blob: 464621d18eb249fde1403741e0f75b193aa0c18a [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>
Arun Sharma60063492011-07-26 16:09:06 -070029#include <linux/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);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700378 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700379 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 */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700409 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700410 if (addr + cached_hole_size < first->va_start)
411 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700412 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700413 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
Nick Piggindb64fe02008-10-18 20:27:03 -0700455static void __free_vmap_area(struct vmap_area *va)
456{
457 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700458
459 if (free_vmap_cache) {
460 if (va->va_end < cached_vstart) {
461 free_vmap_cache = NULL;
462 } else {
463 struct vmap_area *cache;
464 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
465 if (va->va_start <= cache->va_start) {
466 free_vmap_cache = rb_prev(&va->rb_node);
467 /*
468 * We don't try to update cached_hole_size or
469 * cached_align, but it won't go very wrong.
470 */
471 }
472 }
473 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700474 rb_erase(&va->rb_node, &vmap_area_root);
475 RB_CLEAR_NODE(&va->rb_node);
476 list_del_rcu(&va->list);
477
Tejun Heoca23e402009-08-14 15:00:52 +0900478 /*
479 * Track the highest possible candidate for pcpu area
480 * allocation. Areas outside of vmalloc area can be returned
481 * here too, consider only end addresses which fall inside
482 * vmalloc area proper.
483 */
484 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
485 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
486
Lai Jiangshan14769de2011-03-18 12:12:19 +0800487 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700488}
489
490/*
491 * Free a region of KVA allocated by alloc_vmap_area
492 */
493static void free_vmap_area(struct vmap_area *va)
494{
495 spin_lock(&vmap_area_lock);
496 __free_vmap_area(va);
497 spin_unlock(&vmap_area_lock);
498}
499
500/*
501 * Clear the pagetable entries of a given vmap_area
502 */
503static void unmap_vmap_area(struct vmap_area *va)
504{
505 vunmap_page_range(va->va_start, va->va_end);
506}
507
Nick Piggincd528582009-01-06 14:39:20 -0800508static void vmap_debug_free_range(unsigned long start, unsigned long end)
509{
510 /*
511 * Unmap page tables and force a TLB flush immediately if
512 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
513 * bugs similarly to those in linear kernel virtual address
514 * space after a page has been freed.
515 *
516 * All the lazy freeing logic is still retained, in order to
517 * minimise intrusiveness of this debugging feature.
518 *
519 * This is going to be *slow* (linear kernel virtual address
520 * debugging doesn't do a broadcast TLB flush so it is a lot
521 * faster).
522 */
523#ifdef CONFIG_DEBUG_PAGEALLOC
524 vunmap_page_range(start, end);
525 flush_tlb_kernel_range(start, end);
526#endif
527}
528
Nick Piggindb64fe02008-10-18 20:27:03 -0700529/*
530 * lazy_max_pages is the maximum amount of virtual address space we gather up
531 * before attempting to purge with a TLB flush.
532 *
533 * There is a tradeoff here: a larger number will cover more kernel page tables
534 * and take slightly longer to purge, but it will linearly reduce the number of
535 * global TLB flushes that must be performed. It would seem natural to scale
536 * this number up linearly with the number of CPUs (because vmapping activity
537 * could also scale linearly with the number of CPUs), however it is likely
538 * that in practice, workloads might be constrained in other ways that mean
539 * vmap activity will not scale linearly with CPUs. Also, I want to be
540 * conservative and not introduce a big latency on huge systems, so go with
541 * a less aggressive log scale. It will still be an improvement over the old
542 * code, and it will be simple to change the scale factor if we find that it
543 * becomes a problem on bigger systems.
544 */
545static unsigned long lazy_max_pages(void)
546{
547 unsigned int log;
548
549 log = fls(num_online_cpus());
550
551 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
552}
553
554static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
555
Nick Piggin02b709d2010-02-01 22:25:57 +1100556/* for per-CPU blocks */
557static void purge_fragmented_blocks_allcpus(void);
558
Nick Piggindb64fe02008-10-18 20:27:03 -0700559/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500560 * called before a call to iounmap() if the caller wants vm_area_struct's
561 * immediately freed.
562 */
563void set_iounmap_nonlazy(void)
564{
565 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
566}
567
568/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700569 * Purges all lazily-freed vmap areas.
570 *
571 * If sync is 0 then don't purge if there is already a purge in progress.
572 * If force_flush is 1, then flush kernel TLBs between *start and *end even
573 * if we found no lazy vmap areas to unmap (callers can use this to optimise
574 * their own TLB flushing).
575 * Returns with *start = min(*start, lowest purged address)
576 * *end = max(*end, highest purged address)
577 */
578static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
579 int sync, int force_flush)
580{
Andrew Morton46666d82009-01-15 13:51:15 -0800581 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700582 LIST_HEAD(valist);
583 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800584 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700585 int nr = 0;
586
587 /*
588 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
589 * should not expect such behaviour. This just simplifies locking for
590 * the case that isn't actually used at the moment anyway.
591 */
592 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800593 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700594 return;
595 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800596 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700597
Nick Piggin02b709d2010-02-01 22:25:57 +1100598 if (sync)
599 purge_fragmented_blocks_allcpus();
600
Nick Piggindb64fe02008-10-18 20:27:03 -0700601 rcu_read_lock();
602 list_for_each_entry_rcu(va, &vmap_area_list, list) {
603 if (va->flags & VM_LAZY_FREE) {
604 if (va->va_start < *start)
605 *start = va->va_start;
606 if (va->va_end > *end)
607 *end = va->va_end;
608 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700609 list_add_tail(&va->purge_list, &valist);
610 va->flags |= VM_LAZY_FREEING;
611 va->flags &= ~VM_LAZY_FREE;
612 }
613 }
614 rcu_read_unlock();
615
Yongseok Koh88f50042010-01-19 17:33:49 +0900616 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700617 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700618
619 if (nr || force_flush)
620 flush_tlb_kernel_range(*start, *end);
621
622 if (nr) {
623 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800624 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700625 __free_vmap_area(va);
626 spin_unlock(&vmap_area_lock);
627 }
Andrew Morton46666d82009-01-15 13:51:15 -0800628 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700629}
630
631/*
Nick Piggin496850e2008-11-19 15:36:33 -0800632 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
633 * is already purging.
634 */
635static void try_purge_vmap_area_lazy(void)
636{
637 unsigned long start = ULONG_MAX, end = 0;
638
639 __purge_vmap_area_lazy(&start, &end, 0, 0);
640}
641
642/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700643 * Kick off a purge of the outstanding lazy areas.
644 */
645static void purge_vmap_area_lazy(void)
646{
647 unsigned long start = ULONG_MAX, end = 0;
648
Nick Piggin496850e2008-11-19 15:36:33 -0800649 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700650}
651
652/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800653 * Free a vmap area, caller ensuring that the area has been unmapped
654 * and flush_cache_vunmap had been called for the correct range
655 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700656 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800657static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700658{
659 va->flags |= VM_LAZY_FREE;
660 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
661 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800662 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700663}
664
Nick Pigginb29acbd2008-12-01 13:13:47 -0800665/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800666 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
667 * called for the correct range previously.
668 */
669static void free_unmap_vmap_area_noflush(struct vmap_area *va)
670{
671 unmap_vmap_area(va);
672 free_vmap_area_noflush(va);
673}
674
675/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800676 * Free and unmap a vmap area
677 */
678static void free_unmap_vmap_area(struct vmap_area *va)
679{
680 flush_cache_vunmap(va->va_start, va->va_end);
681 free_unmap_vmap_area_noflush(va);
682}
683
Nick Piggindb64fe02008-10-18 20:27:03 -0700684static struct vmap_area *find_vmap_area(unsigned long addr)
685{
686 struct vmap_area *va;
687
688 spin_lock(&vmap_area_lock);
689 va = __find_vmap_area(addr);
690 spin_unlock(&vmap_area_lock);
691
692 return va;
693}
694
695static void free_unmap_vmap_area_addr(unsigned long addr)
696{
697 struct vmap_area *va;
698
699 va = find_vmap_area(addr);
700 BUG_ON(!va);
701 free_unmap_vmap_area(va);
702}
703
704
705/*** Per cpu kva allocator ***/
706
707/*
708 * vmap space is limited especially on 32 bit architectures. Ensure there is
709 * room for at least 16 percpu vmap blocks per CPU.
710 */
711/*
712 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
713 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
714 * instead (we just need a rough idea)
715 */
716#if BITS_PER_LONG == 32
717#define VMALLOC_SPACE (128UL*1024*1024)
718#else
719#define VMALLOC_SPACE (128UL*1024*1024*1024)
720#endif
721
722#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
723#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
724#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
725#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
726#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
727#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
728#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
729 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
730 VMALLOC_PAGES / NR_CPUS / 16))
731
732#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
733
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100734static bool vmap_initialized __read_mostly = false;
735
Nick Piggindb64fe02008-10-18 20:27:03 -0700736struct vmap_block_queue {
737 spinlock_t lock;
738 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700739};
740
741struct vmap_block {
742 spinlock_t lock;
743 struct vmap_area *va;
744 struct vmap_block_queue *vbq;
745 unsigned long free, dirty;
746 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
747 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100748 struct list_head free_list;
749 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100750 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700751};
752
753/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
754static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
755
756/*
757 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
758 * in the free path. Could get rid of this if we change the API to return a
759 * "cookie" from alloc, to be passed to free. But no big deal yet.
760 */
761static DEFINE_SPINLOCK(vmap_block_tree_lock);
762static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
763
764/*
765 * We should probably have a fallback mechanism to allocate virtual memory
766 * out of partially filled vmap blocks. However vmap block sizing should be
767 * fairly reasonable according to the vmalloc size, so it shouldn't be a
768 * big problem.
769 */
770
771static unsigned long addr_to_vb_idx(unsigned long addr)
772{
773 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
774 addr /= VMAP_BLOCK_SIZE;
775 return addr;
776}
777
778static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
779{
780 struct vmap_block_queue *vbq;
781 struct vmap_block *vb;
782 struct vmap_area *va;
783 unsigned long vb_idx;
784 int node, err;
785
786 node = numa_node_id();
787
788 vb = kmalloc_node(sizeof(struct vmap_block),
789 gfp_mask & GFP_RECLAIM_MASK, node);
790 if (unlikely(!vb))
791 return ERR_PTR(-ENOMEM);
792
793 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
794 VMALLOC_START, VMALLOC_END,
795 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800796 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700797 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700798 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700799 }
800
801 err = radix_tree_preload(gfp_mask);
802 if (unlikely(err)) {
803 kfree(vb);
804 free_vmap_area(va);
805 return ERR_PTR(err);
806 }
807
808 spin_lock_init(&vb->lock);
809 vb->va = va;
810 vb->free = VMAP_BBMAP_BITS;
811 vb->dirty = 0;
812 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
813 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
814 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700815
816 vb_idx = addr_to_vb_idx(va->va_start);
817 spin_lock(&vmap_block_tree_lock);
818 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
819 spin_unlock(&vmap_block_tree_lock);
820 BUG_ON(err);
821 radix_tree_preload_end();
822
823 vbq = &get_cpu_var(vmap_block_queue);
824 vb->vbq = vbq;
825 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100826 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700827 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900828 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700829
830 return vb;
831}
832
Nick Piggindb64fe02008-10-18 20:27:03 -0700833static void free_vmap_block(struct vmap_block *vb)
834{
835 struct vmap_block *tmp;
836 unsigned long vb_idx;
837
Nick Piggindb64fe02008-10-18 20:27:03 -0700838 vb_idx = addr_to_vb_idx(vb->va->va_start);
839 spin_lock(&vmap_block_tree_lock);
840 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
841 spin_unlock(&vmap_block_tree_lock);
842 BUG_ON(tmp != vb);
843
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800844 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800845 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700846}
847
Nick Piggin02b709d2010-02-01 22:25:57 +1100848static void purge_fragmented_blocks(int cpu)
849{
850 LIST_HEAD(purge);
851 struct vmap_block *vb;
852 struct vmap_block *n_vb;
853 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
854
855 rcu_read_lock();
856 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
857
858 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
859 continue;
860
861 spin_lock(&vb->lock);
862 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
863 vb->free = 0; /* prevent further allocs after releasing lock */
864 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
865 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
866 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
867 spin_lock(&vbq->lock);
868 list_del_rcu(&vb->free_list);
869 spin_unlock(&vbq->lock);
870 spin_unlock(&vb->lock);
871 list_add_tail(&vb->purge, &purge);
872 } else
873 spin_unlock(&vb->lock);
874 }
875 rcu_read_unlock();
876
877 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
878 list_del(&vb->purge);
879 free_vmap_block(vb);
880 }
881}
882
883static void purge_fragmented_blocks_thiscpu(void)
884{
885 purge_fragmented_blocks(smp_processor_id());
886}
887
888static void purge_fragmented_blocks_allcpus(void)
889{
890 int cpu;
891
892 for_each_possible_cpu(cpu)
893 purge_fragmented_blocks(cpu);
894}
895
Nick Piggindb64fe02008-10-18 20:27:03 -0700896static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
897{
898 struct vmap_block_queue *vbq;
899 struct vmap_block *vb;
900 unsigned long addr = 0;
901 unsigned int order;
Nick Piggin02b709d2010-02-01 22:25:57 +1100902 int purge = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700903
904 BUG_ON(size & ~PAGE_MASK);
905 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
906 order = get_order(size);
907
908again:
909 rcu_read_lock();
910 vbq = &get_cpu_var(vmap_block_queue);
911 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
912 int i;
913
914 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100915 if (vb->free < 1UL << order)
916 goto next;
917
Nick Piggindb64fe02008-10-18 20:27:03 -0700918 i = bitmap_find_free_region(vb->alloc_map,
919 VMAP_BBMAP_BITS, order);
920
Nick Piggin02b709d2010-02-01 22:25:57 +1100921 if (i < 0) {
922 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
923 /* fragmented and no outstanding allocations */
924 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
925 purge = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -0700926 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100927 goto next;
928 }
929 addr = vb->va->va_start + (i << PAGE_SHIFT);
930 BUG_ON(addr_to_vb_idx(addr) !=
931 addr_to_vb_idx(vb->va->va_start));
932 vb->free -= 1UL << order;
933 if (vb->free == 0) {
934 spin_lock(&vbq->lock);
935 list_del_rcu(&vb->free_list);
936 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700937 }
938 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100939 break;
940next:
941 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700942 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100943
944 if (purge)
945 purge_fragmented_blocks_thiscpu();
946
Tejun Heo3f04ba82009-10-29 22:34:12 +0900947 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700948 rcu_read_unlock();
949
950 if (!addr) {
951 vb = new_vmap_block(gfp_mask);
952 if (IS_ERR(vb))
953 return vb;
954 goto again;
955 }
956
957 return (void *)addr;
958}
959
960static void vb_free(const void *addr, unsigned long size)
961{
962 unsigned long offset;
963 unsigned long vb_idx;
964 unsigned int order;
965 struct vmap_block *vb;
966
967 BUG_ON(size & ~PAGE_MASK);
968 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800969
970 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
971
Nick Piggindb64fe02008-10-18 20:27:03 -0700972 order = get_order(size);
973
974 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
975
976 vb_idx = addr_to_vb_idx((unsigned long)addr);
977 rcu_read_lock();
978 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
979 rcu_read_unlock();
980 BUG_ON(!vb);
981
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800982 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
983
Nick Piggindb64fe02008-10-18 20:27:03 -0700984 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100985 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700986
Nick Piggindb64fe02008-10-18 20:27:03 -0700987 vb->dirty += 1UL << order;
988 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100989 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700990 spin_unlock(&vb->lock);
991 free_vmap_block(vb);
992 } else
993 spin_unlock(&vb->lock);
994}
995
996/**
997 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
998 *
999 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1000 * to amortize TLB flushing overheads. What this means is that any page you
1001 * have now, may, in a former life, have been mapped into kernel virtual
1002 * address by the vmap layer and so there might be some CPUs with TLB entries
1003 * still referencing that page (additional to the regular 1:1 kernel mapping).
1004 *
1005 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1006 * be sure that none of the pages we have control over will have any aliases
1007 * from the vmap layer.
1008 */
1009void vm_unmap_aliases(void)
1010{
1011 unsigned long start = ULONG_MAX, end = 0;
1012 int cpu;
1013 int flush = 0;
1014
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001015 if (unlikely(!vmap_initialized))
1016 return;
1017
Nick Piggindb64fe02008-10-18 20:27:03 -07001018 for_each_possible_cpu(cpu) {
1019 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1020 struct vmap_block *vb;
1021
1022 rcu_read_lock();
1023 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1024 int i;
1025
1026 spin_lock(&vb->lock);
1027 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1028 while (i < VMAP_BBMAP_BITS) {
1029 unsigned long s, e;
1030 int j;
1031 j = find_next_zero_bit(vb->dirty_map,
1032 VMAP_BBMAP_BITS, i);
1033
1034 s = vb->va->va_start + (i << PAGE_SHIFT);
1035 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001036 flush = 1;
1037
1038 if (s < start)
1039 start = s;
1040 if (e > end)
1041 end = e;
1042
1043 i = j;
1044 i = find_next_bit(vb->dirty_map,
1045 VMAP_BBMAP_BITS, i);
1046 }
1047 spin_unlock(&vb->lock);
1048 }
1049 rcu_read_unlock();
1050 }
1051
1052 __purge_vmap_area_lazy(&start, &end, 1, flush);
1053}
1054EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1055
1056/**
1057 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1058 * @mem: the pointer returned by vm_map_ram
1059 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1060 */
1061void vm_unmap_ram(const void *mem, unsigned int count)
1062{
1063 unsigned long size = count << PAGE_SHIFT;
1064 unsigned long addr = (unsigned long)mem;
1065
1066 BUG_ON(!addr);
1067 BUG_ON(addr < VMALLOC_START);
1068 BUG_ON(addr > VMALLOC_END);
1069 BUG_ON(addr & (PAGE_SIZE-1));
1070
1071 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001072 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001073
1074 if (likely(count <= VMAP_MAX_ALLOC))
1075 vb_free(mem, size);
1076 else
1077 free_unmap_vmap_area_addr(addr);
1078}
1079EXPORT_SYMBOL(vm_unmap_ram);
1080
1081/**
1082 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1083 * @pages: an array of pointers to the pages to be mapped
1084 * @count: number of pages
1085 * @node: prefer to allocate data structures on this node
1086 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001087 *
1088 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001089 */
1090void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1091{
1092 unsigned long size = count << PAGE_SHIFT;
1093 unsigned long addr;
1094 void *mem;
1095
1096 if (likely(count <= VMAP_MAX_ALLOC)) {
1097 mem = vb_alloc(size, GFP_KERNEL);
1098 if (IS_ERR(mem))
1099 return NULL;
1100 addr = (unsigned long)mem;
1101 } else {
1102 struct vmap_area *va;
1103 va = alloc_vmap_area(size, PAGE_SIZE,
1104 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1105 if (IS_ERR(va))
1106 return NULL;
1107
1108 addr = va->va_start;
1109 mem = (void *)addr;
1110 }
1111 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1112 vm_unmap_ram(mem, count);
1113 return NULL;
1114 }
1115 return mem;
1116}
1117EXPORT_SYMBOL(vm_map_ram);
1118
Tejun Heof0aa6612009-02-20 16:29:08 +09001119/**
1120 * vm_area_register_early - register vmap area early during boot
1121 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001122 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001123 *
1124 * This function is used to register kernel vm area before
1125 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1126 * proper values on entry and other fields should be zero. On return,
1127 * vm->addr contains the allocated address.
1128 *
1129 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1130 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001131void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001132{
1133 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001134 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001135
Tejun Heoc0c0a292009-02-24 11:57:21 +09001136 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1137 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1138
1139 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001140
1141 vm->next = vmlist;
1142 vmlist = vm;
1143}
1144
Nick Piggindb64fe02008-10-18 20:27:03 -07001145void __init vmalloc_init(void)
1146{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001147 struct vmap_area *va;
1148 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001149 int i;
1150
1151 for_each_possible_cpu(i) {
1152 struct vmap_block_queue *vbq;
1153
1154 vbq = &per_cpu(vmap_block_queue, i);
1155 spin_lock_init(&vbq->lock);
1156 INIT_LIST_HEAD(&vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001157 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001158
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001159 /* Import existing vmlist entries. */
1160 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001161 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001162 va->flags = tmp->flags | VM_VM_AREA;
1163 va->va_start = (unsigned long)tmp->addr;
1164 va->va_end = va->va_start + tmp->size;
1165 __insert_vmap_area(va);
1166 }
Tejun Heoca23e402009-08-14 15:00:52 +09001167
1168 vmap_area_pcpu_hole = VMALLOC_END;
1169
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001170 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001171}
1172
Tejun Heo8fc48982009-02-20 16:29:08 +09001173/**
1174 * map_kernel_range_noflush - map kernel VM area with the specified pages
1175 * @addr: start of the VM area to map
1176 * @size: size of the VM area to map
1177 * @prot: page protection flags to use
1178 * @pages: pages to map
1179 *
1180 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1181 * specify should have been allocated using get_vm_area() and its
1182 * friends.
1183 *
1184 * NOTE:
1185 * This function does NOT do any cache flushing. The caller is
1186 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1187 * before calling this function.
1188 *
1189 * RETURNS:
1190 * The number of pages mapped on success, -errno on failure.
1191 */
1192int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1193 pgprot_t prot, struct page **pages)
1194{
1195 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1196}
1197
1198/**
1199 * unmap_kernel_range_noflush - unmap kernel VM area
1200 * @addr: start of the VM area to unmap
1201 * @size: size of the VM area to unmap
1202 *
1203 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1204 * specify should have been allocated using get_vm_area() and its
1205 * friends.
1206 *
1207 * NOTE:
1208 * This function does NOT do any cache flushing. The caller is
1209 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1210 * before calling this function and flush_tlb_kernel_range() after.
1211 */
1212void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1213{
1214 vunmap_page_range(addr, addr + size);
1215}
Huang Ying81e88fd2011-01-12 14:44:55 +08001216EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001217
1218/**
1219 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1220 * @addr: start of the VM area to unmap
1221 * @size: size of the VM area to unmap
1222 *
1223 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1224 * the unmapping and tlb after.
1225 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001226void unmap_kernel_range(unsigned long addr, unsigned long size)
1227{
1228 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001229
1230 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001231 vunmap_page_range(addr, end);
1232 flush_tlb_kernel_range(addr, end);
1233}
1234
1235int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1236{
1237 unsigned long addr = (unsigned long)area->addr;
1238 unsigned long end = addr + area->size - PAGE_SIZE;
1239 int err;
1240
1241 err = vmap_page_range(addr, end, prot, *pages);
1242 if (err > 0) {
1243 *pages += err;
1244 err = 0;
1245 }
1246
1247 return err;
1248}
1249EXPORT_SYMBOL_GPL(map_vm_area);
1250
1251/*** Old vmalloc interfaces ***/
1252DEFINE_RWLOCK(vmlist_lock);
1253struct vm_struct *vmlist;
1254
Tejun Heocf88c792009-08-14 15:00:52 +09001255static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1256 unsigned long flags, void *caller)
1257{
1258 struct vm_struct *tmp, **p;
1259
1260 vm->flags = flags;
1261 vm->addr = (void *)va->va_start;
1262 vm->size = va->va_end - va->va_start;
1263 vm->caller = caller;
1264 va->private = vm;
1265 va->flags |= VM_VM_AREA;
1266
1267 write_lock(&vmlist_lock);
1268 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1269 if (tmp->addr >= vm->addr)
1270 break;
1271 }
1272 vm->next = *p;
1273 *p = vm;
1274 write_unlock(&vmlist_lock);
1275}
1276
Nick Piggindb64fe02008-10-18 20:27:03 -07001277static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001278 unsigned long align, unsigned long flags, unsigned long start,
1279 unsigned long end, int node, gfp_t gfp_mask, void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001280{
1281 static struct vmap_area *va;
1282 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001284 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 if (flags & VM_IOREMAP) {
1286 int bit = fls(size);
1287
1288 if (bit > IOREMAP_MAX_ORDER)
1289 bit = IOREMAP_MAX_ORDER;
1290 else if (bit < PAGE_SHIFT)
1291 bit = PAGE_SHIFT;
1292
1293 align = 1ul << bit;
1294 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001297 if (unlikely(!size))
1298 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Tejun Heocf88c792009-08-14 15:00:52 +09001300 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if (unlikely(!area))
1302 return NULL;
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /*
1305 * We always allocate a guard page.
1306 */
1307 size += PAGE_SIZE;
1308
Nick Piggindb64fe02008-10-18 20:27:03 -07001309 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1310 if (IS_ERR(va)) {
1311 kfree(area);
1312 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Tejun Heocf88c792009-08-14 15:00:52 +09001315 insert_vmalloc_vm(area, va, flags, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
Christoph Lameter930fc452005-10-29 18:15:41 -07001319struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1320 unsigned long start, unsigned long end)
1321{
David Miller2dca6992009-09-21 12:22:34 -07001322 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001323 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001324}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001325EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001326
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001327struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1328 unsigned long start, unsigned long end,
1329 void *caller)
1330{
David Miller2dca6992009-09-21 12:22:34 -07001331 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001332 caller);
1333}
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335/**
Simon Arlott183ff222007-10-20 01:27:18 +02001336 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * @size: size of the area
1338 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1339 *
1340 * Search an area of @size in the kernel virtual mapping area,
1341 * and reserved it for out purposes. Returns the area descriptor
1342 * on success or %NULL on failure.
1343 */
1344struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1345{
David Miller2dca6992009-09-21 12:22:34 -07001346 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001347 -1, GFP_KERNEL, __builtin_return_address(0));
1348}
1349
1350struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1351 void *caller)
1352{
David Miller2dca6992009-09-21 12:22:34 -07001353 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001354 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355}
1356
Nick Piggindb64fe02008-10-18 20:27:03 -07001357static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001358{
Nick Piggindb64fe02008-10-18 20:27:03 -07001359 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001360
Nick Piggindb64fe02008-10-18 20:27:03 -07001361 va = find_vmap_area((unsigned long)addr);
1362 if (va && va->flags & VM_VM_AREA)
1363 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001364
Andi Kleen7856dfe2005-05-20 14:27:57 -07001365 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001366}
1367
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368/**
Simon Arlott183ff222007-10-20 01:27:18 +02001369 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 * @addr: base address
1371 *
1372 * Search for the kernel VM area starting at @addr, and remove it.
1373 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001374 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001376struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
Nick Piggindb64fe02008-10-18 20:27:03 -07001378 struct vmap_area *va;
1379
1380 va = find_vmap_area((unsigned long)addr);
1381 if (va && va->flags & VM_VM_AREA) {
1382 struct vm_struct *vm = va->private;
1383 struct vm_struct *tmp, **p;
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001384 /*
1385 * remove from list and disallow access to this vm_struct
1386 * before unmap. (address range confliction is maintained by
1387 * vmap.)
1388 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001389 write_lock(&vmlist_lock);
1390 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1391 ;
1392 *p = tmp->next;
1393 write_unlock(&vmlist_lock);
1394
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001395 vmap_debug_free_range(va->va_start, va->va_end);
1396 free_unmap_vmap_area(va);
1397 vm->size -= PAGE_SIZE;
1398
Nick Piggindb64fe02008-10-18 20:27:03 -07001399 return vm;
1400 }
1401 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402}
1403
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001404static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405{
1406 struct vm_struct *area;
1407
1408 if (!addr)
1409 return;
1410
1411 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001412 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return;
1414 }
1415
1416 area = remove_vm_area(addr);
1417 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001418 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 return;
1421 }
1422
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001423 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001424 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (deallocate_pages) {
1427 int i;
1428
1429 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001430 struct page *page = area->pages[i];
1431
1432 BUG_ON(!page);
1433 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 }
1435
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001436 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 vfree(area->pages);
1438 else
1439 kfree(area->pages);
1440 }
1441
1442 kfree(area);
1443 return;
1444}
1445
1446/**
1447 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * @addr: memory base address
1449 *
Simon Arlott183ff222007-10-20 01:27:18 +02001450 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001451 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1452 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001454 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001456void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
1458 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001459
1460 kmemleak_free(addr);
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 __vunmap(addr, 1);
1463}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464EXPORT_SYMBOL(vfree);
1465
1466/**
1467 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 * @addr: memory base address
1469 *
1470 * Free the virtually contiguous memory area starting at @addr,
1471 * which was created from the page array passed to vmap().
1472 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001473 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001475void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476{
1477 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001478 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 __vunmap(addr, 0);
1480}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481EXPORT_SYMBOL(vunmap);
1482
1483/**
1484 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 * @pages: array of page pointers
1486 * @count: number of pages to map
1487 * @flags: vm_area->flags
1488 * @prot: page protection for the mapping
1489 *
1490 * Maps @count pages from @pages into contiguous kernel virtual
1491 * space.
1492 */
1493void *vmap(struct page **pages, unsigned int count,
1494 unsigned long flags, pgprot_t prot)
1495{
1496 struct vm_struct *area;
1497
Peter Zijlstra34754b62009-02-25 16:04:03 +01001498 might_sleep();
1499
Jan Beulich44813742009-09-21 17:03:05 -07001500 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 return NULL;
1502
Christoph Lameter23016962008-04-28 02:12:42 -07001503 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1504 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 if (!area)
1506 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (map_vm_area(area, prot, &pages)) {
1509 vunmap(area->addr);
1510 return NULL;
1511 }
1512
1513 return area->addr;
1514}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515EXPORT_SYMBOL(vmap);
1516
David Miller2dca6992009-09-21 12:22:34 -07001517static void *__vmalloc_node(unsigned long size, unsigned long align,
1518 gfp_t gfp_mask, pgprot_t prot,
Nick Piggindb64fe02008-10-18 20:27:03 -07001519 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001520static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001521 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
Dave Hansen22943ab2011-05-24 17:12:18 -07001523 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 struct page **pages;
1525 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001526 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1529 array_size = (nr_pages * sizeof(struct page *));
1530
1531 area->nr_pages = nr_pages;
1532 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001533 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001534 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Christoph Lameter23016962008-04-28 02:12:42 -07001535 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001536 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001537 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001538 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001539 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001541 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 if (!area->pages) {
1543 remove_vm_area(area->addr);
1544 kfree(area);
1545 return NULL;
1546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001549 struct page *page;
Dave Hansen22943ab2011-05-24 17:12:18 -07001550 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001551
Christoph Lameter930fc452005-10-29 18:15:41 -07001552 if (node < 0)
Dave Hansen22943ab2011-05-24 17:12:18 -07001553 page = alloc_page(tmp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001554 else
Dave Hansen22943ab2011-05-24 17:12:18 -07001555 page = alloc_pages_node(node, tmp_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001556
1557 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /* Successfully allocated i pages, free them in __vunmap() */
1559 area->nr_pages = i;
1560 goto fail;
1561 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001562 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 }
1564
1565 if (map_vm_area(area, prot, &pages))
1566 goto fail;
1567 return area->addr;
1568
1569fail:
Dave Hansen22943ab2011-05-24 17:12:18 -07001570 warn_alloc_failed(gfp_mask, order, "vmalloc: allocation failure, "
1571 "allocated %ld of %ld bytes\n",
1572 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 vfree(area->addr);
1574 return NULL;
1575}
1576
David Rientjesd0a21262011-01-13 15:46:02 -08001577/**
1578 * __vmalloc_node_range - allocate virtually contiguous memory
1579 * @size: allocation size
1580 * @align: desired alignment
1581 * @start: vm area range start
1582 * @end: vm area range end
1583 * @gfp_mask: flags for the page level allocator
1584 * @prot: protection mask for the allocated pages
1585 * @node: node to use for allocation or -1
1586 * @caller: caller's return address
1587 *
1588 * Allocate enough pages to cover @size from the page level
1589 * allocator with @gfp_mask flags. Map them into contiguous
1590 * kernel virtual space, using a pagetable protection of @prot.
1591 */
1592void *__vmalloc_node_range(unsigned long size, unsigned long align,
1593 unsigned long start, unsigned long end, gfp_t gfp_mask,
1594 pgprot_t prot, int node, void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001595{
David Rientjesd0a21262011-01-13 15:46:02 -08001596 struct vm_struct *area;
1597 void *addr;
1598 unsigned long real_size = size;
1599
1600 size = PAGE_ALIGN(size);
1601 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1602 return NULL;
1603
1604 area = __get_vm_area_node(size, align, VM_ALLOC, start, end, node,
1605 gfp_mask, caller);
1606
1607 if (!area)
1608 return NULL;
1609
1610 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Catalin Marinas89219d32009-06-11 13:23:19 +01001611
1612 /*
1613 * A ref_count = 3 is needed because the vm_struct and vmap_area
1614 * structures allocated in the __get_vm_area_node() function contain
1615 * references to the virtual address of the vmalloc'ed block.
1616 */
David Rientjesd0a21262011-01-13 15:46:02 -08001617 kmemleak_alloc(addr, real_size, 3, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001618
1619 return addr;
Christoph Lameter930fc452005-10-29 18:15:41 -07001620}
1621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001623 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001625 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 * @gfp_mask: flags for the page level allocator
1627 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001628 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001629 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 *
1631 * Allocate enough pages to cover @size from the page level
1632 * allocator with @gfp_mask flags. Map them into contiguous
1633 * kernel virtual space, using a pagetable protection of @prot.
1634 */
David Miller2dca6992009-09-21 12:22:34 -07001635static void *__vmalloc_node(unsigned long size, unsigned long align,
1636 gfp_t gfp_mask, pgprot_t prot,
1637 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
David Rientjesd0a21262011-01-13 15:46:02 -08001639 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1640 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641}
1642
Christoph Lameter930fc452005-10-29 18:15:41 -07001643void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1644{
David Miller2dca6992009-09-21 12:22:34 -07001645 return __vmalloc_node(size, 1, gfp_mask, prot, -1,
Christoph Lameter23016962008-04-28 02:12:42 -07001646 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001647}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648EXPORT_SYMBOL(__vmalloc);
1649
Dave Younge1ca7782010-10-26 14:22:06 -07001650static inline void *__vmalloc_node_flags(unsigned long size,
1651 int node, gfp_t flags)
1652{
1653 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1654 node, __builtin_return_address(0));
1655}
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657/**
1658 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 * Allocate enough pages to cover @size from the page level
1661 * allocator and map them into contiguous kernel virtual space.
1662 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001663 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 * use __vmalloc() instead.
1665 */
1666void *vmalloc(unsigned long size)
1667{
Dave Younge1ca7782010-10-26 14:22:06 -07001668 return __vmalloc_node_flags(size, -1, GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670EXPORT_SYMBOL(vmalloc);
1671
Christoph Lameter930fc452005-10-29 18:15:41 -07001672/**
Dave Younge1ca7782010-10-26 14:22:06 -07001673 * vzalloc - allocate virtually contiguous memory with zero fill
1674 * @size: allocation size
1675 * Allocate enough pages to cover @size from the page level
1676 * allocator and map them into contiguous kernel virtual space.
1677 * The memory allocated is set to zero.
1678 *
1679 * For tight control over page level allocator and protection flags
1680 * use __vmalloc() instead.
1681 */
1682void *vzalloc(unsigned long size)
1683{
1684 return __vmalloc_node_flags(size, -1,
1685 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1686}
1687EXPORT_SYMBOL(vzalloc);
1688
1689/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001690 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1691 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001692 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001693 * The resulting memory area is zeroed so it can be mapped to userspace
1694 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001695 */
1696void *vmalloc_user(unsigned long size)
1697{
1698 struct vm_struct *area;
1699 void *ret;
1700
David Miller2dca6992009-09-21 12:22:34 -07001701 ret = __vmalloc_node(size, SHMLBA,
1702 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
Glauber Costa84877842009-01-06 14:39:19 -08001703 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001704 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001705 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001706 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001707 }
Nick Piggin83342312006-06-23 02:03:20 -07001708 return ret;
1709}
1710EXPORT_SYMBOL(vmalloc_user);
1711
1712/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001713 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001714 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001715 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001716 *
1717 * Allocate enough pages to cover @size from the page level
1718 * allocator and map them into contiguous kernel virtual space.
1719 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001720 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001721 * use __vmalloc() instead.
1722 */
1723void *vmalloc_node(unsigned long size, int node)
1724{
David Miller2dca6992009-09-21 12:22:34 -07001725 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001726 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001727}
1728EXPORT_SYMBOL(vmalloc_node);
1729
Dave Younge1ca7782010-10-26 14:22:06 -07001730/**
1731 * vzalloc_node - allocate memory on a specific node with zero fill
1732 * @size: allocation size
1733 * @node: numa node
1734 *
1735 * Allocate enough pages to cover @size from the page level
1736 * allocator and map them into contiguous kernel virtual space.
1737 * The memory allocated is set to zero.
1738 *
1739 * For tight control over page level allocator and protection flags
1740 * use __vmalloc_node() instead.
1741 */
1742void *vzalloc_node(unsigned long size, int node)
1743{
1744 return __vmalloc_node_flags(size, node,
1745 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1746}
1747EXPORT_SYMBOL(vzalloc_node);
1748
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001749#ifndef PAGE_KERNEL_EXEC
1750# define PAGE_KERNEL_EXEC PAGE_KERNEL
1751#endif
1752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753/**
1754 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 * @size: allocation size
1756 *
1757 * Kernel-internal function to allocate enough pages to cover @size
1758 * the page level allocator and map them into contiguous and
1759 * executable kernel virtual space.
1760 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001761 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 * use __vmalloc() instead.
1763 */
1764
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765void *vmalloc_exec(unsigned long size)
1766{
David Miller2dca6992009-09-21 12:22:34 -07001767 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
Glauber Costa84877842009-01-06 14:39:19 -08001768 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769}
1770
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001771#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001772#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001773#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001774#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001775#else
1776#define GFP_VMALLOC32 GFP_KERNEL
1777#endif
1778
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779/**
1780 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 * @size: allocation size
1782 *
1783 * Allocate enough 32bit PA addressable pages to cover @size from the
1784 * page level allocator and map them into contiguous kernel virtual space.
1785 */
1786void *vmalloc_32(unsigned long size)
1787{
David Miller2dca6992009-09-21 12:22:34 -07001788 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001789 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791EXPORT_SYMBOL(vmalloc_32);
1792
Nick Piggin83342312006-06-23 02:03:20 -07001793/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001794 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001795 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001796 *
1797 * The resulting memory area is 32bit addressable and zeroed so it can be
1798 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001799 */
1800void *vmalloc_32_user(unsigned long size)
1801{
1802 struct vm_struct *area;
1803 void *ret;
1804
David Miller2dca6992009-09-21 12:22:34 -07001805 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001806 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001807 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001808 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001809 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001810 }
Nick Piggin83342312006-06-23 02:03:20 -07001811 return ret;
1812}
1813EXPORT_SYMBOL(vmalloc_32_user);
1814
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001815/*
1816 * small helper routine , copy contents to buf from addr.
1817 * If the page is not present, fill zero.
1818 */
1819
1820static int aligned_vread(char *buf, char *addr, unsigned long count)
1821{
1822 struct page *p;
1823 int copied = 0;
1824
1825 while (count) {
1826 unsigned long offset, length;
1827
1828 offset = (unsigned long)addr & ~PAGE_MASK;
1829 length = PAGE_SIZE - offset;
1830 if (length > count)
1831 length = count;
1832 p = vmalloc_to_page(addr);
1833 /*
1834 * To do safe access to this _mapped_ area, we need
1835 * lock. But adding lock here means that we need to add
1836 * overhead of vmalloc()/vfree() calles for this _debug_
1837 * interface, rarely used. Instead of that, we'll use
1838 * kmap() and get small overhead in this access function.
1839 */
1840 if (p) {
1841 /*
1842 * we can expect USER0 is not used (see vread/vwrite's
1843 * function description)
1844 */
1845 void *map = kmap_atomic(p, KM_USER0);
1846 memcpy(buf, map + offset, length);
1847 kunmap_atomic(map, KM_USER0);
1848 } else
1849 memset(buf, 0, length);
1850
1851 addr += length;
1852 buf += length;
1853 copied += length;
1854 count -= length;
1855 }
1856 return copied;
1857}
1858
1859static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1860{
1861 struct page *p;
1862 int copied = 0;
1863
1864 while (count) {
1865 unsigned long offset, length;
1866
1867 offset = (unsigned long)addr & ~PAGE_MASK;
1868 length = PAGE_SIZE - offset;
1869 if (length > count)
1870 length = count;
1871 p = vmalloc_to_page(addr);
1872 /*
1873 * To do safe access to this _mapped_ area, we need
1874 * lock. But adding lock here means that we need to add
1875 * overhead of vmalloc()/vfree() calles for this _debug_
1876 * interface, rarely used. Instead of that, we'll use
1877 * kmap() and get small overhead in this access function.
1878 */
1879 if (p) {
1880 /*
1881 * we can expect USER0 is not used (see vread/vwrite's
1882 * function description)
1883 */
1884 void *map = kmap_atomic(p, KM_USER0);
1885 memcpy(map + offset, buf, length);
1886 kunmap_atomic(map, KM_USER0);
1887 }
1888 addr += length;
1889 buf += length;
1890 copied += length;
1891 count -= length;
1892 }
1893 return copied;
1894}
1895
1896/**
1897 * vread() - read vmalloc area in a safe way.
1898 * @buf: buffer for reading data
1899 * @addr: vm address.
1900 * @count: number of bytes to be read.
1901 *
1902 * Returns # of bytes which addr and buf should be increased.
1903 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1904 * includes any intersect with alive vmalloc area.
1905 *
1906 * This function checks that addr is a valid vmalloc'ed area, and
1907 * copy data from that area to a given buffer. If the given memory range
1908 * of [addr...addr+count) includes some valid address, data is copied to
1909 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1910 * IOREMAP area is treated as memory hole and no copy is done.
1911 *
1912 * If [addr...addr+count) doesn't includes any intersects with alive
1913 * vm_struct area, returns 0.
1914 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1915 * the caller should guarantee KM_USER0 is not used.
1916 *
1917 * Note: In usual ops, vread() is never necessary because the caller
1918 * should know vmalloc() area is valid and can use memcpy().
1919 * This is for routines which have to access vmalloc area without
1920 * any informaion, as /dev/kmem.
1921 *
1922 */
1923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924long vread(char *buf, char *addr, unsigned long count)
1925{
1926 struct vm_struct *tmp;
1927 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001928 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 unsigned long n;
1930
1931 /* Don't allow overflow */
1932 if ((unsigned long) addr + count < count)
1933 count = -(unsigned long) addr;
1934
1935 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001936 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 vaddr = (char *) tmp->addr;
1938 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1939 continue;
1940 while (addr < vaddr) {
1941 if (count == 0)
1942 goto finished;
1943 *buf = '\0';
1944 buf++;
1945 addr++;
1946 count--;
1947 }
1948 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001949 if (n > count)
1950 n = count;
1951 if (!(tmp->flags & VM_IOREMAP))
1952 aligned_vread(buf, addr, n);
1953 else /* IOREMAP area is treated as memory hole */
1954 memset(buf, 0, n);
1955 buf += n;
1956 addr += n;
1957 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 }
1959finished:
1960 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001961
1962 if (buf == buf_start)
1963 return 0;
1964 /* zero-fill memory holes */
1965 if (buf != buf_start + buflen)
1966 memset(buf, 0, buflen - (buf - buf_start));
1967
1968 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
1970
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001971/**
1972 * vwrite() - write vmalloc area in a safe way.
1973 * @buf: buffer for source data
1974 * @addr: vm address.
1975 * @count: number of bytes to be read.
1976 *
1977 * Returns # of bytes which addr and buf should be incresed.
1978 * (same number to @count).
1979 * If [addr...addr+count) doesn't includes any intersect with valid
1980 * vmalloc area, returns 0.
1981 *
1982 * This function checks that addr is a valid vmalloc'ed area, and
1983 * copy data from a buffer to the given addr. If specified range of
1984 * [addr...addr+count) includes some valid address, data is copied from
1985 * proper area of @buf. If there are memory holes, no copy to hole.
1986 * IOREMAP area is treated as memory hole and no copy is done.
1987 *
1988 * If [addr...addr+count) doesn't includes any intersects with alive
1989 * vm_struct area, returns 0.
1990 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1991 * the caller should guarantee KM_USER0 is not used.
1992 *
1993 * Note: In usual ops, vwrite() is never necessary because the caller
1994 * should know vmalloc() area is valid and can use memcpy().
1995 * This is for routines which have to access vmalloc area without
1996 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001997 */
1998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999long vwrite(char *buf, char *addr, unsigned long count)
2000{
2001 struct vm_struct *tmp;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002002 char *vaddr;
2003 unsigned long n, buflen;
2004 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006 /* Don't allow overflow */
2007 if ((unsigned long) addr + count < count)
2008 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002009 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
2011 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002012 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 vaddr = (char *) tmp->addr;
2014 if (addr >= vaddr + tmp->size - PAGE_SIZE)
2015 continue;
2016 while (addr < vaddr) {
2017 if (count == 0)
2018 goto finished;
2019 buf++;
2020 addr++;
2021 count--;
2022 }
2023 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002024 if (n > count)
2025 n = count;
2026 if (!(tmp->flags & VM_IOREMAP)) {
2027 aligned_vwrite(buf, addr, n);
2028 copied++;
2029 }
2030 buf += n;
2031 addr += n;
2032 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 }
2034finished:
2035 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002036 if (!copied)
2037 return 0;
2038 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039}
Nick Piggin83342312006-06-23 02:03:20 -07002040
2041/**
2042 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002043 * @vma: vma to cover (map full range of vma)
2044 * @addr: vmalloc memory
2045 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002046 *
2047 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002048 *
2049 * This function checks that addr is a valid vmalloc'ed area, and
2050 * that it is big enough to cover the vma. Will return failure if
2051 * that criteria isn't met.
2052 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002053 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002054 */
2055int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2056 unsigned long pgoff)
2057{
2058 struct vm_struct *area;
2059 unsigned long uaddr = vma->vm_start;
2060 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07002061
2062 if ((PAGE_SIZE-1) & (unsigned long)addr)
2063 return -EINVAL;
2064
Nick Piggindb64fe02008-10-18 20:27:03 -07002065 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07002066 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07002067 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002068
2069 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07002070 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002071
2072 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07002073 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002074
2075 addr += pgoff << PAGE_SHIFT;
2076 do {
2077 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07002078 int ret;
2079
Nick Piggin83342312006-06-23 02:03:20 -07002080 ret = vm_insert_page(vma, uaddr, page);
2081 if (ret)
2082 return ret;
2083
2084 uaddr += PAGE_SIZE;
2085 addr += PAGE_SIZE;
2086 usize -= PAGE_SIZE;
2087 } while (usize > 0);
2088
2089 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
2090 vma->vm_flags |= VM_RESERVED;
2091
Nick Piggindb64fe02008-10-18 20:27:03 -07002092 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07002093}
2094EXPORT_SYMBOL(remap_vmalloc_range);
2095
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002096/*
2097 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2098 * have one.
2099 */
2100void __attribute__((weak)) vmalloc_sync_all(void)
2101{
2102}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002103
2104
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002105static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002106{
2107 /* apply_to_page_range() does all the hard work. */
2108 return 0;
2109}
2110
2111/**
2112 * alloc_vm_area - allocate a range of kernel address space
2113 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07002114 *
2115 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002116 *
2117 * This function reserves a range of kernel address space, and
2118 * allocates pagetables to map that range. No actual mappings
2119 * are created. If the kernel address space is not shared
2120 * between processes, it syncs the pagetable across all
2121 * processes.
2122 */
2123struct vm_struct *alloc_vm_area(size_t size)
2124{
2125 struct vm_struct *area;
2126
Christoph Lameter23016962008-04-28 02:12:42 -07002127 area = get_vm_area_caller(size, VM_IOREMAP,
2128 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002129 if (area == NULL)
2130 return NULL;
2131
2132 /*
2133 * This ensures that page tables are constructed for this region
2134 * of kernel virtual address space and mapped into init_mm.
2135 */
2136 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2137 area->size, f, NULL)) {
2138 free_vm_area(area);
2139 return NULL;
2140 }
2141
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002142 return area;
2143}
2144EXPORT_SYMBOL_GPL(alloc_vm_area);
2145
2146void free_vm_area(struct vm_struct *area)
2147{
2148 struct vm_struct *ret;
2149 ret = remove_vm_area(area->addr);
2150 BUG_ON(ret != area);
2151 kfree(area);
2152}
2153EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002154
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002155#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002156static struct vmap_area *node_to_va(struct rb_node *n)
2157{
2158 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2159}
2160
2161/**
2162 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2163 * @end: target address
2164 * @pnext: out arg for the next vmap_area
2165 * @pprev: out arg for the previous vmap_area
2166 *
2167 * Returns: %true if either or both of next and prev are found,
2168 * %false if no vmap_area exists
2169 *
2170 * Find vmap_areas end addresses of which enclose @end. ie. if not
2171 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2172 */
2173static bool pvm_find_next_prev(unsigned long end,
2174 struct vmap_area **pnext,
2175 struct vmap_area **pprev)
2176{
2177 struct rb_node *n = vmap_area_root.rb_node;
2178 struct vmap_area *va = NULL;
2179
2180 while (n) {
2181 va = rb_entry(n, struct vmap_area, rb_node);
2182 if (end < va->va_end)
2183 n = n->rb_left;
2184 else if (end > va->va_end)
2185 n = n->rb_right;
2186 else
2187 break;
2188 }
2189
2190 if (!va)
2191 return false;
2192
2193 if (va->va_end > end) {
2194 *pnext = va;
2195 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2196 } else {
2197 *pprev = va;
2198 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2199 }
2200 return true;
2201}
2202
2203/**
2204 * pvm_determine_end - find the highest aligned address between two vmap_areas
2205 * @pnext: in/out arg for the next vmap_area
2206 * @pprev: in/out arg for the previous vmap_area
2207 * @align: alignment
2208 *
2209 * Returns: determined end address
2210 *
2211 * Find the highest aligned address between *@pnext and *@pprev below
2212 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2213 * down address is between the end addresses of the two vmap_areas.
2214 *
2215 * Please note that the address returned by this function may fall
2216 * inside *@pnext vmap_area. The caller is responsible for checking
2217 * that.
2218 */
2219static unsigned long pvm_determine_end(struct vmap_area **pnext,
2220 struct vmap_area **pprev,
2221 unsigned long align)
2222{
2223 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2224 unsigned long addr;
2225
2226 if (*pnext)
2227 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2228 else
2229 addr = vmalloc_end;
2230
2231 while (*pprev && (*pprev)->va_end > addr) {
2232 *pnext = *pprev;
2233 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2234 }
2235
2236 return addr;
2237}
2238
2239/**
2240 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2241 * @offsets: array containing offset of each area
2242 * @sizes: array containing size of each area
2243 * @nr_vms: the number of areas to allocate
2244 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002245 *
2246 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2247 * vm_structs on success, %NULL on failure
2248 *
2249 * Percpu allocator wants to use congruent vm areas so that it can
2250 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002251 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2252 * be scattered pretty far, distance between two areas easily going up
2253 * to gigabytes. To avoid interacting with regular vmallocs, these
2254 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002255 *
2256 * Despite its complicated look, this allocator is rather simple. It
2257 * does everything top-down and scans areas from the end looking for
2258 * matching slot. While scanning, if any of the areas overlaps with
2259 * existing vmap_area, the base address is pulled down to fit the
2260 * area. Scanning is repeated till all the areas fit and then all
2261 * necessary data structres are inserted and the result is returned.
2262 */
2263struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2264 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002265 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002266{
2267 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2268 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2269 struct vmap_area **vas, *prev, *next;
2270 struct vm_struct **vms;
2271 int area, area2, last_area, term_area;
2272 unsigned long base, start, end, last_end;
2273 bool purged = false;
2274
Tejun Heoca23e402009-08-14 15:00:52 +09002275 /* verify parameters and allocate data structures */
2276 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2277 for (last_area = 0, area = 0; area < nr_vms; area++) {
2278 start = offsets[area];
2279 end = start + sizes[area];
2280
2281 /* is everything aligned properly? */
2282 BUG_ON(!IS_ALIGNED(offsets[area], align));
2283 BUG_ON(!IS_ALIGNED(sizes[area], align));
2284
2285 /* detect the area with the highest address */
2286 if (start > offsets[last_area])
2287 last_area = area;
2288
2289 for (area2 = 0; area2 < nr_vms; area2++) {
2290 unsigned long start2 = offsets[area2];
2291 unsigned long end2 = start2 + sizes[area2];
2292
2293 if (area2 == area)
2294 continue;
2295
2296 BUG_ON(start2 >= start && start2 < end);
2297 BUG_ON(end2 <= end && end2 > start);
2298 }
2299 }
2300 last_end = offsets[last_area] + sizes[last_area];
2301
2302 if (vmalloc_end - vmalloc_start < last_end) {
2303 WARN_ON(true);
2304 return NULL;
2305 }
2306
David Rientjesec3f64f2011-01-13 15:46:01 -08002307 vms = kzalloc(sizeof(vms[0]) * nr_vms, GFP_KERNEL);
2308 vas = kzalloc(sizeof(vas[0]) * nr_vms, GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002309 if (!vas || !vms)
2310 goto err_free;
2311
2312 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002313 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2314 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002315 if (!vas[area] || !vms[area])
2316 goto err_free;
2317 }
2318retry:
2319 spin_lock(&vmap_area_lock);
2320
2321 /* start scanning - we scan from the top, begin with the last area */
2322 area = term_area = last_area;
2323 start = offsets[area];
2324 end = start + sizes[area];
2325
2326 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2327 base = vmalloc_end - last_end;
2328 goto found;
2329 }
2330 base = pvm_determine_end(&next, &prev, align) - end;
2331
2332 while (true) {
2333 BUG_ON(next && next->va_end <= base + end);
2334 BUG_ON(prev && prev->va_end > base + end);
2335
2336 /*
2337 * base might have underflowed, add last_end before
2338 * comparing.
2339 */
2340 if (base + last_end < vmalloc_start + last_end) {
2341 spin_unlock(&vmap_area_lock);
2342 if (!purged) {
2343 purge_vmap_area_lazy();
2344 purged = true;
2345 goto retry;
2346 }
2347 goto err_free;
2348 }
2349
2350 /*
2351 * If next overlaps, move base downwards so that it's
2352 * right below next and then recheck.
2353 */
2354 if (next && next->va_start < base + end) {
2355 base = pvm_determine_end(&next, &prev, align) - end;
2356 term_area = area;
2357 continue;
2358 }
2359
2360 /*
2361 * If prev overlaps, shift down next and prev and move
2362 * base so that it's right below new next and then
2363 * recheck.
2364 */
2365 if (prev && prev->va_end > base + start) {
2366 next = prev;
2367 prev = node_to_va(rb_prev(&next->rb_node));
2368 base = pvm_determine_end(&next, &prev, align) - end;
2369 term_area = area;
2370 continue;
2371 }
2372
2373 /*
2374 * This area fits, move on to the previous one. If
2375 * the previous one is the terminal one, we're done.
2376 */
2377 area = (area + nr_vms - 1) % nr_vms;
2378 if (area == term_area)
2379 break;
2380 start = offsets[area];
2381 end = start + sizes[area];
2382 pvm_find_next_prev(base + end, &next, &prev);
2383 }
2384found:
2385 /* we've found a fitting base, insert all va's */
2386 for (area = 0; area < nr_vms; area++) {
2387 struct vmap_area *va = vas[area];
2388
2389 va->va_start = base + offsets[area];
2390 va->va_end = va->va_start + sizes[area];
2391 __insert_vmap_area(va);
2392 }
2393
2394 vmap_area_pcpu_hole = base + offsets[last_area];
2395
2396 spin_unlock(&vmap_area_lock);
2397
2398 /* insert all vm's */
2399 for (area = 0; area < nr_vms; area++)
2400 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2401 pcpu_get_vm_areas);
2402
2403 kfree(vas);
2404 return vms;
2405
2406err_free:
2407 for (area = 0; area < nr_vms; area++) {
2408 if (vas)
2409 kfree(vas[area]);
2410 if (vms)
2411 kfree(vms[area]);
2412 }
2413 kfree(vas);
2414 kfree(vms);
2415 return NULL;
2416}
2417
2418/**
2419 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2420 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2421 * @nr_vms: the number of allocated areas
2422 *
2423 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2424 */
2425void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2426{
2427 int i;
2428
2429 for (i = 0; i < nr_vms; i++)
2430 free_vm_area(vms[i]);
2431 kfree(vms);
2432}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002433#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002434
2435#ifdef CONFIG_PROC_FS
2436static void *s_start(struct seq_file *m, loff_t *pos)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002437 __acquires(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002438{
2439 loff_t n = *pos;
2440 struct vm_struct *v;
2441
2442 read_lock(&vmlist_lock);
2443 v = vmlist;
2444 while (n > 0 && v) {
2445 n--;
2446 v = v->next;
2447 }
2448 if (!n)
2449 return v;
2450
2451 return NULL;
2452
2453}
2454
2455static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2456{
2457 struct vm_struct *v = p;
2458
2459 ++*pos;
2460 return v->next;
2461}
2462
2463static void s_stop(struct seq_file *m, void *p)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002464 __releases(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002465{
2466 read_unlock(&vmlist_lock);
2467}
2468
Eric Dumazeta47a1262008-07-23 21:27:38 -07002469static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2470{
2471 if (NUMA_BUILD) {
2472 unsigned int nr, *counters = m->private;
2473
2474 if (!counters)
2475 return;
2476
2477 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2478
2479 for (nr = 0; nr < v->nr_pages; nr++)
2480 counters[page_to_nid(v->pages[nr])]++;
2481
2482 for_each_node_state(nr, N_HIGH_MEMORY)
2483 if (counters[nr])
2484 seq_printf(m, " N%u=%u", nr, counters[nr]);
2485 }
2486}
2487
Christoph Lametera10aa572008-04-28 02:12:40 -07002488static int s_show(struct seq_file *m, void *p)
2489{
2490 struct vm_struct *v = p;
2491
2492 seq_printf(m, "0x%p-0x%p %7ld",
2493 v->addr, v->addr + v->size, v->size);
2494
Joe Perches62c70bc2011-01-13 15:45:52 -08002495 if (v->caller)
2496 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002497
Christoph Lametera10aa572008-04-28 02:12:40 -07002498 if (v->nr_pages)
2499 seq_printf(m, " pages=%d", v->nr_pages);
2500
2501 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002502 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002503
2504 if (v->flags & VM_IOREMAP)
2505 seq_printf(m, " ioremap");
2506
2507 if (v->flags & VM_ALLOC)
2508 seq_printf(m, " vmalloc");
2509
2510 if (v->flags & VM_MAP)
2511 seq_printf(m, " vmap");
2512
2513 if (v->flags & VM_USERMAP)
2514 seq_printf(m, " user");
2515
2516 if (v->flags & VM_VPAGES)
2517 seq_printf(m, " vpages");
2518
Eric Dumazeta47a1262008-07-23 21:27:38 -07002519 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002520 seq_putc(m, '\n');
2521 return 0;
2522}
2523
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002524static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002525 .start = s_start,
2526 .next = s_next,
2527 .stop = s_stop,
2528 .show = s_show,
2529};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002530
2531static int vmalloc_open(struct inode *inode, struct file *file)
2532{
2533 unsigned int *ptr = NULL;
2534 int ret;
2535
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002536 if (NUMA_BUILD) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002537 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002538 if (ptr == NULL)
2539 return -ENOMEM;
2540 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002541 ret = seq_open(file, &vmalloc_op);
2542 if (!ret) {
2543 struct seq_file *m = file->private_data;
2544 m->private = ptr;
2545 } else
2546 kfree(ptr);
2547 return ret;
2548}
2549
2550static const struct file_operations proc_vmalloc_operations = {
2551 .open = vmalloc_open,
2552 .read = seq_read,
2553 .llseek = seq_lseek,
2554 .release = seq_release_private,
2555};
2556
2557static int __init proc_vmalloc_init(void)
2558{
2559 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2560 return 0;
2561}
2562module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07002563#endif
2564