blob: cf7ad1a53be03b866ecee6325c47abef278d70fc [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
Nick Piggindb64fe02008-10-18 20:27:03 -0700278#define VM_VM_AREA 0x04
279
Nick Piggindb64fe02008-10-18 20:27:03 -0700280static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700281/* Export for kexec only */
282LIST_HEAD(vmap_area_list);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700283static LLIST_HEAD(vmap_purge_list);
Nick Piggin89699602011-03-22 16:30:36 -0700284static struct rb_root vmap_area_root = RB_ROOT;
285
286/* The vmap cache globals are protected by vmap_area_lock */
287static struct rb_node *free_vmap_cache;
288static unsigned long cached_hole_size;
289static unsigned long cached_vstart;
290static unsigned long cached_align;
291
Tejun Heoca23e402009-08-14 15:00:52 +0900292static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700293
294static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
Nick Piggindb64fe02008-10-18 20:27:03 -0700296 struct rb_node *n = vmap_area_root.rb_node;
297
298 while (n) {
299 struct vmap_area *va;
300
301 va = rb_entry(n, struct vmap_area, rb_node);
302 if (addr < va->va_start)
303 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700304 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700305 n = n->rb_right;
306 else
307 return va;
308 }
309
310 return NULL;
311}
312
313static void __insert_vmap_area(struct vmap_area *va)
314{
315 struct rb_node **p = &vmap_area_root.rb_node;
316 struct rb_node *parent = NULL;
317 struct rb_node *tmp;
318
319 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700320 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700321
322 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700323 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
324 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700325 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700326 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700327 p = &(*p)->rb_right;
328 else
329 BUG();
330 }
331
332 rb_link_node(&va->rb_node, parent, p);
333 rb_insert_color(&va->rb_node, &vmap_area_root);
334
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700335 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700336 tmp = rb_prev(&va->rb_node);
337 if (tmp) {
338 struct vmap_area *prev;
339 prev = rb_entry(tmp, struct vmap_area, rb_node);
340 list_add_rcu(&va->list, &prev->list);
341 } else
342 list_add_rcu(&va->list, &vmap_area_list);
343}
344
345static void purge_vmap_area_lazy(void);
346
Chris Wilson4da56b92016-04-04 14:46:42 +0100347static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
348
Nick Piggindb64fe02008-10-18 20:27:03 -0700349/*
350 * Allocate a region of KVA of the specified size and alignment, within the
351 * vstart and vend.
352 */
353static struct vmap_area *alloc_vmap_area(unsigned long size,
354 unsigned long align,
355 unsigned long vstart, unsigned long vend,
356 int node, gfp_t gfp_mask)
357{
358 struct vmap_area *va;
359 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700361 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700362 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700363
Nick Piggin77669702009-02-27 14:03:03 -0800364 BUG_ON(!size);
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800365 BUG_ON(offset_in_page(size));
Nick Piggin89699602011-03-22 16:30:36 -0700366 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700367
Chris Wilson4da56b92016-04-04 14:46:42 +0100368 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
369
Nick Piggindb64fe02008-10-18 20:27:03 -0700370 va = kmalloc_node(sizeof(struct vmap_area),
371 gfp_mask & GFP_RECLAIM_MASK, node);
372 if (unlikely(!va))
373 return ERR_PTR(-ENOMEM);
374
Catalin Marinas7f88f882013-11-12 15:07:45 -0800375 /*
376 * Only scan the relevant parts containing pointers to other objects
377 * to avoid false negatives.
378 */
379 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
380
Nick Piggindb64fe02008-10-18 20:27:03 -0700381retry:
382 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700383 /*
384 * Invalidate cache if we have more permissive parameters.
385 * cached_hole_size notes the largest hole noticed _below_
386 * the vmap_area cached in free_vmap_cache: if size fits
387 * into that hole, we want to scan from vstart to reuse
388 * the hole instead of allocating above free_vmap_cache.
389 * Note that __free_vmap_area may update free_vmap_cache
390 * without updating cached_hole_size or cached_align.
391 */
392 if (!free_vmap_cache ||
393 size < cached_hole_size ||
394 vstart < cached_vstart ||
395 align < cached_align) {
396nocache:
397 cached_hole_size = 0;
398 free_vmap_cache = NULL;
399 }
400 /* record if we encounter less permissive parameters */
401 cached_vstart = vstart;
402 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800403
Nick Piggin89699602011-03-22 16:30:36 -0700404 /* find starting point for our search */
405 if (free_vmap_cache) {
406 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700407 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700408 if (addr < vstart)
409 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700410 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700411 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700412
Nick Piggin89699602011-03-22 16:30:36 -0700413 } else {
414 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700415 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700416 goto overflow;
417
418 n = vmap_area_root.rb_node;
419 first = NULL;
420
421 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700422 struct vmap_area *tmp;
423 tmp = rb_entry(n, struct vmap_area, rb_node);
424 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700425 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700426 if (tmp->va_start <= addr)
427 break;
428 n = n->rb_left;
429 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700430 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700431 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700432
433 if (!first)
434 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700435 }
Nick Piggin89699602011-03-22 16:30:36 -0700436
437 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700438 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700439 if (addr + cached_hole_size < first->va_start)
440 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700441 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700442 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700443 goto overflow;
444
Hong zhi guo92ca9222012-07-31 16:41:35 -0700445 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700446 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700447
Geliang Tang6219c2a2016-01-14 15:19:08 -0800448 first = list_next_entry(first, list);
Nick Piggin89699602011-03-22 16:30:36 -0700449 }
450
Nick Piggindb64fe02008-10-18 20:27:03 -0700451found:
Nick Piggin89699602011-03-22 16:30:36 -0700452 if (addr + size > vend)
453 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700454
455 va->va_start = addr;
456 va->va_end = addr + size;
457 va->flags = 0;
458 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700459 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700460 spin_unlock(&vmap_area_lock);
461
Wang Xiaoqiang61e16552016-01-15 16:57:19 -0800462 BUG_ON(!IS_ALIGNED(va->va_start, align));
Nick Piggin89699602011-03-22 16:30:36 -0700463 BUG_ON(va->va_start < vstart);
464 BUG_ON(va->va_end > vend);
465
Nick Piggindb64fe02008-10-18 20:27:03 -0700466 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700467
468overflow:
469 spin_unlock(&vmap_area_lock);
470 if (!purged) {
471 purge_vmap_area_lazy();
472 purged = 1;
473 goto retry;
474 }
Chris Wilson4da56b92016-04-04 14:46:42 +0100475
476 if (gfpflags_allow_blocking(gfp_mask)) {
477 unsigned long freed = 0;
478 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
479 if (freed > 0) {
480 purged = 0;
481 goto retry;
482 }
483 }
484
Nick Piggin89699602011-03-22 16:30:36 -0700485 if (printk_ratelimit())
Joe Perches756a025f02016-03-17 14:19:47 -0700486 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
487 size);
Nick Piggin89699602011-03-22 16:30:36 -0700488 kfree(va);
489 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700490}
491
Chris Wilson4da56b92016-04-04 14:46:42 +0100492int register_vmap_purge_notifier(struct notifier_block *nb)
493{
494 return blocking_notifier_chain_register(&vmap_notify_list, nb);
495}
496EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
497
498int unregister_vmap_purge_notifier(struct notifier_block *nb)
499{
500 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
501}
502EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
503
Nick Piggindb64fe02008-10-18 20:27:03 -0700504static void __free_vmap_area(struct vmap_area *va)
505{
506 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700507
508 if (free_vmap_cache) {
509 if (va->va_end < cached_vstart) {
510 free_vmap_cache = NULL;
511 } else {
512 struct vmap_area *cache;
513 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
514 if (va->va_start <= cache->va_start) {
515 free_vmap_cache = rb_prev(&va->rb_node);
516 /*
517 * We don't try to update cached_hole_size or
518 * cached_align, but it won't go very wrong.
519 */
520 }
521 }
522 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700523 rb_erase(&va->rb_node, &vmap_area_root);
524 RB_CLEAR_NODE(&va->rb_node);
525 list_del_rcu(&va->list);
526
Tejun Heoca23e402009-08-14 15:00:52 +0900527 /*
528 * Track the highest possible candidate for pcpu area
529 * allocation. Areas outside of vmalloc area can be returned
530 * here too, consider only end addresses which fall inside
531 * vmalloc area proper.
532 */
533 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
534 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
535
Lai Jiangshan14769de2011-03-18 12:12:19 +0800536 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700537}
538
539/*
540 * Free a region of KVA allocated by alloc_vmap_area
541 */
542static void free_vmap_area(struct vmap_area *va)
543{
544 spin_lock(&vmap_area_lock);
545 __free_vmap_area(va);
546 spin_unlock(&vmap_area_lock);
547}
548
549/*
550 * Clear the pagetable entries of a given vmap_area
551 */
552static void unmap_vmap_area(struct vmap_area *va)
553{
554 vunmap_page_range(va->va_start, va->va_end);
555}
556
Nick Piggincd528582009-01-06 14:39:20 -0800557static void vmap_debug_free_range(unsigned long start, unsigned long end)
558{
559 /*
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700560 * Unmap page tables and force a TLB flush immediately if pagealloc
561 * debugging is enabled. This catches use after free bugs similarly to
562 * those in linear kernel virtual address space after a page has been
563 * freed.
Nick Piggincd528582009-01-06 14:39:20 -0800564 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700565 * All the lazy freeing logic is still retained, in order to minimise
566 * intrusiveness of this debugging feature.
Nick Piggincd528582009-01-06 14:39:20 -0800567 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700568 * This is going to be *slow* (linear kernel virtual address debugging
569 * doesn't do a broadcast TLB flush so it is a lot faster).
Nick Piggincd528582009-01-06 14:39:20 -0800570 */
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700571 if (debug_pagealloc_enabled()) {
572 vunmap_page_range(start, end);
573 flush_tlb_kernel_range(start, end);
574 }
Nick Piggincd528582009-01-06 14:39:20 -0800575}
576
Nick Piggindb64fe02008-10-18 20:27:03 -0700577/*
578 * lazy_max_pages is the maximum amount of virtual address space we gather up
579 * before attempting to purge with a TLB flush.
580 *
581 * There is a tradeoff here: a larger number will cover more kernel page tables
582 * and take slightly longer to purge, but it will linearly reduce the number of
583 * global TLB flushes that must be performed. It would seem natural to scale
584 * this number up linearly with the number of CPUs (because vmapping activity
585 * could also scale linearly with the number of CPUs), however it is likely
586 * that in practice, workloads might be constrained in other ways that mean
587 * vmap activity will not scale linearly with CPUs. Also, I want to be
588 * conservative and not introduce a big latency on huge systems, so go with
589 * a less aggressive log scale. It will still be an improvement over the old
590 * code, and it will be simple to change the scale factor if we find that it
591 * becomes a problem on bigger systems.
592 */
593static unsigned long lazy_max_pages(void)
594{
595 unsigned int log;
596
597 log = fls(num_online_cpus());
598
599 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
600}
601
602static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
603
Nick Piggin02b709d2010-02-01 22:25:57 +1100604/* for per-CPU blocks */
605static void purge_fragmented_blocks_allcpus(void);
606
Nick Piggindb64fe02008-10-18 20:27:03 -0700607/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500608 * called before a call to iounmap() if the caller wants vm_area_struct's
609 * immediately freed.
610 */
611void set_iounmap_nonlazy(void)
612{
613 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
614}
615
616/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700617 * Purges all lazily-freed vmap areas.
618 *
619 * If sync is 0 then don't purge if there is already a purge in progress.
620 * If force_flush is 1, then flush kernel TLBs between *start and *end even
621 * if we found no lazy vmap areas to unmap (callers can use this to optimise
622 * their own TLB flushing).
623 * Returns with *start = min(*start, lowest purged address)
624 * *end = max(*end, highest purged address)
625 */
626static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
627 int sync, int force_flush)
628{
Andrew Morton46666d82009-01-15 13:51:15 -0800629 static DEFINE_SPINLOCK(purge_lock);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700630 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -0700631 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800632 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700633 int nr = 0;
634
635 /*
636 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
637 * should not expect such behaviour. This just simplifies locking for
638 * the case that isn't actually used at the moment anyway.
639 */
640 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800641 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700642 return;
643 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800644 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700645
Nick Piggin02b709d2010-02-01 22:25:57 +1100646 if (sync)
647 purge_fragmented_blocks_allcpus();
648
Chris Wilson80c4bd72016-05-20 16:57:38 -0700649 valist = llist_del_all(&vmap_purge_list);
650 llist_for_each_entry(va, valist, purge_list) {
651 if (va->va_start < *start)
652 *start = va->va_start;
653 if (va->va_end > *end)
654 *end = va->va_end;
655 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700656 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700657
Yongseok Koh88f50042010-01-19 17:33:49 +0900658 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700659 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700660
661 if (nr || force_flush)
662 flush_tlb_kernel_range(*start, *end);
663
664 if (nr) {
665 spin_lock(&vmap_area_lock);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700666 llist_for_each_entry_safe(va, n_va, valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700667 __free_vmap_area(va);
668 spin_unlock(&vmap_area_lock);
669 }
Andrew Morton46666d82009-01-15 13:51:15 -0800670 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700671}
672
673/*
Nick Piggin496850e2008-11-19 15:36:33 -0800674 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
675 * is already purging.
676 */
677static void try_purge_vmap_area_lazy(void)
678{
679 unsigned long start = ULONG_MAX, end = 0;
680
681 __purge_vmap_area_lazy(&start, &end, 0, 0);
682}
683
684/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700685 * Kick off a purge of the outstanding lazy areas.
686 */
687static void purge_vmap_area_lazy(void)
688{
689 unsigned long start = ULONG_MAX, end = 0;
690
Nick Piggin496850e2008-11-19 15:36:33 -0800691 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700692}
693
694/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800695 * Free a vmap area, caller ensuring that the area has been unmapped
696 * and flush_cache_vunmap had been called for the correct range
697 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700698 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800699static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700700{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700701 int nr_lazy;
702
703 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
704 &vmap_lazy_nr);
705
706 /* After this point, we may free va at any time */
707 llist_add(&va->purge_list, &vmap_purge_list);
708
709 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800710 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700711}
712
Nick Pigginb29acbd2008-12-01 13:13:47 -0800713/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800714 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
715 * called for the correct range previously.
716 */
717static void free_unmap_vmap_area_noflush(struct vmap_area *va)
718{
719 unmap_vmap_area(va);
720 free_vmap_area_noflush(va);
721}
722
723/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800724 * Free and unmap a vmap area
725 */
726static void free_unmap_vmap_area(struct vmap_area *va)
727{
728 flush_cache_vunmap(va->va_start, va->va_end);
729 free_unmap_vmap_area_noflush(va);
730}
731
Nick Piggindb64fe02008-10-18 20:27:03 -0700732static struct vmap_area *find_vmap_area(unsigned long addr)
733{
734 struct vmap_area *va;
735
736 spin_lock(&vmap_area_lock);
737 va = __find_vmap_area(addr);
738 spin_unlock(&vmap_area_lock);
739
740 return va;
741}
742
743static void free_unmap_vmap_area_addr(unsigned long addr)
744{
745 struct vmap_area *va;
746
747 va = find_vmap_area(addr);
748 BUG_ON(!va);
749 free_unmap_vmap_area(va);
750}
751
752
753/*** Per cpu kva allocator ***/
754
755/*
756 * vmap space is limited especially on 32 bit architectures. Ensure there is
757 * room for at least 16 percpu vmap blocks per CPU.
758 */
759/*
760 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
761 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
762 * instead (we just need a rough idea)
763 */
764#if BITS_PER_LONG == 32
765#define VMALLOC_SPACE (128UL*1024*1024)
766#else
767#define VMALLOC_SPACE (128UL*1024*1024*1024)
768#endif
769
770#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
771#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
772#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
773#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
774#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
775#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200776#define VMAP_BBMAP_BITS \
777 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
778 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
779 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700780
781#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
782
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100783static bool vmap_initialized __read_mostly = false;
784
Nick Piggindb64fe02008-10-18 20:27:03 -0700785struct vmap_block_queue {
786 spinlock_t lock;
787 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700788};
789
790struct vmap_block {
791 spinlock_t lock;
792 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700793 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700794 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100795 struct list_head free_list;
796 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100797 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700798};
799
800/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
801static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
802
803/*
804 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
805 * in the free path. Could get rid of this if we change the API to return a
806 * "cookie" from alloc, to be passed to free. But no big deal yet.
807 */
808static DEFINE_SPINLOCK(vmap_block_tree_lock);
809static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
810
811/*
812 * We should probably have a fallback mechanism to allocate virtual memory
813 * out of partially filled vmap blocks. However vmap block sizing should be
814 * fairly reasonable according to the vmalloc size, so it shouldn't be a
815 * big problem.
816 */
817
818static unsigned long addr_to_vb_idx(unsigned long addr)
819{
820 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
821 addr /= VMAP_BLOCK_SIZE;
822 return addr;
823}
824
Roman Pencf725ce2015-04-15 16:13:52 -0700825static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
826{
827 unsigned long addr;
828
829 addr = va_start + (pages_off << PAGE_SHIFT);
830 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
831 return (void *)addr;
832}
833
834/**
835 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
836 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
837 * @order: how many 2^order pages should be occupied in newly allocated block
838 * @gfp_mask: flags for the page level allocator
839 *
840 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
841 */
842static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700843{
844 struct vmap_block_queue *vbq;
845 struct vmap_block *vb;
846 struct vmap_area *va;
847 unsigned long vb_idx;
848 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700849 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700850
851 node = numa_node_id();
852
853 vb = kmalloc_node(sizeof(struct vmap_block),
854 gfp_mask & GFP_RECLAIM_MASK, node);
855 if (unlikely(!vb))
856 return ERR_PTR(-ENOMEM);
857
858 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
859 VMALLOC_START, VMALLOC_END,
860 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800861 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700862 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700863 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700864 }
865
866 err = radix_tree_preload(gfp_mask);
867 if (unlikely(err)) {
868 kfree(vb);
869 free_vmap_area(va);
870 return ERR_PTR(err);
871 }
872
Roman Pencf725ce2015-04-15 16:13:52 -0700873 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700874 spin_lock_init(&vb->lock);
875 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700876 /* At least something should be left free */
877 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
878 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700879 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700880 vb->dirty_min = VMAP_BBMAP_BITS;
881 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700882 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700883
884 vb_idx = addr_to_vb_idx(va->va_start);
885 spin_lock(&vmap_block_tree_lock);
886 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
887 spin_unlock(&vmap_block_tree_lock);
888 BUG_ON(err);
889 radix_tree_preload_end();
890
891 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700892 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700893 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700894 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900895 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700896
Roman Pencf725ce2015-04-15 16:13:52 -0700897 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700898}
899
Nick Piggindb64fe02008-10-18 20:27:03 -0700900static void free_vmap_block(struct vmap_block *vb)
901{
902 struct vmap_block *tmp;
903 unsigned long vb_idx;
904
Nick Piggindb64fe02008-10-18 20:27:03 -0700905 vb_idx = addr_to_vb_idx(vb->va->va_start);
906 spin_lock(&vmap_block_tree_lock);
907 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
908 spin_unlock(&vmap_block_tree_lock);
909 BUG_ON(tmp != vb);
910
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800911 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800912 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700913}
914
Nick Piggin02b709d2010-02-01 22:25:57 +1100915static void purge_fragmented_blocks(int cpu)
916{
917 LIST_HEAD(purge);
918 struct vmap_block *vb;
919 struct vmap_block *n_vb;
920 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
921
922 rcu_read_lock();
923 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
924
925 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
926 continue;
927
928 spin_lock(&vb->lock);
929 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
930 vb->free = 0; /* prevent further allocs after releasing lock */
931 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700932 vb->dirty_min = 0;
933 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100934 spin_lock(&vbq->lock);
935 list_del_rcu(&vb->free_list);
936 spin_unlock(&vbq->lock);
937 spin_unlock(&vb->lock);
938 list_add_tail(&vb->purge, &purge);
939 } else
940 spin_unlock(&vb->lock);
941 }
942 rcu_read_unlock();
943
944 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
945 list_del(&vb->purge);
946 free_vmap_block(vb);
947 }
948}
949
Nick Piggin02b709d2010-02-01 22:25:57 +1100950static void purge_fragmented_blocks_allcpus(void)
951{
952 int cpu;
953
954 for_each_possible_cpu(cpu)
955 purge_fragmented_blocks(cpu);
956}
957
Nick Piggindb64fe02008-10-18 20:27:03 -0700958static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
959{
960 struct vmap_block_queue *vbq;
961 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700962 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700963 unsigned int order;
964
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800965 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700966 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700967 if (WARN_ON(size == 0)) {
968 /*
969 * Allocating 0 bytes isn't what caller wants since
970 * get_order(0) returns funny result. Just warn and terminate
971 * early.
972 */
973 return NULL;
974 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700975 order = get_order(size);
976
Nick Piggindb64fe02008-10-18 20:27:03 -0700977 rcu_read_lock();
978 vbq = &get_cpu_var(vmap_block_queue);
979 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -0700980 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -0700981
982 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -0700983 if (vb->free < (1UL << order)) {
984 spin_unlock(&vb->lock);
985 continue;
986 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100987
Roman Pencf725ce2015-04-15 16:13:52 -0700988 pages_off = VMAP_BBMAP_BITS - vb->free;
989 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +1100990 vb->free -= 1UL << order;
991 if (vb->free == 0) {
992 spin_lock(&vbq->lock);
993 list_del_rcu(&vb->free_list);
994 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700995 }
Roman Pencf725ce2015-04-15 16:13:52 -0700996
Nick Piggindb64fe02008-10-18 20:27:03 -0700997 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100998 break;
Nick Piggindb64fe02008-10-18 20:27:03 -0700999 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001000
Tejun Heo3f04ba82009-10-29 22:34:12 +09001001 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001002 rcu_read_unlock();
1003
Roman Pencf725ce2015-04-15 16:13:52 -07001004 /* Allocate new block if nothing was found */
1005 if (!vaddr)
1006 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001007
Roman Pencf725ce2015-04-15 16:13:52 -07001008 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001009}
1010
1011static void vb_free(const void *addr, unsigned long size)
1012{
1013 unsigned long offset;
1014 unsigned long vb_idx;
1015 unsigned int order;
1016 struct vmap_block *vb;
1017
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001018 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001019 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001020
1021 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1022
Nick Piggindb64fe02008-10-18 20:27:03 -07001023 order = get_order(size);
1024
1025 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001026 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001027
1028 vb_idx = addr_to_vb_idx((unsigned long)addr);
1029 rcu_read_lock();
1030 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1031 rcu_read_unlock();
1032 BUG_ON(!vb);
1033
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001034 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1035
Nick Piggindb64fe02008-10-18 20:27:03 -07001036 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001037
1038 /* Expand dirty range */
1039 vb->dirty_min = min(vb->dirty_min, offset);
1040 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001041
Nick Piggindb64fe02008-10-18 20:27:03 -07001042 vb->dirty += 1UL << order;
1043 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001044 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001045 spin_unlock(&vb->lock);
1046 free_vmap_block(vb);
1047 } else
1048 spin_unlock(&vb->lock);
1049}
1050
1051/**
1052 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1053 *
1054 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1055 * to amortize TLB flushing overheads. What this means is that any page you
1056 * have now, may, in a former life, have been mapped into kernel virtual
1057 * address by the vmap layer and so there might be some CPUs with TLB entries
1058 * still referencing that page (additional to the regular 1:1 kernel mapping).
1059 *
1060 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1061 * be sure that none of the pages we have control over will have any aliases
1062 * from the vmap layer.
1063 */
1064void vm_unmap_aliases(void)
1065{
1066 unsigned long start = ULONG_MAX, end = 0;
1067 int cpu;
1068 int flush = 0;
1069
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001070 if (unlikely(!vmap_initialized))
1071 return;
1072
Nick Piggindb64fe02008-10-18 20:27:03 -07001073 for_each_possible_cpu(cpu) {
1074 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1075 struct vmap_block *vb;
1076
1077 rcu_read_lock();
1078 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001079 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001080 if (vb->dirty) {
1081 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001082 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001083
Roman Pen7d61bfe2015-04-15 16:13:55 -07001084 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1085 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001086
Roman Pen7d61bfe2015-04-15 16:13:55 -07001087 start = min(s, start);
1088 end = max(e, end);
1089
Nick Piggindb64fe02008-10-18 20:27:03 -07001090 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001091 }
1092 spin_unlock(&vb->lock);
1093 }
1094 rcu_read_unlock();
1095 }
1096
1097 __purge_vmap_area_lazy(&start, &end, 1, flush);
1098}
1099EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1100
1101/**
1102 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1103 * @mem: the pointer returned by vm_map_ram
1104 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1105 */
1106void vm_unmap_ram(const void *mem, unsigned int count)
1107{
1108 unsigned long size = count << PAGE_SHIFT;
1109 unsigned long addr = (unsigned long)mem;
1110
1111 BUG_ON(!addr);
1112 BUG_ON(addr < VMALLOC_START);
1113 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001114 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001115
1116 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001117 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001118
1119 if (likely(count <= VMAP_MAX_ALLOC))
1120 vb_free(mem, size);
1121 else
1122 free_unmap_vmap_area_addr(addr);
1123}
1124EXPORT_SYMBOL(vm_unmap_ram);
1125
1126/**
1127 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1128 * @pages: an array of pointers to the pages to be mapped
1129 * @count: number of pages
1130 * @node: prefer to allocate data structures on this node
1131 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001132 *
Gioh Kim36437632014-04-07 15:37:37 -07001133 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1134 * faster than vmap so it's good. But if you mix long-life and short-life
1135 * objects with vm_map_ram(), it could consume lots of address space through
1136 * fragmentation (especially on a 32bit machine). You could see failures in
1137 * the end. Please use this function for short-lived objects.
1138 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001139 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001140 */
1141void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1142{
1143 unsigned long size = count << PAGE_SHIFT;
1144 unsigned long addr;
1145 void *mem;
1146
1147 if (likely(count <= VMAP_MAX_ALLOC)) {
1148 mem = vb_alloc(size, GFP_KERNEL);
1149 if (IS_ERR(mem))
1150 return NULL;
1151 addr = (unsigned long)mem;
1152 } else {
1153 struct vmap_area *va;
1154 va = alloc_vmap_area(size, PAGE_SIZE,
1155 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1156 if (IS_ERR(va))
1157 return NULL;
1158
1159 addr = va->va_start;
1160 mem = (void *)addr;
1161 }
1162 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1163 vm_unmap_ram(mem, count);
1164 return NULL;
1165 }
1166 return mem;
1167}
1168EXPORT_SYMBOL(vm_map_ram);
1169
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001170static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001171/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001172 * vm_area_add_early - add vmap area early during boot
1173 * @vm: vm_struct to add
1174 *
1175 * This function is used to add fixed kernel vm area to vmlist before
1176 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1177 * should contain proper values and the other fields should be zero.
1178 *
1179 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1180 */
1181void __init vm_area_add_early(struct vm_struct *vm)
1182{
1183 struct vm_struct *tmp, **p;
1184
1185 BUG_ON(vmap_initialized);
1186 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1187 if (tmp->addr >= vm->addr) {
1188 BUG_ON(tmp->addr < vm->addr + vm->size);
1189 break;
1190 } else
1191 BUG_ON(tmp->addr + tmp->size > vm->addr);
1192 }
1193 vm->next = *p;
1194 *p = vm;
1195}
1196
1197/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001198 * vm_area_register_early - register vmap area early during boot
1199 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001200 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001201 *
1202 * This function is used to register kernel vm area before
1203 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1204 * proper values on entry and other fields should be zero. On return,
1205 * vm->addr contains the allocated address.
1206 *
1207 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1208 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001209void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001210{
1211 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001212 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001213
Tejun Heoc0c0a292009-02-24 11:57:21 +09001214 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1215 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1216
1217 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001218
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001219 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001220}
1221
Nick Piggindb64fe02008-10-18 20:27:03 -07001222void __init vmalloc_init(void)
1223{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001224 struct vmap_area *va;
1225 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001226 int i;
1227
1228 for_each_possible_cpu(i) {
1229 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001230 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001231
1232 vbq = &per_cpu(vmap_block_queue, i);
1233 spin_lock_init(&vbq->lock);
1234 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001235 p = &per_cpu(vfree_deferred, i);
1236 init_llist_head(&p->list);
1237 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001238 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001239
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001240 /* Import existing vmlist entries. */
1241 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001242 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001243 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001244 va->va_start = (unsigned long)tmp->addr;
1245 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001246 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001247 __insert_vmap_area(va);
1248 }
Tejun Heoca23e402009-08-14 15:00:52 +09001249
1250 vmap_area_pcpu_hole = VMALLOC_END;
1251
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001252 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001253}
1254
Tejun Heo8fc48982009-02-20 16:29:08 +09001255/**
1256 * map_kernel_range_noflush - map kernel VM area with the specified pages
1257 * @addr: start of the VM area to map
1258 * @size: size of the VM area to map
1259 * @prot: page protection flags to use
1260 * @pages: pages to map
1261 *
1262 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1263 * specify should have been allocated using get_vm_area() and its
1264 * friends.
1265 *
1266 * NOTE:
1267 * This function does NOT do any cache flushing. The caller is
1268 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1269 * before calling this function.
1270 *
1271 * RETURNS:
1272 * The number of pages mapped on success, -errno on failure.
1273 */
1274int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1275 pgprot_t prot, struct page **pages)
1276{
1277 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1278}
1279
1280/**
1281 * unmap_kernel_range_noflush - unmap kernel VM area
1282 * @addr: start of the VM area to unmap
1283 * @size: size of the VM area to unmap
1284 *
1285 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1286 * specify should have been allocated using get_vm_area() and its
1287 * friends.
1288 *
1289 * NOTE:
1290 * This function does NOT do any cache flushing. The caller is
1291 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1292 * before calling this function and flush_tlb_kernel_range() after.
1293 */
1294void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1295{
1296 vunmap_page_range(addr, addr + size);
1297}
Huang Ying81e88fd2011-01-12 14:44:55 +08001298EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001299
1300/**
1301 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1302 * @addr: start of the VM area to unmap
1303 * @size: size of the VM area to unmap
1304 *
1305 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1306 * the unmapping and tlb after.
1307 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001308void unmap_kernel_range(unsigned long addr, unsigned long size)
1309{
1310 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001311
1312 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001313 vunmap_page_range(addr, end);
1314 flush_tlb_kernel_range(addr, end);
1315}
Minchan Kim93ef6d62014-06-04 16:11:09 -07001316EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001317
WANG Chaof6f8ed42014-08-06 16:06:58 -07001318int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001319{
1320 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001321 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001322 int err;
1323
WANG Chaof6f8ed42014-08-06 16:06:58 -07001324 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001325
WANG Chaof6f8ed42014-08-06 16:06:58 -07001326 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001327}
1328EXPORT_SYMBOL_GPL(map_vm_area);
1329
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001330static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001331 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001332{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001333 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001334 vm->flags = flags;
1335 vm->addr = (void *)va->va_start;
1336 vm->size = va->va_end - va->va_start;
1337 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001338 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001339 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001340 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001341}
Tejun Heocf88c792009-08-14 15:00:52 +09001342
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001343static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001344{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001345 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001346 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001347 * we should make sure that vm has proper values.
1348 * Pair with smp_rmb() in show_numa_info().
1349 */
1350 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001351 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001352}
1353
Nick Piggindb64fe02008-10-18 20:27:03 -07001354static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001355 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001356 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001357{
Kautuk Consul00065262011-12-19 17:12:04 -08001358 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001359 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001361 BUG_ON(in_interrupt());
Zhang Yanfei0f2d4a82013-07-03 15:04:50 -07001362 if (flags & VM_IOREMAP)
Toshi Kani0f616be2015-04-14 15:47:17 -07001363 align = 1ul << clamp_t(int, fls_long(size),
1364 PAGE_SHIFT, IOREMAP_MAX_ORDER);
Nick Piggindb64fe02008-10-18 20:27:03 -07001365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001367 if (unlikely(!size))
1368 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Tejun Heocf88c792009-08-14 15:00:52 +09001370 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (unlikely(!area))
1372 return NULL;
1373
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001374 if (!(flags & VM_NO_GUARD))
1375 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Nick Piggindb64fe02008-10-18 20:27:03 -07001377 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1378 if (IS_ERR(va)) {
1379 kfree(area);
1380 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001383 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386}
1387
Christoph Lameter930fc452005-10-29 18:15:41 -07001388struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1389 unsigned long start, unsigned long end)
1390{
David Rientjes00ef2d22013-02-22 16:35:36 -08001391 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1392 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001393}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001394EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001395
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001396struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1397 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001398 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001399{
David Rientjes00ef2d22013-02-22 16:35:36 -08001400 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1401 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001402}
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404/**
Simon Arlott183ff222007-10-20 01:27:18 +02001405 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 * @size: size of the area
1407 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1408 *
1409 * Search an area of @size in the kernel virtual mapping area,
1410 * and reserved it for out purposes. Returns the area descriptor
1411 * on success or %NULL on failure.
1412 */
1413struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1414{
David Miller2dca6992009-09-21 12:22:34 -07001415 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001416 NUMA_NO_NODE, GFP_KERNEL,
1417 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001418}
1419
1420struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001421 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001422{
David Miller2dca6992009-09-21 12:22:34 -07001423 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001424 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425}
1426
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001427/**
1428 * find_vm_area - find a continuous kernel virtual area
1429 * @addr: base address
1430 *
1431 * Search for the kernel VM area starting at @addr, and return it.
1432 * It is up to the caller to do all required locking to keep the returned
1433 * pointer valid.
1434 */
1435struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001436{
Nick Piggindb64fe02008-10-18 20:27:03 -07001437 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001438
Nick Piggindb64fe02008-10-18 20:27:03 -07001439 va = find_vmap_area((unsigned long)addr);
1440 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001441 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001442
Andi Kleen7856dfe2005-05-20 14:27:57 -07001443 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001444}
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446/**
Simon Arlott183ff222007-10-20 01:27:18 +02001447 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * @addr: base address
1449 *
1450 * Search for the kernel VM area starting at @addr, and remove it.
1451 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001452 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001454struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455{
Nick Piggindb64fe02008-10-18 20:27:03 -07001456 struct vmap_area *va;
1457
1458 va = find_vmap_area((unsigned long)addr);
1459 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001460 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001461
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001462 spin_lock(&vmap_area_lock);
1463 va->vm = NULL;
1464 va->flags &= ~VM_VM_AREA;
1465 spin_unlock(&vmap_area_lock);
1466
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001467 vmap_debug_free_range(va->va_start, va->va_end);
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001468 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001469 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001470
Nick Piggindb64fe02008-10-18 20:27:03 -07001471 return vm;
1472 }
1473 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001476static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
1478 struct vm_struct *area;
1479
1480 if (!addr)
1481 return;
1482
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001483 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001484 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
1487 area = remove_vm_area(addr);
1488 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001489 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 return;
1492 }
1493
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001494 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1495 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 if (deallocate_pages) {
1498 int i;
1499
1500 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001501 struct page *page = area->pages[i];
1502
1503 BUG_ON(!page);
Vladimir Davydov37f08dd2016-01-14 15:18:18 -08001504 __free_kmem_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 }
1506
David Rientjes244d63e2016-01-14 15:19:35 -08001507 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
1509
1510 kfree(area);
1511 return;
1512}
Al Viro32fcfd42013-03-10 20:14:08 -04001513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514/**
1515 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 * @addr: memory base address
1517 *
Simon Arlott183ff222007-10-20 01:27:18 +02001518 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001519 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1520 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
Al Viro32fcfd42013-03-10 20:14:08 -04001522 * Must not be called in NMI context (strictly speaking, only if we don't
1523 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1524 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001525 *
1526 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001528void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
Al Viro32fcfd42013-03-10 20:14:08 -04001530 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001531
1532 kmemleak_free(addr);
1533
Al Viro32fcfd42013-03-10 20:14:08 -04001534 if (!addr)
1535 return;
1536 if (unlikely(in_interrupt())) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001537 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
Oleg Nesterov59d31322013-07-08 16:00:08 -07001538 if (llist_add((struct llist_node *)addr, &p->list))
1539 schedule_work(&p->wq);
Al Viro32fcfd42013-03-10 20:14:08 -04001540 } else
1541 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543EXPORT_SYMBOL(vfree);
1544
1545/**
1546 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 * @addr: memory base address
1548 *
1549 * Free the virtually contiguous memory area starting at @addr,
1550 * which was created from the page array passed to vmap().
1551 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001552 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001554void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555{
1556 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001557 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001558 if (addr)
1559 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561EXPORT_SYMBOL(vunmap);
1562
1563/**
1564 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 * @pages: array of page pointers
1566 * @count: number of pages to map
1567 * @flags: vm_area->flags
1568 * @prot: page protection for the mapping
1569 *
1570 * Maps @count pages from @pages into contiguous kernel virtual
1571 * space.
1572 */
1573void *vmap(struct page **pages, unsigned int count,
1574 unsigned long flags, pgprot_t prot)
1575{
1576 struct vm_struct *area;
1577
Peter Zijlstra34754b62009-02-25 16:04:03 +01001578 might_sleep();
1579
Jan Beulich44813742009-09-21 17:03:05 -07001580 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return NULL;
1582
Christoph Lameter23016962008-04-28 02:12:42 -07001583 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1584 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 if (!area)
1586 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001587
WANG Chaof6f8ed42014-08-06 16:06:58 -07001588 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 vunmap(area->addr);
1590 return NULL;
1591 }
1592
1593 return area->addr;
1594}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595EXPORT_SYMBOL(vmap);
1596
David Miller2dca6992009-09-21 12:22:34 -07001597static void *__vmalloc_node(unsigned long size, unsigned long align,
1598 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001599 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001600static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001601 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602{
Dave Hansen22943ab2011-05-24 17:12:18 -07001603 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 struct page **pages;
1605 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001606 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1607 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
Wanpeng Li762216a2013-09-11 14:22:42 -07001609 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 array_size = (nr_pages * sizeof(struct page *));
1611
1612 area->nr_pages = nr_pages;
1613 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001614 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001615 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001616 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001617 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001618 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 area->pages = pages;
1621 if (!area->pages) {
1622 remove_vm_area(area->addr);
1623 kfree(area);
1624 return NULL;
1625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
1627 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001628 struct page *page;
1629
Jianguo Wu4b909512013-11-12 15:07:11 -08001630 if (node == NUMA_NO_NODE)
Vladimir Davydov37f08dd2016-01-14 15:18:18 -08001631 page = alloc_kmem_pages(alloc_mask, order);
Christoph Lameter930fc452005-10-29 18:15:41 -07001632 else
Vladimir Davydov37f08dd2016-01-14 15:18:18 -08001633 page = alloc_kmem_pages_node(node, alloc_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001634
1635 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 /* Successfully allocated i pages, free them in __vunmap() */
1637 area->nr_pages = i;
1638 goto fail;
1639 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001640 area->pages[i] = page;
Mel Gormand0164ad2015-11-06 16:28:21 -08001641 if (gfpflags_allow_blocking(gfp_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001642 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 }
1644
WANG Chaof6f8ed42014-08-06 16:06:58 -07001645 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 goto fail;
1647 return area->addr;
1648
1649fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001650 warn_alloc_failed(gfp_mask, order,
1651 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001652 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 vfree(area->addr);
1654 return NULL;
1655}
1656
David Rientjesd0a21262011-01-13 15:46:02 -08001657/**
1658 * __vmalloc_node_range - allocate virtually contiguous memory
1659 * @size: allocation size
1660 * @align: desired alignment
1661 * @start: vm area range start
1662 * @end: vm area range end
1663 * @gfp_mask: flags for the page level allocator
1664 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001665 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001666 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001667 * @caller: caller's return address
1668 *
1669 * Allocate enough pages to cover @size from the page level
1670 * allocator with @gfp_mask flags. Map them into contiguous
1671 * kernel virtual space, using a pagetable protection of @prot.
1672 */
1673void *__vmalloc_node_range(unsigned long size, unsigned long align,
1674 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001675 pgprot_t prot, unsigned long vm_flags, int node,
1676 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001677{
David Rientjesd0a21262011-01-13 15:46:02 -08001678 struct vm_struct *area;
1679 void *addr;
1680 unsigned long real_size = size;
1681
1682 size = PAGE_ALIGN(size);
1683 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001684 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001685
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001686 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1687 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001688 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001689 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001690
Wanpeng Li3722e132013-11-12 15:07:29 -08001691 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001692 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001693 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001694
1695 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001696 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1697 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001698 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001699 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001700 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001701
1702 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001703 * A ref_count = 2 is needed because vm_struct allocated in
1704 * __get_vm_area_node() contains a reference to the virtual address of
1705 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001706 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001707 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001708
1709 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001710
1711fail:
1712 warn_alloc_failed(gfp_mask, 0,
1713 "vmalloc: allocation failure: %lu bytes\n",
1714 real_size);
1715 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001716}
1717
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001719 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001721 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 * @gfp_mask: flags for the page level allocator
1723 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001724 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001725 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 *
1727 * Allocate enough pages to cover @size from the page level
1728 * allocator with @gfp_mask flags. Map them into contiguous
1729 * kernel virtual space, using a pagetable protection of @prot.
1730 */
David Miller2dca6992009-09-21 12:22:34 -07001731static void *__vmalloc_node(unsigned long size, unsigned long align,
1732 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001733 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
David Rientjesd0a21262011-01-13 15:46:02 -08001735 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001736 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737}
1738
Christoph Lameter930fc452005-10-29 18:15:41 -07001739void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1740{
David Rientjes00ef2d22013-02-22 16:35:36 -08001741 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001742 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001743}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744EXPORT_SYMBOL(__vmalloc);
1745
Dave Younge1ca7782010-10-26 14:22:06 -07001746static inline void *__vmalloc_node_flags(unsigned long size,
1747 int node, gfp_t flags)
1748{
1749 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1750 node, __builtin_return_address(0));
1751}
1752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753/**
1754 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * Allocate enough pages to cover @size from the page level
1757 * allocator and map them into contiguous kernel virtual space.
1758 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001759 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 * use __vmalloc() instead.
1761 */
1762void *vmalloc(unsigned long size)
1763{
David Rientjes00ef2d22013-02-22 16:35:36 -08001764 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1765 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767EXPORT_SYMBOL(vmalloc);
1768
Christoph Lameter930fc452005-10-29 18:15:41 -07001769/**
Dave Younge1ca7782010-10-26 14:22:06 -07001770 * vzalloc - allocate virtually contiguous memory with zero fill
1771 * @size: allocation size
1772 * Allocate enough pages to cover @size from the page level
1773 * allocator and map them into contiguous kernel virtual space.
1774 * The memory allocated is set to zero.
1775 *
1776 * For tight control over page level allocator and protection flags
1777 * use __vmalloc() instead.
1778 */
1779void *vzalloc(unsigned long size)
1780{
David Rientjes00ef2d22013-02-22 16:35:36 -08001781 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001782 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1783}
1784EXPORT_SYMBOL(vzalloc);
1785
1786/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001787 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1788 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001789 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001790 * The resulting memory area is zeroed so it can be mapped to userspace
1791 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001792 */
1793void *vmalloc_user(unsigned long size)
1794{
1795 struct vm_struct *area;
1796 void *ret;
1797
David Miller2dca6992009-09-21 12:22:34 -07001798 ret = __vmalloc_node(size, SHMLBA,
1799 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001800 PAGE_KERNEL, NUMA_NO_NODE,
1801 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001802 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001803 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001804 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001805 }
Nick Piggin83342312006-06-23 02:03:20 -07001806 return ret;
1807}
1808EXPORT_SYMBOL(vmalloc_user);
1809
1810/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001811 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001812 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001813 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001814 *
1815 * Allocate enough pages to cover @size from the page level
1816 * allocator and map them into contiguous kernel virtual space.
1817 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001818 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001819 * use __vmalloc() instead.
1820 */
1821void *vmalloc_node(unsigned long size, int node)
1822{
David Miller2dca6992009-09-21 12:22:34 -07001823 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001824 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001825}
1826EXPORT_SYMBOL(vmalloc_node);
1827
Dave Younge1ca7782010-10-26 14:22:06 -07001828/**
1829 * vzalloc_node - allocate memory on a specific node with zero fill
1830 * @size: allocation size
1831 * @node: numa node
1832 *
1833 * Allocate enough pages to cover @size from the page level
1834 * allocator and map them into contiguous kernel virtual space.
1835 * The memory allocated is set to zero.
1836 *
1837 * For tight control over page level allocator and protection flags
1838 * use __vmalloc_node() instead.
1839 */
1840void *vzalloc_node(unsigned long size, int node)
1841{
1842 return __vmalloc_node_flags(size, node,
1843 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1844}
1845EXPORT_SYMBOL(vzalloc_node);
1846
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001847#ifndef PAGE_KERNEL_EXEC
1848# define PAGE_KERNEL_EXEC PAGE_KERNEL
1849#endif
1850
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851/**
1852 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 * @size: allocation size
1854 *
1855 * Kernel-internal function to allocate enough pages to cover @size
1856 * the page level allocator and map them into contiguous and
1857 * executable kernel virtual space.
1858 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001859 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 * use __vmalloc() instead.
1861 */
1862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863void *vmalloc_exec(unsigned long size)
1864{
David Miller2dca6992009-09-21 12:22:34 -07001865 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001866 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867}
1868
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001869#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001870#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001871#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001872#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001873#else
1874#define GFP_VMALLOC32 GFP_KERNEL
1875#endif
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877/**
1878 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 * @size: allocation size
1880 *
1881 * Allocate enough 32bit PA addressable pages to cover @size from the
1882 * page level allocator and map them into contiguous kernel virtual space.
1883 */
1884void *vmalloc_32(unsigned long size)
1885{
David Miller2dca6992009-09-21 12:22:34 -07001886 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001887 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889EXPORT_SYMBOL(vmalloc_32);
1890
Nick Piggin83342312006-06-23 02:03:20 -07001891/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001892 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001893 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001894 *
1895 * The resulting memory area is 32bit addressable and zeroed so it can be
1896 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001897 */
1898void *vmalloc_32_user(unsigned long size)
1899{
1900 struct vm_struct *area;
1901 void *ret;
1902
David Miller2dca6992009-09-21 12:22:34 -07001903 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001904 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001905 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001906 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001907 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001908 }
Nick Piggin83342312006-06-23 02:03:20 -07001909 return ret;
1910}
1911EXPORT_SYMBOL(vmalloc_32_user);
1912
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001913/*
1914 * small helper routine , copy contents to buf from addr.
1915 * If the page is not present, fill zero.
1916 */
1917
1918static int aligned_vread(char *buf, char *addr, unsigned long count)
1919{
1920 struct page *p;
1921 int copied = 0;
1922
1923 while (count) {
1924 unsigned long offset, length;
1925
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001926 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001927 length = PAGE_SIZE - offset;
1928 if (length > count)
1929 length = count;
1930 p = vmalloc_to_page(addr);
1931 /*
1932 * To do safe access to this _mapped_ area, we need
1933 * lock. But adding lock here means that we need to add
1934 * overhead of vmalloc()/vfree() calles for this _debug_
1935 * interface, rarely used. Instead of that, we'll use
1936 * kmap() and get small overhead in this access function.
1937 */
1938 if (p) {
1939 /*
1940 * we can expect USER0 is not used (see vread/vwrite's
1941 * function description)
1942 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001943 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001944 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001945 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001946 } else
1947 memset(buf, 0, length);
1948
1949 addr += length;
1950 buf += length;
1951 copied += length;
1952 count -= length;
1953 }
1954 return copied;
1955}
1956
1957static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1958{
1959 struct page *p;
1960 int copied = 0;
1961
1962 while (count) {
1963 unsigned long offset, length;
1964
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001965 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001966 length = PAGE_SIZE - offset;
1967 if (length > count)
1968 length = count;
1969 p = vmalloc_to_page(addr);
1970 /*
1971 * To do safe access to this _mapped_ area, we need
1972 * lock. But adding lock here means that we need to add
1973 * overhead of vmalloc()/vfree() calles for this _debug_
1974 * interface, rarely used. Instead of that, we'll use
1975 * kmap() and get small overhead in this access function.
1976 */
1977 if (p) {
1978 /*
1979 * we can expect USER0 is not used (see vread/vwrite's
1980 * function description)
1981 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001982 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001983 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001984 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001985 }
1986 addr += length;
1987 buf += length;
1988 copied += length;
1989 count -= length;
1990 }
1991 return copied;
1992}
1993
1994/**
1995 * vread() - read vmalloc area in a safe way.
1996 * @buf: buffer for reading data
1997 * @addr: vm address.
1998 * @count: number of bytes to be read.
1999 *
2000 * Returns # of bytes which addr and buf should be increased.
2001 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2002 * includes any intersect with alive vmalloc area.
2003 *
2004 * This function checks that addr is a valid vmalloc'ed area, and
2005 * copy data from that area to a given buffer. If the given memory range
2006 * of [addr...addr+count) includes some valid address, data is copied to
2007 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2008 * IOREMAP area is treated as memory hole and no copy is done.
2009 *
2010 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002011 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002012 *
2013 * Note: In usual ops, vread() is never necessary because the caller
2014 * should know vmalloc() area is valid and can use memcpy().
2015 * This is for routines which have to access vmalloc area without
2016 * any informaion, as /dev/kmem.
2017 *
2018 */
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020long vread(char *buf, char *addr, unsigned long count)
2021{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002022 struct vmap_area *va;
2023 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002025 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 unsigned long n;
2027
2028 /* Don't allow overflow */
2029 if ((unsigned long) addr + count < count)
2030 count = -(unsigned long) addr;
2031
Joonsoo Kime81ce852013-04-29 15:07:32 -07002032 spin_lock(&vmap_area_lock);
2033 list_for_each_entry(va, &vmap_area_list, list) {
2034 if (!count)
2035 break;
2036
2037 if (!(va->flags & VM_VM_AREA))
2038 continue;
2039
2040 vm = va->vm;
2041 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002042 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 continue;
2044 while (addr < vaddr) {
2045 if (count == 0)
2046 goto finished;
2047 *buf = '\0';
2048 buf++;
2049 addr++;
2050 count--;
2051 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002052 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002053 if (n > count)
2054 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002055 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002056 aligned_vread(buf, addr, n);
2057 else /* IOREMAP area is treated as memory hole */
2058 memset(buf, 0, n);
2059 buf += n;
2060 addr += n;
2061 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 }
2063finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002064 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002065
2066 if (buf == buf_start)
2067 return 0;
2068 /* zero-fill memory holes */
2069 if (buf != buf_start + buflen)
2070 memset(buf, 0, buflen - (buf - buf_start));
2071
2072 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073}
2074
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002075/**
2076 * vwrite() - write vmalloc area in a safe way.
2077 * @buf: buffer for source data
2078 * @addr: vm address.
2079 * @count: number of bytes to be read.
2080 *
2081 * Returns # of bytes which addr and buf should be incresed.
2082 * (same number to @count).
2083 * If [addr...addr+count) doesn't includes any intersect with valid
2084 * vmalloc area, returns 0.
2085 *
2086 * This function checks that addr is a valid vmalloc'ed area, and
2087 * copy data from a buffer to the given addr. If specified range of
2088 * [addr...addr+count) includes some valid address, data is copied from
2089 * proper area of @buf. If there are memory holes, no copy to hole.
2090 * IOREMAP area is treated as memory hole and no copy is done.
2091 *
2092 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002093 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002094 *
2095 * Note: In usual ops, vwrite() is never necessary because the caller
2096 * should know vmalloc() area is valid and can use memcpy().
2097 * This is for routines which have to access vmalloc area without
2098 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002099 */
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101long vwrite(char *buf, char *addr, unsigned long count)
2102{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002103 struct vmap_area *va;
2104 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002105 char *vaddr;
2106 unsigned long n, buflen;
2107 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109 /* Don't allow overflow */
2110 if ((unsigned long) addr + count < count)
2111 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002112 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Joonsoo Kime81ce852013-04-29 15:07:32 -07002114 spin_lock(&vmap_area_lock);
2115 list_for_each_entry(va, &vmap_area_list, list) {
2116 if (!count)
2117 break;
2118
2119 if (!(va->flags & VM_VM_AREA))
2120 continue;
2121
2122 vm = va->vm;
2123 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002124 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 continue;
2126 while (addr < vaddr) {
2127 if (count == 0)
2128 goto finished;
2129 buf++;
2130 addr++;
2131 count--;
2132 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002133 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002134 if (n > count)
2135 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002136 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002137 aligned_vwrite(buf, addr, n);
2138 copied++;
2139 }
2140 buf += n;
2141 addr += n;
2142 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 }
2144finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002145 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002146 if (!copied)
2147 return 0;
2148 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149}
Nick Piggin83342312006-06-23 02:03:20 -07002150
2151/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002152 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2153 * @vma: vma to cover
2154 * @uaddr: target user address to start at
2155 * @kaddr: virtual address of vmalloc kernel memory
2156 * @size: size of map area
2157 *
2158 * Returns: 0 for success, -Exxx on failure
2159 *
2160 * This function checks that @kaddr is a valid vmalloc'ed area,
2161 * and that it is big enough to cover the range starting at
2162 * @uaddr in @vma. Will return failure if that criteria isn't
2163 * met.
2164 *
2165 * Similar to remap_pfn_range() (see mm/memory.c)
2166 */
2167int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2168 void *kaddr, unsigned long size)
2169{
2170 struct vm_struct *area;
2171
2172 size = PAGE_ALIGN(size);
2173
2174 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2175 return -EINVAL;
2176
2177 area = find_vm_area(kaddr);
2178 if (!area)
2179 return -EINVAL;
2180
2181 if (!(area->flags & VM_USERMAP))
2182 return -EINVAL;
2183
2184 if (kaddr + size > area->addr + area->size)
2185 return -EINVAL;
2186
2187 do {
2188 struct page *page = vmalloc_to_page(kaddr);
2189 int ret;
2190
2191 ret = vm_insert_page(vma, uaddr, page);
2192 if (ret)
2193 return ret;
2194
2195 uaddr += PAGE_SIZE;
2196 kaddr += PAGE_SIZE;
2197 size -= PAGE_SIZE;
2198 } while (size > 0);
2199
2200 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2201
2202 return 0;
2203}
2204EXPORT_SYMBOL(remap_vmalloc_range_partial);
2205
2206/**
Nick Piggin83342312006-06-23 02:03:20 -07002207 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002208 * @vma: vma to cover (map full range of vma)
2209 * @addr: vmalloc memory
2210 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002211 *
2212 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002213 *
2214 * This function checks that addr is a valid vmalloc'ed area, and
2215 * that it is big enough to cover the vma. Will return failure if
2216 * that criteria isn't met.
2217 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002218 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002219 */
2220int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2221 unsigned long pgoff)
2222{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002223 return remap_vmalloc_range_partial(vma, vma->vm_start,
2224 addr + (pgoff << PAGE_SHIFT),
2225 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002226}
2227EXPORT_SYMBOL(remap_vmalloc_range);
2228
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002229/*
2230 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2231 * have one.
2232 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002233void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002234{
2235}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002236
2237
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002238static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002239{
David Vrabelcd129092011-09-29 16:53:32 +01002240 pte_t ***p = data;
2241
2242 if (p) {
2243 *(*p) = pte;
2244 (*p)++;
2245 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002246 return 0;
2247}
2248
2249/**
2250 * alloc_vm_area - allocate a range of kernel address space
2251 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002252 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002253 *
2254 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002255 *
2256 * This function reserves a range of kernel address space, and
2257 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002258 * are created.
2259 *
2260 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2261 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002262 */
David Vrabelcd129092011-09-29 16:53:32 +01002263struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002264{
2265 struct vm_struct *area;
2266
Christoph Lameter23016962008-04-28 02:12:42 -07002267 area = get_vm_area_caller(size, VM_IOREMAP,
2268 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002269 if (area == NULL)
2270 return NULL;
2271
2272 /*
2273 * This ensures that page tables are constructed for this region
2274 * of kernel virtual address space and mapped into init_mm.
2275 */
2276 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002277 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002278 free_vm_area(area);
2279 return NULL;
2280 }
2281
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002282 return area;
2283}
2284EXPORT_SYMBOL_GPL(alloc_vm_area);
2285
2286void free_vm_area(struct vm_struct *area)
2287{
2288 struct vm_struct *ret;
2289 ret = remove_vm_area(area->addr);
2290 BUG_ON(ret != area);
2291 kfree(area);
2292}
2293EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002294
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002295#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002296static struct vmap_area *node_to_va(struct rb_node *n)
2297{
2298 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2299}
2300
2301/**
2302 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2303 * @end: target address
2304 * @pnext: out arg for the next vmap_area
2305 * @pprev: out arg for the previous vmap_area
2306 *
2307 * Returns: %true if either or both of next and prev are found,
2308 * %false if no vmap_area exists
2309 *
2310 * Find vmap_areas end addresses of which enclose @end. ie. if not
2311 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2312 */
2313static bool pvm_find_next_prev(unsigned long end,
2314 struct vmap_area **pnext,
2315 struct vmap_area **pprev)
2316{
2317 struct rb_node *n = vmap_area_root.rb_node;
2318 struct vmap_area *va = NULL;
2319
2320 while (n) {
2321 va = rb_entry(n, struct vmap_area, rb_node);
2322 if (end < va->va_end)
2323 n = n->rb_left;
2324 else if (end > va->va_end)
2325 n = n->rb_right;
2326 else
2327 break;
2328 }
2329
2330 if (!va)
2331 return false;
2332
2333 if (va->va_end > end) {
2334 *pnext = va;
2335 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2336 } else {
2337 *pprev = va;
2338 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2339 }
2340 return true;
2341}
2342
2343/**
2344 * pvm_determine_end - find the highest aligned address between two vmap_areas
2345 * @pnext: in/out arg for the next vmap_area
2346 * @pprev: in/out arg for the previous vmap_area
2347 * @align: alignment
2348 *
2349 * Returns: determined end address
2350 *
2351 * Find the highest aligned address between *@pnext and *@pprev below
2352 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2353 * down address is between the end addresses of the two vmap_areas.
2354 *
2355 * Please note that the address returned by this function may fall
2356 * inside *@pnext vmap_area. The caller is responsible for checking
2357 * that.
2358 */
2359static unsigned long pvm_determine_end(struct vmap_area **pnext,
2360 struct vmap_area **pprev,
2361 unsigned long align)
2362{
2363 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2364 unsigned long addr;
2365
2366 if (*pnext)
2367 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2368 else
2369 addr = vmalloc_end;
2370
2371 while (*pprev && (*pprev)->va_end > addr) {
2372 *pnext = *pprev;
2373 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2374 }
2375
2376 return addr;
2377}
2378
2379/**
2380 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2381 * @offsets: array containing offset of each area
2382 * @sizes: array containing size of each area
2383 * @nr_vms: the number of areas to allocate
2384 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002385 *
2386 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2387 * vm_structs on success, %NULL on failure
2388 *
2389 * Percpu allocator wants to use congruent vm areas so that it can
2390 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002391 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2392 * be scattered pretty far, distance between two areas easily going up
2393 * to gigabytes. To avoid interacting with regular vmallocs, these
2394 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002395 *
2396 * Despite its complicated look, this allocator is rather simple. It
2397 * does everything top-down and scans areas from the end looking for
2398 * matching slot. While scanning, if any of the areas overlaps with
2399 * existing vmap_area, the base address is pulled down to fit the
2400 * area. Scanning is repeated till all the areas fit and then all
2401 * necessary data structres are inserted and the result is returned.
2402 */
2403struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2404 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002405 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002406{
2407 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2408 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2409 struct vmap_area **vas, *prev, *next;
2410 struct vm_struct **vms;
2411 int area, area2, last_area, term_area;
2412 unsigned long base, start, end, last_end;
2413 bool purged = false;
2414
Tejun Heoca23e402009-08-14 15:00:52 +09002415 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002416 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002417 for (last_area = 0, area = 0; area < nr_vms; area++) {
2418 start = offsets[area];
2419 end = start + sizes[area];
2420
2421 /* is everything aligned properly? */
2422 BUG_ON(!IS_ALIGNED(offsets[area], align));
2423 BUG_ON(!IS_ALIGNED(sizes[area], align));
2424
2425 /* detect the area with the highest address */
2426 if (start > offsets[last_area])
2427 last_area = area;
2428
2429 for (area2 = 0; area2 < nr_vms; area2++) {
2430 unsigned long start2 = offsets[area2];
2431 unsigned long end2 = start2 + sizes[area2];
2432
2433 if (area2 == area)
2434 continue;
2435
2436 BUG_ON(start2 >= start && start2 < end);
2437 BUG_ON(end2 <= end && end2 > start);
2438 }
2439 }
2440 last_end = offsets[last_area] + sizes[last_area];
2441
2442 if (vmalloc_end - vmalloc_start < last_end) {
2443 WARN_ON(true);
2444 return NULL;
2445 }
2446
Thomas Meyer4d67d862012-05-29 15:06:21 -07002447 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2448 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002449 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002450 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002451
2452 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002453 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2454 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002455 if (!vas[area] || !vms[area])
2456 goto err_free;
2457 }
2458retry:
2459 spin_lock(&vmap_area_lock);
2460
2461 /* start scanning - we scan from the top, begin with the last area */
2462 area = term_area = last_area;
2463 start = offsets[area];
2464 end = start + sizes[area];
2465
2466 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2467 base = vmalloc_end - last_end;
2468 goto found;
2469 }
2470 base = pvm_determine_end(&next, &prev, align) - end;
2471
2472 while (true) {
2473 BUG_ON(next && next->va_end <= base + end);
2474 BUG_ON(prev && prev->va_end > base + end);
2475
2476 /*
2477 * base might have underflowed, add last_end before
2478 * comparing.
2479 */
2480 if (base + last_end < vmalloc_start + last_end) {
2481 spin_unlock(&vmap_area_lock);
2482 if (!purged) {
2483 purge_vmap_area_lazy();
2484 purged = true;
2485 goto retry;
2486 }
2487 goto err_free;
2488 }
2489
2490 /*
2491 * If next overlaps, move base downwards so that it's
2492 * right below next and then recheck.
2493 */
2494 if (next && next->va_start < base + end) {
2495 base = pvm_determine_end(&next, &prev, align) - end;
2496 term_area = area;
2497 continue;
2498 }
2499
2500 /*
2501 * If prev overlaps, shift down next and prev and move
2502 * base so that it's right below new next and then
2503 * recheck.
2504 */
2505 if (prev && prev->va_end > base + start) {
2506 next = prev;
2507 prev = node_to_va(rb_prev(&next->rb_node));
2508 base = pvm_determine_end(&next, &prev, align) - end;
2509 term_area = area;
2510 continue;
2511 }
2512
2513 /*
2514 * This area fits, move on to the previous one. If
2515 * the previous one is the terminal one, we're done.
2516 */
2517 area = (area + nr_vms - 1) % nr_vms;
2518 if (area == term_area)
2519 break;
2520 start = offsets[area];
2521 end = start + sizes[area];
2522 pvm_find_next_prev(base + end, &next, &prev);
2523 }
2524found:
2525 /* we've found a fitting base, insert all va's */
2526 for (area = 0; area < nr_vms; area++) {
2527 struct vmap_area *va = vas[area];
2528
2529 va->va_start = base + offsets[area];
2530 va->va_end = va->va_start + sizes[area];
2531 __insert_vmap_area(va);
2532 }
2533
2534 vmap_area_pcpu_hole = base + offsets[last_area];
2535
2536 spin_unlock(&vmap_area_lock);
2537
2538 /* insert all vm's */
2539 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002540 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2541 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002542
2543 kfree(vas);
2544 return vms;
2545
2546err_free:
2547 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002548 kfree(vas[area]);
2549 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002550 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002551err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002552 kfree(vas);
2553 kfree(vms);
2554 return NULL;
2555}
2556
2557/**
2558 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2559 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2560 * @nr_vms: the number of allocated areas
2561 *
2562 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2563 */
2564void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2565{
2566 int i;
2567
2568 for (i = 0; i < nr_vms; i++)
2569 free_vm_area(vms[i]);
2570 kfree(vms);
2571}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002572#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002573
2574#ifdef CONFIG_PROC_FS
2575static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002576 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002577{
2578 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002579 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002580
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002581 spin_lock(&vmap_area_lock);
Geliang Tang6219c2a2016-01-14 15:19:08 -08002582 va = list_first_entry(&vmap_area_list, typeof(*va), list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002583 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002584 n--;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002585 va = list_next_entry(va, list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002586 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002587 if (!n && &va->list != &vmap_area_list)
2588 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002589
2590 return NULL;
2591
2592}
2593
2594static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2595{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002596 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002597
2598 ++*pos;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002599 next = list_next_entry(va, list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002600 if (&next->list != &vmap_area_list)
2601 return next;
2602
2603 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002604}
2605
2606static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002607 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002608{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002609 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002610}
2611
Eric Dumazeta47a1262008-07-23 21:27:38 -07002612static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2613{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002614 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002615 unsigned int nr, *counters = m->private;
2616
2617 if (!counters)
2618 return;
2619
Wanpeng Liaf123462013-11-12 15:07:32 -08002620 if (v->flags & VM_UNINITIALIZED)
2621 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002622 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2623 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002624
Eric Dumazeta47a1262008-07-23 21:27:38 -07002625 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2626
2627 for (nr = 0; nr < v->nr_pages; nr++)
2628 counters[page_to_nid(v->pages[nr])]++;
2629
2630 for_each_node_state(nr, N_HIGH_MEMORY)
2631 if (counters[nr])
2632 seq_printf(m, " N%u=%u", nr, counters[nr]);
2633 }
2634}
2635
Christoph Lametera10aa572008-04-28 02:12:40 -07002636static int s_show(struct seq_file *m, void *p)
2637{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002638 struct vmap_area *va = p;
2639 struct vm_struct *v;
2640
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002641 /*
2642 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2643 * behalf of vmap area is being tear down or vm_map_ram allocation.
2644 */
2645 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002646 return 0;
2647
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002648 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002649
Kees Cook45ec1692012-10-08 16:34:09 -07002650 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002651 v->addr, v->addr + v->size, v->size);
2652
Joe Perches62c70bc2011-01-13 15:45:52 -08002653 if (v->caller)
2654 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002655
Christoph Lametera10aa572008-04-28 02:12:40 -07002656 if (v->nr_pages)
2657 seq_printf(m, " pages=%d", v->nr_pages);
2658
2659 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002660 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002661
2662 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002663 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002664
2665 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002666 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002667
2668 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002669 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002670
2671 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002672 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002673
David Rientjes244d63e2016-01-14 15:19:35 -08002674 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002675 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002676
Eric Dumazeta47a1262008-07-23 21:27:38 -07002677 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002678 seq_putc(m, '\n');
2679 return 0;
2680}
2681
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002682static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002683 .start = s_start,
2684 .next = s_next,
2685 .stop = s_stop,
2686 .show = s_show,
2687};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002688
2689static int vmalloc_open(struct inode *inode, struct file *file)
2690{
Rob Jones703394c2014-10-09 15:28:01 -07002691 if (IS_ENABLED(CONFIG_NUMA))
2692 return seq_open_private(file, &vmalloc_op,
2693 nr_node_ids * sizeof(unsigned int));
2694 else
2695 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002696}
2697
2698static const struct file_operations proc_vmalloc_operations = {
2699 .open = vmalloc_open,
2700 .read = seq_read,
2701 .llseek = seq_lseek,
2702 .release = seq_release_private,
2703};
2704
2705static int __init proc_vmalloc_init(void)
2706{
2707 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2708 return 0;
2709}
2710module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002711
Christoph Lametera10aa572008-04-28 02:12:40 -07002712#endif
2713