]>
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 | |
78c72746 | 332 | #define VM_LAZY_FREE 0x02 |
db64fe02 NP |
333 | #define VM_VM_AREA 0x04 |
334 | ||
db64fe02 | 335 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
336 | /* Export for kexec only */ |
337 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 338 | static LLIST_HEAD(vmap_purge_list); |
89699605 | 339 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 340 | static bool vmap_initialized __read_mostly; |
89699605 | 341 | |
68ad4a33 URS |
342 | /* |
343 | * This kmem_cache is used for vmap_area objects. Instead of | |
344 | * allocating from slab we reuse an object from this cache to | |
345 | * make things faster. Especially in "no edge" splitting of | |
346 | * free block. | |
347 | */ | |
348 | static struct kmem_cache *vmap_area_cachep; | |
349 | ||
350 | /* | |
351 | * This linked list is used in pair with free_vmap_area_root. | |
352 | * It gives O(1) access to prev/next to perform fast coalescing. | |
353 | */ | |
354 | static LIST_HEAD(free_vmap_area_list); | |
355 | ||
356 | /* | |
357 | * This augment red-black tree represents the free vmap space. | |
358 | * All vmap_area objects in this tree are sorted by va->va_start | |
359 | * address. It is used for allocation and merging when a vmap | |
360 | * object is released. | |
361 | * | |
362 | * Each vmap_area node contains a maximum available free block | |
363 | * of its sub-tree, right or left. Therefore it is possible to | |
364 | * find a lowest match of free area. | |
365 | */ | |
366 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
367 | ||
82dd23e8 URS |
368 | /* |
369 | * Preload a CPU with one object for "no edge" split case. The | |
370 | * aim is to get rid of allocations from the atomic context, thus | |
371 | * to use more permissive allocation masks. | |
372 | */ | |
373 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
374 | ||
68ad4a33 URS |
375 | static __always_inline unsigned long |
376 | va_size(struct vmap_area *va) | |
377 | { | |
378 | return (va->va_end - va->va_start); | |
379 | } | |
380 | ||
381 | static __always_inline unsigned long | |
382 | get_subtree_max_size(struct rb_node *node) | |
383 | { | |
384 | struct vmap_area *va; | |
385 | ||
386 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
387 | return va ? va->subtree_max_size : 0; | |
388 | } | |
89699605 | 389 | |
68ad4a33 URS |
390 | /* |
391 | * Gets called when remove the node and rotate. | |
392 | */ | |
393 | static __always_inline unsigned long | |
394 | compute_subtree_max_size(struct vmap_area *va) | |
395 | { | |
396 | return max3(va_size(va), | |
397 | get_subtree_max_size(va->rb_node.rb_left), | |
398 | get_subtree_max_size(va->rb_node.rb_right)); | |
399 | } | |
400 | ||
401 | RB_DECLARE_CALLBACKS(static, free_vmap_area_rb_augment_cb, | |
402 | struct vmap_area, rb_node, unsigned long, subtree_max_size, | |
403 | compute_subtree_max_size) | |
404 | ||
405 | static void purge_vmap_area_lazy(void); | |
406 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
407 | static unsigned long lazy_max_pages(void); | |
db64fe02 NP |
408 | |
409 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 410 | { |
db64fe02 NP |
411 | struct rb_node *n = vmap_area_root.rb_node; |
412 | ||
413 | while (n) { | |
414 | struct vmap_area *va; | |
415 | ||
416 | va = rb_entry(n, struct vmap_area, rb_node); | |
417 | if (addr < va->va_start) | |
418 | n = n->rb_left; | |
cef2ac3f | 419 | else if (addr >= va->va_end) |
db64fe02 NP |
420 | n = n->rb_right; |
421 | else | |
422 | return va; | |
423 | } | |
424 | ||
425 | return NULL; | |
426 | } | |
427 | ||
68ad4a33 URS |
428 | /* |
429 | * This function returns back addresses of parent node | |
430 | * and its left or right link for further processing. | |
431 | */ | |
432 | static __always_inline struct rb_node ** | |
433 | find_va_links(struct vmap_area *va, | |
434 | struct rb_root *root, struct rb_node *from, | |
435 | struct rb_node **parent) | |
436 | { | |
437 | struct vmap_area *tmp_va; | |
438 | struct rb_node **link; | |
439 | ||
440 | if (root) { | |
441 | link = &root->rb_node; | |
442 | if (unlikely(!*link)) { | |
443 | *parent = NULL; | |
444 | return link; | |
445 | } | |
446 | } else { | |
447 | link = &from; | |
448 | } | |
db64fe02 | 449 | |
68ad4a33 URS |
450 | /* |
451 | * Go to the bottom of the tree. When we hit the last point | |
452 | * we end up with parent rb_node and correct direction, i name | |
453 | * it link, where the new va->rb_node will be attached to. | |
454 | */ | |
455 | do { | |
456 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 457 | |
68ad4a33 URS |
458 | /* |
459 | * During the traversal we also do some sanity check. | |
460 | * Trigger the BUG() if there are sides(left/right) | |
461 | * or full overlaps. | |
462 | */ | |
463 | if (va->va_start < tmp_va->va_end && | |
464 | va->va_end <= tmp_va->va_start) | |
465 | link = &(*link)->rb_left; | |
466 | else if (va->va_end > tmp_va->va_start && | |
467 | va->va_start >= tmp_va->va_end) | |
468 | link = &(*link)->rb_right; | |
db64fe02 NP |
469 | else |
470 | BUG(); | |
68ad4a33 URS |
471 | } while (*link); |
472 | ||
473 | *parent = &tmp_va->rb_node; | |
474 | return link; | |
475 | } | |
476 | ||
477 | static __always_inline struct list_head * | |
478 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
479 | { | |
480 | struct list_head *list; | |
481 | ||
482 | if (unlikely(!parent)) | |
483 | /* | |
484 | * The red-black tree where we try to find VA neighbors | |
485 | * before merging or inserting is empty, i.e. it means | |
486 | * there is no free vmap space. Normally it does not | |
487 | * happen but we handle this case anyway. | |
488 | */ | |
489 | return NULL; | |
490 | ||
491 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
492 | return (&parent->rb_right == link ? list->next : list); | |
493 | } | |
494 | ||
495 | static __always_inline void | |
496 | link_va(struct vmap_area *va, struct rb_root *root, | |
497 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
498 | { | |
499 | /* | |
500 | * VA is still not in the list, but we can | |
501 | * identify its future previous list_head node. | |
502 | */ | |
503 | if (likely(parent)) { | |
504 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
505 | if (&parent->rb_right != link) | |
506 | head = head->prev; | |
db64fe02 NP |
507 | } |
508 | ||
68ad4a33 URS |
509 | /* Insert to the rb-tree */ |
510 | rb_link_node(&va->rb_node, parent, link); | |
511 | if (root == &free_vmap_area_root) { | |
512 | /* | |
513 | * Some explanation here. Just perform simple insertion | |
514 | * to the tree. We do not set va->subtree_max_size to | |
515 | * its current size before calling rb_insert_augmented(). | |
516 | * It is because of we populate the tree from the bottom | |
517 | * to parent levels when the node _is_ in the tree. | |
518 | * | |
519 | * Therefore we set subtree_max_size to zero after insertion, | |
520 | * to let __augment_tree_propagate_from() puts everything to | |
521 | * the correct order later on. | |
522 | */ | |
523 | rb_insert_augmented(&va->rb_node, | |
524 | root, &free_vmap_area_rb_augment_cb); | |
525 | va->subtree_max_size = 0; | |
526 | } else { | |
527 | rb_insert_color(&va->rb_node, root); | |
528 | } | |
db64fe02 | 529 | |
68ad4a33 URS |
530 | /* Address-sort this list */ |
531 | list_add(&va->list, head); | |
db64fe02 NP |
532 | } |
533 | ||
68ad4a33 URS |
534 | static __always_inline void |
535 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
536 | { | |
460e42d1 URS |
537 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
538 | return; | |
db64fe02 | 539 | |
460e42d1 URS |
540 | if (root == &free_vmap_area_root) |
541 | rb_erase_augmented(&va->rb_node, | |
542 | root, &free_vmap_area_rb_augment_cb); | |
543 | else | |
544 | rb_erase(&va->rb_node, root); | |
545 | ||
546 | list_del(&va->list); | |
547 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
548 | } |
549 | ||
bb850f4d URS |
550 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
551 | static void | |
552 | augment_tree_propagate_check(struct rb_node *n) | |
553 | { | |
554 | struct vmap_area *va; | |
555 | struct rb_node *node; | |
556 | unsigned long size; | |
557 | bool found = false; | |
558 | ||
559 | if (n == NULL) | |
560 | return; | |
561 | ||
562 | va = rb_entry(n, struct vmap_area, rb_node); | |
563 | size = va->subtree_max_size; | |
564 | node = n; | |
565 | ||
566 | while (node) { | |
567 | va = rb_entry(node, struct vmap_area, rb_node); | |
568 | ||
569 | if (get_subtree_max_size(node->rb_left) == size) { | |
570 | node = node->rb_left; | |
571 | } else { | |
572 | if (va_size(va) == size) { | |
573 | found = true; | |
574 | break; | |
575 | } | |
576 | ||
577 | node = node->rb_right; | |
578 | } | |
579 | } | |
580 | ||
581 | if (!found) { | |
582 | va = rb_entry(n, struct vmap_area, rb_node); | |
583 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
584 | va_size(va), va->subtree_max_size); | |
585 | } | |
586 | ||
587 | augment_tree_propagate_check(n->rb_left); | |
588 | augment_tree_propagate_check(n->rb_right); | |
589 | } | |
590 | #endif | |
591 | ||
68ad4a33 URS |
592 | /* |
593 | * This function populates subtree_max_size from bottom to upper | |
594 | * levels starting from VA point. The propagation must be done | |
595 | * when VA size is modified by changing its va_start/va_end. Or | |
596 | * in case of newly inserting of VA to the tree. | |
597 | * | |
598 | * It means that __augment_tree_propagate_from() must be called: | |
599 | * - After VA has been inserted to the tree(free path); | |
600 | * - After VA has been shrunk(allocation path); | |
601 | * - After VA has been increased(merging path). | |
602 | * | |
603 | * Please note that, it does not mean that upper parent nodes | |
604 | * and their subtree_max_size are recalculated all the time up | |
605 | * to the root node. | |
606 | * | |
607 | * 4--8 | |
608 | * /\ | |
609 | * / \ | |
610 | * / \ | |
611 | * 2--2 8--8 | |
612 | * | |
613 | * For example if we modify the node 4, shrinking it to 2, then | |
614 | * no any modification is required. If we shrink the node 2 to 1 | |
615 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
616 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
617 | * node becomes 4--6. | |
618 | */ | |
619 | static __always_inline void | |
620 | augment_tree_propagate_from(struct vmap_area *va) | |
621 | { | |
622 | struct rb_node *node = &va->rb_node; | |
623 | unsigned long new_va_sub_max_size; | |
624 | ||
625 | while (node) { | |
626 | va = rb_entry(node, struct vmap_area, rb_node); | |
627 | new_va_sub_max_size = compute_subtree_max_size(va); | |
628 | ||
629 | /* | |
630 | * If the newly calculated maximum available size of the | |
631 | * subtree is equal to the current one, then it means that | |
632 | * the tree is propagated correctly. So we have to stop at | |
633 | * this point to save cycles. | |
634 | */ | |
635 | if (va->subtree_max_size == new_va_sub_max_size) | |
636 | break; | |
637 | ||
638 | va->subtree_max_size = new_va_sub_max_size; | |
639 | node = rb_parent(&va->rb_node); | |
640 | } | |
bb850f4d URS |
641 | |
642 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
643 | augment_tree_propagate_check(free_vmap_area_root.rb_node); | |
644 | #endif | |
68ad4a33 URS |
645 | } |
646 | ||
647 | static void | |
648 | insert_vmap_area(struct vmap_area *va, | |
649 | struct rb_root *root, struct list_head *head) | |
650 | { | |
651 | struct rb_node **link; | |
652 | struct rb_node *parent; | |
653 | ||
654 | link = find_va_links(va, root, NULL, &parent); | |
655 | link_va(va, root, parent, link, head); | |
656 | } | |
657 | ||
658 | static void | |
659 | insert_vmap_area_augment(struct vmap_area *va, | |
660 | struct rb_node *from, struct rb_root *root, | |
661 | struct list_head *head) | |
662 | { | |
663 | struct rb_node **link; | |
664 | struct rb_node *parent; | |
665 | ||
666 | if (from) | |
667 | link = find_va_links(va, NULL, from, &parent); | |
668 | else | |
669 | link = find_va_links(va, root, NULL, &parent); | |
670 | ||
671 | link_va(va, root, parent, link, head); | |
672 | augment_tree_propagate_from(va); | |
673 | } | |
674 | ||
675 | /* | |
676 | * Merge de-allocated chunk of VA memory with previous | |
677 | * and next free blocks. If coalesce is not done a new | |
678 | * free area is inserted. If VA has been merged, it is | |
679 | * freed. | |
680 | */ | |
681 | static __always_inline void | |
682 | merge_or_add_vmap_area(struct vmap_area *va, | |
683 | struct rb_root *root, struct list_head *head) | |
684 | { | |
685 | struct vmap_area *sibling; | |
686 | struct list_head *next; | |
687 | struct rb_node **link; | |
688 | struct rb_node *parent; | |
689 | bool merged = false; | |
690 | ||
691 | /* | |
692 | * Find a place in the tree where VA potentially will be | |
693 | * inserted, unless it is merged with its sibling/siblings. | |
694 | */ | |
695 | link = find_va_links(va, root, NULL, &parent); | |
696 | ||
697 | /* | |
698 | * Get next node of VA to check if merging can be done. | |
699 | */ | |
700 | next = get_va_next_sibling(parent, link); | |
701 | if (unlikely(next == NULL)) | |
702 | goto insert; | |
703 | ||
704 | /* | |
705 | * start end | |
706 | * | | | |
707 | * |<------VA------>|<-----Next----->| | |
708 | * | | | |
709 | * start end | |
710 | */ | |
711 | if (next != head) { | |
712 | sibling = list_entry(next, struct vmap_area, list); | |
713 | if (sibling->va_start == va->va_end) { | |
714 | sibling->va_start = va->va_start; | |
715 | ||
716 | /* Check and update the tree if needed. */ | |
717 | augment_tree_propagate_from(sibling); | |
718 | ||
68ad4a33 URS |
719 | /* Free vmap_area object. */ |
720 | kmem_cache_free(vmap_area_cachep, va); | |
721 | ||
722 | /* Point to the new merged area. */ | |
723 | va = sibling; | |
724 | merged = true; | |
725 | } | |
726 | } | |
727 | ||
728 | /* | |
729 | * start end | |
730 | * | | | |
731 | * |<-----Prev----->|<------VA------>| | |
732 | * | | | |
733 | * start end | |
734 | */ | |
735 | if (next->prev != head) { | |
736 | sibling = list_entry(next->prev, struct vmap_area, list); | |
737 | if (sibling->va_end == va->va_start) { | |
738 | sibling->va_end = va->va_end; | |
739 | ||
740 | /* Check and update the tree if needed. */ | |
741 | augment_tree_propagate_from(sibling); | |
742 | ||
54f63d9d URS |
743 | if (merged) |
744 | unlink_va(va, root); | |
68ad4a33 URS |
745 | |
746 | /* Free vmap_area object. */ | |
747 | kmem_cache_free(vmap_area_cachep, va); | |
68ad4a33 URS |
748 | return; |
749 | } | |
750 | } | |
751 | ||
752 | insert: | |
753 | if (!merged) { | |
754 | link_va(va, root, parent, link, head); | |
755 | augment_tree_propagate_from(va); | |
756 | } | |
757 | } | |
758 | ||
759 | static __always_inline bool | |
760 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
761 | unsigned long align, unsigned long vstart) | |
762 | { | |
763 | unsigned long nva_start_addr; | |
764 | ||
765 | if (va->va_start > vstart) | |
766 | nva_start_addr = ALIGN(va->va_start, align); | |
767 | else | |
768 | nva_start_addr = ALIGN(vstart, align); | |
769 | ||
770 | /* Can be overflowed due to big size or alignment. */ | |
771 | if (nva_start_addr + size < nva_start_addr || | |
772 | nva_start_addr < vstart) | |
773 | return false; | |
774 | ||
775 | return (nva_start_addr + size <= va->va_end); | |
776 | } | |
777 | ||
778 | /* | |
779 | * Find the first free block(lowest start address) in the tree, | |
780 | * that will accomplish the request corresponding to passing | |
781 | * parameters. | |
782 | */ | |
783 | static __always_inline struct vmap_area * | |
784 | find_vmap_lowest_match(unsigned long size, | |
785 | unsigned long align, unsigned long vstart) | |
786 | { | |
787 | struct vmap_area *va; | |
788 | struct rb_node *node; | |
789 | unsigned long length; | |
790 | ||
791 | /* Start from the root. */ | |
792 | node = free_vmap_area_root.rb_node; | |
793 | ||
794 | /* Adjust the search size for alignment overhead. */ | |
795 | length = size + align - 1; | |
796 | ||
797 | while (node) { | |
798 | va = rb_entry(node, struct vmap_area, rb_node); | |
799 | ||
800 | if (get_subtree_max_size(node->rb_left) >= length && | |
801 | vstart < va->va_start) { | |
802 | node = node->rb_left; | |
803 | } else { | |
804 | if (is_within_this_va(va, size, align, vstart)) | |
805 | return va; | |
806 | ||
807 | /* | |
808 | * Does not make sense to go deeper towards the right | |
809 | * sub-tree if it does not have a free block that is | |
810 | * equal or bigger to the requested search length. | |
811 | */ | |
812 | if (get_subtree_max_size(node->rb_right) >= length) { | |
813 | node = node->rb_right; | |
814 | continue; | |
815 | } | |
816 | ||
817 | /* | |
3806b041 | 818 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
819 | * that will satisfy the search criteria. It can happen |
820 | * only once due to "vstart" restriction. | |
821 | */ | |
822 | while ((node = rb_parent(node))) { | |
823 | va = rb_entry(node, struct vmap_area, rb_node); | |
824 | if (is_within_this_va(va, size, align, vstart)) | |
825 | return va; | |
826 | ||
827 | if (get_subtree_max_size(node->rb_right) >= length && | |
828 | vstart <= va->va_start) { | |
829 | node = node->rb_right; | |
830 | break; | |
831 | } | |
832 | } | |
833 | } | |
834 | } | |
835 | ||
836 | return NULL; | |
837 | } | |
838 | ||
a6cf4e0f URS |
839 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
840 | #include <linux/random.h> | |
841 | ||
842 | static struct vmap_area * | |
843 | find_vmap_lowest_linear_match(unsigned long size, | |
844 | unsigned long align, unsigned long vstart) | |
845 | { | |
846 | struct vmap_area *va; | |
847 | ||
848 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
849 | if (!is_within_this_va(va, size, align, vstart)) | |
850 | continue; | |
851 | ||
852 | return va; | |
853 | } | |
854 | ||
855 | return NULL; | |
856 | } | |
857 | ||
858 | static void | |
859 | find_vmap_lowest_match_check(unsigned long size) | |
860 | { | |
861 | struct vmap_area *va_1, *va_2; | |
862 | unsigned long vstart; | |
863 | unsigned int rnd; | |
864 | ||
865 | get_random_bytes(&rnd, sizeof(rnd)); | |
866 | vstart = VMALLOC_START + rnd; | |
867 | ||
868 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
869 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
870 | ||
871 | if (va_1 != va_2) | |
872 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
873 | va_1, va_2, vstart); | |
874 | } | |
875 | #endif | |
876 | ||
68ad4a33 URS |
877 | enum fit_type { |
878 | NOTHING_FIT = 0, | |
879 | FL_FIT_TYPE = 1, /* full fit */ | |
880 | LE_FIT_TYPE = 2, /* left edge fit */ | |
881 | RE_FIT_TYPE = 3, /* right edge fit */ | |
882 | NE_FIT_TYPE = 4 /* no edge fit */ | |
883 | }; | |
884 | ||
885 | static __always_inline enum fit_type | |
886 | classify_va_fit_type(struct vmap_area *va, | |
887 | unsigned long nva_start_addr, unsigned long size) | |
888 | { | |
889 | enum fit_type type; | |
890 | ||
891 | /* Check if it is within VA. */ | |
892 | if (nva_start_addr < va->va_start || | |
893 | nva_start_addr + size > va->va_end) | |
894 | return NOTHING_FIT; | |
895 | ||
896 | /* Now classify. */ | |
897 | if (va->va_start == nva_start_addr) { | |
898 | if (va->va_end == nva_start_addr + size) | |
899 | type = FL_FIT_TYPE; | |
900 | else | |
901 | type = LE_FIT_TYPE; | |
902 | } else if (va->va_end == nva_start_addr + size) { | |
903 | type = RE_FIT_TYPE; | |
904 | } else { | |
905 | type = NE_FIT_TYPE; | |
906 | } | |
907 | ||
908 | return type; | |
909 | } | |
910 | ||
911 | static __always_inline int | |
912 | adjust_va_to_fit_type(struct vmap_area *va, | |
913 | unsigned long nva_start_addr, unsigned long size, | |
914 | enum fit_type type) | |
915 | { | |
2c929233 | 916 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
917 | |
918 | if (type == FL_FIT_TYPE) { | |
919 | /* | |
920 | * No need to split VA, it fully fits. | |
921 | * | |
922 | * | | | |
923 | * V NVA V | |
924 | * |---------------| | |
925 | */ | |
926 | unlink_va(va, &free_vmap_area_root); | |
927 | kmem_cache_free(vmap_area_cachep, va); | |
928 | } else if (type == LE_FIT_TYPE) { | |
929 | /* | |
930 | * Split left edge of fit VA. | |
931 | * | |
932 | * | | | |
933 | * V NVA V R | |
934 | * |-------|-------| | |
935 | */ | |
936 | va->va_start += size; | |
937 | } else if (type == RE_FIT_TYPE) { | |
938 | /* | |
939 | * Split right edge of fit VA. | |
940 | * | |
941 | * | | | |
942 | * L V NVA V | |
943 | * |-------|-------| | |
944 | */ | |
945 | va->va_end = nva_start_addr; | |
946 | } else if (type == NE_FIT_TYPE) { | |
947 | /* | |
948 | * Split no edge of fit VA. | |
949 | * | |
950 | * | | | |
951 | * L V NVA V R | |
952 | * |---|-------|---| | |
953 | */ | |
82dd23e8 URS |
954 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
955 | if (unlikely(!lva)) { | |
956 | /* | |
957 | * For percpu allocator we do not do any pre-allocation | |
958 | * and leave it as it is. The reason is it most likely | |
959 | * never ends up with NE_FIT_TYPE splitting. In case of | |
960 | * percpu allocations offsets and sizes are aligned to | |
961 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
962 | * are its main fitting cases. | |
963 | * | |
964 | * There are a few exceptions though, as an example it is | |
965 | * a first allocation (early boot up) when we have "one" | |
966 | * big free space that has to be split. | |
967 | */ | |
968 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
969 | if (!lva) | |
970 | return -1; | |
971 | } | |
68ad4a33 URS |
972 | |
973 | /* | |
974 | * Build the remainder. | |
975 | */ | |
976 | lva->va_start = va->va_start; | |
977 | lva->va_end = nva_start_addr; | |
978 | ||
979 | /* | |
980 | * Shrink this VA to remaining size. | |
981 | */ | |
982 | va->va_start = nva_start_addr + size; | |
983 | } else { | |
984 | return -1; | |
985 | } | |
986 | ||
987 | if (type != FL_FIT_TYPE) { | |
988 | augment_tree_propagate_from(va); | |
989 | ||
2c929233 | 990 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
991 | insert_vmap_area_augment(lva, &va->rb_node, |
992 | &free_vmap_area_root, &free_vmap_area_list); | |
993 | } | |
994 | ||
995 | return 0; | |
996 | } | |
997 | ||
998 | /* | |
999 | * Returns a start address of the newly allocated area, if success. | |
1000 | * Otherwise a vend is returned that indicates failure. | |
1001 | */ | |
1002 | static __always_inline unsigned long | |
1003 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1004 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1005 | { |
1006 | unsigned long nva_start_addr; | |
1007 | struct vmap_area *va; | |
1008 | enum fit_type type; | |
1009 | int ret; | |
1010 | ||
1011 | va = find_vmap_lowest_match(size, align, vstart); | |
1012 | if (unlikely(!va)) | |
1013 | return vend; | |
1014 | ||
1015 | if (va->va_start > vstart) | |
1016 | nva_start_addr = ALIGN(va->va_start, align); | |
1017 | else | |
1018 | nva_start_addr = ALIGN(vstart, align); | |
1019 | ||
1020 | /* Check the "vend" restriction. */ | |
1021 | if (nva_start_addr + size > vend) | |
1022 | return vend; | |
1023 | ||
1024 | /* Classify what we have found. */ | |
1025 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1026 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1027 | return vend; | |
1028 | ||
1029 | /* Update the free vmap_area. */ | |
1030 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1031 | if (ret) | |
1032 | return vend; | |
1033 | ||
a6cf4e0f URS |
1034 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1035 | find_vmap_lowest_match_check(size); | |
1036 | #endif | |
1037 | ||
68ad4a33 URS |
1038 | return nva_start_addr; |
1039 | } | |
4da56b99 | 1040 | |
db64fe02 NP |
1041 | /* |
1042 | * Allocate a region of KVA of the specified size and alignment, within the | |
1043 | * vstart and vend. | |
1044 | */ | |
1045 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1046 | unsigned long align, | |
1047 | unsigned long vstart, unsigned long vend, | |
1048 | int node, gfp_t gfp_mask) | |
1049 | { | |
82dd23e8 | 1050 | struct vmap_area *va, *pva; |
1da177e4 | 1051 | unsigned long addr; |
db64fe02 NP |
1052 | int purged = 0; |
1053 | ||
7766970c | 1054 | BUG_ON(!size); |
891c49ab | 1055 | BUG_ON(offset_in_page(size)); |
89699605 | 1056 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1057 | |
68ad4a33 URS |
1058 | if (unlikely(!vmap_initialized)) |
1059 | return ERR_PTR(-EBUSY); | |
1060 | ||
5803ed29 | 1061 | might_sleep(); |
4da56b99 | 1062 | |
68ad4a33 | 1063 | va = kmem_cache_alloc_node(vmap_area_cachep, |
db64fe02 NP |
1064 | gfp_mask & GFP_RECLAIM_MASK, node); |
1065 | if (unlikely(!va)) | |
1066 | return ERR_PTR(-ENOMEM); | |
1067 | ||
7f88f88f CM |
1068 | /* |
1069 | * Only scan the relevant parts containing pointers to other objects | |
1070 | * to avoid false negatives. | |
1071 | */ | |
1072 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK); | |
1073 | ||
db64fe02 | 1074 | retry: |
82dd23e8 URS |
1075 | /* |
1076 | * Preload this CPU with one extra vmap_area object to ensure | |
1077 | * that we have it available when fit type of free area is | |
1078 | * NE_FIT_TYPE. | |
1079 | * | |
1080 | * The preload is done in non-atomic context, thus it allows us | |
1081 | * to use more permissive allocation masks to be more stable under | |
1082 | * low memory condition and high memory pressure. | |
1083 | * | |
1084 | * Even if it fails we do not really care about that. Just proceed | |
1085 | * as it is. "overflow" path will refill the cache we allocate from. | |
1086 | */ | |
1087 | preempt_disable(); | |
1088 | if (!__this_cpu_read(ne_fit_preload_node)) { | |
1089 | preempt_enable(); | |
1090 | pva = kmem_cache_alloc_node(vmap_area_cachep, GFP_KERNEL, node); | |
1091 | preempt_disable(); | |
1092 | ||
1093 | if (__this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) { | |
1094 | if (pva) | |
1095 | kmem_cache_free(vmap_area_cachep, pva); | |
1096 | } | |
1097 | } | |
1098 | ||
db64fe02 | 1099 | spin_lock(&vmap_area_lock); |
82dd23e8 | 1100 | preempt_enable(); |
89699605 | 1101 | |
afd07389 | 1102 | /* |
68ad4a33 URS |
1103 | * If an allocation fails, the "vend" address is |
1104 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1105 | */ |
cacca6ba | 1106 | addr = __alloc_vmap_area(size, align, vstart, vend); |
68ad4a33 | 1107 | if (unlikely(addr == vend)) |
89699605 | 1108 | goto overflow; |
db64fe02 NP |
1109 | |
1110 | va->va_start = addr; | |
1111 | va->va_end = addr + size; | |
1112 | va->flags = 0; | |
68ad4a33 URS |
1113 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
1114 | ||
db64fe02 NP |
1115 | spin_unlock(&vmap_area_lock); |
1116 | ||
61e16557 | 1117 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1118 | BUG_ON(va->va_start < vstart); |
1119 | BUG_ON(va->va_end > vend); | |
1120 | ||
db64fe02 | 1121 | return va; |
89699605 NP |
1122 | |
1123 | overflow: | |
1124 | spin_unlock(&vmap_area_lock); | |
1125 | if (!purged) { | |
1126 | purge_vmap_area_lazy(); | |
1127 | purged = 1; | |
1128 | goto retry; | |
1129 | } | |
4da56b99 CW |
1130 | |
1131 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1132 | unsigned long freed = 0; | |
1133 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1134 | if (freed > 0) { | |
1135 | purged = 0; | |
1136 | goto retry; | |
1137 | } | |
1138 | } | |
1139 | ||
03497d76 | 1140 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1141 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1142 | size); | |
68ad4a33 URS |
1143 | |
1144 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1145 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1146 | } |
1147 | ||
4da56b99 CW |
1148 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1149 | { | |
1150 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1151 | } | |
1152 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1153 | ||
1154 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1155 | { | |
1156 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1157 | } | |
1158 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1159 | ||
db64fe02 NP |
1160 | static void __free_vmap_area(struct vmap_area *va) |
1161 | { | |
ca23e405 | 1162 | /* |
68ad4a33 | 1163 | * Remove from the busy tree/list. |
ca23e405 | 1164 | */ |
68ad4a33 | 1165 | unlink_va(va, &vmap_area_root); |
ca23e405 | 1166 | |
68ad4a33 URS |
1167 | /* |
1168 | * Merge VA with its neighbors, otherwise just add it. | |
1169 | */ | |
1170 | merge_or_add_vmap_area(va, | |
1171 | &free_vmap_area_root, &free_vmap_area_list); | |
db64fe02 NP |
1172 | } |
1173 | ||
1174 | /* | |
1175 | * Free a region of KVA allocated by alloc_vmap_area | |
1176 | */ | |
1177 | static void free_vmap_area(struct vmap_area *va) | |
1178 | { | |
1179 | spin_lock(&vmap_area_lock); | |
1180 | __free_vmap_area(va); | |
1181 | spin_unlock(&vmap_area_lock); | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * Clear the pagetable entries of a given vmap_area | |
1186 | */ | |
1187 | static void unmap_vmap_area(struct vmap_area *va) | |
1188 | { | |
1189 | vunmap_page_range(va->va_start, va->va_end); | |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1194 | * before attempting to purge with a TLB flush. | |
1195 | * | |
1196 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1197 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1198 | * global TLB flushes that must be performed. It would seem natural to scale | |
1199 | * this number up linearly with the number of CPUs (because vmapping activity | |
1200 | * could also scale linearly with the number of CPUs), however it is likely | |
1201 | * that in practice, workloads might be constrained in other ways that mean | |
1202 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1203 | * conservative and not introduce a big latency on huge systems, so go with | |
1204 | * a less aggressive log scale. It will still be an improvement over the old | |
1205 | * code, and it will be simple to change the scale factor if we find that it | |
1206 | * becomes a problem on bigger systems. | |
1207 | */ | |
1208 | static unsigned long lazy_max_pages(void) | |
1209 | { | |
1210 | unsigned int log; | |
1211 | ||
1212 | log = fls(num_online_cpus()); | |
1213 | ||
1214 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1215 | } | |
1216 | ||
4d36e6f8 | 1217 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1218 | |
0574ecd1 CH |
1219 | /* |
1220 | * Serialize vmap purging. There is no actual criticial section protected | |
1221 | * by this look, but we want to avoid concurrent calls for performance | |
1222 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1223 | */ | |
f9e09977 | 1224 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1225 | |
02b709df NP |
1226 | /* for per-CPU blocks */ |
1227 | static void purge_fragmented_blocks_allcpus(void); | |
1228 | ||
3ee48b6a CW |
1229 | /* |
1230 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1231 | * immediately freed. | |
1232 | */ | |
1233 | void set_iounmap_nonlazy(void) | |
1234 | { | |
4d36e6f8 | 1235 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1236 | } |
1237 | ||
db64fe02 NP |
1238 | /* |
1239 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1240 | */ |
0574ecd1 | 1241 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1242 | { |
4d36e6f8 | 1243 | unsigned long resched_threshold; |
80c4bd7a | 1244 | struct llist_node *valist; |
db64fe02 | 1245 | struct vmap_area *va; |
cbb76676 | 1246 | struct vmap_area *n_va; |
db64fe02 | 1247 | |
0574ecd1 | 1248 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1249 | |
80c4bd7a | 1250 | valist = llist_del_all(&vmap_purge_list); |
68571be9 URS |
1251 | if (unlikely(valist == NULL)) |
1252 | return false; | |
1253 | ||
1254 | /* | |
1255 | * TODO: to calculate a flush range without looping. | |
1256 | * The list can be up to lazy_max_pages() elements. | |
1257 | */ | |
80c4bd7a | 1258 | llist_for_each_entry(va, valist, purge_list) { |
0574ecd1 CH |
1259 | if (va->va_start < start) |
1260 | start = va->va_start; | |
1261 | if (va->va_end > end) | |
1262 | end = va->va_end; | |
db64fe02 | 1263 | } |
db64fe02 | 1264 | |
0574ecd1 | 1265 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1266 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1267 | |
0574ecd1 | 1268 | spin_lock(&vmap_area_lock); |
763b218d | 1269 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
4d36e6f8 | 1270 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
763b218d | 1271 | |
0574ecd1 | 1272 | __free_vmap_area(va); |
4d36e6f8 | 1273 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1274 | |
4d36e6f8 | 1275 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
68571be9 | 1276 | cond_resched_lock(&vmap_area_lock); |
763b218d | 1277 | } |
0574ecd1 CH |
1278 | spin_unlock(&vmap_area_lock); |
1279 | return true; | |
db64fe02 NP |
1280 | } |
1281 | ||
496850e5 NP |
1282 | /* |
1283 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1284 | * is already purging. | |
1285 | */ | |
1286 | static void try_purge_vmap_area_lazy(void) | |
1287 | { | |
f9e09977 | 1288 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1289 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1290 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1291 | } |
496850e5 NP |
1292 | } |
1293 | ||
db64fe02 NP |
1294 | /* |
1295 | * Kick off a purge of the outstanding lazy areas. | |
1296 | */ | |
1297 | static void purge_vmap_area_lazy(void) | |
1298 | { | |
f9e09977 | 1299 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1300 | purge_fragmented_blocks_allcpus(); |
1301 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1302 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1303 | } |
1304 | ||
1305 | /* | |
64141da5 JF |
1306 | * Free a vmap area, caller ensuring that the area has been unmapped |
1307 | * and flush_cache_vunmap had been called for the correct range | |
1308 | * previously. | |
db64fe02 | 1309 | */ |
64141da5 | 1310 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1311 | { |
4d36e6f8 | 1312 | unsigned long nr_lazy; |
80c4bd7a | 1313 | |
4d36e6f8 URS |
1314 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1315 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a CW |
1316 | |
1317 | /* After this point, we may free va at any time */ | |
1318 | llist_add(&va->purge_list, &vmap_purge_list); | |
1319 | ||
1320 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 1321 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1322 | } |
1323 | ||
b29acbdc NP |
1324 | /* |
1325 | * Free and unmap a vmap area | |
1326 | */ | |
1327 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1328 | { | |
1329 | flush_cache_vunmap(va->va_start, va->va_end); | |
c8eef01e | 1330 | unmap_vmap_area(va); |
82a2e924 CP |
1331 | if (debug_pagealloc_enabled()) |
1332 | flush_tlb_kernel_range(va->va_start, va->va_end); | |
1333 | ||
c8eef01e | 1334 | free_vmap_area_noflush(va); |
b29acbdc NP |
1335 | } |
1336 | ||
db64fe02 NP |
1337 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1338 | { | |
1339 | struct vmap_area *va; | |
1340 | ||
1341 | spin_lock(&vmap_area_lock); | |
1342 | va = __find_vmap_area(addr); | |
1343 | spin_unlock(&vmap_area_lock); | |
1344 | ||
1345 | return va; | |
1346 | } | |
1347 | ||
db64fe02 NP |
1348 | /*** Per cpu kva allocator ***/ |
1349 | ||
1350 | /* | |
1351 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1352 | * room for at least 16 percpu vmap blocks per CPU. | |
1353 | */ | |
1354 | /* | |
1355 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1356 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1357 | * instead (we just need a rough idea) | |
1358 | */ | |
1359 | #if BITS_PER_LONG == 32 | |
1360 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1361 | #else | |
1362 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1363 | #endif | |
1364 | ||
1365 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1366 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1367 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1368 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1369 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1370 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1371 | #define VMAP_BBMAP_BITS \ |
1372 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1373 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1374 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1375 | |
1376 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1377 | ||
1378 | struct vmap_block_queue { | |
1379 | spinlock_t lock; | |
1380 | struct list_head free; | |
db64fe02 NP |
1381 | }; |
1382 | ||
1383 | struct vmap_block { | |
1384 | spinlock_t lock; | |
1385 | struct vmap_area *va; | |
db64fe02 | 1386 | unsigned long free, dirty; |
7d61bfe8 | 1387 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1388 | struct list_head free_list; |
1389 | struct rcu_head rcu_head; | |
02b709df | 1390 | struct list_head purge; |
db64fe02 NP |
1391 | }; |
1392 | ||
1393 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1394 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1395 | ||
1396 | /* | |
1397 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
1398 | * in the free path. Could get rid of this if we change the API to return a | |
1399 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1400 | */ | |
1401 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
1402 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
1403 | ||
1404 | /* | |
1405 | * We should probably have a fallback mechanism to allocate virtual memory | |
1406 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1407 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1408 | * big problem. | |
1409 | */ | |
1410 | ||
1411 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1412 | { | |
1413 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1414 | addr /= VMAP_BLOCK_SIZE; | |
1415 | return addr; | |
1416 | } | |
1417 | ||
cf725ce2 RP |
1418 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1419 | { | |
1420 | unsigned long addr; | |
1421 | ||
1422 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1423 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1424 | return (void *)addr; | |
1425 | } | |
1426 | ||
1427 | /** | |
1428 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1429 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1430 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1431 | * @gfp_mask: flags for the page level allocator | |
1432 | * | |
a862f68a | 1433 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1434 | */ |
1435 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1436 | { |
1437 | struct vmap_block_queue *vbq; | |
1438 | struct vmap_block *vb; | |
1439 | struct vmap_area *va; | |
1440 | unsigned long vb_idx; | |
1441 | int node, err; | |
cf725ce2 | 1442 | void *vaddr; |
db64fe02 NP |
1443 | |
1444 | node = numa_node_id(); | |
1445 | ||
1446 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1447 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1448 | if (unlikely(!vb)) | |
1449 | return ERR_PTR(-ENOMEM); | |
1450 | ||
1451 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1452 | VMALLOC_START, VMALLOC_END, | |
1453 | node, gfp_mask); | |
ddf9c6d4 | 1454 | if (IS_ERR(va)) { |
db64fe02 | 1455 | kfree(vb); |
e7d86340 | 1456 | return ERR_CAST(va); |
db64fe02 NP |
1457 | } |
1458 | ||
1459 | err = radix_tree_preload(gfp_mask); | |
1460 | if (unlikely(err)) { | |
1461 | kfree(vb); | |
1462 | free_vmap_area(va); | |
1463 | return ERR_PTR(err); | |
1464 | } | |
1465 | ||
cf725ce2 | 1466 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1467 | spin_lock_init(&vb->lock); |
1468 | vb->va = va; | |
cf725ce2 RP |
1469 | /* At least something should be left free */ |
1470 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1471 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1472 | vb->dirty = 0; |
7d61bfe8 RP |
1473 | vb->dirty_min = VMAP_BBMAP_BITS; |
1474 | vb->dirty_max = 0; | |
db64fe02 | 1475 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1476 | |
1477 | vb_idx = addr_to_vb_idx(va->va_start); | |
1478 | spin_lock(&vmap_block_tree_lock); | |
1479 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
1480 | spin_unlock(&vmap_block_tree_lock); | |
1481 | BUG_ON(err); | |
1482 | radix_tree_preload_end(); | |
1483 | ||
1484 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1485 | spin_lock(&vbq->lock); |
68ac546f | 1486 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1487 | spin_unlock(&vbq->lock); |
3f04ba85 | 1488 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1489 | |
cf725ce2 | 1490 | return vaddr; |
db64fe02 NP |
1491 | } |
1492 | ||
db64fe02 NP |
1493 | static void free_vmap_block(struct vmap_block *vb) |
1494 | { | |
1495 | struct vmap_block *tmp; | |
1496 | unsigned long vb_idx; | |
1497 | ||
db64fe02 NP |
1498 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
1499 | spin_lock(&vmap_block_tree_lock); | |
1500 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
1501 | spin_unlock(&vmap_block_tree_lock); | |
1502 | BUG_ON(tmp != vb); | |
1503 | ||
64141da5 | 1504 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1505 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1506 | } |
1507 | ||
02b709df NP |
1508 | static void purge_fragmented_blocks(int cpu) |
1509 | { | |
1510 | LIST_HEAD(purge); | |
1511 | struct vmap_block *vb; | |
1512 | struct vmap_block *n_vb; | |
1513 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1514 | ||
1515 | rcu_read_lock(); | |
1516 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1517 | ||
1518 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1519 | continue; | |
1520 | ||
1521 | spin_lock(&vb->lock); | |
1522 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1523 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1524 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1525 | vb->dirty_min = 0; |
1526 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1527 | spin_lock(&vbq->lock); |
1528 | list_del_rcu(&vb->free_list); | |
1529 | spin_unlock(&vbq->lock); | |
1530 | spin_unlock(&vb->lock); | |
1531 | list_add_tail(&vb->purge, &purge); | |
1532 | } else | |
1533 | spin_unlock(&vb->lock); | |
1534 | } | |
1535 | rcu_read_unlock(); | |
1536 | ||
1537 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1538 | list_del(&vb->purge); | |
1539 | free_vmap_block(vb); | |
1540 | } | |
1541 | } | |
1542 | ||
02b709df NP |
1543 | static void purge_fragmented_blocks_allcpus(void) |
1544 | { | |
1545 | int cpu; | |
1546 | ||
1547 | for_each_possible_cpu(cpu) | |
1548 | purge_fragmented_blocks(cpu); | |
1549 | } | |
1550 | ||
db64fe02 NP |
1551 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1552 | { | |
1553 | struct vmap_block_queue *vbq; | |
1554 | struct vmap_block *vb; | |
cf725ce2 | 1555 | void *vaddr = NULL; |
db64fe02 NP |
1556 | unsigned int order; |
1557 | ||
891c49ab | 1558 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1559 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1560 | if (WARN_ON(size == 0)) { |
1561 | /* | |
1562 | * Allocating 0 bytes isn't what caller wants since | |
1563 | * get_order(0) returns funny result. Just warn and terminate | |
1564 | * early. | |
1565 | */ | |
1566 | return NULL; | |
1567 | } | |
db64fe02 NP |
1568 | order = get_order(size); |
1569 | ||
db64fe02 NP |
1570 | rcu_read_lock(); |
1571 | vbq = &get_cpu_var(vmap_block_queue); | |
1572 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1573 | unsigned long pages_off; |
db64fe02 NP |
1574 | |
1575 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1576 | if (vb->free < (1UL << order)) { |
1577 | spin_unlock(&vb->lock); | |
1578 | continue; | |
1579 | } | |
02b709df | 1580 | |
cf725ce2 RP |
1581 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1582 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1583 | vb->free -= 1UL << order; |
1584 | if (vb->free == 0) { | |
1585 | spin_lock(&vbq->lock); | |
1586 | list_del_rcu(&vb->free_list); | |
1587 | spin_unlock(&vbq->lock); | |
1588 | } | |
cf725ce2 | 1589 | |
02b709df NP |
1590 | spin_unlock(&vb->lock); |
1591 | break; | |
db64fe02 | 1592 | } |
02b709df | 1593 | |
3f04ba85 | 1594 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1595 | rcu_read_unlock(); |
1596 | ||
cf725ce2 RP |
1597 | /* Allocate new block if nothing was found */ |
1598 | if (!vaddr) | |
1599 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1600 | |
cf725ce2 | 1601 | return vaddr; |
db64fe02 NP |
1602 | } |
1603 | ||
1604 | static void vb_free(const void *addr, unsigned long size) | |
1605 | { | |
1606 | unsigned long offset; | |
1607 | unsigned long vb_idx; | |
1608 | unsigned int order; | |
1609 | struct vmap_block *vb; | |
1610 | ||
891c49ab | 1611 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1612 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc NP |
1613 | |
1614 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
1615 | ||
db64fe02 NP |
1616 | order = get_order(size); |
1617 | ||
1618 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
7d61bfe8 | 1619 | offset >>= PAGE_SHIFT; |
db64fe02 NP |
1620 | |
1621 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
1622 | rcu_read_lock(); | |
1623 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1624 | rcu_read_unlock(); | |
1625 | BUG_ON(!vb); | |
1626 | ||
64141da5 JF |
1627 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
1628 | ||
82a2e924 CP |
1629 | if (debug_pagealloc_enabled()) |
1630 | flush_tlb_kernel_range((unsigned long)addr, | |
1631 | (unsigned long)addr + size); | |
1632 | ||
db64fe02 | 1633 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1634 | |
1635 | /* Expand dirty range */ | |
1636 | vb->dirty_min = min(vb->dirty_min, offset); | |
1637 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1638 | |
db64fe02 NP |
1639 | vb->dirty += 1UL << order; |
1640 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1641 | BUG_ON(vb->free); |
db64fe02 NP |
1642 | spin_unlock(&vb->lock); |
1643 | free_vmap_block(vb); | |
1644 | } else | |
1645 | spin_unlock(&vb->lock); | |
1646 | } | |
1647 | ||
868b104d | 1648 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1649 | { |
db64fe02 | 1650 | int cpu; |
db64fe02 | 1651 | |
9b463334 JF |
1652 | if (unlikely(!vmap_initialized)) |
1653 | return; | |
1654 | ||
5803ed29 CH |
1655 | might_sleep(); |
1656 | ||
db64fe02 NP |
1657 | for_each_possible_cpu(cpu) { |
1658 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1659 | struct vmap_block *vb; | |
1660 | ||
1661 | rcu_read_lock(); | |
1662 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1663 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1664 | if (vb->dirty) { |
1665 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1666 | unsigned long s, e; |
b136be5e | 1667 | |
7d61bfe8 RP |
1668 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1669 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1670 | |
7d61bfe8 RP |
1671 | start = min(s, start); |
1672 | end = max(e, end); | |
db64fe02 | 1673 | |
7d61bfe8 | 1674 | flush = 1; |
db64fe02 NP |
1675 | } |
1676 | spin_unlock(&vb->lock); | |
1677 | } | |
1678 | rcu_read_unlock(); | |
1679 | } | |
1680 | ||
f9e09977 | 1681 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1682 | purge_fragmented_blocks_allcpus(); |
1683 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1684 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1685 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 1686 | } |
868b104d RE |
1687 | |
1688 | /** | |
1689 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1690 | * | |
1691 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1692 | * to amortize TLB flushing overheads. What this means is that any page you | |
1693 | * have now, may, in a former life, have been mapped into kernel virtual | |
1694 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1695 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1696 | * | |
1697 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1698 | * be sure that none of the pages we have control over will have any aliases | |
1699 | * from the vmap layer. | |
1700 | */ | |
1701 | void vm_unmap_aliases(void) | |
1702 | { | |
1703 | unsigned long start = ULONG_MAX, end = 0; | |
1704 | int flush = 0; | |
1705 | ||
1706 | _vm_unmap_aliases(start, end, flush); | |
1707 | } | |
db64fe02 NP |
1708 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
1709 | ||
1710 | /** | |
1711 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1712 | * @mem: the pointer returned by vm_map_ram | |
1713 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1714 | */ | |
1715 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1716 | { | |
65ee03c4 | 1717 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1718 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1719 | struct vmap_area *va; |
db64fe02 | 1720 | |
5803ed29 | 1721 | might_sleep(); |
db64fe02 NP |
1722 | BUG_ON(!addr); |
1723 | BUG_ON(addr < VMALLOC_START); | |
1724 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1725 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 1726 | |
9c3acf60 | 1727 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 1728 | debug_check_no_locks_freed(mem, size); |
db64fe02 | 1729 | vb_free(mem, size); |
9c3acf60 CH |
1730 | return; |
1731 | } | |
1732 | ||
1733 | va = find_vmap_area(addr); | |
1734 | BUG_ON(!va); | |
05e3ff95 CP |
1735 | debug_check_no_locks_freed((void *)va->va_start, |
1736 | (va->va_end - va->va_start)); | |
9c3acf60 | 1737 | free_unmap_vmap_area(va); |
db64fe02 NP |
1738 | } |
1739 | EXPORT_SYMBOL(vm_unmap_ram); | |
1740 | ||
1741 | /** | |
1742 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1743 | * @pages: an array of pointers to the pages to be mapped | |
1744 | * @count: number of pages | |
1745 | * @node: prefer to allocate data structures on this node | |
1746 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1747 | * |
36437638 GK |
1748 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1749 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1750 | * objects with vm_map_ram(), it could consume lots of address space through | |
1751 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1752 | * the end. Please use this function for short-lived objects. | |
1753 | * | |
e99c97ad | 1754 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1755 | */ |
1756 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1757 | { | |
65ee03c4 | 1758 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1759 | unsigned long addr; |
1760 | void *mem; | |
1761 | ||
1762 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1763 | mem = vb_alloc(size, GFP_KERNEL); | |
1764 | if (IS_ERR(mem)) | |
1765 | return NULL; | |
1766 | addr = (unsigned long)mem; | |
1767 | } else { | |
1768 | struct vmap_area *va; | |
1769 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1770 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1771 | if (IS_ERR(va)) | |
1772 | return NULL; | |
1773 | ||
1774 | addr = va->va_start; | |
1775 | mem = (void *)addr; | |
1776 | } | |
1777 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1778 | vm_unmap_ram(mem, count); | |
1779 | return NULL; | |
1780 | } | |
1781 | return mem; | |
1782 | } | |
1783 | EXPORT_SYMBOL(vm_map_ram); | |
1784 | ||
4341fa45 | 1785 | static struct vm_struct *vmlist __initdata; |
92eac168 | 1786 | |
be9b7335 NP |
1787 | /** |
1788 | * vm_area_add_early - add vmap area early during boot | |
1789 | * @vm: vm_struct to add | |
1790 | * | |
1791 | * This function is used to add fixed kernel vm area to vmlist before | |
1792 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1793 | * should contain proper values and the other fields should be zero. | |
1794 | * | |
1795 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1796 | */ | |
1797 | void __init vm_area_add_early(struct vm_struct *vm) | |
1798 | { | |
1799 | struct vm_struct *tmp, **p; | |
1800 | ||
1801 | BUG_ON(vmap_initialized); | |
1802 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1803 | if (tmp->addr >= vm->addr) { | |
1804 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1805 | break; | |
1806 | } else | |
1807 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1808 | } | |
1809 | vm->next = *p; | |
1810 | *p = vm; | |
1811 | } | |
1812 | ||
f0aa6617 TH |
1813 | /** |
1814 | * vm_area_register_early - register vmap area early during boot | |
1815 | * @vm: vm_struct to register | |
c0c0a293 | 1816 | * @align: requested alignment |
f0aa6617 TH |
1817 | * |
1818 | * This function is used to register kernel vm area before | |
1819 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1820 | * proper values on entry and other fields should be zero. On return, | |
1821 | * vm->addr contains the allocated address. | |
1822 | * | |
1823 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1824 | */ | |
c0c0a293 | 1825 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1826 | { |
1827 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1828 | unsigned long addr; |
1829 | ||
1830 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1831 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1832 | |
c0c0a293 | 1833 | vm->addr = (void *)addr; |
f0aa6617 | 1834 | |
be9b7335 | 1835 | vm_area_add_early(vm); |
f0aa6617 TH |
1836 | } |
1837 | ||
68ad4a33 URS |
1838 | static void vmap_init_free_space(void) |
1839 | { | |
1840 | unsigned long vmap_start = 1; | |
1841 | const unsigned long vmap_end = ULONG_MAX; | |
1842 | struct vmap_area *busy, *free; | |
1843 | ||
1844 | /* | |
1845 | * B F B B B F | |
1846 | * -|-----|.....|-----|-----|-----|.....|- | |
1847 | * | The KVA space | | |
1848 | * |<--------------------------------->| | |
1849 | */ | |
1850 | list_for_each_entry(busy, &vmap_area_list, list) { | |
1851 | if (busy->va_start - vmap_start > 0) { | |
1852 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1853 | if (!WARN_ON_ONCE(!free)) { | |
1854 | free->va_start = vmap_start; | |
1855 | free->va_end = busy->va_start; | |
1856 | ||
1857 | insert_vmap_area_augment(free, NULL, | |
1858 | &free_vmap_area_root, | |
1859 | &free_vmap_area_list); | |
1860 | } | |
1861 | } | |
1862 | ||
1863 | vmap_start = busy->va_end; | |
1864 | } | |
1865 | ||
1866 | if (vmap_end - vmap_start > 0) { | |
1867 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1868 | if (!WARN_ON_ONCE(!free)) { | |
1869 | free->va_start = vmap_start; | |
1870 | free->va_end = vmap_end; | |
1871 | ||
1872 | insert_vmap_area_augment(free, NULL, | |
1873 | &free_vmap_area_root, | |
1874 | &free_vmap_area_list); | |
1875 | } | |
1876 | } | |
1877 | } | |
1878 | ||
db64fe02 NP |
1879 | void __init vmalloc_init(void) |
1880 | { | |
822c18f2 IK |
1881 | struct vmap_area *va; |
1882 | struct vm_struct *tmp; | |
db64fe02 NP |
1883 | int i; |
1884 | ||
68ad4a33 URS |
1885 | /* |
1886 | * Create the cache for vmap_area objects. | |
1887 | */ | |
1888 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
1889 | ||
db64fe02 NP |
1890 | for_each_possible_cpu(i) { |
1891 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1892 | struct vfree_deferred *p; |
db64fe02 NP |
1893 | |
1894 | vbq = &per_cpu(vmap_block_queue, i); | |
1895 | spin_lock_init(&vbq->lock); | |
1896 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1897 | p = &per_cpu(vfree_deferred, i); |
1898 | init_llist_head(&p->list); | |
1899 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1900 | } |
9b463334 | 1901 | |
822c18f2 IK |
1902 | /* Import existing vmlist entries. */ |
1903 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
1904 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
1905 | if (WARN_ON_ONCE(!va)) | |
1906 | continue; | |
1907 | ||
dbda591d | 1908 | va->flags = VM_VM_AREA; |
822c18f2 IK |
1909 | va->va_start = (unsigned long)tmp->addr; |
1910 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1911 | va->vm = tmp; |
68ad4a33 | 1912 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 1913 | } |
ca23e405 | 1914 | |
68ad4a33 URS |
1915 | /* |
1916 | * Now we can initialize a free vmap space. | |
1917 | */ | |
1918 | vmap_init_free_space(); | |
9b463334 | 1919 | vmap_initialized = true; |
db64fe02 NP |
1920 | } |
1921 | ||
8fc48985 TH |
1922 | /** |
1923 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1924 | * @addr: start of the VM area to map | |
1925 | * @size: size of the VM area to map | |
1926 | * @prot: page protection flags to use | |
1927 | * @pages: pages to map | |
1928 | * | |
1929 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1930 | * specify should have been allocated using get_vm_area() and its | |
1931 | * friends. | |
1932 | * | |
1933 | * NOTE: | |
1934 | * This function does NOT do any cache flushing. The caller is | |
1935 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1936 | * before calling this function. | |
1937 | * | |
1938 | * RETURNS: | |
1939 | * The number of pages mapped on success, -errno on failure. | |
1940 | */ | |
1941 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1942 | pgprot_t prot, struct page **pages) | |
1943 | { | |
1944 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1945 | } | |
1946 | ||
1947 | /** | |
1948 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1949 | * @addr: start of the VM area to unmap | |
1950 | * @size: size of the VM area to unmap | |
1951 | * | |
1952 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1953 | * specify should have been allocated using get_vm_area() and its | |
1954 | * friends. | |
1955 | * | |
1956 | * NOTE: | |
1957 | * This function does NOT do any cache flushing. The caller is | |
1958 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1959 | * before calling this function and flush_tlb_kernel_range() after. | |
1960 | */ | |
1961 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1962 | { | |
1963 | vunmap_page_range(addr, addr + size); | |
1964 | } | |
81e88fdc | 1965 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1966 | |
1967 | /** | |
1968 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1969 | * @addr: start of the VM area to unmap | |
1970 | * @size: size of the VM area to unmap | |
1971 | * | |
1972 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1973 | * the unmapping and tlb after. | |
1974 | */ | |
db64fe02 NP |
1975 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1976 | { | |
1977 | unsigned long end = addr + size; | |
f6fcba70 TH |
1978 | |
1979 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1980 | vunmap_page_range(addr, end); |
1981 | flush_tlb_kernel_range(addr, end); | |
1982 | } | |
93ef6d6c | 1983 | EXPORT_SYMBOL_GPL(unmap_kernel_range); |
db64fe02 | 1984 | |
f6f8ed47 | 1985 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
1986 | { |
1987 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 1988 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
1989 | int err; |
1990 | ||
f6f8ed47 | 1991 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 1992 | |
f6f8ed47 | 1993 | return err > 0 ? 0 : err; |
db64fe02 NP |
1994 | } |
1995 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1996 | ||
f5252e00 | 1997 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1998 | unsigned long flags, const void *caller) |
cf88c790 | 1999 | { |
c69480ad | 2000 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
2001 | vm->flags = flags; |
2002 | vm->addr = (void *)va->va_start; | |
2003 | vm->size = va->va_end - va->va_start; | |
2004 | vm->caller = caller; | |
db1aecaf | 2005 | va->vm = vm; |
cf88c790 | 2006 | va->flags |= VM_VM_AREA; |
c69480ad | 2007 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2008 | } |
cf88c790 | 2009 | |
20fc02b4 | 2010 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2011 | { |
d4033afd | 2012 | /* |
20fc02b4 | 2013 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2014 | * we should make sure that vm has proper values. |
2015 | * Pair with smp_rmb() in show_numa_info(). | |
2016 | */ | |
2017 | smp_wmb(); | |
20fc02b4 | 2018 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2019 | } |
2020 | ||
db64fe02 | 2021 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2022 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2023 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2024 | { |
0006526d | 2025 | struct vmap_area *va; |
db64fe02 | 2026 | struct vm_struct *area; |
1da177e4 | 2027 | |
52fd24ca | 2028 | BUG_ON(in_interrupt()); |
1da177e4 | 2029 | size = PAGE_ALIGN(size); |
31be8309 OH |
2030 | if (unlikely(!size)) |
2031 | return NULL; | |
1da177e4 | 2032 | |
252e5c6e | 2033 | if (flags & VM_IOREMAP) |
2034 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2035 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2036 | ||
cf88c790 | 2037 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2038 | if (unlikely(!area)) |
2039 | return NULL; | |
2040 | ||
71394fe5 AR |
2041 | if (!(flags & VM_NO_GUARD)) |
2042 | size += PAGE_SIZE; | |
1da177e4 | 2043 | |
db64fe02 NP |
2044 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2045 | if (IS_ERR(va)) { | |
2046 | kfree(area); | |
2047 | return NULL; | |
1da177e4 | 2048 | } |
1da177e4 | 2049 | |
d82b1d85 | 2050 | setup_vmalloc_vm(area, va, flags, caller); |
f5252e00 | 2051 | |
1da177e4 | 2052 | return area; |
1da177e4 LT |
2053 | } |
2054 | ||
930fc45a CL |
2055 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
2056 | unsigned long start, unsigned long end) | |
2057 | { | |
00ef2d2f DR |
2058 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2059 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 2060 | } |
5992b6da | 2061 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 2062 | |
c2968612 BH |
2063 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2064 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2065 | const void *caller) |
c2968612 | 2066 | { |
00ef2d2f DR |
2067 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2068 | GFP_KERNEL, caller); | |
c2968612 BH |
2069 | } |
2070 | ||
1da177e4 | 2071 | /** |
92eac168 MR |
2072 | * get_vm_area - reserve a contiguous kernel virtual area |
2073 | * @size: size of the area | |
2074 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2075 | * |
92eac168 MR |
2076 | * Search an area of @size in the kernel virtual mapping area, |
2077 | * and reserved it for out purposes. Returns the area descriptor | |
2078 | * on success or %NULL on failure. | |
a862f68a MR |
2079 | * |
2080 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2081 | */ |
2082 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2083 | { | |
2dca6999 | 2084 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2085 | NUMA_NO_NODE, GFP_KERNEL, |
2086 | __builtin_return_address(0)); | |
23016969 CL |
2087 | } |
2088 | ||
2089 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2090 | const void *caller) |
23016969 | 2091 | { |
2dca6999 | 2092 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2093 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2094 | } |
2095 | ||
e9da6e99 | 2096 | /** |
92eac168 MR |
2097 | * find_vm_area - find a continuous kernel virtual area |
2098 | * @addr: base address | |
e9da6e99 | 2099 | * |
92eac168 MR |
2100 | * Search for the kernel VM area starting at @addr, and return it. |
2101 | * It is up to the caller to do all required locking to keep the returned | |
2102 | * pointer valid. | |
a862f68a MR |
2103 | * |
2104 | * Return: pointer to the found area or %NULL on faulure | |
e9da6e99 MS |
2105 | */ |
2106 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2107 | { |
db64fe02 | 2108 | struct vmap_area *va; |
83342314 | 2109 | |
db64fe02 NP |
2110 | va = find_vmap_area((unsigned long)addr); |
2111 | if (va && va->flags & VM_VM_AREA) | |
db1aecaf | 2112 | return va->vm; |
1da177e4 | 2113 | |
1da177e4 | 2114 | return NULL; |
1da177e4 LT |
2115 | } |
2116 | ||
7856dfeb | 2117 | /** |
92eac168 MR |
2118 | * remove_vm_area - find and remove a continuous kernel virtual area |
2119 | * @addr: base address | |
7856dfeb | 2120 | * |
92eac168 MR |
2121 | * Search for the kernel VM area starting at @addr, and remove it. |
2122 | * This function returns the found VM area, but using it is NOT safe | |
2123 | * on SMP machines, except for its size or flags. | |
a862f68a MR |
2124 | * |
2125 | * Return: pointer to the found area or %NULL on faulure | |
7856dfeb | 2126 | */ |
b3bdda02 | 2127 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2128 | { |
db64fe02 NP |
2129 | struct vmap_area *va; |
2130 | ||
5803ed29 CH |
2131 | might_sleep(); |
2132 | ||
db64fe02 NP |
2133 | va = find_vmap_area((unsigned long)addr); |
2134 | if (va && va->flags & VM_VM_AREA) { | |
db1aecaf | 2135 | struct vm_struct *vm = va->vm; |
f5252e00 | 2136 | |
c69480ad JK |
2137 | spin_lock(&vmap_area_lock); |
2138 | va->vm = NULL; | |
2139 | va->flags &= ~VM_VM_AREA; | |
78c72746 | 2140 | va->flags |= VM_LAZY_FREE; |
c69480ad JK |
2141 | spin_unlock(&vmap_area_lock); |
2142 | ||
a5af5aa8 | 2143 | kasan_free_shadow(vm); |
dd32c279 | 2144 | free_unmap_vmap_area(va); |
dd32c279 | 2145 | |
db64fe02 NP |
2146 | return vm; |
2147 | } | |
2148 | return NULL; | |
7856dfeb AK |
2149 | } |
2150 | ||
868b104d RE |
2151 | static inline void set_area_direct_map(const struct vm_struct *area, |
2152 | int (*set_direct_map)(struct page *page)) | |
2153 | { | |
2154 | int i; | |
2155 | ||
2156 | for (i = 0; i < area->nr_pages; i++) | |
2157 | if (page_address(area->pages[i])) | |
2158 | set_direct_map(area->pages[i]); | |
2159 | } | |
2160 | ||
2161 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2162 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2163 | { | |
868b104d RE |
2164 | unsigned long start = ULONG_MAX, end = 0; |
2165 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
31e67340 | 2166 | int flush_dmap = 0; |
868b104d RE |
2167 | int i; |
2168 | ||
868b104d RE |
2169 | remove_vm_area(area->addr); |
2170 | ||
2171 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2172 | if (!flush_reset) | |
2173 | return; | |
2174 | ||
2175 | /* | |
2176 | * If not deallocating pages, just do the flush of the VM area and | |
2177 | * return. | |
2178 | */ | |
2179 | if (!deallocate_pages) { | |
2180 | vm_unmap_aliases(); | |
2181 | return; | |
2182 | } | |
2183 | ||
2184 | /* | |
2185 | * If execution gets here, flush the vm mapping and reset the direct | |
2186 | * map. Find the start and end range of the direct mappings to make sure | |
2187 | * the vm_unmap_aliases() flush includes the direct map. | |
2188 | */ | |
2189 | for (i = 0; i < area->nr_pages; i++) { | |
8e41f872 RE |
2190 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2191 | if (addr) { | |
868b104d | 2192 | start = min(addr, start); |
8e41f872 | 2193 | end = max(addr + PAGE_SIZE, end); |
31e67340 | 2194 | flush_dmap = 1; |
868b104d RE |
2195 | } |
2196 | } | |
2197 | ||
2198 | /* | |
2199 | * Set direct map to something invalid so that it won't be cached if | |
2200 | * there are any accesses after the TLB flush, then flush the TLB and | |
2201 | * reset the direct map permissions to the default. | |
2202 | */ | |
2203 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2204 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2205 | set_area_direct_map(area, set_direct_map_default_noflush); |
2206 | } | |
2207 | ||
b3bdda02 | 2208 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2209 | { |
2210 | struct vm_struct *area; | |
2211 | ||
2212 | if (!addr) | |
2213 | return; | |
2214 | ||
e69e9d4a | 2215 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2216 | addr)) |
1da177e4 | 2217 | return; |
1da177e4 | 2218 | |
6ade2032 | 2219 | area = find_vm_area(addr); |
1da177e4 | 2220 | if (unlikely(!area)) { |
4c8573e2 | 2221 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2222 | addr); |
1da177e4 LT |
2223 | return; |
2224 | } | |
2225 | ||
05e3ff95 CP |
2226 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2227 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2228 | |
868b104d RE |
2229 | vm_remove_mappings(area, deallocate_pages); |
2230 | ||
1da177e4 LT |
2231 | if (deallocate_pages) { |
2232 | int i; | |
2233 | ||
2234 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2235 | struct page *page = area->pages[i]; |
2236 | ||
2237 | BUG_ON(!page); | |
4949148a | 2238 | __free_pages(page, 0); |
1da177e4 LT |
2239 | } |
2240 | ||
244d63ee | 2241 | kvfree(area->pages); |
1da177e4 LT |
2242 | } |
2243 | ||
2244 | kfree(area); | |
2245 | return; | |
2246 | } | |
bf22e37a AR |
2247 | |
2248 | static inline void __vfree_deferred(const void *addr) | |
2249 | { | |
2250 | /* | |
2251 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2252 | * context. Preemption is absolutely fine here, because the llist_add() | |
2253 | * implementation is lockless, so it works even if we are adding to | |
2254 | * nother cpu's list. schedule_work() should be fine with this too. | |
2255 | */ | |
2256 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2257 | ||
2258 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2259 | schedule_work(&p->wq); | |
2260 | } | |
2261 | ||
2262 | /** | |
92eac168 MR |
2263 | * vfree_atomic - release memory allocated by vmalloc() |
2264 | * @addr: memory base address | |
bf22e37a | 2265 | * |
92eac168 MR |
2266 | * This one is just like vfree() but can be called in any atomic context |
2267 | * except NMIs. | |
bf22e37a AR |
2268 | */ |
2269 | void vfree_atomic(const void *addr) | |
2270 | { | |
2271 | BUG_ON(in_nmi()); | |
2272 | ||
2273 | kmemleak_free(addr); | |
2274 | ||
2275 | if (!addr) | |
2276 | return; | |
2277 | __vfree_deferred(addr); | |
2278 | } | |
2279 | ||
c67dc624 RP |
2280 | static void __vfree(const void *addr) |
2281 | { | |
2282 | if (unlikely(in_interrupt())) | |
2283 | __vfree_deferred(addr); | |
2284 | else | |
2285 | __vunmap(addr, 1); | |
2286 | } | |
2287 | ||
1da177e4 | 2288 | /** |
92eac168 MR |
2289 | * vfree - release memory allocated by vmalloc() |
2290 | * @addr: memory base address | |
1da177e4 | 2291 | * |
92eac168 MR |
2292 | * Free the virtually continuous memory area starting at @addr, as |
2293 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is | |
2294 | * NULL, no operation is performed. | |
1da177e4 | 2295 | * |
92eac168 MR |
2296 | * Must not be called in NMI context (strictly speaking, only if we don't |
2297 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2298 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 2299 | * |
92eac168 | 2300 | * May sleep if called *not* from interrupt context. |
3ca4ea3a | 2301 | * |
92eac168 | 2302 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 2303 | */ |
b3bdda02 | 2304 | void vfree(const void *addr) |
1da177e4 | 2305 | { |
32fcfd40 | 2306 | BUG_ON(in_nmi()); |
89219d37 CM |
2307 | |
2308 | kmemleak_free(addr); | |
2309 | ||
a8dda165 AR |
2310 | might_sleep_if(!in_interrupt()); |
2311 | ||
32fcfd40 AV |
2312 | if (!addr) |
2313 | return; | |
c67dc624 RP |
2314 | |
2315 | __vfree(addr); | |
1da177e4 | 2316 | } |
1da177e4 LT |
2317 | EXPORT_SYMBOL(vfree); |
2318 | ||
2319 | /** | |
92eac168 MR |
2320 | * vunmap - release virtual mapping obtained by vmap() |
2321 | * @addr: memory base address | |
1da177e4 | 2322 | * |
92eac168 MR |
2323 | * Free the virtually contiguous memory area starting at @addr, |
2324 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2325 | * |
92eac168 | 2326 | * Must not be called in interrupt context. |
1da177e4 | 2327 | */ |
b3bdda02 | 2328 | void vunmap(const void *addr) |
1da177e4 LT |
2329 | { |
2330 | BUG_ON(in_interrupt()); | |
34754b69 | 2331 | might_sleep(); |
32fcfd40 AV |
2332 | if (addr) |
2333 | __vunmap(addr, 0); | |
1da177e4 | 2334 | } |
1da177e4 LT |
2335 | EXPORT_SYMBOL(vunmap); |
2336 | ||
2337 | /** | |
92eac168 MR |
2338 | * vmap - map an array of pages into virtually contiguous space |
2339 | * @pages: array of page pointers | |
2340 | * @count: number of pages to map | |
2341 | * @flags: vm_area->flags | |
2342 | * @prot: page protection for the mapping | |
2343 | * | |
2344 | * Maps @count pages from @pages into contiguous kernel virtual | |
2345 | * space. | |
a862f68a MR |
2346 | * |
2347 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2348 | */ |
2349 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2350 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2351 | { |
2352 | struct vm_struct *area; | |
65ee03c4 | 2353 | unsigned long size; /* In bytes */ |
1da177e4 | 2354 | |
34754b69 PZ |
2355 | might_sleep(); |
2356 | ||
ca79b0c2 | 2357 | if (count > totalram_pages()) |
1da177e4 LT |
2358 | return NULL; |
2359 | ||
65ee03c4 GJM |
2360 | size = (unsigned long)count << PAGE_SHIFT; |
2361 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2362 | if (!area) |
2363 | return NULL; | |
23016969 | 2364 | |
f6f8ed47 | 2365 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
2366 | vunmap(area->addr); |
2367 | return NULL; | |
2368 | } | |
2369 | ||
2370 | return area->addr; | |
2371 | } | |
1da177e4 LT |
2372 | EXPORT_SYMBOL(vmap); |
2373 | ||
8594a21c MH |
2374 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
2375 | gfp_t gfp_mask, pgprot_t prot, | |
2376 | int node, const void *caller); | |
e31d9eb5 | 2377 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 2378 | pgprot_t prot, int node) |
1da177e4 LT |
2379 | { |
2380 | struct page **pages; | |
2381 | unsigned int nr_pages, array_size, i; | |
930f036b | 2382 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
704b862f LA |
2383 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; |
2384 | const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ? | |
2385 | 0 : | |
2386 | __GFP_HIGHMEM; | |
1da177e4 | 2387 | |
762216ab | 2388 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
2389 | array_size = (nr_pages * sizeof(struct page *)); |
2390 | ||
2391 | area->nr_pages = nr_pages; | |
2392 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 2393 | if (array_size > PAGE_SIZE) { |
704b862f | 2394 | pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask, |
3722e13c | 2395 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 2396 | } else { |
976d6dfb | 2397 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2398 | } |
1da177e4 LT |
2399 | area->pages = pages; |
2400 | if (!area->pages) { | |
2401 | remove_vm_area(area->addr); | |
2402 | kfree(area); | |
2403 | return NULL; | |
2404 | } | |
1da177e4 LT |
2405 | |
2406 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2407 | struct page *page; |
2408 | ||
4b90951c | 2409 | if (node == NUMA_NO_NODE) |
704b862f | 2410 | page = alloc_page(alloc_mask|highmem_mask); |
930fc45a | 2411 | else |
704b862f | 2412 | page = alloc_pages_node(node, alloc_mask|highmem_mask, 0); |
bf53d6f8 CL |
2413 | |
2414 | if (unlikely(!page)) { | |
1da177e4 LT |
2415 | /* Successfully allocated i pages, free them in __vunmap() */ |
2416 | area->nr_pages = i; | |
2417 | goto fail; | |
2418 | } | |
bf53d6f8 | 2419 | area->pages[i] = page; |
704b862f | 2420 | if (gfpflags_allow_blocking(gfp_mask|highmem_mask)) |
660654f9 | 2421 | cond_resched(); |
1da177e4 LT |
2422 | } |
2423 | ||
f6f8ed47 | 2424 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
2425 | goto fail; |
2426 | return area->addr; | |
2427 | ||
2428 | fail: | |
a8e99259 | 2429 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2430 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 2431 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 2432 | __vfree(area->addr); |
1da177e4 LT |
2433 | return NULL; |
2434 | } | |
2435 | ||
2436 | /** | |
92eac168 MR |
2437 | * __vmalloc_node_range - allocate virtually contiguous memory |
2438 | * @size: allocation size | |
2439 | * @align: desired alignment | |
2440 | * @start: vm area range start | |
2441 | * @end: vm area range end | |
2442 | * @gfp_mask: flags for the page level allocator | |
2443 | * @prot: protection mask for the allocated pages | |
2444 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2445 | * @node: node to use for allocation or NUMA_NO_NODE | |
2446 | * @caller: caller's return address | |
2447 | * | |
2448 | * Allocate enough pages to cover @size from the page level | |
2449 | * allocator with @gfp_mask flags. Map them into contiguous | |
2450 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2451 | * |
2452 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2453 | */ |
d0a21265 DR |
2454 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2455 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2456 | pgprot_t prot, unsigned long vm_flags, int node, |
2457 | const void *caller) | |
1da177e4 LT |
2458 | { |
2459 | struct vm_struct *area; | |
89219d37 CM |
2460 | void *addr; |
2461 | unsigned long real_size = size; | |
1da177e4 LT |
2462 | |
2463 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 2464 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 2465 | goto fail; |
1da177e4 | 2466 | |
cb9e3c29 AR |
2467 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | |
2468 | vm_flags, start, end, node, gfp_mask, caller); | |
1da177e4 | 2469 | if (!area) |
de7d2b56 | 2470 | goto fail; |
1da177e4 | 2471 | |
3722e13c | 2472 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 2473 | if (!addr) |
b82225f3 | 2474 | return NULL; |
89219d37 | 2475 | |
f5252e00 | 2476 | /* |
20fc02b4 ZY |
2477 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2478 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2479 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2480 | */ |
20fc02b4 | 2481 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2482 | |
94f4a161 | 2483 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2484 | |
2485 | return addr; | |
de7d2b56 JP |
2486 | |
2487 | fail: | |
a8e99259 | 2488 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2489 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 2490 | return NULL; |
1da177e4 LT |
2491 | } |
2492 | ||
153178ed URS |
2493 | /* |
2494 | * This is only for performance analysis of vmalloc and stress purpose. | |
2495 | * It is required by vmalloc test module, therefore do not use it other | |
2496 | * than that. | |
2497 | */ | |
2498 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2499 | EXPORT_SYMBOL_GPL(__vmalloc_node_range); | |
2500 | #endif | |
2501 | ||
d0a21265 | 2502 | /** |
92eac168 MR |
2503 | * __vmalloc_node - allocate virtually contiguous memory |
2504 | * @size: allocation size | |
2505 | * @align: desired alignment | |
2506 | * @gfp_mask: flags for the page level allocator | |
2507 | * @prot: protection mask for the allocated pages | |
2508 | * @node: node to use for allocation or NUMA_NO_NODE | |
2509 | * @caller: caller's return address | |
a7c3e901 | 2510 | * |
92eac168 MR |
2511 | * Allocate enough pages to cover @size from the page level |
2512 | * allocator with @gfp_mask flags. Map them into contiguous | |
2513 | * kernel virtual space, using a pagetable protection of @prot. | |
a7c3e901 | 2514 | * |
92eac168 MR |
2515 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2516 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2517 | * |
92eac168 MR |
2518 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2519 | * with mm people. | |
a862f68a MR |
2520 | * |
2521 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2522 | */ |
8594a21c | 2523 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
d0a21265 | 2524 | gfp_t gfp_mask, pgprot_t prot, |
5e6cafc8 | 2525 | int node, const void *caller) |
d0a21265 DR |
2526 | { |
2527 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 2528 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
2529 | } |
2530 | ||
930fc45a CL |
2531 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
2532 | { | |
00ef2d2f | 2533 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 2534 | __builtin_return_address(0)); |
930fc45a | 2535 | } |
1da177e4 LT |
2536 | EXPORT_SYMBOL(__vmalloc); |
2537 | ||
8594a21c MH |
2538 | static inline void *__vmalloc_node_flags(unsigned long size, |
2539 | int node, gfp_t flags) | |
2540 | { | |
2541 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
2542 | node, __builtin_return_address(0)); | |
2543 | } | |
2544 | ||
2545 | ||
2546 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags, | |
2547 | void *caller) | |
2548 | { | |
2549 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller); | |
2550 | } | |
2551 | ||
1da177e4 | 2552 | /** |
92eac168 MR |
2553 | * vmalloc - allocate virtually contiguous memory |
2554 | * @size: allocation size | |
2555 | * | |
2556 | * Allocate enough pages to cover @size from the page level | |
2557 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2558 | * |
92eac168 MR |
2559 | * For tight control over page level allocator and protection flags |
2560 | * use __vmalloc() instead. | |
a862f68a MR |
2561 | * |
2562 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2563 | */ |
2564 | void *vmalloc(unsigned long size) | |
2565 | { | |
00ef2d2f | 2566 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2567 | GFP_KERNEL); |
1da177e4 | 2568 | } |
1da177e4 LT |
2569 | EXPORT_SYMBOL(vmalloc); |
2570 | ||
e1ca7788 | 2571 | /** |
92eac168 MR |
2572 | * vzalloc - allocate virtually contiguous memory with zero fill |
2573 | * @size: allocation size | |
2574 | * | |
2575 | * Allocate enough pages to cover @size from the page level | |
2576 | * allocator and map them into contiguous kernel virtual space. | |
2577 | * The memory allocated is set to zero. | |
2578 | * | |
2579 | * For tight control over page level allocator and protection flags | |
2580 | * use __vmalloc() instead. | |
a862f68a MR |
2581 | * |
2582 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2583 | */ |
2584 | void *vzalloc(unsigned long size) | |
2585 | { | |
00ef2d2f | 2586 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2587 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2588 | } |
2589 | EXPORT_SYMBOL(vzalloc); | |
2590 | ||
83342314 | 2591 | /** |
ead04089 REB |
2592 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
2593 | * @size: allocation size | |
83342314 | 2594 | * |
ead04089 REB |
2595 | * The resulting memory area is zeroed so it can be mapped to userspace |
2596 | * without leaking data. | |
a862f68a MR |
2597 | * |
2598 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2599 | */ |
2600 | void *vmalloc_user(unsigned long size) | |
2601 | { | |
bc84c535 RP |
2602 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2603 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
2604 | VM_USERMAP, NUMA_NO_NODE, | |
2605 | __builtin_return_address(0)); | |
83342314 NP |
2606 | } |
2607 | EXPORT_SYMBOL(vmalloc_user); | |
2608 | ||
930fc45a | 2609 | /** |
92eac168 MR |
2610 | * vmalloc_node - allocate memory on a specific node |
2611 | * @size: allocation size | |
2612 | * @node: numa node | |
930fc45a | 2613 | * |
92eac168 MR |
2614 | * Allocate enough pages to cover @size from the page level |
2615 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 2616 | * |
92eac168 MR |
2617 | * For tight control over page level allocator and protection flags |
2618 | * use __vmalloc() instead. | |
a862f68a MR |
2619 | * |
2620 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2621 | */ |
2622 | void *vmalloc_node(unsigned long size, int node) | |
2623 | { | |
19809c2d | 2624 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL, |
23016969 | 2625 | node, __builtin_return_address(0)); |
930fc45a CL |
2626 | } |
2627 | EXPORT_SYMBOL(vmalloc_node); | |
2628 | ||
e1ca7788 DY |
2629 | /** |
2630 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2631 | * @size: allocation size | |
2632 | * @node: numa node | |
2633 | * | |
2634 | * Allocate enough pages to cover @size from the page level | |
2635 | * allocator and map them into contiguous kernel virtual space. | |
2636 | * The memory allocated is set to zero. | |
2637 | * | |
2638 | * For tight control over page level allocator and protection flags | |
2639 | * use __vmalloc_node() instead. | |
a862f68a MR |
2640 | * |
2641 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2642 | */ |
2643 | void *vzalloc_node(unsigned long size, int node) | |
2644 | { | |
2645 | return __vmalloc_node_flags(size, node, | |
19809c2d | 2646 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2647 | } |
2648 | EXPORT_SYMBOL(vzalloc_node); | |
2649 | ||
1da177e4 | 2650 | /** |
92eac168 MR |
2651 | * vmalloc_exec - allocate virtually contiguous, executable memory |
2652 | * @size: allocation size | |
1da177e4 | 2653 | * |
92eac168 MR |
2654 | * Kernel-internal function to allocate enough pages to cover @size |
2655 | * the page level allocator and map them into contiguous and | |
2656 | * executable kernel virtual space. | |
1da177e4 | 2657 | * |
92eac168 MR |
2658 | * For tight control over page level allocator and protection flags |
2659 | * use __vmalloc() instead. | |
a862f68a MR |
2660 | * |
2661 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 | 2662 | */ |
1da177e4 LT |
2663 | void *vmalloc_exec(unsigned long size) |
2664 | { | |
868b104d RE |
2665 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, |
2666 | GFP_KERNEL, PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS, | |
2667 | NUMA_NO_NODE, __builtin_return_address(0)); | |
1da177e4 LT |
2668 | } |
2669 | ||
0d08e0d3 | 2670 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2671 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2672 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2673 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2674 | #else |
698d0831 MH |
2675 | /* |
2676 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2677 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2678 | */ | |
2679 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2680 | #endif |
2681 | ||
1da177e4 | 2682 | /** |
92eac168 MR |
2683 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2684 | * @size: allocation size | |
1da177e4 | 2685 | * |
92eac168 MR |
2686 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2687 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2688 | * |
2689 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2690 | */ |
2691 | void *vmalloc_32(unsigned long size) | |
2692 | { | |
2dca6999 | 2693 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 2694 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 2695 | } |
1da177e4 LT |
2696 | EXPORT_SYMBOL(vmalloc_32); |
2697 | ||
83342314 | 2698 | /** |
ead04089 | 2699 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2700 | * @size: allocation size |
ead04089 REB |
2701 | * |
2702 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2703 | * mapped to userspace without leaking data. | |
a862f68a MR |
2704 | * |
2705 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2706 | */ |
2707 | void *vmalloc_32_user(unsigned long size) | |
2708 | { | |
bc84c535 RP |
2709 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2710 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2711 | VM_USERMAP, NUMA_NO_NODE, | |
2712 | __builtin_return_address(0)); | |
83342314 NP |
2713 | } |
2714 | EXPORT_SYMBOL(vmalloc_32_user); | |
2715 | ||
d0107eb0 KH |
2716 | /* |
2717 | * small helper routine , copy contents to buf from addr. | |
2718 | * If the page is not present, fill zero. | |
2719 | */ | |
2720 | ||
2721 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2722 | { | |
2723 | struct page *p; | |
2724 | int copied = 0; | |
2725 | ||
2726 | while (count) { | |
2727 | unsigned long offset, length; | |
2728 | ||
891c49ab | 2729 | offset = offset_in_page(addr); |
d0107eb0 KH |
2730 | length = PAGE_SIZE - offset; |
2731 | if (length > count) | |
2732 | length = count; | |
2733 | p = vmalloc_to_page(addr); | |
2734 | /* | |
2735 | * To do safe access to this _mapped_ area, we need | |
2736 | * lock. But adding lock here means that we need to add | |
2737 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2738 | * interface, rarely used. Instead of that, we'll use | |
2739 | * kmap() and get small overhead in this access function. | |
2740 | */ | |
2741 | if (p) { | |
2742 | /* | |
2743 | * we can expect USER0 is not used (see vread/vwrite's | |
2744 | * function description) | |
2745 | */ | |
9b04c5fe | 2746 | void *map = kmap_atomic(p); |
d0107eb0 | 2747 | memcpy(buf, map + offset, length); |
9b04c5fe | 2748 | kunmap_atomic(map); |
d0107eb0 KH |
2749 | } else |
2750 | memset(buf, 0, length); | |
2751 | ||
2752 | addr += length; | |
2753 | buf += length; | |
2754 | copied += length; | |
2755 | count -= length; | |
2756 | } | |
2757 | return copied; | |
2758 | } | |
2759 | ||
2760 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2761 | { | |
2762 | struct page *p; | |
2763 | int copied = 0; | |
2764 | ||
2765 | while (count) { | |
2766 | unsigned long offset, length; | |
2767 | ||
891c49ab | 2768 | offset = offset_in_page(addr); |
d0107eb0 KH |
2769 | length = PAGE_SIZE - offset; |
2770 | if (length > count) | |
2771 | length = count; | |
2772 | p = vmalloc_to_page(addr); | |
2773 | /* | |
2774 | * To do safe access to this _mapped_ area, we need | |
2775 | * lock. But adding lock here means that we need to add | |
2776 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2777 | * interface, rarely used. Instead of that, we'll use | |
2778 | * kmap() and get small overhead in this access function. | |
2779 | */ | |
2780 | if (p) { | |
2781 | /* | |
2782 | * we can expect USER0 is not used (see vread/vwrite's | |
2783 | * function description) | |
2784 | */ | |
9b04c5fe | 2785 | void *map = kmap_atomic(p); |
d0107eb0 | 2786 | memcpy(map + offset, buf, length); |
9b04c5fe | 2787 | kunmap_atomic(map); |
d0107eb0 KH |
2788 | } |
2789 | addr += length; | |
2790 | buf += length; | |
2791 | copied += length; | |
2792 | count -= length; | |
2793 | } | |
2794 | return copied; | |
2795 | } | |
2796 | ||
2797 | /** | |
92eac168 MR |
2798 | * vread() - read vmalloc area in a safe way. |
2799 | * @buf: buffer for reading data | |
2800 | * @addr: vm address. | |
2801 | * @count: number of bytes to be read. | |
2802 | * | |
92eac168 MR |
2803 | * This function checks that addr is a valid vmalloc'ed area, and |
2804 | * copy data from that area to a given buffer. If the given memory range | |
2805 | * of [addr...addr+count) includes some valid address, data is copied to | |
2806 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2807 | * IOREMAP area is treated as memory hole and no copy is done. | |
2808 | * | |
2809 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2810 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2811 | * | |
2812 | * Note: In usual ops, vread() is never necessary because the caller | |
2813 | * should know vmalloc() area is valid and can use memcpy(). | |
2814 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2815 | * any information, as /dev/kmem. |
a862f68a MR |
2816 | * |
2817 | * Return: number of bytes for which addr and buf should be increased | |
2818 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
2819 | * include any intersection with valid vmalloc area | |
d0107eb0 | 2820 | */ |
1da177e4 LT |
2821 | long vread(char *buf, char *addr, unsigned long count) |
2822 | { | |
e81ce85f JK |
2823 | struct vmap_area *va; |
2824 | struct vm_struct *vm; | |
1da177e4 | 2825 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2826 | unsigned long buflen = count; |
1da177e4 LT |
2827 | unsigned long n; |
2828 | ||
2829 | /* Don't allow overflow */ | |
2830 | if ((unsigned long) addr + count < count) | |
2831 | count = -(unsigned long) addr; | |
2832 | ||
e81ce85f JK |
2833 | spin_lock(&vmap_area_lock); |
2834 | list_for_each_entry(va, &vmap_area_list, list) { | |
2835 | if (!count) | |
2836 | break; | |
2837 | ||
2838 | if (!(va->flags & VM_VM_AREA)) | |
2839 | continue; | |
2840 | ||
2841 | vm = va->vm; | |
2842 | vaddr = (char *) vm->addr; | |
762216ab | 2843 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2844 | continue; |
2845 | while (addr < vaddr) { | |
2846 | if (count == 0) | |
2847 | goto finished; | |
2848 | *buf = '\0'; | |
2849 | buf++; | |
2850 | addr++; | |
2851 | count--; | |
2852 | } | |
762216ab | 2853 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2854 | if (n > count) |
2855 | n = count; | |
e81ce85f | 2856 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2857 | aligned_vread(buf, addr, n); |
2858 | else /* IOREMAP area is treated as memory hole */ | |
2859 | memset(buf, 0, n); | |
2860 | buf += n; | |
2861 | addr += n; | |
2862 | count -= n; | |
1da177e4 LT |
2863 | } |
2864 | finished: | |
e81ce85f | 2865 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2866 | |
2867 | if (buf == buf_start) | |
2868 | return 0; | |
2869 | /* zero-fill memory holes */ | |
2870 | if (buf != buf_start + buflen) | |
2871 | memset(buf, 0, buflen - (buf - buf_start)); | |
2872 | ||
2873 | return buflen; | |
1da177e4 LT |
2874 | } |
2875 | ||
d0107eb0 | 2876 | /** |
92eac168 MR |
2877 | * vwrite() - write vmalloc area in a safe way. |
2878 | * @buf: buffer for source data | |
2879 | * @addr: vm address. | |
2880 | * @count: number of bytes to be read. | |
2881 | * | |
92eac168 MR |
2882 | * This function checks that addr is a valid vmalloc'ed area, and |
2883 | * copy data from a buffer to the given addr. If specified range of | |
2884 | * [addr...addr+count) includes some valid address, data is copied from | |
2885 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2886 | * IOREMAP area is treated as memory hole and no copy is done. | |
2887 | * | |
2888 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2889 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2890 | * | |
2891 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2892 | * should know vmalloc() area is valid and can use memcpy(). | |
2893 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2894 | * any information, as /dev/kmem. |
a862f68a MR |
2895 | * |
2896 | * Return: number of bytes for which addr and buf should be | |
2897 | * increased (same number as @count) or %0 if [addr...addr+count) | |
2898 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 2899 | */ |
1da177e4 LT |
2900 | long vwrite(char *buf, char *addr, unsigned long count) |
2901 | { | |
e81ce85f JK |
2902 | struct vmap_area *va; |
2903 | struct vm_struct *vm; | |
d0107eb0 KH |
2904 | char *vaddr; |
2905 | unsigned long n, buflen; | |
2906 | int copied = 0; | |
1da177e4 LT |
2907 | |
2908 | /* Don't allow overflow */ | |
2909 | if ((unsigned long) addr + count < count) | |
2910 | count = -(unsigned long) addr; | |
d0107eb0 | 2911 | buflen = count; |
1da177e4 | 2912 | |
e81ce85f JK |
2913 | spin_lock(&vmap_area_lock); |
2914 | list_for_each_entry(va, &vmap_area_list, list) { | |
2915 | if (!count) | |
2916 | break; | |
2917 | ||
2918 | if (!(va->flags & VM_VM_AREA)) | |
2919 | continue; | |
2920 | ||
2921 | vm = va->vm; | |
2922 | vaddr = (char *) vm->addr; | |
762216ab | 2923 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2924 | continue; |
2925 | while (addr < vaddr) { | |
2926 | if (count == 0) | |
2927 | goto finished; | |
2928 | buf++; | |
2929 | addr++; | |
2930 | count--; | |
2931 | } | |
762216ab | 2932 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2933 | if (n > count) |
2934 | n = count; | |
e81ce85f | 2935 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2936 | aligned_vwrite(buf, addr, n); |
2937 | copied++; | |
2938 | } | |
2939 | buf += n; | |
2940 | addr += n; | |
2941 | count -= n; | |
1da177e4 LT |
2942 | } |
2943 | finished: | |
e81ce85f | 2944 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2945 | if (!copied) |
2946 | return 0; | |
2947 | return buflen; | |
1da177e4 | 2948 | } |
83342314 NP |
2949 | |
2950 | /** | |
92eac168 MR |
2951 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
2952 | * @vma: vma to cover | |
2953 | * @uaddr: target user address to start at | |
2954 | * @kaddr: virtual address of vmalloc kernel memory | |
2955 | * @size: size of map area | |
7682486b | 2956 | * |
92eac168 | 2957 | * Returns: 0 for success, -Exxx on failure |
83342314 | 2958 | * |
92eac168 MR |
2959 | * This function checks that @kaddr is a valid vmalloc'ed area, |
2960 | * and that it is big enough to cover the range starting at | |
2961 | * @uaddr in @vma. Will return failure if that criteria isn't | |
2962 | * met. | |
83342314 | 2963 | * |
92eac168 | 2964 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 2965 | */ |
e69e9d4a HD |
2966 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
2967 | void *kaddr, unsigned long size) | |
83342314 NP |
2968 | { |
2969 | struct vm_struct *area; | |
83342314 | 2970 | |
e69e9d4a HD |
2971 | size = PAGE_ALIGN(size); |
2972 | ||
2973 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
2974 | return -EINVAL; |
2975 | ||
e69e9d4a | 2976 | area = find_vm_area(kaddr); |
83342314 | 2977 | if (!area) |
db64fe02 | 2978 | return -EINVAL; |
83342314 NP |
2979 | |
2980 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2981 | return -EINVAL; |
83342314 | 2982 | |
401592d2 | 2983 | if (kaddr + size > area->addr + get_vm_area_size(area)) |
db64fe02 | 2984 | return -EINVAL; |
83342314 | 2985 | |
83342314 | 2986 | do { |
e69e9d4a | 2987 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
2988 | int ret; |
2989 | ||
83342314 NP |
2990 | ret = vm_insert_page(vma, uaddr, page); |
2991 | if (ret) | |
2992 | return ret; | |
2993 | ||
2994 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
2995 | kaddr += PAGE_SIZE; |
2996 | size -= PAGE_SIZE; | |
2997 | } while (size > 0); | |
83342314 | 2998 | |
314e51b9 | 2999 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3000 | |
db64fe02 | 3001 | return 0; |
83342314 | 3002 | } |
e69e9d4a HD |
3003 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
3004 | ||
3005 | /** | |
92eac168 MR |
3006 | * remap_vmalloc_range - map vmalloc pages to userspace |
3007 | * @vma: vma to cover (map full range of vma) | |
3008 | * @addr: vmalloc memory | |
3009 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3010 | * |
92eac168 | 3011 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3012 | * |
92eac168 MR |
3013 | * This function checks that addr is a valid vmalloc'ed area, and |
3014 | * that it is big enough to cover the vma. Will return failure if | |
3015 | * that criteria isn't met. | |
e69e9d4a | 3016 | * |
92eac168 | 3017 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3018 | */ |
3019 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3020 | unsigned long pgoff) | |
3021 | { | |
3022 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
3023 | addr + (pgoff << PAGE_SHIFT), | |
3024 | vma->vm_end - vma->vm_start); | |
3025 | } | |
83342314 NP |
3026 | EXPORT_SYMBOL(remap_vmalloc_range); |
3027 | ||
1eeb66a1 CH |
3028 | /* |
3029 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
3030 | * have one. | |
3031 | */ | |
3b32123d | 3032 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
3033 | { |
3034 | } | |
5f4352fb JF |
3035 | |
3036 | ||
8b1e0f81 | 3037 | static int f(pte_t *pte, unsigned long addr, void *data) |
5f4352fb | 3038 | { |
cd12909c DV |
3039 | pte_t ***p = data; |
3040 | ||
3041 | if (p) { | |
3042 | *(*p) = pte; | |
3043 | (*p)++; | |
3044 | } | |
5f4352fb JF |
3045 | return 0; |
3046 | } | |
3047 | ||
3048 | /** | |
92eac168 MR |
3049 | * alloc_vm_area - allocate a range of kernel address space |
3050 | * @size: size of the area | |
3051 | * @ptes: returns the PTEs for the address space | |
7682486b | 3052 | * |
92eac168 | 3053 | * Returns: NULL on failure, vm_struct on success |
5f4352fb | 3054 | * |
92eac168 MR |
3055 | * This function reserves a range of kernel address space, and |
3056 | * allocates pagetables to map that range. No actual mappings | |
3057 | * are created. | |
cd12909c | 3058 | * |
92eac168 MR |
3059 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
3060 | * allocated for the VM area are returned. | |
5f4352fb | 3061 | */ |
cd12909c | 3062 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
3063 | { |
3064 | struct vm_struct *area; | |
3065 | ||
23016969 CL |
3066 | area = get_vm_area_caller(size, VM_IOREMAP, |
3067 | __builtin_return_address(0)); | |
5f4352fb JF |
3068 | if (area == NULL) |
3069 | return NULL; | |
3070 | ||
3071 | /* | |
3072 | * This ensures that page tables are constructed for this region | |
3073 | * of kernel virtual address space and mapped into init_mm. | |
3074 | */ | |
3075 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 3076 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
3077 | free_vm_area(area); |
3078 | return NULL; | |
3079 | } | |
3080 | ||
5f4352fb JF |
3081 | return area; |
3082 | } | |
3083 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
3084 | ||
3085 | void free_vm_area(struct vm_struct *area) | |
3086 | { | |
3087 | struct vm_struct *ret; | |
3088 | ret = remove_vm_area(area->addr); | |
3089 | BUG_ON(ret != area); | |
3090 | kfree(area); | |
3091 | } | |
3092 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3093 | |
4f8b02b4 | 3094 | #ifdef CONFIG_SMP |
ca23e405 TH |
3095 | static struct vmap_area *node_to_va(struct rb_node *n) |
3096 | { | |
4583e773 | 3097 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3098 | } |
3099 | ||
3100 | /** | |
68ad4a33 URS |
3101 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3102 | * @addr: target address | |
ca23e405 | 3103 | * |
68ad4a33 URS |
3104 | * Returns: vmap_area if it is found. If there is no such area |
3105 | * the first highest(reverse order) vmap_area is returned | |
3106 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3107 | * if there are no any areas before @addr. | |
ca23e405 | 3108 | */ |
68ad4a33 URS |
3109 | static struct vmap_area * |
3110 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3111 | { |
68ad4a33 URS |
3112 | struct vmap_area *va, *tmp; |
3113 | struct rb_node *n; | |
3114 | ||
3115 | n = free_vmap_area_root.rb_node; | |
3116 | va = NULL; | |
ca23e405 TH |
3117 | |
3118 | while (n) { | |
68ad4a33 URS |
3119 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3120 | if (tmp->va_start <= addr) { | |
3121 | va = tmp; | |
3122 | if (tmp->va_end >= addr) | |
3123 | break; | |
3124 | ||
ca23e405 | 3125 | n = n->rb_right; |
68ad4a33 URS |
3126 | } else { |
3127 | n = n->rb_left; | |
3128 | } | |
ca23e405 TH |
3129 | } |
3130 | ||
68ad4a33 | 3131 | return va; |
ca23e405 TH |
3132 | } |
3133 | ||
3134 | /** | |
68ad4a33 URS |
3135 | * pvm_determine_end_from_reverse - find the highest aligned address |
3136 | * of free block below VMALLOC_END | |
3137 | * @va: | |
3138 | * in - the VA we start the search(reverse order); | |
3139 | * out - the VA with the highest aligned end address. | |
ca23e405 | 3140 | * |
68ad4a33 | 3141 | * Returns: determined end address within vmap_area |
ca23e405 | 3142 | */ |
68ad4a33 URS |
3143 | static unsigned long |
3144 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3145 | { |
68ad4a33 | 3146 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3147 | unsigned long addr; |
3148 | ||
68ad4a33 URS |
3149 | if (likely(*va)) { |
3150 | list_for_each_entry_from_reverse((*va), | |
3151 | &free_vmap_area_list, list) { | |
3152 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3153 | if ((*va)->va_start < addr) | |
3154 | return addr; | |
3155 | } | |
ca23e405 TH |
3156 | } |
3157 | ||
68ad4a33 | 3158 | return 0; |
ca23e405 TH |
3159 | } |
3160 | ||
3161 | /** | |
3162 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3163 | * @offsets: array containing offset of each area | |
3164 | * @sizes: array containing size of each area | |
3165 | * @nr_vms: the number of areas to allocate | |
3166 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3167 | * |
3168 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3169 | * vm_structs on success, %NULL on failure | |
3170 | * | |
3171 | * Percpu allocator wants to use congruent vm areas so that it can | |
3172 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3173 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3174 | * be scattered pretty far, distance between two areas easily going up | |
3175 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3176 | * areas are allocated from top. | |
ca23e405 | 3177 | * |
68ad4a33 URS |
3178 | * Despite its complicated look, this allocator is rather simple. It |
3179 | * does everything top-down and scans free blocks from the end looking | |
3180 | * for matching base. While scanning, if any of the areas do not fit the | |
3181 | * base address is pulled down to fit the area. Scanning is repeated till | |
3182 | * all the areas fit and then all necessary data structures are inserted | |
3183 | * and the result is returned. | |
ca23e405 TH |
3184 | */ |
3185 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3186 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3187 | size_t align) |
ca23e405 TH |
3188 | { |
3189 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3190 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3191 | struct vmap_area **vas, *va; |
ca23e405 TH |
3192 | struct vm_struct **vms; |
3193 | int area, area2, last_area, term_area; | |
68ad4a33 | 3194 | unsigned long base, start, size, end, last_end; |
ca23e405 | 3195 | bool purged = false; |
68ad4a33 | 3196 | enum fit_type type; |
ca23e405 | 3197 | |
ca23e405 | 3198 | /* verify parameters and allocate data structures */ |
891c49ab | 3199 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3200 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3201 | start = offsets[area]; | |
3202 | end = start + sizes[area]; | |
3203 | ||
3204 | /* is everything aligned properly? */ | |
3205 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3206 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3207 | ||
3208 | /* detect the area with the highest address */ | |
3209 | if (start > offsets[last_area]) | |
3210 | last_area = area; | |
3211 | ||
c568da28 | 3212 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3213 | unsigned long start2 = offsets[area2]; |
3214 | unsigned long end2 = start2 + sizes[area2]; | |
3215 | ||
c568da28 | 3216 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3217 | } |
3218 | } | |
3219 | last_end = offsets[last_area] + sizes[last_area]; | |
3220 | ||
3221 | if (vmalloc_end - vmalloc_start < last_end) { | |
3222 | WARN_ON(true); | |
3223 | return NULL; | |
3224 | } | |
3225 | ||
4d67d860 TM |
3226 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3227 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3228 | if (!vas || !vms) |
f1db7afd | 3229 | goto err_free2; |
ca23e405 TH |
3230 | |
3231 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3232 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3233 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3234 | if (!vas[area] || !vms[area]) |
3235 | goto err_free; | |
3236 | } | |
3237 | retry: | |
3238 | spin_lock(&vmap_area_lock); | |
3239 | ||
3240 | /* start scanning - we scan from the top, begin with the last area */ | |
3241 | area = term_area = last_area; | |
3242 | start = offsets[area]; | |
3243 | end = start + sizes[area]; | |
3244 | ||
68ad4a33 URS |
3245 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3246 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3247 | |
3248 | while (true) { | |
ca23e405 TH |
3249 | /* |
3250 | * base might have underflowed, add last_end before | |
3251 | * comparing. | |
3252 | */ | |
68ad4a33 URS |
3253 | if (base + last_end < vmalloc_start + last_end) |
3254 | goto overflow; | |
ca23e405 TH |
3255 | |
3256 | /* | |
68ad4a33 | 3257 | * Fitting base has not been found. |
ca23e405 | 3258 | */ |
68ad4a33 URS |
3259 | if (va == NULL) |
3260 | goto overflow; | |
ca23e405 TH |
3261 | |
3262 | /* | |
68ad4a33 | 3263 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3264 | */ |
68ad4a33 URS |
3265 | if (base + start < va->va_start || base + end > va->va_end) { |
3266 | va = node_to_va(rb_prev(&va->rb_node)); | |
3267 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3268 | term_area = area; |
3269 | continue; | |
3270 | } | |
3271 | ||
3272 | /* | |
3273 | * This area fits, move on to the previous one. If | |
3274 | * the previous one is the terminal one, we're done. | |
3275 | */ | |
3276 | area = (area + nr_vms - 1) % nr_vms; | |
3277 | if (area == term_area) | |
3278 | break; | |
68ad4a33 | 3279 | |
ca23e405 TH |
3280 | start = offsets[area]; |
3281 | end = start + sizes[area]; | |
68ad4a33 | 3282 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3283 | } |
68ad4a33 | 3284 | |
ca23e405 TH |
3285 | /* we've found a fitting base, insert all va's */ |
3286 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3287 | int ret; |
ca23e405 | 3288 | |
68ad4a33 URS |
3289 | start = base + offsets[area]; |
3290 | size = sizes[area]; | |
ca23e405 | 3291 | |
68ad4a33 URS |
3292 | va = pvm_find_va_enclose_addr(start); |
3293 | if (WARN_ON_ONCE(va == NULL)) | |
3294 | /* It is a BUG(), but trigger recovery instead. */ | |
3295 | goto recovery; | |
3296 | ||
3297 | type = classify_va_fit_type(va, start, size); | |
3298 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3299 | /* It is a BUG(), but trigger recovery instead. */ | |
3300 | goto recovery; | |
3301 | ||
3302 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3303 | if (unlikely(ret)) | |
3304 | goto recovery; | |
3305 | ||
3306 | /* Allocated area. */ | |
3307 | va = vas[area]; | |
3308 | va->va_start = start; | |
3309 | va->va_end = start + size; | |
3310 | ||
3311 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
3312 | } | |
ca23e405 TH |
3313 | |
3314 | spin_unlock(&vmap_area_lock); | |
3315 | ||
3316 | /* insert all vm's */ | |
3317 | for (area = 0; area < nr_vms; area++) | |
3645cb4a ZY |
3318 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
3319 | pcpu_get_vm_areas); | |
ca23e405 TH |
3320 | |
3321 | kfree(vas); | |
3322 | return vms; | |
3323 | ||
68ad4a33 URS |
3324 | recovery: |
3325 | /* Remove previously inserted areas. */ | |
3326 | while (area--) { | |
3327 | __free_vmap_area(vas[area]); | |
3328 | vas[area] = NULL; | |
3329 | } | |
3330 | ||
3331 | overflow: | |
3332 | spin_unlock(&vmap_area_lock); | |
3333 | if (!purged) { | |
3334 | purge_vmap_area_lazy(); | |
3335 | purged = true; | |
3336 | ||
3337 | /* Before "retry", check if we recover. */ | |
3338 | for (area = 0; area < nr_vms; area++) { | |
3339 | if (vas[area]) | |
3340 | continue; | |
3341 | ||
3342 | vas[area] = kmem_cache_zalloc( | |
3343 | vmap_area_cachep, GFP_KERNEL); | |
3344 | if (!vas[area]) | |
3345 | goto err_free; | |
3346 | } | |
3347 | ||
3348 | goto retry; | |
3349 | } | |
3350 | ||
ca23e405 TH |
3351 | err_free: |
3352 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3353 | if (vas[area]) |
3354 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3355 | ||
f1db7afd | 3356 | kfree(vms[area]); |
ca23e405 | 3357 | } |
f1db7afd | 3358 | err_free2: |
ca23e405 TH |
3359 | kfree(vas); |
3360 | kfree(vms); | |
3361 | return NULL; | |
3362 | } | |
3363 | ||
3364 | /** | |
3365 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3366 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3367 | * @nr_vms: the number of allocated areas | |
3368 | * | |
3369 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3370 | */ | |
3371 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3372 | { | |
3373 | int i; | |
3374 | ||
3375 | for (i = 0; i < nr_vms; i++) | |
3376 | free_vm_area(vms[i]); | |
3377 | kfree(vms); | |
3378 | } | |
4f8b02b4 | 3379 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
3380 | |
3381 | #ifdef CONFIG_PROC_FS | |
3382 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 3383 | __acquires(&vmap_area_lock) |
a10aa579 | 3384 | { |
d4033afd | 3385 | spin_lock(&vmap_area_lock); |
3f500069 | 3386 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3387 | } |
3388 | ||
3389 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3390 | { | |
3f500069 | 3391 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3392 | } |
3393 | ||
3394 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 3395 | __releases(&vmap_area_lock) |
a10aa579 | 3396 | { |
d4033afd | 3397 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
3398 | } |
3399 | ||
a47a126a ED |
3400 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3401 | { | |
e5adfffc | 3402 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3403 | unsigned int nr, *counters = m->private; |
3404 | ||
3405 | if (!counters) | |
3406 | return; | |
3407 | ||
af12346c WL |
3408 | if (v->flags & VM_UNINITIALIZED) |
3409 | return; | |
7e5b528b DV |
3410 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3411 | smp_rmb(); | |
af12346c | 3412 | |
a47a126a ED |
3413 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3414 | ||
3415 | for (nr = 0; nr < v->nr_pages; nr++) | |
3416 | counters[page_to_nid(v->pages[nr])]++; | |
3417 | ||
3418 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3419 | if (counters[nr]) | |
3420 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3421 | } | |
3422 | } | |
3423 | ||
a10aa579 CL |
3424 | static int s_show(struct seq_file *m, void *p) |
3425 | { | |
3f500069 | 3426 | struct vmap_area *va; |
d4033afd JK |
3427 | struct vm_struct *v; |
3428 | ||
3f500069 | 3429 | va = list_entry(p, struct vmap_area, list); |
3430 | ||
c2ce8c14 WL |
3431 | /* |
3432 | * s_show can encounter race with remove_vm_area, !VM_VM_AREA on | |
3433 | * behalf of vmap area is being tear down or vm_map_ram allocation. | |
3434 | */ | |
78c72746 YX |
3435 | if (!(va->flags & VM_VM_AREA)) { |
3436 | seq_printf(m, "0x%pK-0x%pK %7ld %s\n", | |
3437 | (void *)va->va_start, (void *)va->va_end, | |
3438 | va->va_end - va->va_start, | |
3439 | va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram"); | |
3440 | ||
d4033afd | 3441 | return 0; |
78c72746 | 3442 | } |
d4033afd JK |
3443 | |
3444 | v = va->vm; | |
a10aa579 | 3445 | |
45ec1690 | 3446 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3447 | v->addr, v->addr + v->size, v->size); |
3448 | ||
62c70bce JP |
3449 | if (v->caller) |
3450 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3451 | |
a10aa579 CL |
3452 | if (v->nr_pages) |
3453 | seq_printf(m, " pages=%d", v->nr_pages); | |
3454 | ||
3455 | if (v->phys_addr) | |
199eaa05 | 3456 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3457 | |
3458 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3459 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3460 | |
3461 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3462 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3463 | |
3464 | if (v->flags & VM_MAP) | |
f4527c90 | 3465 | seq_puts(m, " vmap"); |
a10aa579 CL |
3466 | |
3467 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3468 | seq_puts(m, " user"); |
a10aa579 | 3469 | |
244d63ee | 3470 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3471 | seq_puts(m, " vpages"); |
a10aa579 | 3472 | |
a47a126a | 3473 | show_numa_info(m, v); |
a10aa579 CL |
3474 | seq_putc(m, '\n'); |
3475 | return 0; | |
3476 | } | |
3477 | ||
5f6a6a9c | 3478 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3479 | .start = s_start, |
3480 | .next = s_next, | |
3481 | .stop = s_stop, | |
3482 | .show = s_show, | |
3483 | }; | |
5f6a6a9c | 3484 | |
5f6a6a9c AD |
3485 | static int __init proc_vmalloc_init(void) |
3486 | { | |
fddda2b7 | 3487 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3488 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3489 | &vmalloc_op, |
3490 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3491 | else |
0825a6f9 | 3492 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3493 | return 0; |
3494 | } | |
3495 | module_init(proc_vmalloc_init); | |
db3808c1 | 3496 | |
a10aa579 | 3497 | #endif |