blob: 293889d7f482995daf8d39de6174698fbeea20be [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
Christoph Lameter48667e72008-02-04 22:28:31 -0800247 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700248 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800249 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700250 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800251 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700252 pte_t *ptep, pte;
253
Christoph Lameter48667e72008-02-04 22:28:31 -0800254 ptep = pte_offset_map(pmd, addr);
255 pte = *ptep;
256 if (pte_present(pte))
malcadd688f2014-01-27 17:06:53 -0800257 page = pte_page(pte);
Christoph Lameter48667e72008-02-04 22:28:31 -0800258 pte_unmap(ptep);
259 }
260 }
261 }
malcadd688f2014-01-27 17:06:53 -0800262 return page;
Jianyu Zhanece86e222014-01-21 15:49:12 -0800263}
264EXPORT_SYMBOL(vmalloc_to_page);
265
malcadd688f2014-01-27 17:06:53 -0800266/*
267 * Map a vmalloc()-space virtual address to the physical page frame number.
268 */
269unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
270{
271 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
272}
273EXPORT_SYMBOL(vmalloc_to_pfn);
274
Nick Piggindb64fe02008-10-18 20:27:03 -0700275
276/*** Global kva allocator ***/
277
278#define VM_LAZY_FREE 0x01
279#define VM_LAZY_FREEING 0x02
280#define VM_VM_AREA 0x04
281
Nick Piggindb64fe02008-10-18 20:27:03 -0700282static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700283/* Export for kexec only */
284LIST_HEAD(vmap_area_list);
Nick Piggin89699602011-03-22 16:30:36 -0700285static struct rb_root vmap_area_root = RB_ROOT;
286
287/* The vmap cache globals are protected by vmap_area_lock */
288static struct rb_node *free_vmap_cache;
289static unsigned long cached_hole_size;
290static unsigned long cached_vstart;
291static unsigned long cached_align;
292
Tejun Heoca23e402009-08-14 15:00:52 +0900293static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700294
295static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Nick Piggindb64fe02008-10-18 20:27:03 -0700297 struct rb_node *n = vmap_area_root.rb_node;
298
299 while (n) {
300 struct vmap_area *va;
301
302 va = rb_entry(n, struct vmap_area, rb_node);
303 if (addr < va->va_start)
304 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700305 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700306 n = n->rb_right;
307 else
308 return va;
309 }
310
311 return NULL;
312}
313
314static void __insert_vmap_area(struct vmap_area *va)
315{
316 struct rb_node **p = &vmap_area_root.rb_node;
317 struct rb_node *parent = NULL;
318 struct rb_node *tmp;
319
320 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700321 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700322
323 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700324 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
325 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700326 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700327 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700328 p = &(*p)->rb_right;
329 else
330 BUG();
331 }
332
333 rb_link_node(&va->rb_node, parent, p);
334 rb_insert_color(&va->rb_node, &vmap_area_root);
335
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700336 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700337 tmp = rb_prev(&va->rb_node);
338 if (tmp) {
339 struct vmap_area *prev;
340 prev = rb_entry(tmp, struct vmap_area, rb_node);
341 list_add_rcu(&va->list, &prev->list);
342 } else
343 list_add_rcu(&va->list, &vmap_area_list);
344}
345
346static void purge_vmap_area_lazy(void);
347
Chris Wilson4da56b92016-04-04 14:46:42 +0100348static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
349
Nick Piggindb64fe02008-10-18 20:27:03 -0700350/*
351 * Allocate a region of KVA of the specified size and alignment, within the
352 * vstart and vend.
353 */
354static struct vmap_area *alloc_vmap_area(unsigned long size,
355 unsigned long align,
356 unsigned long vstart, unsigned long vend,
357 int node, gfp_t gfp_mask)
358{
359 struct vmap_area *va;
360 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700362 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700363 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700364
Nick Piggin77669702009-02-27 14:03:03 -0800365 BUG_ON(!size);
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800366 BUG_ON(offset_in_page(size));
Nick Piggin89699602011-03-22 16:30:36 -0700367 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700368
Chris Wilson4da56b92016-04-04 14:46:42 +0100369 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
370
Nick Piggindb64fe02008-10-18 20:27:03 -0700371 va = kmalloc_node(sizeof(struct vmap_area),
372 gfp_mask & GFP_RECLAIM_MASK, node);
373 if (unlikely(!va))
374 return ERR_PTR(-ENOMEM);
375
Catalin Marinas7f88f882013-11-12 15:07:45 -0800376 /*
377 * Only scan the relevant parts containing pointers to other objects
378 * to avoid false negatives.
379 */
380 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
381
Nick Piggindb64fe02008-10-18 20:27:03 -0700382retry:
383 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700384 /*
385 * Invalidate cache if we have more permissive parameters.
386 * cached_hole_size notes the largest hole noticed _below_
387 * the vmap_area cached in free_vmap_cache: if size fits
388 * into that hole, we want to scan from vstart to reuse
389 * the hole instead of allocating above free_vmap_cache.
390 * Note that __free_vmap_area may update free_vmap_cache
391 * without updating cached_hole_size or cached_align.
392 */
393 if (!free_vmap_cache ||
394 size < cached_hole_size ||
395 vstart < cached_vstart ||
396 align < cached_align) {
397nocache:
398 cached_hole_size = 0;
399 free_vmap_cache = NULL;
400 }
401 /* record if we encounter less permissive parameters */
402 cached_vstart = vstart;
403 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800404
Nick Piggin89699602011-03-22 16:30:36 -0700405 /* find starting point for our search */
406 if (free_vmap_cache) {
407 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700408 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700409 if (addr < vstart)
410 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700411 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700412 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700413
Nick Piggin89699602011-03-22 16:30:36 -0700414 } else {
415 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700416 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700417 goto overflow;
418
419 n = vmap_area_root.rb_node;
420 first = NULL;
421
422 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700423 struct vmap_area *tmp;
424 tmp = rb_entry(n, struct vmap_area, rb_node);
425 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700426 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700427 if (tmp->va_start <= addr)
428 break;
429 n = n->rb_left;
430 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700431 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700432 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700433
434 if (!first)
435 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700436 }
Nick Piggin89699602011-03-22 16:30:36 -0700437
438 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700439 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700440 if (addr + cached_hole_size < first->va_start)
441 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700442 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700443 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700444 goto overflow;
445
Hong zhi guo92ca9222012-07-31 16:41:35 -0700446 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700447 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700448
Geliang Tang6219c2a2016-01-14 15:19:08 -0800449 first = list_next_entry(first, list);
Nick Piggin89699602011-03-22 16:30:36 -0700450 }
451
Nick Piggindb64fe02008-10-18 20:27:03 -0700452found:
Nick Piggin89699602011-03-22 16:30:36 -0700453 if (addr + size > vend)
454 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700455
456 va->va_start = addr;
457 va->va_end = addr + size;
458 va->flags = 0;
459 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700460 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700461 spin_unlock(&vmap_area_lock);
462
Wang Xiaoqiang61e16552016-01-15 16:57:19 -0800463 BUG_ON(!IS_ALIGNED(va->va_start, align));
Nick Piggin89699602011-03-22 16:30:36 -0700464 BUG_ON(va->va_start < vstart);
465 BUG_ON(va->va_end > vend);
466
Nick Piggindb64fe02008-10-18 20:27:03 -0700467 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700468
469overflow:
470 spin_unlock(&vmap_area_lock);
471 if (!purged) {
472 purge_vmap_area_lazy();
473 purged = 1;
474 goto retry;
475 }
Chris Wilson4da56b92016-04-04 14:46:42 +0100476
477 if (gfpflags_allow_blocking(gfp_mask)) {
478 unsigned long freed = 0;
479 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
480 if (freed > 0) {
481 purged = 0;
482 goto retry;
483 }
484 }
485
Nick Piggin89699602011-03-22 16:30:36 -0700486 if (printk_ratelimit())
Joe Perches756a025f02016-03-17 14:19:47 -0700487 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
488 size);
Nick Piggin89699602011-03-22 16:30:36 -0700489 kfree(va);
490 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700491}
492
Chris Wilson4da56b92016-04-04 14:46:42 +0100493int register_vmap_purge_notifier(struct notifier_block *nb)
494{
495 return blocking_notifier_chain_register(&vmap_notify_list, nb);
496}
497EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
498
499int unregister_vmap_purge_notifier(struct notifier_block *nb)
500{
501 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
502}
503EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
504
Nick Piggindb64fe02008-10-18 20:27:03 -0700505static void __free_vmap_area(struct vmap_area *va)
506{
507 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700508
509 if (free_vmap_cache) {
510 if (va->va_end < cached_vstart) {
511 free_vmap_cache = NULL;
512 } else {
513 struct vmap_area *cache;
514 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
515 if (va->va_start <= cache->va_start) {
516 free_vmap_cache = rb_prev(&va->rb_node);
517 /*
518 * We don't try to update cached_hole_size or
519 * cached_align, but it won't go very wrong.
520 */
521 }
522 }
523 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700524 rb_erase(&va->rb_node, &vmap_area_root);
525 RB_CLEAR_NODE(&va->rb_node);
526 list_del_rcu(&va->list);
527
Tejun Heoca23e402009-08-14 15:00:52 +0900528 /*
529 * Track the highest possible candidate for pcpu area
530 * allocation. Areas outside of vmalloc area can be returned
531 * here too, consider only end addresses which fall inside
532 * vmalloc area proper.
533 */
534 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
535 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
536
Lai Jiangshan14769de2011-03-18 12:12:19 +0800537 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700538}
539
540/*
541 * Free a region of KVA allocated by alloc_vmap_area
542 */
543static void free_vmap_area(struct vmap_area *va)
544{
545 spin_lock(&vmap_area_lock);
546 __free_vmap_area(va);
547 spin_unlock(&vmap_area_lock);
548}
549
550/*
551 * Clear the pagetable entries of a given vmap_area
552 */
553static void unmap_vmap_area(struct vmap_area *va)
554{
555 vunmap_page_range(va->va_start, va->va_end);
556}
557
Nick Piggincd528582009-01-06 14:39:20 -0800558static void vmap_debug_free_range(unsigned long start, unsigned long end)
559{
560 /*
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700561 * Unmap page tables and force a TLB flush immediately if pagealloc
562 * debugging is enabled. This catches use after free bugs similarly to
563 * those in linear kernel virtual address space after a page has been
564 * freed.
Nick Piggincd528582009-01-06 14:39:20 -0800565 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700566 * All the lazy freeing logic is still retained, in order to minimise
567 * intrusiveness of this debugging feature.
Nick Piggincd528582009-01-06 14:39:20 -0800568 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700569 * This is going to be *slow* (linear kernel virtual address debugging
570 * doesn't do a broadcast TLB flush so it is a lot faster).
Nick Piggincd528582009-01-06 14:39:20 -0800571 */
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700572 if (debug_pagealloc_enabled()) {
573 vunmap_page_range(start, end);
574 flush_tlb_kernel_range(start, end);
575 }
Nick Piggincd528582009-01-06 14:39:20 -0800576}
577
Nick Piggindb64fe02008-10-18 20:27:03 -0700578/*
579 * lazy_max_pages is the maximum amount of virtual address space we gather up
580 * before attempting to purge with a TLB flush.
581 *
582 * There is a tradeoff here: a larger number will cover more kernel page tables
583 * and take slightly longer to purge, but it will linearly reduce the number of
584 * global TLB flushes that must be performed. It would seem natural to scale
585 * this number up linearly with the number of CPUs (because vmapping activity
586 * could also scale linearly with the number of CPUs), however it is likely
587 * that in practice, workloads might be constrained in other ways that mean
588 * vmap activity will not scale linearly with CPUs. Also, I want to be
589 * conservative and not introduce a big latency on huge systems, so go with
590 * a less aggressive log scale. It will still be an improvement over the old
591 * code, and it will be simple to change the scale factor if we find that it
592 * becomes a problem on bigger systems.
593 */
594static unsigned long lazy_max_pages(void)
595{
596 unsigned int log;
597
598 log = fls(num_online_cpus());
599
600 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
601}
602
603static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
604
Nick Piggin02b709d2010-02-01 22:25:57 +1100605/* for per-CPU blocks */
606static void purge_fragmented_blocks_allcpus(void);
607
Nick Piggindb64fe02008-10-18 20:27:03 -0700608/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500609 * called before a call to iounmap() if the caller wants vm_area_struct's
610 * immediately freed.
611 */
612void set_iounmap_nonlazy(void)
613{
614 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
615}
616
617/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700618 * Purges all lazily-freed vmap areas.
619 *
620 * If sync is 0 then don't purge if there is already a purge in progress.
621 * If force_flush is 1, then flush kernel TLBs between *start and *end even
622 * if we found no lazy vmap areas to unmap (callers can use this to optimise
623 * their own TLB flushing).
624 * Returns with *start = min(*start, lowest purged address)
625 * *end = max(*end, highest purged address)
626 */
627static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
628 int sync, int force_flush)
629{
Andrew Morton46666d82009-01-15 13:51:15 -0800630 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700631 LIST_HEAD(valist);
632 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800633 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700634 int nr = 0;
635
636 /*
637 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
638 * should not expect such behaviour. This just simplifies locking for
639 * the case that isn't actually used at the moment anyway.
640 */
641 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800642 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700643 return;
644 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800645 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700646
Nick Piggin02b709d2010-02-01 22:25:57 +1100647 if (sync)
648 purge_fragmented_blocks_allcpus();
649
Nick Piggindb64fe02008-10-18 20:27:03 -0700650 rcu_read_lock();
651 list_for_each_entry_rcu(va, &vmap_area_list, list) {
652 if (va->flags & VM_LAZY_FREE) {
653 if (va->va_start < *start)
654 *start = va->va_start;
655 if (va->va_end > *end)
656 *end = va->va_end;
657 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700658 list_add_tail(&va->purge_list, &valist);
659 va->flags |= VM_LAZY_FREEING;
660 va->flags &= ~VM_LAZY_FREE;
661 }
662 }
663 rcu_read_unlock();
664
Yongseok Koh88f50042010-01-19 17:33:49 +0900665 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700666 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700667
668 if (nr || force_flush)
669 flush_tlb_kernel_range(*start, *end);
670
671 if (nr) {
672 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800673 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700674 __free_vmap_area(va);
675 spin_unlock(&vmap_area_lock);
676 }
Andrew Morton46666d82009-01-15 13:51:15 -0800677 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700678}
679
680/*
Nick Piggin496850e2008-11-19 15:36:33 -0800681 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
682 * is already purging.
683 */
684static void try_purge_vmap_area_lazy(void)
685{
686 unsigned long start = ULONG_MAX, end = 0;
687
688 __purge_vmap_area_lazy(&start, &end, 0, 0);
689}
690
691/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700692 * Kick off a purge of the outstanding lazy areas.
693 */
694static void purge_vmap_area_lazy(void)
695{
696 unsigned long start = ULONG_MAX, end = 0;
697
Nick Piggin496850e2008-11-19 15:36:33 -0800698 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700699}
700
701/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800702 * Free a vmap area, caller ensuring that the area has been unmapped
703 * and flush_cache_vunmap had been called for the correct range
704 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700705 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800706static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700707{
708 va->flags |= VM_LAZY_FREE;
709 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
710 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800711 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700712}
713
Nick Pigginb29acbd2008-12-01 13:13:47 -0800714/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800715 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
716 * called for the correct range previously.
717 */
718static void free_unmap_vmap_area_noflush(struct vmap_area *va)
719{
720 unmap_vmap_area(va);
721 free_vmap_area_noflush(va);
722}
723
724/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800725 * Free and unmap a vmap area
726 */
727static void free_unmap_vmap_area(struct vmap_area *va)
728{
729 flush_cache_vunmap(va->va_start, va->va_end);
730 free_unmap_vmap_area_noflush(va);
731}
732
Nick Piggindb64fe02008-10-18 20:27:03 -0700733static struct vmap_area *find_vmap_area(unsigned long addr)
734{
735 struct vmap_area *va;
736
737 spin_lock(&vmap_area_lock);
738 va = __find_vmap_area(addr);
739 spin_unlock(&vmap_area_lock);
740
741 return va;
742}
743
744static void free_unmap_vmap_area_addr(unsigned long addr)
745{
746 struct vmap_area *va;
747
748 va = find_vmap_area(addr);
749 BUG_ON(!va);
750 free_unmap_vmap_area(va);
751}
752
753
754/*** Per cpu kva allocator ***/
755
756/*
757 * vmap space is limited especially on 32 bit architectures. Ensure there is
758 * room for at least 16 percpu vmap blocks per CPU.
759 */
760/*
761 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
762 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
763 * instead (we just need a rough idea)
764 */
765#if BITS_PER_LONG == 32
766#define VMALLOC_SPACE (128UL*1024*1024)
767#else
768#define VMALLOC_SPACE (128UL*1024*1024*1024)
769#endif
770
771#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
772#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
773#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
774#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
775#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
776#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200777#define VMAP_BBMAP_BITS \
778 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
779 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
780 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700781
782#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
783
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100784static bool vmap_initialized __read_mostly = false;
785
Nick Piggindb64fe02008-10-18 20:27:03 -0700786struct vmap_block_queue {
787 spinlock_t lock;
788 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700789};
790
791struct vmap_block {
792 spinlock_t lock;
793 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700794 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700795 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100796 struct list_head free_list;
797 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100798 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700799};
800
801/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
802static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
803
804/*
805 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
806 * in the free path. Could get rid of this if we change the API to return a
807 * "cookie" from alloc, to be passed to free. But no big deal yet.
808 */
809static DEFINE_SPINLOCK(vmap_block_tree_lock);
810static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
811
812/*
813 * We should probably have a fallback mechanism to allocate virtual memory
814 * out of partially filled vmap blocks. However vmap block sizing should be
815 * fairly reasonable according to the vmalloc size, so it shouldn't be a
816 * big problem.
817 */
818
819static unsigned long addr_to_vb_idx(unsigned long addr)
820{
821 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
822 addr /= VMAP_BLOCK_SIZE;
823 return addr;
824}
825
Roman Pencf725ce2015-04-15 16:13:52 -0700826static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
827{
828 unsigned long addr;
829
830 addr = va_start + (pages_off << PAGE_SHIFT);
831 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
832 return (void *)addr;
833}
834
835/**
836 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
837 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
838 * @order: how many 2^order pages should be occupied in newly allocated block
839 * @gfp_mask: flags for the page level allocator
840 *
841 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
842 */
843static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700844{
845 struct vmap_block_queue *vbq;
846 struct vmap_block *vb;
847 struct vmap_area *va;
848 unsigned long vb_idx;
849 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700850 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700851
852 node = numa_node_id();
853
854 vb = kmalloc_node(sizeof(struct vmap_block),
855 gfp_mask & GFP_RECLAIM_MASK, node);
856 if (unlikely(!vb))
857 return ERR_PTR(-ENOMEM);
858
859 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
860 VMALLOC_START, VMALLOC_END,
861 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800862 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700863 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700864 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700865 }
866
867 err = radix_tree_preload(gfp_mask);
868 if (unlikely(err)) {
869 kfree(vb);
870 free_vmap_area(va);
871 return ERR_PTR(err);
872 }
873
Roman Pencf725ce2015-04-15 16:13:52 -0700874 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700875 spin_lock_init(&vb->lock);
876 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700877 /* At least something should be left free */
878 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
879 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700880 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700881 vb->dirty_min = VMAP_BBMAP_BITS;
882 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700883 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700884
885 vb_idx = addr_to_vb_idx(va->va_start);
886 spin_lock(&vmap_block_tree_lock);
887 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
888 spin_unlock(&vmap_block_tree_lock);
889 BUG_ON(err);
890 radix_tree_preload_end();
891
892 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700893 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700894 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700895 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900896 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700897
Roman Pencf725ce2015-04-15 16:13:52 -0700898 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700899}
900
Nick Piggindb64fe02008-10-18 20:27:03 -0700901static void free_vmap_block(struct vmap_block *vb)
902{
903 struct vmap_block *tmp;
904 unsigned long vb_idx;
905
Nick Piggindb64fe02008-10-18 20:27:03 -0700906 vb_idx = addr_to_vb_idx(vb->va->va_start);
907 spin_lock(&vmap_block_tree_lock);
908 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
909 spin_unlock(&vmap_block_tree_lock);
910 BUG_ON(tmp != vb);
911
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800912 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800913 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700914}
915
Nick Piggin02b709d2010-02-01 22:25:57 +1100916static void purge_fragmented_blocks(int cpu)
917{
918 LIST_HEAD(purge);
919 struct vmap_block *vb;
920 struct vmap_block *n_vb;
921 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
922
923 rcu_read_lock();
924 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
925
926 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
927 continue;
928
929 spin_lock(&vb->lock);
930 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
931 vb->free = 0; /* prevent further allocs after releasing lock */
932 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700933 vb->dirty_min = 0;
934 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100935 spin_lock(&vbq->lock);
936 list_del_rcu(&vb->free_list);
937 spin_unlock(&vbq->lock);
938 spin_unlock(&vb->lock);
939 list_add_tail(&vb->purge, &purge);
940 } else
941 spin_unlock(&vb->lock);
942 }
943 rcu_read_unlock();
944
945 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
946 list_del(&vb->purge);
947 free_vmap_block(vb);
948 }
949}
950
Nick Piggin02b709d2010-02-01 22:25:57 +1100951static void purge_fragmented_blocks_allcpus(void)
952{
953 int cpu;
954
955 for_each_possible_cpu(cpu)
956 purge_fragmented_blocks(cpu);
957}
958
Nick Piggindb64fe02008-10-18 20:27:03 -0700959static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
960{
961 struct vmap_block_queue *vbq;
962 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700963 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700964 unsigned int order;
965
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800966 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700967 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700968 if (WARN_ON(size == 0)) {
969 /*
970 * Allocating 0 bytes isn't what caller wants since
971 * get_order(0) returns funny result. Just warn and terminate
972 * early.
973 */
974 return NULL;
975 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700976 order = get_order(size);
977
Nick Piggindb64fe02008-10-18 20:27:03 -0700978 rcu_read_lock();
979 vbq = &get_cpu_var(vmap_block_queue);
980 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -0700981 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -0700982
983 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -0700984 if (vb->free < (1UL << order)) {
985 spin_unlock(&vb->lock);
986 continue;
987 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100988
Roman Pencf725ce2015-04-15 16:13:52 -0700989 pages_off = VMAP_BBMAP_BITS - vb->free;
990 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +1100991 vb->free -= 1UL << order;
992 if (vb->free == 0) {
993 spin_lock(&vbq->lock);
994 list_del_rcu(&vb->free_list);
995 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700996 }
Roman Pencf725ce2015-04-15 16:13:52 -0700997
Nick Piggindb64fe02008-10-18 20:27:03 -0700998 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100999 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001000 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001001
Tejun Heo3f04ba82009-10-29 22:34:12 +09001002 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001003 rcu_read_unlock();
1004
Roman Pencf725ce2015-04-15 16:13:52 -07001005 /* Allocate new block if nothing was found */
1006 if (!vaddr)
1007 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001008
Roman Pencf725ce2015-04-15 16:13:52 -07001009 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001010}
1011
1012static void vb_free(const void *addr, unsigned long size)
1013{
1014 unsigned long offset;
1015 unsigned long vb_idx;
1016 unsigned int order;
1017 struct vmap_block *vb;
1018
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001019 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001020 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001021
1022 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1023
Nick Piggindb64fe02008-10-18 20:27:03 -07001024 order = get_order(size);
1025
1026 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001027 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001028
1029 vb_idx = addr_to_vb_idx((unsigned long)addr);
1030 rcu_read_lock();
1031 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1032 rcu_read_unlock();
1033 BUG_ON(!vb);
1034
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001035 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1036
Nick Piggindb64fe02008-10-18 20:27:03 -07001037 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001038
1039 /* Expand dirty range */
1040 vb->dirty_min = min(vb->dirty_min, offset);
1041 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001042
Nick Piggindb64fe02008-10-18 20:27:03 -07001043 vb->dirty += 1UL << order;
1044 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001045 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001046 spin_unlock(&vb->lock);
1047 free_vmap_block(vb);
1048 } else
1049 spin_unlock(&vb->lock);
1050}
1051
1052/**
1053 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1054 *
1055 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1056 * to amortize TLB flushing overheads. What this means is that any page you
1057 * have now, may, in a former life, have been mapped into kernel virtual
1058 * address by the vmap layer and so there might be some CPUs with TLB entries
1059 * still referencing that page (additional to the regular 1:1 kernel mapping).
1060 *
1061 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1062 * be sure that none of the pages we have control over will have any aliases
1063 * from the vmap layer.
1064 */
1065void vm_unmap_aliases(void)
1066{
1067 unsigned long start = ULONG_MAX, end = 0;
1068 int cpu;
1069 int flush = 0;
1070
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001071 if (unlikely(!vmap_initialized))
1072 return;
1073
Nick Piggindb64fe02008-10-18 20:27:03 -07001074 for_each_possible_cpu(cpu) {
1075 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1076 struct vmap_block *vb;
1077
1078 rcu_read_lock();
1079 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001080 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001081 if (vb->dirty) {
1082 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001083 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001084
Roman Pen7d61bfe2015-04-15 16:13:55 -07001085 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1086 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001087
Roman Pen7d61bfe2015-04-15 16:13:55 -07001088 start = min(s, start);
1089 end = max(e, end);
1090
Nick Piggindb64fe02008-10-18 20:27:03 -07001091 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001092 }
1093 spin_unlock(&vb->lock);
1094 }
1095 rcu_read_unlock();
1096 }
1097
1098 __purge_vmap_area_lazy(&start, &end, 1, flush);
1099}
1100EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1101
1102/**
1103 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1104 * @mem: the pointer returned by vm_map_ram
1105 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1106 */
1107void vm_unmap_ram(const void *mem, unsigned int count)
1108{
1109 unsigned long size = count << PAGE_SHIFT;
1110 unsigned long addr = (unsigned long)mem;
1111
1112 BUG_ON(!addr);
1113 BUG_ON(addr < VMALLOC_START);
1114 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001115 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001116
1117 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001118 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001119
1120 if (likely(count <= VMAP_MAX_ALLOC))
1121 vb_free(mem, size);
1122 else
1123 free_unmap_vmap_area_addr(addr);
1124}
1125EXPORT_SYMBOL(vm_unmap_ram);
1126
1127/**
1128 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1129 * @pages: an array of pointers to the pages to be mapped
1130 * @count: number of pages
1131 * @node: prefer to allocate data structures on this node
1132 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001133 *
Gioh Kim36437632014-04-07 15:37:37 -07001134 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1135 * faster than vmap so it's good. But if you mix long-life and short-life
1136 * objects with vm_map_ram(), it could consume lots of address space through
1137 * fragmentation (especially on a 32bit machine). You could see failures in
1138 * the end. Please use this function for short-lived objects.
1139 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001140 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001141 */
1142void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1143{
1144 unsigned long size = count << PAGE_SHIFT;
1145 unsigned long addr;
1146 void *mem;
1147
1148 if (likely(count <= VMAP_MAX_ALLOC)) {
1149 mem = vb_alloc(size, GFP_KERNEL);
1150 if (IS_ERR(mem))
1151 return NULL;
1152 addr = (unsigned long)mem;
1153 } else {
1154 struct vmap_area *va;
1155 va = alloc_vmap_area(size, PAGE_SIZE,
1156 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1157 if (IS_ERR(va))
1158 return NULL;
1159
1160 addr = va->va_start;
1161 mem = (void *)addr;
1162 }
1163 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1164 vm_unmap_ram(mem, count);
1165 return NULL;
1166 }
1167 return mem;
1168}
1169EXPORT_SYMBOL(vm_map_ram);
1170
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001171static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001172/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001173 * vm_area_add_early - add vmap area early during boot
1174 * @vm: vm_struct to add
1175 *
1176 * This function is used to add fixed kernel vm area to vmlist before
1177 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1178 * should contain proper values and the other fields should be zero.
1179 *
1180 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1181 */
1182void __init vm_area_add_early(struct vm_struct *vm)
1183{
1184 struct vm_struct *tmp, **p;
1185
1186 BUG_ON(vmap_initialized);
1187 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1188 if (tmp->addr >= vm->addr) {
1189 BUG_ON(tmp->addr < vm->addr + vm->size);
1190 break;
1191 } else
1192 BUG_ON(tmp->addr + tmp->size > vm->addr);
1193 }
1194 vm->next = *p;
1195 *p = vm;
1196}
1197
1198/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001199 * vm_area_register_early - register vmap area early during boot
1200 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001201 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001202 *
1203 * This function is used to register kernel vm area before
1204 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1205 * proper values on entry and other fields should be zero. On return,
1206 * vm->addr contains the allocated address.
1207 *
1208 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1209 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001210void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001211{
1212 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001213 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001214
Tejun Heoc0c0a292009-02-24 11:57:21 +09001215 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1216 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1217
1218 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001219
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001220 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001221}
1222
Nick Piggindb64fe02008-10-18 20:27:03 -07001223void __init vmalloc_init(void)
1224{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001225 struct vmap_area *va;
1226 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001227 int i;
1228
1229 for_each_possible_cpu(i) {
1230 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001231 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001232
1233 vbq = &per_cpu(vmap_block_queue, i);
1234 spin_lock_init(&vbq->lock);
1235 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001236 p = &per_cpu(vfree_deferred, i);
1237 init_llist_head(&p->list);
1238 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001239 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001240
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001241 /* Import existing vmlist entries. */
1242 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001243 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001244 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001245 va->va_start = (unsigned long)tmp->addr;
1246 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001247 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001248 __insert_vmap_area(va);
1249 }
Tejun Heoca23e402009-08-14 15:00:52 +09001250
1251 vmap_area_pcpu_hole = VMALLOC_END;
1252
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001253 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001254}
1255
Tejun Heo8fc48982009-02-20 16:29:08 +09001256/**
1257 * map_kernel_range_noflush - map kernel VM area with the specified pages
1258 * @addr: start of the VM area to map
1259 * @size: size of the VM area to map
1260 * @prot: page protection flags to use
1261 * @pages: pages to map
1262 *
1263 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1264 * specify should have been allocated using get_vm_area() and its
1265 * friends.
1266 *
1267 * NOTE:
1268 * This function does NOT do any cache flushing. The caller is
1269 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1270 * before calling this function.
1271 *
1272 * RETURNS:
1273 * The number of pages mapped on success, -errno on failure.
1274 */
1275int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1276 pgprot_t prot, struct page **pages)
1277{
1278 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1279}
1280
1281/**
1282 * unmap_kernel_range_noflush - unmap kernel VM area
1283 * @addr: start of the VM area to unmap
1284 * @size: size of the VM area to unmap
1285 *
1286 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1287 * specify should have been allocated using get_vm_area() and its
1288 * friends.
1289 *
1290 * NOTE:
1291 * This function does NOT do any cache flushing. The caller is
1292 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1293 * before calling this function and flush_tlb_kernel_range() after.
1294 */
1295void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1296{
1297 vunmap_page_range(addr, addr + size);
1298}
Huang Ying81e88fd2011-01-12 14:44:55 +08001299EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001300
1301/**
1302 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1303 * @addr: start of the VM area to unmap
1304 * @size: size of the VM area to unmap
1305 *
1306 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1307 * the unmapping and tlb after.
1308 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001309void unmap_kernel_range(unsigned long addr, unsigned long size)
1310{
1311 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001312
1313 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001314 vunmap_page_range(addr, end);
1315 flush_tlb_kernel_range(addr, end);
1316}
Minchan Kim93ef6d62014-06-04 16:11:09 -07001317EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001318
WANG Chaof6f8ed42014-08-06 16:06:58 -07001319int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001320{
1321 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001322 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001323 int err;
1324
WANG Chaof6f8ed42014-08-06 16:06:58 -07001325 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001326
WANG Chaof6f8ed42014-08-06 16:06:58 -07001327 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001328}
1329EXPORT_SYMBOL_GPL(map_vm_area);
1330
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001331static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001332 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001333{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001334 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001335 vm->flags = flags;
1336 vm->addr = (void *)va->va_start;
1337 vm->size = va->va_end - va->va_start;
1338 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001339 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001340 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001341 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001342}
Tejun Heocf88c792009-08-14 15:00:52 +09001343
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001344static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001345{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001346 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001347 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001348 * we should make sure that vm has proper values.
1349 * Pair with smp_rmb() in show_numa_info().
1350 */
1351 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001352 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001353}
1354
Nick Piggindb64fe02008-10-18 20:27:03 -07001355static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001356 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001357 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001358{
Kautuk Consul00065262011-12-19 17:12:04 -08001359 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001360 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001362 BUG_ON(in_interrupt());
Zhang Yanfei0f2d4a82013-07-03 15:04:50 -07001363 if (flags & VM_IOREMAP)
Toshi Kani0f616be2015-04-14 15:47:17 -07001364 align = 1ul << clamp_t(int, fls_long(size),
1365 PAGE_SHIFT, IOREMAP_MAX_ORDER);
Nick Piggindb64fe02008-10-18 20:27:03 -07001366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001368 if (unlikely(!size))
1369 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Tejun Heocf88c792009-08-14 15:00:52 +09001371 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (unlikely(!area))
1373 return NULL;
1374
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001375 if (!(flags & VM_NO_GUARD))
1376 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Nick Piggindb64fe02008-10-18 20:27:03 -07001378 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1379 if (IS_ERR(va)) {
1380 kfree(area);
1381 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001384 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
Christoph Lameter930fc452005-10-29 18:15:41 -07001389struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1390 unsigned long start, unsigned long end)
1391{
David Rientjes00ef2d22013-02-22 16:35:36 -08001392 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1393 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001394}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001395EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001396
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001397struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1398 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001399 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001400{
David Rientjes00ef2d22013-02-22 16:35:36 -08001401 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1402 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001403}
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405/**
Simon Arlott183ff222007-10-20 01:27:18 +02001406 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 * @size: size of the area
1408 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1409 *
1410 * Search an area of @size in the kernel virtual mapping area,
1411 * and reserved it for out purposes. Returns the area descriptor
1412 * on success or %NULL on failure.
1413 */
1414struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1415{
David Miller2dca6992009-09-21 12:22:34 -07001416 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001417 NUMA_NO_NODE, GFP_KERNEL,
1418 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001419}
1420
1421struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001422 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001423{
David Miller2dca6992009-09-21 12:22:34 -07001424 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001425 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426}
1427
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001428/**
1429 * find_vm_area - find a continuous kernel virtual area
1430 * @addr: base address
1431 *
1432 * Search for the kernel VM area starting at @addr, and return it.
1433 * It is up to the caller to do all required locking to keep the returned
1434 * pointer valid.
1435 */
1436struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001437{
Nick Piggindb64fe02008-10-18 20:27:03 -07001438 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001439
Nick Piggindb64fe02008-10-18 20:27:03 -07001440 va = find_vmap_area((unsigned long)addr);
1441 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001442 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001443
Andi Kleen7856dfe2005-05-20 14:27:57 -07001444 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001445}
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447/**
Simon Arlott183ff222007-10-20 01:27:18 +02001448 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 * @addr: base address
1450 *
1451 * Search for the kernel VM area starting at @addr, and remove it.
1452 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001453 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001455struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
Nick Piggindb64fe02008-10-18 20:27:03 -07001457 struct vmap_area *va;
1458
1459 va = find_vmap_area((unsigned long)addr);
1460 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001461 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001462
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001463 spin_lock(&vmap_area_lock);
1464 va->vm = NULL;
1465 va->flags &= ~VM_VM_AREA;
1466 spin_unlock(&vmap_area_lock);
1467
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001468 vmap_debug_free_range(va->va_start, va->va_end);
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001469 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001470 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001471
Nick Piggindb64fe02008-10-18 20:27:03 -07001472 return vm;
1473 }
1474 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475}
1476
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001477static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
1479 struct vm_struct *area;
1480
1481 if (!addr)
1482 return;
1483
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001484 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001485 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 area = remove_vm_area(addr);
1489 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001490 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 return;
1493 }
1494
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001495 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1496 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 if (deallocate_pages) {
1499 int i;
1500
1501 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001502 struct page *page = area->pages[i];
1503
1504 BUG_ON(!page);
Vladimir Davydov37f08dd2016-01-14 15:18:18 -08001505 __free_kmem_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
1507
David Rientjes244d63e2016-01-14 15:19:35 -08001508 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 }
1510
1511 kfree(area);
1512 return;
1513}
Al Viro32fcfd42013-03-10 20:14:08 -04001514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515/**
1516 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 * @addr: memory base address
1518 *
Simon Arlott183ff222007-10-20 01:27:18 +02001519 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001520 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1521 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 *
Al Viro32fcfd42013-03-10 20:14:08 -04001523 * Must not be called in NMI context (strictly speaking, only if we don't
1524 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1525 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001526 *
1527 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001529void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
Al Viro32fcfd42013-03-10 20:14:08 -04001531 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001532
1533 kmemleak_free(addr);
1534
Al Viro32fcfd42013-03-10 20:14:08 -04001535 if (!addr)
1536 return;
1537 if (unlikely(in_interrupt())) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001538 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
Oleg Nesterov59d31322013-07-08 16:00:08 -07001539 if (llist_add((struct llist_node *)addr, &p->list))
1540 schedule_work(&p->wq);
Al Viro32fcfd42013-03-10 20:14:08 -04001541 } else
1542 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544EXPORT_SYMBOL(vfree);
1545
1546/**
1547 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 * @addr: memory base address
1549 *
1550 * Free the virtually contiguous memory area starting at @addr,
1551 * which was created from the page array passed to vmap().
1552 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001553 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001555void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556{
1557 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001558 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001559 if (addr)
1560 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562EXPORT_SYMBOL(vunmap);
1563
1564/**
1565 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 * @pages: array of page pointers
1567 * @count: number of pages to map
1568 * @flags: vm_area->flags
1569 * @prot: page protection for the mapping
1570 *
1571 * Maps @count pages from @pages into contiguous kernel virtual
1572 * space.
1573 */
1574void *vmap(struct page **pages, unsigned int count,
1575 unsigned long flags, pgprot_t prot)
1576{
1577 struct vm_struct *area;
1578
Peter Zijlstra34754b62009-02-25 16:04:03 +01001579 might_sleep();
1580
Jan Beulich44813742009-09-21 17:03:05 -07001581 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 return NULL;
1583
Christoph Lameter23016962008-04-28 02:12:42 -07001584 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1585 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (!area)
1587 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001588
WANG Chaof6f8ed42014-08-06 16:06:58 -07001589 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 vunmap(area->addr);
1591 return NULL;
1592 }
1593
1594 return area->addr;
1595}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596EXPORT_SYMBOL(vmap);
1597
David Miller2dca6992009-09-21 12:22:34 -07001598static void *__vmalloc_node(unsigned long size, unsigned long align,
1599 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001600 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001601static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001602 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
Dave Hansen22943ab2011-05-24 17:12:18 -07001604 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 struct page **pages;
1606 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001607 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1608 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Wanpeng Li762216a2013-09-11 14:22:42 -07001610 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 array_size = (nr_pages * sizeof(struct page *));
1612
1613 area->nr_pages = nr_pages;
1614 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001615 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001616 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001617 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001618 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001619 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 area->pages = pages;
1622 if (!area->pages) {
1623 remove_vm_area(area->addr);
1624 kfree(area);
1625 return NULL;
1626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001629 struct page *page;
1630
Jianguo Wu4b909512013-11-12 15:07:11 -08001631 if (node == NUMA_NO_NODE)
Vladimir Davydov37f08dd2016-01-14 15:18:18 -08001632 page = alloc_kmem_pages(alloc_mask, order);
Christoph Lameter930fc452005-10-29 18:15:41 -07001633 else
Vladimir Davydov37f08dd2016-01-14 15:18:18 -08001634 page = alloc_kmem_pages_node(node, alloc_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001635
1636 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 /* Successfully allocated i pages, free them in __vunmap() */
1638 area->nr_pages = i;
1639 goto fail;
1640 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001641 area->pages[i] = page;
Mel Gormand0164ad2015-11-06 16:28:21 -08001642 if (gfpflags_allow_blocking(gfp_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001643 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 }
1645
WANG Chaof6f8ed42014-08-06 16:06:58 -07001646 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 goto fail;
1648 return area->addr;
1649
1650fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001651 warn_alloc_failed(gfp_mask, order,
1652 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001653 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 vfree(area->addr);
1655 return NULL;
1656}
1657
David Rientjesd0a21262011-01-13 15:46:02 -08001658/**
1659 * __vmalloc_node_range - allocate virtually contiguous memory
1660 * @size: allocation size
1661 * @align: desired alignment
1662 * @start: vm area range start
1663 * @end: vm area range end
1664 * @gfp_mask: flags for the page level allocator
1665 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001666 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001667 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001668 * @caller: caller's return address
1669 *
1670 * Allocate enough pages to cover @size from the page level
1671 * allocator with @gfp_mask flags. Map them into contiguous
1672 * kernel virtual space, using a pagetable protection of @prot.
1673 */
1674void *__vmalloc_node_range(unsigned long size, unsigned long align,
1675 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001676 pgprot_t prot, unsigned long vm_flags, int node,
1677 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001678{
David Rientjesd0a21262011-01-13 15:46:02 -08001679 struct vm_struct *area;
1680 void *addr;
1681 unsigned long real_size = size;
1682
1683 size = PAGE_ALIGN(size);
1684 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001685 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001686
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001687 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1688 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001689 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001690 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001691
Wanpeng Li3722e132013-11-12 15:07:29 -08001692 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001693 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001694 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001695
1696 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001697 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1698 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001699 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001700 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001701 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001702
1703 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001704 * A ref_count = 2 is needed because vm_struct allocated in
1705 * __get_vm_area_node() contains a reference to the virtual address of
1706 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001707 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001708 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001709
1710 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001711
1712fail:
1713 warn_alloc_failed(gfp_mask, 0,
1714 "vmalloc: allocation failure: %lu bytes\n",
1715 real_size);
1716 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001717}
1718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001720 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001722 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 * @gfp_mask: flags for the page level allocator
1724 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001725 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001726 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 *
1728 * Allocate enough pages to cover @size from the page level
1729 * allocator with @gfp_mask flags. Map them into contiguous
1730 * kernel virtual space, using a pagetable protection of @prot.
1731 */
David Miller2dca6992009-09-21 12:22:34 -07001732static void *__vmalloc_node(unsigned long size, unsigned long align,
1733 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001734 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735{
David Rientjesd0a21262011-01-13 15:46:02 -08001736 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001737 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
1739
Christoph Lameter930fc452005-10-29 18:15:41 -07001740void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1741{
David Rientjes00ef2d22013-02-22 16:35:36 -08001742 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001743 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001744}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745EXPORT_SYMBOL(__vmalloc);
1746
Dave Younge1ca7782010-10-26 14:22:06 -07001747static inline void *__vmalloc_node_flags(unsigned long size,
1748 int node, gfp_t flags)
1749{
1750 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1751 node, __builtin_return_address(0));
1752}
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754/**
1755 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 * Allocate enough pages to cover @size from the page level
1758 * allocator and map them into contiguous kernel virtual space.
1759 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001760 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 * use __vmalloc() instead.
1762 */
1763void *vmalloc(unsigned long size)
1764{
David Rientjes00ef2d22013-02-22 16:35:36 -08001765 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1766 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768EXPORT_SYMBOL(vmalloc);
1769
Christoph Lameter930fc452005-10-29 18:15:41 -07001770/**
Dave Younge1ca7782010-10-26 14:22:06 -07001771 * vzalloc - allocate virtually contiguous memory with zero fill
1772 * @size: allocation size
1773 * Allocate enough pages to cover @size from the page level
1774 * allocator and map them into contiguous kernel virtual space.
1775 * The memory allocated is set to zero.
1776 *
1777 * For tight control over page level allocator and protection flags
1778 * use __vmalloc() instead.
1779 */
1780void *vzalloc(unsigned long size)
1781{
David Rientjes00ef2d22013-02-22 16:35:36 -08001782 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001783 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1784}
1785EXPORT_SYMBOL(vzalloc);
1786
1787/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001788 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1789 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001790 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001791 * The resulting memory area is zeroed so it can be mapped to userspace
1792 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001793 */
1794void *vmalloc_user(unsigned long size)
1795{
1796 struct vm_struct *area;
1797 void *ret;
1798
David Miller2dca6992009-09-21 12:22:34 -07001799 ret = __vmalloc_node(size, SHMLBA,
1800 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001801 PAGE_KERNEL, NUMA_NO_NODE,
1802 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001803 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001804 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001805 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001806 }
Nick Piggin83342312006-06-23 02:03:20 -07001807 return ret;
1808}
1809EXPORT_SYMBOL(vmalloc_user);
1810
1811/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001812 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001813 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001814 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001815 *
1816 * Allocate enough pages to cover @size from the page level
1817 * allocator and map them into contiguous kernel virtual space.
1818 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001819 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001820 * use __vmalloc() instead.
1821 */
1822void *vmalloc_node(unsigned long size, int node)
1823{
David Miller2dca6992009-09-21 12:22:34 -07001824 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001825 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001826}
1827EXPORT_SYMBOL(vmalloc_node);
1828
Dave Younge1ca7782010-10-26 14:22:06 -07001829/**
1830 * vzalloc_node - allocate memory on a specific node with zero fill
1831 * @size: allocation size
1832 * @node: numa node
1833 *
1834 * Allocate enough pages to cover @size from the page level
1835 * allocator and map them into contiguous kernel virtual space.
1836 * The memory allocated is set to zero.
1837 *
1838 * For tight control over page level allocator and protection flags
1839 * use __vmalloc_node() instead.
1840 */
1841void *vzalloc_node(unsigned long size, int node)
1842{
1843 return __vmalloc_node_flags(size, node,
1844 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1845}
1846EXPORT_SYMBOL(vzalloc_node);
1847
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001848#ifndef PAGE_KERNEL_EXEC
1849# define PAGE_KERNEL_EXEC PAGE_KERNEL
1850#endif
1851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852/**
1853 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 * @size: allocation size
1855 *
1856 * Kernel-internal function to allocate enough pages to cover @size
1857 * the page level allocator and map them into contiguous and
1858 * executable kernel virtual space.
1859 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001860 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 * use __vmalloc() instead.
1862 */
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864void *vmalloc_exec(unsigned long size)
1865{
David Miller2dca6992009-09-21 12:22:34 -07001866 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001867 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868}
1869
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001870#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001871#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001872#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001873#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001874#else
1875#define GFP_VMALLOC32 GFP_KERNEL
1876#endif
1877
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878/**
1879 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 * @size: allocation size
1881 *
1882 * Allocate enough 32bit PA addressable pages to cover @size from the
1883 * page level allocator and map them into contiguous kernel virtual space.
1884 */
1885void *vmalloc_32(unsigned long size)
1886{
David Miller2dca6992009-09-21 12:22:34 -07001887 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001888 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890EXPORT_SYMBOL(vmalloc_32);
1891
Nick Piggin83342312006-06-23 02:03:20 -07001892/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001893 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001894 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001895 *
1896 * The resulting memory area is 32bit addressable and zeroed so it can be
1897 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001898 */
1899void *vmalloc_32_user(unsigned long size)
1900{
1901 struct vm_struct *area;
1902 void *ret;
1903
David Miller2dca6992009-09-21 12:22:34 -07001904 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001905 NUMA_NO_NODE, __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_32_user);
1913
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001914/*
1915 * small helper routine , copy contents to buf from addr.
1916 * If the page is not present, fill zero.
1917 */
1918
1919static int aligned_vread(char *buf, char *addr, unsigned long count)
1920{
1921 struct page *p;
1922 int copied = 0;
1923
1924 while (count) {
1925 unsigned long offset, length;
1926
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001927 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001928 length = PAGE_SIZE - offset;
1929 if (length > count)
1930 length = count;
1931 p = vmalloc_to_page(addr);
1932 /*
1933 * To do safe access to this _mapped_ area, we need
1934 * lock. But adding lock here means that we need to add
1935 * overhead of vmalloc()/vfree() calles for this _debug_
1936 * interface, rarely used. Instead of that, we'll use
1937 * kmap() and get small overhead in this access function.
1938 */
1939 if (p) {
1940 /*
1941 * we can expect USER0 is not used (see vread/vwrite's
1942 * function description)
1943 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001944 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001945 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001946 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001947 } else
1948 memset(buf, 0, length);
1949
1950 addr += length;
1951 buf += length;
1952 copied += length;
1953 count -= length;
1954 }
1955 return copied;
1956}
1957
1958static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1959{
1960 struct page *p;
1961 int copied = 0;
1962
1963 while (count) {
1964 unsigned long offset, length;
1965
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001966 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001967 length = PAGE_SIZE - offset;
1968 if (length > count)
1969 length = count;
1970 p = vmalloc_to_page(addr);
1971 /*
1972 * To do safe access to this _mapped_ area, we need
1973 * lock. But adding lock here means that we need to add
1974 * overhead of vmalloc()/vfree() calles for this _debug_
1975 * interface, rarely used. Instead of that, we'll use
1976 * kmap() and get small overhead in this access function.
1977 */
1978 if (p) {
1979 /*
1980 * we can expect USER0 is not used (see vread/vwrite's
1981 * function description)
1982 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001983 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001984 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001985 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001986 }
1987 addr += length;
1988 buf += length;
1989 copied += length;
1990 count -= length;
1991 }
1992 return copied;
1993}
1994
1995/**
1996 * vread() - read vmalloc area in a safe way.
1997 * @buf: buffer for reading data
1998 * @addr: vm address.
1999 * @count: number of bytes to be read.
2000 *
2001 * Returns # of bytes which addr and buf should be increased.
2002 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2003 * includes any intersect with alive vmalloc area.
2004 *
2005 * This function checks that addr is a valid vmalloc'ed area, and
2006 * copy data from that area to a given buffer. If the given memory range
2007 * of [addr...addr+count) includes some valid address, data is copied to
2008 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2009 * IOREMAP area is treated as memory hole and no copy is done.
2010 *
2011 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002012 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002013 *
2014 * Note: In usual ops, vread() is never necessary because the caller
2015 * should know vmalloc() area is valid and can use memcpy().
2016 * This is for routines which have to access vmalloc area without
2017 * any informaion, as /dev/kmem.
2018 *
2019 */
2020
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021long vread(char *buf, char *addr, unsigned long count)
2022{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002023 struct vmap_area *va;
2024 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002026 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 unsigned long n;
2028
2029 /* Don't allow overflow */
2030 if ((unsigned long) addr + count < count)
2031 count = -(unsigned long) addr;
2032
Joonsoo Kime81ce852013-04-29 15:07:32 -07002033 spin_lock(&vmap_area_lock);
2034 list_for_each_entry(va, &vmap_area_list, list) {
2035 if (!count)
2036 break;
2037
2038 if (!(va->flags & VM_VM_AREA))
2039 continue;
2040
2041 vm = va->vm;
2042 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002043 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 continue;
2045 while (addr < vaddr) {
2046 if (count == 0)
2047 goto finished;
2048 *buf = '\0';
2049 buf++;
2050 addr++;
2051 count--;
2052 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002053 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002054 if (n > count)
2055 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002056 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002057 aligned_vread(buf, addr, n);
2058 else /* IOREMAP area is treated as memory hole */
2059 memset(buf, 0, n);
2060 buf += n;
2061 addr += n;
2062 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
2064finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002065 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002066
2067 if (buf == buf_start)
2068 return 0;
2069 /* zero-fill memory holes */
2070 if (buf != buf_start + buflen)
2071 memset(buf, 0, buflen - (buf - buf_start));
2072
2073 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074}
2075
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002076/**
2077 * vwrite() - write vmalloc area in a safe way.
2078 * @buf: buffer for source data
2079 * @addr: vm address.
2080 * @count: number of bytes to be read.
2081 *
2082 * Returns # of bytes which addr and buf should be incresed.
2083 * (same number to @count).
2084 * If [addr...addr+count) doesn't includes any intersect with valid
2085 * vmalloc area, returns 0.
2086 *
2087 * This function checks that addr is a valid vmalloc'ed area, and
2088 * copy data from a buffer to the given addr. If specified range of
2089 * [addr...addr+count) includes some valid address, data is copied from
2090 * proper area of @buf. If there are memory holes, no copy to hole.
2091 * IOREMAP area is treated as memory hole and no copy is done.
2092 *
2093 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002094 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002095 *
2096 * Note: In usual ops, vwrite() is never necessary because the caller
2097 * should know vmalloc() area is valid and can use memcpy().
2098 * This is for routines which have to access vmalloc area without
2099 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002100 */
2101
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102long vwrite(char *buf, char *addr, unsigned long count)
2103{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002104 struct vmap_area *va;
2105 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002106 char *vaddr;
2107 unsigned long n, buflen;
2108 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
2110 /* Don't allow overflow */
2111 if ((unsigned long) addr + count < count)
2112 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002113 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Joonsoo Kime81ce852013-04-29 15:07:32 -07002115 spin_lock(&vmap_area_lock);
2116 list_for_each_entry(va, &vmap_area_list, list) {
2117 if (!count)
2118 break;
2119
2120 if (!(va->flags & VM_VM_AREA))
2121 continue;
2122
2123 vm = va->vm;
2124 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002125 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 continue;
2127 while (addr < vaddr) {
2128 if (count == 0)
2129 goto finished;
2130 buf++;
2131 addr++;
2132 count--;
2133 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002134 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002135 if (n > count)
2136 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002137 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002138 aligned_vwrite(buf, addr, n);
2139 copied++;
2140 }
2141 buf += n;
2142 addr += n;
2143 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 }
2145finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002146 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002147 if (!copied)
2148 return 0;
2149 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150}
Nick Piggin83342312006-06-23 02:03:20 -07002151
2152/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002153 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2154 * @vma: vma to cover
2155 * @uaddr: target user address to start at
2156 * @kaddr: virtual address of vmalloc kernel memory
2157 * @size: size of map area
2158 *
2159 * Returns: 0 for success, -Exxx on failure
2160 *
2161 * This function checks that @kaddr is a valid vmalloc'ed area,
2162 * and that it is big enough to cover the range starting at
2163 * @uaddr in @vma. Will return failure if that criteria isn't
2164 * met.
2165 *
2166 * Similar to remap_pfn_range() (see mm/memory.c)
2167 */
2168int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2169 void *kaddr, unsigned long size)
2170{
2171 struct vm_struct *area;
2172
2173 size = PAGE_ALIGN(size);
2174
2175 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2176 return -EINVAL;
2177
2178 area = find_vm_area(kaddr);
2179 if (!area)
2180 return -EINVAL;
2181
2182 if (!(area->flags & VM_USERMAP))
2183 return -EINVAL;
2184
2185 if (kaddr + size > area->addr + area->size)
2186 return -EINVAL;
2187
2188 do {
2189 struct page *page = vmalloc_to_page(kaddr);
2190 int ret;
2191
2192 ret = vm_insert_page(vma, uaddr, page);
2193 if (ret)
2194 return ret;
2195
2196 uaddr += PAGE_SIZE;
2197 kaddr += PAGE_SIZE;
2198 size -= PAGE_SIZE;
2199 } while (size > 0);
2200
2201 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2202
2203 return 0;
2204}
2205EXPORT_SYMBOL(remap_vmalloc_range_partial);
2206
2207/**
Nick Piggin83342312006-06-23 02:03:20 -07002208 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002209 * @vma: vma to cover (map full range of vma)
2210 * @addr: vmalloc memory
2211 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002212 *
2213 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002214 *
2215 * This function checks that addr is a valid vmalloc'ed area, and
2216 * that it is big enough to cover the vma. Will return failure if
2217 * that criteria isn't met.
2218 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002219 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002220 */
2221int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2222 unsigned long pgoff)
2223{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002224 return remap_vmalloc_range_partial(vma, vma->vm_start,
2225 addr + (pgoff << PAGE_SHIFT),
2226 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002227}
2228EXPORT_SYMBOL(remap_vmalloc_range);
2229
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002230/*
2231 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2232 * have one.
2233 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002234void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002235{
2236}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002237
2238
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002239static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002240{
David Vrabelcd129092011-09-29 16:53:32 +01002241 pte_t ***p = data;
2242
2243 if (p) {
2244 *(*p) = pte;
2245 (*p)++;
2246 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002247 return 0;
2248}
2249
2250/**
2251 * alloc_vm_area - allocate a range of kernel address space
2252 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002253 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002254 *
2255 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002256 *
2257 * This function reserves a range of kernel address space, and
2258 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002259 * are created.
2260 *
2261 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2262 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002263 */
David Vrabelcd129092011-09-29 16:53:32 +01002264struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002265{
2266 struct vm_struct *area;
2267
Christoph Lameter23016962008-04-28 02:12:42 -07002268 area = get_vm_area_caller(size, VM_IOREMAP,
2269 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002270 if (area == NULL)
2271 return NULL;
2272
2273 /*
2274 * This ensures that page tables are constructed for this region
2275 * of kernel virtual address space and mapped into init_mm.
2276 */
2277 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002278 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002279 free_vm_area(area);
2280 return NULL;
2281 }
2282
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002283 return area;
2284}
2285EXPORT_SYMBOL_GPL(alloc_vm_area);
2286
2287void free_vm_area(struct vm_struct *area)
2288{
2289 struct vm_struct *ret;
2290 ret = remove_vm_area(area->addr);
2291 BUG_ON(ret != area);
2292 kfree(area);
2293}
2294EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002295
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002296#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002297static struct vmap_area *node_to_va(struct rb_node *n)
2298{
2299 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2300}
2301
2302/**
2303 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2304 * @end: target address
2305 * @pnext: out arg for the next vmap_area
2306 * @pprev: out arg for the previous vmap_area
2307 *
2308 * Returns: %true if either or both of next and prev are found,
2309 * %false if no vmap_area exists
2310 *
2311 * Find vmap_areas end addresses of which enclose @end. ie. if not
2312 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2313 */
2314static bool pvm_find_next_prev(unsigned long end,
2315 struct vmap_area **pnext,
2316 struct vmap_area **pprev)
2317{
2318 struct rb_node *n = vmap_area_root.rb_node;
2319 struct vmap_area *va = NULL;
2320
2321 while (n) {
2322 va = rb_entry(n, struct vmap_area, rb_node);
2323 if (end < va->va_end)
2324 n = n->rb_left;
2325 else if (end > va->va_end)
2326 n = n->rb_right;
2327 else
2328 break;
2329 }
2330
2331 if (!va)
2332 return false;
2333
2334 if (va->va_end > end) {
2335 *pnext = va;
2336 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2337 } else {
2338 *pprev = va;
2339 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2340 }
2341 return true;
2342}
2343
2344/**
2345 * pvm_determine_end - find the highest aligned address between two vmap_areas
2346 * @pnext: in/out arg for the next vmap_area
2347 * @pprev: in/out arg for the previous vmap_area
2348 * @align: alignment
2349 *
2350 * Returns: determined end address
2351 *
2352 * Find the highest aligned address between *@pnext and *@pprev below
2353 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2354 * down address is between the end addresses of the two vmap_areas.
2355 *
2356 * Please note that the address returned by this function may fall
2357 * inside *@pnext vmap_area. The caller is responsible for checking
2358 * that.
2359 */
2360static unsigned long pvm_determine_end(struct vmap_area **pnext,
2361 struct vmap_area **pprev,
2362 unsigned long align)
2363{
2364 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2365 unsigned long addr;
2366
2367 if (*pnext)
2368 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2369 else
2370 addr = vmalloc_end;
2371
2372 while (*pprev && (*pprev)->va_end > addr) {
2373 *pnext = *pprev;
2374 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2375 }
2376
2377 return addr;
2378}
2379
2380/**
2381 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2382 * @offsets: array containing offset of each area
2383 * @sizes: array containing size of each area
2384 * @nr_vms: the number of areas to allocate
2385 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002386 *
2387 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2388 * vm_structs on success, %NULL on failure
2389 *
2390 * Percpu allocator wants to use congruent vm areas so that it can
2391 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002392 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2393 * be scattered pretty far, distance between two areas easily going up
2394 * to gigabytes. To avoid interacting with regular vmallocs, these
2395 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002396 *
2397 * Despite its complicated look, this allocator is rather simple. It
2398 * does everything top-down and scans areas from the end looking for
2399 * matching slot. While scanning, if any of the areas overlaps with
2400 * existing vmap_area, the base address is pulled down to fit the
2401 * area. Scanning is repeated till all the areas fit and then all
2402 * necessary data structres are inserted and the result is returned.
2403 */
2404struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2405 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002406 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002407{
2408 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2409 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2410 struct vmap_area **vas, *prev, *next;
2411 struct vm_struct **vms;
2412 int area, area2, last_area, term_area;
2413 unsigned long base, start, end, last_end;
2414 bool purged = false;
2415
Tejun Heoca23e402009-08-14 15:00:52 +09002416 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002417 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002418 for (last_area = 0, area = 0; area < nr_vms; area++) {
2419 start = offsets[area];
2420 end = start + sizes[area];
2421
2422 /* is everything aligned properly? */
2423 BUG_ON(!IS_ALIGNED(offsets[area], align));
2424 BUG_ON(!IS_ALIGNED(sizes[area], align));
2425
2426 /* detect the area with the highest address */
2427 if (start > offsets[last_area])
2428 last_area = area;
2429
2430 for (area2 = 0; area2 < nr_vms; area2++) {
2431 unsigned long start2 = offsets[area2];
2432 unsigned long end2 = start2 + sizes[area2];
2433
2434 if (area2 == area)
2435 continue;
2436
2437 BUG_ON(start2 >= start && start2 < end);
2438 BUG_ON(end2 <= end && end2 > start);
2439 }
2440 }
2441 last_end = offsets[last_area] + sizes[last_area];
2442
2443 if (vmalloc_end - vmalloc_start < last_end) {
2444 WARN_ON(true);
2445 return NULL;
2446 }
2447
Thomas Meyer4d67d862012-05-29 15:06:21 -07002448 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2449 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002450 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002451 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002452
2453 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002454 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2455 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002456 if (!vas[area] || !vms[area])
2457 goto err_free;
2458 }
2459retry:
2460 spin_lock(&vmap_area_lock);
2461
2462 /* start scanning - we scan from the top, begin with the last area */
2463 area = term_area = last_area;
2464 start = offsets[area];
2465 end = start + sizes[area];
2466
2467 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2468 base = vmalloc_end - last_end;
2469 goto found;
2470 }
2471 base = pvm_determine_end(&next, &prev, align) - end;
2472
2473 while (true) {
2474 BUG_ON(next && next->va_end <= base + end);
2475 BUG_ON(prev && prev->va_end > base + end);
2476
2477 /*
2478 * base might have underflowed, add last_end before
2479 * comparing.
2480 */
2481 if (base + last_end < vmalloc_start + last_end) {
2482 spin_unlock(&vmap_area_lock);
2483 if (!purged) {
2484 purge_vmap_area_lazy();
2485 purged = true;
2486 goto retry;
2487 }
2488 goto err_free;
2489 }
2490
2491 /*
2492 * If next overlaps, move base downwards so that it's
2493 * right below next and then recheck.
2494 */
2495 if (next && next->va_start < base + end) {
2496 base = pvm_determine_end(&next, &prev, align) - end;
2497 term_area = area;
2498 continue;
2499 }
2500
2501 /*
2502 * If prev overlaps, shift down next and prev and move
2503 * base so that it's right below new next and then
2504 * recheck.
2505 */
2506 if (prev && prev->va_end > base + start) {
2507 next = prev;
2508 prev = node_to_va(rb_prev(&next->rb_node));
2509 base = pvm_determine_end(&next, &prev, align) - end;
2510 term_area = area;
2511 continue;
2512 }
2513
2514 /*
2515 * This area fits, move on to the previous one. If
2516 * the previous one is the terminal one, we're done.
2517 */
2518 area = (area + nr_vms - 1) % nr_vms;
2519 if (area == term_area)
2520 break;
2521 start = offsets[area];
2522 end = start + sizes[area];
2523 pvm_find_next_prev(base + end, &next, &prev);
2524 }
2525found:
2526 /* we've found a fitting base, insert all va's */
2527 for (area = 0; area < nr_vms; area++) {
2528 struct vmap_area *va = vas[area];
2529
2530 va->va_start = base + offsets[area];
2531 va->va_end = va->va_start + sizes[area];
2532 __insert_vmap_area(va);
2533 }
2534
2535 vmap_area_pcpu_hole = base + offsets[last_area];
2536
2537 spin_unlock(&vmap_area_lock);
2538
2539 /* insert all vm's */
2540 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002541 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2542 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002543
2544 kfree(vas);
2545 return vms;
2546
2547err_free:
2548 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002549 kfree(vas[area]);
2550 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002551 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002552err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002553 kfree(vas);
2554 kfree(vms);
2555 return NULL;
2556}
2557
2558/**
2559 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2560 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2561 * @nr_vms: the number of allocated areas
2562 *
2563 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2564 */
2565void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2566{
2567 int i;
2568
2569 for (i = 0; i < nr_vms; i++)
2570 free_vm_area(vms[i]);
2571 kfree(vms);
2572}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002573#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002574
2575#ifdef CONFIG_PROC_FS
2576static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002577 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002578{
2579 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002580 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002581
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002582 spin_lock(&vmap_area_lock);
Geliang Tang6219c2a2016-01-14 15:19:08 -08002583 va = list_first_entry(&vmap_area_list, typeof(*va), list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002584 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002585 n--;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002586 va = list_next_entry(va, list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002587 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002588 if (!n && &va->list != &vmap_area_list)
2589 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002590
2591 return NULL;
2592
2593}
2594
2595static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2596{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002597 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002598
2599 ++*pos;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002600 next = list_next_entry(va, list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002601 if (&next->list != &vmap_area_list)
2602 return next;
2603
2604 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002605}
2606
2607static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002608 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002609{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002610 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002611}
2612
Eric Dumazeta47a1262008-07-23 21:27:38 -07002613static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2614{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002615 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002616 unsigned int nr, *counters = m->private;
2617
2618 if (!counters)
2619 return;
2620
Wanpeng Liaf123462013-11-12 15:07:32 -08002621 if (v->flags & VM_UNINITIALIZED)
2622 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002623 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2624 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002625
Eric Dumazeta47a1262008-07-23 21:27:38 -07002626 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2627
2628 for (nr = 0; nr < v->nr_pages; nr++)
2629 counters[page_to_nid(v->pages[nr])]++;
2630
2631 for_each_node_state(nr, N_HIGH_MEMORY)
2632 if (counters[nr])
2633 seq_printf(m, " N%u=%u", nr, counters[nr]);
2634 }
2635}
2636
Christoph Lametera10aa572008-04-28 02:12:40 -07002637static int s_show(struct seq_file *m, void *p)
2638{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002639 struct vmap_area *va = p;
2640 struct vm_struct *v;
2641
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002642 /*
2643 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2644 * behalf of vmap area is being tear down or vm_map_ram allocation.
2645 */
2646 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002647 return 0;
2648
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002649 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002650
Kees Cook45ec1692012-10-08 16:34:09 -07002651 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002652 v->addr, v->addr + v->size, v->size);
2653
Joe Perches62c70bc2011-01-13 15:45:52 -08002654 if (v->caller)
2655 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002656
Christoph Lametera10aa572008-04-28 02:12:40 -07002657 if (v->nr_pages)
2658 seq_printf(m, " pages=%d", v->nr_pages);
2659
2660 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002661 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002662
2663 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002664 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002665
2666 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002667 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002668
2669 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002670 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002671
2672 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002673 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002674
David Rientjes244d63e2016-01-14 15:19:35 -08002675 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002676 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002677
Eric Dumazeta47a1262008-07-23 21:27:38 -07002678 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002679 seq_putc(m, '\n');
2680 return 0;
2681}
2682
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002683static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002684 .start = s_start,
2685 .next = s_next,
2686 .stop = s_stop,
2687 .show = s_show,
2688};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002689
2690static int vmalloc_open(struct inode *inode, struct file *file)
2691{
Rob Jones703394c2014-10-09 15:28:01 -07002692 if (IS_ENABLED(CONFIG_NUMA))
2693 return seq_open_private(file, &vmalloc_op,
2694 nr_node_ids * sizeof(unsigned int));
2695 else
2696 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002697}
2698
2699static const struct file_operations proc_vmalloc_operations = {
2700 .open = vmalloc_open,
2701 .read = seq_read,
2702 .llseek = seq_lseek,
2703 .release = seq_release_private,
2704};
2705
2706static int __init proc_vmalloc_init(void)
2707{
2708 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2709 return 0;
2710}
2711module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002712
Christoph Lametera10aa572008-04-28 02:12:40 -07002713#endif
2714