blob: 5d6030235d7a8e4462498e6d9154ca97029069ed [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);
Nick Piggindb64fe02008-10-18 20:27:03 -0700264static LIST_HEAD(vmap_area_list);
Nick Piggin89699602011-03-22 16:30:36 -0700265static struct rb_root vmap_area_root = RB_ROOT;
266
267/* The vmap cache globals are protected by vmap_area_lock */
268static struct rb_node *free_vmap_cache;
269static unsigned long cached_hole_size;
270static unsigned long cached_vstart;
271static unsigned long cached_align;
272
Tejun Heoca23e402009-08-14 15:00:52 +0900273static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700274
275static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
Nick Piggindb64fe02008-10-18 20:27:03 -0700277 struct rb_node *n = vmap_area_root.rb_node;
278
279 while (n) {
280 struct vmap_area *va;
281
282 va = rb_entry(n, struct vmap_area, rb_node);
283 if (addr < va->va_start)
284 n = n->rb_left;
285 else if (addr > va->va_start)
286 n = n->rb_right;
287 else
288 return va;
289 }
290
291 return NULL;
292}
293
294static void __insert_vmap_area(struct vmap_area *va)
295{
296 struct rb_node **p = &vmap_area_root.rb_node;
297 struct rb_node *parent = NULL;
298 struct rb_node *tmp;
299
300 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700301 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700302
303 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700304 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
305 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700306 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700307 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700308 p = &(*p)->rb_right;
309 else
310 BUG();
311 }
312
313 rb_link_node(&va->rb_node, parent, p);
314 rb_insert_color(&va->rb_node, &vmap_area_root);
315
316 /* address-sort this list so it is usable like the vmlist */
317 tmp = rb_prev(&va->rb_node);
318 if (tmp) {
319 struct vmap_area *prev;
320 prev = rb_entry(tmp, struct vmap_area, rb_node);
321 list_add_rcu(&va->list, &prev->list);
322 } else
323 list_add_rcu(&va->list, &vmap_area_list);
324}
325
326static void purge_vmap_area_lazy(void);
327
328/*
329 * Allocate a region of KVA of the specified size and alignment, within the
330 * vstart and vend.
331 */
332static struct vmap_area *alloc_vmap_area(unsigned long size,
333 unsigned long align,
334 unsigned long vstart, unsigned long vend,
335 int node, gfp_t gfp_mask)
336{
337 struct vmap_area *va;
338 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700340 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700341 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700342
Nick Piggin77669702009-02-27 14:03:03 -0800343 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700344 BUG_ON(size & ~PAGE_MASK);
Nick Piggin89699602011-03-22 16:30:36 -0700345 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700346
Nick Piggindb64fe02008-10-18 20:27:03 -0700347 va = kmalloc_node(sizeof(struct vmap_area),
348 gfp_mask & GFP_RECLAIM_MASK, node);
349 if (unlikely(!va))
350 return ERR_PTR(-ENOMEM);
351
352retry:
353 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700354 /*
355 * Invalidate cache if we have more permissive parameters.
356 * cached_hole_size notes the largest hole noticed _below_
357 * the vmap_area cached in free_vmap_cache: if size fits
358 * into that hole, we want to scan from vstart to reuse
359 * the hole instead of allocating above free_vmap_cache.
360 * Note that __free_vmap_area may update free_vmap_cache
361 * without updating cached_hole_size or cached_align.
362 */
363 if (!free_vmap_cache ||
364 size < cached_hole_size ||
365 vstart < cached_vstart ||
366 align < cached_align) {
367nocache:
368 cached_hole_size = 0;
369 free_vmap_cache = NULL;
370 }
371 /* record if we encounter less permissive parameters */
372 cached_vstart = vstart;
373 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800374
Nick Piggin89699602011-03-22 16:30:36 -0700375 /* find starting point for our search */
376 if (free_vmap_cache) {
377 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
378 addr = ALIGN(first->va_end + PAGE_SIZE, align);
379 if (addr < vstart)
380 goto nocache;
381 if (addr + size - 1 < addr)
382 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700383
Nick Piggin89699602011-03-22 16:30:36 -0700384 } else {
385 addr = ALIGN(vstart, align);
386 if (addr + size - 1 < addr)
387 goto overflow;
388
389 n = vmap_area_root.rb_node;
390 first = NULL;
391
392 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700393 struct vmap_area *tmp;
394 tmp = rb_entry(n, struct vmap_area, rb_node);
395 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700396 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700397 if (tmp->va_start <= addr)
398 break;
399 n = n->rb_left;
400 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700401 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700402 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700403
404 if (!first)
405 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700406 }
Nick Piggin89699602011-03-22 16:30:36 -0700407
408 /* from the starting point, walk areas until a suitable hole is found */
409 while (addr + size >= first->va_start && addr + size <= vend) {
410 if (addr + cached_hole_size < first->va_start)
411 cached_hole_size = first->va_start - addr;
412 addr = ALIGN(first->va_end + PAGE_SIZE, align);
413 if (addr + size - 1 < addr)
414 goto overflow;
415
416 n = rb_next(&first->rb_node);
417 if (n)
418 first = rb_entry(n, struct vmap_area, rb_node);
419 else
420 goto found;
421 }
422
Nick Piggindb64fe02008-10-18 20:27:03 -0700423found:
Nick Piggin89699602011-03-22 16:30:36 -0700424 if (addr + size > vend)
425 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700426
427 va->va_start = addr;
428 va->va_end = addr + size;
429 va->flags = 0;
430 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700431 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700432 spin_unlock(&vmap_area_lock);
433
Nick Piggin89699602011-03-22 16:30:36 -0700434 BUG_ON(va->va_start & (align-1));
435 BUG_ON(va->va_start < vstart);
436 BUG_ON(va->va_end > vend);
437
Nick Piggindb64fe02008-10-18 20:27:03 -0700438 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700439
440overflow:
441 spin_unlock(&vmap_area_lock);
442 if (!purged) {
443 purge_vmap_area_lazy();
444 purged = 1;
445 goto retry;
446 }
447 if (printk_ratelimit())
448 printk(KERN_WARNING
449 "vmap allocation for size %lu failed: "
450 "use vmalloc=<size> to increase size.\n", size);
451 kfree(va);
452 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700453}
454
455static void rcu_free_va(struct rcu_head *head)
456{
457 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
458
459 kfree(va);
460}
461
462static void __free_vmap_area(struct vmap_area *va)
463{
464 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700465
466 if (free_vmap_cache) {
467 if (va->va_end < cached_vstart) {
468 free_vmap_cache = NULL;
469 } else {
470 struct vmap_area *cache;
471 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
472 if (va->va_start <= cache->va_start) {
473 free_vmap_cache = rb_prev(&va->rb_node);
474 /*
475 * We don't try to update cached_hole_size or
476 * cached_align, but it won't go very wrong.
477 */
478 }
479 }
480 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700481 rb_erase(&va->rb_node, &vmap_area_root);
482 RB_CLEAR_NODE(&va->rb_node);
483 list_del_rcu(&va->list);
484
Tejun Heoca23e402009-08-14 15:00:52 +0900485 /*
486 * Track the highest possible candidate for pcpu area
487 * allocation. Areas outside of vmalloc area can be returned
488 * here too, consider only end addresses which fall inside
489 * vmalloc area proper.
490 */
491 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
492 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
493
Nick Piggindb64fe02008-10-18 20:27:03 -0700494 call_rcu(&va->rcu_head, rcu_free_va);
495}
496
497/*
498 * Free a region of KVA allocated by alloc_vmap_area
499 */
500static void free_vmap_area(struct vmap_area *va)
501{
502 spin_lock(&vmap_area_lock);
503 __free_vmap_area(va);
504 spin_unlock(&vmap_area_lock);
505}
506
507/*
508 * Clear the pagetable entries of a given vmap_area
509 */
510static void unmap_vmap_area(struct vmap_area *va)
511{
512 vunmap_page_range(va->va_start, va->va_end);
513}
514
Nick Piggincd528582009-01-06 14:39:20 -0800515static void vmap_debug_free_range(unsigned long start, unsigned long end)
516{
517 /*
518 * Unmap page tables and force a TLB flush immediately if
519 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
520 * bugs similarly to those in linear kernel virtual address
521 * space after a page has been freed.
522 *
523 * All the lazy freeing logic is still retained, in order to
524 * minimise intrusiveness of this debugging feature.
525 *
526 * This is going to be *slow* (linear kernel virtual address
527 * debugging doesn't do a broadcast TLB flush so it is a lot
528 * faster).
529 */
530#ifdef CONFIG_DEBUG_PAGEALLOC
531 vunmap_page_range(start, end);
532 flush_tlb_kernel_range(start, end);
533#endif
534}
535
Nick Piggindb64fe02008-10-18 20:27:03 -0700536/*
537 * lazy_max_pages is the maximum amount of virtual address space we gather up
538 * before attempting to purge with a TLB flush.
539 *
540 * There is a tradeoff here: a larger number will cover more kernel page tables
541 * and take slightly longer to purge, but it will linearly reduce the number of
542 * global TLB flushes that must be performed. It would seem natural to scale
543 * this number up linearly with the number of CPUs (because vmapping activity
544 * could also scale linearly with the number of CPUs), however it is likely
545 * that in practice, workloads might be constrained in other ways that mean
546 * vmap activity will not scale linearly with CPUs. Also, I want to be
547 * conservative and not introduce a big latency on huge systems, so go with
548 * a less aggressive log scale. It will still be an improvement over the old
549 * code, and it will be simple to change the scale factor if we find that it
550 * becomes a problem on bigger systems.
551 */
552static unsigned long lazy_max_pages(void)
553{
554 unsigned int log;
555
556 log = fls(num_online_cpus());
557
558 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
559}
560
561static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
562
Nick Piggin02b709d2010-02-01 22:25:57 +1100563/* for per-CPU blocks */
564static void purge_fragmented_blocks_allcpus(void);
565
Nick Piggindb64fe02008-10-18 20:27:03 -0700566/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500567 * called before a call to iounmap() if the caller wants vm_area_struct's
568 * immediately freed.
569 */
570void set_iounmap_nonlazy(void)
571{
572 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
573}
574
575/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700576 * Purges all lazily-freed vmap areas.
577 *
578 * If sync is 0 then don't purge if there is already a purge in progress.
579 * If force_flush is 1, then flush kernel TLBs between *start and *end even
580 * if we found no lazy vmap areas to unmap (callers can use this to optimise
581 * their own TLB flushing).
582 * Returns with *start = min(*start, lowest purged address)
583 * *end = max(*end, highest purged address)
584 */
585static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
586 int sync, int force_flush)
587{
Andrew Morton46666d82009-01-15 13:51:15 -0800588 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700589 LIST_HEAD(valist);
590 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800591 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700592 int nr = 0;
593
594 /*
595 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
596 * should not expect such behaviour. This just simplifies locking for
597 * the case that isn't actually used at the moment anyway.
598 */
599 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800600 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700601 return;
602 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800603 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700604
Nick Piggin02b709d2010-02-01 22:25:57 +1100605 if (sync)
606 purge_fragmented_blocks_allcpus();
607
Nick Piggindb64fe02008-10-18 20:27:03 -0700608 rcu_read_lock();
609 list_for_each_entry_rcu(va, &vmap_area_list, list) {
610 if (va->flags & VM_LAZY_FREE) {
611 if (va->va_start < *start)
612 *start = va->va_start;
613 if (va->va_end > *end)
614 *end = va->va_end;
615 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700616 list_add_tail(&va->purge_list, &valist);
617 va->flags |= VM_LAZY_FREEING;
618 va->flags &= ~VM_LAZY_FREE;
619 }
620 }
621 rcu_read_unlock();
622
Yongseok Koh88f50042010-01-19 17:33:49 +0900623 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700624 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700625
626 if (nr || force_flush)
627 flush_tlb_kernel_range(*start, *end);
628
629 if (nr) {
630 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800631 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700632 __free_vmap_area(va);
633 spin_unlock(&vmap_area_lock);
634 }
Andrew Morton46666d82009-01-15 13:51:15 -0800635 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700636}
637
638/*
Nick Piggin496850e2008-11-19 15:36:33 -0800639 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
640 * is already purging.
641 */
642static void try_purge_vmap_area_lazy(void)
643{
644 unsigned long start = ULONG_MAX, end = 0;
645
646 __purge_vmap_area_lazy(&start, &end, 0, 0);
647}
648
649/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700650 * Kick off a purge of the outstanding lazy areas.
651 */
652static void purge_vmap_area_lazy(void)
653{
654 unsigned long start = ULONG_MAX, end = 0;
655
Nick Piggin496850e2008-11-19 15:36:33 -0800656 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700657}
658
659/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800660 * Free a vmap area, caller ensuring that the area has been unmapped
661 * and flush_cache_vunmap had been called for the correct range
662 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700663 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800664static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700665{
666 va->flags |= VM_LAZY_FREE;
667 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
668 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800669 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700670}
671
Nick Pigginb29acbd2008-12-01 13:13:47 -0800672/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800673 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
674 * called for the correct range previously.
675 */
676static void free_unmap_vmap_area_noflush(struct vmap_area *va)
677{
678 unmap_vmap_area(va);
679 free_vmap_area_noflush(va);
680}
681
682/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800683 * Free and unmap a vmap area
684 */
685static void free_unmap_vmap_area(struct vmap_area *va)
686{
687 flush_cache_vunmap(va->va_start, va->va_end);
688 free_unmap_vmap_area_noflush(va);
689}
690
Nick Piggindb64fe02008-10-18 20:27:03 -0700691static struct vmap_area *find_vmap_area(unsigned long addr)
692{
693 struct vmap_area *va;
694
695 spin_lock(&vmap_area_lock);
696 va = __find_vmap_area(addr);
697 spin_unlock(&vmap_area_lock);
698
699 return va;
700}
701
702static void free_unmap_vmap_area_addr(unsigned long addr)
703{
704 struct vmap_area *va;
705
706 va = find_vmap_area(addr);
707 BUG_ON(!va);
708 free_unmap_vmap_area(va);
709}
710
711
712/*** Per cpu kva allocator ***/
713
714/*
715 * vmap space is limited especially on 32 bit architectures. Ensure there is
716 * room for at least 16 percpu vmap blocks per CPU.
717 */
718/*
719 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
720 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
721 * instead (we just need a rough idea)
722 */
723#if BITS_PER_LONG == 32
724#define VMALLOC_SPACE (128UL*1024*1024)
725#else
726#define VMALLOC_SPACE (128UL*1024*1024*1024)
727#endif
728
729#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
730#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
731#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
732#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
733#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
734#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
735#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
736 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
737 VMALLOC_PAGES / NR_CPUS / 16))
738
739#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
740
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100741static bool vmap_initialized __read_mostly = false;
742
Nick Piggindb64fe02008-10-18 20:27:03 -0700743struct vmap_block_queue {
744 spinlock_t lock;
745 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700746};
747
748struct vmap_block {
749 spinlock_t lock;
750 struct vmap_area *va;
751 struct vmap_block_queue *vbq;
752 unsigned long free, dirty;
753 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
754 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100755 struct list_head free_list;
756 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100757 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700758};
759
760/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
761static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
762
763/*
764 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
765 * in the free path. Could get rid of this if we change the API to return a
766 * "cookie" from alloc, to be passed to free. But no big deal yet.
767 */
768static DEFINE_SPINLOCK(vmap_block_tree_lock);
769static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
770
771/*
772 * We should probably have a fallback mechanism to allocate virtual memory
773 * out of partially filled vmap blocks. However vmap block sizing should be
774 * fairly reasonable according to the vmalloc size, so it shouldn't be a
775 * big problem.
776 */
777
778static unsigned long addr_to_vb_idx(unsigned long addr)
779{
780 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
781 addr /= VMAP_BLOCK_SIZE;
782 return addr;
783}
784
785static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
786{
787 struct vmap_block_queue *vbq;
788 struct vmap_block *vb;
789 struct vmap_area *va;
790 unsigned long vb_idx;
791 int node, err;
792
793 node = numa_node_id();
794
795 vb = kmalloc_node(sizeof(struct vmap_block),
796 gfp_mask & GFP_RECLAIM_MASK, node);
797 if (unlikely(!vb))
798 return ERR_PTR(-ENOMEM);
799
800 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
801 VMALLOC_START, VMALLOC_END,
802 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800803 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700804 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700805 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700806 }
807
808 err = radix_tree_preload(gfp_mask);
809 if (unlikely(err)) {
810 kfree(vb);
811 free_vmap_area(va);
812 return ERR_PTR(err);
813 }
814
815 spin_lock_init(&vb->lock);
816 vb->va = va;
817 vb->free = VMAP_BBMAP_BITS;
818 vb->dirty = 0;
819 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
820 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
821 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700822
823 vb_idx = addr_to_vb_idx(va->va_start);
824 spin_lock(&vmap_block_tree_lock);
825 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
826 spin_unlock(&vmap_block_tree_lock);
827 BUG_ON(err);
828 radix_tree_preload_end();
829
830 vbq = &get_cpu_var(vmap_block_queue);
831 vb->vbq = vbq;
832 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100833 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700834 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900835 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700836
837 return vb;
838}
839
840static void rcu_free_vb(struct rcu_head *head)
841{
842 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
843
844 kfree(vb);
845}
846
847static void free_vmap_block(struct vmap_block *vb)
848{
849 struct vmap_block *tmp;
850 unsigned long vb_idx;
851
Nick Piggindb64fe02008-10-18 20:27:03 -0700852 vb_idx = addr_to_vb_idx(vb->va->va_start);
853 spin_lock(&vmap_block_tree_lock);
854 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
855 spin_unlock(&vmap_block_tree_lock);
856 BUG_ON(tmp != vb);
857
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800858 free_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700859 call_rcu(&vb->rcu_head, rcu_free_vb);
860}
861
Nick Piggin02b709d2010-02-01 22:25:57 +1100862static void purge_fragmented_blocks(int cpu)
863{
864 LIST_HEAD(purge);
865 struct vmap_block *vb;
866 struct vmap_block *n_vb;
867 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
868
869 rcu_read_lock();
870 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
871
872 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
873 continue;
874
875 spin_lock(&vb->lock);
876 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
877 vb->free = 0; /* prevent further allocs after releasing lock */
878 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
879 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
880 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
881 spin_lock(&vbq->lock);
882 list_del_rcu(&vb->free_list);
883 spin_unlock(&vbq->lock);
884 spin_unlock(&vb->lock);
885 list_add_tail(&vb->purge, &purge);
886 } else
887 spin_unlock(&vb->lock);
888 }
889 rcu_read_unlock();
890
891 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
892 list_del(&vb->purge);
893 free_vmap_block(vb);
894 }
895}
896
897static void purge_fragmented_blocks_thiscpu(void)
898{
899 purge_fragmented_blocks(smp_processor_id());
900}
901
902static void purge_fragmented_blocks_allcpus(void)
903{
904 int cpu;
905
906 for_each_possible_cpu(cpu)
907 purge_fragmented_blocks(cpu);
908}
909
Nick Piggindb64fe02008-10-18 20:27:03 -0700910static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
911{
912 struct vmap_block_queue *vbq;
913 struct vmap_block *vb;
914 unsigned long addr = 0;
915 unsigned int order;
Nick Piggin02b709d2010-02-01 22:25:57 +1100916 int purge = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700917
918 BUG_ON(size & ~PAGE_MASK);
919 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
920 order = get_order(size);
921
922again:
923 rcu_read_lock();
924 vbq = &get_cpu_var(vmap_block_queue);
925 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
926 int i;
927
928 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100929 if (vb->free < 1UL << order)
930 goto next;
931
Nick Piggindb64fe02008-10-18 20:27:03 -0700932 i = bitmap_find_free_region(vb->alloc_map,
933 VMAP_BBMAP_BITS, order);
934
Nick Piggin02b709d2010-02-01 22:25:57 +1100935 if (i < 0) {
936 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
937 /* fragmented and no outstanding allocations */
938 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
939 purge = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -0700940 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100941 goto next;
942 }
943 addr = vb->va->va_start + (i << PAGE_SHIFT);
944 BUG_ON(addr_to_vb_idx(addr) !=
945 addr_to_vb_idx(vb->va->va_start));
946 vb->free -= 1UL << order;
947 if (vb->free == 0) {
948 spin_lock(&vbq->lock);
949 list_del_rcu(&vb->free_list);
950 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700951 }
952 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100953 break;
954next:
955 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700956 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100957
958 if (purge)
959 purge_fragmented_blocks_thiscpu();
960
Tejun Heo3f04ba82009-10-29 22:34:12 +0900961 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700962 rcu_read_unlock();
963
964 if (!addr) {
965 vb = new_vmap_block(gfp_mask);
966 if (IS_ERR(vb))
967 return vb;
968 goto again;
969 }
970
971 return (void *)addr;
972}
973
974static void vb_free(const void *addr, unsigned long size)
975{
976 unsigned long offset;
977 unsigned long vb_idx;
978 unsigned int order;
979 struct vmap_block *vb;
980
981 BUG_ON(size & ~PAGE_MASK);
982 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800983
984 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
985
Nick Piggindb64fe02008-10-18 20:27:03 -0700986 order = get_order(size);
987
988 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
989
990 vb_idx = addr_to_vb_idx((unsigned long)addr);
991 rcu_read_lock();
992 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
993 rcu_read_unlock();
994 BUG_ON(!vb);
995
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800996 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
997
Nick Piggindb64fe02008-10-18 20:27:03 -0700998 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100999 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -07001000
Nick Piggindb64fe02008-10-18 20:27:03 -07001001 vb->dirty += 1UL << order;
1002 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001003 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001004 spin_unlock(&vb->lock);
1005 free_vmap_block(vb);
1006 } else
1007 spin_unlock(&vb->lock);
1008}
1009
1010/**
1011 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1012 *
1013 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1014 * to amortize TLB flushing overheads. What this means is that any page you
1015 * have now, may, in a former life, have been mapped into kernel virtual
1016 * address by the vmap layer and so there might be some CPUs with TLB entries
1017 * still referencing that page (additional to the regular 1:1 kernel mapping).
1018 *
1019 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1020 * be sure that none of the pages we have control over will have any aliases
1021 * from the vmap layer.
1022 */
1023void vm_unmap_aliases(void)
1024{
1025 unsigned long start = ULONG_MAX, end = 0;
1026 int cpu;
1027 int flush = 0;
1028
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001029 if (unlikely(!vmap_initialized))
1030 return;
1031
Nick Piggindb64fe02008-10-18 20:27:03 -07001032 for_each_possible_cpu(cpu) {
1033 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1034 struct vmap_block *vb;
1035
1036 rcu_read_lock();
1037 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1038 int i;
1039
1040 spin_lock(&vb->lock);
1041 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1042 while (i < VMAP_BBMAP_BITS) {
1043 unsigned long s, e;
1044 int j;
1045 j = find_next_zero_bit(vb->dirty_map,
1046 VMAP_BBMAP_BITS, i);
1047
1048 s = vb->va->va_start + (i << PAGE_SHIFT);
1049 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001050 flush = 1;
1051
1052 if (s < start)
1053 start = s;
1054 if (e > end)
1055 end = e;
1056
1057 i = j;
1058 i = find_next_bit(vb->dirty_map,
1059 VMAP_BBMAP_BITS, i);
1060 }
1061 spin_unlock(&vb->lock);
1062 }
1063 rcu_read_unlock();
1064 }
1065
1066 __purge_vmap_area_lazy(&start, &end, 1, flush);
1067}
1068EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1069
1070/**
1071 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1072 * @mem: the pointer returned by vm_map_ram
1073 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1074 */
1075void vm_unmap_ram(const void *mem, unsigned int count)
1076{
1077 unsigned long size = count << PAGE_SHIFT;
1078 unsigned long addr = (unsigned long)mem;
1079
1080 BUG_ON(!addr);
1081 BUG_ON(addr < VMALLOC_START);
1082 BUG_ON(addr > VMALLOC_END);
1083 BUG_ON(addr & (PAGE_SIZE-1));
1084
1085 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001086 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001087
1088 if (likely(count <= VMAP_MAX_ALLOC))
1089 vb_free(mem, size);
1090 else
1091 free_unmap_vmap_area_addr(addr);
1092}
1093EXPORT_SYMBOL(vm_unmap_ram);
1094
1095/**
1096 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1097 * @pages: an array of pointers to the pages to be mapped
1098 * @count: number of pages
1099 * @node: prefer to allocate data structures on this node
1100 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001101 *
1102 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001103 */
1104void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1105{
1106 unsigned long size = count << PAGE_SHIFT;
1107 unsigned long addr;
1108 void *mem;
1109
1110 if (likely(count <= VMAP_MAX_ALLOC)) {
1111 mem = vb_alloc(size, GFP_KERNEL);
1112 if (IS_ERR(mem))
1113 return NULL;
1114 addr = (unsigned long)mem;
1115 } else {
1116 struct vmap_area *va;
1117 va = alloc_vmap_area(size, PAGE_SIZE,
1118 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1119 if (IS_ERR(va))
1120 return NULL;
1121
1122 addr = va->va_start;
1123 mem = (void *)addr;
1124 }
1125 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1126 vm_unmap_ram(mem, count);
1127 return NULL;
1128 }
1129 return mem;
1130}
1131EXPORT_SYMBOL(vm_map_ram);
1132
Tejun Heof0aa6612009-02-20 16:29:08 +09001133/**
1134 * vm_area_register_early - register vmap area early during boot
1135 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001136 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001137 *
1138 * This function is used to register kernel vm area before
1139 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1140 * proper values on entry and other fields should be zero. On return,
1141 * vm->addr contains the allocated address.
1142 *
1143 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1144 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001145void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001146{
1147 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001148 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001149
Tejun Heoc0c0a292009-02-24 11:57:21 +09001150 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1151 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1152
1153 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001154
1155 vm->next = vmlist;
1156 vmlist = vm;
1157}
1158
Nick Piggindb64fe02008-10-18 20:27:03 -07001159void __init vmalloc_init(void)
1160{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001161 struct vmap_area *va;
1162 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001163 int i;
1164
1165 for_each_possible_cpu(i) {
1166 struct vmap_block_queue *vbq;
1167
1168 vbq = &per_cpu(vmap_block_queue, i);
1169 spin_lock_init(&vbq->lock);
1170 INIT_LIST_HEAD(&vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001171 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001172
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001173 /* Import existing vmlist entries. */
1174 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001175 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001176 va->flags = tmp->flags | VM_VM_AREA;
1177 va->va_start = (unsigned long)tmp->addr;
1178 va->va_end = va->va_start + tmp->size;
1179 __insert_vmap_area(va);
1180 }
Tejun Heoca23e402009-08-14 15:00:52 +09001181
1182 vmap_area_pcpu_hole = VMALLOC_END;
1183
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001184 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001185}
1186
Tejun Heo8fc48982009-02-20 16:29:08 +09001187/**
1188 * map_kernel_range_noflush - map kernel VM area with the specified pages
1189 * @addr: start of the VM area to map
1190 * @size: size of the VM area to map
1191 * @prot: page protection flags to use
1192 * @pages: pages to map
1193 *
1194 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1195 * specify should have been allocated using get_vm_area() and its
1196 * friends.
1197 *
1198 * NOTE:
1199 * This function does NOT do any cache flushing. The caller is
1200 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1201 * before calling this function.
1202 *
1203 * RETURNS:
1204 * The number of pages mapped on success, -errno on failure.
1205 */
1206int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1207 pgprot_t prot, struct page **pages)
1208{
1209 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1210}
1211
1212/**
1213 * unmap_kernel_range_noflush - unmap kernel VM area
1214 * @addr: start of the VM area to unmap
1215 * @size: size of the VM area to unmap
1216 *
1217 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1218 * specify should have been allocated using get_vm_area() and its
1219 * friends.
1220 *
1221 * NOTE:
1222 * This function does NOT do any cache flushing. The caller is
1223 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1224 * before calling this function and flush_tlb_kernel_range() after.
1225 */
1226void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1227{
1228 vunmap_page_range(addr, addr + size);
1229}
Huang Ying81e88fd2011-01-12 14:44:55 +08001230EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001231
1232/**
1233 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1234 * @addr: start of the VM area to unmap
1235 * @size: size of the VM area to unmap
1236 *
1237 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1238 * the unmapping and tlb after.
1239 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001240void unmap_kernel_range(unsigned long addr, unsigned long size)
1241{
1242 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001243
1244 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001245 vunmap_page_range(addr, end);
1246 flush_tlb_kernel_range(addr, end);
1247}
1248
1249int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1250{
1251 unsigned long addr = (unsigned long)area->addr;
1252 unsigned long end = addr + area->size - PAGE_SIZE;
1253 int err;
1254
1255 err = vmap_page_range(addr, end, prot, *pages);
1256 if (err > 0) {
1257 *pages += err;
1258 err = 0;
1259 }
1260
1261 return err;
1262}
1263EXPORT_SYMBOL_GPL(map_vm_area);
1264
1265/*** Old vmalloc interfaces ***/
1266DEFINE_RWLOCK(vmlist_lock);
1267struct vm_struct *vmlist;
1268
Tejun Heocf88c792009-08-14 15:00:52 +09001269static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1270 unsigned long flags, void *caller)
1271{
1272 struct vm_struct *tmp, **p;
1273
1274 vm->flags = flags;
1275 vm->addr = (void *)va->va_start;
1276 vm->size = va->va_end - va->va_start;
1277 vm->caller = caller;
1278 va->private = vm;
1279 va->flags |= VM_VM_AREA;
1280
1281 write_lock(&vmlist_lock);
1282 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1283 if (tmp->addr >= vm->addr)
1284 break;
1285 }
1286 vm->next = *p;
1287 *p = vm;
1288 write_unlock(&vmlist_lock);
1289}
1290
Nick Piggindb64fe02008-10-18 20:27:03 -07001291static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001292 unsigned long align, unsigned long flags, unsigned long start,
1293 unsigned long end, int node, gfp_t gfp_mask, void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001294{
1295 static struct vmap_area *va;
1296 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001298 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 if (flags & VM_IOREMAP) {
1300 int bit = fls(size);
1301
1302 if (bit > IOREMAP_MAX_ORDER)
1303 bit = IOREMAP_MAX_ORDER;
1304 else if (bit < PAGE_SHIFT)
1305 bit = PAGE_SHIFT;
1306
1307 align = 1ul << bit;
1308 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001311 if (unlikely(!size))
1312 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Tejun Heocf88c792009-08-14 15:00:52 +09001314 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 if (unlikely(!area))
1316 return NULL;
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 /*
1319 * We always allocate a guard page.
1320 */
1321 size += PAGE_SIZE;
1322
Nick Piggindb64fe02008-10-18 20:27:03 -07001323 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1324 if (IS_ERR(va)) {
1325 kfree(area);
1326 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Tejun Heocf88c792009-08-14 15:00:52 +09001329 insert_vmalloc_vm(area, va, flags, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331}
1332
Christoph Lameter930fc452005-10-29 18:15:41 -07001333struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1334 unsigned long start, unsigned long end)
1335{
David Miller2dca6992009-09-21 12:22:34 -07001336 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001337 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001338}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001339EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001340
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001341struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1342 unsigned long start, unsigned long end,
1343 void *caller)
1344{
David Miller2dca6992009-09-21 12:22:34 -07001345 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001346 caller);
1347}
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349/**
Simon Arlott183ff222007-10-20 01:27:18 +02001350 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 * @size: size of the area
1352 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1353 *
1354 * Search an area of @size in the kernel virtual mapping area,
1355 * and reserved it for out purposes. Returns the area descriptor
1356 * on success or %NULL on failure.
1357 */
1358struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1359{
David Miller2dca6992009-09-21 12:22:34 -07001360 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001361 -1, GFP_KERNEL, __builtin_return_address(0));
1362}
1363
1364struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1365 void *caller)
1366{
David Miller2dca6992009-09-21 12:22:34 -07001367 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001368 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369}
1370
Nick Piggindb64fe02008-10-18 20:27:03 -07001371static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001372{
Nick Piggindb64fe02008-10-18 20:27:03 -07001373 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001374
Nick Piggindb64fe02008-10-18 20:27:03 -07001375 va = find_vmap_area((unsigned long)addr);
1376 if (va && va->flags & VM_VM_AREA)
1377 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001378
Andi Kleen7856dfe2005-05-20 14:27:57 -07001379 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001380}
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382/**
Simon Arlott183ff222007-10-20 01:27:18 +02001383 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 * @addr: base address
1385 *
1386 * Search for the kernel VM area starting at @addr, and remove it.
1387 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001388 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001390struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391{
Nick Piggindb64fe02008-10-18 20:27:03 -07001392 struct vmap_area *va;
1393
1394 va = find_vmap_area((unsigned long)addr);
1395 if (va && va->flags & VM_VM_AREA) {
1396 struct vm_struct *vm = va->private;
1397 struct vm_struct *tmp, **p;
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001398 /*
1399 * remove from list and disallow access to this vm_struct
1400 * before unmap. (address range confliction is maintained by
1401 * vmap.)
1402 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001403 write_lock(&vmlist_lock);
1404 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1405 ;
1406 *p = tmp->next;
1407 write_unlock(&vmlist_lock);
1408
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001409 vmap_debug_free_range(va->va_start, va->va_end);
1410 free_unmap_vmap_area(va);
1411 vm->size -= PAGE_SIZE;
1412
Nick Piggindb64fe02008-10-18 20:27:03 -07001413 return vm;
1414 }
1415 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416}
1417
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001418static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
1420 struct vm_struct *area;
1421
1422 if (!addr)
1423 return;
1424
1425 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001426 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return;
1428 }
1429
1430 area = remove_vm_area(addr);
1431 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001432 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 return;
1435 }
1436
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001437 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001438 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (deallocate_pages) {
1441 int i;
1442
1443 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001444 struct page *page = area->pages[i];
1445
1446 BUG_ON(!page);
1447 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 }
1449
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001450 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 vfree(area->pages);
1452 else
1453 kfree(area->pages);
1454 }
1455
1456 kfree(area);
1457 return;
1458}
1459
1460/**
1461 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 * @addr: memory base address
1463 *
Simon Arlott183ff222007-10-20 01:27:18 +02001464 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001465 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1466 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001468 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001470void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
1472 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001473
1474 kmemleak_free(addr);
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 __vunmap(addr, 1);
1477}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478EXPORT_SYMBOL(vfree);
1479
1480/**
1481 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 * @addr: memory base address
1483 *
1484 * Free the virtually contiguous memory area starting at @addr,
1485 * which was created from the page array passed to vmap().
1486 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001487 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001489void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
1491 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001492 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 __vunmap(addr, 0);
1494}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495EXPORT_SYMBOL(vunmap);
1496
1497/**
1498 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 * @pages: array of page pointers
1500 * @count: number of pages to map
1501 * @flags: vm_area->flags
1502 * @prot: page protection for the mapping
1503 *
1504 * Maps @count pages from @pages into contiguous kernel virtual
1505 * space.
1506 */
1507void *vmap(struct page **pages, unsigned int count,
1508 unsigned long flags, pgprot_t prot)
1509{
1510 struct vm_struct *area;
1511
Peter Zijlstra34754b62009-02-25 16:04:03 +01001512 might_sleep();
1513
Jan Beulich44813742009-09-21 17:03:05 -07001514 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 return NULL;
1516
Christoph Lameter23016962008-04-28 02:12:42 -07001517 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1518 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 if (!area)
1520 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001521
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 if (map_vm_area(area, prot, &pages)) {
1523 vunmap(area->addr);
1524 return NULL;
1525 }
1526
1527 return area->addr;
1528}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529EXPORT_SYMBOL(vmap);
1530
David Miller2dca6992009-09-21 12:22:34 -07001531static void *__vmalloc_node(unsigned long size, unsigned long align,
1532 gfp_t gfp_mask, pgprot_t prot,
Nick Piggindb64fe02008-10-18 20:27:03 -07001533 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001534static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001535 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 struct page **pages;
1538 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001539 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1542 array_size = (nr_pages * sizeof(struct page *));
1543
1544 area->nr_pages = nr_pages;
1545 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001546 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001547 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Christoph Lameter23016962008-04-28 02:12:42 -07001548 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001549 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001550 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001551 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001554 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (!area->pages) {
1556 remove_vm_area(area->addr);
1557 kfree(area);
1558 return NULL;
1559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
1561 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001562 struct page *page;
1563
Christoph Lameter930fc452005-10-29 18:15:41 -07001564 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001565 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001566 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001567 page = alloc_pages_node(node, gfp_mask, 0);
1568
1569 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 /* Successfully allocated i pages, free them in __vunmap() */
1571 area->nr_pages = i;
1572 goto fail;
1573 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001574 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 }
1576
1577 if (map_vm_area(area, prot, &pages))
1578 goto fail;
1579 return area->addr;
1580
1581fail:
1582 vfree(area->addr);
1583 return NULL;
1584}
1585
David Rientjesd0a21262011-01-13 15:46:02 -08001586/**
1587 * __vmalloc_node_range - allocate virtually contiguous memory
1588 * @size: allocation size
1589 * @align: desired alignment
1590 * @start: vm area range start
1591 * @end: vm area range end
1592 * @gfp_mask: flags for the page level allocator
1593 * @prot: protection mask for the allocated pages
1594 * @node: node to use for allocation or -1
1595 * @caller: caller's return address
1596 *
1597 * Allocate enough pages to cover @size from the page level
1598 * allocator with @gfp_mask flags. Map them into contiguous
1599 * kernel virtual space, using a pagetable protection of @prot.
1600 */
1601void *__vmalloc_node_range(unsigned long size, unsigned long align,
1602 unsigned long start, unsigned long end, gfp_t gfp_mask,
1603 pgprot_t prot, int node, void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001604{
David Rientjesd0a21262011-01-13 15:46:02 -08001605 struct vm_struct *area;
1606 void *addr;
1607 unsigned long real_size = size;
1608
1609 size = PAGE_ALIGN(size);
1610 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1611 return NULL;
1612
1613 area = __get_vm_area_node(size, align, VM_ALLOC, start, end, node,
1614 gfp_mask, caller);
1615
1616 if (!area)
1617 return NULL;
1618
1619 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Catalin Marinas89219d32009-06-11 13:23:19 +01001620
1621 /*
1622 * A ref_count = 3 is needed because the vm_struct and vmap_area
1623 * structures allocated in the __get_vm_area_node() function contain
1624 * references to the virtual address of the vmalloc'ed block.
1625 */
David Rientjesd0a21262011-01-13 15:46:02 -08001626 kmemleak_alloc(addr, real_size, 3, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001627
1628 return addr;
Christoph Lameter930fc452005-10-29 18:15:41 -07001629}
1630
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001632 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001634 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 * @gfp_mask: flags for the page level allocator
1636 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001637 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001638 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 *
1640 * Allocate enough pages to cover @size from the page level
1641 * allocator with @gfp_mask flags. Map them into contiguous
1642 * kernel virtual space, using a pagetable protection of @prot.
1643 */
David Miller2dca6992009-09-21 12:22:34 -07001644static void *__vmalloc_node(unsigned long size, unsigned long align,
1645 gfp_t gfp_mask, pgprot_t prot,
1646 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
David Rientjesd0a21262011-01-13 15:46:02 -08001648 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1649 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650}
1651
Christoph Lameter930fc452005-10-29 18:15:41 -07001652void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1653{
David Miller2dca6992009-09-21 12:22:34 -07001654 return __vmalloc_node(size, 1, gfp_mask, prot, -1,
Christoph Lameter23016962008-04-28 02:12:42 -07001655 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001656}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657EXPORT_SYMBOL(__vmalloc);
1658
Dave Younge1ca7782010-10-26 14:22:06 -07001659static inline void *__vmalloc_node_flags(unsigned long size,
1660 int node, gfp_t flags)
1661{
1662 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1663 node, __builtin_return_address(0));
1664}
1665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666/**
1667 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 * Allocate enough pages to cover @size from the page level
1670 * allocator and map them into contiguous kernel virtual space.
1671 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001672 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 * use __vmalloc() instead.
1674 */
1675void *vmalloc(unsigned long size)
1676{
Dave Younge1ca7782010-10-26 14:22:06 -07001677 return __vmalloc_node_flags(size, -1, GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679EXPORT_SYMBOL(vmalloc);
1680
Christoph Lameter930fc452005-10-29 18:15:41 -07001681/**
Dave Younge1ca7782010-10-26 14:22:06 -07001682 * vzalloc - allocate virtually contiguous memory with zero fill
1683 * @size: allocation size
1684 * Allocate enough pages to cover @size from the page level
1685 * allocator and map them into contiguous kernel virtual space.
1686 * The memory allocated is set to zero.
1687 *
1688 * For tight control over page level allocator and protection flags
1689 * use __vmalloc() instead.
1690 */
1691void *vzalloc(unsigned long size)
1692{
1693 return __vmalloc_node_flags(size, -1,
1694 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1695}
1696EXPORT_SYMBOL(vzalloc);
1697
1698/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001699 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1700 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001701 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001702 * The resulting memory area is zeroed so it can be mapped to userspace
1703 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001704 */
1705void *vmalloc_user(unsigned long size)
1706{
1707 struct vm_struct *area;
1708 void *ret;
1709
David Miller2dca6992009-09-21 12:22:34 -07001710 ret = __vmalloc_node(size, SHMLBA,
1711 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
Glauber Costa84877842009-01-06 14:39:19 -08001712 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001713 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001714 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001715 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001716 }
Nick Piggin83342312006-06-23 02:03:20 -07001717 return ret;
1718}
1719EXPORT_SYMBOL(vmalloc_user);
1720
1721/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001722 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001723 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001724 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001725 *
1726 * Allocate enough pages to cover @size from the page level
1727 * allocator and map them into contiguous kernel virtual space.
1728 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001729 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001730 * use __vmalloc() instead.
1731 */
1732void *vmalloc_node(unsigned long size, int node)
1733{
David Miller2dca6992009-09-21 12:22:34 -07001734 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001735 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001736}
1737EXPORT_SYMBOL(vmalloc_node);
1738
Dave Younge1ca7782010-10-26 14:22:06 -07001739/**
1740 * vzalloc_node - allocate memory on a specific node with zero fill
1741 * @size: allocation size
1742 * @node: numa node
1743 *
1744 * Allocate enough pages to cover @size from the page level
1745 * allocator and map them into contiguous kernel virtual space.
1746 * The memory allocated is set to zero.
1747 *
1748 * For tight control over page level allocator and protection flags
1749 * use __vmalloc_node() instead.
1750 */
1751void *vzalloc_node(unsigned long size, int node)
1752{
1753 return __vmalloc_node_flags(size, node,
1754 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1755}
1756EXPORT_SYMBOL(vzalloc_node);
1757
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001758#ifndef PAGE_KERNEL_EXEC
1759# define PAGE_KERNEL_EXEC PAGE_KERNEL
1760#endif
1761
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762/**
1763 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 * @size: allocation size
1765 *
1766 * Kernel-internal function to allocate enough pages to cover @size
1767 * the page level allocator and map them into contiguous and
1768 * executable kernel virtual space.
1769 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001770 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 * use __vmalloc() instead.
1772 */
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774void *vmalloc_exec(unsigned long size)
1775{
David Miller2dca6992009-09-21 12:22:34 -07001776 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
Glauber Costa84877842009-01-06 14:39:19 -08001777 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778}
1779
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001780#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001781#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001782#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001783#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001784#else
1785#define GFP_VMALLOC32 GFP_KERNEL
1786#endif
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788/**
1789 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 * @size: allocation size
1791 *
1792 * Allocate enough 32bit PA addressable pages to cover @size from the
1793 * page level allocator and map them into contiguous kernel virtual space.
1794 */
1795void *vmalloc_32(unsigned long size)
1796{
David Miller2dca6992009-09-21 12:22:34 -07001797 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001798 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800EXPORT_SYMBOL(vmalloc_32);
1801
Nick Piggin83342312006-06-23 02:03:20 -07001802/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001803 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001804 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001805 *
1806 * The resulting memory area is 32bit addressable and zeroed so it can be
1807 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001808 */
1809void *vmalloc_32_user(unsigned long size)
1810{
1811 struct vm_struct *area;
1812 void *ret;
1813
David Miller2dca6992009-09-21 12:22:34 -07001814 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001815 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001816 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001817 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001818 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001819 }
Nick Piggin83342312006-06-23 02:03:20 -07001820 return ret;
1821}
1822EXPORT_SYMBOL(vmalloc_32_user);
1823
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001824/*
1825 * small helper routine , copy contents to buf from addr.
1826 * If the page is not present, fill zero.
1827 */
1828
1829static int aligned_vread(char *buf, char *addr, unsigned long count)
1830{
1831 struct page *p;
1832 int copied = 0;
1833
1834 while (count) {
1835 unsigned long offset, length;
1836
1837 offset = (unsigned long)addr & ~PAGE_MASK;
1838 length = PAGE_SIZE - offset;
1839 if (length > count)
1840 length = count;
1841 p = vmalloc_to_page(addr);
1842 /*
1843 * To do safe access to this _mapped_ area, we need
1844 * lock. But adding lock here means that we need to add
1845 * overhead of vmalloc()/vfree() calles for this _debug_
1846 * interface, rarely used. Instead of that, we'll use
1847 * kmap() and get small overhead in this access function.
1848 */
1849 if (p) {
1850 /*
1851 * we can expect USER0 is not used (see vread/vwrite's
1852 * function description)
1853 */
1854 void *map = kmap_atomic(p, KM_USER0);
1855 memcpy(buf, map + offset, length);
1856 kunmap_atomic(map, KM_USER0);
1857 } else
1858 memset(buf, 0, length);
1859
1860 addr += length;
1861 buf += length;
1862 copied += length;
1863 count -= length;
1864 }
1865 return copied;
1866}
1867
1868static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1869{
1870 struct page *p;
1871 int copied = 0;
1872
1873 while (count) {
1874 unsigned long offset, length;
1875
1876 offset = (unsigned long)addr & ~PAGE_MASK;
1877 length = PAGE_SIZE - offset;
1878 if (length > count)
1879 length = count;
1880 p = vmalloc_to_page(addr);
1881 /*
1882 * To do safe access to this _mapped_ area, we need
1883 * lock. But adding lock here means that we need to add
1884 * overhead of vmalloc()/vfree() calles for this _debug_
1885 * interface, rarely used. Instead of that, we'll use
1886 * kmap() and get small overhead in this access function.
1887 */
1888 if (p) {
1889 /*
1890 * we can expect USER0 is not used (see vread/vwrite's
1891 * function description)
1892 */
1893 void *map = kmap_atomic(p, KM_USER0);
1894 memcpy(map + offset, buf, length);
1895 kunmap_atomic(map, KM_USER0);
1896 }
1897 addr += length;
1898 buf += length;
1899 copied += length;
1900 count -= length;
1901 }
1902 return copied;
1903}
1904
1905/**
1906 * vread() - read vmalloc area in a safe way.
1907 * @buf: buffer for reading data
1908 * @addr: vm address.
1909 * @count: number of bytes to be read.
1910 *
1911 * Returns # of bytes which addr and buf should be increased.
1912 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1913 * includes any intersect with alive vmalloc area.
1914 *
1915 * This function checks that addr is a valid vmalloc'ed area, and
1916 * copy data from that area to a given buffer. If the given memory range
1917 * of [addr...addr+count) includes some valid address, data is copied to
1918 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1919 * IOREMAP area is treated as memory hole and no copy is done.
1920 *
1921 * If [addr...addr+count) doesn't includes any intersects with alive
1922 * vm_struct area, returns 0.
1923 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1924 * the caller should guarantee KM_USER0 is not used.
1925 *
1926 * Note: In usual ops, vread() is never necessary because the caller
1927 * should know vmalloc() area is valid and can use memcpy().
1928 * This is for routines which have to access vmalloc area without
1929 * any informaion, as /dev/kmem.
1930 *
1931 */
1932
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933long vread(char *buf, char *addr, unsigned long count)
1934{
1935 struct vm_struct *tmp;
1936 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001937 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 unsigned long n;
1939
1940 /* Don't allow overflow */
1941 if ((unsigned long) addr + count < count)
1942 count = -(unsigned long) addr;
1943
1944 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001945 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 vaddr = (char *) tmp->addr;
1947 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1948 continue;
1949 while (addr < vaddr) {
1950 if (count == 0)
1951 goto finished;
1952 *buf = '\0';
1953 buf++;
1954 addr++;
1955 count--;
1956 }
1957 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001958 if (n > count)
1959 n = count;
1960 if (!(tmp->flags & VM_IOREMAP))
1961 aligned_vread(buf, addr, n);
1962 else /* IOREMAP area is treated as memory hole */
1963 memset(buf, 0, n);
1964 buf += n;
1965 addr += n;
1966 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 }
1968finished:
1969 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001970
1971 if (buf == buf_start)
1972 return 0;
1973 /* zero-fill memory holes */
1974 if (buf != buf_start + buflen)
1975 memset(buf, 0, buflen - (buf - buf_start));
1976
1977 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978}
1979
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001980/**
1981 * vwrite() - write vmalloc area in a safe way.
1982 * @buf: buffer for source data
1983 * @addr: vm address.
1984 * @count: number of bytes to be read.
1985 *
1986 * Returns # of bytes which addr and buf should be incresed.
1987 * (same number to @count).
1988 * If [addr...addr+count) doesn't includes any intersect with valid
1989 * vmalloc area, returns 0.
1990 *
1991 * This function checks that addr is a valid vmalloc'ed area, and
1992 * copy data from a buffer to the given addr. If specified range of
1993 * [addr...addr+count) includes some valid address, data is copied from
1994 * proper area of @buf. If there are memory holes, no copy to hole.
1995 * IOREMAP area is treated as memory hole and no copy is done.
1996 *
1997 * If [addr...addr+count) doesn't includes any intersects with alive
1998 * vm_struct area, returns 0.
1999 * @buf should be kernel's buffer. Because this function uses KM_USER0,
2000 * the caller should guarantee KM_USER0 is not used.
2001 *
2002 * Note: In usual ops, vwrite() is never necessary because the caller
2003 * should know vmalloc() area is valid and can use memcpy().
2004 * This is for routines which have to access vmalloc area without
2005 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002006 */
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008long vwrite(char *buf, char *addr, unsigned long count)
2009{
2010 struct vm_struct *tmp;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002011 char *vaddr;
2012 unsigned long n, buflen;
2013 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
2015 /* Don't allow overflow */
2016 if ((unsigned long) addr + count < count)
2017 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002018 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
2020 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002021 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 vaddr = (char *) tmp->addr;
2023 if (addr >= vaddr + tmp->size - PAGE_SIZE)
2024 continue;
2025 while (addr < vaddr) {
2026 if (count == 0)
2027 goto finished;
2028 buf++;
2029 addr++;
2030 count--;
2031 }
2032 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002033 if (n > count)
2034 n = count;
2035 if (!(tmp->flags & VM_IOREMAP)) {
2036 aligned_vwrite(buf, addr, n);
2037 copied++;
2038 }
2039 buf += n;
2040 addr += n;
2041 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 }
2043finished:
2044 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002045 if (!copied)
2046 return 0;
2047 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048}
Nick Piggin83342312006-06-23 02:03:20 -07002049
2050/**
2051 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002052 * @vma: vma to cover (map full range of vma)
2053 * @addr: vmalloc memory
2054 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002055 *
2056 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002057 *
2058 * This function checks that addr is a valid vmalloc'ed area, and
2059 * that it is big enough to cover the vma. Will return failure if
2060 * that criteria isn't met.
2061 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002062 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002063 */
2064int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2065 unsigned long pgoff)
2066{
2067 struct vm_struct *area;
2068 unsigned long uaddr = vma->vm_start;
2069 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07002070
2071 if ((PAGE_SIZE-1) & (unsigned long)addr)
2072 return -EINVAL;
2073
Nick Piggindb64fe02008-10-18 20:27:03 -07002074 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07002075 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07002076 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002077
2078 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07002079 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002080
2081 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07002082 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002083
2084 addr += pgoff << PAGE_SHIFT;
2085 do {
2086 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07002087 int ret;
2088
Nick Piggin83342312006-06-23 02:03:20 -07002089 ret = vm_insert_page(vma, uaddr, page);
2090 if (ret)
2091 return ret;
2092
2093 uaddr += PAGE_SIZE;
2094 addr += PAGE_SIZE;
2095 usize -= PAGE_SIZE;
2096 } while (usize > 0);
2097
2098 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
2099 vma->vm_flags |= VM_RESERVED;
2100
Nick Piggindb64fe02008-10-18 20:27:03 -07002101 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07002102}
2103EXPORT_SYMBOL(remap_vmalloc_range);
2104
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002105/*
2106 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2107 * have one.
2108 */
2109void __attribute__((weak)) vmalloc_sync_all(void)
2110{
2111}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002112
2113
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002114static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002115{
2116 /* apply_to_page_range() does all the hard work. */
2117 return 0;
2118}
2119
2120/**
2121 * alloc_vm_area - allocate a range of kernel address space
2122 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07002123 *
2124 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002125 *
2126 * This function reserves a range of kernel address space, and
2127 * allocates pagetables to map that range. No actual mappings
2128 * are created. If the kernel address space is not shared
2129 * between processes, it syncs the pagetable across all
2130 * processes.
2131 */
2132struct vm_struct *alloc_vm_area(size_t size)
2133{
2134 struct vm_struct *area;
2135
Christoph Lameter23016962008-04-28 02:12:42 -07002136 area = get_vm_area_caller(size, VM_IOREMAP,
2137 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002138 if (area == NULL)
2139 return NULL;
2140
2141 /*
2142 * This ensures that page tables are constructed for this region
2143 * of kernel virtual address space and mapped into init_mm.
2144 */
2145 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2146 area->size, f, NULL)) {
2147 free_vm_area(area);
2148 return NULL;
2149 }
2150
2151 /* Make sure the pagetables are constructed in process kernel
2152 mappings */
2153 vmalloc_sync_all();
2154
2155 return area;
2156}
2157EXPORT_SYMBOL_GPL(alloc_vm_area);
2158
2159void free_vm_area(struct vm_struct *area)
2160{
2161 struct vm_struct *ret;
2162 ret = remove_vm_area(area->addr);
2163 BUG_ON(ret != area);
2164 kfree(area);
2165}
2166EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002167
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002168#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002169static struct vmap_area *node_to_va(struct rb_node *n)
2170{
2171 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2172}
2173
2174/**
2175 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2176 * @end: target address
2177 * @pnext: out arg for the next vmap_area
2178 * @pprev: out arg for the previous vmap_area
2179 *
2180 * Returns: %true if either or both of next and prev are found,
2181 * %false if no vmap_area exists
2182 *
2183 * Find vmap_areas end addresses of which enclose @end. ie. if not
2184 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2185 */
2186static bool pvm_find_next_prev(unsigned long end,
2187 struct vmap_area **pnext,
2188 struct vmap_area **pprev)
2189{
2190 struct rb_node *n = vmap_area_root.rb_node;
2191 struct vmap_area *va = NULL;
2192
2193 while (n) {
2194 va = rb_entry(n, struct vmap_area, rb_node);
2195 if (end < va->va_end)
2196 n = n->rb_left;
2197 else if (end > va->va_end)
2198 n = n->rb_right;
2199 else
2200 break;
2201 }
2202
2203 if (!va)
2204 return false;
2205
2206 if (va->va_end > end) {
2207 *pnext = va;
2208 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2209 } else {
2210 *pprev = va;
2211 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2212 }
2213 return true;
2214}
2215
2216/**
2217 * pvm_determine_end - find the highest aligned address between two vmap_areas
2218 * @pnext: in/out arg for the next vmap_area
2219 * @pprev: in/out arg for the previous vmap_area
2220 * @align: alignment
2221 *
2222 * Returns: determined end address
2223 *
2224 * Find the highest aligned address between *@pnext and *@pprev below
2225 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2226 * down address is between the end addresses of the two vmap_areas.
2227 *
2228 * Please note that the address returned by this function may fall
2229 * inside *@pnext vmap_area. The caller is responsible for checking
2230 * that.
2231 */
2232static unsigned long pvm_determine_end(struct vmap_area **pnext,
2233 struct vmap_area **pprev,
2234 unsigned long align)
2235{
2236 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2237 unsigned long addr;
2238
2239 if (*pnext)
2240 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2241 else
2242 addr = vmalloc_end;
2243
2244 while (*pprev && (*pprev)->va_end > addr) {
2245 *pnext = *pprev;
2246 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2247 }
2248
2249 return addr;
2250}
2251
2252/**
2253 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2254 * @offsets: array containing offset of each area
2255 * @sizes: array containing size of each area
2256 * @nr_vms: the number of areas to allocate
2257 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002258 *
2259 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2260 * vm_structs on success, %NULL on failure
2261 *
2262 * Percpu allocator wants to use congruent vm areas so that it can
2263 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002264 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2265 * be scattered pretty far, distance between two areas easily going up
2266 * to gigabytes. To avoid interacting with regular vmallocs, these
2267 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002268 *
2269 * Despite its complicated look, this allocator is rather simple. It
2270 * does everything top-down and scans areas from the end looking for
2271 * matching slot. While scanning, if any of the areas overlaps with
2272 * existing vmap_area, the base address is pulled down to fit the
2273 * area. Scanning is repeated till all the areas fit and then all
2274 * necessary data structres are inserted and the result is returned.
2275 */
2276struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2277 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002278 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002279{
2280 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2281 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2282 struct vmap_area **vas, *prev, *next;
2283 struct vm_struct **vms;
2284 int area, area2, last_area, term_area;
2285 unsigned long base, start, end, last_end;
2286 bool purged = false;
2287
Tejun Heoca23e402009-08-14 15:00:52 +09002288 /* verify parameters and allocate data structures */
2289 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2290 for (last_area = 0, area = 0; area < nr_vms; area++) {
2291 start = offsets[area];
2292 end = start + sizes[area];
2293
2294 /* is everything aligned properly? */
2295 BUG_ON(!IS_ALIGNED(offsets[area], align));
2296 BUG_ON(!IS_ALIGNED(sizes[area], align));
2297
2298 /* detect the area with the highest address */
2299 if (start > offsets[last_area])
2300 last_area = area;
2301
2302 for (area2 = 0; area2 < nr_vms; area2++) {
2303 unsigned long start2 = offsets[area2];
2304 unsigned long end2 = start2 + sizes[area2];
2305
2306 if (area2 == area)
2307 continue;
2308
2309 BUG_ON(start2 >= start && start2 < end);
2310 BUG_ON(end2 <= end && end2 > start);
2311 }
2312 }
2313 last_end = offsets[last_area] + sizes[last_area];
2314
2315 if (vmalloc_end - vmalloc_start < last_end) {
2316 WARN_ON(true);
2317 return NULL;
2318 }
2319
David Rientjesec3f64f2011-01-13 15:46:01 -08002320 vms = kzalloc(sizeof(vms[0]) * nr_vms, GFP_KERNEL);
2321 vas = kzalloc(sizeof(vas[0]) * nr_vms, GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002322 if (!vas || !vms)
2323 goto err_free;
2324
2325 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002326 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2327 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002328 if (!vas[area] || !vms[area])
2329 goto err_free;
2330 }
2331retry:
2332 spin_lock(&vmap_area_lock);
2333
2334 /* start scanning - we scan from the top, begin with the last area */
2335 area = term_area = last_area;
2336 start = offsets[area];
2337 end = start + sizes[area];
2338
2339 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2340 base = vmalloc_end - last_end;
2341 goto found;
2342 }
2343 base = pvm_determine_end(&next, &prev, align) - end;
2344
2345 while (true) {
2346 BUG_ON(next && next->va_end <= base + end);
2347 BUG_ON(prev && prev->va_end > base + end);
2348
2349 /*
2350 * base might have underflowed, add last_end before
2351 * comparing.
2352 */
2353 if (base + last_end < vmalloc_start + last_end) {
2354 spin_unlock(&vmap_area_lock);
2355 if (!purged) {
2356 purge_vmap_area_lazy();
2357 purged = true;
2358 goto retry;
2359 }
2360 goto err_free;
2361 }
2362
2363 /*
2364 * If next overlaps, move base downwards so that it's
2365 * right below next and then recheck.
2366 */
2367 if (next && next->va_start < base + end) {
2368 base = pvm_determine_end(&next, &prev, align) - end;
2369 term_area = area;
2370 continue;
2371 }
2372
2373 /*
2374 * If prev overlaps, shift down next and prev and move
2375 * base so that it's right below new next and then
2376 * recheck.
2377 */
2378 if (prev && prev->va_end > base + start) {
2379 next = prev;
2380 prev = node_to_va(rb_prev(&next->rb_node));
2381 base = pvm_determine_end(&next, &prev, align) - end;
2382 term_area = area;
2383 continue;
2384 }
2385
2386 /*
2387 * This area fits, move on to the previous one. If
2388 * the previous one is the terminal one, we're done.
2389 */
2390 area = (area + nr_vms - 1) % nr_vms;
2391 if (area == term_area)
2392 break;
2393 start = offsets[area];
2394 end = start + sizes[area];
2395 pvm_find_next_prev(base + end, &next, &prev);
2396 }
2397found:
2398 /* we've found a fitting base, insert all va's */
2399 for (area = 0; area < nr_vms; area++) {
2400 struct vmap_area *va = vas[area];
2401
2402 va->va_start = base + offsets[area];
2403 va->va_end = va->va_start + sizes[area];
2404 __insert_vmap_area(va);
2405 }
2406
2407 vmap_area_pcpu_hole = base + offsets[last_area];
2408
2409 spin_unlock(&vmap_area_lock);
2410
2411 /* insert all vm's */
2412 for (area = 0; area < nr_vms; area++)
2413 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2414 pcpu_get_vm_areas);
2415
2416 kfree(vas);
2417 return vms;
2418
2419err_free:
2420 for (area = 0; area < nr_vms; area++) {
2421 if (vas)
2422 kfree(vas[area]);
2423 if (vms)
2424 kfree(vms[area]);
2425 }
2426 kfree(vas);
2427 kfree(vms);
2428 return NULL;
2429}
2430
2431/**
2432 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2433 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2434 * @nr_vms: the number of allocated areas
2435 *
2436 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2437 */
2438void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2439{
2440 int i;
2441
2442 for (i = 0; i < nr_vms; i++)
2443 free_vm_area(vms[i]);
2444 kfree(vms);
2445}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002446#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002447
2448#ifdef CONFIG_PROC_FS
2449static void *s_start(struct seq_file *m, loff_t *pos)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002450 __acquires(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002451{
2452 loff_t n = *pos;
2453 struct vm_struct *v;
2454
2455 read_lock(&vmlist_lock);
2456 v = vmlist;
2457 while (n > 0 && v) {
2458 n--;
2459 v = v->next;
2460 }
2461 if (!n)
2462 return v;
2463
2464 return NULL;
2465
2466}
2467
2468static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2469{
2470 struct vm_struct *v = p;
2471
2472 ++*pos;
2473 return v->next;
2474}
2475
2476static void s_stop(struct seq_file *m, void *p)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002477 __releases(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002478{
2479 read_unlock(&vmlist_lock);
2480}
2481
Eric Dumazeta47a1262008-07-23 21:27:38 -07002482static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2483{
2484 if (NUMA_BUILD) {
2485 unsigned int nr, *counters = m->private;
2486
2487 if (!counters)
2488 return;
2489
2490 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2491
2492 for (nr = 0; nr < v->nr_pages; nr++)
2493 counters[page_to_nid(v->pages[nr])]++;
2494
2495 for_each_node_state(nr, N_HIGH_MEMORY)
2496 if (counters[nr])
2497 seq_printf(m, " N%u=%u", nr, counters[nr]);
2498 }
2499}
2500
Christoph Lametera10aa572008-04-28 02:12:40 -07002501static int s_show(struct seq_file *m, void *p)
2502{
2503 struct vm_struct *v = p;
2504
2505 seq_printf(m, "0x%p-0x%p %7ld",
2506 v->addr, v->addr + v->size, v->size);
2507
Joe Perches62c70bc2011-01-13 15:45:52 -08002508 if (v->caller)
2509 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002510
Christoph Lametera10aa572008-04-28 02:12:40 -07002511 if (v->nr_pages)
2512 seq_printf(m, " pages=%d", v->nr_pages);
2513
2514 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002515 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002516
2517 if (v->flags & VM_IOREMAP)
2518 seq_printf(m, " ioremap");
2519
2520 if (v->flags & VM_ALLOC)
2521 seq_printf(m, " vmalloc");
2522
2523 if (v->flags & VM_MAP)
2524 seq_printf(m, " vmap");
2525
2526 if (v->flags & VM_USERMAP)
2527 seq_printf(m, " user");
2528
2529 if (v->flags & VM_VPAGES)
2530 seq_printf(m, " vpages");
2531
Eric Dumazeta47a1262008-07-23 21:27:38 -07002532 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002533 seq_putc(m, '\n');
2534 return 0;
2535}
2536
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002537static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002538 .start = s_start,
2539 .next = s_next,
2540 .stop = s_stop,
2541 .show = s_show,
2542};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002543
2544static int vmalloc_open(struct inode *inode, struct file *file)
2545{
2546 unsigned int *ptr = NULL;
2547 int ret;
2548
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002549 if (NUMA_BUILD) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002550 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002551 if (ptr == NULL)
2552 return -ENOMEM;
2553 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002554 ret = seq_open(file, &vmalloc_op);
2555 if (!ret) {
2556 struct seq_file *m = file->private_data;
2557 m->private = ptr;
2558 } else
2559 kfree(ptr);
2560 return ret;
2561}
2562
2563static const struct file_operations proc_vmalloc_operations = {
2564 .open = vmalloc_open,
2565 .read = seq_read,
2566 .llseek = seq_lseek,
2567 .release = seq_release_private,
2568};
2569
2570static int __init proc_vmalloc_init(void)
2571{
2572 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2573 return 0;
2574}
2575module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07002576#endif
2577