blob: fab19876b4d178986979c2b5d2cb353285507b46 [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>
Tejun Heof0aa6612009-02-20 16:29:08 +090027#include <linux/pfn.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);
Nick Piggindb64fe02008-10-18 20:27:03 -0700405 return ERR_PTR(-EBUSY);
406 }
407
408 BUG_ON(addr & (align-1));
409
410 va->va_start = addr;
411 va->va_end = addr + size;
412 va->flags = 0;
413 __insert_vmap_area(va);
414 spin_unlock(&vmap_area_lock);
415
416 return va;
417}
418
419static void rcu_free_va(struct rcu_head *head)
420{
421 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
422
423 kfree(va);
424}
425
426static void __free_vmap_area(struct vmap_area *va)
427{
428 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
429 rb_erase(&va->rb_node, &vmap_area_root);
430 RB_CLEAR_NODE(&va->rb_node);
431 list_del_rcu(&va->list);
432
433 call_rcu(&va->rcu_head, rcu_free_va);
434}
435
436/*
437 * Free a region of KVA allocated by alloc_vmap_area
438 */
439static void free_vmap_area(struct vmap_area *va)
440{
441 spin_lock(&vmap_area_lock);
442 __free_vmap_area(va);
443 spin_unlock(&vmap_area_lock);
444}
445
446/*
447 * Clear the pagetable entries of a given vmap_area
448 */
449static void unmap_vmap_area(struct vmap_area *va)
450{
451 vunmap_page_range(va->va_start, va->va_end);
452}
453
Nick Piggincd528582009-01-06 14:39:20 -0800454static void vmap_debug_free_range(unsigned long start, unsigned long end)
455{
456 /*
457 * Unmap page tables and force a TLB flush immediately if
458 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
459 * bugs similarly to those in linear kernel virtual address
460 * space after a page has been freed.
461 *
462 * All the lazy freeing logic is still retained, in order to
463 * minimise intrusiveness of this debugging feature.
464 *
465 * This is going to be *slow* (linear kernel virtual address
466 * debugging doesn't do a broadcast TLB flush so it is a lot
467 * faster).
468 */
469#ifdef CONFIG_DEBUG_PAGEALLOC
470 vunmap_page_range(start, end);
471 flush_tlb_kernel_range(start, end);
472#endif
473}
474
Nick Piggindb64fe02008-10-18 20:27:03 -0700475/*
476 * lazy_max_pages is the maximum amount of virtual address space we gather up
477 * before attempting to purge with a TLB flush.
478 *
479 * There is a tradeoff here: a larger number will cover more kernel page tables
480 * and take slightly longer to purge, but it will linearly reduce the number of
481 * global TLB flushes that must be performed. It would seem natural to scale
482 * this number up linearly with the number of CPUs (because vmapping activity
483 * could also scale linearly with the number of CPUs), however it is likely
484 * that in practice, workloads might be constrained in other ways that mean
485 * vmap activity will not scale linearly with CPUs. Also, I want to be
486 * conservative and not introduce a big latency on huge systems, so go with
487 * a less aggressive log scale. It will still be an improvement over the old
488 * code, and it will be simple to change the scale factor if we find that it
489 * becomes a problem on bigger systems.
490 */
491static unsigned long lazy_max_pages(void)
492{
493 unsigned int log;
494
495 log = fls(num_online_cpus());
496
497 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
498}
499
500static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
501
502/*
503 * Purges all lazily-freed vmap areas.
504 *
505 * If sync is 0 then don't purge if there is already a purge in progress.
506 * If force_flush is 1, then flush kernel TLBs between *start and *end even
507 * if we found no lazy vmap areas to unmap (callers can use this to optimise
508 * their own TLB flushing).
509 * Returns with *start = min(*start, lowest purged address)
510 * *end = max(*end, highest purged address)
511 */
512static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
513 int sync, int force_flush)
514{
Andrew Morton46666d82009-01-15 13:51:15 -0800515 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700516 LIST_HEAD(valist);
517 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800518 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700519 int nr = 0;
520
521 /*
522 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
523 * should not expect such behaviour. This just simplifies locking for
524 * the case that isn't actually used at the moment anyway.
525 */
526 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800527 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700528 return;
529 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800530 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700531
532 rcu_read_lock();
533 list_for_each_entry_rcu(va, &vmap_area_list, list) {
534 if (va->flags & VM_LAZY_FREE) {
535 if (va->va_start < *start)
536 *start = va->va_start;
537 if (va->va_end > *end)
538 *end = va->va_end;
539 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
540 unmap_vmap_area(va);
541 list_add_tail(&va->purge_list, &valist);
542 va->flags |= VM_LAZY_FREEING;
543 va->flags &= ~VM_LAZY_FREE;
544 }
545 }
546 rcu_read_unlock();
547
548 if (nr) {
549 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
550 atomic_sub(nr, &vmap_lazy_nr);
551 }
552
553 if (nr || force_flush)
554 flush_tlb_kernel_range(*start, *end);
555
556 if (nr) {
557 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800558 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700559 __free_vmap_area(va);
560 spin_unlock(&vmap_area_lock);
561 }
Andrew Morton46666d82009-01-15 13:51:15 -0800562 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700563}
564
565/*
Nick Piggin496850e2008-11-19 15:36:33 -0800566 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
567 * is already purging.
568 */
569static void try_purge_vmap_area_lazy(void)
570{
571 unsigned long start = ULONG_MAX, end = 0;
572
573 __purge_vmap_area_lazy(&start, &end, 0, 0);
574}
575
576/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700577 * Kick off a purge of the outstanding lazy areas.
578 */
579static void purge_vmap_area_lazy(void)
580{
581 unsigned long start = ULONG_MAX, end = 0;
582
Nick Piggin496850e2008-11-19 15:36:33 -0800583 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700584}
585
586/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800587 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
588 * called for the correct range previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700589 */
Nick Pigginb29acbd2008-12-01 13:13:47 -0800590static void free_unmap_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700591{
592 va->flags |= VM_LAZY_FREE;
593 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
594 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800595 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700596}
597
Nick Pigginb29acbd2008-12-01 13:13:47 -0800598/*
599 * Free and unmap a vmap area
600 */
601static void free_unmap_vmap_area(struct vmap_area *va)
602{
603 flush_cache_vunmap(va->va_start, va->va_end);
604 free_unmap_vmap_area_noflush(va);
605}
606
Nick Piggindb64fe02008-10-18 20:27:03 -0700607static struct vmap_area *find_vmap_area(unsigned long addr)
608{
609 struct vmap_area *va;
610
611 spin_lock(&vmap_area_lock);
612 va = __find_vmap_area(addr);
613 spin_unlock(&vmap_area_lock);
614
615 return va;
616}
617
618static void free_unmap_vmap_area_addr(unsigned long addr)
619{
620 struct vmap_area *va;
621
622 va = find_vmap_area(addr);
623 BUG_ON(!va);
624 free_unmap_vmap_area(va);
625}
626
627
628/*** Per cpu kva allocator ***/
629
630/*
631 * vmap space is limited especially on 32 bit architectures. Ensure there is
632 * room for at least 16 percpu vmap blocks per CPU.
633 */
634/*
635 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
636 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
637 * instead (we just need a rough idea)
638 */
639#if BITS_PER_LONG == 32
640#define VMALLOC_SPACE (128UL*1024*1024)
641#else
642#define VMALLOC_SPACE (128UL*1024*1024*1024)
643#endif
644
645#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
646#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
647#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
648#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
649#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
650#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
651#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
652 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
653 VMALLOC_PAGES / NR_CPUS / 16))
654
655#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
656
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100657static bool vmap_initialized __read_mostly = false;
658
Nick Piggindb64fe02008-10-18 20:27:03 -0700659struct vmap_block_queue {
660 spinlock_t lock;
661 struct list_head free;
662 struct list_head dirty;
663 unsigned int nr_dirty;
664};
665
666struct vmap_block {
667 spinlock_t lock;
668 struct vmap_area *va;
669 struct vmap_block_queue *vbq;
670 unsigned long free, dirty;
671 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
672 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
673 union {
MinChan Kimd0868172009-03-31 15:19:26 -0700674 struct list_head free_list;
Nick Piggindb64fe02008-10-18 20:27:03 -0700675 struct rcu_head rcu_head;
676 };
677};
678
679/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
680static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
681
682/*
683 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
684 * in the free path. Could get rid of this if we change the API to return a
685 * "cookie" from alloc, to be passed to free. But no big deal yet.
686 */
687static DEFINE_SPINLOCK(vmap_block_tree_lock);
688static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
689
690/*
691 * We should probably have a fallback mechanism to allocate virtual memory
692 * out of partially filled vmap blocks. However vmap block sizing should be
693 * fairly reasonable according to the vmalloc size, so it shouldn't be a
694 * big problem.
695 */
696
697static unsigned long addr_to_vb_idx(unsigned long addr)
698{
699 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
700 addr /= VMAP_BLOCK_SIZE;
701 return addr;
702}
703
704static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
705{
706 struct vmap_block_queue *vbq;
707 struct vmap_block *vb;
708 struct vmap_area *va;
709 unsigned long vb_idx;
710 int node, err;
711
712 node = numa_node_id();
713
714 vb = kmalloc_node(sizeof(struct vmap_block),
715 gfp_mask & GFP_RECLAIM_MASK, node);
716 if (unlikely(!vb))
717 return ERR_PTR(-ENOMEM);
718
719 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
720 VMALLOC_START, VMALLOC_END,
721 node, gfp_mask);
722 if (unlikely(IS_ERR(va))) {
723 kfree(vb);
724 return ERR_PTR(PTR_ERR(va));
725 }
726
727 err = radix_tree_preload(gfp_mask);
728 if (unlikely(err)) {
729 kfree(vb);
730 free_vmap_area(va);
731 return ERR_PTR(err);
732 }
733
734 spin_lock_init(&vb->lock);
735 vb->va = va;
736 vb->free = VMAP_BBMAP_BITS;
737 vb->dirty = 0;
738 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
739 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
740 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700741
742 vb_idx = addr_to_vb_idx(va->va_start);
743 spin_lock(&vmap_block_tree_lock);
744 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
745 spin_unlock(&vmap_block_tree_lock);
746 BUG_ON(err);
747 radix_tree_preload_end();
748
749 vbq = &get_cpu_var(vmap_block_queue);
750 vb->vbq = vbq;
751 spin_lock(&vbq->lock);
752 list_add(&vb->free_list, &vbq->free);
753 spin_unlock(&vbq->lock);
754 put_cpu_var(vmap_cpu_blocks);
755
756 return vb;
757}
758
759static void rcu_free_vb(struct rcu_head *head)
760{
761 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
762
763 kfree(vb);
764}
765
766static void free_vmap_block(struct vmap_block *vb)
767{
768 struct vmap_block *tmp;
769 unsigned long vb_idx;
770
MinChan Kimd0868172009-03-31 15:19:26 -0700771 BUG_ON(!list_empty(&vb->free_list));
Nick Piggindb64fe02008-10-18 20:27:03 -0700772
773 vb_idx = addr_to_vb_idx(vb->va->va_start);
774 spin_lock(&vmap_block_tree_lock);
775 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
776 spin_unlock(&vmap_block_tree_lock);
777 BUG_ON(tmp != vb);
778
Nick Pigginb29acbd2008-12-01 13:13:47 -0800779 free_unmap_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700780 call_rcu(&vb->rcu_head, rcu_free_vb);
781}
782
783static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
784{
785 struct vmap_block_queue *vbq;
786 struct vmap_block *vb;
787 unsigned long addr = 0;
788 unsigned int order;
789
790 BUG_ON(size & ~PAGE_MASK);
791 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
792 order = get_order(size);
793
794again:
795 rcu_read_lock();
796 vbq = &get_cpu_var(vmap_block_queue);
797 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
798 int i;
799
800 spin_lock(&vb->lock);
801 i = bitmap_find_free_region(vb->alloc_map,
802 VMAP_BBMAP_BITS, order);
803
804 if (i >= 0) {
805 addr = vb->va->va_start + (i << PAGE_SHIFT);
806 BUG_ON(addr_to_vb_idx(addr) !=
807 addr_to_vb_idx(vb->va->va_start));
808 vb->free -= 1UL << order;
809 if (vb->free == 0) {
810 spin_lock(&vbq->lock);
811 list_del_init(&vb->free_list);
812 spin_unlock(&vbq->lock);
813 }
814 spin_unlock(&vb->lock);
815 break;
816 }
817 spin_unlock(&vb->lock);
818 }
819 put_cpu_var(vmap_cpu_blocks);
820 rcu_read_unlock();
821
822 if (!addr) {
823 vb = new_vmap_block(gfp_mask);
824 if (IS_ERR(vb))
825 return vb;
826 goto again;
827 }
828
829 return (void *)addr;
830}
831
832static void vb_free(const void *addr, unsigned long size)
833{
834 unsigned long offset;
835 unsigned long vb_idx;
836 unsigned int order;
837 struct vmap_block *vb;
838
839 BUG_ON(size & ~PAGE_MASK);
840 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800841
842 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
843
Nick Piggindb64fe02008-10-18 20:27:03 -0700844 order = get_order(size);
845
846 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
847
848 vb_idx = addr_to_vb_idx((unsigned long)addr);
849 rcu_read_lock();
850 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
851 rcu_read_unlock();
852 BUG_ON(!vb);
853
854 spin_lock(&vb->lock);
855 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
MinChan Kimd0868172009-03-31 15:19:26 -0700856
Nick Piggindb64fe02008-10-18 20:27:03 -0700857 vb->dirty += 1UL << order;
858 if (vb->dirty == VMAP_BBMAP_BITS) {
859 BUG_ON(vb->free || !list_empty(&vb->free_list));
860 spin_unlock(&vb->lock);
861 free_vmap_block(vb);
862 } else
863 spin_unlock(&vb->lock);
864}
865
866/**
867 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
868 *
869 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
870 * to amortize TLB flushing overheads. What this means is that any page you
871 * have now, may, in a former life, have been mapped into kernel virtual
872 * address by the vmap layer and so there might be some CPUs with TLB entries
873 * still referencing that page (additional to the regular 1:1 kernel mapping).
874 *
875 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
876 * be sure that none of the pages we have control over will have any aliases
877 * from the vmap layer.
878 */
879void vm_unmap_aliases(void)
880{
881 unsigned long start = ULONG_MAX, end = 0;
882 int cpu;
883 int flush = 0;
884
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100885 if (unlikely(!vmap_initialized))
886 return;
887
Nick Piggindb64fe02008-10-18 20:27:03 -0700888 for_each_possible_cpu(cpu) {
889 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
890 struct vmap_block *vb;
891
892 rcu_read_lock();
893 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
894 int i;
895
896 spin_lock(&vb->lock);
897 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
898 while (i < VMAP_BBMAP_BITS) {
899 unsigned long s, e;
900 int j;
901 j = find_next_zero_bit(vb->dirty_map,
902 VMAP_BBMAP_BITS, i);
903
904 s = vb->va->va_start + (i << PAGE_SHIFT);
905 e = vb->va->va_start + (j << PAGE_SHIFT);
906 vunmap_page_range(s, e);
907 flush = 1;
908
909 if (s < start)
910 start = s;
911 if (e > end)
912 end = e;
913
914 i = j;
915 i = find_next_bit(vb->dirty_map,
916 VMAP_BBMAP_BITS, i);
917 }
918 spin_unlock(&vb->lock);
919 }
920 rcu_read_unlock();
921 }
922
923 __purge_vmap_area_lazy(&start, &end, 1, flush);
924}
925EXPORT_SYMBOL_GPL(vm_unmap_aliases);
926
927/**
928 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
929 * @mem: the pointer returned by vm_map_ram
930 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
931 */
932void vm_unmap_ram(const void *mem, unsigned int count)
933{
934 unsigned long size = count << PAGE_SHIFT;
935 unsigned long addr = (unsigned long)mem;
936
937 BUG_ON(!addr);
938 BUG_ON(addr < VMALLOC_START);
939 BUG_ON(addr > VMALLOC_END);
940 BUG_ON(addr & (PAGE_SIZE-1));
941
942 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -0800943 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700944
945 if (likely(count <= VMAP_MAX_ALLOC))
946 vb_free(mem, size);
947 else
948 free_unmap_vmap_area_addr(addr);
949}
950EXPORT_SYMBOL(vm_unmap_ram);
951
952/**
953 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
954 * @pages: an array of pointers to the pages to be mapped
955 * @count: number of pages
956 * @node: prefer to allocate data structures on this node
957 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -0700958 *
959 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -0700960 */
961void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
962{
963 unsigned long size = count << PAGE_SHIFT;
964 unsigned long addr;
965 void *mem;
966
967 if (likely(count <= VMAP_MAX_ALLOC)) {
968 mem = vb_alloc(size, GFP_KERNEL);
969 if (IS_ERR(mem))
970 return NULL;
971 addr = (unsigned long)mem;
972 } else {
973 struct vmap_area *va;
974 va = alloc_vmap_area(size, PAGE_SIZE,
975 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
976 if (IS_ERR(va))
977 return NULL;
978
979 addr = va->va_start;
980 mem = (void *)addr;
981 }
982 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
983 vm_unmap_ram(mem, count);
984 return NULL;
985 }
986 return mem;
987}
988EXPORT_SYMBOL(vm_map_ram);
989
Tejun Heof0aa6612009-02-20 16:29:08 +0900990/**
991 * vm_area_register_early - register vmap area early during boot
992 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +0900993 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +0900994 *
995 * This function is used to register kernel vm area before
996 * vmalloc_init() is called. @vm->size and @vm->flags should contain
997 * proper values on entry and other fields should be zero. On return,
998 * vm->addr contains the allocated address.
999 *
1000 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1001 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001002void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001003{
1004 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001005 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001006
Tejun Heoc0c0a292009-02-24 11:57:21 +09001007 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1008 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1009
1010 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001011
1012 vm->next = vmlist;
1013 vmlist = vm;
1014}
1015
Nick Piggindb64fe02008-10-18 20:27:03 -07001016void __init vmalloc_init(void)
1017{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001018 struct vmap_area *va;
1019 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001020 int i;
1021
1022 for_each_possible_cpu(i) {
1023 struct vmap_block_queue *vbq;
1024
1025 vbq = &per_cpu(vmap_block_queue, i);
1026 spin_lock_init(&vbq->lock);
1027 INIT_LIST_HEAD(&vbq->free);
1028 INIT_LIST_HEAD(&vbq->dirty);
1029 vbq->nr_dirty = 0;
1030 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001031
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001032 /* Import existing vmlist entries. */
1033 for (tmp = vmlist; tmp; tmp = tmp->next) {
1034 va = alloc_bootmem(sizeof(struct vmap_area));
1035 va->flags = tmp->flags | VM_VM_AREA;
1036 va->va_start = (unsigned long)tmp->addr;
1037 va->va_end = va->va_start + tmp->size;
1038 __insert_vmap_area(va);
1039 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001040 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001041}
1042
Tejun Heo8fc48982009-02-20 16:29:08 +09001043/**
1044 * map_kernel_range_noflush - map kernel VM area with the specified pages
1045 * @addr: start of the VM area to map
1046 * @size: size of the VM area to map
1047 * @prot: page protection flags to use
1048 * @pages: pages to map
1049 *
1050 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1051 * specify should have been allocated using get_vm_area() and its
1052 * friends.
1053 *
1054 * NOTE:
1055 * This function does NOT do any cache flushing. The caller is
1056 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1057 * before calling this function.
1058 *
1059 * RETURNS:
1060 * The number of pages mapped on success, -errno on failure.
1061 */
1062int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1063 pgprot_t prot, struct page **pages)
1064{
1065 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1066}
1067
1068/**
1069 * unmap_kernel_range_noflush - unmap kernel VM area
1070 * @addr: start of the VM area to unmap
1071 * @size: size of the VM area to unmap
1072 *
1073 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1074 * specify should have been allocated using get_vm_area() and its
1075 * friends.
1076 *
1077 * NOTE:
1078 * This function does NOT do any cache flushing. The caller is
1079 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1080 * before calling this function and flush_tlb_kernel_range() after.
1081 */
1082void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1083{
1084 vunmap_page_range(addr, addr + size);
1085}
1086
1087/**
1088 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1089 * @addr: start of the VM area to unmap
1090 * @size: size of the VM area to unmap
1091 *
1092 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1093 * the unmapping and tlb after.
1094 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001095void unmap_kernel_range(unsigned long addr, unsigned long size)
1096{
1097 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001098
1099 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001100 vunmap_page_range(addr, end);
1101 flush_tlb_kernel_range(addr, end);
1102}
1103
1104int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1105{
1106 unsigned long addr = (unsigned long)area->addr;
1107 unsigned long end = addr + area->size - PAGE_SIZE;
1108 int err;
1109
1110 err = vmap_page_range(addr, end, prot, *pages);
1111 if (err > 0) {
1112 *pages += err;
1113 err = 0;
1114 }
1115
1116 return err;
1117}
1118EXPORT_SYMBOL_GPL(map_vm_area);
1119
1120/*** Old vmalloc interfaces ***/
1121DEFINE_RWLOCK(vmlist_lock);
1122struct vm_struct *vmlist;
1123
1124static struct vm_struct *__get_vm_area_node(unsigned long size,
1125 unsigned long flags, unsigned long start, unsigned long end,
1126 int node, gfp_t gfp_mask, void *caller)
1127{
1128 static struct vmap_area *va;
1129 struct vm_struct *area;
1130 struct vm_struct *tmp, **p;
1131 unsigned long align = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001133 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (flags & VM_IOREMAP) {
1135 int bit = fls(size);
1136
1137 if (bit > IOREMAP_MAX_ORDER)
1138 bit = IOREMAP_MAX_ORDER;
1139 else if (bit < PAGE_SHIFT)
1140 bit = PAGE_SHIFT;
1141
1142 align = 1ul << bit;
1143 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001146 if (unlikely(!size))
1147 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Christoph Lameter6cb06222007-10-16 01:25:41 -07001149 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (unlikely(!area))
1151 return NULL;
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 /*
1154 * We always allocate a guard page.
1155 */
1156 size += PAGE_SIZE;
1157
Nick Piggindb64fe02008-10-18 20:27:03 -07001158 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1159 if (IS_ERR(va)) {
1160 kfree(area);
1161 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 area->flags = flags;
Nick Piggindb64fe02008-10-18 20:27:03 -07001165 area->addr = (void *)va->va_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 area->size = size;
1167 area->pages = NULL;
1168 area->nr_pages = 0;
1169 area->phys_addr = 0;
Christoph Lameter23016962008-04-28 02:12:42 -07001170 area->caller = caller;
Nick Piggindb64fe02008-10-18 20:27:03 -07001171 va->private = area;
1172 va->flags |= VM_VM_AREA;
1173
1174 write_lock(&vmlist_lock);
1175 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1176 if (tmp->addr >= area->addr)
1177 break;
1178 }
1179 area->next = *p;
1180 *p = area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 write_unlock(&vmlist_lock);
1182
1183 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184}
1185
Christoph Lameter930fc452005-10-29 18:15:41 -07001186struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1187 unsigned long start, unsigned long end)
1188{
Christoph Lameter23016962008-04-28 02:12:42 -07001189 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1190 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001191}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001192EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001193
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001194struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1195 unsigned long start, unsigned long end,
1196 void *caller)
1197{
1198 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1199 caller);
1200}
1201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202/**
Simon Arlott183ff222007-10-20 01:27:18 +02001203 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 * @size: size of the area
1205 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1206 *
1207 * Search an area of @size in the kernel virtual mapping area,
1208 * and reserved it for out purposes. Returns the area descriptor
1209 * on success or %NULL on failure.
1210 */
1211struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1212{
Christoph Lameter23016962008-04-28 02:12:42 -07001213 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1214 -1, GFP_KERNEL, __builtin_return_address(0));
1215}
1216
1217struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1218 void *caller)
1219{
1220 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1221 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001224struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1225 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07001226{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001227 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
Christoph Lameter23016962008-04-28 02:12:42 -07001228 gfp_mask, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001229}
1230
Nick Piggindb64fe02008-10-18 20:27:03 -07001231static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001232{
Nick Piggindb64fe02008-10-18 20:27:03 -07001233 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001234
Nick Piggindb64fe02008-10-18 20:27:03 -07001235 va = find_vmap_area((unsigned long)addr);
1236 if (va && va->flags & VM_VM_AREA)
1237 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001238
Andi Kleen7856dfe2005-05-20 14:27:57 -07001239 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001240}
1241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242/**
Simon Arlott183ff222007-10-20 01:27:18 +02001243 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 * @addr: base address
1245 *
1246 * Search for the kernel VM area starting at @addr, and remove it.
1247 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001248 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001250struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
Nick Piggindb64fe02008-10-18 20:27:03 -07001252 struct vmap_area *va;
1253
1254 va = find_vmap_area((unsigned long)addr);
1255 if (va && va->flags & VM_VM_AREA) {
1256 struct vm_struct *vm = va->private;
1257 struct vm_struct *tmp, **p;
Nick Piggincd528582009-01-06 14:39:20 -08001258
1259 vmap_debug_free_range(va->va_start, va->va_end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001260 free_unmap_vmap_area(va);
1261 vm->size -= PAGE_SIZE;
1262
1263 write_lock(&vmlist_lock);
1264 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1265 ;
1266 *p = tmp->next;
1267 write_unlock(&vmlist_lock);
1268
1269 return vm;
1270 }
1271 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272}
1273
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001274static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275{
1276 struct vm_struct *area;
1277
1278 if (!addr)
1279 return;
1280
1281 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001282 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 return;
1284 }
1285
1286 area = remove_vm_area(addr);
1287 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001288 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 return;
1291 }
1292
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001293 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001294 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 if (deallocate_pages) {
1297 int i;
1298
1299 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001300 struct page *page = area->pages[i];
1301
1302 BUG_ON(!page);
1303 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001306 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 vfree(area->pages);
1308 else
1309 kfree(area->pages);
1310 }
1311
1312 kfree(area);
1313 return;
1314}
1315
1316/**
1317 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * @addr: memory base address
1319 *
Simon Arlott183ff222007-10-20 01:27:18 +02001320 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001321 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1322 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001324 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001326void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
1328 BUG_ON(in_interrupt());
1329 __vunmap(addr, 1);
1330}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331EXPORT_SYMBOL(vfree);
1332
1333/**
1334 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * @addr: memory base address
1336 *
1337 * Free the virtually contiguous memory area starting at @addr,
1338 * which was created from the page array passed to vmap().
1339 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001340 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001342void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
1344 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001345 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 __vunmap(addr, 0);
1347}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348EXPORT_SYMBOL(vunmap);
1349
1350/**
1351 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 * @pages: array of page pointers
1353 * @count: number of pages to map
1354 * @flags: vm_area->flags
1355 * @prot: page protection for the mapping
1356 *
1357 * Maps @count pages from @pages into contiguous kernel virtual
1358 * space.
1359 */
1360void *vmap(struct page **pages, unsigned int count,
1361 unsigned long flags, pgprot_t prot)
1362{
1363 struct vm_struct *area;
1364
Peter Zijlstra34754b62009-02-25 16:04:03 +01001365 might_sleep();
1366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (count > num_physpages)
1368 return NULL;
1369
Christoph Lameter23016962008-04-28 02:12:42 -07001370 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1371 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (!area)
1373 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 if (map_vm_area(area, prot, &pages)) {
1376 vunmap(area->addr);
1377 return NULL;
1378 }
1379
1380 return area->addr;
1381}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382EXPORT_SYMBOL(vmap);
1383
Nick Piggindb64fe02008-10-18 20:27:03 -07001384static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1385 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001386static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001387 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 struct page **pages;
1390 unsigned int nr_pages, array_size, i;
1391
1392 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1393 array_size = (nr_pages * sizeof(struct page *));
1394
1395 area->nr_pages = nr_pages;
1396 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001397 if (array_size > PAGE_SIZE) {
Christoph Lameter94f60302007-07-17 04:03:29 -07001398 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
Christoph Lameter23016962008-04-28 02:12:42 -07001399 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001400 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001401 } else {
1402 pages = kmalloc_node(array_size,
Christoph Lameter6cb06222007-10-16 01:25:41 -07001403 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
Andrew Morton286e1ea2006-10-17 00:09:57 -07001404 node);
1405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001407 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if (!area->pages) {
1409 remove_vm_area(area->addr);
1410 kfree(area);
1411 return NULL;
1412 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001415 struct page *page;
1416
Christoph Lameter930fc452005-10-29 18:15:41 -07001417 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001418 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001419 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001420 page = alloc_pages_node(node, gfp_mask, 0);
1421
1422 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 /* Successfully allocated i pages, free them in __vunmap() */
1424 area->nr_pages = i;
1425 goto fail;
1426 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001427 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 }
1429
1430 if (map_vm_area(area, prot, &pages))
1431 goto fail;
1432 return area->addr;
1433
1434fail:
1435 vfree(area->addr);
1436 return NULL;
1437}
1438
Christoph Lameter930fc452005-10-29 18:15:41 -07001439void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1440{
Christoph Lameter23016962008-04-28 02:12:42 -07001441 return __vmalloc_area_node(area, gfp_mask, prot, -1,
1442 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001443}
1444
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001446 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 * @size: allocation size
1448 * @gfp_mask: flags for the page level allocator
1449 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001450 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001451 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 *
1453 * Allocate enough pages to cover @size from the page level
1454 * allocator with @gfp_mask flags. Map them into contiguous
1455 * kernel virtual space, using a pagetable protection of @prot.
1456 */
Adrian Bunkb2213852006-09-25 23:31:02 -07001457static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
Christoph Lameter23016962008-04-28 02:12:42 -07001458 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459{
1460 struct vm_struct *area;
1461
1462 size = PAGE_ALIGN(size);
1463 if (!size || (size >> PAGE_SHIFT) > num_physpages)
1464 return NULL;
1465
Christoph Lameter23016962008-04-28 02:12:42 -07001466 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1467 node, gfp_mask, caller);
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (!area)
1470 return NULL;
1471
Christoph Lameter23016962008-04-28 02:12:42 -07001472 return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
Christoph Lameter930fc452005-10-29 18:15:41 -07001475void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1476{
Christoph Lameter23016962008-04-28 02:12:42 -07001477 return __vmalloc_node(size, gfp_mask, prot, -1,
1478 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001479}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480EXPORT_SYMBOL(__vmalloc);
1481
1482/**
1483 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 * Allocate enough pages to cover @size from the page level
1486 * allocator and map them into contiguous kernel virtual space.
1487 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001488 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 * use __vmalloc() instead.
1490 */
1491void *vmalloc(unsigned long size)
1492{
Christoph Lameter23016962008-04-28 02:12:42 -07001493 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1494 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496EXPORT_SYMBOL(vmalloc);
1497
Christoph Lameter930fc452005-10-29 18:15:41 -07001498/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001499 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1500 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001501 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001502 * The resulting memory area is zeroed so it can be mapped to userspace
1503 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001504 */
1505void *vmalloc_user(unsigned long size)
1506{
1507 struct vm_struct *area;
1508 void *ret;
1509
Glauber Costa84877842009-01-06 14:39:19 -08001510 ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1511 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001512 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001513 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001514 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001515 }
Nick Piggin83342312006-06-23 02:03:20 -07001516 return ret;
1517}
1518EXPORT_SYMBOL(vmalloc_user);
1519
1520/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001521 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001522 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001523 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001524 *
1525 * Allocate enough pages to cover @size from the page level
1526 * allocator and map them into contiguous kernel virtual space.
1527 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001528 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001529 * use __vmalloc() instead.
1530 */
1531void *vmalloc_node(unsigned long size, int node)
1532{
Christoph Lameter23016962008-04-28 02:12:42 -07001533 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1534 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001535}
1536EXPORT_SYMBOL(vmalloc_node);
1537
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001538#ifndef PAGE_KERNEL_EXEC
1539# define PAGE_KERNEL_EXEC PAGE_KERNEL
1540#endif
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542/**
1543 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 * @size: allocation size
1545 *
1546 * Kernel-internal function to allocate enough pages to cover @size
1547 * the page level allocator and map them into contiguous and
1548 * executable kernel virtual space.
1549 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001550 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 * use __vmalloc() instead.
1552 */
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554void *vmalloc_exec(unsigned long size)
1555{
Glauber Costa84877842009-01-06 14:39:19 -08001556 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1557 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
1559
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001560#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001561#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001562#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001563#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001564#else
1565#define GFP_VMALLOC32 GFP_KERNEL
1566#endif
1567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568/**
1569 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 * @size: allocation size
1571 *
1572 * Allocate enough 32bit PA addressable pages to cover @size from the
1573 * page level allocator and map them into contiguous kernel virtual space.
1574 */
1575void *vmalloc_32(unsigned long size)
1576{
Glauber Costa84877842009-01-06 14:39:19 -08001577 return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1578 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580EXPORT_SYMBOL(vmalloc_32);
1581
Nick Piggin83342312006-06-23 02:03:20 -07001582/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001583 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001584 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001585 *
1586 * The resulting memory area is 32bit addressable and zeroed so it can be
1587 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001588 */
1589void *vmalloc_32_user(unsigned long size)
1590{
1591 struct vm_struct *area;
1592 void *ret;
1593
Glauber Costa84877842009-01-06 14:39:19 -08001594 ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1595 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001596 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001597 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001598 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001599 }
Nick Piggin83342312006-06-23 02:03:20 -07001600 return ret;
1601}
1602EXPORT_SYMBOL(vmalloc_32_user);
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604long vread(char *buf, char *addr, unsigned long count)
1605{
1606 struct vm_struct *tmp;
1607 char *vaddr, *buf_start = buf;
1608 unsigned long n;
1609
1610 /* Don't allow overflow */
1611 if ((unsigned long) addr + count < count)
1612 count = -(unsigned long) addr;
1613
1614 read_lock(&vmlist_lock);
1615 for (tmp = vmlist; tmp; tmp = tmp->next) {
1616 vaddr = (char *) tmp->addr;
1617 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1618 continue;
1619 while (addr < vaddr) {
1620 if (count == 0)
1621 goto finished;
1622 *buf = '\0';
1623 buf++;
1624 addr++;
1625 count--;
1626 }
1627 n = vaddr + tmp->size - PAGE_SIZE - addr;
1628 do {
1629 if (count == 0)
1630 goto finished;
1631 *buf = *addr;
1632 buf++;
1633 addr++;
1634 count--;
1635 } while (--n > 0);
1636 }
1637finished:
1638 read_unlock(&vmlist_lock);
1639 return buf - buf_start;
1640}
1641
1642long vwrite(char *buf, char *addr, unsigned long count)
1643{
1644 struct vm_struct *tmp;
1645 char *vaddr, *buf_start = buf;
1646 unsigned long n;
1647
1648 /* Don't allow overflow */
1649 if ((unsigned long) addr + count < count)
1650 count = -(unsigned long) addr;
1651
1652 read_lock(&vmlist_lock);
1653 for (tmp = vmlist; tmp; tmp = tmp->next) {
1654 vaddr = (char *) tmp->addr;
1655 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1656 continue;
1657 while (addr < vaddr) {
1658 if (count == 0)
1659 goto finished;
1660 buf++;
1661 addr++;
1662 count--;
1663 }
1664 n = vaddr + tmp->size - PAGE_SIZE - addr;
1665 do {
1666 if (count == 0)
1667 goto finished;
1668 *addr = *buf;
1669 buf++;
1670 addr++;
1671 count--;
1672 } while (--n > 0);
1673 }
1674finished:
1675 read_unlock(&vmlist_lock);
1676 return buf - buf_start;
1677}
Nick Piggin83342312006-06-23 02:03:20 -07001678
1679/**
1680 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07001681 * @vma: vma to cover (map full range of vma)
1682 * @addr: vmalloc memory
1683 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07001684 *
1685 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07001686 *
1687 * This function checks that addr is a valid vmalloc'ed area, and
1688 * that it is big enough to cover the vma. Will return failure if
1689 * that criteria isn't met.
1690 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001691 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07001692 */
1693int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1694 unsigned long pgoff)
1695{
1696 struct vm_struct *area;
1697 unsigned long uaddr = vma->vm_start;
1698 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07001699
1700 if ((PAGE_SIZE-1) & (unsigned long)addr)
1701 return -EINVAL;
1702
Nick Piggindb64fe02008-10-18 20:27:03 -07001703 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07001704 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07001705 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001706
1707 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07001708 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001709
1710 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07001711 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001712
1713 addr += pgoff << PAGE_SHIFT;
1714 do {
1715 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07001716 int ret;
1717
Nick Piggin83342312006-06-23 02:03:20 -07001718 ret = vm_insert_page(vma, uaddr, page);
1719 if (ret)
1720 return ret;
1721
1722 uaddr += PAGE_SIZE;
1723 addr += PAGE_SIZE;
1724 usize -= PAGE_SIZE;
1725 } while (usize > 0);
1726
1727 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1728 vma->vm_flags |= VM_RESERVED;
1729
Nick Piggindb64fe02008-10-18 20:27:03 -07001730 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07001731}
1732EXPORT_SYMBOL(remap_vmalloc_range);
1733
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07001734/*
1735 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1736 * have one.
1737 */
1738void __attribute__((weak)) vmalloc_sync_all(void)
1739{
1740}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001741
1742
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001743static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001744{
1745 /* apply_to_page_range() does all the hard work. */
1746 return 0;
1747}
1748
1749/**
1750 * alloc_vm_area - allocate a range of kernel address space
1751 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07001752 *
1753 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001754 *
1755 * This function reserves a range of kernel address space, and
1756 * allocates pagetables to map that range. No actual mappings
1757 * are created. If the kernel address space is not shared
1758 * between processes, it syncs the pagetable across all
1759 * processes.
1760 */
1761struct vm_struct *alloc_vm_area(size_t size)
1762{
1763 struct vm_struct *area;
1764
Christoph Lameter23016962008-04-28 02:12:42 -07001765 area = get_vm_area_caller(size, VM_IOREMAP,
1766 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001767 if (area == NULL)
1768 return NULL;
1769
1770 /*
1771 * This ensures that page tables are constructed for this region
1772 * of kernel virtual address space and mapped into init_mm.
1773 */
1774 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1775 area->size, f, NULL)) {
1776 free_vm_area(area);
1777 return NULL;
1778 }
1779
1780 /* Make sure the pagetables are constructed in process kernel
1781 mappings */
1782 vmalloc_sync_all();
1783
1784 return area;
1785}
1786EXPORT_SYMBOL_GPL(alloc_vm_area);
1787
1788void free_vm_area(struct vm_struct *area)
1789{
1790 struct vm_struct *ret;
1791 ret = remove_vm_area(area->addr);
1792 BUG_ON(ret != area);
1793 kfree(area);
1794}
1795EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07001796
1797
1798#ifdef CONFIG_PROC_FS
1799static void *s_start(struct seq_file *m, loff_t *pos)
1800{
1801 loff_t n = *pos;
1802 struct vm_struct *v;
1803
1804 read_lock(&vmlist_lock);
1805 v = vmlist;
1806 while (n > 0 && v) {
1807 n--;
1808 v = v->next;
1809 }
1810 if (!n)
1811 return v;
1812
1813 return NULL;
1814
1815}
1816
1817static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1818{
1819 struct vm_struct *v = p;
1820
1821 ++*pos;
1822 return v->next;
1823}
1824
1825static void s_stop(struct seq_file *m, void *p)
1826{
1827 read_unlock(&vmlist_lock);
1828}
1829
Eric Dumazeta47a1262008-07-23 21:27:38 -07001830static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1831{
1832 if (NUMA_BUILD) {
1833 unsigned int nr, *counters = m->private;
1834
1835 if (!counters)
1836 return;
1837
1838 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1839
1840 for (nr = 0; nr < v->nr_pages; nr++)
1841 counters[page_to_nid(v->pages[nr])]++;
1842
1843 for_each_node_state(nr, N_HIGH_MEMORY)
1844 if (counters[nr])
1845 seq_printf(m, " N%u=%u", nr, counters[nr]);
1846 }
1847}
1848
Christoph Lametera10aa572008-04-28 02:12:40 -07001849static int s_show(struct seq_file *m, void *p)
1850{
1851 struct vm_struct *v = p;
1852
1853 seq_printf(m, "0x%p-0x%p %7ld",
1854 v->addr, v->addr + v->size, v->size);
1855
Christoph Lameter23016962008-04-28 02:12:42 -07001856 if (v->caller) {
Hugh Dickins9c246242008-12-09 13:14:27 -08001857 char buff[KSYM_SYMBOL_LEN];
Christoph Lameter23016962008-04-28 02:12:42 -07001858
1859 seq_putc(m, ' ');
1860 sprint_symbol(buff, (unsigned long)v->caller);
1861 seq_puts(m, buff);
1862 }
1863
Christoph Lametera10aa572008-04-28 02:12:40 -07001864 if (v->nr_pages)
1865 seq_printf(m, " pages=%d", v->nr_pages);
1866
1867 if (v->phys_addr)
1868 seq_printf(m, " phys=%lx", v->phys_addr);
1869
1870 if (v->flags & VM_IOREMAP)
1871 seq_printf(m, " ioremap");
1872
1873 if (v->flags & VM_ALLOC)
1874 seq_printf(m, " vmalloc");
1875
1876 if (v->flags & VM_MAP)
1877 seq_printf(m, " vmap");
1878
1879 if (v->flags & VM_USERMAP)
1880 seq_printf(m, " user");
1881
1882 if (v->flags & VM_VPAGES)
1883 seq_printf(m, " vpages");
1884
Eric Dumazeta47a1262008-07-23 21:27:38 -07001885 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07001886 seq_putc(m, '\n');
1887 return 0;
1888}
1889
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001890static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07001891 .start = s_start,
1892 .next = s_next,
1893 .stop = s_stop,
1894 .show = s_show,
1895};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04001896
1897static int vmalloc_open(struct inode *inode, struct file *file)
1898{
1899 unsigned int *ptr = NULL;
1900 int ret;
1901
1902 if (NUMA_BUILD)
1903 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1904 ret = seq_open(file, &vmalloc_op);
1905 if (!ret) {
1906 struct seq_file *m = file->private_data;
1907 m->private = ptr;
1908 } else
1909 kfree(ptr);
1910 return ret;
1911}
1912
1913static const struct file_operations proc_vmalloc_operations = {
1914 .open = vmalloc_open,
1915 .read = seq_read,
1916 .llseek = seq_lseek,
1917 .release = seq_release_private,
1918};
1919
1920static int __init proc_vmalloc_init(void)
1921{
1922 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1923 return 0;
1924}
1925module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07001926#endif
1927