blob: 673942094328a710b059b2b50e149ce7eb3d5f11 [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>
Ingo Molnarc3edc402017-02-02 08:35:14 +010015#include <linux/sched/signal.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>
Chris Wilson4da56b92016-04-04 14:46:42 +010024#include <linux/notifier.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070025#include <linux/rbtree.h>
26#include <linux/radix-tree.h>
27#include <linux/rcupdate.h>
Tejun Heof0aa6612009-02-20 16:29:08 +090028#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010029#include <linux/kmemleak.h>
Arun Sharma600634972011-07-26 16:09:06 -070030#include <linux/atomic.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070031#include <linux/compiler.h>
Al Viro32fcfd42013-03-10 20:14:08 -040032#include <linux/llist.h>
Toshi Kani0f616be2015-04-14 15:47:17 -070033#include <linux/bitops.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070034
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080035#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/tlbflush.h>
David Miller2dca6992009-09-21 12:22:34 -070037#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Mel Gormandd56b042015-11-06 16:28:43 -080039#include "internal.h"
40
Al Viro32fcfd42013-03-10 20:14:08 -040041struct vfree_deferred {
42 struct llist_head list;
43 struct work_struct wq;
44};
45static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
46
47static void __vunmap(const void *, int);
48
49static void free_work(struct work_struct *w)
50{
51 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
Byungchul Park894e58c2017-09-06 16:24:26 -070052 struct llist_node *t, *llnode;
53
54 llist_for_each_safe(llnode, t, llist_del_all(&p->list))
55 __vunmap((void *)llnode, 1);
Al Viro32fcfd42013-03-10 20:14:08 -040056}
57
Nick Piggindb64fe02008-10-18 20:27:03 -070058/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070059
Linus Torvalds1da177e2005-04-16 15:20:36 -070060static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
61{
62 pte_t *pte;
63
64 pte = pte_offset_kernel(pmd, addr);
65 do {
66 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
67 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
68 } while (pte++, addr += PAGE_SIZE, addr != end);
69}
70
Nick Piggindb64fe02008-10-18 20:27:03 -070071static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 pmd_t *pmd;
74 unsigned long next;
75
76 pmd = pmd_offset(pud, addr);
77 do {
78 next = pmd_addr_end(addr, end);
Toshi Kanib9820d82015-04-14 15:47:26 -070079 if (pmd_clear_huge(pmd))
80 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 if (pmd_none_or_clear_bad(pmd))
82 continue;
83 vunmap_pte_range(pmd, addr, next);
84 } while (pmd++, addr = next, addr != end);
85}
86
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +030087static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 pud_t *pud;
90 unsigned long next;
91
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +030092 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 do {
94 next = pud_addr_end(addr, end);
Toshi Kanib9820d82015-04-14 15:47:26 -070095 if (pud_clear_huge(pud))
96 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 if (pud_none_or_clear_bad(pud))
98 continue;
99 vunmap_pmd_range(pud, addr, next);
100 } while (pud++, addr = next, addr != end);
101}
102
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300103static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end)
104{
105 p4d_t *p4d;
106 unsigned long next;
107
108 p4d = p4d_offset(pgd, addr);
109 do {
110 next = p4d_addr_end(addr, end);
111 if (p4d_clear_huge(p4d))
112 continue;
113 if (p4d_none_or_clear_bad(p4d))
114 continue;
115 vunmap_pud_range(p4d, addr, next);
116 } while (p4d++, addr = next, addr != end);
117}
118
Nick Piggindb64fe02008-10-18 20:27:03 -0700119static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
121 pgd_t *pgd;
122 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 BUG_ON(addr >= end);
125 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 do {
127 next = pgd_addr_end(addr, end);
128 if (pgd_none_or_clear_bad(pgd))
129 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300130 vunmap_p4d_range(pgd, addr, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132}
133
134static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -0700135 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 pte_t *pte;
138
Nick Piggindb64fe02008-10-18 20:27:03 -0700139 /*
140 * nr is a running index into the array which helps higher level
141 * callers keep track of where we're up to.
142 */
143
Hugh Dickins872fec12005-10-29 18:16:21 -0700144 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (!pte)
146 return -ENOMEM;
147 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700148 struct page *page = pages[*nr];
149
150 if (WARN_ON(!pte_none(*pte)))
151 return -EBUSY;
152 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 return -ENOMEM;
154 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700155 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 } while (pte++, addr += PAGE_SIZE, addr != end);
157 return 0;
158}
159
Nick Piggindb64fe02008-10-18 20:27:03 -0700160static int vmap_pmd_range(pud_t *pud, unsigned long addr,
161 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
163 pmd_t *pmd;
164 unsigned long next;
165
166 pmd = pmd_alloc(&init_mm, pud, addr);
167 if (!pmd)
168 return -ENOMEM;
169 do {
170 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700171 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 return -ENOMEM;
173 } while (pmd++, addr = next, addr != end);
174 return 0;
175}
176
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300177static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -0700178 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
180 pud_t *pud;
181 unsigned long next;
182
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300183 pud = pud_alloc(&init_mm, p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 if (!pud)
185 return -ENOMEM;
186 do {
187 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700188 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return -ENOMEM;
190 } while (pud++, addr = next, addr != end);
191 return 0;
192}
193
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300194static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
195 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
196{
197 p4d_t *p4d;
198 unsigned long next;
199
200 p4d = p4d_alloc(&init_mm, pgd, addr);
201 if (!p4d)
202 return -ENOMEM;
203 do {
204 next = p4d_addr_end(addr, end);
205 if (vmap_pud_range(p4d, addr, next, prot, pages, nr))
206 return -ENOMEM;
207 } while (p4d++, addr = next, addr != end);
208 return 0;
209}
210
Nick Piggindb64fe02008-10-18 20:27:03 -0700211/*
212 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
213 * will have pfns corresponding to the "pages" array.
214 *
215 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
216 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900217static int vmap_page_range_noflush(unsigned long start, unsigned long end,
218 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 pgd_t *pgd;
221 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800222 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700223 int err = 0;
224 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
226 BUG_ON(addr >= end);
227 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 do {
229 next = pgd_addr_end(addr, end);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300230 err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700232 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700234
Nick Piggindb64fe02008-10-18 20:27:03 -0700235 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
Tejun Heo8fc48982009-02-20 16:29:08 +0900238static int vmap_page_range(unsigned long start, unsigned long end,
239 pgprot_t prot, struct page **pages)
240{
241 int ret;
242
243 ret = vmap_page_range_noflush(start, end, prot, pages);
244 flush_cache_vmap(start, end);
245 return ret;
246}
247
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700248int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700249{
250 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000251 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700252 * and fall back on vmalloc() if that fails. Others
253 * just put it in the vmalloc space.
254 */
255#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
256 unsigned long addr = (unsigned long)x;
257 if (addr >= MODULES_VADDR && addr < MODULES_END)
258 return 1;
259#endif
260 return is_vmalloc_addr(x);
261}
262
Christoph Lameter48667e72008-02-04 22:28:31 -0800263/*
malcadd688f2014-01-27 17:06:53 -0800264 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800265 */
malcadd688f2014-01-27 17:06:53 -0800266struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800267{
268 unsigned long addr = (unsigned long) vmalloc_addr;
malcadd688f2014-01-27 17:06:53 -0800269 struct page *page = NULL;
Christoph Lameter48667e72008-02-04 22:28:31 -0800270 pgd_t *pgd = pgd_offset_k(addr);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300271 p4d_t *p4d;
272 pud_t *pud;
273 pmd_t *pmd;
274 pte_t *ptep, pte;
Christoph Lameter48667e72008-02-04 22:28:31 -0800275
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200276 /*
277 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
278 * architectures that do not vmalloc module space
279 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700280 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200281
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300282 if (pgd_none(*pgd))
283 return NULL;
284 p4d = p4d_offset(pgd, addr);
285 if (p4d_none(*p4d))
286 return NULL;
287 pud = pud_offset(p4d, addr);
Ard Biesheuvel029c54b2017-06-23 15:08:41 -0700288
289 /*
290 * Don't dereference bad PUD or PMD (below) entries. This will also
291 * identify huge mappings, which we may encounter on architectures
292 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
293 * identified as vmalloc addresses by is_vmalloc_addr(), but are
294 * not [unambiguously] associated with a struct page, so there is
295 * no correct value to return for them.
296 */
297 WARN_ON_ONCE(pud_bad(*pud));
298 if (pud_none(*pud) || pud_bad(*pud))
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300299 return NULL;
300 pmd = pmd_offset(pud, addr);
Ard Biesheuvel029c54b2017-06-23 15:08:41 -0700301 WARN_ON_ONCE(pmd_bad(*pmd));
302 if (pmd_none(*pmd) || pmd_bad(*pmd))
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300303 return NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700304
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300305 ptep = pte_offset_map(pmd, addr);
306 pte = *ptep;
307 if (pte_present(pte))
308 page = pte_page(pte);
309 pte_unmap(ptep);
malcadd688f2014-01-27 17:06:53 -0800310 return page;
Jianyu Zhanece86e222014-01-21 15:49:12 -0800311}
312EXPORT_SYMBOL(vmalloc_to_page);
313
malcadd688f2014-01-27 17:06:53 -0800314/*
315 * Map a vmalloc()-space virtual address to the physical page frame number.
316 */
317unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
318{
319 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
320}
321EXPORT_SYMBOL(vmalloc_to_pfn);
322
Nick Piggindb64fe02008-10-18 20:27:03 -0700323
324/*** Global kva allocator ***/
325
Yisheng Xie78c72742017-07-10 15:48:09 -0700326#define VM_LAZY_FREE 0x02
Nick Piggindb64fe02008-10-18 20:27:03 -0700327#define VM_VM_AREA 0x04
328
Nick Piggindb64fe02008-10-18 20:27:03 -0700329static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700330/* Export for kexec only */
331LIST_HEAD(vmap_area_list);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700332static LLIST_HEAD(vmap_purge_list);
Nick Piggin89699602011-03-22 16:30:36 -0700333static struct rb_root vmap_area_root = RB_ROOT;
334
335/* The vmap cache globals are protected by vmap_area_lock */
336static struct rb_node *free_vmap_cache;
337static unsigned long cached_hole_size;
338static unsigned long cached_vstart;
339static unsigned long cached_align;
340
Tejun Heoca23e402009-08-14 15:00:52 +0900341static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700342
343static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
Nick Piggindb64fe02008-10-18 20:27:03 -0700345 struct rb_node *n = vmap_area_root.rb_node;
346
347 while (n) {
348 struct vmap_area *va;
349
350 va = rb_entry(n, struct vmap_area, rb_node);
351 if (addr < va->va_start)
352 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700353 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700354 n = n->rb_right;
355 else
356 return va;
357 }
358
359 return NULL;
360}
361
362static void __insert_vmap_area(struct vmap_area *va)
363{
364 struct rb_node **p = &vmap_area_root.rb_node;
365 struct rb_node *parent = NULL;
366 struct rb_node *tmp;
367
368 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700369 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700370
371 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700372 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
373 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700374 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700375 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700376 p = &(*p)->rb_right;
377 else
378 BUG();
379 }
380
381 rb_link_node(&va->rb_node, parent, p);
382 rb_insert_color(&va->rb_node, &vmap_area_root);
383
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700384 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700385 tmp = rb_prev(&va->rb_node);
386 if (tmp) {
387 struct vmap_area *prev;
388 prev = rb_entry(tmp, struct vmap_area, rb_node);
389 list_add_rcu(&va->list, &prev->list);
390 } else
391 list_add_rcu(&va->list, &vmap_area_list);
392}
393
394static void purge_vmap_area_lazy(void);
395
Chris Wilson4da56b92016-04-04 14:46:42 +0100396static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
397
Nick Piggindb64fe02008-10-18 20:27:03 -0700398/*
399 * Allocate a region of KVA of the specified size and alignment, within the
400 * vstart and vend.
401 */
402static struct vmap_area *alloc_vmap_area(unsigned long size,
403 unsigned long align,
404 unsigned long vstart, unsigned long vend,
405 int node, gfp_t gfp_mask)
406{
407 struct vmap_area *va;
408 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700410 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700411 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700412
Nick Piggin77669702009-02-27 14:03:03 -0800413 BUG_ON(!size);
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800414 BUG_ON(offset_in_page(size));
Nick Piggin89699602011-03-22 16:30:36 -0700415 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700416
Christoph Hellwig5803ed22016-12-12 16:44:20 -0800417 might_sleep();
Chris Wilson4da56b92016-04-04 14:46:42 +0100418
Nick Piggindb64fe02008-10-18 20:27:03 -0700419 va = kmalloc_node(sizeof(struct vmap_area),
420 gfp_mask & GFP_RECLAIM_MASK, node);
421 if (unlikely(!va))
422 return ERR_PTR(-ENOMEM);
423
Catalin Marinas7f88f882013-11-12 15:07:45 -0800424 /*
425 * Only scan the relevant parts containing pointers to other objects
426 * to avoid false negatives.
427 */
428 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
429
Nick Piggindb64fe02008-10-18 20:27:03 -0700430retry:
431 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700432 /*
433 * Invalidate cache if we have more permissive parameters.
434 * cached_hole_size notes the largest hole noticed _below_
435 * the vmap_area cached in free_vmap_cache: if size fits
436 * into that hole, we want to scan from vstart to reuse
437 * the hole instead of allocating above free_vmap_cache.
438 * Note that __free_vmap_area may update free_vmap_cache
439 * without updating cached_hole_size or cached_align.
440 */
441 if (!free_vmap_cache ||
442 size < cached_hole_size ||
443 vstart < cached_vstart ||
444 align < cached_align) {
445nocache:
446 cached_hole_size = 0;
447 free_vmap_cache = NULL;
448 }
449 /* record if we encounter less permissive parameters */
450 cached_vstart = vstart;
451 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800452
Nick Piggin89699602011-03-22 16:30:36 -0700453 /* find starting point for our search */
454 if (free_vmap_cache) {
455 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700456 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700457 if (addr < vstart)
458 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700459 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700460 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700461
Nick Piggin89699602011-03-22 16:30:36 -0700462 } else {
463 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700464 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700465 goto overflow;
466
467 n = vmap_area_root.rb_node;
468 first = NULL;
469
470 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700471 struct vmap_area *tmp;
472 tmp = rb_entry(n, struct vmap_area, rb_node);
473 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700474 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700475 if (tmp->va_start <= addr)
476 break;
477 n = n->rb_left;
478 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700479 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700480 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700481
482 if (!first)
483 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700484 }
Nick Piggin89699602011-03-22 16:30:36 -0700485
486 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700487 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700488 if (addr + cached_hole_size < first->va_start)
489 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700490 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700491 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700492 goto overflow;
493
Hong zhi guo92ca9222012-07-31 16:41:35 -0700494 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700495 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700496
Geliang Tang6219c2a2016-01-14 15:19:08 -0800497 first = list_next_entry(first, list);
Nick Piggin89699602011-03-22 16:30:36 -0700498 }
499
Nick Piggindb64fe02008-10-18 20:27:03 -0700500found:
Nick Piggin89699602011-03-22 16:30:36 -0700501 if (addr + size > vend)
502 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700503
504 va->va_start = addr;
505 va->va_end = addr + size;
506 va->flags = 0;
507 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700508 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700509 spin_unlock(&vmap_area_lock);
510
Wang Xiaoqiang61e16552016-01-15 16:57:19 -0800511 BUG_ON(!IS_ALIGNED(va->va_start, align));
Nick Piggin89699602011-03-22 16:30:36 -0700512 BUG_ON(va->va_start < vstart);
513 BUG_ON(va->va_end > vend);
514
Nick Piggindb64fe02008-10-18 20:27:03 -0700515 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700516
517overflow:
518 spin_unlock(&vmap_area_lock);
519 if (!purged) {
520 purge_vmap_area_lazy();
521 purged = 1;
522 goto retry;
523 }
Chris Wilson4da56b92016-04-04 14:46:42 +0100524
525 if (gfpflags_allow_blocking(gfp_mask)) {
526 unsigned long freed = 0;
527 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
528 if (freed > 0) {
529 purged = 0;
530 goto retry;
531 }
532 }
533
Florian Fainelli03497d72017-04-27 11:19:00 -0700534 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
Joe Perches756a0252016-03-17 14:19:47 -0700535 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
536 size);
Nick Piggin89699602011-03-22 16:30:36 -0700537 kfree(va);
538 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700539}
540
Chris Wilson4da56b92016-04-04 14:46:42 +0100541int register_vmap_purge_notifier(struct notifier_block *nb)
542{
543 return blocking_notifier_chain_register(&vmap_notify_list, nb);
544}
545EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
546
547int unregister_vmap_purge_notifier(struct notifier_block *nb)
548{
549 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
550}
551EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
552
Nick Piggindb64fe02008-10-18 20:27:03 -0700553static void __free_vmap_area(struct vmap_area *va)
554{
555 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700556
557 if (free_vmap_cache) {
558 if (va->va_end < cached_vstart) {
559 free_vmap_cache = NULL;
560 } else {
561 struct vmap_area *cache;
562 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
563 if (va->va_start <= cache->va_start) {
564 free_vmap_cache = rb_prev(&va->rb_node);
565 /*
566 * We don't try to update cached_hole_size or
567 * cached_align, but it won't go very wrong.
568 */
569 }
570 }
571 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700572 rb_erase(&va->rb_node, &vmap_area_root);
573 RB_CLEAR_NODE(&va->rb_node);
574 list_del_rcu(&va->list);
575
Tejun Heoca23e402009-08-14 15:00:52 +0900576 /*
577 * Track the highest possible candidate for pcpu area
578 * allocation. Areas outside of vmalloc area can be returned
579 * here too, consider only end addresses which fall inside
580 * vmalloc area proper.
581 */
582 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
583 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
584
Lai Jiangshan14769de2011-03-18 12:12:19 +0800585 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700586}
587
588/*
589 * Free a region of KVA allocated by alloc_vmap_area
590 */
591static void free_vmap_area(struct vmap_area *va)
592{
593 spin_lock(&vmap_area_lock);
594 __free_vmap_area(va);
595 spin_unlock(&vmap_area_lock);
596}
597
598/*
599 * Clear the pagetable entries of a given vmap_area
600 */
601static void unmap_vmap_area(struct vmap_area *va)
602{
603 vunmap_page_range(va->va_start, va->va_end);
604}
605
Nick Piggincd528582009-01-06 14:39:20 -0800606static void vmap_debug_free_range(unsigned long start, unsigned long end)
607{
608 /*
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700609 * Unmap page tables and force a TLB flush immediately if pagealloc
610 * debugging is enabled. This catches use after free bugs similarly to
611 * those in linear kernel virtual address space after a page has been
612 * freed.
Nick Piggincd528582009-01-06 14:39:20 -0800613 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700614 * All the lazy freeing logic is still retained, in order to minimise
615 * intrusiveness of this debugging feature.
Nick Piggincd528582009-01-06 14:39:20 -0800616 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700617 * This is going to be *slow* (linear kernel virtual address debugging
618 * doesn't do a broadcast TLB flush so it is a lot faster).
Nick Piggincd528582009-01-06 14:39:20 -0800619 */
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700620 if (debug_pagealloc_enabled()) {
621 vunmap_page_range(start, end);
622 flush_tlb_kernel_range(start, end);
623 }
Nick Piggincd528582009-01-06 14:39:20 -0800624}
625
Nick Piggindb64fe02008-10-18 20:27:03 -0700626/*
627 * lazy_max_pages is the maximum amount of virtual address space we gather up
628 * before attempting to purge with a TLB flush.
629 *
630 * There is a tradeoff here: a larger number will cover more kernel page tables
631 * and take slightly longer to purge, but it will linearly reduce the number of
632 * global TLB flushes that must be performed. It would seem natural to scale
633 * this number up linearly with the number of CPUs (because vmapping activity
634 * could also scale linearly with the number of CPUs), however it is likely
635 * that in practice, workloads might be constrained in other ways that mean
636 * vmap activity will not scale linearly with CPUs. Also, I want to be
637 * conservative and not introduce a big latency on huge systems, so go with
638 * a less aggressive log scale. It will still be an improvement over the old
639 * code, and it will be simple to change the scale factor if we find that it
640 * becomes a problem on bigger systems.
641 */
642static unsigned long lazy_max_pages(void)
643{
644 unsigned int log;
645
646 log = fls(num_online_cpus());
647
648 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
649}
650
651static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
652
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800653/*
654 * Serialize vmap purging. There is no actual criticial section protected
655 * by this look, but we want to avoid concurrent calls for performance
656 * reasons and to make the pcpu_get_vm_areas more deterministic.
657 */
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800658static DEFINE_MUTEX(vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800659
Nick Piggin02b709d2010-02-01 22:25:57 +1100660/* for per-CPU blocks */
661static void purge_fragmented_blocks_allcpus(void);
662
Nick Piggindb64fe02008-10-18 20:27:03 -0700663/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500664 * called before a call to iounmap() if the caller wants vm_area_struct's
665 * immediately freed.
666 */
667void set_iounmap_nonlazy(void)
668{
669 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
670}
671
672/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700673 * Purges all lazily-freed vmap areas.
Nick Piggindb64fe02008-10-18 20:27:03 -0700674 */
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800675static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700676{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700677 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -0700678 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800679 struct vmap_area *n_va;
Joel Fernandes763b2182016-12-12 16:44:26 -0800680 bool do_free = false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700681
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800682 lockdep_assert_held(&vmap_purge_lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100683
Chris Wilson80c4bd72016-05-20 16:57:38 -0700684 valist = llist_del_all(&vmap_purge_list);
685 llist_for_each_entry(va, valist, purge_list) {
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800686 if (va->va_start < start)
687 start = va->va_start;
688 if (va->va_end > end)
689 end = va->va_end;
Joel Fernandes763b2182016-12-12 16:44:26 -0800690 do_free = true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700691 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700692
Joel Fernandes763b2182016-12-12 16:44:26 -0800693 if (!do_free)
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800694 return false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700695
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800696 flush_tlb_kernel_range(start, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700697
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800698 spin_lock(&vmap_area_lock);
Joel Fernandes763b2182016-12-12 16:44:26 -0800699 llist_for_each_entry_safe(va, n_va, valist, purge_list) {
700 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
701
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800702 __free_vmap_area(va);
Joel Fernandes763b2182016-12-12 16:44:26 -0800703 atomic_sub(nr, &vmap_lazy_nr);
704 cond_resched_lock(&vmap_area_lock);
705 }
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800706 spin_unlock(&vmap_area_lock);
707 return true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700708}
709
710/*
Nick Piggin496850e2008-11-19 15:36:33 -0800711 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
712 * is already purging.
713 */
714static void try_purge_vmap_area_lazy(void)
715{
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800716 if (mutex_trylock(&vmap_purge_lock)) {
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800717 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800718 mutex_unlock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800719 }
Nick Piggin496850e2008-11-19 15:36:33 -0800720}
721
722/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700723 * Kick off a purge of the outstanding lazy areas.
724 */
725static void purge_vmap_area_lazy(void)
726{
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800727 mutex_lock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800728 purge_fragmented_blocks_allcpus();
729 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800730 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700731}
732
733/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800734 * Free a vmap area, caller ensuring that the area has been unmapped
735 * and flush_cache_vunmap had been called for the correct range
736 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700737 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800738static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700739{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700740 int nr_lazy;
741
742 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
743 &vmap_lazy_nr);
744
745 /* After this point, we may free va at any time */
746 llist_add(&va->purge_list, &vmap_purge_list);
747
748 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800749 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700750}
751
Nick Pigginb29acbd2008-12-01 13:13:47 -0800752/*
753 * Free and unmap a vmap area
754 */
755static void free_unmap_vmap_area(struct vmap_area *va)
756{
757 flush_cache_vunmap(va->va_start, va->va_end);
Christoph Hellwigc8eef012016-12-12 16:44:01 -0800758 unmap_vmap_area(va);
759 free_vmap_area_noflush(va);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800760}
761
Nick Piggindb64fe02008-10-18 20:27:03 -0700762static struct vmap_area *find_vmap_area(unsigned long addr)
763{
764 struct vmap_area *va;
765
766 spin_lock(&vmap_area_lock);
767 va = __find_vmap_area(addr);
768 spin_unlock(&vmap_area_lock);
769
770 return va;
771}
772
Nick Piggindb64fe02008-10-18 20:27:03 -0700773/*** Per cpu kva allocator ***/
774
775/*
776 * vmap space is limited especially on 32 bit architectures. Ensure there is
777 * room for at least 16 percpu vmap blocks per CPU.
778 */
779/*
780 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
781 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
782 * instead (we just need a rough idea)
783 */
784#if BITS_PER_LONG == 32
785#define VMALLOC_SPACE (128UL*1024*1024)
786#else
787#define VMALLOC_SPACE (128UL*1024*1024*1024)
788#endif
789
790#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
791#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
792#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
793#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
794#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
795#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200796#define VMAP_BBMAP_BITS \
797 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
798 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
799 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700800
801#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
802
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100803static bool vmap_initialized __read_mostly = false;
804
Nick Piggindb64fe02008-10-18 20:27:03 -0700805struct vmap_block_queue {
806 spinlock_t lock;
807 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700808};
809
810struct vmap_block {
811 spinlock_t lock;
812 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700813 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700814 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100815 struct list_head free_list;
816 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100817 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700818};
819
820/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
821static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
822
823/*
824 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
825 * in the free path. Could get rid of this if we change the API to return a
826 * "cookie" from alloc, to be passed to free. But no big deal yet.
827 */
828static DEFINE_SPINLOCK(vmap_block_tree_lock);
829static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
830
831/*
832 * We should probably have a fallback mechanism to allocate virtual memory
833 * out of partially filled vmap blocks. However vmap block sizing should be
834 * fairly reasonable according to the vmalloc size, so it shouldn't be a
835 * big problem.
836 */
837
838static unsigned long addr_to_vb_idx(unsigned long addr)
839{
840 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
841 addr /= VMAP_BLOCK_SIZE;
842 return addr;
843}
844
Roman Pencf725ce2015-04-15 16:13:52 -0700845static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
846{
847 unsigned long addr;
848
849 addr = va_start + (pages_off << PAGE_SHIFT);
850 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
851 return (void *)addr;
852}
853
854/**
855 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
856 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
857 * @order: how many 2^order pages should be occupied in newly allocated block
858 * @gfp_mask: flags for the page level allocator
859 *
860 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
861 */
862static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700863{
864 struct vmap_block_queue *vbq;
865 struct vmap_block *vb;
866 struct vmap_area *va;
867 unsigned long vb_idx;
868 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700869 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700870
871 node = numa_node_id();
872
873 vb = kmalloc_node(sizeof(struct vmap_block),
874 gfp_mask & GFP_RECLAIM_MASK, node);
875 if (unlikely(!vb))
876 return ERR_PTR(-ENOMEM);
877
878 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
879 VMALLOC_START, VMALLOC_END,
880 node, gfp_mask);
Tobias Klauserddf9c6d42011-01-13 15:46:15 -0800881 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700882 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700883 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700884 }
885
886 err = radix_tree_preload(gfp_mask);
887 if (unlikely(err)) {
888 kfree(vb);
889 free_vmap_area(va);
890 return ERR_PTR(err);
891 }
892
Roman Pencf725ce2015-04-15 16:13:52 -0700893 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700894 spin_lock_init(&vb->lock);
895 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700896 /* At least something should be left free */
897 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
898 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700899 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700900 vb->dirty_min = VMAP_BBMAP_BITS;
901 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700902 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700903
904 vb_idx = addr_to_vb_idx(va->va_start);
905 spin_lock(&vmap_block_tree_lock);
906 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
907 spin_unlock(&vmap_block_tree_lock);
908 BUG_ON(err);
909 radix_tree_preload_end();
910
911 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700912 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700913 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700914 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900915 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700916
Roman Pencf725ce2015-04-15 16:13:52 -0700917 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700918}
919
Nick Piggindb64fe02008-10-18 20:27:03 -0700920static void free_vmap_block(struct vmap_block *vb)
921{
922 struct vmap_block *tmp;
923 unsigned long vb_idx;
924
Nick Piggindb64fe02008-10-18 20:27:03 -0700925 vb_idx = addr_to_vb_idx(vb->va->va_start);
926 spin_lock(&vmap_block_tree_lock);
927 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
928 spin_unlock(&vmap_block_tree_lock);
929 BUG_ON(tmp != vb);
930
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800931 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800932 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700933}
934
Nick Piggin02b709d2010-02-01 22:25:57 +1100935static void purge_fragmented_blocks(int cpu)
936{
937 LIST_HEAD(purge);
938 struct vmap_block *vb;
939 struct vmap_block *n_vb;
940 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
941
942 rcu_read_lock();
943 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
944
945 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
946 continue;
947
948 spin_lock(&vb->lock);
949 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
950 vb->free = 0; /* prevent further allocs after releasing lock */
951 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700952 vb->dirty_min = 0;
953 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100954 spin_lock(&vbq->lock);
955 list_del_rcu(&vb->free_list);
956 spin_unlock(&vbq->lock);
957 spin_unlock(&vb->lock);
958 list_add_tail(&vb->purge, &purge);
959 } else
960 spin_unlock(&vb->lock);
961 }
962 rcu_read_unlock();
963
964 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
965 list_del(&vb->purge);
966 free_vmap_block(vb);
967 }
968}
969
Nick Piggin02b709d2010-02-01 22:25:57 +1100970static void purge_fragmented_blocks_allcpus(void)
971{
972 int cpu;
973
974 for_each_possible_cpu(cpu)
975 purge_fragmented_blocks(cpu);
976}
977
Nick Piggindb64fe02008-10-18 20:27:03 -0700978static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
979{
980 struct vmap_block_queue *vbq;
981 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700982 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700983 unsigned int order;
984
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800985 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700986 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700987 if (WARN_ON(size == 0)) {
988 /*
989 * Allocating 0 bytes isn't what caller wants since
990 * get_order(0) returns funny result. Just warn and terminate
991 * early.
992 */
993 return NULL;
994 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700995 order = get_order(size);
996
Nick Piggindb64fe02008-10-18 20:27:03 -0700997 rcu_read_lock();
998 vbq = &get_cpu_var(vmap_block_queue);
999 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -07001000 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -07001001
1002 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -07001003 if (vb->free < (1UL << order)) {
1004 spin_unlock(&vb->lock);
1005 continue;
1006 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001007
Roman Pencf725ce2015-04-15 16:13:52 -07001008 pages_off = VMAP_BBMAP_BITS - vb->free;
1009 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +11001010 vb->free -= 1UL << order;
1011 if (vb->free == 0) {
1012 spin_lock(&vbq->lock);
1013 list_del_rcu(&vb->free_list);
1014 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001015 }
Roman Pencf725ce2015-04-15 16:13:52 -07001016
Nick Piggindb64fe02008-10-18 20:27:03 -07001017 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001018 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001019 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001020
Tejun Heo3f04ba82009-10-29 22:34:12 +09001021 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001022 rcu_read_unlock();
1023
Roman Pencf725ce2015-04-15 16:13:52 -07001024 /* Allocate new block if nothing was found */
1025 if (!vaddr)
1026 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001027
Roman Pencf725ce2015-04-15 16:13:52 -07001028 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001029}
1030
1031static void vb_free(const void *addr, unsigned long size)
1032{
1033 unsigned long offset;
1034 unsigned long vb_idx;
1035 unsigned int order;
1036 struct vmap_block *vb;
1037
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001038 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001039 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001040
1041 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1042
Nick Piggindb64fe02008-10-18 20:27:03 -07001043 order = get_order(size);
1044
1045 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001046 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001047
1048 vb_idx = addr_to_vb_idx((unsigned long)addr);
1049 rcu_read_lock();
1050 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1051 rcu_read_unlock();
1052 BUG_ON(!vb);
1053
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001054 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1055
Nick Piggindb64fe02008-10-18 20:27:03 -07001056 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001057
1058 /* Expand dirty range */
1059 vb->dirty_min = min(vb->dirty_min, offset);
1060 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001061
Nick Piggindb64fe02008-10-18 20:27:03 -07001062 vb->dirty += 1UL << order;
1063 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001064 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001065 spin_unlock(&vb->lock);
1066 free_vmap_block(vb);
1067 } else
1068 spin_unlock(&vb->lock);
1069}
1070
1071/**
1072 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1073 *
1074 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1075 * to amortize TLB flushing overheads. What this means is that any page you
1076 * have now, may, in a former life, have been mapped into kernel virtual
1077 * address by the vmap layer and so there might be some CPUs with TLB entries
1078 * still referencing that page (additional to the regular 1:1 kernel mapping).
1079 *
1080 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1081 * be sure that none of the pages we have control over will have any aliases
1082 * from the vmap layer.
1083 */
1084void vm_unmap_aliases(void)
1085{
1086 unsigned long start = ULONG_MAX, end = 0;
1087 int cpu;
1088 int flush = 0;
1089
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001090 if (unlikely(!vmap_initialized))
1091 return;
1092
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001093 might_sleep();
1094
Nick Piggindb64fe02008-10-18 20:27:03 -07001095 for_each_possible_cpu(cpu) {
1096 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1097 struct vmap_block *vb;
1098
1099 rcu_read_lock();
1100 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001101 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001102 if (vb->dirty) {
1103 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001104 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001105
Roman Pen7d61bfe2015-04-15 16:13:55 -07001106 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1107 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001108
Roman Pen7d61bfe2015-04-15 16:13:55 -07001109 start = min(s, start);
1110 end = max(e, end);
1111
Nick Piggindb64fe02008-10-18 20:27:03 -07001112 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001113 }
1114 spin_unlock(&vb->lock);
1115 }
1116 rcu_read_unlock();
1117 }
1118
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001119 mutex_lock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001120 purge_fragmented_blocks_allcpus();
1121 if (!__purge_vmap_area_lazy(start, end) && flush)
1122 flush_tlb_kernel_range(start, end);
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001123 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001124}
1125EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1126
1127/**
1128 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1129 * @mem: the pointer returned by vm_map_ram
1130 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1131 */
1132void vm_unmap_ram(const void *mem, unsigned int count)
1133{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001134 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001135 unsigned long addr = (unsigned long)mem;
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001136 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001137
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001138 might_sleep();
Nick Piggindb64fe02008-10-18 20:27:03 -07001139 BUG_ON(!addr);
1140 BUG_ON(addr < VMALLOC_START);
1141 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001142 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001143
1144 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001145 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001146
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001147 if (likely(count <= VMAP_MAX_ALLOC)) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001148 vb_free(mem, size);
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001149 return;
1150 }
1151
1152 va = find_vmap_area(addr);
1153 BUG_ON(!va);
1154 free_unmap_vmap_area(va);
Nick Piggindb64fe02008-10-18 20:27:03 -07001155}
1156EXPORT_SYMBOL(vm_unmap_ram);
1157
1158/**
1159 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1160 * @pages: an array of pointers to the pages to be mapped
1161 * @count: number of pages
1162 * @node: prefer to allocate data structures on this node
1163 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001164 *
Gioh Kim36437632014-04-07 15:37:37 -07001165 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1166 * faster than vmap so it's good. But if you mix long-life and short-life
1167 * objects with vm_map_ram(), it could consume lots of address space through
1168 * fragmentation (especially on a 32bit machine). You could see failures in
1169 * the end. Please use this function for short-lived objects.
1170 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001171 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001172 */
1173void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1174{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001175 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001176 unsigned long addr;
1177 void *mem;
1178
1179 if (likely(count <= VMAP_MAX_ALLOC)) {
1180 mem = vb_alloc(size, GFP_KERNEL);
1181 if (IS_ERR(mem))
1182 return NULL;
1183 addr = (unsigned long)mem;
1184 } else {
1185 struct vmap_area *va;
1186 va = alloc_vmap_area(size, PAGE_SIZE,
1187 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1188 if (IS_ERR(va))
1189 return NULL;
1190
1191 addr = va->va_start;
1192 mem = (void *)addr;
1193 }
1194 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1195 vm_unmap_ram(mem, count);
1196 return NULL;
1197 }
1198 return mem;
1199}
1200EXPORT_SYMBOL(vm_map_ram);
1201
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001202static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001203/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001204 * vm_area_add_early - add vmap area early during boot
1205 * @vm: vm_struct to add
1206 *
1207 * This function is used to add fixed kernel vm area to vmlist before
1208 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1209 * should contain proper values and the other fields should be zero.
1210 *
1211 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1212 */
1213void __init vm_area_add_early(struct vm_struct *vm)
1214{
1215 struct vm_struct *tmp, **p;
1216
1217 BUG_ON(vmap_initialized);
1218 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1219 if (tmp->addr >= vm->addr) {
1220 BUG_ON(tmp->addr < vm->addr + vm->size);
1221 break;
1222 } else
1223 BUG_ON(tmp->addr + tmp->size > vm->addr);
1224 }
1225 vm->next = *p;
1226 *p = vm;
1227}
1228
1229/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001230 * vm_area_register_early - register vmap area early during boot
1231 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001232 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001233 *
1234 * This function is used to register kernel vm area before
1235 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1236 * proper values on entry and other fields should be zero. On return,
1237 * vm->addr contains the allocated address.
1238 *
1239 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1240 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001241void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001242{
1243 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001244 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001245
Tejun Heoc0c0a292009-02-24 11:57:21 +09001246 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1247 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1248
1249 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001250
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001251 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001252}
1253
Nick Piggindb64fe02008-10-18 20:27:03 -07001254void __init vmalloc_init(void)
1255{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001256 struct vmap_area *va;
1257 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001258 int i;
1259
1260 for_each_possible_cpu(i) {
1261 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001262 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001263
1264 vbq = &per_cpu(vmap_block_queue, i);
1265 spin_lock_init(&vbq->lock);
1266 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001267 p = &per_cpu(vfree_deferred, i);
1268 init_llist_head(&p->list);
1269 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001270 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001271
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001272 /* Import existing vmlist entries. */
1273 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001274 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001275 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001276 va->va_start = (unsigned long)tmp->addr;
1277 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001278 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001279 __insert_vmap_area(va);
1280 }
Tejun Heoca23e402009-08-14 15:00:52 +09001281
1282 vmap_area_pcpu_hole = VMALLOC_END;
1283
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001284 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001285}
1286
Tejun Heo8fc48982009-02-20 16:29:08 +09001287/**
1288 * map_kernel_range_noflush - map kernel VM area with the specified pages
1289 * @addr: start of the VM area to map
1290 * @size: size of the VM area to map
1291 * @prot: page protection flags to use
1292 * @pages: pages to map
1293 *
1294 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1295 * specify should have been allocated using get_vm_area() and its
1296 * friends.
1297 *
1298 * NOTE:
1299 * This function does NOT do any cache flushing. The caller is
1300 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1301 * before calling this function.
1302 *
1303 * RETURNS:
1304 * The number of pages mapped on success, -errno on failure.
1305 */
1306int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1307 pgprot_t prot, struct page **pages)
1308{
1309 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1310}
1311
1312/**
1313 * unmap_kernel_range_noflush - unmap kernel VM area
1314 * @addr: start of the VM area to unmap
1315 * @size: size of the VM area to unmap
1316 *
1317 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1318 * specify should have been allocated using get_vm_area() and its
1319 * friends.
1320 *
1321 * NOTE:
1322 * This function does NOT do any cache flushing. The caller is
1323 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1324 * before calling this function and flush_tlb_kernel_range() after.
1325 */
1326void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1327{
1328 vunmap_page_range(addr, addr + size);
1329}
Huang Ying81e88fd2011-01-12 14:44:55 +08001330EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001331
1332/**
1333 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1334 * @addr: start of the VM area to unmap
1335 * @size: size of the VM area to unmap
1336 *
1337 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1338 * the unmapping and tlb after.
1339 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001340void unmap_kernel_range(unsigned long addr, unsigned long size)
1341{
1342 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001343
1344 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001345 vunmap_page_range(addr, end);
1346 flush_tlb_kernel_range(addr, end);
1347}
Minchan Kim93ef6d6c2014-06-04 16:11:09 -07001348EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001349
WANG Chaof6f8ed42014-08-06 16:06:58 -07001350int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001351{
1352 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001353 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001354 int err;
1355
WANG Chaof6f8ed42014-08-06 16:06:58 -07001356 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001357
WANG Chaof6f8ed42014-08-06 16:06:58 -07001358 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001359}
1360EXPORT_SYMBOL_GPL(map_vm_area);
1361
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001362static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001363 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001364{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001365 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001366 vm->flags = flags;
1367 vm->addr = (void *)va->va_start;
1368 vm->size = va->va_end - va->va_start;
1369 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001370 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001371 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001372 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001373}
Tejun Heocf88c792009-08-14 15:00:52 +09001374
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001375static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001376{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001377 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001378 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001379 * we should make sure that vm has proper values.
1380 * Pair with smp_rmb() in show_numa_info().
1381 */
1382 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001383 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001384}
1385
Nick Piggindb64fe02008-10-18 20:27:03 -07001386static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001387 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001388 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001389{
Kautuk Consul00065262011-12-19 17:12:04 -08001390 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001391 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001393 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001395 if (unlikely(!size))
1396 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
zijun_hu252e5c62016-10-07 16:57:26 -07001398 if (flags & VM_IOREMAP)
1399 align = 1ul << clamp_t(int, get_count_order_long(size),
1400 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1401
Tejun Heocf88c792009-08-14 15:00:52 +09001402 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (unlikely(!area))
1404 return NULL;
1405
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001406 if (!(flags & VM_NO_GUARD))
1407 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Nick Piggindb64fe02008-10-18 20:27:03 -07001409 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1410 if (IS_ERR(va)) {
1411 kfree(area);
1412 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001415 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418}
1419
Christoph Lameter930fc452005-10-29 18:15:41 -07001420struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1421 unsigned long start, unsigned long end)
1422{
David Rientjes00ef2d22013-02-22 16:35:36 -08001423 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1424 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001425}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001426EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001427
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001428struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1429 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001430 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001431{
David Rientjes00ef2d22013-02-22 16:35:36 -08001432 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1433 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001434}
1435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436/**
Simon Arlott183ff222007-10-20 01:27:18 +02001437 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 * @size: size of the area
1439 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1440 *
1441 * Search an area of @size in the kernel virtual mapping area,
1442 * and reserved it for out purposes. Returns the area descriptor
1443 * on success or %NULL on failure.
1444 */
1445struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1446{
David Miller2dca6992009-09-21 12:22:34 -07001447 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001448 NUMA_NO_NODE, GFP_KERNEL,
1449 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001450}
1451
1452struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001453 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001454{
David Miller2dca6992009-09-21 12:22:34 -07001455 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001456 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001459/**
1460 * find_vm_area - find a continuous kernel virtual area
1461 * @addr: base address
1462 *
1463 * Search for the kernel VM area starting at @addr, and return it.
1464 * It is up to the caller to do all required locking to keep the returned
1465 * pointer valid.
1466 */
1467struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001468{
Nick Piggindb64fe02008-10-18 20:27:03 -07001469 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001470
Nick Piggindb64fe02008-10-18 20:27:03 -07001471 va = find_vmap_area((unsigned long)addr);
1472 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001473 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001474
Andi Kleen7856dfe2005-05-20 14:27:57 -07001475 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001476}
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478/**
Simon Arlott183ff222007-10-20 01:27:18 +02001479 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 * @addr: base address
1481 *
1482 * Search for the kernel VM area starting at @addr, and remove it.
1483 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001484 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001486struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
Nick Piggindb64fe02008-10-18 20:27:03 -07001488 struct vmap_area *va;
1489
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001490 might_sleep();
1491
Nick Piggindb64fe02008-10-18 20:27:03 -07001492 va = find_vmap_area((unsigned long)addr);
1493 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001494 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001495
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001496 spin_lock(&vmap_area_lock);
1497 va->vm = NULL;
1498 va->flags &= ~VM_VM_AREA;
Yisheng Xie78c72742017-07-10 15:48:09 -07001499 va->flags |= VM_LAZY_FREE;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001500 spin_unlock(&vmap_area_lock);
1501
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001502 vmap_debug_free_range(va->va_start, va->va_end);
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001503 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001504 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001505
Nick Piggindb64fe02008-10-18 20:27:03 -07001506 return vm;
1507 }
1508 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509}
1510
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001511static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
1513 struct vm_struct *area;
1514
1515 if (!addr)
1516 return;
1517
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001518 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001519 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
1522 area = remove_vm_area(addr);
1523 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001524 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 return;
1527 }
1528
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001529 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1530 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 if (deallocate_pages) {
1533 int i;
1534
1535 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001536 struct page *page = area->pages[i];
1537
1538 BUG_ON(!page);
Vladimir Davydov49491482016-07-26 15:24:24 -07001539 __free_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 }
1541
David Rientjes244d63e2016-01-14 15:19:35 -08001542 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 }
1544
1545 kfree(area);
1546 return;
1547}
Andrey Ryabininbf22e372016-12-12 16:44:10 -08001548
1549static inline void __vfree_deferred(const void *addr)
1550{
1551 /*
1552 * Use raw_cpu_ptr() because this can be called from preemptible
1553 * context. Preemption is absolutely fine here, because the llist_add()
1554 * implementation is lockless, so it works even if we are adding to
1555 * nother cpu's list. schedule_work() should be fine with this too.
1556 */
1557 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1558
1559 if (llist_add((struct llist_node *)addr, &p->list))
1560 schedule_work(&p->wq);
1561}
1562
1563/**
1564 * vfree_atomic - release memory allocated by vmalloc()
1565 * @addr: memory base address
1566 *
1567 * This one is just like vfree() but can be called in any atomic context
1568 * except NMIs.
1569 */
1570void vfree_atomic(const void *addr)
1571{
1572 BUG_ON(in_nmi());
1573
1574 kmemleak_free(addr);
1575
1576 if (!addr)
1577 return;
1578 __vfree_deferred(addr);
1579}
1580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581/**
1582 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 * @addr: memory base address
1584 *
Simon Arlott183ff222007-10-20 01:27:18 +02001585 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001586 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1587 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 *
Al Viro32fcfd42013-03-10 20:14:08 -04001589 * Must not be called in NMI context (strictly speaking, only if we don't
1590 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1591 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001592 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03001593 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001595void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
Al Viro32fcfd42013-03-10 20:14:08 -04001597 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001598
1599 kmemleak_free(addr);
1600
Al Viro32fcfd42013-03-10 20:14:08 -04001601 if (!addr)
1602 return;
Andrey Ryabininbf22e372016-12-12 16:44:10 -08001603 if (unlikely(in_interrupt()))
1604 __vfree_deferred(addr);
1605 else
Al Viro32fcfd42013-03-10 20:14:08 -04001606 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608EXPORT_SYMBOL(vfree);
1609
1610/**
1611 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 * @addr: memory base address
1613 *
1614 * Free the virtually contiguous memory area starting at @addr,
1615 * which was created from the page array passed to vmap().
1616 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001617 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001619void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620{
1621 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001622 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001623 if (addr)
1624 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626EXPORT_SYMBOL(vunmap);
1627
1628/**
1629 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 * @pages: array of page pointers
1631 * @count: number of pages to map
1632 * @flags: vm_area->flags
1633 * @prot: page protection for the mapping
1634 *
1635 * Maps @count pages from @pages into contiguous kernel virtual
1636 * space.
1637 */
1638void *vmap(struct page **pages, unsigned int count,
1639 unsigned long flags, pgprot_t prot)
1640{
1641 struct vm_struct *area;
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001642 unsigned long size; /* In bytes */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Peter Zijlstra34754b62009-02-25 16:04:03 +01001644 might_sleep();
1645
Jan Beulich44813742009-09-21 17:03:05 -07001646 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 return NULL;
1648
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001649 size = (unsigned long)count << PAGE_SHIFT;
1650 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (!area)
1652 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001653
WANG Chaof6f8ed42014-08-06 16:06:58 -07001654 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 vunmap(area->addr);
1656 return NULL;
1657 }
1658
1659 return area->addr;
1660}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661EXPORT_SYMBOL(vmap);
1662
Michal Hocko8594a212017-05-12 15:46:41 -07001663static void *__vmalloc_node(unsigned long size, unsigned long align,
1664 gfp_t gfp_mask, pgprot_t prot,
1665 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001666static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001667 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668{
1669 struct page **pages;
1670 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001671 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Laura Abbott704b8622017-08-18 15:16:27 -07001672 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1673 const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1674 0 :
1675 __GFP_HIGHMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Wanpeng Li762216a2013-09-11 14:22:42 -07001677 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 array_size = (nr_pages * sizeof(struct page *));
1679
1680 area->nr_pages = nr_pages;
1681 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001682 if (array_size > PAGE_SIZE) {
Laura Abbott704b8622017-08-18 15:16:27 -07001683 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001684 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001685 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001686 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 area->pages = pages;
1689 if (!area->pages) {
1690 remove_vm_area(area->addr);
1691 kfree(area);
1692 return NULL;
1693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
1695 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001696 struct page *page;
1697
Jianguo Wu4b909512013-11-12 15:07:11 -08001698 if (node == NUMA_NO_NODE)
Laura Abbott704b8622017-08-18 15:16:27 -07001699 page = alloc_page(alloc_mask|highmem_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001700 else
Laura Abbott704b8622017-08-18 15:16:27 -07001701 page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001702
1703 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 /* Successfully allocated i pages, free them in __vunmap() */
1705 area->nr_pages = i;
1706 goto fail;
1707 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001708 area->pages[i] = page;
Laura Abbott704b8622017-08-18 15:16:27 -07001709 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001710 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 }
1712
WANG Chaof6f8ed42014-08-06 16:06:58 -07001713 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 goto fail;
1715 return area->addr;
1716
1717fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08001718 warn_alloc(gfp_mask, NULL,
Michal Hocko7877cdc2016-10-07 17:01:55 -07001719 "vmalloc: allocation failure, allocated %ld of %ld bytes",
Dave Hansen22943ab2011-05-24 17:12:18 -07001720 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 vfree(area->addr);
1722 return NULL;
1723}
1724
David Rientjesd0a21262011-01-13 15:46:02 -08001725/**
1726 * __vmalloc_node_range - allocate virtually contiguous memory
1727 * @size: allocation size
1728 * @align: desired alignment
1729 * @start: vm area range start
1730 * @end: vm area range end
1731 * @gfp_mask: flags for the page level allocator
1732 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001733 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001734 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001735 * @caller: caller's return address
1736 *
1737 * Allocate enough pages to cover @size from the page level
1738 * allocator with @gfp_mask flags. Map them into contiguous
1739 * kernel virtual space, using a pagetable protection of @prot.
1740 */
1741void *__vmalloc_node_range(unsigned long size, unsigned long align,
1742 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001743 pgprot_t prot, unsigned long vm_flags, int node,
1744 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001745{
David Rientjesd0a21262011-01-13 15:46:02 -08001746 struct vm_struct *area;
1747 void *addr;
1748 unsigned long real_size = size;
1749
1750 size = PAGE_ALIGN(size);
1751 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001752 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001753
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001754 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1755 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001756 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001757 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001758
Wanpeng Li3722e132013-11-12 15:07:29 -08001759 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001760 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001761 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001762
1763 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001764 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1765 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001766 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001767 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001768 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001769
Catalin Marinas94f4a162017-07-06 15:40:22 -07001770 kmemleak_vmalloc(area, size, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001771
1772 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001773
1774fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08001775 warn_alloc(gfp_mask, NULL,
Michal Hocko7877cdc2016-10-07 17:01:55 -07001776 "vmalloc: allocation failure: %lu bytes", real_size);
Joe Perchesde7d2b52011-10-31 17:08:48 -07001777 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001778}
1779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001781 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001783 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 * @gfp_mask: flags for the page level allocator
1785 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001786 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001787 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 *
1789 * Allocate enough pages to cover @size from the page level
1790 * allocator with @gfp_mask flags. Map them into contiguous
1791 * kernel virtual space, using a pagetable protection of @prot.
Michal Hockoa7c3e902017-05-08 15:57:09 -07001792 *
Michal Hockodcda9b02017-07-12 14:36:45 -07001793 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
Michal Hockoa7c3e902017-05-08 15:57:09 -07001794 * and __GFP_NOFAIL are not supported
1795 *
1796 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1797 * with mm people.
1798 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 */
Michal Hocko8594a212017-05-12 15:46:41 -07001800static void *__vmalloc_node(unsigned long size, unsigned long align,
David Miller2dca6992009-09-21 12:22:34 -07001801 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001802 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
David Rientjesd0a21262011-01-13 15:46:02 -08001804 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001805 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806}
1807
Christoph Lameter930fc452005-10-29 18:15:41 -07001808void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1809{
David Rientjes00ef2d22013-02-22 16:35:36 -08001810 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001811 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001812}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813EXPORT_SYMBOL(__vmalloc);
1814
Michal Hocko8594a212017-05-12 15:46:41 -07001815static inline void *__vmalloc_node_flags(unsigned long size,
1816 int node, gfp_t flags)
1817{
1818 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1819 node, __builtin_return_address(0));
1820}
1821
1822
1823void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
1824 void *caller)
1825{
1826 return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
1827}
1828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829/**
1830 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * Allocate enough pages to cover @size from the page level
1833 * allocator and map them into contiguous kernel virtual space.
1834 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001835 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 * use __vmalloc() instead.
1837 */
1838void *vmalloc(unsigned long size)
1839{
David Rientjes00ef2d22013-02-22 16:35:36 -08001840 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Michal Hocko19809c22017-05-08 15:57:44 -07001841 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843EXPORT_SYMBOL(vmalloc);
1844
Christoph Lameter930fc452005-10-29 18:15:41 -07001845/**
Dave Younge1ca7782010-10-26 14:22:06 -07001846 * vzalloc - allocate virtually contiguous memory with zero fill
1847 * @size: allocation size
1848 * Allocate enough pages to cover @size from the page level
1849 * allocator and map them into contiguous kernel virtual space.
1850 * The memory allocated is set to zero.
1851 *
1852 * For tight control over page level allocator and protection flags
1853 * use __vmalloc() instead.
1854 */
1855void *vzalloc(unsigned long size)
1856{
David Rientjes00ef2d22013-02-22 16:35:36 -08001857 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Michal Hocko19809c22017-05-08 15:57:44 -07001858 GFP_KERNEL | __GFP_ZERO);
Dave Younge1ca7782010-10-26 14:22:06 -07001859}
1860EXPORT_SYMBOL(vzalloc);
1861
1862/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001863 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1864 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001865 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001866 * The resulting memory area is zeroed so it can be mapped to userspace
1867 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001868 */
1869void *vmalloc_user(unsigned long size)
1870{
1871 struct vm_struct *area;
1872 void *ret;
1873
David Miller2dca6992009-09-21 12:22:34 -07001874 ret = __vmalloc_node(size, SHMLBA,
Michal Hocko19809c22017-05-08 15:57:44 -07001875 GFP_KERNEL | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001876 PAGE_KERNEL, NUMA_NO_NODE,
1877 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001878 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001879 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001880 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001881 }
Nick Piggin83342312006-06-23 02:03:20 -07001882 return ret;
1883}
1884EXPORT_SYMBOL(vmalloc_user);
1885
1886/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001887 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001888 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001889 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001890 *
1891 * Allocate enough pages to cover @size from the page level
1892 * allocator and map them into contiguous kernel virtual space.
1893 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001894 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001895 * use __vmalloc() instead.
1896 */
1897void *vmalloc_node(unsigned long size, int node)
1898{
Michal Hocko19809c22017-05-08 15:57:44 -07001899 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001900 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001901}
1902EXPORT_SYMBOL(vmalloc_node);
1903
Dave Younge1ca7782010-10-26 14:22:06 -07001904/**
1905 * vzalloc_node - allocate memory on a specific node with zero fill
1906 * @size: allocation size
1907 * @node: numa node
1908 *
1909 * Allocate enough pages to cover @size from the page level
1910 * allocator and map them into contiguous kernel virtual space.
1911 * The memory allocated is set to zero.
1912 *
1913 * For tight control over page level allocator and protection flags
1914 * use __vmalloc_node() instead.
1915 */
1916void *vzalloc_node(unsigned long size, int node)
1917{
1918 return __vmalloc_node_flags(size, node,
Michal Hocko19809c22017-05-08 15:57:44 -07001919 GFP_KERNEL | __GFP_ZERO);
Dave Younge1ca7782010-10-26 14:22:06 -07001920}
1921EXPORT_SYMBOL(vzalloc_node);
1922
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001923#ifndef PAGE_KERNEL_EXEC
1924# define PAGE_KERNEL_EXEC PAGE_KERNEL
1925#endif
1926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927/**
1928 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 * @size: allocation size
1930 *
1931 * Kernel-internal function to allocate enough pages to cover @size
1932 * the page level allocator and map them into contiguous and
1933 * executable kernel virtual space.
1934 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001935 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 * use __vmalloc() instead.
1937 */
1938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939void *vmalloc_exec(unsigned long size)
1940{
Michal Hocko19809c22017-05-08 15:57:44 -07001941 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001942 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943}
1944
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001945#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001946#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001947#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001948#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001949#else
1950#define GFP_VMALLOC32 GFP_KERNEL
1951#endif
1952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953/**
1954 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 * @size: allocation size
1956 *
1957 * Allocate enough 32bit PA addressable pages to cover @size from the
1958 * page level allocator and map them into contiguous kernel virtual space.
1959 */
1960void *vmalloc_32(unsigned long size)
1961{
David Miller2dca6992009-09-21 12:22:34 -07001962 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001963 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965EXPORT_SYMBOL(vmalloc_32);
1966
Nick Piggin83342312006-06-23 02:03:20 -07001967/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001968 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001969 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001970 *
1971 * The resulting memory area is 32bit addressable and zeroed so it can be
1972 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001973 */
1974void *vmalloc_32_user(unsigned long size)
1975{
1976 struct vm_struct *area;
1977 void *ret;
1978
David Miller2dca6992009-09-21 12:22:34 -07001979 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001980 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001981 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001982 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001983 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001984 }
Nick Piggin83342312006-06-23 02:03:20 -07001985 return ret;
1986}
1987EXPORT_SYMBOL(vmalloc_32_user);
1988
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001989/*
1990 * small helper routine , copy contents to buf from addr.
1991 * If the page is not present, fill zero.
1992 */
1993
1994static int aligned_vread(char *buf, char *addr, unsigned long count)
1995{
1996 struct page *p;
1997 int copied = 0;
1998
1999 while (count) {
2000 unsigned long offset, length;
2001
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002002 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002003 length = PAGE_SIZE - offset;
2004 if (length > count)
2005 length = count;
2006 p = vmalloc_to_page(addr);
2007 /*
2008 * To do safe access to this _mapped_ area, we need
2009 * lock. But adding lock here means that we need to add
2010 * overhead of vmalloc()/vfree() calles for this _debug_
2011 * interface, rarely used. Instead of that, we'll use
2012 * kmap() and get small overhead in this access function.
2013 */
2014 if (p) {
2015 /*
2016 * we can expect USER0 is not used (see vread/vwrite's
2017 * function description)
2018 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002019 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002020 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002021 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002022 } else
2023 memset(buf, 0, length);
2024
2025 addr += length;
2026 buf += length;
2027 copied += length;
2028 count -= length;
2029 }
2030 return copied;
2031}
2032
2033static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2034{
2035 struct page *p;
2036 int copied = 0;
2037
2038 while (count) {
2039 unsigned long offset, length;
2040
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002041 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002042 length = PAGE_SIZE - offset;
2043 if (length > count)
2044 length = count;
2045 p = vmalloc_to_page(addr);
2046 /*
2047 * To do safe access to this _mapped_ area, we need
2048 * lock. But adding lock here means that we need to add
2049 * overhead of vmalloc()/vfree() calles for this _debug_
2050 * interface, rarely used. Instead of that, we'll use
2051 * kmap() and get small overhead in this access function.
2052 */
2053 if (p) {
2054 /*
2055 * we can expect USER0 is not used (see vread/vwrite's
2056 * function description)
2057 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002058 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002059 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002060 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002061 }
2062 addr += length;
2063 buf += length;
2064 copied += length;
2065 count -= length;
2066 }
2067 return copied;
2068}
2069
2070/**
2071 * vread() - read vmalloc area in a safe way.
2072 * @buf: buffer for reading data
2073 * @addr: vm address.
2074 * @count: number of bytes to be read.
2075 *
2076 * Returns # of bytes which addr and buf should be increased.
2077 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2078 * includes any intersect with alive vmalloc area.
2079 *
2080 * This function checks that addr is a valid vmalloc'ed area, and
2081 * copy data from that area to a given buffer. If the given memory range
2082 * of [addr...addr+count) includes some valid address, data is copied to
2083 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2084 * IOREMAP area is treated as memory hole and no copy is done.
2085 *
2086 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002087 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002088 *
2089 * Note: In usual ops, vread() is never necessary because the caller
2090 * should know vmalloc() area is valid and can use memcpy().
2091 * This is for routines which have to access vmalloc area without
2092 * any informaion, as /dev/kmem.
2093 *
2094 */
2095
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096long vread(char *buf, char *addr, unsigned long count)
2097{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002098 struct vmap_area *va;
2099 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002101 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 unsigned long n;
2103
2104 /* Don't allow overflow */
2105 if ((unsigned long) addr + count < count)
2106 count = -(unsigned long) addr;
2107
Joonsoo Kime81ce852013-04-29 15:07:32 -07002108 spin_lock(&vmap_area_lock);
2109 list_for_each_entry(va, &vmap_area_list, list) {
2110 if (!count)
2111 break;
2112
2113 if (!(va->flags & VM_VM_AREA))
2114 continue;
2115
2116 vm = va->vm;
2117 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002118 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 continue;
2120 while (addr < vaddr) {
2121 if (count == 0)
2122 goto finished;
2123 *buf = '\0';
2124 buf++;
2125 addr++;
2126 count--;
2127 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002128 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002129 if (n > count)
2130 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002131 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002132 aligned_vread(buf, addr, n);
2133 else /* IOREMAP area is treated as memory hole */
2134 memset(buf, 0, n);
2135 buf += n;
2136 addr += n;
2137 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
2139finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002140 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002141
2142 if (buf == buf_start)
2143 return 0;
2144 /* zero-fill memory holes */
2145 if (buf != buf_start + buflen)
2146 memset(buf, 0, buflen - (buf - buf_start));
2147
2148 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149}
2150
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002151/**
2152 * vwrite() - write vmalloc area in a safe way.
2153 * @buf: buffer for source data
2154 * @addr: vm address.
2155 * @count: number of bytes to be read.
2156 *
2157 * Returns # of bytes which addr and buf should be incresed.
2158 * (same number to @count).
2159 * If [addr...addr+count) doesn't includes any intersect with valid
2160 * vmalloc area, returns 0.
2161 *
2162 * This function checks that addr is a valid vmalloc'ed area, and
2163 * copy data from a buffer to the given addr. If specified range of
2164 * [addr...addr+count) includes some valid address, data is copied from
2165 * proper area of @buf. If there are memory holes, no copy to hole.
2166 * IOREMAP area is treated as memory hole and no copy is done.
2167 *
2168 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002169 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002170 *
2171 * Note: In usual ops, vwrite() is never necessary because the caller
2172 * should know vmalloc() area is valid and can use memcpy().
2173 * This is for routines which have to access vmalloc area without
2174 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002175 */
2176
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177long vwrite(char *buf, char *addr, unsigned long count)
2178{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002179 struct vmap_area *va;
2180 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002181 char *vaddr;
2182 unsigned long n, buflen;
2183 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
2185 /* Don't allow overflow */
2186 if ((unsigned long) addr + count < count)
2187 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002188 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Joonsoo Kime81ce852013-04-29 15:07:32 -07002190 spin_lock(&vmap_area_lock);
2191 list_for_each_entry(va, &vmap_area_list, list) {
2192 if (!count)
2193 break;
2194
2195 if (!(va->flags & VM_VM_AREA))
2196 continue;
2197
2198 vm = va->vm;
2199 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002200 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 continue;
2202 while (addr < vaddr) {
2203 if (count == 0)
2204 goto finished;
2205 buf++;
2206 addr++;
2207 count--;
2208 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002209 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002210 if (n > count)
2211 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002212 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002213 aligned_vwrite(buf, addr, n);
2214 copied++;
2215 }
2216 buf += n;
2217 addr += n;
2218 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 }
2220finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002221 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002222 if (!copied)
2223 return 0;
2224 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225}
Nick Piggin83342312006-06-23 02:03:20 -07002226
2227/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002228 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2229 * @vma: vma to cover
2230 * @uaddr: target user address to start at
2231 * @kaddr: virtual address of vmalloc kernel memory
2232 * @size: size of map area
2233 *
2234 * Returns: 0 for success, -Exxx on failure
2235 *
2236 * This function checks that @kaddr is a valid vmalloc'ed area,
2237 * and that it is big enough to cover the range starting at
2238 * @uaddr in @vma. Will return failure if that criteria isn't
2239 * met.
2240 *
2241 * Similar to remap_pfn_range() (see mm/memory.c)
2242 */
2243int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2244 void *kaddr, unsigned long size)
2245{
2246 struct vm_struct *area;
2247
2248 size = PAGE_ALIGN(size);
2249
2250 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2251 return -EINVAL;
2252
2253 area = find_vm_area(kaddr);
2254 if (!area)
2255 return -EINVAL;
2256
2257 if (!(area->flags & VM_USERMAP))
2258 return -EINVAL;
2259
2260 if (kaddr + size > area->addr + area->size)
2261 return -EINVAL;
2262
2263 do {
2264 struct page *page = vmalloc_to_page(kaddr);
2265 int ret;
2266
2267 ret = vm_insert_page(vma, uaddr, page);
2268 if (ret)
2269 return ret;
2270
2271 uaddr += PAGE_SIZE;
2272 kaddr += PAGE_SIZE;
2273 size -= PAGE_SIZE;
2274 } while (size > 0);
2275
2276 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2277
2278 return 0;
2279}
2280EXPORT_SYMBOL(remap_vmalloc_range_partial);
2281
2282/**
Nick Piggin83342312006-06-23 02:03:20 -07002283 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002284 * @vma: vma to cover (map full range of vma)
2285 * @addr: vmalloc memory
2286 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002287 *
2288 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002289 *
2290 * This function checks that addr is a valid vmalloc'ed area, and
2291 * that it is big enough to cover the vma. Will return failure if
2292 * that criteria isn't met.
2293 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002294 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002295 */
2296int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2297 unsigned long pgoff)
2298{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002299 return remap_vmalloc_range_partial(vma, vma->vm_start,
2300 addr + (pgoff << PAGE_SHIFT),
2301 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002302}
2303EXPORT_SYMBOL(remap_vmalloc_range);
2304
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002305/*
2306 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2307 * have one.
2308 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002309void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002310{
2311}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002312
2313
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002314static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002315{
David Vrabelcd129092011-09-29 16:53:32 +01002316 pte_t ***p = data;
2317
2318 if (p) {
2319 *(*p) = pte;
2320 (*p)++;
2321 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002322 return 0;
2323}
2324
2325/**
2326 * alloc_vm_area - allocate a range of kernel address space
2327 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002328 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002329 *
2330 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002331 *
2332 * This function reserves a range of kernel address space, and
2333 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002334 * are created.
2335 *
2336 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2337 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002338 */
David Vrabelcd129092011-09-29 16:53:32 +01002339struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002340{
2341 struct vm_struct *area;
2342
Christoph Lameter23016962008-04-28 02:12:42 -07002343 area = get_vm_area_caller(size, VM_IOREMAP,
2344 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002345 if (area == NULL)
2346 return NULL;
2347
2348 /*
2349 * This ensures that page tables are constructed for this region
2350 * of kernel virtual address space and mapped into init_mm.
2351 */
2352 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002353 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002354 free_vm_area(area);
2355 return NULL;
2356 }
2357
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002358 return area;
2359}
2360EXPORT_SYMBOL_GPL(alloc_vm_area);
2361
2362void free_vm_area(struct vm_struct *area)
2363{
2364 struct vm_struct *ret;
2365 ret = remove_vm_area(area->addr);
2366 BUG_ON(ret != area);
2367 kfree(area);
2368}
2369EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002370
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002371#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002372static struct vmap_area *node_to_va(struct rb_node *n)
2373{
Geliang Tang4583e772017-02-22 15:41:54 -08002374 return rb_entry_safe(n, struct vmap_area, rb_node);
Tejun Heoca23e402009-08-14 15:00:52 +09002375}
2376
2377/**
2378 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2379 * @end: target address
2380 * @pnext: out arg for the next vmap_area
2381 * @pprev: out arg for the previous vmap_area
2382 *
2383 * Returns: %true if either or both of next and prev are found,
2384 * %false if no vmap_area exists
2385 *
2386 * Find vmap_areas end addresses of which enclose @end. ie. if not
2387 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2388 */
2389static bool pvm_find_next_prev(unsigned long end,
2390 struct vmap_area **pnext,
2391 struct vmap_area **pprev)
2392{
2393 struct rb_node *n = vmap_area_root.rb_node;
2394 struct vmap_area *va = NULL;
2395
2396 while (n) {
2397 va = rb_entry(n, struct vmap_area, rb_node);
2398 if (end < va->va_end)
2399 n = n->rb_left;
2400 else if (end > va->va_end)
2401 n = n->rb_right;
2402 else
2403 break;
2404 }
2405
2406 if (!va)
2407 return false;
2408
2409 if (va->va_end > end) {
2410 *pnext = va;
2411 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2412 } else {
2413 *pprev = va;
2414 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2415 }
2416 return true;
2417}
2418
2419/**
2420 * pvm_determine_end - find the highest aligned address between two vmap_areas
2421 * @pnext: in/out arg for the next vmap_area
2422 * @pprev: in/out arg for the previous vmap_area
2423 * @align: alignment
2424 *
2425 * Returns: determined end address
2426 *
2427 * Find the highest aligned address between *@pnext and *@pprev below
2428 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2429 * down address is between the end addresses of the two vmap_areas.
2430 *
2431 * Please note that the address returned by this function may fall
2432 * inside *@pnext vmap_area. The caller is responsible for checking
2433 * that.
2434 */
2435static unsigned long pvm_determine_end(struct vmap_area **pnext,
2436 struct vmap_area **pprev,
2437 unsigned long align)
2438{
2439 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2440 unsigned long addr;
2441
2442 if (*pnext)
2443 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2444 else
2445 addr = vmalloc_end;
2446
2447 while (*pprev && (*pprev)->va_end > addr) {
2448 *pnext = *pprev;
2449 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2450 }
2451
2452 return addr;
2453}
2454
2455/**
2456 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2457 * @offsets: array containing offset of each area
2458 * @sizes: array containing size of each area
2459 * @nr_vms: the number of areas to allocate
2460 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002461 *
2462 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2463 * vm_structs on success, %NULL on failure
2464 *
2465 * Percpu allocator wants to use congruent vm areas so that it can
2466 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002467 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2468 * be scattered pretty far, distance between two areas easily going up
2469 * to gigabytes. To avoid interacting with regular vmallocs, these
2470 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002471 *
2472 * Despite its complicated look, this allocator is rather simple. It
2473 * does everything top-down and scans areas from the end looking for
2474 * matching slot. While scanning, if any of the areas overlaps with
2475 * existing vmap_area, the base address is pulled down to fit the
2476 * area. Scanning is repeated till all the areas fit and then all
Wei Yangc568da22017-09-06 16:24:09 -07002477 * necessary data structures are inserted and the result is returned.
Tejun Heoca23e402009-08-14 15:00:52 +09002478 */
2479struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2480 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002481 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002482{
2483 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2484 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2485 struct vmap_area **vas, *prev, *next;
2486 struct vm_struct **vms;
2487 int area, area2, last_area, term_area;
2488 unsigned long base, start, end, last_end;
2489 bool purged = false;
2490
Tejun Heoca23e402009-08-14 15:00:52 +09002491 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002492 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002493 for (last_area = 0, area = 0; area < nr_vms; area++) {
2494 start = offsets[area];
2495 end = start + sizes[area];
2496
2497 /* is everything aligned properly? */
2498 BUG_ON(!IS_ALIGNED(offsets[area], align));
2499 BUG_ON(!IS_ALIGNED(sizes[area], align));
2500
2501 /* detect the area with the highest address */
2502 if (start > offsets[last_area])
2503 last_area = area;
2504
Wei Yangc568da22017-09-06 16:24:09 -07002505 for (area2 = area + 1; area2 < nr_vms; area2++) {
Tejun Heoca23e402009-08-14 15:00:52 +09002506 unsigned long start2 = offsets[area2];
2507 unsigned long end2 = start2 + sizes[area2];
2508
Wei Yangc568da22017-09-06 16:24:09 -07002509 BUG_ON(start2 < end && start < end2);
Tejun Heoca23e402009-08-14 15:00:52 +09002510 }
2511 }
2512 last_end = offsets[last_area] + sizes[last_area];
2513
2514 if (vmalloc_end - vmalloc_start < last_end) {
2515 WARN_ON(true);
2516 return NULL;
2517 }
2518
Thomas Meyer4d67d862012-05-29 15:06:21 -07002519 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2520 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002521 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002522 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002523
2524 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002525 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2526 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002527 if (!vas[area] || !vms[area])
2528 goto err_free;
2529 }
2530retry:
2531 spin_lock(&vmap_area_lock);
2532
2533 /* start scanning - we scan from the top, begin with the last area */
2534 area = term_area = last_area;
2535 start = offsets[area];
2536 end = start + sizes[area];
2537
2538 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2539 base = vmalloc_end - last_end;
2540 goto found;
2541 }
2542 base = pvm_determine_end(&next, &prev, align) - end;
2543
2544 while (true) {
2545 BUG_ON(next && next->va_end <= base + end);
2546 BUG_ON(prev && prev->va_end > base + end);
2547
2548 /*
2549 * base might have underflowed, add last_end before
2550 * comparing.
2551 */
2552 if (base + last_end < vmalloc_start + last_end) {
2553 spin_unlock(&vmap_area_lock);
2554 if (!purged) {
2555 purge_vmap_area_lazy();
2556 purged = true;
2557 goto retry;
2558 }
2559 goto err_free;
2560 }
2561
2562 /*
2563 * If next overlaps, move base downwards so that it's
2564 * right below next and then recheck.
2565 */
2566 if (next && next->va_start < base + end) {
2567 base = pvm_determine_end(&next, &prev, align) - end;
2568 term_area = area;
2569 continue;
2570 }
2571
2572 /*
2573 * If prev overlaps, shift down next and prev and move
2574 * base so that it's right below new next and then
2575 * recheck.
2576 */
2577 if (prev && prev->va_end > base + start) {
2578 next = prev;
2579 prev = node_to_va(rb_prev(&next->rb_node));
2580 base = pvm_determine_end(&next, &prev, align) - end;
2581 term_area = area;
2582 continue;
2583 }
2584
2585 /*
2586 * This area fits, move on to the previous one. If
2587 * the previous one is the terminal one, we're done.
2588 */
2589 area = (area + nr_vms - 1) % nr_vms;
2590 if (area == term_area)
2591 break;
2592 start = offsets[area];
2593 end = start + sizes[area];
2594 pvm_find_next_prev(base + end, &next, &prev);
2595 }
2596found:
2597 /* we've found a fitting base, insert all va's */
2598 for (area = 0; area < nr_vms; area++) {
2599 struct vmap_area *va = vas[area];
2600
2601 va->va_start = base + offsets[area];
2602 va->va_end = va->va_start + sizes[area];
2603 __insert_vmap_area(va);
2604 }
2605
2606 vmap_area_pcpu_hole = base + offsets[last_area];
2607
2608 spin_unlock(&vmap_area_lock);
2609
2610 /* insert all vm's */
2611 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002612 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2613 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002614
2615 kfree(vas);
2616 return vms;
2617
2618err_free:
2619 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002620 kfree(vas[area]);
2621 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002622 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002623err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002624 kfree(vas);
2625 kfree(vms);
2626 return NULL;
2627}
2628
2629/**
2630 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2631 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2632 * @nr_vms: the number of allocated areas
2633 *
2634 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2635 */
2636void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2637{
2638 int i;
2639
2640 for (i = 0; i < nr_vms; i++)
2641 free_vm_area(vms[i]);
2642 kfree(vms);
2643}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002644#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002645
2646#ifdef CONFIG_PROC_FS
2647static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002648 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002649{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002650 spin_lock(&vmap_area_lock);
zijun_hu3f500062016-12-12 16:42:17 -08002651 return seq_list_start(&vmap_area_list, *pos);
Christoph Lametera10aa572008-04-28 02:12:40 -07002652}
2653
2654static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2655{
zijun_hu3f500062016-12-12 16:42:17 -08002656 return seq_list_next(p, &vmap_area_list, pos);
Christoph Lametera10aa572008-04-28 02:12:40 -07002657}
2658
2659static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002660 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002661{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002662 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002663}
2664
Eric Dumazeta47a1262008-07-23 21:27:38 -07002665static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2666{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002667 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002668 unsigned int nr, *counters = m->private;
2669
2670 if (!counters)
2671 return;
2672
Wanpeng Liaf123462013-11-12 15:07:32 -08002673 if (v->flags & VM_UNINITIALIZED)
2674 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002675 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2676 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002677
Eric Dumazeta47a1262008-07-23 21:27:38 -07002678 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2679
2680 for (nr = 0; nr < v->nr_pages; nr++)
2681 counters[page_to_nid(v->pages[nr])]++;
2682
2683 for_each_node_state(nr, N_HIGH_MEMORY)
2684 if (counters[nr])
2685 seq_printf(m, " N%u=%u", nr, counters[nr]);
2686 }
2687}
2688
Christoph Lametera10aa572008-04-28 02:12:40 -07002689static int s_show(struct seq_file *m, void *p)
2690{
zijun_hu3f500062016-12-12 16:42:17 -08002691 struct vmap_area *va;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002692 struct vm_struct *v;
2693
zijun_hu3f500062016-12-12 16:42:17 -08002694 va = list_entry(p, struct vmap_area, list);
2695
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002696 /*
2697 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2698 * behalf of vmap area is being tear down or vm_map_ram allocation.
2699 */
Yisheng Xie78c72742017-07-10 15:48:09 -07002700 if (!(va->flags & VM_VM_AREA)) {
2701 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2702 (void *)va->va_start, (void *)va->va_end,
2703 va->va_end - va->va_start,
2704 va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2705
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002706 return 0;
Yisheng Xie78c72742017-07-10 15:48:09 -07002707 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002708
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002709 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002710
Kees Cook45ec1692012-10-08 16:34:09 -07002711 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002712 v->addr, v->addr + v->size, v->size);
2713
Joe Perches62c70bc2011-01-13 15:45:52 -08002714 if (v->caller)
2715 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002716
Christoph Lametera10aa572008-04-28 02:12:40 -07002717 if (v->nr_pages)
2718 seq_printf(m, " pages=%d", v->nr_pages);
2719
2720 if (v->phys_addr)
Miles Chen199eaa02017-02-24 14:59:51 -08002721 seq_printf(m, " phys=%pa", &v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002722
2723 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002724 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002725
2726 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002727 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002728
2729 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002730 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002731
2732 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002733 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002734
David Rientjes244d63e2016-01-14 15:19:35 -08002735 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002736 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002737
Eric Dumazeta47a1262008-07-23 21:27:38 -07002738 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002739 seq_putc(m, '\n');
2740 return 0;
2741}
2742
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002743static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002744 .start = s_start,
2745 .next = s_next,
2746 .stop = s_stop,
2747 .show = s_show,
2748};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002749
2750static int vmalloc_open(struct inode *inode, struct file *file)
2751{
Rob Jones703394c2014-10-09 15:28:01 -07002752 if (IS_ENABLED(CONFIG_NUMA))
2753 return seq_open_private(file, &vmalloc_op,
2754 nr_node_ids * sizeof(unsigned int));
2755 else
2756 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002757}
2758
2759static const struct file_operations proc_vmalloc_operations = {
2760 .open = vmalloc_open,
2761 .read = seq_read,
2762 .llseek = seq_lseek,
2763 .release = seq_release_private,
2764};
2765
2766static int __init proc_vmalloc_init(void)
2767{
2768 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2769 return 0;
2770}
2771module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002772
Christoph Lametera10aa572008-04-28 02:12:40 -07002773#endif
2774