]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/vmalloc.c | |
4 | * | |
5 | * Copyright (C) 1993 Linus Torvalds | |
6 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
7 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
8 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 9 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
10 | */ |
11 | ||
db64fe02 | 12 | #include <linux/vmalloc.h> |
1da177e4 LT |
13 | #include <linux/mm.h> |
14 | #include <linux/module.h> | |
15 | #include <linux/highmem.h> | |
c3edc401 | 16 | #include <linux/sched/signal.h> |
1da177e4 LT |
17 | #include <linux/slab.h> |
18 | #include <linux/spinlock.h> | |
19 | #include <linux/interrupt.h> | |
5f6a6a9c | 20 | #include <linux/proc_fs.h> |
a10aa579 | 21 | #include <linux/seq_file.h> |
868b104d | 22 | #include <linux/set_memory.h> |
3ac7fe5a | 23 | #include <linux/debugobjects.h> |
23016969 | 24 | #include <linux/kallsyms.h> |
db64fe02 | 25 | #include <linux/list.h> |
4da56b99 | 26 | #include <linux/notifier.h> |
db64fe02 NP |
27 | #include <linux/rbtree.h> |
28 | #include <linux/radix-tree.h> | |
29 | #include <linux/rcupdate.h> | |
f0aa6617 | 30 | #include <linux/pfn.h> |
89219d37 | 31 | #include <linux/kmemleak.h> |
60063497 | 32 | #include <linux/atomic.h> |
3b32123d | 33 | #include <linux/compiler.h> |
32fcfd40 | 34 | #include <linux/llist.h> |
0f616be1 | 35 | #include <linux/bitops.h> |
68ad4a33 | 36 | #include <linux/rbtree_augmented.h> |
3b32123d | 37 | |
7c0f6ba6 | 38 | #include <linux/uaccess.h> |
1da177e4 | 39 | #include <asm/tlbflush.h> |
2dca6999 | 40 | #include <asm/shmparam.h> |
1da177e4 | 41 | |
dd56b046 MG |
42 | #include "internal.h" |
43 | ||
32fcfd40 AV |
44 | struct vfree_deferred { |
45 | struct llist_head list; | |
46 | struct work_struct wq; | |
47 | }; | |
48 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
49 | ||
50 | static void __vunmap(const void *, int); | |
51 | ||
52 | static void free_work(struct work_struct *w) | |
53 | { | |
54 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
55 | struct llist_node *t, *llnode; |
56 | ||
57 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
58 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
59 | } |
60 | ||
db64fe02 | 61 | /*** Page table manipulation functions ***/ |
b221385b | 62 | |
1da177e4 LT |
63 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
64 | { | |
65 | pte_t *pte; | |
66 | ||
67 | pte = pte_offset_kernel(pmd, addr); | |
68 | do { | |
69 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
70 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
71 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
72 | } | |
73 | ||
db64fe02 | 74 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
75 | { |
76 | pmd_t *pmd; | |
77 | unsigned long next; | |
78 | ||
79 | pmd = pmd_offset(pud, addr); | |
80 | do { | |
81 | next = pmd_addr_end(addr, end); | |
b9820d8f TK |
82 | if (pmd_clear_huge(pmd)) |
83 | continue; | |
1da177e4 LT |
84 | if (pmd_none_or_clear_bad(pmd)) |
85 | continue; | |
86 | vunmap_pte_range(pmd, addr, next); | |
87 | } while (pmd++, addr = next, addr != end); | |
88 | } | |
89 | ||
c2febafc | 90 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end) |
1da177e4 LT |
91 | { |
92 | pud_t *pud; | |
93 | unsigned long next; | |
94 | ||
c2febafc | 95 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
96 | do { |
97 | next = pud_addr_end(addr, end); | |
b9820d8f TK |
98 | if (pud_clear_huge(pud)) |
99 | continue; | |
1da177e4 LT |
100 | if (pud_none_or_clear_bad(pud)) |
101 | continue; | |
102 | vunmap_pmd_range(pud, addr, next); | |
103 | } while (pud++, addr = next, addr != end); | |
104 | } | |
105 | ||
c2febafc KS |
106 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
107 | { | |
108 | p4d_t *p4d; | |
109 | unsigned long next; | |
110 | ||
111 | p4d = p4d_offset(pgd, addr); | |
112 | do { | |
113 | next = p4d_addr_end(addr, end); | |
114 | if (p4d_clear_huge(p4d)) | |
115 | continue; | |
116 | if (p4d_none_or_clear_bad(p4d)) | |
117 | continue; | |
118 | vunmap_pud_range(p4d, addr, next); | |
119 | } while (p4d++, addr = next, addr != end); | |
120 | } | |
121 | ||
db64fe02 | 122 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
123 | { |
124 | pgd_t *pgd; | |
125 | unsigned long next; | |
1da177e4 LT |
126 | |
127 | BUG_ON(addr >= end); | |
128 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
129 | do { |
130 | next = pgd_addr_end(addr, end); | |
131 | if (pgd_none_or_clear_bad(pgd)) | |
132 | continue; | |
c2febafc | 133 | vunmap_p4d_range(pgd, addr, next); |
1da177e4 | 134 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
135 | } |
136 | ||
137 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 138 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
139 | { |
140 | pte_t *pte; | |
141 | ||
db64fe02 NP |
142 | /* |
143 | * nr is a running index into the array which helps higher level | |
144 | * callers keep track of where we're up to. | |
145 | */ | |
146 | ||
872fec16 | 147 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
148 | if (!pte) |
149 | return -ENOMEM; | |
150 | do { | |
db64fe02 NP |
151 | struct page *page = pages[*nr]; |
152 | ||
153 | if (WARN_ON(!pte_none(*pte))) | |
154 | return -EBUSY; | |
155 | if (WARN_ON(!page)) | |
1da177e4 LT |
156 | return -ENOMEM; |
157 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 158 | (*nr)++; |
1da177e4 LT |
159 | } while (pte++, addr += PAGE_SIZE, addr != end); |
160 | return 0; | |
161 | } | |
162 | ||
db64fe02 NP |
163 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
164 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
165 | { |
166 | pmd_t *pmd; | |
167 | unsigned long next; | |
168 | ||
169 | pmd = pmd_alloc(&init_mm, pud, addr); | |
170 | if (!pmd) | |
171 | return -ENOMEM; | |
172 | do { | |
173 | next = pmd_addr_end(addr, end); | |
db64fe02 | 174 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
175 | return -ENOMEM; |
176 | } while (pmd++, addr = next, addr != end); | |
177 | return 0; | |
178 | } | |
179 | ||
c2febafc | 180 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, |
db64fe02 | 181 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
182 | { |
183 | pud_t *pud; | |
184 | unsigned long next; | |
185 | ||
c2febafc | 186 | pud = pud_alloc(&init_mm, p4d, addr); |
1da177e4 LT |
187 | if (!pud) |
188 | return -ENOMEM; | |
189 | do { | |
190 | next = pud_addr_end(addr, end); | |
db64fe02 | 191 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
192 | return -ENOMEM; |
193 | } while (pud++, addr = next, addr != end); | |
194 | return 0; | |
195 | } | |
196 | ||
c2febafc KS |
197 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, |
198 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
199 | { | |
200 | p4d_t *p4d; | |
201 | unsigned long next; | |
202 | ||
203 | p4d = p4d_alloc(&init_mm, pgd, addr); | |
204 | if (!p4d) | |
205 | return -ENOMEM; | |
206 | do { | |
207 | next = p4d_addr_end(addr, end); | |
208 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr)) | |
209 | return -ENOMEM; | |
210 | } while (p4d++, addr = next, addr != end); | |
211 | return 0; | |
212 | } | |
213 | ||
db64fe02 NP |
214 | /* |
215 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
216 | * will have pfns corresponding to the "pages" array. | |
217 | * | |
218 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
219 | */ | |
8fc48985 TH |
220 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
221 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
222 | { |
223 | pgd_t *pgd; | |
224 | unsigned long next; | |
2e4e27c7 | 225 | unsigned long addr = start; |
db64fe02 NP |
226 | int err = 0; |
227 | int nr = 0; | |
1da177e4 LT |
228 | |
229 | BUG_ON(addr >= end); | |
230 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
231 | do { |
232 | next = pgd_addr_end(addr, end); | |
c2febafc | 233 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 234 | if (err) |
bf88c8c8 | 235 | return err; |
1da177e4 | 236 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 237 | |
db64fe02 | 238 | return nr; |
1da177e4 LT |
239 | } |
240 | ||
8fc48985 TH |
241 | static int vmap_page_range(unsigned long start, unsigned long end, |
242 | pgprot_t prot, struct page **pages) | |
243 | { | |
244 | int ret; | |
245 | ||
246 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
247 | flush_cache_vmap(start, end); | |
248 | return ret; | |
249 | } | |
250 | ||
81ac3ad9 | 251 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
252 | { |
253 | /* | |
ab4f2ee1 | 254 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
255 | * and fall back on vmalloc() if that fails. Others |
256 | * just put it in the vmalloc space. | |
257 | */ | |
258 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
259 | unsigned long addr = (unsigned long)x; | |
260 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
261 | return 1; | |
262 | #endif | |
263 | return is_vmalloc_addr(x); | |
264 | } | |
265 | ||
48667e7a | 266 | /* |
add688fb | 267 | * Walk a vmap address to the struct page it maps. |
48667e7a | 268 | */ |
add688fb | 269 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
270 | { |
271 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 272 | struct page *page = NULL; |
48667e7a | 273 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
274 | p4d_t *p4d; |
275 | pud_t *pud; | |
276 | pmd_t *pmd; | |
277 | pte_t *ptep, pte; | |
48667e7a | 278 | |
7aa413de IM |
279 | /* |
280 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
281 | * architectures that do not vmalloc module space | |
282 | */ | |
73bdf0a6 | 283 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 284 | |
c2febafc KS |
285 | if (pgd_none(*pgd)) |
286 | return NULL; | |
287 | p4d = p4d_offset(pgd, addr); | |
288 | if (p4d_none(*p4d)) | |
289 | return NULL; | |
290 | pud = pud_offset(p4d, addr); | |
029c54b0 AB |
291 | |
292 | /* | |
293 | * Don't dereference bad PUD or PMD (below) entries. This will also | |
294 | * identify huge mappings, which we may encounter on architectures | |
295 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be | |
296 | * identified as vmalloc addresses by is_vmalloc_addr(), but are | |
297 | * not [unambiguously] associated with a struct page, so there is | |
298 | * no correct value to return for them. | |
299 | */ | |
300 | WARN_ON_ONCE(pud_bad(*pud)); | |
301 | if (pud_none(*pud) || pud_bad(*pud)) | |
c2febafc KS |
302 | return NULL; |
303 | pmd = pmd_offset(pud, addr); | |
029c54b0 AB |
304 | WARN_ON_ONCE(pmd_bad(*pmd)); |
305 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
c2febafc KS |
306 | return NULL; |
307 | ||
308 | ptep = pte_offset_map(pmd, addr); | |
309 | pte = *ptep; | |
310 | if (pte_present(pte)) | |
311 | page = pte_page(pte); | |
312 | pte_unmap(ptep); | |
add688fb | 313 | return page; |
48667e7a | 314 | } |
add688fb | 315 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
316 | |
317 | /* | |
add688fb | 318 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 319 | */ |
add688fb | 320 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 321 | { |
add688fb | 322 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 323 | } |
add688fb | 324 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 325 | |
db64fe02 NP |
326 | |
327 | /*** Global kva allocator ***/ | |
328 | ||
bb850f4d | 329 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 330 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 331 | |
db64fe02 | 332 | |
db64fe02 | 333 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
334 | /* Export for kexec only */ |
335 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 336 | static LLIST_HEAD(vmap_purge_list); |
89699605 | 337 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 338 | static bool vmap_initialized __read_mostly; |
89699605 | 339 | |
68ad4a33 URS |
340 | /* |
341 | * This kmem_cache is used for vmap_area objects. Instead of | |
342 | * allocating from slab we reuse an object from this cache to | |
343 | * make things faster. Especially in "no edge" splitting of | |
344 | * free block. | |
345 | */ | |
346 | static struct kmem_cache *vmap_area_cachep; | |
347 | ||
348 | /* | |
349 | * This linked list is used in pair with free_vmap_area_root. | |
350 | * It gives O(1) access to prev/next to perform fast coalescing. | |
351 | */ | |
352 | static LIST_HEAD(free_vmap_area_list); | |
353 | ||
354 | /* | |
355 | * This augment red-black tree represents the free vmap space. | |
356 | * All vmap_area objects in this tree are sorted by va->va_start | |
357 | * address. It is used for allocation and merging when a vmap | |
358 | * object is released. | |
359 | * | |
360 | * Each vmap_area node contains a maximum available free block | |
361 | * of its sub-tree, right or left. Therefore it is possible to | |
362 | * find a lowest match of free area. | |
363 | */ | |
364 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
365 | ||
82dd23e8 URS |
366 | /* |
367 | * Preload a CPU with one object for "no edge" split case. The | |
368 | * aim is to get rid of allocations from the atomic context, thus | |
369 | * to use more permissive allocation masks. | |
370 | */ | |
371 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
372 | ||
68ad4a33 URS |
373 | static __always_inline unsigned long |
374 | va_size(struct vmap_area *va) | |
375 | { | |
376 | return (va->va_end - va->va_start); | |
377 | } | |
378 | ||
379 | static __always_inline unsigned long | |
380 | get_subtree_max_size(struct rb_node *node) | |
381 | { | |
382 | struct vmap_area *va; | |
383 | ||
384 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
385 | return va ? va->subtree_max_size : 0; | |
386 | } | |
89699605 | 387 | |
68ad4a33 URS |
388 | /* |
389 | * Gets called when remove the node and rotate. | |
390 | */ | |
391 | static __always_inline unsigned long | |
392 | compute_subtree_max_size(struct vmap_area *va) | |
393 | { | |
394 | return max3(va_size(va), | |
395 | get_subtree_max_size(va->rb_node.rb_left), | |
396 | get_subtree_max_size(va->rb_node.rb_right)); | |
397 | } | |
398 | ||
399 | RB_DECLARE_CALLBACKS(static, free_vmap_area_rb_augment_cb, | |
400 | struct vmap_area, rb_node, unsigned long, subtree_max_size, | |
401 | compute_subtree_max_size) | |
402 | ||
403 | static void purge_vmap_area_lazy(void); | |
404 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
405 | static unsigned long lazy_max_pages(void); | |
db64fe02 | 406 | |
97105f0a RG |
407 | static atomic_long_t nr_vmalloc_pages; |
408 | ||
409 | unsigned long vmalloc_nr_pages(void) | |
410 | { | |
411 | return atomic_long_read(&nr_vmalloc_pages); | |
412 | } | |
413 | ||
db64fe02 | 414 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 415 | { |
db64fe02 NP |
416 | struct rb_node *n = vmap_area_root.rb_node; |
417 | ||
418 | while (n) { | |
419 | struct vmap_area *va; | |
420 | ||
421 | va = rb_entry(n, struct vmap_area, rb_node); | |
422 | if (addr < va->va_start) | |
423 | n = n->rb_left; | |
cef2ac3f | 424 | else if (addr >= va->va_end) |
db64fe02 NP |
425 | n = n->rb_right; |
426 | else | |
427 | return va; | |
428 | } | |
429 | ||
430 | return NULL; | |
431 | } | |
432 | ||
68ad4a33 URS |
433 | /* |
434 | * This function returns back addresses of parent node | |
435 | * and its left or right link for further processing. | |
436 | */ | |
437 | static __always_inline struct rb_node ** | |
438 | find_va_links(struct vmap_area *va, | |
439 | struct rb_root *root, struct rb_node *from, | |
440 | struct rb_node **parent) | |
441 | { | |
442 | struct vmap_area *tmp_va; | |
443 | struct rb_node **link; | |
444 | ||
445 | if (root) { | |
446 | link = &root->rb_node; | |
447 | if (unlikely(!*link)) { | |
448 | *parent = NULL; | |
449 | return link; | |
450 | } | |
451 | } else { | |
452 | link = &from; | |
453 | } | |
db64fe02 | 454 | |
68ad4a33 URS |
455 | /* |
456 | * Go to the bottom of the tree. When we hit the last point | |
457 | * we end up with parent rb_node and correct direction, i name | |
458 | * it link, where the new va->rb_node will be attached to. | |
459 | */ | |
460 | do { | |
461 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 462 | |
68ad4a33 URS |
463 | /* |
464 | * During the traversal we also do some sanity check. | |
465 | * Trigger the BUG() if there are sides(left/right) | |
466 | * or full overlaps. | |
467 | */ | |
468 | if (va->va_start < tmp_va->va_end && | |
469 | va->va_end <= tmp_va->va_start) | |
470 | link = &(*link)->rb_left; | |
471 | else if (va->va_end > tmp_va->va_start && | |
472 | va->va_start >= tmp_va->va_end) | |
473 | link = &(*link)->rb_right; | |
db64fe02 NP |
474 | else |
475 | BUG(); | |
68ad4a33 URS |
476 | } while (*link); |
477 | ||
478 | *parent = &tmp_va->rb_node; | |
479 | return link; | |
480 | } | |
481 | ||
482 | static __always_inline struct list_head * | |
483 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
484 | { | |
485 | struct list_head *list; | |
486 | ||
487 | if (unlikely(!parent)) | |
488 | /* | |
489 | * The red-black tree where we try to find VA neighbors | |
490 | * before merging or inserting is empty, i.e. it means | |
491 | * there is no free vmap space. Normally it does not | |
492 | * happen but we handle this case anyway. | |
493 | */ | |
494 | return NULL; | |
495 | ||
496 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
497 | return (&parent->rb_right == link ? list->next : list); | |
498 | } | |
499 | ||
500 | static __always_inline void | |
501 | link_va(struct vmap_area *va, struct rb_root *root, | |
502 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
503 | { | |
504 | /* | |
505 | * VA is still not in the list, but we can | |
506 | * identify its future previous list_head node. | |
507 | */ | |
508 | if (likely(parent)) { | |
509 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
510 | if (&parent->rb_right != link) | |
511 | head = head->prev; | |
db64fe02 NP |
512 | } |
513 | ||
68ad4a33 URS |
514 | /* Insert to the rb-tree */ |
515 | rb_link_node(&va->rb_node, parent, link); | |
516 | if (root == &free_vmap_area_root) { | |
517 | /* | |
518 | * Some explanation here. Just perform simple insertion | |
519 | * to the tree. We do not set va->subtree_max_size to | |
520 | * its current size before calling rb_insert_augmented(). | |
521 | * It is because of we populate the tree from the bottom | |
522 | * to parent levels when the node _is_ in the tree. | |
523 | * | |
524 | * Therefore we set subtree_max_size to zero after insertion, | |
525 | * to let __augment_tree_propagate_from() puts everything to | |
526 | * the correct order later on. | |
527 | */ | |
528 | rb_insert_augmented(&va->rb_node, | |
529 | root, &free_vmap_area_rb_augment_cb); | |
530 | va->subtree_max_size = 0; | |
531 | } else { | |
532 | rb_insert_color(&va->rb_node, root); | |
533 | } | |
db64fe02 | 534 | |
68ad4a33 URS |
535 | /* Address-sort this list */ |
536 | list_add(&va->list, head); | |
db64fe02 NP |
537 | } |
538 | ||
68ad4a33 URS |
539 | static __always_inline void |
540 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
541 | { | |
460e42d1 URS |
542 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
543 | return; | |
db64fe02 | 544 | |
460e42d1 URS |
545 | if (root == &free_vmap_area_root) |
546 | rb_erase_augmented(&va->rb_node, | |
547 | root, &free_vmap_area_rb_augment_cb); | |
548 | else | |
549 | rb_erase(&va->rb_node, root); | |
550 | ||
551 | list_del(&va->list); | |
552 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
553 | } |
554 | ||
bb850f4d URS |
555 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
556 | static void | |
557 | augment_tree_propagate_check(struct rb_node *n) | |
558 | { | |
559 | struct vmap_area *va; | |
560 | struct rb_node *node; | |
561 | unsigned long size; | |
562 | bool found = false; | |
563 | ||
564 | if (n == NULL) | |
565 | return; | |
566 | ||
567 | va = rb_entry(n, struct vmap_area, rb_node); | |
568 | size = va->subtree_max_size; | |
569 | node = n; | |
570 | ||
571 | while (node) { | |
572 | va = rb_entry(node, struct vmap_area, rb_node); | |
573 | ||
574 | if (get_subtree_max_size(node->rb_left) == size) { | |
575 | node = node->rb_left; | |
576 | } else { | |
577 | if (va_size(va) == size) { | |
578 | found = true; | |
579 | break; | |
580 | } | |
581 | ||
582 | node = node->rb_right; | |
583 | } | |
584 | } | |
585 | ||
586 | if (!found) { | |
587 | va = rb_entry(n, struct vmap_area, rb_node); | |
588 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
589 | va_size(va), va->subtree_max_size); | |
590 | } | |
591 | ||
592 | augment_tree_propagate_check(n->rb_left); | |
593 | augment_tree_propagate_check(n->rb_right); | |
594 | } | |
595 | #endif | |
596 | ||
68ad4a33 URS |
597 | /* |
598 | * This function populates subtree_max_size from bottom to upper | |
599 | * levels starting from VA point. The propagation must be done | |
600 | * when VA size is modified by changing its va_start/va_end. Or | |
601 | * in case of newly inserting of VA to the tree. | |
602 | * | |
603 | * It means that __augment_tree_propagate_from() must be called: | |
604 | * - After VA has been inserted to the tree(free path); | |
605 | * - After VA has been shrunk(allocation path); | |
606 | * - After VA has been increased(merging path). | |
607 | * | |
608 | * Please note that, it does not mean that upper parent nodes | |
609 | * and their subtree_max_size are recalculated all the time up | |
610 | * to the root node. | |
611 | * | |
612 | * 4--8 | |
613 | * /\ | |
614 | * / \ | |
615 | * / \ | |
616 | * 2--2 8--8 | |
617 | * | |
618 | * For example if we modify the node 4, shrinking it to 2, then | |
619 | * no any modification is required. If we shrink the node 2 to 1 | |
620 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
621 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
622 | * node becomes 4--6. | |
623 | */ | |
624 | static __always_inline void | |
625 | augment_tree_propagate_from(struct vmap_area *va) | |
626 | { | |
627 | struct rb_node *node = &va->rb_node; | |
628 | unsigned long new_va_sub_max_size; | |
629 | ||
630 | while (node) { | |
631 | va = rb_entry(node, struct vmap_area, rb_node); | |
632 | new_va_sub_max_size = compute_subtree_max_size(va); | |
633 | ||
634 | /* | |
635 | * If the newly calculated maximum available size of the | |
636 | * subtree is equal to the current one, then it means that | |
637 | * the tree is propagated correctly. So we have to stop at | |
638 | * this point to save cycles. | |
639 | */ | |
640 | if (va->subtree_max_size == new_va_sub_max_size) | |
641 | break; | |
642 | ||
643 | va->subtree_max_size = new_va_sub_max_size; | |
644 | node = rb_parent(&va->rb_node); | |
645 | } | |
bb850f4d URS |
646 | |
647 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
648 | augment_tree_propagate_check(free_vmap_area_root.rb_node); | |
649 | #endif | |
68ad4a33 URS |
650 | } |
651 | ||
652 | static void | |
653 | insert_vmap_area(struct vmap_area *va, | |
654 | struct rb_root *root, struct list_head *head) | |
655 | { | |
656 | struct rb_node **link; | |
657 | struct rb_node *parent; | |
658 | ||
659 | link = find_va_links(va, root, NULL, &parent); | |
660 | link_va(va, root, parent, link, head); | |
661 | } | |
662 | ||
663 | static void | |
664 | insert_vmap_area_augment(struct vmap_area *va, | |
665 | struct rb_node *from, struct rb_root *root, | |
666 | struct list_head *head) | |
667 | { | |
668 | struct rb_node **link; | |
669 | struct rb_node *parent; | |
670 | ||
671 | if (from) | |
672 | link = find_va_links(va, NULL, from, &parent); | |
673 | else | |
674 | link = find_va_links(va, root, NULL, &parent); | |
675 | ||
676 | link_va(va, root, parent, link, head); | |
677 | augment_tree_propagate_from(va); | |
678 | } | |
679 | ||
680 | /* | |
681 | * Merge de-allocated chunk of VA memory with previous | |
682 | * and next free blocks. If coalesce is not done a new | |
683 | * free area is inserted. If VA has been merged, it is | |
684 | * freed. | |
685 | */ | |
686 | static __always_inline void | |
687 | merge_or_add_vmap_area(struct vmap_area *va, | |
688 | struct rb_root *root, struct list_head *head) | |
689 | { | |
690 | struct vmap_area *sibling; | |
691 | struct list_head *next; | |
692 | struct rb_node **link; | |
693 | struct rb_node *parent; | |
694 | bool merged = false; | |
695 | ||
696 | /* | |
697 | * Find a place in the tree where VA potentially will be | |
698 | * inserted, unless it is merged with its sibling/siblings. | |
699 | */ | |
700 | link = find_va_links(va, root, NULL, &parent); | |
701 | ||
702 | /* | |
703 | * Get next node of VA to check if merging can be done. | |
704 | */ | |
705 | next = get_va_next_sibling(parent, link); | |
706 | if (unlikely(next == NULL)) | |
707 | goto insert; | |
708 | ||
709 | /* | |
710 | * start end | |
711 | * | | | |
712 | * |<------VA------>|<-----Next----->| | |
713 | * | | | |
714 | * start end | |
715 | */ | |
716 | if (next != head) { | |
717 | sibling = list_entry(next, struct vmap_area, list); | |
718 | if (sibling->va_start == va->va_end) { | |
719 | sibling->va_start = va->va_start; | |
720 | ||
721 | /* Check and update the tree if needed. */ | |
722 | augment_tree_propagate_from(sibling); | |
723 | ||
68ad4a33 URS |
724 | /* Free vmap_area object. */ |
725 | kmem_cache_free(vmap_area_cachep, va); | |
726 | ||
727 | /* Point to the new merged area. */ | |
728 | va = sibling; | |
729 | merged = true; | |
730 | } | |
731 | } | |
732 | ||
733 | /* | |
734 | * start end | |
735 | * | | | |
736 | * |<-----Prev----->|<------VA------>| | |
737 | * | | | |
738 | * start end | |
739 | */ | |
740 | if (next->prev != head) { | |
741 | sibling = list_entry(next->prev, struct vmap_area, list); | |
742 | if (sibling->va_end == va->va_start) { | |
743 | sibling->va_end = va->va_end; | |
744 | ||
745 | /* Check and update the tree if needed. */ | |
746 | augment_tree_propagate_from(sibling); | |
747 | ||
54f63d9d URS |
748 | if (merged) |
749 | unlink_va(va, root); | |
68ad4a33 URS |
750 | |
751 | /* Free vmap_area object. */ | |
752 | kmem_cache_free(vmap_area_cachep, va); | |
68ad4a33 URS |
753 | return; |
754 | } | |
755 | } | |
756 | ||
757 | insert: | |
758 | if (!merged) { | |
759 | link_va(va, root, parent, link, head); | |
760 | augment_tree_propagate_from(va); | |
761 | } | |
762 | } | |
763 | ||
764 | static __always_inline bool | |
765 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
766 | unsigned long align, unsigned long vstart) | |
767 | { | |
768 | unsigned long nva_start_addr; | |
769 | ||
770 | if (va->va_start > vstart) | |
771 | nva_start_addr = ALIGN(va->va_start, align); | |
772 | else | |
773 | nva_start_addr = ALIGN(vstart, align); | |
774 | ||
775 | /* Can be overflowed due to big size or alignment. */ | |
776 | if (nva_start_addr + size < nva_start_addr || | |
777 | nva_start_addr < vstart) | |
778 | return false; | |
779 | ||
780 | return (nva_start_addr + size <= va->va_end); | |
781 | } | |
782 | ||
783 | /* | |
784 | * Find the first free block(lowest start address) in the tree, | |
785 | * that will accomplish the request corresponding to passing | |
786 | * parameters. | |
787 | */ | |
788 | static __always_inline struct vmap_area * | |
789 | find_vmap_lowest_match(unsigned long size, | |
790 | unsigned long align, unsigned long vstart) | |
791 | { | |
792 | struct vmap_area *va; | |
793 | struct rb_node *node; | |
794 | unsigned long length; | |
795 | ||
796 | /* Start from the root. */ | |
797 | node = free_vmap_area_root.rb_node; | |
798 | ||
799 | /* Adjust the search size for alignment overhead. */ | |
800 | length = size + align - 1; | |
801 | ||
802 | while (node) { | |
803 | va = rb_entry(node, struct vmap_area, rb_node); | |
804 | ||
805 | if (get_subtree_max_size(node->rb_left) >= length && | |
806 | vstart < va->va_start) { | |
807 | node = node->rb_left; | |
808 | } else { | |
809 | if (is_within_this_va(va, size, align, vstart)) | |
810 | return va; | |
811 | ||
812 | /* | |
813 | * Does not make sense to go deeper towards the right | |
814 | * sub-tree if it does not have a free block that is | |
815 | * equal or bigger to the requested search length. | |
816 | */ | |
817 | if (get_subtree_max_size(node->rb_right) >= length) { | |
818 | node = node->rb_right; | |
819 | continue; | |
820 | } | |
821 | ||
822 | /* | |
3806b041 | 823 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
824 | * that will satisfy the search criteria. It can happen |
825 | * only once due to "vstart" restriction. | |
826 | */ | |
827 | while ((node = rb_parent(node))) { | |
828 | va = rb_entry(node, struct vmap_area, rb_node); | |
829 | if (is_within_this_va(va, size, align, vstart)) | |
830 | return va; | |
831 | ||
832 | if (get_subtree_max_size(node->rb_right) >= length && | |
833 | vstart <= va->va_start) { | |
834 | node = node->rb_right; | |
835 | break; | |
836 | } | |
837 | } | |
838 | } | |
839 | } | |
840 | ||
841 | return NULL; | |
842 | } | |
843 | ||
a6cf4e0f URS |
844 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
845 | #include <linux/random.h> | |
846 | ||
847 | static struct vmap_area * | |
848 | find_vmap_lowest_linear_match(unsigned long size, | |
849 | unsigned long align, unsigned long vstart) | |
850 | { | |
851 | struct vmap_area *va; | |
852 | ||
853 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
854 | if (!is_within_this_va(va, size, align, vstart)) | |
855 | continue; | |
856 | ||
857 | return va; | |
858 | } | |
859 | ||
860 | return NULL; | |
861 | } | |
862 | ||
863 | static void | |
864 | find_vmap_lowest_match_check(unsigned long size) | |
865 | { | |
866 | struct vmap_area *va_1, *va_2; | |
867 | unsigned long vstart; | |
868 | unsigned int rnd; | |
869 | ||
870 | get_random_bytes(&rnd, sizeof(rnd)); | |
871 | vstart = VMALLOC_START + rnd; | |
872 | ||
873 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
874 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
875 | ||
876 | if (va_1 != va_2) | |
877 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
878 | va_1, va_2, vstart); | |
879 | } | |
880 | #endif | |
881 | ||
68ad4a33 URS |
882 | enum fit_type { |
883 | NOTHING_FIT = 0, | |
884 | FL_FIT_TYPE = 1, /* full fit */ | |
885 | LE_FIT_TYPE = 2, /* left edge fit */ | |
886 | RE_FIT_TYPE = 3, /* right edge fit */ | |
887 | NE_FIT_TYPE = 4 /* no edge fit */ | |
888 | }; | |
889 | ||
890 | static __always_inline enum fit_type | |
891 | classify_va_fit_type(struct vmap_area *va, | |
892 | unsigned long nva_start_addr, unsigned long size) | |
893 | { | |
894 | enum fit_type type; | |
895 | ||
896 | /* Check if it is within VA. */ | |
897 | if (nva_start_addr < va->va_start || | |
898 | nva_start_addr + size > va->va_end) | |
899 | return NOTHING_FIT; | |
900 | ||
901 | /* Now classify. */ | |
902 | if (va->va_start == nva_start_addr) { | |
903 | if (va->va_end == nva_start_addr + size) | |
904 | type = FL_FIT_TYPE; | |
905 | else | |
906 | type = LE_FIT_TYPE; | |
907 | } else if (va->va_end == nva_start_addr + size) { | |
908 | type = RE_FIT_TYPE; | |
909 | } else { | |
910 | type = NE_FIT_TYPE; | |
911 | } | |
912 | ||
913 | return type; | |
914 | } | |
915 | ||
916 | static __always_inline int | |
917 | adjust_va_to_fit_type(struct vmap_area *va, | |
918 | unsigned long nva_start_addr, unsigned long size, | |
919 | enum fit_type type) | |
920 | { | |
2c929233 | 921 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
922 | |
923 | if (type == FL_FIT_TYPE) { | |
924 | /* | |
925 | * No need to split VA, it fully fits. | |
926 | * | |
927 | * | | | |
928 | * V NVA V | |
929 | * |---------------| | |
930 | */ | |
931 | unlink_va(va, &free_vmap_area_root); | |
932 | kmem_cache_free(vmap_area_cachep, va); | |
933 | } else if (type == LE_FIT_TYPE) { | |
934 | /* | |
935 | * Split left edge of fit VA. | |
936 | * | |
937 | * | | | |
938 | * V NVA V R | |
939 | * |-------|-------| | |
940 | */ | |
941 | va->va_start += size; | |
942 | } else if (type == RE_FIT_TYPE) { | |
943 | /* | |
944 | * Split right edge of fit VA. | |
945 | * | |
946 | * | | | |
947 | * L V NVA V | |
948 | * |-------|-------| | |
949 | */ | |
950 | va->va_end = nva_start_addr; | |
951 | } else if (type == NE_FIT_TYPE) { | |
952 | /* | |
953 | * Split no edge of fit VA. | |
954 | * | |
955 | * | | | |
956 | * L V NVA V R | |
957 | * |---|-------|---| | |
958 | */ | |
82dd23e8 URS |
959 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
960 | if (unlikely(!lva)) { | |
961 | /* | |
962 | * For percpu allocator we do not do any pre-allocation | |
963 | * and leave it as it is. The reason is it most likely | |
964 | * never ends up with NE_FIT_TYPE splitting. In case of | |
965 | * percpu allocations offsets and sizes are aligned to | |
966 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
967 | * are its main fitting cases. | |
968 | * | |
969 | * There are a few exceptions though, as an example it is | |
970 | * a first allocation (early boot up) when we have "one" | |
971 | * big free space that has to be split. | |
972 | */ | |
973 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
974 | if (!lva) | |
975 | return -1; | |
976 | } | |
68ad4a33 URS |
977 | |
978 | /* | |
979 | * Build the remainder. | |
980 | */ | |
981 | lva->va_start = va->va_start; | |
982 | lva->va_end = nva_start_addr; | |
983 | ||
984 | /* | |
985 | * Shrink this VA to remaining size. | |
986 | */ | |
987 | va->va_start = nva_start_addr + size; | |
988 | } else { | |
989 | return -1; | |
990 | } | |
991 | ||
992 | if (type != FL_FIT_TYPE) { | |
993 | augment_tree_propagate_from(va); | |
994 | ||
2c929233 | 995 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
996 | insert_vmap_area_augment(lva, &va->rb_node, |
997 | &free_vmap_area_root, &free_vmap_area_list); | |
998 | } | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Returns a start address of the newly allocated area, if success. | |
1005 | * Otherwise a vend is returned that indicates failure. | |
1006 | */ | |
1007 | static __always_inline unsigned long | |
1008 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1009 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1010 | { |
1011 | unsigned long nva_start_addr; | |
1012 | struct vmap_area *va; | |
1013 | enum fit_type type; | |
1014 | int ret; | |
1015 | ||
1016 | va = find_vmap_lowest_match(size, align, vstart); | |
1017 | if (unlikely(!va)) | |
1018 | return vend; | |
1019 | ||
1020 | if (va->va_start > vstart) | |
1021 | nva_start_addr = ALIGN(va->va_start, align); | |
1022 | else | |
1023 | nva_start_addr = ALIGN(vstart, align); | |
1024 | ||
1025 | /* Check the "vend" restriction. */ | |
1026 | if (nva_start_addr + size > vend) | |
1027 | return vend; | |
1028 | ||
1029 | /* Classify what we have found. */ | |
1030 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1031 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1032 | return vend; | |
1033 | ||
1034 | /* Update the free vmap_area. */ | |
1035 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1036 | if (ret) | |
1037 | return vend; | |
1038 | ||
a6cf4e0f URS |
1039 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1040 | find_vmap_lowest_match_check(size); | |
1041 | #endif | |
1042 | ||
68ad4a33 URS |
1043 | return nva_start_addr; |
1044 | } | |
4da56b99 | 1045 | |
db64fe02 NP |
1046 | /* |
1047 | * Allocate a region of KVA of the specified size and alignment, within the | |
1048 | * vstart and vend. | |
1049 | */ | |
1050 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1051 | unsigned long align, | |
1052 | unsigned long vstart, unsigned long vend, | |
1053 | int node, gfp_t gfp_mask) | |
1054 | { | |
82dd23e8 | 1055 | struct vmap_area *va, *pva; |
1da177e4 | 1056 | unsigned long addr; |
db64fe02 NP |
1057 | int purged = 0; |
1058 | ||
7766970c | 1059 | BUG_ON(!size); |
891c49ab | 1060 | BUG_ON(offset_in_page(size)); |
89699605 | 1061 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1062 | |
68ad4a33 URS |
1063 | if (unlikely(!vmap_initialized)) |
1064 | return ERR_PTR(-EBUSY); | |
1065 | ||
5803ed29 | 1066 | might_sleep(); |
4da56b99 | 1067 | |
68ad4a33 | 1068 | va = kmem_cache_alloc_node(vmap_area_cachep, |
db64fe02 NP |
1069 | gfp_mask & GFP_RECLAIM_MASK, node); |
1070 | if (unlikely(!va)) | |
1071 | return ERR_PTR(-ENOMEM); | |
1072 | ||
7f88f88f CM |
1073 | /* |
1074 | * Only scan the relevant parts containing pointers to other objects | |
1075 | * to avoid false negatives. | |
1076 | */ | |
1077 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK); | |
1078 | ||
db64fe02 | 1079 | retry: |
82dd23e8 URS |
1080 | /* |
1081 | * Preload this CPU with one extra vmap_area object to ensure | |
1082 | * that we have it available when fit type of free area is | |
1083 | * NE_FIT_TYPE. | |
1084 | * | |
1085 | * The preload is done in non-atomic context, thus it allows us | |
1086 | * to use more permissive allocation masks to be more stable under | |
1087 | * low memory condition and high memory pressure. | |
1088 | * | |
1089 | * Even if it fails we do not really care about that. Just proceed | |
1090 | * as it is. "overflow" path will refill the cache we allocate from. | |
1091 | */ | |
1092 | preempt_disable(); | |
1093 | if (!__this_cpu_read(ne_fit_preload_node)) { | |
1094 | preempt_enable(); | |
1095 | pva = kmem_cache_alloc_node(vmap_area_cachep, GFP_KERNEL, node); | |
1096 | preempt_disable(); | |
1097 | ||
1098 | if (__this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) { | |
1099 | if (pva) | |
1100 | kmem_cache_free(vmap_area_cachep, pva); | |
1101 | } | |
1102 | } | |
1103 | ||
db64fe02 | 1104 | spin_lock(&vmap_area_lock); |
82dd23e8 | 1105 | preempt_enable(); |
89699605 | 1106 | |
afd07389 | 1107 | /* |
68ad4a33 URS |
1108 | * If an allocation fails, the "vend" address is |
1109 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1110 | */ |
cacca6ba | 1111 | addr = __alloc_vmap_area(size, align, vstart, vend); |
68ad4a33 | 1112 | if (unlikely(addr == vend)) |
89699605 | 1113 | goto overflow; |
db64fe02 NP |
1114 | |
1115 | va->va_start = addr; | |
1116 | va->va_end = addr + size; | |
688fcbfc | 1117 | va->vm = NULL; |
68ad4a33 URS |
1118 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
1119 | ||
db64fe02 NP |
1120 | spin_unlock(&vmap_area_lock); |
1121 | ||
61e16557 | 1122 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1123 | BUG_ON(va->va_start < vstart); |
1124 | BUG_ON(va->va_end > vend); | |
1125 | ||
db64fe02 | 1126 | return va; |
89699605 NP |
1127 | |
1128 | overflow: | |
1129 | spin_unlock(&vmap_area_lock); | |
1130 | if (!purged) { | |
1131 | purge_vmap_area_lazy(); | |
1132 | purged = 1; | |
1133 | goto retry; | |
1134 | } | |
4da56b99 CW |
1135 | |
1136 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1137 | unsigned long freed = 0; | |
1138 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1139 | if (freed > 0) { | |
1140 | purged = 0; | |
1141 | goto retry; | |
1142 | } | |
1143 | } | |
1144 | ||
03497d76 | 1145 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1146 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1147 | size); | |
68ad4a33 URS |
1148 | |
1149 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1150 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1151 | } |
1152 | ||
4da56b99 CW |
1153 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1154 | { | |
1155 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1156 | } | |
1157 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1158 | ||
1159 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1160 | { | |
1161 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1162 | } | |
1163 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1164 | ||
db64fe02 NP |
1165 | static void __free_vmap_area(struct vmap_area *va) |
1166 | { | |
ca23e405 | 1167 | /* |
68ad4a33 | 1168 | * Remove from the busy tree/list. |
ca23e405 | 1169 | */ |
68ad4a33 | 1170 | unlink_va(va, &vmap_area_root); |
ca23e405 | 1171 | |
68ad4a33 URS |
1172 | /* |
1173 | * Merge VA with its neighbors, otherwise just add it. | |
1174 | */ | |
1175 | merge_or_add_vmap_area(va, | |
1176 | &free_vmap_area_root, &free_vmap_area_list); | |
db64fe02 NP |
1177 | } |
1178 | ||
1179 | /* | |
1180 | * Free a region of KVA allocated by alloc_vmap_area | |
1181 | */ | |
1182 | static void free_vmap_area(struct vmap_area *va) | |
1183 | { | |
1184 | spin_lock(&vmap_area_lock); | |
1185 | __free_vmap_area(va); | |
1186 | spin_unlock(&vmap_area_lock); | |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * Clear the pagetable entries of a given vmap_area | |
1191 | */ | |
1192 | static void unmap_vmap_area(struct vmap_area *va) | |
1193 | { | |
1194 | vunmap_page_range(va->va_start, va->va_end); | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1199 | * before attempting to purge with a TLB flush. | |
1200 | * | |
1201 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1202 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1203 | * global TLB flushes that must be performed. It would seem natural to scale | |
1204 | * this number up linearly with the number of CPUs (because vmapping activity | |
1205 | * could also scale linearly with the number of CPUs), however it is likely | |
1206 | * that in practice, workloads might be constrained in other ways that mean | |
1207 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1208 | * conservative and not introduce a big latency on huge systems, so go with | |
1209 | * a less aggressive log scale. It will still be an improvement over the old | |
1210 | * code, and it will be simple to change the scale factor if we find that it | |
1211 | * becomes a problem on bigger systems. | |
1212 | */ | |
1213 | static unsigned long lazy_max_pages(void) | |
1214 | { | |
1215 | unsigned int log; | |
1216 | ||
1217 | log = fls(num_online_cpus()); | |
1218 | ||
1219 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1220 | } | |
1221 | ||
4d36e6f8 | 1222 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1223 | |
0574ecd1 CH |
1224 | /* |
1225 | * Serialize vmap purging. There is no actual criticial section protected | |
1226 | * by this look, but we want to avoid concurrent calls for performance | |
1227 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1228 | */ | |
f9e09977 | 1229 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1230 | |
02b709df NP |
1231 | /* for per-CPU blocks */ |
1232 | static void purge_fragmented_blocks_allcpus(void); | |
1233 | ||
3ee48b6a CW |
1234 | /* |
1235 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1236 | * immediately freed. | |
1237 | */ | |
1238 | void set_iounmap_nonlazy(void) | |
1239 | { | |
4d36e6f8 | 1240 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1241 | } |
1242 | ||
db64fe02 NP |
1243 | /* |
1244 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1245 | */ |
0574ecd1 | 1246 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1247 | { |
4d36e6f8 | 1248 | unsigned long resched_threshold; |
80c4bd7a | 1249 | struct llist_node *valist; |
db64fe02 | 1250 | struct vmap_area *va; |
cbb76676 | 1251 | struct vmap_area *n_va; |
db64fe02 | 1252 | |
0574ecd1 | 1253 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1254 | |
80c4bd7a | 1255 | valist = llist_del_all(&vmap_purge_list); |
68571be9 URS |
1256 | if (unlikely(valist == NULL)) |
1257 | return false; | |
1258 | ||
3f8fd02b JR |
1259 | /* |
1260 | * First make sure the mappings are removed from all page-tables | |
1261 | * before they are freed. | |
1262 | */ | |
1263 | vmalloc_sync_all(); | |
1264 | ||
68571be9 URS |
1265 | /* |
1266 | * TODO: to calculate a flush range without looping. | |
1267 | * The list can be up to lazy_max_pages() elements. | |
1268 | */ | |
80c4bd7a | 1269 | llist_for_each_entry(va, valist, purge_list) { |
0574ecd1 CH |
1270 | if (va->va_start < start) |
1271 | start = va->va_start; | |
1272 | if (va->va_end > end) | |
1273 | end = va->va_end; | |
db64fe02 | 1274 | } |
db64fe02 | 1275 | |
0574ecd1 | 1276 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1277 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1278 | |
0574ecd1 | 1279 | spin_lock(&vmap_area_lock); |
763b218d | 1280 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
4d36e6f8 | 1281 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
763b218d | 1282 | |
dd3b8353 URS |
1283 | /* |
1284 | * Finally insert or merge lazily-freed area. It is | |
1285 | * detached and there is no need to "unlink" it from | |
1286 | * anything. | |
1287 | */ | |
1288 | merge_or_add_vmap_area(va, | |
1289 | &free_vmap_area_root, &free_vmap_area_list); | |
1290 | ||
4d36e6f8 | 1291 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1292 | |
4d36e6f8 | 1293 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
68571be9 | 1294 | cond_resched_lock(&vmap_area_lock); |
763b218d | 1295 | } |
0574ecd1 CH |
1296 | spin_unlock(&vmap_area_lock); |
1297 | return true; | |
db64fe02 NP |
1298 | } |
1299 | ||
496850e5 NP |
1300 | /* |
1301 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1302 | * is already purging. | |
1303 | */ | |
1304 | static void try_purge_vmap_area_lazy(void) | |
1305 | { | |
f9e09977 | 1306 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1307 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1308 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1309 | } |
496850e5 NP |
1310 | } |
1311 | ||
db64fe02 NP |
1312 | /* |
1313 | * Kick off a purge of the outstanding lazy areas. | |
1314 | */ | |
1315 | static void purge_vmap_area_lazy(void) | |
1316 | { | |
f9e09977 | 1317 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1318 | purge_fragmented_blocks_allcpus(); |
1319 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1320 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1321 | } |
1322 | ||
1323 | /* | |
64141da5 JF |
1324 | * Free a vmap area, caller ensuring that the area has been unmapped |
1325 | * and flush_cache_vunmap had been called for the correct range | |
1326 | * previously. | |
db64fe02 | 1327 | */ |
64141da5 | 1328 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1329 | { |
4d36e6f8 | 1330 | unsigned long nr_lazy; |
80c4bd7a | 1331 | |
dd3b8353 URS |
1332 | spin_lock(&vmap_area_lock); |
1333 | unlink_va(va, &vmap_area_root); | |
1334 | spin_unlock(&vmap_area_lock); | |
1335 | ||
4d36e6f8 URS |
1336 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1337 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a CW |
1338 | |
1339 | /* After this point, we may free va at any time */ | |
1340 | llist_add(&va->purge_list, &vmap_purge_list); | |
1341 | ||
1342 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 1343 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1344 | } |
1345 | ||
b29acbdc NP |
1346 | /* |
1347 | * Free and unmap a vmap area | |
1348 | */ | |
1349 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1350 | { | |
1351 | flush_cache_vunmap(va->va_start, va->va_end); | |
c8eef01e | 1352 | unmap_vmap_area(va); |
82a2e924 CP |
1353 | if (debug_pagealloc_enabled()) |
1354 | flush_tlb_kernel_range(va->va_start, va->va_end); | |
1355 | ||
c8eef01e | 1356 | free_vmap_area_noflush(va); |
b29acbdc NP |
1357 | } |
1358 | ||
db64fe02 NP |
1359 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1360 | { | |
1361 | struct vmap_area *va; | |
1362 | ||
1363 | spin_lock(&vmap_area_lock); | |
1364 | va = __find_vmap_area(addr); | |
1365 | spin_unlock(&vmap_area_lock); | |
1366 | ||
1367 | return va; | |
1368 | } | |
1369 | ||
db64fe02 NP |
1370 | /*** Per cpu kva allocator ***/ |
1371 | ||
1372 | /* | |
1373 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1374 | * room for at least 16 percpu vmap blocks per CPU. | |
1375 | */ | |
1376 | /* | |
1377 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1378 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1379 | * instead (we just need a rough idea) | |
1380 | */ | |
1381 | #if BITS_PER_LONG == 32 | |
1382 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1383 | #else | |
1384 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1385 | #endif | |
1386 | ||
1387 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1388 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1389 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1390 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1391 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1392 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1393 | #define VMAP_BBMAP_BITS \ |
1394 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1395 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1396 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1397 | |
1398 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1399 | ||
1400 | struct vmap_block_queue { | |
1401 | spinlock_t lock; | |
1402 | struct list_head free; | |
db64fe02 NP |
1403 | }; |
1404 | ||
1405 | struct vmap_block { | |
1406 | spinlock_t lock; | |
1407 | struct vmap_area *va; | |
db64fe02 | 1408 | unsigned long free, dirty; |
7d61bfe8 | 1409 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1410 | struct list_head free_list; |
1411 | struct rcu_head rcu_head; | |
02b709df | 1412 | struct list_head purge; |
db64fe02 NP |
1413 | }; |
1414 | ||
1415 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1416 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1417 | ||
1418 | /* | |
1419 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
1420 | * in the free path. Could get rid of this if we change the API to return a | |
1421 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1422 | */ | |
1423 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
1424 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
1425 | ||
1426 | /* | |
1427 | * We should probably have a fallback mechanism to allocate virtual memory | |
1428 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1429 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1430 | * big problem. | |
1431 | */ | |
1432 | ||
1433 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1434 | { | |
1435 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1436 | addr /= VMAP_BLOCK_SIZE; | |
1437 | return addr; | |
1438 | } | |
1439 | ||
cf725ce2 RP |
1440 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1441 | { | |
1442 | unsigned long addr; | |
1443 | ||
1444 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1445 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1446 | return (void *)addr; | |
1447 | } | |
1448 | ||
1449 | /** | |
1450 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1451 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1452 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1453 | * @gfp_mask: flags for the page level allocator | |
1454 | * | |
a862f68a | 1455 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1456 | */ |
1457 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1458 | { |
1459 | struct vmap_block_queue *vbq; | |
1460 | struct vmap_block *vb; | |
1461 | struct vmap_area *va; | |
1462 | unsigned long vb_idx; | |
1463 | int node, err; | |
cf725ce2 | 1464 | void *vaddr; |
db64fe02 NP |
1465 | |
1466 | node = numa_node_id(); | |
1467 | ||
1468 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1469 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1470 | if (unlikely(!vb)) | |
1471 | return ERR_PTR(-ENOMEM); | |
1472 | ||
1473 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1474 | VMALLOC_START, VMALLOC_END, | |
1475 | node, gfp_mask); | |
ddf9c6d4 | 1476 | if (IS_ERR(va)) { |
db64fe02 | 1477 | kfree(vb); |
e7d86340 | 1478 | return ERR_CAST(va); |
db64fe02 NP |
1479 | } |
1480 | ||
1481 | err = radix_tree_preload(gfp_mask); | |
1482 | if (unlikely(err)) { | |
1483 | kfree(vb); | |
1484 | free_vmap_area(va); | |
1485 | return ERR_PTR(err); | |
1486 | } | |
1487 | ||
cf725ce2 | 1488 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1489 | spin_lock_init(&vb->lock); |
1490 | vb->va = va; | |
cf725ce2 RP |
1491 | /* At least something should be left free */ |
1492 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1493 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1494 | vb->dirty = 0; |
7d61bfe8 RP |
1495 | vb->dirty_min = VMAP_BBMAP_BITS; |
1496 | vb->dirty_max = 0; | |
db64fe02 | 1497 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1498 | |
1499 | vb_idx = addr_to_vb_idx(va->va_start); | |
1500 | spin_lock(&vmap_block_tree_lock); | |
1501 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
1502 | spin_unlock(&vmap_block_tree_lock); | |
1503 | BUG_ON(err); | |
1504 | radix_tree_preload_end(); | |
1505 | ||
1506 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1507 | spin_lock(&vbq->lock); |
68ac546f | 1508 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1509 | spin_unlock(&vbq->lock); |
3f04ba85 | 1510 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1511 | |
cf725ce2 | 1512 | return vaddr; |
db64fe02 NP |
1513 | } |
1514 | ||
db64fe02 NP |
1515 | static void free_vmap_block(struct vmap_block *vb) |
1516 | { | |
1517 | struct vmap_block *tmp; | |
1518 | unsigned long vb_idx; | |
1519 | ||
db64fe02 NP |
1520 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
1521 | spin_lock(&vmap_block_tree_lock); | |
1522 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
1523 | spin_unlock(&vmap_block_tree_lock); | |
1524 | BUG_ON(tmp != vb); | |
1525 | ||
64141da5 | 1526 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1527 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1528 | } |
1529 | ||
02b709df NP |
1530 | static void purge_fragmented_blocks(int cpu) |
1531 | { | |
1532 | LIST_HEAD(purge); | |
1533 | struct vmap_block *vb; | |
1534 | struct vmap_block *n_vb; | |
1535 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1536 | ||
1537 | rcu_read_lock(); | |
1538 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1539 | ||
1540 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1541 | continue; | |
1542 | ||
1543 | spin_lock(&vb->lock); | |
1544 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1545 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1546 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1547 | vb->dirty_min = 0; |
1548 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1549 | spin_lock(&vbq->lock); |
1550 | list_del_rcu(&vb->free_list); | |
1551 | spin_unlock(&vbq->lock); | |
1552 | spin_unlock(&vb->lock); | |
1553 | list_add_tail(&vb->purge, &purge); | |
1554 | } else | |
1555 | spin_unlock(&vb->lock); | |
1556 | } | |
1557 | rcu_read_unlock(); | |
1558 | ||
1559 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1560 | list_del(&vb->purge); | |
1561 | free_vmap_block(vb); | |
1562 | } | |
1563 | } | |
1564 | ||
02b709df NP |
1565 | static void purge_fragmented_blocks_allcpus(void) |
1566 | { | |
1567 | int cpu; | |
1568 | ||
1569 | for_each_possible_cpu(cpu) | |
1570 | purge_fragmented_blocks(cpu); | |
1571 | } | |
1572 | ||
db64fe02 NP |
1573 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1574 | { | |
1575 | struct vmap_block_queue *vbq; | |
1576 | struct vmap_block *vb; | |
cf725ce2 | 1577 | void *vaddr = NULL; |
db64fe02 NP |
1578 | unsigned int order; |
1579 | ||
891c49ab | 1580 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1581 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1582 | if (WARN_ON(size == 0)) { |
1583 | /* | |
1584 | * Allocating 0 bytes isn't what caller wants since | |
1585 | * get_order(0) returns funny result. Just warn and terminate | |
1586 | * early. | |
1587 | */ | |
1588 | return NULL; | |
1589 | } | |
db64fe02 NP |
1590 | order = get_order(size); |
1591 | ||
db64fe02 NP |
1592 | rcu_read_lock(); |
1593 | vbq = &get_cpu_var(vmap_block_queue); | |
1594 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1595 | unsigned long pages_off; |
db64fe02 NP |
1596 | |
1597 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1598 | if (vb->free < (1UL << order)) { |
1599 | spin_unlock(&vb->lock); | |
1600 | continue; | |
1601 | } | |
02b709df | 1602 | |
cf725ce2 RP |
1603 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1604 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1605 | vb->free -= 1UL << order; |
1606 | if (vb->free == 0) { | |
1607 | spin_lock(&vbq->lock); | |
1608 | list_del_rcu(&vb->free_list); | |
1609 | spin_unlock(&vbq->lock); | |
1610 | } | |
cf725ce2 | 1611 | |
02b709df NP |
1612 | spin_unlock(&vb->lock); |
1613 | break; | |
db64fe02 | 1614 | } |
02b709df | 1615 | |
3f04ba85 | 1616 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1617 | rcu_read_unlock(); |
1618 | ||
cf725ce2 RP |
1619 | /* Allocate new block if nothing was found */ |
1620 | if (!vaddr) | |
1621 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1622 | |
cf725ce2 | 1623 | return vaddr; |
db64fe02 NP |
1624 | } |
1625 | ||
1626 | static void vb_free(const void *addr, unsigned long size) | |
1627 | { | |
1628 | unsigned long offset; | |
1629 | unsigned long vb_idx; | |
1630 | unsigned int order; | |
1631 | struct vmap_block *vb; | |
1632 | ||
891c49ab | 1633 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1634 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc NP |
1635 | |
1636 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
1637 | ||
db64fe02 NP |
1638 | order = get_order(size); |
1639 | ||
1640 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
7d61bfe8 | 1641 | offset >>= PAGE_SHIFT; |
db64fe02 NP |
1642 | |
1643 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
1644 | rcu_read_lock(); | |
1645 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1646 | rcu_read_unlock(); | |
1647 | BUG_ON(!vb); | |
1648 | ||
64141da5 JF |
1649 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
1650 | ||
82a2e924 CP |
1651 | if (debug_pagealloc_enabled()) |
1652 | flush_tlb_kernel_range((unsigned long)addr, | |
1653 | (unsigned long)addr + size); | |
1654 | ||
db64fe02 | 1655 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1656 | |
1657 | /* Expand dirty range */ | |
1658 | vb->dirty_min = min(vb->dirty_min, offset); | |
1659 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1660 | |
db64fe02 NP |
1661 | vb->dirty += 1UL << order; |
1662 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1663 | BUG_ON(vb->free); |
db64fe02 NP |
1664 | spin_unlock(&vb->lock); |
1665 | free_vmap_block(vb); | |
1666 | } else | |
1667 | spin_unlock(&vb->lock); | |
1668 | } | |
1669 | ||
868b104d | 1670 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1671 | { |
db64fe02 | 1672 | int cpu; |
db64fe02 | 1673 | |
9b463334 JF |
1674 | if (unlikely(!vmap_initialized)) |
1675 | return; | |
1676 | ||
5803ed29 CH |
1677 | might_sleep(); |
1678 | ||
db64fe02 NP |
1679 | for_each_possible_cpu(cpu) { |
1680 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1681 | struct vmap_block *vb; | |
1682 | ||
1683 | rcu_read_lock(); | |
1684 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1685 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1686 | if (vb->dirty) { |
1687 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1688 | unsigned long s, e; |
b136be5e | 1689 | |
7d61bfe8 RP |
1690 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1691 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1692 | |
7d61bfe8 RP |
1693 | start = min(s, start); |
1694 | end = max(e, end); | |
db64fe02 | 1695 | |
7d61bfe8 | 1696 | flush = 1; |
db64fe02 NP |
1697 | } |
1698 | spin_unlock(&vb->lock); | |
1699 | } | |
1700 | rcu_read_unlock(); | |
1701 | } | |
1702 | ||
f9e09977 | 1703 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1704 | purge_fragmented_blocks_allcpus(); |
1705 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1706 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1707 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 1708 | } |
868b104d RE |
1709 | |
1710 | /** | |
1711 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1712 | * | |
1713 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1714 | * to amortize TLB flushing overheads. What this means is that any page you | |
1715 | * have now, may, in a former life, have been mapped into kernel virtual | |
1716 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1717 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1718 | * | |
1719 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1720 | * be sure that none of the pages we have control over will have any aliases | |
1721 | * from the vmap layer. | |
1722 | */ | |
1723 | void vm_unmap_aliases(void) | |
1724 | { | |
1725 | unsigned long start = ULONG_MAX, end = 0; | |
1726 | int flush = 0; | |
1727 | ||
1728 | _vm_unmap_aliases(start, end, flush); | |
1729 | } | |
db64fe02 NP |
1730 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
1731 | ||
1732 | /** | |
1733 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1734 | * @mem: the pointer returned by vm_map_ram | |
1735 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1736 | */ | |
1737 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1738 | { | |
65ee03c4 | 1739 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1740 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1741 | struct vmap_area *va; |
db64fe02 | 1742 | |
5803ed29 | 1743 | might_sleep(); |
db64fe02 NP |
1744 | BUG_ON(!addr); |
1745 | BUG_ON(addr < VMALLOC_START); | |
1746 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1747 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 1748 | |
9c3acf60 | 1749 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 1750 | debug_check_no_locks_freed(mem, size); |
db64fe02 | 1751 | vb_free(mem, size); |
9c3acf60 CH |
1752 | return; |
1753 | } | |
1754 | ||
1755 | va = find_vmap_area(addr); | |
1756 | BUG_ON(!va); | |
05e3ff95 CP |
1757 | debug_check_no_locks_freed((void *)va->va_start, |
1758 | (va->va_end - va->va_start)); | |
9c3acf60 | 1759 | free_unmap_vmap_area(va); |
db64fe02 NP |
1760 | } |
1761 | EXPORT_SYMBOL(vm_unmap_ram); | |
1762 | ||
1763 | /** | |
1764 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1765 | * @pages: an array of pointers to the pages to be mapped | |
1766 | * @count: number of pages | |
1767 | * @node: prefer to allocate data structures on this node | |
1768 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1769 | * |
36437638 GK |
1770 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1771 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1772 | * objects with vm_map_ram(), it could consume lots of address space through | |
1773 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1774 | * the end. Please use this function for short-lived objects. | |
1775 | * | |
e99c97ad | 1776 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1777 | */ |
1778 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1779 | { | |
65ee03c4 | 1780 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1781 | unsigned long addr; |
1782 | void *mem; | |
1783 | ||
1784 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1785 | mem = vb_alloc(size, GFP_KERNEL); | |
1786 | if (IS_ERR(mem)) | |
1787 | return NULL; | |
1788 | addr = (unsigned long)mem; | |
1789 | } else { | |
1790 | struct vmap_area *va; | |
1791 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1792 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1793 | if (IS_ERR(va)) | |
1794 | return NULL; | |
1795 | ||
1796 | addr = va->va_start; | |
1797 | mem = (void *)addr; | |
1798 | } | |
1799 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1800 | vm_unmap_ram(mem, count); | |
1801 | return NULL; | |
1802 | } | |
1803 | return mem; | |
1804 | } | |
1805 | EXPORT_SYMBOL(vm_map_ram); | |
1806 | ||
4341fa45 | 1807 | static struct vm_struct *vmlist __initdata; |
92eac168 | 1808 | |
be9b7335 NP |
1809 | /** |
1810 | * vm_area_add_early - add vmap area early during boot | |
1811 | * @vm: vm_struct to add | |
1812 | * | |
1813 | * This function is used to add fixed kernel vm area to vmlist before | |
1814 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1815 | * should contain proper values and the other fields should be zero. | |
1816 | * | |
1817 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1818 | */ | |
1819 | void __init vm_area_add_early(struct vm_struct *vm) | |
1820 | { | |
1821 | struct vm_struct *tmp, **p; | |
1822 | ||
1823 | BUG_ON(vmap_initialized); | |
1824 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1825 | if (tmp->addr >= vm->addr) { | |
1826 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1827 | break; | |
1828 | } else | |
1829 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1830 | } | |
1831 | vm->next = *p; | |
1832 | *p = vm; | |
1833 | } | |
1834 | ||
f0aa6617 TH |
1835 | /** |
1836 | * vm_area_register_early - register vmap area early during boot | |
1837 | * @vm: vm_struct to register | |
c0c0a293 | 1838 | * @align: requested alignment |
f0aa6617 TH |
1839 | * |
1840 | * This function is used to register kernel vm area before | |
1841 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1842 | * proper values on entry and other fields should be zero. On return, | |
1843 | * vm->addr contains the allocated address. | |
1844 | * | |
1845 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1846 | */ | |
c0c0a293 | 1847 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1848 | { |
1849 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1850 | unsigned long addr; |
1851 | ||
1852 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1853 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1854 | |
c0c0a293 | 1855 | vm->addr = (void *)addr; |
f0aa6617 | 1856 | |
be9b7335 | 1857 | vm_area_add_early(vm); |
f0aa6617 TH |
1858 | } |
1859 | ||
68ad4a33 URS |
1860 | static void vmap_init_free_space(void) |
1861 | { | |
1862 | unsigned long vmap_start = 1; | |
1863 | const unsigned long vmap_end = ULONG_MAX; | |
1864 | struct vmap_area *busy, *free; | |
1865 | ||
1866 | /* | |
1867 | * B F B B B F | |
1868 | * -|-----|.....|-----|-----|-----|.....|- | |
1869 | * | The KVA space | | |
1870 | * |<--------------------------------->| | |
1871 | */ | |
1872 | list_for_each_entry(busy, &vmap_area_list, list) { | |
1873 | if (busy->va_start - vmap_start > 0) { | |
1874 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1875 | if (!WARN_ON_ONCE(!free)) { | |
1876 | free->va_start = vmap_start; | |
1877 | free->va_end = busy->va_start; | |
1878 | ||
1879 | insert_vmap_area_augment(free, NULL, | |
1880 | &free_vmap_area_root, | |
1881 | &free_vmap_area_list); | |
1882 | } | |
1883 | } | |
1884 | ||
1885 | vmap_start = busy->va_end; | |
1886 | } | |
1887 | ||
1888 | if (vmap_end - vmap_start > 0) { | |
1889 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1890 | if (!WARN_ON_ONCE(!free)) { | |
1891 | free->va_start = vmap_start; | |
1892 | free->va_end = vmap_end; | |
1893 | ||
1894 | insert_vmap_area_augment(free, NULL, | |
1895 | &free_vmap_area_root, | |
1896 | &free_vmap_area_list); | |
1897 | } | |
1898 | } | |
1899 | } | |
1900 | ||
db64fe02 NP |
1901 | void __init vmalloc_init(void) |
1902 | { | |
822c18f2 IK |
1903 | struct vmap_area *va; |
1904 | struct vm_struct *tmp; | |
db64fe02 NP |
1905 | int i; |
1906 | ||
68ad4a33 URS |
1907 | /* |
1908 | * Create the cache for vmap_area objects. | |
1909 | */ | |
1910 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
1911 | ||
db64fe02 NP |
1912 | for_each_possible_cpu(i) { |
1913 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1914 | struct vfree_deferred *p; |
db64fe02 NP |
1915 | |
1916 | vbq = &per_cpu(vmap_block_queue, i); | |
1917 | spin_lock_init(&vbq->lock); | |
1918 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1919 | p = &per_cpu(vfree_deferred, i); |
1920 | init_llist_head(&p->list); | |
1921 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1922 | } |
9b463334 | 1923 | |
822c18f2 IK |
1924 | /* Import existing vmlist entries. */ |
1925 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
1926 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
1927 | if (WARN_ON_ONCE(!va)) | |
1928 | continue; | |
1929 | ||
822c18f2 IK |
1930 | va->va_start = (unsigned long)tmp->addr; |
1931 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1932 | va->vm = tmp; |
68ad4a33 | 1933 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 1934 | } |
ca23e405 | 1935 | |
68ad4a33 URS |
1936 | /* |
1937 | * Now we can initialize a free vmap space. | |
1938 | */ | |
1939 | vmap_init_free_space(); | |
9b463334 | 1940 | vmap_initialized = true; |
db64fe02 NP |
1941 | } |
1942 | ||
8fc48985 TH |
1943 | /** |
1944 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1945 | * @addr: start of the VM area to map | |
1946 | * @size: size of the VM area to map | |
1947 | * @prot: page protection flags to use | |
1948 | * @pages: pages to map | |
1949 | * | |
1950 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1951 | * specify should have been allocated using get_vm_area() and its | |
1952 | * friends. | |
1953 | * | |
1954 | * NOTE: | |
1955 | * This function does NOT do any cache flushing. The caller is | |
1956 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1957 | * before calling this function. | |
1958 | * | |
1959 | * RETURNS: | |
1960 | * The number of pages mapped on success, -errno on failure. | |
1961 | */ | |
1962 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1963 | pgprot_t prot, struct page **pages) | |
1964 | { | |
1965 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1966 | } | |
1967 | ||
1968 | /** | |
1969 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1970 | * @addr: start of the VM area to unmap | |
1971 | * @size: size of the VM area to unmap | |
1972 | * | |
1973 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1974 | * specify should have been allocated using get_vm_area() and its | |
1975 | * friends. | |
1976 | * | |
1977 | * NOTE: | |
1978 | * This function does NOT do any cache flushing. The caller is | |
1979 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1980 | * before calling this function and flush_tlb_kernel_range() after. | |
1981 | */ | |
1982 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1983 | { | |
1984 | vunmap_page_range(addr, addr + size); | |
1985 | } | |
81e88fdc | 1986 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1987 | |
1988 | /** | |
1989 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1990 | * @addr: start of the VM area to unmap | |
1991 | * @size: size of the VM area to unmap | |
1992 | * | |
1993 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1994 | * the unmapping and tlb after. | |
1995 | */ | |
db64fe02 NP |
1996 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1997 | { | |
1998 | unsigned long end = addr + size; | |
f6fcba70 TH |
1999 | |
2000 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
2001 | vunmap_page_range(addr, end); |
2002 | flush_tlb_kernel_range(addr, end); | |
2003 | } | |
93ef6d6c | 2004 | EXPORT_SYMBOL_GPL(unmap_kernel_range); |
db64fe02 | 2005 | |
f6f8ed47 | 2006 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
2007 | { |
2008 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 2009 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
2010 | int err; |
2011 | ||
f6f8ed47 | 2012 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 2013 | |
f6f8ed47 | 2014 | return err > 0 ? 0 : err; |
db64fe02 NP |
2015 | } |
2016 | EXPORT_SYMBOL_GPL(map_vm_area); | |
2017 | ||
f5252e00 | 2018 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 2019 | unsigned long flags, const void *caller) |
cf88c790 | 2020 | { |
c69480ad | 2021 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
2022 | vm->flags = flags; |
2023 | vm->addr = (void *)va->va_start; | |
2024 | vm->size = va->va_end - va->va_start; | |
2025 | vm->caller = caller; | |
db1aecaf | 2026 | va->vm = vm; |
c69480ad | 2027 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2028 | } |
cf88c790 | 2029 | |
20fc02b4 | 2030 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2031 | { |
d4033afd | 2032 | /* |
20fc02b4 | 2033 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2034 | * we should make sure that vm has proper values. |
2035 | * Pair with smp_rmb() in show_numa_info(). | |
2036 | */ | |
2037 | smp_wmb(); | |
20fc02b4 | 2038 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2039 | } |
2040 | ||
db64fe02 | 2041 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2042 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2043 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2044 | { |
0006526d | 2045 | struct vmap_area *va; |
db64fe02 | 2046 | struct vm_struct *area; |
1da177e4 | 2047 | |
52fd24ca | 2048 | BUG_ON(in_interrupt()); |
1da177e4 | 2049 | size = PAGE_ALIGN(size); |
31be8309 OH |
2050 | if (unlikely(!size)) |
2051 | return NULL; | |
1da177e4 | 2052 | |
252e5c6e | 2053 | if (flags & VM_IOREMAP) |
2054 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2055 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2056 | ||
cf88c790 | 2057 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2058 | if (unlikely(!area)) |
2059 | return NULL; | |
2060 | ||
71394fe5 AR |
2061 | if (!(flags & VM_NO_GUARD)) |
2062 | size += PAGE_SIZE; | |
1da177e4 | 2063 | |
db64fe02 NP |
2064 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2065 | if (IS_ERR(va)) { | |
2066 | kfree(area); | |
2067 | return NULL; | |
1da177e4 | 2068 | } |
1da177e4 | 2069 | |
d82b1d85 | 2070 | setup_vmalloc_vm(area, va, flags, caller); |
f5252e00 | 2071 | |
1da177e4 | 2072 | return area; |
1da177e4 LT |
2073 | } |
2074 | ||
930fc45a CL |
2075 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
2076 | unsigned long start, unsigned long end) | |
2077 | { | |
00ef2d2f DR |
2078 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2079 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 2080 | } |
5992b6da | 2081 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 2082 | |
c2968612 BH |
2083 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2084 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2085 | const void *caller) |
c2968612 | 2086 | { |
00ef2d2f DR |
2087 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2088 | GFP_KERNEL, caller); | |
c2968612 BH |
2089 | } |
2090 | ||
1da177e4 | 2091 | /** |
92eac168 MR |
2092 | * get_vm_area - reserve a contiguous kernel virtual area |
2093 | * @size: size of the area | |
2094 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2095 | * |
92eac168 MR |
2096 | * Search an area of @size in the kernel virtual mapping area, |
2097 | * and reserved it for out purposes. Returns the area descriptor | |
2098 | * on success or %NULL on failure. | |
a862f68a MR |
2099 | * |
2100 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2101 | */ |
2102 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2103 | { | |
2dca6999 | 2104 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2105 | NUMA_NO_NODE, GFP_KERNEL, |
2106 | __builtin_return_address(0)); | |
23016969 CL |
2107 | } |
2108 | ||
2109 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2110 | const void *caller) |
23016969 | 2111 | { |
2dca6999 | 2112 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2113 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2114 | } |
2115 | ||
e9da6e99 | 2116 | /** |
92eac168 MR |
2117 | * find_vm_area - find a continuous kernel virtual area |
2118 | * @addr: base address | |
e9da6e99 | 2119 | * |
92eac168 MR |
2120 | * Search for the kernel VM area starting at @addr, and return it. |
2121 | * It is up to the caller to do all required locking to keep the returned | |
2122 | * pointer valid. | |
a862f68a MR |
2123 | * |
2124 | * Return: pointer to the found area or %NULL on faulure | |
e9da6e99 MS |
2125 | */ |
2126 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2127 | { |
db64fe02 | 2128 | struct vmap_area *va; |
83342314 | 2129 | |
db64fe02 | 2130 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2131 | if (!va) |
2132 | return NULL; | |
1da177e4 | 2133 | |
688fcbfc | 2134 | return va->vm; |
1da177e4 LT |
2135 | } |
2136 | ||
7856dfeb | 2137 | /** |
92eac168 MR |
2138 | * remove_vm_area - find and remove a continuous kernel virtual area |
2139 | * @addr: base address | |
7856dfeb | 2140 | * |
92eac168 MR |
2141 | * Search for the kernel VM area starting at @addr, and remove it. |
2142 | * This function returns the found VM area, but using it is NOT safe | |
2143 | * on SMP machines, except for its size or flags. | |
a862f68a MR |
2144 | * |
2145 | * Return: pointer to the found area or %NULL on faulure | |
7856dfeb | 2146 | */ |
b3bdda02 | 2147 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2148 | { |
db64fe02 NP |
2149 | struct vmap_area *va; |
2150 | ||
5803ed29 CH |
2151 | might_sleep(); |
2152 | ||
dd3b8353 URS |
2153 | spin_lock(&vmap_area_lock); |
2154 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2155 | if (va && va->vm) { |
db1aecaf | 2156 | struct vm_struct *vm = va->vm; |
f5252e00 | 2157 | |
c69480ad | 2158 | va->vm = NULL; |
c69480ad JK |
2159 | spin_unlock(&vmap_area_lock); |
2160 | ||
a5af5aa8 | 2161 | kasan_free_shadow(vm); |
dd32c279 | 2162 | free_unmap_vmap_area(va); |
dd32c279 | 2163 | |
db64fe02 NP |
2164 | return vm; |
2165 | } | |
dd3b8353 URS |
2166 | |
2167 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2168 | return NULL; |
7856dfeb AK |
2169 | } |
2170 | ||
868b104d RE |
2171 | static inline void set_area_direct_map(const struct vm_struct *area, |
2172 | int (*set_direct_map)(struct page *page)) | |
2173 | { | |
2174 | int i; | |
2175 | ||
2176 | for (i = 0; i < area->nr_pages; i++) | |
2177 | if (page_address(area->pages[i])) | |
2178 | set_direct_map(area->pages[i]); | |
2179 | } | |
2180 | ||
2181 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2182 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2183 | { | |
868b104d RE |
2184 | unsigned long start = ULONG_MAX, end = 0; |
2185 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
31e67340 | 2186 | int flush_dmap = 0; |
868b104d RE |
2187 | int i; |
2188 | ||
868b104d RE |
2189 | remove_vm_area(area->addr); |
2190 | ||
2191 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2192 | if (!flush_reset) | |
2193 | return; | |
2194 | ||
2195 | /* | |
2196 | * If not deallocating pages, just do the flush of the VM area and | |
2197 | * return. | |
2198 | */ | |
2199 | if (!deallocate_pages) { | |
2200 | vm_unmap_aliases(); | |
2201 | return; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * If execution gets here, flush the vm mapping and reset the direct | |
2206 | * map. Find the start and end range of the direct mappings to make sure | |
2207 | * the vm_unmap_aliases() flush includes the direct map. | |
2208 | */ | |
2209 | for (i = 0; i < area->nr_pages; i++) { | |
8e41f872 RE |
2210 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2211 | if (addr) { | |
868b104d | 2212 | start = min(addr, start); |
8e41f872 | 2213 | end = max(addr + PAGE_SIZE, end); |
31e67340 | 2214 | flush_dmap = 1; |
868b104d RE |
2215 | } |
2216 | } | |
2217 | ||
2218 | /* | |
2219 | * Set direct map to something invalid so that it won't be cached if | |
2220 | * there are any accesses after the TLB flush, then flush the TLB and | |
2221 | * reset the direct map permissions to the default. | |
2222 | */ | |
2223 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2224 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2225 | set_area_direct_map(area, set_direct_map_default_noflush); |
2226 | } | |
2227 | ||
b3bdda02 | 2228 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2229 | { |
2230 | struct vm_struct *area; | |
2231 | ||
2232 | if (!addr) | |
2233 | return; | |
2234 | ||
e69e9d4a | 2235 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2236 | addr)) |
1da177e4 | 2237 | return; |
1da177e4 | 2238 | |
6ade2032 | 2239 | area = find_vm_area(addr); |
1da177e4 | 2240 | if (unlikely(!area)) { |
4c8573e2 | 2241 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2242 | addr); |
1da177e4 LT |
2243 | return; |
2244 | } | |
2245 | ||
05e3ff95 CP |
2246 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2247 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2248 | |
868b104d RE |
2249 | vm_remove_mappings(area, deallocate_pages); |
2250 | ||
1da177e4 LT |
2251 | if (deallocate_pages) { |
2252 | int i; | |
2253 | ||
2254 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2255 | struct page *page = area->pages[i]; |
2256 | ||
2257 | BUG_ON(!page); | |
4949148a | 2258 | __free_pages(page, 0); |
1da177e4 | 2259 | } |
97105f0a | 2260 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2261 | |
244d63ee | 2262 | kvfree(area->pages); |
1da177e4 LT |
2263 | } |
2264 | ||
2265 | kfree(area); | |
2266 | return; | |
2267 | } | |
bf22e37a AR |
2268 | |
2269 | static inline void __vfree_deferred(const void *addr) | |
2270 | { | |
2271 | /* | |
2272 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2273 | * context. Preemption is absolutely fine here, because the llist_add() | |
2274 | * implementation is lockless, so it works even if we are adding to | |
2275 | * nother cpu's list. schedule_work() should be fine with this too. | |
2276 | */ | |
2277 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2278 | ||
2279 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2280 | schedule_work(&p->wq); | |
2281 | } | |
2282 | ||
2283 | /** | |
92eac168 MR |
2284 | * vfree_atomic - release memory allocated by vmalloc() |
2285 | * @addr: memory base address | |
bf22e37a | 2286 | * |
92eac168 MR |
2287 | * This one is just like vfree() but can be called in any atomic context |
2288 | * except NMIs. | |
bf22e37a AR |
2289 | */ |
2290 | void vfree_atomic(const void *addr) | |
2291 | { | |
2292 | BUG_ON(in_nmi()); | |
2293 | ||
2294 | kmemleak_free(addr); | |
2295 | ||
2296 | if (!addr) | |
2297 | return; | |
2298 | __vfree_deferred(addr); | |
2299 | } | |
2300 | ||
c67dc624 RP |
2301 | static void __vfree(const void *addr) |
2302 | { | |
2303 | if (unlikely(in_interrupt())) | |
2304 | __vfree_deferred(addr); | |
2305 | else | |
2306 | __vunmap(addr, 1); | |
2307 | } | |
2308 | ||
1da177e4 | 2309 | /** |
92eac168 MR |
2310 | * vfree - release memory allocated by vmalloc() |
2311 | * @addr: memory base address | |
1da177e4 | 2312 | * |
92eac168 MR |
2313 | * Free the virtually continuous memory area starting at @addr, as |
2314 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is | |
2315 | * NULL, no operation is performed. | |
1da177e4 | 2316 | * |
92eac168 MR |
2317 | * Must not be called in NMI context (strictly speaking, only if we don't |
2318 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2319 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 2320 | * |
92eac168 | 2321 | * May sleep if called *not* from interrupt context. |
3ca4ea3a | 2322 | * |
92eac168 | 2323 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 2324 | */ |
b3bdda02 | 2325 | void vfree(const void *addr) |
1da177e4 | 2326 | { |
32fcfd40 | 2327 | BUG_ON(in_nmi()); |
89219d37 CM |
2328 | |
2329 | kmemleak_free(addr); | |
2330 | ||
a8dda165 AR |
2331 | might_sleep_if(!in_interrupt()); |
2332 | ||
32fcfd40 AV |
2333 | if (!addr) |
2334 | return; | |
c67dc624 RP |
2335 | |
2336 | __vfree(addr); | |
1da177e4 | 2337 | } |
1da177e4 LT |
2338 | EXPORT_SYMBOL(vfree); |
2339 | ||
2340 | /** | |
92eac168 MR |
2341 | * vunmap - release virtual mapping obtained by vmap() |
2342 | * @addr: memory base address | |
1da177e4 | 2343 | * |
92eac168 MR |
2344 | * Free the virtually contiguous memory area starting at @addr, |
2345 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2346 | * |
92eac168 | 2347 | * Must not be called in interrupt context. |
1da177e4 | 2348 | */ |
b3bdda02 | 2349 | void vunmap(const void *addr) |
1da177e4 LT |
2350 | { |
2351 | BUG_ON(in_interrupt()); | |
34754b69 | 2352 | might_sleep(); |
32fcfd40 AV |
2353 | if (addr) |
2354 | __vunmap(addr, 0); | |
1da177e4 | 2355 | } |
1da177e4 LT |
2356 | EXPORT_SYMBOL(vunmap); |
2357 | ||
2358 | /** | |
92eac168 MR |
2359 | * vmap - map an array of pages into virtually contiguous space |
2360 | * @pages: array of page pointers | |
2361 | * @count: number of pages to map | |
2362 | * @flags: vm_area->flags | |
2363 | * @prot: page protection for the mapping | |
2364 | * | |
2365 | * Maps @count pages from @pages into contiguous kernel virtual | |
2366 | * space. | |
a862f68a MR |
2367 | * |
2368 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2369 | */ |
2370 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2371 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2372 | { |
2373 | struct vm_struct *area; | |
65ee03c4 | 2374 | unsigned long size; /* In bytes */ |
1da177e4 | 2375 | |
34754b69 PZ |
2376 | might_sleep(); |
2377 | ||
ca79b0c2 | 2378 | if (count > totalram_pages()) |
1da177e4 LT |
2379 | return NULL; |
2380 | ||
65ee03c4 GJM |
2381 | size = (unsigned long)count << PAGE_SHIFT; |
2382 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2383 | if (!area) |
2384 | return NULL; | |
23016969 | 2385 | |
f6f8ed47 | 2386 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
2387 | vunmap(area->addr); |
2388 | return NULL; | |
2389 | } | |
2390 | ||
2391 | return area->addr; | |
2392 | } | |
1da177e4 LT |
2393 | EXPORT_SYMBOL(vmap); |
2394 | ||
8594a21c MH |
2395 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
2396 | gfp_t gfp_mask, pgprot_t prot, | |
2397 | int node, const void *caller); | |
e31d9eb5 | 2398 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 2399 | pgprot_t prot, int node) |
1da177e4 LT |
2400 | { |
2401 | struct page **pages; | |
2402 | unsigned int nr_pages, array_size, i; | |
930f036b | 2403 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
704b862f LA |
2404 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; |
2405 | const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ? | |
2406 | 0 : | |
2407 | __GFP_HIGHMEM; | |
1da177e4 | 2408 | |
762216ab | 2409 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
2410 | array_size = (nr_pages * sizeof(struct page *)); |
2411 | ||
2412 | area->nr_pages = nr_pages; | |
2413 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 2414 | if (array_size > PAGE_SIZE) { |
704b862f | 2415 | pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask, |
3722e13c | 2416 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 2417 | } else { |
976d6dfb | 2418 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2419 | } |
1da177e4 LT |
2420 | area->pages = pages; |
2421 | if (!area->pages) { | |
2422 | remove_vm_area(area->addr); | |
2423 | kfree(area); | |
2424 | return NULL; | |
2425 | } | |
1da177e4 LT |
2426 | |
2427 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2428 | struct page *page; |
2429 | ||
4b90951c | 2430 | if (node == NUMA_NO_NODE) |
704b862f | 2431 | page = alloc_page(alloc_mask|highmem_mask); |
930fc45a | 2432 | else |
704b862f | 2433 | page = alloc_pages_node(node, alloc_mask|highmem_mask, 0); |
bf53d6f8 CL |
2434 | |
2435 | if (unlikely(!page)) { | |
1da177e4 LT |
2436 | /* Successfully allocated i pages, free them in __vunmap() */ |
2437 | area->nr_pages = i; | |
97105f0a | 2438 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 LT |
2439 | goto fail; |
2440 | } | |
bf53d6f8 | 2441 | area->pages[i] = page; |
704b862f | 2442 | if (gfpflags_allow_blocking(gfp_mask|highmem_mask)) |
660654f9 | 2443 | cond_resched(); |
1da177e4 | 2444 | } |
97105f0a | 2445 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2446 | |
f6f8ed47 | 2447 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
2448 | goto fail; |
2449 | return area->addr; | |
2450 | ||
2451 | fail: | |
a8e99259 | 2452 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2453 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 2454 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 2455 | __vfree(area->addr); |
1da177e4 LT |
2456 | return NULL; |
2457 | } | |
2458 | ||
2459 | /** | |
92eac168 MR |
2460 | * __vmalloc_node_range - allocate virtually contiguous memory |
2461 | * @size: allocation size | |
2462 | * @align: desired alignment | |
2463 | * @start: vm area range start | |
2464 | * @end: vm area range end | |
2465 | * @gfp_mask: flags for the page level allocator | |
2466 | * @prot: protection mask for the allocated pages | |
2467 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2468 | * @node: node to use for allocation or NUMA_NO_NODE | |
2469 | * @caller: caller's return address | |
2470 | * | |
2471 | * Allocate enough pages to cover @size from the page level | |
2472 | * allocator with @gfp_mask flags. Map them into contiguous | |
2473 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2474 | * |
2475 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2476 | */ |
d0a21265 DR |
2477 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2478 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2479 | pgprot_t prot, unsigned long vm_flags, int node, |
2480 | const void *caller) | |
1da177e4 LT |
2481 | { |
2482 | struct vm_struct *area; | |
89219d37 CM |
2483 | void *addr; |
2484 | unsigned long real_size = size; | |
1da177e4 LT |
2485 | |
2486 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 2487 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 2488 | goto fail; |
1da177e4 | 2489 | |
cb9e3c29 AR |
2490 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | |
2491 | vm_flags, start, end, node, gfp_mask, caller); | |
1da177e4 | 2492 | if (!area) |
de7d2b56 | 2493 | goto fail; |
1da177e4 | 2494 | |
3722e13c | 2495 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 2496 | if (!addr) |
b82225f3 | 2497 | return NULL; |
89219d37 | 2498 | |
f5252e00 | 2499 | /* |
20fc02b4 ZY |
2500 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2501 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2502 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2503 | */ |
20fc02b4 | 2504 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2505 | |
94f4a161 | 2506 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2507 | |
2508 | return addr; | |
de7d2b56 JP |
2509 | |
2510 | fail: | |
a8e99259 | 2511 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2512 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 2513 | return NULL; |
1da177e4 LT |
2514 | } |
2515 | ||
153178ed URS |
2516 | /* |
2517 | * This is only for performance analysis of vmalloc and stress purpose. | |
2518 | * It is required by vmalloc test module, therefore do not use it other | |
2519 | * than that. | |
2520 | */ | |
2521 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2522 | EXPORT_SYMBOL_GPL(__vmalloc_node_range); | |
2523 | #endif | |
2524 | ||
d0a21265 | 2525 | /** |
92eac168 MR |
2526 | * __vmalloc_node - allocate virtually contiguous memory |
2527 | * @size: allocation size | |
2528 | * @align: desired alignment | |
2529 | * @gfp_mask: flags for the page level allocator | |
2530 | * @prot: protection mask for the allocated pages | |
2531 | * @node: node to use for allocation or NUMA_NO_NODE | |
2532 | * @caller: caller's return address | |
a7c3e901 | 2533 | * |
92eac168 MR |
2534 | * Allocate enough pages to cover @size from the page level |
2535 | * allocator with @gfp_mask flags. Map them into contiguous | |
2536 | * kernel virtual space, using a pagetable protection of @prot. | |
a7c3e901 | 2537 | * |
92eac168 MR |
2538 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2539 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2540 | * |
92eac168 MR |
2541 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2542 | * with mm people. | |
a862f68a MR |
2543 | * |
2544 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2545 | */ |
8594a21c | 2546 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
d0a21265 | 2547 | gfp_t gfp_mask, pgprot_t prot, |
5e6cafc8 | 2548 | int node, const void *caller) |
d0a21265 DR |
2549 | { |
2550 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 2551 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
2552 | } |
2553 | ||
930fc45a CL |
2554 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
2555 | { | |
00ef2d2f | 2556 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 2557 | __builtin_return_address(0)); |
930fc45a | 2558 | } |
1da177e4 LT |
2559 | EXPORT_SYMBOL(__vmalloc); |
2560 | ||
8594a21c MH |
2561 | static inline void *__vmalloc_node_flags(unsigned long size, |
2562 | int node, gfp_t flags) | |
2563 | { | |
2564 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
2565 | node, __builtin_return_address(0)); | |
2566 | } | |
2567 | ||
2568 | ||
2569 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags, | |
2570 | void *caller) | |
2571 | { | |
2572 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller); | |
2573 | } | |
2574 | ||
1da177e4 | 2575 | /** |
92eac168 MR |
2576 | * vmalloc - allocate virtually contiguous memory |
2577 | * @size: allocation size | |
2578 | * | |
2579 | * Allocate enough pages to cover @size from the page level | |
2580 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2581 | * |
92eac168 MR |
2582 | * For tight control over page level allocator and protection flags |
2583 | * use __vmalloc() instead. | |
a862f68a MR |
2584 | * |
2585 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2586 | */ |
2587 | void *vmalloc(unsigned long size) | |
2588 | { | |
00ef2d2f | 2589 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2590 | GFP_KERNEL); |
1da177e4 | 2591 | } |
1da177e4 LT |
2592 | EXPORT_SYMBOL(vmalloc); |
2593 | ||
e1ca7788 | 2594 | /** |
92eac168 MR |
2595 | * vzalloc - allocate virtually contiguous memory with zero fill |
2596 | * @size: allocation size | |
2597 | * | |
2598 | * Allocate enough pages to cover @size from the page level | |
2599 | * allocator and map them into contiguous kernel virtual space. | |
2600 | * The memory allocated is set to zero. | |
2601 | * | |
2602 | * For tight control over page level allocator and protection flags | |
2603 | * use __vmalloc() instead. | |
a862f68a MR |
2604 | * |
2605 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2606 | */ |
2607 | void *vzalloc(unsigned long size) | |
2608 | { | |
00ef2d2f | 2609 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2610 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2611 | } |
2612 | EXPORT_SYMBOL(vzalloc); | |
2613 | ||
83342314 | 2614 | /** |
ead04089 REB |
2615 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
2616 | * @size: allocation size | |
83342314 | 2617 | * |
ead04089 REB |
2618 | * The resulting memory area is zeroed so it can be mapped to userspace |
2619 | * without leaking data. | |
a862f68a MR |
2620 | * |
2621 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2622 | */ |
2623 | void *vmalloc_user(unsigned long size) | |
2624 | { | |
bc84c535 RP |
2625 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2626 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
2627 | VM_USERMAP, NUMA_NO_NODE, | |
2628 | __builtin_return_address(0)); | |
83342314 NP |
2629 | } |
2630 | EXPORT_SYMBOL(vmalloc_user); | |
2631 | ||
930fc45a | 2632 | /** |
92eac168 MR |
2633 | * vmalloc_node - allocate memory on a specific node |
2634 | * @size: allocation size | |
2635 | * @node: numa node | |
930fc45a | 2636 | * |
92eac168 MR |
2637 | * Allocate enough pages to cover @size from the page level |
2638 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 2639 | * |
92eac168 MR |
2640 | * For tight control over page level allocator and protection flags |
2641 | * use __vmalloc() instead. | |
a862f68a MR |
2642 | * |
2643 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2644 | */ |
2645 | void *vmalloc_node(unsigned long size, int node) | |
2646 | { | |
19809c2d | 2647 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL, |
23016969 | 2648 | node, __builtin_return_address(0)); |
930fc45a CL |
2649 | } |
2650 | EXPORT_SYMBOL(vmalloc_node); | |
2651 | ||
e1ca7788 DY |
2652 | /** |
2653 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2654 | * @size: allocation size | |
2655 | * @node: numa node | |
2656 | * | |
2657 | * Allocate enough pages to cover @size from the page level | |
2658 | * allocator and map them into contiguous kernel virtual space. | |
2659 | * The memory allocated is set to zero. | |
2660 | * | |
2661 | * For tight control over page level allocator and protection flags | |
2662 | * use __vmalloc_node() instead. | |
a862f68a MR |
2663 | * |
2664 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2665 | */ |
2666 | void *vzalloc_node(unsigned long size, int node) | |
2667 | { | |
2668 | return __vmalloc_node_flags(size, node, | |
19809c2d | 2669 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2670 | } |
2671 | EXPORT_SYMBOL(vzalloc_node); | |
2672 | ||
1da177e4 | 2673 | /** |
92eac168 MR |
2674 | * vmalloc_exec - allocate virtually contiguous, executable memory |
2675 | * @size: allocation size | |
1da177e4 | 2676 | * |
92eac168 MR |
2677 | * Kernel-internal function to allocate enough pages to cover @size |
2678 | * the page level allocator and map them into contiguous and | |
2679 | * executable kernel virtual space. | |
1da177e4 | 2680 | * |
92eac168 MR |
2681 | * For tight control over page level allocator and protection flags |
2682 | * use __vmalloc() instead. | |
a862f68a MR |
2683 | * |
2684 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 | 2685 | */ |
1da177e4 LT |
2686 | void *vmalloc_exec(unsigned long size) |
2687 | { | |
868b104d RE |
2688 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, |
2689 | GFP_KERNEL, PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS, | |
2690 | NUMA_NO_NODE, __builtin_return_address(0)); | |
1da177e4 LT |
2691 | } |
2692 | ||
0d08e0d3 | 2693 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2694 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2695 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2696 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2697 | #else |
698d0831 MH |
2698 | /* |
2699 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2700 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2701 | */ | |
2702 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2703 | #endif |
2704 | ||
1da177e4 | 2705 | /** |
92eac168 MR |
2706 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2707 | * @size: allocation size | |
1da177e4 | 2708 | * |
92eac168 MR |
2709 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2710 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2711 | * |
2712 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2713 | */ |
2714 | void *vmalloc_32(unsigned long size) | |
2715 | { | |
2dca6999 | 2716 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 2717 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 2718 | } |
1da177e4 LT |
2719 | EXPORT_SYMBOL(vmalloc_32); |
2720 | ||
83342314 | 2721 | /** |
ead04089 | 2722 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2723 | * @size: allocation size |
ead04089 REB |
2724 | * |
2725 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2726 | * mapped to userspace without leaking data. | |
a862f68a MR |
2727 | * |
2728 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2729 | */ |
2730 | void *vmalloc_32_user(unsigned long size) | |
2731 | { | |
bc84c535 RP |
2732 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2733 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2734 | VM_USERMAP, NUMA_NO_NODE, | |
2735 | __builtin_return_address(0)); | |
83342314 NP |
2736 | } |
2737 | EXPORT_SYMBOL(vmalloc_32_user); | |
2738 | ||
d0107eb0 KH |
2739 | /* |
2740 | * small helper routine , copy contents to buf from addr. | |
2741 | * If the page is not present, fill zero. | |
2742 | */ | |
2743 | ||
2744 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2745 | { | |
2746 | struct page *p; | |
2747 | int copied = 0; | |
2748 | ||
2749 | while (count) { | |
2750 | unsigned long offset, length; | |
2751 | ||
891c49ab | 2752 | offset = offset_in_page(addr); |
d0107eb0 KH |
2753 | length = PAGE_SIZE - offset; |
2754 | if (length > count) | |
2755 | length = count; | |
2756 | p = vmalloc_to_page(addr); | |
2757 | /* | |
2758 | * To do safe access to this _mapped_ area, we need | |
2759 | * lock. But adding lock here means that we need to add | |
2760 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2761 | * interface, rarely used. Instead of that, we'll use | |
2762 | * kmap() and get small overhead in this access function. | |
2763 | */ | |
2764 | if (p) { | |
2765 | /* | |
2766 | * we can expect USER0 is not used (see vread/vwrite's | |
2767 | * function description) | |
2768 | */ | |
9b04c5fe | 2769 | void *map = kmap_atomic(p); |
d0107eb0 | 2770 | memcpy(buf, map + offset, length); |
9b04c5fe | 2771 | kunmap_atomic(map); |
d0107eb0 KH |
2772 | } else |
2773 | memset(buf, 0, length); | |
2774 | ||
2775 | addr += length; | |
2776 | buf += length; | |
2777 | copied += length; | |
2778 | count -= length; | |
2779 | } | |
2780 | return copied; | |
2781 | } | |
2782 | ||
2783 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2784 | { | |
2785 | struct page *p; | |
2786 | int copied = 0; | |
2787 | ||
2788 | while (count) { | |
2789 | unsigned long offset, length; | |
2790 | ||
891c49ab | 2791 | offset = offset_in_page(addr); |
d0107eb0 KH |
2792 | length = PAGE_SIZE - offset; |
2793 | if (length > count) | |
2794 | length = count; | |
2795 | p = vmalloc_to_page(addr); | |
2796 | /* | |
2797 | * To do safe access to this _mapped_ area, we need | |
2798 | * lock. But adding lock here means that we need to add | |
2799 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2800 | * interface, rarely used. Instead of that, we'll use | |
2801 | * kmap() and get small overhead in this access function. | |
2802 | */ | |
2803 | if (p) { | |
2804 | /* | |
2805 | * we can expect USER0 is not used (see vread/vwrite's | |
2806 | * function description) | |
2807 | */ | |
9b04c5fe | 2808 | void *map = kmap_atomic(p); |
d0107eb0 | 2809 | memcpy(map + offset, buf, length); |
9b04c5fe | 2810 | kunmap_atomic(map); |
d0107eb0 KH |
2811 | } |
2812 | addr += length; | |
2813 | buf += length; | |
2814 | copied += length; | |
2815 | count -= length; | |
2816 | } | |
2817 | return copied; | |
2818 | } | |
2819 | ||
2820 | /** | |
92eac168 MR |
2821 | * vread() - read vmalloc area in a safe way. |
2822 | * @buf: buffer for reading data | |
2823 | * @addr: vm address. | |
2824 | * @count: number of bytes to be read. | |
2825 | * | |
92eac168 MR |
2826 | * This function checks that addr is a valid vmalloc'ed area, and |
2827 | * copy data from that area to a given buffer. If the given memory range | |
2828 | * of [addr...addr+count) includes some valid address, data is copied to | |
2829 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2830 | * IOREMAP area is treated as memory hole and no copy is done. | |
2831 | * | |
2832 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2833 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2834 | * | |
2835 | * Note: In usual ops, vread() is never necessary because the caller | |
2836 | * should know vmalloc() area is valid and can use memcpy(). | |
2837 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2838 | * any information, as /dev/kmem. |
a862f68a MR |
2839 | * |
2840 | * Return: number of bytes for which addr and buf should be increased | |
2841 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
2842 | * include any intersection with valid vmalloc area | |
d0107eb0 | 2843 | */ |
1da177e4 LT |
2844 | long vread(char *buf, char *addr, unsigned long count) |
2845 | { | |
e81ce85f JK |
2846 | struct vmap_area *va; |
2847 | struct vm_struct *vm; | |
1da177e4 | 2848 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2849 | unsigned long buflen = count; |
1da177e4 LT |
2850 | unsigned long n; |
2851 | ||
2852 | /* Don't allow overflow */ | |
2853 | if ((unsigned long) addr + count < count) | |
2854 | count = -(unsigned long) addr; | |
2855 | ||
e81ce85f JK |
2856 | spin_lock(&vmap_area_lock); |
2857 | list_for_each_entry(va, &vmap_area_list, list) { | |
2858 | if (!count) | |
2859 | break; | |
2860 | ||
688fcbfc | 2861 | if (!va->vm) |
e81ce85f JK |
2862 | continue; |
2863 | ||
2864 | vm = va->vm; | |
2865 | vaddr = (char *) vm->addr; | |
762216ab | 2866 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2867 | continue; |
2868 | while (addr < vaddr) { | |
2869 | if (count == 0) | |
2870 | goto finished; | |
2871 | *buf = '\0'; | |
2872 | buf++; | |
2873 | addr++; | |
2874 | count--; | |
2875 | } | |
762216ab | 2876 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2877 | if (n > count) |
2878 | n = count; | |
e81ce85f | 2879 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2880 | aligned_vread(buf, addr, n); |
2881 | else /* IOREMAP area is treated as memory hole */ | |
2882 | memset(buf, 0, n); | |
2883 | buf += n; | |
2884 | addr += n; | |
2885 | count -= n; | |
1da177e4 LT |
2886 | } |
2887 | finished: | |
e81ce85f | 2888 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2889 | |
2890 | if (buf == buf_start) | |
2891 | return 0; | |
2892 | /* zero-fill memory holes */ | |
2893 | if (buf != buf_start + buflen) | |
2894 | memset(buf, 0, buflen - (buf - buf_start)); | |
2895 | ||
2896 | return buflen; | |
1da177e4 LT |
2897 | } |
2898 | ||
d0107eb0 | 2899 | /** |
92eac168 MR |
2900 | * vwrite() - write vmalloc area in a safe way. |
2901 | * @buf: buffer for source data | |
2902 | * @addr: vm address. | |
2903 | * @count: number of bytes to be read. | |
2904 | * | |
92eac168 MR |
2905 | * This function checks that addr is a valid vmalloc'ed area, and |
2906 | * copy data from a buffer to the given addr. If specified range of | |
2907 | * [addr...addr+count) includes some valid address, data is copied from | |
2908 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2909 | * IOREMAP area is treated as memory hole and no copy is done. | |
2910 | * | |
2911 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2912 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2913 | * | |
2914 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2915 | * should know vmalloc() area is valid and can use memcpy(). | |
2916 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2917 | * any information, as /dev/kmem. |
a862f68a MR |
2918 | * |
2919 | * Return: number of bytes for which addr and buf should be | |
2920 | * increased (same number as @count) or %0 if [addr...addr+count) | |
2921 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 2922 | */ |
1da177e4 LT |
2923 | long vwrite(char *buf, char *addr, unsigned long count) |
2924 | { | |
e81ce85f JK |
2925 | struct vmap_area *va; |
2926 | struct vm_struct *vm; | |
d0107eb0 KH |
2927 | char *vaddr; |
2928 | unsigned long n, buflen; | |
2929 | int copied = 0; | |
1da177e4 LT |
2930 | |
2931 | /* Don't allow overflow */ | |
2932 | if ((unsigned long) addr + count < count) | |
2933 | count = -(unsigned long) addr; | |
d0107eb0 | 2934 | buflen = count; |
1da177e4 | 2935 | |
e81ce85f JK |
2936 | spin_lock(&vmap_area_lock); |
2937 | list_for_each_entry(va, &vmap_area_list, list) { | |
2938 | if (!count) | |
2939 | break; | |
2940 | ||
688fcbfc | 2941 | if (!va->vm) |
e81ce85f JK |
2942 | continue; |
2943 | ||
2944 | vm = va->vm; | |
2945 | vaddr = (char *) vm->addr; | |
762216ab | 2946 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2947 | continue; |
2948 | while (addr < vaddr) { | |
2949 | if (count == 0) | |
2950 | goto finished; | |
2951 | buf++; | |
2952 | addr++; | |
2953 | count--; | |
2954 | } | |
762216ab | 2955 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2956 | if (n > count) |
2957 | n = count; | |
e81ce85f | 2958 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2959 | aligned_vwrite(buf, addr, n); |
2960 | copied++; | |
2961 | } | |
2962 | buf += n; | |
2963 | addr += n; | |
2964 | count -= n; | |
1da177e4 LT |
2965 | } |
2966 | finished: | |
e81ce85f | 2967 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2968 | if (!copied) |
2969 | return 0; | |
2970 | return buflen; | |
1da177e4 | 2971 | } |
83342314 NP |
2972 | |
2973 | /** | |
92eac168 MR |
2974 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
2975 | * @vma: vma to cover | |
2976 | * @uaddr: target user address to start at | |
2977 | * @kaddr: virtual address of vmalloc kernel memory | |
2978 | * @size: size of map area | |
7682486b | 2979 | * |
92eac168 | 2980 | * Returns: 0 for success, -Exxx on failure |
83342314 | 2981 | * |
92eac168 MR |
2982 | * This function checks that @kaddr is a valid vmalloc'ed area, |
2983 | * and that it is big enough to cover the range starting at | |
2984 | * @uaddr in @vma. Will return failure if that criteria isn't | |
2985 | * met. | |
83342314 | 2986 | * |
92eac168 | 2987 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 2988 | */ |
e69e9d4a HD |
2989 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
2990 | void *kaddr, unsigned long size) | |
83342314 NP |
2991 | { |
2992 | struct vm_struct *area; | |
83342314 | 2993 | |
e69e9d4a HD |
2994 | size = PAGE_ALIGN(size); |
2995 | ||
2996 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
2997 | return -EINVAL; |
2998 | ||
e69e9d4a | 2999 | area = find_vm_area(kaddr); |
83342314 | 3000 | if (!area) |
db64fe02 | 3001 | return -EINVAL; |
83342314 | 3002 | |
fe9041c2 | 3003 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3004 | return -EINVAL; |
83342314 | 3005 | |
401592d2 | 3006 | if (kaddr + size > area->addr + get_vm_area_size(area)) |
db64fe02 | 3007 | return -EINVAL; |
83342314 | 3008 | |
83342314 | 3009 | do { |
e69e9d4a | 3010 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3011 | int ret; |
3012 | ||
83342314 NP |
3013 | ret = vm_insert_page(vma, uaddr, page); |
3014 | if (ret) | |
3015 | return ret; | |
3016 | ||
3017 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3018 | kaddr += PAGE_SIZE; |
3019 | size -= PAGE_SIZE; | |
3020 | } while (size > 0); | |
83342314 | 3021 | |
314e51b9 | 3022 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3023 | |
db64fe02 | 3024 | return 0; |
83342314 | 3025 | } |
e69e9d4a HD |
3026 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
3027 | ||
3028 | /** | |
92eac168 MR |
3029 | * remap_vmalloc_range - map vmalloc pages to userspace |
3030 | * @vma: vma to cover (map full range of vma) | |
3031 | * @addr: vmalloc memory | |
3032 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3033 | * |
92eac168 | 3034 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3035 | * |
92eac168 MR |
3036 | * This function checks that addr is a valid vmalloc'ed area, and |
3037 | * that it is big enough to cover the vma. Will return failure if | |
3038 | * that criteria isn't met. | |
e69e9d4a | 3039 | * |
92eac168 | 3040 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3041 | */ |
3042 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3043 | unsigned long pgoff) | |
3044 | { | |
3045 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
3046 | addr + (pgoff << PAGE_SHIFT), | |
3047 | vma->vm_end - vma->vm_start); | |
3048 | } | |
83342314 NP |
3049 | EXPORT_SYMBOL(remap_vmalloc_range); |
3050 | ||
1eeb66a1 CH |
3051 | /* |
3052 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
3053 | * have one. | |
3f8fd02b JR |
3054 | * |
3055 | * The purpose of this function is to make sure the vmalloc area | |
3056 | * mappings are identical in all page-tables in the system. | |
1eeb66a1 | 3057 | */ |
3b32123d | 3058 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
3059 | { |
3060 | } | |
5f4352fb JF |
3061 | |
3062 | ||
8b1e0f81 | 3063 | static int f(pte_t *pte, unsigned long addr, void *data) |
5f4352fb | 3064 | { |
cd12909c DV |
3065 | pte_t ***p = data; |
3066 | ||
3067 | if (p) { | |
3068 | *(*p) = pte; | |
3069 | (*p)++; | |
3070 | } | |
5f4352fb JF |
3071 | return 0; |
3072 | } | |
3073 | ||
3074 | /** | |
92eac168 MR |
3075 | * alloc_vm_area - allocate a range of kernel address space |
3076 | * @size: size of the area | |
3077 | * @ptes: returns the PTEs for the address space | |
7682486b | 3078 | * |
92eac168 | 3079 | * Returns: NULL on failure, vm_struct on success |
5f4352fb | 3080 | * |
92eac168 MR |
3081 | * This function reserves a range of kernel address space, and |
3082 | * allocates pagetables to map that range. No actual mappings | |
3083 | * are created. | |
cd12909c | 3084 | * |
92eac168 MR |
3085 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
3086 | * allocated for the VM area are returned. | |
5f4352fb | 3087 | */ |
cd12909c | 3088 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
3089 | { |
3090 | struct vm_struct *area; | |
3091 | ||
23016969 CL |
3092 | area = get_vm_area_caller(size, VM_IOREMAP, |
3093 | __builtin_return_address(0)); | |
5f4352fb JF |
3094 | if (area == NULL) |
3095 | return NULL; | |
3096 | ||
3097 | /* | |
3098 | * This ensures that page tables are constructed for this region | |
3099 | * of kernel virtual address space and mapped into init_mm. | |
3100 | */ | |
3101 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 3102 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
3103 | free_vm_area(area); |
3104 | return NULL; | |
3105 | } | |
3106 | ||
5f4352fb JF |
3107 | return area; |
3108 | } | |
3109 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
3110 | ||
3111 | void free_vm_area(struct vm_struct *area) | |
3112 | { | |
3113 | struct vm_struct *ret; | |
3114 | ret = remove_vm_area(area->addr); | |
3115 | BUG_ON(ret != area); | |
3116 | kfree(area); | |
3117 | } | |
3118 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3119 | |
4f8b02b4 | 3120 | #ifdef CONFIG_SMP |
ca23e405 TH |
3121 | static struct vmap_area *node_to_va(struct rb_node *n) |
3122 | { | |
4583e773 | 3123 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3124 | } |
3125 | ||
3126 | /** | |
68ad4a33 URS |
3127 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3128 | * @addr: target address | |
ca23e405 | 3129 | * |
68ad4a33 URS |
3130 | * Returns: vmap_area if it is found. If there is no such area |
3131 | * the first highest(reverse order) vmap_area is returned | |
3132 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3133 | * if there are no any areas before @addr. | |
ca23e405 | 3134 | */ |
68ad4a33 URS |
3135 | static struct vmap_area * |
3136 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3137 | { |
68ad4a33 URS |
3138 | struct vmap_area *va, *tmp; |
3139 | struct rb_node *n; | |
3140 | ||
3141 | n = free_vmap_area_root.rb_node; | |
3142 | va = NULL; | |
ca23e405 TH |
3143 | |
3144 | while (n) { | |
68ad4a33 URS |
3145 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3146 | if (tmp->va_start <= addr) { | |
3147 | va = tmp; | |
3148 | if (tmp->va_end >= addr) | |
3149 | break; | |
3150 | ||
ca23e405 | 3151 | n = n->rb_right; |
68ad4a33 URS |
3152 | } else { |
3153 | n = n->rb_left; | |
3154 | } | |
ca23e405 TH |
3155 | } |
3156 | ||
68ad4a33 | 3157 | return va; |
ca23e405 TH |
3158 | } |
3159 | ||
3160 | /** | |
68ad4a33 URS |
3161 | * pvm_determine_end_from_reverse - find the highest aligned address |
3162 | * of free block below VMALLOC_END | |
3163 | * @va: | |
3164 | * in - the VA we start the search(reverse order); | |
3165 | * out - the VA with the highest aligned end address. | |
ca23e405 | 3166 | * |
68ad4a33 | 3167 | * Returns: determined end address within vmap_area |
ca23e405 | 3168 | */ |
68ad4a33 URS |
3169 | static unsigned long |
3170 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3171 | { |
68ad4a33 | 3172 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3173 | unsigned long addr; |
3174 | ||
68ad4a33 URS |
3175 | if (likely(*va)) { |
3176 | list_for_each_entry_from_reverse((*va), | |
3177 | &free_vmap_area_list, list) { | |
3178 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3179 | if ((*va)->va_start < addr) | |
3180 | return addr; | |
3181 | } | |
ca23e405 TH |
3182 | } |
3183 | ||
68ad4a33 | 3184 | return 0; |
ca23e405 TH |
3185 | } |
3186 | ||
3187 | /** | |
3188 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3189 | * @offsets: array containing offset of each area | |
3190 | * @sizes: array containing size of each area | |
3191 | * @nr_vms: the number of areas to allocate | |
3192 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3193 | * |
3194 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3195 | * vm_structs on success, %NULL on failure | |
3196 | * | |
3197 | * Percpu allocator wants to use congruent vm areas so that it can | |
3198 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3199 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3200 | * be scattered pretty far, distance between two areas easily going up | |
3201 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3202 | * areas are allocated from top. | |
ca23e405 | 3203 | * |
68ad4a33 URS |
3204 | * Despite its complicated look, this allocator is rather simple. It |
3205 | * does everything top-down and scans free blocks from the end looking | |
3206 | * for matching base. While scanning, if any of the areas do not fit the | |
3207 | * base address is pulled down to fit the area. Scanning is repeated till | |
3208 | * all the areas fit and then all necessary data structures are inserted | |
3209 | * and the result is returned. | |
ca23e405 TH |
3210 | */ |
3211 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3212 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3213 | size_t align) |
ca23e405 TH |
3214 | { |
3215 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3216 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3217 | struct vmap_area **vas, *va; |
ca23e405 TH |
3218 | struct vm_struct **vms; |
3219 | int area, area2, last_area, term_area; | |
68ad4a33 | 3220 | unsigned long base, start, size, end, last_end; |
ca23e405 | 3221 | bool purged = false; |
68ad4a33 | 3222 | enum fit_type type; |
ca23e405 | 3223 | |
ca23e405 | 3224 | /* verify parameters and allocate data structures */ |
891c49ab | 3225 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3226 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3227 | start = offsets[area]; | |
3228 | end = start + sizes[area]; | |
3229 | ||
3230 | /* is everything aligned properly? */ | |
3231 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3232 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3233 | ||
3234 | /* detect the area with the highest address */ | |
3235 | if (start > offsets[last_area]) | |
3236 | last_area = area; | |
3237 | ||
c568da28 | 3238 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3239 | unsigned long start2 = offsets[area2]; |
3240 | unsigned long end2 = start2 + sizes[area2]; | |
3241 | ||
c568da28 | 3242 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3243 | } |
3244 | } | |
3245 | last_end = offsets[last_area] + sizes[last_area]; | |
3246 | ||
3247 | if (vmalloc_end - vmalloc_start < last_end) { | |
3248 | WARN_ON(true); | |
3249 | return NULL; | |
3250 | } | |
3251 | ||
4d67d860 TM |
3252 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3253 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3254 | if (!vas || !vms) |
f1db7afd | 3255 | goto err_free2; |
ca23e405 TH |
3256 | |
3257 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3258 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3259 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3260 | if (!vas[area] || !vms[area]) |
3261 | goto err_free; | |
3262 | } | |
3263 | retry: | |
3264 | spin_lock(&vmap_area_lock); | |
3265 | ||
3266 | /* start scanning - we scan from the top, begin with the last area */ | |
3267 | area = term_area = last_area; | |
3268 | start = offsets[area]; | |
3269 | end = start + sizes[area]; | |
3270 | ||
68ad4a33 URS |
3271 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3272 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3273 | |
3274 | while (true) { | |
ca23e405 TH |
3275 | /* |
3276 | * base might have underflowed, add last_end before | |
3277 | * comparing. | |
3278 | */ | |
68ad4a33 URS |
3279 | if (base + last_end < vmalloc_start + last_end) |
3280 | goto overflow; | |
ca23e405 TH |
3281 | |
3282 | /* | |
68ad4a33 | 3283 | * Fitting base has not been found. |
ca23e405 | 3284 | */ |
68ad4a33 URS |
3285 | if (va == NULL) |
3286 | goto overflow; | |
ca23e405 | 3287 | |
5336e52c KS |
3288 | /* |
3289 | * If required width exeeds current VA block, move | |
3290 | * base downwards and then recheck. | |
3291 | */ | |
3292 | if (base + end > va->va_end) { | |
3293 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3294 | term_area = area; | |
3295 | continue; | |
3296 | } | |
3297 | ||
ca23e405 | 3298 | /* |
68ad4a33 | 3299 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3300 | */ |
5336e52c | 3301 | if (base + start < va->va_start) { |
68ad4a33 URS |
3302 | va = node_to_va(rb_prev(&va->rb_node)); |
3303 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3304 | term_area = area; |
3305 | continue; | |
3306 | } | |
3307 | ||
3308 | /* | |
3309 | * This area fits, move on to the previous one. If | |
3310 | * the previous one is the terminal one, we're done. | |
3311 | */ | |
3312 | area = (area + nr_vms - 1) % nr_vms; | |
3313 | if (area == term_area) | |
3314 | break; | |
68ad4a33 | 3315 | |
ca23e405 TH |
3316 | start = offsets[area]; |
3317 | end = start + sizes[area]; | |
68ad4a33 | 3318 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3319 | } |
68ad4a33 | 3320 | |
ca23e405 TH |
3321 | /* we've found a fitting base, insert all va's */ |
3322 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3323 | int ret; |
ca23e405 | 3324 | |
68ad4a33 URS |
3325 | start = base + offsets[area]; |
3326 | size = sizes[area]; | |
ca23e405 | 3327 | |
68ad4a33 URS |
3328 | va = pvm_find_va_enclose_addr(start); |
3329 | if (WARN_ON_ONCE(va == NULL)) | |
3330 | /* It is a BUG(), but trigger recovery instead. */ | |
3331 | goto recovery; | |
3332 | ||
3333 | type = classify_va_fit_type(va, start, size); | |
3334 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3335 | /* It is a BUG(), but trigger recovery instead. */ | |
3336 | goto recovery; | |
3337 | ||
3338 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3339 | if (unlikely(ret)) | |
3340 | goto recovery; | |
3341 | ||
3342 | /* Allocated area. */ | |
3343 | va = vas[area]; | |
3344 | va->va_start = start; | |
3345 | va->va_end = start + size; | |
3346 | ||
3347 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
3348 | } | |
ca23e405 TH |
3349 | |
3350 | spin_unlock(&vmap_area_lock); | |
3351 | ||
3352 | /* insert all vm's */ | |
3353 | for (area = 0; area < nr_vms; area++) | |
3645cb4a ZY |
3354 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
3355 | pcpu_get_vm_areas); | |
ca23e405 TH |
3356 | |
3357 | kfree(vas); | |
3358 | return vms; | |
3359 | ||
68ad4a33 URS |
3360 | recovery: |
3361 | /* Remove previously inserted areas. */ | |
3362 | while (area--) { | |
3363 | __free_vmap_area(vas[area]); | |
3364 | vas[area] = NULL; | |
3365 | } | |
3366 | ||
3367 | overflow: | |
3368 | spin_unlock(&vmap_area_lock); | |
3369 | if (!purged) { | |
3370 | purge_vmap_area_lazy(); | |
3371 | purged = true; | |
3372 | ||
3373 | /* Before "retry", check if we recover. */ | |
3374 | for (area = 0; area < nr_vms; area++) { | |
3375 | if (vas[area]) | |
3376 | continue; | |
3377 | ||
3378 | vas[area] = kmem_cache_zalloc( | |
3379 | vmap_area_cachep, GFP_KERNEL); | |
3380 | if (!vas[area]) | |
3381 | goto err_free; | |
3382 | } | |
3383 | ||
3384 | goto retry; | |
3385 | } | |
3386 | ||
ca23e405 TH |
3387 | err_free: |
3388 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3389 | if (vas[area]) |
3390 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3391 | ||
f1db7afd | 3392 | kfree(vms[area]); |
ca23e405 | 3393 | } |
f1db7afd | 3394 | err_free2: |
ca23e405 TH |
3395 | kfree(vas); |
3396 | kfree(vms); | |
3397 | return NULL; | |
3398 | } | |
3399 | ||
3400 | /** | |
3401 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3402 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3403 | * @nr_vms: the number of allocated areas | |
3404 | * | |
3405 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3406 | */ | |
3407 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3408 | { | |
3409 | int i; | |
3410 | ||
3411 | for (i = 0; i < nr_vms; i++) | |
3412 | free_vm_area(vms[i]); | |
3413 | kfree(vms); | |
3414 | } | |
4f8b02b4 | 3415 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
3416 | |
3417 | #ifdef CONFIG_PROC_FS | |
3418 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 3419 | __acquires(&vmap_area_lock) |
a10aa579 | 3420 | { |
d4033afd | 3421 | spin_lock(&vmap_area_lock); |
3f500069 | 3422 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3423 | } |
3424 | ||
3425 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3426 | { | |
3f500069 | 3427 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3428 | } |
3429 | ||
3430 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 3431 | __releases(&vmap_area_lock) |
a10aa579 | 3432 | { |
d4033afd | 3433 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
3434 | } |
3435 | ||
a47a126a ED |
3436 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3437 | { | |
e5adfffc | 3438 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3439 | unsigned int nr, *counters = m->private; |
3440 | ||
3441 | if (!counters) | |
3442 | return; | |
3443 | ||
af12346c WL |
3444 | if (v->flags & VM_UNINITIALIZED) |
3445 | return; | |
7e5b528b DV |
3446 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3447 | smp_rmb(); | |
af12346c | 3448 | |
a47a126a ED |
3449 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3450 | ||
3451 | for (nr = 0; nr < v->nr_pages; nr++) | |
3452 | counters[page_to_nid(v->pages[nr])]++; | |
3453 | ||
3454 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3455 | if (counters[nr]) | |
3456 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3457 | } | |
3458 | } | |
3459 | ||
dd3b8353 URS |
3460 | static void show_purge_info(struct seq_file *m) |
3461 | { | |
3462 | struct llist_node *head; | |
3463 | struct vmap_area *va; | |
3464 | ||
3465 | head = READ_ONCE(vmap_purge_list.first); | |
3466 | if (head == NULL) | |
3467 | return; | |
3468 | ||
3469 | llist_for_each_entry(va, head, purge_list) { | |
3470 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", | |
3471 | (void *)va->va_start, (void *)va->va_end, | |
3472 | va->va_end - va->va_start); | |
3473 | } | |
3474 | } | |
3475 | ||
a10aa579 CL |
3476 | static int s_show(struct seq_file *m, void *p) |
3477 | { | |
3f500069 | 3478 | struct vmap_area *va; |
d4033afd JK |
3479 | struct vm_struct *v; |
3480 | ||
3f500069 | 3481 | va = list_entry(p, struct vmap_area, list); |
3482 | ||
c2ce8c14 | 3483 | /* |
688fcbfc PL |
3484 | * s_show can encounter race with remove_vm_area, !vm on behalf |
3485 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 3486 | */ |
688fcbfc | 3487 | if (!va->vm) { |
dd3b8353 | 3488 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 3489 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 3490 | va->va_end - va->va_start); |
78c72746 | 3491 | |
d4033afd | 3492 | return 0; |
78c72746 | 3493 | } |
d4033afd JK |
3494 | |
3495 | v = va->vm; | |
a10aa579 | 3496 | |
45ec1690 | 3497 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3498 | v->addr, v->addr + v->size, v->size); |
3499 | ||
62c70bce JP |
3500 | if (v->caller) |
3501 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3502 | |
a10aa579 CL |
3503 | if (v->nr_pages) |
3504 | seq_printf(m, " pages=%d", v->nr_pages); | |
3505 | ||
3506 | if (v->phys_addr) | |
199eaa05 | 3507 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3508 | |
3509 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3510 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3511 | |
3512 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3513 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3514 | |
3515 | if (v->flags & VM_MAP) | |
f4527c90 | 3516 | seq_puts(m, " vmap"); |
a10aa579 CL |
3517 | |
3518 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3519 | seq_puts(m, " user"); |
a10aa579 | 3520 | |
fe9041c2 CH |
3521 | if (v->flags & VM_DMA_COHERENT) |
3522 | seq_puts(m, " dma-coherent"); | |
3523 | ||
244d63ee | 3524 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3525 | seq_puts(m, " vpages"); |
a10aa579 | 3526 | |
a47a126a | 3527 | show_numa_info(m, v); |
a10aa579 | 3528 | seq_putc(m, '\n'); |
dd3b8353 URS |
3529 | |
3530 | /* | |
3531 | * As a final step, dump "unpurged" areas. Note, | |
3532 | * that entire "/proc/vmallocinfo" output will not | |
3533 | * be address sorted, because the purge list is not | |
3534 | * sorted. | |
3535 | */ | |
3536 | if (list_is_last(&va->list, &vmap_area_list)) | |
3537 | show_purge_info(m); | |
3538 | ||
a10aa579 CL |
3539 | return 0; |
3540 | } | |
3541 | ||
5f6a6a9c | 3542 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3543 | .start = s_start, |
3544 | .next = s_next, | |
3545 | .stop = s_stop, | |
3546 | .show = s_show, | |
3547 | }; | |
5f6a6a9c | 3548 | |
5f6a6a9c AD |
3549 | static int __init proc_vmalloc_init(void) |
3550 | { | |
fddda2b7 | 3551 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3552 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3553 | &vmalloc_op, |
3554 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3555 | else |
0825a6f9 | 3556 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3557 | return 0; |
3558 | } | |
3559 | module_init(proc_vmalloc_init); | |
db3808c1 | 3560 | |
a10aa579 | 3561 | #endif |