blob: f8189a4b3e135e4c4158bb80082a49434fcb54af [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>
Tejun Heof0aa6612009-02-20 16:29:08 +090026#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010027#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Nick Piggindb64fe02008-10-18 20:27:03 -070029#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/uaccess.h>
31#include <asm/tlbflush.h>
32
33
Nick Piggindb64fe02008-10-18 20:27:03 -070034/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070035
Linus Torvalds1da177e2005-04-16 15:20:36 -070036static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
37{
38 pte_t *pte;
39
40 pte = pte_offset_kernel(pmd, addr);
41 do {
42 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
43 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
44 } while (pte++, addr += PAGE_SIZE, addr != end);
45}
46
Nick Piggindb64fe02008-10-18 20:27:03 -070047static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048{
49 pmd_t *pmd;
50 unsigned long next;
51
52 pmd = pmd_offset(pud, addr);
53 do {
54 next = pmd_addr_end(addr, end);
55 if (pmd_none_or_clear_bad(pmd))
56 continue;
57 vunmap_pte_range(pmd, addr, next);
58 } while (pmd++, addr = next, addr != end);
59}
60
Nick Piggindb64fe02008-10-18 20:27:03 -070061static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062{
63 pud_t *pud;
64 unsigned long next;
65
66 pud = pud_offset(pgd, addr);
67 do {
68 next = pud_addr_end(addr, end);
69 if (pud_none_or_clear_bad(pud))
70 continue;
71 vunmap_pmd_range(pud, addr, next);
72 } while (pud++, addr = next, addr != end);
73}
74
Nick Piggindb64fe02008-10-18 20:27:03 -070075static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
77 pgd_t *pgd;
78 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80 BUG_ON(addr >= end);
81 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 do {
83 next = pgd_addr_end(addr, end);
84 if (pgd_none_or_clear_bad(pgd))
85 continue;
86 vunmap_pud_range(pgd, addr, next);
87 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088}
89
90static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -070091 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 pte_t *pte;
94
Nick Piggindb64fe02008-10-18 20:27:03 -070095 /*
96 * nr is a running index into the array which helps higher level
97 * callers keep track of where we're up to.
98 */
99
Hugh Dickins872fec12005-10-29 18:16:21 -0700100 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 if (!pte)
102 return -ENOMEM;
103 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700104 struct page *page = pages[*nr];
105
106 if (WARN_ON(!pte_none(*pte)))
107 return -EBUSY;
108 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 return -ENOMEM;
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700111 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 } while (pte++, addr += PAGE_SIZE, addr != end);
113 return 0;
114}
115
Nick Piggindb64fe02008-10-18 20:27:03 -0700116static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 pmd_t *pmd;
120 unsigned long next;
121
122 pmd = pmd_alloc(&init_mm, pud, addr);
123 if (!pmd)
124 return -ENOMEM;
125 do {
126 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700127 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return -ENOMEM;
129 } while (pmd++, addr = next, addr != end);
130 return 0;
131}
132
Nick Piggindb64fe02008-10-18 20:27:03 -0700133static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
136 pud_t *pud;
137 unsigned long next;
138
139 pud = pud_alloc(&init_mm, pgd, addr);
140 if (!pud)
141 return -ENOMEM;
142 do {
143 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700144 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 return -ENOMEM;
146 } while (pud++, addr = next, addr != end);
147 return 0;
148}
149
Nick Piggindb64fe02008-10-18 20:27:03 -0700150/*
151 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152 * will have pfns corresponding to the "pages" array.
153 *
154 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
155 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900156static int vmap_page_range_noflush(unsigned long start, unsigned long end,
157 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
159 pgd_t *pgd;
160 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800161 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700162 int err = 0;
163 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 BUG_ON(addr >= end);
166 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 do {
168 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700169 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 if (err)
171 break;
172 } while (pgd++, addr = next, addr != 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
Tejun Heo8fc48982009-02-20 16:29:08 +0900179static int vmap_page_range(unsigned long start, unsigned long end,
180 pgprot_t prot, struct page **pages)
181{
182 int ret;
183
184 ret = vmap_page_range_noflush(start, end, prot, pages);
185 flush_cache_vmap(start, end);
186 return ret;
187}
188
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700189static inline int is_vmalloc_or_module_addr(const void *x)
190{
191 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000192 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700193 * and fall back on vmalloc() if that fails. Others
194 * just put it in the vmalloc space.
195 */
196#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
197 unsigned long addr = (unsigned long)x;
198 if (addr >= MODULES_VADDR && addr < MODULES_END)
199 return 1;
200#endif
201 return is_vmalloc_addr(x);
202}
203
Christoph Lameter48667e72008-02-04 22:28:31 -0800204/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700205 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800206 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800207struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800208{
209 unsigned long addr = (unsigned long) vmalloc_addr;
210 struct page *page = NULL;
211 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800212
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200213 /*
214 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
215 * architectures that do not vmalloc module space
216 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700217 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200218
Christoph Lameter48667e72008-02-04 22:28:31 -0800219 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700220 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800221 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700222 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800223 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700224 pte_t *ptep, pte;
225
Christoph Lameter48667e72008-02-04 22:28:31 -0800226 ptep = pte_offset_map(pmd, addr);
227 pte = *ptep;
228 if (pte_present(pte))
229 page = pte_page(pte);
230 pte_unmap(ptep);
231 }
232 }
233 }
234 return page;
235}
236EXPORT_SYMBOL(vmalloc_to_page);
237
238/*
239 * Map a vmalloc()-space virtual address to the physical page frame number.
240 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800241unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800242{
243 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
244}
245EXPORT_SYMBOL(vmalloc_to_pfn);
246
Nick Piggindb64fe02008-10-18 20:27:03 -0700247
248/*** Global kva allocator ***/
249
250#define VM_LAZY_FREE 0x01
251#define VM_LAZY_FREEING 0x02
252#define VM_VM_AREA 0x04
253
254struct vmap_area {
255 unsigned long va_start;
256 unsigned long va_end;
257 unsigned long flags;
258 struct rb_node rb_node; /* address sorted rbtree */
259 struct list_head list; /* address sorted list */
260 struct list_head purge_list; /* "lazy purge" list */
261 void *private;
262 struct rcu_head rcu_head;
263};
264
265static DEFINE_SPINLOCK(vmap_area_lock);
266static struct rb_root vmap_area_root = RB_ROOT;
267static LIST_HEAD(vmap_area_list);
268
269static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Nick Piggindb64fe02008-10-18 20:27:03 -0700271 struct rb_node *n = vmap_area_root.rb_node;
272
273 while (n) {
274 struct vmap_area *va;
275
276 va = rb_entry(n, struct vmap_area, rb_node);
277 if (addr < va->va_start)
278 n = n->rb_left;
279 else if (addr > va->va_start)
280 n = n->rb_right;
281 else
282 return va;
283 }
284
285 return NULL;
286}
287
288static void __insert_vmap_area(struct vmap_area *va)
289{
290 struct rb_node **p = &vmap_area_root.rb_node;
291 struct rb_node *parent = NULL;
292 struct rb_node *tmp;
293
294 while (*p) {
295 struct vmap_area *tmp;
296
297 parent = *p;
298 tmp = rb_entry(parent, struct vmap_area, rb_node);
299 if (va->va_start < tmp->va_end)
300 p = &(*p)->rb_left;
301 else if (va->va_end > tmp->va_start)
302 p = &(*p)->rb_right;
303 else
304 BUG();
305 }
306
307 rb_link_node(&va->rb_node, parent, p);
308 rb_insert_color(&va->rb_node, &vmap_area_root);
309
310 /* address-sort this list so it is usable like the vmlist */
311 tmp = rb_prev(&va->rb_node);
312 if (tmp) {
313 struct vmap_area *prev;
314 prev = rb_entry(tmp, struct vmap_area, rb_node);
315 list_add_rcu(&va->list, &prev->list);
316 } else
317 list_add_rcu(&va->list, &vmap_area_list);
318}
319
320static void purge_vmap_area_lazy(void);
321
322/*
323 * Allocate a region of KVA of the specified size and alignment, within the
324 * vstart and vend.
325 */
326static struct vmap_area *alloc_vmap_area(unsigned long size,
327 unsigned long align,
328 unsigned long vstart, unsigned long vend,
329 int node, gfp_t gfp_mask)
330{
331 struct vmap_area *va;
332 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700334 int purged = 0;
335
Nick Piggin77669702009-02-27 14:03:03 -0800336 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700337 BUG_ON(size & ~PAGE_MASK);
338
Nick Piggindb64fe02008-10-18 20:27:03 -0700339 va = kmalloc_node(sizeof(struct vmap_area),
340 gfp_mask & GFP_RECLAIM_MASK, node);
341 if (unlikely(!va))
342 return ERR_PTR(-ENOMEM);
343
344retry:
Glauber Costa0ae15132008-11-19 15:36:33 -0800345 addr = ALIGN(vstart, align);
346
Nick Piggindb64fe02008-10-18 20:27:03 -0700347 spin_lock(&vmap_area_lock);
Nick Piggin77669702009-02-27 14:03:03 -0800348 if (addr + size - 1 < addr)
349 goto overflow;
350
Nick Piggindb64fe02008-10-18 20:27:03 -0700351 /* XXX: could have a last_hole cache */
352 n = vmap_area_root.rb_node;
353 if (n) {
354 struct vmap_area *first = NULL;
355
356 do {
357 struct vmap_area *tmp;
358 tmp = rb_entry(n, struct vmap_area, rb_node);
359 if (tmp->va_end >= addr) {
360 if (!first && tmp->va_start < addr + size)
361 first = tmp;
362 n = n->rb_left;
363 } else {
364 first = tmp;
365 n = n->rb_right;
366 }
367 } while (n);
368
369 if (!first)
370 goto found;
371
372 if (first->va_end < addr) {
373 n = rb_next(&first->rb_node);
374 if (n)
375 first = rb_entry(n, struct vmap_area, rb_node);
376 else
377 goto found;
378 }
379
Nick Pigginf011c2d2008-11-19 15:36:32 -0800380 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700381 addr = ALIGN(first->va_end + PAGE_SIZE, align);
Nick Piggin77669702009-02-27 14:03:03 -0800382 if (addr + size - 1 < addr)
383 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700384
385 n = rb_next(&first->rb_node);
386 if (n)
387 first = rb_entry(n, struct vmap_area, rb_node);
388 else
389 goto found;
390 }
391 }
392found:
393 if (addr + size > vend) {
Nick Piggin77669702009-02-27 14:03:03 -0800394overflow:
Nick Piggindb64fe02008-10-18 20:27:03 -0700395 spin_unlock(&vmap_area_lock);
396 if (!purged) {
397 purge_vmap_area_lazy();
398 purged = 1;
399 goto retry;
400 }
401 if (printk_ratelimit())
Glauber Costac1279c42009-01-06 14:39:18 -0800402 printk(KERN_WARNING
403 "vmap allocation for size %lu failed: "
404 "use vmalloc=<size> to increase size.\n", size);
Ralph Wuerthner2498ce42009-05-06 16:02:59 -0700405 kfree(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700406 return ERR_PTR(-EBUSY);
407 }
408
409 BUG_ON(addr & (align-1));
410
411 va->va_start = addr;
412 va->va_end = addr + size;
413 va->flags = 0;
414 __insert_vmap_area(va);
415 spin_unlock(&vmap_area_lock);
416
417 return va;
418}
419
420static void rcu_free_va(struct rcu_head *head)
421{
422 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
423
424 kfree(va);
425}
426
427static void __free_vmap_area(struct vmap_area *va)
428{
429 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
430 rb_erase(&va->rb_node, &vmap_area_root);
431 RB_CLEAR_NODE(&va->rb_node);
432 list_del_rcu(&va->list);
433
434 call_rcu(&va->rcu_head, rcu_free_va);
435}
436
437/*
438 * Free a region of KVA allocated by alloc_vmap_area
439 */
440static void free_vmap_area(struct vmap_area *va)
441{
442 spin_lock(&vmap_area_lock);
443 __free_vmap_area(va);
444 spin_unlock(&vmap_area_lock);
445}
446
447/*
448 * Clear the pagetable entries of a given vmap_area
449 */
450static void unmap_vmap_area(struct vmap_area *va)
451{
452 vunmap_page_range(va->va_start, va->va_end);
453}
454
Nick Piggincd528582009-01-06 14:39:20 -0800455static void vmap_debug_free_range(unsigned long start, unsigned long end)
456{
457 /*
458 * Unmap page tables and force a TLB flush immediately if
459 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
460 * bugs similarly to those in linear kernel virtual address
461 * space after a page has been freed.
462 *
463 * All the lazy freeing logic is still retained, in order to
464 * minimise intrusiveness of this debugging feature.
465 *
466 * This is going to be *slow* (linear kernel virtual address
467 * debugging doesn't do a broadcast TLB flush so it is a lot
468 * faster).
469 */
470#ifdef CONFIG_DEBUG_PAGEALLOC
471 vunmap_page_range(start, end);
472 flush_tlb_kernel_range(start, end);
473#endif
474}
475
Nick Piggindb64fe02008-10-18 20:27:03 -0700476/*
477 * lazy_max_pages is the maximum amount of virtual address space we gather up
478 * before attempting to purge with a TLB flush.
479 *
480 * There is a tradeoff here: a larger number will cover more kernel page tables
481 * and take slightly longer to purge, but it will linearly reduce the number of
482 * global TLB flushes that must be performed. It would seem natural to scale
483 * this number up linearly with the number of CPUs (because vmapping activity
484 * could also scale linearly with the number of CPUs), however it is likely
485 * that in practice, workloads might be constrained in other ways that mean
486 * vmap activity will not scale linearly with CPUs. Also, I want to be
487 * conservative and not introduce a big latency on huge systems, so go with
488 * a less aggressive log scale. It will still be an improvement over the old
489 * code, and it will be simple to change the scale factor if we find that it
490 * becomes a problem on bigger systems.
491 */
492static unsigned long lazy_max_pages(void)
493{
494 unsigned int log;
495
496 log = fls(num_online_cpus());
497
498 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
499}
500
501static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
502
503/*
504 * Purges all lazily-freed vmap areas.
505 *
506 * If sync is 0 then don't purge if there is already a purge in progress.
507 * If force_flush is 1, then flush kernel TLBs between *start and *end even
508 * if we found no lazy vmap areas to unmap (callers can use this to optimise
509 * their own TLB flushing).
510 * Returns with *start = min(*start, lowest purged address)
511 * *end = max(*end, highest purged address)
512 */
513static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
514 int sync, int force_flush)
515{
Andrew Morton46666d82009-01-15 13:51:15 -0800516 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700517 LIST_HEAD(valist);
518 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800519 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700520 int nr = 0;
521
522 /*
523 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
524 * should not expect such behaviour. This just simplifies locking for
525 * the case that isn't actually used at the moment anyway.
526 */
527 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800528 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700529 return;
530 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800531 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700532
533 rcu_read_lock();
534 list_for_each_entry_rcu(va, &vmap_area_list, list) {
535 if (va->flags & VM_LAZY_FREE) {
536 if (va->va_start < *start)
537 *start = va->va_start;
538 if (va->va_end > *end)
539 *end = va->va_end;
540 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
541 unmap_vmap_area(va);
542 list_add_tail(&va->purge_list, &valist);
543 va->flags |= VM_LAZY_FREEING;
544 va->flags &= ~VM_LAZY_FREE;
545 }
546 }
547 rcu_read_unlock();
548
549 if (nr) {
550 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
551 atomic_sub(nr, &vmap_lazy_nr);
552 }
553
554 if (nr || force_flush)
555 flush_tlb_kernel_range(*start, *end);
556
557 if (nr) {
558 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800559 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700560 __free_vmap_area(va);
561 spin_unlock(&vmap_area_lock);
562 }
Andrew Morton46666d82009-01-15 13:51:15 -0800563 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700564}
565
566/*
Nick Piggin496850e2008-11-19 15:36:33 -0800567 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
568 * is already purging.
569 */
570static void try_purge_vmap_area_lazy(void)
571{
572 unsigned long start = ULONG_MAX, end = 0;
573
574 __purge_vmap_area_lazy(&start, &end, 0, 0);
575}
576
577/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700578 * Kick off a purge of the outstanding lazy areas.
579 */
580static void purge_vmap_area_lazy(void)
581{
582 unsigned long start = ULONG_MAX, end = 0;
583
Nick Piggin496850e2008-11-19 15:36:33 -0800584 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700585}
586
587/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800588 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
589 * called for the correct range previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700590 */
Nick Pigginb29acbd2008-12-01 13:13:47 -0800591static void free_unmap_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700592{
593 va->flags |= VM_LAZY_FREE;
594 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
595 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800596 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700597}
598
Nick Pigginb29acbd2008-12-01 13:13:47 -0800599/*
600 * Free and unmap a vmap area
601 */
602static void free_unmap_vmap_area(struct vmap_area *va)
603{
604 flush_cache_vunmap(va->va_start, va->va_end);
605 free_unmap_vmap_area_noflush(va);
606}
607
Nick Piggindb64fe02008-10-18 20:27:03 -0700608static struct vmap_area *find_vmap_area(unsigned long addr)
609{
610 struct vmap_area *va;
611
612 spin_lock(&vmap_area_lock);
613 va = __find_vmap_area(addr);
614 spin_unlock(&vmap_area_lock);
615
616 return va;
617}
618
619static void free_unmap_vmap_area_addr(unsigned long addr)
620{
621 struct vmap_area *va;
622
623 va = find_vmap_area(addr);
624 BUG_ON(!va);
625 free_unmap_vmap_area(va);
626}
627
628
629/*** Per cpu kva allocator ***/
630
631/*
632 * vmap space is limited especially on 32 bit architectures. Ensure there is
633 * room for at least 16 percpu vmap blocks per CPU.
634 */
635/*
636 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
637 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
638 * instead (we just need a rough idea)
639 */
640#if BITS_PER_LONG == 32
641#define VMALLOC_SPACE (128UL*1024*1024)
642#else
643#define VMALLOC_SPACE (128UL*1024*1024*1024)
644#endif
645
646#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
647#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
648#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
649#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
650#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
651#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
652#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
653 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
654 VMALLOC_PAGES / NR_CPUS / 16))
655
656#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
657
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100658static bool vmap_initialized __read_mostly = false;
659
Nick Piggindb64fe02008-10-18 20:27:03 -0700660struct vmap_block_queue {
661 spinlock_t lock;
662 struct list_head free;
663 struct list_head dirty;
664 unsigned int nr_dirty;
665};
666
667struct vmap_block {
668 spinlock_t lock;
669 struct vmap_area *va;
670 struct vmap_block_queue *vbq;
671 unsigned long free, dirty;
672 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
673 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
674 union {
MinChan Kimd0868172009-03-31 15:19:26 -0700675 struct list_head free_list;
Nick Piggindb64fe02008-10-18 20:27:03 -0700676 struct rcu_head rcu_head;
677 };
678};
679
680/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
681static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
682
683/*
684 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
685 * in the free path. Could get rid of this if we change the API to return a
686 * "cookie" from alloc, to be passed to free. But no big deal yet.
687 */
688static DEFINE_SPINLOCK(vmap_block_tree_lock);
689static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
690
691/*
692 * We should probably have a fallback mechanism to allocate virtual memory
693 * out of partially filled vmap blocks. However vmap block sizing should be
694 * fairly reasonable according to the vmalloc size, so it shouldn't be a
695 * big problem.
696 */
697
698static unsigned long addr_to_vb_idx(unsigned long addr)
699{
700 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
701 addr /= VMAP_BLOCK_SIZE;
702 return addr;
703}
704
705static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
706{
707 struct vmap_block_queue *vbq;
708 struct vmap_block *vb;
709 struct vmap_area *va;
710 unsigned long vb_idx;
711 int node, err;
712
713 node = numa_node_id();
714
715 vb = kmalloc_node(sizeof(struct vmap_block),
716 gfp_mask & GFP_RECLAIM_MASK, node);
717 if (unlikely(!vb))
718 return ERR_PTR(-ENOMEM);
719
720 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
721 VMALLOC_START, VMALLOC_END,
722 node, gfp_mask);
723 if (unlikely(IS_ERR(va))) {
724 kfree(vb);
725 return ERR_PTR(PTR_ERR(va));
726 }
727
728 err = radix_tree_preload(gfp_mask);
729 if (unlikely(err)) {
730 kfree(vb);
731 free_vmap_area(va);
732 return ERR_PTR(err);
733 }
734
735 spin_lock_init(&vb->lock);
736 vb->va = va;
737 vb->free = VMAP_BBMAP_BITS;
738 vb->dirty = 0;
739 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
740 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
741 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700742
743 vb_idx = addr_to_vb_idx(va->va_start);
744 spin_lock(&vmap_block_tree_lock);
745 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
746 spin_unlock(&vmap_block_tree_lock);
747 BUG_ON(err);
748 radix_tree_preload_end();
749
750 vbq = &get_cpu_var(vmap_block_queue);
751 vb->vbq = vbq;
752 spin_lock(&vbq->lock);
753 list_add(&vb->free_list, &vbq->free);
754 spin_unlock(&vbq->lock);
755 put_cpu_var(vmap_cpu_blocks);
756
757 return vb;
758}
759
760static void rcu_free_vb(struct rcu_head *head)
761{
762 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
763
764 kfree(vb);
765}
766
767static void free_vmap_block(struct vmap_block *vb)
768{
769 struct vmap_block *tmp;
770 unsigned long vb_idx;
771
MinChan Kimd0868172009-03-31 15:19:26 -0700772 BUG_ON(!list_empty(&vb->free_list));
Nick Piggindb64fe02008-10-18 20:27:03 -0700773
774 vb_idx = addr_to_vb_idx(vb->va->va_start);
775 spin_lock(&vmap_block_tree_lock);
776 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
777 spin_unlock(&vmap_block_tree_lock);
778 BUG_ON(tmp != vb);
779
Nick Pigginb29acbd2008-12-01 13:13:47 -0800780 free_unmap_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700781 call_rcu(&vb->rcu_head, rcu_free_vb);
782}
783
784static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
785{
786 struct vmap_block_queue *vbq;
787 struct vmap_block *vb;
788 unsigned long addr = 0;
789 unsigned int order;
790
791 BUG_ON(size & ~PAGE_MASK);
792 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
793 order = get_order(size);
794
795again:
796 rcu_read_lock();
797 vbq = &get_cpu_var(vmap_block_queue);
798 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
799 int i;
800
801 spin_lock(&vb->lock);
802 i = bitmap_find_free_region(vb->alloc_map,
803 VMAP_BBMAP_BITS, order);
804
805 if (i >= 0) {
806 addr = vb->va->va_start + (i << PAGE_SHIFT);
807 BUG_ON(addr_to_vb_idx(addr) !=
808 addr_to_vb_idx(vb->va->va_start));
809 vb->free -= 1UL << order;
810 if (vb->free == 0) {
811 spin_lock(&vbq->lock);
812 list_del_init(&vb->free_list);
813 spin_unlock(&vbq->lock);
814 }
815 spin_unlock(&vb->lock);
816 break;
817 }
818 spin_unlock(&vb->lock);
819 }
820 put_cpu_var(vmap_cpu_blocks);
821 rcu_read_unlock();
822
823 if (!addr) {
824 vb = new_vmap_block(gfp_mask);
825 if (IS_ERR(vb))
826 return vb;
827 goto again;
828 }
829
830 return (void *)addr;
831}
832
833static void vb_free(const void *addr, unsigned long size)
834{
835 unsigned long offset;
836 unsigned long vb_idx;
837 unsigned int order;
838 struct vmap_block *vb;
839
840 BUG_ON(size & ~PAGE_MASK);
841 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800842
843 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
844
Nick Piggindb64fe02008-10-18 20:27:03 -0700845 order = get_order(size);
846
847 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
848
849 vb_idx = addr_to_vb_idx((unsigned long)addr);
850 rcu_read_lock();
851 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
852 rcu_read_unlock();
853 BUG_ON(!vb);
854
855 spin_lock(&vb->lock);
856 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
MinChan Kimd0868172009-03-31 15:19:26 -0700857
Nick Piggindb64fe02008-10-18 20:27:03 -0700858 vb->dirty += 1UL << order;
859 if (vb->dirty == VMAP_BBMAP_BITS) {
860 BUG_ON(vb->free || !list_empty(&vb->free_list));
861 spin_unlock(&vb->lock);
862 free_vmap_block(vb);
863 } else
864 spin_unlock(&vb->lock);
865}
866
867/**
868 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
869 *
870 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
871 * to amortize TLB flushing overheads. What this means is that any page you
872 * have now, may, in a former life, have been mapped into kernel virtual
873 * address by the vmap layer and so there might be some CPUs with TLB entries
874 * still referencing that page (additional to the regular 1:1 kernel mapping).
875 *
876 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
877 * be sure that none of the pages we have control over will have any aliases
878 * from the vmap layer.
879 */
880void vm_unmap_aliases(void)
881{
882 unsigned long start = ULONG_MAX, end = 0;
883 int cpu;
884 int flush = 0;
885
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100886 if (unlikely(!vmap_initialized))
887 return;
888
Nick Piggindb64fe02008-10-18 20:27:03 -0700889 for_each_possible_cpu(cpu) {
890 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
891 struct vmap_block *vb;
892
893 rcu_read_lock();
894 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
895 int i;
896
897 spin_lock(&vb->lock);
898 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
899 while (i < VMAP_BBMAP_BITS) {
900 unsigned long s, e;
901 int j;
902 j = find_next_zero_bit(vb->dirty_map,
903 VMAP_BBMAP_BITS, i);
904
905 s = vb->va->va_start + (i << PAGE_SHIFT);
906 e = vb->va->va_start + (j << PAGE_SHIFT);
907 vunmap_page_range(s, e);
908 flush = 1;
909
910 if (s < start)
911 start = s;
912 if (e > end)
913 end = e;
914
915 i = j;
916 i = find_next_bit(vb->dirty_map,
917 VMAP_BBMAP_BITS, i);
918 }
919 spin_unlock(&vb->lock);
920 }
921 rcu_read_unlock();
922 }
923
924 __purge_vmap_area_lazy(&start, &end, 1, flush);
925}
926EXPORT_SYMBOL_GPL(vm_unmap_aliases);
927
928/**
929 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
930 * @mem: the pointer returned by vm_map_ram
931 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
932 */
933void vm_unmap_ram(const void *mem, unsigned int count)
934{
935 unsigned long size = count << PAGE_SHIFT;
936 unsigned long addr = (unsigned long)mem;
937
938 BUG_ON(!addr);
939 BUG_ON(addr < VMALLOC_START);
940 BUG_ON(addr > VMALLOC_END);
941 BUG_ON(addr & (PAGE_SIZE-1));
942
943 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -0800944 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700945
946 if (likely(count <= VMAP_MAX_ALLOC))
947 vb_free(mem, size);
948 else
949 free_unmap_vmap_area_addr(addr);
950}
951EXPORT_SYMBOL(vm_unmap_ram);
952
953/**
954 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
955 * @pages: an array of pointers to the pages to be mapped
956 * @count: number of pages
957 * @node: prefer to allocate data structures on this node
958 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -0700959 *
960 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -0700961 */
962void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
963{
964 unsigned long size = count << PAGE_SHIFT;
965 unsigned long addr;
966 void *mem;
967
968 if (likely(count <= VMAP_MAX_ALLOC)) {
969 mem = vb_alloc(size, GFP_KERNEL);
970 if (IS_ERR(mem))
971 return NULL;
972 addr = (unsigned long)mem;
973 } else {
974 struct vmap_area *va;
975 va = alloc_vmap_area(size, PAGE_SIZE,
976 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
977 if (IS_ERR(va))
978 return NULL;
979
980 addr = va->va_start;
981 mem = (void *)addr;
982 }
983 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
984 vm_unmap_ram(mem, count);
985 return NULL;
986 }
987 return mem;
988}
989EXPORT_SYMBOL(vm_map_ram);
990
Tejun Heof0aa6612009-02-20 16:29:08 +0900991/**
992 * vm_area_register_early - register vmap area early during boot
993 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +0900994 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +0900995 *
996 * This function is used to register kernel vm area before
997 * vmalloc_init() is called. @vm->size and @vm->flags should contain
998 * proper values on entry and other fields should be zero. On return,
999 * vm->addr contains the allocated address.
1000 *
1001 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1002 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001003void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001004{
1005 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001006 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001007
Tejun Heoc0c0a292009-02-24 11:57:21 +09001008 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1009 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1010
1011 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001012
1013 vm->next = vmlist;
1014 vmlist = vm;
1015}
1016
Nick Piggindb64fe02008-10-18 20:27:03 -07001017void __init vmalloc_init(void)
1018{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001019 struct vmap_area *va;
1020 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001021 int i;
1022
1023 for_each_possible_cpu(i) {
1024 struct vmap_block_queue *vbq;
1025
1026 vbq = &per_cpu(vmap_block_queue, i);
1027 spin_lock_init(&vbq->lock);
1028 INIT_LIST_HEAD(&vbq->free);
1029 INIT_LIST_HEAD(&vbq->dirty);
1030 vbq->nr_dirty = 0;
1031 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001032
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001033 /* Import existing vmlist entries. */
1034 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001035 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001036 va->flags = tmp->flags | VM_VM_AREA;
1037 va->va_start = (unsigned long)tmp->addr;
1038 va->va_end = va->va_start + tmp->size;
1039 __insert_vmap_area(va);
1040 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001041 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001042}
1043
Tejun Heo8fc48982009-02-20 16:29:08 +09001044/**
1045 * map_kernel_range_noflush - map kernel VM area with the specified pages
1046 * @addr: start of the VM area to map
1047 * @size: size of the VM area to map
1048 * @prot: page protection flags to use
1049 * @pages: pages to map
1050 *
1051 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1052 * specify should have been allocated using get_vm_area() and its
1053 * friends.
1054 *
1055 * NOTE:
1056 * This function does NOT do any cache flushing. The caller is
1057 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1058 * before calling this function.
1059 *
1060 * RETURNS:
1061 * The number of pages mapped on success, -errno on failure.
1062 */
1063int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1064 pgprot_t prot, struct page **pages)
1065{
1066 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1067}
1068
1069/**
1070 * unmap_kernel_range_noflush - unmap kernel VM area
1071 * @addr: start of the VM area to unmap
1072 * @size: size of the VM area to unmap
1073 *
1074 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1075 * specify should have been allocated using get_vm_area() and its
1076 * friends.
1077 *
1078 * NOTE:
1079 * This function does NOT do any cache flushing. The caller is
1080 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1081 * before calling this function and flush_tlb_kernel_range() after.
1082 */
1083void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1084{
1085 vunmap_page_range(addr, addr + size);
1086}
1087
1088/**
1089 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1090 * @addr: start of the VM area to unmap
1091 * @size: size of the VM area to unmap
1092 *
1093 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1094 * the unmapping and tlb after.
1095 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001096void unmap_kernel_range(unsigned long addr, unsigned long size)
1097{
1098 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001099
1100 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001101 vunmap_page_range(addr, end);
1102 flush_tlb_kernel_range(addr, end);
1103}
1104
1105int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1106{
1107 unsigned long addr = (unsigned long)area->addr;
1108 unsigned long end = addr + area->size - PAGE_SIZE;
1109 int err;
1110
1111 err = vmap_page_range(addr, end, prot, *pages);
1112 if (err > 0) {
1113 *pages += err;
1114 err = 0;
1115 }
1116
1117 return err;
1118}
1119EXPORT_SYMBOL_GPL(map_vm_area);
1120
1121/*** Old vmalloc interfaces ***/
1122DEFINE_RWLOCK(vmlist_lock);
1123struct vm_struct *vmlist;
1124
1125static struct vm_struct *__get_vm_area_node(unsigned long size,
1126 unsigned long flags, unsigned long start, unsigned long end,
1127 int node, gfp_t gfp_mask, void *caller)
1128{
1129 static struct vmap_area *va;
1130 struct vm_struct *area;
1131 struct vm_struct *tmp, **p;
1132 unsigned long align = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001134 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (flags & VM_IOREMAP) {
1136 int bit = fls(size);
1137
1138 if (bit > IOREMAP_MAX_ORDER)
1139 bit = IOREMAP_MAX_ORDER;
1140 else if (bit < PAGE_SHIFT)
1141 bit = PAGE_SHIFT;
1142
1143 align = 1ul << bit;
1144 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001147 if (unlikely(!size))
1148 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Christoph Lameter6cb06222007-10-16 01:25:41 -07001150 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 if (unlikely(!area))
1152 return NULL;
1153
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 /*
1155 * We always allocate a guard page.
1156 */
1157 size += PAGE_SIZE;
1158
Nick Piggindb64fe02008-10-18 20:27:03 -07001159 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1160 if (IS_ERR(va)) {
1161 kfree(area);
1162 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 area->flags = flags;
Nick Piggindb64fe02008-10-18 20:27:03 -07001166 area->addr = (void *)va->va_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 area->size = size;
1168 area->pages = NULL;
1169 area->nr_pages = 0;
1170 area->phys_addr = 0;
Christoph Lameter23016962008-04-28 02:12:42 -07001171 area->caller = caller;
Nick Piggindb64fe02008-10-18 20:27:03 -07001172 va->private = area;
1173 va->flags |= VM_VM_AREA;
1174
1175 write_lock(&vmlist_lock);
1176 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1177 if (tmp->addr >= area->addr)
1178 break;
1179 }
1180 area->next = *p;
1181 *p = area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 write_unlock(&vmlist_lock);
1183
1184 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185}
1186
Christoph Lameter930fc452005-10-29 18:15:41 -07001187struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1188 unsigned long start, unsigned long end)
1189{
Christoph Lameter23016962008-04-28 02:12:42 -07001190 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1191 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001192}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001193EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001194
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001195struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1196 unsigned long start, unsigned long end,
1197 void *caller)
1198{
1199 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1200 caller);
1201}
1202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203/**
Simon Arlott183ff222007-10-20 01:27:18 +02001204 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 * @size: size of the area
1206 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1207 *
1208 * Search an area of @size in the kernel virtual mapping area,
1209 * and reserved it for out purposes. Returns the area descriptor
1210 * on success or %NULL on failure.
1211 */
1212struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1213{
Christoph Lameter23016962008-04-28 02:12:42 -07001214 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1215 -1, GFP_KERNEL, __builtin_return_address(0));
1216}
1217
1218struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1219 void *caller)
1220{
1221 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1222 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223}
1224
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001225struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1226 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07001227{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001228 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
Christoph Lameter23016962008-04-28 02:12:42 -07001229 gfp_mask, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001230}
1231
Nick Piggindb64fe02008-10-18 20:27:03 -07001232static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001233{
Nick Piggindb64fe02008-10-18 20:27:03 -07001234 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001235
Nick Piggindb64fe02008-10-18 20:27:03 -07001236 va = find_vmap_area((unsigned long)addr);
1237 if (va && va->flags & VM_VM_AREA)
1238 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001239
Andi Kleen7856dfe2005-05-20 14:27:57 -07001240 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001241}
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243/**
Simon Arlott183ff222007-10-20 01:27:18 +02001244 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * @addr: base address
1246 *
1247 * Search for the kernel VM area starting at @addr, and remove it.
1248 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001249 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001251struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252{
Nick Piggindb64fe02008-10-18 20:27:03 -07001253 struct vmap_area *va;
1254
1255 va = find_vmap_area((unsigned long)addr);
1256 if (va && va->flags & VM_VM_AREA) {
1257 struct vm_struct *vm = va->private;
1258 struct vm_struct *tmp, **p;
Nick Piggincd528582009-01-06 14:39:20 -08001259
1260 vmap_debug_free_range(va->va_start, va->va_end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001261 free_unmap_vmap_area(va);
1262 vm->size -= PAGE_SIZE;
1263
1264 write_lock(&vmlist_lock);
1265 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1266 ;
1267 *p = tmp->next;
1268 write_unlock(&vmlist_lock);
1269
1270 return vm;
1271 }
1272 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273}
1274
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001275static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276{
1277 struct vm_struct *area;
1278
1279 if (!addr)
1280 return;
1281
1282 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001283 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return;
1285 }
1286
1287 area = remove_vm_area(addr);
1288 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001289 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 return;
1292 }
1293
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001294 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001295 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001296
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 if (deallocate_pages) {
1298 int i;
1299
1300 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001301 struct page *page = area->pages[i];
1302
1303 BUG_ON(!page);
1304 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 }
1306
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001307 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 vfree(area->pages);
1309 else
1310 kfree(area->pages);
1311 }
1312
1313 kfree(area);
1314 return;
1315}
1316
1317/**
1318 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 * @addr: memory base address
1320 *
Simon Arlott183ff222007-10-20 01:27:18 +02001321 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001322 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1323 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001325 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001327void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
1329 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001330
1331 kmemleak_free(addr);
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 __vunmap(addr, 1);
1334}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335EXPORT_SYMBOL(vfree);
1336
1337/**
1338 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 * @addr: memory base address
1340 *
1341 * Free the virtually contiguous memory area starting at @addr,
1342 * which was created from the page array passed to vmap().
1343 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001344 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001346void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001349 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 __vunmap(addr, 0);
1351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352EXPORT_SYMBOL(vunmap);
1353
1354/**
1355 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 * @pages: array of page pointers
1357 * @count: number of pages to map
1358 * @flags: vm_area->flags
1359 * @prot: page protection for the mapping
1360 *
1361 * Maps @count pages from @pages into contiguous kernel virtual
1362 * space.
1363 */
1364void *vmap(struct page **pages, unsigned int count,
1365 unsigned long flags, pgprot_t prot)
1366{
1367 struct vm_struct *area;
1368
Peter Zijlstra34754b62009-02-25 16:04:03 +01001369 might_sleep();
1370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (count > num_physpages)
1372 return NULL;
1373
Christoph Lameter23016962008-04-28 02:12:42 -07001374 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1375 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 if (!area)
1377 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 if (map_vm_area(area, prot, &pages)) {
1380 vunmap(area->addr);
1381 return NULL;
1382 }
1383
1384 return area->addr;
1385}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386EXPORT_SYMBOL(vmap);
1387
Nick Piggindb64fe02008-10-18 20:27:03 -07001388static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1389 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001390static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001391 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 struct page **pages;
1394 unsigned int nr_pages, array_size, i;
1395
1396 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1397 array_size = (nr_pages * sizeof(struct page *));
1398
1399 area->nr_pages = nr_pages;
1400 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001401 if (array_size > PAGE_SIZE) {
Christoph Lameter94f60302007-07-17 04:03:29 -07001402 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
Christoph Lameter23016962008-04-28 02:12:42 -07001403 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001404 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001405 } else {
1406 pages = kmalloc_node(array_size,
Christoph Lameter6cb06222007-10-16 01:25:41 -07001407 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
Andrew Morton286e1ea2006-10-17 00:09:57 -07001408 node);
1409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001411 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if (!area->pages) {
1413 remove_vm_area(area->addr);
1414 kfree(area);
1415 return NULL;
1416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
1418 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001419 struct page *page;
1420
Christoph Lameter930fc452005-10-29 18:15:41 -07001421 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001422 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001423 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001424 page = alloc_pages_node(node, gfp_mask, 0);
1425
1426 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 /* Successfully allocated i pages, free them in __vunmap() */
1428 area->nr_pages = i;
1429 goto fail;
1430 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001431 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 }
1433
1434 if (map_vm_area(area, prot, &pages))
1435 goto fail;
1436 return area->addr;
1437
1438fail:
1439 vfree(area->addr);
1440 return NULL;
1441}
1442
Christoph Lameter930fc452005-10-29 18:15:41 -07001443void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1444{
Catalin Marinas89219d32009-06-11 13:23:19 +01001445 void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1,
1446 __builtin_return_address(0));
1447
1448 /*
1449 * A ref_count = 3 is needed because the vm_struct and vmap_area
1450 * structures allocated in the __get_vm_area_node() function contain
1451 * references to the virtual address of the vmalloc'ed block.
1452 */
1453 kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask);
1454
1455 return addr;
Christoph Lameter930fc452005-10-29 18:15:41 -07001456}
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001459 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 * @size: allocation size
1461 * @gfp_mask: flags for the page level allocator
1462 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001463 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001464 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 *
1466 * Allocate enough pages to cover @size from the page level
1467 * allocator with @gfp_mask flags. Map them into contiguous
1468 * kernel virtual space, using a pagetable protection of @prot.
1469 */
Adrian Bunkb2213852006-09-25 23:31:02 -07001470static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
Christoph Lameter23016962008-04-28 02:12:42 -07001471 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
1473 struct vm_struct *area;
Catalin Marinas89219d32009-06-11 13:23:19 +01001474 void *addr;
1475 unsigned long real_size = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
1477 size = PAGE_ALIGN(size);
1478 if (!size || (size >> PAGE_SHIFT) > num_physpages)
1479 return NULL;
1480
Christoph Lameter23016962008-04-28 02:12:42 -07001481 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1482 node, gfp_mask, caller);
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 if (!area)
1485 return NULL;
1486
Catalin Marinas89219d32009-06-11 13:23:19 +01001487 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1488
1489 /*
1490 * A ref_count = 3 is needed because the vm_struct and vmap_area
1491 * structures allocated in the __get_vm_area_node() function contain
1492 * references to the virtual address of the vmalloc'ed block.
1493 */
1494 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1495
1496 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497}
1498
Christoph Lameter930fc452005-10-29 18:15:41 -07001499void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1500{
Christoph Lameter23016962008-04-28 02:12:42 -07001501 return __vmalloc_node(size, gfp_mask, prot, -1,
1502 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001503}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504EXPORT_SYMBOL(__vmalloc);
1505
1506/**
1507 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 * Allocate enough pages to cover @size from the page level
1510 * allocator and map them into contiguous kernel virtual space.
1511 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001512 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 * use __vmalloc() instead.
1514 */
1515void *vmalloc(unsigned long size)
1516{
Christoph Lameter23016962008-04-28 02:12:42 -07001517 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1518 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520EXPORT_SYMBOL(vmalloc);
1521
Christoph Lameter930fc452005-10-29 18:15:41 -07001522/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001523 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1524 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001525 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001526 * The resulting memory area is zeroed so it can be mapped to userspace
1527 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001528 */
1529void *vmalloc_user(unsigned long size)
1530{
1531 struct vm_struct *area;
1532 void *ret;
1533
Glauber Costa84877842009-01-06 14:39:19 -08001534 ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1535 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001536 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001537 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001538 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001539 }
Nick Piggin83342312006-06-23 02:03:20 -07001540 return ret;
1541}
1542EXPORT_SYMBOL(vmalloc_user);
1543
1544/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001545 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001546 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001547 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001548 *
1549 * Allocate enough pages to cover @size from the page level
1550 * allocator and map them into contiguous kernel virtual space.
1551 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001552 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001553 * use __vmalloc() instead.
1554 */
1555void *vmalloc_node(unsigned long size, int node)
1556{
Christoph Lameter23016962008-04-28 02:12:42 -07001557 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1558 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001559}
1560EXPORT_SYMBOL(vmalloc_node);
1561
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001562#ifndef PAGE_KERNEL_EXEC
1563# define PAGE_KERNEL_EXEC PAGE_KERNEL
1564#endif
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566/**
1567 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 * @size: allocation size
1569 *
1570 * Kernel-internal function to allocate enough pages to cover @size
1571 * the page level allocator and map them into contiguous and
1572 * executable kernel virtual space.
1573 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001574 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 * use __vmalloc() instead.
1576 */
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578void *vmalloc_exec(unsigned long size)
1579{
Glauber Costa84877842009-01-06 14:39:19 -08001580 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1581 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582}
1583
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001584#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001585#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001586#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001587#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001588#else
1589#define GFP_VMALLOC32 GFP_KERNEL
1590#endif
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/**
1593 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 * @size: allocation size
1595 *
1596 * Allocate enough 32bit PA addressable pages to cover @size from the
1597 * page level allocator and map them into contiguous kernel virtual space.
1598 */
1599void *vmalloc_32(unsigned long size)
1600{
Glauber Costa84877842009-01-06 14:39:19 -08001601 return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1602 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604EXPORT_SYMBOL(vmalloc_32);
1605
Nick Piggin83342312006-06-23 02:03:20 -07001606/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001607 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001608 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001609 *
1610 * The resulting memory area is 32bit addressable and zeroed so it can be
1611 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001612 */
1613void *vmalloc_32_user(unsigned long size)
1614{
1615 struct vm_struct *area;
1616 void *ret;
1617
Glauber Costa84877842009-01-06 14:39:19 -08001618 ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1619 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001620 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001621 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001622 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001623 }
Nick Piggin83342312006-06-23 02:03:20 -07001624 return ret;
1625}
1626EXPORT_SYMBOL(vmalloc_32_user);
1627
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628long vread(char *buf, char *addr, unsigned long count)
1629{
1630 struct vm_struct *tmp;
1631 char *vaddr, *buf_start = buf;
1632 unsigned long n;
1633
1634 /* Don't allow overflow */
1635 if ((unsigned long) addr + count < count)
1636 count = -(unsigned long) addr;
1637
1638 read_lock(&vmlist_lock);
1639 for (tmp = vmlist; tmp; tmp = tmp->next) {
1640 vaddr = (char *) tmp->addr;
1641 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1642 continue;
1643 while (addr < vaddr) {
1644 if (count == 0)
1645 goto finished;
1646 *buf = '\0';
1647 buf++;
1648 addr++;
1649 count--;
1650 }
1651 n = vaddr + tmp->size - PAGE_SIZE - addr;
1652 do {
1653 if (count == 0)
1654 goto finished;
1655 *buf = *addr;
1656 buf++;
1657 addr++;
1658 count--;
1659 } while (--n > 0);
1660 }
1661finished:
1662 read_unlock(&vmlist_lock);
1663 return buf - buf_start;
1664}
1665
1666long vwrite(char *buf, char *addr, unsigned long count)
1667{
1668 struct vm_struct *tmp;
1669 char *vaddr, *buf_start = buf;
1670 unsigned long n;
1671
1672 /* Don't allow overflow */
1673 if ((unsigned long) addr + count < count)
1674 count = -(unsigned long) addr;
1675
1676 read_lock(&vmlist_lock);
1677 for (tmp = vmlist; tmp; tmp = tmp->next) {
1678 vaddr = (char *) tmp->addr;
1679 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1680 continue;
1681 while (addr < vaddr) {
1682 if (count == 0)
1683 goto finished;
1684 buf++;
1685 addr++;
1686 count--;
1687 }
1688 n = vaddr + tmp->size - PAGE_SIZE - addr;
1689 do {
1690 if (count == 0)
1691 goto finished;
1692 *addr = *buf;
1693 buf++;
1694 addr++;
1695 count--;
1696 } while (--n > 0);
1697 }
1698finished:
1699 read_unlock(&vmlist_lock);
1700 return buf - buf_start;
1701}
Nick Piggin83342312006-06-23 02:03:20 -07001702
1703/**
1704 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07001705 * @vma: vma to cover (map full range of vma)
1706 * @addr: vmalloc memory
1707 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07001708 *
1709 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07001710 *
1711 * This function checks that addr is a valid vmalloc'ed area, and
1712 * that it is big enough to cover the vma. Will return failure if
1713 * that criteria isn't met.
1714 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001715 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07001716 */
1717int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1718 unsigned long pgoff)
1719{
1720 struct vm_struct *area;
1721 unsigned long uaddr = vma->vm_start;
1722 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07001723
1724 if ((PAGE_SIZE-1) & (unsigned long)addr)
1725 return -EINVAL;
1726
Nick Piggindb64fe02008-10-18 20:27:03 -07001727 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07001728 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07001729 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001730
1731 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07001732 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001733
1734 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07001735 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001736
1737 addr += pgoff << PAGE_SHIFT;
1738 do {
1739 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07001740 int ret;
1741
Nick Piggin83342312006-06-23 02:03:20 -07001742 ret = vm_insert_page(vma, uaddr, page);
1743 if (ret)
1744 return ret;
1745
1746 uaddr += PAGE_SIZE;
1747 addr += PAGE_SIZE;
1748 usize -= PAGE_SIZE;
1749 } while (usize > 0);
1750
1751 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1752 vma->vm_flags |= VM_RESERVED;
1753
Nick Piggindb64fe02008-10-18 20:27:03 -07001754 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07001755}
1756EXPORT_SYMBOL(remap_vmalloc_range);
1757
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07001758/*
1759 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1760 * have one.
1761 */
1762void __attribute__((weak)) vmalloc_sync_all(void)
1763{
1764}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001765
1766
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001767static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001768{
1769 /* apply_to_page_range() does all the hard work. */
1770 return 0;
1771}
1772
1773/**
1774 * alloc_vm_area - allocate a range of kernel address space
1775 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07001776 *
1777 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001778 *
1779 * This function reserves a range of kernel address space, and
1780 * allocates pagetables to map that range. No actual mappings
1781 * are created. If the kernel address space is not shared
1782 * between processes, it syncs the pagetable across all
1783 * processes.
1784 */
1785struct vm_struct *alloc_vm_area(size_t size)
1786{
1787 struct vm_struct *area;
1788
Christoph Lameter23016962008-04-28 02:12:42 -07001789 area = get_vm_area_caller(size, VM_IOREMAP,
1790 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001791 if (area == NULL)
1792 return NULL;
1793
1794 /*
1795 * This ensures that page tables are constructed for this region
1796 * of kernel virtual address space and mapped into init_mm.
1797 */
1798 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1799 area->size, f, NULL)) {
1800 free_vm_area(area);
1801 return NULL;
1802 }
1803
1804 /* Make sure the pagetables are constructed in process kernel
1805 mappings */
1806 vmalloc_sync_all();
1807
1808 return area;
1809}
1810EXPORT_SYMBOL_GPL(alloc_vm_area);
1811
1812void free_vm_area(struct vm_struct *area)
1813{
1814 struct vm_struct *ret;
1815 ret = remove_vm_area(area->addr);
1816 BUG_ON(ret != area);
1817 kfree(area);
1818}
1819EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07001820
1821
1822#ifdef CONFIG_PROC_FS
1823static void *s_start(struct seq_file *m, loff_t *pos)
1824{
1825 loff_t n = *pos;
1826 struct vm_struct *v;
1827
1828 read_lock(&vmlist_lock);
1829 v = vmlist;
1830 while (n > 0 && v) {
1831 n--;
1832 v = v->next;
1833 }
1834 if (!n)
1835 return v;
1836
1837 return NULL;
1838
1839}
1840
1841static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1842{
1843 struct vm_struct *v = p;
1844
1845 ++*pos;
1846 return v->next;
1847}
1848
1849static void s_stop(struct seq_file *m, void *p)
1850{
1851 read_unlock(&vmlist_lock);
1852}
1853
Eric Dumazeta47a1262008-07-23 21:27:38 -07001854static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1855{
1856 if (NUMA_BUILD) {
1857 unsigned int nr, *counters = m->private;
1858
1859 if (!counters)
1860 return;
1861
1862 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1863
1864 for (nr = 0; nr < v->nr_pages; nr++)
1865 counters[page_to_nid(v->pages[nr])]++;
1866
1867 for_each_node_state(nr, N_HIGH_MEMORY)
1868 if (counters[nr])
1869 seq_printf(m, " N%u=%u", nr, counters[nr]);
1870 }
1871}
1872
Christoph Lametera10aa572008-04-28 02:12:40 -07001873static int s_show(struct seq_file *m, void *p)
1874{
1875 struct vm_struct *v = p;
1876
1877 seq_printf(m, "0x%p-0x%p %7ld",
1878 v->addr, v->addr + v->size, v->size);
1879
Christoph Lameter23016962008-04-28 02:12:42 -07001880 if (v->caller) {
Hugh Dickins9c246242008-12-09 13:14:27 -08001881 char buff[KSYM_SYMBOL_LEN];
Christoph Lameter23016962008-04-28 02:12:42 -07001882
1883 seq_putc(m, ' ');
1884 sprint_symbol(buff, (unsigned long)v->caller);
1885 seq_puts(m, buff);
1886 }
1887
Christoph Lametera10aa572008-04-28 02:12:40 -07001888 if (v->nr_pages)
1889 seq_printf(m, " pages=%d", v->nr_pages);
1890
1891 if (v->phys_addr)
1892 seq_printf(m, " phys=%lx", v->phys_addr);
1893
1894 if (v->flags & VM_IOREMAP)
1895 seq_printf(m, " ioremap");
1896
1897 if (v->flags & VM_ALLOC)
1898 seq_printf(m, " vmalloc");
1899
1900 if (v->flags & VM_MAP)
1901 seq_printf(m, " vmap");
1902
1903 if (v->flags & VM_USERMAP)
1904 seq_printf(m, " user");
1905
1906 if (v->flags & VM_VPAGES)
1907 seq_printf(m, " vpages");
1908
Eric Dumazeta47a1262008-07-23 21:27:38 -07001909 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07001910 seq_putc(m, '\n');
1911 return 0;
1912}
1913
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001914static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07001915 .start = s_start,
1916 .next = s_next,
1917 .stop = s_stop,
1918 .show = s_show,
1919};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001920
1921static int vmalloc_open(struct inode *inode, struct file *file)
1922{
1923 unsigned int *ptr = NULL;
1924 int ret;
1925
1926 if (NUMA_BUILD)
1927 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1928 ret = seq_open(file, &vmalloc_op);
1929 if (!ret) {
1930 struct seq_file *m = file->private_data;
1931 m->private = ptr;
1932 } else
1933 kfree(ptr);
1934 return ret;
1935}
1936
1937static const struct file_operations proc_vmalloc_operations = {
1938 .open = vmalloc_open,
1939 .read = seq_read,
1940 .llseek = seq_lseek,
1941 .release = seq_release_private,
1942};
1943
1944static int __init proc_vmalloc_init(void)
1945{
1946 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1947 return 0;
1948}
1949module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07001950#endif
1951