blob: 90520af7f18604ccd825d2595e6103dccbbf3f32 [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>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070030#include <linux/compiler.h>
Al Viro32fcfd42013-03-10 20:14:08 -040031#include <linux/llist.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <asm/uaccess.h>
34#include <asm/tlbflush.h>
David Miller2dca6992009-09-21 12:22:34 -070035#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Al Viro32fcfd42013-03-10 20:14:08 -040037struct vfree_deferred {
38 struct llist_head list;
39 struct work_struct wq;
40};
41static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
42
43static void __vunmap(const void *, int);
44
45static void free_work(struct work_struct *w)
46{
47 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
48 struct llist_node *llnode = llist_del_all(&p->list);
49 while (llnode) {
50 void *p = llnode;
51 llnode = llist_next(llnode);
52 __vunmap(p, 1);
53 }
54}
55
Nick Piggindb64fe02008-10-18 20:27:03 -070056/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070057
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
59{
60 pte_t *pte;
61
62 pte = pte_offset_kernel(pmd, addr);
63 do {
64 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
65 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
66 } while (pte++, addr += PAGE_SIZE, addr != end);
67}
68
Nick Piggindb64fe02008-10-18 20:27:03 -070069static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
71 pmd_t *pmd;
72 unsigned long next;
73
74 pmd = pmd_offset(pud, addr);
75 do {
76 next = pmd_addr_end(addr, end);
77 if (pmd_none_or_clear_bad(pmd))
78 continue;
79 vunmap_pte_range(pmd, addr, next);
80 } while (pmd++, addr = next, addr != end);
81}
82
Nick Piggindb64fe02008-10-18 20:27:03 -070083static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084{
85 pud_t *pud;
86 unsigned long next;
87
88 pud = pud_offset(pgd, addr);
89 do {
90 next = pud_addr_end(addr, end);
91 if (pud_none_or_clear_bad(pud))
92 continue;
93 vunmap_pmd_range(pud, addr, next);
94 } while (pud++, addr = next, addr != end);
95}
96
Nick Piggindb64fe02008-10-18 20:27:03 -070097static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
99 pgd_t *pgd;
100 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102 BUG_ON(addr >= end);
103 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 do {
105 next = pgd_addr_end(addr, end);
106 if (pgd_none_or_clear_bad(pgd))
107 continue;
108 vunmap_pud_range(pgd, addr, next);
109 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110}
111
112static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -0700113 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
115 pte_t *pte;
116
Nick Piggindb64fe02008-10-18 20:27:03 -0700117 /*
118 * nr is a running index into the array which helps higher level
119 * callers keep track of where we're up to.
120 */
121
Hugh Dickins872fec12005-10-29 18:16:21 -0700122 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 if (!pte)
124 return -ENOMEM;
125 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700126 struct page *page = pages[*nr];
127
128 if (WARN_ON(!pte_none(*pte)))
129 return -EBUSY;
130 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 return -ENOMEM;
132 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700133 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 } while (pte++, addr += PAGE_SIZE, addr != end);
135 return 0;
136}
137
Nick Piggindb64fe02008-10-18 20:27:03 -0700138static int vmap_pmd_range(pud_t *pud, unsigned long addr,
139 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 pmd_t *pmd;
142 unsigned long next;
143
144 pmd = pmd_alloc(&init_mm, pud, addr);
145 if (!pmd)
146 return -ENOMEM;
147 do {
148 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700149 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 return -ENOMEM;
151 } while (pmd++, addr = next, addr != end);
152 return 0;
153}
154
Nick Piggindb64fe02008-10-18 20:27:03 -0700155static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
156 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
158 pud_t *pud;
159 unsigned long next;
160
161 pud = pud_alloc(&init_mm, pgd, addr);
162 if (!pud)
163 return -ENOMEM;
164 do {
165 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700166 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 return -ENOMEM;
168 } while (pud++, addr = next, addr != end);
169 return 0;
170}
171
Nick Piggindb64fe02008-10-18 20:27:03 -0700172/*
173 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
174 * will have pfns corresponding to the "pages" array.
175 *
176 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
177 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900178static int vmap_page_range_noflush(unsigned long start, unsigned long end,
179 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
181 pgd_t *pgd;
182 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800183 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700184 int err = 0;
185 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187 BUG_ON(addr >= end);
188 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 do {
190 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700191 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700193 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700195
Nick Piggindb64fe02008-10-18 20:27:03 -0700196 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Tejun Heo8fc48982009-02-20 16:29:08 +0900199static int vmap_page_range(unsigned long start, unsigned long end,
200 pgprot_t prot, struct page **pages)
201{
202 int ret;
203
204 ret = vmap_page_range_noflush(start, end, prot, pages);
205 flush_cache_vmap(start, end);
206 return ret;
207}
208
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700209int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700210{
211 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000212 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700213 * and fall back on vmalloc() if that fails. Others
214 * just put it in the vmalloc space.
215 */
216#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
217 unsigned long addr = (unsigned long)x;
218 if (addr >= MODULES_VADDR && addr < MODULES_END)
219 return 1;
220#endif
221 return is_vmalloc_addr(x);
222}
223
Christoph Lameter48667e72008-02-04 22:28:31 -0800224/*
malcadd688f2014-01-27 17:06:53 -0800225 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800226 */
malcadd688f2014-01-27 17:06:53 -0800227struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800228{
229 unsigned long addr = (unsigned long) vmalloc_addr;
malcadd688f2014-01-27 17:06:53 -0800230 struct page *page = NULL;
Christoph Lameter48667e72008-02-04 22:28:31 -0800231 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800232
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200233 /*
234 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
235 * architectures that do not vmalloc module space
236 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700237 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200238
Christoph Lameter48667e72008-02-04 22:28:31 -0800239 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700240 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800241 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700242 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800243 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700244 pte_t *ptep, pte;
245
Christoph Lameter48667e72008-02-04 22:28:31 -0800246 ptep = pte_offset_map(pmd, addr);
247 pte = *ptep;
248 if (pte_present(pte))
malcadd688f2014-01-27 17:06:53 -0800249 page = pte_page(pte);
Christoph Lameter48667e72008-02-04 22:28:31 -0800250 pte_unmap(ptep);
251 }
252 }
253 }
malcadd688f2014-01-27 17:06:53 -0800254 return page;
Jianyu Zhanece86e222014-01-21 15:49:12 -0800255}
256EXPORT_SYMBOL(vmalloc_to_page);
257
malcadd688f2014-01-27 17:06:53 -0800258/*
259 * Map a vmalloc()-space virtual address to the physical page frame number.
260 */
261unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
262{
263 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
264}
265EXPORT_SYMBOL(vmalloc_to_pfn);
266
Nick Piggindb64fe02008-10-18 20:27:03 -0700267
268/*** Global kva allocator ***/
269
270#define VM_LAZY_FREE 0x01
271#define VM_LAZY_FREEING 0x02
272#define VM_VM_AREA 0x04
273
Nick Piggindb64fe02008-10-18 20:27:03 -0700274static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700275/* Export for kexec only */
276LIST_HEAD(vmap_area_list);
Nick Piggin89699602011-03-22 16:30:36 -0700277static struct rb_root vmap_area_root = RB_ROOT;
278
279/* The vmap cache globals are protected by vmap_area_lock */
280static struct rb_node *free_vmap_cache;
281static unsigned long cached_hole_size;
282static unsigned long cached_vstart;
283static unsigned long cached_align;
284
Tejun Heoca23e402009-08-14 15:00:52 +0900285static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700286
287static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
Nick Piggindb64fe02008-10-18 20:27:03 -0700289 struct rb_node *n = vmap_area_root.rb_node;
290
291 while (n) {
292 struct vmap_area *va;
293
294 va = rb_entry(n, struct vmap_area, rb_node);
295 if (addr < va->va_start)
296 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700297 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700298 n = n->rb_right;
299 else
300 return va;
301 }
302
303 return NULL;
304}
305
306static void __insert_vmap_area(struct vmap_area *va)
307{
308 struct rb_node **p = &vmap_area_root.rb_node;
309 struct rb_node *parent = NULL;
310 struct rb_node *tmp;
311
312 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700313 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700314
315 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700316 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
317 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700318 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700319 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700320 p = &(*p)->rb_right;
321 else
322 BUG();
323 }
324
325 rb_link_node(&va->rb_node, parent, p);
326 rb_insert_color(&va->rb_node, &vmap_area_root);
327
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700328 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700329 tmp = rb_prev(&va->rb_node);
330 if (tmp) {
331 struct vmap_area *prev;
332 prev = rb_entry(tmp, struct vmap_area, rb_node);
333 list_add_rcu(&va->list, &prev->list);
334 } else
335 list_add_rcu(&va->list, &vmap_area_list);
336}
337
338static void purge_vmap_area_lazy(void);
339
340/*
341 * Allocate a region of KVA of the specified size and alignment, within the
342 * vstart and vend.
343 */
344static struct vmap_area *alloc_vmap_area(unsigned long size,
345 unsigned long align,
346 unsigned long vstart, unsigned long vend,
347 int node, gfp_t gfp_mask)
348{
349 struct vmap_area *va;
350 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700352 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700353 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700354
Nick Piggin77669702009-02-27 14:03:03 -0800355 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700356 BUG_ON(size & ~PAGE_MASK);
Nick Piggin89699602011-03-22 16:30:36 -0700357 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700358
Nick Piggindb64fe02008-10-18 20:27:03 -0700359 va = kmalloc_node(sizeof(struct vmap_area),
360 gfp_mask & GFP_RECLAIM_MASK, node);
361 if (unlikely(!va))
362 return ERR_PTR(-ENOMEM);
363
Catalin Marinas7f88f882013-11-12 15:07:45 -0800364 /*
365 * Only scan the relevant parts containing pointers to other objects
366 * to avoid false negatives.
367 */
368 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
369
Nick Piggindb64fe02008-10-18 20:27:03 -0700370retry:
371 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700372 /*
373 * Invalidate cache if we have more permissive parameters.
374 * cached_hole_size notes the largest hole noticed _below_
375 * the vmap_area cached in free_vmap_cache: if size fits
376 * into that hole, we want to scan from vstart to reuse
377 * the hole instead of allocating above free_vmap_cache.
378 * Note that __free_vmap_area may update free_vmap_cache
379 * without updating cached_hole_size or cached_align.
380 */
381 if (!free_vmap_cache ||
382 size < cached_hole_size ||
383 vstart < cached_vstart ||
384 align < cached_align) {
385nocache:
386 cached_hole_size = 0;
387 free_vmap_cache = NULL;
388 }
389 /* record if we encounter less permissive parameters */
390 cached_vstart = vstart;
391 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800392
Nick Piggin89699602011-03-22 16:30:36 -0700393 /* find starting point for our search */
394 if (free_vmap_cache) {
395 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700396 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700397 if (addr < vstart)
398 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700399 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700400 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700401
Nick Piggin89699602011-03-22 16:30:36 -0700402 } else {
403 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700404 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700405 goto overflow;
406
407 n = vmap_area_root.rb_node;
408 first = NULL;
409
410 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700411 struct vmap_area *tmp;
412 tmp = rb_entry(n, struct vmap_area, rb_node);
413 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700414 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700415 if (tmp->va_start <= addr)
416 break;
417 n = n->rb_left;
418 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700419 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700420 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700421
422 if (!first)
423 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700424 }
Nick Piggin89699602011-03-22 16:30:36 -0700425
426 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700427 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700428 if (addr + cached_hole_size < first->va_start)
429 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700430 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700431 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700432 goto overflow;
433
Hong zhi guo92ca9222012-07-31 16:41:35 -0700434 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700435 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700436
437 first = list_entry(first->list.next,
438 struct vmap_area, list);
Nick Piggin89699602011-03-22 16:30:36 -0700439 }
440
Nick Piggindb64fe02008-10-18 20:27:03 -0700441found:
Nick Piggin89699602011-03-22 16:30:36 -0700442 if (addr + size > vend)
443 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700444
445 va->va_start = addr;
446 va->va_end = addr + size;
447 va->flags = 0;
448 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700449 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700450 spin_unlock(&vmap_area_lock);
451
Nick Piggin89699602011-03-22 16:30:36 -0700452 BUG_ON(va->va_start & (align-1));
453 BUG_ON(va->va_start < vstart);
454 BUG_ON(va->va_end > vend);
455
Nick Piggindb64fe02008-10-18 20:27:03 -0700456 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700457
458overflow:
459 spin_unlock(&vmap_area_lock);
460 if (!purged) {
461 purge_vmap_area_lazy();
462 purged = 1;
463 goto retry;
464 }
465 if (printk_ratelimit())
466 printk(KERN_WARNING
467 "vmap allocation for size %lu failed: "
468 "use vmalloc=<size> to increase size.\n", size);
469 kfree(va);
470 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700471}
472
Nick Piggindb64fe02008-10-18 20:27:03 -0700473static void __free_vmap_area(struct vmap_area *va)
474{
475 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700476
477 if (free_vmap_cache) {
478 if (va->va_end < cached_vstart) {
479 free_vmap_cache = NULL;
480 } else {
481 struct vmap_area *cache;
482 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
483 if (va->va_start <= cache->va_start) {
484 free_vmap_cache = rb_prev(&va->rb_node);
485 /*
486 * We don't try to update cached_hole_size or
487 * cached_align, but it won't go very wrong.
488 */
489 }
490 }
491 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700492 rb_erase(&va->rb_node, &vmap_area_root);
493 RB_CLEAR_NODE(&va->rb_node);
494 list_del_rcu(&va->list);
495
Tejun Heoca23e402009-08-14 15:00:52 +0900496 /*
497 * Track the highest possible candidate for pcpu area
498 * allocation. Areas outside of vmalloc area can be returned
499 * here too, consider only end addresses which fall inside
500 * vmalloc area proper.
501 */
502 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
503 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
504
Lai Jiangshan14769de2011-03-18 12:12:19 +0800505 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700506}
507
508/*
509 * Free a region of KVA allocated by alloc_vmap_area
510 */
511static void free_vmap_area(struct vmap_area *va)
512{
513 spin_lock(&vmap_area_lock);
514 __free_vmap_area(va);
515 spin_unlock(&vmap_area_lock);
516}
517
518/*
519 * Clear the pagetable entries of a given vmap_area
520 */
521static void unmap_vmap_area(struct vmap_area *va)
522{
523 vunmap_page_range(va->va_start, va->va_end);
524}
525
Nick Piggincd528582009-01-06 14:39:20 -0800526static void vmap_debug_free_range(unsigned long start, unsigned long end)
527{
528 /*
529 * Unmap page tables and force a TLB flush immediately if
530 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
531 * bugs similarly to those in linear kernel virtual address
532 * space after a page has been freed.
533 *
534 * All the lazy freeing logic is still retained, in order to
535 * minimise intrusiveness of this debugging feature.
536 *
537 * This is going to be *slow* (linear kernel virtual address
538 * debugging doesn't do a broadcast TLB flush so it is a lot
539 * faster).
540 */
541#ifdef CONFIG_DEBUG_PAGEALLOC
542 vunmap_page_range(start, end);
543 flush_tlb_kernel_range(start, end);
544#endif
545}
546
Nick Piggindb64fe02008-10-18 20:27:03 -0700547/*
548 * lazy_max_pages is the maximum amount of virtual address space we gather up
549 * before attempting to purge with a TLB flush.
550 *
551 * There is a tradeoff here: a larger number will cover more kernel page tables
552 * and take slightly longer to purge, but it will linearly reduce the number of
553 * global TLB flushes that must be performed. It would seem natural to scale
554 * this number up linearly with the number of CPUs (because vmapping activity
555 * could also scale linearly with the number of CPUs), however it is likely
556 * that in practice, workloads might be constrained in other ways that mean
557 * vmap activity will not scale linearly with CPUs. Also, I want to be
558 * conservative and not introduce a big latency on huge systems, so go with
559 * a less aggressive log scale. It will still be an improvement over the old
560 * code, and it will be simple to change the scale factor if we find that it
561 * becomes a problem on bigger systems.
562 */
563static unsigned long lazy_max_pages(void)
564{
565 unsigned int log;
566
567 log = fls(num_online_cpus());
568
569 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
570}
571
572static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
573
Nick Piggin02b709d2010-02-01 22:25:57 +1100574/* for per-CPU blocks */
575static void purge_fragmented_blocks_allcpus(void);
576
Nick Piggindb64fe02008-10-18 20:27:03 -0700577/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500578 * called before a call to iounmap() if the caller wants vm_area_struct's
579 * immediately freed.
580 */
581void set_iounmap_nonlazy(void)
582{
583 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
584}
585
586/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700587 * Purges all lazily-freed vmap areas.
588 *
589 * If sync is 0 then don't purge if there is already a purge in progress.
590 * If force_flush is 1, then flush kernel TLBs between *start and *end even
591 * if we found no lazy vmap areas to unmap (callers can use this to optimise
592 * their own TLB flushing).
593 * Returns with *start = min(*start, lowest purged address)
594 * *end = max(*end, highest purged address)
595 */
596static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
597 int sync, int force_flush)
598{
Andrew Morton46666d82009-01-15 13:51:15 -0800599 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700600 LIST_HEAD(valist);
601 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800602 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700603 int nr = 0;
604
605 /*
606 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
607 * should not expect such behaviour. This just simplifies locking for
608 * the case that isn't actually used at the moment anyway.
609 */
610 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800611 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700612 return;
613 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800614 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700615
Nick Piggin02b709d2010-02-01 22:25:57 +1100616 if (sync)
617 purge_fragmented_blocks_allcpus();
618
Nick Piggindb64fe02008-10-18 20:27:03 -0700619 rcu_read_lock();
620 list_for_each_entry_rcu(va, &vmap_area_list, list) {
621 if (va->flags & VM_LAZY_FREE) {
622 if (va->va_start < *start)
623 *start = va->va_start;
624 if (va->va_end > *end)
625 *end = va->va_end;
626 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700627 list_add_tail(&va->purge_list, &valist);
628 va->flags |= VM_LAZY_FREEING;
629 va->flags &= ~VM_LAZY_FREE;
630 }
631 }
632 rcu_read_unlock();
633
Yongseok Koh88f50042010-01-19 17:33:49 +0900634 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700635 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700636
637 if (nr || force_flush)
638 flush_tlb_kernel_range(*start, *end);
639
640 if (nr) {
641 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800642 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700643 __free_vmap_area(va);
644 spin_unlock(&vmap_area_lock);
645 }
Andrew Morton46666d82009-01-15 13:51:15 -0800646 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700647}
648
649/*
Nick Piggin496850e2008-11-19 15:36:33 -0800650 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
651 * is already purging.
652 */
653static void try_purge_vmap_area_lazy(void)
654{
655 unsigned long start = ULONG_MAX, end = 0;
656
657 __purge_vmap_area_lazy(&start, &end, 0, 0);
658}
659
660/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700661 * Kick off a purge of the outstanding lazy areas.
662 */
663static void purge_vmap_area_lazy(void)
664{
665 unsigned long start = ULONG_MAX, end = 0;
666
Nick Piggin496850e2008-11-19 15:36:33 -0800667 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700668}
669
670/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800671 * Free a vmap area, caller ensuring that the area has been unmapped
672 * and flush_cache_vunmap had been called for the correct range
673 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700674 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800675static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700676{
677 va->flags |= VM_LAZY_FREE;
678 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
679 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800680 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700681}
682
Nick Pigginb29acbd2008-12-01 13:13:47 -0800683/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800684 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
685 * called for the correct range previously.
686 */
687static void free_unmap_vmap_area_noflush(struct vmap_area *va)
688{
689 unmap_vmap_area(va);
690 free_vmap_area_noflush(va);
691}
692
693/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800694 * Free and unmap a vmap area
695 */
696static void free_unmap_vmap_area(struct vmap_area *va)
697{
698 flush_cache_vunmap(va->va_start, va->va_end);
699 free_unmap_vmap_area_noflush(va);
700}
701
Nick Piggindb64fe02008-10-18 20:27:03 -0700702static struct vmap_area *find_vmap_area(unsigned long addr)
703{
704 struct vmap_area *va;
705
706 spin_lock(&vmap_area_lock);
707 va = __find_vmap_area(addr);
708 spin_unlock(&vmap_area_lock);
709
710 return va;
711}
712
713static void free_unmap_vmap_area_addr(unsigned long addr)
714{
715 struct vmap_area *va;
716
717 va = find_vmap_area(addr);
718 BUG_ON(!va);
719 free_unmap_vmap_area(va);
720}
721
722
723/*** Per cpu kva allocator ***/
724
725/*
726 * vmap space is limited especially on 32 bit architectures. Ensure there is
727 * room for at least 16 percpu vmap blocks per CPU.
728 */
729/*
730 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
731 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
732 * instead (we just need a rough idea)
733 */
734#if BITS_PER_LONG == 32
735#define VMALLOC_SPACE (128UL*1024*1024)
736#else
737#define VMALLOC_SPACE (128UL*1024*1024*1024)
738#endif
739
740#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
741#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
742#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
743#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
744#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
745#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200746#define VMAP_BBMAP_BITS \
747 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
748 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
749 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700750
751#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
752
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100753static bool vmap_initialized __read_mostly = false;
754
Nick Piggindb64fe02008-10-18 20:27:03 -0700755struct vmap_block_queue {
756 spinlock_t lock;
757 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700758};
759
760struct vmap_block {
761 spinlock_t lock;
762 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700763 unsigned long free, dirty;
Nick Piggindb64fe02008-10-18 20:27:03 -0700764 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100765 struct list_head free_list;
766 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100767 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700768};
769
770/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
771static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
772
773/*
774 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
775 * in the free path. Could get rid of this if we change the API to return a
776 * "cookie" from alloc, to be passed to free. But no big deal yet.
777 */
778static DEFINE_SPINLOCK(vmap_block_tree_lock);
779static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
780
781/*
782 * We should probably have a fallback mechanism to allocate virtual memory
783 * out of partially filled vmap blocks. However vmap block sizing should be
784 * fairly reasonable according to the vmalloc size, so it shouldn't be a
785 * big problem.
786 */
787
788static unsigned long addr_to_vb_idx(unsigned long addr)
789{
790 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
791 addr /= VMAP_BLOCK_SIZE;
792 return addr;
793}
794
795static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
796{
797 struct vmap_block_queue *vbq;
798 struct vmap_block *vb;
799 struct vmap_area *va;
800 unsigned long vb_idx;
801 int node, err;
802
803 node = numa_node_id();
804
805 vb = kmalloc_node(sizeof(struct vmap_block),
806 gfp_mask & GFP_RECLAIM_MASK, node);
807 if (unlikely(!vb))
808 return ERR_PTR(-ENOMEM);
809
810 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
811 VMALLOC_START, VMALLOC_END,
812 node, gfp_mask);
Tobias Klauserddf9c6d42011-01-13 15:46:15 -0800813 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700814 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700815 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700816 }
817
818 err = radix_tree_preload(gfp_mask);
819 if (unlikely(err)) {
820 kfree(vb);
821 free_vmap_area(va);
822 return ERR_PTR(err);
823 }
824
825 spin_lock_init(&vb->lock);
826 vb->va = va;
827 vb->free = VMAP_BBMAP_BITS;
828 vb->dirty = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700829 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
830 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700831
832 vb_idx = addr_to_vb_idx(va->va_start);
833 spin_lock(&vmap_block_tree_lock);
834 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
835 spin_unlock(&vmap_block_tree_lock);
836 BUG_ON(err);
837 radix_tree_preload_end();
838
839 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700840 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100841 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700842 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900843 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700844
845 return vb;
846}
847
Nick Piggindb64fe02008-10-18 20:27:03 -0700848static void free_vmap_block(struct vmap_block *vb)
849{
850 struct vmap_block *tmp;
851 unsigned long vb_idx;
852
Nick Piggindb64fe02008-10-18 20:27:03 -0700853 vb_idx = addr_to_vb_idx(vb->va->va_start);
854 spin_lock(&vmap_block_tree_lock);
855 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
856 spin_unlock(&vmap_block_tree_lock);
857 BUG_ON(tmp != vb);
858
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800859 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800860 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700861}
862
Nick Piggin02b709d2010-02-01 22:25:57 +1100863static void purge_fragmented_blocks(int cpu)
864{
865 LIST_HEAD(purge);
866 struct vmap_block *vb;
867 struct vmap_block *n_vb;
868 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
869
870 rcu_read_lock();
871 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
872
873 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
874 continue;
875
876 spin_lock(&vb->lock);
877 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
878 vb->free = 0; /* prevent further allocs after releasing lock */
879 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Nick Piggin02b709d2010-02-01 22:25:57 +1100880 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
881 spin_lock(&vbq->lock);
882 list_del_rcu(&vb->free_list);
883 spin_unlock(&vbq->lock);
884 spin_unlock(&vb->lock);
885 list_add_tail(&vb->purge, &purge);
886 } else
887 spin_unlock(&vb->lock);
888 }
889 rcu_read_unlock();
890
891 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
892 list_del(&vb->purge);
893 free_vmap_block(vb);
894 }
895}
896
Nick Piggin02b709d2010-02-01 22:25:57 +1100897static void purge_fragmented_blocks_allcpus(void)
898{
899 int cpu;
900
901 for_each_possible_cpu(cpu)
902 purge_fragmented_blocks(cpu);
903}
904
Nick Piggindb64fe02008-10-18 20:27:03 -0700905static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
906{
907 struct vmap_block_queue *vbq;
908 struct vmap_block *vb;
909 unsigned long addr = 0;
910 unsigned int order;
911
912 BUG_ON(size & ~PAGE_MASK);
913 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700914 if (WARN_ON(size == 0)) {
915 /*
916 * Allocating 0 bytes isn't what caller wants since
917 * get_order(0) returns funny result. Just warn and terminate
918 * early.
919 */
920 return NULL;
921 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700922 order = get_order(size);
923
924again:
925 rcu_read_lock();
926 vbq = &get_cpu_var(vmap_block_queue);
927 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
928 int i;
929
930 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100931 if (vb->free < 1UL << order)
932 goto next;
933
Zhang Yanfei3fcd76e2013-07-08 15:59:54 -0700934 i = VMAP_BBMAP_BITS - vb->free;
Nick Piggin02b709d2010-02-01 22:25:57 +1100935 addr = vb->va->va_start + (i << PAGE_SHIFT);
936 BUG_ON(addr_to_vb_idx(addr) !=
937 addr_to_vb_idx(vb->va->va_start));
938 vb->free -= 1UL << order;
939 if (vb->free == 0) {
940 spin_lock(&vbq->lock);
941 list_del_rcu(&vb->free_list);
942 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700943 }
944 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100945 break;
946next:
947 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700948 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100949
Tejun Heo3f04ba82009-10-29 22:34:12 +0900950 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700951 rcu_read_unlock();
952
953 if (!addr) {
954 vb = new_vmap_block(gfp_mask);
955 if (IS_ERR(vb))
956 return vb;
957 goto again;
958 }
959
960 return (void *)addr;
961}
962
963static void vb_free(const void *addr, unsigned long size)
964{
965 unsigned long offset;
966 unsigned long vb_idx;
967 unsigned int order;
968 struct vmap_block *vb;
969
970 BUG_ON(size & ~PAGE_MASK);
971 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800972
973 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
974
Nick Piggindb64fe02008-10-18 20:27:03 -0700975 order = get_order(size);
976
977 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
978
979 vb_idx = addr_to_vb_idx((unsigned long)addr);
980 rcu_read_lock();
981 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
982 rcu_read_unlock();
983 BUG_ON(!vb);
984
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800985 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
986
Nick Piggindb64fe02008-10-18 20:27:03 -0700987 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100988 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700989
Nick Piggindb64fe02008-10-18 20:27:03 -0700990 vb->dirty += 1UL << order;
991 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100992 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700993 spin_unlock(&vb->lock);
994 free_vmap_block(vb);
995 } else
996 spin_unlock(&vb->lock);
997}
998
999/**
1000 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1001 *
1002 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1003 * to amortize TLB flushing overheads. What this means is that any page you
1004 * have now, may, in a former life, have been mapped into kernel virtual
1005 * address by the vmap layer and so there might be some CPUs with TLB entries
1006 * still referencing that page (additional to the regular 1:1 kernel mapping).
1007 *
1008 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1009 * be sure that none of the pages we have control over will have any aliases
1010 * from the vmap layer.
1011 */
1012void vm_unmap_aliases(void)
1013{
1014 unsigned long start = ULONG_MAX, end = 0;
1015 int cpu;
1016 int flush = 0;
1017
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001018 if (unlikely(!vmap_initialized))
1019 return;
1020
Nick Piggindb64fe02008-10-18 20:27:03 -07001021 for_each_possible_cpu(cpu) {
1022 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1023 struct vmap_block *vb;
1024
1025 rcu_read_lock();
1026 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001027 int i, j;
Nick Piggindb64fe02008-10-18 20:27:03 -07001028
1029 spin_lock(&vb->lock);
1030 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001031 if (i < VMAP_BBMAP_BITS) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001032 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001033
1034 j = find_last_bit(vb->dirty_map,
1035 VMAP_BBMAP_BITS);
1036 j = j + 1; /* need exclusive index */
Nick Piggindb64fe02008-10-18 20:27:03 -07001037
1038 s = vb->va->va_start + (i << PAGE_SHIFT);
1039 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001040 flush = 1;
1041
1042 if (s < start)
1043 start = s;
1044 if (e > end)
1045 end = e;
Nick Piggindb64fe02008-10-18 20:27:03 -07001046 }
1047 spin_unlock(&vb->lock);
1048 }
1049 rcu_read_unlock();
1050 }
1051
1052 __purge_vmap_area_lazy(&start, &end, 1, flush);
1053}
1054EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1055
1056/**
1057 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1058 * @mem: the pointer returned by vm_map_ram
1059 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1060 */
1061void vm_unmap_ram(const void *mem, unsigned int count)
1062{
1063 unsigned long size = count << PAGE_SHIFT;
1064 unsigned long addr = (unsigned long)mem;
1065
1066 BUG_ON(!addr);
1067 BUG_ON(addr < VMALLOC_START);
1068 BUG_ON(addr > VMALLOC_END);
1069 BUG_ON(addr & (PAGE_SIZE-1));
1070
1071 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001072 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001073
1074 if (likely(count <= VMAP_MAX_ALLOC))
1075 vb_free(mem, size);
1076 else
1077 free_unmap_vmap_area_addr(addr);
1078}
1079EXPORT_SYMBOL(vm_unmap_ram);
1080
1081/**
1082 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1083 * @pages: an array of pointers to the pages to be mapped
1084 * @count: number of pages
1085 * @node: prefer to allocate data structures on this node
1086 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001087 *
Gioh Kim36437632014-04-07 15:37:37 -07001088 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1089 * faster than vmap so it's good. But if you mix long-life and short-life
1090 * objects with vm_map_ram(), it could consume lots of address space through
1091 * fragmentation (especially on a 32bit machine). You could see failures in
1092 * the end. Please use this function for short-lived objects.
1093 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001094 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001095 */
1096void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1097{
1098 unsigned long size = count << PAGE_SHIFT;
1099 unsigned long addr;
1100 void *mem;
1101
1102 if (likely(count <= VMAP_MAX_ALLOC)) {
1103 mem = vb_alloc(size, GFP_KERNEL);
1104 if (IS_ERR(mem))
1105 return NULL;
1106 addr = (unsigned long)mem;
1107 } else {
1108 struct vmap_area *va;
1109 va = alloc_vmap_area(size, PAGE_SIZE,
1110 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1111 if (IS_ERR(va))
1112 return NULL;
1113
1114 addr = va->va_start;
1115 mem = (void *)addr;
1116 }
1117 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1118 vm_unmap_ram(mem, count);
1119 return NULL;
1120 }
1121 return mem;
1122}
1123EXPORT_SYMBOL(vm_map_ram);
1124
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001125static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001126/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001127 * vm_area_add_early - add vmap area early during boot
1128 * @vm: vm_struct to add
1129 *
1130 * This function is used to add fixed kernel vm area to vmlist before
1131 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1132 * should contain proper values and the other fields should be zero.
1133 *
1134 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1135 */
1136void __init vm_area_add_early(struct vm_struct *vm)
1137{
1138 struct vm_struct *tmp, **p;
1139
1140 BUG_ON(vmap_initialized);
1141 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1142 if (tmp->addr >= vm->addr) {
1143 BUG_ON(tmp->addr < vm->addr + vm->size);
1144 break;
1145 } else
1146 BUG_ON(tmp->addr + tmp->size > vm->addr);
1147 }
1148 vm->next = *p;
1149 *p = vm;
1150}
1151
1152/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001153 * vm_area_register_early - register vmap area early during boot
1154 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001155 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001156 *
1157 * This function is used to register kernel vm area before
1158 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1159 * proper values on entry and other fields should be zero. On return,
1160 * vm->addr contains the allocated address.
1161 *
1162 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1163 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001164void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001165{
1166 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001167 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001168
Tejun Heoc0c0a292009-02-24 11:57:21 +09001169 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1170 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1171
1172 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001173
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001174 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001175}
1176
Nick Piggindb64fe02008-10-18 20:27:03 -07001177void __init vmalloc_init(void)
1178{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001179 struct vmap_area *va;
1180 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001181 int i;
1182
1183 for_each_possible_cpu(i) {
1184 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001185 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001186
1187 vbq = &per_cpu(vmap_block_queue, i);
1188 spin_lock_init(&vbq->lock);
1189 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001190 p = &per_cpu(vfree_deferred, i);
1191 init_llist_head(&p->list);
1192 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001193 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001194
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001195 /* Import existing vmlist entries. */
1196 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001197 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001198 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001199 va->va_start = (unsigned long)tmp->addr;
1200 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001201 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001202 __insert_vmap_area(va);
1203 }
Tejun Heoca23e402009-08-14 15:00:52 +09001204
1205 vmap_area_pcpu_hole = VMALLOC_END;
1206
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001207 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001208}
1209
Tejun Heo8fc48982009-02-20 16:29:08 +09001210/**
1211 * map_kernel_range_noflush - map kernel VM area with the specified pages
1212 * @addr: start of the VM area to map
1213 * @size: size of the VM area to map
1214 * @prot: page protection flags to use
1215 * @pages: pages to map
1216 *
1217 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1218 * specify should have been allocated using get_vm_area() and its
1219 * friends.
1220 *
1221 * NOTE:
1222 * This function does NOT do any cache flushing. The caller is
1223 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1224 * before calling this function.
1225 *
1226 * RETURNS:
1227 * The number of pages mapped on success, -errno on failure.
1228 */
1229int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1230 pgprot_t prot, struct page **pages)
1231{
1232 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1233}
1234
1235/**
1236 * unmap_kernel_range_noflush - unmap kernel VM area
1237 * @addr: start of the VM area to unmap
1238 * @size: size of the VM area to unmap
1239 *
1240 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1241 * specify should have been allocated using get_vm_area() and its
1242 * friends.
1243 *
1244 * NOTE:
1245 * This function does NOT do any cache flushing. The caller is
1246 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1247 * before calling this function and flush_tlb_kernel_range() after.
1248 */
1249void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1250{
1251 vunmap_page_range(addr, addr + size);
1252}
Huang Ying81e88fd2011-01-12 14:44:55 +08001253EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001254
1255/**
1256 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1257 * @addr: start of the VM area to unmap
1258 * @size: size of the VM area to unmap
1259 *
1260 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1261 * the unmapping and tlb after.
1262 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001263void unmap_kernel_range(unsigned long addr, unsigned long size)
1264{
1265 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001266
1267 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001268 vunmap_page_range(addr, end);
1269 flush_tlb_kernel_range(addr, end);
1270}
Minchan Kim93ef6d6c2014-06-04 16:11:09 -07001271EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001272
WANG Chaof6f8ed42014-08-06 16:06:58 -07001273int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001274{
1275 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001276 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001277 int err;
1278
WANG Chaof6f8ed42014-08-06 16:06:58 -07001279 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001280
WANG Chaof6f8ed42014-08-06 16:06:58 -07001281 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001282}
1283EXPORT_SYMBOL_GPL(map_vm_area);
1284
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001285static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001286 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001287{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001288 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001289 vm->flags = flags;
1290 vm->addr = (void *)va->va_start;
1291 vm->size = va->va_end - va->va_start;
1292 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001293 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001294 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001295 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001296}
Tejun Heocf88c792009-08-14 15:00:52 +09001297
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001298static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001299{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001300 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001301 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001302 * we should make sure that vm has proper values.
1303 * Pair with smp_rmb() in show_numa_info().
1304 */
1305 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001306 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001307}
1308
Nick Piggindb64fe02008-10-18 20:27:03 -07001309static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001310 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001311 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001312{
Kautuk Consul00065262011-12-19 17:12:04 -08001313 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001314 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001316 BUG_ON(in_interrupt());
Zhang Yanfei0f2d4a82013-07-03 15:04:50 -07001317 if (flags & VM_IOREMAP)
1318 align = 1ul << clamp(fls(size), PAGE_SHIFT, IOREMAP_MAX_ORDER);
Nick Piggindb64fe02008-10-18 20:27:03 -07001319
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001321 if (unlikely(!size))
1322 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Tejun Heocf88c792009-08-14 15:00:52 +09001324 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 if (unlikely(!area))
1326 return NULL;
1327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 /*
1329 * We always allocate a guard page.
1330 */
1331 size += PAGE_SIZE;
1332
Nick Piggindb64fe02008-10-18 20:27:03 -07001333 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1334 if (IS_ERR(va)) {
1335 kfree(area);
1336 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001339 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342}
1343
Christoph Lameter930fc452005-10-29 18:15:41 -07001344struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1345 unsigned long start, unsigned long end)
1346{
David Rientjes00ef2d22013-02-22 16:35:36 -08001347 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1348 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001349}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001350EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001351
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001352struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1353 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001354 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001355{
David Rientjes00ef2d22013-02-22 16:35:36 -08001356 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1357 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001358}
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360/**
Simon Arlott183ff222007-10-20 01:27:18 +02001361 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 * @size: size of the area
1363 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1364 *
1365 * Search an area of @size in the kernel virtual mapping area,
1366 * and reserved it for out purposes. Returns the area descriptor
1367 * on success or %NULL on failure.
1368 */
1369struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1370{
David Miller2dca6992009-09-21 12:22:34 -07001371 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001372 NUMA_NO_NODE, GFP_KERNEL,
1373 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001374}
1375
1376struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001377 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001378{
David Miller2dca6992009-09-21 12:22:34 -07001379 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001380 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381}
1382
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001383/**
1384 * find_vm_area - find a continuous kernel virtual area
1385 * @addr: base address
1386 *
1387 * Search for the kernel VM area starting at @addr, and return it.
1388 * It is up to the caller to do all required locking to keep the returned
1389 * pointer valid.
1390 */
1391struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001392{
Nick Piggindb64fe02008-10-18 20:27:03 -07001393 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001394
Nick Piggindb64fe02008-10-18 20:27:03 -07001395 va = find_vmap_area((unsigned long)addr);
1396 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001397 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001398
Andi Kleen7856dfe2005-05-20 14:27:57 -07001399 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001400}
1401
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402/**
Simon Arlott183ff222007-10-20 01:27:18 +02001403 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 * @addr: base address
1405 *
1406 * Search for the kernel VM area starting at @addr, and remove it.
1407 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001408 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001410struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
Nick Piggindb64fe02008-10-18 20:27:03 -07001412 struct vmap_area *va;
1413
1414 va = find_vmap_area((unsigned long)addr);
1415 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001416 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001417
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001418 spin_lock(&vmap_area_lock);
1419 va->vm = NULL;
1420 va->flags &= ~VM_VM_AREA;
1421 spin_unlock(&vmap_area_lock);
1422
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001423 vmap_debug_free_range(va->va_start, va->va_end);
1424 free_unmap_vmap_area(va);
1425 vm->size -= PAGE_SIZE;
1426
Nick Piggindb64fe02008-10-18 20:27:03 -07001427 return vm;
1428 }
1429 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430}
1431
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001432static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433{
1434 struct vm_struct *area;
1435
1436 if (!addr)
1437 return;
1438
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001439 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001440 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 area = remove_vm_area(addr);
1444 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001445 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 return;
1448 }
1449
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001450 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001451 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 if (deallocate_pages) {
1454 int i;
1455
1456 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001457 struct page *page = area->pages[i];
1458
1459 BUG_ON(!page);
1460 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 }
1462
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001463 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 vfree(area->pages);
1465 else
1466 kfree(area->pages);
1467 }
1468
1469 kfree(area);
1470 return;
1471}
Al Viro32fcfd42013-03-10 20:14:08 -04001472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473/**
1474 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 * @addr: memory base address
1476 *
Simon Arlott183ff222007-10-20 01:27:18 +02001477 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001478 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1479 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 *
Al Viro32fcfd42013-03-10 20:14:08 -04001481 * Must not be called in NMI context (strictly speaking, only if we don't
1482 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1483 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001484 *
1485 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001487void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488{
Al Viro32fcfd42013-03-10 20:14:08 -04001489 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001490
1491 kmemleak_free(addr);
1492
Al Viro32fcfd42013-03-10 20:14:08 -04001493 if (!addr)
1494 return;
1495 if (unlikely(in_interrupt())) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001496 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
Oleg Nesterov59d31322013-07-08 16:00:08 -07001497 if (llist_add((struct llist_node *)addr, &p->list))
1498 schedule_work(&p->wq);
Al Viro32fcfd42013-03-10 20:14:08 -04001499 } else
1500 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502EXPORT_SYMBOL(vfree);
1503
1504/**
1505 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 * @addr: memory base address
1507 *
1508 * Free the virtually contiguous memory area starting at @addr,
1509 * which was created from the page array passed to vmap().
1510 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001511 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001513void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514{
1515 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001516 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001517 if (addr)
1518 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520EXPORT_SYMBOL(vunmap);
1521
1522/**
1523 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 * @pages: array of page pointers
1525 * @count: number of pages to map
1526 * @flags: vm_area->flags
1527 * @prot: page protection for the mapping
1528 *
1529 * Maps @count pages from @pages into contiguous kernel virtual
1530 * space.
1531 */
1532void *vmap(struct page **pages, unsigned int count,
1533 unsigned long flags, pgprot_t prot)
1534{
1535 struct vm_struct *area;
1536
Peter Zijlstra34754b62009-02-25 16:04:03 +01001537 might_sleep();
1538
Jan Beulich44813742009-09-21 17:03:05 -07001539 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 return NULL;
1541
Christoph Lameter23016962008-04-28 02:12:42 -07001542 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1543 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 if (!area)
1545 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001546
WANG Chaof6f8ed42014-08-06 16:06:58 -07001547 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 vunmap(area->addr);
1549 return NULL;
1550 }
1551
1552 return area->addr;
1553}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554EXPORT_SYMBOL(vmap);
1555
David Miller2dca6992009-09-21 12:22:34 -07001556static void *__vmalloc_node(unsigned long size, unsigned long align,
1557 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001558 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001559static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001560 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
Dave Hansen22943ab2011-05-24 17:12:18 -07001562 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 struct page **pages;
1564 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001565 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1566 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Wanpeng Li762216a2013-09-11 14:22:42 -07001568 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 array_size = (nr_pages * sizeof(struct page *));
1570
1571 area->nr_pages = nr_pages;
1572 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001573 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001574 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001575 PAGE_KERNEL, node, area->caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001576 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001577 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001578 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 area->pages = pages;
1581 if (!area->pages) {
1582 remove_vm_area(area->addr);
1583 kfree(area);
1584 return NULL;
1585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001588 struct page *page;
1589
Jianguo Wu4b909512013-11-12 15:07:11 -08001590 if (node == NUMA_NO_NODE)
David Rientjes930f0362014-08-06 16:06:28 -07001591 page = alloc_page(alloc_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001592 else
David Rientjes930f0362014-08-06 16:06:28 -07001593 page = alloc_pages_node(node, alloc_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001594
1595 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 /* Successfully allocated i pages, free them in __vunmap() */
1597 area->nr_pages = i;
1598 goto fail;
1599 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001600 area->pages[i] = page;
Eric Dumazet660654f2014-08-06 16:06:25 -07001601 if (gfp_mask & __GFP_WAIT)
1602 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 }
1604
WANG Chaof6f8ed42014-08-06 16:06:58 -07001605 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 goto fail;
1607 return area->addr;
1608
1609fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001610 warn_alloc_failed(gfp_mask, order,
1611 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001612 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 vfree(area->addr);
1614 return NULL;
1615}
1616
David Rientjesd0a21262011-01-13 15:46:02 -08001617/**
1618 * __vmalloc_node_range - allocate virtually contiguous memory
1619 * @size: allocation size
1620 * @align: desired alignment
1621 * @start: vm area range start
1622 * @end: vm area range end
1623 * @gfp_mask: flags for the page level allocator
1624 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001625 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001626 * @caller: caller's return address
1627 *
1628 * Allocate enough pages to cover @size from the page level
1629 * allocator with @gfp_mask flags. Map them into contiguous
1630 * kernel virtual space, using a pagetable protection of @prot.
1631 */
1632void *__vmalloc_node_range(unsigned long size, unsigned long align,
1633 unsigned long start, unsigned long end, gfp_t gfp_mask,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001634 pgprot_t prot, int node, const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001635{
David Rientjesd0a21262011-01-13 15:46:02 -08001636 struct vm_struct *area;
1637 void *addr;
1638 unsigned long real_size = size;
1639
1640 size = PAGE_ALIGN(size);
1641 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001642 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001643
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001644 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED,
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001645 start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001646 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001647 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001648
Wanpeng Li3722e132013-11-12 15:07:29 -08001649 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001650 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001651 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001652
1653 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001654 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1655 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001656 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001657 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001658 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001659
1660 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001661 * A ref_count = 2 is needed because vm_struct allocated in
1662 * __get_vm_area_node() contains a reference to the virtual address of
1663 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001664 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001665 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001666
1667 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001668
1669fail:
1670 warn_alloc_failed(gfp_mask, 0,
1671 "vmalloc: allocation failure: %lu bytes\n",
1672 real_size);
1673 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001674}
1675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001677 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001679 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 * @gfp_mask: flags for the page level allocator
1681 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001682 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001683 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 *
1685 * Allocate enough pages to cover @size from the page level
1686 * allocator with @gfp_mask flags. Map them into contiguous
1687 * kernel virtual space, using a pagetable protection of @prot.
1688 */
David Miller2dca6992009-09-21 12:22:34 -07001689static void *__vmalloc_node(unsigned long size, unsigned long align,
1690 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001691 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
David Rientjesd0a21262011-01-13 15:46:02 -08001693 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1694 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695}
1696
Christoph Lameter930fc452005-10-29 18:15:41 -07001697void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1698{
David Rientjes00ef2d22013-02-22 16:35:36 -08001699 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001700 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001701}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702EXPORT_SYMBOL(__vmalloc);
1703
Dave Younge1ca7782010-10-26 14:22:06 -07001704static inline void *__vmalloc_node_flags(unsigned long size,
1705 int node, gfp_t flags)
1706{
1707 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1708 node, __builtin_return_address(0));
1709}
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711/**
1712 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 * Allocate enough pages to cover @size from the page level
1715 * allocator and map them into contiguous kernel virtual space.
1716 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001717 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 * use __vmalloc() instead.
1719 */
1720void *vmalloc(unsigned long size)
1721{
David Rientjes00ef2d22013-02-22 16:35:36 -08001722 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1723 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725EXPORT_SYMBOL(vmalloc);
1726
Christoph Lameter930fc452005-10-29 18:15:41 -07001727/**
Dave Younge1ca7782010-10-26 14:22:06 -07001728 * vzalloc - allocate virtually contiguous memory with zero fill
1729 * @size: allocation size
1730 * Allocate enough pages to cover @size from the page level
1731 * allocator and map them into contiguous kernel virtual space.
1732 * The memory allocated is set to zero.
1733 *
1734 * For tight control over page level allocator and protection flags
1735 * use __vmalloc() instead.
1736 */
1737void *vzalloc(unsigned long size)
1738{
David Rientjes00ef2d22013-02-22 16:35:36 -08001739 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001740 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1741}
1742EXPORT_SYMBOL(vzalloc);
1743
1744/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001745 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1746 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001747 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001748 * The resulting memory area is zeroed so it can be mapped to userspace
1749 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001750 */
1751void *vmalloc_user(unsigned long size)
1752{
1753 struct vm_struct *area;
1754 void *ret;
1755
David Miller2dca6992009-09-21 12:22:34 -07001756 ret = __vmalloc_node(size, SHMLBA,
1757 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001758 PAGE_KERNEL, NUMA_NO_NODE,
1759 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001760 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001761 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001762 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001763 }
Nick Piggin83342312006-06-23 02:03:20 -07001764 return ret;
1765}
1766EXPORT_SYMBOL(vmalloc_user);
1767
1768/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001769 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001770 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001771 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001772 *
1773 * Allocate enough pages to cover @size from the page level
1774 * allocator and map them into contiguous kernel virtual space.
1775 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001776 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001777 * use __vmalloc() instead.
1778 */
1779void *vmalloc_node(unsigned long size, int node)
1780{
David Miller2dca6992009-09-21 12:22:34 -07001781 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001782 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001783}
1784EXPORT_SYMBOL(vmalloc_node);
1785
Dave Younge1ca7782010-10-26 14:22:06 -07001786/**
1787 * vzalloc_node - allocate memory on a specific node with zero fill
1788 * @size: allocation size
1789 * @node: numa node
1790 *
1791 * Allocate enough pages to cover @size from the page level
1792 * allocator and map them into contiguous kernel virtual space.
1793 * The memory allocated is set to zero.
1794 *
1795 * For tight control over page level allocator and protection flags
1796 * use __vmalloc_node() instead.
1797 */
1798void *vzalloc_node(unsigned long size, int node)
1799{
1800 return __vmalloc_node_flags(size, node,
1801 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1802}
1803EXPORT_SYMBOL(vzalloc_node);
1804
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001805#ifndef PAGE_KERNEL_EXEC
1806# define PAGE_KERNEL_EXEC PAGE_KERNEL
1807#endif
1808
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809/**
1810 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 * @size: allocation size
1812 *
1813 * Kernel-internal function to allocate enough pages to cover @size
1814 * the page level allocator and map them into contiguous and
1815 * executable kernel virtual space.
1816 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001817 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 * use __vmalloc() instead.
1819 */
1820
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821void *vmalloc_exec(unsigned long size)
1822{
David Miller2dca6992009-09-21 12:22:34 -07001823 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001824 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825}
1826
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001827#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001828#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001829#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001830#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001831#else
1832#define GFP_VMALLOC32 GFP_KERNEL
1833#endif
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835/**
1836 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 * @size: allocation size
1838 *
1839 * Allocate enough 32bit PA addressable pages to cover @size from the
1840 * page level allocator and map them into contiguous kernel virtual space.
1841 */
1842void *vmalloc_32(unsigned long size)
1843{
David Miller2dca6992009-09-21 12:22:34 -07001844 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001845 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847EXPORT_SYMBOL(vmalloc_32);
1848
Nick Piggin83342312006-06-23 02:03:20 -07001849/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001850 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001851 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001852 *
1853 * The resulting memory area is 32bit addressable and zeroed so it can be
1854 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001855 */
1856void *vmalloc_32_user(unsigned long size)
1857{
1858 struct vm_struct *area;
1859 void *ret;
1860
David Miller2dca6992009-09-21 12:22:34 -07001861 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001862 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001863 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001864 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001865 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001866 }
Nick Piggin83342312006-06-23 02:03:20 -07001867 return ret;
1868}
1869EXPORT_SYMBOL(vmalloc_32_user);
1870
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001871/*
1872 * small helper routine , copy contents to buf from addr.
1873 * If the page is not present, fill zero.
1874 */
1875
1876static int aligned_vread(char *buf, char *addr, unsigned long count)
1877{
1878 struct page *p;
1879 int copied = 0;
1880
1881 while (count) {
1882 unsigned long offset, length;
1883
1884 offset = (unsigned long)addr & ~PAGE_MASK;
1885 length = PAGE_SIZE - offset;
1886 if (length > count)
1887 length = count;
1888 p = vmalloc_to_page(addr);
1889 /*
1890 * To do safe access to this _mapped_ area, we need
1891 * lock. But adding lock here means that we need to add
1892 * overhead of vmalloc()/vfree() calles for this _debug_
1893 * interface, rarely used. Instead of that, we'll use
1894 * kmap() and get small overhead in this access function.
1895 */
1896 if (p) {
1897 /*
1898 * we can expect USER0 is not used (see vread/vwrite's
1899 * function description)
1900 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001901 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001902 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001903 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001904 } else
1905 memset(buf, 0, length);
1906
1907 addr += length;
1908 buf += length;
1909 copied += length;
1910 count -= length;
1911 }
1912 return copied;
1913}
1914
1915static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1916{
1917 struct page *p;
1918 int copied = 0;
1919
1920 while (count) {
1921 unsigned long offset, length;
1922
1923 offset = (unsigned long)addr & ~PAGE_MASK;
1924 length = PAGE_SIZE - offset;
1925 if (length > count)
1926 length = count;
1927 p = vmalloc_to_page(addr);
1928 /*
1929 * To do safe access to this _mapped_ area, we need
1930 * lock. But adding lock here means that we need to add
1931 * overhead of vmalloc()/vfree() calles for this _debug_
1932 * interface, rarely used. Instead of that, we'll use
1933 * kmap() and get small overhead in this access function.
1934 */
1935 if (p) {
1936 /*
1937 * we can expect USER0 is not used (see vread/vwrite's
1938 * function description)
1939 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001940 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001941 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001942 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001943 }
1944 addr += length;
1945 buf += length;
1946 copied += length;
1947 count -= length;
1948 }
1949 return copied;
1950}
1951
1952/**
1953 * vread() - read vmalloc area in a safe way.
1954 * @buf: buffer for reading data
1955 * @addr: vm address.
1956 * @count: number of bytes to be read.
1957 *
1958 * Returns # of bytes which addr and buf should be increased.
1959 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1960 * includes any intersect with alive vmalloc area.
1961 *
1962 * This function checks that addr is a valid vmalloc'ed area, and
1963 * copy data from that area to a given buffer. If the given memory range
1964 * of [addr...addr+count) includes some valid address, data is copied to
1965 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1966 * IOREMAP area is treated as memory hole and no copy is done.
1967 *
1968 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08001969 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001970 *
1971 * Note: In usual ops, vread() is never necessary because the caller
1972 * should know vmalloc() area is valid and can use memcpy().
1973 * This is for routines which have to access vmalloc area without
1974 * any informaion, as /dev/kmem.
1975 *
1976 */
1977
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978long vread(char *buf, char *addr, unsigned long count)
1979{
Joonsoo Kime81ce852013-04-29 15:07:32 -07001980 struct vmap_area *va;
1981 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001983 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 unsigned long n;
1985
1986 /* Don't allow overflow */
1987 if ((unsigned long) addr + count < count)
1988 count = -(unsigned long) addr;
1989
Joonsoo Kime81ce852013-04-29 15:07:32 -07001990 spin_lock(&vmap_area_lock);
1991 list_for_each_entry(va, &vmap_area_list, list) {
1992 if (!count)
1993 break;
1994
1995 if (!(va->flags & VM_VM_AREA))
1996 continue;
1997
1998 vm = va->vm;
1999 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002000 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 continue;
2002 while (addr < vaddr) {
2003 if (count == 0)
2004 goto finished;
2005 *buf = '\0';
2006 buf++;
2007 addr++;
2008 count--;
2009 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002010 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002011 if (n > count)
2012 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002013 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002014 aligned_vread(buf, addr, n);
2015 else /* IOREMAP area is treated as memory hole */
2016 memset(buf, 0, n);
2017 buf += n;
2018 addr += n;
2019 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 }
2021finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002022 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002023
2024 if (buf == buf_start)
2025 return 0;
2026 /* zero-fill memory holes */
2027 if (buf != buf_start + buflen)
2028 memset(buf, 0, buflen - (buf - buf_start));
2029
2030 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031}
2032
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002033/**
2034 * vwrite() - write vmalloc area in a safe way.
2035 * @buf: buffer for source data
2036 * @addr: vm address.
2037 * @count: number of bytes to be read.
2038 *
2039 * Returns # of bytes which addr and buf should be incresed.
2040 * (same number to @count).
2041 * If [addr...addr+count) doesn't includes any intersect with valid
2042 * vmalloc area, returns 0.
2043 *
2044 * This function checks that addr is a valid vmalloc'ed area, and
2045 * copy data from a buffer to the given addr. If specified range of
2046 * [addr...addr+count) includes some valid address, data is copied from
2047 * proper area of @buf. If there are memory holes, no copy to hole.
2048 * IOREMAP area is treated as memory hole and no copy is done.
2049 *
2050 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002051 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002052 *
2053 * Note: In usual ops, vwrite() is never necessary because the caller
2054 * should know vmalloc() area is valid and can use memcpy().
2055 * This is for routines which have to access vmalloc area without
2056 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002057 */
2058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059long vwrite(char *buf, char *addr, unsigned long count)
2060{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002061 struct vmap_area *va;
2062 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002063 char *vaddr;
2064 unsigned long n, buflen;
2065 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
2067 /* Don't allow overflow */
2068 if ((unsigned long) addr + count < count)
2069 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002070 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
Joonsoo Kime81ce852013-04-29 15:07:32 -07002072 spin_lock(&vmap_area_lock);
2073 list_for_each_entry(va, &vmap_area_list, list) {
2074 if (!count)
2075 break;
2076
2077 if (!(va->flags & VM_VM_AREA))
2078 continue;
2079
2080 vm = va->vm;
2081 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002082 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 continue;
2084 while (addr < vaddr) {
2085 if (count == 0)
2086 goto finished;
2087 buf++;
2088 addr++;
2089 count--;
2090 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002091 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002092 if (n > count)
2093 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002094 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002095 aligned_vwrite(buf, addr, n);
2096 copied++;
2097 }
2098 buf += n;
2099 addr += n;
2100 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 }
2102finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002103 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002104 if (!copied)
2105 return 0;
2106 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
Nick Piggin83342312006-06-23 02:03:20 -07002108
2109/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002110 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2111 * @vma: vma to cover
2112 * @uaddr: target user address to start at
2113 * @kaddr: virtual address of vmalloc kernel memory
2114 * @size: size of map area
2115 *
2116 * Returns: 0 for success, -Exxx on failure
2117 *
2118 * This function checks that @kaddr is a valid vmalloc'ed area,
2119 * and that it is big enough to cover the range starting at
2120 * @uaddr in @vma. Will return failure if that criteria isn't
2121 * met.
2122 *
2123 * Similar to remap_pfn_range() (see mm/memory.c)
2124 */
2125int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2126 void *kaddr, unsigned long size)
2127{
2128 struct vm_struct *area;
2129
2130 size = PAGE_ALIGN(size);
2131
2132 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2133 return -EINVAL;
2134
2135 area = find_vm_area(kaddr);
2136 if (!area)
2137 return -EINVAL;
2138
2139 if (!(area->flags & VM_USERMAP))
2140 return -EINVAL;
2141
2142 if (kaddr + size > area->addr + area->size)
2143 return -EINVAL;
2144
2145 do {
2146 struct page *page = vmalloc_to_page(kaddr);
2147 int ret;
2148
2149 ret = vm_insert_page(vma, uaddr, page);
2150 if (ret)
2151 return ret;
2152
2153 uaddr += PAGE_SIZE;
2154 kaddr += PAGE_SIZE;
2155 size -= PAGE_SIZE;
2156 } while (size > 0);
2157
2158 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2159
2160 return 0;
2161}
2162EXPORT_SYMBOL(remap_vmalloc_range_partial);
2163
2164/**
Nick Piggin83342312006-06-23 02:03:20 -07002165 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002166 * @vma: vma to cover (map full range of vma)
2167 * @addr: vmalloc memory
2168 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002169 *
2170 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002171 *
2172 * This function checks that addr is a valid vmalloc'ed area, and
2173 * that it is big enough to cover the vma. Will return failure if
2174 * that criteria isn't met.
2175 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002176 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002177 */
2178int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2179 unsigned long pgoff)
2180{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002181 return remap_vmalloc_range_partial(vma, vma->vm_start,
2182 addr + (pgoff << PAGE_SHIFT),
2183 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002184}
2185EXPORT_SYMBOL(remap_vmalloc_range);
2186
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002187/*
2188 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2189 * have one.
2190 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002191void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002192{
2193}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002194
2195
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002196static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002197{
David Vrabelcd129092011-09-29 16:53:32 +01002198 pte_t ***p = data;
2199
2200 if (p) {
2201 *(*p) = pte;
2202 (*p)++;
2203 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002204 return 0;
2205}
2206
2207/**
2208 * alloc_vm_area - allocate a range of kernel address space
2209 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002210 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002211 *
2212 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002213 *
2214 * This function reserves a range of kernel address space, and
2215 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002216 * are created.
2217 *
2218 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2219 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002220 */
David Vrabelcd129092011-09-29 16:53:32 +01002221struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002222{
2223 struct vm_struct *area;
2224
Christoph Lameter23016962008-04-28 02:12:42 -07002225 area = get_vm_area_caller(size, VM_IOREMAP,
2226 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002227 if (area == NULL)
2228 return NULL;
2229
2230 /*
2231 * This ensures that page tables are constructed for this region
2232 * of kernel virtual address space and mapped into init_mm.
2233 */
2234 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002235 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002236 free_vm_area(area);
2237 return NULL;
2238 }
2239
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002240 return area;
2241}
2242EXPORT_SYMBOL_GPL(alloc_vm_area);
2243
2244void free_vm_area(struct vm_struct *area)
2245{
2246 struct vm_struct *ret;
2247 ret = remove_vm_area(area->addr);
2248 BUG_ON(ret != area);
2249 kfree(area);
2250}
2251EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002252
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002253#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002254static struct vmap_area *node_to_va(struct rb_node *n)
2255{
2256 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2257}
2258
2259/**
2260 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2261 * @end: target address
2262 * @pnext: out arg for the next vmap_area
2263 * @pprev: out arg for the previous vmap_area
2264 *
2265 * Returns: %true if either or both of next and prev are found,
2266 * %false if no vmap_area exists
2267 *
2268 * Find vmap_areas end addresses of which enclose @end. ie. if not
2269 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2270 */
2271static bool pvm_find_next_prev(unsigned long end,
2272 struct vmap_area **pnext,
2273 struct vmap_area **pprev)
2274{
2275 struct rb_node *n = vmap_area_root.rb_node;
2276 struct vmap_area *va = NULL;
2277
2278 while (n) {
2279 va = rb_entry(n, struct vmap_area, rb_node);
2280 if (end < va->va_end)
2281 n = n->rb_left;
2282 else if (end > va->va_end)
2283 n = n->rb_right;
2284 else
2285 break;
2286 }
2287
2288 if (!va)
2289 return false;
2290
2291 if (va->va_end > end) {
2292 *pnext = va;
2293 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2294 } else {
2295 *pprev = va;
2296 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2297 }
2298 return true;
2299}
2300
2301/**
2302 * pvm_determine_end - find the highest aligned address between two vmap_areas
2303 * @pnext: in/out arg for the next vmap_area
2304 * @pprev: in/out arg for the previous vmap_area
2305 * @align: alignment
2306 *
2307 * Returns: determined end address
2308 *
2309 * Find the highest aligned address between *@pnext and *@pprev below
2310 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2311 * down address is between the end addresses of the two vmap_areas.
2312 *
2313 * Please note that the address returned by this function may fall
2314 * inside *@pnext vmap_area. The caller is responsible for checking
2315 * that.
2316 */
2317static unsigned long pvm_determine_end(struct vmap_area **pnext,
2318 struct vmap_area **pprev,
2319 unsigned long align)
2320{
2321 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2322 unsigned long addr;
2323
2324 if (*pnext)
2325 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2326 else
2327 addr = vmalloc_end;
2328
2329 while (*pprev && (*pprev)->va_end > addr) {
2330 *pnext = *pprev;
2331 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2332 }
2333
2334 return addr;
2335}
2336
2337/**
2338 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2339 * @offsets: array containing offset of each area
2340 * @sizes: array containing size of each area
2341 * @nr_vms: the number of areas to allocate
2342 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002343 *
2344 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2345 * vm_structs on success, %NULL on failure
2346 *
2347 * Percpu allocator wants to use congruent vm areas so that it can
2348 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002349 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2350 * be scattered pretty far, distance between two areas easily going up
2351 * to gigabytes. To avoid interacting with regular vmallocs, these
2352 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002353 *
2354 * Despite its complicated look, this allocator is rather simple. It
2355 * does everything top-down and scans areas from the end looking for
2356 * matching slot. While scanning, if any of the areas overlaps with
2357 * existing vmap_area, the base address is pulled down to fit the
2358 * area. Scanning is repeated till all the areas fit and then all
2359 * necessary data structres are inserted and the result is returned.
2360 */
2361struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2362 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002363 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002364{
2365 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2366 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2367 struct vmap_area **vas, *prev, *next;
2368 struct vm_struct **vms;
2369 int area, area2, last_area, term_area;
2370 unsigned long base, start, end, last_end;
2371 bool purged = false;
2372
Tejun Heoca23e402009-08-14 15:00:52 +09002373 /* verify parameters and allocate data structures */
2374 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2375 for (last_area = 0, area = 0; area < nr_vms; area++) {
2376 start = offsets[area];
2377 end = start + sizes[area];
2378
2379 /* is everything aligned properly? */
2380 BUG_ON(!IS_ALIGNED(offsets[area], align));
2381 BUG_ON(!IS_ALIGNED(sizes[area], align));
2382
2383 /* detect the area with the highest address */
2384 if (start > offsets[last_area])
2385 last_area = area;
2386
2387 for (area2 = 0; area2 < nr_vms; area2++) {
2388 unsigned long start2 = offsets[area2];
2389 unsigned long end2 = start2 + sizes[area2];
2390
2391 if (area2 == area)
2392 continue;
2393
2394 BUG_ON(start2 >= start && start2 < end);
2395 BUG_ON(end2 <= end && end2 > start);
2396 }
2397 }
2398 last_end = offsets[last_area] + sizes[last_area];
2399
2400 if (vmalloc_end - vmalloc_start < last_end) {
2401 WARN_ON(true);
2402 return NULL;
2403 }
2404
Thomas Meyer4d67d862012-05-29 15:06:21 -07002405 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2406 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002407 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002408 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002409
2410 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002411 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2412 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002413 if (!vas[area] || !vms[area])
2414 goto err_free;
2415 }
2416retry:
2417 spin_lock(&vmap_area_lock);
2418
2419 /* start scanning - we scan from the top, begin with the last area */
2420 area = term_area = last_area;
2421 start = offsets[area];
2422 end = start + sizes[area];
2423
2424 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2425 base = vmalloc_end - last_end;
2426 goto found;
2427 }
2428 base = pvm_determine_end(&next, &prev, align) - end;
2429
2430 while (true) {
2431 BUG_ON(next && next->va_end <= base + end);
2432 BUG_ON(prev && prev->va_end > base + end);
2433
2434 /*
2435 * base might have underflowed, add last_end before
2436 * comparing.
2437 */
2438 if (base + last_end < vmalloc_start + last_end) {
2439 spin_unlock(&vmap_area_lock);
2440 if (!purged) {
2441 purge_vmap_area_lazy();
2442 purged = true;
2443 goto retry;
2444 }
2445 goto err_free;
2446 }
2447
2448 /*
2449 * If next overlaps, move base downwards so that it's
2450 * right below next and then recheck.
2451 */
2452 if (next && next->va_start < base + end) {
2453 base = pvm_determine_end(&next, &prev, align) - end;
2454 term_area = area;
2455 continue;
2456 }
2457
2458 /*
2459 * If prev overlaps, shift down next and prev and move
2460 * base so that it's right below new next and then
2461 * recheck.
2462 */
2463 if (prev && prev->va_end > base + start) {
2464 next = prev;
2465 prev = node_to_va(rb_prev(&next->rb_node));
2466 base = pvm_determine_end(&next, &prev, align) - end;
2467 term_area = area;
2468 continue;
2469 }
2470
2471 /*
2472 * This area fits, move on to the previous one. If
2473 * the previous one is the terminal one, we're done.
2474 */
2475 area = (area + nr_vms - 1) % nr_vms;
2476 if (area == term_area)
2477 break;
2478 start = offsets[area];
2479 end = start + sizes[area];
2480 pvm_find_next_prev(base + end, &next, &prev);
2481 }
2482found:
2483 /* we've found a fitting base, insert all va's */
2484 for (area = 0; area < nr_vms; area++) {
2485 struct vmap_area *va = vas[area];
2486
2487 va->va_start = base + offsets[area];
2488 va->va_end = va->va_start + sizes[area];
2489 __insert_vmap_area(va);
2490 }
2491
2492 vmap_area_pcpu_hole = base + offsets[last_area];
2493
2494 spin_unlock(&vmap_area_lock);
2495
2496 /* insert all vm's */
2497 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002498 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2499 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002500
2501 kfree(vas);
2502 return vms;
2503
2504err_free:
2505 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002506 kfree(vas[area]);
2507 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002508 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002509err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002510 kfree(vas);
2511 kfree(vms);
2512 return NULL;
2513}
2514
2515/**
2516 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2517 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2518 * @nr_vms: the number of allocated areas
2519 *
2520 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2521 */
2522void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2523{
2524 int i;
2525
2526 for (i = 0; i < nr_vms; i++)
2527 free_vm_area(vms[i]);
2528 kfree(vms);
2529}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002530#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002531
2532#ifdef CONFIG_PROC_FS
2533static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002534 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002535{
2536 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002537 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002538
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002539 spin_lock(&vmap_area_lock);
2540 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2541 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002542 n--;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002543 va = list_entry(va->list.next, typeof(*va), list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002544 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002545 if (!n && &va->list != &vmap_area_list)
2546 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002547
2548 return NULL;
2549
2550}
2551
2552static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2553{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002554 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002555
2556 ++*pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002557 next = list_entry(va->list.next, typeof(*va), list);
2558 if (&next->list != &vmap_area_list)
2559 return next;
2560
2561 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002562}
2563
2564static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002565 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002566{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002567 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002568}
2569
Eric Dumazeta47a1262008-07-23 21:27:38 -07002570static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2571{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002572 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002573 unsigned int nr, *counters = m->private;
2574
2575 if (!counters)
2576 return;
2577
Wanpeng Liaf123462013-11-12 15:07:32 -08002578 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2579 smp_rmb();
2580 if (v->flags & VM_UNINITIALIZED)
2581 return;
2582
Eric Dumazeta47a1262008-07-23 21:27:38 -07002583 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2584
2585 for (nr = 0; nr < v->nr_pages; nr++)
2586 counters[page_to_nid(v->pages[nr])]++;
2587
2588 for_each_node_state(nr, N_HIGH_MEMORY)
2589 if (counters[nr])
2590 seq_printf(m, " N%u=%u", nr, counters[nr]);
2591 }
2592}
2593
Christoph Lametera10aa572008-04-28 02:12:40 -07002594static int s_show(struct seq_file *m, void *p)
2595{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002596 struct vmap_area *va = p;
2597 struct vm_struct *v;
2598
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002599 /*
2600 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2601 * behalf of vmap area is being tear down or vm_map_ram allocation.
2602 */
2603 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002604 return 0;
2605
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002606 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002607
Kees Cook45ec1692012-10-08 16:34:09 -07002608 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002609 v->addr, v->addr + v->size, v->size);
2610
Joe Perches62c70bc2011-01-13 15:45:52 -08002611 if (v->caller)
2612 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002613
Christoph Lametera10aa572008-04-28 02:12:40 -07002614 if (v->nr_pages)
2615 seq_printf(m, " pages=%d", v->nr_pages);
2616
2617 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002618 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002619
2620 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002621 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002622
2623 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002624 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002625
2626 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002627 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002628
2629 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002630 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002631
2632 if (v->flags & VM_VPAGES)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002633 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002634
Eric Dumazeta47a1262008-07-23 21:27:38 -07002635 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002636 seq_putc(m, '\n');
2637 return 0;
2638}
2639
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002640static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002641 .start = s_start,
2642 .next = s_next,
2643 .stop = s_stop,
2644 .show = s_show,
2645};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002646
2647static int vmalloc_open(struct inode *inode, struct file *file)
2648{
Rob Jones703394c2014-10-09 15:28:01 -07002649 if (IS_ENABLED(CONFIG_NUMA))
2650 return seq_open_private(file, &vmalloc_op,
2651 nr_node_ids * sizeof(unsigned int));
2652 else
2653 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002654}
2655
2656static const struct file_operations proc_vmalloc_operations = {
2657 .open = vmalloc_open,
2658 .read = seq_read,
2659 .llseek = seq_lseek,
2660 .release = seq_release_private,
2661};
2662
2663static int __init proc_vmalloc_init(void)
2664{
2665 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2666 return 0;
2667}
2668module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002669
2670void get_vmalloc_info(struct vmalloc_info *vmi)
2671{
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002672 struct vmap_area *va;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002673 unsigned long free_area_size;
2674 unsigned long prev_end;
2675
2676 vmi->used = 0;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002677 vmi->largest_chunk = 0;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002678
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002679 prev_end = VMALLOC_START;
2680
Joonsoo Kim474750a2014-08-06 16:05:06 -07002681 rcu_read_lock();
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002682
2683 if (list_empty(&vmap_area_list)) {
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002684 vmi->largest_chunk = VMALLOC_TOTAL;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002685 goto out;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002686 }
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002687
Joonsoo Kim474750a2014-08-06 16:05:06 -07002688 list_for_each_entry_rcu(va, &vmap_area_list, list) {
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002689 unsigned long addr = va->va_start;
2690
2691 /*
2692 * Some archs keep another range for modules in vmalloc space
2693 */
2694 if (addr < VMALLOC_START)
2695 continue;
2696 if (addr >= VMALLOC_END)
2697 break;
2698
2699 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2700 continue;
2701
2702 vmi->used += (va->va_end - va->va_start);
2703
2704 free_area_size = addr - prev_end;
2705 if (vmi->largest_chunk < free_area_size)
2706 vmi->largest_chunk = free_area_size;
2707
2708 prev_end = va->va_end;
2709 }
2710
2711 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2712 vmi->largest_chunk = VMALLOC_END - prev_end;
2713
2714out:
Joonsoo Kim474750a2014-08-06 16:05:06 -07002715 rcu_read_unlock();
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002716}
Christoph Lametera10aa572008-04-28 02:12:40 -07002717#endif
2718