -0800987 struct vmap_area *va;
988 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -0700989 int i;
990
991 for_each_possible_cpu(i) {
992 struct vmap_block_queue *vbq;
993
994 vbq = &per_cpu(vmap_block_queue, i);
995 spin_lock_init(&vbq->lock);
996 INIT_LIST_HEAD(&vbq->free);
997 INIT_LIST_HEAD(&vbq->dirty);
998 vbq->nr_dirty = 0;
999 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001000
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001001 /* Import existing vmlist entries. */
1002 for (tmp = vmlist; tmp; tmp = tmp->next) {
1003 va = alloc_bootmem(sizeof(struct vmap_area));
1004 va->flags = tmp->flags | VM_VM_AREA;
1005 va->va_start = (unsigned long)tmp->addr;
1006 va->va_end = va->va_start + tmp->size;
1007 __insert_vmap_area(va);
1008 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001009 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001010}
1011
1012void unmap_kernel_range(unsigned long addr, unsigned long size)
1013{
1014 unsigned long end = addr + size;
1015 vunmap_page_range(addr, end);
1016 flush_tlb_kernel_range(addr, end);
1017}
1018
1019int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1020{
1021 unsigned long addr = (unsigned long)area->addr;
1022 unsigned long end = addr + area->size - PAGE_SIZE;
1023 int err;
1024
1025 err = vmap_page_range(addr, end, prot, *pages);
1026 if (err > 0) {
1027 *pages += err;
1028 err = 0;
1029 }
1030
1031 return err;
1032}
1033EXPORT_SYMBOL_GPL(map_vm_area);
1034
1035/*** Old vmalloc interfaces ***/
1036DEFINE_RWLOCK(vmlist_lock);
1037struct vm_struct *vmlist;
1038
1039static struct vm_struct *__get_vm_area_node(unsigned long size,
1040 unsigned long flags, unsigned long start, unsigned long end,
1041 int node, gfp_t gfp_mask, void *caller)
1042{
1043 static struct vmap_area *va;
1044 struct vm_struct *area;
1045 struct vm_struct *tmp, **p;
1046 unsigned long align = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001048 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 if (flags & VM_IOREMAP) {
1050 int bit = fls(size);
1051
1052 if (bit > IOREMAP_MAX_ORDER)
1053 bit = IOREMAP_MAX_ORDER;
1054 else if (bit < PAGE_SHIFT)
1055 bit = PAGE_SHIFT;
1056
1057 align = 1ul << bit;
1058 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001061 if (unlikely(!size))
1062 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Christoph Lameter6cb06222007-10-16 01:25:41 -07001064 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if (unlikely(!area))
1066 return NULL;
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 /*
1069 * We always allocate a guard page.
1070 */
1071 size += PAGE_SIZE;
1072
Nick Piggindb64fe02008-10-18 20:27:03 -07001073 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1074 if (IS_ERR(va)) {
1075 kfree(area);
1076 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 area->flags = flags;
Nick Piggindb64fe02008-10-18 20:27:03 -07001080 area->addr = (void *)va->va_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 area->size = size;
1082 area->pages = NULL;
1083 area->nr_pages = 0;
1084 area->phys_addr = 0;
Christoph Lameter23016962008-04-28 02:12:42 -07001085 area->caller = caller;
Nick Piggindb64fe02008-10-18 20:27:03 -07001086 va->private = area;
1087 va->flags |= VM_VM_AREA;
1088
1089 write_lock(&vmlist_lock);
1090 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1091 if (tmp->addr >= area->addr)
1092 break;
1093 }
1094 area->next = *p;
1095 *p = area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 write_unlock(&vmlist_lock);
1097
1098 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099}
1100
Christoph Lameter930fc452005-10-29 18:15:41 -07001101struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1102 unsigned long start, unsigned long end)
1103{
Christoph Lameter23016962008-04-28 02:12:42 -07001104 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1105 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001106}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001107EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109/**
Simon Arlott183ff222007-10-20 01:27:18 +02001110 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 * @size: size of the area
1112 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1113 *
1114 * Search an area of @size in the kernel virtual mapping area,
1115 * and reserved it for out purposes. Returns the area descriptor
1116 * on success or %NULL on failure.
1117 */
1118struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1119{
Christoph Lameter23016962008-04-28 02:12:42 -07001120 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1121 -1, GFP_KERNEL, __builtin_return_address(0));
1122}
1123
1124struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1125 void *caller)
1126{
1127 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1128 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001131struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1132 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07001133{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001134 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
Christoph Lameter23016962008-04-28 02:12:42 -07001135 gfp_mask, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001136}
1137
Nick Piggindb64fe02008-10-18 20:27:03 -07001138static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001139{
Nick Piggindb64fe02008-10-18 20:27:03 -07001140 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001141
Nick Piggindb64fe02008-10-18 20:27:03 -07001142 va = find_vmap_area((unsigned long)addr);
1143 if (va && va->flags & VM_VM_AREA)
1144 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001145
Andi Kleen7856dfe2005-05-20 14:27:57 -07001146 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001147}
1148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149/**
Simon Arlott183ff222007-10-20 01:27:18 +02001150 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 * @addr: base address
1152 *
1153 * Search for the kernel VM area starting at @addr, and remove it.
1154 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001155 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001157struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
Nick Piggindb64fe02008-10-18 20:27:03 -07001159 struct vmap_area *va;
1160
1161 va = find_vmap_area((unsigned long)addr);
1162 if (va && va->flags & VM_VM_AREA) {
1163 struct vm_struct *vm = va->private;
1164 struct vm_struct *tmp, **p;
Nick Piggincd528582009-01-06 14:39:20 -08001165
1166 vmap_debug_free_range(va->va_start, va->va_end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001167 free_unmap_vmap_area(va);
1168 vm->size -= PAGE_SIZE;
1169
1170 write_lock(&vmlist_lock);
1171 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1172 ;
1173 *p = tmp->next;
1174 write_unlock(&vmlist_lock);
1175
1176 return vm;
1177 }
1178 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179}
1180
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001181static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
1183 struct vm_struct *area;
1184
1185 if (!addr)
1186 return;
1187
1188 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001189 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 return;
1191 }
1192
1193 area = remove_vm_area(addr);
1194 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001195 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return;
1198 }
1199
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001200 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001201 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 if (deallocate_pages) {
1204 int i;
1205
1206 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001207 struct page *page = area->pages[i];
1208
1209 BUG_ON(!page);
1210 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 }
1212
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001213 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 vfree(area->pages);
1215 else
1216 kfree(area->pages);
1217 }
1218
1219 kfree(area);
1220 return;
1221}
1222
1223/**
1224 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 * @addr: memory base address
1226 *
Simon Arlott183ff222007-10-20 01:27:18 +02001227 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001228 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1229 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001231 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001233void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 BUG_ON(in_interrupt());
1236 __vunmap(addr, 1);
1237}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238EXPORT_SYMBOL(vfree);
1239
1240/**
1241 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 * @addr: memory base address
1243 *
1244 * Free the virtually contiguous memory area starting at @addr,
1245 * which was created from the page array passed to vmap().
1246 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001247 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001249void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250{
1251 BUG_ON(in_interrupt());
1252 __vunmap(addr, 0);
1253}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254EXPORT_SYMBOL(vunmap);
1255
1256/**
1257 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 * @pages: array of page pointers
1259 * @count: number of pages to map
1260 * @flags: vm_area->flags
1261 * @prot: page protection for the mapping
1262 *
1263 * Maps @count pages from @pages into contiguous kernel virtual
1264 * space.
1265 */
1266void *vmap(struct page **pages, unsigned int count,
1267 unsigned long flags, pgprot_t prot)
1268{
1269 struct vm_struct *area;
1270
1271 if (count > num_physpages)
1272 return NULL;
1273
Christoph Lameter23016962008-04-28 02:12:42 -07001274 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1275 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (!area)
1277 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (map_vm_area(area, prot, &pages)) {
1280 vunmap(area->addr);
1281 return NULL;
1282 }
1283
1284 return area->addr;
1285}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286EXPORT_SYMBOL(vmap);
1287
Nick Piggindb64fe02008-10-18 20:27:03 -07001288static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1289 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001290static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001291 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292{
1293 struct page **pages;
1294 unsigned int nr_pages, array_size, i;
1295
1296 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1297 array_size = (nr_pages * sizeof(struct page *));
1298
1299 area->nr_pages = nr_pages;
1300 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001301 if (array_size > PAGE_SIZE) {
Christoph Lameter94f60302007-07-17 04:03:29 -07001302 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
Christoph Lameter23016962008-04-28 02:12:42 -07001303 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001304 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001305 } else {
1306 pages = kmalloc_node(array_size,
Christoph Lameter6cb06222007-10-16 01:25:41 -07001307 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
Andrew Morton286e1ea2006-10-17 00:09:57 -07001308 node);
1309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001311 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if (!area->pages) {
1313 remove_vm_area(area->addr);
1314 kfree(area);
1315 return NULL;
1316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001319 struct page *page;
1320
Christoph Lameter930fc452005-10-29 18:15:41 -07001321 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001322 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001323 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001324 page = alloc_pages_node(node, gfp_mask, 0);
1325
1326 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 /* Successfully allocated i pages, free them in __vunmap() */
1328 area->nr_pages = i;
1329 goto fail;
1330 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001331 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
1333
1334 if (map_vm_area(area, prot, &pages))
1335 goto fail;
1336 return area->addr;
1337
1338fail:
1339 vfree(area->addr);
1340 return NULL;
1341}
1342
Christoph Lameter930fc452005-10-29 18:15:41 -07001343void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1344{
Christoph Lameter23016962008-04-28 02:12:42 -07001345 return __vmalloc_area_node(area, gfp_mask, prot, -1,
1346 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001347}
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001350 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 * @size: allocation size
1352 * @gfp_mask: flags for the page level allocator
1353 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001354 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001355 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 *
1357 * Allocate enough pages to cover @size from the page level
1358 * allocator with @gfp_mask flags. Map them into contiguous
1359 * kernel virtual space, using a pagetable protection of @prot.
1360 */
Adrian Bunkb2213852006-09-25 23:31:02 -07001361static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
Christoph Lameter23016962008-04-28 02:12:42 -07001362 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363{
1364 struct vm_struct *area;
1365
1366 size = PAGE_ALIGN(size);
1367 if (!size || (size >> PAGE_SHIFT) > num_physpages)
1368 return NULL;
1369
Christoph Lameter23016962008-04-28 02:12:42 -07001370 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1371 node, gfp_mask, caller);
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (!area)
1374 return NULL;
1375
Christoph Lameter23016962008-04-28 02:12:42 -07001376 return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377}
1378
Christoph Lameter930fc452005-10-29 18:15:41 -07001379void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1380{
Christoph Lameter23016962008-04-28 02:12:42 -07001381 return __vmalloc_node(size, gfp_mask, prot, -1,
1382 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001383}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384EXPORT_SYMBOL(__vmalloc);
1385
1386/**
1387 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 * Allocate enough pages to cover @size from the page level
1390 * allocator and map them into contiguous kernel virtual space.
1391 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001392 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 * use __vmalloc() instead.
1394 */
1395void *vmalloc(unsigned long size)
1396{
Christoph Lameter23016962008-04-28 02:12:42 -07001397 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1398 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400EXPORT_SYMBOL(vmalloc);
1401
Christoph Lameter930fc452005-10-29 18:15:41 -07001402/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001403 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1404 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001405 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001406 * The resulting memory area is zeroed so it can be mapped to userspace
1407 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001408 */
1409void *vmalloc_user(unsigned long size)
1410{
1411 struct vm_struct *area;
1412 void *ret;
1413
Glauber Costa84877842009-01-06 14:39:19 -08001414 ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1415 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001416 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001417 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001418 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001419 }
Nick Piggin83342312006-06-23 02:03:20 -07001420 return ret;
1421}
1422EXPORT_SYMBOL(vmalloc_user);
1423
1424/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001425 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001426 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001427 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001428 *
1429 * Allocate enough pages to cover @size from the page level
1430 * allocator and map them into contiguous kernel virtual space.
1431 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001432 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001433 * use __vmalloc() instead.
1434 */
1435void *vmalloc_node(unsigned long size, int node)
1436{
Christoph Lameter23016962008-04-28 02:12:42 -07001437 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1438 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001439}
1440EXPORT_SYMBOL(vmalloc_node);
1441
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001442#ifndef PAGE_KERNEL_EXEC
1443# define PAGE_KERNEL_EXEC PAGE_KERNEL
1444#endif
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446/**
1447 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * @size: allocation size
1449 *
1450 * Kernel-internal function to allocate enough pages to cover @size
1451 * the page level allocator and map them into contiguous and
1452 * executable kernel virtual space.
1453 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001454 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 * use __vmalloc() instead.
1456 */
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458void *vmalloc_exec(unsigned long size)
1459{
Glauber Costa84877842009-01-06 14:39:19 -08001460 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1461 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001464#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001465#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001466#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001467#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001468#else
1469#define GFP_VMALLOC32 GFP_KERNEL
1470#endif
1471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472/**
1473 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 * @size: allocation size
1475 *
1476 * Allocate enough 32bit PA addressable pages to cover @size from the
1477 * page level allocator and map them into contiguous kernel virtual space.
1478 */
1479void *vmalloc_32(unsigned long size)
1480{
Glauber Costa84877842009-01-06 14:39:19 -08001481 return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1482 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484EXPORT_SYMBOL(vmalloc_32);
1485
Nick Piggin83342312006-06-23 02:03:20 -07001486/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001487 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001488 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001489 *
1490 * The resulting memory area is 32bit addressable and zeroed so it can be
1491 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001492 */
1493void *vmalloc_32_user(unsigned long size)
1494{
1495 struct vm_struct *area;
1496 void *ret;
1497
Glauber Costa84877842009-01-06 14:39:19 -08001498 ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1499 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001500 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001501 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001502 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001503 }
Nick Piggin83342312006-06-23 02:03:20 -07001504 return ret;
1505}
1506EXPORT_SYMBOL(vmalloc_32_user);
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508long vread(char *buf, char *addr, unsigned long count)
1509{
1510 struct vm_struct *tmp;
1511 char *vaddr, *buf_start = buf;
1512 unsigned long n;
1513
1514 /* Don't allow overflow */
1515 if ((unsigned long) addr + count < count)
1516 count = -(unsigned long) addr;
1517
1518 read_lock(&vmlist_lock);
1519 for (tmp = vmlist; tmp; tmp = tmp->next) {
1520 vaddr = (char *) tmp->addr;
1521 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1522 continue;
1523 while (addr < vaddr) {
1524 if (count == 0)
1525 goto finished;
1526 *buf = '\0';
1527 buf++;
1528 addr++;
1529 count--;
1530 }
1531 n = vaddr + tmp->size - PAGE_SIZE - addr;
1532 do {
1533 if (count == 0)
1534 goto finished;
1535 *buf = *addr;
1536 buf++;
1537 addr++;
1538 count--;
1539 } while (--n > 0);
1540 }
1541finished:
1542 read_unlock(&vmlist_lock);
1543 return buf - buf_start;
1544}
1545
1546long vwrite(char *buf, char *addr, unsigned long count)
1547{
1548 struct vm_struct *tmp;
1549 char *vaddr, *buf_start = buf;
1550 unsigned long n;
1551
1552 /* Don't allow overflow */
1553 if ((unsigned long) addr + count < count)
1554 count = -(unsigned long) addr;
1555
1556 read_lock(&vmlist_lock);
1557 for (tmp = vmlist; tmp; tmp = tmp->next) {
1558 vaddr = (char *) tmp->addr;
1559 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1560 continue;
1561 while (addr < vaddr) {
1562 if (count == 0)
1563 goto finished;
1564 buf++;
1565 addr++;
1566 count--;
1567 }
1568 n = vaddr + tmp->size - PAGE_SIZE - addr;
1569 do {
1570 if (count == 0)
1571 goto finished;
1572 *addr = *buf;
1573 buf++;
1574 addr++;
1575 count--;
1576 } while (--n > 0);
1577 }
1578finished:
1579 read_unlock(&vmlist_lock);
1580 return buf - buf_start;
1581}
Nick Piggin83342312006-06-23 02:03:20 -07001582
1583/**
1584 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07001585 * @vma: vma to cover (map full range of vma)
1586 * @addr: vmalloc memory
1587 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07001588 *
1589 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07001590 *
1591 * This function checks that addr is a valid vmalloc'ed area, and
1592 * that it is big enough to cover the vma. Will return failure if
1593 * that criteria isn't met.
1594 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001595 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07001596 */
1597int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1598 unsigned long pgoff)
1599{
1600 struct vm_struct *area;
1601 unsigned long uaddr = vma->vm_start;
1602 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07001603
1604 if ((PAGE_SIZE-1) & (unsigned long)addr)
1605 return -EINVAL;
1606
Nick Piggindb64fe02008-10-18 20:27:03 -07001607 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07001608 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07001609 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001610
1611 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07001612 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001613
1614 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07001615 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001616
1617 addr += pgoff << PAGE_SHIFT;
1618 do {
1619 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07001620 int ret;
1621
Nick Piggin83342312006-06-23 02:03:20 -07001622 ret = vm_insert_page(vma, uaddr, page);
1623 if (ret)
1624 return ret;
1625
1626 uaddr += PAGE_SIZE;
1627 addr += PAGE_SIZE;
1628 usize -= PAGE_SIZE;
1629 } while (usize > 0);
1630
1631 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1632 vma->vm_flags |= VM_RESERVED;
1633
Nick Piggindb64fe02008-10-18 20:27:03 -07001634 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07001635}
1636EXPORT_SYMBOL(remap_vmalloc_range);
1637
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07001638/*
1639 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1640 * have one.
1641 */
1642void __attribute__((weak)) vmalloc_sync_all(void)
1643{
1644}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001645
1646
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001647static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001648{
1649 /* apply_to_page_range() does all the hard work. */
1650 return 0;
1651}
1652
1653/**
1654 * alloc_vm_area - allocate a range of kernel address space
1655 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07001656 *
1657 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001658 *
1659 * This function reserves a range of kernel address space, and
1660 * allocates pagetables to map that range. No actual mappings
1661 * are created. If the kernel address space is not shared
1662 * between processes, it syncs the pagetable across all
1663 * processes.
1664 */
1665struct vm_struct *alloc_vm_area(size_t size)
1666{
1667 struct vm_struct *area;
1668
Christoph Lameter23016962008-04-28 02:12:42 -07001669 area = get_vm_area_caller(size, VM_IOREMAP,
1670 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001671 if (area == NULL)
1672 return NULL;
1673
1674 /*
1675 * This ensures that page tables are constructed for this region
1676 * of kernel virtual address space and mapped into init_mm.
1677 */
1678 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1679 area->size, f, NULL)) {
1680 free_vm_area(area);
1681 return NULL;
1682 }
1683
1684 /* Make sure the pagetables are constructed in process kernel
1685 mappings */
1686 vmalloc_sync_all();
1687
1688 return area;
1689}
1690EXPORT_SYMBOL_GPL(alloc_vm_area);
1691
1692void free_vm_area(struct vm_struct *area)
1693{
1694 struct vm_struct *ret;
1695 ret = remove_vm_area(area->addr);
1696 BUG_ON(ret != area);
1697 kfree(area);
1698}
1699EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07001700
1701
1702#ifdef CONFIG_PROC_FS
1703static void *s_start(struct seq_file *m, loff_t *pos)
1704{
1705 loff_t n = *pos;
1706 struct vm_struct *v;
1707
1708 read_lock(&vmlist_lock);
1709 v = vmlist;
1710 while (n > 0 && v) {
1711 n--;
1712 v = v->next;
1713 }
1714 if (!n)
1715 return v;
1716
1717 return NULL;
1718
1719}
1720
1721static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1722{
1723 struct vm_struct *v = p;
1724
1725 ++*pos;
1726 return v->next;
1727}
1728
1729static void s_stop(struct seq_file *m, void *p)
1730{
1731 read_unlock(&vmlist_lock);
1732}
1733
Eric Dumazeta47a1262008-07-23 21:27:38 -07001734static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1735{
1736 if (NUMA_BUILD) {
1737 unsigned int nr, *counters = m->private;
1738
1739 if (!counters)
1740 return;
1741
1742 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1743
1744 for (nr = 0; nr < v->nr_pages; nr++)
1745 counters[page_to_nid(v->pages[nr])]++;
1746
1747 for_each_node_state(nr, N_HIGH_MEMORY)
1748 if (counters[nr])
1749 seq_printf(m, " N%u=%u", nr, counters[nr]);
1750 }
1751}
1752
Christoph Lametera10aa572008-04-28 02:12:40 -07001753static int s_show(struct seq_file *m, void *p)
1754{
1755 struct vm_struct *v = p;
1756
1757 seq_printf(m, "0x%p-0x%p %7ld",
1758 v->addr, v->addr + v->size, v->size);
1759
Christoph Lameter23016962008-04-28 02:12:42 -07001760 if (v->caller) {
Hugh Dickins9c246242008-12-09 13:14:27 -08001761 char buff[KSYM_SYMBOL_LEN];
Christoph Lameter23016962008-04-28 02:12:42 -07001762
1763 seq_putc(m, ' ');
1764 sprint_symbol(buff, (unsigned long)v->caller);
1765 seq_puts(m, buff);
1766 }
1767
Christoph Lametera10aa572008-04-28 02:12:40 -07001768 if (v->nr_pages)
1769 seq_printf(m, " pages=%d", v->nr_pages);
1770
1771 if (v->phys_addr)
1772 seq_printf(m, " phys=%lx", v->phys_addr);
1773
1774 if (v->flags & VM_IOREMAP)
1775 seq_printf(m, " ioremap");
1776
1777 if (v->flags & VM_ALLOC)
1778 seq_printf(m, " vmalloc");
1779
1780 if (v->flags & VM_MAP)
1781 seq_printf(m, " vmap");
1782
1783 if (v->flags & VM_USERMAP)
1784 seq_printf(m, " user");
1785
1786 if (v->flags & VM_VPAGES)
1787 seq_printf(m, " vpages");
1788
Eric Dumazeta47a1262008-07-23 21:27:38 -07001789 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07001790 seq_putc(m, '\n');
1791 return 0;
1792}
1793
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001794static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07001795 .start = s_start,
1796 .next = s_next,
1797 .stop = s_stop,
1798 .show = s_show,
1799};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001800
1801static int vmalloc_open(struct inode *inode, struct file *file)
1802{
1803 unsigned int *ptr = NULL;
1804 int ret;
1805
1806 if (NUMA_BUILD)
1807 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1808 ret = seq_open(file, &vmalloc_op);
1809 if (!ret) {
1810 struct seq_file *m = file->private_data;
1811 m->private = ptr;
1812 } else
1813 kfree(ptr);
1814 return ret;
1815}
1816
1817static const struct file_operations proc_vmalloc_operations = {
1818 .open = vmalloc_open,
1819 .read = seq_read,
1820 .llseek = seq_lseek,
1821 .release = seq_release_private,
1822};
1823
1824static int __init proc_vmalloc_init(void)
1825{
1826 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1827 return 0;
1828}
1829module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07001830#endif
1831