blob: 6b5f0bce0bd2ac40048a998f9f062d59087a0e53 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/vmalloc.c
3 *
4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
Christoph Lameter930fc452005-10-29 18:15:41 -07008 * Numa awareness, Christoph Lameter, SGI, June 2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Nick Piggindb64fe02008-10-18 20:27:03 -070011#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mm.h>
13#include <linux/module.h>
14#include <linux/highmem.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040015#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/spinlock.h>
18#include <linux/interrupt.h>
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +040019#include <linux/proc_fs.h>
Christoph Lametera10aa572008-04-28 02:12:40 -070020#include <linux/seq_file.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070021#include <linux/debugobjects.h>
Christoph Lameter23016962008-04-28 02:12:42 -070022#include <linux/kallsyms.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070023#include <linux/list.h>
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 Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36#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);
52 struct llist_node *llnode = llist_del_all(&p->list);
53 while (llnode) {
54 void *p = llnode;
55 llnode = llist_next(llnode);
56 __vunmap(p, 1);
57 }
58}
59
Nick Piggindb64fe02008-10-18 20:27:03 -070060/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070061
Linus Torvalds1da177e2005-04-16 15:20:36 -070062static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
63{
64 pte_t *pte;
65
66 pte = pte_offset_kernel(pmd, addr);
67 do {
68 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
69 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
70 } while (pte++, addr += PAGE_SIZE, addr != end);
71}
72
Nick Piggindb64fe02008-10-18 20:27:03 -070073static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 pmd_t *pmd;
76 unsigned long next;
77
78 pmd = pmd_offset(pud, addr);
79 do {
80 next = pmd_addr_end(addr, end);
Toshi Kanib9820d82015-04-14 15:47:26 -070081 if (pmd_clear_huge(pmd))
82 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 if (pmd_none_or_clear_bad(pmd))
84 continue;
85 vunmap_pte_range(pmd, addr, next);
86 } while (pmd++, addr = next, addr != end);
87}
88
Nick Piggindb64fe02008-10-18 20:27:03 -070089static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
91 pud_t *pud;
92 unsigned long next;
93
94 pud = pud_offset(pgd, addr);
95 do {
96 next = pud_addr_end(addr, end);
Toshi Kanib9820d82015-04-14 15:47:26 -070097 if (pud_clear_huge(pud))
98 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 if (pud_none_or_clear_bad(pud))
100 continue;
101 vunmap_pmd_range(pud, addr, next);
102 } while (pud++, addr = next, addr != end);
103}
104
Nick Piggindb64fe02008-10-18 20:27:03 -0700105static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
107 pgd_t *pgd;
108 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110 BUG_ON(addr >= end);
111 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 do {
113 next = pgd_addr_end(addr, end);
114 if (pgd_none_or_clear_bad(pgd))
115 continue;
116 vunmap_pud_range(pgd, addr, next);
117 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
119
120static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -0700121 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 pte_t *pte;
124
Nick Piggindb64fe02008-10-18 20:27:03 -0700125 /*
126 * nr is a running index into the array which helps higher level
127 * callers keep track of where we're up to.
128 */
129
Hugh Dickins872fec12005-10-29 18:16:21 -0700130 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (!pte)
132 return -ENOMEM;
133 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700134 struct page *page = pages[*nr];
135
136 if (WARN_ON(!pte_none(*pte)))
137 return -EBUSY;
138 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return -ENOMEM;
140 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700141 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 } while (pte++, addr += PAGE_SIZE, addr != end);
143 return 0;
144}
145
Nick Piggindb64fe02008-10-18 20:27:03 -0700146static int vmap_pmd_range(pud_t *pud, unsigned long addr,
147 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 pmd_t *pmd;
150 unsigned long next;
151
152 pmd = pmd_alloc(&init_mm, pud, addr);
153 if (!pmd)
154 return -ENOMEM;
155 do {
156 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700157 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 return -ENOMEM;
159 } while (pmd++, addr = next, addr != end);
160 return 0;
161}
162
Nick Piggindb64fe02008-10-18 20:27:03 -0700163static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
164 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
166 pud_t *pud;
167 unsigned long next;
168
169 pud = pud_alloc(&init_mm, pgd, addr);
170 if (!pud)
171 return -ENOMEM;
172 do {
173 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700174 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return -ENOMEM;
176 } while (pud++, addr = next, addr != end);
177 return 0;
178}
179
Nick Piggindb64fe02008-10-18 20:27:03 -0700180/*
181 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
182 * will have pfns corresponding to the "pages" array.
183 *
184 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
185 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900186static int vmap_page_range_noflush(unsigned long start, unsigned long end,
187 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
189 pgd_t *pgd;
190 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800191 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700192 int err = 0;
193 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 BUG_ON(addr >= end);
196 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 do {
198 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700199 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700201 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700203
Nick Piggindb64fe02008-10-18 20:27:03 -0700204 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
Tejun Heo8fc48982009-02-20 16:29:08 +0900207static int vmap_page_range(unsigned long start, unsigned long end,
208 pgprot_t prot, struct page **pages)
209{
210 int ret;
211
212 ret = vmap_page_range_noflush(start, end, prot, pages);
213 flush_cache_vmap(start, end);
214 return ret;
215}
216
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700217int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700218{
219 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000220 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700221 * and fall back on vmalloc() if that fails. Others
222 * just put it in the vmalloc space.
223 */
224#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
225 unsigned long addr = (unsigned long)x;
226 if (addr >= MODULES_VADDR && addr < MODULES_END)
227 return 1;
228#endif
229 return is_vmalloc_addr(x);
230}
231
Christoph Lameter48667e72008-02-04 22:28:31 -0800232/*
malcadd688f2014-01-27 17:06:53 -0800233 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800234 */
malcadd688f2014-01-27 17:06:53 -0800235struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800236{
237 unsigned long addr = (unsigned long) vmalloc_addr;
malcadd688f2014-01-27 17:06:53 -0800238 struct page *page = NULL;
Christoph Lameter48667e72008-02-04 22:28:31 -0800239 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800240
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200241 /*
242 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
243 * architectures that do not vmalloc module space
244 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700245 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200246
Ard Biesheuvel647f6052017-06-23 15:08:41 -0700247 /*
248 * Don't dereference bad PUD or PMD (below) entries. This will also
249 * identify huge mappings, which we may encounter on architectures
250 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
251 * identified as vmalloc addresses by is_vmalloc_addr(), but are
252 * not [unambiguously] associated with a struct page, so there is
253 * no correct value to return for them.
254 */
Christoph Lameter48667e72008-02-04 22:28:31 -0800255 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700256 pud_t *pud = pud_offset(pgd, addr);
Ard Biesheuvel647f6052017-06-23 15:08:41 -0700257 WARN_ON_ONCE(pud_bad(*pud));
258 if (!pud_none(*pud) && !pud_bad(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700259 pmd_t *pmd = pmd_offset(pud, addr);
Ard Biesheuvel647f6052017-06-23 15:08:41 -0700260 WARN_ON_ONCE(pmd_bad(*pmd));
261 if (!pmd_none(*pmd) && !pmd_bad(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700262 pte_t *ptep, pte;
263
Christoph Lameter48667e72008-02-04 22:28:31 -0800264 ptep = pte_offset_map(pmd, addr);
265 pte = *ptep;
266 if (pte_present(pte))
malcadd688f2014-01-27 17:06:53 -0800267 page = pte_page(pte);
Christoph Lameter48667e72008-02-04 22:28:31 -0800268 pte_unmap(ptep);
269 }
270 }
271 }
malcadd688f2014-01-27 17:06:53 -0800272 return page;
Jianyu Zhanece86e222014-01-21 15:49:12 -0800273}
274EXPORT_SYMBOL(vmalloc_to_page);
275
malcadd688f2014-01-27 17:06:53 -0800276/*
277 * Map a vmalloc()-space virtual address to the physical page frame number.
278 */
279unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
280{
281 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
282}
283EXPORT_SYMBOL(vmalloc_to_pfn);
284
Nick Piggindb64fe02008-10-18 20:27:03 -0700285
286/*** Global kva allocator ***/
287
Nick Piggindb64fe02008-10-18 20:27:03 -0700288#define VM_VM_AREA 0x04
289
Nick Piggindb64fe02008-10-18 20:27:03 -0700290static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700291/* Export for kexec only */
292LIST_HEAD(vmap_area_list);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700293static LLIST_HEAD(vmap_purge_list);
Nick Piggin89699602011-03-22 16:30:36 -0700294static struct rb_root vmap_area_root = RB_ROOT;
295
296/* The vmap cache globals are protected by vmap_area_lock */
297static struct rb_node *free_vmap_cache;
298static unsigned long cached_hole_size;
299static unsigned long cached_vstart;
300static unsigned long cached_align;
301
Tejun Heoca23e402009-08-14 15:00:52 +0900302static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700303
Susheel Khiani8db21e12013-08-22 13:46:07 -0700304#ifdef CONFIG_ENABLE_VMALLOC_SAVING
305#define POSSIBLE_VMALLOC_START PAGE_OFFSET
306
307#define VMALLOC_BITMAP_SIZE ((VMALLOC_END - PAGE_OFFSET) >> \
308 PAGE_SHIFT)
309#define VMALLOC_TO_BIT(addr) ((addr - PAGE_OFFSET) >> PAGE_SHIFT)
310#define BIT_TO_VMALLOC(i) (PAGE_OFFSET + i * PAGE_SIZE)
311
312unsigned long total_vmalloc_size;
313unsigned long vmalloc_reserved;
314
315DECLARE_BITMAP(possible_areas, VMALLOC_BITMAP_SIZE);
316
317void mark_vmalloc_reserved_area(void *x, unsigned long size)
318{
319 unsigned long addr = (unsigned long)x;
320
321 bitmap_set(possible_areas, VMALLOC_TO_BIT(addr), size >> PAGE_SHIFT);
322 vmalloc_reserved += size;
323}
324
325int is_vmalloc_addr(const void *x)
326{
327 unsigned long addr = (unsigned long)x;
328
329 if (addr < POSSIBLE_VMALLOC_START || addr >= VMALLOC_END)
330 return 0;
331
332 if (test_bit(VMALLOC_TO_BIT(addr), possible_areas))
333 return 0;
334
335 return 1;
336}
337
338static void calc_total_vmalloc_size(void)
339{
340 total_vmalloc_size = VMALLOC_END - POSSIBLE_VMALLOC_START -
341 vmalloc_reserved;
342}
343#else
344int is_vmalloc_addr(const void *x)
345{
346 unsigned long addr = (unsigned long)x;
347
348 return addr >= VMALLOC_START && addr < VMALLOC_END;
349}
350
351static void calc_total_vmalloc_size(void) { }
352#endif
353EXPORT_SYMBOL(is_vmalloc_addr);
354
Nick Piggindb64fe02008-10-18 20:27:03 -0700355static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
Nick Piggindb64fe02008-10-18 20:27:03 -0700357 struct rb_node *n = vmap_area_root.rb_node;
358
359 while (n) {
360 struct vmap_area *va;
361
362 va = rb_entry(n, struct vmap_area, rb_node);
363 if (addr < va->va_start)
364 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700365 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700366 n = n->rb_right;
367 else
368 return va;
369 }
370
371 return NULL;
372}
373
374static void __insert_vmap_area(struct vmap_area *va)
375{
376 struct rb_node **p = &vmap_area_root.rb_node;
377 struct rb_node *parent = NULL;
378 struct rb_node *tmp;
379
380 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700381 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700382
383 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700384 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
385 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700386 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700387 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700388 p = &(*p)->rb_right;
389 else
390 BUG();
391 }
392
393 rb_link_node(&va->rb_node, parent, p);
394 rb_insert_color(&va->rb_node, &vmap_area_root);
395
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700396 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700397 tmp = rb_prev(&va->rb_node);
398 if (tmp) {
399 struct vmap_area *prev;
400 prev = rb_entry(tmp, struct vmap_area, rb_node);
401 list_add_rcu(&va->list, &prev->list);
402 } else
403 list_add_rcu(&va->list, &vmap_area_list);
404}
405
406static void purge_vmap_area_lazy(void);
407
Chris Wilson4da56b92016-04-04 14:46:42 +0100408static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
409
Nick Piggindb64fe02008-10-18 20:27:03 -0700410/*
411 * Allocate a region of KVA of the specified size and alignment, within the
412 * vstart and vend.
413 */
414static struct vmap_area *alloc_vmap_area(unsigned long size,
415 unsigned long align,
416 unsigned long vstart, unsigned long vend,
417 int node, gfp_t gfp_mask)
418{
419 struct vmap_area *va;
420 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700422 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700423 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700424
Nick Piggin77669702009-02-27 14:03:03 -0800425 BUG_ON(!size);
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800426 BUG_ON(offset_in_page(size));
Nick Piggin89699602011-03-22 16:30:36 -0700427 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700428
Christoph Hellwig4a4394c2016-12-12 16:44:20 -0800429 might_sleep();
Chris Wilson4da56b92016-04-04 14:46:42 +0100430
Nick Piggindb64fe02008-10-18 20:27:03 -0700431 va = kmalloc_node(sizeof(struct vmap_area),
432 gfp_mask & GFP_RECLAIM_MASK, node);
433 if (unlikely(!va))
434 return ERR_PTR(-ENOMEM);
435
Catalin Marinas7f88f882013-11-12 15:07:45 -0800436 /*
437 * Only scan the relevant parts containing pointers to other objects
438 * to avoid false negatives.
439 */
440 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
441
Nick Piggindb64fe02008-10-18 20:27:03 -0700442retry:
443 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700444 /*
445 * Invalidate cache if we have more permissive parameters.
446 * cached_hole_size notes the largest hole noticed _below_
447 * the vmap_area cached in free_vmap_cache: if size fits
448 * into that hole, we want to scan from vstart to reuse
449 * the hole instead of allocating above free_vmap_cache.
450 * Note that __free_vmap_area may update free_vmap_cache
451 * without updating cached_hole_size or cached_align.
452 */
453 if (!free_vmap_cache ||
454 size < cached_hole_size ||
455 vstart < cached_vstart ||
456 align < cached_align) {
457nocache:
458 cached_hole_size = 0;
459 free_vmap_cache = NULL;
460 }
461 /* record if we encounter less permissive parameters */
462 cached_vstart = vstart;
463 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800464
Nick Piggin89699602011-03-22 16:30:36 -0700465 /* find starting point for our search */
466 if (free_vmap_cache) {
467 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700468 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700469 if (addr < vstart)
470 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700471 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700472 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700473
Nick Piggin89699602011-03-22 16:30:36 -0700474 } else {
475 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700476 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700477 goto overflow;
478
479 n = vmap_area_root.rb_node;
480 first = NULL;
481
482 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700483 struct vmap_area *tmp;
484 tmp = rb_entry(n, struct vmap_area, rb_node);
485 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700486 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700487 if (tmp->va_start <= addr)
488 break;
489 n = n->rb_left;
490 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700491 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700492 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700493
494 if (!first)
495 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700496 }
Nick Piggin89699602011-03-22 16:30:36 -0700497
498 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700499 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700500 if (addr + cached_hole_size < first->va_start)
501 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700502 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700503 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700504 goto overflow;
505
Hong zhi guo92ca9222012-07-31 16:41:35 -0700506 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700507 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700508
Geliang Tang6219c2a2016-01-14 15:19:08 -0800509 first = list_next_entry(first, list);
Nick Piggin89699602011-03-22 16:30:36 -0700510 }
511
Nick Piggindb64fe02008-10-18 20:27:03 -0700512found:
Nick Piggin89699602011-03-22 16:30:36 -0700513 if (addr + size > vend)
514 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700515
516 va->va_start = addr;
517 va->va_end = addr + size;
518 va->flags = 0;
519 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700520 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700521 spin_unlock(&vmap_area_lock);
522
Wang Xiaoqiang61e16552016-01-15 16:57:19 -0800523 BUG_ON(!IS_ALIGNED(va->va_start, align));
Nick Piggin89699602011-03-22 16:30:36 -0700524 BUG_ON(va->va_start < vstart);
525 BUG_ON(va->va_end > vend);
526
Nick Piggindb64fe02008-10-18 20:27:03 -0700527 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700528
529overflow:
530 spin_unlock(&vmap_area_lock);
531 if (!purged) {
532 purge_vmap_area_lazy();
533 purged = 1;
534 goto retry;
535 }
Chris Wilson4da56b92016-04-04 14:46:42 +0100536
537 if (gfpflags_allow_blocking(gfp_mask)) {
538 unsigned long freed = 0;
539 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
540 if (freed > 0) {
541 purged = 0;
542 goto retry;
543 }
544 }
545
Florian Fainellicd3caa62017-04-27 11:19:00 -0700546 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
Joe Perches756a025f02016-03-17 14:19:47 -0700547 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
548 size);
Nick Piggin89699602011-03-22 16:30:36 -0700549 kfree(va);
550 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700551}
552
Chris Wilson4da56b92016-04-04 14:46:42 +0100553int register_vmap_purge_notifier(struct notifier_block *nb)
554{
555 return blocking_notifier_chain_register(&vmap_notify_list, nb);
556}
557EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
558
559int unregister_vmap_purge_notifier(struct notifier_block *nb)
560{
561 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
562}
563EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
564
Nick Piggindb64fe02008-10-18 20:27:03 -0700565static void __free_vmap_area(struct vmap_area *va)
566{
567 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700568
569 if (free_vmap_cache) {
570 if (va->va_end < cached_vstart) {
571 free_vmap_cache = NULL;
572 } else {
573 struct vmap_area *cache;
574 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
575 if (va->va_start <= cache->va_start) {
576 free_vmap_cache = rb_prev(&va->rb_node);
577 /*
578 * We don't try to update cached_hole_size or
579 * cached_align, but it won't go very wrong.
580 */
581 }
582 }
583 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700584 rb_erase(&va->rb_node, &vmap_area_root);
585 RB_CLEAR_NODE(&va->rb_node);
586 list_del_rcu(&va->list);
587
Tejun Heoca23e402009-08-14 15:00:52 +0900588 /*
589 * Track the highest possible candidate for pcpu area
590 * allocation. Areas outside of vmalloc area can be returned
591 * here too, consider only end addresses which fall inside
592 * vmalloc area proper.
593 */
594 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
595 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
596
Lai Jiangshan14769de2011-03-18 12:12:19 +0800597 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700598}
599
600/*
601 * Free a region of KVA allocated by alloc_vmap_area
602 */
603static void free_vmap_area(struct vmap_area *va)
604{
605 spin_lock(&vmap_area_lock);
606 __free_vmap_area(va);
607 spin_unlock(&vmap_area_lock);
608}
609
610/*
611 * Clear the pagetable entries of a given vmap_area
612 */
613static void unmap_vmap_area(struct vmap_area *va)
614{
615 vunmap_page_range(va->va_start, va->va_end);
616}
617
Nick Piggincd528582009-01-06 14:39:20 -0800618static void vmap_debug_free_range(unsigned long start, unsigned long end)
619{
620 /*
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700621 * Unmap page tables and force a TLB flush immediately if pagealloc
622 * debugging is enabled. This catches use after free bugs similarly to
623 * those in linear kernel virtual address space after a page has been
624 * freed.
Nick Piggincd528582009-01-06 14:39:20 -0800625 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700626 * All the lazy freeing logic is still retained, in order to minimise
627 * intrusiveness of this debugging feature.
Nick Piggincd528582009-01-06 14:39:20 -0800628 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700629 * This is going to be *slow* (linear kernel virtual address debugging
630 * doesn't do a broadcast TLB flush so it is a lot faster).
Nick Piggincd528582009-01-06 14:39:20 -0800631 */
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700632 if (debug_pagealloc_enabled()) {
633 vunmap_page_range(start, end);
634 flush_tlb_kernel_range(start, end);
635 }
Nick Piggincd528582009-01-06 14:39:20 -0800636}
637
Nick Piggindb64fe02008-10-18 20:27:03 -0700638/*
639 * lazy_max_pages is the maximum amount of virtual address space we gather up
640 * before attempting to purge with a TLB flush.
641 *
642 * There is a tradeoff here: a larger number will cover more kernel page tables
643 * and take slightly longer to purge, but it will linearly reduce the number of
644 * global TLB flushes that must be performed. It would seem natural to scale
645 * this number up linearly with the number of CPUs (because vmapping activity
646 * could also scale linearly with the number of CPUs), however it is likely
647 * that in practice, workloads might be constrained in other ways that mean
648 * vmap activity will not scale linearly with CPUs. Also, I want to be
649 * conservative and not introduce a big latency on huge systems, so go with
650 * a less aggressive log scale. It will still be an improvement over the old
651 * code, and it will be simple to change the scale factor if we find that it
652 * becomes a problem on bigger systems.
653 */
654static unsigned long lazy_max_pages(void)
655{
656 unsigned int log;
657
658 log = fls(num_online_cpus());
659
Zhenhua Huang78701f12018-06-21 12:54:40 +0800660 return log * (1UL * CONFIG_VMAP_LAZY_PURGING_FACTOR *
661 1024 * 1024 / PAGE_SIZE);
Nick Piggindb64fe02008-10-18 20:27:03 -0700662}
663
664static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
665
Christoph Hellwig835db092016-12-12 16:44:07 -0800666/*
667 * Serialize vmap purging. There is no actual criticial section protected
668 * by this look, but we want to avoid concurrent calls for performance
669 * reasons and to make the pcpu_get_vm_areas more deterministic.
670 */
Christoph Hellwig9fd2c022016-12-12 16:44:23 -0800671static DEFINE_MUTEX(vmap_purge_lock);
Christoph Hellwig835db092016-12-12 16:44:07 -0800672
Nick Piggin02b709d2010-02-01 22:25:57 +1100673/* for per-CPU blocks */
674static void purge_fragmented_blocks_allcpus(void);
675
Nick Piggindb64fe02008-10-18 20:27:03 -0700676/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500677 * called before a call to iounmap() if the caller wants vm_area_struct's
678 * immediately freed.
679 */
680void set_iounmap_nonlazy(void)
681{
682 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
683}
684
685/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700686 * Purges all lazily-freed vmap areas.
Nick Piggindb64fe02008-10-18 20:27:03 -0700687 */
Christoph Hellwig835db092016-12-12 16:44:07 -0800688static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700689{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700690 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -0700691 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800692 struct vmap_area *n_va;
Joel Fernandes68b90432016-12-12 16:44:26 -0800693 bool do_free = false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700694
Christoph Hellwig835db092016-12-12 16:44:07 -0800695 lockdep_assert_held(&vmap_purge_lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100696
Chris Wilson80c4bd72016-05-20 16:57:38 -0700697 valist = llist_del_all(&vmap_purge_list);
698 llist_for_each_entry(va, valist, purge_list) {
Christoph Hellwig835db092016-12-12 16:44:07 -0800699 if (va->va_start < start)
700 start = va->va_start;
701 if (va->va_end > end)
702 end = va->va_end;
Joel Fernandes68b90432016-12-12 16:44:26 -0800703 do_free = true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700704 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700705
Joel Fernandes68b90432016-12-12 16:44:26 -0800706 if (!do_free)
Christoph Hellwig835db092016-12-12 16:44:07 -0800707 return false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700708
Christoph Hellwig835db092016-12-12 16:44:07 -0800709 flush_tlb_kernel_range(start, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700710
Christoph Hellwig835db092016-12-12 16:44:07 -0800711 spin_lock(&vmap_area_lock);
Joel Fernandes68b90432016-12-12 16:44:26 -0800712 llist_for_each_entry_safe(va, n_va, valist, purge_list) {
713 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
714
Christoph Hellwig835db092016-12-12 16:44:07 -0800715 __free_vmap_area(va);
Joel Fernandes68b90432016-12-12 16:44:26 -0800716 atomic_sub(nr, &vmap_lazy_nr);
717 cond_resched_lock(&vmap_area_lock);
718 }
Christoph Hellwig835db092016-12-12 16:44:07 -0800719 spin_unlock(&vmap_area_lock);
720 return true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700721}
722
723/*
Nick Piggin496850e2008-11-19 15:36:33 -0800724 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
725 * is already purging.
726 */
727static void try_purge_vmap_area_lazy(void)
728{
Christoph Hellwig9fd2c022016-12-12 16:44:23 -0800729 if (mutex_trylock(&vmap_purge_lock)) {
Christoph Hellwig835db092016-12-12 16:44:07 -0800730 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwig9fd2c022016-12-12 16:44:23 -0800731 mutex_unlock(&vmap_purge_lock);
Christoph Hellwig835db092016-12-12 16:44:07 -0800732 }
Nick Piggin496850e2008-11-19 15:36:33 -0800733}
734
735/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700736 * Kick off a purge of the outstanding lazy areas.
737 */
738static void purge_vmap_area_lazy(void)
739{
Christoph Hellwig9fd2c022016-12-12 16:44:23 -0800740 mutex_lock(&vmap_purge_lock);
Christoph Hellwig835db092016-12-12 16:44:07 -0800741 purge_fragmented_blocks_allcpus();
742 __purge_vmap_area_lazy(ULONG_MAX, 0);
Christoph Hellwig9fd2c022016-12-12 16:44:23 -0800743 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700744}
745
746/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800747 * Free a vmap area, caller ensuring that the area has been unmapped
748 * and flush_cache_vunmap had been called for the correct range
749 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700750 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800751static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700752{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700753 int nr_lazy;
754
755 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
756 &vmap_lazy_nr);
757
758 /* After this point, we may free va at any time */
759 llist_add(&va->purge_list, &vmap_purge_list);
760
761 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800762 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700763}
764
Nick Pigginb29acbd2008-12-01 13:13:47 -0800765/*
766 * Free and unmap a vmap area
767 */
768static void free_unmap_vmap_area(struct vmap_area *va)
769{
770 flush_cache_vunmap(va->va_start, va->va_end);
Christoph Hellwig71aadd02016-12-12 16:44:01 -0800771 unmap_vmap_area(va);
772 free_vmap_area_noflush(va);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800773}
774
Nick Piggindb64fe02008-10-18 20:27:03 -0700775static struct vmap_area *find_vmap_area(unsigned long addr)
776{
777 struct vmap_area *va;
778
779 spin_lock(&vmap_area_lock);
780 va = __find_vmap_area(addr);
781 spin_unlock(&vmap_area_lock);
782
783 return va;
784}
785
Nick Piggindb64fe02008-10-18 20:27:03 -0700786/*** Per cpu kva allocator ***/
787
788/*
789 * vmap space is limited especially on 32 bit architectures. Ensure there is
790 * room for at least 16 percpu vmap blocks per CPU.
791 */
792/*
793 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
794 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
795 * instead (we just need a rough idea)
796 */
797#if BITS_PER_LONG == 32
798#define VMALLOC_SPACE (128UL*1024*1024)
799#else
800#define VMALLOC_SPACE (128UL*1024*1024*1024)
801#endif
802
803#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
804#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
805#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
806#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
807#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
808#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200809#define VMAP_BBMAP_BITS \
810 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
811 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
812 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700813
814#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
815
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100816static bool vmap_initialized __read_mostly = false;
817
Nick Piggindb64fe02008-10-18 20:27:03 -0700818struct vmap_block_queue {
819 spinlock_t lock;
820 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700821};
822
823struct vmap_block {
824 spinlock_t lock;
825 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700826 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700827 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100828 struct list_head free_list;
829 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100830 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700831};
832
833/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
834static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
835
836/*
837 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
838 * in the free path. Could get rid of this if we change the API to return a
839 * "cookie" from alloc, to be passed to free. But no big deal yet.
840 */
841static DEFINE_SPINLOCK(vmap_block_tree_lock);
842static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
843
844/*
845 * We should probably have a fallback mechanism to allocate virtual memory
846 * out of partially filled vmap blocks. However vmap block sizing should be
847 * fairly reasonable according to the vmalloc size, so it shouldn't be a
848 * big problem.
849 */
850
851static unsigned long addr_to_vb_idx(unsigned long addr)
852{
853 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
854 addr /= VMAP_BLOCK_SIZE;
855 return addr;
856}
857
Roman Pencf725ce2015-04-15 16:13:52 -0700858static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
859{
860 unsigned long addr;
861
862 addr = va_start + (pages_off << PAGE_SHIFT);
863 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
864 return (void *)addr;
865}
866
867/**
868 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
869 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
870 * @order: how many 2^order pages should be occupied in newly allocated block
871 * @gfp_mask: flags for the page level allocator
872 *
873 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
874 */
875static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700876{
877 struct vmap_block_queue *vbq;
878 struct vmap_block *vb;
879 struct vmap_area *va;
880 unsigned long vb_idx;
881 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700882 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700883
884 node = numa_node_id();
885
886 vb = kmalloc_node(sizeof(struct vmap_block),
887 gfp_mask & GFP_RECLAIM_MASK, node);
888 if (unlikely(!vb))
889 return ERR_PTR(-ENOMEM);
890
891 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
892 VMALLOC_START, VMALLOC_END,
893 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800894 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700895 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700896 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700897 }
898
899 err = radix_tree_preload(gfp_mask);
900 if (unlikely(err)) {
901 kfree(vb);
902 free_vmap_area(va);
903 return ERR_PTR(err);
904 }
905
Roman Pencf725ce2015-04-15 16:13:52 -0700906 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700907 spin_lock_init(&vb->lock);
908 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700909 /* At least something should be left free */
910 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
911 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700912 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700913 vb->dirty_min = VMAP_BBMAP_BITS;
914 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700915 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700916
917 vb_idx = addr_to_vb_idx(va->va_start);
918 spin_lock(&vmap_block_tree_lock);
919 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
920 spin_unlock(&vmap_block_tree_lock);
921 BUG_ON(err);
922 radix_tree_preload_end();
923
924 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700925 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700926 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700927 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900928 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700929
Roman Pencf725ce2015-04-15 16:13:52 -0700930 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700931}
932
Nick Piggindb64fe02008-10-18 20:27:03 -0700933static void free_vmap_block(struct vmap_block *vb)
934{
935 struct vmap_block *tmp;
936 unsigned long vb_idx;
937
Nick Piggindb64fe02008-10-18 20:27:03 -0700938 vb_idx = addr_to_vb_idx(vb->va->va_start);
939 spin_lock(&vmap_block_tree_lock);
940 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
941 spin_unlock(&vmap_block_tree_lock);
942 BUG_ON(tmp != vb);
943
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800944 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800945 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700946}
947
Nick Piggin02b709d2010-02-01 22:25:57 +1100948static void purge_fragmented_blocks(int cpu)
949{
950 LIST_HEAD(purge);
951 struct vmap_block *vb;
952 struct vmap_block *n_vb;
953 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
954
955 rcu_read_lock();
956 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
957
958 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
959 continue;
960
961 spin_lock(&vb->lock);
962 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
963 vb->free = 0; /* prevent further allocs after releasing lock */
964 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700965 vb->dirty_min = 0;
966 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100967 spin_lock(&vbq->lock);
968 list_del_rcu(&vb->free_list);
969 spin_unlock(&vbq->lock);
970 spin_unlock(&vb->lock);
971 list_add_tail(&vb->purge, &purge);
972 } else
973 spin_unlock(&vb->lock);
974 }
975 rcu_read_unlock();
976
977 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
978 list_del(&vb->purge);
979 free_vmap_block(vb);
980 }
981}
982
Nick Piggin02b709d2010-02-01 22:25:57 +1100983static void purge_fragmented_blocks_allcpus(void)
984{
985 int cpu;
986
987 for_each_possible_cpu(cpu)
988 purge_fragmented_blocks(cpu);
989}
990
Nick Piggindb64fe02008-10-18 20:27:03 -0700991static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
992{
993 struct vmap_block_queue *vbq;
994 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700995 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700996 unsigned int order;
997
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800998 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700999 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -07001000 if (WARN_ON(size == 0)) {
1001 /*
1002 * Allocating 0 bytes isn't what caller wants since
1003 * get_order(0) returns funny result. Just warn and terminate
1004 * early.
1005 */
1006 return NULL;
1007 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001008 order = get_order(size);
1009
Nick Piggindb64fe02008-10-18 20:27:03 -07001010 rcu_read_lock();
1011 vbq = &get_cpu_var(vmap_block_queue);
1012 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -07001013 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -07001014
1015 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -07001016 if (vb->free < (1UL << order)) {
1017 spin_unlock(&vb->lock);
1018 continue;
1019 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001020
Roman Pencf725ce2015-04-15 16:13:52 -07001021 pages_off = VMAP_BBMAP_BITS - vb->free;
1022 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +11001023 vb->free -= 1UL << order;
1024 if (vb->free == 0) {
1025 spin_lock(&vbq->lock);
1026 list_del_rcu(&vb->free_list);
1027 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001028 }
Roman Pencf725ce2015-04-15 16:13:52 -07001029
Nick Piggindb64fe02008-10-18 20:27:03 -07001030 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001031 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001032 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001033
Tejun Heo3f04ba82009-10-29 22:34:12 +09001034 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001035 rcu_read_unlock();
1036
Roman Pencf725ce2015-04-15 16:13:52 -07001037 /* Allocate new block if nothing was found */
1038 if (!vaddr)
1039 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001040
Roman Pencf725ce2015-04-15 16:13:52 -07001041 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001042}
1043
1044static void vb_free(const void *addr, unsigned long size)
1045{
1046 unsigned long offset;
1047 unsigned long vb_idx;
1048 unsigned int order;
1049 struct vmap_block *vb;
1050
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001051 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001052 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001053
1054 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1055
Nick Piggindb64fe02008-10-18 20:27:03 -07001056 order = get_order(size);
1057
1058 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001059 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001060
1061 vb_idx = addr_to_vb_idx((unsigned long)addr);
1062 rcu_read_lock();
1063 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1064 rcu_read_unlock();
1065 BUG_ON(!vb);
1066
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001067 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1068
Nick Piggindb64fe02008-10-18 20:27:03 -07001069 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001070
1071 /* Expand dirty range */
1072 vb->dirty_min = min(vb->dirty_min, offset);
1073 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001074
Nick Piggindb64fe02008-10-18 20:27:03 -07001075 vb->dirty += 1UL << order;
1076 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001077 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001078 spin_unlock(&vb->lock);
1079 free_vmap_block(vb);
1080 } else
1081 spin_unlock(&vb->lock);
1082}
1083
1084/**
1085 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1086 *
1087 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1088 * to amortize TLB flushing overheads. What this means is that any page you
1089 * have now, may, in a former life, have been mapped into kernel virtual
1090 * address by the vmap layer and so there might be some CPUs with TLB entries
1091 * still referencing that page (additional to the regular 1:1 kernel mapping).
1092 *
1093 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1094 * be sure that none of the pages we have control over will have any aliases
1095 * from the vmap layer.
1096 */
1097void vm_unmap_aliases(void)
1098{
1099 unsigned long start = ULONG_MAX, end = 0;
1100 int cpu;
1101 int flush = 0;
1102
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001103 if (unlikely(!vmap_initialized))
1104 return;
1105
Christoph Hellwig4a4394c2016-12-12 16:44:20 -08001106 might_sleep();
1107
Nick Piggindb64fe02008-10-18 20:27:03 -07001108 for_each_possible_cpu(cpu) {
1109 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1110 struct vmap_block *vb;
1111
1112 rcu_read_lock();
1113 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001114 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001115 if (vb->dirty) {
1116 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001117 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001118
Roman Pen7d61bfe2015-04-15 16:13:55 -07001119 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1120 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001121
Roman Pen7d61bfe2015-04-15 16:13:55 -07001122 start = min(s, start);
1123 end = max(e, end);
1124
Nick Piggindb64fe02008-10-18 20:27:03 -07001125 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001126 }
1127 spin_unlock(&vb->lock);
1128 }
1129 rcu_read_unlock();
1130 }
1131
Christoph Hellwig9fd2c022016-12-12 16:44:23 -08001132 mutex_lock(&vmap_purge_lock);
Christoph Hellwig835db092016-12-12 16:44:07 -08001133 purge_fragmented_blocks_allcpus();
1134 if (!__purge_vmap_area_lazy(start, end) && flush)
1135 flush_tlb_kernel_range(start, end);
Christoph Hellwig9fd2c022016-12-12 16:44:23 -08001136 mutex_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001137}
1138EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1139
1140/**
1141 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1142 * @mem: the pointer returned by vm_map_ram
1143 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1144 */
1145void vm_unmap_ram(const void *mem, unsigned int count)
1146{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001147 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001148 unsigned long addr = (unsigned long)mem;
Christoph Hellwig7df98752016-12-12 16:44:04 -08001149 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001150
Christoph Hellwig4a4394c2016-12-12 16:44:20 -08001151 might_sleep();
Nick Piggindb64fe02008-10-18 20:27:03 -07001152 BUG_ON(!addr);
1153 BUG_ON(addr < VMALLOC_START);
1154 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001155 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001156
1157 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001158 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001159
Christoph Hellwig7df98752016-12-12 16:44:04 -08001160 if (likely(count <= VMAP_MAX_ALLOC)) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001161 vb_free(mem, size);
Christoph Hellwig7df98752016-12-12 16:44:04 -08001162 return;
1163 }
1164
1165 va = find_vmap_area(addr);
1166 BUG_ON(!va);
1167 free_unmap_vmap_area(va);
Nick Piggindb64fe02008-10-18 20:27:03 -07001168}
1169EXPORT_SYMBOL(vm_unmap_ram);
1170
1171/**
1172 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1173 * @pages: an array of pointers to the pages to be mapped
1174 * @count: number of pages
1175 * @node: prefer to allocate data structures on this node
1176 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001177 *
Gioh Kim36437632014-04-07 15:37:37 -07001178 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1179 * faster than vmap so it's good. But if you mix long-life and short-life
1180 * objects with vm_map_ram(), it could consume lots of address space through
1181 * fragmentation (especially on a 32bit machine). You could see failures in
1182 * the end. Please use this function for short-lived objects.
1183 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001184 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001185 */
1186void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1187{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001188 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001189 unsigned long addr;
1190 void *mem;
1191
1192 if (likely(count <= VMAP_MAX_ALLOC)) {
1193 mem = vb_alloc(size, GFP_KERNEL);
1194 if (IS_ERR(mem))
1195 return NULL;
1196 addr = (unsigned long)mem;
1197 } else {
1198 struct vmap_area *va;
1199 va = alloc_vmap_area(size, PAGE_SIZE,
1200 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1201 if (IS_ERR(va))
1202 return NULL;
1203
1204 addr = va->va_start;
1205 mem = (void *)addr;
1206 }
1207 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1208 vm_unmap_ram(mem, count);
1209 return NULL;
1210 }
1211 return mem;
1212}
1213EXPORT_SYMBOL(vm_map_ram);
1214
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001215static struct vm_struct *vmlist __initdata;
Susheel Khiani432f1662015-09-08 15:05:43 +05301216
1217/**
1218 * vm_area_check_early - check if vmap area is already mapped
1219 * @vm: vm_struct to be checked
1220 *
1221 * This function is used to check if the vmap area has been
1222 * mapped already. @vm->addr, @vm->size and @vm->flags should
1223 * contain proper values.
1224 *
1225 */
1226int __init vm_area_check_early(struct vm_struct *vm)
1227{
1228 struct vm_struct *tmp, **p;
1229
1230 BUG_ON(vmap_initialized);
1231 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1232 if (tmp->addr >= vm->addr) {
1233 if (tmp->addr < vm->addr + vm->size)
1234 return 1;
1235 } else {
1236 if (tmp->addr + tmp->size > vm->addr)
1237 return 1;
1238 }
1239 }
1240 return 0;
1241}
1242
Tejun Heof0aa6612009-02-20 16:29:08 +09001243/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001244 * vm_area_add_early - add vmap area early during boot
1245 * @vm: vm_struct to add
1246 *
1247 * This function is used to add fixed kernel vm area to vmlist before
1248 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1249 * should contain proper values and the other fields should be zero.
1250 *
1251 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1252 */
1253void __init vm_area_add_early(struct vm_struct *vm)
1254{
1255 struct vm_struct *tmp, **p;
1256
1257 BUG_ON(vmap_initialized);
1258 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1259 if (tmp->addr >= vm->addr) {
1260 BUG_ON(tmp->addr < vm->addr + vm->size);
1261 break;
1262 } else
1263 BUG_ON(tmp->addr + tmp->size > vm->addr);
1264 }
1265 vm->next = *p;
1266 *p = vm;
1267}
1268
1269/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001270 * vm_area_register_early - register vmap area early during boot
1271 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001272 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001273 *
1274 * This function is used to register kernel vm area before
1275 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1276 * proper values on entry and other fields should be zero. On return,
1277 * vm->addr contains the allocated address.
1278 *
1279 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1280 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001281void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001282{
1283 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001284 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001285
Tejun Heoc0c0a292009-02-24 11:57:21 +09001286 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1287 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1288
1289 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001290
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001291 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001292}
1293
Nick Piggindb64fe02008-10-18 20:27:03 -07001294void __init vmalloc_init(void)
1295{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001296 struct vmap_area *va;
1297 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001298 int i;
1299
1300 for_each_possible_cpu(i) {
1301 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001302 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001303
1304 vbq = &per_cpu(vmap_block_queue, i);
1305 spin_lock_init(&vbq->lock);
1306 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001307 p = &per_cpu(vfree_deferred, i);
1308 init_llist_head(&p->list);
1309 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001310 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001311
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001312 /* Import existing vmlist entries. */
1313 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001314 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001315 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001316 va->va_start = (unsigned long)tmp->addr;
1317 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001318 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001319 __insert_vmap_area(va);
1320 }
Tejun Heoca23e402009-08-14 15:00:52 +09001321
1322 vmap_area_pcpu_hole = VMALLOC_END;
Susheel Khiani8db21e12013-08-22 13:46:07 -07001323 calc_total_vmalloc_size();
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001324 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001325}
1326
Tejun Heo8fc48982009-02-20 16:29:08 +09001327/**
1328 * map_kernel_range_noflush - map kernel VM area with the specified pages
1329 * @addr: start of the VM area to map
1330 * @size: size of the VM area to map
1331 * @prot: page protection flags to use
1332 * @pages: pages to map
1333 *
1334 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1335 * specify should have been allocated using get_vm_area() and its
1336 * friends.
1337 *
1338 * NOTE:
1339 * This function does NOT do any cache flushing. The caller is
1340 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1341 * before calling this function.
1342 *
1343 * RETURNS:
1344 * The number of pages mapped on success, -errno on failure.
1345 */
1346int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1347 pgprot_t prot, struct page **pages)
1348{
1349 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1350}
1351
1352/**
1353 * unmap_kernel_range_noflush - unmap kernel VM area
1354 * @addr: start of the VM area to unmap
1355 * @size: size of the VM area to unmap
1356 *
1357 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1358 * specify should have been allocated using get_vm_area() and its
1359 * friends.
1360 *
1361 * NOTE:
1362 * This function does NOT do any cache flushing. The caller is
1363 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1364 * before calling this function and flush_tlb_kernel_range() after.
1365 */
1366void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1367{
1368 vunmap_page_range(addr, addr + size);
1369}
Huang Ying81e88fd2011-01-12 14:44:55 +08001370EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001371
1372/**
1373 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1374 * @addr: start of the VM area to unmap
1375 * @size: size of the VM area to unmap
1376 *
1377 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1378 * the unmapping and tlb after.
1379 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001380void unmap_kernel_range(unsigned long addr, unsigned long size)
1381{
1382 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001383
1384 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001385 vunmap_page_range(addr, end);
1386 flush_tlb_kernel_range(addr, end);
1387}
Minchan Kim93ef6d62014-06-04 16:11:09 -07001388EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001389
WANG Chaof6f8ed42014-08-06 16:06:58 -07001390int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001391{
1392 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001393 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001394 int err;
1395
WANG Chaof6f8ed42014-08-06 16:06:58 -07001396 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001397
WANG Chaof6f8ed42014-08-06 16:06:58 -07001398 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001399}
1400EXPORT_SYMBOL_GPL(map_vm_area);
1401
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001402static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001403 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001404{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001405 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001406 vm->flags = flags;
1407 vm->addr = (void *)va->va_start;
1408 vm->size = va->va_end - va->va_start;
1409 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001410 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001411 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001412 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001413}
Tejun Heocf88c792009-08-14 15:00:52 +09001414
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001415static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001416{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001417 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001418 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001419 * we should make sure that vm has proper values.
1420 * Pair with smp_rmb() in show_numa_info().
1421 */
1422 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001423 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001424}
1425
Nick Piggindb64fe02008-10-18 20:27:03 -07001426static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001427 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001428 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001429{
Kautuk Consul00065262011-12-19 17:12:04 -08001430 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001431 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001433 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001435 if (unlikely(!size))
1436 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
zijun_hu252e5c62016-10-07 16:57:26 -07001438 if (flags & VM_IOREMAP)
1439 align = 1ul << clamp_t(int, get_count_order_long(size),
1440 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1441
Tejun Heocf88c792009-08-14 15:00:52 +09001442 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 if (unlikely(!area))
1444 return NULL;
1445
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001446 if (!(flags & VM_NO_GUARD))
1447 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Nick Piggindb64fe02008-10-18 20:27:03 -07001449 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1450 if (IS_ERR(va)) {
1451 kfree(area);
1452 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001455 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458}
1459
Christoph Lameter930fc452005-10-29 18:15:41 -07001460struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1461 unsigned long start, unsigned long end)
1462{
David Rientjes00ef2d22013-02-22 16:35:36 -08001463 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1464 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001465}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001466EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001467
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001468struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1469 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001470 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001471{
David Rientjes00ef2d22013-02-22 16:35:36 -08001472 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1473 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001474}
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476/**
Simon Arlott183ff222007-10-20 01:27:18 +02001477 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 * @size: size of the area
1479 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1480 *
1481 * Search an area of @size in the kernel virtual mapping area,
1482 * and reserved it for out purposes. Returns the area descriptor
1483 * on success or %NULL on failure.
1484 */
1485struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1486{
Susheel Khianid0bd1992015-09-03 18:21:23 +05301487#ifdef CONFIG_ENABLE_VMALLOC_SAVING
1488 return __get_vm_area_node(size, 1, flags, PAGE_OFFSET, VMALLOC_END,
1489 NUMA_NO_NODE, GFP_KERNEL,
1490 __builtin_return_address(0));
1491#else
David Miller2dca6992009-09-21 12:22:34 -07001492 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001493 NUMA_NO_NODE, GFP_KERNEL,
1494 __builtin_return_address(0));
Susheel Khianid0bd1992015-09-03 18:21:23 +05301495#endif
Christoph Lameter23016962008-04-28 02:12:42 -07001496}
1497
1498struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001499 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001500{
Susheel Khianid0bd1992015-09-03 18:21:23 +05301501#ifdef CONFIG_ENABLE_VMALLOC_SAVING
1502 return __get_vm_area_node(size, 1, flags, PAGE_OFFSET, VMALLOC_END,
1503 NUMA_NO_NODE, GFP_KERNEL, caller);
1504#else
David Miller2dca6992009-09-21 12:22:34 -07001505 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001506 NUMA_NO_NODE, GFP_KERNEL, caller);
Susheel Khianid0bd1992015-09-03 18:21:23 +05301507#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508}
1509
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001510/**
1511 * find_vm_area - find a continuous kernel virtual area
1512 * @addr: base address
1513 *
1514 * Search for the kernel VM area starting at @addr, and return it.
1515 * It is up to the caller to do all required locking to keep the returned
1516 * pointer valid.
1517 */
1518struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001519{
Nick Piggindb64fe02008-10-18 20:27:03 -07001520 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001521
Nick Piggindb64fe02008-10-18 20:27:03 -07001522 va = find_vmap_area((unsigned long)addr);
1523 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001524 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001525
Andi Kleen7856dfe2005-05-20 14:27:57 -07001526 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001527}
1528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529/**
Simon Arlott183ff222007-10-20 01:27:18 +02001530 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 * @addr: base address
1532 *
1533 * Search for the kernel VM area starting at @addr, and remove it.
1534 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001535 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001537struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538{
Nick Piggindb64fe02008-10-18 20:27:03 -07001539 struct vmap_area *va;
1540
Christoph Hellwig4a4394c2016-12-12 16:44:20 -08001541 might_sleep();
1542
Nick Piggindb64fe02008-10-18 20:27:03 -07001543 va = find_vmap_area((unsigned long)addr);
1544 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001545 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001546
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001547 spin_lock(&vmap_area_lock);
1548 va->vm = NULL;
1549 va->flags &= ~VM_VM_AREA;
1550 spin_unlock(&vmap_area_lock);
1551
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001552 vmap_debug_free_range(va->va_start, va->va_end);
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001553 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001554 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001555
Nick Piggindb64fe02008-10-18 20:27:03 -07001556 return vm;
1557 }
1558 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001561static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
1563 struct vm_struct *area;
1564
1565 if (!addr)
1566 return;
1567
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001568 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001569 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572 area = remove_vm_area(addr);
1573 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001574 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 return;
1577 }
1578
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001579 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1580 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 if (deallocate_pages) {
1583 int i;
1584
1585 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001586 struct page *page = area->pages[i];
1587
1588 BUG_ON(!page);
Vladimir Davydov49491482016-07-26 15:24:24 -07001589 __free_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
1591
David Rientjes244d63e2016-01-14 15:19:35 -08001592 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 }
1594
1595 kfree(area);
1596 return;
1597}
Andrey Ryabinin3af922c2016-12-12 16:44:10 -08001598
1599static inline void __vfree_deferred(const void *addr)
1600{
1601 /*
1602 * Use raw_cpu_ptr() because this can be called from preemptible
1603 * context. Preemption is absolutely fine here, because the llist_add()
1604 * implementation is lockless, so it works even if we are adding to
1605 * nother cpu's list. schedule_work() should be fine with this too.
1606 */
1607 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1608
1609 if (llist_add((struct llist_node *)addr, &p->list))
1610 schedule_work(&p->wq);
1611}
1612
1613/**
1614 * vfree_atomic - release memory allocated by vmalloc()
1615 * @addr: memory base address
1616 *
1617 * This one is just like vfree() but can be called in any atomic context
1618 * except NMIs.
1619 */
1620void vfree_atomic(const void *addr)
1621{
1622 BUG_ON(in_nmi());
1623
1624 kmemleak_free(addr);
1625
1626 if (!addr)
1627 return;
1628 __vfree_deferred(addr);
1629}
1630
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631/**
1632 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 * @addr: memory base address
1634 *
Simon Arlott183ff222007-10-20 01:27:18 +02001635 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001636 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1637 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 *
Al Viro32fcfd42013-03-10 20:14:08 -04001639 * Must not be called in NMI context (strictly speaking, only if we don't
1640 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1641 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001642 *
1643 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001645void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646{
Al Viro32fcfd42013-03-10 20:14:08 -04001647 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001648
1649 kmemleak_free(addr);
1650
Al Viro32fcfd42013-03-10 20:14:08 -04001651 if (!addr)
1652 return;
Andrey Ryabinin3af922c2016-12-12 16:44:10 -08001653 if (unlikely(in_interrupt()))
1654 __vfree_deferred(addr);
1655 else
Al Viro32fcfd42013-03-10 20:14:08 -04001656 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658EXPORT_SYMBOL(vfree);
1659
1660/**
1661 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 * @addr: memory base address
1663 *
1664 * Free the virtually contiguous memory area starting at @addr,
1665 * which was created from the page array passed to vmap().
1666 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001667 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001669void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
1671 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001672 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001673 if (addr)
1674 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676EXPORT_SYMBOL(vunmap);
1677
1678/**
1679 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 * @pages: array of page pointers
1681 * @count: number of pages to map
1682 * @flags: vm_area->flags
1683 * @prot: page protection for the mapping
1684 *
1685 * Maps @count pages from @pages into contiguous kernel virtual
1686 * space.
1687 */
1688void *vmap(struct page **pages, unsigned int count,
1689 unsigned long flags, pgprot_t prot)
1690{
1691 struct vm_struct *area;
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001692 unsigned long size; /* In bytes */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693
Peter Zijlstra34754b62009-02-25 16:04:03 +01001694 might_sleep();
1695
Jan Beulich44813742009-09-21 17:03:05 -07001696 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 return NULL;
1698
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001699 size = (unsigned long)count << PAGE_SHIFT;
1700 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 if (!area)
1702 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001703
WANG Chaof6f8ed42014-08-06 16:06:58 -07001704 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 vunmap(area->addr);
1706 return NULL;
1707 }
1708
1709 return area->addr;
1710}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711EXPORT_SYMBOL(vmap);
1712
David Miller2dca6992009-09-21 12:22:34 -07001713static void *__vmalloc_node(unsigned long size, unsigned long align,
1714 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001715 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001716static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001717 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718{
1719 struct page **pages;
1720 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001721 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1722 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Wanpeng Li762216a2013-09-11 14:22:42 -07001724 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 array_size = (nr_pages * sizeof(struct page *));
1726
1727 area->nr_pages = nr_pages;
1728 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001729 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001730 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001731 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001732 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001733 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 area->pages = pages;
1736 if (!area->pages) {
1737 remove_vm_area(area->addr);
1738 kfree(area);
1739 return NULL;
1740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001743 struct page *page;
1744
Jianguo Wu4b909512013-11-12 15:07:11 -08001745 if (node == NUMA_NO_NODE)
Michal Hocko7877cdc2016-10-07 17:01:55 -07001746 page = alloc_page(alloc_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001747 else
Michal Hocko7877cdc2016-10-07 17:01:55 -07001748 page = alloc_pages_node(node, alloc_mask, 0);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001749
1750 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 /* Successfully allocated i pages, free them in __vunmap() */
1752 area->nr_pages = i;
1753 goto fail;
1754 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001755 area->pages[i] = page;
Mel Gormand0164ad2015-11-06 16:28:21 -08001756 if (gfpflags_allow_blocking(gfp_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001757 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 }
1759
WANG Chaof6f8ed42014-08-06 16:06:58 -07001760 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 goto fail;
1762 return area->addr;
1763
1764fail:
Michal Hocko7877cdc2016-10-07 17:01:55 -07001765 warn_alloc(gfp_mask,
1766 "vmalloc: allocation failure, allocated %ld of %ld bytes",
Dave Hansen22943ab2011-05-24 17:12:18 -07001767 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 vfree(area->addr);
1769 return NULL;
1770}
1771
David Rientjesd0a21262011-01-13 15:46:02 -08001772/**
1773 * __vmalloc_node_range - allocate virtually contiguous memory
1774 * @size: allocation size
1775 * @align: desired alignment
1776 * @start: vm area range start
1777 * @end: vm area range end
1778 * @gfp_mask: flags for the page level allocator
1779 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001780 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001781 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001782 * @caller: caller's return address
1783 *
1784 * Allocate enough pages to cover @size from the page level
1785 * allocator with @gfp_mask flags. Map them into contiguous
1786 * kernel virtual space, using a pagetable protection of @prot.
1787 */
1788void *__vmalloc_node_range(unsigned long size, unsigned long align,
1789 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001790 pgprot_t prot, unsigned long vm_flags, int node,
1791 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001792{
David Rientjesd0a21262011-01-13 15:46:02 -08001793 struct vm_struct *area;
1794 void *addr;
1795 unsigned long real_size = size;
1796
1797 size = PAGE_ALIGN(size);
1798 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001799 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001800
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001801 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1802 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001803 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001804 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001805
Wanpeng Li3722e132013-11-12 15:07:29 -08001806 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001807 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001808 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001809
1810 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001811 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1812 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001813 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001814 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001815 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001816
1817 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001818 * A ref_count = 2 is needed because vm_struct allocated in
1819 * __get_vm_area_node() contains a reference to the virtual address of
1820 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001821 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001822 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001823
1824 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001825
1826fail:
Michal Hocko7877cdc2016-10-07 17:01:55 -07001827 warn_alloc(gfp_mask,
1828 "vmalloc: allocation failure: %lu bytes", real_size);
Joe Perchesde7d2b52011-10-31 17:08:48 -07001829 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001830}
1831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001833 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001835 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 * @gfp_mask: flags for the page level allocator
1837 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001838 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001839 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 *
1841 * Allocate enough pages to cover @size from the page level
1842 * allocator with @gfp_mask flags. Map them into contiguous
1843 * kernel virtual space, using a pagetable protection of @prot.
1844 */
David Miller2dca6992009-09-21 12:22:34 -07001845static void *__vmalloc_node(unsigned long size, unsigned long align,
1846 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001847 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848{
David Rientjesd0a21262011-01-13 15:46:02 -08001849 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001850 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851}
1852
Christoph Lameter930fc452005-10-29 18:15:41 -07001853void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1854{
David Rientjes00ef2d22013-02-22 16:35:36 -08001855 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001856 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001857}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858EXPORT_SYMBOL(__vmalloc);
1859
Dave Younge1ca7782010-10-26 14:22:06 -07001860static inline void *__vmalloc_node_flags(unsigned long size,
1861 int node, gfp_t flags)
1862{
1863 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1864 node, __builtin_return_address(0));
1865}
1866
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867/**
1868 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 * Allocate enough pages to cover @size from the page level
1871 * allocator and map them into contiguous kernel virtual space.
1872 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001873 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 * use __vmalloc() instead.
1875 */
1876void *vmalloc(unsigned long size)
1877{
David Rientjes00ef2d22013-02-22 16:35:36 -08001878 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1879 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881EXPORT_SYMBOL(vmalloc);
1882
Christoph Lameter930fc452005-10-29 18:15:41 -07001883/**
Dave Younge1ca7782010-10-26 14:22:06 -07001884 * vzalloc - allocate virtually contiguous memory with zero fill
1885 * @size: allocation size
1886 * Allocate enough pages to cover @size from the page level
1887 * allocator and map them into contiguous kernel virtual space.
1888 * The memory allocated is set to zero.
1889 *
1890 * For tight control over page level allocator and protection flags
1891 * use __vmalloc() instead.
1892 */
1893void *vzalloc(unsigned long size)
1894{
David Rientjes00ef2d22013-02-22 16:35:36 -08001895 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001896 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1897}
1898EXPORT_SYMBOL(vzalloc);
1899
1900/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001901 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1902 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001903 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001904 * The resulting memory area is zeroed so it can be mapped to userspace
1905 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001906 */
1907void *vmalloc_user(unsigned long size)
1908{
1909 struct vm_struct *area;
1910 void *ret;
1911
David Miller2dca6992009-09-21 12:22:34 -07001912 ret = __vmalloc_node(size, SHMLBA,
1913 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001914 PAGE_KERNEL, NUMA_NO_NODE,
1915 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001916 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001917 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001918 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001919 }
Nick Piggin83342312006-06-23 02:03:20 -07001920 return ret;
1921}
1922EXPORT_SYMBOL(vmalloc_user);
1923
1924/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001925 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001926 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001927 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001928 *
1929 * Allocate enough pages to cover @size from the page level
1930 * allocator and map them into contiguous kernel virtual space.
1931 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001932 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001933 * use __vmalloc() instead.
1934 */
1935void *vmalloc_node(unsigned long size, int node)
1936{
David Miller2dca6992009-09-21 12:22:34 -07001937 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001938 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001939}
1940EXPORT_SYMBOL(vmalloc_node);
1941
Dave Younge1ca7782010-10-26 14:22:06 -07001942/**
1943 * vzalloc_node - allocate memory on a specific node with zero fill
1944 * @size: allocation size
1945 * @node: numa node
1946 *
1947 * Allocate enough pages to cover @size from the page level
1948 * allocator and map them into contiguous kernel virtual space.
1949 * The memory allocated is set to zero.
1950 *
1951 * For tight control over page level allocator and protection flags
1952 * use __vmalloc_node() instead.
1953 */
1954void *vzalloc_node(unsigned long size, int node)
1955{
1956 return __vmalloc_node_flags(size, node,
1957 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1958}
1959EXPORT_SYMBOL(vzalloc_node);
1960
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001961#ifndef PAGE_KERNEL_EXEC
1962# define PAGE_KERNEL_EXEC PAGE_KERNEL
1963#endif
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965/**
1966 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 * @size: allocation size
1968 *
1969 * Kernel-internal function to allocate enough pages to cover @size
1970 * the page level allocator and map them into contiguous and
1971 * executable kernel virtual space.
1972 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001973 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 * use __vmalloc() instead.
1975 */
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977void *vmalloc_exec(unsigned long size)
1978{
David Miller2dca6992009-09-21 12:22:34 -07001979 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001980 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981}
1982
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001983#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001984#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001985#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001986#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001987#else
1988#define GFP_VMALLOC32 GFP_KERNEL
1989#endif
1990
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991/**
1992 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 * @size: allocation size
1994 *
1995 * Allocate enough 32bit PA addressable pages to cover @size from the
1996 * page level allocator and map them into contiguous kernel virtual space.
1997 */
1998void *vmalloc_32(unsigned long size)
1999{
David Miller2dca6992009-09-21 12:22:34 -07002000 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08002001 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003EXPORT_SYMBOL(vmalloc_32);
2004
Nick Piggin83342312006-06-23 02:03:20 -07002005/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07002006 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07002007 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07002008 *
2009 * The resulting memory area is 32bit addressable and zeroed so it can be
2010 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07002011 */
2012void *vmalloc_32_user(unsigned long size)
2013{
2014 struct vm_struct *area;
2015 void *ret;
2016
David Miller2dca6992009-09-21 12:22:34 -07002017 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08002018 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08002019 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07002020 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08002021 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08002022 }
Nick Piggin83342312006-06-23 02:03:20 -07002023 return ret;
2024}
2025EXPORT_SYMBOL(vmalloc_32_user);
2026
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002027/*
2028 * small helper routine , copy contents to buf from addr.
2029 * If the page is not present, fill zero.
2030 */
2031
2032static int aligned_vread(char *buf, char *addr, unsigned long count)
2033{
2034 struct page *p;
2035 int copied = 0;
2036
2037 while (count) {
2038 unsigned long offset, length;
2039
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002040 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002041 length = PAGE_SIZE - offset;
2042 if (length > count)
2043 length = count;
2044 p = vmalloc_to_page(addr);
2045 /*
2046 * To do safe access to this _mapped_ area, we need
2047 * lock. But adding lock here means that we need to add
2048 * overhead of vmalloc()/vfree() calles for this _debug_
2049 * interface, rarely used. Instead of that, we'll use
2050 * kmap() and get small overhead in this access function.
2051 */
2052 if (p) {
2053 /*
2054 * we can expect USER0 is not used (see vread/vwrite's
2055 * function description)
2056 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002057 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002058 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002059 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002060 } else
2061 memset(buf, 0, length);
2062
2063 addr += length;
2064 buf += length;
2065 copied += length;
2066 count -= length;
2067 }
2068 return copied;
2069}
2070
2071static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2072{
2073 struct page *p;
2074 int copied = 0;
2075
2076 while (count) {
2077 unsigned long offset, length;
2078
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002079 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002080 length = PAGE_SIZE - offset;
2081 if (length > count)
2082 length = count;
2083 p = vmalloc_to_page(addr);
2084 /*
2085 * To do safe access to this _mapped_ area, we need
2086 * lock. But adding lock here means that we need to add
2087 * overhead of vmalloc()/vfree() calles for this _debug_
2088 * interface, rarely used. Instead of that, we'll use
2089 * kmap() and get small overhead in this access function.
2090 */
2091 if (p) {
2092 /*
2093 * we can expect USER0 is not used (see vread/vwrite's
2094 * function description)
2095 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002096 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002097 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002098 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002099 }
2100 addr += length;
2101 buf += length;
2102 copied += length;
2103 count -= length;
2104 }
2105 return copied;
2106}
2107
2108/**
2109 * vread() - read vmalloc area in a safe way.
2110 * @buf: buffer for reading data
2111 * @addr: vm address.
2112 * @count: number of bytes to be read.
2113 *
2114 * Returns # of bytes which addr and buf should be increased.
2115 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2116 * includes any intersect with alive vmalloc area.
2117 *
2118 * This function checks that addr is a valid vmalloc'ed area, and
2119 * copy data from that area to a given buffer. If the given memory range
2120 * of [addr...addr+count) includes some valid address, data is copied to
2121 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2122 * IOREMAP area is treated as memory hole and no copy is done.
2123 *
2124 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002125 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002126 *
2127 * Note: In usual ops, vread() is never necessary because the caller
2128 * should know vmalloc() area is valid and can use memcpy().
2129 * This is for routines which have to access vmalloc area without
2130 * any informaion, as /dev/kmem.
2131 *
2132 */
2133
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134long vread(char *buf, char *addr, unsigned long count)
2135{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002136 struct vmap_area *va;
2137 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002139 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 unsigned long n;
2141
2142 /* Don't allow overflow */
2143 if ((unsigned long) addr + count < count)
2144 count = -(unsigned long) addr;
2145
Joonsoo Kime81ce852013-04-29 15:07:32 -07002146 spin_lock(&vmap_area_lock);
2147 list_for_each_entry(va, &vmap_area_list, list) {
2148 if (!count)
2149 break;
2150
2151 if (!(va->flags & VM_VM_AREA))
2152 continue;
2153
2154 vm = va->vm;
2155 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002156 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 continue;
2158 while (addr < vaddr) {
2159 if (count == 0)
2160 goto finished;
2161 *buf = '\0';
2162 buf++;
2163 addr++;
2164 count--;
2165 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002166 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002167 if (n > count)
2168 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002169 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002170 aligned_vread(buf, addr, n);
2171 else /* IOREMAP area is treated as memory hole */
2172 memset(buf, 0, n);
2173 buf += n;
2174 addr += n;
2175 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 }
2177finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002178 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002179
2180 if (buf == buf_start)
2181 return 0;
2182 /* zero-fill memory holes */
2183 if (buf != buf_start + buflen)
2184 memset(buf, 0, buflen - (buf - buf_start));
2185
2186 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187}
2188
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002189/**
2190 * vwrite() - write vmalloc area in a safe way.
2191 * @buf: buffer for source data
2192 * @addr: vm address.
2193 * @count: number of bytes to be read.
2194 *
2195 * Returns # of bytes which addr and buf should be incresed.
2196 * (same number to @count).
2197 * If [addr...addr+count) doesn't includes any intersect with valid
2198 * vmalloc area, returns 0.
2199 *
2200 * This function checks that addr is a valid vmalloc'ed area, and
2201 * copy data from a buffer to the given addr. If specified range of
2202 * [addr...addr+count) includes some valid address, data is copied from
2203 * proper area of @buf. If there are memory holes, no copy to hole.
2204 * IOREMAP area is treated as memory hole and no copy is done.
2205 *
2206 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002207 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002208 *
2209 * Note: In usual ops, vwrite() is never necessary because the caller
2210 * should know vmalloc() area is valid and can use memcpy().
2211 * This is for routines which have to access vmalloc area without
2212 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002213 */
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215long vwrite(char *buf, char *addr, unsigned long count)
2216{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002217 struct vmap_area *va;
2218 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002219 char *vaddr;
2220 unsigned long n, buflen;
2221 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
2223 /* Don't allow overflow */
2224 if ((unsigned long) addr + count < count)
2225 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002226 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Joonsoo Kime81ce852013-04-29 15:07:32 -07002228 spin_lock(&vmap_area_lock);
2229 list_for_each_entry(va, &vmap_area_list, list) {
2230 if (!count)
2231 break;
2232
2233 if (!(va->flags & VM_VM_AREA))
2234 continue;
2235
2236 vm = va->vm;
2237 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002238 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 continue;
2240 while (addr < vaddr) {
2241 if (count == 0)
2242 goto finished;
2243 buf++;
2244 addr++;
2245 count--;
2246 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002247 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002248 if (n > count)
2249 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002250 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002251 aligned_vwrite(buf, addr, n);
2252 copied++;
2253 }
2254 buf += n;
2255 addr += n;
2256 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 }
2258finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002259 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002260 if (!copied)
2261 return 0;
2262 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263}
Nick Piggin83342312006-06-23 02:03:20 -07002264
2265/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002266 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2267 * @vma: vma to cover
2268 * @uaddr: target user address to start at
2269 * @kaddr: virtual address of vmalloc kernel memory
2270 * @size: size of map area
2271 *
2272 * Returns: 0 for success, -Exxx on failure
2273 *
2274 * This function checks that @kaddr is a valid vmalloc'ed area,
2275 * and that it is big enough to cover the range starting at
2276 * @uaddr in @vma. Will return failure if that criteria isn't
2277 * met.
2278 *
2279 * Similar to remap_pfn_range() (see mm/memory.c)
2280 */
2281int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2282 void *kaddr, unsigned long size)
2283{
2284 struct vm_struct *area;
2285
2286 size = PAGE_ALIGN(size);
2287
2288 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2289 return -EINVAL;
2290
2291 area = find_vm_area(kaddr);
2292 if (!area)
2293 return -EINVAL;
2294
2295 if (!(area->flags & VM_USERMAP))
2296 return -EINVAL;
2297
2298 if (kaddr + size > area->addr + area->size)
2299 return -EINVAL;
2300
2301 do {
2302 struct page *page = vmalloc_to_page(kaddr);
2303 int ret;
2304
2305 ret = vm_insert_page(vma, uaddr, page);
2306 if (ret)
2307 return ret;
2308
2309 uaddr += PAGE_SIZE;
2310 kaddr += PAGE_SIZE;
2311 size -= PAGE_SIZE;
2312 } while (size > 0);
2313
2314 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2315
2316 return 0;
2317}
2318EXPORT_SYMBOL(remap_vmalloc_range_partial);
2319
2320/**
Nick Piggin83342312006-06-23 02:03:20 -07002321 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002322 * @vma: vma to cover (map full range of vma)
2323 * @addr: vmalloc memory
2324 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002325 *
2326 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002327 *
2328 * This function checks that addr is a valid vmalloc'ed area, and
2329 * that it is big enough to cover the vma. Will return failure if
2330 * that criteria isn't met.
2331 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002332 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002333 */
2334int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2335 unsigned long pgoff)
2336{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002337 return remap_vmalloc_range_partial(vma, vma->vm_start,
2338 addr + (pgoff << PAGE_SHIFT),
2339 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002340}
2341EXPORT_SYMBOL(remap_vmalloc_range);
2342
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002343/*
2344 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2345 * have one.
2346 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002347void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002348{
2349}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002350
2351
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002352static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002353{
David Vrabelcd129092011-09-29 16:53:32 +01002354 pte_t ***p = data;
2355
2356 if (p) {
2357 *(*p) = pte;
2358 (*p)++;
2359 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002360 return 0;
2361}
2362
2363/**
2364 * alloc_vm_area - allocate a range of kernel address space
2365 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002366 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002367 *
2368 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002369 *
2370 * This function reserves a range of kernel address space, and
2371 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002372 * are created.
2373 *
2374 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2375 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002376 */
David Vrabelcd129092011-09-29 16:53:32 +01002377struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002378{
2379 struct vm_struct *area;
2380
Christoph Lameter23016962008-04-28 02:12:42 -07002381 area = get_vm_area_caller(size, VM_IOREMAP,
2382 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002383 if (area == NULL)
2384 return NULL;
2385
2386 /*
2387 * This ensures that page tables are constructed for this region
2388 * of kernel virtual address space and mapped into init_mm.
2389 */
2390 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002391 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002392 free_vm_area(area);
2393 return NULL;
2394 }
2395
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002396 return area;
2397}
2398EXPORT_SYMBOL_GPL(alloc_vm_area);
2399
2400void free_vm_area(struct vm_struct *area)
2401{
2402 struct vm_struct *ret;
2403 ret = remove_vm_area(area->addr);
2404 BUG_ON(ret != area);
2405 kfree(area);
2406}
2407EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002408
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002409#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002410static struct vmap_area *node_to_va(struct rb_node *n)
2411{
2412 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2413}
2414
2415/**
2416 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2417 * @end: target address
2418 * @pnext: out arg for the next vmap_area
2419 * @pprev: out arg for the previous vmap_area
2420 *
2421 * Returns: %true if either or both of next and prev are found,
2422 * %false if no vmap_area exists
2423 *
2424 * Find vmap_areas end addresses of which enclose @end. ie. if not
2425 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2426 */
2427static bool pvm_find_next_prev(unsigned long end,
2428 struct vmap_area **pnext,
2429 struct vmap_area **pprev)
2430{
2431 struct rb_node *n = vmap_area_root.rb_node;
2432 struct vmap_area *va = NULL;
2433
2434 while (n) {
2435 va = rb_entry(n, struct vmap_area, rb_node);
2436 if (end < va->va_end)
2437 n = n->rb_left;
2438 else if (end > va->va_end)
2439 n = n->rb_right;
2440 else
2441 break;
2442 }
2443
2444 if (!va)
2445 return false;
2446
2447 if (va->va_end > end) {
2448 *pnext = va;
2449 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2450 } else {
2451 *pprev = va;
2452 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2453 }
2454 return true;
2455}
2456
2457/**
2458 * pvm_determine_end - find the highest aligned address between two vmap_areas
2459 * @pnext: in/out arg for the next vmap_area
2460 * @pprev: in/out arg for the previous vmap_area
2461 * @align: alignment
2462 *
2463 * Returns: determined end address
2464 *
2465 * Find the highest aligned address between *@pnext and *@pprev below
2466 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2467 * down address is between the end addresses of the two vmap_areas.
2468 *
2469 * Please note that the address returned by this function may fall
2470 * inside *@pnext vmap_area. The caller is responsible for checking
2471 * that.
2472 */
2473static unsigned long pvm_determine_end(struct vmap_area **pnext,
2474 struct vmap_area **pprev,
2475 unsigned long align)
2476{
2477 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2478 unsigned long addr;
2479
2480 if (*pnext)
2481 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2482 else
2483 addr = vmalloc_end;
2484
2485 while (*pprev && (*pprev)->va_end > addr) {
2486 *pnext = *pprev;
2487 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2488 }
2489
2490 return addr;
2491}
2492
2493/**
2494 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2495 * @offsets: array containing offset of each area
2496 * @sizes: array containing size of each area
2497 * @nr_vms: the number of areas to allocate
2498 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002499 *
2500 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2501 * vm_structs on success, %NULL on failure
2502 *
2503 * Percpu allocator wants to use congruent vm areas so that it can
2504 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002505 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2506 * be scattered pretty far, distance between two areas easily going up
2507 * to gigabytes. To avoid interacting with regular vmallocs, these
2508 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002509 *
2510 * Despite its complicated look, this allocator is rather simple. It
2511 * does everything top-down and scans areas from the end looking for
2512 * matching slot. While scanning, if any of the areas overlaps with
2513 * existing vmap_area, the base address is pulled down to fit the
2514 * area. Scanning is repeated till all the areas fit and then all
2515 * necessary data structres are inserted and the result is returned.
2516 */
2517struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2518 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002519 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002520{
2521 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2522 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2523 struct vmap_area **vas, *prev, *next;
2524 struct vm_struct **vms;
2525 int area, area2, last_area, term_area;
2526 unsigned long base, start, end, last_end;
2527 bool purged = false;
2528
Tejun Heoca23e402009-08-14 15:00:52 +09002529 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002530 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002531 for (last_area = 0, area = 0; area < nr_vms; area++) {
2532 start = offsets[area];
2533 end = start + sizes[area];
2534
2535 /* is everything aligned properly? */
2536 BUG_ON(!IS_ALIGNED(offsets[area], align));
2537 BUG_ON(!IS_ALIGNED(sizes[area], align));
2538
2539 /* detect the area with the highest address */
2540 if (start > offsets[last_area])
2541 last_area = area;
2542
2543 for (area2 = 0; area2 < nr_vms; area2++) {
2544 unsigned long start2 = offsets[area2];
2545 unsigned long end2 = start2 + sizes[area2];
2546
2547 if (area2 == area)
2548 continue;
2549
2550 BUG_ON(start2 >= start && start2 < end);
2551 BUG_ON(end2 <= end && end2 > start);
2552 }
2553 }
2554 last_end = offsets[last_area] + sizes[last_area];
2555
2556 if (vmalloc_end - vmalloc_start < last_end) {
2557 WARN_ON(true);
2558 return NULL;
2559 }
2560
Thomas Meyer4d67d862012-05-29 15:06:21 -07002561 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2562 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002563 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002564 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002565
2566 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002567 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2568 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002569 if (!vas[area] || !vms[area])
2570 goto err_free;
2571 }
2572retry:
2573 spin_lock(&vmap_area_lock);
2574
2575 /* start scanning - we scan from the top, begin with the last area */
2576 area = term_area = last_area;
2577 start = offsets[area];
2578 end = start + sizes[area];
2579
2580 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2581 base = vmalloc_end - last_end;
2582 goto found;
2583 }
2584 base = pvm_determine_end(&next, &prev, align) - end;
2585
2586 while (true) {
2587 BUG_ON(next && next->va_end <= base + end);
2588 BUG_ON(prev && prev->va_end > base + end);
2589
2590 /*
2591 * base might have underflowed, add last_end before
2592 * comparing.
2593 */
2594 if (base + last_end < vmalloc_start + last_end) {
2595 spin_unlock(&vmap_area_lock);
2596 if (!purged) {
2597 purge_vmap_area_lazy();
2598 purged = true;
2599 goto retry;
2600 }
2601 goto err_free;
2602 }
2603
2604 /*
2605 * If next overlaps, move base downwards so that it's
2606 * right below next and then recheck.
2607 */
2608 if (next && next->va_start < base + end) {
2609 base = pvm_determine_end(&next, &prev, align) - end;
2610 term_area = area;
2611 continue;
2612 }
2613
2614 /*
2615 * If prev overlaps, shift down next and prev and move
2616 * base so that it's right below new next and then
2617 * recheck.
2618 */
2619 if (prev && prev->va_end > base + start) {
2620 next = prev;
2621 prev = node_to_va(rb_prev(&next->rb_node));
2622 base = pvm_determine_end(&next, &prev, align) - end;
2623 term_area = area;
2624 continue;
2625 }
2626
2627 /*
2628 * This area fits, move on to the previous one. If
2629 * the previous one is the terminal one, we're done.
2630 */
2631 area = (area + nr_vms - 1) % nr_vms;
2632 if (area == term_area)
2633 break;
2634 start = offsets[area];
2635 end = start + sizes[area];
2636 pvm_find_next_prev(base + end, &next, &prev);
2637 }
2638found:
2639 /* we've found a fitting base, insert all va's */
2640 for (area = 0; area < nr_vms; area++) {
2641 struct vmap_area *va = vas[area];
2642
2643 va->va_start = base + offsets[area];
2644 va->va_end = va->va_start + sizes[area];
2645 __insert_vmap_area(va);
2646 }
2647
2648 vmap_area_pcpu_hole = base + offsets[last_area];
2649
2650 spin_unlock(&vmap_area_lock);
2651
2652 /* insert all vm's */
2653 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002654 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2655 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002656
2657 kfree(vas);
2658 return vms;
2659
2660err_free:
2661 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002662 kfree(vas[area]);
2663 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002664 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002665err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002666 kfree(vas);
2667 kfree(vms);
2668 return NULL;
2669}
2670
2671/**
2672 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2673 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2674 * @nr_vms: the number of allocated areas
2675 *
2676 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2677 */
2678void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2679{
2680 int i;
2681
2682 for (i = 0; i < nr_vms; i++)
2683 free_vm_area(vms[i]);
2684 kfree(vms);
2685}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002686#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002687
2688#ifdef CONFIG_PROC_FS
2689static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002690 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002691{
2692 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002693 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002694
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002695 spin_lock(&vmap_area_lock);
Geliang Tang6219c2a2016-01-14 15:19:08 -08002696 va = list_first_entry(&vmap_area_list, typeof(*va), list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002697 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002698 n--;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002699 va = list_next_entry(va, list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002700 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002701 if (!n && &va->list != &vmap_area_list)
2702 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002703
2704 return NULL;
2705
2706}
2707
2708static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2709{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002710 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002711
2712 ++*pos;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002713 next = list_next_entry(va, list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002714 if (&next->list != &vmap_area_list)
2715 return next;
2716
2717 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002718}
2719
2720static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002721 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002722{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002723 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002724}
2725
Eric Dumazeta47a1262008-07-23 21:27:38 -07002726static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2727{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002728 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002729 unsigned int nr, *counters = m->private;
2730
2731 if (!counters)
2732 return;
2733
Wanpeng Liaf123462013-11-12 15:07:32 -08002734 if (v->flags & VM_UNINITIALIZED)
2735 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002736 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2737 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002738
Eric Dumazeta47a1262008-07-23 21:27:38 -07002739 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2740
2741 for (nr = 0; nr < v->nr_pages; nr++)
2742 counters[page_to_nid(v->pages[nr])]++;
2743
2744 for_each_node_state(nr, N_HIGH_MEMORY)
2745 if (counters[nr])
2746 seq_printf(m, " N%u=%u", nr, counters[nr]);
2747 }
2748}
2749
Christoph Lametera10aa572008-04-28 02:12:40 -07002750static int s_show(struct seq_file *m, void *p)
2751{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002752 struct vmap_area *va = p;
2753 struct vm_struct *v;
2754
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002755 /*
2756 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2757 * behalf of vmap area is being tear down or vm_map_ram allocation.
2758 */
2759 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002760 return 0;
2761
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002762 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002763
Kees Cook45ec1692012-10-08 16:34:09 -07002764 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002765 v->addr, v->addr + v->size, v->size);
2766
Joe Perches62c70bc2011-01-13 15:45:52 -08002767 if (v->caller)
2768 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002769
Christoph Lametera10aa572008-04-28 02:12:40 -07002770 if (v->nr_pages)
2771 seq_printf(m, " pages=%d", v->nr_pages);
2772
2773 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002774 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002775
2776 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002777 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002778
2779 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002780 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002781
2782 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002783 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002784
2785 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002786 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002787
David Rientjes244d63e2016-01-14 15:19:35 -08002788 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002789 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002790
Susheel Khiani8db21e12013-08-22 13:46:07 -07002791 if (v->flags & VM_LOWMEM)
2792 seq_puts(m, " lowmem");
2793
Eric Dumazeta47a1262008-07-23 21:27:38 -07002794 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002795 seq_putc(m, '\n');
2796 return 0;
2797}
2798
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002799static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002800 .start = s_start,
2801 .next = s_next,
2802 .stop = s_stop,
2803 .show = s_show,
2804};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002805
2806static int vmalloc_open(struct inode *inode, struct file *file)
2807{
Rob Jones703394c2014-10-09 15:28:01 -07002808 if (IS_ENABLED(CONFIG_NUMA))
2809 return seq_open_private(file, &vmalloc_op,
2810 nr_node_ids * sizeof(unsigned int));
2811 else
2812 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002813}
2814
2815static const struct file_operations proc_vmalloc_operations = {
2816 .open = vmalloc_open,
2817 .read = seq_read,
2818 .llseek = seq_lseek,
2819 .release = seq_release_private,
2820};
2821
2822static int __init proc_vmalloc_init(void)
2823{
2824 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2825 return 0;
2826}
2827module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002828
Christoph Lametera10aa572008-04-28 02:12:40 -07002829#endif
2830