blob: 69511e663234482228909f373e247d21082e6ada [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>
15#include <linux/slab.h>
16#include <linux/spinlock.h>
17#include <linux/interrupt.h>
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +040018#include <linux/proc_fs.h>
Christoph Lametera10aa572008-04-28 02:12:40 -070019#include <linux/seq_file.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070020#include <linux/debugobjects.h>
Christoph Lameter23016962008-04-28 02:12:42 -070021#include <linux/kallsyms.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070022#include <linux/list.h>
23#include <linux/rbtree.h>
24#include <linux/radix-tree.h>
25#include <linux/rcupdate.h>
Tejun Heof0aa6612009-02-20 16:29:08 +090026#include <linux/pfn.h>
Catalin Marinas89219d32009-06-11 13:23:19 +010027#include <linux/kmemleak.h>
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -070028#include <linux/highmem.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070029#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/uaccess.h>
31#include <asm/tlbflush.h>
32
33
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 */
259 void *private;
260 struct rcu_head rcu_head;
261};
262
263static DEFINE_SPINLOCK(vmap_area_lock);
264static struct rb_root vmap_area_root = RB_ROOT;
265static LIST_HEAD(vmap_area_list);
Tejun Heoca23e402009-08-14 15:00:52 +0900266static unsigned long vmap_area_pcpu_hole;
Nick Piggindb64fe02008-10-18 20:27:03 -0700267
268static struct vmap_area *__find_vmap_area(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Nick Piggindb64fe02008-10-18 20:27:03 -0700270 struct rb_node *n = vmap_area_root.rb_node;
271
272 while (n) {
273 struct vmap_area *va;
274
275 va = rb_entry(n, struct vmap_area, rb_node);
276 if (addr < va->va_start)
277 n = n->rb_left;
278 else if (addr > va->va_start)
279 n = n->rb_right;
280 else
281 return va;
282 }
283
284 return NULL;
285}
286
287static void __insert_vmap_area(struct vmap_area *va)
288{
289 struct rb_node **p = &vmap_area_root.rb_node;
290 struct rb_node *parent = NULL;
291 struct rb_node *tmp;
292
293 while (*p) {
294 struct vmap_area *tmp;
295
296 parent = *p;
297 tmp = rb_entry(parent, struct vmap_area, rb_node);
298 if (va->va_start < tmp->va_end)
299 p = &(*p)->rb_left;
300 else if (va->va_end > tmp->va_start)
301 p = &(*p)->rb_right;
302 else
303 BUG();
304 }
305
306 rb_link_node(&va->rb_node, parent, p);
307 rb_insert_color(&va->rb_node, &vmap_area_root);
308
309 /* address-sort this list so it is usable like the vmlist */
310 tmp = rb_prev(&va->rb_node);
311 if (tmp) {
312 struct vmap_area *prev;
313 prev = rb_entry(tmp, struct vmap_area, rb_node);
314 list_add_rcu(&va->list, &prev->list);
315 } else
316 list_add_rcu(&va->list, &vmap_area_list);
317}
318
319static void purge_vmap_area_lazy(void);
320
321/*
322 * Allocate a region of KVA of the specified size and alignment, within the
323 * vstart and vend.
324 */
325static struct vmap_area *alloc_vmap_area(unsigned long size,
326 unsigned long align,
327 unsigned long vstart, unsigned long vend,
328 int node, gfp_t gfp_mask)
329{
330 struct vmap_area *va;
331 struct rb_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 unsigned long addr;
Nick Piggindb64fe02008-10-18 20:27:03 -0700333 int purged = 0;
334
Nick Piggin77669702009-02-27 14:03:03 -0800335 BUG_ON(!size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700336 BUG_ON(size & ~PAGE_MASK);
337
Nick Piggindb64fe02008-10-18 20:27:03 -0700338 va = kmalloc_node(sizeof(struct vmap_area),
339 gfp_mask & GFP_RECLAIM_MASK, node);
340 if (unlikely(!va))
341 return ERR_PTR(-ENOMEM);
342
343retry:
Glauber Costa0ae15132008-11-19 15:36:33 -0800344 addr = ALIGN(vstart, align);
345
Nick Piggindb64fe02008-10-18 20:27:03 -0700346 spin_lock(&vmap_area_lock);
Nick Piggin77669702009-02-27 14:03:03 -0800347 if (addr + size - 1 < addr)
348 goto overflow;
349
Nick Piggindb64fe02008-10-18 20:27:03 -0700350 /* XXX: could have a last_hole cache */
351 n = vmap_area_root.rb_node;
352 if (n) {
353 struct vmap_area *first = NULL;
354
355 do {
356 struct vmap_area *tmp;
357 tmp = rb_entry(n, struct vmap_area, rb_node);
358 if (tmp->va_end >= addr) {
359 if (!first && tmp->va_start < addr + size)
360 first = tmp;
361 n = n->rb_left;
362 } else {
363 first = tmp;
364 n = n->rb_right;
365 }
366 } while (n);
367
368 if (!first)
369 goto found;
370
371 if (first->va_end < addr) {
372 n = rb_next(&first->rb_node);
373 if (n)
374 first = rb_entry(n, struct vmap_area, rb_node);
375 else
376 goto found;
377 }
378
Nick Pigginf011c2d2008-11-19 15:36:32 -0800379 while (addr + size > first->va_start && addr + size <= vend) {
Nick Piggindb64fe02008-10-18 20:27:03 -0700380 addr = ALIGN(first->va_end + PAGE_SIZE, align);
Nick Piggin77669702009-02-27 14:03:03 -0800381 if (addr + size - 1 < addr)
382 goto overflow;
Nick Piggindb64fe02008-10-18 20:27:03 -0700383
384 n = rb_next(&first->rb_node);
385 if (n)
386 first = rb_entry(n, struct vmap_area, rb_node);
387 else
388 goto found;
389 }
390 }
391found:
392 if (addr + size > vend) {
Nick Piggin77669702009-02-27 14:03:03 -0800393overflow:
Nick Piggindb64fe02008-10-18 20:27:03 -0700394 spin_unlock(&vmap_area_lock);
395 if (!purged) {
396 purge_vmap_area_lazy();
397 purged = 1;
398 goto retry;
399 }
400 if (printk_ratelimit())
Glauber Costac1279c42009-01-06 14:39:18 -0800401 printk(KERN_WARNING
402 "vmap allocation for size %lu failed: "
403 "use vmalloc=<size> to increase size.\n", size);
Ralph Wuerthner2498ce42009-05-06 16:02:59 -0700404 kfree(va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700405 return ERR_PTR(-EBUSY);
406 }
407
408 BUG_ON(addr & (align-1));
409
410 va->va_start = addr;
411 va->va_end = addr + size;
412 va->flags = 0;
413 __insert_vmap_area(va);
414 spin_unlock(&vmap_area_lock);
415
416 return va;
417}
418
419static void rcu_free_va(struct rcu_head *head)
420{
421 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
422
423 kfree(va);
424}
425
426static void __free_vmap_area(struct vmap_area *va)
427{
428 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
429 rb_erase(&va->rb_node, &vmap_area_root);
430 RB_CLEAR_NODE(&va->rb_node);
431 list_del_rcu(&va->list);
432
Tejun Heoca23e402009-08-14 15:00:52 +0900433 /*
434 * Track the highest possible candidate for pcpu area
435 * allocation. Areas outside of vmalloc area can be returned
436 * here too, consider only end addresses which fall inside
437 * vmalloc area proper.
438 */
439 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
440 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
441
Nick Piggindb64fe02008-10-18 20:27:03 -0700442 call_rcu(&va->rcu_head, rcu_free_va);
443}
444
445/*
446 * Free a region of KVA allocated by alloc_vmap_area
447 */
448static void free_vmap_area(struct vmap_area *va)
449{
450 spin_lock(&vmap_area_lock);
451 __free_vmap_area(va);
452 spin_unlock(&vmap_area_lock);
453}
454
455/*
456 * Clear the pagetable entries of a given vmap_area
457 */
458static void unmap_vmap_area(struct vmap_area *va)
459{
460 vunmap_page_range(va->va_start, va->va_end);
461}
462
Nick Piggincd528582009-01-06 14:39:20 -0800463static void vmap_debug_free_range(unsigned long start, unsigned long end)
464{
465 /*
466 * Unmap page tables and force a TLB flush immediately if
467 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
468 * bugs similarly to those in linear kernel virtual address
469 * space after a page has been freed.
470 *
471 * All the lazy freeing logic is still retained, in order to
472 * minimise intrusiveness of this debugging feature.
473 *
474 * This is going to be *slow* (linear kernel virtual address
475 * debugging doesn't do a broadcast TLB flush so it is a lot
476 * faster).
477 */
478#ifdef CONFIG_DEBUG_PAGEALLOC
479 vunmap_page_range(start, end);
480 flush_tlb_kernel_range(start, end);
481#endif
482}
483
Nick Piggindb64fe02008-10-18 20:27:03 -0700484/*
485 * lazy_max_pages is the maximum amount of virtual address space we gather up
486 * before attempting to purge with a TLB flush.
487 *
488 * There is a tradeoff here: a larger number will cover more kernel page tables
489 * and take slightly longer to purge, but it will linearly reduce the number of
490 * global TLB flushes that must be performed. It would seem natural to scale
491 * this number up linearly with the number of CPUs (because vmapping activity
492 * could also scale linearly with the number of CPUs), however it is likely
493 * that in practice, workloads might be constrained in other ways that mean
494 * vmap activity will not scale linearly with CPUs. Also, I want to be
495 * conservative and not introduce a big latency on huge systems, so go with
496 * a less aggressive log scale. It will still be an improvement over the old
497 * code, and it will be simple to change the scale factor if we find that it
498 * becomes a problem on bigger systems.
499 */
500static unsigned long lazy_max_pages(void)
501{
502 unsigned int log;
503
504 log = fls(num_online_cpus());
505
506 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
507}
508
509static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
510
511/*
512 * Purges all lazily-freed vmap areas.
513 *
514 * If sync is 0 then don't purge if there is already a purge in progress.
515 * If force_flush is 1, then flush kernel TLBs between *start and *end even
516 * if we found no lazy vmap areas to unmap (callers can use this to optimise
517 * their own TLB flushing).
518 * Returns with *start = min(*start, lowest purged address)
519 * *end = max(*end, highest purged address)
520 */
521static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
522 int sync, int force_flush)
523{
Andrew Morton46666d82009-01-15 13:51:15 -0800524 static DEFINE_SPINLOCK(purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700525 LIST_HEAD(valist);
526 struct vmap_area *va;
Vegard Nossumcbb76672009-02-27 14:03:04 -0800527 struct vmap_area *n_va;
Nick Piggindb64fe02008-10-18 20:27:03 -0700528 int nr = 0;
529
530 /*
531 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
532 * should not expect such behaviour. This just simplifies locking for
533 * the case that isn't actually used at the moment anyway.
534 */
535 if (!sync && !force_flush) {
Andrew Morton46666d82009-01-15 13:51:15 -0800536 if (!spin_trylock(&purge_lock))
Nick Piggindb64fe02008-10-18 20:27:03 -0700537 return;
538 } else
Andrew Morton46666d82009-01-15 13:51:15 -0800539 spin_lock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700540
541 rcu_read_lock();
542 list_for_each_entry_rcu(va, &vmap_area_list, list) {
543 if (va->flags & VM_LAZY_FREE) {
544 if (va->va_start < *start)
545 *start = va->va_start;
546 if (va->va_end > *end)
547 *end = va->va_end;
548 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
549 unmap_vmap_area(va);
550 list_add_tail(&va->purge_list, &valist);
551 va->flags |= VM_LAZY_FREEING;
552 va->flags &= ~VM_LAZY_FREE;
553 }
554 }
555 rcu_read_unlock();
556
557 if (nr) {
558 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
559 atomic_sub(nr, &vmap_lazy_nr);
560 }
561
562 if (nr || force_flush)
563 flush_tlb_kernel_range(*start, *end);
564
565 if (nr) {
566 spin_lock(&vmap_area_lock);
Vegard Nossumcbb76672009-02-27 14:03:04 -0800567 list_for_each_entry_safe(va, n_va, &valist, purge_list)
Nick Piggindb64fe02008-10-18 20:27:03 -0700568 __free_vmap_area(va);
569 spin_unlock(&vmap_area_lock);
570 }
Andrew Morton46666d82009-01-15 13:51:15 -0800571 spin_unlock(&purge_lock);
Nick Piggindb64fe02008-10-18 20:27:03 -0700572}
573
574/*
Nick Piggin496850e2008-11-19 15:36:33 -0800575 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
576 * is already purging.
577 */
578static void try_purge_vmap_area_lazy(void)
579{
580 unsigned long start = ULONG_MAX, end = 0;
581
582 __purge_vmap_area_lazy(&start, &end, 0, 0);
583}
584
585/*
Nick Piggindb64fe02008-10-18 20:27:03 -0700586 * Kick off a purge of the outstanding lazy areas.
587 */
588static void purge_vmap_area_lazy(void)
589{
590 unsigned long start = ULONG_MAX, end = 0;
591
Nick Piggin496850e2008-11-19 15:36:33 -0800592 __purge_vmap_area_lazy(&start, &end, 1, 0);
Nick Piggindb64fe02008-10-18 20:27:03 -0700593}
594
595/*
Nick Pigginb29acbd2008-12-01 13:13:47 -0800596 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
597 * called for the correct range previously.
Nick Piggindb64fe02008-10-18 20:27:03 -0700598 */
Nick Pigginb29acbd2008-12-01 13:13:47 -0800599static void free_unmap_vmap_area_noflush(struct vmap_area *va)
Nick Piggindb64fe02008-10-18 20:27:03 -0700600{
601 va->flags |= VM_LAZY_FREE;
602 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
603 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
Nick Piggin496850e2008-11-19 15:36:33 -0800604 try_purge_vmap_area_lazy();
Nick Piggindb64fe02008-10-18 20:27:03 -0700605}
606
Nick Pigginb29acbd2008-12-01 13:13:47 -0800607/*
608 * Free and unmap a vmap area
609 */
610static void free_unmap_vmap_area(struct vmap_area *va)
611{
612 flush_cache_vunmap(va->va_start, va->va_end);
613 free_unmap_vmap_area_noflush(va);
614}
615
Nick Piggindb64fe02008-10-18 20:27:03 -0700616static struct vmap_area *find_vmap_area(unsigned long addr)
617{
618 struct vmap_area *va;
619
620 spin_lock(&vmap_area_lock);
621 va = __find_vmap_area(addr);
622 spin_unlock(&vmap_area_lock);
623
624 return va;
625}
626
627static void free_unmap_vmap_area_addr(unsigned long addr)
628{
629 struct vmap_area *va;
630
631 va = find_vmap_area(addr);
632 BUG_ON(!va);
633 free_unmap_vmap_area(va);
634}
635
636
637/*** Per cpu kva allocator ***/
638
639/*
640 * vmap space is limited especially on 32 bit architectures. Ensure there is
641 * room for at least 16 percpu vmap blocks per CPU.
642 */
643/*
644 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
645 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
646 * instead (we just need a rough idea)
647 */
648#if BITS_PER_LONG == 32
649#define VMALLOC_SPACE (128UL*1024*1024)
650#else
651#define VMALLOC_SPACE (128UL*1024*1024*1024)
652#endif
653
654#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
655#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
656#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
657#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
658#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
659#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
660#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
661 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
662 VMALLOC_PAGES / NR_CPUS / 16))
663
664#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
665
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100666static bool vmap_initialized __read_mostly = false;
667
Nick Piggindb64fe02008-10-18 20:27:03 -0700668struct vmap_block_queue {
669 spinlock_t lock;
670 struct list_head free;
671 struct list_head dirty;
672 unsigned int nr_dirty;
673};
674
675struct vmap_block {
676 spinlock_t lock;
677 struct vmap_area *va;
678 struct vmap_block_queue *vbq;
679 unsigned long free, dirty;
680 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
681 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
682 union {
MinChan Kimd0868172009-03-31 15:19:26 -0700683 struct list_head free_list;
Nick Piggindb64fe02008-10-18 20:27:03 -0700684 struct rcu_head rcu_head;
685 };
686};
687
688/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
689static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
690
691/*
692 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
693 * in the free path. Could get rid of this if we change the API to return a
694 * "cookie" from alloc, to be passed to free. But no big deal yet.
695 */
696static DEFINE_SPINLOCK(vmap_block_tree_lock);
697static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
698
699/*
700 * We should probably have a fallback mechanism to allocate virtual memory
701 * out of partially filled vmap blocks. However vmap block sizing should be
702 * fairly reasonable according to the vmalloc size, so it shouldn't be a
703 * big problem.
704 */
705
706static unsigned long addr_to_vb_idx(unsigned long addr)
707{
708 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
709 addr /= VMAP_BLOCK_SIZE;
710 return addr;
711}
712
713static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
714{
715 struct vmap_block_queue *vbq;
716 struct vmap_block *vb;
717 struct vmap_area *va;
718 unsigned long vb_idx;
719 int node, err;
720
721 node = numa_node_id();
722
723 vb = kmalloc_node(sizeof(struct vmap_block),
724 gfp_mask & GFP_RECLAIM_MASK, node);
725 if (unlikely(!vb))
726 return ERR_PTR(-ENOMEM);
727
728 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
729 VMALLOC_START, VMALLOC_END,
730 node, gfp_mask);
731 if (unlikely(IS_ERR(va))) {
732 kfree(vb);
733 return ERR_PTR(PTR_ERR(va));
734 }
735
736 err = radix_tree_preload(gfp_mask);
737 if (unlikely(err)) {
738 kfree(vb);
739 free_vmap_area(va);
740 return ERR_PTR(err);
741 }
742
743 spin_lock_init(&vb->lock);
744 vb->va = va;
745 vb->free = VMAP_BBMAP_BITS;
746 vb->dirty = 0;
747 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
748 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
749 INIT_LIST_HEAD(&vb->free_list);
Nick Piggindb64fe02008-10-18 20:27:03 -0700750
751 vb_idx = addr_to_vb_idx(va->va_start);
752 spin_lock(&vmap_block_tree_lock);
753 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
754 spin_unlock(&vmap_block_tree_lock);
755 BUG_ON(err);
756 radix_tree_preload_end();
757
758 vbq = &get_cpu_var(vmap_block_queue);
759 vb->vbq = vbq;
760 spin_lock(&vbq->lock);
761 list_add(&vb->free_list, &vbq->free);
762 spin_unlock(&vbq->lock);
763 put_cpu_var(vmap_cpu_blocks);
764
765 return vb;
766}
767
768static void rcu_free_vb(struct rcu_head *head)
769{
770 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
771
772 kfree(vb);
773}
774
775static void free_vmap_block(struct vmap_block *vb)
776{
777 struct vmap_block *tmp;
778 unsigned long vb_idx;
779
MinChan Kimd0868172009-03-31 15:19:26 -0700780 BUG_ON(!list_empty(&vb->free_list));
Nick Piggindb64fe02008-10-18 20:27:03 -0700781
782 vb_idx = addr_to_vb_idx(vb->va->va_start);
783 spin_lock(&vmap_block_tree_lock);
784 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
785 spin_unlock(&vmap_block_tree_lock);
786 BUG_ON(tmp != vb);
787
Nick Pigginb29acbd2008-12-01 13:13:47 -0800788 free_unmap_vmap_area_noflush(vb->va);
Nick Piggindb64fe02008-10-18 20:27:03 -0700789 call_rcu(&vb->rcu_head, rcu_free_vb);
790}
791
792static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
793{
794 struct vmap_block_queue *vbq;
795 struct vmap_block *vb;
796 unsigned long addr = 0;
797 unsigned int order;
798
799 BUG_ON(size & ~PAGE_MASK);
800 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
801 order = get_order(size);
802
803again:
804 rcu_read_lock();
805 vbq = &get_cpu_var(vmap_block_queue);
806 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
807 int i;
808
809 spin_lock(&vb->lock);
810 i = bitmap_find_free_region(vb->alloc_map,
811 VMAP_BBMAP_BITS, order);
812
813 if (i >= 0) {
814 addr = vb->va->va_start + (i << PAGE_SHIFT);
815 BUG_ON(addr_to_vb_idx(addr) !=
816 addr_to_vb_idx(vb->va->va_start));
817 vb->free -= 1UL << order;
818 if (vb->free == 0) {
819 spin_lock(&vbq->lock);
820 list_del_init(&vb->free_list);
821 spin_unlock(&vbq->lock);
822 }
823 spin_unlock(&vb->lock);
824 break;
825 }
826 spin_unlock(&vb->lock);
827 }
828 put_cpu_var(vmap_cpu_blocks);
829 rcu_read_unlock();
830
831 if (!addr) {
832 vb = new_vmap_block(gfp_mask);
833 if (IS_ERR(vb))
834 return vb;
835 goto again;
836 }
837
838 return (void *)addr;
839}
840
841static void vb_free(const void *addr, unsigned long size)
842{
843 unsigned long offset;
844 unsigned long vb_idx;
845 unsigned int order;
846 struct vmap_block *vb;
847
848 BUG_ON(size & ~PAGE_MASK);
849 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
Nick Pigginb29acbd2008-12-01 13:13:47 -0800850
851 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
852
Nick Piggindb64fe02008-10-18 20:27:03 -0700853 order = get_order(size);
854
855 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
856
857 vb_idx = addr_to_vb_idx((unsigned long)addr);
858 rcu_read_lock();
859 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
860 rcu_read_unlock();
861 BUG_ON(!vb);
862
863 spin_lock(&vb->lock);
864 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
MinChan Kimd0868172009-03-31 15:19:26 -0700865
Nick Piggindb64fe02008-10-18 20:27:03 -0700866 vb->dirty += 1UL << order;
867 if (vb->dirty == VMAP_BBMAP_BITS) {
868 BUG_ON(vb->free || !list_empty(&vb->free_list));
869 spin_unlock(&vb->lock);
870 free_vmap_block(vb);
871 } else
872 spin_unlock(&vb->lock);
873}
874
875/**
876 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
877 *
878 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
879 * to amortize TLB flushing overheads. What this means is that any page you
880 * have now, may, in a former life, have been mapped into kernel virtual
881 * address by the vmap layer and so there might be some CPUs with TLB entries
882 * still referencing that page (additional to the regular 1:1 kernel mapping).
883 *
884 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
885 * be sure that none of the pages we have control over will have any aliases
886 * from the vmap layer.
887 */
888void vm_unmap_aliases(void)
889{
890 unsigned long start = ULONG_MAX, end = 0;
891 int cpu;
892 int flush = 0;
893
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +1100894 if (unlikely(!vmap_initialized))
895 return;
896
Nick Piggindb64fe02008-10-18 20:27:03 -0700897 for_each_possible_cpu(cpu) {
898 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
899 struct vmap_block *vb;
900
901 rcu_read_lock();
902 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
903 int i;
904
905 spin_lock(&vb->lock);
906 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
907 while (i < VMAP_BBMAP_BITS) {
908 unsigned long s, e;
909 int j;
910 j = find_next_zero_bit(vb->dirty_map,
911 VMAP_BBMAP_BITS, i);
912
913 s = vb->va->va_start + (i << PAGE_SHIFT);
914 e = vb->va->va_start + (j << PAGE_SHIFT);
915 vunmap_page_range(s, e);
916 flush = 1;
917
918 if (s < start)
919 start = s;
920 if (e > end)
921 end = e;
922
923 i = j;
924 i = find_next_bit(vb->dirty_map,
925 VMAP_BBMAP_BITS, i);
926 }
927 spin_unlock(&vb->lock);
928 }
929 rcu_read_unlock();
930 }
931
932 __purge_vmap_area_lazy(&start, &end, 1, flush);
933}
934EXPORT_SYMBOL_GPL(vm_unmap_aliases);
935
936/**
937 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
938 * @mem: the pointer returned by vm_map_ram
939 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
940 */
941void vm_unmap_ram(const void *mem, unsigned int count)
942{
943 unsigned long size = count << PAGE_SHIFT;
944 unsigned long addr = (unsigned long)mem;
945
946 BUG_ON(!addr);
947 BUG_ON(addr < VMALLOC_START);
948 BUG_ON(addr > VMALLOC_END);
949 BUG_ON(addr & (PAGE_SIZE-1));
950
951 debug_check_no_locks_freed(mem, size);
Nick Piggincd528582009-01-06 14:39:20 -0800952 vmap_debug_free_range(addr, addr+size);
Nick Piggindb64fe02008-10-18 20:27:03 -0700953
954 if (likely(count <= VMAP_MAX_ALLOC))
955 vb_free(mem, size);
956 else
957 free_unmap_vmap_area_addr(addr);
958}
959EXPORT_SYMBOL(vm_unmap_ram);
960
961/**
962 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
963 * @pages: an array of pointers to the pages to be mapped
964 * @count: number of pages
965 * @node: prefer to allocate data structures on this node
966 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
Randy Dunlape99c97a2008-10-29 14:01:09 -0700967 *
968 * Returns: a pointer to the address that has been mapped, or %NULL on failure
Nick Piggindb64fe02008-10-18 20:27:03 -0700969 */
970void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
971{
972 unsigned long size = count << PAGE_SHIFT;
973 unsigned long addr;
974 void *mem;
975
976 if (likely(count <= VMAP_MAX_ALLOC)) {
977 mem = vb_alloc(size, GFP_KERNEL);
978 if (IS_ERR(mem))
979 return NULL;
980 addr = (unsigned long)mem;
981 } else {
982 struct vmap_area *va;
983 va = alloc_vmap_area(size, PAGE_SIZE,
984 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
985 if (IS_ERR(va))
986 return NULL;
987
988 addr = va->va_start;
989 mem = (void *)addr;
990 }
991 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
992 vm_unmap_ram(mem, count);
993 return NULL;
994 }
995 return mem;
996}
997EXPORT_SYMBOL(vm_map_ram);
998
Tejun Heof0aa6612009-02-20 16:29:08 +0900999/**
1000 * vm_area_register_early - register vmap area early during boot
1001 * @vm: vm_struct to register
Tejun Heoc0c0a292009-02-24 11:57:21 +09001002 * @align: requested alignment
Tejun Heof0aa6612009-02-20 16:29:08 +09001003 *
1004 * This function is used to register kernel vm area before
1005 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1006 * proper values on entry and other fields should be zero. On return,
1007 * vm->addr contains the allocated address.
1008 *
1009 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1010 */
Tejun Heoc0c0a292009-02-24 11:57:21 +09001011void __init vm_area_register_early(struct vm_struct *vm, size_t align)
Tejun Heof0aa6612009-02-20 16:29:08 +09001012{
1013 static size_t vm_init_off __initdata;
Tejun Heoc0c0a292009-02-24 11:57:21 +09001014 unsigned long addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001015
Tejun Heoc0c0a292009-02-24 11:57:21 +09001016 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1017 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1018
1019 vm->addr = (void *)addr;
Tejun Heof0aa6612009-02-20 16:29:08 +09001020
1021 vm->next = vmlist;
1022 vmlist = vm;
1023}
1024
Nick Piggindb64fe02008-10-18 20:27:03 -07001025void __init vmalloc_init(void)
1026{
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001027 struct vmap_area *va;
1028 struct vm_struct *tmp;
Nick Piggindb64fe02008-10-18 20:27:03 -07001029 int i;
1030
1031 for_each_possible_cpu(i) {
1032 struct vmap_block_queue *vbq;
1033
1034 vbq = &per_cpu(vmap_block_queue, i);
1035 spin_lock_init(&vbq->lock);
1036 INIT_LIST_HEAD(&vbq->free);
1037 INIT_LIST_HEAD(&vbq->dirty);
1038 vbq->nr_dirty = 0;
1039 }
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001040
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001041 /* Import existing vmlist entries. */
1042 for (tmp = vmlist; tmp; tmp = tmp->next) {
Pekka Enberg43ebdac2009-05-25 15:01:35 +03001043 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
Ivan Kokshaysky822c18f2009-01-15 13:50:48 -08001044 va->flags = tmp->flags | VM_VM_AREA;
1045 va->va_start = (unsigned long)tmp->addr;
1046 va->va_end = va->va_start + tmp->size;
1047 __insert_vmap_area(va);
1048 }
Tejun Heoca23e402009-08-14 15:00:52 +09001049
1050 vmap_area_pcpu_hole = VMALLOC_END;
1051
Jeremy Fitzhardinge9b463332008-10-28 19:22:34 +11001052 vmap_initialized = true;
Nick Piggindb64fe02008-10-18 20:27:03 -07001053}
1054
Tejun Heo8fc48982009-02-20 16:29:08 +09001055/**
1056 * map_kernel_range_noflush - map kernel VM area with the specified pages
1057 * @addr: start of the VM area to map
1058 * @size: size of the VM area to map
1059 * @prot: page protection flags to use
1060 * @pages: pages to map
1061 *
1062 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1063 * specify should have been allocated using get_vm_area() and its
1064 * friends.
1065 *
1066 * NOTE:
1067 * This function does NOT do any cache flushing. The caller is
1068 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1069 * before calling this function.
1070 *
1071 * RETURNS:
1072 * The number of pages mapped on success, -errno on failure.
1073 */
1074int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1075 pgprot_t prot, struct page **pages)
1076{
1077 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1078}
1079
1080/**
1081 * unmap_kernel_range_noflush - unmap kernel VM area
1082 * @addr: start of the VM area to unmap
1083 * @size: size of the VM area to unmap
1084 *
1085 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1086 * specify should have been allocated using get_vm_area() and its
1087 * friends.
1088 *
1089 * NOTE:
1090 * This function does NOT do any cache flushing. The caller is
1091 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1092 * before calling this function and flush_tlb_kernel_range() after.
1093 */
1094void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1095{
1096 vunmap_page_range(addr, addr + size);
1097}
1098
1099/**
1100 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1101 * @addr: start of the VM area to unmap
1102 * @size: size of the VM area to unmap
1103 *
1104 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1105 * the unmapping and tlb after.
1106 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001107void unmap_kernel_range(unsigned long addr, unsigned long size)
1108{
1109 unsigned long end = addr + size;
Tejun Heof6fcba72009-02-20 15:38:48 -08001110
1111 flush_cache_vunmap(addr, end);
Nick Piggindb64fe02008-10-18 20:27:03 -07001112 vunmap_page_range(addr, end);
1113 flush_tlb_kernel_range(addr, end);
1114}
1115
1116int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1117{
1118 unsigned long addr = (unsigned long)area->addr;
1119 unsigned long end = addr + area->size - PAGE_SIZE;
1120 int err;
1121
1122 err = vmap_page_range(addr, end, prot, *pages);
1123 if (err > 0) {
1124 *pages += err;
1125 err = 0;
1126 }
1127
1128 return err;
1129}
1130EXPORT_SYMBOL_GPL(map_vm_area);
1131
1132/*** Old vmalloc interfaces ***/
1133DEFINE_RWLOCK(vmlist_lock);
1134struct vm_struct *vmlist;
1135
Tejun Heocf88c792009-08-14 15:00:52 +09001136static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1137 unsigned long flags, void *caller)
1138{
1139 struct vm_struct *tmp, **p;
1140
1141 vm->flags = flags;
1142 vm->addr = (void *)va->va_start;
1143 vm->size = va->va_end - va->va_start;
1144 vm->caller = caller;
1145 va->private = vm;
1146 va->flags |= VM_VM_AREA;
1147
1148 write_lock(&vmlist_lock);
1149 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1150 if (tmp->addr >= vm->addr)
1151 break;
1152 }
1153 vm->next = *p;
1154 *p = vm;
1155 write_unlock(&vmlist_lock);
1156}
1157
Nick Piggindb64fe02008-10-18 20:27:03 -07001158static struct vm_struct *__get_vm_area_node(unsigned long size,
1159 unsigned long flags, unsigned long start, unsigned long end,
1160 int node, gfp_t gfp_mask, void *caller)
1161{
1162 static struct vmap_area *va;
1163 struct vm_struct *area;
Nick Piggindb64fe02008-10-18 20:27:03 -07001164 unsigned long align = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001166 BUG_ON(in_interrupt());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 if (flags & VM_IOREMAP) {
1168 int bit = fls(size);
1169
1170 if (bit > IOREMAP_MAX_ORDER)
1171 bit = IOREMAP_MAX_ORDER;
1172 else if (bit < PAGE_SHIFT)
1173 bit = PAGE_SHIFT;
1174
1175 align = 1ul << bit;
1176 }
Nick Piggindb64fe02008-10-18 20:27:03 -07001177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 size = PAGE_ALIGN(size);
OGAWA Hirofumi31be8302006-11-16 01:19:29 -08001179 if (unlikely(!size))
1180 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Tejun Heocf88c792009-08-14 15:00:52 +09001182 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 if (unlikely(!area))
1184 return NULL;
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 /*
1187 * We always allocate a guard page.
1188 */
1189 size += PAGE_SIZE;
1190
Nick Piggindb64fe02008-10-18 20:27:03 -07001191 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1192 if (IS_ERR(va)) {
1193 kfree(area);
1194 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Tejun Heocf88c792009-08-14 15:00:52 +09001197 insert_vmalloc_vm(area, va, flags, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 return area;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199}
1200
Christoph Lameter930fc452005-10-29 18:15:41 -07001201struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1202 unsigned long start, unsigned long end)
1203{
Christoph Lameter23016962008-04-28 02:12:42 -07001204 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1205 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001206}
Rusty Russell5992b6d2007-07-19 01:49:21 -07001207EXPORT_SYMBOL_GPL(__get_vm_area);
Christoph Lameter930fc452005-10-29 18:15:41 -07001208
Benjamin Herrenschmidtc2968612009-02-18 14:48:12 -08001209struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1210 unsigned long start, unsigned long end,
1211 void *caller)
1212{
1213 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1214 caller);
1215}
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217/**
Simon Arlott183ff222007-10-20 01:27:18 +02001218 * get_vm_area - reserve a contiguous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 * @size: size of the area
1220 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1221 *
1222 * Search an area of @size in the kernel virtual mapping area,
1223 * and reserved it for out purposes. Returns the area descriptor
1224 * on success or %NULL on failure.
1225 */
1226struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1227{
Christoph Lameter23016962008-04-28 02:12:42 -07001228 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1229 -1, GFP_KERNEL, __builtin_return_address(0));
1230}
1231
1232struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1233 void *caller)
1234{
1235 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1236 -1, GFP_KERNEL, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237}
1238
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001239struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1240 int node, gfp_t gfp_mask)
Christoph Lameter930fc452005-10-29 18:15:41 -07001241{
Giridhar Pemmasani52fd24c2006-10-28 10:38:34 -07001242 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
Christoph Lameter23016962008-04-28 02:12:42 -07001243 gfp_mask, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001244}
1245
Nick Piggindb64fe02008-10-18 20:27:03 -07001246static struct vm_struct *find_vm_area(const void *addr)
Nick Piggin83342312006-06-23 02:03:20 -07001247{
Nick Piggindb64fe02008-10-18 20:27:03 -07001248 struct vmap_area *va;
Nick Piggin83342312006-06-23 02:03:20 -07001249
Nick Piggindb64fe02008-10-18 20:27:03 -07001250 va = find_vmap_area((unsigned long)addr);
1251 if (va && va->flags & VM_VM_AREA)
1252 return va->private;
Nick Piggin83342312006-06-23 02:03:20 -07001253
Andi Kleen7856dfe2005-05-20 14:27:57 -07001254 return NULL;
Andi Kleen7856dfe2005-05-20 14:27:57 -07001255}
1256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257/**
Simon Arlott183ff222007-10-20 01:27:18 +02001258 * remove_vm_area - find and remove a continuous kernel virtual area
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 * @addr: base address
1260 *
1261 * Search for the kernel VM area starting at @addr, and remove it.
1262 * This function returns the found VM area, but using it is NOT safe
Andi Kleen7856dfe2005-05-20 14:27:57 -07001263 * on SMP machines, except for its size or flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001265struct vm_struct *remove_vm_area(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
Nick Piggindb64fe02008-10-18 20:27:03 -07001267 struct vmap_area *va;
1268
1269 va = find_vmap_area((unsigned long)addr);
1270 if (va && va->flags & VM_VM_AREA) {
1271 struct vm_struct *vm = va->private;
1272 struct vm_struct *tmp, **p;
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001273 /*
1274 * remove from list and disallow access to this vm_struct
1275 * before unmap. (address range confliction is maintained by
1276 * vmap.)
1277 */
Nick Piggindb64fe02008-10-18 20:27:03 -07001278 write_lock(&vmlist_lock);
1279 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1280 ;
1281 *p = tmp->next;
1282 write_unlock(&vmlist_lock);
1283
KAMEZAWA Hiroyukidd32c272009-09-21 17:02:32 -07001284 vmap_debug_free_range(va->va_start, va->va_end);
1285 free_unmap_vmap_area(va);
1286 vm->size -= PAGE_SIZE;
1287
Nick Piggindb64fe02008-10-18 20:27:03 -07001288 return vm;
1289 }
1290 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291}
1292
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001293static void __vunmap(const void *addr, int deallocate_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
1295 struct vm_struct *area;
1296
1297 if (!addr)
1298 return;
1299
1300 if ((PAGE_SIZE-1) & (unsigned long)addr) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001301 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 return;
1303 }
1304
1305 area = remove_vm_area(addr);
1306 if (unlikely(!area)) {
Arjan van de Ven4c8573e2008-07-25 19:45:37 -07001307 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 return;
1310 }
1311
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001312 debug_check_no_locks_freed(addr, area->size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001313 debug_check_no_obj_freed(addr, area->size);
Ingo Molnar9a11b49a2006-07-03 00:24:33 -07001314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 if (deallocate_pages) {
1316 int i;
1317
1318 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001319 struct page *page = area->pages[i];
1320
1321 BUG_ON(!page);
1322 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 }
1324
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001325 if (area->flags & VM_VPAGES)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 vfree(area->pages);
1327 else
1328 kfree(area->pages);
1329 }
1330
1331 kfree(area);
1332 return;
1333}
1334
1335/**
1336 * vfree - release memory allocated by vmalloc()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * @addr: memory base address
1338 *
Simon Arlott183ff222007-10-20 01:27:18 +02001339 * Free the virtually continuous memory area starting at @addr, as
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001340 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1341 * NULL, no operation is performed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001343 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001345void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
1347 BUG_ON(in_interrupt());
Catalin Marinas89219d32009-06-11 13:23:19 +01001348
1349 kmemleak_free(addr);
1350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 __vunmap(addr, 1);
1352}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353EXPORT_SYMBOL(vfree);
1354
1355/**
1356 * vunmap - release virtual mapping obtained by vmap()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 * @addr: memory base address
1358 *
1359 * Free the virtually contiguous memory area starting at @addr,
1360 * which was created from the page array passed to vmap().
1361 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07001362 * Must not be called in interrupt context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
Christoph Lameterb3bdda02008-02-04 22:28:32 -08001364void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365{
1366 BUG_ON(in_interrupt());
Peter Zijlstra34754b62009-02-25 16:04:03 +01001367 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 __vunmap(addr, 0);
1369}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370EXPORT_SYMBOL(vunmap);
1371
1372/**
1373 * vmap - map an array of pages into virtually contiguous space
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 * @pages: array of page pointers
1375 * @count: number of pages to map
1376 * @flags: vm_area->flags
1377 * @prot: page protection for the mapping
1378 *
1379 * Maps @count pages from @pages into contiguous kernel virtual
1380 * space.
1381 */
1382void *vmap(struct page **pages, unsigned int count,
1383 unsigned long flags, pgprot_t prot)
1384{
1385 struct vm_struct *area;
1386
Peter Zijlstra34754b62009-02-25 16:04:03 +01001387 might_sleep();
1388
Jan Beulich44813742009-09-21 17:03:05 -07001389 if (count > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 return NULL;
1391
Christoph Lameter23016962008-04-28 02:12:42 -07001392 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1393 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if (!area)
1395 return NULL;
Christoph Lameter23016962008-04-28 02:12:42 -07001396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 if (map_vm_area(area, prot, &pages)) {
1398 vunmap(area->addr);
1399 return NULL;
1400 }
1401
1402 return area->addr;
1403}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404EXPORT_SYMBOL(vmap);
1405
Nick Piggindb64fe02008-10-18 20:27:03 -07001406static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1407 int node, void *caller);
Adrian Bunke31d9eb2008-02-04 22:29:09 -08001408static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
Christoph Lameter23016962008-04-28 02:12:42 -07001409 pgprot_t prot, int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410{
1411 struct page **pages;
1412 unsigned int nr_pages, array_size, i;
1413
1414 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1415 array_size = (nr_pages * sizeof(struct page *));
1416
1417 area->nr_pages = nr_pages;
1418 /* Please note that the recursion is strictly bounded. */
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001419 if (array_size > PAGE_SIZE) {
Christoph Lameter94f60302007-07-17 04:03:29 -07001420 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
Christoph Lameter23016962008-04-28 02:12:42 -07001421 PAGE_KERNEL, node, caller);
Jan Kiszka8757d5f2006-07-14 00:23:56 -07001422 area->flags |= VM_VPAGES;
Andrew Morton286e1ea2006-10-17 00:09:57 -07001423 } else {
1424 pages = kmalloc_node(array_size,
Christoph Lameter6cb06222007-10-16 01:25:41 -07001425 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
Andrew Morton286e1ea2006-10-17 00:09:57 -07001426 node);
1427 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 area->pages = pages;
Christoph Lameter23016962008-04-28 02:12:42 -07001429 area->caller = caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (!area->pages) {
1431 remove_vm_area(area->addr);
1432 kfree(area);
1433 return NULL;
1434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 for (i = 0; i < area->nr_pages; i++) {
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001437 struct page *page;
1438
Christoph Lameter930fc452005-10-29 18:15:41 -07001439 if (node < 0)
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001440 page = alloc_page(gfp_mask);
Christoph Lameter930fc452005-10-29 18:15:41 -07001441 else
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001442 page = alloc_pages_node(node, gfp_mask, 0);
1443
1444 if (unlikely(!page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 /* Successfully allocated i pages, free them in __vunmap() */
1446 area->nr_pages = i;
1447 goto fail;
1448 }
Christoph Lameterbf53d6f2008-02-04 22:28:34 -08001449 area->pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 }
1451
1452 if (map_vm_area(area, prot, &pages))
1453 goto fail;
1454 return area->addr;
1455
1456fail:
1457 vfree(area->addr);
1458 return NULL;
1459}
1460
Christoph Lameter930fc452005-10-29 18:15:41 -07001461void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1462{
Catalin Marinas89219d32009-06-11 13:23:19 +01001463 void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1,
1464 __builtin_return_address(0));
1465
1466 /*
1467 * A ref_count = 3 is needed because the vm_struct and vmap_area
1468 * structures allocated in the __get_vm_area_node() function contain
1469 * references to the virtual address of the vmalloc'ed block.
1470 */
1471 kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask);
1472
1473 return addr;
Christoph Lameter930fc452005-10-29 18:15:41 -07001474}
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001477 * __vmalloc_node - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 * @size: allocation size
1479 * @gfp_mask: flags for the page level allocator
1480 * @prot: protection mask for the allocated pages
Randy Dunlapd44e0782005-11-07 01:01:10 -08001481 * @node: node to use for allocation or -1
Randy Dunlapc85d1942008-05-01 04:34:48 -07001482 * @caller: caller's return address
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 *
1484 * Allocate enough pages to cover @size from the page level
1485 * allocator with @gfp_mask flags. Map them into contiguous
1486 * kernel virtual space, using a pagetable protection of @prot.
1487 */
Adrian Bunkb2213852006-09-25 23:31:02 -07001488static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
Christoph Lameter23016962008-04-28 02:12:42 -07001489 int node, void *caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
1491 struct vm_struct *area;
Catalin Marinas89219d32009-06-11 13:23:19 +01001492 void *addr;
1493 unsigned long real_size = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 size = PAGE_ALIGN(size);
Jan Beulich44813742009-09-21 17:03:05 -07001496 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return NULL;
1498
Christoph Lameter23016962008-04-28 02:12:42 -07001499 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1500 node, gfp_mask, caller);
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if (!area)
1503 return NULL;
1504
Catalin Marinas89219d32009-06-11 13:23:19 +01001505 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1506
1507 /*
1508 * A ref_count = 3 is needed because the vm_struct and vmap_area
1509 * structures allocated in the __get_vm_area_node() function contain
1510 * references to the virtual address of the vmalloc'ed block.
1511 */
1512 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1513
1514 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515}
1516
Christoph Lameter930fc452005-10-29 18:15:41 -07001517void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1518{
Christoph Lameter23016962008-04-28 02:12:42 -07001519 return __vmalloc_node(size, gfp_mask, prot, -1,
1520 __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001521}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522EXPORT_SYMBOL(__vmalloc);
1523
1524/**
1525 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 * @size: allocation size
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * Allocate enough pages to cover @size from the page level
1528 * allocator and map them into contiguous kernel virtual space.
1529 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001530 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 * use __vmalloc() instead.
1532 */
1533void *vmalloc(unsigned long size)
1534{
Christoph Lameter23016962008-04-28 02:12:42 -07001535 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1536 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538EXPORT_SYMBOL(vmalloc);
1539
Christoph Lameter930fc452005-10-29 18:15:41 -07001540/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001541 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1542 * @size: allocation size
Nick Piggin83342312006-06-23 02:03:20 -07001543 *
Rolf Eike Beeread04082006-09-27 01:50:13 -07001544 * The resulting memory area is zeroed so it can be mapped to userspace
1545 * without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001546 */
1547void *vmalloc_user(unsigned long size)
1548{
1549 struct vm_struct *area;
1550 void *ret;
1551
Glauber Costa84877842009-01-06 14:39:19 -08001552 ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1553 PAGE_KERNEL, -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001554 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001555 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001556 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001557 }
Nick Piggin83342312006-06-23 02:03:20 -07001558 return ret;
1559}
1560EXPORT_SYMBOL(vmalloc_user);
1561
1562/**
Christoph Lameter930fc452005-10-29 18:15:41 -07001563 * vmalloc_node - allocate memory on a specific node
Christoph Lameter930fc452005-10-29 18:15:41 -07001564 * @size: allocation size
Randy Dunlapd44e0782005-11-07 01:01:10 -08001565 * @node: numa node
Christoph Lameter930fc452005-10-29 18:15:41 -07001566 *
1567 * Allocate enough pages to cover @size from the page level
1568 * allocator and map them into contiguous kernel virtual space.
1569 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001570 * For tight control over page level allocator and protection flags
Christoph Lameter930fc452005-10-29 18:15:41 -07001571 * use __vmalloc() instead.
1572 */
1573void *vmalloc_node(unsigned long size, int node)
1574{
Christoph Lameter23016962008-04-28 02:12:42 -07001575 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1576 node, __builtin_return_address(0));
Christoph Lameter930fc452005-10-29 18:15:41 -07001577}
1578EXPORT_SYMBOL(vmalloc_node);
1579
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001580#ifndef PAGE_KERNEL_EXEC
1581# define PAGE_KERNEL_EXEC PAGE_KERNEL
1582#endif
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584/**
1585 * vmalloc_exec - allocate virtually contiguous, executable memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 * @size: allocation size
1587 *
1588 * Kernel-internal function to allocate enough pages to cover @size
1589 * the page level allocator and map them into contiguous and
1590 * executable kernel virtual space.
1591 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +02001592 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 * use __vmalloc() instead.
1594 */
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596void *vmalloc_exec(unsigned long size)
1597{
Glauber Costa84877842009-01-06 14:39:19 -08001598 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1599 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600}
1601
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001602#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001603#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001604#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
Benjamin Herrenschmidt7ac674f2007-07-19 01:49:10 -07001605#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
Andi Kleen0d08e0d2007-05-02 19:27:12 +02001606#else
1607#define GFP_VMALLOC32 GFP_KERNEL
1608#endif
1609
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610/**
1611 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 * @size: allocation size
1613 *
1614 * Allocate enough 32bit PA addressable pages to cover @size from the
1615 * page level allocator and map them into contiguous kernel virtual space.
1616 */
1617void *vmalloc_32(unsigned long size)
1618{
Glauber Costa84877842009-01-06 14:39:19 -08001619 return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1620 -1, __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622EXPORT_SYMBOL(vmalloc_32);
1623
Nick Piggin83342312006-06-23 02:03:20 -07001624/**
Rolf Eike Beeread04082006-09-27 01:50:13 -07001625 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
Nick Piggin83342312006-06-23 02:03:20 -07001626 * @size: allocation size
Rolf Eike Beeread04082006-09-27 01:50:13 -07001627 *
1628 * The resulting memory area is 32bit addressable and zeroed so it can be
1629 * mapped to userspace without leaking data.
Nick Piggin83342312006-06-23 02:03:20 -07001630 */
1631void *vmalloc_32_user(unsigned long size)
1632{
1633 struct vm_struct *area;
1634 void *ret;
1635
Glauber Costa84877842009-01-06 14:39:19 -08001636 ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1637 -1, __builtin_return_address(0));
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001638 if (ret) {
Nick Piggindb64fe02008-10-18 20:27:03 -07001639 area = find_vm_area(ret);
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001640 area->flags |= VM_USERMAP;
Eric Dumazet2b4ac442006-11-10 12:27:48 -08001641 }
Nick Piggin83342312006-06-23 02:03:20 -07001642 return ret;
1643}
1644EXPORT_SYMBOL(vmalloc_32_user);
1645
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001646/*
1647 * small helper routine , copy contents to buf from addr.
1648 * If the page is not present, fill zero.
1649 */
1650
1651static int aligned_vread(char *buf, char *addr, unsigned long count)
1652{
1653 struct page *p;
1654 int copied = 0;
1655
1656 while (count) {
1657 unsigned long offset, length;
1658
1659 offset = (unsigned long)addr & ~PAGE_MASK;
1660 length = PAGE_SIZE - offset;
1661 if (length > count)
1662 length = count;
1663 p = vmalloc_to_page(addr);
1664 /*
1665 * To do safe access to this _mapped_ area, we need
1666 * lock. But adding lock here means that we need to add
1667 * overhead of vmalloc()/vfree() calles for this _debug_
1668 * interface, rarely used. Instead of that, we'll use
1669 * kmap() and get small overhead in this access function.
1670 */
1671 if (p) {
1672 /*
1673 * we can expect USER0 is not used (see vread/vwrite's
1674 * function description)
1675 */
1676 void *map = kmap_atomic(p, KM_USER0);
1677 memcpy(buf, map + offset, length);
1678 kunmap_atomic(map, KM_USER0);
1679 } else
1680 memset(buf, 0, length);
1681
1682 addr += length;
1683 buf += length;
1684 copied += length;
1685 count -= length;
1686 }
1687 return copied;
1688}
1689
1690static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1691{
1692 struct page *p;
1693 int copied = 0;
1694
1695 while (count) {
1696 unsigned long offset, length;
1697
1698 offset = (unsigned long)addr & ~PAGE_MASK;
1699 length = PAGE_SIZE - offset;
1700 if (length > count)
1701 length = count;
1702 p = vmalloc_to_page(addr);
1703 /*
1704 * To do safe access to this _mapped_ area, we need
1705 * lock. But adding lock here means that we need to add
1706 * overhead of vmalloc()/vfree() calles for this _debug_
1707 * interface, rarely used. Instead of that, we'll use
1708 * kmap() and get small overhead in this access function.
1709 */
1710 if (p) {
1711 /*
1712 * we can expect USER0 is not used (see vread/vwrite's
1713 * function description)
1714 */
1715 void *map = kmap_atomic(p, KM_USER0);
1716 memcpy(map + offset, buf, length);
1717 kunmap_atomic(map, KM_USER0);
1718 }
1719 addr += length;
1720 buf += length;
1721 copied += length;
1722 count -= length;
1723 }
1724 return copied;
1725}
1726
1727/**
1728 * vread() - read vmalloc area in a safe way.
1729 * @buf: buffer for reading data
1730 * @addr: vm address.
1731 * @count: number of bytes to be read.
1732 *
1733 * Returns # of bytes which addr and buf should be increased.
1734 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1735 * includes any intersect with alive vmalloc area.
1736 *
1737 * This function checks that addr is a valid vmalloc'ed area, and
1738 * copy data from that area to a given buffer. If the given memory range
1739 * of [addr...addr+count) includes some valid address, data is copied to
1740 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1741 * IOREMAP area is treated as memory hole and no copy is done.
1742 *
1743 * If [addr...addr+count) doesn't includes any intersects with alive
1744 * vm_struct area, returns 0.
1745 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1746 * the caller should guarantee KM_USER0 is not used.
1747 *
1748 * Note: In usual ops, vread() is never necessary because the caller
1749 * should know vmalloc() area is valid and can use memcpy().
1750 * This is for routines which have to access vmalloc area without
1751 * any informaion, as /dev/kmem.
1752 *
1753 */
1754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755long vread(char *buf, char *addr, unsigned long count)
1756{
1757 struct vm_struct *tmp;
1758 char *vaddr, *buf_start = buf;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001759 unsigned long buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 unsigned long n;
1761
1762 /* Don't allow overflow */
1763 if ((unsigned long) addr + count < count)
1764 count = -(unsigned long) addr;
1765
1766 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001767 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 vaddr = (char *) tmp->addr;
1769 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1770 continue;
1771 while (addr < vaddr) {
1772 if (count == 0)
1773 goto finished;
1774 *buf = '\0';
1775 buf++;
1776 addr++;
1777 count--;
1778 }
1779 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001780 if (n > count)
1781 n = count;
1782 if (!(tmp->flags & VM_IOREMAP))
1783 aligned_vread(buf, addr, n);
1784 else /* IOREMAP area is treated as memory hole */
1785 memset(buf, 0, n);
1786 buf += n;
1787 addr += n;
1788 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 }
1790finished:
1791 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001792
1793 if (buf == buf_start)
1794 return 0;
1795 /* zero-fill memory holes */
1796 if (buf != buf_start + buflen)
1797 memset(buf, 0, buflen - (buf - buf_start));
1798
1799 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800}
1801
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001802/**
1803 * vwrite() - write vmalloc area in a safe way.
1804 * @buf: buffer for source data
1805 * @addr: vm address.
1806 * @count: number of bytes to be read.
1807 *
1808 * Returns # of bytes which addr and buf should be incresed.
1809 * (same number to @count).
1810 * If [addr...addr+count) doesn't includes any intersect with valid
1811 * vmalloc area, returns 0.
1812 *
1813 * This function checks that addr is a valid vmalloc'ed area, and
1814 * copy data from a buffer to the given addr. If specified range of
1815 * [addr...addr+count) includes some valid address, data is copied from
1816 * proper area of @buf. If there are memory holes, no copy to hole.
1817 * IOREMAP area is treated as memory hole and no copy is done.
1818 *
1819 * If [addr...addr+count) doesn't includes any intersects with alive
1820 * vm_struct area, returns 0.
1821 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1822 * the caller should guarantee KM_USER0 is not used.
1823 *
1824 * Note: In usual ops, vwrite() is never necessary because the caller
1825 * should know vmalloc() area is valid and can use memcpy().
1826 * This is for routines which have to access vmalloc area without
1827 * any informaion, as /dev/kmem.
1828 *
1829 * The caller should guarantee KM_USER1 is not used.
1830 */
1831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832long vwrite(char *buf, char *addr, unsigned long count)
1833{
1834 struct vm_struct *tmp;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001835 char *vaddr;
1836 unsigned long n, buflen;
1837 int copied = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
1839 /* Don't allow overflow */
1840 if ((unsigned long) addr + count < count)
1841 count = -(unsigned long) addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001842 buflen = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
1844 read_lock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001845 for (tmp = vmlist; count && tmp; tmp = tmp->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 vaddr = (char *) tmp->addr;
1847 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1848 continue;
1849 while (addr < vaddr) {
1850 if (count == 0)
1851 goto finished;
1852 buf++;
1853 addr++;
1854 count--;
1855 }
1856 n = vaddr + tmp->size - PAGE_SIZE - addr;
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001857 if (n > count)
1858 n = count;
1859 if (!(tmp->flags & VM_IOREMAP)) {
1860 aligned_vwrite(buf, addr, n);
1861 copied++;
1862 }
1863 buf += n;
1864 addr += n;
1865 count -= n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 }
1867finished:
1868 read_unlock(&vmlist_lock);
KAMEZAWA Hiroyukid0107eb2009-09-21 17:02:34 -07001869 if (!copied)
1870 return 0;
1871 return buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872}
Nick Piggin83342312006-06-23 02:03:20 -07001873
1874/**
1875 * remap_vmalloc_range - map vmalloc pages to userspace
Nick Piggin83342312006-06-23 02:03:20 -07001876 * @vma: vma to cover (map full range of vma)
1877 * @addr: vmalloc memory
1878 * @pgoff: number of pages into addr before first page to map
Randy Dunlap76824862008-03-19 17:00:40 -07001879 *
1880 * Returns: 0 for success, -Exxx on failure
Nick Piggin83342312006-06-23 02:03:20 -07001881 *
1882 * This function checks that addr is a valid vmalloc'ed area, and
1883 * that it is big enough to cover the vma. Will return failure if
1884 * that criteria isn't met.
1885 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001886 * Similar to remap_pfn_range() (see mm/memory.c)
Nick Piggin83342312006-06-23 02:03:20 -07001887 */
1888int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1889 unsigned long pgoff)
1890{
1891 struct vm_struct *area;
1892 unsigned long uaddr = vma->vm_start;
1893 unsigned long usize = vma->vm_end - vma->vm_start;
Nick Piggin83342312006-06-23 02:03:20 -07001894
1895 if ((PAGE_SIZE-1) & (unsigned long)addr)
1896 return -EINVAL;
1897
Nick Piggindb64fe02008-10-18 20:27:03 -07001898 area = find_vm_area(addr);
Nick Piggin83342312006-06-23 02:03:20 -07001899 if (!area)
Nick Piggindb64fe02008-10-18 20:27:03 -07001900 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001901
1902 if (!(area->flags & VM_USERMAP))
Nick Piggindb64fe02008-10-18 20:27:03 -07001903 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001904
1905 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
Nick Piggindb64fe02008-10-18 20:27:03 -07001906 return -EINVAL;
Nick Piggin83342312006-06-23 02:03:20 -07001907
1908 addr += pgoff << PAGE_SHIFT;
1909 do {
1910 struct page *page = vmalloc_to_page(addr);
Nick Piggindb64fe02008-10-18 20:27:03 -07001911 int ret;
1912
Nick Piggin83342312006-06-23 02:03:20 -07001913 ret = vm_insert_page(vma, uaddr, page);
1914 if (ret)
1915 return ret;
1916
1917 uaddr += PAGE_SIZE;
1918 addr += PAGE_SIZE;
1919 usize -= PAGE_SIZE;
1920 } while (usize > 0);
1921
1922 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1923 vma->vm_flags |= VM_RESERVED;
1924
Nick Piggindb64fe02008-10-18 20:27:03 -07001925 return 0;
Nick Piggin83342312006-06-23 02:03:20 -07001926}
1927EXPORT_SYMBOL(remap_vmalloc_range);
1928
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -07001929/*
1930 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1931 * have one.
1932 */
1933void __attribute__((weak)) vmalloc_sync_all(void)
1934{
1935}
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001936
1937
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001938static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001939{
1940 /* apply_to_page_range() does all the hard work. */
1941 return 0;
1942}
1943
1944/**
1945 * alloc_vm_area - allocate a range of kernel address space
1946 * @size: size of the area
Randy Dunlap76824862008-03-19 17:00:40 -07001947 *
1948 * Returns: NULL on failure, vm_struct on success
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001949 *
1950 * This function reserves a range of kernel address space, and
1951 * allocates pagetables to map that range. No actual mappings
1952 * are created. If the kernel address space is not shared
1953 * between processes, it syncs the pagetable across all
1954 * processes.
1955 */
1956struct vm_struct *alloc_vm_area(size_t size)
1957{
1958 struct vm_struct *area;
1959
Christoph Lameter23016962008-04-28 02:12:42 -07001960 area = get_vm_area_caller(size, VM_IOREMAP,
1961 __builtin_return_address(0));
Jeremy Fitzhardinge5f4352f2007-07-17 18:37:04 -07001962 if (area == NULL)
1963 return NULL;
1964
1965 /*
1966 * This ensures that page tables are constructed for this region
1967 * of kernel virtual address space and mapped into init_mm.
1968 */
1969 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1970 area->size, f, NULL)) {
1971 free_vm_area(area);
1972 return NULL;
1973 }
1974
1975 /* Make sure the pagetables are constructed in process kernel
1976 mappings */
1977 vmalloc_sync_all();
1978
1979 return area;
1980}
1981EXPORT_SYMBOL_GPL(alloc_vm_area);
1982
1983void free_vm_area(struct vm_struct *area)
1984{
1985 struct vm_struct *ret;
1986 ret = remove_vm_area(area->addr);
1987 BUG_ON(ret != area);
1988 kfree(area);
1989}
1990EXPORT_SYMBOL_GPL(free_vm_area);
Christoph Lametera10aa572008-04-28 02:12:40 -07001991
Tejun Heoca23e402009-08-14 15:00:52 +09001992static struct vmap_area *node_to_va(struct rb_node *n)
1993{
1994 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
1995}
1996
1997/**
1998 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
1999 * @end: target address
2000 * @pnext: out arg for the next vmap_area
2001 * @pprev: out arg for the previous vmap_area
2002 *
2003 * Returns: %true if either or both of next and prev are found,
2004 * %false if no vmap_area exists
2005 *
2006 * Find vmap_areas end addresses of which enclose @end. ie. if not
2007 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2008 */
2009static bool pvm_find_next_prev(unsigned long end,
2010 struct vmap_area **pnext,
2011 struct vmap_area **pprev)
2012{
2013 struct rb_node *n = vmap_area_root.rb_node;
2014 struct vmap_area *va = NULL;
2015
2016 while (n) {
2017 va = rb_entry(n, struct vmap_area, rb_node);
2018 if (end < va->va_end)
2019 n = n->rb_left;
2020 else if (end > va->va_end)
2021 n = n->rb_right;
2022 else
2023 break;
2024 }
2025
2026 if (!va)
2027 return false;
2028
2029 if (va->va_end > end) {
2030 *pnext = va;
2031 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2032 } else {
2033 *pprev = va;
2034 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2035 }
2036 return true;
2037}
2038
2039/**
2040 * pvm_determine_end - find the highest aligned address between two vmap_areas
2041 * @pnext: in/out arg for the next vmap_area
2042 * @pprev: in/out arg for the previous vmap_area
2043 * @align: alignment
2044 *
2045 * Returns: determined end address
2046 *
2047 * Find the highest aligned address between *@pnext and *@pprev below
2048 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2049 * down address is between the end addresses of the two vmap_areas.
2050 *
2051 * Please note that the address returned by this function may fall
2052 * inside *@pnext vmap_area. The caller is responsible for checking
2053 * that.
2054 */
2055static unsigned long pvm_determine_end(struct vmap_area **pnext,
2056 struct vmap_area **pprev,
2057 unsigned long align)
2058{
2059 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2060 unsigned long addr;
2061
2062 if (*pnext)
2063 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2064 else
2065 addr = vmalloc_end;
2066
2067 while (*pprev && (*pprev)->va_end > addr) {
2068 *pnext = *pprev;
2069 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2070 }
2071
2072 return addr;
2073}
2074
2075/**
2076 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2077 * @offsets: array containing offset of each area
2078 * @sizes: array containing size of each area
2079 * @nr_vms: the number of areas to allocate
2080 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2081 * @gfp_mask: allocation mask
2082 *
2083 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2084 * vm_structs on success, %NULL on failure
2085 *
2086 * Percpu allocator wants to use congruent vm areas so that it can
2087 * maintain the offsets among percpu areas. This function allocates
2088 * congruent vmalloc areas for it. These areas tend to be scattered
2089 * pretty far, distance between two areas easily going up to
2090 * gigabytes. To avoid interacting with regular vmallocs, these areas
2091 * are allocated from top.
2092 *
2093 * Despite its complicated look, this allocator is rather simple. It
2094 * does everything top-down and scans areas from the end looking for
2095 * matching slot. While scanning, if any of the areas overlaps with
2096 * existing vmap_area, the base address is pulled down to fit the
2097 * area. Scanning is repeated till all the areas fit and then all
2098 * necessary data structres are inserted and the result is returned.
2099 */
2100struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2101 const size_t *sizes, int nr_vms,
2102 size_t align, gfp_t gfp_mask)
2103{
2104 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2105 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2106 struct vmap_area **vas, *prev, *next;
2107 struct vm_struct **vms;
2108 int area, area2, last_area, term_area;
2109 unsigned long base, start, end, last_end;
2110 bool purged = false;
2111
2112 gfp_mask &= GFP_RECLAIM_MASK;
2113
2114 /* verify parameters and allocate data structures */
2115 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2116 for (last_area = 0, area = 0; area < nr_vms; area++) {
2117 start = offsets[area];
2118 end = start + sizes[area];
2119
2120 /* is everything aligned properly? */
2121 BUG_ON(!IS_ALIGNED(offsets[area], align));
2122 BUG_ON(!IS_ALIGNED(sizes[area], align));
2123
2124 /* detect the area with the highest address */
2125 if (start > offsets[last_area])
2126 last_area = area;
2127
2128 for (area2 = 0; area2 < nr_vms; area2++) {
2129 unsigned long start2 = offsets[area2];
2130 unsigned long end2 = start2 + sizes[area2];
2131
2132 if (area2 == area)
2133 continue;
2134
2135 BUG_ON(start2 >= start && start2 < end);
2136 BUG_ON(end2 <= end && end2 > start);
2137 }
2138 }
2139 last_end = offsets[last_area] + sizes[last_area];
2140
2141 if (vmalloc_end - vmalloc_start < last_end) {
2142 WARN_ON(true);
2143 return NULL;
2144 }
2145
2146 vms = kzalloc(sizeof(vms[0]) * nr_vms, gfp_mask);
2147 vas = kzalloc(sizeof(vas[0]) * nr_vms, gfp_mask);
2148 if (!vas || !vms)
2149 goto err_free;
2150
2151 for (area = 0; area < nr_vms; area++) {
2152 vas[area] = kzalloc(sizeof(struct vmap_area), gfp_mask);
2153 vms[area] = kzalloc(sizeof(struct vm_struct), gfp_mask);
2154 if (!vas[area] || !vms[area])
2155 goto err_free;
2156 }
2157retry:
2158 spin_lock(&vmap_area_lock);
2159
2160 /* start scanning - we scan from the top, begin with the last area */
2161 area = term_area = last_area;
2162 start = offsets[area];
2163 end = start + sizes[area];
2164
2165 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2166 base = vmalloc_end - last_end;
2167 goto found;
2168 }
2169 base = pvm_determine_end(&next, &prev, align) - end;
2170
2171 while (true) {
2172 BUG_ON(next && next->va_end <= base + end);
2173 BUG_ON(prev && prev->va_end > base + end);
2174
2175 /*
2176 * base might have underflowed, add last_end before
2177 * comparing.
2178 */
2179 if (base + last_end < vmalloc_start + last_end) {
2180 spin_unlock(&vmap_area_lock);
2181 if (!purged) {
2182 purge_vmap_area_lazy();
2183 purged = true;
2184 goto retry;
2185 }
2186 goto err_free;
2187 }
2188
2189 /*
2190 * If next overlaps, move base downwards so that it's
2191 * right below next and then recheck.
2192 */
2193 if (next && next->va_start < base + end) {
2194 base = pvm_determine_end(&next, &prev, align) - end;
2195 term_area = area;
2196 continue;
2197 }
2198
2199 /*
2200 * If prev overlaps, shift down next and prev and move
2201 * base so that it's right below new next and then
2202 * recheck.
2203 */
2204 if (prev && prev->va_end > base + start) {
2205 next = prev;
2206 prev = node_to_va(rb_prev(&next->rb_node));
2207 base = pvm_determine_end(&next, &prev, align) - end;
2208 term_area = area;
2209 continue;
2210 }
2211
2212 /*
2213 * This area fits, move on to the previous one. If
2214 * the previous one is the terminal one, we're done.
2215 */
2216 area = (area + nr_vms - 1) % nr_vms;
2217 if (area == term_area)
2218 break;
2219 start = offsets[area];
2220 end = start + sizes[area];
2221 pvm_find_next_prev(base + end, &next, &prev);
2222 }
2223found:
2224 /* we've found a fitting base, insert all va's */
2225 for (area = 0; area < nr_vms; area++) {
2226 struct vmap_area *va = vas[area];
2227
2228 va->va_start = base + offsets[area];
2229 va->va_end = va->va_start + sizes[area];
2230 __insert_vmap_area(va);
2231 }
2232
2233 vmap_area_pcpu_hole = base + offsets[last_area];
2234
2235 spin_unlock(&vmap_area_lock);
2236
2237 /* insert all vm's */
2238 for (area = 0; area < nr_vms; area++)
2239 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2240 pcpu_get_vm_areas);
2241
2242 kfree(vas);
2243 return vms;
2244
2245err_free:
2246 for (area = 0; area < nr_vms; area++) {
2247 if (vas)
2248 kfree(vas[area]);
2249 if (vms)
2250 kfree(vms[area]);
2251 }
2252 kfree(vas);
2253 kfree(vms);
2254 return NULL;
2255}
2256
2257/**
2258 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2259 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2260 * @nr_vms: the number of allocated areas
2261 *
2262 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2263 */
2264void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2265{
2266 int i;
2267
2268 for (i = 0; i < nr_vms; i++)
2269 free_vm_area(vms[i]);
2270 kfree(vms);
2271}
Christoph Lametera10aa572008-04-28 02:12:40 -07002272
2273#ifdef CONFIG_PROC_FS
2274static void *s_start(struct seq_file *m, loff_t *pos)
2275{
2276 loff_t n = *pos;
2277 struct vm_struct *v;
2278
2279 read_lock(&vmlist_lock);
2280 v = vmlist;
2281 while (n > 0 && v) {
2282 n--;
2283 v = v->next;
2284 }
2285 if (!n)
2286 return v;
2287
2288 return NULL;
2289
2290}
2291
2292static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2293{
2294 struct vm_struct *v = p;
2295
2296 ++*pos;
2297 return v->next;
2298}
2299
2300static void s_stop(struct seq_file *m, void *p)
2301{
2302 read_unlock(&vmlist_lock);
2303}
2304
Eric Dumazeta47a1262008-07-23 21:27:38 -07002305static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2306{
2307 if (NUMA_BUILD) {
2308 unsigned int nr, *counters = m->private;
2309
2310 if (!counters)
2311 return;
2312
2313 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2314
2315 for (nr = 0; nr < v->nr_pages; nr++)
2316 counters[page_to_nid(v->pages[nr])]++;
2317
2318 for_each_node_state(nr, N_HIGH_MEMORY)
2319 if (counters[nr])
2320 seq_printf(m, " N%u=%u", nr, counters[nr]);
2321 }
2322}
2323
Christoph Lametera10aa572008-04-28 02:12:40 -07002324static int s_show(struct seq_file *m, void *p)
2325{
2326 struct vm_struct *v = p;
2327
2328 seq_printf(m, "0x%p-0x%p %7ld",
2329 v->addr, v->addr + v->size, v->size);
2330
Christoph Lameter23016962008-04-28 02:12:42 -07002331 if (v->caller) {
Hugh Dickins9c246242008-12-09 13:14:27 -08002332 char buff[KSYM_SYMBOL_LEN];
Christoph Lameter23016962008-04-28 02:12:42 -07002333
2334 seq_putc(m, ' ');
2335 sprint_symbol(buff, (unsigned long)v->caller);
2336 seq_puts(m, buff);
2337 }
2338
Christoph Lametera10aa572008-04-28 02:12:40 -07002339 if (v->nr_pages)
2340 seq_printf(m, " pages=%d", v->nr_pages);
2341
2342 if (v->phys_addr)
2343 seq_printf(m, " phys=%lx", v->phys_addr);
2344
2345 if (v->flags & VM_IOREMAP)
2346 seq_printf(m, " ioremap");
2347
2348 if (v->flags & VM_ALLOC)
2349 seq_printf(m, " vmalloc");
2350
2351 if (v->flags & VM_MAP)
2352 seq_printf(m, " vmap");
2353
2354 if (v->flags & VM_USERMAP)
2355 seq_printf(m, " user");
2356
2357 if (v->flags & VM_VPAGES)
2358 seq_printf(m, " vpages");
2359
Eric Dumazeta47a1262008-07-23 21:27:38 -07002360 show_numa_info(m, v);
Christoph Lametera10aa572008-04-28 02:12:40 -07002361 seq_putc(m, '\n');
2362 return 0;
2363}
2364
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002365static const struct seq_operations vmalloc_op = {
Christoph Lametera10aa572008-04-28 02:12:40 -07002366 .start = s_start,
2367 .next = s_next,
2368 .stop = s_stop,
2369 .show = s_show,
2370};
Alexey Dobriyan5f6a6a92008-10-06 03:50:47 +04002371
2372static int vmalloc_open(struct inode *inode, struct file *file)
2373{
2374 unsigned int *ptr = NULL;
2375 int ret;
2376
2377 if (NUMA_BUILD)
2378 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
2379 ret = seq_open(file, &vmalloc_op);
2380 if (!ret) {
2381 struct seq_file *m = file->private_data;
2382 m->private = ptr;
2383 } else
2384 kfree(ptr);
2385 return ret;
2386}
2387
2388static const struct file_operations proc_vmalloc_operations = {
2389 .open = vmalloc_open,
2390 .read = seq_read,
2391 .llseek = seq_lseek,
2392 .release = seq_release_private,
2393};
2394
2395static int __init proc_vmalloc_init(void)
2396{
2397 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2398 return 0;
2399}
2400module_init(proc_vmalloc_init);
Christoph Lametera10aa572008-04-28 02:12:40 -07002401#endif
2402