blob: eb5cc7d00c5a7c0443f9ff663317cf229ab74353 [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>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040015#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/spinlock.h>
18#include <linux/interrupt.h>
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +040019#include <linux/proc_fs.h>
Christoph Lametera10aa572008-04-28 02:12:40 -070020#include <linux/seq_file.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070021#include <linux/debugobjects.h>
Christoph Lameter23016962008-04-28 02:12:42 -070022#include <linux/kallsyms.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070023#include <linux/list.h>
24#include <linux/rbtree.h>
25#include <linux/radix-tree.h>
26#include <linux/rcupdate.h>
Tejun Heof0aa6612009-02-20 16:29:08 +090027#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010028#include <linux/kmemleak.h>
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>
David Miller2dca6992009-09-21 12:22:34 -070032#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
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)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700171 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700173
Nick Piggindb64fe02008-10-18 20:27:03 -0700174 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
Tejun Heo8fc48982009-02-20 16:29:08 +0900177static int vmap_page_range(unsigned long start, unsigned long end,
178 pgprot_t prot, struct page **pages)
179{
180 int ret;
181
182 ret = vmap_page_range_noflush(start, end, prot, pages);
183 flush_cache_vmap(start, end);
184 return ret;
185}
186
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700187int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700188{
189 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000190 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700191 * and fall back on vmalloc() if that fails. Others
192 * just put it in the vmalloc space.
193 */
194#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
195 unsigned long addr = (unsigned long)x;
196 if (addr >= MODULES_VADDR && addr < MODULES_END)
197 return 1;
198#endif
199 return is_vmalloc_addr(x);
200}
201
Christoph Lameter48667e72008-02-04 22:28:31 -0800202/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700203 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800204 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800205struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800206{
207 unsigned long addr = (unsigned long) vmalloc_addr;
208 struct page *page = NULL;
209 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800210
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200211 /*
212 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
213 * architectures that do not vmalloc module space
214 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700215 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200216
Christoph Lameter48667e72008-02-04 22:28:31 -0800217 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700218 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800219 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700220 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800221 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700222 pte_t *ptep, pte;
223
Christoph Lameter48667e72008-02-04 22:28:31 -0800224 ptep = pte_offset_map(pmd, addr);
225 pte = *ptep;
226 if (pte_present(pte))
227 page = pte_page(pte);
228 pte_unmap(ptep);
229 }
230 }
231 }
232 return page;
233}
234EXPORT_SYMBOL(vmalloc_to_page);
235
236/*
237 * Map a vmalloc()-space virtual address to the physical page frame number.
238 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800239unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800240{
241 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
242}
243EXPORT_SYMBOL(vmalloc_to_pfn);
244
Nick Piggindb64fe02008-10-18 20:27:03 -0700245
246/*** Global kva allocator ***/
247
248#define VM_LAZY_FREE 0x01
249#define VM_LAZY_FREEING 0x02
250#define VM_VM_AREA 0x04
251
252struct vmap_area {
253 unsigned long va_start;
254 unsigned long va_end;
255 unsigned long flags;
256 struct rb_node rb_node; /* address sorted rbtree */
257 struct list_head list; /* address sorted list */
258 struct list_head purge_list; /* "lazy purge" list */
259 void *private;
260 struct rcu_head rcu_head;
261};
262
263static DEFINE_SPINLOCK(vmap_area_lock);
264static struct rb_root vmap_area_root = RB_ROOT;
265static LIST_HEAD(vmap_area_list);
Tejun Heoca23e402009-08-14 15:00:52 +0900266static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700267
268static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Nick Piggindb64fe02008-10-18 20:27:03 -0700270 struct rb_node *n = vmap_area_root.rb_node;
271
272 while (n) {
273 struct vmap_area *va;
274
275 va = rb_entry(n, struct vmap_area, rb_node);
276 if (addr < va->va_start)
277 n = n->rb_left;
278 else if (addr > va->va_start)
279 n = n->rb_right;
280 else
281 return va;
282 }
283
284 return NULL;
285}
286
287static void __insert_vmap_area(struct vmap_area *va)
288{
289 struct rb_node **p = &vmap_area_root.rb_node;
290 struct rb_node *parent = NULL;
291 struct rb_node *tmp;
292
293 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700294 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700295
296 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700297 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
298 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700299 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700300 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700301 p = &(*p)->rb_right;
302 else
303 BUG();
304 }
305
306 rb_link_node(&va->rb_node, parent, p);
307 rb_insert_color(&va->rb_node, &vmap_area_root);
308
309 /* address-sort this list so it is usable like the vmlist */
310 tmp = rb_prev(&va->rb_node);
311 if (tmp) {
312 struct vmap_area *prev;
313 prev = rb_entry(tmp, struct vmap_area, rb_node);
314 list_add_rcu(&va->list, &prev->list);
315 } else
316 list_add_rcu(&va->list, &vmap_area_list);
317}
318
319static void purge_vmap_area_lazy(void);
320
321/*
322 * Allocate a region of KVA of the specified size and alignment, within the
323 * vstart and vend.
324 */
325static struct vmap_area *alloc_vmap_area(unsigned long size,
326 unsigned long align,
327 unsigned long vstart, unsigned long vend,
328 int node, gfp_t gfp_mask)
329{
330 struct vmap_area *va;
331 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700333 int purged = 0;
334
Nick Piggin77669702009-02-27 14:03:03 -0800335 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700336 BUG_ON(size & ~PAGE_MASK);
337
Nick Piggindb64fe02008-10-18 20:27:03 -0700338 va = kmalloc_node(sizeof(struct vmap_area),
339 gfp_mask & GFP_RECLAIM_MASK, node);
340 if (unlikely(!va))
341 return ERR_PTR(-ENOMEM);
342
343retry:
Glauber Costa0ae15132008-11-19 15:36:33 -0800344 addr = ALIGN(vstart, align);
345
Nick Piggindb64fe02008-10-18 20:27:03 -0700346 spin_lock(&vmap_area_lock);
Nick Piggin77669702009-02-27 14:03:03 -0800347 if (addr + size - 1 < addr)
348 goto overflow;
349
Nick Piggindb64fe02008-10-18 20:27:03 -0700350 /* XXX: could have a last_hole cache */
351 n = vmap_area_root.rb_node;
352 if (n) {
353 struct vmap_area *first = NULL;
354
355 do {
356 struct vmap_area *tmp;
357 tmp = rb_entry(n, struct vmap_area, rb_node);
358 if (tmp->va_end >= addr) {
359 if (!first && tmp->va_start < addr + size)
360 first = tmp;
361 n = n->rb_left;
362 } else {
363 first = tmp;
364 n = n->rb_right;
365 }
366 } while (n);
367
368 if (!first)
369 goto found;
370
371 if (first->va_end < addr) {
372 n = rb_next(&first->rb_node);
373 if (n)
374 first = rb_entry(n, struct vmap_area, rb_node);
375 else
376 goto found;
377 }
378
Nick Pigginf011c2d2008-11-19 15:36:32 -0800379 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700380 addr = ALIGN(first->va_end + PAGE_SIZE, align);
Nick Piggin77669702009-02-27 14:03:03 -0800381 if (addr + size - 1 < addr)
382 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700383
384 n = rb_next(&first->rb_node);
385 if (n)
386 first = rb_entry(n, struct vmap_area, rb_node);
387 else
388 goto found;
389 }
390 }
391found:
392 if (addr + size > vend) {
Nick Piggin77669702009-02-27 14:03:03 -0800393overflow:
Nick Piggindb64fe02008-10-18 20:27:03 -0700394 spin_unlock(&vmap_area_lock);
395 if (!purged) {
396 purge_vmap_area_lazy();
397 purged = 1;
398 goto retry;
399 }
400 if (printk_ratelimit())
Glauber Costac1279c42009-01-06 14:39:18 -0800401 printk(KERN_WARNING
402 "vmap allocation for size %lu failed: "
403 "use vmalloc=<size> to increase size.\n", size);
Ralph Wuerthner2498ce42009-05-06 16:02:59 -0700404 kfree(va);
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
Tejun Heoca23e402009-08-14 15:00:52 +0900433 /*
434 * Track the highest possible candidate for pcpu area
435 * allocation. Areas outside of vmalloc area can be returned
436 * here too, consider only end addresses which fall inside
437 * vmalloc area proper.
438 */
439 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
440 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
441
Nick Piggindb64fe02008-10-18 20:27:03 -0700442 call_rcu(&va->rcu_head, rcu_free_va);
443}
444
445/*
446 * Free a region of KVA allocated by alloc_vmap_area
447 */
448static void free_vmap_area(struct vmap_area *va)
449{
450 spin_lock(&vmap_area_lock);
451 __free_vmap_area(va);
452 spin_unlock(&vmap_area_lock);
453}
454
455/*
456 * Clear the pagetable entries of a given vmap_area
457 */
458static void unmap_vmap_area(struct vmap_area *va)
459{
460 vunmap_page_range(va->va_start, va->va_end);
461}
462
Nick Piggincd528582009-01-06 14:39:20 -0800463static void vmap_debug_free_range(unsigned long start, unsigned long end)
464{
465 /*
466 * Unmap page tables and force a TLB flush immediately if
467 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
468 * bugs similarly to those in linear kernel virtual address
469 * space after a page has been freed.
470 *
471 * All the lazy freeing logic is still retained, in order to
472 * minimise intrusiveness of this debugging feature.
473 *
474 * This is going to be *slow* (linear kernel virtual address
475 * debugging doesn't do a broadcast TLB flush so it is a lot
476 * faster).
477 */
478#ifdef CONFIG_DEBUG_PAGEALLOC
479 vunmap_page_range(start, end);
480 flush_tlb_kernel_range(start, end);
481#endif
482}
483
Nick Piggindb64fe02008-10-18 20:27:03 -0700484/*
485 * lazy_max_pages is the maximum amount of virtual address space we gather up
486 * before attempting to purge with a TLB flush.
487 *
488 * There is a tradeoff here: a larger number will cover more kernel page tables
489 * and take slightly longer to purge, but it will linearly reduce the number of
490 * global TLB flushes that must be performed. It would seem natural to scale
491 * this number up linearly with the number of CPUs (because vmapping activity
492 * could also scale linearly with the number of CPUs), however it is likely
493 * that in practice, workloads might be constrained in other ways that mean
494 * vmap activity will not scale linearly with CPUs. Also, I want to be
495 * conservative and not introduce a big latency on huge systems, so go with
496 * a less aggressive log scale. It will still be an improvement over the old
497 * code, and it will be simple to change the scale factor if we find that it
498 * becomes a problem on bigger systems.
499 */
500static unsigned long lazy_max_pages(void)
501{
502 unsigned int log;
503
504 log = fls(num_online_cpus());
505
506 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
507}
508
509static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
510
Nick Piggin02b709d2010-02-01 22:25:57 +1100511/* for per-CPU blocks */
512static void purge_fragmented_blocks_allcpus(void);
513
Nick Piggindb64fe02008-10-18 20:27:03 -0700514/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500515 * called before a call to iounmap() if the caller wants vm_area_struct's
516 * immediately freed.
517 */
518void set_iounmap_nonlazy(void)
519{
520 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
521}
522
523/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700524 * Purges all lazily-freed vmap areas.
525 *
526 * If sync is 0 then don't purge if there is already a purge in progress.
527 * If force_flush is 1, then flush kernel TLBs between *start and *end even
528 * if we found no lazy vmap areas to unmap (callers can use this to optimise
529 * their own TLB flushing).
530 * Returns with *start = min(*start, lowest purged address)
531 * *end = max(*end, highest purged address)
532 */
533static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
534 int sync, int force_flush)
535{
Andrew Morton46666d82009-01-15 13:51:15 -0800536 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700537 LIST_HEAD(valist);
538 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800539 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700540 int nr = 0;
541
542 /*
543 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
544 * should not expect such behaviour. This just simplifies locking for
545 * the case that isn't actually used at the moment anyway.
546 */
547 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800548 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700549 return;
550 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800551 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700552
Nick Piggin02b709d2010-02-01 22:25:57 +1100553 if (sync)
554 purge_fragmented_blocks_allcpus();
555
Nick Piggindb64fe02008-10-18 20:27:03 -0700556 rcu_read_lock();
557 list_for_each_entry_rcu(va, &vmap_area_list, list) {
558 if (va->flags & VM_LAZY_FREE) {
559 if (va->va_start < *start)
560 *start = va->va_start;
561 if (va->va_end > *end)
562 *end = va->va_end;
563 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700564 list_add_tail(&va->purge_list, &valist);
565 va->flags |= VM_LAZY_FREEING;
566 va->flags &= ~VM_LAZY_FREE;
567 }
568 }
569 rcu_read_unlock();
570
Yongseok Koh88f50042010-01-19 17:33:49 +0900571 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700572 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700573
574 if (nr || force_flush)
575 flush_tlb_kernel_range(*start, *end);
576
577 if (nr) {
578 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800579 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700580 __free_vmap_area(va);
581 spin_unlock(&vmap_area_lock);
582 }
Andrew Morton46666d82009-01-15 13:51:15 -0800583 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700584}
585
586/*
Nick Piggin496850e2008-11-19 15:36:33 -0800587 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
588 * is already purging.
589 */
590static void try_purge_vmap_area_lazy(void)
591{
592 unsigned long start = ULONG_MAX, end = 0;
593
594 __purge_vmap_area_lazy(&start, &end, 0, 0);
595}
596
597/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700598 * Kick off a purge of the outstanding lazy areas.
599 */
600static void purge_vmap_area_lazy(void)
601{
602 unsigned long start = ULONG_MAX, end = 0;
603
Nick Piggin496850e2008-11-19 15:36:33 -0800604 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700605}
606
607/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800608 * Free a vmap area, caller ensuring that the area has been unmapped
609 * and flush_cache_vunmap had been called for the correct range
610 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700611 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800612static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700613{
614 va->flags |= VM_LAZY_FREE;
615 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
616 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800617 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700618}
619
Nick Pigginb29acbd2008-12-01 13:13:47 -0800620/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800621 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
622 * called for the correct range previously.
623 */
624static void free_unmap_vmap_area_noflush(struct vmap_area *va)
625{
626 unmap_vmap_area(va);
627 free_vmap_area_noflush(va);
628}
629
630/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800631 * Free and unmap a vmap area
632 */
633static void free_unmap_vmap_area(struct vmap_area *va)
634{
635 flush_cache_vunmap(va->va_start, va->va_end);
636 free_unmap_vmap_area_noflush(va);
637}
638
Nick Piggindb64fe02008-10-18 20:27:03 -0700639static struct vmap_area *find_vmap_area(unsigned long addr)
640{
641 struct vmap_area *va;
642
643 spin_lock(&vmap_area_lock);
644 va = __find_vmap_area(addr);
645 spin_unlock(&vmap_area_lock);
646
647 return va;
648}
649
650static void free_unmap_vmap_area_addr(unsigned long addr)
651{
652 struct vmap_area *va;
653
654 va = find_vmap_area(addr);
655 BUG_ON(!va);
656 free_unmap_vmap_area(va);
657}
658
659
660/*** Per cpu kva allocator ***/
661
662/*
663 * vmap space is limited especially on 32 bit architectures. Ensure there is
664 * room for at least 16 percpu vmap blocks per CPU.
665 */
666/*
667 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
668 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
669 * instead (we just need a rough idea)
670 */
671#if BITS_PER_LONG == 32
672#define VMALLOC_SPACE (128UL*1024*1024)
673#else
674#define VMALLOC_SPACE (128UL*1024*1024*1024)
675#endif
676
677#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
678#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
679#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
680#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
681#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
682#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
683#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
684 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
685 VMALLOC_PAGES / NR_CPUS / 16))
686
687#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
688
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100689static bool vmap_initialized __read_mostly = false;
690
Nick Piggindb64fe02008-10-18 20:27:03 -0700691struct vmap_block_queue {
692 spinlock_t lock;
693 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700694};
695
696struct vmap_block {
697 spinlock_t lock;
698 struct vmap_area *va;
699 struct vmap_block_queue *vbq;
700 unsigned long free, dirty;
701 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
702 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100703 struct list_head free_list;
704 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100705 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700706};
707
708/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
709static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
710
711/*
712 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
713 * in the free path. Could get rid of this if we change the API to return a
714 * "cookie" from alloc, to be passed to free. But no big deal yet.
715 */
716static DEFINE_SPINLOCK(vmap_block_tree_lock);
717static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
718
719/*
720 * We should probably have a fallback mechanism to allocate virtual memory
721 * out of partially filled vmap blocks. However vmap block sizing should be
722 * fairly reasonable according to the vmalloc size, so it shouldn't be a
723 * big problem.
724 */
725
726static unsigned long addr_to_vb_idx(unsigned long addr)
727{
728 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
729 addr /= VMAP_BLOCK_SIZE;
730 return addr;
731}
732
733static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
734{
735 struct vmap_block_queue *vbq;
736 struct vmap_block *vb;
737 struct vmap_area *va;
738 unsigned long vb_idx;
739 int node, err;
740
741 node = numa_node_id();
742
743 vb = kmalloc_node(sizeof(struct vmap_block),
744 gfp_mask & GFP_RECLAIM_MASK, node);
745 if (unlikely(!vb))
746 return ERR_PTR(-ENOMEM);
747
748 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
749 VMALLOC_START, VMALLOC_END,
750 node, gfp_mask);
751 if (unlikely(IS_ERR(va))) {
752 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700753 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700754 }
755
756 err = radix_tree_preload(gfp_mask);
757 if (unlikely(err)) {
758 kfree(vb);
759 free_vmap_area(va);
760 return ERR_PTR(err);
761 }
762
763 spin_lock_init(&vb->lock);
764 vb->va = va;
765 vb->free = VMAP_BBMAP_BITS;
766 vb->dirty = 0;
767 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
768 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
769 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700770
771 vb_idx = addr_to_vb_idx(va->va_start);
772 spin_lock(&vmap_block_tree_lock);
773 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
774 spin_unlock(&vmap_block_tree_lock);
775 BUG_ON(err);
776 radix_tree_preload_end();
777
778 vbq = &get_cpu_var(vmap_block_queue);
779 vb->vbq = vbq;
780 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100781 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700782 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900783 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700784
785 return vb;
786}
787
788static void rcu_free_vb(struct rcu_head *head)
789{
790 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
791
792 kfree(vb);
793}
794
795static void free_vmap_block(struct vmap_block *vb)
796{
797 struct vmap_block *tmp;
798 unsigned long vb_idx;
799
Nick Piggindb64fe02008-10-18 20:27:03 -0700800 vb_idx = addr_to_vb_idx(vb->va->va_start);
801 spin_lock(&vmap_block_tree_lock);
802 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
803 spin_unlock(&vmap_block_tree_lock);
804 BUG_ON(tmp != vb);
805
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800806 free_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700807 call_rcu(&vb->rcu_head, rcu_free_vb);
808}
809
Nick Piggin02b709d2010-02-01 22:25:57 +1100810static void purge_fragmented_blocks(int cpu)
811{
812 LIST_HEAD(purge);
813 struct vmap_block *vb;
814 struct vmap_block *n_vb;
815 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
816
817 rcu_read_lock();
818 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
819
820 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
821 continue;
822
823 spin_lock(&vb->lock);
824 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
825 vb->free = 0; /* prevent further allocs after releasing lock */
826 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
827 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
828 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
829 spin_lock(&vbq->lock);
830 list_del_rcu(&vb->free_list);
831 spin_unlock(&vbq->lock);
832 spin_unlock(&vb->lock);
833 list_add_tail(&vb->purge, &purge);
834 } else
835 spin_unlock(&vb->lock);
836 }
837 rcu_read_unlock();
838
839 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
840 list_del(&vb->purge);
841 free_vmap_block(vb);
842 }
843}
844
845static void purge_fragmented_blocks_thiscpu(void)
846{
847 purge_fragmented_blocks(smp_processor_id());
848}
849
850static void purge_fragmented_blocks_allcpus(void)
851{
852 int cpu;
853
854 for_each_possible_cpu(cpu)
855 purge_fragmented_blocks(cpu);
856}
857
Nick Piggindb64fe02008-10-18 20:27:03 -0700858static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
859{
860 struct vmap_block_queue *vbq;
861 struct vmap_block *vb;
862 unsigned long addr = 0;
863 unsigned int order;
Nick Piggin02b709d2010-02-01 22:25:57 +1100864 int purge = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700865
866 BUG_ON(size & ~PAGE_MASK);
867 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
868 order = get_order(size);
869
870again:
871 rcu_read_lock();
872 vbq = &get_cpu_var(vmap_block_queue);
873 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
874 int i;
875
876 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100877 if (vb->free < 1UL << order)
878 goto next;
879
Nick Piggindb64fe02008-10-18 20:27:03 -0700880 i = bitmap_find_free_region(vb->alloc_map,
881 VMAP_BBMAP_BITS, order);
882
Nick Piggin02b709d2010-02-01 22:25:57 +1100883 if (i < 0) {
884 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
885 /* fragmented and no outstanding allocations */
886 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
887 purge = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -0700888 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100889 goto next;
890 }
891 addr = vb->va->va_start + (i << PAGE_SHIFT);
892 BUG_ON(addr_to_vb_idx(addr) !=
893 addr_to_vb_idx(vb->va->va_start));
894 vb->free -= 1UL << order;
895 if (vb->free == 0) {
896 spin_lock(&vbq->lock);
897 list_del_rcu(&vb->free_list);
898 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700899 }
900 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100901 break;
902next:
903 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700904 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100905
906 if (purge)
907 purge_fragmented_blocks_thiscpu();
908
Tejun Heo3f04ba82009-10-29 22:34:12 +0900909 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700910 rcu_read_unlock();
911
912 if (!addr) {
913 vb = new_vmap_block(gfp_mask);
914 if (IS_ERR(vb))
915 return vb;
916 goto again;
917 }
918
919 return (void *)addr;
920}
921
922static void vb_free(const void *addr, unsigned long size)
923{
924 unsigned long offset;
925 unsigned long vb_idx;
926 unsigned int order;
927 struct vmap_block *vb;
928
929 BUG_ON(size & ~PAGE_MASK);
930 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800931
932 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
933
Nick Piggindb64fe02008-10-18 20:27:03 -0700934 order = get_order(size);
935
936 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
937
938 vb_idx = addr_to_vb_idx((unsigned long)addr);
939 rcu_read_lock();
940 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
941 rcu_read_unlock();
942 BUG_ON(!vb);
943
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800944 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
945
Nick Piggindb64fe02008-10-18 20:27:03 -0700946 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100947 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700948
Nick Piggindb64fe02008-10-18 20:27:03 -0700949 vb->dirty += 1UL << order;
950 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100951 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700952 spin_unlock(&vb->lock);
953 free_vmap_block(vb);
954 } else
955 spin_unlock(&vb->lock);
956}
957
958/**
959 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
960 *
961 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
962 * to amortize TLB flushing overheads. What this means is that any page you
963 * have now, may, in a former life, have been mapped into kernel virtual
964 * address by the vmap layer and so there might be some CPUs with TLB entries
965 * still referencing that page (additional to the regular 1:1 kernel mapping).
966 *
967 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
968 * be sure that none of the pages we have control over will have any aliases
969 * from the vmap layer.
970 */
971void vm_unmap_aliases(void)
972{
973 unsigned long start = ULONG_MAX, end = 0;
974 int cpu;
975 int flush = 0;
976
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100977 if (unlikely(!vmap_initialized))
978 return;
979
Nick Piggindb64fe02008-10-18 20:27:03 -0700980 for_each_possible_cpu(cpu) {
981 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
982 struct vmap_block *vb;
983
984 rcu_read_lock();
985 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
986 int i;
987
988 spin_lock(&vb->lock);
989 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
990 while (i < VMAP_BBMAP_BITS) {
991 unsigned long s, e;
992 int j;
993 j = find_next_zero_bit(vb->dirty_map,
994 VMAP_BBMAP_BITS, i);
995
996 s = vb->va->va_start + (i << PAGE_SHIFT);
997 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -0700998 flush = 1;
999
1000 if (s < start)
1001 start = s;
1002 if (e > end)
1003 end = e;
1004
1005 i = j;
1006 i = find_next_bit(vb->dirty_map,
1007 VMAP_BBMAP_BITS, i);
1008 }
1009 spin_unlock(&vb->lock);
1010 }
1011 rcu_read_unlock();
1012 }
1013
1014 __purge_vmap_area_lazy(&start, &end, 1, flush);
1015}
1016EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1017
1018/**
1019 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1020 * @mem: the pointer returned by vm_map_ram
1021 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1022 */
1023void vm_unmap_ram(const void *mem, unsigned int count)
1024{
1025 unsigned long size = count << PAGE_SHIFT;
1026 unsigned long addr = (unsigned long)mem;
1027
1028 BUG_ON(!addr);
1029 BUG_ON(addr < VMALLOC_START);
1030 BUG_ON(addr > VMALLOC_END);
1031 BUG_ON(addr & (PAGE_SIZE-1));
1032
1033 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001034 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001035
1036 if (likely(count <= VMAP_MAX_ALLOC))
1037 vb_free(mem, size);
1038 else
1039 free_unmap_vmap_area_addr(addr);
1040}
1041EXPORT_SYMBOL(vm_unmap_ram);
1042
1043/**
1044 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1045 * @pages: an array of pointers to the pages to be mapped
1046 * @count: number of pages
1047 * @node: prefer to allocate data structures on this node
1048 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001049 *
1050 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001051 */
1052void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1053{
1054 unsigned long size = count << PAGE_SHIFT;
1055 unsigned long addr;
1056 void *mem;
1057
1058 if (likely(count <= VMAP_MAX_ALLOC)) {
1059 mem = vb_alloc(size, GFP_KERNEL);
1060 if (IS_ERR(mem))
1061 return NULL;
1062 addr = (unsigned long)mem;
1063 } else {
1064 struct vmap_area *va;
1065 va = alloc_vmap_area(size, PAGE_SIZE,
1066 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1067 if (IS_ERR(va))
1068 return NULL;
1069
1070 addr = va->va_start;
1071 mem = (void *)addr;
1072 }
1073 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1074 vm_unmap_ram(mem, count);
1075 return NULL;
1076 }
1077 return mem;
1078}
1079EXPORT_SYMBOL(vm_map_ram);
1080
Tejun Heof0aa6612009-02-20 16:29:08 +09001081/**
1082 * vm_area_register_early - register vmap area early during boot
1083 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001084 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001085 *
1086 * This function is used to register kernel vm area before
1087 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1088 * proper values on entry and other fields should be zero. On return,
1089 * vm->addr contains the allocated address.
1090 *
1091 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1092 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001093void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001094{
1095 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001096 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001097
Tejun Heoc0c0a292009-02-24 11:57:21 +09001098 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1099 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1100
1101 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001102
1103 vm->next = vmlist;
1104 vmlist = vm;
1105}
1106
Nick Piggindb64fe02008-10-18 20:27:03 -07001107void __init vmalloc_init(void)
1108{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001109 struct vmap_area *va;
1110 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001111 int i;
1112
1113 for_each_possible_cpu(i) {
1114 struct vmap_block_queue *vbq;
1115
1116 vbq = &per_cpu(vmap_block_queue, i);
1117 spin_lock_init(&vbq->lock);
1118 INIT_LIST_HEAD(&vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001119 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001120
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001121 /* Import existing vmlist entries. */
1122 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001123 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001124 va->flags = tmp->flags | VM_VM_AREA;
1125 va->va_start = (unsigned long)tmp->addr;
1126 va->va_end = va->va_start + tmp->size;
1127 __insert_vmap_area(va);
1128 }
Tejun Heoca23e402009-08-14 15:00:52 +09001129
1130 vmap_area_pcpu_hole = VMALLOC_END;
1131
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001132 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001133}
1134
Tejun Heo8fc48982009-02-20 16:29:08 +09001135/**
1136 * map_kernel_range_noflush - map kernel VM area with the specified pages
1137 * @addr: start of the VM area to map
1138 * @size: size of the VM area to map
1139 * @prot: page protection flags to use
1140 * @pages: pages to map
1141 *
1142 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1143 * specify should have been allocated using get_vm_area() and its
1144 * friends.
1145 *
1146 * NOTE:
1147 * This function does NOT do any cache flushing. The caller is
1148 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1149 * before calling this function.
1150 *
1151 * RETURNS:
1152 * The number of pages mapped on success, -errno on failure.
1153 */
1154int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1155 pgprot_t prot, struct page **pages)
1156{
1157 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1158}
1159
1160/**
1161 * unmap_kernel_range_noflush - unmap kernel VM area
1162 * @addr: start of the VM area to unmap
1163 * @size: size of the VM area to unmap
1164 *
1165 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1166 * specify should have been allocated using get_vm_area() and its
1167 * friends.
1168 *
1169 * NOTE:
1170 * This function does NOT do any cache flushing. The caller is
1171 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1172 * before calling this function and flush_tlb_kernel_range() after.
1173 */
1174void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1175{
1176 vunmap_page_range(addr, addr + size);
1177}
1178
1179/**
1180 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1181 * @addr: start of the VM area to unmap
1182 * @size: size of the VM area to unmap
1183 *
1184 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1185 * the unmapping and tlb after.
1186 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001187void unmap_kernel_range(unsigned long addr, unsigned long size)
1188{
1189 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001190
1191 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001192 vunmap_page_range(addr, end);
1193 flush_tlb_kernel_range(addr, end);
1194}
1195
1196int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1197{
1198 unsigned long addr = (unsigned long)area->addr;
1199 unsigned long end = addr + area->size - PAGE_SIZE;
1200 int err;
1201
1202 err = vmap_page_range(addr, end, prot, *pages);
1203 if (err > 0) {
1204 *pages += err;
1205 err = 0;
1206 }
1207
1208 return err;
1209}
1210EXPORT_SYMBOL_GPL(map_vm_area);
1211
1212/*** Old vmalloc interfaces ***/
1213DEFINE_RWLOCK(vmlist_lock);
1214struct vm_struct *vmlist;
1215
Tejun Heocf88c792009-08-14 15:00:52 +09001216static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1217 unsigned long flags, void *caller)
1218{
1219 struct vm_struct *tmp, **p;
1220
1221 vm->flags = flags;
1222 vm->addr = (void *)va->va_start;
1223 vm->size = va->va_end - va->va_start;
1224 vm->caller = caller;
1225 va->private = vm;
1226 va->flags |= VM_VM_AREA;
1227
1228 write_lock(&vmlist_lock);
1229 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1230 if (tmp->addr >= vm->addr)
1231 break;
1232 }
1233 vm->next = *p;
1234 *p = vm;
1235 write_unlock(&vmlist_lock);
1236}
1237
Nick Piggindb64fe02008-10-18 20:27:03 -07001238static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001239 unsigned long align, unsigned long flags, unsigned long start,
1240 unsigned long end, int node, gfp_t gfp_mask, void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001241{
1242 static struct vmap_area *va;
1243 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001245 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 if (flags & VM_IOREMAP) {
1247 int bit = fls(size);
1248
1249 if (bit > IOREMAP_MAX_ORDER)
1250 bit = IOREMAP_MAX_ORDER;
1251 else if (bit < PAGE_SHIFT)
1252 bit = PAGE_SHIFT;
1253
1254 align = 1ul << bit;
1255 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001258 if (unlikely(!size))
1259 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Tejun Heocf88c792009-08-14 15:00:52 +09001261 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 if (unlikely(!area))
1263 return NULL;
1264
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 /*
1266 * We always allocate a guard page.
1267 */
1268 size += PAGE_SIZE;
1269
Nick Piggindb64fe02008-10-18 20:27:03 -07001270 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1271 if (IS_ERR(va)) {
1272 kfree(area);
1273 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Tejun Heocf88c792009-08-14 15:00:52 +09001276 insert_vmalloc_vm(area, va, flags, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278}
1279
Christoph Lameter930fc452005-10-29 18:15:41 -07001280struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1281 unsigned long start, unsigned long end)
1282{
David Miller2dca6992009-09-21 12:22:34 -07001283 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001284 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001285}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001286EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001287
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001288struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1289 unsigned long start, unsigned long end,
1290 void *caller)
1291{
David Miller2dca6992009-09-21 12:22:34 -07001292 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001293 caller);
1294}
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296/**
Simon Arlott183ff222007-10-20 01:27:18 +02001297 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 * @size: size of the area
1299 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1300 *
1301 * Search an area of @size in the kernel virtual mapping area,
1302 * and reserved it for out purposes. Returns the area descriptor
1303 * on success or %NULL on failure.
1304 */
1305struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1306{
David Miller2dca6992009-09-21 12:22:34 -07001307 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001308 -1, GFP_KERNEL, __builtin_return_address(0));
1309}
1310
1311struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1312 void *caller)
1313{
David Miller2dca6992009-09-21 12:22:34 -07001314 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001315 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316}
1317
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001318struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1319 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07001320{
David Miller2dca6992009-09-21 12:22:34 -07001321 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1322 node, gfp_mask, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001323}
1324
Nick Piggindb64fe02008-10-18 20:27:03 -07001325static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001326{
Nick Piggindb64fe02008-10-18 20:27:03 -07001327 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001328
Nick Piggindb64fe02008-10-18 20:27:03 -07001329 va = find_vmap_area((unsigned long)addr);
1330 if (va && va->flags & VM_VM_AREA)
1331 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001332
Andi Kleen7856dfe2005-05-20 14:27:57 -07001333 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001334}
1335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336/**
Simon Arlott183ff222007-10-20 01:27:18 +02001337 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 * @addr: base address
1339 *
1340 * Search for the kernel VM area starting at @addr, and remove it.
1341 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001342 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001344struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Nick Piggindb64fe02008-10-18 20:27:03 -07001346 struct vmap_area *va;
1347
1348 va = find_vmap_area((unsigned long)addr);
1349 if (va && va->flags & VM_VM_AREA) {
1350 struct vm_struct *vm = va->private;
1351 struct vm_struct *tmp, **p;
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001352 /*
1353 * remove from list and disallow access to this vm_struct
1354 * before unmap. (address range confliction is maintained by
1355 * vmap.)
1356 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001357 write_lock(&vmlist_lock);
1358 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1359 ;
1360 *p = tmp->next;
1361 write_unlock(&vmlist_lock);
1362
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001363 vmap_debug_free_range(va->va_start, va->va_end);
1364 free_unmap_vmap_area(va);
1365 vm->size -= PAGE_SIZE;
1366
Nick Piggindb64fe02008-10-18 20:27:03 -07001367 return vm;
1368 }
1369 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370}
1371
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001372static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
1374 struct vm_struct *area;
1375
1376 if (!addr)
1377 return;
1378
1379 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001380 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 return;
1382 }
1383
1384 area = remove_vm_area(addr);
1385 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001386 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 return;
1389 }
1390
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001391 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001392 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if (deallocate_pages) {
1395 int i;
1396
1397 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001398 struct page *page = area->pages[i];
1399
1400 BUG_ON(!page);
1401 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001404 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 vfree(area->pages);
1406 else
1407 kfree(area->pages);
1408 }
1409
1410 kfree(area);
1411 return;
1412}
1413
1414/**
1415 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 * @addr: memory base address
1417 *
Simon Arlott183ff222007-10-20 01:27:18 +02001418 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001419 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1420 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001422 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001424void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001427
1428 kmemleak_free(addr);
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 __vunmap(addr, 1);
1431}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432EXPORT_SYMBOL(vfree);
1433
1434/**
1435 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 * @addr: memory base address
1437 *
1438 * Free the virtually contiguous memory area starting at @addr,
1439 * which was created from the page array passed to vmap().
1440 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001441 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001443void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
1445 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001446 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 __vunmap(addr, 0);
1448}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449EXPORT_SYMBOL(vunmap);
1450
1451/**
1452 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 * @pages: array of page pointers
1454 * @count: number of pages to map
1455 * @flags: vm_area->flags
1456 * @prot: page protection for the mapping
1457 *
1458 * Maps @count pages from @pages into contiguous kernel virtual
1459 * space.
1460 */
1461void *vmap(struct page **pages, unsigned int count,
1462 unsigned long flags, pgprot_t prot)
1463{
1464 struct vm_struct *area;
1465
Peter Zijlstra34754b62009-02-25 16:04:03 +01001466 might_sleep();
1467
Jan Beulich44813742009-09-21 17:03:05 -07001468 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 return NULL;
1470
Christoph Lameter23016962008-04-28 02:12:42 -07001471 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1472 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (!area)
1474 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (map_vm_area(area, prot, &pages)) {
1477 vunmap(area->addr);
1478 return NULL;
1479 }
1480
1481 return area->addr;
1482}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483EXPORT_SYMBOL(vmap);
1484
David Miller2dca6992009-09-21 12:22:34 -07001485static void *__vmalloc_node(unsigned long size, unsigned long align,
1486 gfp_t gfp_mask, pgprot_t prot,
Nick Piggindb64fe02008-10-18 20:27:03 -07001487 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001488static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001489 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
1491 struct page **pages;
1492 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001493 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1496 array_size = (nr_pages * sizeof(struct page *));
1497
1498 area->nr_pages = nr_pages;
1499 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001500 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001501 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Christoph Lameter23016962008-04-28 02:12:42 -07001502 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001503 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001504 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001505 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001506 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001508 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 if (!area->pages) {
1510 remove_vm_area(area->addr);
1511 kfree(area);
1512 return NULL;
1513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
1515 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001516 struct page *page;
1517
Christoph Lameter930fc452005-10-29 18:15:41 -07001518 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001519 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001520 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001521 page = alloc_pages_node(node, gfp_mask, 0);
1522
1523 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 /* Successfully allocated i pages, free them in __vunmap() */
1525 area->nr_pages = i;
1526 goto fail;
1527 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001528 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 }
1530
1531 if (map_vm_area(area, prot, &pages))
1532 goto fail;
1533 return area->addr;
1534
1535fail:
1536 vfree(area->addr);
1537 return NULL;
1538}
1539
Christoph Lameter930fc452005-10-29 18:15:41 -07001540void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1541{
Catalin Marinas89219d32009-06-11 13:23:19 +01001542 void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1,
1543 __builtin_return_address(0));
1544
1545 /*
1546 * A ref_count = 3 is needed because the vm_struct and vmap_area
1547 * structures allocated in the __get_vm_area_node() function contain
1548 * references to the virtual address of the vmalloc'ed block.
1549 */
1550 kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask);
1551
1552 return addr;
Christoph Lameter930fc452005-10-29 18:15:41 -07001553}
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001556 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001558 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 * @gfp_mask: flags for the page level allocator
1560 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001561 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001562 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 *
1564 * Allocate enough pages to cover @size from the page level
1565 * allocator with @gfp_mask flags. Map them into contiguous
1566 * kernel virtual space, using a pagetable protection of @prot.
1567 */
David Miller2dca6992009-09-21 12:22:34 -07001568static void *__vmalloc_node(unsigned long size, unsigned long align,
1569 gfp_t gfp_mask, pgprot_t prot,
1570 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571{
1572 struct vm_struct *area;
Catalin Marinas89219d32009-06-11 13:23:19 +01001573 void *addr;
1574 unsigned long real_size = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 size = PAGE_ALIGN(size);
Jan Beulich44813742009-09-21 17:03:05 -07001577 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return NULL;
1579
David Miller2dca6992009-09-21 12:22:34 -07001580 area = __get_vm_area_node(size, align, VM_ALLOC, VMALLOC_START,
1581 VMALLOC_END, node, gfp_mask, caller);
Christoph Lameter23016962008-04-28 02:12:42 -07001582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 if (!area)
1584 return NULL;
1585
Catalin Marinas89219d32009-06-11 13:23:19 +01001586 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1587
1588 /*
1589 * A ref_count = 3 is needed because the vm_struct and vmap_area
1590 * structures allocated in the __get_vm_area_node() function contain
1591 * references to the virtual address of the vmalloc'ed block.
1592 */
1593 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1594
1595 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
1597
Christoph Lameter930fc452005-10-29 18:15:41 -07001598void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1599{
David Miller2dca6992009-09-21 12:22:34 -07001600 return __vmalloc_node(size, 1, gfp_mask, prot, -1,
Christoph Lameter23016962008-04-28 02:12:42 -07001601 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001602}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603EXPORT_SYMBOL(__vmalloc);
1604
Dave Younge1ca7782010-10-26 14:22:06 -07001605static inline void *__vmalloc_node_flags(unsigned long size,
1606 int node, gfp_t flags)
1607{
1608 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1609 node, __builtin_return_address(0));
1610}
1611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612/**
1613 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 * Allocate enough pages to cover @size from the page level
1616 * allocator and map them into contiguous kernel virtual space.
1617 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001618 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 * use __vmalloc() instead.
1620 */
1621void *vmalloc(unsigned long size)
1622{
Dave Younge1ca7782010-10-26 14:22:06 -07001623 return __vmalloc_node_flags(size, -1, GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625EXPORT_SYMBOL(vmalloc);
1626
Christoph Lameter930fc452005-10-29 18:15:41 -07001627/**
Dave Younge1ca7782010-10-26 14:22:06 -07001628 * vzalloc - allocate virtually contiguous memory with zero fill
1629 * @size: allocation size
1630 * Allocate enough pages to cover @size from the page level
1631 * allocator and map them into contiguous kernel virtual space.
1632 * The memory allocated is set to zero.
1633 *
1634 * For tight control over page level allocator and protection flags
1635 * use __vmalloc() instead.
1636 */
1637void *vzalloc(unsigned long size)
1638{
1639 return __vmalloc_node_flags(size, -1,
1640 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1641}
1642EXPORT_SYMBOL(vzalloc);
1643
1644/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001645 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1646 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001647 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001648 * The resulting memory area is zeroed so it can be mapped to userspace
1649 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001650 */
1651void *vmalloc_user(unsigned long size)
1652{
1653 struct vm_struct *area;
1654 void *ret;
1655
David Miller2dca6992009-09-21 12:22:34 -07001656 ret = __vmalloc_node(size, SHMLBA,
1657 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
Glauber Costa84877842009-01-06 14:39:19 -08001658 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001659 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001660 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001661 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001662 }
Nick Piggin83342312006-06-23 02:03:20 -07001663 return ret;
1664}
1665EXPORT_SYMBOL(vmalloc_user);
1666
1667/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001668 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001669 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001670 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001671 *
1672 * Allocate enough pages to cover @size from the page level
1673 * allocator and map them into contiguous kernel virtual space.
1674 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001675 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001676 * use __vmalloc() instead.
1677 */
1678void *vmalloc_node(unsigned long size, int node)
1679{
David Miller2dca6992009-09-21 12:22:34 -07001680 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001681 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001682}
1683EXPORT_SYMBOL(vmalloc_node);
1684
Dave Younge1ca7782010-10-26 14:22:06 -07001685/**
1686 * vzalloc_node - allocate memory on a specific node with zero fill
1687 * @size: allocation size
1688 * @node: numa node
1689 *
1690 * Allocate enough pages to cover @size from the page level
1691 * allocator and map them into contiguous kernel virtual space.
1692 * The memory allocated is set to zero.
1693 *
1694 * For tight control over page level allocator and protection flags
1695 * use __vmalloc_node() instead.
1696 */
1697void *vzalloc_node(unsigned long size, int node)
1698{
1699 return __vmalloc_node_flags(size, node,
1700 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1701}
1702EXPORT_SYMBOL(vzalloc_node);
1703
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001704#ifndef PAGE_KERNEL_EXEC
1705# define PAGE_KERNEL_EXEC PAGE_KERNEL
1706#endif
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708/**
1709 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 * @size: allocation size
1711 *
1712 * Kernel-internal function to allocate enough pages to cover @size
1713 * the page level allocator and map them into contiguous and
1714 * executable kernel virtual space.
1715 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001716 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 * use __vmalloc() instead.
1718 */
1719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720void *vmalloc_exec(unsigned long size)
1721{
David Miller2dca6992009-09-21 12:22:34 -07001722 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
Glauber Costa84877842009-01-06 14:39:19 -08001723 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
1725
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001726#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001727#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001728#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001729#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001730#else
1731#define GFP_VMALLOC32 GFP_KERNEL
1732#endif
1733
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734/**
1735 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 * @size: allocation size
1737 *
1738 * Allocate enough 32bit PA addressable pages to cover @size from the
1739 * page level allocator and map them into contiguous kernel virtual space.
1740 */
1741void *vmalloc_32(unsigned long size)
1742{
David Miller2dca6992009-09-21 12:22:34 -07001743 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001744 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746EXPORT_SYMBOL(vmalloc_32);
1747
Nick Piggin83342312006-06-23 02:03:20 -07001748/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001749 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001750 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001751 *
1752 * The resulting memory area is 32bit addressable and zeroed so it can be
1753 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001754 */
1755void *vmalloc_32_user(unsigned long size)
1756{
1757 struct vm_struct *area;
1758 void *ret;
1759
David Miller2dca6992009-09-21 12:22:34 -07001760 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001761 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001762 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001763 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001764 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001765 }
Nick Piggin83342312006-06-23 02:03:20 -07001766 return ret;
1767}
1768EXPORT_SYMBOL(vmalloc_32_user);
1769
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001770/*
1771 * small helper routine , copy contents to buf from addr.
1772 * If the page is not present, fill zero.
1773 */
1774
1775static int aligned_vread(char *buf, char *addr, unsigned long count)
1776{
1777 struct page *p;
1778 int copied = 0;
1779
1780 while (count) {
1781 unsigned long offset, length;
1782
1783 offset = (unsigned long)addr & ~PAGE_MASK;
1784 length = PAGE_SIZE - offset;
1785 if (length > count)
1786 length = count;
1787 p = vmalloc_to_page(addr);
1788 /*
1789 * To do safe access to this _mapped_ area, we need
1790 * lock. But adding lock here means that we need to add
1791 * overhead of vmalloc()/vfree() calles for this _debug_
1792 * interface, rarely used. Instead of that, we'll use
1793 * kmap() and get small overhead in this access function.
1794 */
1795 if (p) {
1796 /*
1797 * we can expect USER0 is not used (see vread/vwrite's
1798 * function description)
1799 */
1800 void *map = kmap_atomic(p, KM_USER0);
1801 memcpy(buf, map + offset, length);
1802 kunmap_atomic(map, KM_USER0);
1803 } else
1804 memset(buf, 0, length);
1805
1806 addr += length;
1807 buf += length;
1808 copied += length;
1809 count -= length;
1810 }
1811 return copied;
1812}
1813
1814static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1815{
1816 struct page *p;
1817 int copied = 0;
1818
1819 while (count) {
1820 unsigned long offset, length;
1821
1822 offset = (unsigned long)addr & ~PAGE_MASK;
1823 length = PAGE_SIZE - offset;
1824 if (length > count)
1825 length = count;
1826 p = vmalloc_to_page(addr);
1827 /*
1828 * To do safe access to this _mapped_ area, we need
1829 * lock. But adding lock here means that we need to add
1830 * overhead of vmalloc()/vfree() calles for this _debug_
1831 * interface, rarely used. Instead of that, we'll use
1832 * kmap() and get small overhead in this access function.
1833 */
1834 if (p) {
1835 /*
1836 * we can expect USER0 is not used (see vread/vwrite's
1837 * function description)
1838 */
1839 void *map = kmap_atomic(p, KM_USER0);
1840 memcpy(map + offset, buf, length);
1841 kunmap_atomic(map, KM_USER0);
1842 }
1843 addr += length;
1844 buf += length;
1845 copied += length;
1846 count -= length;
1847 }
1848 return copied;
1849}
1850
1851/**
1852 * vread() - read vmalloc area in a safe way.
1853 * @buf: buffer for reading data
1854 * @addr: vm address.
1855 * @count: number of bytes to be read.
1856 *
1857 * Returns # of bytes which addr and buf should be increased.
1858 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1859 * includes any intersect with alive vmalloc area.
1860 *
1861 * This function checks that addr is a valid vmalloc'ed area, and
1862 * copy data from that area to a given buffer. If the given memory range
1863 * of [addr...addr+count) includes some valid address, data is copied to
1864 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1865 * IOREMAP area is treated as memory hole and no copy is done.
1866 *
1867 * If [addr...addr+count) doesn't includes any intersects with alive
1868 * vm_struct area, returns 0.
1869 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1870 * the caller should guarantee KM_USER0 is not used.
1871 *
1872 * Note: In usual ops, vread() is never necessary because the caller
1873 * should know vmalloc() area is valid and can use memcpy().
1874 * This is for routines which have to access vmalloc area without
1875 * any informaion, as /dev/kmem.
1876 *
1877 */
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879long vread(char *buf, char *addr, unsigned long count)
1880{
1881 struct vm_struct *tmp;
1882 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001883 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 unsigned long n;
1885
1886 /* Don't allow overflow */
1887 if ((unsigned long) addr + count < count)
1888 count = -(unsigned long) addr;
1889
1890 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001891 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 vaddr = (char *) tmp->addr;
1893 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1894 continue;
1895 while (addr < vaddr) {
1896 if (count == 0)
1897 goto finished;
1898 *buf = '\0';
1899 buf++;
1900 addr++;
1901 count--;
1902 }
1903 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001904 if (n > count)
1905 n = count;
1906 if (!(tmp->flags & VM_IOREMAP))
1907 aligned_vread(buf, addr, n);
1908 else /* IOREMAP area is treated as memory hole */
1909 memset(buf, 0, n);
1910 buf += n;
1911 addr += n;
1912 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 }
1914finished:
1915 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001916
1917 if (buf == buf_start)
1918 return 0;
1919 /* zero-fill memory holes */
1920 if (buf != buf_start + buflen)
1921 memset(buf, 0, buflen - (buf - buf_start));
1922
1923 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924}
1925
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001926/**
1927 * vwrite() - write vmalloc area in a safe way.
1928 * @buf: buffer for source data
1929 * @addr: vm address.
1930 * @count: number of bytes to be read.
1931 *
1932 * Returns # of bytes which addr and buf should be incresed.
1933 * (same number to @count).
1934 * If [addr...addr+count) doesn't includes any intersect with valid
1935 * vmalloc area, returns 0.
1936 *
1937 * This function checks that addr is a valid vmalloc'ed area, and
1938 * copy data from a buffer to the given addr. If specified range of
1939 * [addr...addr+count) includes some valid address, data is copied from
1940 * proper area of @buf. If there are memory holes, no copy to hole.
1941 * IOREMAP area is treated as memory hole and no copy is done.
1942 *
1943 * If [addr...addr+count) doesn't includes any intersects with alive
1944 * vm_struct area, returns 0.
1945 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1946 * the caller should guarantee KM_USER0 is not used.
1947 *
1948 * Note: In usual ops, vwrite() is never necessary because the caller
1949 * should know vmalloc() area is valid and can use memcpy().
1950 * This is for routines which have to access vmalloc area without
1951 * any informaion, as /dev/kmem.
1952 *
1953 * The caller should guarantee KM_USER1 is not used.
1954 */
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956long vwrite(char *buf, char *addr, unsigned long count)
1957{
1958 struct vm_struct *tmp;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001959 char *vaddr;
1960 unsigned long n, buflen;
1961 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 /* Don't allow overflow */
1964 if ((unsigned long) addr + count < count)
1965 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001966 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001969 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 vaddr = (char *) tmp->addr;
1971 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1972 continue;
1973 while (addr < vaddr) {
1974 if (count == 0)
1975 goto finished;
1976 buf++;
1977 addr++;
1978 count--;
1979 }
1980 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001981 if (n > count)
1982 n = count;
1983 if (!(tmp->flags & VM_IOREMAP)) {
1984 aligned_vwrite(buf, addr, n);
1985 copied++;
1986 }
1987 buf += n;
1988 addr += n;
1989 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 }
1991finished:
1992 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001993 if (!copied)
1994 return 0;
1995 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996}
Nick Piggin83342312006-06-23 02:03:20 -07001997
1998/**
1999 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002000 * @vma: vma to cover (map full range of vma)
2001 * @addr: vmalloc memory
2002 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002003 *
2004 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002005 *
2006 * This function checks that addr is a valid vmalloc'ed area, and
2007 * that it is big enough to cover the vma. Will return failure if
2008 * that criteria isn't met.
2009 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002010 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002011 */
2012int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2013 unsigned long pgoff)
2014{
2015 struct vm_struct *area;
2016 unsigned long uaddr = vma->vm_start;
2017 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07002018
2019 if ((PAGE_SIZE-1) & (unsigned long)addr)
2020 return -EINVAL;
2021
Nick Piggindb64fe02008-10-18 20:27:03 -07002022 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07002023 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07002024 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002025
2026 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07002027 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002028
2029 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07002030 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002031
2032 addr += pgoff << PAGE_SHIFT;
2033 do {
2034 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07002035 int ret;
2036
Nick Piggin83342312006-06-23 02:03:20 -07002037 ret = vm_insert_page(vma, uaddr, page);
2038 if (ret)
2039 return ret;
2040
2041 uaddr += PAGE_SIZE;
2042 addr += PAGE_SIZE;
2043 usize -= PAGE_SIZE;
2044 } while (usize > 0);
2045
2046 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
2047 vma->vm_flags |= VM_RESERVED;
2048
Nick Piggindb64fe02008-10-18 20:27:03 -07002049 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07002050}
2051EXPORT_SYMBOL(remap_vmalloc_range);
2052
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002053/*
2054 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2055 * have one.
2056 */
2057void __attribute__((weak)) vmalloc_sync_all(void)
2058{
2059}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002060
2061
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002062static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002063{
2064 /* apply_to_page_range() does all the hard work. */
2065 return 0;
2066}
2067
2068/**
2069 * alloc_vm_area - allocate a range of kernel address space
2070 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07002071 *
2072 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002073 *
2074 * This function reserves a range of kernel address space, and
2075 * allocates pagetables to map that range. No actual mappings
2076 * are created. If the kernel address space is not shared
2077 * between processes, it syncs the pagetable across all
2078 * processes.
2079 */
2080struct vm_struct *alloc_vm_area(size_t size)
2081{
2082 struct vm_struct *area;
2083
Christoph Lameter23016962008-04-28 02:12:42 -07002084 area = get_vm_area_caller(size, VM_IOREMAP,
2085 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002086 if (area == NULL)
2087 return NULL;
2088
2089 /*
2090 * This ensures that page tables are constructed for this region
2091 * of kernel virtual address space and mapped into init_mm.
2092 */
2093 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2094 area->size, f, NULL)) {
2095 free_vm_area(area);
2096 return NULL;
2097 }
2098
2099 /* Make sure the pagetables are constructed in process kernel
2100 mappings */
2101 vmalloc_sync_all();
2102
2103 return area;
2104}
2105EXPORT_SYMBOL_GPL(alloc_vm_area);
2106
2107void free_vm_area(struct vm_struct *area)
2108{
2109 struct vm_struct *ret;
2110 ret = remove_vm_area(area->addr);
2111 BUG_ON(ret != area);
2112 kfree(area);
2113}
2114EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002115
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002116#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002117static struct vmap_area *node_to_va(struct rb_node *n)
2118{
2119 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2120}
2121
2122/**
2123 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2124 * @end: target address
2125 * @pnext: out arg for the next vmap_area
2126 * @pprev: out arg for the previous vmap_area
2127 *
2128 * Returns: %true if either or both of next and prev are found,
2129 * %false if no vmap_area exists
2130 *
2131 * Find vmap_areas end addresses of which enclose @end. ie. if not
2132 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2133 */
2134static bool pvm_find_next_prev(unsigned long end,
2135 struct vmap_area **pnext,
2136 struct vmap_area **pprev)
2137{
2138 struct rb_node *n = vmap_area_root.rb_node;
2139 struct vmap_area *va = NULL;
2140
2141 while (n) {
2142 va = rb_entry(n, struct vmap_area, rb_node);
2143 if (end < va->va_end)
2144 n = n->rb_left;
2145 else if (end > va->va_end)
2146 n = n->rb_right;
2147 else
2148 break;
2149 }
2150
2151 if (!va)
2152 return false;
2153
2154 if (va->va_end > end) {
2155 *pnext = va;
2156 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2157 } else {
2158 *pprev = va;
2159 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2160 }
2161 return true;
2162}
2163
2164/**
2165 * pvm_determine_end - find the highest aligned address between two vmap_areas
2166 * @pnext: in/out arg for the next vmap_area
2167 * @pprev: in/out arg for the previous vmap_area
2168 * @align: alignment
2169 *
2170 * Returns: determined end address
2171 *
2172 * Find the highest aligned address between *@pnext and *@pprev below
2173 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2174 * down address is between the end addresses of the two vmap_areas.
2175 *
2176 * Please note that the address returned by this function may fall
2177 * inside *@pnext vmap_area. The caller is responsible for checking
2178 * that.
2179 */
2180static unsigned long pvm_determine_end(struct vmap_area **pnext,
2181 struct vmap_area **pprev,
2182 unsigned long align)
2183{
2184 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2185 unsigned long addr;
2186
2187 if (*pnext)
2188 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2189 else
2190 addr = vmalloc_end;
2191
2192 while (*pprev && (*pprev)->va_end > addr) {
2193 *pnext = *pprev;
2194 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2195 }
2196
2197 return addr;
2198}
2199
2200/**
2201 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2202 * @offsets: array containing offset of each area
2203 * @sizes: array containing size of each area
2204 * @nr_vms: the number of areas to allocate
2205 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2206 * @gfp_mask: allocation mask
2207 *
2208 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2209 * vm_structs on success, %NULL on failure
2210 *
2211 * Percpu allocator wants to use congruent vm areas so that it can
2212 * maintain the offsets among percpu areas. This function allocates
2213 * congruent vmalloc areas for it. These areas tend to be scattered
2214 * pretty far, distance between two areas easily going up to
2215 * gigabytes. To avoid interacting with regular vmallocs, these areas
2216 * are allocated from top.
2217 *
2218 * Despite its complicated look, this allocator is rather simple. It
2219 * does everything top-down and scans areas from the end looking for
2220 * matching slot. While scanning, if any of the areas overlaps with
2221 * existing vmap_area, the base address is pulled down to fit the
2222 * area. Scanning is repeated till all the areas fit and then all
2223 * necessary data structres are inserted and the result is returned.
2224 */
2225struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2226 const size_t *sizes, int nr_vms,
2227 size_t align, gfp_t gfp_mask)
2228{
2229 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2230 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2231 struct vmap_area **vas, *prev, *next;
2232 struct vm_struct **vms;
2233 int area, area2, last_area, term_area;
2234 unsigned long base, start, end, last_end;
2235 bool purged = false;
2236
2237 gfp_mask &= GFP_RECLAIM_MASK;
2238
2239 /* verify parameters and allocate data structures */
2240 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2241 for (last_area = 0, area = 0; area < nr_vms; area++) {
2242 start = offsets[area];
2243 end = start + sizes[area];
2244
2245 /* is everything aligned properly? */
2246 BUG_ON(!IS_ALIGNED(offsets[area], align));
2247 BUG_ON(!IS_ALIGNED(sizes[area], align));
2248
2249 /* detect the area with the highest address */
2250 if (start > offsets[last_area])
2251 last_area = area;
2252
2253 for (area2 = 0; area2 < nr_vms; area2++) {
2254 unsigned long start2 = offsets[area2];
2255 unsigned long end2 = start2 + sizes[area2];
2256
2257 if (area2 == area)
2258 continue;
2259
2260 BUG_ON(start2 >= start && start2 < end);
2261 BUG_ON(end2 <= end && end2 > start);
2262 }
2263 }
2264 last_end = offsets[last_area] + sizes[last_area];
2265
2266 if (vmalloc_end - vmalloc_start < last_end) {
2267 WARN_ON(true);
2268 return NULL;
2269 }
2270
2271 vms = kzalloc(sizeof(vms[0]) * nr_vms, gfp_mask);
2272 vas = kzalloc(sizeof(vas[0]) * nr_vms, gfp_mask);
2273 if (!vas || !vms)
2274 goto err_free;
2275
2276 for (area = 0; area < nr_vms; area++) {
2277 vas[area] = kzalloc(sizeof(struct vmap_area), gfp_mask);
2278 vms[area] = kzalloc(sizeof(struct vm_struct), gfp_mask);
2279 if (!vas[area] || !vms[area])
2280 goto err_free;
2281 }
2282retry:
2283 spin_lock(&vmap_area_lock);
2284
2285 /* start scanning - we scan from the top, begin with the last area */
2286 area = term_area = last_area;
2287 start = offsets[area];
2288 end = start + sizes[area];
2289
2290 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2291 base = vmalloc_end - last_end;
2292 goto found;
2293 }
2294 base = pvm_determine_end(&next, &prev, align) - end;
2295
2296 while (true) {
2297 BUG_ON(next && next->va_end <= base + end);
2298 BUG_ON(prev && prev->va_end > base + end);
2299
2300 /*
2301 * base might have underflowed, add last_end before
2302 * comparing.
2303 */
2304 if (base + last_end < vmalloc_start + last_end) {
2305 spin_unlock(&vmap_area_lock);
2306 if (!purged) {
2307 purge_vmap_area_lazy();
2308 purged = true;
2309 goto retry;
2310 }
2311 goto err_free;
2312 }
2313
2314 /*
2315 * If next overlaps, move base downwards so that it's
2316 * right below next and then recheck.
2317 */
2318 if (next && next->va_start < base + end) {
2319 base = pvm_determine_end(&next, &prev, align) - end;
2320 term_area = area;
2321 continue;
2322 }
2323
2324 /*
2325 * If prev overlaps, shift down next and prev and move
2326 * base so that it's right below new next and then
2327 * recheck.
2328 */
2329 if (prev && prev->va_end > base + start) {
2330 next = prev;
2331 prev = node_to_va(rb_prev(&next->rb_node));
2332 base = pvm_determine_end(&next, &prev, align) - end;
2333 term_area = area;
2334 continue;
2335 }
2336
2337 /*
2338 * This area fits, move on to the previous one. If
2339 * the previous one is the terminal one, we're done.
2340 */
2341 area = (area + nr_vms - 1) % nr_vms;
2342 if (area == term_area)
2343 break;
2344 start = offsets[area];
2345 end = start + sizes[area];
2346 pvm_find_next_prev(base + end, &next, &prev);
2347 }
2348found:
2349 /* we've found a fitting base, insert all va's */
2350 for (area = 0; area < nr_vms; area++) {
2351 struct vmap_area *va = vas[area];
2352
2353 va->va_start = base + offsets[area];
2354 va->va_end = va->va_start + sizes[area];
2355 __insert_vmap_area(va);
2356 }
2357
2358 vmap_area_pcpu_hole = base + offsets[last_area];
2359
2360 spin_unlock(&vmap_area_lock);
2361
2362 /* insert all vm's */
2363 for (area = 0; area < nr_vms; area++)
2364 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2365 pcpu_get_vm_areas);
2366
2367 kfree(vas);
2368 return vms;
2369
2370err_free:
2371 for (area = 0; area < nr_vms; area++) {
2372 if (vas)
2373 kfree(vas[area]);
2374 if (vms)
2375 kfree(vms[area]);
2376 }
2377 kfree(vas);
2378 kfree(vms);
2379 return NULL;
2380}
2381
2382/**
2383 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2384 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2385 * @nr_vms: the number of allocated areas
2386 *
2387 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2388 */
2389void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2390{
2391 int i;
2392
2393 for (i = 0; i < nr_vms; i++)
2394 free_vm_area(vms[i]);
2395 kfree(vms);
2396}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002397#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002398
2399#ifdef CONFIG_PROC_FS
2400static void *s_start(struct seq_file *m, loff_t *pos)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002401 __acquires(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002402{
2403 loff_t n = *pos;
2404 struct vm_struct *v;
2405
2406 read_lock(&vmlist_lock);
2407 v = vmlist;
2408 while (n > 0 && v) {
2409 n--;
2410 v = v->next;
2411 }
2412 if (!n)
2413 return v;
2414
2415 return NULL;
2416
2417}
2418
2419static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2420{
2421 struct vm_struct *v = p;
2422
2423 ++*pos;
2424 return v->next;
2425}
2426
2427static void s_stop(struct seq_file *m, void *p)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002428 __releases(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002429{
2430 read_unlock(&vmlist_lock);
2431}
2432
Eric Dumazeta47a1262008-07-23 21:27:38 -07002433static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2434{
2435 if (NUMA_BUILD) {
2436 unsigned int nr, *counters = m->private;
2437
2438 if (!counters)
2439 return;
2440
2441 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2442
2443 for (nr = 0; nr < v->nr_pages; nr++)
2444 counters[page_to_nid(v->pages[nr])]++;
2445
2446 for_each_node_state(nr, N_HIGH_MEMORY)
2447 if (counters[nr])
2448 seq_printf(m, " N%u=%u", nr, counters[nr]);
2449 }
2450}
2451
Christoph Lametera10aa572008-04-28 02:12:40 -07002452static int s_show(struct seq_file *m, void *p)
2453{
2454 struct vm_struct *v = p;
2455
2456 seq_printf(m, "0x%p-0x%p %7ld",
2457 v->addr, v->addr + v->size, v->size);
2458
Christoph Lameter23016962008-04-28 02:12:42 -07002459 if (v->caller) {
Hugh Dickins9c246242008-12-09 13:14:27 -08002460 char buff[KSYM_SYMBOL_LEN];
Christoph Lameter23016962008-04-28 02:12:42 -07002461
2462 seq_putc(m, ' ');
2463 sprint_symbol(buff, (unsigned long)v->caller);
2464 seq_puts(m, buff);
2465 }
2466
Christoph Lametera10aa572008-04-28 02:12:40 -07002467 if (v->nr_pages)
2468 seq_printf(m, " pages=%d", v->nr_pages);
2469
2470 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002471 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002472
2473 if (v->flags & VM_IOREMAP)
2474 seq_printf(m, " ioremap");
2475
2476 if (v->flags & VM_ALLOC)
2477 seq_printf(m, " vmalloc");
2478
2479 if (v->flags & VM_MAP)
2480 seq_printf(m, " vmap");
2481
2482 if (v->flags & VM_USERMAP)
2483 seq_printf(m, " user");
2484
2485 if (v->flags & VM_VPAGES)
2486 seq_printf(m, " vpages");
2487
Eric Dumazeta47a1262008-07-23 21:27:38 -07002488 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002489 seq_putc(m, '\n');
2490 return 0;
2491}
2492
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002493static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002494 .start = s_start,
2495 .next = s_next,
2496 .stop = s_stop,
2497 .show = s_show,
2498};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002499
2500static int vmalloc_open(struct inode *inode, struct file *file)
2501{
2502 unsigned int *ptr = NULL;
2503 int ret;
2504
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002505 if (NUMA_BUILD) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002506 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002507 if (ptr == NULL)
2508 return -ENOMEM;
2509 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002510 ret = seq_open(file, &vmalloc_op);
2511 if (!ret) {
2512 struct seq_file *m = file->private_data;
2513 m->private = ptr;
2514 } else
2515 kfree(ptr);
2516 return ret;
2517}
2518
2519static const struct file_operations proc_vmalloc_operations = {
2520 .open = vmalloc_open,
2521 .read = seq_read,
2522 .llseek = seq_lseek,
2523 .release = seq_release_private,
2524};
2525
2526static int __init proc_vmalloc_init(void)
2527{
2528 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2529 return 0;
2530}
2531module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07002532#endif
2533