blob: 25fc1eaa594d6b97dcc245e318639891c8f31ecc [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
Chris Wilson4da56b92016-04-04 14:46:42 +0100429 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
430
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
Nick Piggin89699602011-03-22 16:30:36 -0700546 if (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
660 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
661}
662
663static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
664
Christoph Hellwig835db092016-12-12 16:44:07 -0800665/*
666 * Serialize vmap purging. There is no actual criticial section protected
667 * by this look, but we want to avoid concurrent calls for performance
668 * reasons and to make the pcpu_get_vm_areas more deterministic.
669 */
670static DEFINE_SPINLOCK(vmap_purge_lock);
671
Nick Piggin02b709d2010-02-01 22:25:57 +1100672/* for per-CPU blocks */
673static void purge_fragmented_blocks_allcpus(void);
674
Nick Piggindb64fe02008-10-18 20:27:03 -0700675/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500676 * called before a call to iounmap() if the caller wants vm_area_struct's
677 * immediately freed.
678 */
679void set_iounmap_nonlazy(void)
680{
681 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
682}
683
684/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700685 * Purges all lazily-freed vmap areas.
Nick Piggindb64fe02008-10-18 20:27:03 -0700686 */
Christoph Hellwig835db092016-12-12 16:44:07 -0800687static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700688{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700689 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -0700690 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800691 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700692 int nr = 0;
693
Christoph Hellwig835db092016-12-12 16:44:07 -0800694 lockdep_assert_held(&vmap_purge_lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100695
Chris Wilson80c4bd72016-05-20 16:57:38 -0700696 valist = llist_del_all(&vmap_purge_list);
697 llist_for_each_entry(va, valist, purge_list) {
Christoph Hellwig835db092016-12-12 16:44:07 -0800698 if (va->va_start < start)
699 start = va->va_start;
700 if (va->va_end > end)
701 end = va->va_end;
Chris Wilson80c4bd72016-05-20 16:57:38 -0700702 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700703 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700704
Christoph Hellwig835db092016-12-12 16:44:07 -0800705 if (!nr)
706 return false;
Nick Piggindb64fe02008-10-18 20:27:03 -0700707
Christoph Hellwig835db092016-12-12 16:44:07 -0800708 atomic_sub(nr, &vmap_lazy_nr);
709 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);
712 llist_for_each_entry_safe(va, n_va, valist, purge_list)
713 __free_vmap_area(va);
714 spin_unlock(&vmap_area_lock);
715 return true;
Nick Piggindb64fe02008-10-18 20:27:03 -0700716}
717
718/*
Nick Piggin496850e2008-11-19 15:36:33 -0800719 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
720 * is already purging.
721 */
722static void try_purge_vmap_area_lazy(void)
723{
Christoph Hellwig835db092016-12-12 16:44:07 -0800724 if (spin_trylock(&vmap_purge_lock)) {
725 __purge_vmap_area_lazy(ULONG_MAX, 0);
726 spin_unlock(&vmap_purge_lock);
727 }
Nick Piggin496850e2008-11-19 15:36:33 -0800728}
729
730/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700731 * Kick off a purge of the outstanding lazy areas.
732 */
733static void purge_vmap_area_lazy(void)
734{
Christoph Hellwig835db092016-12-12 16:44:07 -0800735 spin_lock(&vmap_purge_lock);
736 purge_fragmented_blocks_allcpus();
737 __purge_vmap_area_lazy(ULONG_MAX, 0);
738 spin_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700739}
740
741/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800742 * Free a vmap area, caller ensuring that the area has been unmapped
743 * and flush_cache_vunmap had been called for the correct range
744 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700745 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800746static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700747{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700748 int nr_lazy;
749
750 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
751 &vmap_lazy_nr);
752
753 /* After this point, we may free va at any time */
754 llist_add(&va->purge_list, &vmap_purge_list);
755
756 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800757 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700758}
759
Nick Pigginb29acbd2008-12-01 13:13:47 -0800760/*
761 * Free and unmap a vmap area
762 */
763static void free_unmap_vmap_area(struct vmap_area *va)
764{
765 flush_cache_vunmap(va->va_start, va->va_end);
Christoph Hellwig71aadd02016-12-12 16:44:01 -0800766 unmap_vmap_area(va);
767 free_vmap_area_noflush(va);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800768}
769
Nick Piggindb64fe02008-10-18 20:27:03 -0700770static struct vmap_area *find_vmap_area(unsigned long addr)
771{
772 struct vmap_area *va;
773
774 spin_lock(&vmap_area_lock);
775 va = __find_vmap_area(addr);
776 spin_unlock(&vmap_area_lock);
777
778 return va;
779}
780
Nick Piggindb64fe02008-10-18 20:27:03 -0700781/*** Per cpu kva allocator ***/
782
783/*
784 * vmap space is limited especially on 32 bit architectures. Ensure there is
785 * room for at least 16 percpu vmap blocks per CPU.
786 */
787/*
788 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
789 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
790 * instead (we just need a rough idea)
791 */
792#if BITS_PER_LONG == 32
793#define VMALLOC_SPACE (128UL*1024*1024)
794#else
795#define VMALLOC_SPACE (128UL*1024*1024*1024)
796#endif
797
798#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
799#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
800#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
801#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
802#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
803#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200804#define VMAP_BBMAP_BITS \
805 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
806 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
807 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700808
809#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
810
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100811static bool vmap_initialized __read_mostly = false;
812
Nick Piggindb64fe02008-10-18 20:27:03 -0700813struct vmap_block_queue {
814 spinlock_t lock;
815 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700816};
817
818struct vmap_block {
819 spinlock_t lock;
820 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700821 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700822 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100823 struct list_head free_list;
824 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100825 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700826};
827
828/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
829static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
830
831/*
832 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
833 * in the free path. Could get rid of this if we change the API to return a
834 * "cookie" from alloc, to be passed to free. But no big deal yet.
835 */
836static DEFINE_SPINLOCK(vmap_block_tree_lock);
837static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
838
839/*
840 * We should probably have a fallback mechanism to allocate virtual memory
841 * out of partially filled vmap blocks. However vmap block sizing should be
842 * fairly reasonable according to the vmalloc size, so it shouldn't be a
843 * big problem.
844 */
845
846static unsigned long addr_to_vb_idx(unsigned long addr)
847{
848 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
849 addr /= VMAP_BLOCK_SIZE;
850 return addr;
851}
852
Roman Pencf725ce2015-04-15 16:13:52 -0700853static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
854{
855 unsigned long addr;
856
857 addr = va_start + (pages_off << PAGE_SHIFT);
858 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
859 return (void *)addr;
860}
861
862/**
863 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
864 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
865 * @order: how many 2^order pages should be occupied in newly allocated block
866 * @gfp_mask: flags for the page level allocator
867 *
868 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
869 */
870static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700871{
872 struct vmap_block_queue *vbq;
873 struct vmap_block *vb;
874 struct vmap_area *va;
875 unsigned long vb_idx;
876 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700877 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700878
879 node = numa_node_id();
880
881 vb = kmalloc_node(sizeof(struct vmap_block),
882 gfp_mask & GFP_RECLAIM_MASK, node);
883 if (unlikely(!vb))
884 return ERR_PTR(-ENOMEM);
885
886 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
887 VMALLOC_START, VMALLOC_END,
888 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800889 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700890 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700891 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700892 }
893
894 err = radix_tree_preload(gfp_mask);
895 if (unlikely(err)) {
896 kfree(vb);
897 free_vmap_area(va);
898 return ERR_PTR(err);
899 }
900
Roman Pencf725ce2015-04-15 16:13:52 -0700901 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700902 spin_lock_init(&vb->lock);
903 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700904 /* At least something should be left free */
905 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
906 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700907 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700908 vb->dirty_min = VMAP_BBMAP_BITS;
909 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700910 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700911
912 vb_idx = addr_to_vb_idx(va->va_start);
913 spin_lock(&vmap_block_tree_lock);
914 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
915 spin_unlock(&vmap_block_tree_lock);
916 BUG_ON(err);
917 radix_tree_preload_end();
918
919 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700920 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700921 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700922 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900923 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700924
Roman Pencf725ce2015-04-15 16:13:52 -0700925 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700926}
927
Nick Piggindb64fe02008-10-18 20:27:03 -0700928static void free_vmap_block(struct vmap_block *vb)
929{
930 struct vmap_block *tmp;
931 unsigned long vb_idx;
932
Nick Piggindb64fe02008-10-18 20:27:03 -0700933 vb_idx = addr_to_vb_idx(vb->va->va_start);
934 spin_lock(&vmap_block_tree_lock);
935 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
936 spin_unlock(&vmap_block_tree_lock);
937 BUG_ON(tmp != vb);
938
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800939 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800940 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700941}
942
Nick Piggin02b709d2010-02-01 22:25:57 +1100943static void purge_fragmented_blocks(int cpu)
944{
945 LIST_HEAD(purge);
946 struct vmap_block *vb;
947 struct vmap_block *n_vb;
948 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
949
950 rcu_read_lock();
951 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
952
953 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
954 continue;
955
956 spin_lock(&vb->lock);
957 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
958 vb->free = 0; /* prevent further allocs after releasing lock */
959 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700960 vb->dirty_min = 0;
961 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100962 spin_lock(&vbq->lock);
963 list_del_rcu(&vb->free_list);
964 spin_unlock(&vbq->lock);
965 spin_unlock(&vb->lock);
966 list_add_tail(&vb->purge, &purge);
967 } else
968 spin_unlock(&vb->lock);
969 }
970 rcu_read_unlock();
971
972 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
973 list_del(&vb->purge);
974 free_vmap_block(vb);
975 }
976}
977
Nick Piggin02b709d2010-02-01 22:25:57 +1100978static void purge_fragmented_blocks_allcpus(void)
979{
980 int cpu;
981
982 for_each_possible_cpu(cpu)
983 purge_fragmented_blocks(cpu);
984}
985
Nick Piggindb64fe02008-10-18 20:27:03 -0700986static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
987{
988 struct vmap_block_queue *vbq;
989 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700990 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700991 unsigned int order;
992
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800993 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700994 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700995 if (WARN_ON(size == 0)) {
996 /*
997 * Allocating 0 bytes isn't what caller wants since
998 * get_order(0) returns funny result. Just warn and terminate
999 * early.
1000 */
1001 return NULL;
1002 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001003 order = get_order(size);
1004
Nick Piggindb64fe02008-10-18 20:27:03 -07001005 rcu_read_lock();
1006 vbq = &get_cpu_var(vmap_block_queue);
1007 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -07001008 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -07001009
1010 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -07001011 if (vb->free < (1UL << order)) {
1012 spin_unlock(&vb->lock);
1013 continue;
1014 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001015
Roman Pencf725ce2015-04-15 16:13:52 -07001016 pages_off = VMAP_BBMAP_BITS - vb->free;
1017 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +11001018 vb->free -= 1UL << order;
1019 if (vb->free == 0) {
1020 spin_lock(&vbq->lock);
1021 list_del_rcu(&vb->free_list);
1022 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001023 }
Roman Pencf725ce2015-04-15 16:13:52 -07001024
Nick Piggindb64fe02008-10-18 20:27:03 -07001025 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001026 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001027 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001028
Tejun Heo3f04ba82009-10-29 22:34:12 +09001029 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001030 rcu_read_unlock();
1031
Roman Pencf725ce2015-04-15 16:13:52 -07001032 /* Allocate new block if nothing was found */
1033 if (!vaddr)
1034 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001035
Roman Pencf725ce2015-04-15 16:13:52 -07001036 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001037}
1038
1039static void vb_free(const void *addr, unsigned long size)
1040{
1041 unsigned long offset;
1042 unsigned long vb_idx;
1043 unsigned int order;
1044 struct vmap_block *vb;
1045
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001046 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001047 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001048
1049 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1050
Nick Piggindb64fe02008-10-18 20:27:03 -07001051 order = get_order(size);
1052
1053 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001054 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001055
1056 vb_idx = addr_to_vb_idx((unsigned long)addr);
1057 rcu_read_lock();
1058 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1059 rcu_read_unlock();
1060 BUG_ON(!vb);
1061
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001062 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1063
Nick Piggindb64fe02008-10-18 20:27:03 -07001064 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001065
1066 /* Expand dirty range */
1067 vb->dirty_min = min(vb->dirty_min, offset);
1068 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001069
Nick Piggindb64fe02008-10-18 20:27:03 -07001070 vb->dirty += 1UL << order;
1071 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001072 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001073 spin_unlock(&vb->lock);
1074 free_vmap_block(vb);
1075 } else
1076 spin_unlock(&vb->lock);
1077}
1078
1079/**
1080 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1081 *
1082 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1083 * to amortize TLB flushing overheads. What this means is that any page you
1084 * have now, may, in a former life, have been mapped into kernel virtual
1085 * address by the vmap layer and so there might be some CPUs with TLB entries
1086 * still referencing that page (additional to the regular 1:1 kernel mapping).
1087 *
1088 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1089 * be sure that none of the pages we have control over will have any aliases
1090 * from the vmap layer.
1091 */
1092void vm_unmap_aliases(void)
1093{
1094 unsigned long start = ULONG_MAX, end = 0;
1095 int cpu;
1096 int flush = 0;
1097
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001098 if (unlikely(!vmap_initialized))
1099 return;
1100
Nick Piggindb64fe02008-10-18 20:27:03 -07001101 for_each_possible_cpu(cpu) {
1102 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1103 struct vmap_block *vb;
1104
1105 rcu_read_lock();
1106 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001107 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001108 if (vb->dirty) {
1109 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001110 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001111
Roman Pen7d61bfe2015-04-15 16:13:55 -07001112 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1113 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001114
Roman Pen7d61bfe2015-04-15 16:13:55 -07001115 start = min(s, start);
1116 end = max(e, end);
1117
Nick Piggindb64fe02008-10-18 20:27:03 -07001118 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001119 }
1120 spin_unlock(&vb->lock);
1121 }
1122 rcu_read_unlock();
1123 }
1124
Christoph Hellwig835db092016-12-12 16:44:07 -08001125 spin_lock(&vmap_purge_lock);
1126 purge_fragmented_blocks_allcpus();
1127 if (!__purge_vmap_area_lazy(start, end) && flush)
1128 flush_tlb_kernel_range(start, end);
1129 spin_unlock(&vmap_purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001130}
1131EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1132
1133/**
1134 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1135 * @mem: the pointer returned by vm_map_ram
1136 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1137 */
1138void vm_unmap_ram(const void *mem, unsigned int count)
1139{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001140 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001141 unsigned long addr = (unsigned long)mem;
Christoph Hellwig7df98752016-12-12 16:44:04 -08001142 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001143
1144 BUG_ON(!addr);
1145 BUG_ON(addr < VMALLOC_START);
1146 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001147 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001148
1149 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001150 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001151
Christoph Hellwig7df98752016-12-12 16:44:04 -08001152 if (likely(count <= VMAP_MAX_ALLOC)) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001153 vb_free(mem, size);
Christoph Hellwig7df98752016-12-12 16:44:04 -08001154 return;
1155 }
1156
1157 va = find_vmap_area(addr);
1158 BUG_ON(!va);
1159 free_unmap_vmap_area(va);
Nick Piggindb64fe02008-10-18 20:27:03 -07001160}
1161EXPORT_SYMBOL(vm_unmap_ram);
1162
1163/**
1164 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1165 * @pages: an array of pointers to the pages to be mapped
1166 * @count: number of pages
1167 * @node: prefer to allocate data structures on this node
1168 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001169 *
Gioh Kim36437632014-04-07 15:37:37 -07001170 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1171 * faster than vmap so it's good. But if you mix long-life and short-life
1172 * objects with vm_map_ram(), it could consume lots of address space through
1173 * fragmentation (especially on a 32bit machine). You could see failures in
1174 * the end. Please use this function for short-lived objects.
1175 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001176 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001177 */
1178void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1179{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001180 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001181 unsigned long addr;
1182 void *mem;
1183
1184 if (likely(count <= VMAP_MAX_ALLOC)) {
1185 mem = vb_alloc(size, GFP_KERNEL);
1186 if (IS_ERR(mem))
1187 return NULL;
1188 addr = (unsigned long)mem;
1189 } else {
1190 struct vmap_area *va;
1191 va = alloc_vmap_area(size, PAGE_SIZE,
1192 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1193 if (IS_ERR(va))
1194 return NULL;
1195
1196 addr = va->va_start;
1197 mem = (void *)addr;
1198 }
1199 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1200 vm_unmap_ram(mem, count);
1201 return NULL;
1202 }
1203 return mem;
1204}
1205EXPORT_SYMBOL(vm_map_ram);
1206
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001207static struct vm_struct *vmlist __initdata;
Susheel Khiani432f1662015-09-08 15:05:43 +05301208
1209/**
1210 * vm_area_check_early - check if vmap area is already mapped
1211 * @vm: vm_struct to be checked
1212 *
1213 * This function is used to check if the vmap area has been
1214 * mapped already. @vm->addr, @vm->size and @vm->flags should
1215 * contain proper values.
1216 *
1217 */
1218int __init vm_area_check_early(struct vm_struct *vm)
1219{
1220 struct vm_struct *tmp, **p;
1221
1222 BUG_ON(vmap_initialized);
1223 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1224 if (tmp->addr >= vm->addr) {
1225 if (tmp->addr < vm->addr + vm->size)
1226 return 1;
1227 } else {
1228 if (tmp->addr + tmp->size > vm->addr)
1229 return 1;
1230 }
1231 }
1232 return 0;
1233}
1234
Tejun Heof0aa6612009-02-20 16:29:08 +09001235/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001236 * vm_area_add_early - add vmap area early during boot
1237 * @vm: vm_struct to add
1238 *
1239 * This function is used to add fixed kernel vm area to vmlist before
1240 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1241 * should contain proper values and the other fields should be zero.
1242 *
1243 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1244 */
1245void __init vm_area_add_early(struct vm_struct *vm)
1246{
1247 struct vm_struct *tmp, **p;
1248
1249 BUG_ON(vmap_initialized);
1250 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1251 if (tmp->addr >= vm->addr) {
1252 BUG_ON(tmp->addr < vm->addr + vm->size);
1253 break;
1254 } else
1255 BUG_ON(tmp->addr + tmp->size > vm->addr);
1256 }
1257 vm->next = *p;
1258 *p = vm;
1259}
1260
1261/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001262 * vm_area_register_early - register vmap area early during boot
1263 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001264 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001265 *
1266 * This function is used to register kernel vm area before
1267 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1268 * proper values on entry and other fields should be zero. On return,
1269 * vm->addr contains the allocated address.
1270 *
1271 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1272 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001273void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001274{
1275 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001276 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001277
Tejun Heoc0c0a292009-02-24 11:57:21 +09001278 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1279 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1280
1281 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001282
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001283 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001284}
1285
Nick Piggindb64fe02008-10-18 20:27:03 -07001286void __init vmalloc_init(void)
1287{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001288 struct vmap_area *va;
1289 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001290 int i;
1291
1292 for_each_possible_cpu(i) {
1293 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001294 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001295
1296 vbq = &per_cpu(vmap_block_queue, i);
1297 spin_lock_init(&vbq->lock);
1298 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001299 p = &per_cpu(vfree_deferred, i);
1300 init_llist_head(&p->list);
1301 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001302 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001303
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001304 /* Import existing vmlist entries. */
1305 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001306 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001307 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001308 va->va_start = (unsigned long)tmp->addr;
1309 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001310 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001311 __insert_vmap_area(va);
1312 }
Tejun Heoca23e402009-08-14 15:00:52 +09001313
1314 vmap_area_pcpu_hole = VMALLOC_END;
Susheel Khiani8db21e12013-08-22 13:46:07 -07001315 calc_total_vmalloc_size();
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001316 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001317}
1318
Tejun Heo8fc48982009-02-20 16:29:08 +09001319/**
1320 * map_kernel_range_noflush - map kernel VM area with the specified pages
1321 * @addr: start of the VM area to map
1322 * @size: size of the VM area to map
1323 * @prot: page protection flags to use
1324 * @pages: pages to map
1325 *
1326 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1327 * specify should have been allocated using get_vm_area() and its
1328 * friends.
1329 *
1330 * NOTE:
1331 * This function does NOT do any cache flushing. The caller is
1332 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1333 * before calling this function.
1334 *
1335 * RETURNS:
1336 * The number of pages mapped on success, -errno on failure.
1337 */
1338int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1339 pgprot_t prot, struct page **pages)
1340{
1341 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1342}
1343
1344/**
1345 * unmap_kernel_range_noflush - unmap kernel VM area
1346 * @addr: start of the VM area to unmap
1347 * @size: size of the VM area to unmap
1348 *
1349 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1350 * specify should have been allocated using get_vm_area() and its
1351 * friends.
1352 *
1353 * NOTE:
1354 * This function does NOT do any cache flushing. The caller is
1355 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1356 * before calling this function and flush_tlb_kernel_range() after.
1357 */
1358void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1359{
1360 vunmap_page_range(addr, addr + size);
1361}
Huang Ying81e88fd2011-01-12 14:44:55 +08001362EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001363
1364/**
1365 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1366 * @addr: start of the VM area to unmap
1367 * @size: size of the VM area to unmap
1368 *
1369 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1370 * the unmapping and tlb after.
1371 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001372void unmap_kernel_range(unsigned long addr, unsigned long size)
1373{
1374 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001375
1376 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001377 vunmap_page_range(addr, end);
1378 flush_tlb_kernel_range(addr, end);
1379}
Minchan Kim93ef6d62014-06-04 16:11:09 -07001380EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001381
WANG Chaof6f8ed42014-08-06 16:06:58 -07001382int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001383{
1384 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001385 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001386 int err;
1387
WANG Chaof6f8ed42014-08-06 16:06:58 -07001388 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001389
WANG Chaof6f8ed42014-08-06 16:06:58 -07001390 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001391}
1392EXPORT_SYMBOL_GPL(map_vm_area);
1393
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001394static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001395 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001396{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001397 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001398 vm->flags = flags;
1399 vm->addr = (void *)va->va_start;
1400 vm->size = va->va_end - va->va_start;
1401 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001402 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001403 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001404 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001405}
Tejun Heocf88c792009-08-14 15:00:52 +09001406
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001407static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001408{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001409 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001410 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001411 * we should make sure that vm has proper values.
1412 * Pair with smp_rmb() in show_numa_info().
1413 */
1414 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001415 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001416}
1417
Nick Piggindb64fe02008-10-18 20:27:03 -07001418static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001419 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001420 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001421{
Kautuk Consul00065262011-12-19 17:12:04 -08001422 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001423 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001425 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001427 if (unlikely(!size))
1428 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
zijun_hu252e5c62016-10-07 16:57:26 -07001430 if (flags & VM_IOREMAP)
1431 align = 1ul << clamp_t(int, get_count_order_long(size),
1432 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1433
Tejun Heocf88c792009-08-14 15:00:52 +09001434 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 if (unlikely(!area))
1436 return NULL;
1437
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001438 if (!(flags & VM_NO_GUARD))
1439 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Nick Piggindb64fe02008-10-18 20:27:03 -07001441 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1442 if (IS_ERR(va)) {
1443 kfree(area);
1444 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001447 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450}
1451
Christoph Lameter930fc452005-10-29 18:15:41 -07001452struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1453 unsigned long start, unsigned long end)
1454{
David Rientjes00ef2d22013-02-22 16:35:36 -08001455 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1456 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001457}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001458EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001459
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001460struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1461 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001462 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001463{
David Rientjes00ef2d22013-02-22 16:35:36 -08001464 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1465 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001466}
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468/**
Simon Arlott183ff222007-10-20 01:27:18 +02001469 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 * @size: size of the area
1471 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1472 *
1473 * Search an area of @size in the kernel virtual mapping area,
1474 * and reserved it for out purposes. Returns the area descriptor
1475 * on success or %NULL on failure.
1476 */
1477struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1478{
Susheel Khianid0bd1992015-09-03 18:21:23 +05301479#ifdef CONFIG_ENABLE_VMALLOC_SAVING
1480 return __get_vm_area_node(size, 1, flags, PAGE_OFFSET, VMALLOC_END,
1481 NUMA_NO_NODE, GFP_KERNEL,
1482 __builtin_return_address(0));
1483#else
David Miller2dca6992009-09-21 12:22:34 -07001484 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001485 NUMA_NO_NODE, GFP_KERNEL,
1486 __builtin_return_address(0));
Susheel Khianid0bd1992015-09-03 18:21:23 +05301487#endif
Christoph Lameter23016962008-04-28 02:12:42 -07001488}
1489
1490struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001491 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001492{
Susheel Khianid0bd1992015-09-03 18:21:23 +05301493#ifdef CONFIG_ENABLE_VMALLOC_SAVING
1494 return __get_vm_area_node(size, 1, flags, PAGE_OFFSET, VMALLOC_END,
1495 NUMA_NO_NODE, GFP_KERNEL, caller);
1496#else
David Miller2dca6992009-09-21 12:22:34 -07001497 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001498 NUMA_NO_NODE, GFP_KERNEL, caller);
Susheel Khianid0bd1992015-09-03 18:21:23 +05301499#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001502/**
1503 * find_vm_area - find a continuous kernel virtual area
1504 * @addr: base address
1505 *
1506 * Search for the kernel VM area starting at @addr, and return it.
1507 * It is up to the caller to do all required locking to keep the returned
1508 * pointer valid.
1509 */
1510struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001511{
Nick Piggindb64fe02008-10-18 20:27:03 -07001512 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001513
Nick Piggindb64fe02008-10-18 20:27:03 -07001514 va = find_vmap_area((unsigned long)addr);
1515 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001516 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001517
Andi Kleen7856dfe2005-05-20 14:27:57 -07001518 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001519}
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521/**
Simon Arlott183ff222007-10-20 01:27:18 +02001522 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 * @addr: base address
1524 *
1525 * Search for the kernel VM area starting at @addr, and remove it.
1526 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001527 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001529struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
Nick Piggindb64fe02008-10-18 20:27:03 -07001531 struct vmap_area *va;
1532
1533 va = find_vmap_area((unsigned long)addr);
1534 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001535 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001536
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001537 spin_lock(&vmap_area_lock);
1538 va->vm = NULL;
1539 va->flags &= ~VM_VM_AREA;
1540 spin_unlock(&vmap_area_lock);
1541
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001542 vmap_debug_free_range(va->va_start, va->va_end);
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001543 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001544 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001545
Nick Piggindb64fe02008-10-18 20:27:03 -07001546 return vm;
1547 }
1548 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001551static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552{
1553 struct vm_struct *area;
1554
1555 if (!addr)
1556 return;
1557
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001558 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001559 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 area = remove_vm_area(addr);
1563 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001564 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 return;
1567 }
1568
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001569 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1570 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (deallocate_pages) {
1573 int i;
1574
1575 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001576 struct page *page = area->pages[i];
1577
1578 BUG_ON(!page);
Vladimir Davydov49491482016-07-26 15:24:24 -07001579 __free_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
David Rientjes244d63e2016-01-14 15:19:35 -08001582 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 }
1584
1585 kfree(area);
1586 return;
1587}
Andrey Ryabinin3af922c2016-12-12 16:44:10 -08001588
1589static inline void __vfree_deferred(const void *addr)
1590{
1591 /*
1592 * Use raw_cpu_ptr() because this can be called from preemptible
1593 * context. Preemption is absolutely fine here, because the llist_add()
1594 * implementation is lockless, so it works even if we are adding to
1595 * nother cpu's list. schedule_work() should be fine with this too.
1596 */
1597 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1598
1599 if (llist_add((struct llist_node *)addr, &p->list))
1600 schedule_work(&p->wq);
1601}
1602
1603/**
1604 * vfree_atomic - release memory allocated by vmalloc()
1605 * @addr: memory base address
1606 *
1607 * This one is just like vfree() but can be called in any atomic context
1608 * except NMIs.
1609 */
1610void vfree_atomic(const void *addr)
1611{
1612 BUG_ON(in_nmi());
1613
1614 kmemleak_free(addr);
1615
1616 if (!addr)
1617 return;
1618 __vfree_deferred(addr);
1619}
1620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621/**
1622 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 * @addr: memory base address
1624 *
Simon Arlott183ff222007-10-20 01:27:18 +02001625 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001626 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1627 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 *
Al Viro32fcfd42013-03-10 20:14:08 -04001629 * Must not be called in NMI context (strictly speaking, only if we don't
1630 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1631 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001632 *
1633 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001635void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
Al Viro32fcfd42013-03-10 20:14:08 -04001637 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001638
1639 kmemleak_free(addr);
1640
Al Viro32fcfd42013-03-10 20:14:08 -04001641 if (!addr)
1642 return;
Andrey Ryabinin3af922c2016-12-12 16:44:10 -08001643 if (unlikely(in_interrupt()))
1644 __vfree_deferred(addr);
1645 else
Al Viro32fcfd42013-03-10 20:14:08 -04001646 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648EXPORT_SYMBOL(vfree);
1649
1650/**
1651 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 * @addr: memory base address
1653 *
1654 * Free the virtually contiguous memory area starting at @addr,
1655 * which was created from the page array passed to vmap().
1656 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001657 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001659void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660{
1661 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001662 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001663 if (addr)
1664 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666EXPORT_SYMBOL(vunmap);
1667
1668/**
1669 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 * @pages: array of page pointers
1671 * @count: number of pages to map
1672 * @flags: vm_area->flags
1673 * @prot: page protection for the mapping
1674 *
1675 * Maps @count pages from @pages into contiguous kernel virtual
1676 * space.
1677 */
1678void *vmap(struct page **pages, unsigned int count,
1679 unsigned long flags, pgprot_t prot)
1680{
1681 struct vm_struct *area;
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001682 unsigned long size; /* In bytes */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Peter Zijlstra34754b62009-02-25 16:04:03 +01001684 might_sleep();
1685
Jan Beulich44813742009-09-21 17:03:05 -07001686 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return NULL;
1688
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001689 size = (unsigned long)count << PAGE_SHIFT;
1690 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 if (!area)
1692 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001693
WANG Chaof6f8ed42014-08-06 16:06:58 -07001694 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 vunmap(area->addr);
1696 return NULL;
1697 }
1698
1699 return area->addr;
1700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701EXPORT_SYMBOL(vmap);
1702
David Miller2dca6992009-09-21 12:22:34 -07001703static void *__vmalloc_node(unsigned long size, unsigned long align,
1704 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001705 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001706static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001707 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
1709 struct page **pages;
1710 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001711 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1712 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713
Wanpeng Li762216a2013-09-11 14:22:42 -07001714 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 array_size = (nr_pages * sizeof(struct page *));
1716
1717 area->nr_pages = nr_pages;
1718 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001719 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001720 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001721 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001722 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001723 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 area->pages = pages;
1726 if (!area->pages) {
1727 remove_vm_area(area->addr);
1728 kfree(area);
1729 return NULL;
1730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
1732 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001733 struct page *page;
1734
Jianguo Wu4b909512013-11-12 15:07:11 -08001735 if (node == NUMA_NO_NODE)
Michal Hocko7877cdc2016-10-07 17:01:55 -07001736 page = alloc_page(alloc_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001737 else
Michal Hocko7877cdc2016-10-07 17:01:55 -07001738 page = alloc_pages_node(node, alloc_mask, 0);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001739
1740 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 /* Successfully allocated i pages, free them in __vunmap() */
1742 area->nr_pages = i;
1743 goto fail;
1744 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001745 area->pages[i] = page;
Mel Gormand0164ad2015-11-06 16:28:21 -08001746 if (gfpflags_allow_blocking(gfp_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001747 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 }
1749
WANG Chaof6f8ed42014-08-06 16:06:58 -07001750 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 goto fail;
1752 return area->addr;
1753
1754fail:
Michal Hocko7877cdc2016-10-07 17:01:55 -07001755 warn_alloc(gfp_mask,
1756 "vmalloc: allocation failure, allocated %ld of %ld bytes",
Dave Hansen22943ab2011-05-24 17:12:18 -07001757 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 vfree(area->addr);
1759 return NULL;
1760}
1761
David Rientjesd0a21262011-01-13 15:46:02 -08001762/**
1763 * __vmalloc_node_range - allocate virtually contiguous memory
1764 * @size: allocation size
1765 * @align: desired alignment
1766 * @start: vm area range start
1767 * @end: vm area range end
1768 * @gfp_mask: flags for the page level allocator
1769 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001770 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001771 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001772 * @caller: caller's return address
1773 *
1774 * Allocate enough pages to cover @size from the page level
1775 * allocator with @gfp_mask flags. Map them into contiguous
1776 * kernel virtual space, using a pagetable protection of @prot.
1777 */
1778void *__vmalloc_node_range(unsigned long size, unsigned long align,
1779 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001780 pgprot_t prot, unsigned long vm_flags, int node,
1781 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001782{
David Rientjesd0a21262011-01-13 15:46:02 -08001783 struct vm_struct *area;
1784 void *addr;
1785 unsigned long real_size = size;
1786
1787 size = PAGE_ALIGN(size);
1788 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001789 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001790
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001791 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1792 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001793 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001794 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001795
Wanpeng Li3722e132013-11-12 15:07:29 -08001796 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001797 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001798 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001799
1800 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001801 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1802 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001803 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001804 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001805 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001806
1807 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001808 * A ref_count = 2 is needed because vm_struct allocated in
1809 * __get_vm_area_node() contains a reference to the virtual address of
1810 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001811 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001812 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001813
1814 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001815
1816fail:
Michal Hocko7877cdc2016-10-07 17:01:55 -07001817 warn_alloc(gfp_mask,
1818 "vmalloc: allocation failure: %lu bytes", real_size);
Joe Perchesde7d2b52011-10-31 17:08:48 -07001819 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001820}
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001823 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001825 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 * @gfp_mask: flags for the page level allocator
1827 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001828 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001829 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 *
1831 * Allocate enough pages to cover @size from the page level
1832 * allocator with @gfp_mask flags. Map them into contiguous
1833 * kernel virtual space, using a pagetable protection of @prot.
1834 */
David Miller2dca6992009-09-21 12:22:34 -07001835static void *__vmalloc_node(unsigned long size, unsigned long align,
1836 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001837 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838{
David Rientjesd0a21262011-01-13 15:46:02 -08001839 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001840 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841}
1842
Christoph Lameter930fc452005-10-29 18:15:41 -07001843void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1844{
David Rientjes00ef2d22013-02-22 16:35:36 -08001845 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001846 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001847}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848EXPORT_SYMBOL(__vmalloc);
1849
Dave Younge1ca7782010-10-26 14:22:06 -07001850static inline void *__vmalloc_node_flags(unsigned long size,
1851 int node, gfp_t flags)
1852{
1853 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1854 node, __builtin_return_address(0));
1855}
1856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857/**
1858 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 * Allocate enough pages to cover @size from the page level
1861 * allocator and map them into contiguous kernel virtual space.
1862 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001863 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 * use __vmalloc() instead.
1865 */
1866void *vmalloc(unsigned long size)
1867{
David Rientjes00ef2d22013-02-22 16:35:36 -08001868 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1869 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871EXPORT_SYMBOL(vmalloc);
1872
Christoph Lameter930fc452005-10-29 18:15:41 -07001873/**
Dave Younge1ca7782010-10-26 14:22:06 -07001874 * vzalloc - allocate virtually contiguous memory with zero fill
1875 * @size: allocation size
1876 * Allocate enough pages to cover @size from the page level
1877 * allocator and map them into contiguous kernel virtual space.
1878 * The memory allocated is set to zero.
1879 *
1880 * For tight control over page level allocator and protection flags
1881 * use __vmalloc() instead.
1882 */
1883void *vzalloc(unsigned long size)
1884{
David Rientjes00ef2d22013-02-22 16:35:36 -08001885 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001886 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1887}
1888EXPORT_SYMBOL(vzalloc);
1889
1890/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001891 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1892 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001893 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001894 * The resulting memory area is zeroed so it can be mapped to userspace
1895 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001896 */
1897void *vmalloc_user(unsigned long size)
1898{
1899 struct vm_struct *area;
1900 void *ret;
1901
David Miller2dca6992009-09-21 12:22:34 -07001902 ret = __vmalloc_node(size, SHMLBA,
1903 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001904 PAGE_KERNEL, NUMA_NO_NODE,
1905 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001906 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001907 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001908 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001909 }
Nick Piggin83342312006-06-23 02:03:20 -07001910 return ret;
1911}
1912EXPORT_SYMBOL(vmalloc_user);
1913
1914/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001915 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001916 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001917 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001918 *
1919 * Allocate enough pages to cover @size from the page level
1920 * allocator and map them into contiguous kernel virtual space.
1921 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001922 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001923 * use __vmalloc() instead.
1924 */
1925void *vmalloc_node(unsigned long size, int node)
1926{
David Miller2dca6992009-09-21 12:22:34 -07001927 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001928 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001929}
1930EXPORT_SYMBOL(vmalloc_node);
1931
Dave Younge1ca7782010-10-26 14:22:06 -07001932/**
1933 * vzalloc_node - allocate memory on a specific node with zero fill
1934 * @size: allocation size
1935 * @node: numa node
1936 *
1937 * Allocate enough pages to cover @size from the page level
1938 * allocator and map them into contiguous kernel virtual space.
1939 * The memory allocated is set to zero.
1940 *
1941 * For tight control over page level allocator and protection flags
1942 * use __vmalloc_node() instead.
1943 */
1944void *vzalloc_node(unsigned long size, int node)
1945{
1946 return __vmalloc_node_flags(size, node,
1947 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1948}
1949EXPORT_SYMBOL(vzalloc_node);
1950
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001951#ifndef PAGE_KERNEL_EXEC
1952# define PAGE_KERNEL_EXEC PAGE_KERNEL
1953#endif
1954
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955/**
1956 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 * @size: allocation size
1958 *
1959 * Kernel-internal function to allocate enough pages to cover @size
1960 * the page level allocator and map them into contiguous and
1961 * executable kernel virtual space.
1962 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001963 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 * use __vmalloc() instead.
1965 */
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967void *vmalloc_exec(unsigned long size)
1968{
David Miller2dca6992009-09-21 12:22:34 -07001969 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001970 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}
1972
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001973#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001974#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001975#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001976#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001977#else
1978#define GFP_VMALLOC32 GFP_KERNEL
1979#endif
1980
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981/**
1982 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 * @size: allocation size
1984 *
1985 * Allocate enough 32bit PA addressable pages to cover @size from the
1986 * page level allocator and map them into contiguous kernel virtual space.
1987 */
1988void *vmalloc_32(unsigned long size)
1989{
David Miller2dca6992009-09-21 12:22:34 -07001990 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001991 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993EXPORT_SYMBOL(vmalloc_32);
1994
Nick Piggin83342312006-06-23 02:03:20 -07001995/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001996 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001997 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001998 *
1999 * The resulting memory area is 32bit addressable and zeroed so it can be
2000 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07002001 */
2002void *vmalloc_32_user(unsigned long size)
2003{
2004 struct vm_struct *area;
2005 void *ret;
2006
David Miller2dca6992009-09-21 12:22:34 -07002007 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08002008 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08002009 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07002010 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08002011 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08002012 }
Nick Piggin83342312006-06-23 02:03:20 -07002013 return ret;
2014}
2015EXPORT_SYMBOL(vmalloc_32_user);
2016
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002017/*
2018 * small helper routine , copy contents to buf from addr.
2019 * If the page is not present, fill zero.
2020 */
2021
2022static int aligned_vread(char *buf, char *addr, unsigned long count)
2023{
2024 struct page *p;
2025 int copied = 0;
2026
2027 while (count) {
2028 unsigned long offset, length;
2029
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002030 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002031 length = PAGE_SIZE - offset;
2032 if (length > count)
2033 length = count;
2034 p = vmalloc_to_page(addr);
2035 /*
2036 * To do safe access to this _mapped_ area, we need
2037 * lock. But adding lock here means that we need to add
2038 * overhead of vmalloc()/vfree() calles for this _debug_
2039 * interface, rarely used. Instead of that, we'll use
2040 * kmap() and get small overhead in this access function.
2041 */
2042 if (p) {
2043 /*
2044 * we can expect USER0 is not used (see vread/vwrite's
2045 * function description)
2046 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002047 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002048 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002049 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002050 } else
2051 memset(buf, 0, length);
2052
2053 addr += length;
2054 buf += length;
2055 copied += length;
2056 count -= length;
2057 }
2058 return copied;
2059}
2060
2061static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2062{
2063 struct page *p;
2064 int copied = 0;
2065
2066 while (count) {
2067 unsigned long offset, length;
2068
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002069 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002070 length = PAGE_SIZE - offset;
2071 if (length > count)
2072 length = count;
2073 p = vmalloc_to_page(addr);
2074 /*
2075 * To do safe access to this _mapped_ area, we need
2076 * lock. But adding lock here means that we need to add
2077 * overhead of vmalloc()/vfree() calles for this _debug_
2078 * interface, rarely used. Instead of that, we'll use
2079 * kmap() and get small overhead in this access function.
2080 */
2081 if (p) {
2082 /*
2083 * we can expect USER0 is not used (see vread/vwrite's
2084 * function description)
2085 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08002086 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002087 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002088 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002089 }
2090 addr += length;
2091 buf += length;
2092 copied += length;
2093 count -= length;
2094 }
2095 return copied;
2096}
2097
2098/**
2099 * vread() - read vmalloc area in a safe way.
2100 * @buf: buffer for reading data
2101 * @addr: vm address.
2102 * @count: number of bytes to be read.
2103 *
2104 * Returns # of bytes which addr and buf should be increased.
2105 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2106 * includes any intersect with alive vmalloc area.
2107 *
2108 * This function checks that addr is a valid vmalloc'ed area, and
2109 * copy data from that area to a given buffer. If the given memory range
2110 * of [addr...addr+count) includes some valid address, data is copied to
2111 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2112 * IOREMAP area is treated as memory hole and no copy is done.
2113 *
2114 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002115 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002116 *
2117 * Note: In usual ops, vread() is never necessary because the caller
2118 * should know vmalloc() area is valid and can use memcpy().
2119 * This is for routines which have to access vmalloc area without
2120 * any informaion, as /dev/kmem.
2121 *
2122 */
2123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124long vread(char *buf, char *addr, unsigned long count)
2125{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002126 struct vmap_area *va;
2127 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002129 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 unsigned long n;
2131
2132 /* Don't allow overflow */
2133 if ((unsigned long) addr + count < count)
2134 count = -(unsigned long) addr;
2135
Joonsoo Kime81ce852013-04-29 15:07:32 -07002136 spin_lock(&vmap_area_lock);
2137 list_for_each_entry(va, &vmap_area_list, list) {
2138 if (!count)
2139 break;
2140
2141 if (!(va->flags & VM_VM_AREA))
2142 continue;
2143
2144 vm = va->vm;
2145 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002146 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 continue;
2148 while (addr < vaddr) {
2149 if (count == 0)
2150 goto finished;
2151 *buf = '\0';
2152 buf++;
2153 addr++;
2154 count--;
2155 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002156 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002157 if (n > count)
2158 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002159 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002160 aligned_vread(buf, addr, n);
2161 else /* IOREMAP area is treated as memory hole */
2162 memset(buf, 0, n);
2163 buf += n;
2164 addr += n;
2165 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 }
2167finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002168 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002169
2170 if (buf == buf_start)
2171 return 0;
2172 /* zero-fill memory holes */
2173 if (buf != buf_start + buflen)
2174 memset(buf, 0, buflen - (buf - buf_start));
2175
2176 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177}
2178
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002179/**
2180 * vwrite() - write vmalloc area in a safe way.
2181 * @buf: buffer for source data
2182 * @addr: vm address.
2183 * @count: number of bytes to be read.
2184 *
2185 * Returns # of bytes which addr and buf should be incresed.
2186 * (same number to @count).
2187 * If [addr...addr+count) doesn't includes any intersect with valid
2188 * vmalloc area, returns 0.
2189 *
2190 * This function checks that addr is a valid vmalloc'ed area, and
2191 * copy data from a buffer to the given addr. If specified range of
2192 * [addr...addr+count) includes some valid address, data is copied from
2193 * proper area of @buf. If there are memory holes, no copy to hole.
2194 * IOREMAP area is treated as memory hole and no copy is done.
2195 *
2196 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002197 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002198 *
2199 * Note: In usual ops, vwrite() is never necessary because the caller
2200 * should know vmalloc() area is valid and can use memcpy().
2201 * This is for routines which have to access vmalloc area without
2202 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002203 */
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205long vwrite(char *buf, char *addr, unsigned long count)
2206{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002207 struct vmap_area *va;
2208 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002209 char *vaddr;
2210 unsigned long n, buflen;
2211 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212
2213 /* Don't allow overflow */
2214 if ((unsigned long) addr + count < count)
2215 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002216 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Joonsoo Kime81ce852013-04-29 15:07:32 -07002218 spin_lock(&vmap_area_lock);
2219 list_for_each_entry(va, &vmap_area_list, list) {
2220 if (!count)
2221 break;
2222
2223 if (!(va->flags & VM_VM_AREA))
2224 continue;
2225
2226 vm = va->vm;
2227 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002228 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 continue;
2230 while (addr < vaddr) {
2231 if (count == 0)
2232 goto finished;
2233 buf++;
2234 addr++;
2235 count--;
2236 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002237 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002238 if (n > count)
2239 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002240 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002241 aligned_vwrite(buf, addr, n);
2242 copied++;
2243 }
2244 buf += n;
2245 addr += n;
2246 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 }
2248finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002249 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002250 if (!copied)
2251 return 0;
2252 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253}
Nick Piggin83342312006-06-23 02:03:20 -07002254
2255/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002256 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2257 * @vma: vma to cover
2258 * @uaddr: target user address to start at
2259 * @kaddr: virtual address of vmalloc kernel memory
2260 * @size: size of map area
2261 *
2262 * Returns: 0 for success, -Exxx on failure
2263 *
2264 * This function checks that @kaddr is a valid vmalloc'ed area,
2265 * and that it is big enough to cover the range starting at
2266 * @uaddr in @vma. Will return failure if that criteria isn't
2267 * met.
2268 *
2269 * Similar to remap_pfn_range() (see mm/memory.c)
2270 */
2271int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2272 void *kaddr, unsigned long size)
2273{
2274 struct vm_struct *area;
2275
2276 size = PAGE_ALIGN(size);
2277
2278 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2279 return -EINVAL;
2280
2281 area = find_vm_area(kaddr);
2282 if (!area)
2283 return -EINVAL;
2284
2285 if (!(area->flags & VM_USERMAP))
2286 return -EINVAL;
2287
2288 if (kaddr + size > area->addr + area->size)
2289 return -EINVAL;
2290
2291 do {
2292 struct page *page = vmalloc_to_page(kaddr);
2293 int ret;
2294
2295 ret = vm_insert_page(vma, uaddr, page);
2296 if (ret)
2297 return ret;
2298
2299 uaddr += PAGE_SIZE;
2300 kaddr += PAGE_SIZE;
2301 size -= PAGE_SIZE;
2302 } while (size > 0);
2303
2304 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2305
2306 return 0;
2307}
2308EXPORT_SYMBOL(remap_vmalloc_range_partial);
2309
2310/**
Nick Piggin83342312006-06-23 02:03:20 -07002311 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002312 * @vma: vma to cover (map full range of vma)
2313 * @addr: vmalloc memory
2314 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002315 *
2316 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002317 *
2318 * This function checks that addr is a valid vmalloc'ed area, and
2319 * that it is big enough to cover the vma. Will return failure if
2320 * that criteria isn't met.
2321 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002322 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002323 */
2324int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2325 unsigned long pgoff)
2326{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002327 return remap_vmalloc_range_partial(vma, vma->vm_start,
2328 addr + (pgoff << PAGE_SHIFT),
2329 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002330}
2331EXPORT_SYMBOL(remap_vmalloc_range);
2332
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002333/*
2334 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2335 * have one.
2336 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002337void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002338{
2339}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002340
2341
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002342static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002343{
David Vrabelcd129092011-09-29 16:53:32 +01002344 pte_t ***p = data;
2345
2346 if (p) {
2347 *(*p) = pte;
2348 (*p)++;
2349 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002350 return 0;
2351}
2352
2353/**
2354 * alloc_vm_area - allocate a range of kernel address space
2355 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002356 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002357 *
2358 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002359 *
2360 * This function reserves a range of kernel address space, and
2361 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002362 * are created.
2363 *
2364 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2365 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002366 */
David Vrabelcd129092011-09-29 16:53:32 +01002367struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002368{
2369 struct vm_struct *area;
2370
Christoph Lameter23016962008-04-28 02:12:42 -07002371 area = get_vm_area_caller(size, VM_IOREMAP,
2372 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002373 if (area == NULL)
2374 return NULL;
2375
2376 /*
2377 * This ensures that page tables are constructed for this region
2378 * of kernel virtual address space and mapped into init_mm.
2379 */
2380 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002381 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002382 free_vm_area(area);
2383 return NULL;
2384 }
2385
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002386 return area;
2387}
2388EXPORT_SYMBOL_GPL(alloc_vm_area);
2389
2390void free_vm_area(struct vm_struct *area)
2391{
2392 struct vm_struct *ret;
2393 ret = remove_vm_area(area->addr);
2394 BUG_ON(ret != area);
2395 kfree(area);
2396}
2397EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002398
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002399#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002400static struct vmap_area *node_to_va(struct rb_node *n)
2401{
2402 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2403}
2404
2405/**
2406 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2407 * @end: target address
2408 * @pnext: out arg for the next vmap_area
2409 * @pprev: out arg for the previous vmap_area
2410 *
2411 * Returns: %true if either or both of next and prev are found,
2412 * %false if no vmap_area exists
2413 *
2414 * Find vmap_areas end addresses of which enclose @end. ie. if not
2415 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2416 */
2417static bool pvm_find_next_prev(unsigned long end,
2418 struct vmap_area **pnext,
2419 struct vmap_area **pprev)
2420{
2421 struct rb_node *n = vmap_area_root.rb_node;
2422 struct vmap_area *va = NULL;
2423
2424 while (n) {
2425 va = rb_entry(n, struct vmap_area, rb_node);
2426 if (end < va->va_end)
2427 n = n->rb_left;
2428 else if (end > va->va_end)
2429 n = n->rb_right;
2430 else
2431 break;
2432 }
2433
2434 if (!va)
2435 return false;
2436
2437 if (va->va_end > end) {
2438 *pnext = va;
2439 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2440 } else {
2441 *pprev = va;
2442 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2443 }
2444 return true;
2445}
2446
2447/**
2448 * pvm_determine_end - find the highest aligned address between two vmap_areas
2449 * @pnext: in/out arg for the next vmap_area
2450 * @pprev: in/out arg for the previous vmap_area
2451 * @align: alignment
2452 *
2453 * Returns: determined end address
2454 *
2455 * Find the highest aligned address between *@pnext and *@pprev below
2456 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2457 * down address is between the end addresses of the two vmap_areas.
2458 *
2459 * Please note that the address returned by this function may fall
2460 * inside *@pnext vmap_area. The caller is responsible for checking
2461 * that.
2462 */
2463static unsigned long pvm_determine_end(struct vmap_area **pnext,
2464 struct vmap_area **pprev,
2465 unsigned long align)
2466{
2467 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2468 unsigned long addr;
2469
2470 if (*pnext)
2471 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2472 else
2473 addr = vmalloc_end;
2474
2475 while (*pprev && (*pprev)->va_end > addr) {
2476 *pnext = *pprev;
2477 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2478 }
2479
2480 return addr;
2481}
2482
2483/**
2484 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2485 * @offsets: array containing offset of each area
2486 * @sizes: array containing size of each area
2487 * @nr_vms: the number of areas to allocate
2488 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002489 *
2490 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2491 * vm_structs on success, %NULL on failure
2492 *
2493 * Percpu allocator wants to use congruent vm areas so that it can
2494 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002495 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2496 * be scattered pretty far, distance between two areas easily going up
2497 * to gigabytes. To avoid interacting with regular vmallocs, these
2498 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002499 *
2500 * Despite its complicated look, this allocator is rather simple. It
2501 * does everything top-down and scans areas from the end looking for
2502 * matching slot. While scanning, if any of the areas overlaps with
2503 * existing vmap_area, the base address is pulled down to fit the
2504 * area. Scanning is repeated till all the areas fit and then all
2505 * necessary data structres are inserted and the result is returned.
2506 */
2507struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2508 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002509 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002510{
2511 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2512 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2513 struct vmap_area **vas, *prev, *next;
2514 struct vm_struct **vms;
2515 int area, area2, last_area, term_area;
2516 unsigned long base, start, end, last_end;
2517 bool purged = false;
2518
Tejun Heoca23e402009-08-14 15:00:52 +09002519 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002520 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002521 for (last_area = 0, area = 0; area < nr_vms; area++) {
2522 start = offsets[area];
2523 end = start + sizes[area];
2524
2525 /* is everything aligned properly? */
2526 BUG_ON(!IS_ALIGNED(offsets[area], align));
2527 BUG_ON(!IS_ALIGNED(sizes[area], align));
2528
2529 /* detect the area with the highest address */
2530 if (start > offsets[last_area])
2531 last_area = area;
2532
2533 for (area2 = 0; area2 < nr_vms; area2++) {
2534 unsigned long start2 = offsets[area2];
2535 unsigned long end2 = start2 + sizes[area2];
2536
2537 if (area2 == area)
2538 continue;
2539
2540 BUG_ON(start2 >= start && start2 < end);
2541 BUG_ON(end2 <= end && end2 > start);
2542 }
2543 }
2544 last_end = offsets[last_area] + sizes[last_area];
2545
2546 if (vmalloc_end - vmalloc_start < last_end) {
2547 WARN_ON(true);
2548 return NULL;
2549 }
2550
Thomas Meyer4d67d862012-05-29 15:06:21 -07002551 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2552 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002553 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002554 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002555
2556 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002557 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2558 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002559 if (!vas[area] || !vms[area])
2560 goto err_free;
2561 }
2562retry:
2563 spin_lock(&vmap_area_lock);
2564
2565 /* start scanning - we scan from the top, begin with the last area */
2566 area = term_area = last_area;
2567 start = offsets[area];
2568 end = start + sizes[area];
2569
2570 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2571 base = vmalloc_end - last_end;
2572 goto found;
2573 }
2574 base = pvm_determine_end(&next, &prev, align) - end;
2575
2576 while (true) {
2577 BUG_ON(next && next->va_end <= base + end);
2578 BUG_ON(prev && prev->va_end > base + end);
2579
2580 /*
2581 * base might have underflowed, add last_end before
2582 * comparing.
2583 */
2584 if (base + last_end < vmalloc_start + last_end) {
2585 spin_unlock(&vmap_area_lock);
2586 if (!purged) {
2587 purge_vmap_area_lazy();
2588 purged = true;
2589 goto retry;
2590 }
2591 goto err_free;
2592 }
2593
2594 /*
2595 * If next overlaps, move base downwards so that it's
2596 * right below next and then recheck.
2597 */
2598 if (next && next->va_start < base + end) {
2599 base = pvm_determine_end(&next, &prev, align) - end;
2600 term_area = area;
2601 continue;
2602 }
2603
2604 /*
2605 * If prev overlaps, shift down next and prev and move
2606 * base so that it's right below new next and then
2607 * recheck.
2608 */
2609 if (prev && prev->va_end > base + start) {
2610 next = prev;
2611 prev = node_to_va(rb_prev(&next->rb_node));
2612 base = pvm_determine_end(&next, &prev, align) - end;
2613 term_area = area;
2614 continue;
2615 }
2616
2617 /*
2618 * This area fits, move on to the previous one. If
2619 * the previous one is the terminal one, we're done.
2620 */
2621 area = (area + nr_vms - 1) % nr_vms;
2622 if (area == term_area)
2623 break;
2624 start = offsets[area];
2625 end = start + sizes[area];
2626 pvm_find_next_prev(base + end, &next, &prev);
2627 }
2628found:
2629 /* we've found a fitting base, insert all va's */
2630 for (area = 0; area < nr_vms; area++) {
2631 struct vmap_area *va = vas[area];
2632
2633 va->va_start = base + offsets[area];
2634 va->va_end = va->va_start + sizes[area];
2635 __insert_vmap_area(va);
2636 }
2637
2638 vmap_area_pcpu_hole = base + offsets[last_area];
2639
2640 spin_unlock(&vmap_area_lock);
2641
2642 /* insert all vm's */
2643 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002644 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2645 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002646
2647 kfree(vas);
2648 return vms;
2649
2650err_free:
2651 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002652 kfree(vas[area]);
2653 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002654 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002655err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002656 kfree(vas);
2657 kfree(vms);
2658 return NULL;
2659}
2660
2661/**
2662 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2663 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2664 * @nr_vms: the number of allocated areas
2665 *
2666 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2667 */
2668void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2669{
2670 int i;
2671
2672 for (i = 0; i < nr_vms; i++)
2673 free_vm_area(vms[i]);
2674 kfree(vms);
2675}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002676#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002677
2678#ifdef CONFIG_PROC_FS
2679static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002680 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002681{
2682 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002683 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002684
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002685 spin_lock(&vmap_area_lock);
Geliang Tang6219c2a2016-01-14 15:19:08 -08002686 va = list_first_entry(&vmap_area_list, typeof(*va), list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002687 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002688 n--;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002689 va = list_next_entry(va, list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002690 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002691 if (!n && &va->list != &vmap_area_list)
2692 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002693
2694 return NULL;
2695
2696}
2697
2698static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2699{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002700 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002701
2702 ++*pos;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002703 next = list_next_entry(va, list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002704 if (&next->list != &vmap_area_list)
2705 return next;
2706
2707 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002708}
2709
2710static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002711 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002712{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002713 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002714}
2715
Eric Dumazeta47a1262008-07-23 21:27:38 -07002716static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2717{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002718 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002719 unsigned int nr, *counters = m->private;
2720
2721 if (!counters)
2722 return;
2723
Wanpeng Liaf123462013-11-12 15:07:32 -08002724 if (v->flags & VM_UNINITIALIZED)
2725 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002726 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2727 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002728
Eric Dumazeta47a1262008-07-23 21:27:38 -07002729 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2730
2731 for (nr = 0; nr < v->nr_pages; nr++)
2732 counters[page_to_nid(v->pages[nr])]++;
2733
2734 for_each_node_state(nr, N_HIGH_MEMORY)
2735 if (counters[nr])
2736 seq_printf(m, " N%u=%u", nr, counters[nr]);
2737 }
2738}
2739
Christoph Lametera10aa572008-04-28 02:12:40 -07002740static int s_show(struct seq_file *m, void *p)
2741{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002742 struct vmap_area *va = p;
2743 struct vm_struct *v;
2744
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002745 /*
2746 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2747 * behalf of vmap area is being tear down or vm_map_ram allocation.
2748 */
2749 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002750 return 0;
2751
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002752 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002753
Kees Cook45ec1692012-10-08 16:34:09 -07002754 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002755 v->addr, v->addr + v->size, v->size);
2756
Joe Perches62c70bc2011-01-13 15:45:52 -08002757 if (v->caller)
2758 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002759
Christoph Lametera10aa572008-04-28 02:12:40 -07002760 if (v->nr_pages)
2761 seq_printf(m, " pages=%d", v->nr_pages);
2762
2763 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002764 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002765
2766 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002767 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002768
2769 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002770 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002771
2772 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002773 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002774
2775 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002776 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002777
David Rientjes244d63e2016-01-14 15:19:35 -08002778 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002779 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002780
Susheel Khiani8db21e12013-08-22 13:46:07 -07002781 if (v->flags & VM_LOWMEM)
2782 seq_puts(m, " lowmem");
2783
Eric Dumazeta47a1262008-07-23 21:27:38 -07002784 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002785 seq_putc(m, '\n');
2786 return 0;
2787}
2788
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002789static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002790 .start = s_start,
2791 .next = s_next,
2792 .stop = s_stop,
2793 .show = s_show,
2794};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002795
2796static int vmalloc_open(struct inode *inode, struct file *file)
2797{
Rob Jones703394c2014-10-09 15:28:01 -07002798 if (IS_ENABLED(CONFIG_NUMA))
2799 return seq_open_private(file, &vmalloc_op,
2800 nr_node_ids * sizeof(unsigned int));
2801 else
2802 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002803}
2804
2805static const struct file_operations proc_vmalloc_operations = {
2806 .open = vmalloc_open,
2807 .read = seq_read,
2808 .llseek = seq_lseek,
2809 .release = seq_release_private,
2810};
2811
2812static int __init proc_vmalloc_init(void)
2813{
2814 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2815 return 0;
2816}
2817module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002818
Christoph Lametera10aa572008-04-28 02:12:40 -07002819#endif
2820