blob: 1bf94ad452b6ab3c47b1ccce4d83422f2cb42cb9 [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>
24#include <linux/rbtree.h>
25#include <linux/radix-tree.h>
26#include <linux/rcupdate.h>
Tejun Heof0aa6612009-02-20 16:29:08 +090027#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010028#include <linux/kmemleak.h>
Arun Sharma600634972011-07-26 16:09:06 -070029#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/uaccess.h>
31#include <asm/tlbflush.h>
David Miller2dca6992009-09-21 12:22:34 -070032#include <asm/shmparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Nick Piggindb64fe02008-10-18 20:27:03 -070034/*** Page table manipulation functions ***/
Adrian Bunkb2213852006-09-25 23:31:02 -070035
Linus Torvalds1da177e2005-04-16 15:20:36 -070036static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
37{
38 pte_t *pte;
39
40 pte = pte_offset_kernel(pmd, addr);
41 do {
42 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
43 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
44 } while (pte++, addr += PAGE_SIZE, addr != end);
45}
46
Nick Piggindb64fe02008-10-18 20:27:03 -070047static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048{
49 pmd_t *pmd;
50 unsigned long next;
51
52 pmd = pmd_offset(pud, addr);
53 do {
54 next = pmd_addr_end(addr, end);
55 if (pmd_none_or_clear_bad(pmd))
56 continue;
57 vunmap_pte_range(pmd, addr, next);
58 } while (pmd++, addr = next, addr != end);
59}
60
Nick Piggindb64fe02008-10-18 20:27:03 -070061static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062{
63 pud_t *pud;
64 unsigned long next;
65
66 pud = pud_offset(pgd, addr);
67 do {
68 next = pud_addr_end(addr, end);
69 if (pud_none_or_clear_bad(pud))
70 continue;
71 vunmap_pmd_range(pud, addr, next);
72 } while (pud++, addr = next, addr != end);
73}
74
Nick Piggindb64fe02008-10-18 20:27:03 -070075static void vunmap_page_range(unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
77 pgd_t *pgd;
78 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80 BUG_ON(addr >= end);
81 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 do {
83 next = pgd_addr_end(addr, end);
84 if (pgd_none_or_clear_bad(pgd))
85 continue;
86 vunmap_pud_range(pgd, addr, next);
87 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088}
89
90static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
Nick Piggindb64fe02008-10-18 20:27:03 -070091 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 pte_t *pte;
94
Nick Piggindb64fe02008-10-18 20:27:03 -070095 /*
96 * nr is a running index into the array which helps higher level
97 * callers keep track of where we're up to.
98 */
99
Hugh Dickins872fec12005-10-29 18:16:21 -0700100 pte = pte_alloc_kernel(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 if (!pte)
102 return -ENOMEM;
103 do {
Nick Piggindb64fe02008-10-18 20:27:03 -0700104 struct page *page = pages[*nr];
105
106 if (WARN_ON(!pte_none(*pte)))
107 return -EBUSY;
108 if (WARN_ON(!page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 return -ENOMEM;
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
Nick Piggindb64fe02008-10-18 20:27:03 -0700111 (*nr)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 } while (pte++, addr += PAGE_SIZE, addr != end);
113 return 0;
114}
115
Nick Piggindb64fe02008-10-18 20:27:03 -0700116static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 pmd_t *pmd;
120 unsigned long next;
121
122 pmd = pmd_alloc(&init_mm, pud, addr);
123 if (!pmd)
124 return -ENOMEM;
125 do {
126 next = pmd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700127 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return -ENOMEM;
129 } while (pmd++, addr = next, addr != end);
130 return 0;
131}
132
Nick Piggindb64fe02008-10-18 20:27:03 -0700133static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
136 pud_t *pud;
137 unsigned long next;
138
139 pud = pud_alloc(&init_mm, pgd, addr);
140 if (!pud)
141 return -ENOMEM;
142 do {
143 next = pud_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700144 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 return -ENOMEM;
146 } while (pud++, addr = next, addr != end);
147 return 0;
148}
149
Nick Piggindb64fe02008-10-18 20:27:03 -0700150/*
151 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152 * will have pfns corresponding to the "pages" array.
153 *
154 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
155 */
Tejun Heo8fc48982009-02-20 16:29:08 +0900156static int vmap_page_range_noflush(unsigned long start, unsigned long end,
157 pgprot_t prot, struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
159 pgd_t *pgd;
160 unsigned long next;
Adam Lackorzynski2e4e27c2009-01-04 12:00:46 -0800161 unsigned long addr = start;
Nick Piggindb64fe02008-10-18 20:27:03 -0700162 int err = 0;
163 int nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 BUG_ON(addr >= end);
166 pgd = pgd_offset_k(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 do {
168 next = pgd_addr_end(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700169 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 if (err)
Figo.zhangbf88c8c2009-09-21 17:01:47 -0700171 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 } while (pgd++, addr = next, addr != end);
Nick Piggindb64fe02008-10-18 20:27:03 -0700173
Nick Piggindb64fe02008-10-18 20:27:03 -0700174 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
Tejun Heo8fc48982009-02-20 16:29:08 +0900177static int vmap_page_range(unsigned long start, unsigned long end,
178 pgprot_t prot, struct page **pages)
179{
180 int ret;
181
182 ret = vmap_page_range_noflush(start, end, prot, pages);
183 flush_cache_vmap(start, end);
184 return ret;
185}
186
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700187int is_vmalloc_or_module_addr(const void *x)
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700188{
189 /*
Russell Kingab4f2ee2008-11-06 17:11:07 +0000190 * ARM, x86-64 and sparc64 put modules in a special place,
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700191 * and fall back on vmalloc() if that fails. Others
192 * just put it in the vmalloc space.
193 */
194#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
195 unsigned long addr = (unsigned long)x;
196 if (addr >= MODULES_VADDR && addr < MODULES_END)
197 return 1;
198#endif
199 return is_vmalloc_addr(x);
200}
201
Christoph Lameter48667e72008-02-04 22:28:31 -0800202/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700203 * Walk a vmap address to the struct page it maps.
Christoph Lameter48667e72008-02-04 22:28:31 -0800204 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800205struct page *vmalloc_to_page(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800206{
207 unsigned long addr = (unsigned long) vmalloc_addr;
208 struct page *page = NULL;
209 pgd_t *pgd = pgd_offset_k(addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800210
Ingo Molnar7aa413d2008-06-19 13:28:11 +0200211 /*
212 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
213 * architectures that do not vmalloc module space
214 */
Linus Torvalds73bdf0a2008-10-15 08:35:12 -0700215 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
Jiri Slaby59ea7462008-06-12 13:56:40 +0200216
Christoph Lameter48667e72008-02-04 22:28:31 -0800217 if (!pgd_none(*pgd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700218 pud_t *pud = pud_offset(pgd, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800219 if (!pud_none(*pud)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700220 pmd_t *pmd = pmd_offset(pud, addr);
Christoph Lameter48667e72008-02-04 22:28:31 -0800221 if (!pmd_none(*pmd)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700222 pte_t *ptep, pte;
223
Christoph Lameter48667e72008-02-04 22:28:31 -0800224 ptep = pte_offset_map(pmd, addr);
225 pte = *ptep;
226 if (pte_present(pte))
227 page = pte_page(pte);
228 pte_unmap(ptep);
229 }
230 }
231 }
232 return page;
233}
234EXPORT_SYMBOL(vmalloc_to_page);
235
236/*
237 * Map a vmalloc()-space virtual address to the physical page frame number.
238 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800239unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
Christoph Lameter48667e72008-02-04 22:28:31 -0800240{
241 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
242}
243EXPORT_SYMBOL(vmalloc_to_pfn);
244
Nick Piggindb64fe02008-10-18 20:27:03 -0700245
246/*** Global kva allocator ***/
247
248#define VM_LAZY_FREE 0x01
249#define VM_LAZY_FREEING 0x02
250#define VM_VM_AREA 0x04
251
252struct vmap_area {
253 unsigned long va_start;
254 unsigned long va_end;
255 unsigned long flags;
256 struct rb_node rb_node; /* address sorted rbtree */
257 struct list_head list; /* address sorted list */
258 struct list_head purge_list; /* "lazy purge" list */
Minchan Kimdb1aeca2012-01-10 15:08:39 -0800259 struct vm_struct *vm;
Nick Piggindb64fe02008-10-18 20:27:03 -0700260 struct rcu_head rcu_head;
261};
262
263static DEFINE_SPINLOCK(vmap_area_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700264static LIST_HEAD(vmap_area_list);
Nick Piggin89699602011-03-22 16:30:36 -0700265static struct rb_root vmap_area_root = RB_ROOT;
266
267/* The vmap cache globals are protected by vmap_area_lock */
268static struct rb_node *free_vmap_cache;
269static unsigned long cached_hole_size;
270static unsigned long cached_vstart;
271static unsigned long cached_align;
272
Tejun Heoca23e402009-08-14 15:00:52 +0900273static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700274
275static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
Nick Piggindb64fe02008-10-18 20:27:03 -0700277 struct rb_node *n = vmap_area_root.rb_node;
278
279 while (n) {
280 struct vmap_area *va;
281
282 va = rb_entry(n, struct vmap_area, rb_node);
283 if (addr < va->va_start)
284 n = n->rb_left;
285 else if (addr > va->va_start)
286 n = n->rb_right;
287 else
288 return va;
289 }
290
291 return NULL;
292}
293
294static void __insert_vmap_area(struct vmap_area *va)
295{
296 struct rb_node **p = &vmap_area_root.rb_node;
297 struct rb_node *parent = NULL;
298 struct rb_node *tmp;
299
300 while (*p) {
Namhyung Kim170168d2010-10-26 14:22:02 -0700301 struct vmap_area *tmp_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700302
303 parent = *p;
Namhyung Kim170168d2010-10-26 14:22:02 -0700304 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
305 if (va->va_start < tmp_va->va_end)
Nick Piggindb64fe02008-10-18 20:27:03 -0700306 p = &(*p)->rb_left;
Namhyung Kim170168d2010-10-26 14:22:02 -0700307 else if (va->va_end > tmp_va->va_start)
Nick Piggindb64fe02008-10-18 20:27:03 -0700308 p = &(*p)->rb_right;
309 else
310 BUG();
311 }
312
313 rb_link_node(&va->rb_node, parent, p);
314 rb_insert_color(&va->rb_node, &vmap_area_root);
315
316 /* address-sort this list so it is usable like the vmlist */
317 tmp = rb_prev(&va->rb_node);
318 if (tmp) {
319 struct vmap_area *prev;
320 prev = rb_entry(tmp, struct vmap_area, rb_node);
321 list_add_rcu(&va->list, &prev->list);
322 } else
323 list_add_rcu(&va->list, &vmap_area_list);
324}
325
326static void purge_vmap_area_lazy(void);
327
328/*
329 * Allocate a region of KVA of the specified size and alignment, within the
330 * vstart and vend.
331 */
332static struct vmap_area *alloc_vmap_area(unsigned long size,
333 unsigned long align,
334 unsigned long vstart, unsigned long vend,
335 int node, gfp_t gfp_mask)
336{
337 struct vmap_area *va;
338 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700340 int purged = 0;
Nick Piggin89699602011-03-22 16:30:36 -0700341 struct vmap_area *first;
Nick Piggindb64fe02008-10-18 20:27:03 -0700342
Nick Piggin77669702009-02-27 14:03:03 -0800343 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700344 BUG_ON(size & ~PAGE_MASK);
Nick Piggin89699602011-03-22 16:30:36 -0700345 BUG_ON(!is_power_of_2(align));
Nick Piggindb64fe02008-10-18 20:27:03 -0700346
Nick Piggindb64fe02008-10-18 20:27:03 -0700347 va = kmalloc_node(sizeof(struct vmap_area),
348 gfp_mask & GFP_RECLAIM_MASK, node);
349 if (unlikely(!va))
350 return ERR_PTR(-ENOMEM);
351
352retry:
353 spin_lock(&vmap_area_lock);
Nick Piggin89699602011-03-22 16:30:36 -0700354 /*
355 * Invalidate cache if we have more permissive parameters.
356 * cached_hole_size notes the largest hole noticed _below_
357 * the vmap_area cached in free_vmap_cache: if size fits
358 * into that hole, we want to scan from vstart to reuse
359 * the hole instead of allocating above free_vmap_cache.
360 * Note that __free_vmap_area may update free_vmap_cache
361 * without updating cached_hole_size or cached_align.
362 */
363 if (!free_vmap_cache ||
364 size < cached_hole_size ||
365 vstart < cached_vstart ||
366 align < cached_align) {
367nocache:
368 cached_hole_size = 0;
369 free_vmap_cache = NULL;
370 }
371 /* record if we encounter less permissive parameters */
372 cached_vstart = vstart;
373 cached_align = align;
Nick Piggin77669702009-02-27 14:03:03 -0800374
Nick Piggin89699602011-03-22 16:30:36 -0700375 /* find starting point for our search */
376 if (free_vmap_cache) {
377 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700378 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700379 if (addr < vstart)
380 goto nocache;
381 if (addr + size - 1 < addr)
382 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700383
Nick Piggin89699602011-03-22 16:30:36 -0700384 } else {
385 addr = ALIGN(vstart, align);
386 if (addr + size - 1 < addr)
387 goto overflow;
388
389 n = vmap_area_root.rb_node;
390 first = NULL;
391
392 while (n) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700393 struct vmap_area *tmp;
394 tmp = rb_entry(n, struct vmap_area, rb_node);
395 if (tmp->va_end >= addr) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700396 first = tmp;
Nick Piggin89699602011-03-22 16:30:36 -0700397 if (tmp->va_start <= addr)
398 break;
399 n = n->rb_left;
400 } else
Nick Piggindb64fe02008-10-18 20:27:03 -0700401 n = n->rb_right;
Nick Piggin89699602011-03-22 16:30:36 -0700402 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700403
404 if (!first)
405 goto found;
Nick Piggindb64fe02008-10-18 20:27:03 -0700406 }
Nick Piggin89699602011-03-22 16:30:36 -0700407
408 /* from the starting point, walk areas until a suitable hole is found */
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700409 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggin89699602011-03-22 16:30:36 -0700410 if (addr + cached_hole_size < first->va_start)
411 cached_hole_size = first->va_start - addr;
Johannes Weiner248ac0e2011-05-24 17:11:43 -0700412 addr = ALIGN(first->va_end, align);
Nick Piggin89699602011-03-22 16:30:36 -0700413 if (addr + size - 1 < addr)
414 goto overflow;
415
Hong zhi guo92ca9222012-07-31 16:41:35 -0700416 if (list_is_last(&first->list, &vmap_area_list))
Nick Piggin89699602011-03-22 16:30:36 -0700417 goto found;
Hong zhi guo92ca9222012-07-31 16:41:35 -0700418
419 first = list_entry(first->list.next,
420 struct vmap_area, list);
Nick Piggin89699602011-03-22 16:30:36 -0700421 }
422
Nick Piggindb64fe02008-10-18 20:27:03 -0700423found:
Nick Piggin89699602011-03-22 16:30:36 -0700424 if (addr + size > vend)
425 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700426
427 va->va_start = addr;
428 va->va_end = addr + size;
429 va->flags = 0;
430 __insert_vmap_area(va);
Nick Piggin89699602011-03-22 16:30:36 -0700431 free_vmap_cache = &va->rb_node;
Nick Piggindb64fe02008-10-18 20:27:03 -0700432 spin_unlock(&vmap_area_lock);
433
Nick Piggin89699602011-03-22 16:30:36 -0700434 BUG_ON(va->va_start & (align-1));
435 BUG_ON(va->va_start < vstart);
436 BUG_ON(va->va_end > vend);
437
Nick Piggindb64fe02008-10-18 20:27:03 -0700438 return va;
Nick Piggin89699602011-03-22 16:30:36 -0700439
440overflow:
441 spin_unlock(&vmap_area_lock);
442 if (!purged) {
443 purge_vmap_area_lazy();
444 purged = 1;
445 goto retry;
446 }
447 if (printk_ratelimit())
448 printk(KERN_WARNING
449 "vmap allocation for size %lu failed: "
450 "use vmalloc=<size> to increase size.\n", size);
451 kfree(va);
452 return ERR_PTR(-EBUSY);
Nick Piggindb64fe02008-10-18 20:27:03 -0700453}
454
Nick Piggindb64fe02008-10-18 20:27:03 -0700455static void __free_vmap_area(struct vmap_area *va)
456{
457 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
Nick Piggin89699602011-03-22 16:30:36 -0700458
459 if (free_vmap_cache) {
460 if (va->va_end < cached_vstart) {
461 free_vmap_cache = NULL;
462 } else {
463 struct vmap_area *cache;
464 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
465 if (va->va_start <= cache->va_start) {
466 free_vmap_cache = rb_prev(&va->rb_node);
467 /*
468 * We don't try to update cached_hole_size or
469 * cached_align, but it won't go very wrong.
470 */
471 }
472 }
473 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700474 rb_erase(&va->rb_node, &vmap_area_root);
475 RB_CLEAR_NODE(&va->rb_node);
476 list_del_rcu(&va->list);
477
Tejun Heoca23e402009-08-14 15:00:52 +0900478 /*
479 * Track the highest possible candidate for pcpu area
480 * allocation. Areas outside of vmalloc area can be returned
481 * here too, consider only end addresses which fall inside
482 * vmalloc area proper.
483 */
484 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
485 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
486
Lai Jiangshan14769de2011-03-18 12:12:19 +0800487 kfree_rcu(va, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700488}
489
490/*
491 * Free a region of KVA allocated by alloc_vmap_area
492 */
493static void free_vmap_area(struct vmap_area *va)
494{
495 spin_lock(&vmap_area_lock);
496 __free_vmap_area(va);
497 spin_unlock(&vmap_area_lock);
498}
499
500/*
501 * Clear the pagetable entries of a given vmap_area
502 */
503static void unmap_vmap_area(struct vmap_area *va)
504{
505 vunmap_page_range(va->va_start, va->va_end);
506}
507
Nick Piggincd528582009-01-06 14:39:20 -0800508static void vmap_debug_free_range(unsigned long start, unsigned long end)
509{
510 /*
511 * Unmap page tables and force a TLB flush immediately if
512 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
513 * bugs similarly to those in linear kernel virtual address
514 * space after a page has been freed.
515 *
516 * All the lazy freeing logic is still retained, in order to
517 * minimise intrusiveness of this debugging feature.
518 *
519 * This is going to be *slow* (linear kernel virtual address
520 * debugging doesn't do a broadcast TLB flush so it is a lot
521 * faster).
522 */
523#ifdef CONFIG_DEBUG_PAGEALLOC
524 vunmap_page_range(start, end);
525 flush_tlb_kernel_range(start, end);
526#endif
527}
528
Nick Piggindb64fe02008-10-18 20:27:03 -0700529/*
530 * lazy_max_pages is the maximum amount of virtual address space we gather up
531 * before attempting to purge with a TLB flush.
532 *
533 * There is a tradeoff here: a larger number will cover more kernel page tables
534 * and take slightly longer to purge, but it will linearly reduce the number of
535 * global TLB flushes that must be performed. It would seem natural to scale
536 * this number up linearly with the number of CPUs (because vmapping activity
537 * could also scale linearly with the number of CPUs), however it is likely
538 * that in practice, workloads might be constrained in other ways that mean
539 * vmap activity will not scale linearly with CPUs. Also, I want to be
540 * conservative and not introduce a big latency on huge systems, so go with
541 * a less aggressive log scale. It will still be an improvement over the old
542 * code, and it will be simple to change the scale factor if we find that it
543 * becomes a problem on bigger systems.
544 */
545static unsigned long lazy_max_pages(void)
546{
547 unsigned int log;
548
549 log = fls(num_online_cpus());
550
551 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
552}
553
554static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
555
Nick Piggin02b709d2010-02-01 22:25:57 +1100556/* for per-CPU blocks */
557static void purge_fragmented_blocks_allcpus(void);
558
Nick Piggindb64fe02008-10-18 20:27:03 -0700559/*
Cliff Wickman3ee48b62010-09-16 11:44:02 -0500560 * called before a call to iounmap() if the caller wants vm_area_struct's
561 * immediately freed.
562 */
563void set_iounmap_nonlazy(void)
564{
565 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
566}
567
568/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700569 * Purges all lazily-freed vmap areas.
570 *
571 * If sync is 0 then don't purge if there is already a purge in progress.
572 * If force_flush is 1, then flush kernel TLBs between *start and *end even
573 * if we found no lazy vmap areas to unmap (callers can use this to optimise
574 * their own TLB flushing).
575 * Returns with *start = min(*start, lowest purged address)
576 * *end = max(*end, highest purged address)
577 */
578static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
579 int sync, int force_flush)
580{
Andrew Morton46666d82009-01-15 13:51:15 -0800581 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700582 LIST_HEAD(valist);
583 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800584 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700585 int nr = 0;
586
587 /*
588 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
589 * should not expect such behaviour. This just simplifies locking for
590 * the case that isn't actually used at the moment anyway.
591 */
592 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800593 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700594 return;
595 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800596 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700597
Nick Piggin02b709d2010-02-01 22:25:57 +1100598 if (sync)
599 purge_fragmented_blocks_allcpus();
600
Nick Piggindb64fe02008-10-18 20:27:03 -0700601 rcu_read_lock();
602 list_for_each_entry_rcu(va, &vmap_area_list, list) {
603 if (va->flags & VM_LAZY_FREE) {
604 if (va->va_start < *start)
605 *start = va->va_start;
606 if (va->va_end > *end)
607 *end = va->va_end;
608 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
Nick Piggindb64fe02008-10-18 20:27:03 -0700609 list_add_tail(&va->purge_list, &valist);
610 va->flags |= VM_LAZY_FREEING;
611 va->flags &= ~VM_LAZY_FREE;
612 }
613 }
614 rcu_read_unlock();
615
Yongseok Koh88f50042010-01-19 17:33:49 +0900616 if (nr)
Nick Piggindb64fe02008-10-18 20:27:03 -0700617 atomic_sub(nr, &vmap_lazy_nr);
Nick Piggindb64fe02008-10-18 20:27:03 -0700618
619 if (nr || force_flush)
620 flush_tlb_kernel_range(*start, *end);
621
622 if (nr) {
623 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800624 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700625 __free_vmap_area(va);
626 spin_unlock(&vmap_area_lock);
627 }
Andrew Morton46666d82009-01-15 13:51:15 -0800628 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700629}
630
631/*
Nick Piggin496850e2008-11-19 15:36:33 -0800632 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
633 * is already purging.
634 */
635static void try_purge_vmap_area_lazy(void)
636{
637 unsigned long start = ULONG_MAX, end = 0;
638
639 __purge_vmap_area_lazy(&start, &end, 0, 0);
640}
641
642/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700643 * Kick off a purge of the outstanding lazy areas.
644 */
645static void purge_vmap_area_lazy(void)
646{
647 unsigned long start = ULONG_MAX, end = 0;
648
Nick Piggin496850e2008-11-19 15:36:33 -0800649 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700650}
651
652/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800653 * Free a vmap area, caller ensuring that the area has been unmapped
654 * and flush_cache_vunmap had been called for the correct range
655 * previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700656 */
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800657static void free_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700658{
659 va->flags |= VM_LAZY_FREE;
660 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
661 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800662 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700663}
664
Nick Pigginb29acbd2008-12-01 13:13:47 -0800665/*
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800666 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
667 * called for the correct range previously.
668 */
669static void free_unmap_vmap_area_noflush(struct vmap_area *va)
670{
671 unmap_vmap_area(va);
672 free_vmap_area_noflush(va);
673}
674
675/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800676 * Free and unmap a vmap area
677 */
678static void free_unmap_vmap_area(struct vmap_area *va)
679{
680 flush_cache_vunmap(va->va_start, va->va_end);
681 free_unmap_vmap_area_noflush(va);
682}
683
Nick Piggindb64fe02008-10-18 20:27:03 -0700684static struct vmap_area *find_vmap_area(unsigned long addr)
685{
686 struct vmap_area *va;
687
688 spin_lock(&vmap_area_lock);
689 va = __find_vmap_area(addr);
690 spin_unlock(&vmap_area_lock);
691
692 return va;
693}
694
695static void free_unmap_vmap_area_addr(unsigned long addr)
696{
697 struct vmap_area *va;
698
699 va = find_vmap_area(addr);
700 BUG_ON(!va);
701 free_unmap_vmap_area(va);
702}
703
704
705/*** Per cpu kva allocator ***/
706
707/*
708 * vmap space is limited especially on 32 bit architectures. Ensure there is
709 * room for at least 16 percpu vmap blocks per CPU.
710 */
711/*
712 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
713 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
714 * instead (we just need a rough idea)
715 */
716#if BITS_PER_LONG == 32
717#define VMALLOC_SPACE (128UL*1024*1024)
718#else
719#define VMALLOC_SPACE (128UL*1024*1024*1024)
720#endif
721
722#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
723#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
724#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
725#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
726#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
727#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
Clemens Ladischf982f912011-06-21 22:09:50 +0200728#define VMAP_BBMAP_BITS \
729 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
730 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
731 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
Nick Piggindb64fe02008-10-18 20:27:03 -0700732
733#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
734
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100735static bool vmap_initialized __read_mostly = false;
736
Nick Piggindb64fe02008-10-18 20:27:03 -0700737struct vmap_block_queue {
738 spinlock_t lock;
739 struct list_head free;
Nick Piggindb64fe02008-10-18 20:27:03 -0700740};
741
742struct vmap_block {
743 spinlock_t lock;
744 struct vmap_area *va;
745 struct vmap_block_queue *vbq;
746 unsigned long free, dirty;
747 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
748 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
Nick Pigginde560422010-02-01 22:24:18 +1100749 struct list_head free_list;
750 struct rcu_head rcu_head;
Nick Piggin02b709d2010-02-01 22:25:57 +1100751 struct list_head purge;
Nick Piggindb64fe02008-10-18 20:27:03 -0700752};
753
754/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
755static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
756
757/*
758 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
759 * in the free path. Could get rid of this if we change the API to return a
760 * "cookie" from alloc, to be passed to free. But no big deal yet.
761 */
762static DEFINE_SPINLOCK(vmap_block_tree_lock);
763static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
764
765/*
766 * We should probably have a fallback mechanism to allocate virtual memory
767 * out of partially filled vmap blocks. However vmap block sizing should be
768 * fairly reasonable according to the vmalloc size, so it shouldn't be a
769 * big problem.
770 */
771
772static unsigned long addr_to_vb_idx(unsigned long addr)
773{
774 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
775 addr /= VMAP_BLOCK_SIZE;
776 return addr;
777}
778
779static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
780{
781 struct vmap_block_queue *vbq;
782 struct vmap_block *vb;
783 struct vmap_area *va;
784 unsigned long vb_idx;
785 int node, err;
786
787 node = numa_node_id();
788
789 vb = kmalloc_node(sizeof(struct vmap_block),
790 gfp_mask & GFP_RECLAIM_MASK, node);
791 if (unlikely(!vb))
792 return ERR_PTR(-ENOMEM);
793
794 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
795 VMALLOC_START, VMALLOC_END,
796 node, gfp_mask);
Tobias Klauserddf9c6d2011-01-13 15:46:15 -0800797 if (IS_ERR(va)) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700798 kfree(vb);
Julia Lawalle7d86342010-08-09 17:18:28 -0700799 return ERR_CAST(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700800 }
801
802 err = radix_tree_preload(gfp_mask);
803 if (unlikely(err)) {
804 kfree(vb);
805 free_vmap_area(va);
806 return ERR_PTR(err);
807 }
808
809 spin_lock_init(&vb->lock);
810 vb->va = va;
811 vb->free = VMAP_BBMAP_BITS;
812 vb->dirty = 0;
813 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
814 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
815 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700816
817 vb_idx = addr_to_vb_idx(va->va_start);
818 spin_lock(&vmap_block_tree_lock);
819 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
820 spin_unlock(&vmap_block_tree_lock);
821 BUG_ON(err);
822 radix_tree_preload_end();
823
824 vbq = &get_cpu_var(vmap_block_queue);
825 vb->vbq = vbq;
826 spin_lock(&vbq->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100827 list_add_rcu(&vb->free_list, &vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700828 spin_unlock(&vbq->lock);
Tejun Heo3f04ba82009-10-29 22:34:12 +0900829 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700830
831 return vb;
832}
833
Nick Piggindb64fe02008-10-18 20:27:03 -0700834static void free_vmap_block(struct vmap_block *vb)
835{
836 struct vmap_block *tmp;
837 unsigned long vb_idx;
838
Nick Piggindb64fe02008-10-18 20:27:03 -0700839 vb_idx = addr_to_vb_idx(vb->va->va_start);
840 spin_lock(&vmap_block_tree_lock);
841 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
842 spin_unlock(&vmap_block_tree_lock);
843 BUG_ON(tmp != vb);
844
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800845 free_vmap_area_noflush(vb->va);
Lai Jiangshan22a3c7d2011-03-18 12:13:08 +0800846 kfree_rcu(vb, rcu_head);
Nick Piggindb64fe02008-10-18 20:27:03 -0700847}
848
Nick Piggin02b709d2010-02-01 22:25:57 +1100849static void purge_fragmented_blocks(int cpu)
850{
851 LIST_HEAD(purge);
852 struct vmap_block *vb;
853 struct vmap_block *n_vb;
854 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
855
856 rcu_read_lock();
857 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
858
859 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
860 continue;
861
862 spin_lock(&vb->lock);
863 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
864 vb->free = 0; /* prevent further allocs after releasing lock */
865 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
866 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
867 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
868 spin_lock(&vbq->lock);
869 list_del_rcu(&vb->free_list);
870 spin_unlock(&vbq->lock);
871 spin_unlock(&vb->lock);
872 list_add_tail(&vb->purge, &purge);
873 } else
874 spin_unlock(&vb->lock);
875 }
876 rcu_read_unlock();
877
878 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
879 list_del(&vb->purge);
880 free_vmap_block(vb);
881 }
882}
883
884static void purge_fragmented_blocks_thiscpu(void)
885{
886 purge_fragmented_blocks(smp_processor_id());
887}
888
889static void purge_fragmented_blocks_allcpus(void)
890{
891 int cpu;
892
893 for_each_possible_cpu(cpu)
894 purge_fragmented_blocks(cpu);
895}
896
Nick Piggindb64fe02008-10-18 20:27:03 -0700897static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
898{
899 struct vmap_block_queue *vbq;
900 struct vmap_block *vb;
901 unsigned long addr = 0;
902 unsigned int order;
Nick Piggin02b709d2010-02-01 22:25:57 +1100903 int purge = 0;
Nick Piggindb64fe02008-10-18 20:27:03 -0700904
905 BUG_ON(size & ~PAGE_MASK);
906 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Jan Karaaa91c4d2012-07-31 16:41:37 -0700907 if (WARN_ON(size == 0)) {
908 /*
909 * Allocating 0 bytes isn't what caller wants since
910 * get_order(0) returns funny result. Just warn and terminate
911 * early.
912 */
913 return NULL;
914 }
Nick Piggindb64fe02008-10-18 20:27:03 -0700915 order = get_order(size);
916
917again:
918 rcu_read_lock();
919 vbq = &get_cpu_var(vmap_block_queue);
920 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
921 int i;
922
923 spin_lock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100924 if (vb->free < 1UL << order)
925 goto next;
926
Nick Piggindb64fe02008-10-18 20:27:03 -0700927 i = bitmap_find_free_region(vb->alloc_map,
928 VMAP_BBMAP_BITS, order);
929
Nick Piggin02b709d2010-02-01 22:25:57 +1100930 if (i < 0) {
931 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
932 /* fragmented and no outstanding allocations */
933 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
934 purge = 1;
Nick Piggindb64fe02008-10-18 20:27:03 -0700935 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100936 goto next;
937 }
938 addr = vb->va->va_start + (i << PAGE_SHIFT);
939 BUG_ON(addr_to_vb_idx(addr) !=
940 addr_to_vb_idx(vb->va->va_start));
941 vb->free -= 1UL << order;
942 if (vb->free == 0) {
943 spin_lock(&vbq->lock);
944 list_del_rcu(&vb->free_list);
945 spin_unlock(&vbq->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700946 }
947 spin_unlock(&vb->lock);
Nick Piggin02b709d2010-02-01 22:25:57 +1100948 break;
949next:
950 spin_unlock(&vb->lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700951 }
Nick Piggin02b709d2010-02-01 22:25:57 +1100952
953 if (purge)
954 purge_fragmented_blocks_thiscpu();
955
Tejun Heo3f04ba82009-10-29 22:34:12 +0900956 put_cpu_var(vmap_block_queue);
Nick Piggindb64fe02008-10-18 20:27:03 -0700957 rcu_read_unlock();
958
959 if (!addr) {
960 vb = new_vmap_block(gfp_mask);
961 if (IS_ERR(vb))
962 return vb;
963 goto again;
964 }
965
966 return (void *)addr;
967}
968
969static void vb_free(const void *addr, unsigned long size)
970{
971 unsigned long offset;
972 unsigned long vb_idx;
973 unsigned int order;
974 struct vmap_block *vb;
975
976 BUG_ON(size & ~PAGE_MASK);
977 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800978
979 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
980
Nick Piggindb64fe02008-10-18 20:27:03 -0700981 order = get_order(size);
982
983 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
984
985 vb_idx = addr_to_vb_idx((unsigned long)addr);
986 rcu_read_lock();
987 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
988 rcu_read_unlock();
989 BUG_ON(!vb);
990
Jeremy Fitzhardinge64141da2010-12-02 14:31:18 -0800991 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
992
Nick Piggindb64fe02008-10-18 20:27:03 -0700993 spin_lock(&vb->lock);
Nick Pigginde560422010-02-01 22:24:18 +1100994 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
MinChan Kimd0868172009-03-31 15:19:26 -0700995
Nick Piggindb64fe02008-10-18 20:27:03 -0700996 vb->dirty += 1UL << order;
997 if (vb->dirty == VMAP_BBMAP_BITS) {
Nick Pigginde560422010-02-01 22:24:18 +1100998 BUG_ON(vb->free);
Nick Piggindb64fe02008-10-18 20:27:03 -0700999 spin_unlock(&vb->lock);
1000 free_vmap_block(vb);
1001 } else
1002 spin_unlock(&vb->lock);
1003}
1004
1005/**
1006 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1007 *
1008 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1009 * to amortize TLB flushing overheads. What this means is that any page you
1010 * have now, may, in a former life, have been mapped into kernel virtual
1011 * address by the vmap layer and so there might be some CPUs with TLB entries
1012 * still referencing that page (additional to the regular 1:1 kernel mapping).
1013 *
1014 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1015 * be sure that none of the pages we have control over will have any aliases
1016 * from the vmap layer.
1017 */
1018void vm_unmap_aliases(void)
1019{
1020 unsigned long start = ULONG_MAX, end = 0;
1021 int cpu;
1022 int flush = 0;
1023
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001024 if (unlikely(!vmap_initialized))
1025 return;
1026
Nick Piggindb64fe02008-10-18 20:27:03 -07001027 for_each_possible_cpu(cpu) {
1028 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1029 struct vmap_block *vb;
1030
1031 rcu_read_lock();
1032 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1033 int i;
1034
1035 spin_lock(&vb->lock);
1036 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1037 while (i < VMAP_BBMAP_BITS) {
1038 unsigned long s, e;
1039 int j;
1040 j = find_next_zero_bit(vb->dirty_map,
1041 VMAP_BBMAP_BITS, i);
1042
1043 s = vb->va->va_start + (i << PAGE_SHIFT);
1044 e = vb->va->va_start + (j << PAGE_SHIFT);
Nick Piggindb64fe02008-10-18 20:27:03 -07001045 flush = 1;
1046
1047 if (s < start)
1048 start = s;
1049 if (e > end)
1050 end = e;
1051
1052 i = j;
1053 i = find_next_bit(vb->dirty_map,
1054 VMAP_BBMAP_BITS, i);
1055 }
1056 spin_unlock(&vb->lock);
1057 }
1058 rcu_read_unlock();
1059 }
1060
1061 __purge_vmap_area_lazy(&start, &end, 1, flush);
1062}
1063EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1064
1065/**
1066 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1067 * @mem: the pointer returned by vm_map_ram
1068 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1069 */
1070void vm_unmap_ram(const void *mem, unsigned int count)
1071{
1072 unsigned long size = count << PAGE_SHIFT;
1073 unsigned long addr = (unsigned long)mem;
1074
1075 BUG_ON(!addr);
1076 BUG_ON(addr < VMALLOC_START);
1077 BUG_ON(addr > VMALLOC_END);
1078 BUG_ON(addr & (PAGE_SIZE-1));
1079
1080 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -08001081 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -07001082
1083 if (likely(count <= VMAP_MAX_ALLOC))
1084 vb_free(mem, size);
1085 else
1086 free_unmap_vmap_area_addr(addr);
1087}
1088EXPORT_SYMBOL(vm_unmap_ram);
1089
1090/**
1091 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1092 * @pages: an array of pointers to the pages to be mapped
1093 * @count: number of pages
1094 * @node: prefer to allocate data structures on this node
1095 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -07001096 *
1097 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -07001098 */
1099void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1100{
1101 unsigned long size = count << PAGE_SHIFT;
1102 unsigned long addr;
1103 void *mem;
1104
1105 if (likely(count <= VMAP_MAX_ALLOC)) {
1106 mem = vb_alloc(size, GFP_KERNEL);
1107 if (IS_ERR(mem))
1108 return NULL;
1109 addr = (unsigned long)mem;
1110 } else {
1111 struct vmap_area *va;
1112 va = alloc_vmap_area(size, PAGE_SIZE,
1113 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1114 if (IS_ERR(va))
1115 return NULL;
1116
1117 addr = va->va_start;
1118 mem = (void *)addr;
1119 }
1120 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1121 vm_unmap_ram(mem, count);
1122 return NULL;
1123 }
1124 return mem;
1125}
1126EXPORT_SYMBOL(vm_map_ram);
1127
Tejun Heof0aa6612009-02-20 16:29:08 +09001128/**
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001129 * vm_area_add_early - add vmap area early during boot
1130 * @vm: vm_struct to add
1131 *
1132 * This function is used to add fixed kernel vm area to vmlist before
1133 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1134 * should contain proper values and the other fields should be zero.
1135 *
1136 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1137 */
1138void __init vm_area_add_early(struct vm_struct *vm)
1139{
1140 struct vm_struct *tmp, **p;
1141
1142 BUG_ON(vmap_initialized);
1143 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1144 if (tmp->addr >= vm->addr) {
1145 BUG_ON(tmp->addr < vm->addr + vm->size);
1146 break;
1147 } else
1148 BUG_ON(tmp->addr + tmp->size > vm->addr);
1149 }
1150 vm->next = *p;
1151 *p = vm;
1152}
1153
1154/**
Tejun Heof0aa6612009-02-20 16:29:08 +09001155 * vm_area_register_early - register vmap area early during boot
1156 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001157 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001158 *
1159 * This function is used to register kernel vm area before
1160 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1161 * proper values on entry and other fields should be zero. On return,
1162 * vm->addr contains the allocated address.
1163 *
1164 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1165 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001166void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001167{
1168 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001169 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001170
Tejun Heoc0c0a292009-02-24 11:57:21 +09001171 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1172 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1173
1174 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001175
Nicolas Pitrebe9b7332011-08-25 00:24:21 -04001176 vm_area_add_early(vm);
Tejun Heof0aa6612009-02-20 16:29:08 +09001177}
1178
Nick Piggindb64fe02008-10-18 20:27:03 -07001179void __init vmalloc_init(void)
1180{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001181 struct vmap_area *va;
1182 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001183 int i;
1184
1185 for_each_possible_cpu(i) {
1186 struct vmap_block_queue *vbq;
1187
1188 vbq = &per_cpu(vmap_block_queue, i);
1189 spin_lock_init(&vbq->lock);
1190 INIT_LIST_HEAD(&vbq->free);
Nick Piggindb64fe02008-10-18 20:27:03 -07001191 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001192
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001193 /* Import existing vmlist entries. */
1194 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001195 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
KyongHodbda5912012-05-29 15:06:49 -07001196 va->flags = VM_VM_AREA;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001197 va->va_start = (unsigned long)tmp->addr;
1198 va->va_end = va->va_start + tmp->size;
KyongHodbda5912012-05-29 15:06:49 -07001199 va->vm = tmp;
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001200 __insert_vmap_area(va);
1201 }
Tejun Heoca23e402009-08-14 15:00:52 +09001202
1203 vmap_area_pcpu_hole = VMALLOC_END;
1204
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001205 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001206}
1207
Tejun Heo8fc48982009-02-20 16:29:08 +09001208/**
1209 * map_kernel_range_noflush - map kernel VM area with the specified pages
1210 * @addr: start of the VM area to map
1211 * @size: size of the VM area to map
1212 * @prot: page protection flags to use
1213 * @pages: pages to map
1214 *
1215 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1216 * specify should have been allocated using get_vm_area() and its
1217 * friends.
1218 *
1219 * NOTE:
1220 * This function does NOT do any cache flushing. The caller is
1221 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1222 * before calling this function.
1223 *
1224 * RETURNS:
1225 * The number of pages mapped on success, -errno on failure.
1226 */
1227int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1228 pgprot_t prot, struct page **pages)
1229{
1230 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1231}
1232
1233/**
1234 * unmap_kernel_range_noflush - unmap kernel VM area
1235 * @addr: start of the VM area to unmap
1236 * @size: size of the VM area to unmap
1237 *
1238 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1239 * specify should have been allocated using get_vm_area() and its
1240 * friends.
1241 *
1242 * NOTE:
1243 * This function does NOT do any cache flushing. The caller is
1244 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1245 * before calling this function and flush_tlb_kernel_range() after.
1246 */
1247void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1248{
1249 vunmap_page_range(addr, addr + size);
1250}
Huang Ying81e88fd2011-01-12 14:44:55 +08001251EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
Tejun Heo8fc48982009-02-20 16:29:08 +09001252
1253/**
1254 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1255 * @addr: start of the VM area to unmap
1256 * @size: size of the VM area to unmap
1257 *
1258 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1259 * the unmapping and tlb after.
1260 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001261void unmap_kernel_range(unsigned long addr, unsigned long size)
1262{
1263 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001264
1265 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001266 vunmap_page_range(addr, end);
1267 flush_tlb_kernel_range(addr, end);
1268}
1269
1270int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1271{
1272 unsigned long addr = (unsigned long)area->addr;
1273 unsigned long end = addr + area->size - PAGE_SIZE;
1274 int err;
1275
1276 err = vmap_page_range(addr, end, prot, *pages);
1277 if (err > 0) {
1278 *pages += err;
1279 err = 0;
1280 }
1281
1282 return err;
1283}
1284EXPORT_SYMBOL_GPL(map_vm_area);
1285
1286/*** Old vmalloc interfaces ***/
1287DEFINE_RWLOCK(vmlist_lock);
1288struct vm_struct *vmlist;
1289
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001290static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001291 unsigned long flags, const void *caller)
Tejun Heocf88c792009-08-14 15:00:52 +09001292{
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001293 spin_lock(&vmap_area_lock);
Tejun Heocf88c792009-08-14 15:00:52 +09001294 vm->flags = flags;
1295 vm->addr = (void *)va->va_start;
1296 vm->size = va->va_end - va->va_start;
1297 vm->caller = caller;
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001298 va->vm = vm;
Tejun Heocf88c792009-08-14 15:00:52 +09001299 va->flags |= VM_VM_AREA;
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001300 spin_unlock(&vmap_area_lock);
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001301}
Tejun Heocf88c792009-08-14 15:00:52 +09001302
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001303static void insert_vmalloc_vmlist(struct vm_struct *vm)
1304{
1305 struct vm_struct *tmp, **p;
1306
1307 vm->flags &= ~VM_UNLIST;
Tejun Heocf88c792009-08-14 15:00:52 +09001308 write_lock(&vmlist_lock);
1309 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1310 if (tmp->addr >= vm->addr)
1311 break;
1312 }
1313 vm->next = *p;
1314 *p = vm;
1315 write_unlock(&vmlist_lock);
1316}
1317
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001318static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001319 unsigned long flags, const void *caller)
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001320{
1321 setup_vmalloc_vm(vm, va, flags, caller);
1322 insert_vmalloc_vmlist(vm);
1323}
1324
Nick Piggindb64fe02008-10-18 20:27:03 -07001325static struct vm_struct *__get_vm_area_node(unsigned long size,
David Miller2dca6992009-09-21 12:22:34 -07001326 unsigned long align, unsigned long flags, unsigned long start,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001327 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
Nick Piggindb64fe02008-10-18 20:27:03 -07001328{
Kautuk Consul00065262011-12-19 17:12:04 -08001329 struct vmap_area *va;
Nick Piggindb64fe02008-10-18 20:27:03 -07001330 struct vm_struct *area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001332 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 if (flags & VM_IOREMAP) {
1334 int bit = fls(size);
1335
1336 if (bit > IOREMAP_MAX_ORDER)
1337 bit = IOREMAP_MAX_ORDER;
1338 else if (bit < PAGE_SHIFT)
1339 bit = PAGE_SHIFT;
1340
1341 align = 1ul << bit;
1342 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001345 if (unlikely(!size))
1346 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Tejun Heocf88c792009-08-14 15:00:52 +09001348 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 if (unlikely(!area))
1350 return NULL;
1351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 /*
1353 * We always allocate a guard page.
1354 */
1355 size += PAGE_SIZE;
1356
Nick Piggindb64fe02008-10-18 20:27:03 -07001357 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1358 if (IS_ERR(va)) {
1359 kfree(area);
1360 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001363 /*
1364 * When this function is called from __vmalloc_node_range,
1365 * we do not add vm_struct to vmlist here to avoid
1366 * accessing uninitialized members of vm_struct such as
1367 * pages and nr_pages fields. They will be set later.
1368 * To distinguish it from others, we use a VM_UNLIST flag.
1369 */
1370 if (flags & VM_UNLIST)
1371 setup_vmalloc_vm(area, va, flags, caller);
1372 else
1373 insert_vmalloc_vm(area, va, flags, caller);
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
Christoph Lameter930fc452005-10-29 18:15:41 -07001378struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1379 unsigned long start, unsigned long end)
1380{
David Rientjes00ef2d22013-02-22 16:35:36 -08001381 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1382 GFP_KERNEL, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001383}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001384EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001385
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001386struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1387 unsigned long start, unsigned long end,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001388 const void *caller)
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001389{
David Rientjes00ef2d22013-02-22 16:35:36 -08001390 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1391 GFP_KERNEL, caller);
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001392}
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394/**
Simon Arlott183ff222007-10-20 01:27:18 +02001395 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 * @size: size of the area
1397 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1398 *
1399 * Search an area of @size in the kernel virtual mapping area,
1400 * and reserved it for out purposes. Returns the area descriptor
1401 * on success or %NULL on failure.
1402 */
1403struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1404{
David Miller2dca6992009-09-21 12:22:34 -07001405 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001406 NUMA_NO_NODE, GFP_KERNEL,
1407 __builtin_return_address(0));
Christoph Lameter23016962008-04-28 02:12:42 -07001408}
1409
1410struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001411 const void *caller)
Christoph Lameter23016962008-04-28 02:12:42 -07001412{
David Miller2dca6992009-09-21 12:22:34 -07001413 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
David Rientjes00ef2d22013-02-22 16:35:36 -08001414 NUMA_NO_NODE, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
1416
Marek Szyprowskie9da6e92012-07-30 09:11:33 +02001417/**
1418 * find_vm_area - find a continuous kernel virtual area
1419 * @addr: base address
1420 *
1421 * Search for the kernel VM area starting at @addr, and return it.
1422 * It is up to the caller to do all required locking to keep the returned
1423 * pointer valid.
1424 */
1425struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001426{
Nick Piggindb64fe02008-10-18 20:27:03 -07001427 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001428
Nick Piggindb64fe02008-10-18 20:27:03 -07001429 va = find_vmap_area((unsigned long)addr);
1430 if (va && va->flags & VM_VM_AREA)
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001431 return va->vm;
Nick Piggin83342312006-06-23 02:03:20 -07001432
Andi Kleen7856dfe2005-05-20 14:27:57 -07001433 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001434}
1435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436/**
Simon Arlott183ff222007-10-20 01:27:18 +02001437 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 * @addr: base address
1439 *
1440 * Search for the kernel VM area starting at @addr, and remove it.
1441 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001442 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001444struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
Nick Piggindb64fe02008-10-18 20:27:03 -07001446 struct vmap_area *va;
1447
1448 va = find_vmap_area((unsigned long)addr);
1449 if (va && va->flags & VM_VM_AREA) {
Minchan Kimdb1aeca2012-01-10 15:08:39 -08001450 struct vm_struct *vm = va->vm;
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001451
Joonsoo Kimc69480a2013-04-29 15:07:30 -07001452 spin_lock(&vmap_area_lock);
1453 va->vm = NULL;
1454 va->flags &= ~VM_VM_AREA;
1455 spin_unlock(&vmap_area_lock);
1456
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001457 if (!(vm->flags & VM_UNLIST)) {
1458 struct vm_struct *tmp, **p;
1459 /*
1460 * remove from list and disallow access to
1461 * this vm_struct before unmap. (address range
1462 * confliction is maintained by vmap.)
1463 */
1464 write_lock(&vmlist_lock);
1465 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1466 ;
1467 *p = tmp->next;
1468 write_unlock(&vmlist_lock);
1469 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001470
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001471 vmap_debug_free_range(va->va_start, va->va_end);
1472 free_unmap_vmap_area(va);
1473 vm->size -= PAGE_SIZE;
1474
Nick Piggindb64fe02008-10-18 20:27:03 -07001475 return vm;
1476 }
1477 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001480static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481{
1482 struct vm_struct *area;
1483
1484 if (!addr)
1485 return;
1486
1487 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001488 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return;
1490 }
1491
1492 area = remove_vm_area(addr);
1493 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001494 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 return;
1497 }
1498
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001499 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001500 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if (deallocate_pages) {
1503 int i;
1504
1505 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001506 struct page *page = area->pages[i];
1507
1508 BUG_ON(!page);
1509 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
1511
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001512 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 vfree(area->pages);
1514 else
1515 kfree(area->pages);
1516 }
1517
1518 kfree(area);
1519 return;
1520}
1521
1522/**
1523 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 * @addr: memory base address
1525 *
Simon Arlott183ff222007-10-20 01:27:18 +02001526 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001527 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1528 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001530 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001532void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
1534 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001535
1536 kmemleak_free(addr);
1537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 __vunmap(addr, 1);
1539}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540EXPORT_SYMBOL(vfree);
1541
1542/**
1543 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 * @addr: memory base address
1545 *
1546 * Free the virtually contiguous memory area starting at @addr,
1547 * which was created from the page array passed to vmap().
1548 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001549 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001551void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552{
1553 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001554 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 __vunmap(addr, 0);
1556}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557EXPORT_SYMBOL(vunmap);
1558
1559/**
1560 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 * @pages: array of page pointers
1562 * @count: number of pages to map
1563 * @flags: vm_area->flags
1564 * @prot: page protection for the mapping
1565 *
1566 * Maps @count pages from @pages into contiguous kernel virtual
1567 * space.
1568 */
1569void *vmap(struct page **pages, unsigned int count,
1570 unsigned long flags, pgprot_t prot)
1571{
1572 struct vm_struct *area;
1573
Peter Zijlstra34754b62009-02-25 16:04:03 +01001574 might_sleep();
1575
Jan Beulich44813742009-09-21 17:03:05 -07001576 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return NULL;
1578
Christoph Lameter23016962008-04-28 02:12:42 -07001579 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1580 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 if (!area)
1582 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 if (map_vm_area(area, prot, &pages)) {
1585 vunmap(area->addr);
1586 return NULL;
1587 }
1588
1589 return area->addr;
1590}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591EXPORT_SYMBOL(vmap);
1592
David Miller2dca6992009-09-21 12:22:34 -07001593static void *__vmalloc_node(unsigned long size, unsigned long align,
1594 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001595 int node, const void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001596static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001597 pgprot_t prot, int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598{
Dave Hansen22943ab2011-05-24 17:12:18 -07001599 const int order = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 struct page **pages;
1601 unsigned int nr_pages, array_size, i;
Jan Beulich976d6df2009-12-14 17:58:39 -08001602 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
1604 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1605 array_size = (nr_pages * sizeof(struct page *));
1606
1607 area->nr_pages = nr_pages;
1608 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001609 if (array_size > PAGE_SIZE) {
Jan Beulich976d6df2009-12-14 17:58:39 -08001610 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
Christoph Lameter23016962008-04-28 02:12:42 -07001611 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001612 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001613 } else {
Jan Beulich976d6df2009-12-14 17:58:39 -08001614 pages = kmalloc_node(array_size, nested_gfp, node);
Andrew Morton286e1ea2006-10-17 00:09:57 -07001615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001617 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 if (!area->pages) {
1619 remove_vm_area(area->addr);
1620 kfree(area);
1621 return NULL;
1622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001625 struct page *page;
Dave Hansen22943ab2011-05-24 17:12:18 -07001626 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001627
Christoph Lameter930fc452005-10-29 18:15:41 -07001628 if (node < 0)
Dave Hansen22943ab2011-05-24 17:12:18 -07001629 page = alloc_page(tmp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001630 else
Dave Hansen22943ab2011-05-24 17:12:18 -07001631 page = alloc_pages_node(node, tmp_mask, order);
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001632
1633 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 /* Successfully allocated i pages, free them in __vunmap() */
1635 area->nr_pages = i;
1636 goto fail;
1637 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001638 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 }
1640
1641 if (map_vm_area(area, prot, &pages))
1642 goto fail;
1643 return area->addr;
1644
1645fail:
Joe Perches3ee9a4f2011-10-31 17:08:35 -07001646 warn_alloc_failed(gfp_mask, order,
1647 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
Dave Hansen22943ab2011-05-24 17:12:18 -07001648 (area->nr_pages*PAGE_SIZE), area->size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 vfree(area->addr);
1650 return NULL;
1651}
1652
David Rientjesd0a21262011-01-13 15:46:02 -08001653/**
1654 * __vmalloc_node_range - allocate virtually contiguous memory
1655 * @size: allocation size
1656 * @align: desired alignment
1657 * @start: vm area range start
1658 * @end: vm area range end
1659 * @gfp_mask: flags for the page level allocator
1660 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001661 * @node: node to use for allocation or NUMA_NO_NODE
David Rientjesd0a21262011-01-13 15:46:02 -08001662 * @caller: caller's return address
1663 *
1664 * Allocate enough pages to cover @size from the page level
1665 * allocator with @gfp_mask flags. Map them into contiguous
1666 * kernel virtual space, using a pagetable protection of @prot.
1667 */
1668void *__vmalloc_node_range(unsigned long size, unsigned long align,
1669 unsigned long start, unsigned long end, gfp_t gfp_mask,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001670 pgprot_t prot, int node, const void *caller)
Christoph Lameter930fc452005-10-29 18:15:41 -07001671{
David Rientjesd0a21262011-01-13 15:46:02 -08001672 struct vm_struct *area;
1673 void *addr;
1674 unsigned long real_size = size;
1675
1676 size = PAGE_ALIGN(size);
1677 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001678 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001679
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001680 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1681 start, end, node, gfp_mask, caller);
David Rientjesd0a21262011-01-13 15:46:02 -08001682 if (!area)
Joe Perchesde7d2b52011-10-31 17:08:48 -07001683 goto fail;
David Rientjesd0a21262011-01-13 15:46:02 -08001684
1685 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
Mel Gorman1368edf2011-12-08 14:34:30 -08001686 if (!addr)
1687 return NULL;
Catalin Marinas89219d32009-06-11 13:23:19 +01001688
1689 /*
Mitsuo Hayasakaf5252e02011-10-31 17:08:13 -07001690 * In this function, newly allocated vm_struct is not added
1691 * to vmlist at __get_vm_area_node(). so, it is added here.
1692 */
1693 insert_vmalloc_vmlist(area);
1694
1695 /*
Catalin Marinas89219d32009-06-11 13:23:19 +01001696 * A ref_count = 3 is needed because the vm_struct and vmap_area
1697 * structures allocated in the __get_vm_area_node() function contain
1698 * references to the virtual address of the vmalloc'ed block.
1699 */
David Rientjesd0a21262011-01-13 15:46:02 -08001700 kmemleak_alloc(addr, real_size, 3, gfp_mask);
Catalin Marinas89219d32009-06-11 13:23:19 +01001701
1702 return addr;
Joe Perchesde7d2b52011-10-31 17:08:48 -07001703
1704fail:
1705 warn_alloc_failed(gfp_mask, 0,
1706 "vmalloc: allocation failure: %lu bytes\n",
1707 real_size);
1708 return NULL;
Christoph Lameter930fc452005-10-29 18:15:41 -07001709}
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001712 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 * @size: allocation size
David Miller2dca6992009-09-21 12:22:34 -07001714 * @align: desired alignment
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 * @gfp_mask: flags for the page level allocator
1716 * @prot: protection mask for the allocated pages
David Rientjes00ef2d22013-02-22 16:35:36 -08001717 * @node: node to use for allocation or NUMA_NO_NODE
Randy Dunlapc85d1942008-05-01 04:34:48 -07001718 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 *
1720 * Allocate enough pages to cover @size from the page level
1721 * allocator with @gfp_mask flags. Map them into contiguous
1722 * kernel virtual space, using a pagetable protection of @prot.
1723 */
David Miller2dca6992009-09-21 12:22:34 -07001724static void *__vmalloc_node(unsigned long size, unsigned long align,
1725 gfp_t gfp_mask, pgprot_t prot,
Marek Szyprowski5e6cafc2012-04-13 12:32:09 +02001726 int node, const void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727{
David Rientjesd0a21262011-01-13 15:46:02 -08001728 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1729 gfp_mask, prot, node, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730}
1731
Christoph Lameter930fc452005-10-29 18:15:41 -07001732void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1733{
David Rientjes00ef2d22013-02-22 16:35:36 -08001734 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
Christoph Lameter23016962008-04-28 02:12:42 -07001735 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001736}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737EXPORT_SYMBOL(__vmalloc);
1738
Dave Younge1ca7782010-10-26 14:22:06 -07001739static inline void *__vmalloc_node_flags(unsigned long size,
1740 int node, gfp_t flags)
1741{
1742 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1743 node, __builtin_return_address(0));
1744}
1745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746/**
1747 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 * Allocate enough pages to cover @size from the page level
1750 * allocator and map them into contiguous kernel virtual space.
1751 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001752 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 * use __vmalloc() instead.
1754 */
1755void *vmalloc(unsigned long size)
1756{
David Rientjes00ef2d22013-02-22 16:35:36 -08001757 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1758 GFP_KERNEL | __GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760EXPORT_SYMBOL(vmalloc);
1761
Christoph Lameter930fc452005-10-29 18:15:41 -07001762/**
Dave Younge1ca7782010-10-26 14:22:06 -07001763 * vzalloc - allocate virtually contiguous memory with zero fill
1764 * @size: allocation size
1765 * Allocate enough pages to cover @size from the page level
1766 * allocator and map them into contiguous kernel virtual space.
1767 * The memory allocated is set to zero.
1768 *
1769 * For tight control over page level allocator and protection flags
1770 * use __vmalloc() instead.
1771 */
1772void *vzalloc(unsigned long size)
1773{
David Rientjes00ef2d22013-02-22 16:35:36 -08001774 return __vmalloc_node_flags(size, NUMA_NO_NODE,
Dave Younge1ca7782010-10-26 14:22:06 -07001775 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1776}
1777EXPORT_SYMBOL(vzalloc);
1778
1779/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001780 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1781 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001782 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001783 * The resulting memory area is zeroed so it can be mapped to userspace
1784 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001785 */
1786void *vmalloc_user(unsigned long size)
1787{
1788 struct vm_struct *area;
1789 void *ret;
1790
David Miller2dca6992009-09-21 12:22:34 -07001791 ret = __vmalloc_node(size, SHMLBA,
1792 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
David Rientjes00ef2d22013-02-22 16:35:36 -08001793 PAGE_KERNEL, NUMA_NO_NODE,
1794 __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001795 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001796 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001797 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001798 }
Nick Piggin83342312006-06-23 02:03:20 -07001799 return ret;
1800}
1801EXPORT_SYMBOL(vmalloc_user);
1802
1803/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001804 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001805 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001806 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001807 *
1808 * Allocate enough pages to cover @size from the page level
1809 * allocator and map them into contiguous kernel virtual space.
1810 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001811 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001812 * use __vmalloc() instead.
1813 */
1814void *vmalloc_node(unsigned long size, int node)
1815{
David Miller2dca6992009-09-21 12:22:34 -07001816 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
Christoph Lameter23016962008-04-28 02:12:42 -07001817 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001818}
1819EXPORT_SYMBOL(vmalloc_node);
1820
Dave Younge1ca7782010-10-26 14:22:06 -07001821/**
1822 * vzalloc_node - allocate memory on a specific node with zero fill
1823 * @size: allocation size
1824 * @node: numa node
1825 *
1826 * Allocate enough pages to cover @size from the page level
1827 * allocator and map them into contiguous kernel virtual space.
1828 * The memory allocated is set to zero.
1829 *
1830 * For tight control over page level allocator and protection flags
1831 * use __vmalloc_node() instead.
1832 */
1833void *vzalloc_node(unsigned long size, int node)
1834{
1835 return __vmalloc_node_flags(size, node,
1836 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1837}
1838EXPORT_SYMBOL(vzalloc_node);
1839
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001840#ifndef PAGE_KERNEL_EXEC
1841# define PAGE_KERNEL_EXEC PAGE_KERNEL
1842#endif
1843
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844/**
1845 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 * @size: allocation size
1847 *
1848 * Kernel-internal function to allocate enough pages to cover @size
1849 * the page level allocator and map them into contiguous and
1850 * executable kernel virtual space.
1851 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001852 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 * use __vmalloc() instead.
1854 */
1855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856void *vmalloc_exec(unsigned long size)
1857{
David Miller2dca6992009-09-21 12:22:34 -07001858 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
David Rientjes00ef2d22013-02-22 16:35:36 -08001859 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860}
1861
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001862#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001863#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001864#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001865#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001866#else
1867#define GFP_VMALLOC32 GFP_KERNEL
1868#endif
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870/**
1871 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 * @size: allocation size
1873 *
1874 * Allocate enough 32bit PA addressable pages to cover @size from the
1875 * page level allocator and map them into contiguous kernel virtual space.
1876 */
1877void *vmalloc_32(unsigned long size)
1878{
David Miller2dca6992009-09-21 12:22:34 -07001879 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001880 NUMA_NO_NODE, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882EXPORT_SYMBOL(vmalloc_32);
1883
Nick Piggin83342312006-06-23 02:03:20 -07001884/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001885 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001886 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001887 *
1888 * The resulting memory area is 32bit addressable and zeroed so it can be
1889 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001890 */
1891void *vmalloc_32_user(unsigned long size)
1892{
1893 struct vm_struct *area;
1894 void *ret;
1895
David Miller2dca6992009-09-21 12:22:34 -07001896 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
David Rientjes00ef2d22013-02-22 16:35:36 -08001897 NUMA_NO_NODE, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001898 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001899 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001900 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001901 }
Nick Piggin83342312006-06-23 02:03:20 -07001902 return ret;
1903}
1904EXPORT_SYMBOL(vmalloc_32_user);
1905
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001906/*
1907 * small helper routine , copy contents to buf from addr.
1908 * If the page is not present, fill zero.
1909 */
1910
1911static int aligned_vread(char *buf, char *addr, unsigned long count)
1912{
1913 struct page *p;
1914 int copied = 0;
1915
1916 while (count) {
1917 unsigned long offset, length;
1918
1919 offset = (unsigned long)addr & ~PAGE_MASK;
1920 length = PAGE_SIZE - offset;
1921 if (length > count)
1922 length = count;
1923 p = vmalloc_to_page(addr);
1924 /*
1925 * To do safe access to this _mapped_ area, we need
1926 * lock. But adding lock here means that we need to add
1927 * overhead of vmalloc()/vfree() calles for this _debug_
1928 * interface, rarely used. Instead of that, we'll use
1929 * kmap() and get small overhead in this access function.
1930 */
1931 if (p) {
1932 /*
1933 * we can expect USER0 is not used (see vread/vwrite's
1934 * function description)
1935 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001936 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001937 memcpy(buf, map + offset, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001938 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001939 } else
1940 memset(buf, 0, length);
1941
1942 addr += length;
1943 buf += length;
1944 copied += length;
1945 count -= length;
1946 }
1947 return copied;
1948}
1949
1950static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1951{
1952 struct page *p;
1953 int copied = 0;
1954
1955 while (count) {
1956 unsigned long offset, length;
1957
1958 offset = (unsigned long)addr & ~PAGE_MASK;
1959 length = PAGE_SIZE - offset;
1960 if (length > count)
1961 length = count;
1962 p = vmalloc_to_page(addr);
1963 /*
1964 * To do safe access to this _mapped_ area, we need
1965 * lock. But adding lock here means that we need to add
1966 * overhead of vmalloc()/vfree() calles for this _debug_
1967 * interface, rarely used. Instead of that, we'll use
1968 * kmap() and get small overhead in this access function.
1969 */
1970 if (p) {
1971 /*
1972 * we can expect USER0 is not used (see vread/vwrite's
1973 * function description)
1974 */
Cong Wang9b04c5f2011-11-25 23:14:39 +08001975 void *map = kmap_atomic(p);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001976 memcpy(map + offset, buf, length);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001977 kunmap_atomic(map);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001978 }
1979 addr += length;
1980 buf += length;
1981 copied += length;
1982 count -= length;
1983 }
1984 return copied;
1985}
1986
1987/**
1988 * vread() - read vmalloc area in a safe way.
1989 * @buf: buffer for reading data
1990 * @addr: vm address.
1991 * @count: number of bytes to be read.
1992 *
1993 * Returns # of bytes which addr and buf should be increased.
1994 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1995 * includes any intersect with alive vmalloc area.
1996 *
1997 * This function checks that addr is a valid vmalloc'ed area, and
1998 * copy data from that area to a given buffer. If the given memory range
1999 * of [addr...addr+count) includes some valid address, data is copied to
2000 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2001 * IOREMAP area is treated as memory hole and no copy is done.
2002 *
2003 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002004 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002005 *
2006 * Note: In usual ops, vread() is never necessary because the caller
2007 * should know vmalloc() area is valid and can use memcpy().
2008 * This is for routines which have to access vmalloc area without
2009 * any informaion, as /dev/kmem.
2010 *
2011 */
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013long vread(char *buf, char *addr, unsigned long count)
2014{
2015 struct vm_struct *tmp;
2016 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002017 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 unsigned long n;
2019
2020 /* Don't allow overflow */
2021 if ((unsigned long) addr + count < count)
2022 count = -(unsigned long) addr;
2023
2024 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002025 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 vaddr = (char *) tmp->addr;
2027 if (addr >= vaddr + tmp->size - PAGE_SIZE)
2028 continue;
2029 while (addr < vaddr) {
2030 if (count == 0)
2031 goto finished;
2032 *buf = '\0';
2033 buf++;
2034 addr++;
2035 count--;
2036 }
2037 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002038 if (n > count)
2039 n = count;
2040 if (!(tmp->flags & VM_IOREMAP))
2041 aligned_vread(buf, addr, n);
2042 else /* IOREMAP area is treated as memory hole */
2043 memset(buf, 0, n);
2044 buf += n;
2045 addr += n;
2046 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 }
2048finished:
2049 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002050
2051 if (buf == buf_start)
2052 return 0;
2053 /* zero-fill memory holes */
2054 if (buf != buf_start + buflen)
2055 memset(buf, 0, buflen - (buf - buf_start));
2056
2057 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058}
2059
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002060/**
2061 * vwrite() - write vmalloc area in a safe way.
2062 * @buf: buffer for source data
2063 * @addr: vm address.
2064 * @count: number of bytes to be read.
2065 *
2066 * Returns # of bytes which addr and buf should be incresed.
2067 * (same number to @count).
2068 * If [addr...addr+count) doesn't includes any intersect with valid
2069 * vmalloc area, returns 0.
2070 *
2071 * This function checks that addr is a valid vmalloc'ed area, and
2072 * copy data from a buffer to the given addr. If specified range of
2073 * [addr...addr+count) includes some valid address, data is copied from
2074 * proper area of @buf. If there are memory holes, no copy to hole.
2075 * IOREMAP area is treated as memory hole and no copy is done.
2076 *
2077 * If [addr...addr+count) doesn't includes any intersects with alive
Cong Wanga8e52022012-06-23 11:30:16 +08002078 * vm_struct area, returns 0. @buf should be kernel's buffer.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002079 *
2080 * Note: In usual ops, vwrite() is never necessary because the caller
2081 * should know vmalloc() area is valid and can use memcpy().
2082 * This is for routines which have to access vmalloc area without
2083 * any informaion, as /dev/kmem.
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002084 */
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086long vwrite(char *buf, char *addr, unsigned long count)
2087{
2088 struct vm_struct *tmp;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002089 char *vaddr;
2090 unsigned long n, buflen;
2091 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
2093 /* Don't allow overflow */
2094 if ((unsigned long) addr + count < count)
2095 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002096 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
2098 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002099 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 vaddr = (char *) tmp->addr;
2101 if (addr >= vaddr + tmp->size - PAGE_SIZE)
2102 continue;
2103 while (addr < vaddr) {
2104 if (count == 0)
2105 goto finished;
2106 buf++;
2107 addr++;
2108 count--;
2109 }
2110 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002111 if (n > count)
2112 n = count;
2113 if (!(tmp->flags & VM_IOREMAP)) {
2114 aligned_vwrite(buf, addr, n);
2115 copied++;
2116 }
2117 buf += n;
2118 addr += n;
2119 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 }
2121finished:
2122 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07002123 if (!copied)
2124 return 0;
2125 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
Nick Piggin83342312006-06-23 02:03:20 -07002127
2128/**
2129 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07002130 * @vma: vma to cover (map full range of vma)
2131 * @addr: vmalloc memory
2132 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07002133 *
2134 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07002135 *
2136 * This function checks that addr is a valid vmalloc'ed area, and
2137 * that it is big enough to cover the vma. Will return failure if
2138 * that criteria isn't met.
2139 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002140 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07002141 */
2142int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2143 unsigned long pgoff)
2144{
2145 struct vm_struct *area;
2146 unsigned long uaddr = vma->vm_start;
2147 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07002148
2149 if ((PAGE_SIZE-1) & (unsigned long)addr)
2150 return -EINVAL;
2151
Nick Piggindb64fe02008-10-18 20:27:03 -07002152 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07002153 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07002154 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002155
2156 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07002157 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002158
2159 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07002160 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07002161
2162 addr += pgoff << PAGE_SHIFT;
2163 do {
2164 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07002165 int ret;
2166
Nick Piggin83342312006-06-23 02:03:20 -07002167 ret = vm_insert_page(vma, uaddr, page);
2168 if (ret)
2169 return ret;
2170
2171 uaddr += PAGE_SIZE;
2172 addr += PAGE_SIZE;
2173 usize -= PAGE_SIZE;
2174 } while (usize > 0);
2175
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07002176 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
Nick Piggin83342312006-06-23 02:03:20 -07002177
Nick Piggindb64fe02008-10-18 20:27:03 -07002178 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07002179}
2180EXPORT_SYMBOL(remap_vmalloc_range);
2181
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07002182/*
2183 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2184 * have one.
2185 */
2186void __attribute__((weak)) vmalloc_sync_all(void)
2187{
2188}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002189
2190
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002191static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002192{
David Vrabelcd129092011-09-29 16:53:32 +01002193 pte_t ***p = data;
2194
2195 if (p) {
2196 *(*p) = pte;
2197 (*p)++;
2198 }
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002199 return 0;
2200}
2201
2202/**
2203 * alloc_vm_area - allocate a range of kernel address space
2204 * @size: size of the area
David Vrabelcd129092011-09-29 16:53:32 +01002205 * @ptes: returns the PTEs for the address space
Randy Dunlap76824862008-03-19 17:00:40 -07002206 *
2207 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002208 *
2209 * This function reserves a range of kernel address space, and
2210 * allocates pagetables to map that range. No actual mappings
David Vrabelcd129092011-09-29 16:53:32 +01002211 * are created.
2212 *
2213 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2214 * allocated for the VM area are returned.
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002215 */
David Vrabelcd129092011-09-29 16:53:32 +01002216struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002217{
2218 struct vm_struct *area;
2219
Christoph Lameter23016962008-04-28 02:12:42 -07002220 area = get_vm_area_caller(size, VM_IOREMAP,
2221 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002222 if (area == NULL)
2223 return NULL;
2224
2225 /*
2226 * This ensures that page tables are constructed for this region
2227 * of kernel virtual address space and mapped into init_mm.
2228 */
2229 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
David Vrabelcd129092011-09-29 16:53:32 +01002230 size, f, ptes ? &ptes : NULL)) {
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002231 free_vm_area(area);
2232 return NULL;
2233 }
2234
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07002235 return area;
2236}
2237EXPORT_SYMBOL_GPL(alloc_vm_area);
2238
2239void free_vm_area(struct vm_struct *area)
2240{
2241 struct vm_struct *ret;
2242 ret = remove_vm_area(area->addr);
2243 BUG_ON(ret != area);
2244 kfree(area);
2245}
2246EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07002247
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002248#ifdef CONFIG_SMP
Tejun Heoca23e402009-08-14 15:00:52 +09002249static struct vmap_area *node_to_va(struct rb_node *n)
2250{
2251 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2252}
2253
2254/**
2255 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2256 * @end: target address
2257 * @pnext: out arg for the next vmap_area
2258 * @pprev: out arg for the previous vmap_area
2259 *
2260 * Returns: %true if either or both of next and prev are found,
2261 * %false if no vmap_area exists
2262 *
2263 * Find vmap_areas end addresses of which enclose @end. ie. if not
2264 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2265 */
2266static bool pvm_find_next_prev(unsigned long end,
2267 struct vmap_area **pnext,
2268 struct vmap_area **pprev)
2269{
2270 struct rb_node *n = vmap_area_root.rb_node;
2271 struct vmap_area *va = NULL;
2272
2273 while (n) {
2274 va = rb_entry(n, struct vmap_area, rb_node);
2275 if (end < va->va_end)
2276 n = n->rb_left;
2277 else if (end > va->va_end)
2278 n = n->rb_right;
2279 else
2280 break;
2281 }
2282
2283 if (!va)
2284 return false;
2285
2286 if (va->va_end > end) {
2287 *pnext = va;
2288 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2289 } else {
2290 *pprev = va;
2291 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2292 }
2293 return true;
2294}
2295
2296/**
2297 * pvm_determine_end - find the highest aligned address between two vmap_areas
2298 * @pnext: in/out arg for the next vmap_area
2299 * @pprev: in/out arg for the previous vmap_area
2300 * @align: alignment
2301 *
2302 * Returns: determined end address
2303 *
2304 * Find the highest aligned address between *@pnext and *@pprev below
2305 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2306 * down address is between the end addresses of the two vmap_areas.
2307 *
2308 * Please note that the address returned by this function may fall
2309 * inside *@pnext vmap_area. The caller is responsible for checking
2310 * that.
2311 */
2312static unsigned long pvm_determine_end(struct vmap_area **pnext,
2313 struct vmap_area **pprev,
2314 unsigned long align)
2315{
2316 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2317 unsigned long addr;
2318
2319 if (*pnext)
2320 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2321 else
2322 addr = vmalloc_end;
2323
2324 while (*pprev && (*pprev)->va_end > addr) {
2325 *pnext = *pprev;
2326 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2327 }
2328
2329 return addr;
2330}
2331
2332/**
2333 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2334 * @offsets: array containing offset of each area
2335 * @sizes: array containing size of each area
2336 * @nr_vms: the number of areas to allocate
2337 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
Tejun Heoca23e402009-08-14 15:00:52 +09002338 *
2339 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2340 * vm_structs on success, %NULL on failure
2341 *
2342 * Percpu allocator wants to use congruent vm areas so that it can
2343 * maintain the offsets among percpu areas. This function allocates
David Rientjesec3f64f2011-01-13 15:46:01 -08002344 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2345 * be scattered pretty far, distance between two areas easily going up
2346 * to gigabytes. To avoid interacting with regular vmallocs, these
2347 * areas are allocated from top.
Tejun Heoca23e402009-08-14 15:00:52 +09002348 *
2349 * Despite its complicated look, this allocator is rather simple. It
2350 * does everything top-down and scans areas from the end looking for
2351 * matching slot. While scanning, if any of the areas overlaps with
2352 * existing vmap_area, the base address is pulled down to fit the
2353 * area. Scanning is repeated till all the areas fit and then all
2354 * necessary data structres are inserted and the result is returned.
2355 */
2356struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2357 const size_t *sizes, int nr_vms,
David Rientjesec3f64f2011-01-13 15:46:01 -08002358 size_t align)
Tejun Heoca23e402009-08-14 15:00:52 +09002359{
2360 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2361 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2362 struct vmap_area **vas, *prev, *next;
2363 struct vm_struct **vms;
2364 int area, area2, last_area, term_area;
2365 unsigned long base, start, end, last_end;
2366 bool purged = false;
2367
Tejun Heoca23e402009-08-14 15:00:52 +09002368 /* verify parameters and allocate data structures */
2369 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2370 for (last_area = 0, area = 0; area < nr_vms; area++) {
2371 start = offsets[area];
2372 end = start + sizes[area];
2373
2374 /* is everything aligned properly? */
2375 BUG_ON(!IS_ALIGNED(offsets[area], align));
2376 BUG_ON(!IS_ALIGNED(sizes[area], align));
2377
2378 /* detect the area with the highest address */
2379 if (start > offsets[last_area])
2380 last_area = area;
2381
2382 for (area2 = 0; area2 < nr_vms; area2++) {
2383 unsigned long start2 = offsets[area2];
2384 unsigned long end2 = start2 + sizes[area2];
2385
2386 if (area2 == area)
2387 continue;
2388
2389 BUG_ON(start2 >= start && start2 < end);
2390 BUG_ON(end2 <= end && end2 > start);
2391 }
2392 }
2393 last_end = offsets[last_area] + sizes[last_area];
2394
2395 if (vmalloc_end - vmalloc_start < last_end) {
2396 WARN_ON(true);
2397 return NULL;
2398 }
2399
Thomas Meyer4d67d862012-05-29 15:06:21 -07002400 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2401 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002402 if (!vas || !vms)
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002403 goto err_free2;
Tejun Heoca23e402009-08-14 15:00:52 +09002404
2405 for (area = 0; area < nr_vms; area++) {
David Rientjesec3f64f2011-01-13 15:46:01 -08002406 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2407 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
Tejun Heoca23e402009-08-14 15:00:52 +09002408 if (!vas[area] || !vms[area])
2409 goto err_free;
2410 }
2411retry:
2412 spin_lock(&vmap_area_lock);
2413
2414 /* start scanning - we scan from the top, begin with the last area */
2415 area = term_area = last_area;
2416 start = offsets[area];
2417 end = start + sizes[area];
2418
2419 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2420 base = vmalloc_end - last_end;
2421 goto found;
2422 }
2423 base = pvm_determine_end(&next, &prev, align) - end;
2424
2425 while (true) {
2426 BUG_ON(next && next->va_end <= base + end);
2427 BUG_ON(prev && prev->va_end > base + end);
2428
2429 /*
2430 * base might have underflowed, add last_end before
2431 * comparing.
2432 */
2433 if (base + last_end < vmalloc_start + last_end) {
2434 spin_unlock(&vmap_area_lock);
2435 if (!purged) {
2436 purge_vmap_area_lazy();
2437 purged = true;
2438 goto retry;
2439 }
2440 goto err_free;
2441 }
2442
2443 /*
2444 * If next overlaps, move base downwards so that it's
2445 * right below next and then recheck.
2446 */
2447 if (next && next->va_start < base + end) {
2448 base = pvm_determine_end(&next, &prev, align) - end;
2449 term_area = area;
2450 continue;
2451 }
2452
2453 /*
2454 * If prev overlaps, shift down next and prev and move
2455 * base so that it's right below new next and then
2456 * recheck.
2457 */
2458 if (prev && prev->va_end > base + start) {
2459 next = prev;
2460 prev = node_to_va(rb_prev(&next->rb_node));
2461 base = pvm_determine_end(&next, &prev, align) - end;
2462 term_area = area;
2463 continue;
2464 }
2465
2466 /*
2467 * This area fits, move on to the previous one. If
2468 * the previous one is the terminal one, we're done.
2469 */
2470 area = (area + nr_vms - 1) % nr_vms;
2471 if (area == term_area)
2472 break;
2473 start = offsets[area];
2474 end = start + sizes[area];
2475 pvm_find_next_prev(base + end, &next, &prev);
2476 }
2477found:
2478 /* we've found a fitting base, insert all va's */
2479 for (area = 0; area < nr_vms; area++) {
2480 struct vmap_area *va = vas[area];
2481
2482 va->va_start = base + offsets[area];
2483 va->va_end = va->va_start + sizes[area];
2484 __insert_vmap_area(va);
2485 }
2486
2487 vmap_area_pcpu_hole = base + offsets[last_area];
2488
2489 spin_unlock(&vmap_area_lock);
2490
2491 /* insert all vm's */
2492 for (area = 0; area < nr_vms; area++)
2493 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2494 pcpu_get_vm_areas);
2495
2496 kfree(vas);
2497 return vms;
2498
2499err_free:
2500 for (area = 0; area < nr_vms; area++) {
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002501 kfree(vas[area]);
2502 kfree(vms[area]);
Tejun Heoca23e402009-08-14 15:00:52 +09002503 }
Kautuk Consulf1db7af2012-01-12 17:20:08 -08002504err_free2:
Tejun Heoca23e402009-08-14 15:00:52 +09002505 kfree(vas);
2506 kfree(vms);
2507 return NULL;
2508}
2509
2510/**
2511 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2512 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2513 * @nr_vms: the number of allocated areas
2514 *
2515 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2516 */
2517void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2518{
2519 int i;
2520
2521 for (i = 0; i < nr_vms; i++)
2522 free_vm_area(vms[i]);
2523 kfree(vms);
2524}
Tejun Heo4f8b02b2010-09-03 18:22:47 +02002525#endif /* CONFIG_SMP */
Christoph Lametera10aa572008-04-28 02:12:40 -07002526
2527#ifdef CONFIG_PROC_FS
2528static void *s_start(struct seq_file *m, loff_t *pos)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002529 __acquires(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002530{
2531 loff_t n = *pos;
2532 struct vm_struct *v;
2533
2534 read_lock(&vmlist_lock);
2535 v = vmlist;
2536 while (n > 0 && v) {
2537 n--;
2538 v = v->next;
2539 }
2540 if (!n)
2541 return v;
2542
2543 return NULL;
2544
2545}
2546
2547static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2548{
2549 struct vm_struct *v = p;
2550
2551 ++*pos;
2552 return v->next;
2553}
2554
2555static void s_stop(struct seq_file *m, void *p)
Namhyung Kime199b5d2010-10-26 14:22:03 -07002556 __releases(&vmlist_lock)
Christoph Lametera10aa572008-04-28 02:12:40 -07002557{
2558 read_unlock(&vmlist_lock);
2559}
2560
Eric Dumazeta47a1262008-07-23 21:27:38 -07002561static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2562{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002563 if (IS_ENABLED(CONFIG_NUMA)) {
Eric Dumazeta47a1262008-07-23 21:27:38 -07002564 unsigned int nr, *counters = m->private;
2565
2566 if (!counters)
2567 return;
2568
2569 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2570
2571 for (nr = 0; nr < v->nr_pages; nr++)
2572 counters[page_to_nid(v->pages[nr])]++;
2573
2574 for_each_node_state(nr, N_HIGH_MEMORY)
2575 if (counters[nr])
2576 seq_printf(m, " N%u=%u", nr, counters[nr]);
2577 }
2578}
2579
Christoph Lametera10aa572008-04-28 02:12:40 -07002580static int s_show(struct seq_file *m, void *p)
2581{
2582 struct vm_struct *v = p;
2583
Kees Cook45ec1692012-10-08 16:34:09 -07002584 seq_printf(m, "0x%pK-0x%pK %7ld",
Christoph Lametera10aa572008-04-28 02:12:40 -07002585 v->addr, v->addr + v->size, v->size);
2586
Joe Perches62c70bc2011-01-13 15:45:52 -08002587 if (v->caller)
2588 seq_printf(m, " %pS", v->caller);
Christoph Lameter23016962008-04-28 02:12:42 -07002589
Christoph Lametera10aa572008-04-28 02:12:40 -07002590 if (v->nr_pages)
2591 seq_printf(m, " pages=%d", v->nr_pages);
2592
2593 if (v->phys_addr)
Kenji Kaneshigeffa71f32010-06-18 12:22:40 +09002594 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
Christoph Lametera10aa572008-04-28 02:12:40 -07002595
2596 if (v->flags & VM_IOREMAP)
2597 seq_printf(m, " ioremap");
2598
2599 if (v->flags & VM_ALLOC)
2600 seq_printf(m, " vmalloc");
2601
2602 if (v->flags & VM_MAP)
2603 seq_printf(m, " vmap");
2604
2605 if (v->flags & VM_USERMAP)
2606 seq_printf(m, " user");
2607
2608 if (v->flags & VM_VPAGES)
2609 seq_printf(m, " vpages");
2610
Eric Dumazeta47a1262008-07-23 21:27:38 -07002611 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002612 seq_putc(m, '\n');
2613 return 0;
2614}
2615
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002616static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002617 .start = s_start,
2618 .next = s_next,
2619 .stop = s_stop,
2620 .show = s_show,
2621};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002622
2623static int vmalloc_open(struct inode *inode, struct file *file)
2624{
2625 unsigned int *ptr = NULL;
2626 int ret;
2627
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002628 if (IS_ENABLED(CONFIG_NUMA)) {
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002629 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
Kulikov Vasiliy51980ac2010-08-09 17:19:58 -07002630 if (ptr == NULL)
2631 return -ENOMEM;
2632 }
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002633 ret = seq_open(file, &vmalloc_op);
2634 if (!ret) {
2635 struct seq_file *m = file->private_data;
2636 m->private = ptr;
2637 } else
2638 kfree(ptr);
2639 return ret;
2640}
2641
2642static const struct file_operations proc_vmalloc_operations = {
2643 .open = vmalloc_open,
2644 .read = seq_read,
2645 .llseek = seq_lseek,
2646 .release = seq_release_private,
2647};
2648
2649static int __init proc_vmalloc_init(void)
2650{
2651 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2652 return 0;
2653}
2654module_init(proc_vmalloc_init);
Joonsoo Kimdb3808c2013-04-29 15:07:28 -07002655
2656void get_vmalloc_info(struct vmalloc_info *vmi)
2657{
2658 struct vm_struct *vma;
2659 unsigned long free_area_size;
2660 unsigned long prev_end;
2661
2662 vmi->used = 0;
2663
2664 if (!vmlist) {
2665 vmi->largest_chunk = VMALLOC_TOTAL;
2666 } else {
2667 vmi->largest_chunk = 0;
2668
2669 prev_end = VMALLOC_START;
2670
2671 read_lock(&vmlist_lock);
2672
2673 for (vma = vmlist; vma; vma = vma->next) {
2674 unsigned long addr = (unsigned long) vma->addr;
2675
2676 /*
2677 * Some archs keep another range for modules in vmlist
2678 */
2679 if (addr < VMALLOC_START)
2680 continue;
2681 if (addr >= VMALLOC_END)
2682 break;
2683
2684 vmi->used += vma->size;
2685
2686 free_area_size = addr - prev_end;
2687 if (vmi->largest_chunk < free_area_size)
2688 vmi->largest_chunk = free_area_size;
2689
2690 prev_end = vma->size + addr;
2691 }
2692
2693 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2694 vmi->largest_chunk = VMALLOC_END - prev_end;
2695
2696 read_unlock(&vmlist_lock);
2697 }
2698}
Christoph Lametera10aa572008-04-28 02:12:40 -07002699#endif
2700