blob: 07a3858a9c1bfbd0d2b86828dbb858814a60a404 [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 */
Minchan Kimdb1aeca2012-01-10 15:08:39 -0800259 struct vm_struct *vm;
Nick Piggindb64fe02008-10-18 20:27:03 -0700260 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() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200728#define VMAP_BBMAP_BITS \
729 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
730 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
731 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700732
733#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
734
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100735static bool vmap_initialized __read_mostly = false;
736
Nick Piggindb64fe02008-10-18 20:27:03 -0700737struct vmap_block_queue {
738 spinlock_t lock;
739 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700740};
741
742struct vmap_block {
743 spinlock_t lock;
744 struct vmap_area *va;
745 struct vmap_block_queue *vbq;
746 unsigned long free, dirty;
747 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
748 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100749 struct list_head free_list;
750 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100751 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700752};
753
754/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
755static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
756
757/*
758 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
759 * in the free path. Could get rid of this if we change the API to return a
760 * "cookie" from alloc, to be passed to free. But no big deal yet.
761 */
762static DEFINE_SPINLOCK(vmap_block_tree_lock);
763static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
764
765/*
766 * We should probably have a fallback mechanism to allocate virtual memory
767 * out of partially filled vmap blocks. However vmap block sizing should be
768 * fairly reasonable according to the vmalloc size, so it shouldn't be a
769 * big problem.
770 */
771
772static unsigned long addr_to_vb_idx(unsigned long addr)
773{
774 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
775 addr /= VMAP_BLOCK_SIZE;
776 return addr;
777}
778
779static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
780{
781 struct vmap_block_queue *vbq;
782 struct vmap_block *vb;
783 struct vmap_area *va;
784 unsigned long vb_idx;
785 int node, err;
786
787 node = numa_node_id();
788
789 vb = kmalloc_node(sizeof(struct vmap_block),
790 gfp_mask & GFP_RECLAIM_MASK, node);
791 if (unlikely(!vb))
792 return ERR_PTR(-ENOMEM);
793
794 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
795 VMALLOC_START, VMALLOC_END,
796 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800797 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700798 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700799 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700800 }
801
802 err = radix_tree_preload(gfp_mask);
803 if (unlikely(err)) {
804 kfree(vb);
805 free_vmap_area(va);
806 return ERR_PTR(err);
807 }
808
809 spin_lock_init(&vb->lock);
810 vb->va = va;
811 vb->free = VMAP_BBMAP_BITS;
812 vb->dirty = 0;
813 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
814 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
815 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700816
817 vb_idx = addr_to_vb_idx(va->va_start);
818 spin_lock(&vmap_block_tree_lock);
819 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
820 spin_unlock(&vmap_block_tree_lock);
821 BUG_ON(err);
822 radix_tree_preload_end();
823
824 vbq = &get_cpu_var(vmap_block_queue);
825 vb->vbq = vbq;
826 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100827 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700828 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900829 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700830
831 return vb;
832}
833
Nick Piggindb64fe02008-10-18 20:27:03 -0700834static void free_vmap_block(struct vmap_block *vb)
835{
836 struct vmap_block *tmp;
837 unsigned long vb_idx;
838
Nick Piggindb64fe02008-10-18 20:27:03 -0700839 vb_idx = addr_to_vb_idx(vb->va->va_start);
840 spin_lock(&vmap_block_tree_lock);
841 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
842 spin_unlock(&vmap_block_tree_lock);
843 BUG_ON(tmp != vb);
844
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800845 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800846 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700847}
848
Nick Piggin02b709d2010-02-01 22:25:57 +1100849static void purge_fragmented_blocks(int cpu)
850{
851 LIST_HEAD(purge);
852 struct vmap_block *vb;
853 struct vmap_block *n_vb;
854 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
855
856 rcu_read_lock();
857 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
858
859 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
860 continue;
861
862 spin_lock(&vb->lock);
863 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
864 vb->free = 0; /* prevent further allocs after releasing lock */
865 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
866 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
867 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
868 spin_lock(&vbq->lock);
869 list_del_rcu(&vb->free_list);
870 spin_unlock(&vbq->lock);
871 spin_unlock(&vb->lock);
872 list_add_tail(&vb->purge, &purge);
873 } else
874 spin_unlock(&vb->lock);
875 }
876 rcu_read_unlock();
877
878 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
879 list_del(&vb->purge);
880 free_vmap_block(vb);
881 }
882}
883
884static void purge_fragmented_blocks_thiscpu(void)
885{
886 purge_fragmented_blocks(smp_processor_id());
887}
888
889static void purge_fragmented_blocks_allcpus(void)
890{
891 int cpu;
892
893 for_each_possible_cpu(cpu)
894 purge_fragmented_blocks(cpu);
895}
896
Nick Piggindb64fe02008-10-18 20:27:03 -0700897static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
898{
899 struct vmap_block_queue *vbq;
900 struct vmap_block *vb;
901 unsigned long addr = 0;
902 unsigned int order;
Nick Piggin02b709d2010-02-01 22:25:57 +1100903 int purge = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700904
905 BUG_ON(size & ~PAGE_MASK);
906 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
907 order = get_order(size);
908
909again:
910 rcu_read_lock();
911 vbq = &get_cpu_var(vmap_block_queue);
912 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
913 int i;
914
915 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100916 if (vb->free < 1UL << order)
917 goto next;
918
Nick Piggindb64fe02008-10-18 20:27:03 -0700919 i = bitmap_find_free_region(vb->alloc_map,
920 VMAP_BBMAP_BITS, order);
921
Nick Piggin02b709d2010-02-01 22:25:57 +1100922 if (i < 0) {
923 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
924 /* fragmented and no outstanding allocations */
925 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
926 purge = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -0700927 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100928 goto next;
929 }
930 addr = vb->va->va_start + (i << PAGE_SHIFT);
931 BUG_ON(addr_to_vb_idx(addr) !=
932 addr_to_vb_idx(vb->va->va_start));
933 vb->free -= 1UL << order;
934 if (vb->free == 0) {
935 spin_lock(&vbq->lock);
936 list_del_rcu(&vb->free_list);
937 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700938 }
939 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100940 break;
941next:
942 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700943 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100944
945 if (purge)
946 purge_fragmented_blocks_thiscpu();
947
Tejun Heo3f04ba82009-10-29 22:34:12 +0900948 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700949 rcu_read_unlock();
950
951 if (!addr) {
952 vb = new_vmap_block(gfp_mask);
953 if (IS_ERR(vb))
954 return vb;
955 goto again;
956 }
957
958 return (void *)addr;
959}
960
961static void vb_free(const void *addr, unsigned long size)
962{
963 unsigned long offset;
964 unsigned long vb_idx;
965 unsigned int order;
966 struct vmap_block *vb;
967
968 BUG_ON(size & ~PAGE_MASK);
969 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800970
971 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
972
Nick Piggindb64fe02008-10-18 20:27:03 -0700973 order = get_order(size);
974
975 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
976
977 vb_idx = addr_to_vb_idx((unsigned long)addr);
978 rcu_read_lock();
979 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
980 rcu_read_unlock();
981 BUG_ON(!vb);
982
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800983 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
984
Nick Piggindb64fe02008-10-18 20:27:03 -0700985 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100986 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700987
Nick Piggindb64fe02008-10-18 20:27:03 -0700988 vb->dirty += 1UL << order;
989 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100990 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700991 spin_unlock(&vb->lock);
992 free_vmap_block(vb);
993 } else
994 spin_unlock(&vb->lock);
995}
996
997/**
998 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
999 *
1000 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1001 * to amortize TLB flushing overheads. What this means is that any page you
1002 * have now, may, in a former life, have been mapped into kernel virtual
1003 * address by the vmap layer and so there might be some CPUs with TLB entries
1004 * still referencing that page (additional to the regular 1:1 kernel mapping).
1005 *
1006 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1007 * be sure that none of the pages we have control over will have any aliases
1008 * from the vmap layer.
1009 */
1010void vm_unmap_aliases(void)
1011{
1012 unsigned long start = ULONG_MAX, end = 0;
1013 int cpu;
1014 int flush = 0;
1015
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001016 if (unlikely(!vmap_initialized))
1017 return;
1018
Nick Piggindb64fe02008-10-18 20:27:03 -07001019 for_each_possible_cpu(cpu) {
1020 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1021 struct vmap_block *vb;
1022
1023 rcu_read_lock();
1024 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1025 int i;
1026
1027 spin_lock(&vb->lock);
1028 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1029 while (i < VMAP_BBMAP_BITS) {
1030 unsigned long s, e;
1031 int j;
1032 j = find_next_zero_bit(vb->dirty_map,
1033 VMAP_BBMAP_BITS, i);
1034
1035 s = vb->va->va_start + (i << PAGE_SHIFT);
1036 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001037 flush = 1;
1038
1039 if (s < start)
1040 start = s;
1041 if (e > end)
1042 end = e;
1043
1044 i = j;
1045 i = find_next_bit(vb->dirty_map,
1046 VMAP_BBMAP_BITS, i);
1047 }
1048 spin_unlock(&vb->lock);
1049 }
1050 rcu_read_unlock();
1051 }
1052
1053 __purge_vmap_area_lazy(&start, &end, 1, flush);
1054}
1055EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1056
1057/**
1058 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1059 * @mem: the pointer returned by vm_map_ram
1060 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1061 */
1062void vm_unmap_ram(const void *mem, unsigned int count)
1063{
1064 unsigned long size = count << PAGE_SHIFT;
1065 unsigned long addr = (unsigned long)mem;
1066
1067 BUG_ON(!addr);
1068 BUG_ON(addr < VMALLOC_START);
1069 BUG_ON(addr > VMALLOC_END);
1070 BUG_ON(addr & (PAGE_SIZE-1));
1071
1072 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001073 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001074
1075 if (likely(count <= VMAP_MAX_ALLOC))
1076 vb_free(mem, size);
1077 else
1078 free_unmap_vmap_area_addr(addr);
1079}
1080EXPORT_SYMBOL(vm_unmap_ram);
1081
1082/**
1083 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1084 * @pages: an array of pointers to the pages to be mapped
1085 * @count: number of pages
1086 * @node: prefer to allocate data structures on this node
1087 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001088 *
1089 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001090 */
1091void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1092{
1093 unsigned long size = count << PAGE_SHIFT;
1094 unsigned long addr;
1095 void *mem;
1096
1097 if (likely(count <= VMAP_MAX_ALLOC)) {
1098 mem = vb_alloc(size, GFP_KERNEL);
1099 if (IS_ERR(mem))
1100 return NULL;
1101 addr = (unsigned long)mem;
1102 } else {
1103 struct vmap_area *va;
1104 va = alloc_vmap_area(size, PAGE_SIZE,
1105 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1106 if (IS_ERR(va))
1107 return NULL;
1108
1109 addr = va->va_start;
1110 mem = (void *)addr;
1111 }
1112 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1113 vm_unmap_ram(mem, count);
1114 return NULL;
1115 }
1116 return mem;
1117}
1118EXPORT_SYMBOL(vm_map_ram);
1119
Tejun Heof0aa6612009-02-20 16:29:08 +09001120/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001121 * vm_area_add_early - add vmap area early during boot
1122 * @vm: vm_struct to add
1123 *
1124 * This function is used to add fixed kernel vm area to vmlist before
1125 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1126 * should contain proper values and the other fields should be zero.
1127 *
1128 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1129 */
1130void __init vm_area_add_early(struct vm_struct *vm)
1131{
1132 struct vm_struct *tmp, **p;
1133
1134 BUG_ON(vmap_initialized);
1135 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1136 if (tmp->addr >= vm->addr) {
1137 BUG_ON(tmp->addr < vm->addr + vm->size);
1138 break;
1139 } else
1140 BUG_ON(tmp->addr + tmp->size > vm->addr);
1141 }
1142 vm->next = *p;
1143 *p = vm;
1144}
1145
1146/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001147 * vm_area_register_early - register vmap area early during boot
1148 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001149 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001150 *
1151 * This function is used to register kernel vm area before
1152 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1153 * proper values on entry and other fields should be zero. On return,
1154 * vm->addr contains the allocated address.
1155 *
1156 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1157 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001158void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001159{
1160 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001161 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001162
Tejun Heoc0c0a292009-02-24 11:57:21 +09001163 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1164 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1165
1166 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001167
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001168 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001169}
1170
Nick Piggindb64fe02008-10-18 20:27:03 -07001171void __init vmalloc_init(void)
1172{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001173 struct vmap_area *va;
1174 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001175 int i;
1176
1177 for_each_possible_cpu(i) {
1178 struct vmap_block_queue *vbq;
1179
1180 vbq = &per_cpu(vmap_block_queue, i);
1181 spin_lock_init(&vbq->lock);
1182 INIT_LIST_HEAD(&vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001183 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001184
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001185 /* Import existing vmlist entries. */
1186 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001187 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHofa002622012-05-29 15:06:49 -07001188 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001189 va->va_start = (unsigned long)tmp->addr;
1190 va->va_end = va->va_start + tmp->size;
KyongHofa002622012-05-29 15:06:49 -07001191 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001192 __insert_vmap_area(va);
1193 }
Tejun Heoca23e402009-08-14 15:00:52 +09001194
1195 vmap_area_pcpu_hole = VMALLOC_END;
1196
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001197 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001198}
1199
Tejun Heo8fc48982009-02-20 16:29:08 +09001200/**
1201 * map_kernel_range_noflush - map kernel VM area with the specified pages
1202 * @addr: start of the VM area to map
1203 * @size: size of the VM area to map
1204 * @prot: page protection flags to use
1205 * @pages: pages to map
1206 *
1207 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1208 * specify should have been allocated using get_vm_area() and its
1209 * friends.
1210 *
1211 * NOTE:
1212 * This function does NOT do any cache flushing. The caller is
1213 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1214 * before calling this function.
1215 *
1216 * RETURNS:
1217 * The number of pages mapped on success, -errno on failure.
1218 */
1219int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1220 pgprot_t prot, struct page **pages)
1221{
1222 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1223}
1224
1225/**
1226 * unmap_kernel_range_noflush - unmap kernel VM area
1227 * @addr: start of the VM area to unmap
1228 * @size: size of the VM area to unmap
1229 *
1230 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1231 * specify should have been allocated using get_vm_area() and its
1232 * friends.
1233 *
1234 * NOTE:
1235 * This function does NOT do any cache flushing. The caller is
1236 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1237 * before calling this function and flush_tlb_kernel_range() after.
1238 */
1239void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1240{
1241 vunmap_page_range(addr, addr + size);
1242}
Huang Ying81e88fd2011-01-12 14:44:55 +08001243EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001244
1245/**
1246 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1247 * @addr: start of the VM area to unmap
1248 * @size: size of the VM area to unmap
1249 *
1250 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1251 * the unmapping and tlb after.
1252 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001253void unmap_kernel_range(unsigned long addr, unsigned long size)
1254{
1255 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001256
1257 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001258 vunmap_page_range(addr, end);
1259 flush_tlb_kernel_range(addr, end);
1260}
1261
1262int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1263{
1264 unsigned long addr = (unsigned long)area->addr;
1265 unsigned long end = addr + area->size - PAGE_SIZE;
1266 int err;
1267
1268 err = vmap_page_range(addr, end, prot, *pages);
1269 if (err > 0) {
1270 *pages += err;
1271 err = 0;
1272 }
1273
1274 return err;
1275}
1276EXPORT_SYMBOL_GPL(map_vm_area);
1277
1278/*** Old vmalloc interfaces ***/
1279DEFINE_RWLOCK(vmlist_lock);
1280struct vm_struct *vmlist;
1281
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001282static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Tejun Heocf88c792009-08-14 15:00:52 +09001283 unsigned long flags, void *caller)
1284{
Tejun Heocf88c792009-08-14 15:00:52 +09001285 vm->flags = flags;
1286 vm->addr = (void *)va->va_start;
1287 vm->size = va->va_end - va->va_start;
1288 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001289 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001290 va->flags |= VM_VM_AREA;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001291}
Tejun Heocf88c792009-08-14 15:00:52 +09001292
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001293static void insert_vmalloc_vmlist(struct vm_struct *vm)
1294{
1295 struct vm_struct *tmp, **p;
1296
1297 vm->flags &= ~VM_UNLIST;
Tejun Heocf88c792009-08-14 15:00:52 +09001298 write_lock(&vmlist_lock);
1299 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1300 if (tmp->addr >= vm->addr)
1301 break;
1302 }
1303 vm->next = *p;
1304 *p = vm;
1305 write_unlock(&vmlist_lock);
1306}
1307
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001308static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1309 unsigned long flags, void *caller)
1310{
1311 setup_vmalloc_vm(vm, va, flags, caller);
1312 insert_vmalloc_vmlist(vm);
1313}
1314
Nick Piggindb64fe02008-10-18 20:27:03 -07001315static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001316 unsigned long align, unsigned long flags, unsigned long start,
1317 unsigned long end, int node, gfp_t gfp_mask, void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001318{
Kautuk Consul00065262011-12-19 17:12:04 -08001319 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001320 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001322 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 if (flags & VM_IOREMAP) {
1324 int bit = fls(size);
1325
1326 if (bit > IOREMAP_MAX_ORDER)
1327 bit = IOREMAP_MAX_ORDER;
1328 else if (bit < PAGE_SHIFT)
1329 bit = PAGE_SHIFT;
1330
1331 align = 1ul << bit;
1332 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001335 if (unlikely(!size))
1336 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
Tejun Heocf88c792009-08-14 15:00:52 +09001338 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 if (unlikely(!area))
1340 return NULL;
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 /*
1343 * We always allocate a guard page.
1344 */
1345 size += PAGE_SIZE;
1346
Nick Piggindb64fe02008-10-18 20:27:03 -07001347 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1348 if (IS_ERR(va)) {
1349 kfree(area);
1350 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001353 /*
1354 * When this function is called from __vmalloc_node_range,
1355 * we do not add vm_struct to vmlist here to avoid
1356 * accessing uninitialized members of vm_struct such as
1357 * pages and nr_pages fields. They will be set later.
1358 * To distinguish it from others, we use a VM_UNLIST flag.
1359 */
1360 if (flags & VM_UNLIST)
1361 setup_vmalloc_vm(area, va, flags, caller);
1362 else
1363 insert_vmalloc_vm(area, va, flags, caller);
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366}
1367
Christoph Lameter930fc452005-10-29 18:15:41 -07001368struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1369 unsigned long start, unsigned long end)
1370{
David Miller2dca6992009-09-21 12:22:34 -07001371 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001372 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001373}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001374EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001375
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001376struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1377 unsigned long start, unsigned long end,
1378 void *caller)
1379{
David Miller2dca6992009-09-21 12:22:34 -07001380 return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL,
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001381 caller);
1382}
1383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384/**
Simon Arlott183ff222007-10-20 01:27:18 +02001385 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 * @size: size of the area
1387 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1388 *
1389 * Search an area of @size in the kernel virtual mapping area,
1390 * and reserved it for out purposes. Returns the area descriptor
1391 * on success or %NULL on failure.
1392 */
1393struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1394{
David Miller2dca6992009-09-21 12:22:34 -07001395 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001396 -1, GFP_KERNEL, __builtin_return_address(0));
1397}
1398
1399struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1400 void *caller)
1401{
David Miller2dca6992009-09-21 12:22:34 -07001402 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
Christoph Lameter23016962008-04-28 02:12:42 -07001403 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404}
1405
Nick Piggindb64fe02008-10-18 20:27:03 -07001406static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001407{
Nick Piggindb64fe02008-10-18 20:27:03 -07001408 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001409
Nick Piggindb64fe02008-10-18 20:27:03 -07001410 va = find_vmap_area((unsigned long)addr);
1411 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001412 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001413
Andi Kleen7856dfe2005-05-20 14:27:57 -07001414 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001415}
1416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417/**
Simon Arlott183ff222007-10-20 01:27:18 +02001418 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 * @addr: base address
1420 *
1421 * Search for the kernel VM area starting at @addr, and remove it.
1422 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001423 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001425struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
Nick Piggindb64fe02008-10-18 20:27:03 -07001427 struct vmap_area *va;
1428
1429 va = find_vmap_area((unsigned long)addr);
1430 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001431 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001432
1433 if (!(vm->flags & VM_UNLIST)) {
1434 struct vm_struct *tmp, **p;
1435 /*
1436 * remove from list and disallow access to
1437 * this vm_struct before unmap. (address range
1438 * confliction is maintained by vmap.)
1439 */
1440 write_lock(&vmlist_lock);
1441 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1442 ;
1443 *p = tmp->next;
1444 write_unlock(&vmlist_lock);
1445 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001446
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001447 vmap_debug_free_range(va->va_start, va->va_end);
1448 free_unmap_vmap_area(va);
1449 vm->size -= PAGE_SIZE;
1450
Nick Piggindb64fe02008-10-18 20:27:03 -07001451 return vm;
1452 }
1453 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454}
1455
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001456static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
1458 struct vm_struct *area;
1459
1460 if (!addr)
1461 return;
1462
1463 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001464 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return;
1466 }
1467
1468 area = remove_vm_area(addr);
1469 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001470 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 return;
1473 }
1474
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001475 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001476 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 if (deallocate_pages) {
1479 int i;
1480
1481 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001482 struct page *page = area->pages[i];
1483
1484 BUG_ON(!page);
1485 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
1487
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001488 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 vfree(area->pages);
1490 else
1491 kfree(area->pages);
1492 }
1493
1494 kfree(area);
1495 return;
1496}
1497
1498/**
1499 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 * @addr: memory base address
1501 *
Simon Arlott183ff222007-10-20 01:27:18 +02001502 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001503 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1504 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001506 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001508void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
1510 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001511
1512 kmemleak_free(addr);
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 __vunmap(addr, 1);
1515}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516EXPORT_SYMBOL(vfree);
1517
1518/**
1519 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 * @addr: memory base address
1521 *
1522 * Free the virtually contiguous memory area starting at @addr,
1523 * which was created from the page array passed to vmap().
1524 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001525 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001527void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
1529 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001530 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 __vunmap(addr, 0);
1532}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533EXPORT_SYMBOL(vunmap);
1534
1535/**
1536 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 * @pages: array of page pointers
1538 * @count: number of pages to map
1539 * @flags: vm_area->flags
1540 * @prot: page protection for the mapping
1541 *
1542 * Maps @count pages from @pages into contiguous kernel virtual
1543 * space.
1544 */
1545void *vmap(struct page **pages, unsigned int count,
1546 unsigned long flags, pgprot_t prot)
1547{
1548 struct vm_struct *area;
1549
Peter Zijlstra34754b62009-02-25 16:04:03 +01001550 might_sleep();
1551
Jan Beulich44813742009-09-21 17:03:05 -07001552 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 return NULL;
1554
Christoph Lameter23016962008-04-28 02:12:42 -07001555 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1556 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (!area)
1558 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 if (map_vm_area(area, prot, &pages)) {
1561 vunmap(area->addr);
1562 return NULL;
1563 }
1564
1565 return area->addr;
1566}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567EXPORT_SYMBOL(vmap);
1568
David Miller2dca6992009-09-21 12:22:34 -07001569static void *__vmalloc_node(unsigned long size, unsigned long align,
1570 gfp_t gfp_mask, pgprot_t prot,
Nick Piggindb64fe02008-10-18 20:27:03 -07001571 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001572static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001573 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574{
Dave Hansen22943ab2011-05-24 17:12:18 -07001575 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 struct page **pages;
1577 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001578 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1581 array_size = (nr_pages * sizeof(struct page *));
1582
1583 area->nr_pages = nr_pages;
1584 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001585 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001586 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Christoph Lameter23016962008-04-28 02:12:42 -07001587 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001588 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001589 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001590 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001593 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (!area->pages) {
1595 remove_vm_area(area->addr);
1596 kfree(area);
1597 return NULL;
1598 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001601 struct page *page;
Dave Hansen22943ab2011-05-24 17:12:18 -07001602 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001603
Christoph Lameter930fc452005-10-29 18:15:41 -07001604 if (node < 0)
Dave Hansen22943ab2011-05-24 17:12:18 -07001605 page = alloc_page(tmp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001606 else
Dave Hansen22943ab2011-05-24 17:12:18 -07001607 page = alloc_pages_node(node, tmp_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001608
1609 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 /* Successfully allocated i pages, free them in __vunmap() */
1611 area->nr_pages = i;
1612 goto fail;
1613 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001614 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
1616
1617 if (map_vm_area(area, prot, &pages))
1618 goto fail;
1619 return area->addr;
1620
1621fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001622 warn_alloc_failed(gfp_mask, order,
1623 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001624 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 vfree(area->addr);
1626 return NULL;
1627}
1628
David Rientjesd0a21262011-01-13 15:46:02 -08001629/**
1630 * __vmalloc_node_range - allocate virtually contiguous memory
1631 * @size: allocation size
1632 * @align: desired alignment
1633 * @start: vm area range start
1634 * @end: vm area range end
1635 * @gfp_mask: flags for the page level allocator
1636 * @prot: protection mask for the allocated pages
1637 * @node: node to use for allocation or -1
1638 * @caller: caller's return address
1639 *
1640 * Allocate enough pages to cover @size from the page level
1641 * allocator with @gfp_mask flags. Map them into contiguous
1642 * kernel virtual space, using a pagetable protection of @prot.
1643 */
1644void *__vmalloc_node_range(unsigned long size, unsigned long align,
1645 unsigned long start, unsigned long end, gfp_t gfp_mask,
1646 pgprot_t prot, int node, void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001647{
David Rientjesd0a21262011-01-13 15:46:02 -08001648 struct vm_struct *area;
1649 void *addr;
1650 unsigned long real_size = size;
Jack Cheung59f9f1c2011-11-29 16:52:49 -08001651#ifdef CONFIG_FIX_MOVABLE_ZONE
1652 unsigned long total_pages = total_unmovable_pages;
1653#else
1654 unsigned long total_pages = totalram_pages;
1655#endif
David Rientjesd0a21262011-01-13 15:46:02 -08001656
1657 size = PAGE_ALIGN(size);
Jack Cheung59f9f1c2011-11-29 16:52:49 -08001658 if (!size || (size >> PAGE_SHIFT) > total_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001659 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001660
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001661 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1662 start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001663 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001664 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001665
1666 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Mel Gorman1368edf2011-12-08 14:34:30 -08001667 if (!addr)
1668 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001669
1670 /*
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001671 * In this function, newly allocated vm_struct is not added
1672 * to vmlist at __get_vm_area_node(). so, it is added here.
1673 */
1674 insert_vmalloc_vmlist(area);
1675
1676 /*
Catalin Marinas89219d32009-06-11 13:23:19 +01001677 * A ref_count = 3 is needed because the vm_struct and vmap_area
1678 * structures allocated in the __get_vm_area_node() function contain
1679 * references to the virtual address of the vmalloc'ed block.
1680 */
David Rientjesd0a21262011-01-13 15:46:02 -08001681 kmemleak_alloc(addr, real_size, 3, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001682
1683 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001684
1685fail:
1686 warn_alloc_failed(gfp_mask, 0,
1687 "vmalloc: allocation failure: %lu bytes\n",
1688 real_size);
1689 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001690}
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001693 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001695 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 * @gfp_mask: flags for the page level allocator
1697 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001698 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001699 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 *
1701 * Allocate enough pages to cover @size from the page level
1702 * allocator with @gfp_mask flags. Map them into contiguous
1703 * kernel virtual space, using a pagetable protection of @prot.
1704 */
David Miller2dca6992009-09-21 12:22:34 -07001705static void *__vmalloc_node(unsigned long size, unsigned long align,
1706 gfp_t gfp_mask, pgprot_t prot,
1707 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
David Rientjesd0a21262011-01-13 15:46:02 -08001709 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1710 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711}
1712
Christoph Lameter930fc452005-10-29 18:15:41 -07001713void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1714{
David Miller2dca6992009-09-21 12:22:34 -07001715 return __vmalloc_node(size, 1, gfp_mask, prot, -1,
Christoph Lameter23016962008-04-28 02:12:42 -07001716 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001717}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718EXPORT_SYMBOL(__vmalloc);
1719
Dave Younge1ca7782010-10-26 14:22:06 -07001720static inline void *__vmalloc_node_flags(unsigned long size,
1721 int node, gfp_t flags)
1722{
1723 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1724 node, __builtin_return_address(0));
1725}
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727/**
1728 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 * Allocate enough pages to cover @size from the page level
1731 * allocator and map them into contiguous kernel virtual space.
1732 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001733 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 * use __vmalloc() instead.
1735 */
1736void *vmalloc(unsigned long size)
1737{
Dave Younge1ca7782010-10-26 14:22:06 -07001738 return __vmalloc_node_flags(size, -1, GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740EXPORT_SYMBOL(vmalloc);
1741
Christoph Lameter930fc452005-10-29 18:15:41 -07001742/**
Dave Younge1ca7782010-10-26 14:22:06 -07001743 * vzalloc - allocate virtually contiguous memory with zero fill
1744 * @size: allocation size
1745 * Allocate enough pages to cover @size from the page level
1746 * allocator and map them into contiguous kernel virtual space.
1747 * The memory allocated is set to zero.
1748 *
1749 * For tight control over page level allocator and protection flags
1750 * use __vmalloc() instead.
1751 */
1752void *vzalloc(unsigned long size)
1753{
1754 return __vmalloc_node_flags(size, -1,
1755 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1756}
1757EXPORT_SYMBOL(vzalloc);
1758
1759/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001760 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1761 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001762 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001763 * The resulting memory area is zeroed so it can be mapped to userspace
1764 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001765 */
1766void *vmalloc_user(unsigned long size)
1767{
1768 struct vm_struct *area;
1769 void *ret;
1770
David Miller2dca6992009-09-21 12:22:34 -07001771 ret = __vmalloc_node(size, SHMLBA,
1772 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
Glauber Costa84877842009-01-06 14:39:19 -08001773 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001774 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001775 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001776 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001777 }
Nick Piggin83342312006-06-23 02:03:20 -07001778 return ret;
1779}
1780EXPORT_SYMBOL(vmalloc_user);
1781
1782/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001783 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001784 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001785 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001786 *
1787 * Allocate enough pages to cover @size from the page level
1788 * allocator and map them into contiguous kernel virtual space.
1789 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001790 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001791 * use __vmalloc() instead.
1792 */
1793void *vmalloc_node(unsigned long size, int node)
1794{
David Miller2dca6992009-09-21 12:22:34 -07001795 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001796 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001797}
1798EXPORT_SYMBOL(vmalloc_node);
1799
Dave Younge1ca7782010-10-26 14:22:06 -07001800/**
1801 * vzalloc_node - allocate memory on a specific node with zero fill
1802 * @size: allocation size
1803 * @node: numa node
1804 *
1805 * Allocate enough pages to cover @size from the page level
1806 * allocator and map them into contiguous kernel virtual space.
1807 * The memory allocated is set to zero.
1808 *
1809 * For tight control over page level allocator and protection flags
1810 * use __vmalloc_node() instead.
1811 */
1812void *vzalloc_node(unsigned long size, int node)
1813{
1814 return __vmalloc_node_flags(size, node,
1815 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1816}
1817EXPORT_SYMBOL(vzalloc_node);
1818
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001819#ifndef PAGE_KERNEL_EXEC
1820# define PAGE_KERNEL_EXEC PAGE_KERNEL
1821#endif
1822
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823/**
1824 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 * @size: allocation size
1826 *
1827 * Kernel-internal function to allocate enough pages to cover @size
1828 * the page level allocator and map them into contiguous and
1829 * executable kernel virtual space.
1830 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001831 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * use __vmalloc() instead.
1833 */
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835void *vmalloc_exec(unsigned long size)
1836{
David Miller2dca6992009-09-21 12:22:34 -07001837 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
Glauber Costa84877842009-01-06 14:39:19 -08001838 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839}
1840
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001841#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001842#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001843#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001844#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001845#else
1846#define GFP_VMALLOC32 GFP_KERNEL
1847#endif
1848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849/**
1850 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 * @size: allocation size
1852 *
1853 * Allocate enough 32bit PA addressable pages to cover @size from the
1854 * page level allocator and map them into contiguous kernel virtual space.
1855 */
1856void *vmalloc_32(unsigned long size)
1857{
David Miller2dca6992009-09-21 12:22:34 -07001858 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001859 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861EXPORT_SYMBOL(vmalloc_32);
1862
Nick Piggin83342312006-06-23 02:03:20 -07001863/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001864 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001865 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001866 *
1867 * The resulting memory area is 32bit addressable and zeroed so it can be
1868 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001869 */
1870void *vmalloc_32_user(unsigned long size)
1871{
1872 struct vm_struct *area;
1873 void *ret;
1874
David Miller2dca6992009-09-21 12:22:34 -07001875 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
Glauber Costa84877842009-01-06 14:39:19 -08001876 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001877 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001878 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001879 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001880 }
Nick Piggin83342312006-06-23 02:03:20 -07001881 return ret;
1882}
1883EXPORT_SYMBOL(vmalloc_32_user);
1884
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001885/*
1886 * small helper routine , copy contents to buf from addr.
1887 * If the page is not present, fill zero.
1888 */
1889
1890static int aligned_vread(char *buf, char *addr, unsigned long count)
1891{
1892 struct page *p;
1893 int copied = 0;
1894
1895 while (count) {
1896 unsigned long offset, length;
1897
1898 offset = (unsigned long)addr & ~PAGE_MASK;
1899 length = PAGE_SIZE - offset;
1900 if (length > count)
1901 length = count;
1902 p = vmalloc_to_page(addr);
1903 /*
1904 * To do safe access to this _mapped_ area, we need
1905 * lock. But adding lock here means that we need to add
1906 * overhead of vmalloc()/vfree() calles for this _debug_
1907 * interface, rarely used. Instead of that, we'll use
1908 * kmap() and get small overhead in this access function.
1909 */
1910 if (p) {
1911 /*
1912 * we can expect USER0 is not used (see vread/vwrite's
1913 * function description)
1914 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001915 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001916 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001917 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001918 } else
1919 memset(buf, 0, length);
1920
1921 addr += length;
1922 buf += length;
1923 copied += length;
1924 count -= length;
1925 }
1926 return copied;
1927}
1928
1929static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1930{
1931 struct page *p;
1932 int copied = 0;
1933
1934 while (count) {
1935 unsigned long offset, length;
1936
1937 offset = (unsigned long)addr & ~PAGE_MASK;
1938 length = PAGE_SIZE - offset;
1939 if (length > count)
1940 length = count;
1941 p = vmalloc_to_page(addr);
1942 /*
1943 * To do safe access to this _mapped_ area, we need
1944 * lock. But adding lock here means that we need to add
1945 * overhead of vmalloc()/vfree() calles for this _debug_
1946 * interface, rarely used. Instead of that, we'll use
1947 * kmap() and get small overhead in this access function.
1948 */
1949 if (p) {
1950 /*
1951 * we can expect USER0 is not used (see vread/vwrite's
1952 * function description)
1953 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001954 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001955 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001956 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001957 }
1958 addr += length;
1959 buf += length;
1960 copied += length;
1961 count -= length;
1962 }
1963 return copied;
1964}
1965
1966/**
1967 * vread() - read vmalloc area in a safe way.
1968 * @buf: buffer for reading data
1969 * @addr: vm address.
1970 * @count: number of bytes to be read.
1971 *
1972 * Returns # of bytes which addr and buf should be increased.
1973 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1974 * includes any intersect with alive vmalloc area.
1975 *
1976 * This function checks that addr is a valid vmalloc'ed area, and
1977 * copy data from that area to a given buffer. If the given memory range
1978 * of [addr...addr+count) includes some valid address, data is copied to
1979 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1980 * IOREMAP area is treated as memory hole and no copy is done.
1981 *
1982 * If [addr...addr+count) doesn't includes any intersects with alive
1983 * vm_struct area, returns 0.
1984 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1985 * the caller should guarantee KM_USER0 is not used.
1986 *
1987 * Note: In usual ops, vread() is never necessary because the caller
1988 * should know vmalloc() area is valid and can use memcpy().
1989 * This is for routines which have to access vmalloc area without
1990 * any informaion, as /dev/kmem.
1991 *
1992 */
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994long vread(char *buf, char *addr, unsigned long count)
1995{
1996 struct vm_struct *tmp;
1997 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001998 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 unsigned long n;
2000
2001 /* Don't allow overflow */
2002 if ((unsigned long) addr + count < count)
2003 count = -(unsigned long) addr;
2004
2005 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002006 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 vaddr = (char *) tmp->addr;
2008 if (addr >= vaddr + tmp->size - PAGE_SIZE)
2009 continue;
2010 while (addr < vaddr) {
2011 if (count == 0)
2012 goto finished;
2013 *buf = '\0';
2014 buf++;
2015 addr++;
2016 count--;
2017 }
2018 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002019 if (n > count)
2020 n = count;
2021 if (!(tmp->flags & VM_IOREMAP))
2022 aligned_vread(buf, addr, n);
2023 else /* IOREMAP area is treated as memory hole */
2024 memset(buf, 0, n);
2025 buf += n;
2026 addr += n;
2027 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 }
2029finished:
2030 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002031
2032 if (buf == buf_start)
2033 return 0;
2034 /* zero-fill memory holes */
2035 if (buf != buf_start + buflen)
2036 memset(buf, 0, buflen - (buf - buf_start));
2037
2038 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039}
2040
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002041/**
2042 * vwrite() - write vmalloc area in a safe way.
2043 * @buf: buffer for source data
2044 * @addr: vm address.
2045 * @count: number of bytes to be read.
2046 *
2047 * Returns # of bytes which addr and buf should be incresed.
2048 * (same number to @count).
2049 * If [addr...addr+count) doesn't includes any intersect with valid
2050 * vmalloc area, returns 0.
2051 *
2052 * This function checks that addr is a valid vmalloc'ed area, and
2053 * copy data from a buffer to the given addr. If specified range of
2054 * [addr...addr+count) includes some valid address, data is copied from
2055 * proper area of @buf. If there are memory holes, no copy to hole.
2056 * IOREMAP area is treated as memory hole and no copy is done.
2057 *
2058 * If [addr...addr+count) doesn't includes any intersects with alive
2059 * vm_struct area, returns 0.
2060 * @buf should be kernel's buffer. Because this function uses KM_USER0,
2061 * the caller should guarantee KM_USER0 is not used.
2062 *
2063 * Note: In usual ops, vwrite() is never necessary because the caller
2064 * should know vmalloc() area is valid and can use memcpy().
2065 * This is for routines which have to access vmalloc area without
2066 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002067 */
2068
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069long vwrite(char *buf, char *addr, unsigned long count)
2070{
2071 struct vm_struct *tmp;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002072 char *vaddr;
2073 unsigned long n, buflen;
2074 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
2076 /* Don't allow overflow */
2077 if ((unsigned long) addr + count < count)
2078 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002079 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
2081 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002082 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 vaddr = (char *) tmp->addr;
2084 if (addr >= vaddr + tmp->size - PAGE_SIZE)
2085 continue;
2086 while (addr < vaddr) {
2087 if (count == 0)
2088 goto finished;
2089 buf++;
2090 addr++;
2091 count--;
2092 }
2093 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002094 if (n > count)
2095 n = count;
2096 if (!(tmp->flags & VM_IOREMAP)) {
2097 aligned_vwrite(buf, addr, n);
2098 copied++;
2099 }
2100 buf += n;
2101 addr += n;
2102 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 }
2104finished:
2105 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002106 if (!copied)
2107 return 0;
2108 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109}
Nick Piggin83342312006-06-23 02:03:20 -07002110
2111/**
2112 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002113 * @vma: vma to cover (map full range of vma)
2114 * @addr: vmalloc memory
2115 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002116 *
2117 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002118 *
2119 * This function checks that addr is a valid vmalloc'ed area, and
2120 * that it is big enough to cover the vma. Will return failure if
2121 * that criteria isn't met.
2122 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002123 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002124 */
2125int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2126 unsigned long pgoff)
2127{
2128 struct vm_struct *area;
2129 unsigned long uaddr = vma->vm_start;
2130 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07002131
2132 if ((PAGE_SIZE-1) & (unsigned long)addr)
2133 return -EINVAL;
2134
Nick Piggindb64fe02008-10-18 20:27:03 -07002135 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07002136 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07002137 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002138
2139 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07002140 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002141
2142 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07002143 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002144
2145 addr += pgoff << PAGE_SHIFT;
2146 do {
2147 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07002148 int ret;
2149
Nick Piggin83342312006-06-23 02:03:20 -07002150 ret = vm_insert_page(vma, uaddr, page);
2151 if (ret)
2152 return ret;
2153
2154 uaddr += PAGE_SIZE;
2155 addr += PAGE_SIZE;
2156 usize -= PAGE_SIZE;
2157 } while (usize > 0);
2158
2159 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
2160 vma->vm_flags |= VM_RESERVED;
2161
Nick Piggindb64fe02008-10-18 20:27:03 -07002162 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07002163}
2164EXPORT_SYMBOL(remap_vmalloc_range);
2165
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002166/*
2167 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2168 * have one.
2169 */
2170void __attribute__((weak)) vmalloc_sync_all(void)
2171{
2172}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002173
2174
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002175static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002176{
David Vrabelcd129092011-09-29 16:53:32 +01002177 pte_t ***p = data;
2178
2179 if (p) {
2180 *(*p) = pte;
2181 (*p)++;
2182 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002183 return 0;
2184}
2185
2186/**
2187 * alloc_vm_area - allocate a range of kernel address space
2188 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002189 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002190 *
2191 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002192 *
2193 * This function reserves a range of kernel address space, and
2194 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002195 * are created.
2196 *
2197 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2198 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002199 */
David Vrabelcd129092011-09-29 16:53:32 +01002200struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002201{
2202 struct vm_struct *area;
2203
Christoph Lameter23016962008-04-28 02:12:42 -07002204 area = get_vm_area_caller(size, VM_IOREMAP,
2205 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002206 if (area == NULL)
2207 return NULL;
2208
2209 /*
2210 * This ensures that page tables are constructed for this region
2211 * of kernel virtual address space and mapped into init_mm.
2212 */
2213 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002214 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002215 free_vm_area(area);
2216 return NULL;
2217 }
2218
David Vrabeld63c8a02011-09-14 16:22:02 -07002219 /*
2220 * If the allocated address space is passed to a hypercall
2221 * before being used then we cannot rely on a page fault to
2222 * trigger an update of the page tables. So sync all the page
2223 * tables here.
2224 */
2225 vmalloc_sync_all();
2226
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002227 return area;
2228}
2229EXPORT_SYMBOL_GPL(alloc_vm_area);
2230
2231void free_vm_area(struct vm_struct *area)
2232{
2233 struct vm_struct *ret;
2234 ret = remove_vm_area(area->addr);
2235 BUG_ON(ret != area);
2236 kfree(area);
2237}
2238EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002239
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002240#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002241static struct vmap_area *node_to_va(struct rb_node *n)
2242{
2243 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2244}
2245
2246/**
2247 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2248 * @end: target address
2249 * @pnext: out arg for the next vmap_area
2250 * @pprev: out arg for the previous vmap_area
2251 *
2252 * Returns: %true if either or both of next and prev are found,
2253 * %false if no vmap_area exists
2254 *
2255 * Find vmap_areas end addresses of which enclose @end. ie. if not
2256 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2257 */
2258static bool pvm_find_next_prev(unsigned long end,
2259 struct vmap_area **pnext,
2260 struct vmap_area **pprev)
2261{
2262 struct rb_node *n = vmap_area_root.rb_node;
2263 struct vmap_area *va = NULL;
2264
2265 while (n) {
2266 va = rb_entry(n, struct vmap_area, rb_node);
2267 if (end < va->va_end)
2268 n = n->rb_left;
2269 else if (end > va->va_end)
2270 n = n->rb_right;
2271 else
2272 break;
2273 }
2274
2275 if (!va)
2276 return false;
2277
2278 if (va->va_end > end) {
2279 *pnext = va;
2280 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2281 } else {
2282 *pprev = va;
2283 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2284 }
2285 return true;
2286}
2287
2288/**
2289 * pvm_determine_end - find the highest aligned address between two vmap_areas
2290 * @pnext: in/out arg for the next vmap_area
2291 * @pprev: in/out arg for the previous vmap_area
2292 * @align: alignment
2293 *
2294 * Returns: determined end address
2295 *
2296 * Find the highest aligned address between *@pnext and *@pprev below
2297 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2298 * down address is between the end addresses of the two vmap_areas.
2299 *
2300 * Please note that the address returned by this function may fall
2301 * inside *@pnext vmap_area. The caller is responsible for checking
2302 * that.
2303 */
2304static unsigned long pvm_determine_end(struct vmap_area **pnext,
2305 struct vmap_area **pprev,
2306 unsigned long align)
2307{
2308 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2309 unsigned long addr;
2310
2311 if (*pnext)
2312 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2313 else
2314 addr = vmalloc_end;
2315
2316 while (*pprev && (*pprev)->va_end > addr) {
2317 *pnext = *pprev;
2318 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2319 }
2320
2321 return addr;
2322}
2323
2324/**
2325 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2326 * @offsets: array containing offset of each area
2327 * @sizes: array containing size of each area
2328 * @nr_vms: the number of areas to allocate
2329 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002330 *
2331 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2332 * vm_structs on success, %NULL on failure
2333 *
2334 * Percpu allocator wants to use congruent vm areas so that it can
2335 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002336 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2337 * be scattered pretty far, distance between two areas easily going up
2338 * to gigabytes. To avoid interacting with regular vmallocs, these
2339 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002340 *
2341 * Despite its complicated look, this allocator is rather simple. It
2342 * does everything top-down and scans areas from the end looking for
2343 * matching slot. While scanning, if any of the areas overlaps with
2344 * existing vmap_area, the base address is pulled down to fit the
2345 * area. Scanning is repeated till all the areas fit and then all
2346 * necessary data structres are inserted and the result is returned.
2347 */
2348struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2349 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002350 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002351{
2352 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2353 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2354 struct vmap_area **vas, *prev, *next;
2355 struct vm_struct **vms;
2356 int area, area2, last_area, term_area;
2357 unsigned long base, start, end, last_end;
2358 bool purged = false;
2359
Tejun Heoca23e402009-08-14 15:00:52 +09002360 /* verify parameters and allocate data structures */
2361 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2362 for (last_area = 0, area = 0; area < nr_vms; area++) {
2363 start = offsets[area];
2364 end = start + sizes[area];
2365
2366 /* is everything aligned properly? */
2367 BUG_ON(!IS_ALIGNED(offsets[area], align));
2368 BUG_ON(!IS_ALIGNED(sizes[area], align));
2369
2370 /* detect the area with the highest address */
2371 if (start > offsets[last_area])
2372 last_area = area;
2373
2374 for (area2 = 0; area2 < nr_vms; area2++) {
2375 unsigned long start2 = offsets[area2];
2376 unsigned long end2 = start2 + sizes[area2];
2377
2378 if (area2 == area)
2379 continue;
2380
2381 BUG_ON(start2 >= start && start2 < end);
2382 BUG_ON(end2 <= end && end2 > start);
2383 }
2384 }
2385 last_end = offsets[last_area] + sizes[last_area];
2386
2387 if (vmalloc_end - vmalloc_start < last_end) {
2388 WARN_ON(true);
2389 return NULL;
2390 }
2391
David Rientjesec3f64f2011-01-13 15:46:01 -08002392 vms = kzalloc(sizeof(vms[0]) * nr_vms, GFP_KERNEL);
2393 vas = kzalloc(sizeof(vas[0]) * nr_vms, GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002394 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002395 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002396
2397 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002398 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2399 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002400 if (!vas[area] || !vms[area])
2401 goto err_free;
2402 }
2403retry:
2404 spin_lock(&vmap_area_lock);
2405
2406 /* start scanning - we scan from the top, begin with the last area */
2407 area = term_area = last_area;
2408 start = offsets[area];
2409 end = start + sizes[area];
2410
2411 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2412 base = vmalloc_end - last_end;
2413 goto found;
2414 }
2415 base = pvm_determine_end(&next, &prev, align) - end;
2416
2417 while (true) {
2418 BUG_ON(next && next->va_end <= base + end);
2419 BUG_ON(prev && prev->va_end > base + end);
2420
2421 /*
2422 * base might have underflowed, add last_end before
2423 * comparing.
2424 */
2425 if (base + last_end < vmalloc_start + last_end) {
2426 spin_unlock(&vmap_area_lock);
2427 if (!purged) {
2428 purge_vmap_area_lazy();
2429 purged = true;
2430 goto retry;
2431 }
2432 goto err_free;
2433 }
2434
2435 /*
2436 * If next overlaps, move base downwards so that it's
2437 * right below next and then recheck.
2438 */
2439 if (next && next->va_start < base + end) {
2440 base = pvm_determine_end(&next, &prev, align) - end;
2441 term_area = area;
2442 continue;
2443 }
2444
2445 /*
2446 * If prev overlaps, shift down next and prev and move
2447 * base so that it's right below new next and then
2448 * recheck.
2449 */
2450 if (prev && prev->va_end > base + start) {
2451 next = prev;
2452 prev = node_to_va(rb_prev(&next->rb_node));
2453 base = pvm_determine_end(&next, &prev, align) - end;
2454 term_area = area;
2455 continue;
2456 }
2457
2458 /*
2459 * This area fits, move on to the previous one. If
2460 * the previous one is the terminal one, we're done.
2461 */
2462 area = (area + nr_vms - 1) % nr_vms;
2463 if (area == term_area)
2464 break;
2465 start = offsets[area];
2466 end = start + sizes[area];
2467 pvm_find_next_prev(base + end, &next, &prev);
2468 }
2469found:
2470 /* we've found a fitting base, insert all va's */
2471 for (area = 0; area < nr_vms; area++) {
2472 struct vmap_area *va = vas[area];
2473
2474 va->va_start = base + offsets[area];
2475 va->va_end = va->va_start + sizes[area];
2476 __insert_vmap_area(va);
2477 }
2478
2479 vmap_area_pcpu_hole = base + offsets[last_area];
2480
2481 spin_unlock(&vmap_area_lock);
2482
2483 /* insert all vm's */
2484 for (area = 0; area < nr_vms; area++)
2485 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2486 pcpu_get_vm_areas);
2487
2488 kfree(vas);
2489 return vms;
2490
2491err_free:
2492 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002493 kfree(vas[area]);
2494 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002495 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002496err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002497 kfree(vas);
2498 kfree(vms);
2499 return NULL;
2500}
2501
2502/**
2503 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2504 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2505 * @nr_vms: the number of allocated areas
2506 *
2507 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2508 */
2509void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2510{
2511 int i;
2512
2513 for (i = 0; i < nr_vms; i++)
2514 free_vm_area(vms[i]);
2515 kfree(vms);
2516}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002517#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002518
2519#ifdef CONFIG_PROC_FS
2520static void *s_start(struct seq_file *m, loff_t *pos)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002521 __acquires(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002522{
2523 loff_t n = *pos;
2524 struct vm_struct *v;
2525
2526 read_lock(&vmlist_lock);
2527 v = vmlist;
2528 while (n > 0 && v) {
2529 n--;
2530 v = v->next;
2531 }
2532 if (!n)
2533 return v;
2534
2535 return NULL;
2536
2537}
2538
2539static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2540{
2541 struct vm_struct *v = p;
2542
2543 ++*pos;
2544 return v->next;
2545}
2546
2547static void s_stop(struct seq_file *m, void *p)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002548 __releases(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002549{
2550 read_unlock(&vmlist_lock);
2551}
2552
Eric Dumazeta47a1262008-07-23 21:27:38 -07002553static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2554{
2555 if (NUMA_BUILD) {
2556 unsigned int nr, *counters = m->private;
2557
2558 if (!counters)
2559 return;
2560
2561 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2562
2563 for (nr = 0; nr < v->nr_pages; nr++)
2564 counters[page_to_nid(v->pages[nr])]++;
2565
2566 for_each_node_state(nr, N_HIGH_MEMORY)
2567 if (counters[nr])
2568 seq_printf(m, " N%u=%u", nr, counters[nr]);
2569 }
2570}
2571
Christoph Lametera10aa572008-04-28 02:12:40 -07002572static int s_show(struct seq_file *m, void *p)
2573{
2574 struct vm_struct *v = p;
2575
2576 seq_printf(m, "0x%p-0x%p %7ld",
2577 v->addr, v->addr + v->size, v->size);
2578
Joe Perches62c70bc2011-01-13 15:45:52 -08002579 if (v->caller)
2580 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002581
Christoph Lametera10aa572008-04-28 02:12:40 -07002582 if (v->nr_pages)
2583 seq_printf(m, " pages=%d", v->nr_pages);
2584
2585 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002586 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002587
2588 if (v->flags & VM_IOREMAP)
2589 seq_printf(m, " ioremap");
2590
2591 if (v->flags & VM_ALLOC)
2592 seq_printf(m, " vmalloc");
2593
2594 if (v->flags & VM_MAP)
2595 seq_printf(m, " vmap");
2596
2597 if (v->flags & VM_USERMAP)
2598 seq_printf(m, " user");
2599
2600 if (v->flags & VM_VPAGES)
2601 seq_printf(m, " vpages");
2602
Eric Dumazeta47a1262008-07-23 21:27:38 -07002603 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002604 seq_putc(m, '\n');
2605 return 0;
2606}
2607
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002608static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002609 .start = s_start,
2610 .next = s_next,
2611 .stop = s_stop,
2612 .show = s_show,
2613};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002614
2615static int vmalloc_open(struct inode *inode, struct file *file)
2616{
2617 unsigned int *ptr = NULL;
2618 int ret;
2619
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002620 if (NUMA_BUILD) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002621 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002622 if (ptr == NULL)
2623 return -ENOMEM;
2624 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002625 ret = seq_open(file, &vmalloc_op);
2626 if (!ret) {
2627 struct seq_file *m = file->private_data;
2628 m->private = ptr;
2629 } else
2630 kfree(ptr);
2631 return ret;
2632}
2633
2634static const struct file_operations proc_vmalloc_operations = {
2635 .open = vmalloc_open,
2636 .read = seq_read,
2637 .llseek = seq_lseek,
2638 .release = seq_release_private,
2639};
2640
2641static int __init proc_vmalloc_init(void)
2642{
2643 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2644 return 0;
2645}
2646module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07002647#endif
2648