blob: e6aa073f01dfd8be2c686f4e7792d653eec4869f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/vmalloc.c
3 *
4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
Christoph Lameter930fc452005-10-29 18:15:41 -07008 * Numa awareness, Christoph Lameter, SGI, June 2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Nick Piggindb64fe02008-10-18 20:27:03 -070011#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mm.h>
13#include <linux/module.h>
14#include <linux/highmem.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040015#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/spinlock.h>
18#include <linux/interrupt.h>
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +040019#include <linux/proc_fs.h>
Christoph Lametera10aa572008-04-28 02:12:40 -070020#include <linux/seq_file.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070021#include <linux/debugobjects.h>
Christoph Lameter23016962008-04-28 02:12:42 -070022#include <linux/kallsyms.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070023#include <linux/list.h>
Chris Wilson4da56b92016-04-04 14:46:42 +010024#include <linux/notifier.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070025#include <linux/rbtree.h>
26#include <linux/radix-tree.h>
27#include <linux/rcupdate.h>
Tejun Heof0aa6612009-02-20 16:29:08 +090028#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010029#include <linux/kmemleak.h>
Arun Sharma600634972011-07-26 16:09:06 -070030#include <linux/atomic.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070031#include <linux/compiler.h>
Al Viro32fcfd42013-03-10 20:14:08 -040032#include <linux/llist.h>
Toshi Kani0f616be2015-04-14 15:47:17 -070033#include <linux/bitops.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36#include <asm/tlbflush.h>
David Miller2dca6992009-09-21 12:22:34 -070037#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Mel Gormandd56b042015-11-06 16:28:43 -080039#include "internal.h"
40
Al Viro32fcfd42013-03-10 20:14:08 -040041struct vfree_deferred {
42 struct llist_head list;
43 struct work_struct wq;
44};
45static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
46
47static void __vunmap(const void *, int);
48
49static void free_work(struct work_struct *w)
50{
51 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
52 struct llist_node *llnode = llist_del_all(&p->list);
53 while (llnode) {
54 void *p = llnode;
55 llnode = llist_next(llnode);
56 __vunmap(p, 1);
57 }
58}
59
Nick Piggindb64fe02008-10-18 20:27:03 -070060/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070061
Linus Torvalds1da177e2005-04-16 15:20:36 -070062static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
63{
64 pte_t *pte;
65
66 pte = pte_offset_kernel(pmd, addr);
67 do {
68 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
69 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
70 } while (pte++, addr += PAGE_SIZE, addr != end);
71}
72
Nick Piggindb64fe02008-10-18 20:27:03 -070073static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 pmd_t *pmd;
76 unsigned long next;
77
78 pmd = pmd_offset(pud, addr);
79 do {
80 next = pmd_addr_end(addr, end);
Toshi Kanib9820d82015-04-14 15:47:26 -070081 if (pmd_clear_huge(pmd))
82 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 if (pmd_none_or_clear_bad(pmd))
84 continue;
85 vunmap_pte_range(pmd, addr, next);
86 } while (pmd++, addr = next, addr != end);
87}
88
Nick Piggindb64fe02008-10-18 20:27:03 -070089static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
91 pud_t *pud;
92 unsigned long next;
93
94 pud = pud_offset(pgd, addr);
95 do {
96 next = pud_addr_end(addr, end);
Toshi Kanib9820d82015-04-14 15:47:26 -070097 if (pud_clear_huge(pud))
98 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 if (pud_none_or_clear_bad(pud))
100 continue;
101 vunmap_pmd_range(pud, addr, next);
102 } while (pud++, addr = next, addr != end);
103}
104
Nick Piggindb64fe02008-10-18 20:27:03 -0700105static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
107 pgd_t *pgd;
108 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110 BUG_ON(addr >= end);
111 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 do {
113 next = pgd_addr_end(addr, end);
114 if (pgd_none_or_clear_bad(pgd))
115 continue;
116 vunmap_pud_range(pgd, addr, next);
117 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
119
120static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -0700121 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 pte_t *pte;
124
Nick Piggindb64fe02008-10-18 20:27:03 -0700125 /*
126 * nr is a running index into the array which helps higher level
127 * callers keep track of where we're up to.
128 */
129
Hugh Dickins872fec12005-10-29 18:16:21 -0700130 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (!pte)
132 return -ENOMEM;
133 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700134 struct page *page = pages[*nr];
135
136 if (WARN_ON(!pte_none(*pte)))
137 return -EBUSY;
138 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return -ENOMEM;
140 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700141 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 } while (pte++, addr += PAGE_SIZE, addr != end);
143 return 0;
144}
145
Nick Piggindb64fe02008-10-18 20:27:03 -0700146static int vmap_pmd_range(pud_t *pud, unsigned long addr,
147 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 pmd_t *pmd;
150 unsigned long next;
151
152 pmd = pmd_alloc(&init_mm, pud, addr);
153 if (!pmd)
154 return -ENOMEM;
155 do {
156 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700157 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 return -ENOMEM;
159 } while (pmd++, addr = next, addr != end);
160 return 0;
161}
162
Nick Piggindb64fe02008-10-18 20:27:03 -0700163static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
164 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
166 pud_t *pud;
167 unsigned long next;
168
169 pud = pud_alloc(&init_mm, pgd, addr);
170 if (!pud)
171 return -ENOMEM;
172 do {
173 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700174 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return -ENOMEM;
176 } while (pud++, addr = next, addr != end);
177 return 0;
178}
179
Nick Piggindb64fe02008-10-18 20:27:03 -0700180/*
181 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
182 * will have pfns corresponding to the "pages" array.
183 *
184 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
185 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900186static int vmap_page_range_noflush(unsigned long start, unsigned long end,
187 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
189 pgd_t *pgd;
190 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800191 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700192 int err = 0;
193 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 BUG_ON(addr >= end);
196 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 do {
198 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700199 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700201 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700203
Nick Piggindb64fe02008-10-18 20:27:03 -0700204 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
Tejun Heo8fc48982009-02-20 16:29:08 +0900207static int vmap_page_range(unsigned long start, unsigned long end,
208 pgprot_t prot, struct page **pages)
209{
210 int ret;
211
212 ret = vmap_page_range_noflush(start, end, prot, pages);
213 flush_cache_vmap(start, end);
214 return ret;
215}
216
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700217int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700218{
219 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000220 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700221 * and fall back on vmalloc() if that fails. Others
222 * just put it in the vmalloc space.
223 */
224#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
225 unsigned long addr = (unsigned long)x;
226 if (addr >= MODULES_VADDR && addr < MODULES_END)
227 return 1;
228#endif
229 return is_vmalloc_addr(x);
230}
231
Christoph Lameter48667e72008-02-04 22:28:31 -0800232/*
malcadd688f2014-01-27 17:06:53 -0800233 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800234 */
malcadd688f2014-01-27 17:06:53 -0800235struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800236{
237 unsigned long addr = (unsigned long) vmalloc_addr;
malcadd688f2014-01-27 17:06:53 -0800238 struct page *page = NULL;
Christoph Lameter48667e72008-02-04 22:28:31 -0800239 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800240
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200241 /*
242 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
243 * architectures that do not vmalloc module space
244 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700245 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200246
Ard Biesheuvel647f6052017-06-23 15:08:41 -0700247 /*
248 * Don't dereference bad PUD or PMD (below) entries. This will also
249 * identify huge mappings, which we may encounter on architectures
250 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
251 * identified as vmalloc addresses by is_vmalloc_addr(), but are
252 * not [unambiguously] associated with a struct page, so there is
253 * no correct value to return for them.
254 */
Christoph Lameter48667e72008-02-04 22:28:31 -0800255 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700256 pud_t *pud = pud_offset(pgd, addr);
Ard Biesheuvel647f6052017-06-23 15:08:41 -0700257 WARN_ON_ONCE(pud_bad(*pud));
258 if (!pud_none(*pud) && !pud_bad(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700259 pmd_t *pmd = pmd_offset(pud, addr);
Ard Biesheuvel647f6052017-06-23 15:08:41 -0700260 WARN_ON_ONCE(pmd_bad(*pmd));
261 if (!pmd_none(*pmd) && !pmd_bad(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700262 pte_t *ptep, pte;
263
Christoph Lameter48667e72008-02-04 22:28:31 -0800264 ptep = pte_offset_map(pmd, addr);
265 pte = *ptep;
266 if (pte_present(pte))
malcadd688f2014-01-27 17:06:53 -0800267 page = pte_page(pte);
Christoph Lameter48667e72008-02-04 22:28:31 -0800268 pte_unmap(ptep);
269 }
270 }
271 }
malcadd688f2014-01-27 17:06:53 -0800272 return page;
Jianyu Zhanece86e222014-01-21 15:49:12 -0800273}
274EXPORT_SYMBOL(vmalloc_to_page);
275
malcadd688f2014-01-27 17:06:53 -0800276/*
277 * Map a vmalloc()-space virtual address to the physical page frame number.
278 */
279unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
280{
281 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
282}
283EXPORT_SYMBOL(vmalloc_to_pfn);
284
Nick Piggindb64fe02008-10-18 20:27:03 -0700285
286/*** Global kva allocator ***/
287
Nick Piggindb64fe02008-10-18 20:27:03 -0700288#define VM_VM_AREA 0x04
289
Nick Piggindb64fe02008-10-18 20:27:03 -0700290static DEFINE_SPINLOCK(vmap_area_lock);
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700291/* Export for kexec only */
292LIST_HEAD(vmap_area_list);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700293static LLIST_HEAD(vmap_purge_list);
Nick Piggin89699602011-03-22 16:30:36 -0700294static struct rb_root vmap_area_root = RB_ROOT;
295
296/* The vmap cache globals are protected by vmap_area_lock */
297static struct rb_node *free_vmap_cache;
298static unsigned long cached_hole_size;
299static unsigned long cached_vstart;
300static unsigned long cached_align;
301
Tejun Heoca23e402009-08-14 15:00:52 +0900302static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700303
304static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Nick Piggindb64fe02008-10-18 20:27:03 -0700306 struct rb_node *n = vmap_area_root.rb_node;
307
308 while (n) {
309 struct vmap_area *va;
310
311 va = rb_entry(n, struct vmap_area, rb_node);
312 if (addr < va->va_start)
313 n = n->rb_left;
HATAYAMA Daisukecef2ac32013-07-03 15:02:17 -0700314 else if (addr >= va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700315 n = n->rb_right;
316 else
317 return va;
318 }
319
320 return NULL;
321}
322
323static void __insert_vmap_area(struct vmap_area *va)
324{
325 struct rb_node **p = &vmap_area_root.rb_node;
326 struct rb_node *parent = NULL;
327 struct rb_node *tmp;
328
329 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700330 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700331
332 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700333 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
334 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700335 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700336 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700337 p = &(*p)->rb_right;
338 else
339 BUG();
340 }
341
342 rb_link_node(&va->rb_node, parent, p);
343 rb_insert_color(&va->rb_node, &vmap_area_root);
344
Joonsoo Kim4341fa42013-04-29 15:07:39 -0700345 /* address-sort this list */
Nick Piggindb64fe02008-10-18 20:27:03 -0700346 tmp = rb_prev(&va->rb_node);
347 if (tmp) {
348 struct vmap_area *prev;
349 prev = rb_entry(tmp, struct vmap_area, rb_node);
350 list_add_rcu(&va->list, &prev->list);
351 } else
352 list_add_rcu(&va->list, &vmap_area_list);
353}
354
355static void purge_vmap_area_lazy(void);
356
Chris Wilson4da56b92016-04-04 14:46:42 +0100357static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
358
Nick Piggindb64fe02008-10-18 20:27:03 -0700359/*
360 * Allocate a region of KVA of the specified size and alignment, within the
361 * vstart and vend.
362 */
363static struct vmap_area *alloc_vmap_area(unsigned long size,
364 unsigned long align,
365 unsigned long vstart, unsigned long vend,
366 int node, gfp_t gfp_mask)
367{
368 struct vmap_area *va;
369 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700371 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700372 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700373
Nick Piggin77669702009-02-27 14:03:03 -0800374 BUG_ON(!size);
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800375 BUG_ON(offset_in_page(size));
Nick Piggin89699602011-03-22 16:30:36 -0700376 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700377
Chris Wilson4da56b92016-04-04 14:46:42 +0100378 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
379
Nick Piggindb64fe02008-10-18 20:27:03 -0700380 va = kmalloc_node(sizeof(struct vmap_area),
381 gfp_mask & GFP_RECLAIM_MASK, node);
382 if (unlikely(!va))
383 return ERR_PTR(-ENOMEM);
384
Catalin Marinas7f88f882013-11-12 15:07:45 -0800385 /*
386 * Only scan the relevant parts containing pointers to other objects
387 * to avoid false negatives.
388 */
389 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
390
Nick Piggindb64fe02008-10-18 20:27:03 -0700391retry:
392 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700393 /*
394 * Invalidate cache if we have more permissive parameters.
395 * cached_hole_size notes the largest hole noticed _below_
396 * the vmap_area cached in free_vmap_cache: if size fits
397 * into that hole, we want to scan from vstart to reuse
398 * the hole instead of allocating above free_vmap_cache.
399 * Note that __free_vmap_area may update free_vmap_cache
400 * without updating cached_hole_size or cached_align.
401 */
402 if (!free_vmap_cache ||
403 size < cached_hole_size ||
404 vstart < cached_vstart ||
405 align < cached_align) {
406nocache:
407 cached_hole_size = 0;
408 free_vmap_cache = NULL;
409 }
410 /* record if we encounter less permissive parameters */
411 cached_vstart = vstart;
412 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800413
Nick Piggin89699602011-03-22 16:30:36 -0700414 /* find starting point for our search */
415 if (free_vmap_cache) {
416 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700417 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700418 if (addr < vstart)
419 goto nocache;
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700420 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700421 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700422
Nick Piggin89699602011-03-22 16:30:36 -0700423 } else {
424 addr = ALIGN(vstart, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700425 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700426 goto overflow;
427
428 n = vmap_area_root.rb_node;
429 first = NULL;
430
431 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700432 struct vmap_area *tmp;
433 tmp = rb_entry(n, struct vmap_area, rb_node);
434 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700435 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700436 if (tmp->va_start <= addr)
437 break;
438 n = n->rb_left;
439 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700440 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700441 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700442
443 if (!first)
444 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700445 }
Nick Piggin89699602011-03-22 16:30:36 -0700446
447 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700448 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700449 if (addr + cached_hole_size < first->va_start)
450 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700451 addr = ALIGN(first->va_end, align);
Zhang Yanfeibcb615a2013-07-08 16:00:19 -0700452 if (addr + size < addr)
Nick Piggin89699602011-03-22 16:30:36 -0700453 goto overflow;
454
Hong zhi guo92ca9222012-07-31 16:41:35 -0700455 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700456 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700457
Geliang Tang6219c2a2016-01-14 15:19:08 -0800458 first = list_next_entry(first, list);
Nick Piggin89699602011-03-22 16:30:36 -0700459 }
460
Nick Piggindb64fe02008-10-18 20:27:03 -0700461found:
Nick Piggin89699602011-03-22 16:30:36 -0700462 if (addr + size > vend)
463 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700464
465 va->va_start = addr;
466 va->va_end = addr + size;
467 va->flags = 0;
468 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700469 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700470 spin_unlock(&vmap_area_lock);
471
Wang Xiaoqiang61e16552016-01-15 16:57:19 -0800472 BUG_ON(!IS_ALIGNED(va->va_start, align));
Nick Piggin89699602011-03-22 16:30:36 -0700473 BUG_ON(va->va_start < vstart);
474 BUG_ON(va->va_end > vend);
475
Nick Piggindb64fe02008-10-18 20:27:03 -0700476 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700477
478overflow:
479 spin_unlock(&vmap_area_lock);
480 if (!purged) {
481 purge_vmap_area_lazy();
482 purged = 1;
483 goto retry;
484 }
Chris Wilson4da56b92016-04-04 14:46:42 +0100485
486 if (gfpflags_allow_blocking(gfp_mask)) {
487 unsigned long freed = 0;
488 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
489 if (freed > 0) {
490 purged = 0;
491 goto retry;
492 }
493 }
494
Nick Piggin89699602011-03-22 16:30:36 -0700495 if (printk_ratelimit())
Joe Perches756a025f02016-03-17 14:19:47 -0700496 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
497 size);
Nick Piggin89699602011-03-22 16:30:36 -0700498 kfree(va);
499 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700500}
501
Chris Wilson4da56b92016-04-04 14:46:42 +0100502int register_vmap_purge_notifier(struct notifier_block *nb)
503{
504 return blocking_notifier_chain_register(&vmap_notify_list, nb);
505}
506EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
507
508int unregister_vmap_purge_notifier(struct notifier_block *nb)
509{
510 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
511}
512EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
513
Nick Piggindb64fe02008-10-18 20:27:03 -0700514static void __free_vmap_area(struct vmap_area *va)
515{
516 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700517
518 if (free_vmap_cache) {
519 if (va->va_end < cached_vstart) {
520 free_vmap_cache = NULL;
521 } else {
522 struct vmap_area *cache;
523 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
524 if (va->va_start <= cache->va_start) {
525 free_vmap_cache = rb_prev(&va->rb_node);
526 /*
527 * We don't try to update cached_hole_size or
528 * cached_align, but it won't go very wrong.
529 */
530 }
531 }
532 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700533 rb_erase(&va->rb_node, &vmap_area_root);
534 RB_CLEAR_NODE(&va->rb_node);
535 list_del_rcu(&va->list);
536
Tejun Heoca23e402009-08-14 15:00:52 +0900537 /*
538 * Track the highest possible candidate for pcpu area
539 * allocation. Areas outside of vmalloc area can be returned
540 * here too, consider only end addresses which fall inside
541 * vmalloc area proper.
542 */
543 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
544 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
545
Lai Jiangshan14769de2011-03-18 12:12:19 +0800546 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700547}
548
549/*
550 * Free a region of KVA allocated by alloc_vmap_area
551 */
552static void free_vmap_area(struct vmap_area *va)
553{
554 spin_lock(&vmap_area_lock);
555 __free_vmap_area(va);
556 spin_unlock(&vmap_area_lock);
557}
558
559/*
560 * Clear the pagetable entries of a given vmap_area
561 */
562static void unmap_vmap_area(struct vmap_area *va)
563{
564 vunmap_page_range(va->va_start, va->va_end);
565}
566
Nick Piggincd528582009-01-06 14:39:20 -0800567static void vmap_debug_free_range(unsigned long start, unsigned long end)
568{
569 /*
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700570 * Unmap page tables and force a TLB flush immediately if pagealloc
571 * debugging is enabled. This catches use after free bugs similarly to
572 * those in linear kernel virtual address space after a page has been
573 * freed.
Nick Piggincd528582009-01-06 14:39:20 -0800574 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700575 * All the lazy freeing logic is still retained, in order to minimise
576 * intrusiveness of this debugging feature.
Nick Piggincd528582009-01-06 14:39:20 -0800577 *
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700578 * This is going to be *slow* (linear kernel virtual address debugging
579 * doesn't do a broadcast TLB flush so it is a lot faster).
Nick Piggincd528582009-01-06 14:39:20 -0800580 */
Joonsoo Kimf48d97f2016-03-17 14:17:49 -0700581 if (debug_pagealloc_enabled()) {
582 vunmap_page_range(start, end);
583 flush_tlb_kernel_range(start, end);
584 }
Nick Piggincd528582009-01-06 14:39:20 -0800585}
586
Nick Piggindb64fe02008-10-18 20:27:03 -0700587/*
588 * lazy_max_pages is the maximum amount of virtual address space we gather up
589 * before attempting to purge with a TLB flush.
590 *
591 * There is a tradeoff here: a larger number will cover more kernel page tables
592 * and take slightly longer to purge, but it will linearly reduce the number of
593 * global TLB flushes that must be performed. It would seem natural to scale
594 * this number up linearly with the number of CPUs (because vmapping activity
595 * could also scale linearly with the number of CPUs), however it is likely
596 * that in practice, workloads might be constrained in other ways that mean
597 * vmap activity will not scale linearly with CPUs. Also, I want to be
598 * conservative and not introduce a big latency on huge systems, so go with
599 * a less aggressive log scale. It will still be an improvement over the old
600 * code, and it will be simple to change the scale factor if we find that it
601 * becomes a problem on bigger systems.
602 */
603static unsigned long lazy_max_pages(void)
604{
605 unsigned int log;
606
607 log = fls(num_online_cpus());
608
609 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
610}
611
612static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
613
Nick Piggin02b709d2010-02-01 22:25:57 +1100614/* for per-CPU blocks */
615static void purge_fragmented_blocks_allcpus(void);
616
Nick Piggindb64fe02008-10-18 20:27:03 -0700617/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500618 * called before a call to iounmap() if the caller wants vm_area_struct's
619 * immediately freed.
620 */
621void set_iounmap_nonlazy(void)
622{
623 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
624}
625
626/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700627 * Purges all lazily-freed vmap areas.
628 *
629 * If sync is 0 then don't purge if there is already a purge in progress.
630 * If force_flush is 1, then flush kernel TLBs between *start and *end even
631 * if we found no lazy vmap areas to unmap (callers can use this to optimise
632 * their own TLB flushing).
633 * Returns with *start = min(*start, lowest purged address)
634 * *end = max(*end, highest purged address)
635 */
636static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
637 int sync, int force_flush)
638{
Andrew Morton46666d82009-01-15 13:51:15 -0800639 static DEFINE_SPINLOCK(purge_lock);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700640 struct llist_node *valist;
Nick Piggindb64fe02008-10-18 20:27:03 -0700641 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800642 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700643 int nr = 0;
644
645 /*
646 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
647 * should not expect such behaviour. This just simplifies locking for
648 * the case that isn't actually used at the moment anyway.
649 */
650 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800651 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700652 return;
653 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800654 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700655
Nick Piggin02b709d2010-02-01 22:25:57 +1100656 if (sync)
657 purge_fragmented_blocks_allcpus();
658
Chris Wilson80c4bd72016-05-20 16:57:38 -0700659 valist = llist_del_all(&vmap_purge_list);
660 llist_for_each_entry(va, valist, purge_list) {
661 if (va->va_start < *start)
662 *start = va->va_start;
663 if (va->va_end > *end)
664 *end = va->va_end;
665 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700666 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700667
Yongseok Koh88f50042010-01-19 17:33:49 +0900668 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700669 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700670
671 if (nr || force_flush)
672 flush_tlb_kernel_range(*start, *end);
673
674 if (nr) {
675 spin_lock(&vmap_area_lock);
Chris Wilson80c4bd72016-05-20 16:57:38 -0700676 llist_for_each_entry_safe(va, n_va, valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700677 __free_vmap_area(va);
678 spin_unlock(&vmap_area_lock);
679 }
Andrew Morton46666d82009-01-15 13:51:15 -0800680 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700681}
682
683/*
Nick Piggin496850e2008-11-19 15:36:33 -0800684 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
685 * is already purging.
686 */
687static void try_purge_vmap_area_lazy(void)
688{
689 unsigned long start = ULONG_MAX, end = 0;
690
691 __purge_vmap_area_lazy(&start, &end, 0, 0);
692}
693
694/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700695 * Kick off a purge of the outstanding lazy areas.
696 */
697static void purge_vmap_area_lazy(void)
698{
699 unsigned long start = ULONG_MAX, end = 0;
700
Nick Piggin496850e2008-11-19 15:36:33 -0800701 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700702}
703
704/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800705 * Free a vmap area, caller ensuring that the area has been unmapped
706 * and flush_cache_vunmap had been called for the correct range
707 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700708 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800709static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700710{
Chris Wilson80c4bd72016-05-20 16:57:38 -0700711 int nr_lazy;
712
713 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
714 &vmap_lazy_nr);
715
716 /* After this point, we may free va at any time */
717 llist_add(&va->purge_list, &vmap_purge_list);
718
719 if (unlikely(nr_lazy > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800720 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700721}
722
Nick Pigginb29acbd2008-12-01 13:13:47 -0800723/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800724 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
725 * called for the correct range previously.
726 */
727static void free_unmap_vmap_area_noflush(struct vmap_area *va)
728{
729 unmap_vmap_area(va);
730 free_vmap_area_noflush(va);
731}
732
733/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800734 * Free and unmap a vmap area
735 */
736static void free_unmap_vmap_area(struct vmap_area *va)
737{
738 flush_cache_vunmap(va->va_start, va->va_end);
739 free_unmap_vmap_area_noflush(va);
740}
741
Nick Piggindb64fe02008-10-18 20:27:03 -0700742static struct vmap_area *find_vmap_area(unsigned long addr)
743{
744 struct vmap_area *va;
745
746 spin_lock(&vmap_area_lock);
747 va = __find_vmap_area(addr);
748 spin_unlock(&vmap_area_lock);
749
750 return va;
751}
752
753static void free_unmap_vmap_area_addr(unsigned long addr)
754{
755 struct vmap_area *va;
756
757 va = find_vmap_area(addr);
758 BUG_ON(!va);
759 free_unmap_vmap_area(va);
760}
761
762
763/*** Per cpu kva allocator ***/
764
765/*
766 * vmap space is limited especially on 32 bit architectures. Ensure there is
767 * room for at least 16 percpu vmap blocks per CPU.
768 */
769/*
770 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
771 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
772 * instead (we just need a rough idea)
773 */
774#if BITS_PER_LONG == 32
775#define VMALLOC_SPACE (128UL*1024*1024)
776#else
777#define VMALLOC_SPACE (128UL*1024*1024*1024)
778#endif
779
780#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
781#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
782#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
783#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
784#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
785#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200786#define VMAP_BBMAP_BITS \
787 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
788 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
789 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700790
791#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
792
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100793static bool vmap_initialized __read_mostly = false;
794
Nick Piggindb64fe02008-10-18 20:27:03 -0700795struct vmap_block_queue {
796 spinlock_t lock;
797 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700798};
799
800struct vmap_block {
801 spinlock_t lock;
802 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700803 unsigned long free, dirty;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700804 unsigned long dirty_min, dirty_max; /*< dirty range */
Nick Pigginde560422010-02-01 22:24:18 +1100805 struct list_head free_list;
806 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100807 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700808};
809
810/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
811static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
812
813/*
814 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
815 * in the free path. Could get rid of this if we change the API to return a
816 * "cookie" from alloc, to be passed to free. But no big deal yet.
817 */
818static DEFINE_SPINLOCK(vmap_block_tree_lock);
819static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
820
821/*
822 * We should probably have a fallback mechanism to allocate virtual memory
823 * out of partially filled vmap blocks. However vmap block sizing should be
824 * fairly reasonable according to the vmalloc size, so it shouldn't be a
825 * big problem.
826 */
827
828static unsigned long addr_to_vb_idx(unsigned long addr)
829{
830 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
831 addr /= VMAP_BLOCK_SIZE;
832 return addr;
833}
834
Roman Pencf725ce2015-04-15 16:13:52 -0700835static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
836{
837 unsigned long addr;
838
839 addr = va_start + (pages_off << PAGE_SHIFT);
840 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
841 return (void *)addr;
842}
843
844/**
845 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
846 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
847 * @order: how many 2^order pages should be occupied in newly allocated block
848 * @gfp_mask: flags for the page level allocator
849 *
850 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
851 */
852static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
Nick Piggindb64fe02008-10-18 20:27:03 -0700853{
854 struct vmap_block_queue *vbq;
855 struct vmap_block *vb;
856 struct vmap_area *va;
857 unsigned long vb_idx;
858 int node, err;
Roman Pencf725ce2015-04-15 16:13:52 -0700859 void *vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700860
861 node = numa_node_id();
862
863 vb = kmalloc_node(sizeof(struct vmap_block),
864 gfp_mask & GFP_RECLAIM_MASK, node);
865 if (unlikely(!vb))
866 return ERR_PTR(-ENOMEM);
867
868 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
869 VMALLOC_START, VMALLOC_END,
870 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800871 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700872 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700873 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700874 }
875
876 err = radix_tree_preload(gfp_mask);
877 if (unlikely(err)) {
878 kfree(vb);
879 free_vmap_area(va);
880 return ERR_PTR(err);
881 }
882
Roman Pencf725ce2015-04-15 16:13:52 -0700883 vaddr = vmap_block_vaddr(va->va_start, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700884 spin_lock_init(&vb->lock);
885 vb->va = va;
Roman Pencf725ce2015-04-15 16:13:52 -0700886 /* At least something should be left free */
887 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
888 vb->free = VMAP_BBMAP_BITS - (1UL << order);
Nick Piggindb64fe02008-10-18 20:27:03 -0700889 vb->dirty = 0;
Roman Pen7d61bfe2015-04-15 16:13:55 -0700890 vb->dirty_min = VMAP_BBMAP_BITS;
891 vb->dirty_max = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700892 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700893
894 vb_idx = addr_to_vb_idx(va->va_start);
895 spin_lock(&vmap_block_tree_lock);
896 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
897 spin_unlock(&vmap_block_tree_lock);
898 BUG_ON(err);
899 radix_tree_preload_end();
900
901 vbq = &get_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700902 spin_lock(&vbq->lock);
Roman Pen68ac5462015-04-15 16:13:48 -0700903 list_add_tail_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700904 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900905 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700906
Roman Pencf725ce2015-04-15 16:13:52 -0700907 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700908}
909
Nick Piggindb64fe02008-10-18 20:27:03 -0700910static void free_vmap_block(struct vmap_block *vb)
911{
912 struct vmap_block *tmp;
913 unsigned long vb_idx;
914
Nick Piggindb64fe02008-10-18 20:27:03 -0700915 vb_idx = addr_to_vb_idx(vb->va->va_start);
916 spin_lock(&vmap_block_tree_lock);
917 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
918 spin_unlock(&vmap_block_tree_lock);
919 BUG_ON(tmp != vb);
920
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800921 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800922 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700923}
924
Nick Piggin02b709d2010-02-01 22:25:57 +1100925static void purge_fragmented_blocks(int cpu)
926{
927 LIST_HEAD(purge);
928 struct vmap_block *vb;
929 struct vmap_block *n_vb;
930 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
931
932 rcu_read_lock();
933 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
934
935 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
936 continue;
937
938 spin_lock(&vb->lock);
939 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
940 vb->free = 0; /* prevent further allocs after releasing lock */
941 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
Roman Pen7d61bfe2015-04-15 16:13:55 -0700942 vb->dirty_min = 0;
943 vb->dirty_max = VMAP_BBMAP_BITS;
Nick Piggin02b709d2010-02-01 22:25:57 +1100944 spin_lock(&vbq->lock);
945 list_del_rcu(&vb->free_list);
946 spin_unlock(&vbq->lock);
947 spin_unlock(&vb->lock);
948 list_add_tail(&vb->purge, &purge);
949 } else
950 spin_unlock(&vb->lock);
951 }
952 rcu_read_unlock();
953
954 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
955 list_del(&vb->purge);
956 free_vmap_block(vb);
957 }
958}
959
Nick Piggin02b709d2010-02-01 22:25:57 +1100960static void purge_fragmented_blocks_allcpus(void)
961{
962 int cpu;
963
964 for_each_possible_cpu(cpu)
965 purge_fragmented_blocks(cpu);
966}
967
Nick Piggindb64fe02008-10-18 20:27:03 -0700968static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
969{
970 struct vmap_block_queue *vbq;
971 struct vmap_block *vb;
Roman Pencf725ce2015-04-15 16:13:52 -0700972 void *vaddr = NULL;
Nick Piggindb64fe02008-10-18 20:27:03 -0700973 unsigned int order;
974
Alexander Kuleshov891c49a2015-11-05 18:46:51 -0800975 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -0700976 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700977 if (WARN_ON(size == 0)) {
978 /*
979 * Allocating 0 bytes isn't what caller wants since
980 * get_order(0) returns funny result. Just warn and terminate
981 * early.
982 */
983 return NULL;
984 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700985 order = get_order(size);
986
Nick Piggindb64fe02008-10-18 20:27:03 -0700987 rcu_read_lock();
988 vbq = &get_cpu_var(vmap_block_queue);
989 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Roman Pencf725ce2015-04-15 16:13:52 -0700990 unsigned long pages_off;
Nick Piggindb64fe02008-10-18 20:27:03 -0700991
992 spin_lock(&vb->lock);
Roman Pencf725ce2015-04-15 16:13:52 -0700993 if (vb->free < (1UL << order)) {
994 spin_unlock(&vb->lock);
995 continue;
996 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100997
Roman Pencf725ce2015-04-15 16:13:52 -0700998 pages_off = VMAP_BBMAP_BITS - vb->free;
999 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
Nick Piggin02b709d2010-02-01 22:25:57 +11001000 vb->free -= 1UL << order;
1001 if (vb->free == 0) {
1002 spin_lock(&vbq->lock);
1003 list_del_rcu(&vb->free_list);
1004 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -07001005 }
Roman Pencf725ce2015-04-15 16:13:52 -07001006
Nick Piggindb64fe02008-10-18 20:27:03 -07001007 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +11001008 break;
Nick Piggindb64fe02008-10-18 20:27:03 -07001009 }
Nick Piggin02b709d2010-02-01 22:25:57 +11001010
Tejun Heo3f04ba82009-10-29 22:34:12 +09001011 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -07001012 rcu_read_unlock();
1013
Roman Pencf725ce2015-04-15 16:13:52 -07001014 /* Allocate new block if nothing was found */
1015 if (!vaddr)
1016 vaddr = new_vmap_block(order, gfp_mask);
Nick Piggindb64fe02008-10-18 20:27:03 -07001017
Roman Pencf725ce2015-04-15 16:13:52 -07001018 return vaddr;
Nick Piggindb64fe02008-10-18 20:27:03 -07001019}
1020
1021static void vb_free(const void *addr, unsigned long size)
1022{
1023 unsigned long offset;
1024 unsigned long vb_idx;
1025 unsigned int order;
1026 struct vmap_block *vb;
1027
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001028 BUG_ON(offset_in_page(size));
Nick Piggindb64fe02008-10-18 20:27:03 -07001029 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -08001030
1031 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1032
Nick Piggindb64fe02008-10-18 20:27:03 -07001033 order = get_order(size);
1034
1035 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001036 offset >>= PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001037
1038 vb_idx = addr_to_vb_idx((unsigned long)addr);
1039 rcu_read_lock();
1040 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1041 rcu_read_unlock();
1042 BUG_ON(!vb);
1043
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -08001044 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1045
Nick Piggindb64fe02008-10-18 20:27:03 -07001046 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001047
1048 /* Expand dirty range */
1049 vb->dirty_min = min(vb->dirty_min, offset);
1050 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
MinChan Kimd0868172009-03-31 15:19:26 -07001051
Nick Piggindb64fe02008-10-18 20:27:03 -07001052 vb->dirty += 1UL << order;
1053 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +11001054 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001055 spin_unlock(&vb->lock);
1056 free_vmap_block(vb);
1057 } else
1058 spin_unlock(&vb->lock);
1059}
1060
1061/**
1062 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1063 *
1064 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1065 * to amortize TLB flushing overheads. What this means is that any page you
1066 * have now, may, in a former life, have been mapped into kernel virtual
1067 * address by the vmap layer and so there might be some CPUs with TLB entries
1068 * still referencing that page (additional to the regular 1:1 kernel mapping).
1069 *
1070 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1071 * be sure that none of the pages we have control over will have any aliases
1072 * from the vmap layer.
1073 */
1074void vm_unmap_aliases(void)
1075{
1076 unsigned long start = ULONG_MAX, end = 0;
1077 int cpu;
1078 int flush = 0;
1079
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001080 if (unlikely(!vmap_initialized))
1081 return;
1082
Nick Piggindb64fe02008-10-18 20:27:03 -07001083 for_each_possible_cpu(cpu) {
1084 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1085 struct vmap_block *vb;
1086
1087 rcu_read_lock();
1088 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001089 spin_lock(&vb->lock);
Roman Pen7d61bfe2015-04-15 16:13:55 -07001090 if (vb->dirty) {
1091 unsigned long va_start = vb->va->va_start;
Nick Piggindb64fe02008-10-18 20:27:03 -07001092 unsigned long s, e;
Joonsoo Kimb136be5e2013-09-11 14:21:40 -07001093
Roman Pen7d61bfe2015-04-15 16:13:55 -07001094 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1095 e = va_start + (vb->dirty_max << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001096
Roman Pen7d61bfe2015-04-15 16:13:55 -07001097 start = min(s, start);
1098 end = max(e, end);
1099
Nick Piggindb64fe02008-10-18 20:27:03 -07001100 flush = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -07001101 }
1102 spin_unlock(&vb->lock);
1103 }
1104 rcu_read_unlock();
1105 }
1106
1107 __purge_vmap_area_lazy(&start, &end, 1, flush);
1108}
1109EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1110
1111/**
1112 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1113 * @mem: the pointer returned by vm_map_ram
1114 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1115 */
1116void vm_unmap_ram(const void *mem, unsigned int count)
1117{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001118 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001119 unsigned long addr = (unsigned long)mem;
1120
1121 BUG_ON(!addr);
1122 BUG_ON(addr < VMALLOC_START);
1123 BUG_ON(addr > VMALLOC_END);
Shawn Lina1c0b1a2016-03-17 14:20:37 -07001124 BUG_ON(!PAGE_ALIGNED(addr));
Nick Piggindb64fe02008-10-18 20:27:03 -07001125
1126 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001127 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001128
1129 if (likely(count <= VMAP_MAX_ALLOC))
1130 vb_free(mem, size);
1131 else
1132 free_unmap_vmap_area_addr(addr);
1133}
1134EXPORT_SYMBOL(vm_unmap_ram);
1135
1136/**
1137 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1138 * @pages: an array of pointers to the pages to be mapped
1139 * @count: number of pages
1140 * @node: prefer to allocate data structures on this node
1141 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001142 *
Gioh Kim36437632014-04-07 15:37:37 -07001143 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1144 * faster than vmap so it's good. But if you mix long-life and short-life
1145 * objects with vm_map_ram(), it could consume lots of address space through
1146 * fragmentation (especially on a 32bit machine). You could see failures in
1147 * the end. Please use this function for short-lived objects.
1148 *
Randy Dunlape99c97a2008-10-29 14:01:09 -07001149 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001150 */
1151void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1152{
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001153 unsigned long size = (unsigned long)count << PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -07001154 unsigned long addr;
1155 void *mem;
1156
1157 if (likely(count <= VMAP_MAX_ALLOC)) {
1158 mem = vb_alloc(size, GFP_KERNEL);
1159 if (IS_ERR(mem))
1160 return NULL;
1161 addr = (unsigned long)mem;
1162 } else {
1163 struct vmap_area *va;
1164 va = alloc_vmap_area(size, PAGE_SIZE,
1165 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1166 if (IS_ERR(va))
1167 return NULL;
1168
1169 addr = va->va_start;
1170 mem = (void *)addr;
1171 }
1172 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1173 vm_unmap_ram(mem, count);
1174 return NULL;
1175 }
1176 return mem;
1177}
1178EXPORT_SYMBOL(vm_map_ram);
1179
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001180static struct vm_struct *vmlist __initdata;
Tejun Heof0aa6612009-02-20 16:29:08 +09001181/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001182 * vm_area_add_early - add vmap area early during boot
1183 * @vm: vm_struct to add
1184 *
1185 * This function is used to add fixed kernel vm area to vmlist before
1186 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1187 * should contain proper values and the other fields should be zero.
1188 *
1189 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1190 */
1191void __init vm_area_add_early(struct vm_struct *vm)
1192{
1193 struct vm_struct *tmp, **p;
1194
1195 BUG_ON(vmap_initialized);
1196 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1197 if (tmp->addr >= vm->addr) {
1198 BUG_ON(tmp->addr < vm->addr + vm->size);
1199 break;
1200 } else
1201 BUG_ON(tmp->addr + tmp->size > vm->addr);
1202 }
1203 vm->next = *p;
1204 *p = vm;
1205}
1206
1207/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001208 * vm_area_register_early - register vmap area early during boot
1209 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001210 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001211 *
1212 * This function is used to register kernel vm area before
1213 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1214 * proper values on entry and other fields should be zero. On return,
1215 * vm->addr contains the allocated address.
1216 *
1217 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1218 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001219void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001220{
1221 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001222 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001223
Tejun Heoc0c0a292009-02-24 11:57:21 +09001224 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1225 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1226
1227 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001228
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001229 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001230}
1231
Nick Piggindb64fe02008-10-18 20:27:03 -07001232void __init vmalloc_init(void)
1233{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001234 struct vmap_area *va;
1235 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001236 int i;
1237
1238 for_each_possible_cpu(i) {
1239 struct vmap_block_queue *vbq;
Al Viro32fcfd42013-03-10 20:14:08 -04001240 struct vfree_deferred *p;
Nick Piggindb64fe02008-10-18 20:27:03 -07001241
1242 vbq = &per_cpu(vmap_block_queue, i);
1243 spin_lock_init(&vbq->lock);
1244 INIT_LIST_HEAD(&vbq->free);
Al Viro32fcfd42013-03-10 20:14:08 -04001245 p = &per_cpu(vfree_deferred, i);
1246 init_llist_head(&p->list);
1247 INIT_WORK(&p->wq, free_work);
Nick Piggindb64fe02008-10-18 20:27:03 -07001248 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001249
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001250 /* Import existing vmlist entries. */
1251 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001252 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001253 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001254 va->va_start = (unsigned long)tmp->addr;
1255 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001256 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001257 __insert_vmap_area(va);
1258 }
Tejun Heoca23e402009-08-14 15:00:52 +09001259
1260 vmap_area_pcpu_hole = VMALLOC_END;
1261
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001262 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001263}
1264
Tejun Heo8fc48982009-02-20 16:29:08 +09001265/**
1266 * map_kernel_range_noflush - map kernel VM area with the specified pages
1267 * @addr: start of the VM area to map
1268 * @size: size of the VM area to map
1269 * @prot: page protection flags to use
1270 * @pages: pages to map
1271 *
1272 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1273 * specify should have been allocated using get_vm_area() and its
1274 * friends.
1275 *
1276 * NOTE:
1277 * This function does NOT do any cache flushing. The caller is
1278 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1279 * before calling this function.
1280 *
1281 * RETURNS:
1282 * The number of pages mapped on success, -errno on failure.
1283 */
1284int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1285 pgprot_t prot, struct page **pages)
1286{
1287 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1288}
1289
1290/**
1291 * unmap_kernel_range_noflush - unmap kernel VM area
1292 * @addr: start of the VM area to unmap
1293 * @size: size of the VM area to unmap
1294 *
1295 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1296 * specify should have been allocated using get_vm_area() and its
1297 * friends.
1298 *
1299 * NOTE:
1300 * This function does NOT do any cache flushing. The caller is
1301 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1302 * before calling this function and flush_tlb_kernel_range() after.
1303 */
1304void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1305{
1306 vunmap_page_range(addr, addr + size);
1307}
Huang Ying81e88fd2011-01-12 14:44:55 +08001308EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001309
1310/**
1311 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1312 * @addr: start of the VM area to unmap
1313 * @size: size of the VM area to unmap
1314 *
1315 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1316 * the unmapping and tlb after.
1317 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001318void unmap_kernel_range(unsigned long addr, unsigned long size)
1319{
1320 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001321
1322 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001323 vunmap_page_range(addr, end);
1324 flush_tlb_kernel_range(addr, end);
1325}
Minchan Kim93ef6d62014-06-04 16:11:09 -07001326EXPORT_SYMBOL_GPL(unmap_kernel_range);
Nick Piggindb64fe02008-10-18 20:27:03 -07001327
WANG Chaof6f8ed42014-08-06 16:06:58 -07001328int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
Nick Piggindb64fe02008-10-18 20:27:03 -07001329{
1330 unsigned long addr = (unsigned long)area->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07001331 unsigned long end = addr + get_vm_area_size(area);
Nick Piggindb64fe02008-10-18 20:27:03 -07001332 int err;
1333
WANG Chaof6f8ed42014-08-06 16:06:58 -07001334 err = vmap_page_range(addr, end, prot, pages);
Nick Piggindb64fe02008-10-18 20:27:03 -07001335
WANG Chaof6f8ed42014-08-06 16:06:58 -07001336 return err > 0 ? 0 : err;
Nick Piggindb64fe02008-10-18 20:27:03 -07001337}
1338EXPORT_SYMBOL_GPL(map_vm_area);
1339
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001340static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001341 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001342{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001343 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001344 vm->flags = flags;
1345 vm->addr = (void *)va->va_start;
1346 vm->size = va->va_end - va->va_start;
1347 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001348 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001349 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001350 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001351}
Tejun Heocf88c792009-08-14 15:00:52 +09001352
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001353static void clear_vm_uninitialized_flag(struct vm_struct *vm)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001354{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001355 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001356 * Before removing VM_UNINITIALIZED,
Joonsoo Kimd4033af2013-04-29 15:07:35 -07001357 * we should make sure that vm has proper values.
1358 * Pair with smp_rmb() in show_numa_info().
1359 */
1360 smp_wmb();
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001361 vm->flags &= ~VM_UNINITIALIZED;
Tejun Heocf88c792009-08-14 15:00:52 +09001362}
1363
Nick Piggindb64fe02008-10-18 20:27:03 -07001364static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001365 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001366 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001367{
Kautuk Consul00065262011-12-19 17:12:04 -08001368 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001369 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001371 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001373 if (unlikely(!size))
1374 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
zijun_hu252e5c62016-10-07 16:57:26 -07001376 if (flags & VM_IOREMAP)
1377 align = 1ul << clamp_t(int, get_count_order_long(size),
1378 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1379
Tejun Heocf88c792009-08-14 15:00:52 +09001380 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 if (unlikely(!area))
1382 return NULL;
1383
Andrey Ryabinin71394fe2015-02-13 14:40:03 -08001384 if (!(flags & VM_NO_GUARD))
1385 size += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Nick Piggindb64fe02008-10-18 20:27:03 -07001387 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1388 if (IS_ERR(va)) {
1389 kfree(area);
1390 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Zhang Yanfeid82b1d82013-07-03 15:04:47 -07001393 setup_vmalloc_vm(area, va, flags, caller);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396}
1397
Christoph Lameter930fc452005-10-29 18:15:41 -07001398struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1399 unsigned long start, unsigned long end)
1400{
David Rientjes00ef2d22013-02-22 16:35:36 -08001401 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1402 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001403}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001404EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001405
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001406struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1407 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001408 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001409{
David Rientjes00ef2d22013-02-22 16:35:36 -08001410 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1411 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001412}
1413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414/**
Simon Arlott183ff222007-10-20 01:27:18 +02001415 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 * @size: size of the area
1417 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1418 *
1419 * Search an area of @size in the kernel virtual mapping area,
1420 * and reserved it for out purposes. Returns the area descriptor
1421 * on success or %NULL on failure.
1422 */
1423struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1424{
David Miller2dca6992009-09-21 12:22:34 -07001425 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001426 NUMA_NO_NODE, GFP_KERNEL,
1427 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001428}
1429
1430struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001431 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001432{
David Miller2dca6992009-09-21 12:22:34 -07001433 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001434 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
1436
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001437/**
1438 * find_vm_area - find a continuous kernel virtual area
1439 * @addr: base address
1440 *
1441 * Search for the kernel VM area starting at @addr, and return it.
1442 * It is up to the caller to do all required locking to keep the returned
1443 * pointer valid.
1444 */
1445struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001446{
Nick Piggindb64fe02008-10-18 20:27:03 -07001447 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001448
Nick Piggindb64fe02008-10-18 20:27:03 -07001449 va = find_vmap_area((unsigned long)addr);
1450 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001451 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001452
Andi Kleen7856dfe2005-05-20 14:27:57 -07001453 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001454}
1455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456/**
Simon Arlott183ff222007-10-20 01:27:18 +02001457 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 * @addr: base address
1459 *
1460 * Search for the kernel VM area starting at @addr, and remove it.
1461 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001462 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001464struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
Nick Piggindb64fe02008-10-18 20:27:03 -07001466 struct vmap_area *va;
1467
1468 va = find_vmap_area((unsigned long)addr);
1469 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001470 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001471
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001472 spin_lock(&vmap_area_lock);
1473 va->vm = NULL;
1474 va->flags &= ~VM_VM_AREA;
1475 spin_unlock(&vmap_area_lock);
1476
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001477 vmap_debug_free_range(va->va_start, va->va_end);
Andrey Ryabinina5af5aa2015-03-12 16:26:11 -07001478 kasan_free_shadow(vm);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001479 free_unmap_vmap_area(va);
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001480
Nick Piggindb64fe02008-10-18 20:27:03 -07001481 return vm;
1482 }
1483 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001486static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
1488 struct vm_struct *area;
1489
1490 if (!addr)
1491 return;
1492
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07001493 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
Dan Carpenterab15d9b2013-07-08 15:59:53 -07001494 addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
Chintan Pandyae18d3282018-06-07 17:06:50 -07001497 area = find_vmap_area((unsigned long)addr)->vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001499 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 return;
1502 }
1503
Jerome Marchand7511c3e2015-11-20 15:57:02 -08001504 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1505 debug_check_no_obj_freed(addr, get_vm_area_size(area));
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001506
Chintan Pandyae18d3282018-06-07 17:06:50 -07001507 remove_vm_area(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (deallocate_pages) {
1509 int i;
1510
1511 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001512 struct page *page = area->pages[i];
1513
1514 BUG_ON(!page);
Vladimir Davydov49491482016-07-26 15:24:24 -07001515 __free_pages(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 }
1517
David Rientjes244d63e2016-01-14 15:19:35 -08001518 kvfree(area->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 }
1520
1521 kfree(area);
1522 return;
1523}
Al Viro32fcfd42013-03-10 20:14:08 -04001524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525/**
1526 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * @addr: memory base address
1528 *
Simon Arlott183ff222007-10-20 01:27:18 +02001529 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001530 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1531 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 *
Al Viro32fcfd42013-03-10 20:14:08 -04001533 * Must not be called in NMI context (strictly speaking, only if we don't
1534 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1535 * conventions for vfree() arch-depenedent would be a really bad idea)
Andrew Mortonc9fcee52013-05-07 16:18:18 -07001536 *
1537 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001539void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
Al Viro32fcfd42013-03-10 20:14:08 -04001541 BUG_ON(in_nmi());
Catalin Marinas89219d32009-06-11 13:23:19 +01001542
1543 kmemleak_free(addr);
1544
Al Viro32fcfd42013-03-10 20:14:08 -04001545 if (!addr)
1546 return;
1547 if (unlikely(in_interrupt())) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001548 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
Oleg Nesterov59d31322013-07-08 16:00:08 -07001549 if (llist_add((struct llist_node *)addr, &p->list))
1550 schedule_work(&p->wq);
Al Viro32fcfd42013-03-10 20:14:08 -04001551 } else
1552 __vunmap(addr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554EXPORT_SYMBOL(vfree);
1555
1556/**
1557 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 * @addr: memory base address
1559 *
1560 * Free the virtually contiguous memory area starting at @addr,
1561 * which was created from the page array passed to vmap().
1562 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001563 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001565void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
1567 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001568 might_sleep();
Al Viro32fcfd42013-03-10 20:14:08 -04001569 if (addr)
1570 __vunmap(addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572EXPORT_SYMBOL(vunmap);
1573
1574/**
1575 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 * @pages: array of page pointers
1577 * @count: number of pages to map
1578 * @flags: vm_area->flags
1579 * @prot: page protection for the mapping
1580 *
1581 * Maps @count pages from @pages into contiguous kernel virtual
1582 * space.
1583 */
1584void *vmap(struct page **pages, unsigned int count,
1585 unsigned long flags, pgprot_t prot)
1586{
1587 struct vm_struct *area;
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001588 unsigned long size; /* In bytes */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
Peter Zijlstra34754b62009-02-25 16:04:03 +01001590 might_sleep();
1591
Jan Beulich44813742009-09-21 17:03:05 -07001592 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 return NULL;
1594
Guillermo Julián Moreno65ee03c2016-06-03 14:55:33 -07001595 size = (unsigned long)count << PAGE_SHIFT;
1596 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 if (!area)
1598 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001599
WANG Chaof6f8ed42014-08-06 16:06:58 -07001600 if (map_vm_area(area, prot, pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 vunmap(area->addr);
1602 return NULL;
1603 }
1604
1605 return area->addr;
1606}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607EXPORT_SYMBOL(vmap);
1608
David Miller2dca6992009-09-21 12:22:34 -07001609static void *__vmalloc_node(unsigned long size, unsigned long align,
1610 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001611 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001612static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Wanpeng Li3722e132013-11-12 15:07:29 -08001613 pgprot_t prot, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614{
1615 struct page **pages;
1616 unsigned int nr_pages, array_size, i;
David Rientjes930f0362014-08-06 16:06:28 -07001617 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1618 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Wanpeng Li762216a2013-09-11 14:22:42 -07001620 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 array_size = (nr_pages * sizeof(struct page *));
1622
1623 area->nr_pages = nr_pages;
1624 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001625 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001626 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Wanpeng Li3722e132013-11-12 15:07:29 -08001627 PAGE_KERNEL, node, area->caller);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001628 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001629 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 area->pages = pages;
1632 if (!area->pages) {
1633 remove_vm_area(area->addr);
1634 kfree(area);
1635 return NULL;
1636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
1638 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001639 struct page *page;
1640
Jianguo Wu4b909512013-11-12 15:07:11 -08001641 if (node == NUMA_NO_NODE)
Michal Hocko7877cdc2016-10-07 17:01:55 -07001642 page = alloc_page(alloc_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001643 else
Michal Hocko7877cdc2016-10-07 17:01:55 -07001644 page = alloc_pages_node(node, alloc_mask, 0);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001645
1646 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 /* Successfully allocated i pages, free them in __vunmap() */
1648 area->nr_pages = i;
1649 goto fail;
1650 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001651 area->pages[i] = page;
Mel Gormand0164ad2015-11-06 16:28:21 -08001652 if (gfpflags_allow_blocking(gfp_mask))
Eric Dumazet660654f2014-08-06 16:06:25 -07001653 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 }
1655
WANG Chaof6f8ed42014-08-06 16:06:58 -07001656 if (map_vm_area(area, prot, pages))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 goto fail;
1658 return area->addr;
1659
1660fail:
Michal Hocko7877cdc2016-10-07 17:01:55 -07001661 warn_alloc(gfp_mask,
1662 "vmalloc: allocation failure, allocated %ld of %ld bytes",
Dave Hansen22943ab2011-05-24 17:12:18 -07001663 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 vfree(area->addr);
1665 return NULL;
1666}
1667
David Rientjesd0a21262011-01-13 15:46:02 -08001668/**
1669 * __vmalloc_node_range - allocate virtually contiguous memory
1670 * @size: allocation size
1671 * @align: desired alignment
1672 * @start: vm area range start
1673 * @end: vm area range end
1674 * @gfp_mask: flags for the page level allocator
1675 * @prot: protection mask for the allocated pages
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001676 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
David Rientjes00ef2d22013-02-22 16:35:36 -08001677 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001678 * @caller: caller's return address
1679 *
1680 * Allocate enough pages to cover @size from the page level
1681 * allocator with @gfp_mask flags. Map them into contiguous
1682 * kernel virtual space, using a pagetable protection of @prot.
1683 */
1684void *__vmalloc_node_range(unsigned long size, unsigned long align,
1685 unsigned long start, unsigned long end, gfp_t gfp_mask,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001686 pgprot_t prot, unsigned long vm_flags, int node,
1687 const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001688{
David Rientjesd0a21262011-01-13 15:46:02 -08001689 struct vm_struct *area;
1690 void *addr;
1691 unsigned long real_size = size;
1692
1693 size = PAGE_ALIGN(size);
1694 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001695 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001696
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001697 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1698 vm_flags, start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001699 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001700 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001701
Wanpeng Li3722e132013-11-12 15:07:29 -08001702 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
Mel Gorman1368edf2011-12-08 14:34:30 -08001703 if (!addr)
Wanpeng Lib82225f32013-11-12 15:07:33 -08001704 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001705
1706 /*
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001707 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1708 * flag. It means that vm_struct is not fully initialized.
Joonsoo Kim4341fa42013-04-29 15:07:39 -07001709 * Now, it is fully initialized, so remove this flag here.
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001710 */
Zhang Yanfei20fc02b2013-07-08 15:59:58 -07001711 clear_vm_uninitialized_flag(area);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001712
1713 /*
Catalin Marinas7f88f882013-11-12 15:07:45 -08001714 * A ref_count = 2 is needed because vm_struct allocated in
1715 * __get_vm_area_node() contains a reference to the virtual address of
1716 * the vmalloc'ed block.
Catalin Marinas89219d32009-06-11 13:23:19 +01001717 */
Catalin Marinas7f88f882013-11-12 15:07:45 -08001718 kmemleak_alloc(addr, real_size, 2, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001719
1720 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001721
1722fail:
Michal Hocko7877cdc2016-10-07 17:01:55 -07001723 warn_alloc(gfp_mask,
1724 "vmalloc: allocation failure: %lu bytes", real_size);
Joe Perchesde7d2b52011-10-31 17:08:48 -07001725 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001726}
1727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001729 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001731 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 * @gfp_mask: flags for the page level allocator
1733 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001734 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001735 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 *
1737 * Allocate enough pages to cover @size from the page level
1738 * allocator with @gfp_mask flags. Map them into contiguous
1739 * kernel virtual space, using a pagetable protection of @prot.
1740 */
David Miller2dca6992009-09-21 12:22:34 -07001741static void *__vmalloc_node(unsigned long size, unsigned long align,
1742 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001743 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
David Rientjesd0a21262011-01-13 15:46:02 -08001745 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
Andrey Ryabinincb9e3c22015-02-13 14:40:07 -08001746 gfp_mask, prot, 0, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747}
1748
Christoph Lameter930fc452005-10-29 18:15:41 -07001749void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1750{
David Rientjes00ef2d22013-02-22 16:35:36 -08001751 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001752 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001753}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754EXPORT_SYMBOL(__vmalloc);
1755
Dave Younge1ca7782010-10-26 14:22:06 -07001756static inline void *__vmalloc_node_flags(unsigned long size,
1757 int node, gfp_t flags)
1758{
1759 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1760 node, __builtin_return_address(0));
1761}
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763/**
1764 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 * Allocate enough pages to cover @size from the page level
1767 * allocator and map them into contiguous kernel virtual space.
1768 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001769 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 * use __vmalloc() instead.
1771 */
1772void *vmalloc(unsigned long size)
1773{
David Rientjes00ef2d22013-02-22 16:35:36 -08001774 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1775 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777EXPORT_SYMBOL(vmalloc);
1778
Christoph Lameter930fc452005-10-29 18:15:41 -07001779/**
Dave Younge1ca7782010-10-26 14:22:06 -07001780 * vzalloc - allocate virtually contiguous memory with zero fill
1781 * @size: allocation size
1782 * Allocate enough pages to cover @size from the page level
1783 * allocator and map them into contiguous kernel virtual space.
1784 * The memory allocated is set to zero.
1785 *
1786 * For tight control over page level allocator and protection flags
1787 * use __vmalloc() instead.
1788 */
1789void *vzalloc(unsigned long size)
1790{
David Rientjes00ef2d22013-02-22 16:35:36 -08001791 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001792 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1793}
1794EXPORT_SYMBOL(vzalloc);
1795
1796/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001797 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1798 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001799 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001800 * The resulting memory area is zeroed so it can be mapped to userspace
1801 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001802 */
1803void *vmalloc_user(unsigned long size)
1804{
1805 struct vm_struct *area;
1806 void *ret;
1807
David Miller2dca6992009-09-21 12:22:34 -07001808 ret = __vmalloc_node(size, SHMLBA,
1809 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001810 PAGE_KERNEL, NUMA_NO_NODE,
1811 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001812 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001813 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001814 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001815 }
Nick Piggin83342312006-06-23 02:03:20 -07001816 return ret;
1817}
1818EXPORT_SYMBOL(vmalloc_user);
1819
1820/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001821 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001822 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001823 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001824 *
1825 * Allocate enough pages to cover @size from the page level
1826 * allocator and map them into contiguous kernel virtual space.
1827 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001828 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001829 * use __vmalloc() instead.
1830 */
1831void *vmalloc_node(unsigned long size, int node)
1832{
David Miller2dca6992009-09-21 12:22:34 -07001833 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001834 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001835}
1836EXPORT_SYMBOL(vmalloc_node);
1837
Dave Younge1ca7782010-10-26 14:22:06 -07001838/**
1839 * vzalloc_node - allocate memory on a specific node with zero fill
1840 * @size: allocation size
1841 * @node: numa node
1842 *
1843 * Allocate enough pages to cover @size from the page level
1844 * allocator and map them into contiguous kernel virtual space.
1845 * The memory allocated is set to zero.
1846 *
1847 * For tight control over page level allocator and protection flags
1848 * use __vmalloc_node() instead.
1849 */
1850void *vzalloc_node(unsigned long size, int node)
1851{
1852 return __vmalloc_node_flags(size, node,
1853 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1854}
1855EXPORT_SYMBOL(vzalloc_node);
1856
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001857#ifndef PAGE_KERNEL_EXEC
1858# define PAGE_KERNEL_EXEC PAGE_KERNEL
1859#endif
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861/**
1862 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 * @size: allocation size
1864 *
1865 * Kernel-internal function to allocate enough pages to cover @size
1866 * the page level allocator and map them into contiguous and
1867 * executable kernel virtual space.
1868 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001869 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 * use __vmalloc() instead.
1871 */
1872
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873void *vmalloc_exec(unsigned long size)
1874{
David Miller2dca6992009-09-21 12:22:34 -07001875 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001876 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877}
1878
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001879#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001880#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001881#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001882#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001883#else
1884#define GFP_VMALLOC32 GFP_KERNEL
1885#endif
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887/**
1888 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 * @size: allocation size
1890 *
1891 * Allocate enough 32bit PA addressable pages to cover @size from the
1892 * page level allocator and map them into contiguous kernel virtual space.
1893 */
1894void *vmalloc_32(unsigned long size)
1895{
David Miller2dca6992009-09-21 12:22:34 -07001896 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001897 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899EXPORT_SYMBOL(vmalloc_32);
1900
Nick Piggin83342312006-06-23 02:03:20 -07001901/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001902 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001903 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001904 *
1905 * The resulting memory area is 32bit addressable and zeroed so it can be
1906 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001907 */
1908void *vmalloc_32_user(unsigned long size)
1909{
1910 struct vm_struct *area;
1911 void *ret;
1912
David Miller2dca6992009-09-21 12:22:34 -07001913 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001914 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001915 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001916 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001917 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001918 }
Nick Piggin83342312006-06-23 02:03:20 -07001919 return ret;
1920}
1921EXPORT_SYMBOL(vmalloc_32_user);
1922
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001923/*
1924 * small helper routine , copy contents to buf from addr.
1925 * If the page is not present, fill zero.
1926 */
1927
1928static int aligned_vread(char *buf, char *addr, unsigned long count)
1929{
1930 struct page *p;
1931 int copied = 0;
1932
1933 while (count) {
1934 unsigned long offset, length;
1935
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001936 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001937 length = PAGE_SIZE - offset;
1938 if (length > count)
1939 length = count;
1940 p = vmalloc_to_page(addr);
1941 /*
1942 * To do safe access to this _mapped_ area, we need
1943 * lock. But adding lock here means that we need to add
1944 * overhead of vmalloc()/vfree() calles for this _debug_
1945 * interface, rarely used. Instead of that, we'll use
1946 * kmap() and get small overhead in this access function.
1947 */
1948 if (p) {
1949 /*
1950 * we can expect USER0 is not used (see vread/vwrite's
1951 * function description)
1952 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001953 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001954 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001955 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001956 } else
1957 memset(buf, 0, length);
1958
1959 addr += length;
1960 buf += length;
1961 copied += length;
1962 count -= length;
1963 }
1964 return copied;
1965}
1966
1967static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1968{
1969 struct page *p;
1970 int copied = 0;
1971
1972 while (count) {
1973 unsigned long offset, length;
1974
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08001975 offset = offset_in_page(addr);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001976 length = PAGE_SIZE - offset;
1977 if (length > count)
1978 length = count;
1979 p = vmalloc_to_page(addr);
1980 /*
1981 * To do safe access to this _mapped_ area, we need
1982 * lock. But adding lock here means that we need to add
1983 * overhead of vmalloc()/vfree() calles for this _debug_
1984 * interface, rarely used. Instead of that, we'll use
1985 * kmap() and get small overhead in this access function.
1986 */
1987 if (p) {
1988 /*
1989 * we can expect USER0 is not used (see vread/vwrite's
1990 * function description)
1991 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001992 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001993 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001994 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001995 }
1996 addr += length;
1997 buf += length;
1998 copied += length;
1999 count -= length;
2000 }
2001 return copied;
2002}
2003
2004/**
2005 * vread() - read vmalloc area in a safe way.
2006 * @buf: buffer for reading data
2007 * @addr: vm address.
2008 * @count: number of bytes to be read.
2009 *
2010 * Returns # of bytes which addr and buf should be increased.
2011 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2012 * includes any intersect with alive vmalloc area.
2013 *
2014 * This function checks that addr is a valid vmalloc'ed area, and
2015 * copy data from that area to a given buffer. If the given memory range
2016 * of [addr...addr+count) includes some valid address, data is copied to
2017 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2018 * IOREMAP area is treated as memory hole and no copy is done.
2019 *
2020 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002021 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002022 *
2023 * Note: In usual ops, vread() is never necessary because the caller
2024 * should know vmalloc() area is valid and can use memcpy().
2025 * This is for routines which have to access vmalloc area without
2026 * any informaion, as /dev/kmem.
2027 *
2028 */
2029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030long vread(char *buf, char *addr, unsigned long count)
2031{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002032 struct vmap_area *va;
2033 struct vm_struct *vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002035 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 unsigned long n;
2037
2038 /* Don't allow overflow */
2039 if ((unsigned long) addr + count < count)
2040 count = -(unsigned long) addr;
2041
Joonsoo Kime81ce852013-04-29 15:07:32 -07002042 spin_lock(&vmap_area_lock);
2043 list_for_each_entry(va, &vmap_area_list, list) {
2044 if (!count)
2045 break;
2046
2047 if (!(va->flags & VM_VM_AREA))
2048 continue;
2049
2050 vm = va->vm;
2051 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002052 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 continue;
2054 while (addr < vaddr) {
2055 if (count == 0)
2056 goto finished;
2057 *buf = '\0';
2058 buf++;
2059 addr++;
2060 count--;
2061 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002062 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002063 if (n > count)
2064 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002065 if (!(vm->flags & VM_IOREMAP))
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002066 aligned_vread(buf, addr, n);
2067 else /* IOREMAP area is treated as memory hole */
2068 memset(buf, 0, n);
2069 buf += n;
2070 addr += n;
2071 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002074 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002075
2076 if (buf == buf_start)
2077 return 0;
2078 /* zero-fill memory holes */
2079 if (buf != buf_start + buflen)
2080 memset(buf, 0, buflen - (buf - buf_start));
2081
2082 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083}
2084
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002085/**
2086 * vwrite() - write vmalloc area in a safe way.
2087 * @buf: buffer for source data
2088 * @addr: vm address.
2089 * @count: number of bytes to be read.
2090 *
2091 * Returns # of bytes which addr and buf should be incresed.
2092 * (same number to @count).
2093 * If [addr...addr+count) doesn't includes any intersect with valid
2094 * vmalloc area, returns 0.
2095 *
2096 * This function checks that addr is a valid vmalloc'ed area, and
2097 * copy data from a buffer to the given addr. If specified range of
2098 * [addr...addr+count) includes some valid address, data is copied from
2099 * proper area of @buf. If there are memory holes, no copy to hole.
2100 * IOREMAP area is treated as memory hole and no copy is done.
2101 *
2102 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002103 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002104 *
2105 * Note: In usual ops, vwrite() is never necessary because the caller
2106 * should know vmalloc() area is valid and can use memcpy().
2107 * This is for routines which have to access vmalloc area without
2108 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002109 */
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111long vwrite(char *buf, char *addr, unsigned long count)
2112{
Joonsoo Kime81ce852013-04-29 15:07:32 -07002113 struct vmap_area *va;
2114 struct vm_struct *vm;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002115 char *vaddr;
2116 unsigned long n, buflen;
2117 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 /* Don't allow overflow */
2120 if ((unsigned long) addr + count < count)
2121 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002122 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
Joonsoo Kime81ce852013-04-29 15:07:32 -07002124 spin_lock(&vmap_area_lock);
2125 list_for_each_entry(va, &vmap_area_list, list) {
2126 if (!count)
2127 break;
2128
2129 if (!(va->flags & VM_VM_AREA))
2130 continue;
2131
2132 vm = va->vm;
2133 vaddr = (char *) vm->addr;
Wanpeng Li762216a2013-09-11 14:22:42 -07002134 if (addr >= vaddr + get_vm_area_size(vm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 continue;
2136 while (addr < vaddr) {
2137 if (count == 0)
2138 goto finished;
2139 buf++;
2140 addr++;
2141 count--;
2142 }
Wanpeng Li762216a2013-09-11 14:22:42 -07002143 n = vaddr + get_vm_area_size(vm) - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002144 if (n > count)
2145 n = count;
Joonsoo Kime81ce852013-04-29 15:07:32 -07002146 if (!(vm->flags & VM_IOREMAP)) {
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002147 aligned_vwrite(buf, addr, n);
2148 copied++;
2149 }
2150 buf += n;
2151 addr += n;
2152 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 }
2154finished:
Joonsoo Kime81ce852013-04-29 15:07:32 -07002155 spin_unlock(&vmap_area_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002156 if (!copied)
2157 return 0;
2158 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159}
Nick Piggin83342312006-06-23 02:03:20 -07002160
2161/**
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002162 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2163 * @vma: vma to cover
2164 * @uaddr: target user address to start at
2165 * @kaddr: virtual address of vmalloc kernel memory
2166 * @size: size of map area
2167 *
2168 * Returns: 0 for success, -Exxx on failure
2169 *
2170 * This function checks that @kaddr is a valid vmalloc'ed area,
2171 * and that it is big enough to cover the range starting at
2172 * @uaddr in @vma. Will return failure if that criteria isn't
2173 * met.
2174 *
2175 * Similar to remap_pfn_range() (see mm/memory.c)
2176 */
2177int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2178 void *kaddr, unsigned long size)
2179{
2180 struct vm_struct *area;
2181
2182 size = PAGE_ALIGN(size);
2183
2184 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2185 return -EINVAL;
2186
2187 area = find_vm_area(kaddr);
2188 if (!area)
2189 return -EINVAL;
2190
2191 if (!(area->flags & VM_USERMAP))
2192 return -EINVAL;
2193
Roman Penyaev5b4e7792019-03-05 15:43:20 -08002194 if (kaddr + size > area->addr + get_vm_area_size(area))
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002195 return -EINVAL;
2196
2197 do {
2198 struct page *page = vmalloc_to_page(kaddr);
2199 int ret;
2200
2201 ret = vm_insert_page(vma, uaddr, page);
2202 if (ret)
2203 return ret;
2204
2205 uaddr += PAGE_SIZE;
2206 kaddr += PAGE_SIZE;
2207 size -= PAGE_SIZE;
2208 } while (size > 0);
2209
2210 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2211
2212 return 0;
2213}
2214EXPORT_SYMBOL(remap_vmalloc_range_partial);
2215
2216/**
Nick Piggin83342312006-06-23 02:03:20 -07002217 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002218 * @vma: vma to cover (map full range of vma)
2219 * @addr: vmalloc memory
2220 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002221 *
2222 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002223 *
2224 * This function checks that addr is a valid vmalloc'ed area, and
2225 * that it is big enough to cover the vma. Will return failure if
2226 * that criteria isn't met.
2227 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002228 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002229 */
2230int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2231 unsigned long pgoff)
2232{
HATAYAMA Daisukee69e9d4a2013-07-03 15:02:18 -07002233 return remap_vmalloc_range_partial(vma, vma->vm_start,
2234 addr + (pgoff << PAGE_SHIFT),
2235 vma->vm_end - vma->vm_start);
Nick Piggin83342312006-06-23 02:03:20 -07002236}
2237EXPORT_SYMBOL(remap_vmalloc_range);
2238
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002239/*
2240 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2241 * have one.
2242 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07002243void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002244{
2245}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002246
2247
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002248static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002249{
David Vrabelcd129092011-09-29 16:53:32 +01002250 pte_t ***p = data;
2251
2252 if (p) {
2253 *(*p) = pte;
2254 (*p)++;
2255 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002256 return 0;
2257}
2258
2259/**
2260 * alloc_vm_area - allocate a range of kernel address space
2261 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002262 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002263 *
2264 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002265 *
2266 * This function reserves a range of kernel address space, and
2267 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002268 * are created.
2269 *
2270 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2271 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002272 */
David Vrabelcd129092011-09-29 16:53:32 +01002273struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002274{
2275 struct vm_struct *area;
2276
Christoph Lameter23016962008-04-28 02:12:42 -07002277 area = get_vm_area_caller(size, VM_IOREMAP,
2278 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002279 if (area == NULL)
2280 return NULL;
2281
2282 /*
2283 * This ensures that page tables are constructed for this region
2284 * of kernel virtual address space and mapped into init_mm.
2285 */
2286 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002287 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002288 free_vm_area(area);
2289 return NULL;
2290 }
2291
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002292 return area;
2293}
2294EXPORT_SYMBOL_GPL(alloc_vm_area);
2295
2296void free_vm_area(struct vm_struct *area)
2297{
2298 struct vm_struct *ret;
2299 ret = remove_vm_area(area->addr);
2300 BUG_ON(ret != area);
2301 kfree(area);
2302}
2303EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002304
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002305#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002306static struct vmap_area *node_to_va(struct rb_node *n)
2307{
2308 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2309}
2310
2311/**
2312 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2313 * @end: target address
2314 * @pnext: out arg for the next vmap_area
2315 * @pprev: out arg for the previous vmap_area
2316 *
2317 * Returns: %true if either or both of next and prev are found,
2318 * %false if no vmap_area exists
2319 *
2320 * Find vmap_areas end addresses of which enclose @end. ie. if not
2321 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2322 */
2323static bool pvm_find_next_prev(unsigned long end,
2324 struct vmap_area **pnext,
2325 struct vmap_area **pprev)
2326{
2327 struct rb_node *n = vmap_area_root.rb_node;
2328 struct vmap_area *va = NULL;
2329
2330 while (n) {
2331 va = rb_entry(n, struct vmap_area, rb_node);
2332 if (end < va->va_end)
2333 n = n->rb_left;
2334 else if (end > va->va_end)
2335 n = n->rb_right;
2336 else
2337 break;
2338 }
2339
2340 if (!va)
2341 return false;
2342
2343 if (va->va_end > end) {
2344 *pnext = va;
2345 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2346 } else {
2347 *pprev = va;
2348 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2349 }
2350 return true;
2351}
2352
2353/**
2354 * pvm_determine_end - find the highest aligned address between two vmap_areas
2355 * @pnext: in/out arg for the next vmap_area
2356 * @pprev: in/out arg for the previous vmap_area
2357 * @align: alignment
2358 *
2359 * Returns: determined end address
2360 *
2361 * Find the highest aligned address between *@pnext and *@pprev below
2362 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2363 * down address is between the end addresses of the two vmap_areas.
2364 *
2365 * Please note that the address returned by this function may fall
2366 * inside *@pnext vmap_area. The caller is responsible for checking
2367 * that.
2368 */
2369static unsigned long pvm_determine_end(struct vmap_area **pnext,
2370 struct vmap_area **pprev,
2371 unsigned long align)
2372{
2373 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2374 unsigned long addr;
2375
2376 if (*pnext)
2377 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2378 else
2379 addr = vmalloc_end;
2380
2381 while (*pprev && (*pprev)->va_end > addr) {
2382 *pnext = *pprev;
2383 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2384 }
2385
2386 return addr;
2387}
2388
2389/**
2390 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2391 * @offsets: array containing offset of each area
2392 * @sizes: array containing size of each area
2393 * @nr_vms: the number of areas to allocate
2394 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002395 *
2396 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2397 * vm_structs on success, %NULL on failure
2398 *
2399 * Percpu allocator wants to use congruent vm areas so that it can
2400 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002401 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2402 * be scattered pretty far, distance between two areas easily going up
2403 * to gigabytes. To avoid interacting with regular vmallocs, these
2404 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002405 *
2406 * Despite its complicated look, this allocator is rather simple. It
2407 * does everything top-down and scans areas from the end looking for
2408 * matching slot. While scanning, if any of the areas overlaps with
2409 * existing vmap_area, the base address is pulled down to fit the
2410 * area. Scanning is repeated till all the areas fit and then all
2411 * necessary data structres are inserted and the result is returned.
2412 */
2413struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2414 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002415 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002416{
2417 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2418 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2419 struct vmap_area **vas, *prev, *next;
2420 struct vm_struct **vms;
2421 int area, area2, last_area, term_area;
2422 unsigned long base, start, end, last_end;
2423 bool purged = false;
2424
Tejun Heoca23e402009-08-14 15:00:52 +09002425 /* verify parameters and allocate data structures */
Alexander Kuleshov891c49a2015-11-05 18:46:51 -08002426 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
Tejun Heoca23e402009-08-14 15:00:52 +09002427 for (last_area = 0, area = 0; area < nr_vms; area++) {
2428 start = offsets[area];
2429 end = start + sizes[area];
2430
2431 /* is everything aligned properly? */
2432 BUG_ON(!IS_ALIGNED(offsets[area], align));
2433 BUG_ON(!IS_ALIGNED(sizes[area], align));
2434
2435 /* detect the area with the highest address */
2436 if (start > offsets[last_area])
2437 last_area = area;
2438
2439 for (area2 = 0; area2 < nr_vms; area2++) {
2440 unsigned long start2 = offsets[area2];
2441 unsigned long end2 = start2 + sizes[area2];
2442
2443 if (area2 == area)
2444 continue;
2445
2446 BUG_ON(start2 >= start && start2 < end);
2447 BUG_ON(end2 <= end && end2 > start);
2448 }
2449 }
2450 last_end = offsets[last_area] + sizes[last_area];
2451
2452 if (vmalloc_end - vmalloc_start < last_end) {
2453 WARN_ON(true);
2454 return NULL;
2455 }
2456
Thomas Meyer4d67d862012-05-29 15:06:21 -07002457 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2458 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002459 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002460 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002461
2462 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002463 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2464 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002465 if (!vas[area] || !vms[area])
2466 goto err_free;
2467 }
2468retry:
2469 spin_lock(&vmap_area_lock);
2470
2471 /* start scanning - we scan from the top, begin with the last area */
2472 area = term_area = last_area;
2473 start = offsets[area];
2474 end = start + sizes[area];
2475
2476 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2477 base = vmalloc_end - last_end;
2478 goto found;
2479 }
2480 base = pvm_determine_end(&next, &prev, align) - end;
2481
2482 while (true) {
2483 BUG_ON(next && next->va_end <= base + end);
2484 BUG_ON(prev && prev->va_end > base + end);
2485
2486 /*
2487 * base might have underflowed, add last_end before
2488 * comparing.
2489 */
2490 if (base + last_end < vmalloc_start + last_end) {
2491 spin_unlock(&vmap_area_lock);
2492 if (!purged) {
2493 purge_vmap_area_lazy();
2494 purged = true;
2495 goto retry;
2496 }
2497 goto err_free;
2498 }
2499
2500 /*
2501 * If next overlaps, move base downwards so that it's
2502 * right below next and then recheck.
2503 */
2504 if (next && next->va_start < base + end) {
2505 base = pvm_determine_end(&next, &prev, align) - end;
2506 term_area = area;
2507 continue;
2508 }
2509
2510 /*
2511 * If prev overlaps, shift down next and prev and move
2512 * base so that it's right below new next and then
2513 * recheck.
2514 */
2515 if (prev && prev->va_end > base + start) {
2516 next = prev;
2517 prev = node_to_va(rb_prev(&next->rb_node));
2518 base = pvm_determine_end(&next, &prev, align) - end;
2519 term_area = area;
2520 continue;
2521 }
2522
2523 /*
2524 * This area fits, move on to the previous one. If
2525 * the previous one is the terminal one, we're done.
2526 */
2527 area = (area + nr_vms - 1) % nr_vms;
2528 if (area == term_area)
2529 break;
2530 start = offsets[area];
2531 end = start + sizes[area];
2532 pvm_find_next_prev(base + end, &next, &prev);
2533 }
2534found:
2535 /* we've found a fitting base, insert all va's */
2536 for (area = 0; area < nr_vms; area++) {
2537 struct vmap_area *va = vas[area];
2538
2539 va->va_start = base + offsets[area];
2540 va->va_end = va->va_start + sizes[area];
2541 __insert_vmap_area(va);
2542 }
2543
2544 vmap_area_pcpu_hole = base + offsets[last_area];
2545
2546 spin_unlock(&vmap_area_lock);
2547
2548 /* insert all vm's */
2549 for (area = 0; area < nr_vms; area++)
Zhang Yanfei3645cb42013-07-03 15:04:48 -07002550 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2551 pcpu_get_vm_areas);
Tejun Heoca23e402009-08-14 15:00:52 +09002552
2553 kfree(vas);
2554 return vms;
2555
2556err_free:
2557 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002558 kfree(vas[area]);
2559 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002560 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002561err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002562 kfree(vas);
2563 kfree(vms);
2564 return NULL;
2565}
2566
2567/**
2568 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2569 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2570 * @nr_vms: the number of allocated areas
2571 *
2572 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2573 */
2574void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2575{
2576 int i;
2577
2578 for (i = 0; i < nr_vms; i++)
2579 free_vm_area(vms[i]);
2580 kfree(vms);
2581}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002582#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002583
2584#ifdef CONFIG_PROC_FS
2585static void *s_start(struct seq_file *m, loff_t *pos)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002586 __acquires(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002587{
2588 loff_t n = *pos;
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002589 struct vmap_area *va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002590
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002591 spin_lock(&vmap_area_lock);
Geliang Tang6219c2a2016-01-14 15:19:08 -08002592 va = list_first_entry(&vmap_area_list, typeof(*va), list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002593 while (n > 0 && &va->list != &vmap_area_list) {
Christoph Lametera10aa572008-04-28 02:12:40 -07002594 n--;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002595 va = list_next_entry(va, list);
Christoph Lametera10aa572008-04-28 02:12:40 -07002596 }
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002597 if (!n && &va->list != &vmap_area_list)
2598 return va;
Christoph Lametera10aa572008-04-28 02:12:40 -07002599
2600 return NULL;
2601
2602}
2603
2604static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2605{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002606 struct vmap_area *va = p, *next;
Christoph Lametera10aa572008-04-28 02:12:40 -07002607
2608 ++*pos;
Geliang Tang6219c2a2016-01-14 15:19:08 -08002609 next = list_next_entry(va, list);
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002610 if (&next->list != &vmap_area_list)
2611 return next;
2612
2613 return NULL;
Christoph Lametera10aa572008-04-28 02:12:40 -07002614}
2615
2616static void s_stop(struct seq_file *m, void *p)
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002617 __releases(&vmap_area_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002618{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002619 spin_unlock(&vmap_area_lock);
Christoph Lametera10aa572008-04-28 02:12:40 -07002620}
2621
Eric Dumazeta47a1262008-07-23 21:27:38 -07002622static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2623{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002624 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002625 unsigned int nr, *counters = m->private;
2626
2627 if (!counters)
2628 return;
2629
Wanpeng Liaf123462013-11-12 15:07:32 -08002630 if (v->flags & VM_UNINITIALIZED)
2631 return;
Dmitry Vyukov7e5b5282014-12-12 16:56:30 -08002632 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2633 smp_rmb();
Wanpeng Liaf123462013-11-12 15:07:32 -08002634
Eric Dumazeta47a1262008-07-23 21:27:38 -07002635 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2636
2637 for (nr = 0; nr < v->nr_pages; nr++)
2638 counters[page_to_nid(v->pages[nr])]++;
2639
2640 for_each_node_state(nr, N_HIGH_MEMORY)
2641 if (counters[nr])
2642 seq_printf(m, " N%u=%u", nr, counters[nr]);
2643 }
2644}
2645
Christoph Lametera10aa572008-04-28 02:12:40 -07002646static int s_show(struct seq_file *m, void *p)
2647{
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002648 struct vmap_area *va = p;
2649 struct vm_struct *v;
2650
Wanpeng Lic2ce8c12013-11-12 15:07:31 -08002651 /*
2652 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2653 * behalf of vmap area is being tear down or vm_map_ram allocation.
2654 */
2655 if (!(va->flags & VM_VM_AREA))
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002656 return 0;
2657
Joonsoo Kimd4033af2013-04-29 15:07:35 -07002658 v = va->vm;
Christoph Lametera10aa572008-04-28 02:12:40 -07002659
Kees Cook45ec1692012-10-08 16:34:09 -07002660 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002661 v->addr, v->addr + v->size, v->size);
2662
Joe Perches62c70bc2011-01-13 15:45:52 -08002663 if (v->caller)
2664 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002665
Christoph Lametera10aa572008-04-28 02:12:40 -07002666 if (v->nr_pages)
2667 seq_printf(m, " pages=%d", v->nr_pages);
2668
2669 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002670 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002671
2672 if (v->flags & VM_IOREMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002673 seq_puts(m, " ioremap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002674
2675 if (v->flags & VM_ALLOC)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002676 seq_puts(m, " vmalloc");
Christoph Lametera10aa572008-04-28 02:12:40 -07002677
2678 if (v->flags & VM_MAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002679 seq_puts(m, " vmap");
Christoph Lametera10aa572008-04-28 02:12:40 -07002680
2681 if (v->flags & VM_USERMAP)
Fabian Frederickf4527c92014-06-04 16:08:09 -07002682 seq_puts(m, " user");
Christoph Lametera10aa572008-04-28 02:12:40 -07002683
David Rientjes244d63e2016-01-14 15:19:35 -08002684 if (is_vmalloc_addr(v->pages))
Fabian Frederickf4527c92014-06-04 16:08:09 -07002685 seq_puts(m, " vpages");
Christoph Lametera10aa572008-04-28 02:12:40 -07002686
Eric Dumazeta47a1262008-07-23 21:27:38 -07002687 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002688 seq_putc(m, '\n');
2689 return 0;
2690}
2691
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002692static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002693 .start = s_start,
2694 .next = s_next,
2695 .stop = s_stop,
2696 .show = s_show,
2697};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002698
2699static int vmalloc_open(struct inode *inode, struct file *file)
2700{
Rob Jones703394c2014-10-09 15:28:01 -07002701 if (IS_ENABLED(CONFIG_NUMA))
2702 return seq_open_private(file, &vmalloc_op,
2703 nr_node_ids * sizeof(unsigned int));
2704 else
2705 return seq_open(file, &vmalloc_op);
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002706}
2707
2708static const struct file_operations proc_vmalloc_operations = {
2709 .open = vmalloc_open,
2710 .read = seq_read,
2711 .llseek = seq_lseek,
2712 .release = seq_release_private,
2713};
2714
2715static int __init proc_vmalloc_init(void)
2716{
2717 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2718 return 0;
2719}
2720module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002721
Christoph Lametera10aa572008-04-28 02:12:40 -07002722#endif
2723