blob: d365724feb05206fc0b37c852f5859499f1cb6b3 [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 Sharma600634972011-07-26 16:09:06 -070029#include <linux/atomic.h>
Al Viro32fcfd42013-03-10 20:14:08 -040030#include <linux/llist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/uaccess.h>
32#include <asm/tlbflush.h>
David Miller2dca6992009-09-21 12:22:34 -070033#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Al Viro32fcfd42013-03-10 20:14:08 -040035struct vfree_deferred {
36 struct llist_head list;
37 struct work_struct wq;
38};
39static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
40
41static void __vunmap(const void *, int);
42
43static void free_work(struct work_struct *w)
44{
45 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
46 struct llist_node *llnode = llist_del_all(&p->list);
47 while (llnode) {
48 void *p = llnode;
49 llnode = llist_next(llnode);
50 __vunmap(p, 1);
51 }
52}
53
Nick Piggindb64fe02008-10-18 20:27:03 -070054/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
57{
58 pte_t *pte;
59
60 pte = pte_offset_kernel(pmd, addr);
61 do {
62 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
63 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
64 } while (pte++, addr += PAGE_SIZE, addr != end);
65}
66
Nick Piggindb64fe02008-10-18 20:27:03 -070067static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
69 pmd_t *pmd;
70 unsigned long next;
71
72 pmd = pmd_offset(pud, addr);
73 do {
74 next = pmd_addr_end(addr, end);
75 if (pmd_none_or_clear_bad(pmd))
76 continue;
77 vunmap_pte_range(pmd, addr, next);
78 } while (pmd++, addr = next, addr != end);
79}
80
Nick Piggindb64fe02008-10-18 20:27:03 -070081static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
83 pud_t *pud;
84 unsigned long next;
85
86 pud = pud_offset(pgd, addr);
87 do {
88 next = pud_addr_end(addr, end);
89 if (pud_none_or_clear_bad(pud))
90 continue;
91 vunmap_pmd_range(pud, addr, next);
92 } while (pud++, addr = next, addr != end);
93}
94
Nick Piggindb64fe02008-10-18 20:27:03 -070095static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
97 pgd_t *pgd;
98 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100 BUG_ON(addr >= end);
101 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 do {
103 next = pgd_addr_end(addr, end);
104 if (pgd_none_or_clear_bad(pgd))
105 continue;
106 vunmap_pud_range(pgd, addr, next);
107 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109
110static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -0700111 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
113 pte_t *pte;
114
Nick Piggindb64fe02008-10-18 20:27:03 -0700115 /*
116 * nr is a running index into the array which helps higher level
117 * callers keep track of where we're up to.
118 */
119
Hugh Dickins872fec12005-10-29 18:16:21 -0700120 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 if (!pte)
122 return -ENOMEM;
123 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700124 struct page *page = pages[*nr];
125
126 if (WARN_ON(!pte_none(*pte)))
127 return -EBUSY;
128 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 return -ENOMEM;
130 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700131 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 } while (pte++, addr += PAGE_SIZE, addr != end);
133 return 0;
134}
135
Nick Piggindb64fe02008-10-18 20:27:03 -0700136static int vmap_pmd_range(pud_t *pud, unsigned long addr,
137 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
139 pmd_t *pmd;
140 unsigned long next;
141
142 pmd = pmd_alloc(&init_mm, pud, addr);
143 if (!pmd)
144 return -ENOMEM;
145 do {
146 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700147 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 return -ENOMEM;
149 } while (pmd++, addr = next, addr != end);
150 return 0;
151}
152
Nick Piggindb64fe02008-10-18 20:27:03 -0700153static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
154 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
156 pud_t *pud;
157 unsigned long next;
158
159 pud = pud_alloc(&init_mm, pgd, addr);
160 if (!pud)
161 return -ENOMEM;
162 do {
163 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700164 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 return -ENOMEM;
166 } while (pud++, addr = next, addr != end);
167 return 0;
168}
169
Nick Piggindb64fe02008-10-18 20:27:03 -0700170/*
171 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
172 * will have pfns corresponding to the "pages" array.
173 *
174 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
175 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900176static int vmap_page_range_noflush(unsigned long start, unsigned long end,
177 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
179 pgd_t *pgd;
180 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800181 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700182 int err = 0;
183 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
185 BUG_ON(addr >= end);
186 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 do {
188 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700189 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700191 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700193
Nick Piggindb64fe02008-10-18 20:27:03 -0700194 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195}
196
Tejun Heo8fc48982009-02-20 16:29:08 +0900197static int vmap_page_range(unsigned long start, unsigned long end,
198 pgprot_t prot, struct page **pages)
199{
200 int ret;
201
202 ret = vmap_page_range_noflush(start, end, prot, pages);
203 flush_cache_vmap(start, end);
204 return ret;
205}
206
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700207int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700208{
209 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000210 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700211 * and fall back on vmalloc() if that fails. Others
212 * just put it in the vmalloc space.
213 */
214#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
215 unsigned long addr = (unsigned long)x;
216 if (addr >= MODULES_VADDR && addr < MODULES_END)
217 return 1;
218#endif
219 return is_vmalloc_addr(x);
220}
221
Christoph Lameter48667e72008-02-04 22:28:31 -0800222/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700223 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800224 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800225struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800226{
227 unsigned long addr = (unsigned long) vmalloc_addr;
228 struct page *page = NULL;
229 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800230
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200231 /*
232 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
233 * architectures that do not vmalloc module space
234 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700235 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200236
Christoph Lameter48667e72008-02-04 22:28:31 -0800237 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700238 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800239 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700240 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800241 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700242 pte_t *ptep, pte;
243
Christoph Lameter48667e72008-02-04 22:28:31 -0800244 ptep = pte_offset_map(pmd, addr);
245 pte = *ptep;
246 if (pte_present(pte))
247 page = pte_page(pte);
248 pte_unmap(ptep);
249 }
250 }
251 }
252 return page;
253}
254EXPORT_SYMBOL(vmalloc_to_page);
255
256/*
257 * Map a vmalloc()-space virtual address to the physical page frame number.
258 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800259unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800260{
261 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
262}
263EXPORT_SYMBOL(vmalloc_to_pfn);
264
Nick Piggindb64fe02008-10-18 20:27:03 -0700265
266/*** Global kva allocator ***/
267
268#define VM_LAZY_FREE 0x01
269#define VM_LAZY_FREEING 0x02
270#define VM_VM_AREA 0x04
271
Nick Piggindb64fe02008-10-18 20:27:03 -0700272static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700273/* Export for kexec only */
274LIST_HEAD(vmap_area_list);
Nick Piggin89699602011-03-22 16:30:36 -0700275static struct rb_root vmap_area_root = RB_ROOT;
276
277/* The vmap cache globals are protected by vmap_area_lock */
278static struct rb_node *free_vmap_cache;
279static unsigned long cached_hole_size;
280static unsigned long cached_vstart;
281static unsigned long cached_align;
282
Tejun Heoca23e402009-08-14 15:00:52 +0900283static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700284
285static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
Nick Piggindb64fe02008-10-18 20:27:03 -0700287 struct rb_node *n = vmap_area_root.rb_node;
288
289 while (n) {
290 struct vmap_area *va;
291
292 va = rb_entry(n, struct vmap_area, rb_node);
293 if (addr < va->va_start)
294 n = n->rb_left;
295 else if (addr > va->va_start)
296 n = n->rb_right;
297 else
298 return va;
299 }
300
301 return NULL;
302}
303
304static void __insert_vmap_area(struct vmap_area *va)
305{
306 struct rb_node **p = &vmap_area_root.rb_node;
307 struct rb_node *parent = NULL;
308 struct rb_node *tmp;
309
310 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700311 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700312
313 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700314 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
315 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700316 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700317 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700318 p = &(*p)->rb_right;
319 else
320 BUG();
321 }
322
323 rb_link_node(&va->rb_node, parent, p);
324 rb_insert_color(&va->rb_node, &vmap_area_root);
325
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700326 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700327 tmp = rb_prev(&va->rb_node);
328 if (tmp) {
329 struct vmap_area *prev;
330 prev = rb_entry(tmp, struct vmap_area, rb_node);
331 list_add_rcu(&va->list, &prev->list);
332 } else
333 list_add_rcu(&va->list, &vmap_area_list);
334}
335
336static void purge_vmap_area_lazy(void);
337
338/*
339 * Allocate a region of KVA of the specified size and alignment, within the
340 * vstart and vend.
341 */
342static struct vmap_area *alloc_vmap_area(unsigned long size,
343 unsigned long align,
344 unsigned long vstart, unsigned long vend,
345 int node, gfp_t gfp_mask)
346{
347 struct vmap_area *va;
348 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700350 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700351 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700352
Nick Piggin77669702009-02-27 14:03:03 -0800353 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700354 BUG_ON(size & ~PAGE_MASK);
Nick Piggin89699602011-03-22 16:30:36 -0700355 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700356
Nick Piggindb64fe02008-10-18 20:27:03 -0700357 va = kmalloc_node(sizeof(struct vmap_area),
358 gfp_mask & GFP_RECLAIM_MASK, node);
359 if (unlikely(!va))
360 return ERR_PTR(-ENOMEM);
361
362retry:
363 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700364 /*
365 * Invalidate cache if we have more permissive parameters.
366 * cached_hole_size notes the largest hole noticed _below_
367 * the vmap_area cached in free_vmap_cache: if size fits
368 * into that hole, we want to scan from vstart to reuse
369 * the hole instead of allocating above free_vmap_cache.
370 * Note that __free_vmap_area may update free_vmap_cache
371 * without updating cached_hole_size or cached_align.
372 */
373 if (!free_vmap_cache ||
374 size < cached_hole_size ||
375 vstart < cached_vstart ||
376 align < cached_align) {
377nocache:
378 cached_hole_size = 0;
379 free_vmap_cache = NULL;
380 }
381 /* record if we encounter less permissive parameters */
382 cached_vstart = vstart;
383 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800384
Nick Piggin89699602011-03-22 16:30:36 -0700385 /* find starting point for our search */
386 if (free_vmap_cache) {
387 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700388 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700389 if (addr < vstart)
390 goto nocache;
391 if (addr + size - 1 < addr)
392 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700393
Nick Piggin89699602011-03-22 16:30:36 -0700394 } else {
395 addr = ALIGN(vstart, align);
396 if (addr + size - 1 < addr)
397 goto overflow;
398
399 n = vmap_area_root.rb_node;
400 first = NULL;
401
402 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700403 struct vmap_area *tmp;
404 tmp = rb_entry(n, struct vmap_area, rb_node);
405 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700406 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700407 if (tmp->va_start <= addr)
408 break;
409 n = n->rb_left;
410 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700411 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700412 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700413
414 if (!first)
415 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700416 }
Nick Piggin89699602011-03-22 16:30:36 -0700417
418 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700419 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700420 if (addr + cached_hole_size < first->va_start)
421 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700422 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700423 if (addr + size - 1 < addr)
424 goto overflow;
425
Hong zhi guo92ca9222012-07-31 16:41:35 -0700426 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700427 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700428
429 first = list_entry(first->list.next,
430 struct vmap_area, list);
Nick Piggin89699602011-03-22 16:30:36 -0700431 }
432
Nick Piggindb64fe02008-10-18 20:27:03 -0700433found:
Nick Piggin89699602011-03-22 16:30:36 -0700434 if (addr + size > vend)
435 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700436
437 va->va_start = addr;
438 va->va_end = addr + size;
439 va->flags = 0;
440 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700441 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700442 spin_unlock(&vmap_area_lock);
443
Nick Piggin89699602011-03-22 16:30:36 -0700444 BUG_ON(va->va_start & (align-1));
445 BUG_ON(va->va_start < vstart);
446 BUG_ON(va->va_end > vend);
447
Nick Piggindb64fe02008-10-18 20:27:03 -0700448 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700449
450overflow:
451 spin_unlock(&vmap_area_lock);
452 if (!purged) {
453 purge_vmap_area_lazy();
454 purged = 1;
455 goto retry;
456 }
457 if (printk_ratelimit())
458 printk(KERN_WARNING
459 "vmap allocation for size %lu failed: "
460 "use vmalloc=<size> to increase size.\n", size);
461 kfree(va);
462 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700463}
464
Nick Piggindb64fe02008-10-18 20:27:03 -0700465static void __free_vmap_area(struct vmap_area *va)
466{
467 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700468
469 if (free_vmap_cache) {
470 if (va->va_end < cached_vstart) {
471 free_vmap_cache = NULL;
472 } else {
473 struct vmap_area *cache;
474 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
475 if (va->va_start <= cache->va_start) {
476 free_vmap_cache = rb_prev(&va->rb_node);
477 /*
478 * We don't try to update cached_hole_size or
479 * cached_align, but it won't go very wrong.
480 */
481 }
482 }
483 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700484 rb_erase(&va->rb_node, &vmap_area_root);
485 RB_CLEAR_NODE(&va->rb_node);
486 list_del_rcu(&va->list);
487
Tejun Heoca23e402009-08-14 15:00:52 +0900488 /*
489 * Track the highest possible candidate for pcpu area
490 * allocation. Areas outside of vmalloc area can be returned
491 * here too, consider only end addresses which fall inside
492 * vmalloc area proper.
493 */
494 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
495 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
496
Lai Jiangshan14769de2011-03-18 12:12:19 +0800497 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700498}
499
500/*
501 * Free a region of KVA allocated by alloc_vmap_area
502 */
503static void free_vmap_area(struct vmap_area *va)
504{
505 spin_lock(&vmap_area_lock);
506 __free_vmap_area(va);
507 spin_unlock(&vmap_area_lock);
508}
509
510/*
511 * Clear the pagetable entries of a given vmap_area
512 */
513static void unmap_vmap_area(struct vmap_area *va)
514{
515 vunmap_page_range(va->va_start, va->va_end);
516}
517
Nick Piggincd528582009-01-06 14:39:20 -0800518static void vmap_debug_free_range(unsigned long start, unsigned long end)
519{
520 /*
521 * Unmap page tables and force a TLB flush immediately if
522 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
523 * bugs similarly to those in linear kernel virtual address
524 * space after a page has been freed.
525 *
526 * All the lazy freeing logic is still retained, in order to
527 * minimise intrusiveness of this debugging feature.
528 *
529 * This is going to be *slow* (linear kernel virtual address
530 * debugging doesn't do a broadcast TLB flush so it is a lot
531 * faster).
532 */
533#ifdef CONFIG_DEBUG_PAGEALLOC
534 vunmap_page_range(start, end);
535 flush_tlb_kernel_range(start, end);
536#endif
537}
538
Nick Piggindb64fe02008-10-18 20:27:03 -0700539/*
540 * lazy_max_pages is the maximum amount of virtual address space we gather up
541 * before attempting to purge with a TLB flush.
542 *
543 * There is a tradeoff here: a larger number will cover more kernel page tables
544 * and take slightly longer to purge, but it will linearly reduce the number of
545 * global TLB flushes that must be performed. It would seem natural to scale
546 * this number up linearly with the number of CPUs (because vmapping activity
547 * could also scale linearly with the number of CPUs), however it is likely
548 * that in practice, workloads might be constrained in other ways that mean
549 * vmap activity will not scale linearly with CPUs. Also, I want to be
550 * conservative and not introduce a big latency on huge systems, so go with
551 * a less aggressive log scale. It will still be an improvement over the old
552 * code, and it will be simple to change the scale factor if we find that it
553 * becomes a problem on bigger systems.
554 */
555static unsigned long lazy_max_pages(void)
556{
557 unsigned int log;
558
559 log = fls(num_online_cpus());
560
561 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
562}
563
564static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
565
Nick Piggin02b709d2010-02-01 22:25:57 +1100566/* for per-CPU blocks */
567static void purge_fragmented_blocks_allcpus(void);
568
Nick Piggindb64fe02008-10-18 20:27:03 -0700569/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500570 * called before a call to iounmap() if the caller wants vm_area_struct's
571 * immediately freed.
572 */
573void set_iounmap_nonlazy(void)
574{
575 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
576}
577
578/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700579 * Purges all lazily-freed vmap areas.
580 *
581 * If sync is 0 then don't purge if there is already a purge in progress.
582 * If force_flush is 1, then flush kernel TLBs between *start and *end even
583 * if we found no lazy vmap areas to unmap (callers can use this to optimise
584 * their own TLB flushing).
585 * Returns with *start = min(*start, lowest purged address)
586 * *end = max(*end, highest purged address)
587 */
588static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
589 int sync, int force_flush)
590{
Andrew Morton46666d82009-01-15 13:51:15 -0800591 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700592 LIST_HEAD(valist);
593 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800594 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700595 int nr = 0;
596
597 /*
598 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
599 * should not expect such behaviour. This just simplifies locking for
600 * the case that isn't actually used at the moment anyway.
601 */
602 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800603 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700604 return;
605 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800606 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700607
Nick Piggin02b709d2010-02-01 22:25:57 +1100608 if (sync)
609 purge_fragmented_blocks_allcpus();
610
Nick Piggindb64fe02008-10-18 20:27:03 -0700611 rcu_read_lock();
612 list_for_each_entry_rcu(va, &vmap_area_list, list) {
613 if (va->flags & VM_LAZY_FREE) {
614 if (va->va_start < *start)
615 *start = va->va_start;
616 if (va->va_end > *end)
617 *end = va->va_end;
618 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700619 list_add_tail(&va->purge_list, &valist);
620 va->flags |= VM_LAZY_FREEING;
621 va->flags &= ~VM_LAZY_FREE;
622 }
623 }
624 rcu_read_unlock();
625
Yongseok Koh88f50042010-01-19 17:33:49 +0900626 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700627 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700628
629 if (nr || force_flush)
630 flush_tlb_kernel_range(*start, *end);
631
632 if (nr) {
633 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800634 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700635 __free_vmap_area(va);
636 spin_unlock(&vmap_area_lock);
637 }
Andrew Morton46666d82009-01-15 13:51:15 -0800638 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700639}
640
641/*
Nick Piggin496850e2008-11-19 15:36:33 -0800642 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
643 * is already purging.
644 */
645static void try_purge_vmap_area_lazy(void)
646{
647 unsigned long start = ULONG_MAX, end = 0;
648
649 __purge_vmap_area_lazy(&start, &end, 0, 0);
650}
651
652/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700653 * Kick off a purge of the outstanding lazy areas.
654 */
655static void purge_vmap_area_lazy(void)
656{
657 unsigned long start = ULONG_MAX, end = 0;
658
Nick Piggin496850e2008-11-19 15:36:33 -0800659 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700660}
661
662/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800663 * Free a vmap area, caller ensuring that the area has been unmapped
664 * and flush_cache_vunmap had been called for the correct range
665 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700666 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800667static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700668{
669 va->flags |= VM_LAZY_FREE;
670 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
671 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800672 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700673}
674
Nick Pigginb29acbd2008-12-01 13:13:47 -0800675/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800676 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
677 * called for the correct range previously.
678 */
679static void free_unmap_vmap_area_noflush(struct vmap_area *va)
680{
681 unmap_vmap_area(va);
682 free_vmap_area_noflush(va);
683}
684
685/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800686 * Free and unmap a vmap area
687 */
688static void free_unmap_vmap_area(struct vmap_area *va)
689{
690 flush_cache_vunmap(va->va_start, va->va_end);
691 free_unmap_vmap_area_noflush(va);
692}
693
Nick Piggindb64fe02008-10-18 20:27:03 -0700694static struct vmap_area *find_vmap_area(unsigned long addr)
695{
696 struct vmap_area *va;
697
698 spin_lock(&vmap_area_lock);
699 va = __find_vmap_area(addr);
700 spin_unlock(&vmap_area_lock);
701
702 return va;
703}
704
705static void free_unmap_vmap_area_addr(unsigned long addr)
706{
707 struct vmap_area *va;
708
709 va = find_vmap_area(addr);
710 BUG_ON(!va);
711 free_unmap_vmap_area(va);
712}
713
714
715/*** Per cpu kva allocator ***/
716
717/*
718 * vmap space is limited especially on 32 bit architectures. Ensure there is
719 * room for at least 16 percpu vmap blocks per CPU.
720 */
721/*
722 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
723 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
724 * instead (we just need a rough idea)
725 */
726#if BITS_PER_LONG == 32
727#define VMALLOC_SPACE (128UL*1024*1024)
728#else
729#define VMALLOC_SPACE (128UL*1024*1024*1024)
730#endif
731
732#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
733#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
734#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
735#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
736#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
737#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200738#define VMAP_BBMAP_BITS \
739 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
740 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
741 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700742
743#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
744
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100745static bool vmap_initialized __read_mostly = false;
746
Nick Piggindb64fe02008-10-18 20:27:03 -0700747struct vmap_block_queue {
748 spinlock_t lock;
749 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700750};
751
752struct vmap_block {
753 spinlock_t lock;
754 struct vmap_area *va;
755 struct vmap_block_queue *vbq;
756 unsigned long free, dirty;
757 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
758 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100759 struct list_head free_list;
760 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100761 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700762};
763
764/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
765static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
766
767/*
768 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
769 * in the free path. Could get rid of this if we change the API to return a
770 * "cookie" from alloc, to be passed to free. But no big deal yet.
771 */
772static DEFINE_SPINLOCK(vmap_block_tree_lock);
773static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
774
775/*
776 * We should probably have a fallback mechanism to allocate virtual memory
777 * out of partially filled vmap blocks. However vmap block sizing should be
778 * fairly reasonable according to the vmalloc size, so it shouldn't be a
779 * big problem.
780 */
781
782static unsigned long addr_to_vb_idx(unsigned long addr)
783{
784 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
785 addr /= VMAP_BLOCK_SIZE;
786 return addr;
787}
788
789static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
790{
791 struct vmap_block_queue *vbq;
792 struct vmap_block *vb;
793 struct vmap_area *va;
794 unsigned long vb_idx;
795 int node, err;
796
797 node = numa_node_id();
798
799 vb = kmalloc_node(sizeof(struct vmap_block),
800 gfp_mask & GFP_RECLAIM_MASK, node);
801 if (unlikely(!vb))
802 return ERR_PTR(-ENOMEM);
803
804 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
805 VMALLOC_START, VMALLOC_END,
806 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800807 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700808 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700809 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700810 }
811
812 err = radix_tree_preload(gfp_mask);
813 if (unlikely(err)) {
814 kfree(vb);
815 free_vmap_area(va);
816 return ERR_PTR(err);
817 }
818
819 spin_lock_init(&vb->lock);
820 vb->va = va;
821 vb->free = VMAP_BBMAP_BITS;
822 vb->dirty = 0;
823 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
824 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
825 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700826
827 vb_idx = addr_to_vb_idx(va->va_start);
828 spin_lock(&vmap_block_tree_lock);
829 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
830 spin_unlock(&vmap_block_tree_lock);
831 BUG_ON(err);
832 radix_tree_preload_end();
833
834 vbq = &get_cpu_var(vmap_block_queue);
835 vb->vbq = vbq;
836 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100837 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700838 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900839 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700840
841 return vb;
842}
843
Nick Piggindb64fe02008-10-18 20:27:03 -0700844static void free_vmap_block(struct vmap_block *vb)
845{
846 struct vmap_block *tmp;
847 unsigned long vb_idx;
848
Nick Piggindb64fe02008-10-18 20:27:03 -0700849 vb_idx = addr_to_vb_idx(vb->va->va_start);
850 spin_lock(&vmap_block_tree_lock);
851 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
852 spin_unlock(&vmap_block_tree_lock);
853 BUG_ON(tmp != vb);
854
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800855 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800856 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700857}
858
Nick Piggin02b709d2010-02-01 22:25:57 +1100859static void purge_fragmented_blocks(int cpu)
860{
861 LIST_HEAD(purge);
862 struct vmap_block *vb;
863 struct vmap_block *n_vb;
864 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
865
866 rcu_read_lock();
867 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
868
869 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
870 continue;
871
872 spin_lock(&vb->lock);
873 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
874 vb->free = 0; /* prevent further allocs after releasing lock */
875 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
876 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
877 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
878 spin_lock(&vbq->lock);
879 list_del_rcu(&vb->free_list);
880 spin_unlock(&vbq->lock);
881 spin_unlock(&vb->lock);
882 list_add_tail(&vb->purge, &purge);
883 } else
884 spin_unlock(&vb->lock);
885 }
886 rcu_read_unlock();
887
888 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
889 list_del(&vb->purge);
890 free_vmap_block(vb);
891 }
892}
893
894static void purge_fragmented_blocks_thiscpu(void)
895{
896 purge_fragmented_blocks(smp_processor_id());
897}
898
899static void purge_fragmented_blocks_allcpus(void)
900{
901 int cpu;
902
903 for_each_possible_cpu(cpu)
904 purge_fragmented_blocks(cpu);
905}
906
Nick Piggindb64fe02008-10-18 20:27:03 -0700907static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
908{
909 struct vmap_block_queue *vbq;
910 struct vmap_block *vb;
911 unsigned long addr = 0;
912 unsigned int order;
Nick Piggin02b709d2010-02-01 22:25:57 +1100913 int purge = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700914
915 BUG_ON(size & ~PAGE_MASK);
916 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700917 if (WARN_ON(size == 0)) {
918 /*
919 * Allocating 0 bytes isn't what caller wants since
920 * get_order(0) returns funny result. Just warn and terminate
921 * early.
922 */
923 return NULL;
924 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700925 order = get_order(size);
926
927again:
928 rcu_read_lock();
929 vbq = &get_cpu_var(vmap_block_queue);
930 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
931 int i;
932
933 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100934 if (vb->free < 1UL << order)
935 goto next;
936
Nick Piggindb64fe02008-10-18 20:27:03 -0700937 i = bitmap_find_free_region(vb->alloc_map,
938 VMAP_BBMAP_BITS, order);
939
Nick Piggin02b709d2010-02-01 22:25:57 +1100940 if (i < 0) {
941 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
942 /* fragmented and no outstanding allocations */
943 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
944 purge = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -0700945 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100946 goto next;
947 }
948 addr = vb->va->va_start + (i << PAGE_SHIFT);
949 BUG_ON(addr_to_vb_idx(addr) !=
950 addr_to_vb_idx(vb->va->va_start));
951 vb->free -= 1UL << order;
952 if (vb->free == 0) {
953 spin_lock(&vbq->lock);
954 list_del_rcu(&vb->free_list);
955 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700956 }
957 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100958 break;
959next:
960 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700961 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100962
963 if (purge)
964 purge_fragmented_blocks_thiscpu();
965
Tejun Heo3f04ba82009-10-29 22:34:12 +0900966 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700967 rcu_read_unlock();
968
969 if (!addr) {
970 vb = new_vmap_block(gfp_mask);
971 if (IS_ERR(vb))
972 return vb;
973 goto again;
974 }
975
976 return (void *)addr;
977}
978
979static void vb_free(const void *addr, unsigned long size)
980{
981 unsigned long offset;
982 unsigned long vb_idx;
983 unsigned int order;
984 struct vmap_block *vb;
985
986 BUG_ON(size & ~PAGE_MASK);
987 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800988
989 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
990
Nick Piggindb64fe02008-10-18 20:27:03 -0700991 order = get_order(size);
992
993 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
994
995 vb_idx = addr_to_vb_idx((unsigned long)addr);
996 rcu_read_lock();
997 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
998 rcu_read_unlock();
999 BUG_ON(!vb);
1000
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001001 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1002
Nick Piggindb64fe02008-10-18 20:27:03 -07001003 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +11001004 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -07001005
Nick Piggindb64fe02008-10-18 20:27:03 -07001006 vb->dirty += 1UL << order;
1007 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001008 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001009 spin_unlock(&vb->lock);
1010 free_vmap_block(vb);
1011 } else
1012 spin_unlock(&vb->lock);
1013}
1014
1015/**
1016 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1017 *
1018 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1019 * to amortize TLB flushing overheads. What this means is that any page you
1020 * have now, may, in a former life, have been mapped into kernel virtual
1021 * address by the vmap layer and so there might be some CPUs with TLB entries
1022 * still referencing that page (additional to the regular 1:1 kernel mapping).
1023 *
1024 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1025 * be sure that none of the pages we have control over will have any aliases
1026 * from the vmap layer.
1027 */
1028void vm_unmap_aliases(void)
1029{
1030 unsigned long start = ULONG_MAX, end = 0;
1031 int cpu;
1032 int flush = 0;
1033
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001034 if (unlikely(!vmap_initialized))
1035 return;
1036
Nick Piggindb64fe02008-10-18 20:27:03 -07001037 for_each_possible_cpu(cpu) {
1038 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1039 struct vmap_block *vb;
1040
1041 rcu_read_lock();
1042 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1043 int i;
1044
1045 spin_lock(&vb->lock);
1046 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1047 while (i < VMAP_BBMAP_BITS) {
1048 unsigned long s, e;
1049 int j;
1050 j = find_next_zero_bit(vb->dirty_map,
1051 VMAP_BBMAP_BITS, i);
1052
1053 s = vb->va->va_start + (i << PAGE_SHIFT);
1054 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001055 flush = 1;
1056
1057 if (s < start)
1058 start = s;
1059 if (e > end)
1060 end = e;
1061
1062 i = j;
1063 i = find_next_bit(vb->dirty_map,
1064 VMAP_BBMAP_BITS, i);
1065 }
1066 spin_unlock(&vb->lock);
1067 }
1068 rcu_read_unlock();
1069 }
1070
1071 __purge_vmap_area_lazy(&start, &end, 1, flush);
1072}
1073EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1074
1075/**
1076 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1077 * @mem: the pointer returned by vm_map_ram
1078 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1079 */
1080void vm_unmap_ram(const void *mem, unsigned int count)
1081{
1082 unsigned long size = count << PAGE_SHIFT;
1083 unsigned long addr = (unsigned long)mem;
1084
1085 BUG_ON(!addr);
1086 BUG_ON(addr < VMALLOC_START);
1087 BUG_ON(addr > VMALLOC_END);
1088 BUG_ON(addr & (PAGE_SIZE-1));
1089
1090 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001091 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001092
1093 if (likely(count <= VMAP_MAX_ALLOC))
1094 vb_free(mem, size);
1095 else
1096 free_unmap_vmap_area_addr(addr);
1097}
1098EXPORT_SYMBOL(vm_unmap_ram);
1099
1100/**
1101 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1102 * @pages: an array of pointers to the pages to be mapped
1103 * @count: number of pages
1104 * @node: prefer to allocate data structures on this node
1105 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001106 *
1107 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001108 */
1109void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1110{
1111 unsigned long size = count << PAGE_SHIFT;
1112 unsigned long addr;
1113 void *mem;
1114
1115 if (likely(count <= VMAP_MAX_ALLOC)) {
1116 mem = vb_alloc(size, GFP_KERNEL);
1117 if (IS_ERR(mem))
1118 return NULL;
1119 addr = (unsigned long)mem;
1120 } else {
1121 struct vmap_area *va;
1122 va = alloc_vmap_area(size, PAGE_SIZE,
1123 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1124 if (IS_ERR(va))
1125 return NULL;
1126
1127 addr = va->va_start;
1128 mem = (void *)addr;
1129 }
1130 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1131 vm_unmap_ram(mem, count);
1132 return NULL;
1133 }
1134 return mem;
1135}
1136EXPORT_SYMBOL(vm_map_ram);
1137
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001138static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001139/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001140 * vm_area_add_early - add vmap area early during boot
1141 * @vm: vm_struct to add
1142 *
1143 * This function is used to add fixed kernel vm area to vmlist before
1144 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1145 * should contain proper values and the other fields should be zero.
1146 *
1147 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1148 */
1149void __init vm_area_add_early(struct vm_struct *vm)
1150{
1151 struct vm_struct *tmp, **p;
1152
1153 BUG_ON(vmap_initialized);
1154 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1155 if (tmp->addr >= vm->addr) {
1156 BUG_ON(tmp->addr < vm->addr + vm->size);
1157 break;
1158 } else
1159 BUG_ON(tmp->addr + tmp->size > vm->addr);
1160 }
1161 vm->next = *p;
1162 *p = vm;
1163}
1164
1165/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001166 * vm_area_register_early - register vmap area early during boot
1167 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001168 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001169 *
1170 * This function is used to register kernel vm area before
1171 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1172 * proper values on entry and other fields should be zero. On return,
1173 * vm->addr contains the allocated address.
1174 *
1175 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1176 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001177void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001178{
1179 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001180 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001181
Tejun Heoc0c0a292009-02-24 11:57:21 +09001182 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1183 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1184
1185 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001186
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001187 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001188}
1189
Nick Piggindb64fe02008-10-18 20:27:03 -07001190void __init vmalloc_init(void)
1191{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001192 struct vmap_area *va;
1193 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001194 int i;
1195
1196 for_each_possible_cpu(i) {
1197 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001198 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001199
1200 vbq = &per_cpu(vmap_block_queue, i);
1201 spin_lock_init(&vbq->lock);
1202 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001203 p = &per_cpu(vfree_deferred, i);
1204 init_llist_head(&p->list);
1205 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001206 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001207
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001208 /* Import existing vmlist entries. */
1209 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001210 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001211 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001212 va->va_start = (unsigned long)tmp->addr;
1213 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001214 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001215 __insert_vmap_area(va);
1216 }
Tejun Heoca23e402009-08-14 15:00:52 +09001217
1218 vmap_area_pcpu_hole = VMALLOC_END;
1219
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001220 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001221}
1222
Tejun Heo8fc48982009-02-20 16:29:08 +09001223/**
1224 * map_kernel_range_noflush - map kernel VM area with the specified pages
1225 * @addr: start of the VM area to map
1226 * @size: size of the VM area to map
1227 * @prot: page protection flags to use
1228 * @pages: pages to map
1229 *
1230 * Map 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_vmap() on to-be-mapped areas
1237 * before calling this function.
1238 *
1239 * RETURNS:
1240 * The number of pages mapped on success, -errno on failure.
1241 */
1242int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1243 pgprot_t prot, struct page **pages)
1244{
1245 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1246}
1247
1248/**
1249 * unmap_kernel_range_noflush - unmap kernel VM area
1250 * @addr: start of the VM area to unmap
1251 * @size: size of the VM area to unmap
1252 *
1253 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1254 * specify should have been allocated using get_vm_area() and its
1255 * friends.
1256 *
1257 * NOTE:
1258 * This function does NOT do any cache flushing. The caller is
1259 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1260 * before calling this function and flush_tlb_kernel_range() after.
1261 */
1262void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1263{
1264 vunmap_page_range(addr, addr + size);
1265}
Huang Ying81e88fd2011-01-12 14:44:55 +08001266EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001267
1268/**
1269 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1270 * @addr: start of the VM area to unmap
1271 * @size: size of the VM area to unmap
1272 *
1273 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1274 * the unmapping and tlb after.
1275 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001276void unmap_kernel_range(unsigned long addr, unsigned long size)
1277{
1278 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001279
1280 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001281 vunmap_page_range(addr, end);
1282 flush_tlb_kernel_range(addr, end);
1283}
1284
1285int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1286{
1287 unsigned long addr = (unsigned long)area->addr;
1288 unsigned long end = addr + area->size - PAGE_SIZE;
1289 int err;
1290
1291 err = vmap_page_range(addr, end, prot, *pages);
1292 if (err > 0) {
1293 *pages += err;
1294 err = 0;
1295 }
1296
1297 return err;
1298}
1299EXPORT_SYMBOL_GPL(map_vm_area);
1300
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001301static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001302 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001303{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001304 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001305 vm->flags = flags;
1306 vm->addr = (void *)va->va_start;
1307 vm->size = va->va_end - va->va_start;
1308 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001309 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001310 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001311 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001312}
Tejun Heocf88c792009-08-14 15:00:52 +09001313
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001314static void clear_vm_unlist(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001315{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001316 /*
1317 * Before removing VM_UNLIST,
1318 * we should make sure that vm has proper values.
1319 * Pair with smp_rmb() in show_numa_info().
1320 */
1321 smp_wmb();
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001322 vm->flags &= ~VM_UNLIST;
Tejun Heocf88c792009-08-14 15:00:52 +09001323}
1324
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001325static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001326 unsigned long flags, const void *caller)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001327{
1328 setup_vmalloc_vm(vm, va, flags, caller);
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001329 clear_vm_unlist(vm);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001330}
1331
Nick Piggindb64fe02008-10-18 20:27:03 -07001332static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001333 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001334 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001335{
Kautuk Consul00065262011-12-19 17:12:04 -08001336 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001337 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001339 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (flags & VM_IOREMAP) {
1341 int bit = fls(size);
1342
1343 if (bit > IOREMAP_MAX_ORDER)
1344 bit = IOREMAP_MAX_ORDER;
1345 else if (bit < PAGE_SHIFT)
1346 bit = PAGE_SHIFT;
1347
1348 align = 1ul << bit;
1349 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001352 if (unlikely(!size))
1353 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Tejun Heocf88c792009-08-14 15:00:52 +09001355 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 if (unlikely(!area))
1357 return NULL;
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /*
1360 * We always allocate a guard page.
1361 */
1362 size += PAGE_SIZE;
1363
Nick Piggindb64fe02008-10-18 20:27:03 -07001364 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1365 if (IS_ERR(va)) {
1366 kfree(area);
1367 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001370 /*
1371 * When this function is called from __vmalloc_node_range,
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001372 * we add VM_UNLIST flag to avoid accessing uninitialized
1373 * members of vm_struct such as pages and nr_pages fields.
1374 * They will be set later.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001375 */
1376 if (flags & VM_UNLIST)
1377 setup_vmalloc_vm(area, va, flags, caller);
1378 else
1379 insert_vmalloc_vm(area, va, flags, caller);
1380
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382}
1383
Christoph Lameter930fc452005-10-29 18:15:41 -07001384struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1385 unsigned long start, unsigned long end)
1386{
David Rientjes00ef2d22013-02-22 16:35:36 -08001387 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1388 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001389}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001390EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001391
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001392struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1393 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001394 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001395{
David Rientjes00ef2d22013-02-22 16:35:36 -08001396 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1397 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001398}
1399
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400/**
Simon Arlott183ff222007-10-20 01:27:18 +02001401 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 * @size: size of the area
1403 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1404 *
1405 * Search an area of @size in the kernel virtual mapping area,
1406 * and reserved it for out purposes. Returns the area descriptor
1407 * on success or %NULL on failure.
1408 */
1409struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1410{
David Miller2dca6992009-09-21 12:22:34 -07001411 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001412 NUMA_NO_NODE, GFP_KERNEL,
1413 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001414}
1415
1416struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001417 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001418{
David Miller2dca6992009-09-21 12:22:34 -07001419 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001420 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421}
1422
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001423/**
1424 * find_vm_area - find a continuous kernel virtual area
1425 * @addr: base address
1426 *
1427 * Search for the kernel VM area starting at @addr, and return it.
1428 * It is up to the caller to do all required locking to keep the returned
1429 * pointer valid.
1430 */
1431struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001432{
Nick Piggindb64fe02008-10-18 20:27:03 -07001433 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001434
Nick Piggindb64fe02008-10-18 20:27:03 -07001435 va = find_vmap_area((unsigned long)addr);
1436 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001437 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001438
Andi Kleen7856dfe2005-05-20 14:27:57 -07001439 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001440}
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442/**
Simon Arlott183ff222007-10-20 01:27:18 +02001443 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 * @addr: base address
1445 *
1446 * Search for the kernel VM area starting at @addr, and remove it.
1447 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001448 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001450struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
Nick Piggindb64fe02008-10-18 20:27:03 -07001452 struct vmap_area *va;
1453
1454 va = find_vmap_area((unsigned long)addr);
1455 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001456 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001457
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001458 spin_lock(&vmap_area_lock);
1459 va->vm = NULL;
1460 va->flags &= ~VM_VM_AREA;
1461 spin_unlock(&vmap_area_lock);
1462
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001463 vmap_debug_free_range(va->va_start, va->va_end);
1464 free_unmap_vmap_area(va);
1465 vm->size -= PAGE_SIZE;
1466
Nick Piggindb64fe02008-10-18 20:27:03 -07001467 return vm;
1468 }
1469 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001472static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
1474 struct vm_struct *area;
1475
1476 if (!addr)
1477 return;
1478
1479 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001480 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 return;
1482 }
1483
1484 area = remove_vm_area(addr);
1485 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001486 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 return;
1489 }
1490
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001491 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001492 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if (deallocate_pages) {
1495 int i;
1496
1497 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001498 struct page *page = area->pages[i];
1499
1500 BUG_ON(!page);
1501 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 }
1503
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001504 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 vfree(area->pages);
1506 else
1507 kfree(area->pages);
1508 }
1509
1510 kfree(area);
1511 return;
1512}
Al Viro32fcfd42013-03-10 20:14:08 -04001513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514/**
1515 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 * @addr: memory base address
1517 *
Simon Arlott183ff222007-10-20 01:27:18 +02001518 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001519 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1520 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
Al Viro32fcfd42013-03-10 20:14:08 -04001522 * Must not be called in NMI context (strictly speaking, only if we don't
1523 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1524 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001525 *
1526 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Al Viro32fcfd42013-03-10 20:14:08 -04001527 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001529void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
Al Viro32fcfd42013-03-10 20:14:08 -04001531 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001532
1533 kmemleak_free(addr);
1534
Al Viro32fcfd42013-03-10 20:14:08 -04001535 if (!addr)
1536 return;
1537 if (unlikely(in_interrupt())) {
1538 struct vfree_deferred *p = &__get_cpu_var(vfree_deferred);
1539 llist_add((struct llist_node *)addr, &p->list);
1540 schedule_work(&p->wq);
1541 } else
1542 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544EXPORT_SYMBOL(vfree);
1545
1546/**
1547 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 * @addr: memory base address
1549 *
1550 * Free the virtually contiguous memory area starting at @addr,
1551 * which was created from the page array passed to vmap().
1552 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001553 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001555void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556{
1557 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001558 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001559 if (addr)
1560 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562EXPORT_SYMBOL(vunmap);
1563
1564/**
1565 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 * @pages: array of page pointers
1567 * @count: number of pages to map
1568 * @flags: vm_area->flags
1569 * @prot: page protection for the mapping
1570 *
1571 * Maps @count pages from @pages into contiguous kernel virtual
1572 * space.
1573 */
1574void *vmap(struct page **pages, unsigned int count,
1575 unsigned long flags, pgprot_t prot)
1576{
1577 struct vm_struct *area;
1578
Peter Zijlstra34754b62009-02-25 16:04:03 +01001579 might_sleep();
1580
Jan Beulich44813742009-09-21 17:03:05 -07001581 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 return NULL;
1583
Christoph Lameter23016962008-04-28 02:12:42 -07001584 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1585 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (!area)
1587 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 if (map_vm_area(area, prot, &pages)) {
1590 vunmap(area->addr);
1591 return NULL;
1592 }
1593
1594 return area->addr;
1595}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596EXPORT_SYMBOL(vmap);
1597
David Miller2dca6992009-09-21 12:22:34 -07001598static void *__vmalloc_node(unsigned long size, unsigned long align,
1599 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001600 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001601static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001602 pgprot_t prot, int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
Dave Hansen22943ab2011-05-24 17:12:18 -07001604 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 struct page **pages;
1606 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001607 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1610 array_size = (nr_pages * sizeof(struct page *));
1611
1612 area->nr_pages = nr_pages;
1613 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001614 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001615 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Christoph Lameter23016962008-04-28 02:12:42 -07001616 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001617 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001618 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001619 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001622 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 if (!area->pages) {
1624 remove_vm_area(area->addr);
1625 kfree(area);
1626 return NULL;
1627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001630 struct page *page;
Dave Hansen22943ab2011-05-24 17:12:18 -07001631 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001632
Christoph Lameter930fc452005-10-29 18:15:41 -07001633 if (node < 0)
Dave Hansen22943ab2011-05-24 17:12:18 -07001634 page = alloc_page(tmp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001635 else
Dave Hansen22943ab2011-05-24 17:12:18 -07001636 page = alloc_pages_node(node, tmp_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001637
1638 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 /* Successfully allocated i pages, free them in __vunmap() */
1640 area->nr_pages = i;
1641 goto fail;
1642 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001643 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 }
1645
1646 if (map_vm_area(area, prot, &pages))
1647 goto fail;
1648 return area->addr;
1649
1650fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001651 warn_alloc_failed(gfp_mask, order,
1652 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001653 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 vfree(area->addr);
1655 return NULL;
1656}
1657
David Rientjesd0a21262011-01-13 15:46:02 -08001658/**
1659 * __vmalloc_node_range - allocate virtually contiguous memory
1660 * @size: allocation size
1661 * @align: desired alignment
1662 * @start: vm area range start
1663 * @end: vm area range end
1664 * @gfp_mask: flags for the page level allocator
1665 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001666 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001667 * @caller: caller's return address
1668 *
1669 * Allocate enough pages to cover @size from the page level
1670 * allocator with @gfp_mask flags. Map them into contiguous
1671 * kernel virtual space, using a pagetable protection of @prot.
1672 */
1673void *__vmalloc_node_range(unsigned long size, unsigned long align,
1674 unsigned long start, unsigned long end, gfp_t gfp_mask,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001675 pgprot_t prot, int node, const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001676{
David Rientjesd0a21262011-01-13 15:46:02 -08001677 struct vm_struct *area;
1678 void *addr;
1679 unsigned long real_size = size;
1680
1681 size = PAGE_ALIGN(size);
1682 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001683 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001684
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001685 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1686 start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001687 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001688 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001689
1690 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Mel Gorman1368edf2011-12-08 14:34:30 -08001691 if (!addr)
1692 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001693
1694 /*
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001695 * In this function, newly allocated vm_struct has VM_UNLIST flag.
1696 * It means that vm_struct is not fully initialized.
1697 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001698 */
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001699 clear_vm_unlist(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001700
1701 /*
Catalin Marinas89219d32009-06-11 13:23:19 +01001702 * A ref_count = 3 is needed because the vm_struct and vmap_area
1703 * structures allocated in the __get_vm_area_node() function contain
1704 * references to the virtual address of the vmalloc'ed block.
1705 */
David Rientjesd0a21262011-01-13 15:46:02 -08001706 kmemleak_alloc(addr, real_size, 3, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001707
1708 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001709
1710fail:
1711 warn_alloc_failed(gfp_mask, 0,
1712 "vmalloc: allocation failure: %lu bytes\n",
1713 real_size);
1714 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001715}
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001718 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001720 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 * @gfp_mask: flags for the page level allocator
1722 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001723 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001724 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 *
1726 * Allocate enough pages to cover @size from the page level
1727 * allocator with @gfp_mask flags. Map them into contiguous
1728 * kernel virtual space, using a pagetable protection of @prot.
1729 */
David Miller2dca6992009-09-21 12:22:34 -07001730static void *__vmalloc_node(unsigned long size, unsigned long align,
1731 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001732 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733{
David Rientjesd0a21262011-01-13 15:46:02 -08001734 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1735 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736}
1737
Christoph Lameter930fc452005-10-29 18:15:41 -07001738void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1739{
David Rientjes00ef2d22013-02-22 16:35:36 -08001740 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001741 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001742}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743EXPORT_SYMBOL(__vmalloc);
1744
Dave Younge1ca7782010-10-26 14:22:06 -07001745static inline void *__vmalloc_node_flags(unsigned long size,
1746 int node, gfp_t flags)
1747{
1748 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1749 node, __builtin_return_address(0));
1750}
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752/**
1753 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 * Allocate enough pages to cover @size from the page level
1756 * allocator and map them into contiguous kernel virtual space.
1757 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001758 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 * use __vmalloc() instead.
1760 */
1761void *vmalloc(unsigned long size)
1762{
David Rientjes00ef2d22013-02-22 16:35:36 -08001763 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1764 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766EXPORT_SYMBOL(vmalloc);
1767
Christoph Lameter930fc452005-10-29 18:15:41 -07001768/**
Dave Younge1ca7782010-10-26 14:22:06 -07001769 * vzalloc - allocate virtually contiguous memory with zero fill
1770 * @size: allocation size
1771 * Allocate enough pages to cover @size from the page level
1772 * allocator and map them into contiguous kernel virtual space.
1773 * The memory allocated is set to zero.
1774 *
1775 * For tight control over page level allocator and protection flags
1776 * use __vmalloc() instead.
1777 */
1778void *vzalloc(unsigned long size)
1779{
David Rientjes00ef2d22013-02-22 16:35:36 -08001780 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001781 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1782}
1783EXPORT_SYMBOL(vzalloc);
1784
1785/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001786 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1787 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001788 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001789 * The resulting memory area is zeroed so it can be mapped to userspace
1790 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001791 */
1792void *vmalloc_user(unsigned long size)
1793{
1794 struct vm_struct *area;
1795 void *ret;
1796
David Miller2dca6992009-09-21 12:22:34 -07001797 ret = __vmalloc_node(size, SHMLBA,
1798 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001799 PAGE_KERNEL, NUMA_NO_NODE,
1800 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001801 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001802 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001803 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001804 }
Nick Piggin83342312006-06-23 02:03:20 -07001805 return ret;
1806}
1807EXPORT_SYMBOL(vmalloc_user);
1808
1809/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001810 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001811 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001812 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001813 *
1814 * Allocate enough pages to cover @size from the page level
1815 * allocator and map them into contiguous kernel virtual space.
1816 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001817 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001818 * use __vmalloc() instead.
1819 */
1820void *vmalloc_node(unsigned long size, int node)
1821{
David Miller2dca6992009-09-21 12:22:34 -07001822 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001823 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001824}
1825EXPORT_SYMBOL(vmalloc_node);
1826
Dave Younge1ca7782010-10-26 14:22:06 -07001827/**
1828 * vzalloc_node - allocate memory on a specific node with zero fill
1829 * @size: allocation size
1830 * @node: numa node
1831 *
1832 * Allocate enough pages to cover @size from the page level
1833 * allocator and map them into contiguous kernel virtual space.
1834 * The memory allocated is set to zero.
1835 *
1836 * For tight control over page level allocator and protection flags
1837 * use __vmalloc_node() instead.
1838 */
1839void *vzalloc_node(unsigned long size, int node)
1840{
1841 return __vmalloc_node_flags(size, node,
1842 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1843}
1844EXPORT_SYMBOL(vzalloc_node);
1845
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001846#ifndef PAGE_KERNEL_EXEC
1847# define PAGE_KERNEL_EXEC PAGE_KERNEL
1848#endif
1849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850/**
1851 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 * @size: allocation size
1853 *
1854 * Kernel-internal function to allocate enough pages to cover @size
1855 * the page level allocator and map them into contiguous and
1856 * executable kernel virtual space.
1857 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001858 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 * use __vmalloc() instead.
1860 */
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862void *vmalloc_exec(unsigned long size)
1863{
David Miller2dca6992009-09-21 12:22:34 -07001864 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001865 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001868#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001869#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001870#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001871#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001872#else
1873#define GFP_VMALLOC32 GFP_KERNEL
1874#endif
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876/**
1877 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 * @size: allocation size
1879 *
1880 * Allocate enough 32bit PA addressable pages to cover @size from the
1881 * page level allocator and map them into contiguous kernel virtual space.
1882 */
1883void *vmalloc_32(unsigned long size)
1884{
David Miller2dca6992009-09-21 12:22:34 -07001885 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001886 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888EXPORT_SYMBOL(vmalloc_32);
1889
Nick Piggin83342312006-06-23 02:03:20 -07001890/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001891 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001892 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001893 *
1894 * The resulting memory area is 32bit addressable and zeroed so it can be
1895 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001896 */
1897void *vmalloc_32_user(unsigned long size)
1898{
1899 struct vm_struct *area;
1900 void *ret;
1901
David Miller2dca6992009-09-21 12:22:34 -07001902 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001903 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001904 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001905 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001906 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001907 }
Nick Piggin83342312006-06-23 02:03:20 -07001908 return ret;
1909}
1910EXPORT_SYMBOL(vmalloc_32_user);
1911
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001912/*
1913 * small helper routine , copy contents to buf from addr.
1914 * If the page is not present, fill zero.
1915 */
1916
1917static int aligned_vread(char *buf, char *addr, unsigned long count)
1918{
1919 struct page *p;
1920 int copied = 0;
1921
1922 while (count) {
1923 unsigned long offset, length;
1924
1925 offset = (unsigned long)addr & ~PAGE_MASK;
1926 length = PAGE_SIZE - offset;
1927 if (length > count)
1928 length = count;
1929 p = vmalloc_to_page(addr);
1930 /*
1931 * To do safe access to this _mapped_ area, we need
1932 * lock. But adding lock here means that we need to add
1933 * overhead of vmalloc()/vfree() calles for this _debug_
1934 * interface, rarely used. Instead of that, we'll use
1935 * kmap() and get small overhead in this access function.
1936 */
1937 if (p) {
1938 /*
1939 * we can expect USER0 is not used (see vread/vwrite's
1940 * function description)
1941 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001942 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001943 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001944 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001945 } else
1946 memset(buf, 0, length);
1947
1948 addr += length;
1949 buf += length;
1950 copied += length;
1951 count -= length;
1952 }
1953 return copied;
1954}
1955
1956static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1957{
1958 struct page *p;
1959 int copied = 0;
1960
1961 while (count) {
1962 unsigned long offset, length;
1963
1964 offset = (unsigned long)addr & ~PAGE_MASK;
1965 length = PAGE_SIZE - offset;
1966 if (length > count)
1967 length = count;
1968 p = vmalloc_to_page(addr);
1969 /*
1970 * To do safe access to this _mapped_ area, we need
1971 * lock. But adding lock here means that we need to add
1972 * overhead of vmalloc()/vfree() calles for this _debug_
1973 * interface, rarely used. Instead of that, we'll use
1974 * kmap() and get small overhead in this access function.
1975 */
1976 if (p) {
1977 /*
1978 * we can expect USER0 is not used (see vread/vwrite's
1979 * function description)
1980 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001981 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001982 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001983 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001984 }
1985 addr += length;
1986 buf += length;
1987 copied += length;
1988 count -= length;
1989 }
1990 return copied;
1991}
1992
1993/**
1994 * vread() - read vmalloc area in a safe way.
1995 * @buf: buffer for reading data
1996 * @addr: vm address.
1997 * @count: number of bytes to be read.
1998 *
1999 * Returns # of bytes which addr and buf should be increased.
2000 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2001 * includes any intersect with alive vmalloc area.
2002 *
2003 * This function checks that addr is a valid vmalloc'ed area, and
2004 * copy data from that area to a given buffer. If the given memory range
2005 * of [addr...addr+count) includes some valid address, data is copied to
2006 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2007 * IOREMAP area is treated as memory hole and no copy is done.
2008 *
2009 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002010 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002011 *
2012 * Note: In usual ops, vread() is never necessary because the caller
2013 * should know vmalloc() area is valid and can use memcpy().
2014 * This is for routines which have to access vmalloc area without
2015 * any informaion, as /dev/kmem.
2016 *
2017 */
2018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019long vread(char *buf, char *addr, unsigned long count)
2020{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002021 struct vmap_area *va;
2022 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002024 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 unsigned long n;
2026
2027 /* Don't allow overflow */
2028 if ((unsigned long) addr + count < count)
2029 count = -(unsigned long) addr;
2030
Joonsoo Kime81ce852013-04-29 15:07:32 -07002031 spin_lock(&vmap_area_lock);
2032 list_for_each_entry(va, &vmap_area_list, list) {
2033 if (!count)
2034 break;
2035
2036 if (!(va->flags & VM_VM_AREA))
2037 continue;
2038
2039 vm = va->vm;
2040 vaddr = (char *) vm->addr;
2041 if (addr >= vaddr + vm->size - PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 continue;
2043 while (addr < vaddr) {
2044 if (count == 0)
2045 goto finished;
2046 *buf = '\0';
2047 buf++;
2048 addr++;
2049 count--;
2050 }
Joonsoo Kime81ce852013-04-29 15:07:32 -07002051 n = vaddr + vm->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002052 if (n > count)
2053 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002054 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002055 aligned_vread(buf, addr, n);
2056 else /* IOREMAP area is treated as memory hole */
2057 memset(buf, 0, n);
2058 buf += n;
2059 addr += n;
2060 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 }
2062finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002063 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002064
2065 if (buf == buf_start)
2066 return 0;
2067 /* zero-fill memory holes */
2068 if (buf != buf_start + buflen)
2069 memset(buf, 0, buflen - (buf - buf_start));
2070
2071 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072}
2073
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002074/**
2075 * vwrite() - write vmalloc area in a safe way.
2076 * @buf: buffer for source data
2077 * @addr: vm address.
2078 * @count: number of bytes to be read.
2079 *
2080 * Returns # of bytes which addr and buf should be incresed.
2081 * (same number to @count).
2082 * If [addr...addr+count) doesn't includes any intersect with valid
2083 * vmalloc area, returns 0.
2084 *
2085 * This function checks that addr is a valid vmalloc'ed area, and
2086 * copy data from a buffer to the given addr. If specified range of
2087 * [addr...addr+count) includes some valid address, data is copied from
2088 * proper area of @buf. If there are memory holes, no copy to hole.
2089 * IOREMAP area is treated as memory hole and no copy is done.
2090 *
2091 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002092 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002093 *
2094 * Note: In usual ops, vwrite() is never necessary because the caller
2095 * should know vmalloc() area is valid and can use memcpy().
2096 * This is for routines which have to access vmalloc area without
2097 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002098 */
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100long vwrite(char *buf, char *addr, unsigned long count)
2101{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002102 struct vmap_area *va;
2103 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002104 char *vaddr;
2105 unsigned long n, buflen;
2106 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
2108 /* Don't allow overflow */
2109 if ((unsigned long) addr + count < count)
2110 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002111 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
Joonsoo Kime81ce852013-04-29 15:07:32 -07002113 spin_lock(&vmap_area_lock);
2114 list_for_each_entry(va, &vmap_area_list, list) {
2115 if (!count)
2116 break;
2117
2118 if (!(va->flags & VM_VM_AREA))
2119 continue;
2120
2121 vm = va->vm;
2122 vaddr = (char *) vm->addr;
2123 if (addr >= vaddr + vm->size - PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 continue;
2125 while (addr < vaddr) {
2126 if (count == 0)
2127 goto finished;
2128 buf++;
2129 addr++;
2130 count--;
2131 }
Joonsoo Kime81ce852013-04-29 15:07:32 -07002132 n = vaddr + vm->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002133 if (n > count)
2134 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002135 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002136 aligned_vwrite(buf, addr, n);
2137 copied++;
2138 }
2139 buf += n;
2140 addr += n;
2141 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 }
2143finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002144 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002145 if (!copied)
2146 return 0;
2147 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148}
Nick Piggin83342312006-06-23 02:03:20 -07002149
2150/**
2151 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002152 * @vma: vma to cover (map full range of vma)
2153 * @addr: vmalloc memory
2154 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002155 *
2156 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002157 *
2158 * This function checks that addr is a valid vmalloc'ed area, and
2159 * that it is big enough to cover the vma. Will return failure if
2160 * that criteria isn't met.
2161 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002162 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002163 */
2164int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2165 unsigned long pgoff)
2166{
2167 struct vm_struct *area;
2168 unsigned long uaddr = vma->vm_start;
2169 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07002170
2171 if ((PAGE_SIZE-1) & (unsigned long)addr)
2172 return -EINVAL;
2173
Nick Piggindb64fe02008-10-18 20:27:03 -07002174 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07002175 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07002176 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002177
2178 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07002179 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002180
2181 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07002182 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002183
2184 addr += pgoff << PAGE_SHIFT;
2185 do {
2186 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07002187 int ret;
2188
Nick Piggin83342312006-06-23 02:03:20 -07002189 ret = vm_insert_page(vma, uaddr, page);
2190 if (ret)
2191 return ret;
2192
2193 uaddr += PAGE_SIZE;
2194 addr += PAGE_SIZE;
2195 usize -= PAGE_SIZE;
2196 } while (usize > 0);
2197
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07002198 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
Nick Piggin83342312006-06-23 02:03:20 -07002199
Nick Piggindb64fe02008-10-18 20:27:03 -07002200 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07002201}
2202EXPORT_SYMBOL(remap_vmalloc_range);
2203
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002204/*
2205 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2206 * have one.
2207 */
2208void __attribute__((weak)) vmalloc_sync_all(void)
2209{
2210}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002211
2212
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002213static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002214{
David Vrabelcd129092011-09-29 16:53:32 +01002215 pte_t ***p = data;
2216
2217 if (p) {
2218 *(*p) = pte;
2219 (*p)++;
2220 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002221 return 0;
2222}
2223
2224/**
2225 * alloc_vm_area - allocate a range of kernel address space
2226 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002227 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002228 *
2229 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002230 *
2231 * This function reserves a range of kernel address space, and
2232 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002233 * are created.
2234 *
2235 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2236 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002237 */
David Vrabelcd129092011-09-29 16:53:32 +01002238struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002239{
2240 struct vm_struct *area;
2241
Christoph Lameter23016962008-04-28 02:12:42 -07002242 area = get_vm_area_caller(size, VM_IOREMAP,
2243 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002244 if (area == NULL)
2245 return NULL;
2246
2247 /*
2248 * This ensures that page tables are constructed for this region
2249 * of kernel virtual address space and mapped into init_mm.
2250 */
2251 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002252 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002253 free_vm_area(area);
2254 return NULL;
2255 }
2256
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002257 return area;
2258}
2259EXPORT_SYMBOL_GPL(alloc_vm_area);
2260
2261void free_vm_area(struct vm_struct *area)
2262{
2263 struct vm_struct *ret;
2264 ret = remove_vm_area(area->addr);
2265 BUG_ON(ret != area);
2266 kfree(area);
2267}
2268EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002269
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002270#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002271static struct vmap_area *node_to_va(struct rb_node *n)
2272{
2273 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2274}
2275
2276/**
2277 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2278 * @end: target address
2279 * @pnext: out arg for the next vmap_area
2280 * @pprev: out arg for the previous vmap_area
2281 *
2282 * Returns: %true if either or both of next and prev are found,
2283 * %false if no vmap_area exists
2284 *
2285 * Find vmap_areas end addresses of which enclose @end. ie. if not
2286 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2287 */
2288static bool pvm_find_next_prev(unsigned long end,
2289 struct vmap_area **pnext,
2290 struct vmap_area **pprev)
2291{
2292 struct rb_node *n = vmap_area_root.rb_node;
2293 struct vmap_area *va = NULL;
2294
2295 while (n) {
2296 va = rb_entry(n, struct vmap_area, rb_node);
2297 if (end < va->va_end)
2298 n = n->rb_left;
2299 else if (end > va->va_end)
2300 n = n->rb_right;
2301 else
2302 break;
2303 }
2304
2305 if (!va)
2306 return false;
2307
2308 if (va->va_end > end) {
2309 *pnext = va;
2310 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2311 } else {
2312 *pprev = va;
2313 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2314 }
2315 return true;
2316}
2317
2318/**
2319 * pvm_determine_end - find the highest aligned address between two vmap_areas
2320 * @pnext: in/out arg for the next vmap_area
2321 * @pprev: in/out arg for the previous vmap_area
2322 * @align: alignment
2323 *
2324 * Returns: determined end address
2325 *
2326 * Find the highest aligned address between *@pnext and *@pprev below
2327 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2328 * down address is between the end addresses of the two vmap_areas.
2329 *
2330 * Please note that the address returned by this function may fall
2331 * inside *@pnext vmap_area. The caller is responsible for checking
2332 * that.
2333 */
2334static unsigned long pvm_determine_end(struct vmap_area **pnext,
2335 struct vmap_area **pprev,
2336 unsigned long align)
2337{
2338 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2339 unsigned long addr;
2340
2341 if (*pnext)
2342 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2343 else
2344 addr = vmalloc_end;
2345
2346 while (*pprev && (*pprev)->va_end > addr) {
2347 *pnext = *pprev;
2348 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2349 }
2350
2351 return addr;
2352}
2353
2354/**
2355 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2356 * @offsets: array containing offset of each area
2357 * @sizes: array containing size of each area
2358 * @nr_vms: the number of areas to allocate
2359 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002360 *
2361 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2362 * vm_structs on success, %NULL on failure
2363 *
2364 * Percpu allocator wants to use congruent vm areas so that it can
2365 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002366 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2367 * be scattered pretty far, distance between two areas easily going up
2368 * to gigabytes. To avoid interacting with regular vmallocs, these
2369 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002370 *
2371 * Despite its complicated look, this allocator is rather simple. It
2372 * does everything top-down and scans areas from the end looking for
2373 * matching slot. While scanning, if any of the areas overlaps with
2374 * existing vmap_area, the base address is pulled down to fit the
2375 * area. Scanning is repeated till all the areas fit and then all
2376 * necessary data structres are inserted and the result is returned.
2377 */
2378struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2379 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002380 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002381{
2382 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2383 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2384 struct vmap_area **vas, *prev, *next;
2385 struct vm_struct **vms;
2386 int area, area2, last_area, term_area;
2387 unsigned long base, start, end, last_end;
2388 bool purged = false;
2389
Tejun Heoca23e402009-08-14 15:00:52 +09002390 /* verify parameters and allocate data structures */
2391 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2392 for (last_area = 0, area = 0; area < nr_vms; area++) {
2393 start = offsets[area];
2394 end = start + sizes[area];
2395
2396 /* is everything aligned properly? */
2397 BUG_ON(!IS_ALIGNED(offsets[area], align));
2398 BUG_ON(!IS_ALIGNED(sizes[area], align));
2399
2400 /* detect the area with the highest address */
2401 if (start > offsets[last_area])
2402 last_area = area;
2403
2404 for (area2 = 0; area2 < nr_vms; area2++) {
2405 unsigned long start2 = offsets[area2];
2406 unsigned long end2 = start2 + sizes[area2];
2407
2408 if (area2 == area)
2409 continue;
2410
2411 BUG_ON(start2 >= start && start2 < end);
2412 BUG_ON(end2 <= end && end2 > start);
2413 }
2414 }
2415 last_end = offsets[last_area] + sizes[last_area];
2416
2417 if (vmalloc_end - vmalloc_start < last_end) {
2418 WARN_ON(true);
2419 return NULL;
2420 }
2421
Thomas Meyer4d67d862012-05-29 15:06:21 -07002422 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2423 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002424 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002425 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002426
2427 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002428 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2429 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002430 if (!vas[area] || !vms[area])
2431 goto err_free;
2432 }
2433retry:
2434 spin_lock(&vmap_area_lock);
2435
2436 /* start scanning - we scan from the top, begin with the last area */
2437 area = term_area = last_area;
2438 start = offsets[area];
2439 end = start + sizes[area];
2440
2441 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2442 base = vmalloc_end - last_end;
2443 goto found;
2444 }
2445 base = pvm_determine_end(&next, &prev, align) - end;
2446
2447 while (true) {
2448 BUG_ON(next && next->va_end <= base + end);
2449 BUG_ON(prev && prev->va_end > base + end);
2450
2451 /*
2452 * base might have underflowed, add last_end before
2453 * comparing.
2454 */
2455 if (base + last_end < vmalloc_start + last_end) {
2456 spin_unlock(&vmap_area_lock);
2457 if (!purged) {
2458 purge_vmap_area_lazy();
2459 purged = true;
2460 goto retry;
2461 }
2462 goto err_free;
2463 }
2464
2465 /*
2466 * If next overlaps, move base downwards so that it's
2467 * right below next and then recheck.
2468 */
2469 if (next && next->va_start < base + end) {
2470 base = pvm_determine_end(&next, &prev, align) - end;
2471 term_area = area;
2472 continue;
2473 }
2474
2475 /*
2476 * If prev overlaps, shift down next and prev and move
2477 * base so that it's right below new next and then
2478 * recheck.
2479 */
2480 if (prev && prev->va_end > base + start) {
2481 next = prev;
2482 prev = node_to_va(rb_prev(&next->rb_node));
2483 base = pvm_determine_end(&next, &prev, align) - end;
2484 term_area = area;
2485 continue;
2486 }
2487
2488 /*
2489 * This area fits, move on to the previous one. If
2490 * the previous one is the terminal one, we're done.
2491 */
2492 area = (area + nr_vms - 1) % nr_vms;
2493 if (area == term_area)
2494 break;
2495 start = offsets[area];
2496 end = start + sizes[area];
2497 pvm_find_next_prev(base + end, &next, &prev);
2498 }
2499found:
2500 /* we've found a fitting base, insert all va's */
2501 for (area = 0; area < nr_vms; area++) {
2502 struct vmap_area *va = vas[area];
2503
2504 va->va_start = base + offsets[area];
2505 va->va_end = va->va_start + sizes[area];
2506 __insert_vmap_area(va);
2507 }
2508
2509 vmap_area_pcpu_hole = base + offsets[last_area];
2510
2511 spin_unlock(&vmap_area_lock);
2512
2513 /* insert all vm's */
2514 for (area = 0; area < nr_vms; area++)
2515 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2516 pcpu_get_vm_areas);
2517
2518 kfree(vas);
2519 return vms;
2520
2521err_free:
2522 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002523 kfree(vas[area]);
2524 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002525 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002526err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002527 kfree(vas);
2528 kfree(vms);
2529 return NULL;
2530}
2531
2532/**
2533 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2534 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2535 * @nr_vms: the number of allocated areas
2536 *
2537 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2538 */
2539void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2540{
2541 int i;
2542
2543 for (i = 0; i < nr_vms; i++)
2544 free_vm_area(vms[i]);
2545 kfree(vms);
2546}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002547#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002548
2549#ifdef CONFIG_PROC_FS
2550static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002551 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002552{
2553 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002554 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002555
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002556 spin_lock(&vmap_area_lock);
2557 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2558 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002559 n--;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002560 va = list_entry(va->list.next, typeof(*va), list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002561 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002562 if (!n && &va->list != &vmap_area_list)
2563 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002564
2565 return NULL;
2566
2567}
2568
2569static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2570{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002571 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002572
2573 ++*pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002574 next = list_entry(va->list.next, typeof(*va), list);
2575 if (&next->list != &vmap_area_list)
2576 return next;
2577
2578 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002579}
2580
2581static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002582 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002583{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002584 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002585}
2586
Eric Dumazeta47a1262008-07-23 21:27:38 -07002587static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2588{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002589 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002590 unsigned int nr, *counters = m->private;
2591
2592 if (!counters)
2593 return;
2594
Joonsoo Kim4341fa42013-04-29 15:07:39 -07002595 /* Pair with smp_wmb() in clear_vm_unlist() */
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002596 smp_rmb();
2597 if (v->flags & VM_UNLIST)
2598 return;
2599
Eric Dumazeta47a1262008-07-23 21:27:38 -07002600 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2601
2602 for (nr = 0; nr < v->nr_pages; nr++)
2603 counters[page_to_nid(v->pages[nr])]++;
2604
2605 for_each_node_state(nr, N_HIGH_MEMORY)
2606 if (counters[nr])
2607 seq_printf(m, " N%u=%u", nr, counters[nr]);
2608 }
2609}
2610
Christoph Lametera10aa572008-04-28 02:12:40 -07002611static int s_show(struct seq_file *m, void *p)
2612{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002613 struct vmap_area *va = p;
2614 struct vm_struct *v;
2615
2616 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2617 return 0;
2618
2619 if (!(va->flags & VM_VM_AREA)) {
2620 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
2621 (void *)va->va_start, (void *)va->va_end,
2622 va->va_end - va->va_start);
2623 return 0;
2624 }
2625
2626 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002627
Kees Cook45ec1692012-10-08 16:34:09 -07002628 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002629 v->addr, v->addr + v->size, v->size);
2630
Joe Perches62c70bc2011-01-13 15:45:52 -08002631 if (v->caller)
2632 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002633
Christoph Lametera10aa572008-04-28 02:12:40 -07002634 if (v->nr_pages)
2635 seq_printf(m, " pages=%d", v->nr_pages);
2636
2637 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002638 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002639
2640 if (v->flags & VM_IOREMAP)
2641 seq_printf(m, " ioremap");
2642
2643 if (v->flags & VM_ALLOC)
2644 seq_printf(m, " vmalloc");
2645
2646 if (v->flags & VM_MAP)
2647 seq_printf(m, " vmap");
2648
2649 if (v->flags & VM_USERMAP)
2650 seq_printf(m, " user");
2651
2652 if (v->flags & VM_VPAGES)
2653 seq_printf(m, " vpages");
2654
Eric Dumazeta47a1262008-07-23 21:27:38 -07002655 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002656 seq_putc(m, '\n');
2657 return 0;
2658}
2659
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002660static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002661 .start = s_start,
2662 .next = s_next,
2663 .stop = s_stop,
2664 .show = s_show,
2665};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002666
2667static int vmalloc_open(struct inode *inode, struct file *file)
2668{
2669 unsigned int *ptr = NULL;
2670 int ret;
2671
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002672 if (IS_ENABLED(CONFIG_NUMA)) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002673 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002674 if (ptr == NULL)
2675 return -ENOMEM;
2676 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002677 ret = seq_open(file, &vmalloc_op);
2678 if (!ret) {
2679 struct seq_file *m = file->private_data;
2680 m->private = ptr;
2681 } else
2682 kfree(ptr);
2683 return ret;
2684}
2685
2686static const struct file_operations proc_vmalloc_operations = {
2687 .open = vmalloc_open,
2688 .read = seq_read,
2689 .llseek = seq_lseek,
2690 .release = seq_release_private,
2691};
2692
2693static int __init proc_vmalloc_init(void)
2694{
2695 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2696 return 0;
2697}
2698module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002699
2700void get_vmalloc_info(struct vmalloc_info *vmi)
2701{
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002702 struct vmap_area *va;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002703 unsigned long free_area_size;
2704 unsigned long prev_end;
2705
2706 vmi->used = 0;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002707 vmi->largest_chunk = 0;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002708
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002709 prev_end = VMALLOC_START;
2710
2711 spin_lock(&vmap_area_lock);
2712
2713 if (list_empty(&vmap_area_list)) {
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002714 vmi->largest_chunk = VMALLOC_TOTAL;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002715 goto out;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002716 }
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002717
2718 list_for_each_entry(va, &vmap_area_list, list) {
2719 unsigned long addr = va->va_start;
2720
2721 /*
2722 * Some archs keep another range for modules in vmalloc space
2723 */
2724 if (addr < VMALLOC_START)
2725 continue;
2726 if (addr >= VMALLOC_END)
2727 break;
2728
2729 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2730 continue;
2731
2732 vmi->used += (va->va_end - va->va_start);
2733
2734 free_area_size = addr - prev_end;
2735 if (vmi->largest_chunk < free_area_size)
2736 vmi->largest_chunk = free_area_size;
2737
2738 prev_end = va->va_end;
2739 }
2740
2741 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2742 vmi->largest_chunk = VMALLOC_END - prev_end;
2743
2744out:
2745 spin_unlock(&vmap_area_lock);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002746}
Christoph Lametera10aa572008-04-28 02:12:40 -07002747#endif
2748