blob: 0fdf96803c5b59623792a24e57015fb0e25098bb [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;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700295 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700296 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
Catalin Marinas7f88f882013-11-12 15:07:45 -0800362 /*
363 * Only scan the relevant parts containing pointers to other objects
364 * to avoid false negatives.
365 */
366 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
367
Nick Piggindb64fe02008-10-18 20:27:03 -0700368retry:
369 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700370 /*
371 * Invalidate cache if we have more permissive parameters.
372 * cached_hole_size notes the largest hole noticed _below_
373 * the vmap_area cached in free_vmap_cache: if size fits
374 * into that hole, we want to scan from vstart to reuse
375 * the hole instead of allocating above free_vmap_cache.
376 * Note that __free_vmap_area may update free_vmap_cache
377 * without updating cached_hole_size or cached_align.
378 */
379 if (!free_vmap_cache ||
380 size < cached_hole_size ||
381 vstart < cached_vstart ||
382 align < cached_align) {
383nocache:
384 cached_hole_size = 0;
385 free_vmap_cache = NULL;
386 }
387 /* record if we encounter less permissive parameters */
388 cached_vstart = vstart;
389 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800390
Nick Piggin89699602011-03-22 16:30:36 -0700391 /* find starting point for our search */
392 if (free_vmap_cache) {
393 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700394 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700395 if (addr < vstart)
396 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700397 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700398 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700399
Nick Piggin89699602011-03-22 16:30:36 -0700400 } else {
401 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700402 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700403 goto overflow;
404
405 n = vmap_area_root.rb_node;
406 first = NULL;
407
408 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700409 struct vmap_area *tmp;
410 tmp = rb_entry(n, struct vmap_area, rb_node);
411 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700412 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700413 if (tmp->va_start <= addr)
414 break;
415 n = n->rb_left;
416 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700417 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700418 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700419
420 if (!first)
421 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700422 }
Nick Piggin89699602011-03-22 16:30:36 -0700423
424 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700425 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700426 if (addr + cached_hole_size < first->va_start)
427 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700428 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700429 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700430 goto overflow;
431
Hong zhi guo92ca9222012-07-31 16:41:35 -0700432 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700433 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700434
435 first = list_entry(first->list.next,
436 struct vmap_area, list);
Nick Piggin89699602011-03-22 16:30:36 -0700437 }
438
Nick Piggindb64fe02008-10-18 20:27:03 -0700439found:
Nick Piggin89699602011-03-22 16:30:36 -0700440 if (addr + size > vend)
441 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700442
443 va->va_start = addr;
444 va->va_end = addr + size;
445 va->flags = 0;
446 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700447 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700448 spin_unlock(&vmap_area_lock);
449
Nick Piggin89699602011-03-22 16:30:36 -0700450 BUG_ON(va->va_start & (align-1));
451 BUG_ON(va->va_start < vstart);
452 BUG_ON(va->va_end > vend);
453
Nick Piggindb64fe02008-10-18 20:27:03 -0700454 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700455
456overflow:
457 spin_unlock(&vmap_area_lock);
458 if (!purged) {
459 purge_vmap_area_lazy();
460 purged = 1;
461 goto retry;
462 }
463 if (printk_ratelimit())
464 printk(KERN_WARNING
465 "vmap allocation for size %lu failed: "
466 "use vmalloc=<size> to increase size.\n", size);
467 kfree(va);
468 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700469}
470
Nick Piggindb64fe02008-10-18 20:27:03 -0700471static void __free_vmap_area(struct vmap_area *va)
472{
473 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700474
475 if (free_vmap_cache) {
476 if (va->va_end < cached_vstart) {
477 free_vmap_cache = NULL;
478 } else {
479 struct vmap_area *cache;
480 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
481 if (va->va_start <= cache->va_start) {
482 free_vmap_cache = rb_prev(&va->rb_node);
483 /*
484 * We don't try to update cached_hole_size or
485 * cached_align, but it won't go very wrong.
486 */
487 }
488 }
489 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700490 rb_erase(&va->rb_node, &vmap_area_root);
491 RB_CLEAR_NODE(&va->rb_node);
492 list_del_rcu(&va->list);
493
Tejun Heoca23e402009-08-14 15:00:52 +0900494 /*
495 * Track the highest possible candidate for pcpu area
496 * allocation. Areas outside of vmalloc area can be returned
497 * here too, consider only end addresses which fall inside
498 * vmalloc area proper.
499 */
500 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
501 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
502
Lai Jiangshan14769de2011-03-18 12:12:19 +0800503 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700504}
505
506/*
507 * Free a region of KVA allocated by alloc_vmap_area
508 */
509static void free_vmap_area(struct vmap_area *va)
510{
511 spin_lock(&vmap_area_lock);
512 __free_vmap_area(va);
513 spin_unlock(&vmap_area_lock);
514}
515
516/*
517 * Clear the pagetable entries of a given vmap_area
518 */
519static void unmap_vmap_area(struct vmap_area *va)
520{
521 vunmap_page_range(va->va_start, va->va_end);
522}
523
Nick Piggincd528582009-01-06 14:39:20 -0800524static void vmap_debug_free_range(unsigned long start, unsigned long end)
525{
526 /*
527 * Unmap page tables and force a TLB flush immediately if
528 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
529 * bugs similarly to those in linear kernel virtual address
530 * space after a page has been freed.
531 *
532 * All the lazy freeing logic is still retained, in order to
533 * minimise intrusiveness of this debugging feature.
534 *
535 * This is going to be *slow* (linear kernel virtual address
536 * debugging doesn't do a broadcast TLB flush so it is a lot
537 * faster).
538 */
539#ifdef CONFIG_DEBUG_PAGEALLOC
540 vunmap_page_range(start, end);
541 flush_tlb_kernel_range(start, end);
542#endif
543}
544
Nick Piggindb64fe02008-10-18 20:27:03 -0700545/*
546 * lazy_max_pages is the maximum amount of virtual address space we gather up
547 * before attempting to purge with a TLB flush.
548 *
549 * There is a tradeoff here: a larger number will cover more kernel page tables
550 * and take slightly longer to purge, but it will linearly reduce the number of
551 * global TLB flushes that must be performed. It would seem natural to scale
552 * this number up linearly with the number of CPUs (because vmapping activity
553 * could also scale linearly with the number of CPUs), however it is likely
554 * that in practice, workloads might be constrained in other ways that mean
555 * vmap activity will not scale linearly with CPUs. Also, I want to be
556 * conservative and not introduce a big latency on huge systems, so go with
557 * a less aggressive log scale. It will still be an improvement over the old
558 * code, and it will be simple to change the scale factor if we find that it
559 * becomes a problem on bigger systems.
560 */
561static unsigned long lazy_max_pages(void)
562{
563 unsigned int log;
564
565 log = fls(num_online_cpus());
566
567 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
568}
569
570static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
571
Nick Piggin02b709d2010-02-01 22:25:57 +1100572/* for per-CPU blocks */
573static void purge_fragmented_blocks_allcpus(void);
574
Nick Piggindb64fe02008-10-18 20:27:03 -0700575/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500576 * called before a call to iounmap() if the caller wants vm_area_struct's
577 * immediately freed.
578 */
579void set_iounmap_nonlazy(void)
580{
581 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
582}
583
584/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700585 * Purges all lazily-freed vmap areas.
586 *
587 * If sync is 0 then don't purge if there is already a purge in progress.
588 * If force_flush is 1, then flush kernel TLBs between *start and *end even
589 * if we found no lazy vmap areas to unmap (callers can use this to optimise
590 * their own TLB flushing).
591 * Returns with *start = min(*start, lowest purged address)
592 * *end = max(*end, highest purged address)
593 */
594static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
595 int sync, int force_flush)
596{
Andrew Morton46666d82009-01-15 13:51:15 -0800597 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700598 LIST_HEAD(valist);
599 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800600 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700601 int nr = 0;
602
603 /*
604 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
605 * should not expect such behaviour. This just simplifies locking for
606 * the case that isn't actually used at the moment anyway.
607 */
608 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800609 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700610 return;
611 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800612 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700613
Nick Piggin02b709d2010-02-01 22:25:57 +1100614 if (sync)
615 purge_fragmented_blocks_allcpus();
616
Nick Piggindb64fe02008-10-18 20:27:03 -0700617 rcu_read_lock();
618 list_for_each_entry_rcu(va, &vmap_area_list, list) {
619 if (va->flags & VM_LAZY_FREE) {
620 if (va->va_start < *start)
621 *start = va->va_start;
622 if (va->va_end > *end)
623 *end = va->va_end;
624 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700625 list_add_tail(&va->purge_list, &valist);
626 va->flags |= VM_LAZY_FREEING;
627 va->flags &= ~VM_LAZY_FREE;
628 }
629 }
630 rcu_read_unlock();
631
Yongseok Koh88f50042010-01-19 17:33:49 +0900632 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700633 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700634
635 if (nr || force_flush)
636 flush_tlb_kernel_range(*start, *end);
637
638 if (nr) {
639 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800640 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700641 __free_vmap_area(va);
642 spin_unlock(&vmap_area_lock);
643 }
Andrew Morton46666d82009-01-15 13:51:15 -0800644 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700645}
646
647/*
Nick Piggin496850e2008-11-19 15:36:33 -0800648 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
649 * is already purging.
650 */
651static void try_purge_vmap_area_lazy(void)
652{
653 unsigned long start = ULONG_MAX, end = 0;
654
655 __purge_vmap_area_lazy(&start, &end, 0, 0);
656}
657
658/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700659 * Kick off a purge of the outstanding lazy areas.
660 */
661static void purge_vmap_area_lazy(void)
662{
663 unsigned long start = ULONG_MAX, end = 0;
664
Nick Piggin496850e2008-11-19 15:36:33 -0800665 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700666}
667
668/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800669 * Free a vmap area, caller ensuring that the area has been unmapped
670 * and flush_cache_vunmap had been called for the correct range
671 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700672 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800673static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700674{
675 va->flags |= VM_LAZY_FREE;
676 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
677 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800678 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700679}
680
Nick Pigginb29acbd2008-12-01 13:13:47 -0800681/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800682 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
683 * called for the correct range previously.
684 */
685static void free_unmap_vmap_area_noflush(struct vmap_area *va)
686{
687 unmap_vmap_area(va);
688 free_vmap_area_noflush(va);
689}
690
691/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800692 * Free and unmap a vmap area
693 */
694static void free_unmap_vmap_area(struct vmap_area *va)
695{
696 flush_cache_vunmap(va->va_start, va->va_end);
697 free_unmap_vmap_area_noflush(va);
698}
699
Nick Piggindb64fe02008-10-18 20:27:03 -0700700static struct vmap_area *find_vmap_area(unsigned long addr)
701{
702 struct vmap_area *va;
703
704 spin_lock(&vmap_area_lock);
705 va = __find_vmap_area(addr);
706 spin_unlock(&vmap_area_lock);
707
708 return va;
709}
710
711static void free_unmap_vmap_area_addr(unsigned long addr)
712{
713 struct vmap_area *va;
714
715 va = find_vmap_area(addr);
716 BUG_ON(!va);
717 free_unmap_vmap_area(va);
718}
719
720
721/*** Per cpu kva allocator ***/
722
723/*
724 * vmap space is limited especially on 32 bit architectures. Ensure there is
725 * room for at least 16 percpu vmap blocks per CPU.
726 */
727/*
728 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
729 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
730 * instead (we just need a rough idea)
731 */
732#if BITS_PER_LONG == 32
733#define VMALLOC_SPACE (128UL*1024*1024)
734#else
735#define VMALLOC_SPACE (128UL*1024*1024*1024)
736#endif
737
738#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
739#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
740#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
741#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
742#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
743#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200744#define VMAP_BBMAP_BITS \
745 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
746 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
747 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700748
749#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
750
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100751static bool vmap_initialized __read_mostly = false;
752
Nick Piggindb64fe02008-10-18 20:27:03 -0700753struct vmap_block_queue {
754 spinlock_t lock;
755 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700756};
757
758struct vmap_block {
759 spinlock_t lock;
760 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700761 unsigned long free, dirty;
Nick Piggindb64fe02008-10-18 20:27:03 -0700762 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100763 struct list_head free_list;
764 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100765 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700766};
767
768/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
769static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
770
771/*
772 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
773 * in the free path. Could get rid of this if we change the API to return a
774 * "cookie" from alloc, to be passed to free. But no big deal yet.
775 */
776static DEFINE_SPINLOCK(vmap_block_tree_lock);
777static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
778
779/*
780 * We should probably have a fallback mechanism to allocate virtual memory
781 * out of partially filled vmap blocks. However vmap block sizing should be
782 * fairly reasonable according to the vmalloc size, so it shouldn't be a
783 * big problem.
784 */
785
786static unsigned long addr_to_vb_idx(unsigned long addr)
787{
788 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
789 addr /= VMAP_BLOCK_SIZE;
790 return addr;
791}
792
793static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
794{
795 struct vmap_block_queue *vbq;
796 struct vmap_block *vb;
797 struct vmap_area *va;
798 unsigned long vb_idx;
799 int node, err;
800
801 node = numa_node_id();
802
803 vb = kmalloc_node(sizeof(struct vmap_block),
804 gfp_mask & GFP_RECLAIM_MASK, node);
805 if (unlikely(!vb))
806 return ERR_PTR(-ENOMEM);
807
808 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
809 VMALLOC_START, VMALLOC_END,
810 node, gfp_mask);
Tobias Klauserddf9c6d42011-01-13 15:46:15 -0800811 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700812 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700813 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700814 }
815
816 err = radix_tree_preload(gfp_mask);
817 if (unlikely(err)) {
818 kfree(vb);
819 free_vmap_area(va);
820 return ERR_PTR(err);
821 }
822
823 spin_lock_init(&vb->lock);
824 vb->va = va;
825 vb->free = VMAP_BBMAP_BITS;
826 vb->dirty = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700827 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
828 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700829
830 vb_idx = addr_to_vb_idx(va->va_start);
831 spin_lock(&vmap_block_tree_lock);
832 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
833 spin_unlock(&vmap_block_tree_lock);
834 BUG_ON(err);
835 radix_tree_preload_end();
836
837 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700838 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100839 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700840 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900841 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700842
843 return vb;
844}
845
Nick Piggindb64fe02008-10-18 20:27:03 -0700846static void free_vmap_block(struct vmap_block *vb)
847{
848 struct vmap_block *tmp;
849 unsigned long vb_idx;
850
Nick Piggindb64fe02008-10-18 20:27:03 -0700851 vb_idx = addr_to_vb_idx(vb->va->va_start);
852 spin_lock(&vmap_block_tree_lock);
853 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
854 spin_unlock(&vmap_block_tree_lock);
855 BUG_ON(tmp != vb);
856
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800857 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800858 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700859}
860
Nick Piggin02b709d2010-02-01 22:25:57 +1100861static void purge_fragmented_blocks(int cpu)
862{
863 LIST_HEAD(purge);
864 struct vmap_block *vb;
865 struct vmap_block *n_vb;
866 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
867
868 rcu_read_lock();
869 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
870
871 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
872 continue;
873
874 spin_lock(&vb->lock);
875 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
876 vb->free = 0; /* prevent further allocs after releasing lock */
877 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Nick Piggin02b709d2010-02-01 22:25:57 +1100878 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
879 spin_lock(&vbq->lock);
880 list_del_rcu(&vb->free_list);
881 spin_unlock(&vbq->lock);
882 spin_unlock(&vb->lock);
883 list_add_tail(&vb->purge, &purge);
884 } else
885 spin_unlock(&vb->lock);
886 }
887 rcu_read_unlock();
888
889 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
890 list_del(&vb->purge);
891 free_vmap_block(vb);
892 }
893}
894
Nick Piggin02b709d2010-02-01 22:25:57 +1100895static void purge_fragmented_blocks_allcpus(void)
896{
897 int cpu;
898
899 for_each_possible_cpu(cpu)
900 purge_fragmented_blocks(cpu);
901}
902
Nick Piggindb64fe02008-10-18 20:27:03 -0700903static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
904{
905 struct vmap_block_queue *vbq;
906 struct vmap_block *vb;
907 unsigned long addr = 0;
908 unsigned int order;
909
910 BUG_ON(size & ~PAGE_MASK);
911 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700912 if (WARN_ON(size == 0)) {
913 /*
914 * Allocating 0 bytes isn't what caller wants since
915 * get_order(0) returns funny result. Just warn and terminate
916 * early.
917 */
918 return NULL;
919 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700920 order = get_order(size);
921
922again:
923 rcu_read_lock();
924 vbq = &get_cpu_var(vmap_block_queue);
925 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
926 int i;
927
928 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100929 if (vb->free < 1UL << order)
930 goto next;
931
Zhang Yanfei3fcd76e2013-07-08 15:59:54 -0700932 i = VMAP_BBMAP_BITS - vb->free;
Nick Piggin02b709d2010-02-01 22:25:57 +1100933 addr = vb->va->va_start + (i << PAGE_SHIFT);
934 BUG_ON(addr_to_vb_idx(addr) !=
935 addr_to_vb_idx(vb->va->va_start));
936 vb->free -= 1UL << order;
937 if (vb->free == 0) {
938 spin_lock(&vbq->lock);
939 list_del_rcu(&vb->free_list);
940 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700941 }
942 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100943 break;
944next:
945 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700946 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100947
Tejun Heo3f04ba82009-10-29 22:34:12 +0900948 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700949 rcu_read_unlock();
950
951 if (!addr) {
952 vb = new_vmap_block(gfp_mask);
953 if (IS_ERR(vb))
954 return vb;
955 goto again;
956 }
957
958 return (void *)addr;
959}
960
961static void vb_free(const void *addr, unsigned long size)
962{
963 unsigned long offset;
964 unsigned long vb_idx;
965 unsigned int order;
966 struct vmap_block *vb;
967
968 BUG_ON(size & ~PAGE_MASK);
969 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800970
971 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
972
Nick Piggindb64fe02008-10-18 20:27:03 -0700973 order = get_order(size);
974
975 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
976
977 vb_idx = addr_to_vb_idx((unsigned long)addr);
978 rcu_read_lock();
979 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
980 rcu_read_unlock();
981 BUG_ON(!vb);
982
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800983 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
984
Nick Piggindb64fe02008-10-18 20:27:03 -0700985 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100986 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700987
Nick Piggindb64fe02008-10-18 20:27:03 -0700988 vb->dirty += 1UL << order;
989 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100990 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700991 spin_unlock(&vb->lock);
992 free_vmap_block(vb);
993 } else
994 spin_unlock(&vb->lock);
995}
996
997/**
998 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
999 *
1000 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1001 * to amortize TLB flushing overheads. What this means is that any page you
1002 * have now, may, in a former life, have been mapped into kernel virtual
1003 * address by the vmap layer and so there might be some CPUs with TLB entries
1004 * still referencing that page (additional to the regular 1:1 kernel mapping).
1005 *
1006 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1007 * be sure that none of the pages we have control over will have any aliases
1008 * from the vmap layer.
1009 */
1010void vm_unmap_aliases(void)
1011{
1012 unsigned long start = ULONG_MAX, end = 0;
1013 int cpu;
1014 int flush = 0;
1015
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001016 if (unlikely(!vmap_initialized))
1017 return;
1018
Nick Piggindb64fe02008-10-18 20:27:03 -07001019 for_each_possible_cpu(cpu) {
1020 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1021 struct vmap_block *vb;
1022
1023 rcu_read_lock();
1024 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001025 int i, j;
Nick Piggindb64fe02008-10-18 20:27:03 -07001026
1027 spin_lock(&vb->lock);
1028 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001029 if (i < VMAP_BBMAP_BITS) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001030 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001031
1032 j = find_last_bit(vb->dirty_map,
1033 VMAP_BBMAP_BITS);
1034 j = j + 1; /* need exclusive index */
Nick Piggindb64fe02008-10-18 20:27:03 -07001035
1036 s = vb->va->va_start + (i << PAGE_SHIFT);
1037 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001038 flush = 1;
1039
1040 if (s < start)
1041 start = s;
1042 if (e > end)
1043 end = e;
Nick Piggindb64fe02008-10-18 20:27:03 -07001044 }
1045 spin_unlock(&vb->lock);
1046 }
1047 rcu_read_unlock();
1048 }
1049
1050 __purge_vmap_area_lazy(&start, &end, 1, flush);
1051}
1052EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1053
1054/**
1055 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1056 * @mem: the pointer returned by vm_map_ram
1057 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1058 */
1059void vm_unmap_ram(const void *mem, unsigned int count)
1060{
1061 unsigned long size = count << PAGE_SHIFT;
1062 unsigned long addr = (unsigned long)mem;
1063
1064 BUG_ON(!addr);
1065 BUG_ON(addr < VMALLOC_START);
1066 BUG_ON(addr > VMALLOC_END);
1067 BUG_ON(addr & (PAGE_SIZE-1));
1068
1069 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001070 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001071
1072 if (likely(count <= VMAP_MAX_ALLOC))
1073 vb_free(mem, size);
1074 else
1075 free_unmap_vmap_area_addr(addr);
1076}
1077EXPORT_SYMBOL(vm_unmap_ram);
1078
1079/**
1080 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1081 * @pages: an array of pointers to the pages to be mapped
1082 * @count: number of pages
1083 * @node: prefer to allocate data structures on this node
1084 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001085 *
1086 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001087 */
1088void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1089{
1090 unsigned long size = count << PAGE_SHIFT;
1091 unsigned long addr;
1092 void *mem;
1093
1094 if (likely(count <= VMAP_MAX_ALLOC)) {
1095 mem = vb_alloc(size, GFP_KERNEL);
1096 if (IS_ERR(mem))
1097 return NULL;
1098 addr = (unsigned long)mem;
1099 } else {
1100 struct vmap_area *va;
1101 va = alloc_vmap_area(size, PAGE_SIZE,
1102 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1103 if (IS_ERR(va))
1104 return NULL;
1105
1106 addr = va->va_start;
1107 mem = (void *)addr;
1108 }
1109 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1110 vm_unmap_ram(mem, count);
1111 return NULL;
1112 }
1113 return mem;
1114}
1115EXPORT_SYMBOL(vm_map_ram);
1116
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001117static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001118/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001119 * vm_area_add_early - add vmap area early during boot
1120 * @vm: vm_struct to add
1121 *
1122 * This function is used to add fixed kernel vm area to vmlist before
1123 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1124 * should contain proper values and the other fields should be zero.
1125 *
1126 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1127 */
1128void __init vm_area_add_early(struct vm_struct *vm)
1129{
1130 struct vm_struct *tmp, **p;
1131
1132 BUG_ON(vmap_initialized);
1133 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1134 if (tmp->addr >= vm->addr) {
1135 BUG_ON(tmp->addr < vm->addr + vm->size);
1136 break;
1137 } else
1138 BUG_ON(tmp->addr + tmp->size > vm->addr);
1139 }
1140 vm->next = *p;
1141 *p = vm;
1142}
1143
1144/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001145 * vm_area_register_early - register vmap area early during boot
1146 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001147 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001148 *
1149 * This function is used to register kernel vm area before
1150 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1151 * proper values on entry and other fields should be zero. On return,
1152 * vm->addr contains the allocated address.
1153 *
1154 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1155 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001156void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001157{
1158 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001159 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001160
Tejun Heoc0c0a292009-02-24 11:57:21 +09001161 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1162 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1163
1164 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001165
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001166 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001167}
1168
Nick Piggindb64fe02008-10-18 20:27:03 -07001169void __init vmalloc_init(void)
1170{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001171 struct vmap_area *va;
1172 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001173 int i;
1174
1175 for_each_possible_cpu(i) {
1176 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001177 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001178
1179 vbq = &per_cpu(vmap_block_queue, i);
1180 spin_lock_init(&vbq->lock);
1181 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001182 p = &per_cpu(vfree_deferred, i);
1183 init_llist_head(&p->list);
1184 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001185 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001186
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001187 /* Import existing vmlist entries. */
1188 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001189 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001190 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001191 va->va_start = (unsigned long)tmp->addr;
1192 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001193 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001194 __insert_vmap_area(va);
1195 }
Tejun Heoca23e402009-08-14 15:00:52 +09001196
1197 vmap_area_pcpu_hole = VMALLOC_END;
1198
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001199 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001200}
1201
Tejun Heo8fc48982009-02-20 16:29:08 +09001202/**
1203 * map_kernel_range_noflush - map kernel VM area with the specified pages
1204 * @addr: start of the VM area to map
1205 * @size: size of the VM area to map
1206 * @prot: page protection flags to use
1207 * @pages: pages to map
1208 *
1209 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1210 * specify should have been allocated using get_vm_area() and its
1211 * friends.
1212 *
1213 * NOTE:
1214 * This function does NOT do any cache flushing. The caller is
1215 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1216 * before calling this function.
1217 *
1218 * RETURNS:
1219 * The number of pages mapped on success, -errno on failure.
1220 */
1221int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1222 pgprot_t prot, struct page **pages)
1223{
1224 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1225}
1226
1227/**
1228 * unmap_kernel_range_noflush - unmap kernel VM area
1229 * @addr: start of the VM area to unmap
1230 * @size: size of the VM area to unmap
1231 *
1232 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1233 * specify should have been allocated using get_vm_area() and its
1234 * friends.
1235 *
1236 * NOTE:
1237 * This function does NOT do any cache flushing. The caller is
1238 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1239 * before calling this function and flush_tlb_kernel_range() after.
1240 */
1241void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1242{
1243 vunmap_page_range(addr, addr + size);
1244}
Huang Ying81e88fd2011-01-12 14:44:55 +08001245EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001246
1247/**
1248 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1249 * @addr: start of the VM area to unmap
1250 * @size: size of the VM area to unmap
1251 *
1252 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1253 * the unmapping and tlb after.
1254 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001255void unmap_kernel_range(unsigned long addr, unsigned long size)
1256{
1257 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001258
1259 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001260 vunmap_page_range(addr, end);
1261 flush_tlb_kernel_range(addr, end);
1262}
1263
1264int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1265{
1266 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001267 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001268 int err;
1269
1270 err = vmap_page_range(addr, end, prot, *pages);
1271 if (err > 0) {
1272 *pages += err;
1273 err = 0;
1274 }
1275
1276 return err;
1277}
1278EXPORT_SYMBOL_GPL(map_vm_area);
1279
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001280static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001281 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001282{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001283 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001284 vm->flags = flags;
1285 vm->addr = (void *)va->va_start;
1286 vm->size = va->va_end - va->va_start;
1287 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001288 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001289 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001290 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001291}
Tejun Heocf88c792009-08-14 15:00:52 +09001292
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001293static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001294{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001295 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001296 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001297 * we should make sure that vm has proper values.
1298 * Pair with smp_rmb() in show_numa_info().
1299 */
1300 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001301 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001302}
1303
Nick Piggindb64fe02008-10-18 20:27:03 -07001304static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001305 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001306 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001307{
Kautuk Consul00065262011-12-19 17:12:04 -08001308 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001309 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001311 BUG_ON(in_interrupt());
Zhang Yanfei0f2d4a82013-07-03 15:04:50 -07001312 if (flags & VM_IOREMAP)
1313 align = 1ul << clamp(fls(size), PAGE_SHIFT, IOREMAP_MAX_ORDER);
Nick Piggindb64fe02008-10-18 20:27:03 -07001314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001316 if (unlikely(!size))
1317 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Tejun Heocf88c792009-08-14 15:00:52 +09001319 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 if (unlikely(!area))
1321 return NULL;
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 /*
1324 * We always allocate a guard page.
1325 */
1326 size += PAGE_SIZE;
1327
Nick Piggindb64fe02008-10-18 20:27:03 -07001328 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1329 if (IS_ERR(va)) {
1330 kfree(area);
1331 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001334 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337}
1338
Christoph Lameter930fc452005-10-29 18:15:41 -07001339struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1340 unsigned long start, unsigned long end)
1341{
David Rientjes00ef2d22013-02-22 16:35:36 -08001342 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1343 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001344}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001345EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001346
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001347struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1348 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001349 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001350{
David Rientjes00ef2d22013-02-22 16:35:36 -08001351 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1352 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001353}
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355/**
Simon Arlott183ff222007-10-20 01:27:18 +02001356 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 * @size: size of the area
1358 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1359 *
1360 * Search an area of @size in the kernel virtual mapping area,
1361 * and reserved it for out purposes. Returns the area descriptor
1362 * on success or %NULL on failure.
1363 */
1364struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1365{
David Miller2dca6992009-09-21 12:22:34 -07001366 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001367 NUMA_NO_NODE, GFP_KERNEL,
1368 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001369}
1370
1371struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001372 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001373{
David Miller2dca6992009-09-21 12:22:34 -07001374 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001375 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001378/**
1379 * find_vm_area - find a continuous kernel virtual area
1380 * @addr: base address
1381 *
1382 * Search for the kernel VM area starting at @addr, and return it.
1383 * It is up to the caller to do all required locking to keep the returned
1384 * pointer valid.
1385 */
1386struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001387{
Nick Piggindb64fe02008-10-18 20:27:03 -07001388 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001389
Nick Piggindb64fe02008-10-18 20:27:03 -07001390 va = find_vmap_area((unsigned long)addr);
1391 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001392 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001393
Andi Kleen7856dfe2005-05-20 14:27:57 -07001394 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001395}
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397/**
Simon Arlott183ff222007-10-20 01:27:18 +02001398 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 * @addr: base address
1400 *
1401 * Search for the kernel VM area starting at @addr, and remove it.
1402 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001403 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001405struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
Nick Piggindb64fe02008-10-18 20:27:03 -07001407 struct vmap_area *va;
1408
1409 va = find_vmap_area((unsigned long)addr);
1410 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001411 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001412
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001413 spin_lock(&vmap_area_lock);
1414 va->vm = NULL;
1415 va->flags &= ~VM_VM_AREA;
1416 spin_unlock(&vmap_area_lock);
1417
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001418 vmap_debug_free_range(va->va_start, va->va_end);
1419 free_unmap_vmap_area(va);
1420 vm->size -= PAGE_SIZE;
1421
Nick Piggindb64fe02008-10-18 20:27:03 -07001422 return vm;
1423 }
1424 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425}
1426
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001427static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
1429 struct vm_struct *area;
1430
1431 if (!addr)
1432 return;
1433
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001434 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001435 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438 area = remove_vm_area(addr);
1439 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001440 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 return;
1443 }
1444
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001445 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001446 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 if (deallocate_pages) {
1449 int i;
1450
1451 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001452 struct page *page = area->pages[i];
1453
1454 BUG_ON(!page);
1455 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
1457
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001458 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 vfree(area->pages);
1460 else
1461 kfree(area->pages);
1462 }
1463
1464 kfree(area);
1465 return;
1466}
Al Viro32fcfd42013-03-10 20:14:08 -04001467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468/**
1469 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 * @addr: memory base address
1471 *
Simon Arlott183ff222007-10-20 01:27:18 +02001472 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001473 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1474 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 *
Al Viro32fcfd42013-03-10 20:14:08 -04001476 * Must not be called in NMI context (strictly speaking, only if we don't
1477 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1478 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001479 *
1480 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001482void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
Al Viro32fcfd42013-03-10 20:14:08 -04001484 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001485
1486 kmemleak_free(addr);
1487
Al Viro32fcfd42013-03-10 20:14:08 -04001488 if (!addr)
1489 return;
1490 if (unlikely(in_interrupt())) {
1491 struct vfree_deferred *p = &__get_cpu_var(vfree_deferred);
Oleg Nesterov59d31322013-07-08 16:00:08 -07001492 if (llist_add((struct llist_node *)addr, &p->list))
1493 schedule_work(&p->wq);
Al Viro32fcfd42013-03-10 20:14:08 -04001494 } else
1495 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497EXPORT_SYMBOL(vfree);
1498
1499/**
1500 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 * @addr: memory base address
1502 *
1503 * Free the virtually contiguous memory area starting at @addr,
1504 * which was created from the page array passed to vmap().
1505 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001506 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001508void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
1510 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001511 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001512 if (addr)
1513 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515EXPORT_SYMBOL(vunmap);
1516
1517/**
1518 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 * @pages: array of page pointers
1520 * @count: number of pages to map
1521 * @flags: vm_area->flags
1522 * @prot: page protection for the mapping
1523 *
1524 * Maps @count pages from @pages into contiguous kernel virtual
1525 * space.
1526 */
1527void *vmap(struct page **pages, unsigned int count,
1528 unsigned long flags, pgprot_t prot)
1529{
1530 struct vm_struct *area;
1531
Peter Zijlstra34754b62009-02-25 16:04:03 +01001532 might_sleep();
1533
Jan Beulich44813742009-09-21 17:03:05 -07001534 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 return NULL;
1536
Christoph Lameter23016962008-04-28 02:12:42 -07001537 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1538 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (!area)
1540 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 if (map_vm_area(area, prot, &pages)) {
1543 vunmap(area->addr);
1544 return NULL;
1545 }
1546
1547 return area->addr;
1548}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549EXPORT_SYMBOL(vmap);
1550
David Miller2dca6992009-09-21 12:22:34 -07001551static void *__vmalloc_node(unsigned long size, unsigned long align,
1552 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001553 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001554static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001555 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556{
Dave Hansen22943ab2011-05-24 17:12:18 -07001557 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 struct page **pages;
1559 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001560 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
Wanpeng Li762216a2013-09-11 14:22:42 -07001562 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 array_size = (nr_pages * sizeof(struct page *));
1564
1565 area->nr_pages = nr_pages;
1566 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001567 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001568 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001569 PAGE_KERNEL, node, area->caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001570 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001571 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001572 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 area->pages = pages;
1575 if (!area->pages) {
1576 remove_vm_area(area->addr);
1577 kfree(area);
1578 return NULL;
1579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
1581 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001582 struct page *page;
Dave Hansen22943ab2011-05-24 17:12:18 -07001583 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001584
Jianguo Wu4b909512013-11-12 15:07:11 -08001585 if (node == NUMA_NO_NODE)
Dave Hansen22943ab2011-05-24 17:12:18 -07001586 page = alloc_page(tmp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001587 else
Dave Hansen22943ab2011-05-24 17:12:18 -07001588 page = alloc_pages_node(node, tmp_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001589
1590 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 /* Successfully allocated i pages, free them in __vunmap() */
1592 area->nr_pages = i;
1593 goto fail;
1594 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001595 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
1597
1598 if (map_vm_area(area, prot, &pages))
1599 goto fail;
1600 return area->addr;
1601
1602fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001603 warn_alloc_failed(gfp_mask, order,
1604 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001605 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 vfree(area->addr);
1607 return NULL;
1608}
1609
David Rientjesd0a21262011-01-13 15:46:02 -08001610/**
1611 * __vmalloc_node_range - allocate virtually contiguous memory
1612 * @size: allocation size
1613 * @align: desired alignment
1614 * @start: vm area range start
1615 * @end: vm area range end
1616 * @gfp_mask: flags for the page level allocator
1617 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001618 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001619 * @caller: caller's return address
1620 *
1621 * Allocate enough pages to cover @size from the page level
1622 * allocator with @gfp_mask flags. Map them into contiguous
1623 * kernel virtual space, using a pagetable protection of @prot.
1624 */
1625void *__vmalloc_node_range(unsigned long size, unsigned long align,
1626 unsigned long start, unsigned long end, gfp_t gfp_mask,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001627 pgprot_t prot, int node, const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001628{
David Rientjesd0a21262011-01-13 15:46:02 -08001629 struct vm_struct *area;
1630 void *addr;
1631 unsigned long real_size = size;
1632
1633 size = PAGE_ALIGN(size);
1634 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001635 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001636
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001637 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED,
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001638 start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001639 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001640 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001641
Wanpeng Li3722e132013-11-12 15:07:29 -08001642 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001643 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001644 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001645
1646 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001647 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1648 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001649 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001650 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001651 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001652
1653 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001654 * A ref_count = 2 is needed because vm_struct allocated in
1655 * __get_vm_area_node() contains a reference to the virtual address of
1656 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001657 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001658 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001659
1660 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001661
1662fail:
1663 warn_alloc_failed(gfp_mask, 0,
1664 "vmalloc: allocation failure: %lu bytes\n",
1665 real_size);
1666 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001667}
1668
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001670 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001672 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 * @gfp_mask: flags for the page level allocator
1674 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001675 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001676 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 *
1678 * Allocate enough pages to cover @size from the page level
1679 * allocator with @gfp_mask flags. Map them into contiguous
1680 * kernel virtual space, using a pagetable protection of @prot.
1681 */
David Miller2dca6992009-09-21 12:22:34 -07001682static void *__vmalloc_node(unsigned long size, unsigned long align,
1683 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001684 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685{
David Rientjesd0a21262011-01-13 15:46:02 -08001686 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1687 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
Christoph Lameter930fc452005-10-29 18:15:41 -07001690void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1691{
David Rientjes00ef2d22013-02-22 16:35:36 -08001692 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001693 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001694}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695EXPORT_SYMBOL(__vmalloc);
1696
Dave Younge1ca7782010-10-26 14:22:06 -07001697static inline void *__vmalloc_node_flags(unsigned long size,
1698 int node, gfp_t flags)
1699{
1700 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1701 node, __builtin_return_address(0));
1702}
1703
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704/**
1705 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 * Allocate enough pages to cover @size from the page level
1708 * allocator and map them into contiguous kernel virtual space.
1709 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001710 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 * use __vmalloc() instead.
1712 */
1713void *vmalloc(unsigned long size)
1714{
David Rientjes00ef2d22013-02-22 16:35:36 -08001715 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1716 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718EXPORT_SYMBOL(vmalloc);
1719
Christoph Lameter930fc452005-10-29 18:15:41 -07001720/**
Dave Younge1ca7782010-10-26 14:22:06 -07001721 * vzalloc - allocate virtually contiguous memory with zero fill
1722 * @size: allocation size
1723 * Allocate enough pages to cover @size from the page level
1724 * allocator and map them into contiguous kernel virtual space.
1725 * The memory allocated is set to zero.
1726 *
1727 * For tight control over page level allocator and protection flags
1728 * use __vmalloc() instead.
1729 */
1730void *vzalloc(unsigned long size)
1731{
David Rientjes00ef2d22013-02-22 16:35:36 -08001732 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001733 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1734}
1735EXPORT_SYMBOL(vzalloc);
1736
1737/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001738 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1739 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001740 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001741 * The resulting memory area is zeroed so it can be mapped to userspace
1742 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001743 */
1744void *vmalloc_user(unsigned long size)
1745{
1746 struct vm_struct *area;
1747 void *ret;
1748
David Miller2dca6992009-09-21 12:22:34 -07001749 ret = __vmalloc_node(size, SHMLBA,
1750 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001751 PAGE_KERNEL, NUMA_NO_NODE,
1752 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001753 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001754 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001755 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001756 }
Nick Piggin83342312006-06-23 02:03:20 -07001757 return ret;
1758}
1759EXPORT_SYMBOL(vmalloc_user);
1760
1761/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001762 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001763 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001764 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001765 *
1766 * Allocate enough pages to cover @size from the page level
1767 * allocator and map them into contiguous kernel virtual space.
1768 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001769 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001770 * use __vmalloc() instead.
1771 */
1772void *vmalloc_node(unsigned long size, int node)
1773{
David Miller2dca6992009-09-21 12:22:34 -07001774 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001775 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001776}
1777EXPORT_SYMBOL(vmalloc_node);
1778
Dave Younge1ca7782010-10-26 14:22:06 -07001779/**
1780 * vzalloc_node - allocate memory on a specific node with zero fill
1781 * @size: allocation size
1782 * @node: numa node
1783 *
1784 * Allocate enough pages to cover @size from the page level
1785 * allocator and map them into contiguous kernel virtual space.
1786 * The memory allocated is set to zero.
1787 *
1788 * For tight control over page level allocator and protection flags
1789 * use __vmalloc_node() instead.
1790 */
1791void *vzalloc_node(unsigned long size, int node)
1792{
1793 return __vmalloc_node_flags(size, node,
1794 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1795}
1796EXPORT_SYMBOL(vzalloc_node);
1797
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001798#ifndef PAGE_KERNEL_EXEC
1799# define PAGE_KERNEL_EXEC PAGE_KERNEL
1800#endif
1801
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802/**
1803 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 * @size: allocation size
1805 *
1806 * Kernel-internal function to allocate enough pages to cover @size
1807 * the page level allocator and map them into contiguous and
1808 * executable kernel virtual space.
1809 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001810 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 * use __vmalloc() instead.
1812 */
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814void *vmalloc_exec(unsigned long size)
1815{
David Miller2dca6992009-09-21 12:22:34 -07001816 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001817 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818}
1819
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001820#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001821#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001822#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001823#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001824#else
1825#define GFP_VMALLOC32 GFP_KERNEL
1826#endif
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828/**
1829 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 * @size: allocation size
1831 *
1832 * Allocate enough 32bit PA addressable pages to cover @size from the
1833 * page level allocator and map them into contiguous kernel virtual space.
1834 */
1835void *vmalloc_32(unsigned long size)
1836{
David Miller2dca6992009-09-21 12:22:34 -07001837 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001838 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840EXPORT_SYMBOL(vmalloc_32);
1841
Nick Piggin83342312006-06-23 02:03:20 -07001842/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001843 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001844 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001845 *
1846 * The resulting memory area is 32bit addressable and zeroed so it can be
1847 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001848 */
1849void *vmalloc_32_user(unsigned long size)
1850{
1851 struct vm_struct *area;
1852 void *ret;
1853
David Miller2dca6992009-09-21 12:22:34 -07001854 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001855 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001856 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001857 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001858 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001859 }
Nick Piggin83342312006-06-23 02:03:20 -07001860 return ret;
1861}
1862EXPORT_SYMBOL(vmalloc_32_user);
1863
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001864/*
1865 * small helper routine , copy contents to buf from addr.
1866 * If the page is not present, fill zero.
1867 */
1868
1869static int aligned_vread(char *buf, char *addr, unsigned long count)
1870{
1871 struct page *p;
1872 int copied = 0;
1873
1874 while (count) {
1875 unsigned long offset, length;
1876
1877 offset = (unsigned long)addr & ~PAGE_MASK;
1878 length = PAGE_SIZE - offset;
1879 if (length > count)
1880 length = count;
1881 p = vmalloc_to_page(addr);
1882 /*
1883 * To do safe access to this _mapped_ area, we need
1884 * lock. But adding lock here means that we need to add
1885 * overhead of vmalloc()/vfree() calles for this _debug_
1886 * interface, rarely used. Instead of that, we'll use
1887 * kmap() and get small overhead in this access function.
1888 */
1889 if (p) {
1890 /*
1891 * we can expect USER0 is not used (see vread/vwrite's
1892 * function description)
1893 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001894 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001895 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001896 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001897 } else
1898 memset(buf, 0, length);
1899
1900 addr += length;
1901 buf += length;
1902 copied += length;
1903 count -= length;
1904 }
1905 return copied;
1906}
1907
1908static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1909{
1910 struct page *p;
1911 int copied = 0;
1912
1913 while (count) {
1914 unsigned long offset, length;
1915
1916 offset = (unsigned long)addr & ~PAGE_MASK;
1917 length = PAGE_SIZE - offset;
1918 if (length > count)
1919 length = count;
1920 p = vmalloc_to_page(addr);
1921 /*
1922 * To do safe access to this _mapped_ area, we need
1923 * lock. But adding lock here means that we need to add
1924 * overhead of vmalloc()/vfree() calles for this _debug_
1925 * interface, rarely used. Instead of that, we'll use
1926 * kmap() and get small overhead in this access function.
1927 */
1928 if (p) {
1929 /*
1930 * we can expect USER0 is not used (see vread/vwrite's
1931 * function description)
1932 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001933 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001934 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001935 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001936 }
1937 addr += length;
1938 buf += length;
1939 copied += length;
1940 count -= length;
1941 }
1942 return copied;
1943}
1944
1945/**
1946 * vread() - read vmalloc area in a safe way.
1947 * @buf: buffer for reading data
1948 * @addr: vm address.
1949 * @count: number of bytes to be read.
1950 *
1951 * Returns # of bytes which addr and buf should be increased.
1952 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1953 * includes any intersect with alive vmalloc area.
1954 *
1955 * This function checks that addr is a valid vmalloc'ed area, and
1956 * copy data from that area to a given buffer. If the given memory range
1957 * of [addr...addr+count) includes some valid address, data is copied to
1958 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1959 * IOREMAP area is treated as memory hole and no copy is done.
1960 *
1961 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08001962 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001963 *
1964 * Note: In usual ops, vread() is never necessary because the caller
1965 * should know vmalloc() area is valid and can use memcpy().
1966 * This is for routines which have to access vmalloc area without
1967 * any informaion, as /dev/kmem.
1968 *
1969 */
1970
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971long vread(char *buf, char *addr, unsigned long count)
1972{
Joonsoo Kime81ce852013-04-29 15:07:32 -07001973 struct vmap_area *va;
1974 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001976 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 unsigned long n;
1978
1979 /* Don't allow overflow */
1980 if ((unsigned long) addr + count < count)
1981 count = -(unsigned long) addr;
1982
Joonsoo Kime81ce852013-04-29 15:07:32 -07001983 spin_lock(&vmap_area_lock);
1984 list_for_each_entry(va, &vmap_area_list, list) {
1985 if (!count)
1986 break;
1987
1988 if (!(va->flags & VM_VM_AREA))
1989 continue;
1990
1991 vm = va->vm;
1992 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001993 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 continue;
1995 while (addr < vaddr) {
1996 if (count == 0)
1997 goto finished;
1998 *buf = '\0';
1999 buf++;
2000 addr++;
2001 count--;
2002 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002003 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002004 if (n > count)
2005 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002006 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002007 aligned_vread(buf, addr, n);
2008 else /* IOREMAP area is treated as memory hole */
2009 memset(buf, 0, n);
2010 buf += n;
2011 addr += n;
2012 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 }
2014finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002015 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002016
2017 if (buf == buf_start)
2018 return 0;
2019 /* zero-fill memory holes */
2020 if (buf != buf_start + buflen)
2021 memset(buf, 0, buflen - (buf - buf_start));
2022
2023 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024}
2025
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002026/**
2027 * vwrite() - write vmalloc area in a safe way.
2028 * @buf: buffer for source data
2029 * @addr: vm address.
2030 * @count: number of bytes to be read.
2031 *
2032 * Returns # of bytes which addr and buf should be incresed.
2033 * (same number to @count).
2034 * If [addr...addr+count) doesn't includes any intersect with valid
2035 * vmalloc area, returns 0.
2036 *
2037 * This function checks that addr is a valid vmalloc'ed area, and
2038 * copy data from a buffer to the given addr. If specified range of
2039 * [addr...addr+count) includes some valid address, data is copied from
2040 * proper area of @buf. If there are memory holes, no copy to hole.
2041 * IOREMAP area is treated as memory hole and no copy is done.
2042 *
2043 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002044 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002045 *
2046 * Note: In usual ops, vwrite() is never necessary because the caller
2047 * should know vmalloc() area is valid and can use memcpy().
2048 * This is for routines which have to access vmalloc area without
2049 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002050 */
2051
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052long vwrite(char *buf, char *addr, unsigned long count)
2053{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002054 struct vmap_area *va;
2055 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002056 char *vaddr;
2057 unsigned long n, buflen;
2058 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
2060 /* Don't allow overflow */
2061 if ((unsigned long) addr + count < count)
2062 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002063 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Joonsoo Kime81ce852013-04-29 15:07:32 -07002065 spin_lock(&vmap_area_lock);
2066 list_for_each_entry(va, &vmap_area_list, list) {
2067 if (!count)
2068 break;
2069
2070 if (!(va->flags & VM_VM_AREA))
2071 continue;
2072
2073 vm = va->vm;
2074 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002075 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 continue;
2077 while (addr < vaddr) {
2078 if (count == 0)
2079 goto finished;
2080 buf++;
2081 addr++;
2082 count--;
2083 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002084 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002085 if (n > count)
2086 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002087 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002088 aligned_vwrite(buf, addr, n);
2089 copied++;
2090 }
2091 buf += n;
2092 addr += n;
2093 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 }
2095finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002096 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002097 if (!copied)
2098 return 0;
2099 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100}
Nick Piggin83342312006-06-23 02:03:20 -07002101
2102/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002103 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2104 * @vma: vma to cover
2105 * @uaddr: target user address to start at
2106 * @kaddr: virtual address of vmalloc kernel memory
2107 * @size: size of map area
2108 *
2109 * Returns: 0 for success, -Exxx on failure
2110 *
2111 * This function checks that @kaddr is a valid vmalloc'ed area,
2112 * and that it is big enough to cover the range starting at
2113 * @uaddr in @vma. Will return failure if that criteria isn't
2114 * met.
2115 *
2116 * Similar to remap_pfn_range() (see mm/memory.c)
2117 */
2118int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2119 void *kaddr, unsigned long size)
2120{
2121 struct vm_struct *area;
2122
2123 size = PAGE_ALIGN(size);
2124
2125 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2126 return -EINVAL;
2127
2128 area = find_vm_area(kaddr);
2129 if (!area)
2130 return -EINVAL;
2131
2132 if (!(area->flags & VM_USERMAP))
2133 return -EINVAL;
2134
2135 if (kaddr + size > area->addr + area->size)
2136 return -EINVAL;
2137
2138 do {
2139 struct page *page = vmalloc_to_page(kaddr);
2140 int ret;
2141
2142 ret = vm_insert_page(vma, uaddr, page);
2143 if (ret)
2144 return ret;
2145
2146 uaddr += PAGE_SIZE;
2147 kaddr += PAGE_SIZE;
2148 size -= PAGE_SIZE;
2149 } while (size > 0);
2150
2151 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2152
2153 return 0;
2154}
2155EXPORT_SYMBOL(remap_vmalloc_range_partial);
2156
2157/**
Nick Piggin83342312006-06-23 02:03:20 -07002158 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002159 * @vma: vma to cover (map full range of vma)
2160 * @addr: vmalloc memory
2161 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002162 *
2163 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002164 *
2165 * This function checks that addr is a valid vmalloc'ed area, and
2166 * that it is big enough to cover the vma. Will return failure if
2167 * that criteria isn't met.
2168 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002169 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002170 */
2171int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2172 unsigned long pgoff)
2173{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002174 return remap_vmalloc_range_partial(vma, vma->vm_start,
2175 addr + (pgoff << PAGE_SHIFT),
2176 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002177}
2178EXPORT_SYMBOL(remap_vmalloc_range);
2179
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002180/*
2181 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2182 * have one.
2183 */
2184void __attribute__((weak)) vmalloc_sync_all(void)
2185{
2186}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002187
2188
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002189static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002190{
David Vrabelcd129092011-09-29 16:53:32 +01002191 pte_t ***p = data;
2192
2193 if (p) {
2194 *(*p) = pte;
2195 (*p)++;
2196 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002197 return 0;
2198}
2199
2200/**
2201 * alloc_vm_area - allocate a range of kernel address space
2202 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002203 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002204 *
2205 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002206 *
2207 * This function reserves a range of kernel address space, and
2208 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002209 * are created.
2210 *
2211 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2212 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002213 */
David Vrabelcd129092011-09-29 16:53:32 +01002214struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002215{
2216 struct vm_struct *area;
2217
Christoph Lameter23016962008-04-28 02:12:42 -07002218 area = get_vm_area_caller(size, VM_IOREMAP,
2219 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002220 if (area == NULL)
2221 return NULL;
2222
2223 /*
2224 * This ensures that page tables are constructed for this region
2225 * of kernel virtual address space and mapped into init_mm.
2226 */
2227 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002228 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002229 free_vm_area(area);
2230 return NULL;
2231 }
2232
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002233 return area;
2234}
2235EXPORT_SYMBOL_GPL(alloc_vm_area);
2236
2237void free_vm_area(struct vm_struct *area)
2238{
2239 struct vm_struct *ret;
2240 ret = remove_vm_area(area->addr);
2241 BUG_ON(ret != area);
2242 kfree(area);
2243}
2244EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002245
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002246#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002247static struct vmap_area *node_to_va(struct rb_node *n)
2248{
2249 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2250}
2251
2252/**
2253 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2254 * @end: target address
2255 * @pnext: out arg for the next vmap_area
2256 * @pprev: out arg for the previous vmap_area
2257 *
2258 * Returns: %true if either or both of next and prev are found,
2259 * %false if no vmap_area exists
2260 *
2261 * Find vmap_areas end addresses of which enclose @end. ie. if not
2262 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2263 */
2264static bool pvm_find_next_prev(unsigned long end,
2265 struct vmap_area **pnext,
2266 struct vmap_area **pprev)
2267{
2268 struct rb_node *n = vmap_area_root.rb_node;
2269 struct vmap_area *va = NULL;
2270
2271 while (n) {
2272 va = rb_entry(n, struct vmap_area, rb_node);
2273 if (end < va->va_end)
2274 n = n->rb_left;
2275 else if (end > va->va_end)
2276 n = n->rb_right;
2277 else
2278 break;
2279 }
2280
2281 if (!va)
2282 return false;
2283
2284 if (va->va_end > end) {
2285 *pnext = va;
2286 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2287 } else {
2288 *pprev = va;
2289 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2290 }
2291 return true;
2292}
2293
2294/**
2295 * pvm_determine_end - find the highest aligned address between two vmap_areas
2296 * @pnext: in/out arg for the next vmap_area
2297 * @pprev: in/out arg for the previous vmap_area
2298 * @align: alignment
2299 *
2300 * Returns: determined end address
2301 *
2302 * Find the highest aligned address between *@pnext and *@pprev below
2303 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2304 * down address is between the end addresses of the two vmap_areas.
2305 *
2306 * Please note that the address returned by this function may fall
2307 * inside *@pnext vmap_area. The caller is responsible for checking
2308 * that.
2309 */
2310static unsigned long pvm_determine_end(struct vmap_area **pnext,
2311 struct vmap_area **pprev,
2312 unsigned long align)
2313{
2314 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2315 unsigned long addr;
2316
2317 if (*pnext)
2318 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2319 else
2320 addr = vmalloc_end;
2321
2322 while (*pprev && (*pprev)->va_end > addr) {
2323 *pnext = *pprev;
2324 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2325 }
2326
2327 return addr;
2328}
2329
2330/**
2331 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2332 * @offsets: array containing offset of each area
2333 * @sizes: array containing size of each area
2334 * @nr_vms: the number of areas to allocate
2335 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002336 *
2337 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2338 * vm_structs on success, %NULL on failure
2339 *
2340 * Percpu allocator wants to use congruent vm areas so that it can
2341 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002342 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2343 * be scattered pretty far, distance between two areas easily going up
2344 * to gigabytes. To avoid interacting with regular vmallocs, these
2345 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002346 *
2347 * Despite its complicated look, this allocator is rather simple. It
2348 * does everything top-down and scans areas from the end looking for
2349 * matching slot. While scanning, if any of the areas overlaps with
2350 * existing vmap_area, the base address is pulled down to fit the
2351 * area. Scanning is repeated till all the areas fit and then all
2352 * necessary data structres are inserted and the result is returned.
2353 */
2354struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2355 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002356 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002357{
2358 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2359 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2360 struct vmap_area **vas, *prev, *next;
2361 struct vm_struct **vms;
2362 int area, area2, last_area, term_area;
2363 unsigned long base, start, end, last_end;
2364 bool purged = false;
2365
Tejun Heoca23e402009-08-14 15:00:52 +09002366 /* verify parameters and allocate data structures */
2367 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2368 for (last_area = 0, area = 0; area < nr_vms; area++) {
2369 start = offsets[area];
2370 end = start + sizes[area];
2371
2372 /* is everything aligned properly? */
2373 BUG_ON(!IS_ALIGNED(offsets[area], align));
2374 BUG_ON(!IS_ALIGNED(sizes[area], align));
2375
2376 /* detect the area with the highest address */
2377 if (start > offsets[last_area])
2378 last_area = area;
2379
2380 for (area2 = 0; area2 < nr_vms; area2++) {
2381 unsigned long start2 = offsets[area2];
2382 unsigned long end2 = start2 + sizes[area2];
2383
2384 if (area2 == area)
2385 continue;
2386
2387 BUG_ON(start2 >= start && start2 < end);
2388 BUG_ON(end2 <= end && end2 > start);
2389 }
2390 }
2391 last_end = offsets[last_area] + sizes[last_area];
2392
2393 if (vmalloc_end - vmalloc_start < last_end) {
2394 WARN_ON(true);
2395 return NULL;
2396 }
2397
Thomas Meyer4d67d862012-05-29 15:06:21 -07002398 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2399 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002400 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002401 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002402
2403 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002404 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2405 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002406 if (!vas[area] || !vms[area])
2407 goto err_free;
2408 }
2409retry:
2410 spin_lock(&vmap_area_lock);
2411
2412 /* start scanning - we scan from the top, begin with the last area */
2413 area = term_area = last_area;
2414 start = offsets[area];
2415 end = start + sizes[area];
2416
2417 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2418 base = vmalloc_end - last_end;
2419 goto found;
2420 }
2421 base = pvm_determine_end(&next, &prev, align) - end;
2422
2423 while (true) {
2424 BUG_ON(next && next->va_end <= base + end);
2425 BUG_ON(prev && prev->va_end > base + end);
2426
2427 /*
2428 * base might have underflowed, add last_end before
2429 * comparing.
2430 */
2431 if (base + last_end < vmalloc_start + last_end) {
2432 spin_unlock(&vmap_area_lock);
2433 if (!purged) {
2434 purge_vmap_area_lazy();
2435 purged = true;
2436 goto retry;
2437 }
2438 goto err_free;
2439 }
2440
2441 /*
2442 * If next overlaps, move base downwards so that it's
2443 * right below next and then recheck.
2444 */
2445 if (next && next->va_start < base + end) {
2446 base = pvm_determine_end(&next, &prev, align) - end;
2447 term_area = area;
2448 continue;
2449 }
2450
2451 /*
2452 * If prev overlaps, shift down next and prev and move
2453 * base so that it's right below new next and then
2454 * recheck.
2455 */
2456 if (prev && prev->va_end > base + start) {
2457 next = prev;
2458 prev = node_to_va(rb_prev(&next->rb_node));
2459 base = pvm_determine_end(&next, &prev, align) - end;
2460 term_area = area;
2461 continue;
2462 }
2463
2464 /*
2465 * This area fits, move on to the previous one. If
2466 * the previous one is the terminal one, we're done.
2467 */
2468 area = (area + nr_vms - 1) % nr_vms;
2469 if (area == term_area)
2470 break;
2471 start = offsets[area];
2472 end = start + sizes[area];
2473 pvm_find_next_prev(base + end, &next, &prev);
2474 }
2475found:
2476 /* we've found a fitting base, insert all va's */
2477 for (area = 0; area < nr_vms; area++) {
2478 struct vmap_area *va = vas[area];
2479
2480 va->va_start = base + offsets[area];
2481 va->va_end = va->va_start + sizes[area];
2482 __insert_vmap_area(va);
2483 }
2484
2485 vmap_area_pcpu_hole = base + offsets[last_area];
2486
2487 spin_unlock(&vmap_area_lock);
2488
2489 /* insert all vm's */
2490 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002491 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2492 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002493
2494 kfree(vas);
2495 return vms;
2496
2497err_free:
2498 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002499 kfree(vas[area]);
2500 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002501 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002502err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002503 kfree(vas);
2504 kfree(vms);
2505 return NULL;
2506}
2507
2508/**
2509 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2510 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2511 * @nr_vms: the number of allocated areas
2512 *
2513 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2514 */
2515void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2516{
2517 int i;
2518
2519 for (i = 0; i < nr_vms; i++)
2520 free_vm_area(vms[i]);
2521 kfree(vms);
2522}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002523#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002524
2525#ifdef CONFIG_PROC_FS
2526static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002527 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002528{
2529 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002530 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002531
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002532 spin_lock(&vmap_area_lock);
2533 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2534 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002535 n--;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002536 va = list_entry(va->list.next, typeof(*va), list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002537 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002538 if (!n && &va->list != &vmap_area_list)
2539 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002540
2541 return NULL;
2542
2543}
2544
2545static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2546{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002547 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002548
2549 ++*pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002550 next = list_entry(va->list.next, typeof(*va), list);
2551 if (&next->list != &vmap_area_list)
2552 return next;
2553
2554 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002555}
2556
2557static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002558 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002559{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002560 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002561}
2562
Eric Dumazeta47a1262008-07-23 21:27:38 -07002563static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2564{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002565 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002566 unsigned int nr, *counters = m->private;
2567
2568 if (!counters)
2569 return;
2570
Wanpeng Liaf123462013-11-12 15:07:32 -08002571 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2572 smp_rmb();
2573 if (v->flags & VM_UNINITIALIZED)
2574 return;
2575
Eric Dumazeta47a1262008-07-23 21:27:38 -07002576 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2577
2578 for (nr = 0; nr < v->nr_pages; nr++)
2579 counters[page_to_nid(v->pages[nr])]++;
2580
2581 for_each_node_state(nr, N_HIGH_MEMORY)
2582 if (counters[nr])
2583 seq_printf(m, " N%u=%u", nr, counters[nr]);
2584 }
2585}
2586
Christoph Lametera10aa572008-04-28 02:12:40 -07002587static int s_show(struct seq_file *m, void *p)
2588{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002589 struct vmap_area *va = p;
2590 struct vm_struct *v;
2591
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002592 /*
2593 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2594 * behalf of vmap area is being tear down or vm_map_ram allocation.
2595 */
2596 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002597 return 0;
2598
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002599 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002600
Kees Cook45ec1692012-10-08 16:34:09 -07002601 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002602 v->addr, v->addr + v->size, v->size);
2603
Joe Perches62c70bc2011-01-13 15:45:52 -08002604 if (v->caller)
2605 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002606
Christoph Lametera10aa572008-04-28 02:12:40 -07002607 if (v->nr_pages)
2608 seq_printf(m, " pages=%d", v->nr_pages);
2609
2610 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002611 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002612
2613 if (v->flags & VM_IOREMAP)
2614 seq_printf(m, " ioremap");
2615
2616 if (v->flags & VM_ALLOC)
2617 seq_printf(m, " vmalloc");
2618
2619 if (v->flags & VM_MAP)
2620 seq_printf(m, " vmap");
2621
2622 if (v->flags & VM_USERMAP)
2623 seq_printf(m, " user");
2624
2625 if (v->flags & VM_VPAGES)
2626 seq_printf(m, " vpages");
2627
Eric Dumazeta47a1262008-07-23 21:27:38 -07002628 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002629 seq_putc(m, '\n');
2630 return 0;
2631}
2632
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002633static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002634 .start = s_start,
2635 .next = s_next,
2636 .stop = s_stop,
2637 .show = s_show,
2638};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002639
2640static int vmalloc_open(struct inode *inode, struct file *file)
2641{
2642 unsigned int *ptr = NULL;
2643 int ret;
2644
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002645 if (IS_ENABLED(CONFIG_NUMA)) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002646 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002647 if (ptr == NULL)
2648 return -ENOMEM;
2649 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002650 ret = seq_open(file, &vmalloc_op);
2651 if (!ret) {
2652 struct seq_file *m = file->private_data;
2653 m->private = ptr;
2654 } else
2655 kfree(ptr);
2656 return ret;
2657}
2658
2659static const struct file_operations proc_vmalloc_operations = {
2660 .open = vmalloc_open,
2661 .read = seq_read,
2662 .llseek = seq_lseek,
2663 .release = seq_release_private,
2664};
2665
2666static int __init proc_vmalloc_init(void)
2667{
2668 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2669 return 0;
2670}
2671module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002672
2673void get_vmalloc_info(struct vmalloc_info *vmi)
2674{
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002675 struct vmap_area *va;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002676 unsigned long free_area_size;
2677 unsigned long prev_end;
2678
2679 vmi->used = 0;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002680 vmi->largest_chunk = 0;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002681
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002682 prev_end = VMALLOC_START;
2683
2684 spin_lock(&vmap_area_lock);
2685
2686 if (list_empty(&vmap_area_list)) {
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002687 vmi->largest_chunk = VMALLOC_TOTAL;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002688 goto out;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002689 }
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002690
2691 list_for_each_entry(va, &vmap_area_list, list) {
2692 unsigned long addr = va->va_start;
2693
2694 /*
2695 * Some archs keep another range for modules in vmalloc space
2696 */
2697 if (addr < VMALLOC_START)
2698 continue;
2699 if (addr >= VMALLOC_END)
2700 break;
2701
2702 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2703 continue;
2704
2705 vmi->used += (va->va_end - va->va_start);
2706
2707 free_area_size = addr - prev_end;
2708 if (vmi->largest_chunk < free_area_size)
2709 vmi->largest_chunk = free_area_size;
2710
2711 prev_end = va->va_end;
2712 }
2713
2714 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2715 vmi->largest_chunk = VMALLOC_END - prev_end;
2716
2717out:
2718 spin_unlock(&vmap_area_lock);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002719}
Christoph Lametera10aa572008-04-28 02:12:40 -07002720#endif
2721