blob: 4df66e1abeb15ab4773605e13bb918d01309a1d2 [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
606/*
607 * lazy_max_pages is the maximum amount of virtual address space we gather up
608 * before attempting to purge with a TLB flush.
609 *
610 * There is a tradeoff here: a larger number will cover more kernel page tables
611 * and take slightly longer to purge, but it will linearly reduce the number of
612 * global TLB flushes that must be performed. It would seem natural to scale
613 * this number up linearly with the number of CPUs (because vmapping activity
614 * could also scale linearly with the number of CPUs), however it is likely
615 * that in practice, workloads might be constrained in other ways that mean
616 * vmap activity will not scale linearly with CPUs. Also, I want to be
617 * conservative and not introduce a big latency on huge systems, so go with
618 * a less aggressive log scale. It will still be an improvement over the old
619 * code, and it will be simple to change the scale factor if we find that it
620 * becomes a problem on bigger systems.
621 */
622static unsigned long lazy_max_pages(void)
623{
624 unsigned int log;
625
626 log = fls(num_online_cpus());
627
628 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
629}
630
631static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
632
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800633/*
634 * Serialize vmap purging. There is no actual criticial section protected
635 * by this look, but we want to avoid concurrent calls for performance
636 * reasons and to make the pcpu_get_vm_areas more deterministic.
637 */
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800638static DEFINE_MUTEX(vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800639
Nick Piggin02b709d2010-02-01 22:25:57 +1100640/* for per-CPU blocks */
641static void purge_fragmented_blocks_allcpus(void);
642
Nick Piggindb64fe02008-10-18 20:27:03 -0700643/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500644 * called before a call to iounmap() if the caller wants vm_area_struct's
645 * immediately freed.
646 */
647void set_iounmap_nonlazy(void)
648{
649 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
650}
651
652/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700653 * Purges all lazily-freed vmap areas.
Nick Piggindb64fe02008-10-18 20:27:03 -0700654 */
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800655static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700656{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700657 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -0700658 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800659 struct vmap_area *n_va;
Joel Fernandes763b2182016-12-12 16:44:26 -0800660 bool do_free = false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700661
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800662 lockdep_assert_held(&vmap_purge_lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100663
Chris Wilson80c4bd72016-05-20 16:57:38 -0700664 valist = llist_del_all(&vmap_purge_list);
665 llist_for_each_entry(va, valist, purge_list) {
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800666 if (va->va_start < start)
667 start = va->va_start;
668 if (va->va_end > end)
669 end = va->va_end;
Joel Fernandes763b2182016-12-12 16:44:26 -0800670 do_free = true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700671 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700672
Joel Fernandes763b2182016-12-12 16:44:26 -0800673 if (!do_free)
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800674 return false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700675
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800676 flush_tlb_kernel_range(start, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700677
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800678 spin_lock(&vmap_area_lock);
Joel Fernandes763b2182016-12-12 16:44:26 -0800679 llist_for_each_entry_safe(va, n_va, valist, purge_list) {
680 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
681
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800682 __free_vmap_area(va);
Joel Fernandes763b2182016-12-12 16:44:26 -0800683 atomic_sub(nr, &vmap_lazy_nr);
684 cond_resched_lock(&vmap_area_lock);
685 }
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800686 spin_unlock(&vmap_area_lock);
687 return true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700688}
689
690/*
Nick Piggin496850e2008-11-19 15:36:33 -0800691 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
692 * is already purging.
693 */
694static void try_purge_vmap_area_lazy(void)
695{
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800696 if (mutex_trylock(&vmap_purge_lock)) {
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800697 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800698 mutex_unlock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800699 }
Nick Piggin496850e2008-11-19 15:36:33 -0800700}
701
702/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700703 * Kick off a purge of the outstanding lazy areas.
704 */
705static void purge_vmap_area_lazy(void)
706{
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800707 mutex_lock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -0800708 purge_fragmented_blocks_allcpus();
709 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwigf9e09972016-12-12 16:44:23 -0800710 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700711}
712
713/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800714 * Free a vmap area, caller ensuring that the area has been unmapped
715 * and flush_cache_vunmap had been called for the correct range
716 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700717 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800718static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700719{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700720 int nr_lazy;
721
722 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
723 &vmap_lazy_nr);
724
725 /* After this point, we may free va at any time */
726 llist_add(&va->purge_list, &vmap_purge_list);
727
728 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800729 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700730}
731
Nick Pigginb29acbd2008-12-01 13:13:47 -0800732/*
733 * Free and unmap a vmap area
734 */
735static void free_unmap_vmap_area(struct vmap_area *va)
736{
737 flush_cache_vunmap(va->va_start, va->va_end);
Christoph Hellwigc8eef012016-12-12 16:44:01 -0800738 unmap_vmap_area(va);
Chintan Pandya82a2e922018-06-07 17:06:46 -0700739 if (debug_pagealloc_enabled())
740 flush_tlb_kernel_range(va->va_start, va->va_end);
741
Christoph Hellwigc8eef012016-12-12 16:44:01 -0800742 free_vmap_area_noflush(va);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800743}
744
Nick Piggindb64fe02008-10-18 20:27:03 -0700745static struct vmap_area *find_vmap_area(unsigned long addr)
746{
747 struct vmap_area *va;
748
749 spin_lock(&vmap_area_lock);
750 va = __find_vmap_area(addr);
751 spin_unlock(&vmap_area_lock);
752
753 return va;
754}
755
Nick Piggindb64fe02008-10-18 20:27:03 -0700756/*** Per cpu kva allocator ***/
757
758/*
759 * vmap space is limited especially on 32 bit architectures. Ensure there is
760 * room for at least 16 percpu vmap blocks per CPU.
761 */
762/*
763 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
764 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
765 * instead (we just need a rough idea)
766 */
767#if BITS_PER_LONG == 32
768#define VMALLOC_SPACE (128UL*1024*1024)
769#else
770#define VMALLOC_SPACE (128UL*1024*1024*1024)
771#endif
772
773#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
774#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
775#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
776#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
777#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
778#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f9152011-06-21 22:09:50 +0200779#define VMAP_BBMAP_BITS \
780 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
781 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
782 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700783
784#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
785
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100786static bool vmap_initialized __read_mostly = false;
787
Nick Piggindb64fe02008-10-18 20:27:03 -0700788struct vmap_block_queue {
789 spinlock_t lock;
790 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700791};
792
793struct vmap_block {
794 spinlock_t lock;
795 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700796 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700797 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100798 struct list_head free_list;
799 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100800 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700801};
802
803/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
804static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
805
806/*
807 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
808 * in the free path. Could get rid of this if we change the API to return a
809 * "cookie" from alloc, to be passed to free. But no big deal yet.
810 */
811static DEFINE_SPINLOCK(vmap_block_tree_lock);
812static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
813
814/*
815 * We should probably have a fallback mechanism to allocate virtual memory
816 * out of partially filled vmap blocks. However vmap block sizing should be
817 * fairly reasonable according to the vmalloc size, so it shouldn't be a
818 * big problem.
819 */
820
821static unsigned long addr_to_vb_idx(unsigned long addr)
822{
823 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
824 addr /= VMAP_BLOCK_SIZE;
825 return addr;
826}
827
Roman Pencf725ce2015-04-15 16:13:52 -0700828static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
829{
830 unsigned long addr;
831
832 addr = va_start + (pages_off << PAGE_SHIFT);
833 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
834 return (void *)addr;
835}
836
837/**
838 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
839 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
840 * @order: how many 2^order pages should be occupied in newly allocated block
841 * @gfp_mask: flags for the page level allocator
842 *
843 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
844 */
845static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700846{
847 struct vmap_block_queue *vbq;
848 struct vmap_block *vb;
849 struct vmap_area *va;
850 unsigned long vb_idx;
851 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700852 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700853
854 node = numa_node_id();
855
856 vb = kmalloc_node(sizeof(struct vmap_block),
857 gfp_mask & GFP_RECLAIM_MASK, node);
858 if (unlikely(!vb))
859 return ERR_PTR(-ENOMEM);
860
861 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
862 VMALLOC_START, VMALLOC_END,
863 node, gfp_mask);
Tobias Klauserddf9c6d42011-01-13 15:46:15 -0800864 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700865 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700866 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700867 }
868
869 err = radix_tree_preload(gfp_mask);
870 if (unlikely(err)) {
871 kfree(vb);
872 free_vmap_area(va);
873 return ERR_PTR(err);
874 }
875
Roman Pencf725ce2015-04-15 16:13:52 -0700876 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700877 spin_lock_init(&vb->lock);
878 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700879 /* At least something should be left free */
880 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
881 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700882 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700883 vb->dirty_min = VMAP_BBMAP_BITS;
884 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700885 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700886
887 vb_idx = addr_to_vb_idx(va->va_start);
888 spin_lock(&vmap_block_tree_lock);
889 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
890 spin_unlock(&vmap_block_tree_lock);
891 BUG_ON(err);
892 radix_tree_preload_end();
893
894 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700895 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700896 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700897 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900898 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700899
Roman Pencf725ce2015-04-15 16:13:52 -0700900 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700901}
902
Nick Piggindb64fe02008-10-18 20:27:03 -0700903static void free_vmap_block(struct vmap_block *vb)
904{
905 struct vmap_block *tmp;
906 unsigned long vb_idx;
907
Nick Piggindb64fe02008-10-18 20:27:03 -0700908 vb_idx = addr_to_vb_idx(vb->va->va_start);
909 spin_lock(&vmap_block_tree_lock);
910 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
911 spin_unlock(&vmap_block_tree_lock);
912 BUG_ON(tmp != vb);
913
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800914 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800915 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700916}
917
Nick Piggin02b709d2010-02-01 22:25:57 +1100918static void purge_fragmented_blocks(int cpu)
919{
920 LIST_HEAD(purge);
921 struct vmap_block *vb;
922 struct vmap_block *n_vb;
923 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
924
925 rcu_read_lock();
926 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
927
928 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
929 continue;
930
931 spin_lock(&vb->lock);
932 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
933 vb->free = 0; /* prevent further allocs after releasing lock */
934 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700935 vb->dirty_min = 0;
936 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100937 spin_lock(&vbq->lock);
938 list_del_rcu(&vb->free_list);
939 spin_unlock(&vbq->lock);
940 spin_unlock(&vb->lock);
941 list_add_tail(&vb->purge, &purge);
942 } else
943 spin_unlock(&vb->lock);
944 }
945 rcu_read_unlock();
946
947 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
948 list_del(&vb->purge);
949 free_vmap_block(vb);
950 }
951}
952
Nick Piggin02b709d2010-02-01 22:25:57 +1100953static void purge_fragmented_blocks_allcpus(void)
954{
955 int cpu;
956
957 for_each_possible_cpu(cpu)
958 purge_fragmented_blocks(cpu);
959}
960
Nick Piggindb64fe02008-10-18 20:27:03 -0700961static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
962{
963 struct vmap_block_queue *vbq;
964 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700965 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700966 unsigned int order;
967
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800968 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700969 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700970 if (WARN_ON(size == 0)) {
971 /*
972 * Allocating 0 bytes isn't what caller wants since
973 * get_order(0) returns funny result. Just warn and terminate
974 * early.
975 */
976 return NULL;
977 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700978 order = get_order(size);
979
Nick Piggindb64fe02008-10-18 20:27:03 -0700980 rcu_read_lock();
981 vbq = &get_cpu_var(vmap_block_queue);
982 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -0700983 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -0700984
985 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -0700986 if (vb->free < (1UL << order)) {
987 spin_unlock(&vb->lock);
988 continue;
989 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100990
Roman Pencf725ce2015-04-15 16:13:52 -0700991 pages_off = VMAP_BBMAP_BITS - vb->free;
992 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +1100993 vb->free -= 1UL << order;
994 if (vb->free == 0) {
995 spin_lock(&vbq->lock);
996 list_del_rcu(&vb->free_list);
997 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700998 }
Roman Pencf725ce2015-04-15 16:13:52 -0700999
Nick Piggindb64fe02008-10-18 20:27:03 -07001000 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001001 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001002 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001003
Tejun Heo3f04ba82009-10-29 22:34:12 +09001004 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001005 rcu_read_unlock();
1006
Roman Pencf725ce2015-04-15 16:13:52 -07001007 /* Allocate new block if nothing was found */
1008 if (!vaddr)
1009 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001010
Roman Pencf725ce2015-04-15 16:13:52 -07001011 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001012}
1013
1014static void vb_free(const void *addr, unsigned long size)
1015{
1016 unsigned long offset;
1017 unsigned long vb_idx;
1018 unsigned int order;
1019 struct vmap_block *vb;
1020
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001021 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001022 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001023
1024 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1025
Nick Piggindb64fe02008-10-18 20:27:03 -07001026 order = get_order(size);
1027
1028 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001029 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001030
1031 vb_idx = addr_to_vb_idx((unsigned long)addr);
1032 rcu_read_lock();
1033 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1034 rcu_read_unlock();
1035 BUG_ON(!vb);
1036
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001037 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1038
Chintan Pandya82a2e922018-06-07 17:06:46 -07001039 if (debug_pagealloc_enabled())
1040 flush_tlb_kernel_range((unsigned long)addr,
1041 (unsigned long)addr + size);
1042
Nick Piggindb64fe02008-10-18 20:27:03 -07001043 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001044
1045 /* Expand dirty range */
1046 vb->dirty_min = min(vb->dirty_min, offset);
1047 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001048
Nick Piggindb64fe02008-10-18 20:27:03 -07001049 vb->dirty += 1UL << order;
1050 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001051 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001052 spin_unlock(&vb->lock);
1053 free_vmap_block(vb);
1054 } else
1055 spin_unlock(&vb->lock);
1056}
1057
1058/**
1059 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1060 *
1061 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1062 * to amortize TLB flushing overheads. What this means is that any page you
1063 * have now, may, in a former life, have been mapped into kernel virtual
1064 * address by the vmap layer and so there might be some CPUs with TLB entries
1065 * still referencing that page (additional to the regular 1:1 kernel mapping).
1066 *
1067 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1068 * be sure that none of the pages we have control over will have any aliases
1069 * from the vmap layer.
1070 */
1071void vm_unmap_aliases(void)
1072{
1073 unsigned long start = ULONG_MAX, end = 0;
1074 int cpu;
1075 int flush = 0;
1076
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001077 if (unlikely(!vmap_initialized))
1078 return;
1079
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001080 might_sleep();
1081
Nick Piggindb64fe02008-10-18 20:27:03 -07001082 for_each_possible_cpu(cpu) {
1083 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1084 struct vmap_block *vb;
1085
1086 rcu_read_lock();
1087 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001088 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001089 if (vb->dirty) {
1090 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001091 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001092
Roman Pen7d61bfe2015-04-15 16:13:55 -07001093 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1094 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001095
Roman Pen7d61bfe2015-04-15 16:13:55 -07001096 start = min(s, start);
1097 end = max(e, end);
1098
Nick Piggindb64fe02008-10-18 20:27:03 -07001099 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001100 }
1101 spin_unlock(&vb->lock);
1102 }
1103 rcu_read_unlock();
1104 }
1105
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001106 mutex_lock(&vmap_purge_lock);
Christoph Hellwig0574ecd2016-12-12 16:44:07 -08001107 purge_fragmented_blocks_allcpus();
1108 if (!__purge_vmap_area_lazy(start, end) && flush)
1109 flush_tlb_kernel_range(start, end);
Christoph Hellwigf9e09972016-12-12 16:44:23 -08001110 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001111}
1112EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1113
1114/**
1115 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1116 * @mem: the pointer returned by vm_map_ram
1117 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1118 */
1119void vm_unmap_ram(const void *mem, unsigned int count)
1120{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001121 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001122 unsigned long addr = (unsigned long)mem;
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001123 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001124
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001125 might_sleep();
Nick Piggindb64fe02008-10-18 20:27:03 -07001126 BUG_ON(!addr);
1127 BUG_ON(addr < VMALLOC_START);
1128 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001129 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001130
1131 debug_check_no_locks_freed(mem, size);
1132
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001133 if (likely(count <= VMAP_MAX_ALLOC)) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001134 vb_free(mem, size);
Christoph Hellwig9c3acf62016-12-12 16:44:04 -08001135 return;
1136 }
1137
1138 va = find_vmap_area(addr);
1139 BUG_ON(!va);
1140 free_unmap_vmap_area(va);
Nick Piggindb64fe02008-10-18 20:27:03 -07001141}
1142EXPORT_SYMBOL(vm_unmap_ram);
1143
1144/**
1145 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1146 * @pages: an array of pointers to the pages to be mapped
1147 * @count: number of pages
1148 * @node: prefer to allocate data structures on this node
1149 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001150 *
Gioh Kim36437632014-04-07 15:37:37 -07001151 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1152 * faster than vmap so it's good. But if you mix long-life and short-life
1153 * objects with vm_map_ram(), it could consume lots of address space through
1154 * fragmentation (especially on a 32bit machine). You could see failures in
1155 * the end. Please use this function for short-lived objects.
1156 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001157 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001158 */
1159void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1160{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001161 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001162 unsigned long addr;
1163 void *mem;
1164
1165 if (likely(count <= VMAP_MAX_ALLOC)) {
1166 mem = vb_alloc(size, GFP_KERNEL);
1167 if (IS_ERR(mem))
1168 return NULL;
1169 addr = (unsigned long)mem;
1170 } else {
1171 struct vmap_area *va;
1172 va = alloc_vmap_area(size, PAGE_SIZE,
1173 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1174 if (IS_ERR(va))
1175 return NULL;
1176
1177 addr = va->va_start;
1178 mem = (void *)addr;
1179 }
1180 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1181 vm_unmap_ram(mem, count);
1182 return NULL;
1183 }
1184 return mem;
1185}
1186EXPORT_SYMBOL(vm_map_ram);
1187
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001188static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001189/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001190 * vm_area_add_early - add vmap area early during boot
1191 * @vm: vm_struct to add
1192 *
1193 * This function is used to add fixed kernel vm area to vmlist before
1194 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1195 * should contain proper values and the other fields should be zero.
1196 *
1197 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1198 */
1199void __init vm_area_add_early(struct vm_struct *vm)
1200{
1201 struct vm_struct *tmp, **p;
1202
1203 BUG_ON(vmap_initialized);
1204 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1205 if (tmp->addr >= vm->addr) {
1206 BUG_ON(tmp->addr < vm->addr + vm->size);
1207 break;
1208 } else
1209 BUG_ON(tmp->addr + tmp->size > vm->addr);
1210 }
1211 vm->next = *p;
1212 *p = vm;
1213}
1214
1215/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001216 * vm_area_register_early - register vmap area early during boot
1217 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001218 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001219 *
1220 * This function is used to register kernel vm area before
1221 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1222 * proper values on entry and other fields should be zero. On return,
1223 * vm->addr contains the allocated address.
1224 *
1225 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1226 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001227void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001228{
1229 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001230 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001231
Tejun Heoc0c0a292009-02-24 11:57:21 +09001232 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1233 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1234
1235 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001236
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001237 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001238}
1239
Nick Piggindb64fe02008-10-18 20:27:03 -07001240void __init vmalloc_init(void)
1241{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001242 struct vmap_area *va;
1243 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001244 int i;
1245
1246 for_each_possible_cpu(i) {
1247 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001248 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001249
1250 vbq = &per_cpu(vmap_block_queue, i);
1251 spin_lock_init(&vbq->lock);
1252 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001253 p = &per_cpu(vfree_deferred, i);
1254 init_llist_head(&p->list);
1255 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001256 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001257
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001258 /* Import existing vmlist entries. */
1259 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001260 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001261 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001262 va->va_start = (unsigned long)tmp->addr;
1263 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001264 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001265 __insert_vmap_area(va);
1266 }
Tejun Heoca23e402009-08-14 15:00:52 +09001267
1268 vmap_area_pcpu_hole = VMALLOC_END;
1269
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001270 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001271}
1272
Tejun Heo8fc48982009-02-20 16:29:08 +09001273/**
1274 * map_kernel_range_noflush - map kernel VM area with the specified pages
1275 * @addr: start of the VM area to map
1276 * @size: size of the VM area to map
1277 * @prot: page protection flags to use
1278 * @pages: pages to map
1279 *
1280 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1281 * specify should have been allocated using get_vm_area() and its
1282 * friends.
1283 *
1284 * NOTE:
1285 * This function does NOT do any cache flushing. The caller is
1286 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1287 * before calling this function.
1288 *
1289 * RETURNS:
1290 * The number of pages mapped on success, -errno on failure.
1291 */
1292int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1293 pgprot_t prot, struct page **pages)
1294{
1295 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1296}
1297
1298/**
1299 * unmap_kernel_range_noflush - unmap kernel VM area
1300 * @addr: start of the VM area to unmap
1301 * @size: size of the VM area to unmap
1302 *
1303 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1304 * specify should have been allocated using get_vm_area() and its
1305 * friends.
1306 *
1307 * NOTE:
1308 * This function does NOT do any cache flushing. The caller is
1309 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1310 * before calling this function and flush_tlb_kernel_range() after.
1311 */
1312void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1313{
1314 vunmap_page_range(addr, addr + size);
1315}
Huang Ying81e88fd2011-01-12 14:44:55 +08001316EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001317
1318/**
1319 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1320 * @addr: start of the VM area to unmap
1321 * @size: size of the VM area to unmap
1322 *
1323 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1324 * the unmapping and tlb after.
1325 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001326void unmap_kernel_range(unsigned long addr, unsigned long size)
1327{
1328 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001329
1330 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001331 vunmap_page_range(addr, end);
1332 flush_tlb_kernel_range(addr, end);
1333}
Minchan Kim93ef6d6c2014-06-04 16:11:09 -07001334EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001335
WANG Chaof6f8ed42014-08-06 16:06:58 -07001336int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001337{
1338 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001339 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001340 int err;
1341
WANG Chaof6f8ed42014-08-06 16:06:58 -07001342 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001343
WANG Chaof6f8ed42014-08-06 16:06:58 -07001344 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001345}
1346EXPORT_SYMBOL_GPL(map_vm_area);
1347
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001348static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001349 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001350{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001351 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001352 vm->flags = flags;
1353 vm->addr = (void *)va->va_start;
1354 vm->size = va->va_end - va->va_start;
1355 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001356 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001357 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001358 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001359}
Tejun Heocf88c792009-08-14 15:00:52 +09001360
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001361static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001362{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001363 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001364 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001365 * we should make sure that vm has proper values.
1366 * Pair with smp_rmb() in show_numa_info().
1367 */
1368 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001369 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001370}
1371
Nick Piggindb64fe02008-10-18 20:27:03 -07001372static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001373 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001374 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001375{
Kautuk Consul00065262011-12-19 17:12:04 -08001376 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001377 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001379 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001381 if (unlikely(!size))
1382 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
zijun_hu252e5c62016-10-07 16:57:26 -07001384 if (flags & VM_IOREMAP)
1385 align = 1ul << clamp_t(int, get_count_order_long(size),
1386 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1387
Tejun Heocf88c792009-08-14 15:00:52 +09001388 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (unlikely(!area))
1390 return NULL;
1391
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001392 if (!(flags & VM_NO_GUARD))
1393 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
Nick Piggindb64fe02008-10-18 20:27:03 -07001395 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1396 if (IS_ERR(va)) {
1397 kfree(area);
1398 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001401 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001402
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404}
1405
Christoph Lameter930fc452005-10-29 18:15:41 -07001406struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1407 unsigned long start, unsigned long end)
1408{
David Rientjes00ef2d22013-02-22 16:35:36 -08001409 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1410 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001411}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001412EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001413
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001414struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1415 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001416 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001417{
David Rientjes00ef2d22013-02-22 16:35:36 -08001418 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1419 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001420}
1421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422/**
Simon Arlott183ff222007-10-20 01:27:18 +02001423 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 * @size: size of the area
1425 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1426 *
1427 * Search an area of @size in the kernel virtual mapping area,
1428 * and reserved it for out purposes. Returns the area descriptor
1429 * on success or %NULL on failure.
1430 */
1431struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1432{
David Miller2dca6992009-09-21 12:22:34 -07001433 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001434 NUMA_NO_NODE, GFP_KERNEL,
1435 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001436}
1437
1438struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001439 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001440{
David Miller2dca6992009-09-21 12:22:34 -07001441 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001442 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001445/**
1446 * find_vm_area - find a continuous kernel virtual area
1447 * @addr: base address
1448 *
1449 * Search for the kernel VM area starting at @addr, and return it.
1450 * It is up to the caller to do all required locking to keep the returned
1451 * pointer valid.
1452 */
1453struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001454{
Nick Piggindb64fe02008-10-18 20:27:03 -07001455 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001456
Nick Piggindb64fe02008-10-18 20:27:03 -07001457 va = find_vmap_area((unsigned long)addr);
1458 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001459 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001460
Andi Kleen7856dfe2005-05-20 14:27:57 -07001461 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001462}
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464/**
Simon Arlott183ff222007-10-20 01:27:18 +02001465 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 * @addr: base address
1467 *
1468 * Search for the kernel VM area starting at @addr, and remove it.
1469 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001470 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001472struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
Nick Piggindb64fe02008-10-18 20:27:03 -07001474 struct vmap_area *va;
1475
Christoph Hellwig5803ed22016-12-12 16:44:20 -08001476 might_sleep();
1477
Nick Piggindb64fe02008-10-18 20:27:03 -07001478 va = find_vmap_area((unsigned long)addr);
1479 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001480 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001481
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001482 spin_lock(&vmap_area_lock);
1483 va->vm = NULL;
1484 va->flags &= ~VM_VM_AREA;
Yisheng Xie78c72742017-07-10 15:48:09 -07001485 va->flags |= VM_LAZY_FREE;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001486 spin_unlock(&vmap_area_lock);
1487
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001488 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001489 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001490
Nick Piggindb64fe02008-10-18 20:27:03 -07001491 return vm;
1492 }
1493 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494}
1495
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001496static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
1498 struct vm_struct *area;
1499
1500 if (!addr)
1501 return;
1502
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001503 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001504 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Chintan Pandyaf3c01d22018-06-07 17:06:50 -07001507 area = find_vmap_area((unsigned long)addr)->vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001509 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return;
1512 }
1513
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001514 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1515 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001516
Chintan Pandyaf3c01d22018-06-07 17:06:50 -07001517 remove_vm_area(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 if (deallocate_pages) {
1519 int i;
1520
1521 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001522 struct page *page = area->pages[i];
1523
1524 BUG_ON(!page);
Vladimir Davydov49491482016-07-26 15:24:24 -07001525 __free_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
1527
David Rientjes244d63e2016-01-14 15:19:35 -08001528 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 }
1530
1531 kfree(area);
1532 return;
1533}
Andrey Ryabininbf22e372016-12-12 16:44:10 -08001534
1535static inline void __vfree_deferred(const void *addr)
1536{
1537 /*
1538 * Use raw_cpu_ptr() because this can be called from preemptible
1539 * context. Preemption is absolutely fine here, because the llist_add()
1540 * implementation is lockless, so it works even if we are adding to
1541 * nother cpu's list. schedule_work() should be fine with this too.
1542 */
1543 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1544
1545 if (llist_add((struct llist_node *)addr, &p->list))
1546 schedule_work(&p->wq);
1547}
1548
1549/**
1550 * vfree_atomic - release memory allocated by vmalloc()
1551 * @addr: memory base address
1552 *
1553 * This one is just like vfree() but can be called in any atomic context
1554 * except NMIs.
1555 */
1556void vfree_atomic(const void *addr)
1557{
1558 BUG_ON(in_nmi());
1559
1560 kmemleak_free(addr);
1561
1562 if (!addr)
1563 return;
1564 __vfree_deferred(addr);
1565}
1566
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567/**
1568 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 * @addr: memory base address
1570 *
Simon Arlott183ff222007-10-20 01:27:18 +02001571 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001572 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1573 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 *
Al Viro32fcfd42013-03-10 20:14:08 -04001575 * Must not be called in NMI context (strictly speaking, only if we don't
1576 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1577 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001578 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03001579 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001581void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
Al Viro32fcfd42013-03-10 20:14:08 -04001583 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001584
1585 kmemleak_free(addr);
1586
Al Viro32fcfd42013-03-10 20:14:08 -04001587 if (!addr)
1588 return;
Andrey Ryabininbf22e372016-12-12 16:44:10 -08001589 if (unlikely(in_interrupt()))
1590 __vfree_deferred(addr);
1591 else
Al Viro32fcfd42013-03-10 20:14:08 -04001592 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594EXPORT_SYMBOL(vfree);
1595
1596/**
1597 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 * @addr: memory base address
1599 *
1600 * Free the virtually contiguous memory area starting at @addr,
1601 * which was created from the page array passed to vmap().
1602 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001603 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001605void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606{
1607 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001608 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001609 if (addr)
1610 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612EXPORT_SYMBOL(vunmap);
1613
1614/**
1615 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 * @pages: array of page pointers
1617 * @count: number of pages to map
1618 * @flags: vm_area->flags
1619 * @prot: page protection for the mapping
1620 *
1621 * Maps @count pages from @pages into contiguous kernel virtual
1622 * space.
1623 */
1624void *vmap(struct page **pages, unsigned int count,
1625 unsigned long flags, pgprot_t prot)
1626{
1627 struct vm_struct *area;
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001628 unsigned long size; /* In bytes */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629
Peter Zijlstra34754b62009-02-25 16:04:03 +01001630 might_sleep();
1631
Jan Beulich44813742009-09-21 17:03:05 -07001632 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 return NULL;
1634
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001635 size = (unsigned long)count << PAGE_SHIFT;
1636 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (!area)
1638 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001639
WANG Chaof6f8ed42014-08-06 16:06:58 -07001640 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 vunmap(area->addr);
1642 return NULL;
1643 }
1644
1645 return area->addr;
1646}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647EXPORT_SYMBOL(vmap);
1648
Michal Hocko8594a212017-05-12 15:46:41 -07001649static void *__vmalloc_node(unsigned long size, unsigned long align,
1650 gfp_t gfp_mask, pgprot_t prot,
1651 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001652static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001653 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 struct page **pages;
1656 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001657 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Laura Abbott704b8622017-08-18 15:16:27 -07001658 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1659 const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1660 0 :
1661 __GFP_HIGHMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
Wanpeng Li762216a2013-09-11 14:22:42 -07001663 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 array_size = (nr_pages * sizeof(struct page *));
1665
1666 area->nr_pages = nr_pages;
1667 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001668 if (array_size > PAGE_SIZE) {
Laura Abbott704b8622017-08-18 15:16:27 -07001669 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001670 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001671 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001672 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 area->pages = pages;
1675 if (!area->pages) {
1676 remove_vm_area(area->addr);
1677 kfree(area);
1678 return NULL;
1679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
1681 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001682 struct page *page;
1683
Jianguo Wu4b909512013-11-12 15:07:11 -08001684 if (node == NUMA_NO_NODE)
Laura Abbott704b8622017-08-18 15:16:27 -07001685 page = alloc_page(alloc_mask|highmem_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001686 else
Laura Abbott704b8622017-08-18 15:16:27 -07001687 page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001688
1689 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 /* Successfully allocated i pages, free them in __vunmap() */
1691 area->nr_pages = i;
1692 goto fail;
1693 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001694 area->pages[i] = page;
Laura Abbott704b8622017-08-18 15:16:27 -07001695 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001696 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698
WANG Chaof6f8ed42014-08-06 16:06:58 -07001699 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 goto fail;
1701 return area->addr;
1702
1703fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08001704 warn_alloc(gfp_mask, NULL,
Michal Hocko7877cdc2016-10-07 17:01:55 -07001705 "vmalloc: allocation failure, allocated %ld of %ld bytes",
Dave Hansen22943ab2011-05-24 17:12:18 -07001706 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 vfree(area->addr);
1708 return NULL;
1709}
1710
David Rientjesd0a21262011-01-13 15:46:02 -08001711/**
1712 * __vmalloc_node_range - allocate virtually contiguous memory
1713 * @size: allocation size
1714 * @align: desired alignment
1715 * @start: vm area range start
1716 * @end: vm area range end
1717 * @gfp_mask: flags for the page level allocator
1718 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001719 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001720 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001721 * @caller: caller's return address
1722 *
1723 * Allocate enough pages to cover @size from the page level
1724 * allocator with @gfp_mask flags. Map them into contiguous
1725 * kernel virtual space, using a pagetable protection of @prot.
1726 */
1727void *__vmalloc_node_range(unsigned long size, unsigned long align,
1728 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001729 pgprot_t prot, unsigned long vm_flags, int node,
1730 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001731{
David Rientjesd0a21262011-01-13 15:46:02 -08001732 struct vm_struct *area;
1733 void *addr;
1734 unsigned long real_size = size;
1735
1736 size = PAGE_ALIGN(size);
1737 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001738 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001739
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001740 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1741 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001742 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001743 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001744
Wanpeng Li3722e132013-11-12 15:07:29 -08001745 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001746 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001747 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001748
1749 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001750 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1751 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001752 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001753 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001754 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001755
Catalin Marinas94f4a162017-07-06 15:40:22 -07001756 kmemleak_vmalloc(area, size, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001757
1758 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001759
1760fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08001761 warn_alloc(gfp_mask, NULL,
Michal Hocko7877cdc2016-10-07 17:01:55 -07001762 "vmalloc: allocation failure: %lu bytes", real_size);
Joe Perchesde7d2b52011-10-31 17:08:48 -07001763 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001764}
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001767 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001769 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 * @gfp_mask: flags for the page level allocator
1771 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001772 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001773 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 *
1775 * Allocate enough pages to cover @size from the page level
1776 * allocator with @gfp_mask flags. Map them into contiguous
1777 * kernel virtual space, using a pagetable protection of @prot.
Michal Hockoa7c3e902017-05-08 15:57:09 -07001778 *
Michal Hockodcda9b02017-07-12 14:36:45 -07001779 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
Michal Hockoa7c3e902017-05-08 15:57:09 -07001780 * and __GFP_NOFAIL are not supported
1781 *
1782 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1783 * with mm people.
1784 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 */
Michal Hocko8594a212017-05-12 15:46:41 -07001786static void *__vmalloc_node(unsigned long size, unsigned long align,
David Miller2dca6992009-09-21 12:22:34 -07001787 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001788 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789{
David Rientjesd0a21262011-01-13 15:46:02 -08001790 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001791 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792}
1793
Christoph Lameter930fc452005-10-29 18:15:41 -07001794void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1795{
David Rientjes00ef2d22013-02-22 16:35:36 -08001796 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001797 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001798}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799EXPORT_SYMBOL(__vmalloc);
1800
Michal Hocko8594a212017-05-12 15:46:41 -07001801static inline void *__vmalloc_node_flags(unsigned long size,
1802 int node, gfp_t flags)
1803{
1804 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1805 node, __builtin_return_address(0));
1806}
1807
1808
1809void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
1810 void *caller)
1811{
1812 return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
1813}
1814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815/**
1816 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 * Allocate enough pages to cover @size from the page level
1819 * allocator and map them into contiguous kernel virtual space.
1820 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001821 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 * use __vmalloc() instead.
1823 */
1824void *vmalloc(unsigned long size)
1825{
David Rientjes00ef2d22013-02-22 16:35:36 -08001826 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Michal Hocko19809c22017-05-08 15:57:44 -07001827 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829EXPORT_SYMBOL(vmalloc);
1830
Christoph Lameter930fc452005-10-29 18:15:41 -07001831/**
Dave Younge1ca7782010-10-26 14:22:06 -07001832 * vzalloc - allocate virtually contiguous memory with zero fill
1833 * @size: allocation size
1834 * Allocate enough pages to cover @size from the page level
1835 * allocator and map them into contiguous kernel virtual space.
1836 * The memory allocated is set to zero.
1837 *
1838 * For tight control over page level allocator and protection flags
1839 * use __vmalloc() instead.
1840 */
1841void *vzalloc(unsigned long size)
1842{
David Rientjes00ef2d22013-02-22 16:35:36 -08001843 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Michal Hocko19809c22017-05-08 15:57:44 -07001844 GFP_KERNEL | __GFP_ZERO);
Dave Younge1ca7782010-10-26 14:22:06 -07001845}
1846EXPORT_SYMBOL(vzalloc);
1847
1848/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001849 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1850 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001851 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001852 * The resulting memory area is zeroed so it can be mapped to userspace
1853 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001854 */
1855void *vmalloc_user(unsigned long size)
1856{
1857 struct vm_struct *area;
1858 void *ret;
1859
David Miller2dca6992009-09-21 12:22:34 -07001860 ret = __vmalloc_node(size, SHMLBA,
Michal Hocko19809c22017-05-08 15:57:44 -07001861 GFP_KERNEL | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001862 PAGE_KERNEL, NUMA_NO_NODE,
1863 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001864 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001865 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001866 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001867 }
Nick Piggin83342312006-06-23 02:03:20 -07001868 return ret;
1869}
1870EXPORT_SYMBOL(vmalloc_user);
1871
1872/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001873 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001874 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001875 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001876 *
1877 * Allocate enough pages to cover @size from the page level
1878 * allocator and map them into contiguous kernel virtual space.
1879 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001880 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001881 * use __vmalloc() instead.
1882 */
1883void *vmalloc_node(unsigned long size, int node)
1884{
Michal Hocko19809c22017-05-08 15:57:44 -07001885 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001886 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001887}
1888EXPORT_SYMBOL(vmalloc_node);
1889
Dave Younge1ca7782010-10-26 14:22:06 -07001890/**
1891 * vzalloc_node - allocate memory on a specific node with zero fill
1892 * @size: allocation size
1893 * @node: numa node
1894 *
1895 * Allocate enough pages to cover @size from the page level
1896 * allocator and map them into contiguous kernel virtual space.
1897 * The memory allocated is set to zero.
1898 *
1899 * For tight control over page level allocator and protection flags
1900 * use __vmalloc_node() instead.
1901 */
1902void *vzalloc_node(unsigned long size, int node)
1903{
1904 return __vmalloc_node_flags(size, node,
Michal Hocko19809c22017-05-08 15:57:44 -07001905 GFP_KERNEL | __GFP_ZERO);
Dave Younge1ca7782010-10-26 14:22:06 -07001906}
1907EXPORT_SYMBOL(vzalloc_node);
1908
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001909#ifndef PAGE_KERNEL_EXEC
1910# define PAGE_KERNEL_EXEC PAGE_KERNEL
1911#endif
1912
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913/**
1914 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 * @size: allocation size
1916 *
1917 * Kernel-internal function to allocate enough pages to cover @size
1918 * the page level allocator and map them into contiguous and
1919 * executable kernel virtual space.
1920 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001921 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 * use __vmalloc() instead.
1923 */
1924
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925void *vmalloc_exec(unsigned long size)
1926{
Michal Hocko19809c22017-05-08 15:57:44 -07001927 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001928 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929}
1930
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001931#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Michal Hocko698d0832018-02-21 14:46:01 -08001932#define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001933#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Michal Hocko698d0832018-02-21 14:46:01 -08001934#define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001935#else
Michal Hocko698d0832018-02-21 14:46:01 -08001936/*
1937 * 64b systems should always have either DMA or DMA32 zones. For others
1938 * GFP_DMA32 should do the right thing and use the normal zone.
1939 */
1940#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001941#endif
1942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943/**
1944 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 * @size: allocation size
1946 *
1947 * Allocate enough 32bit PA addressable pages to cover @size from the
1948 * page level allocator and map them into contiguous kernel virtual space.
1949 */
1950void *vmalloc_32(unsigned long size)
1951{
David Miller2dca6992009-09-21 12:22:34 -07001952 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001953 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955EXPORT_SYMBOL(vmalloc_32);
1956
Nick Piggin83342312006-06-23 02:03:20 -07001957/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001958 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001959 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001960 *
1961 * The resulting memory area is 32bit addressable and zeroed so it can be
1962 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001963 */
1964void *vmalloc_32_user(unsigned long size)
1965{
1966 struct vm_struct *area;
1967 void *ret;
1968
David Miller2dca6992009-09-21 12:22:34 -07001969 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001970 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001971 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001972 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001973 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001974 }
Nick Piggin83342312006-06-23 02:03:20 -07001975 return ret;
1976}
1977EXPORT_SYMBOL(vmalloc_32_user);
1978
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001979/*
1980 * small helper routine , copy contents to buf from addr.
1981 * If the page is not present, fill zero.
1982 */
1983
1984static int aligned_vread(char *buf, char *addr, unsigned long count)
1985{
1986 struct page *p;
1987 int copied = 0;
1988
1989 while (count) {
1990 unsigned long offset, length;
1991
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001992 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001993 length = PAGE_SIZE - offset;
1994 if (length > count)
1995 length = count;
1996 p = vmalloc_to_page(addr);
1997 /*
1998 * To do safe access to this _mapped_ area, we need
1999 * lock. But adding lock here means that we need to add
2000 * overhead of vmalloc()/vfree() calles for this _debug_
2001 * interface, rarely used. Instead of that, we'll use
2002 * kmap() and get small overhead in this access function.
2003 */
2004 if (p) {
2005 /*
2006 * we can expect USER0 is not used (see vread/vwrite's
2007 * function description)
2008 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002009 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002010 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002011 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002012 } else
2013 memset(buf, 0, length);
2014
2015 addr += length;
2016 buf += length;
2017 copied += length;
2018 count -= length;
2019 }
2020 return copied;
2021}
2022
2023static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2024{
2025 struct page *p;
2026 int copied = 0;
2027
2028 while (count) {
2029 unsigned long offset, length;
2030
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002031 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002032 length = PAGE_SIZE - offset;
2033 if (length > count)
2034 length = count;
2035 p = vmalloc_to_page(addr);
2036 /*
2037 * To do safe access to this _mapped_ area, we need
2038 * lock. But adding lock here means that we need to add
2039 * overhead of vmalloc()/vfree() calles for this _debug_
2040 * interface, rarely used. Instead of that, we'll use
2041 * kmap() and get small overhead in this access function.
2042 */
2043 if (p) {
2044 /*
2045 * we can expect USER0 is not used (see vread/vwrite's
2046 * function description)
2047 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002048 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002049 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002050 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002051 }
2052 addr += length;
2053 buf += length;
2054 copied += length;
2055 count -= length;
2056 }
2057 return copied;
2058}
2059
2060/**
2061 * vread() - read vmalloc area in a safe way.
2062 * @buf: buffer for reading data
2063 * @addr: vm address.
2064 * @count: number of bytes to be read.
2065 *
2066 * Returns # of bytes which addr and buf should be increased.
2067 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2068 * includes any intersect with alive vmalloc area.
2069 *
2070 * This function checks that addr is a valid vmalloc'ed area, and
2071 * copy data from that area to a given buffer. If the given memory range
2072 * of [addr...addr+count) includes some valid address, data is copied to
2073 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2074 * IOREMAP area is treated as memory hole and no copy is done.
2075 *
2076 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002077 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002078 *
2079 * Note: In usual ops, vread() is never necessary because the caller
2080 * should know vmalloc() area is valid and can use memcpy().
2081 * This is for routines which have to access vmalloc area without
2082 * any informaion, as /dev/kmem.
2083 *
2084 */
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086long vread(char *buf, char *addr, unsigned long count)
2087{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002088 struct vmap_area *va;
2089 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002091 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 unsigned long n;
2093
2094 /* Don't allow overflow */
2095 if ((unsigned long) addr + count < count)
2096 count = -(unsigned long) addr;
2097
Joonsoo Kime81ce852013-04-29 15:07:32 -07002098 spin_lock(&vmap_area_lock);
2099 list_for_each_entry(va, &vmap_area_list, list) {
2100 if (!count)
2101 break;
2102
2103 if (!(va->flags & VM_VM_AREA))
2104 continue;
2105
2106 vm = va->vm;
2107 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002108 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 continue;
2110 while (addr < vaddr) {
2111 if (count == 0)
2112 goto finished;
2113 *buf = '\0';
2114 buf++;
2115 addr++;
2116 count--;
2117 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002118 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002119 if (n > count)
2120 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002121 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002122 aligned_vread(buf, addr, n);
2123 else /* IOREMAP area is treated as memory hole */
2124 memset(buf, 0, n);
2125 buf += n;
2126 addr += n;
2127 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 }
2129finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002130 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002131
2132 if (buf == buf_start)
2133 return 0;
2134 /* zero-fill memory holes */
2135 if (buf != buf_start + buflen)
2136 memset(buf, 0, buflen - (buf - buf_start));
2137
2138 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139}
2140
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002141/**
2142 * vwrite() - write vmalloc area in a safe way.
2143 * @buf: buffer for source data
2144 * @addr: vm address.
2145 * @count: number of bytes to be read.
2146 *
2147 * Returns # of bytes which addr and buf should be incresed.
2148 * (same number to @count).
2149 * If [addr...addr+count) doesn't includes any intersect with valid
2150 * vmalloc area, returns 0.
2151 *
2152 * This function checks that addr is a valid vmalloc'ed area, and
2153 * copy data from a buffer to the given addr. If specified range of
2154 * [addr...addr+count) includes some valid address, data is copied from
2155 * proper area of @buf. If there are memory holes, no copy to hole.
2156 * IOREMAP area is treated as memory hole and no copy is done.
2157 *
2158 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002159 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002160 *
2161 * Note: In usual ops, vwrite() is never necessary because the caller
2162 * should know vmalloc() area is valid and can use memcpy().
2163 * This is for routines which have to access vmalloc area without
2164 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002165 */
2166
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167long vwrite(char *buf, char *addr, unsigned long count)
2168{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002169 struct vmap_area *va;
2170 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002171 char *vaddr;
2172 unsigned long n, buflen;
2173 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
2175 /* Don't allow overflow */
2176 if ((unsigned long) addr + count < count)
2177 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002178 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Joonsoo Kime81ce852013-04-29 15:07:32 -07002180 spin_lock(&vmap_area_lock);
2181 list_for_each_entry(va, &vmap_area_list, list) {
2182 if (!count)
2183 break;
2184
2185 if (!(va->flags & VM_VM_AREA))
2186 continue;
2187
2188 vm = va->vm;
2189 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002190 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 continue;
2192 while (addr < vaddr) {
2193 if (count == 0)
2194 goto finished;
2195 buf++;
2196 addr++;
2197 count--;
2198 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002199 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002200 if (n > count)
2201 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002202 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002203 aligned_vwrite(buf, addr, n);
2204 copied++;
2205 }
2206 buf += n;
2207 addr += n;
2208 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 }
2210finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002211 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002212 if (!copied)
2213 return 0;
2214 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215}
Nick Piggin83342312006-06-23 02:03:20 -07002216
2217/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002218 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2219 * @vma: vma to cover
2220 * @uaddr: target user address to start at
2221 * @kaddr: virtual address of vmalloc kernel memory
2222 * @size: size of map area
2223 *
2224 * Returns: 0 for success, -Exxx on failure
2225 *
2226 * This function checks that @kaddr is a valid vmalloc'ed area,
2227 * and that it is big enough to cover the range starting at
2228 * @uaddr in @vma. Will return failure if that criteria isn't
2229 * met.
2230 *
2231 * Similar to remap_pfn_range() (see mm/memory.c)
2232 */
2233int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2234 void *kaddr, unsigned long size)
2235{
2236 struct vm_struct *area;
2237
2238 size = PAGE_ALIGN(size);
2239
2240 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2241 return -EINVAL;
2242
2243 area = find_vm_area(kaddr);
2244 if (!area)
2245 return -EINVAL;
2246
2247 if (!(area->flags & VM_USERMAP))
2248 return -EINVAL;
2249
2250 if (kaddr + size > area->addr + area->size)
2251 return -EINVAL;
2252
2253 do {
2254 struct page *page = vmalloc_to_page(kaddr);
2255 int ret;
2256
2257 ret = vm_insert_page(vma, uaddr, page);
2258 if (ret)
2259 return ret;
2260
2261 uaddr += PAGE_SIZE;
2262 kaddr += PAGE_SIZE;
2263 size -= PAGE_SIZE;
2264 } while (size > 0);
2265
2266 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2267
2268 return 0;
2269}
2270EXPORT_SYMBOL(remap_vmalloc_range_partial);
2271
2272/**
Nick Piggin83342312006-06-23 02:03:20 -07002273 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002274 * @vma: vma to cover (map full range of vma)
2275 * @addr: vmalloc memory
2276 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002277 *
2278 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002279 *
2280 * This function checks that addr is a valid vmalloc'ed area, and
2281 * that it is big enough to cover the vma. Will return failure if
2282 * that criteria isn't met.
2283 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002284 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002285 */
2286int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2287 unsigned long pgoff)
2288{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002289 return remap_vmalloc_range_partial(vma, vma->vm_start,
2290 addr + (pgoff << PAGE_SHIFT),
2291 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002292}
2293EXPORT_SYMBOL(remap_vmalloc_range);
2294
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002295/*
2296 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2297 * have one.
2298 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002299void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002300{
2301}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002302
2303
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002304static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002305{
David Vrabelcd129092011-09-29 16:53:32 +01002306 pte_t ***p = data;
2307
2308 if (p) {
2309 *(*p) = pte;
2310 (*p)++;
2311 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002312 return 0;
2313}
2314
2315/**
2316 * alloc_vm_area - allocate a range of kernel address space
2317 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002318 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002319 *
2320 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002321 *
2322 * This function reserves a range of kernel address space, and
2323 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002324 * are created.
2325 *
2326 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2327 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002328 */
David Vrabelcd129092011-09-29 16:53:32 +01002329struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002330{
2331 struct vm_struct *area;
2332
Christoph Lameter23016962008-04-28 02:12:42 -07002333 area = get_vm_area_caller(size, VM_IOREMAP,
2334 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002335 if (area == NULL)
2336 return NULL;
2337
2338 /*
2339 * This ensures that page tables are constructed for this region
2340 * of kernel virtual address space and mapped into init_mm.
2341 */
2342 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002343 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002344 free_vm_area(area);
2345 return NULL;
2346 }
2347
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002348 return area;
2349}
2350EXPORT_SYMBOL_GPL(alloc_vm_area);
2351
2352void free_vm_area(struct vm_struct *area)
2353{
2354 struct vm_struct *ret;
2355 ret = remove_vm_area(area->addr);
2356 BUG_ON(ret != area);
2357 kfree(area);
2358}
2359EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002360
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002361#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002362static struct vmap_area *node_to_va(struct rb_node *n)
2363{
Geliang Tang4583e772017-02-22 15:41:54 -08002364 return rb_entry_safe(n, struct vmap_area, rb_node);
Tejun Heoca23e402009-08-14 15:00:52 +09002365}
2366
2367/**
2368 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2369 * @end: target address
2370 * @pnext: out arg for the next vmap_area
2371 * @pprev: out arg for the previous vmap_area
2372 *
2373 * Returns: %true if either or both of next and prev are found,
2374 * %false if no vmap_area exists
2375 *
2376 * Find vmap_areas end addresses of which enclose @end. ie. if not
2377 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2378 */
2379static bool pvm_find_next_prev(unsigned long end,
2380 struct vmap_area **pnext,
2381 struct vmap_area **pprev)
2382{
2383 struct rb_node *n = vmap_area_root.rb_node;
2384 struct vmap_area *va = NULL;
2385
2386 while (n) {
2387 va = rb_entry(n, struct vmap_area, rb_node);
2388 if (end < va->va_end)
2389 n = n->rb_left;
2390 else if (end > va->va_end)
2391 n = n->rb_right;
2392 else
2393 break;
2394 }
2395
2396 if (!va)
2397 return false;
2398
2399 if (va->va_end > end) {
2400 *pnext = va;
2401 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2402 } else {
2403 *pprev = va;
2404 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2405 }
2406 return true;
2407}
2408
2409/**
2410 * pvm_determine_end - find the highest aligned address between two vmap_areas
2411 * @pnext: in/out arg for the next vmap_area
2412 * @pprev: in/out arg for the previous vmap_area
2413 * @align: alignment
2414 *
2415 * Returns: determined end address
2416 *
2417 * Find the highest aligned address between *@pnext and *@pprev below
2418 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2419 * down address is between the end addresses of the two vmap_areas.
2420 *
2421 * Please note that the address returned by this function may fall
2422 * inside *@pnext vmap_area. The caller is responsible for checking
2423 * that.
2424 */
2425static unsigned long pvm_determine_end(struct vmap_area **pnext,
2426 struct vmap_area **pprev,
2427 unsigned long align)
2428{
2429 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2430 unsigned long addr;
2431
2432 if (*pnext)
2433 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2434 else
2435 addr = vmalloc_end;
2436
2437 while (*pprev && (*pprev)->va_end > addr) {
2438 *pnext = *pprev;
2439 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2440 }
2441
2442 return addr;
2443}
2444
2445/**
2446 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2447 * @offsets: array containing offset of each area
2448 * @sizes: array containing size of each area
2449 * @nr_vms: the number of areas to allocate
2450 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002451 *
2452 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2453 * vm_structs on success, %NULL on failure
2454 *
2455 * Percpu allocator wants to use congruent vm areas so that it can
2456 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002457 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2458 * be scattered pretty far, distance between two areas easily going up
2459 * to gigabytes. To avoid interacting with regular vmallocs, these
2460 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002461 *
2462 * Despite its complicated look, this allocator is rather simple. It
2463 * does everything top-down and scans areas from the end looking for
2464 * matching slot. While scanning, if any of the areas overlaps with
2465 * existing vmap_area, the base address is pulled down to fit the
2466 * area. Scanning is repeated till all the areas fit and then all
Wei Yangc568da22017-09-06 16:24:09 -07002467 * necessary data structures are inserted and the result is returned.
Tejun Heoca23e402009-08-14 15:00:52 +09002468 */
2469struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2470 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002471 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002472{
2473 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2474 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2475 struct vmap_area **vas, *prev, *next;
2476 struct vm_struct **vms;
2477 int area, area2, last_area, term_area;
2478 unsigned long base, start, end, last_end;
2479 bool purged = false;
2480
Tejun Heoca23e402009-08-14 15:00:52 +09002481 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002482 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002483 for (last_area = 0, area = 0; area < nr_vms; area++) {
2484 start = offsets[area];
2485 end = start + sizes[area];
2486
2487 /* is everything aligned properly? */
2488 BUG_ON(!IS_ALIGNED(offsets[area], align));
2489 BUG_ON(!IS_ALIGNED(sizes[area], align));
2490
2491 /* detect the area with the highest address */
2492 if (start > offsets[last_area])
2493 last_area = area;
2494
Wei Yangc568da22017-09-06 16:24:09 -07002495 for (area2 = area + 1; area2 < nr_vms; area2++) {
Tejun Heoca23e402009-08-14 15:00:52 +09002496 unsigned long start2 = offsets[area2];
2497 unsigned long end2 = start2 + sizes[area2];
2498
Wei Yangc568da22017-09-06 16:24:09 -07002499 BUG_ON(start2 < end && start < end2);
Tejun Heoca23e402009-08-14 15:00:52 +09002500 }
2501 }
2502 last_end = offsets[last_area] + sizes[last_area];
2503
2504 if (vmalloc_end - vmalloc_start < last_end) {
2505 WARN_ON(true);
2506 return NULL;
2507 }
2508
Thomas Meyer4d67d862012-05-29 15:06:21 -07002509 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2510 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002511 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002512 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002513
2514 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002515 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2516 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002517 if (!vas[area] || !vms[area])
2518 goto err_free;
2519 }
2520retry:
2521 spin_lock(&vmap_area_lock);
2522
2523 /* start scanning - we scan from the top, begin with the last area */
2524 area = term_area = last_area;
2525 start = offsets[area];
2526 end = start + sizes[area];
2527
2528 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2529 base = vmalloc_end - last_end;
2530 goto found;
2531 }
2532 base = pvm_determine_end(&next, &prev, align) - end;
2533
2534 while (true) {
2535 BUG_ON(next && next->va_end <= base + end);
2536 BUG_ON(prev && prev->va_end > base + end);
2537
2538 /*
2539 * base might have underflowed, add last_end before
2540 * comparing.
2541 */
2542 if (base + last_end < vmalloc_start + last_end) {
2543 spin_unlock(&vmap_area_lock);
2544 if (!purged) {
2545 purge_vmap_area_lazy();
2546 purged = true;
2547 goto retry;
2548 }
2549 goto err_free;
2550 }
2551
2552 /*
2553 * If next overlaps, move base downwards so that it's
2554 * right below next and then recheck.
2555 */
2556 if (next && next->va_start < base + end) {
2557 base = pvm_determine_end(&next, &prev, align) - end;
2558 term_area = area;
2559 continue;
2560 }
2561
2562 /*
2563 * If prev overlaps, shift down next and prev and move
2564 * base so that it's right below new next and then
2565 * recheck.
2566 */
2567 if (prev && prev->va_end > base + start) {
2568 next = prev;
2569 prev = node_to_va(rb_prev(&next->rb_node));
2570 base = pvm_determine_end(&next, &prev, align) - end;
2571 term_area = area;
2572 continue;
2573 }
2574
2575 /*
2576 * This area fits, move on to the previous one. If
2577 * the previous one is the terminal one, we're done.
2578 */
2579 area = (area + nr_vms - 1) % nr_vms;
2580 if (area == term_area)
2581 break;
2582 start = offsets[area];
2583 end = start + sizes[area];
2584 pvm_find_next_prev(base + end, &next, &prev);
2585 }
2586found:
2587 /* we've found a fitting base, insert all va's */
2588 for (area = 0; area < nr_vms; area++) {
2589 struct vmap_area *va = vas[area];
2590
2591 va->va_start = base + offsets[area];
2592 va->va_end = va->va_start + sizes[area];
2593 __insert_vmap_area(va);
2594 }
2595
2596 vmap_area_pcpu_hole = base + offsets[last_area];
2597
2598 spin_unlock(&vmap_area_lock);
2599
2600 /* insert all vm's */
2601 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002602 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2603 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002604
2605 kfree(vas);
2606 return vms;
2607
2608err_free:
2609 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002610 kfree(vas[area]);
2611 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002612 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002613err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002614 kfree(vas);
2615 kfree(vms);
2616 return NULL;
2617}
2618
2619/**
2620 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2621 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2622 * @nr_vms: the number of allocated areas
2623 *
2624 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2625 */
2626void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2627{
2628 int i;
2629
2630 for (i = 0; i < nr_vms; i++)
2631 free_vm_area(vms[i]);
2632 kfree(vms);
2633}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002634#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002635
2636#ifdef CONFIG_PROC_FS
2637static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002638 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002639{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002640 spin_lock(&vmap_area_lock);
zijun_hu3f500062016-12-12 16:42:17 -08002641 return seq_list_start(&vmap_area_list, *pos);
Christoph Lametera10aa572008-04-28 02:12:40 -07002642}
2643
2644static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2645{
zijun_hu3f500062016-12-12 16:42:17 -08002646 return seq_list_next(p, &vmap_area_list, pos);
Christoph Lametera10aa572008-04-28 02:12:40 -07002647}
2648
2649static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002650 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002651{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002652 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002653}
2654
Eric Dumazeta47a1262008-07-23 21:27:38 -07002655static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2656{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002657 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002658 unsigned int nr, *counters = m->private;
2659
2660 if (!counters)
2661 return;
2662
Wanpeng Liaf123462013-11-12 15:07:32 -08002663 if (v->flags & VM_UNINITIALIZED)
2664 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002665 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2666 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002667
Eric Dumazeta47a1262008-07-23 21:27:38 -07002668 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2669
2670 for (nr = 0; nr < v->nr_pages; nr++)
2671 counters[page_to_nid(v->pages[nr])]++;
2672
2673 for_each_node_state(nr, N_HIGH_MEMORY)
2674 if (counters[nr])
2675 seq_printf(m, " N%u=%u", nr, counters[nr]);
2676 }
2677}
2678
Christoph Lametera10aa572008-04-28 02:12:40 -07002679static int s_show(struct seq_file *m, void *p)
2680{
zijun_hu3f500062016-12-12 16:42:17 -08002681 struct vmap_area *va;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002682 struct vm_struct *v;
2683
zijun_hu3f500062016-12-12 16:42:17 -08002684 va = list_entry(p, struct vmap_area, list);
2685
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002686 /*
2687 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2688 * behalf of vmap area is being tear down or vm_map_ram allocation.
2689 */
Yisheng Xie78c72742017-07-10 15:48:09 -07002690 if (!(va->flags & VM_VM_AREA)) {
2691 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2692 (void *)va->va_start, (void *)va->va_end,
2693 va->va_end - va->va_start,
2694 va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2695
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002696 return 0;
Yisheng Xie78c72742017-07-10 15:48:09 -07002697 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002698
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002699 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002700
Kees Cook45ec1692012-10-08 16:34:09 -07002701 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002702 v->addr, v->addr + v->size, v->size);
2703
Joe Perches62c70bc2011-01-13 15:45:52 -08002704 if (v->caller)
2705 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002706
Christoph Lametera10aa572008-04-28 02:12:40 -07002707 if (v->nr_pages)
2708 seq_printf(m, " pages=%d", v->nr_pages);
2709
2710 if (v->phys_addr)
Miles Chen199eaa02017-02-24 14:59:51 -08002711 seq_printf(m, " phys=%pa", &v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002712
2713 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002714 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002715
2716 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002717 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002718
2719 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002720 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002721
2722 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002723 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002724
David Rientjes244d63e2016-01-14 15:19:35 -08002725 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002726 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002727
Eric Dumazeta47a1262008-07-23 21:27:38 -07002728 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002729 seq_putc(m, '\n');
2730 return 0;
2731}
2732
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002733static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002734 .start = s_start,
2735 .next = s_next,
2736 .stop = s_stop,
2737 .show = s_show,
2738};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002739
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002740static int __init proc_vmalloc_init(void)
2741{
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02002742 if (IS_ENABLED(CONFIG_NUMA))
Christoph Hellwig44414d82018-04-24 17:05:17 +02002743 proc_create_seq_private("vmallocinfo", S_IRUSR, NULL,
2744 &vmalloc_op,
2745 nr_node_ids * sizeof(unsigned int), NULL);
Christoph Hellwigfddda2b2018-04-13 19:44:18 +02002746 else
2747 proc_create_seq("vmallocinfo", S_IRUSR, NULL, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002748 return 0;
2749}
2750module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002751
Christoph Lametera10aa572008-04-28 02:12:40 -07002752#endif
2753