blob: 35b25e1340ca49cf0a95bbc80c69490ec68b5554 [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())
Pintu Kumar0cbc8532014-12-10 15:42:56 -0800466 pr_warn("vmap allocation for size %lu failed: "
Nick Piggin89699602011-03-22 16:30:36 -0700467 "use vmalloc=<size> to increase size.\n", size);
468 kfree(va);
469 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700470}
471
Nick Piggindb64fe02008-10-18 20:27:03 -0700472static void __free_vmap_area(struct vmap_area *va)
473{
474 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700475
476 if (free_vmap_cache) {
477 if (va->va_end < cached_vstart) {
478 free_vmap_cache = NULL;
479 } else {
480 struct vmap_area *cache;
481 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
482 if (va->va_start <= cache->va_start) {
483 free_vmap_cache = rb_prev(&va->rb_node);
484 /*
485 * We don't try to update cached_hole_size or
486 * cached_align, but it won't go very wrong.
487 */
488 }
489 }
490 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700491 rb_erase(&va->rb_node, &vmap_area_root);
492 RB_CLEAR_NODE(&va->rb_node);
493 list_del_rcu(&va->list);
494
Tejun Heoca23e402009-08-14 15:00:52 +0900495 /*
496 * Track the highest possible candidate for pcpu area
497 * allocation. Areas outside of vmalloc area can be returned
498 * here too, consider only end addresses which fall inside
499 * vmalloc area proper.
500 */
501 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
502 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
503
Lai Jiangshan14769de2011-03-18 12:12:19 +0800504 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700505}
506
507/*
508 * Free a region of KVA allocated by alloc_vmap_area
509 */
510static void free_vmap_area(struct vmap_area *va)
511{
512 spin_lock(&vmap_area_lock);
513 __free_vmap_area(va);
514 spin_unlock(&vmap_area_lock);
515}
516
517/*
518 * Clear the pagetable entries of a given vmap_area
519 */
520static void unmap_vmap_area(struct vmap_area *va)
521{
522 vunmap_page_range(va->va_start, va->va_end);
523}
524
Nick Piggincd528582009-01-06 14:39:20 -0800525static void vmap_debug_free_range(unsigned long start, unsigned long end)
526{
527 /*
528 * Unmap page tables and force a TLB flush immediately if
529 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
530 * bugs similarly to those in linear kernel virtual address
531 * space after a page has been freed.
532 *
533 * All the lazy freeing logic is still retained, in order to
534 * minimise intrusiveness of this debugging feature.
535 *
536 * This is going to be *slow* (linear kernel virtual address
537 * debugging doesn't do a broadcast TLB flush so it is a lot
538 * faster).
539 */
540#ifdef CONFIG_DEBUG_PAGEALLOC
541 vunmap_page_range(start, end);
542 flush_tlb_kernel_range(start, end);
543#endif
544}
545
Nick Piggindb64fe02008-10-18 20:27:03 -0700546/*
547 * lazy_max_pages is the maximum amount of virtual address space we gather up
548 * before attempting to purge with a TLB flush.
549 *
550 * There is a tradeoff here: a larger number will cover more kernel page tables
551 * and take slightly longer to purge, but it will linearly reduce the number of
552 * global TLB flushes that must be performed. It would seem natural to scale
553 * this number up linearly with the number of CPUs (because vmapping activity
554 * could also scale linearly with the number of CPUs), however it is likely
555 * that in practice, workloads might be constrained in other ways that mean
556 * vmap activity will not scale linearly with CPUs. Also, I want to be
557 * conservative and not introduce a big latency on huge systems, so go with
558 * a less aggressive log scale. It will still be an improvement over the old
559 * code, and it will be simple to change the scale factor if we find that it
560 * becomes a problem on bigger systems.
561 */
562static unsigned long lazy_max_pages(void)
563{
564 unsigned int log;
565
566 log = fls(num_online_cpus());
567
568 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
569}
570
571static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
572
Nick Piggin02b709d2010-02-01 22:25:57 +1100573/* for per-CPU blocks */
574static void purge_fragmented_blocks_allcpus(void);
575
Nick Piggindb64fe02008-10-18 20:27:03 -0700576/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500577 * called before a call to iounmap() if the caller wants vm_area_struct's
578 * immediately freed.
579 */
580void set_iounmap_nonlazy(void)
581{
582 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
583}
584
585/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700586 * Purges all lazily-freed vmap areas.
587 *
588 * If sync is 0 then don't purge if there is already a purge in progress.
589 * If force_flush is 1, then flush kernel TLBs between *start and *end even
590 * if we found no lazy vmap areas to unmap (callers can use this to optimise
591 * their own TLB flushing).
592 * Returns with *start = min(*start, lowest purged address)
593 * *end = max(*end, highest purged address)
594 */
595static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
596 int sync, int force_flush)
597{
Andrew Morton46666d82009-01-15 13:51:15 -0800598 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700599 LIST_HEAD(valist);
600 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800601 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700602 int nr = 0;
603
604 /*
605 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
606 * should not expect such behaviour. This just simplifies locking for
607 * the case that isn't actually used at the moment anyway.
608 */
609 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800610 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700611 return;
612 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800613 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700614
Nick Piggin02b709d2010-02-01 22:25:57 +1100615 if (sync)
616 purge_fragmented_blocks_allcpus();
617
Nick Piggindb64fe02008-10-18 20:27:03 -0700618 rcu_read_lock();
619 list_for_each_entry_rcu(va, &vmap_area_list, list) {
620 if (va->flags & VM_LAZY_FREE) {
621 if (va->va_start < *start)
622 *start = va->va_start;
623 if (va->va_end > *end)
624 *end = va->va_end;
625 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700626 list_add_tail(&va->purge_list, &valist);
627 va->flags |= VM_LAZY_FREEING;
628 va->flags &= ~VM_LAZY_FREE;
629 }
630 }
631 rcu_read_unlock();
632
Yongseok Koh88f50042010-01-19 17:33:49 +0900633 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700634 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700635
636 if (nr || force_flush)
637 flush_tlb_kernel_range(*start, *end);
638
639 if (nr) {
640 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800641 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700642 __free_vmap_area(va);
643 spin_unlock(&vmap_area_lock);
644 }
Andrew Morton46666d82009-01-15 13:51:15 -0800645 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700646}
647
648/*
Nick Piggin496850e2008-11-19 15:36:33 -0800649 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
650 * is already purging.
651 */
652static void try_purge_vmap_area_lazy(void)
653{
654 unsigned long start = ULONG_MAX, end = 0;
655
656 __purge_vmap_area_lazy(&start, &end, 0, 0);
657}
658
659/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700660 * Kick off a purge of the outstanding lazy areas.
661 */
662static void purge_vmap_area_lazy(void)
663{
664 unsigned long start = ULONG_MAX, end = 0;
665
Nick Piggin496850e2008-11-19 15:36:33 -0800666 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700667}
668
669/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800670 * Free a vmap area, caller ensuring that the area has been unmapped
671 * and flush_cache_vunmap had been called for the correct range
672 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700673 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800674static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700675{
676 va->flags |= VM_LAZY_FREE;
677 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
678 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800679 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700680}
681
Nick Pigginb29acbd2008-12-01 13:13:47 -0800682/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800683 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
684 * called for the correct range previously.
685 */
686static void free_unmap_vmap_area_noflush(struct vmap_area *va)
687{
688 unmap_vmap_area(va);
689 free_vmap_area_noflush(va);
690}
691
692/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800693 * Free and unmap a vmap area
694 */
695static void free_unmap_vmap_area(struct vmap_area *va)
696{
697 flush_cache_vunmap(va->va_start, va->va_end);
698 free_unmap_vmap_area_noflush(va);
699}
700
Nick Piggindb64fe02008-10-18 20:27:03 -0700701static struct vmap_area *find_vmap_area(unsigned long addr)
702{
703 struct vmap_area *va;
704
705 spin_lock(&vmap_area_lock);
706 va = __find_vmap_area(addr);
707 spin_unlock(&vmap_area_lock);
708
709 return va;
710}
711
712static void free_unmap_vmap_area_addr(unsigned long addr)
713{
714 struct vmap_area *va;
715
716 va = find_vmap_area(addr);
717 BUG_ON(!va);
718 free_unmap_vmap_area(va);
719}
720
721
722/*** Per cpu kva allocator ***/
723
724/*
725 * vmap space is limited especially on 32 bit architectures. Ensure there is
726 * room for at least 16 percpu vmap blocks per CPU.
727 */
728/*
729 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
730 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
731 * instead (we just need a rough idea)
732 */
733#if BITS_PER_LONG == 32
734#define VMALLOC_SPACE (128UL*1024*1024)
735#else
736#define VMALLOC_SPACE (128UL*1024*1024*1024)
737#endif
738
739#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
740#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
741#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
742#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
743#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
744#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200745#define VMAP_BBMAP_BITS \
746 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
747 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
748 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700749
750#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
751
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100752static bool vmap_initialized __read_mostly = false;
753
Nick Piggindb64fe02008-10-18 20:27:03 -0700754struct vmap_block_queue {
755 spinlock_t lock;
756 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700757};
758
759struct vmap_block {
760 spinlock_t lock;
761 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700762 unsigned long free, dirty;
Nick Piggindb64fe02008-10-18 20:27:03 -0700763 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100764 struct list_head free_list;
765 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100766 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700767};
768
769/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
770static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
771
772/*
773 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
774 * in the free path. Could get rid of this if we change the API to return a
775 * "cookie" from alloc, to be passed to free. But no big deal yet.
776 */
777static DEFINE_SPINLOCK(vmap_block_tree_lock);
778static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
779
780/*
781 * We should probably have a fallback mechanism to allocate virtual memory
782 * out of partially filled vmap blocks. However vmap block sizing should be
783 * fairly reasonable according to the vmalloc size, so it shouldn't be a
784 * big problem.
785 */
786
787static unsigned long addr_to_vb_idx(unsigned long addr)
788{
789 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
790 addr /= VMAP_BLOCK_SIZE;
791 return addr;
792}
793
794static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
795{
796 struct vmap_block_queue *vbq;
797 struct vmap_block *vb;
798 struct vmap_area *va;
799 unsigned long vb_idx;
800 int node, err;
801
802 node = numa_node_id();
803
804 vb = kmalloc_node(sizeof(struct vmap_block),
805 gfp_mask & GFP_RECLAIM_MASK, node);
806 if (unlikely(!vb))
807 return ERR_PTR(-ENOMEM);
808
809 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
810 VMALLOC_START, VMALLOC_END,
811 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800812 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700813 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700814 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700815 }
816
817 err = radix_tree_preload(gfp_mask);
818 if (unlikely(err)) {
819 kfree(vb);
820 free_vmap_area(va);
821 return ERR_PTR(err);
822 }
823
824 spin_lock_init(&vb->lock);
825 vb->va = va;
826 vb->free = VMAP_BBMAP_BITS;
827 vb->dirty = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700828 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
829 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700830
831 vb_idx = addr_to_vb_idx(va->va_start);
832 spin_lock(&vmap_block_tree_lock);
833 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
834 spin_unlock(&vmap_block_tree_lock);
835 BUG_ON(err);
836 radix_tree_preload_end();
837
838 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700839 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100840 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700841 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900842 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700843
844 return vb;
845}
846
Nick Piggindb64fe02008-10-18 20:27:03 -0700847static void free_vmap_block(struct vmap_block *vb)
848{
849 struct vmap_block *tmp;
850 unsigned long vb_idx;
851
Nick Piggindb64fe02008-10-18 20:27:03 -0700852 vb_idx = addr_to_vb_idx(vb->va->va_start);
853 spin_lock(&vmap_block_tree_lock);
854 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
855 spin_unlock(&vmap_block_tree_lock);
856 BUG_ON(tmp != vb);
857
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800858 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800859 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700860}
861
Nick Piggin02b709d2010-02-01 22:25:57 +1100862static void purge_fragmented_blocks(int cpu)
863{
864 LIST_HEAD(purge);
865 struct vmap_block *vb;
866 struct vmap_block *n_vb;
867 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
868
869 rcu_read_lock();
870 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
871
872 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
873 continue;
874
875 spin_lock(&vb->lock);
876 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
877 vb->free = 0; /* prevent further allocs after releasing lock */
878 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Nick Piggin02b709d2010-02-01 22:25:57 +1100879 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
880 spin_lock(&vbq->lock);
881 list_del_rcu(&vb->free_list);
882 spin_unlock(&vbq->lock);
883 spin_unlock(&vb->lock);
884 list_add_tail(&vb->purge, &purge);
885 } else
886 spin_unlock(&vb->lock);
887 }
888 rcu_read_unlock();
889
890 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
891 list_del(&vb->purge);
892 free_vmap_block(vb);
893 }
894}
895
Nick Piggin02b709d2010-02-01 22:25:57 +1100896static void purge_fragmented_blocks_allcpus(void)
897{
898 int cpu;
899
900 for_each_possible_cpu(cpu)
901 purge_fragmented_blocks(cpu);
902}
903
Nick Piggindb64fe02008-10-18 20:27:03 -0700904static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
905{
906 struct vmap_block_queue *vbq;
907 struct vmap_block *vb;
908 unsigned long addr = 0;
909 unsigned int order;
910
911 BUG_ON(size & ~PAGE_MASK);
912 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700913 if (WARN_ON(size == 0)) {
914 /*
915 * Allocating 0 bytes isn't what caller wants since
916 * get_order(0) returns funny result. Just warn and terminate
917 * early.
918 */
919 return NULL;
920 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700921 order = get_order(size);
922
923again:
924 rcu_read_lock();
925 vbq = &get_cpu_var(vmap_block_queue);
926 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
927 int i;
928
929 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100930 if (vb->free < 1UL << order)
931 goto next;
932
Zhang Yanfei3fcd76e2013-07-08 15:59:54 -0700933 i = VMAP_BBMAP_BITS - vb->free;
Nick Piggin02b709d2010-02-01 22:25:57 +1100934 addr = vb->va->va_start + (i << PAGE_SHIFT);
935 BUG_ON(addr_to_vb_idx(addr) !=
936 addr_to_vb_idx(vb->va->va_start));
937 vb->free -= 1UL << order;
938 if (vb->free == 0) {
939 spin_lock(&vbq->lock);
940 list_del_rcu(&vb->free_list);
941 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700942 }
943 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100944 break;
945next:
946 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700947 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100948
Tejun Heo3f04ba82009-10-29 22:34:12 +0900949 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700950 rcu_read_unlock();
951
952 if (!addr) {
953 vb = new_vmap_block(gfp_mask);
954 if (IS_ERR(vb))
955 return vb;
956 goto again;
957 }
958
959 return (void *)addr;
960}
961
962static void vb_free(const void *addr, unsigned long size)
963{
964 unsigned long offset;
965 unsigned long vb_idx;
966 unsigned int order;
967 struct vmap_block *vb;
968
969 BUG_ON(size & ~PAGE_MASK);
970 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800971
972 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
973
Nick Piggindb64fe02008-10-18 20:27:03 -0700974 order = get_order(size);
975
976 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
977
978 vb_idx = addr_to_vb_idx((unsigned long)addr);
979 rcu_read_lock();
980 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
981 rcu_read_unlock();
982 BUG_ON(!vb);
983
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800984 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
985
Nick Piggindb64fe02008-10-18 20:27:03 -0700986 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100987 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700988
Nick Piggindb64fe02008-10-18 20:27:03 -0700989 vb->dirty += 1UL << order;
990 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100991 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700992 spin_unlock(&vb->lock);
993 free_vmap_block(vb);
994 } else
995 spin_unlock(&vb->lock);
996}
997
998/**
999 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1000 *
1001 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1002 * to amortize TLB flushing overheads. What this means is that any page you
1003 * have now, may, in a former life, have been mapped into kernel virtual
1004 * address by the vmap layer and so there might be some CPUs with TLB entries
1005 * still referencing that page (additional to the regular 1:1 kernel mapping).
1006 *
1007 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1008 * be sure that none of the pages we have control over will have any aliases
1009 * from the vmap layer.
1010 */
1011void vm_unmap_aliases(void)
1012{
1013 unsigned long start = ULONG_MAX, end = 0;
1014 int cpu;
1015 int flush = 0;
1016
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001017 if (unlikely(!vmap_initialized))
1018 return;
1019
Nick Piggindb64fe02008-10-18 20:27:03 -07001020 for_each_possible_cpu(cpu) {
1021 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1022 struct vmap_block *vb;
1023
1024 rcu_read_lock();
1025 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001026 int i, j;
Nick Piggindb64fe02008-10-18 20:27:03 -07001027
1028 spin_lock(&vb->lock);
1029 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001030 if (i < VMAP_BBMAP_BITS) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001031 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001032
1033 j = find_last_bit(vb->dirty_map,
1034 VMAP_BBMAP_BITS);
1035 j = j + 1; /* need exclusive index */
Nick Piggindb64fe02008-10-18 20:27:03 -07001036
1037 s = vb->va->va_start + (i << PAGE_SHIFT);
1038 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001039 flush = 1;
1040
1041 if (s < start)
1042 start = s;
1043 if (e > end)
1044 end = e;
Nick Piggindb64fe02008-10-18 20:27:03 -07001045 }
1046 spin_unlock(&vb->lock);
1047 }
1048 rcu_read_unlock();
1049 }
1050
1051 __purge_vmap_area_lazy(&start, &end, 1, flush);
1052}
1053EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1054
1055/**
1056 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1057 * @mem: the pointer returned by vm_map_ram
1058 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1059 */
1060void vm_unmap_ram(const void *mem, unsigned int count)
1061{
1062 unsigned long size = count << PAGE_SHIFT;
1063 unsigned long addr = (unsigned long)mem;
1064
1065 BUG_ON(!addr);
1066 BUG_ON(addr < VMALLOC_START);
1067 BUG_ON(addr > VMALLOC_END);
1068 BUG_ON(addr & (PAGE_SIZE-1));
1069
1070 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001071 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001072
1073 if (likely(count <= VMAP_MAX_ALLOC))
1074 vb_free(mem, size);
1075 else
1076 free_unmap_vmap_area_addr(addr);
1077}
1078EXPORT_SYMBOL(vm_unmap_ram);
1079
1080/**
1081 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1082 * @pages: an array of pointers to the pages to be mapped
1083 * @count: number of pages
1084 * @node: prefer to allocate data structures on this node
1085 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001086 *
Gioh Kim36437632014-04-07 15:37:37 -07001087 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1088 * faster than vmap so it's good. But if you mix long-life and short-life
1089 * objects with vm_map_ram(), it could consume lots of address space through
1090 * fragmentation (especially on a 32bit machine). You could see failures in
1091 * the end. Please use this function for short-lived objects.
1092 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001093 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001094 */
1095void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1096{
1097 unsigned long size = count << PAGE_SHIFT;
1098 unsigned long addr;
1099 void *mem;
1100
1101 if (likely(count <= VMAP_MAX_ALLOC)) {
1102 mem = vb_alloc(size, GFP_KERNEL);
1103 if (IS_ERR(mem))
1104 return NULL;
1105 addr = (unsigned long)mem;
1106 } else {
1107 struct vmap_area *va;
1108 va = alloc_vmap_area(size, PAGE_SIZE,
1109 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1110 if (IS_ERR(va))
1111 return NULL;
1112
1113 addr = va->va_start;
1114 mem = (void *)addr;
1115 }
1116 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1117 vm_unmap_ram(mem, count);
1118 return NULL;
1119 }
1120 return mem;
1121}
1122EXPORT_SYMBOL(vm_map_ram);
1123
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001124static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001125/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001126 * vm_area_add_early - add vmap area early during boot
1127 * @vm: vm_struct to add
1128 *
1129 * This function is used to add fixed kernel vm area to vmlist before
1130 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1131 * should contain proper values and the other fields should be zero.
1132 *
1133 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1134 */
1135void __init vm_area_add_early(struct vm_struct *vm)
1136{
1137 struct vm_struct *tmp, **p;
1138
1139 BUG_ON(vmap_initialized);
1140 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1141 if (tmp->addr >= vm->addr) {
1142 BUG_ON(tmp->addr < vm->addr + vm->size);
1143 break;
1144 } else
1145 BUG_ON(tmp->addr + tmp->size > vm->addr);
1146 }
1147 vm->next = *p;
1148 *p = vm;
1149}
1150
1151/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001152 * vm_area_register_early - register vmap area early during boot
1153 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001154 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001155 *
1156 * This function is used to register kernel vm area before
1157 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1158 * proper values on entry and other fields should be zero. On return,
1159 * vm->addr contains the allocated address.
1160 *
1161 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1162 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001163void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001164{
1165 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001166 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001167
Tejun Heoc0c0a292009-02-24 11:57:21 +09001168 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1169 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1170
1171 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001172
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001173 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001174}
1175
Nick Piggindb64fe02008-10-18 20:27:03 -07001176void __init vmalloc_init(void)
1177{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001178 struct vmap_area *va;
1179 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001180 int i;
1181
1182 for_each_possible_cpu(i) {
1183 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001184 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001185
1186 vbq = &per_cpu(vmap_block_queue, i);
1187 spin_lock_init(&vbq->lock);
1188 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001189 p = &per_cpu(vfree_deferred, i);
1190 init_llist_head(&p->list);
1191 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001192 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001193
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001194 /* Import existing vmlist entries. */
1195 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001196 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001197 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001198 va->va_start = (unsigned long)tmp->addr;
1199 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001200 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001201 __insert_vmap_area(va);
1202 }
Tejun Heoca23e402009-08-14 15:00:52 +09001203
1204 vmap_area_pcpu_hole = VMALLOC_END;
1205
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001206 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001207}
1208
Tejun Heo8fc48982009-02-20 16:29:08 +09001209/**
1210 * map_kernel_range_noflush - map kernel VM area with the specified pages
1211 * @addr: start of the VM area to map
1212 * @size: size of the VM area to map
1213 * @prot: page protection flags to use
1214 * @pages: pages to map
1215 *
1216 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1217 * specify should have been allocated using get_vm_area() and its
1218 * friends.
1219 *
1220 * NOTE:
1221 * This function does NOT do any cache flushing. The caller is
1222 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1223 * before calling this function.
1224 *
1225 * RETURNS:
1226 * The number of pages mapped on success, -errno on failure.
1227 */
1228int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1229 pgprot_t prot, struct page **pages)
1230{
1231 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1232}
1233
1234/**
1235 * unmap_kernel_range_noflush - unmap kernel VM area
1236 * @addr: start of the VM area to unmap
1237 * @size: size of the VM area to unmap
1238 *
1239 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1240 * specify should have been allocated using get_vm_area() and its
1241 * friends.
1242 *
1243 * NOTE:
1244 * This function does NOT do any cache flushing. The caller is
1245 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1246 * before calling this function and flush_tlb_kernel_range() after.
1247 */
1248void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1249{
1250 vunmap_page_range(addr, addr + size);
1251}
Huang Ying81e88fd2011-01-12 14:44:55 +08001252EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001253
1254/**
1255 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1256 * @addr: start of the VM area to unmap
1257 * @size: size of the VM area to unmap
1258 *
1259 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1260 * the unmapping and tlb after.
1261 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001262void unmap_kernel_range(unsigned long addr, unsigned long size)
1263{
1264 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001265
1266 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001267 vunmap_page_range(addr, end);
1268 flush_tlb_kernel_range(addr, end);
1269}
Minchan Kim93ef6d62014-06-04 16:11:09 -07001270EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001271
WANG Chaof6f8ed42014-08-06 16:06:58 -07001272int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001273{
1274 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001275 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001276 int err;
1277
WANG Chaof6f8ed42014-08-06 16:06:58 -07001278 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001279
WANG Chaof6f8ed42014-08-06 16:06:58 -07001280 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001281}
1282EXPORT_SYMBOL_GPL(map_vm_area);
1283
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001284static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001285 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001286{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001287 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001288 vm->flags = flags;
1289 vm->addr = (void *)va->va_start;
1290 vm->size = va->va_end - va->va_start;
1291 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001292 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001293 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001294 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001295}
Tejun Heocf88c792009-08-14 15:00:52 +09001296
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001297static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001298{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001299 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001300 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001301 * we should make sure that vm has proper values.
1302 * Pair with smp_rmb() in show_numa_info().
1303 */
1304 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001305 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001306}
1307
Nick Piggindb64fe02008-10-18 20:27:03 -07001308static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001309 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001310 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001311{
Kautuk Consul00065262011-12-19 17:12:04 -08001312 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001313 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001315 BUG_ON(in_interrupt());
Zhang Yanfei0f2d4a82013-07-03 15:04:50 -07001316 if (flags & VM_IOREMAP)
1317 align = 1ul << clamp(fls(size), PAGE_SHIFT, IOREMAP_MAX_ORDER);
Nick Piggindb64fe02008-10-18 20:27:03 -07001318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001320 if (unlikely(!size))
1321 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Tejun Heocf88c792009-08-14 15:00:52 +09001323 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 if (unlikely(!area))
1325 return NULL;
1326
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001327 if (!(flags & VM_NO_GUARD))
1328 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Nick Piggindb64fe02008-10-18 20:27:03 -07001330 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1331 if (IS_ERR(va)) {
1332 kfree(area);
1333 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001336 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339}
1340
Christoph Lameter930fc452005-10-29 18:15:41 -07001341struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1342 unsigned long start, unsigned long end)
1343{
David Rientjes00ef2d22013-02-22 16:35:36 -08001344 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1345 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001346}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001347EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001348
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001349struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1350 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001351 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001352{
David Rientjes00ef2d22013-02-22 16:35:36 -08001353 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1354 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001355}
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357/**
Simon Arlott183ff222007-10-20 01:27:18 +02001358 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 * @size: size of the area
1360 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1361 *
1362 * Search an area of @size in the kernel virtual mapping area,
1363 * and reserved it for out purposes. Returns the area descriptor
1364 * on success or %NULL on failure.
1365 */
1366struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1367{
David Miller2dca6992009-09-21 12:22:34 -07001368 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001369 NUMA_NO_NODE, GFP_KERNEL,
1370 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001371}
1372
1373struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001374 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001375{
David Miller2dca6992009-09-21 12:22:34 -07001376 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001377 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
1379
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001380/**
1381 * find_vm_area - find a continuous kernel virtual area
1382 * @addr: base address
1383 *
1384 * Search for the kernel VM area starting at @addr, and return it.
1385 * It is up to the caller to do all required locking to keep the returned
1386 * pointer valid.
1387 */
1388struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001389{
Nick Piggindb64fe02008-10-18 20:27:03 -07001390 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001391
Nick Piggindb64fe02008-10-18 20:27:03 -07001392 va = find_vmap_area((unsigned long)addr);
1393 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001394 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001395
Andi Kleen7856dfe2005-05-20 14:27:57 -07001396 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001397}
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399/**
Simon Arlott183ff222007-10-20 01:27:18 +02001400 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * @addr: base address
1402 *
1403 * Search for the kernel VM area starting at @addr, and remove it.
1404 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001405 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001407struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
Nick Piggindb64fe02008-10-18 20:27:03 -07001409 struct vmap_area *va;
1410
1411 va = find_vmap_area((unsigned long)addr);
1412 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001413 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001414
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001415 spin_lock(&vmap_area_lock);
1416 va->vm = NULL;
1417 va->flags &= ~VM_VM_AREA;
1418 spin_unlock(&vmap_area_lock);
1419
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001420 vmap_debug_free_range(va->va_start, va->va_end);
1421 free_unmap_vmap_area(va);
1422 vm->size -= PAGE_SIZE;
1423
Nick Piggindb64fe02008-10-18 20:27:03 -07001424 return vm;
1425 }
1426 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427}
1428
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001429static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
1431 struct vm_struct *area;
1432
1433 if (!addr)
1434 return;
1435
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001436 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001437 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440 area = remove_vm_area(addr);
1441 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001442 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 return;
1445 }
1446
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001447 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001448 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001449
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 if (deallocate_pages) {
1451 int i;
1452
1453 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001454 struct page *page = area->pages[i];
1455
1456 BUG_ON(!page);
1457 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
1459
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001460 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 vfree(area->pages);
1462 else
1463 kfree(area->pages);
1464 }
1465
1466 kfree(area);
1467 return;
1468}
Al Viro32fcfd42013-03-10 20:14:08 -04001469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470/**
1471 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 * @addr: memory base address
1473 *
Simon Arlott183ff222007-10-20 01:27:18 +02001474 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001475 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1476 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 *
Al Viro32fcfd42013-03-10 20:14:08 -04001478 * Must not be called in NMI context (strictly speaking, only if we don't
1479 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1480 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001481 *
1482 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001484void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485{
Al Viro32fcfd42013-03-10 20:14:08 -04001486 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001487
1488 kmemleak_free(addr);
1489
Al Viro32fcfd42013-03-10 20:14:08 -04001490 if (!addr)
1491 return;
1492 if (unlikely(in_interrupt())) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001493 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
Oleg Nesterov59d31322013-07-08 16:00:08 -07001494 if (llist_add((struct llist_node *)addr, &p->list))
1495 schedule_work(&p->wq);
Al Viro32fcfd42013-03-10 20:14:08 -04001496 } else
1497 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499EXPORT_SYMBOL(vfree);
1500
1501/**
1502 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 * @addr: memory base address
1504 *
1505 * Free the virtually contiguous memory area starting at @addr,
1506 * which was created from the page array passed to vmap().
1507 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001508 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001510void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511{
1512 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001513 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001514 if (addr)
1515 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517EXPORT_SYMBOL(vunmap);
1518
1519/**
1520 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 * @pages: array of page pointers
1522 * @count: number of pages to map
1523 * @flags: vm_area->flags
1524 * @prot: page protection for the mapping
1525 *
1526 * Maps @count pages from @pages into contiguous kernel virtual
1527 * space.
1528 */
1529void *vmap(struct page **pages, unsigned int count,
1530 unsigned long flags, pgprot_t prot)
1531{
1532 struct vm_struct *area;
1533
Peter Zijlstra34754b62009-02-25 16:04:03 +01001534 might_sleep();
1535
Jan Beulich44813742009-09-21 17:03:05 -07001536 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 return NULL;
1538
Christoph Lameter23016962008-04-28 02:12:42 -07001539 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1540 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 if (!area)
1542 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001543
WANG Chaof6f8ed42014-08-06 16:06:58 -07001544 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 vunmap(area->addr);
1546 return NULL;
1547 }
1548
1549 return area->addr;
1550}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551EXPORT_SYMBOL(vmap);
1552
David Miller2dca6992009-09-21 12:22:34 -07001553static void *__vmalloc_node(unsigned long size, unsigned long align,
1554 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001555 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001556static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001557 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558{
Dave Hansen22943ab2011-05-24 17:12:18 -07001559 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 struct page **pages;
1561 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001562 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1563 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Wanpeng Li762216a2013-09-11 14:22:42 -07001565 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 array_size = (nr_pages * sizeof(struct page *));
1567
1568 area->nr_pages = nr_pages;
1569 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001570 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001571 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001572 PAGE_KERNEL, node, area->caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001573 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001574 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001575 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 area->pages = pages;
1578 if (!area->pages) {
1579 remove_vm_area(area->addr);
1580 kfree(area);
1581 return NULL;
1582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
1584 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001585 struct page *page;
1586
Jianguo Wu4b909512013-11-12 15:07:11 -08001587 if (node == NUMA_NO_NODE)
David Rientjes930f0362014-08-06 16:06:28 -07001588 page = alloc_page(alloc_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001589 else
David Rientjes930f0362014-08-06 16:06:28 -07001590 page = alloc_pages_node(node, alloc_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001591
1592 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 /* Successfully allocated i pages, free them in __vunmap() */
1594 area->nr_pages = i;
1595 goto fail;
1596 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001597 area->pages[i] = page;
Eric Dumazet660654f2014-08-06 16:06:25 -07001598 if (gfp_mask & __GFP_WAIT)
1599 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 }
1601
WANG Chaof6f8ed42014-08-06 16:06:58 -07001602 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 goto fail;
1604 return area->addr;
1605
1606fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001607 warn_alloc_failed(gfp_mask, order,
1608 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001609 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 vfree(area->addr);
1611 return NULL;
1612}
1613
David Rientjesd0a21262011-01-13 15:46:02 -08001614/**
1615 * __vmalloc_node_range - allocate virtually contiguous memory
1616 * @size: allocation size
1617 * @align: desired alignment
1618 * @start: vm area range start
1619 * @end: vm area range end
1620 * @gfp_mask: flags for the page level allocator
1621 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001622 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001623 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001624 * @caller: caller's return address
1625 *
1626 * Allocate enough pages to cover @size from the page level
1627 * allocator with @gfp_mask flags. Map them into contiguous
1628 * kernel virtual space, using a pagetable protection of @prot.
1629 */
1630void *__vmalloc_node_range(unsigned long size, unsigned long align,
1631 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001632 pgprot_t prot, unsigned long vm_flags, int node,
1633 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001634{
David Rientjesd0a21262011-01-13 15:46:02 -08001635 struct vm_struct *area;
1636 void *addr;
1637 unsigned long real_size = size;
1638
1639 size = PAGE_ALIGN(size);
1640 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001641 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001642
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001643 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1644 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001645 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001646 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001647
Wanpeng Li3722e132013-11-12 15:07:29 -08001648 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001649 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001650 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001651
1652 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001653 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1654 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001655 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001656 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001657 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001658
1659 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001660 * A ref_count = 2 is needed because vm_struct allocated in
1661 * __get_vm_area_node() contains a reference to the virtual address of
1662 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001663 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001664 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001665
1666 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001667
1668fail:
1669 warn_alloc_failed(gfp_mask, 0,
1670 "vmalloc: allocation failure: %lu bytes\n",
1671 real_size);
1672 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001673}
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001676 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001678 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 * @gfp_mask: flags for the page level allocator
1680 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001681 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001682 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 *
1684 * Allocate enough pages to cover @size from the page level
1685 * allocator with @gfp_mask flags. Map them into contiguous
1686 * kernel virtual space, using a pagetable protection of @prot.
1687 */
David Miller2dca6992009-09-21 12:22:34 -07001688static void *__vmalloc_node(unsigned long size, unsigned long align,
1689 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001690 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691{
David Rientjesd0a21262011-01-13 15:46:02 -08001692 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001693 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694}
1695
Christoph Lameter930fc452005-10-29 18:15:41 -07001696void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1697{
David Rientjes00ef2d22013-02-22 16:35:36 -08001698 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001699 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701EXPORT_SYMBOL(__vmalloc);
1702
Dave Younge1ca7782010-10-26 14:22:06 -07001703static inline void *__vmalloc_node_flags(unsigned long size,
1704 int node, gfp_t flags)
1705{
1706 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1707 node, __builtin_return_address(0));
1708}
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710/**
1711 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 * Allocate enough pages to cover @size from the page level
1714 * allocator and map them into contiguous kernel virtual space.
1715 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001716 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 * use __vmalloc() instead.
1718 */
1719void *vmalloc(unsigned long size)
1720{
David Rientjes00ef2d22013-02-22 16:35:36 -08001721 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1722 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724EXPORT_SYMBOL(vmalloc);
1725
Christoph Lameter930fc452005-10-29 18:15:41 -07001726/**
Dave Younge1ca7782010-10-26 14:22:06 -07001727 * vzalloc - allocate virtually contiguous memory with zero fill
1728 * @size: allocation size
1729 * Allocate enough pages to cover @size from the page level
1730 * allocator and map them into contiguous kernel virtual space.
1731 * The memory allocated is set to zero.
1732 *
1733 * For tight control over page level allocator and protection flags
1734 * use __vmalloc() instead.
1735 */
1736void *vzalloc(unsigned long size)
1737{
David Rientjes00ef2d22013-02-22 16:35:36 -08001738 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001739 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1740}
1741EXPORT_SYMBOL(vzalloc);
1742
1743/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001744 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1745 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001746 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001747 * The resulting memory area is zeroed so it can be mapped to userspace
1748 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001749 */
1750void *vmalloc_user(unsigned long size)
1751{
1752 struct vm_struct *area;
1753 void *ret;
1754
David Miller2dca6992009-09-21 12:22:34 -07001755 ret = __vmalloc_node(size, SHMLBA,
1756 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001757 PAGE_KERNEL, NUMA_NO_NODE,
1758 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001759 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001760 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001761 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001762 }
Nick Piggin83342312006-06-23 02:03:20 -07001763 return ret;
1764}
1765EXPORT_SYMBOL(vmalloc_user);
1766
1767/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001768 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001769 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001770 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001771 *
1772 * Allocate enough pages to cover @size from the page level
1773 * allocator and map them into contiguous kernel virtual space.
1774 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001775 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001776 * use __vmalloc() instead.
1777 */
1778void *vmalloc_node(unsigned long size, int node)
1779{
David Miller2dca6992009-09-21 12:22:34 -07001780 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001781 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001782}
1783EXPORT_SYMBOL(vmalloc_node);
1784
Dave Younge1ca7782010-10-26 14:22:06 -07001785/**
1786 * vzalloc_node - allocate memory on a specific node with zero fill
1787 * @size: allocation size
1788 * @node: numa node
1789 *
1790 * Allocate enough pages to cover @size from the page level
1791 * allocator and map them into contiguous kernel virtual space.
1792 * The memory allocated is set to zero.
1793 *
1794 * For tight control over page level allocator and protection flags
1795 * use __vmalloc_node() instead.
1796 */
1797void *vzalloc_node(unsigned long size, int node)
1798{
1799 return __vmalloc_node_flags(size, node,
1800 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1801}
1802EXPORT_SYMBOL(vzalloc_node);
1803
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001804#ifndef PAGE_KERNEL_EXEC
1805# define PAGE_KERNEL_EXEC PAGE_KERNEL
1806#endif
1807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808/**
1809 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 * @size: allocation size
1811 *
1812 * Kernel-internal function to allocate enough pages to cover @size
1813 * the page level allocator and map them into contiguous and
1814 * executable kernel virtual space.
1815 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001816 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 * use __vmalloc() instead.
1818 */
1819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820void *vmalloc_exec(unsigned long size)
1821{
David Miller2dca6992009-09-21 12:22:34 -07001822 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001823 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824}
1825
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001826#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001827#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001828#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001829#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001830#else
1831#define GFP_VMALLOC32 GFP_KERNEL
1832#endif
1833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834/**
1835 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 * @size: allocation size
1837 *
1838 * Allocate enough 32bit PA addressable pages to cover @size from the
1839 * page level allocator and map them into contiguous kernel virtual space.
1840 */
1841void *vmalloc_32(unsigned long size)
1842{
David Miller2dca6992009-09-21 12:22:34 -07001843 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001844 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846EXPORT_SYMBOL(vmalloc_32);
1847
Nick Piggin83342312006-06-23 02:03:20 -07001848/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001849 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001850 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001851 *
1852 * The resulting memory area is 32bit addressable and zeroed so it can be
1853 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001854 */
1855void *vmalloc_32_user(unsigned long size)
1856{
1857 struct vm_struct *area;
1858 void *ret;
1859
David Miller2dca6992009-09-21 12:22:34 -07001860 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001861 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001862 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001863 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001864 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001865 }
Nick Piggin83342312006-06-23 02:03:20 -07001866 return ret;
1867}
1868EXPORT_SYMBOL(vmalloc_32_user);
1869
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001870/*
1871 * small helper routine , copy contents to buf from addr.
1872 * If the page is not present, fill zero.
1873 */
1874
1875static int aligned_vread(char *buf, char *addr, unsigned long count)
1876{
1877 struct page *p;
1878 int copied = 0;
1879
1880 while (count) {
1881 unsigned long offset, length;
1882
1883 offset = (unsigned long)addr & ~PAGE_MASK;
1884 length = PAGE_SIZE - offset;
1885 if (length > count)
1886 length = count;
1887 p = vmalloc_to_page(addr);
1888 /*
1889 * To do safe access to this _mapped_ area, we need
1890 * lock. But adding lock here means that we need to add
1891 * overhead of vmalloc()/vfree() calles for this _debug_
1892 * interface, rarely used. Instead of that, we'll use
1893 * kmap() and get small overhead in this access function.
1894 */
1895 if (p) {
1896 /*
1897 * we can expect USER0 is not used (see vread/vwrite's
1898 * function description)
1899 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001900 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001901 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001902 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001903 } else
1904 memset(buf, 0, length);
1905
1906 addr += length;
1907 buf += length;
1908 copied += length;
1909 count -= length;
1910 }
1911 return copied;
1912}
1913
1914static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1915{
1916 struct page *p;
1917 int copied = 0;
1918
1919 while (count) {
1920 unsigned long offset, length;
1921
1922 offset = (unsigned long)addr & ~PAGE_MASK;
1923 length = PAGE_SIZE - offset;
1924 if (length > count)
1925 length = count;
1926 p = vmalloc_to_page(addr);
1927 /*
1928 * To do safe access to this _mapped_ area, we need
1929 * lock. But adding lock here means that we need to add
1930 * overhead of vmalloc()/vfree() calles for this _debug_
1931 * interface, rarely used. Instead of that, we'll use
1932 * kmap() and get small overhead in this access function.
1933 */
1934 if (p) {
1935 /*
1936 * we can expect USER0 is not used (see vread/vwrite's
1937 * function description)
1938 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001939 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001940 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001941 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001942 }
1943 addr += length;
1944 buf += length;
1945 copied += length;
1946 count -= length;
1947 }
1948 return copied;
1949}
1950
1951/**
1952 * vread() - read vmalloc area in a safe way.
1953 * @buf: buffer for reading data
1954 * @addr: vm address.
1955 * @count: number of bytes to be read.
1956 *
1957 * Returns # of bytes which addr and buf should be increased.
1958 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1959 * includes any intersect with alive vmalloc area.
1960 *
1961 * This function checks that addr is a valid vmalloc'ed area, and
1962 * copy data from that area to a given buffer. If the given memory range
1963 * of [addr...addr+count) includes some valid address, data is copied to
1964 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1965 * IOREMAP area is treated as memory hole and no copy is done.
1966 *
1967 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08001968 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001969 *
1970 * Note: In usual ops, vread() is never necessary because the caller
1971 * should know vmalloc() area is valid and can use memcpy().
1972 * This is for routines which have to access vmalloc area without
1973 * any informaion, as /dev/kmem.
1974 *
1975 */
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977long vread(char *buf, char *addr, unsigned long count)
1978{
Joonsoo Kime81ce852013-04-29 15:07:32 -07001979 struct vmap_area *va;
1980 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001982 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 unsigned long n;
1984
1985 /* Don't allow overflow */
1986 if ((unsigned long) addr + count < count)
1987 count = -(unsigned long) addr;
1988
Joonsoo Kime81ce852013-04-29 15:07:32 -07001989 spin_lock(&vmap_area_lock);
1990 list_for_each_entry(va, &vmap_area_list, list) {
1991 if (!count)
1992 break;
1993
1994 if (!(va->flags & VM_VM_AREA))
1995 continue;
1996
1997 vm = va->vm;
1998 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001999 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 continue;
2001 while (addr < vaddr) {
2002 if (count == 0)
2003 goto finished;
2004 *buf = '\0';
2005 buf++;
2006 addr++;
2007 count--;
2008 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002009 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002010 if (n > count)
2011 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002012 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002013 aligned_vread(buf, addr, n);
2014 else /* IOREMAP area is treated as memory hole */
2015 memset(buf, 0, n);
2016 buf += n;
2017 addr += n;
2018 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
2020finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002021 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002022
2023 if (buf == buf_start)
2024 return 0;
2025 /* zero-fill memory holes */
2026 if (buf != buf_start + buflen)
2027 memset(buf, 0, buflen - (buf - buf_start));
2028
2029 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002032/**
2033 * vwrite() - write vmalloc area in a safe way.
2034 * @buf: buffer for source data
2035 * @addr: vm address.
2036 * @count: number of bytes to be read.
2037 *
2038 * Returns # of bytes which addr and buf should be incresed.
2039 * (same number to @count).
2040 * If [addr...addr+count) doesn't includes any intersect with valid
2041 * vmalloc area, returns 0.
2042 *
2043 * This function checks that addr is a valid vmalloc'ed area, and
2044 * copy data from a buffer to the given addr. If specified range of
2045 * [addr...addr+count) includes some valid address, data is copied from
2046 * proper area of @buf. If there are memory holes, no copy to hole.
2047 * IOREMAP area is treated as memory hole and no copy is done.
2048 *
2049 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002050 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002051 *
2052 * Note: In usual ops, vwrite() is never necessary because the caller
2053 * should know vmalloc() area is valid and can use memcpy().
2054 * This is for routines which have to access vmalloc area without
2055 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002056 */
2057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058long vwrite(char *buf, char *addr, unsigned long count)
2059{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002060 struct vmap_area *va;
2061 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002062 char *vaddr;
2063 unsigned long n, buflen;
2064 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
2066 /* Don't allow overflow */
2067 if ((unsigned long) addr + count < count)
2068 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002069 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Joonsoo Kime81ce852013-04-29 15:07:32 -07002071 spin_lock(&vmap_area_lock);
2072 list_for_each_entry(va, &vmap_area_list, list) {
2073 if (!count)
2074 break;
2075
2076 if (!(va->flags & VM_VM_AREA))
2077 continue;
2078
2079 vm = va->vm;
2080 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002081 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 continue;
2083 while (addr < vaddr) {
2084 if (count == 0)
2085 goto finished;
2086 buf++;
2087 addr++;
2088 count--;
2089 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002090 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002091 if (n > count)
2092 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002093 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002094 aligned_vwrite(buf, addr, n);
2095 copied++;
2096 }
2097 buf += n;
2098 addr += n;
2099 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 }
2101finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002102 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002103 if (!copied)
2104 return 0;
2105 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
Nick Piggin83342312006-06-23 02:03:20 -07002107
2108/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002109 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2110 * @vma: vma to cover
2111 * @uaddr: target user address to start at
2112 * @kaddr: virtual address of vmalloc kernel memory
2113 * @size: size of map area
2114 *
2115 * Returns: 0 for success, -Exxx on failure
2116 *
2117 * This function checks that @kaddr is a valid vmalloc'ed area,
2118 * and that it is big enough to cover the range starting at
2119 * @uaddr in @vma. Will return failure if that criteria isn't
2120 * met.
2121 *
2122 * Similar to remap_pfn_range() (see mm/memory.c)
2123 */
2124int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2125 void *kaddr, unsigned long size)
2126{
2127 struct vm_struct *area;
2128
2129 size = PAGE_ALIGN(size);
2130
2131 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2132 return -EINVAL;
2133
2134 area = find_vm_area(kaddr);
2135 if (!area)
2136 return -EINVAL;
2137
2138 if (!(area->flags & VM_USERMAP))
2139 return -EINVAL;
2140
2141 if (kaddr + size > area->addr + area->size)
2142 return -EINVAL;
2143
2144 do {
2145 struct page *page = vmalloc_to_page(kaddr);
2146 int ret;
2147
2148 ret = vm_insert_page(vma, uaddr, page);
2149 if (ret)
2150 return ret;
2151
2152 uaddr += PAGE_SIZE;
2153 kaddr += PAGE_SIZE;
2154 size -= PAGE_SIZE;
2155 } while (size > 0);
2156
2157 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2158
2159 return 0;
2160}
2161EXPORT_SYMBOL(remap_vmalloc_range_partial);
2162
2163/**
Nick Piggin83342312006-06-23 02:03:20 -07002164 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002165 * @vma: vma to cover (map full range of vma)
2166 * @addr: vmalloc memory
2167 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002168 *
2169 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002170 *
2171 * This function checks that addr is a valid vmalloc'ed area, and
2172 * that it is big enough to cover the vma. Will return failure if
2173 * that criteria isn't met.
2174 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002175 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002176 */
2177int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2178 unsigned long pgoff)
2179{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002180 return remap_vmalloc_range_partial(vma, vma->vm_start,
2181 addr + (pgoff << PAGE_SHIFT),
2182 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002183}
2184EXPORT_SYMBOL(remap_vmalloc_range);
2185
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002186/*
2187 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2188 * have one.
2189 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002190void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002191{
2192}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002193
2194
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002195static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002196{
David Vrabelcd129092011-09-29 16:53:32 +01002197 pte_t ***p = data;
2198
2199 if (p) {
2200 *(*p) = pte;
2201 (*p)++;
2202 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002203 return 0;
2204}
2205
2206/**
2207 * alloc_vm_area - allocate a range of kernel address space
2208 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002209 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002210 *
2211 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002212 *
2213 * This function reserves a range of kernel address space, and
2214 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002215 * are created.
2216 *
2217 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2218 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002219 */
David Vrabelcd129092011-09-29 16:53:32 +01002220struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002221{
2222 struct vm_struct *area;
2223
Christoph Lameter23016962008-04-28 02:12:42 -07002224 area = get_vm_area_caller(size, VM_IOREMAP,
2225 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002226 if (area == NULL)
2227 return NULL;
2228
2229 /*
2230 * This ensures that page tables are constructed for this region
2231 * of kernel virtual address space and mapped into init_mm.
2232 */
2233 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002234 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002235 free_vm_area(area);
2236 return NULL;
2237 }
2238
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002239 return area;
2240}
2241EXPORT_SYMBOL_GPL(alloc_vm_area);
2242
2243void free_vm_area(struct vm_struct *area)
2244{
2245 struct vm_struct *ret;
2246 ret = remove_vm_area(area->addr);
2247 BUG_ON(ret != area);
2248 kfree(area);
2249}
2250EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002251
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002252#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002253static struct vmap_area *node_to_va(struct rb_node *n)
2254{
2255 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2256}
2257
2258/**
2259 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2260 * @end: target address
2261 * @pnext: out arg for the next vmap_area
2262 * @pprev: out arg for the previous vmap_area
2263 *
2264 * Returns: %true if either or both of next and prev are found,
2265 * %false if no vmap_area exists
2266 *
2267 * Find vmap_areas end addresses of which enclose @end. ie. if not
2268 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2269 */
2270static bool pvm_find_next_prev(unsigned long end,
2271 struct vmap_area **pnext,
2272 struct vmap_area **pprev)
2273{
2274 struct rb_node *n = vmap_area_root.rb_node;
2275 struct vmap_area *va = NULL;
2276
2277 while (n) {
2278 va = rb_entry(n, struct vmap_area, rb_node);
2279 if (end < va->va_end)
2280 n = n->rb_left;
2281 else if (end > va->va_end)
2282 n = n->rb_right;
2283 else
2284 break;
2285 }
2286
2287 if (!va)
2288 return false;
2289
2290 if (va->va_end > end) {
2291 *pnext = va;
2292 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2293 } else {
2294 *pprev = va;
2295 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2296 }
2297 return true;
2298}
2299
2300/**
2301 * pvm_determine_end - find the highest aligned address between two vmap_areas
2302 * @pnext: in/out arg for the next vmap_area
2303 * @pprev: in/out arg for the previous vmap_area
2304 * @align: alignment
2305 *
2306 * Returns: determined end address
2307 *
2308 * Find the highest aligned address between *@pnext and *@pprev below
2309 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2310 * down address is between the end addresses of the two vmap_areas.
2311 *
2312 * Please note that the address returned by this function may fall
2313 * inside *@pnext vmap_area. The caller is responsible for checking
2314 * that.
2315 */
2316static unsigned long pvm_determine_end(struct vmap_area **pnext,
2317 struct vmap_area **pprev,
2318 unsigned long align)
2319{
2320 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2321 unsigned long addr;
2322
2323 if (*pnext)
2324 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2325 else
2326 addr = vmalloc_end;
2327
2328 while (*pprev && (*pprev)->va_end > addr) {
2329 *pnext = *pprev;
2330 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2331 }
2332
2333 return addr;
2334}
2335
2336/**
2337 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2338 * @offsets: array containing offset of each area
2339 * @sizes: array containing size of each area
2340 * @nr_vms: the number of areas to allocate
2341 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002342 *
2343 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2344 * vm_structs on success, %NULL on failure
2345 *
2346 * Percpu allocator wants to use congruent vm areas so that it can
2347 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002348 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2349 * be scattered pretty far, distance between two areas easily going up
2350 * to gigabytes. To avoid interacting with regular vmallocs, these
2351 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002352 *
2353 * Despite its complicated look, this allocator is rather simple. It
2354 * does everything top-down and scans areas from the end looking for
2355 * matching slot. While scanning, if any of the areas overlaps with
2356 * existing vmap_area, the base address is pulled down to fit the
2357 * area. Scanning is repeated till all the areas fit and then all
2358 * necessary data structres are inserted and the result is returned.
2359 */
2360struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2361 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002362 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002363{
2364 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2365 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2366 struct vmap_area **vas, *prev, *next;
2367 struct vm_struct **vms;
2368 int area, area2, last_area, term_area;
2369 unsigned long base, start, end, last_end;
2370 bool purged = false;
2371
Tejun Heoca23e402009-08-14 15:00:52 +09002372 /* verify parameters and allocate data structures */
2373 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2374 for (last_area = 0, area = 0; area < nr_vms; area++) {
2375 start = offsets[area];
2376 end = start + sizes[area];
2377
2378 /* is everything aligned properly? */
2379 BUG_ON(!IS_ALIGNED(offsets[area], align));
2380 BUG_ON(!IS_ALIGNED(sizes[area], align));
2381
2382 /* detect the area with the highest address */
2383 if (start > offsets[last_area])
2384 last_area = area;
2385
2386 for (area2 = 0; area2 < nr_vms; area2++) {
2387 unsigned long start2 = offsets[area2];
2388 unsigned long end2 = start2 + sizes[area2];
2389
2390 if (area2 == area)
2391 continue;
2392
2393 BUG_ON(start2 >= start && start2 < end);
2394 BUG_ON(end2 <= end && end2 > start);
2395 }
2396 }
2397 last_end = offsets[last_area] + sizes[last_area];
2398
2399 if (vmalloc_end - vmalloc_start < last_end) {
2400 WARN_ON(true);
2401 return NULL;
2402 }
2403
Thomas Meyer4d67d862012-05-29 15:06:21 -07002404 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2405 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002406 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002407 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002408
2409 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002410 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2411 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002412 if (!vas[area] || !vms[area])
2413 goto err_free;
2414 }
2415retry:
2416 spin_lock(&vmap_area_lock);
2417
2418 /* start scanning - we scan from the top, begin with the last area */
2419 area = term_area = last_area;
2420 start = offsets[area];
2421 end = start + sizes[area];
2422
2423 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2424 base = vmalloc_end - last_end;
2425 goto found;
2426 }
2427 base = pvm_determine_end(&next, &prev, align) - end;
2428
2429 while (true) {
2430 BUG_ON(next && next->va_end <= base + end);
2431 BUG_ON(prev && prev->va_end > base + end);
2432
2433 /*
2434 * base might have underflowed, add last_end before
2435 * comparing.
2436 */
2437 if (base + last_end < vmalloc_start + last_end) {
2438 spin_unlock(&vmap_area_lock);
2439 if (!purged) {
2440 purge_vmap_area_lazy();
2441 purged = true;
2442 goto retry;
2443 }
2444 goto err_free;
2445 }
2446
2447 /*
2448 * If next overlaps, move base downwards so that it's
2449 * right below next and then recheck.
2450 */
2451 if (next && next->va_start < base + end) {
2452 base = pvm_determine_end(&next, &prev, align) - end;
2453 term_area = area;
2454 continue;
2455 }
2456
2457 /*
2458 * If prev overlaps, shift down next and prev and move
2459 * base so that it's right below new next and then
2460 * recheck.
2461 */
2462 if (prev && prev->va_end > base + start) {
2463 next = prev;
2464 prev = node_to_va(rb_prev(&next->rb_node));
2465 base = pvm_determine_end(&next, &prev, align) - end;
2466 term_area = area;
2467 continue;
2468 }
2469
2470 /*
2471 * This area fits, move on to the previous one. If
2472 * the previous one is the terminal one, we're done.
2473 */
2474 area = (area + nr_vms - 1) % nr_vms;
2475 if (area == term_area)
2476 break;
2477 start = offsets[area];
2478 end = start + sizes[area];
2479 pvm_find_next_prev(base + end, &next, &prev);
2480 }
2481found:
2482 /* we've found a fitting base, insert all va's */
2483 for (area = 0; area < nr_vms; area++) {
2484 struct vmap_area *va = vas[area];
2485
2486 va->va_start = base + offsets[area];
2487 va->va_end = va->va_start + sizes[area];
2488 __insert_vmap_area(va);
2489 }
2490
2491 vmap_area_pcpu_hole = base + offsets[last_area];
2492
2493 spin_unlock(&vmap_area_lock);
2494
2495 /* insert all vm's */
2496 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002497 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2498 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002499
2500 kfree(vas);
2501 return vms;
2502
2503err_free:
2504 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002505 kfree(vas[area]);
2506 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002507 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002508err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002509 kfree(vas);
2510 kfree(vms);
2511 return NULL;
2512}
2513
2514/**
2515 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2516 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2517 * @nr_vms: the number of allocated areas
2518 *
2519 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2520 */
2521void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2522{
2523 int i;
2524
2525 for (i = 0; i < nr_vms; i++)
2526 free_vm_area(vms[i]);
2527 kfree(vms);
2528}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002529#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002530
2531#ifdef CONFIG_PROC_FS
2532static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002533 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002534{
2535 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002536 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002537
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002538 spin_lock(&vmap_area_lock);
2539 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2540 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002541 n--;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002542 va = list_entry(va->list.next, typeof(*va), list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002543 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002544 if (!n && &va->list != &vmap_area_list)
2545 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002546
2547 return NULL;
2548
2549}
2550
2551static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2552{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002553 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002554
2555 ++*pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002556 next = list_entry(va->list.next, typeof(*va), list);
2557 if (&next->list != &vmap_area_list)
2558 return next;
2559
2560 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002561}
2562
2563static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002564 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002565{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002566 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002567}
2568
Eric Dumazeta47a1262008-07-23 21:27:38 -07002569static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2570{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002571 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002572 unsigned int nr, *counters = m->private;
2573
2574 if (!counters)
2575 return;
2576
Wanpeng Liaf123462013-11-12 15:07:32 -08002577 if (v->flags & VM_UNINITIALIZED)
2578 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002579 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2580 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002581
Eric Dumazeta47a1262008-07-23 21:27:38 -07002582 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2583
2584 for (nr = 0; nr < v->nr_pages; nr++)
2585 counters[page_to_nid(v->pages[nr])]++;
2586
2587 for_each_node_state(nr, N_HIGH_MEMORY)
2588 if (counters[nr])
2589 seq_printf(m, " N%u=%u", nr, counters[nr]);
2590 }
2591}
2592
Christoph Lametera10aa572008-04-28 02:12:40 -07002593static int s_show(struct seq_file *m, void *p)
2594{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002595 struct vmap_area *va = p;
2596 struct vm_struct *v;
2597
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002598 /*
2599 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2600 * behalf of vmap area is being tear down or vm_map_ram allocation.
2601 */
2602 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002603 return 0;
2604
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002605 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002606
Kees Cook45ec1692012-10-08 16:34:09 -07002607 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002608 v->addr, v->addr + v->size, v->size);
2609
Joe Perches62c70bc2011-01-13 15:45:52 -08002610 if (v->caller)
2611 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002612
Christoph Lametera10aa572008-04-28 02:12:40 -07002613 if (v->nr_pages)
2614 seq_printf(m, " pages=%d", v->nr_pages);
2615
2616 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002617 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002618
2619 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002620 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002621
2622 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002623 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002624
2625 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002626 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002627
2628 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002629 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002630
2631 if (v->flags & VM_VPAGES)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002632 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002633
Eric Dumazeta47a1262008-07-23 21:27:38 -07002634 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002635 seq_putc(m, '\n');
2636 return 0;
2637}
2638
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002639static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002640 .start = s_start,
2641 .next = s_next,
2642 .stop = s_stop,
2643 .show = s_show,
2644};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002645
2646static int vmalloc_open(struct inode *inode, struct file *file)
2647{
Rob Jones703394c2014-10-09 15:28:01 -07002648 if (IS_ENABLED(CONFIG_NUMA))
2649 return seq_open_private(file, &vmalloc_op,
2650 nr_node_ids * sizeof(unsigned int));
2651 else
2652 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002653}
2654
2655static const struct file_operations proc_vmalloc_operations = {
2656 .open = vmalloc_open,
2657 .read = seq_read,
2658 .llseek = seq_lseek,
2659 .release = seq_release_private,
2660};
2661
2662static int __init proc_vmalloc_init(void)
2663{
2664 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2665 return 0;
2666}
2667module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002668
2669void get_vmalloc_info(struct vmalloc_info *vmi)
2670{
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002671 struct vmap_area *va;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002672 unsigned long free_area_size;
2673 unsigned long prev_end;
2674
2675 vmi->used = 0;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002676 vmi->largest_chunk = 0;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002677
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002678 prev_end = VMALLOC_START;
2679
Joonsoo Kim474750a2014-08-06 16:05:06 -07002680 rcu_read_lock();
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002681
2682 if (list_empty(&vmap_area_list)) {
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002683 vmi->largest_chunk = VMALLOC_TOTAL;
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002684 goto out;
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002685 }
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002686
Joonsoo Kim474750a2014-08-06 16:05:06 -07002687 list_for_each_entry_rcu(va, &vmap_area_list, list) {
Joonsoo Kimf98782d2013-04-29 15:07:34 -07002688 unsigned long addr = va->va_start;
2689
2690 /*
2691 * Some archs keep another range for modules in vmalloc space
2692 */
2693 if (addr < VMALLOC_START)
2694 continue;
2695 if (addr >= VMALLOC_END)
2696 break;
2697
2698 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2699 continue;
2700
2701 vmi->used += (va->va_end - va->va_start);
2702
2703 free_area_size = addr - prev_end;
2704 if (vmi->largest_chunk < free_area_size)
2705 vmi->largest_chunk = free_area_size;
2706
2707 prev_end = va->va_end;
2708 }
2709
2710 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2711 vmi->largest_chunk = VMALLOC_END - prev_end;
2712
2713out:
Joonsoo Kim474750a2014-08-06 16:05:06 -07002714 rcu_read_unlock();
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002715}
Christoph Lametera10aa572008-04-28 02:12:40 -07002716#endif
2717