]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/vmalloc.c | |
4 | * | |
5 | * Copyright (C) 1993 Linus Torvalds | |
6 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
7 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
8 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 9 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
10 | */ |
11 | ||
db64fe02 | 12 | #include <linux/vmalloc.h> |
1da177e4 LT |
13 | #include <linux/mm.h> |
14 | #include <linux/module.h> | |
15 | #include <linux/highmem.h> | |
c3edc401 | 16 | #include <linux/sched/signal.h> |
1da177e4 LT |
17 | #include <linux/slab.h> |
18 | #include <linux/spinlock.h> | |
19 | #include <linux/interrupt.h> | |
5f6a6a9c | 20 | #include <linux/proc_fs.h> |
a10aa579 | 21 | #include <linux/seq_file.h> |
868b104d | 22 | #include <linux/set_memory.h> |
3ac7fe5a | 23 | #include <linux/debugobjects.h> |
23016969 | 24 | #include <linux/kallsyms.h> |
db64fe02 | 25 | #include <linux/list.h> |
4da56b99 | 26 | #include <linux/notifier.h> |
db64fe02 NP |
27 | #include <linux/rbtree.h> |
28 | #include <linux/radix-tree.h> | |
29 | #include <linux/rcupdate.h> | |
f0aa6617 | 30 | #include <linux/pfn.h> |
89219d37 | 31 | #include <linux/kmemleak.h> |
60063497 | 32 | #include <linux/atomic.h> |
3b32123d | 33 | #include <linux/compiler.h> |
32fcfd40 | 34 | #include <linux/llist.h> |
0f616be1 | 35 | #include <linux/bitops.h> |
68ad4a33 | 36 | #include <linux/rbtree_augmented.h> |
3b32123d | 37 | |
7c0f6ba6 | 38 | #include <linux/uaccess.h> |
1da177e4 | 39 | #include <asm/tlbflush.h> |
2dca6999 | 40 | #include <asm/shmparam.h> |
1da177e4 | 41 | |
dd56b046 MG |
42 | #include "internal.h" |
43 | ||
32fcfd40 AV |
44 | struct vfree_deferred { |
45 | struct llist_head list; | |
46 | struct work_struct wq; | |
47 | }; | |
48 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
49 | ||
50 | static void __vunmap(const void *, int); | |
51 | ||
52 | static void free_work(struct work_struct *w) | |
53 | { | |
54 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
55 | struct llist_node *t, *llnode; |
56 | ||
57 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
58 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
59 | } |
60 | ||
db64fe02 | 61 | /*** Page table manipulation functions ***/ |
b221385b | 62 | |
1da177e4 LT |
63 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
64 | { | |
65 | pte_t *pte; | |
66 | ||
67 | pte = pte_offset_kernel(pmd, addr); | |
68 | do { | |
69 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
70 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
71 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
72 | } | |
73 | ||
db64fe02 | 74 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
75 | { |
76 | pmd_t *pmd; | |
77 | unsigned long next; | |
78 | ||
79 | pmd = pmd_offset(pud, addr); | |
80 | do { | |
81 | next = pmd_addr_end(addr, end); | |
b9820d8f TK |
82 | if (pmd_clear_huge(pmd)) |
83 | continue; | |
1da177e4 LT |
84 | if (pmd_none_or_clear_bad(pmd)) |
85 | continue; | |
86 | vunmap_pte_range(pmd, addr, next); | |
87 | } while (pmd++, addr = next, addr != end); | |
88 | } | |
89 | ||
c2febafc | 90 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end) |
1da177e4 LT |
91 | { |
92 | pud_t *pud; | |
93 | unsigned long next; | |
94 | ||
c2febafc | 95 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
96 | do { |
97 | next = pud_addr_end(addr, end); | |
b9820d8f TK |
98 | if (pud_clear_huge(pud)) |
99 | continue; | |
1da177e4 LT |
100 | if (pud_none_or_clear_bad(pud)) |
101 | continue; | |
102 | vunmap_pmd_range(pud, addr, next); | |
103 | } while (pud++, addr = next, addr != end); | |
104 | } | |
105 | ||
c2febafc KS |
106 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
107 | { | |
108 | p4d_t *p4d; | |
109 | unsigned long next; | |
110 | ||
111 | p4d = p4d_offset(pgd, addr); | |
112 | do { | |
113 | next = p4d_addr_end(addr, end); | |
114 | if (p4d_clear_huge(p4d)) | |
115 | continue; | |
116 | if (p4d_none_or_clear_bad(p4d)) | |
117 | continue; | |
118 | vunmap_pud_range(p4d, addr, next); | |
119 | } while (p4d++, addr = next, addr != end); | |
120 | } | |
121 | ||
db64fe02 | 122 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
123 | { |
124 | pgd_t *pgd; | |
125 | unsigned long next; | |
1da177e4 LT |
126 | |
127 | BUG_ON(addr >= end); | |
128 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
129 | do { |
130 | next = pgd_addr_end(addr, end); | |
131 | if (pgd_none_or_clear_bad(pgd)) | |
132 | continue; | |
c2febafc | 133 | vunmap_p4d_range(pgd, addr, next); |
1da177e4 | 134 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
135 | } |
136 | ||
137 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 138 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
139 | { |
140 | pte_t *pte; | |
141 | ||
db64fe02 NP |
142 | /* |
143 | * nr is a running index into the array which helps higher level | |
144 | * callers keep track of where we're up to. | |
145 | */ | |
146 | ||
872fec16 | 147 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
148 | if (!pte) |
149 | return -ENOMEM; | |
150 | do { | |
db64fe02 NP |
151 | struct page *page = pages[*nr]; |
152 | ||
153 | if (WARN_ON(!pte_none(*pte))) | |
154 | return -EBUSY; | |
155 | if (WARN_ON(!page)) | |
1da177e4 LT |
156 | return -ENOMEM; |
157 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 158 | (*nr)++; |
1da177e4 LT |
159 | } while (pte++, addr += PAGE_SIZE, addr != end); |
160 | return 0; | |
161 | } | |
162 | ||
db64fe02 NP |
163 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
164 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
165 | { |
166 | pmd_t *pmd; | |
167 | unsigned long next; | |
168 | ||
169 | pmd = pmd_alloc(&init_mm, pud, addr); | |
170 | if (!pmd) | |
171 | return -ENOMEM; | |
172 | do { | |
173 | next = pmd_addr_end(addr, end); | |
db64fe02 | 174 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
175 | return -ENOMEM; |
176 | } while (pmd++, addr = next, addr != end); | |
177 | return 0; | |
178 | } | |
179 | ||
c2febafc | 180 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, |
db64fe02 | 181 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
182 | { |
183 | pud_t *pud; | |
184 | unsigned long next; | |
185 | ||
c2febafc | 186 | pud = pud_alloc(&init_mm, p4d, addr); |
1da177e4 LT |
187 | if (!pud) |
188 | return -ENOMEM; | |
189 | do { | |
190 | next = pud_addr_end(addr, end); | |
db64fe02 | 191 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
192 | return -ENOMEM; |
193 | } while (pud++, addr = next, addr != end); | |
194 | return 0; | |
195 | } | |
196 | ||
c2febafc KS |
197 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, |
198 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
199 | { | |
200 | p4d_t *p4d; | |
201 | unsigned long next; | |
202 | ||
203 | p4d = p4d_alloc(&init_mm, pgd, addr); | |
204 | if (!p4d) | |
205 | return -ENOMEM; | |
206 | do { | |
207 | next = p4d_addr_end(addr, end); | |
208 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr)) | |
209 | return -ENOMEM; | |
210 | } while (p4d++, addr = next, addr != end); | |
211 | return 0; | |
212 | } | |
213 | ||
db64fe02 NP |
214 | /* |
215 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
216 | * will have pfns corresponding to the "pages" array. | |
217 | * | |
218 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
219 | */ | |
8fc48985 TH |
220 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
221 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
222 | { |
223 | pgd_t *pgd; | |
224 | unsigned long next; | |
2e4e27c7 | 225 | unsigned long addr = start; |
db64fe02 NP |
226 | int err = 0; |
227 | int nr = 0; | |
1da177e4 LT |
228 | |
229 | BUG_ON(addr >= end); | |
230 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
231 | do { |
232 | next = pgd_addr_end(addr, end); | |
c2febafc | 233 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 234 | if (err) |
bf88c8c8 | 235 | return err; |
1da177e4 | 236 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 237 | |
db64fe02 | 238 | return nr; |
1da177e4 LT |
239 | } |
240 | ||
8fc48985 TH |
241 | static int vmap_page_range(unsigned long start, unsigned long end, |
242 | pgprot_t prot, struct page **pages) | |
243 | { | |
244 | int ret; | |
245 | ||
246 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
247 | flush_cache_vmap(start, end); | |
248 | return ret; | |
249 | } | |
250 | ||
81ac3ad9 | 251 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
252 | { |
253 | /* | |
ab4f2ee1 | 254 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
255 | * and fall back on vmalloc() if that fails. Others |
256 | * just put it in the vmalloc space. | |
257 | */ | |
258 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
259 | unsigned long addr = (unsigned long)x; | |
260 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
261 | return 1; | |
262 | #endif | |
263 | return is_vmalloc_addr(x); | |
264 | } | |
265 | ||
48667e7a | 266 | /* |
add688fb | 267 | * Walk a vmap address to the struct page it maps. |
48667e7a | 268 | */ |
add688fb | 269 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
270 | { |
271 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 272 | struct page *page = NULL; |
48667e7a | 273 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
274 | p4d_t *p4d; |
275 | pud_t *pud; | |
276 | pmd_t *pmd; | |
277 | pte_t *ptep, pte; | |
48667e7a | 278 | |
7aa413de IM |
279 | /* |
280 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
281 | * architectures that do not vmalloc module space | |
282 | */ | |
73bdf0a6 | 283 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 284 | |
c2febafc KS |
285 | if (pgd_none(*pgd)) |
286 | return NULL; | |
287 | p4d = p4d_offset(pgd, addr); | |
288 | if (p4d_none(*p4d)) | |
289 | return NULL; | |
290 | pud = pud_offset(p4d, addr); | |
029c54b0 AB |
291 | |
292 | /* | |
293 | * Don't dereference bad PUD or PMD (below) entries. This will also | |
294 | * identify huge mappings, which we may encounter on architectures | |
295 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be | |
296 | * identified as vmalloc addresses by is_vmalloc_addr(), but are | |
297 | * not [unambiguously] associated with a struct page, so there is | |
298 | * no correct value to return for them. | |
299 | */ | |
300 | WARN_ON_ONCE(pud_bad(*pud)); | |
301 | if (pud_none(*pud) || pud_bad(*pud)) | |
c2febafc KS |
302 | return NULL; |
303 | pmd = pmd_offset(pud, addr); | |
029c54b0 AB |
304 | WARN_ON_ONCE(pmd_bad(*pmd)); |
305 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
c2febafc KS |
306 | return NULL; |
307 | ||
308 | ptep = pte_offset_map(pmd, addr); | |
309 | pte = *ptep; | |
310 | if (pte_present(pte)) | |
311 | page = pte_page(pte); | |
312 | pte_unmap(ptep); | |
add688fb | 313 | return page; |
48667e7a | 314 | } |
add688fb | 315 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
316 | |
317 | /* | |
add688fb | 318 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 319 | */ |
add688fb | 320 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 321 | { |
add688fb | 322 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 323 | } |
add688fb | 324 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 325 | |
db64fe02 NP |
326 | |
327 | /*** Global kva allocator ***/ | |
328 | ||
bb850f4d | 329 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 330 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 331 | |
db64fe02 | 332 | |
db64fe02 | 333 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
334 | /* Export for kexec only */ |
335 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 336 | static LLIST_HEAD(vmap_purge_list); |
89699605 | 337 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 338 | static bool vmap_initialized __read_mostly; |
89699605 | 339 | |
68ad4a33 URS |
340 | /* |
341 | * This kmem_cache is used for vmap_area objects. Instead of | |
342 | * allocating from slab we reuse an object from this cache to | |
343 | * make things faster. Especially in "no edge" splitting of | |
344 | * free block. | |
345 | */ | |
346 | static struct kmem_cache *vmap_area_cachep; | |
347 | ||
348 | /* | |
349 | * This linked list is used in pair with free_vmap_area_root. | |
350 | * It gives O(1) access to prev/next to perform fast coalescing. | |
351 | */ | |
352 | static LIST_HEAD(free_vmap_area_list); | |
353 | ||
354 | /* | |
355 | * This augment red-black tree represents the free vmap space. | |
356 | * All vmap_area objects in this tree are sorted by va->va_start | |
357 | * address. It is used for allocation and merging when a vmap | |
358 | * object is released. | |
359 | * | |
360 | * Each vmap_area node contains a maximum available free block | |
361 | * of its sub-tree, right or left. Therefore it is possible to | |
362 | * find a lowest match of free area. | |
363 | */ | |
364 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
365 | ||
82dd23e8 URS |
366 | /* |
367 | * Preload a CPU with one object for "no edge" split case. The | |
368 | * aim is to get rid of allocations from the atomic context, thus | |
369 | * to use more permissive allocation masks. | |
370 | */ | |
371 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
372 | ||
68ad4a33 URS |
373 | static __always_inline unsigned long |
374 | va_size(struct vmap_area *va) | |
375 | { | |
376 | return (va->va_end - va->va_start); | |
377 | } | |
378 | ||
379 | static __always_inline unsigned long | |
380 | get_subtree_max_size(struct rb_node *node) | |
381 | { | |
382 | struct vmap_area *va; | |
383 | ||
384 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
385 | return va ? va->subtree_max_size : 0; | |
386 | } | |
89699605 | 387 | |
68ad4a33 URS |
388 | /* |
389 | * Gets called when remove the node and rotate. | |
390 | */ | |
391 | static __always_inline unsigned long | |
392 | compute_subtree_max_size(struct vmap_area *va) | |
393 | { | |
394 | return max3(va_size(va), | |
395 | get_subtree_max_size(va->rb_node.rb_left), | |
396 | get_subtree_max_size(va->rb_node.rb_right)); | |
397 | } | |
398 | ||
315cc066 ML |
399 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
400 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 URS |
401 | |
402 | static void purge_vmap_area_lazy(void); | |
403 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
404 | static unsigned long lazy_max_pages(void); | |
db64fe02 | 405 | |
97105f0a RG |
406 | static atomic_long_t nr_vmalloc_pages; |
407 | ||
408 | unsigned long vmalloc_nr_pages(void) | |
409 | { | |
410 | return atomic_long_read(&nr_vmalloc_pages); | |
411 | } | |
412 | ||
db64fe02 | 413 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 414 | { |
db64fe02 NP |
415 | struct rb_node *n = vmap_area_root.rb_node; |
416 | ||
417 | while (n) { | |
418 | struct vmap_area *va; | |
419 | ||
420 | va = rb_entry(n, struct vmap_area, rb_node); | |
421 | if (addr < va->va_start) | |
422 | n = n->rb_left; | |
cef2ac3f | 423 | else if (addr >= va->va_end) |
db64fe02 NP |
424 | n = n->rb_right; |
425 | else | |
426 | return va; | |
427 | } | |
428 | ||
429 | return NULL; | |
430 | } | |
431 | ||
68ad4a33 URS |
432 | /* |
433 | * This function returns back addresses of parent node | |
434 | * and its left or right link for further processing. | |
435 | */ | |
436 | static __always_inline struct rb_node ** | |
437 | find_va_links(struct vmap_area *va, | |
438 | struct rb_root *root, struct rb_node *from, | |
439 | struct rb_node **parent) | |
440 | { | |
441 | struct vmap_area *tmp_va; | |
442 | struct rb_node **link; | |
443 | ||
444 | if (root) { | |
445 | link = &root->rb_node; | |
446 | if (unlikely(!*link)) { | |
447 | *parent = NULL; | |
448 | return link; | |
449 | } | |
450 | } else { | |
451 | link = &from; | |
452 | } | |
db64fe02 | 453 | |
68ad4a33 URS |
454 | /* |
455 | * Go to the bottom of the tree. When we hit the last point | |
456 | * we end up with parent rb_node and correct direction, i name | |
457 | * it link, where the new va->rb_node will be attached to. | |
458 | */ | |
459 | do { | |
460 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 461 | |
68ad4a33 URS |
462 | /* |
463 | * During the traversal we also do some sanity check. | |
464 | * Trigger the BUG() if there are sides(left/right) | |
465 | * or full overlaps. | |
466 | */ | |
467 | if (va->va_start < tmp_va->va_end && | |
468 | va->va_end <= tmp_va->va_start) | |
469 | link = &(*link)->rb_left; | |
470 | else if (va->va_end > tmp_va->va_start && | |
471 | va->va_start >= tmp_va->va_end) | |
472 | link = &(*link)->rb_right; | |
db64fe02 NP |
473 | else |
474 | BUG(); | |
68ad4a33 URS |
475 | } while (*link); |
476 | ||
477 | *parent = &tmp_va->rb_node; | |
478 | return link; | |
479 | } | |
480 | ||
481 | static __always_inline struct list_head * | |
482 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
483 | { | |
484 | struct list_head *list; | |
485 | ||
486 | if (unlikely(!parent)) | |
487 | /* | |
488 | * The red-black tree where we try to find VA neighbors | |
489 | * before merging or inserting is empty, i.e. it means | |
490 | * there is no free vmap space. Normally it does not | |
491 | * happen but we handle this case anyway. | |
492 | */ | |
493 | return NULL; | |
494 | ||
495 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
496 | return (&parent->rb_right == link ? list->next : list); | |
497 | } | |
498 | ||
499 | static __always_inline void | |
500 | link_va(struct vmap_area *va, struct rb_root *root, | |
501 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
502 | { | |
503 | /* | |
504 | * VA is still not in the list, but we can | |
505 | * identify its future previous list_head node. | |
506 | */ | |
507 | if (likely(parent)) { | |
508 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
509 | if (&parent->rb_right != link) | |
510 | head = head->prev; | |
db64fe02 NP |
511 | } |
512 | ||
68ad4a33 URS |
513 | /* Insert to the rb-tree */ |
514 | rb_link_node(&va->rb_node, parent, link); | |
515 | if (root == &free_vmap_area_root) { | |
516 | /* | |
517 | * Some explanation here. Just perform simple insertion | |
518 | * to the tree. We do not set va->subtree_max_size to | |
519 | * its current size before calling rb_insert_augmented(). | |
520 | * It is because of we populate the tree from the bottom | |
521 | * to parent levels when the node _is_ in the tree. | |
522 | * | |
523 | * Therefore we set subtree_max_size to zero after insertion, | |
524 | * to let __augment_tree_propagate_from() puts everything to | |
525 | * the correct order later on. | |
526 | */ | |
527 | rb_insert_augmented(&va->rb_node, | |
528 | root, &free_vmap_area_rb_augment_cb); | |
529 | va->subtree_max_size = 0; | |
530 | } else { | |
531 | rb_insert_color(&va->rb_node, root); | |
532 | } | |
db64fe02 | 533 | |
68ad4a33 URS |
534 | /* Address-sort this list */ |
535 | list_add(&va->list, head); | |
db64fe02 NP |
536 | } |
537 | ||
68ad4a33 URS |
538 | static __always_inline void |
539 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
540 | { | |
460e42d1 URS |
541 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
542 | return; | |
db64fe02 | 543 | |
460e42d1 URS |
544 | if (root == &free_vmap_area_root) |
545 | rb_erase_augmented(&va->rb_node, | |
546 | root, &free_vmap_area_rb_augment_cb); | |
547 | else | |
548 | rb_erase(&va->rb_node, root); | |
549 | ||
550 | list_del(&va->list); | |
551 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
552 | } |
553 | ||
bb850f4d URS |
554 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
555 | static void | |
556 | augment_tree_propagate_check(struct rb_node *n) | |
557 | { | |
558 | struct vmap_area *va; | |
559 | struct rb_node *node; | |
560 | unsigned long size; | |
561 | bool found = false; | |
562 | ||
563 | if (n == NULL) | |
564 | return; | |
565 | ||
566 | va = rb_entry(n, struct vmap_area, rb_node); | |
567 | size = va->subtree_max_size; | |
568 | node = n; | |
569 | ||
570 | while (node) { | |
571 | va = rb_entry(node, struct vmap_area, rb_node); | |
572 | ||
573 | if (get_subtree_max_size(node->rb_left) == size) { | |
574 | node = node->rb_left; | |
575 | } else { | |
576 | if (va_size(va) == size) { | |
577 | found = true; | |
578 | break; | |
579 | } | |
580 | ||
581 | node = node->rb_right; | |
582 | } | |
583 | } | |
584 | ||
585 | if (!found) { | |
586 | va = rb_entry(n, struct vmap_area, rb_node); | |
587 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
588 | va_size(va), va->subtree_max_size); | |
589 | } | |
590 | ||
591 | augment_tree_propagate_check(n->rb_left); | |
592 | augment_tree_propagate_check(n->rb_right); | |
593 | } | |
594 | #endif | |
595 | ||
68ad4a33 URS |
596 | /* |
597 | * This function populates subtree_max_size from bottom to upper | |
598 | * levels starting from VA point. The propagation must be done | |
599 | * when VA size is modified by changing its va_start/va_end. Or | |
600 | * in case of newly inserting of VA to the tree. | |
601 | * | |
602 | * It means that __augment_tree_propagate_from() must be called: | |
603 | * - After VA has been inserted to the tree(free path); | |
604 | * - After VA has been shrunk(allocation path); | |
605 | * - After VA has been increased(merging path). | |
606 | * | |
607 | * Please note that, it does not mean that upper parent nodes | |
608 | * and their subtree_max_size are recalculated all the time up | |
609 | * to the root node. | |
610 | * | |
611 | * 4--8 | |
612 | * /\ | |
613 | * / \ | |
614 | * / \ | |
615 | * 2--2 8--8 | |
616 | * | |
617 | * For example if we modify the node 4, shrinking it to 2, then | |
618 | * no any modification is required. If we shrink the node 2 to 1 | |
619 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
620 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
621 | * node becomes 4--6. | |
622 | */ | |
623 | static __always_inline void | |
624 | augment_tree_propagate_from(struct vmap_area *va) | |
625 | { | |
626 | struct rb_node *node = &va->rb_node; | |
627 | unsigned long new_va_sub_max_size; | |
628 | ||
629 | while (node) { | |
630 | va = rb_entry(node, struct vmap_area, rb_node); | |
631 | new_va_sub_max_size = compute_subtree_max_size(va); | |
632 | ||
633 | /* | |
634 | * If the newly calculated maximum available size of the | |
635 | * subtree is equal to the current one, then it means that | |
636 | * the tree is propagated correctly. So we have to stop at | |
637 | * this point to save cycles. | |
638 | */ | |
639 | if (va->subtree_max_size == new_va_sub_max_size) | |
640 | break; | |
641 | ||
642 | va->subtree_max_size = new_va_sub_max_size; | |
643 | node = rb_parent(&va->rb_node); | |
644 | } | |
bb850f4d URS |
645 | |
646 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
647 | augment_tree_propagate_check(free_vmap_area_root.rb_node); | |
648 | #endif | |
68ad4a33 URS |
649 | } |
650 | ||
651 | static void | |
652 | insert_vmap_area(struct vmap_area *va, | |
653 | struct rb_root *root, struct list_head *head) | |
654 | { | |
655 | struct rb_node **link; | |
656 | struct rb_node *parent; | |
657 | ||
658 | link = find_va_links(va, root, NULL, &parent); | |
659 | link_va(va, root, parent, link, head); | |
660 | } | |
661 | ||
662 | static void | |
663 | insert_vmap_area_augment(struct vmap_area *va, | |
664 | struct rb_node *from, struct rb_root *root, | |
665 | struct list_head *head) | |
666 | { | |
667 | struct rb_node **link; | |
668 | struct rb_node *parent; | |
669 | ||
670 | if (from) | |
671 | link = find_va_links(va, NULL, from, &parent); | |
672 | else | |
673 | link = find_va_links(va, root, NULL, &parent); | |
674 | ||
675 | link_va(va, root, parent, link, head); | |
676 | augment_tree_propagate_from(va); | |
677 | } | |
678 | ||
679 | /* | |
680 | * Merge de-allocated chunk of VA memory with previous | |
681 | * and next free blocks. If coalesce is not done a new | |
682 | * free area is inserted. If VA has been merged, it is | |
683 | * freed. | |
684 | */ | |
685 | static __always_inline void | |
686 | merge_or_add_vmap_area(struct vmap_area *va, | |
687 | struct rb_root *root, struct list_head *head) | |
688 | { | |
689 | struct vmap_area *sibling; | |
690 | struct list_head *next; | |
691 | struct rb_node **link; | |
692 | struct rb_node *parent; | |
693 | bool merged = false; | |
694 | ||
695 | /* | |
696 | * Find a place in the tree where VA potentially will be | |
697 | * inserted, unless it is merged with its sibling/siblings. | |
698 | */ | |
699 | link = find_va_links(va, root, NULL, &parent); | |
700 | ||
701 | /* | |
702 | * Get next node of VA to check if merging can be done. | |
703 | */ | |
704 | next = get_va_next_sibling(parent, link); | |
705 | if (unlikely(next == NULL)) | |
706 | goto insert; | |
707 | ||
708 | /* | |
709 | * start end | |
710 | * | | | |
711 | * |<------VA------>|<-----Next----->| | |
712 | * | | | |
713 | * start end | |
714 | */ | |
715 | if (next != head) { | |
716 | sibling = list_entry(next, struct vmap_area, list); | |
717 | if (sibling->va_start == va->va_end) { | |
718 | sibling->va_start = va->va_start; | |
719 | ||
720 | /* Check and update the tree if needed. */ | |
721 | augment_tree_propagate_from(sibling); | |
722 | ||
68ad4a33 URS |
723 | /* Free vmap_area object. */ |
724 | kmem_cache_free(vmap_area_cachep, va); | |
725 | ||
726 | /* Point to the new merged area. */ | |
727 | va = sibling; | |
728 | merged = true; | |
729 | } | |
730 | } | |
731 | ||
732 | /* | |
733 | * start end | |
734 | * | | | |
735 | * |<-----Prev----->|<------VA------>| | |
736 | * | | | |
737 | * start end | |
738 | */ | |
739 | if (next->prev != head) { | |
740 | sibling = list_entry(next->prev, struct vmap_area, list); | |
741 | if (sibling->va_end == va->va_start) { | |
742 | sibling->va_end = va->va_end; | |
743 | ||
744 | /* Check and update the tree if needed. */ | |
745 | augment_tree_propagate_from(sibling); | |
746 | ||
54f63d9d URS |
747 | if (merged) |
748 | unlink_va(va, root); | |
68ad4a33 URS |
749 | |
750 | /* Free vmap_area object. */ | |
751 | kmem_cache_free(vmap_area_cachep, va); | |
68ad4a33 URS |
752 | return; |
753 | } | |
754 | } | |
755 | ||
756 | insert: | |
757 | if (!merged) { | |
758 | link_va(va, root, parent, link, head); | |
759 | augment_tree_propagate_from(va); | |
760 | } | |
761 | } | |
762 | ||
763 | static __always_inline bool | |
764 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
765 | unsigned long align, unsigned long vstart) | |
766 | { | |
767 | unsigned long nva_start_addr; | |
768 | ||
769 | if (va->va_start > vstart) | |
770 | nva_start_addr = ALIGN(va->va_start, align); | |
771 | else | |
772 | nva_start_addr = ALIGN(vstart, align); | |
773 | ||
774 | /* Can be overflowed due to big size or alignment. */ | |
775 | if (nva_start_addr + size < nva_start_addr || | |
776 | nva_start_addr < vstart) | |
777 | return false; | |
778 | ||
779 | return (nva_start_addr + size <= va->va_end); | |
780 | } | |
781 | ||
782 | /* | |
783 | * Find the first free block(lowest start address) in the tree, | |
784 | * that will accomplish the request corresponding to passing | |
785 | * parameters. | |
786 | */ | |
787 | static __always_inline struct vmap_area * | |
788 | find_vmap_lowest_match(unsigned long size, | |
789 | unsigned long align, unsigned long vstart) | |
790 | { | |
791 | struct vmap_area *va; | |
792 | struct rb_node *node; | |
793 | unsigned long length; | |
794 | ||
795 | /* Start from the root. */ | |
796 | node = free_vmap_area_root.rb_node; | |
797 | ||
798 | /* Adjust the search size for alignment overhead. */ | |
799 | length = size + align - 1; | |
800 | ||
801 | while (node) { | |
802 | va = rb_entry(node, struct vmap_area, rb_node); | |
803 | ||
804 | if (get_subtree_max_size(node->rb_left) >= length && | |
805 | vstart < va->va_start) { | |
806 | node = node->rb_left; | |
807 | } else { | |
808 | if (is_within_this_va(va, size, align, vstart)) | |
809 | return va; | |
810 | ||
811 | /* | |
812 | * Does not make sense to go deeper towards the right | |
813 | * sub-tree if it does not have a free block that is | |
814 | * equal or bigger to the requested search length. | |
815 | */ | |
816 | if (get_subtree_max_size(node->rb_right) >= length) { | |
817 | node = node->rb_right; | |
818 | continue; | |
819 | } | |
820 | ||
821 | /* | |
3806b041 | 822 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
823 | * that will satisfy the search criteria. It can happen |
824 | * only once due to "vstart" restriction. | |
825 | */ | |
826 | while ((node = rb_parent(node))) { | |
827 | va = rb_entry(node, struct vmap_area, rb_node); | |
828 | if (is_within_this_va(va, size, align, vstart)) | |
829 | return va; | |
830 | ||
831 | if (get_subtree_max_size(node->rb_right) >= length && | |
832 | vstart <= va->va_start) { | |
833 | node = node->rb_right; | |
834 | break; | |
835 | } | |
836 | } | |
837 | } | |
838 | } | |
839 | ||
840 | return NULL; | |
841 | } | |
842 | ||
a6cf4e0f URS |
843 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
844 | #include <linux/random.h> | |
845 | ||
846 | static struct vmap_area * | |
847 | find_vmap_lowest_linear_match(unsigned long size, | |
848 | unsigned long align, unsigned long vstart) | |
849 | { | |
850 | struct vmap_area *va; | |
851 | ||
852 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
853 | if (!is_within_this_va(va, size, align, vstart)) | |
854 | continue; | |
855 | ||
856 | return va; | |
857 | } | |
858 | ||
859 | return NULL; | |
860 | } | |
861 | ||
862 | static void | |
863 | find_vmap_lowest_match_check(unsigned long size) | |
864 | { | |
865 | struct vmap_area *va_1, *va_2; | |
866 | unsigned long vstart; | |
867 | unsigned int rnd; | |
868 | ||
869 | get_random_bytes(&rnd, sizeof(rnd)); | |
870 | vstart = VMALLOC_START + rnd; | |
871 | ||
872 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
873 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
874 | ||
875 | if (va_1 != va_2) | |
876 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
877 | va_1, va_2, vstart); | |
878 | } | |
879 | #endif | |
880 | ||
68ad4a33 URS |
881 | enum fit_type { |
882 | NOTHING_FIT = 0, | |
883 | FL_FIT_TYPE = 1, /* full fit */ | |
884 | LE_FIT_TYPE = 2, /* left edge fit */ | |
885 | RE_FIT_TYPE = 3, /* right edge fit */ | |
886 | NE_FIT_TYPE = 4 /* no edge fit */ | |
887 | }; | |
888 | ||
889 | static __always_inline enum fit_type | |
890 | classify_va_fit_type(struct vmap_area *va, | |
891 | unsigned long nva_start_addr, unsigned long size) | |
892 | { | |
893 | enum fit_type type; | |
894 | ||
895 | /* Check if it is within VA. */ | |
896 | if (nva_start_addr < va->va_start || | |
897 | nva_start_addr + size > va->va_end) | |
898 | return NOTHING_FIT; | |
899 | ||
900 | /* Now classify. */ | |
901 | if (va->va_start == nva_start_addr) { | |
902 | if (va->va_end == nva_start_addr + size) | |
903 | type = FL_FIT_TYPE; | |
904 | else | |
905 | type = LE_FIT_TYPE; | |
906 | } else if (va->va_end == nva_start_addr + size) { | |
907 | type = RE_FIT_TYPE; | |
908 | } else { | |
909 | type = NE_FIT_TYPE; | |
910 | } | |
911 | ||
912 | return type; | |
913 | } | |
914 | ||
915 | static __always_inline int | |
916 | adjust_va_to_fit_type(struct vmap_area *va, | |
917 | unsigned long nva_start_addr, unsigned long size, | |
918 | enum fit_type type) | |
919 | { | |
2c929233 | 920 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
921 | |
922 | if (type == FL_FIT_TYPE) { | |
923 | /* | |
924 | * No need to split VA, it fully fits. | |
925 | * | |
926 | * | | | |
927 | * V NVA V | |
928 | * |---------------| | |
929 | */ | |
930 | unlink_va(va, &free_vmap_area_root); | |
931 | kmem_cache_free(vmap_area_cachep, va); | |
932 | } else if (type == LE_FIT_TYPE) { | |
933 | /* | |
934 | * Split left edge of fit VA. | |
935 | * | |
936 | * | | | |
937 | * V NVA V R | |
938 | * |-------|-------| | |
939 | */ | |
940 | va->va_start += size; | |
941 | } else if (type == RE_FIT_TYPE) { | |
942 | /* | |
943 | * Split right edge of fit VA. | |
944 | * | |
945 | * | | | |
946 | * L V NVA V | |
947 | * |-------|-------| | |
948 | */ | |
949 | va->va_end = nva_start_addr; | |
950 | } else if (type == NE_FIT_TYPE) { | |
951 | /* | |
952 | * Split no edge of fit VA. | |
953 | * | |
954 | * | | | |
955 | * L V NVA V R | |
956 | * |---|-------|---| | |
957 | */ | |
82dd23e8 URS |
958 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
959 | if (unlikely(!lva)) { | |
960 | /* | |
961 | * For percpu allocator we do not do any pre-allocation | |
962 | * and leave it as it is. The reason is it most likely | |
963 | * never ends up with NE_FIT_TYPE splitting. In case of | |
964 | * percpu allocations offsets and sizes are aligned to | |
965 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
966 | * are its main fitting cases. | |
967 | * | |
968 | * There are a few exceptions though, as an example it is | |
969 | * a first allocation (early boot up) when we have "one" | |
970 | * big free space that has to be split. | |
060650a2 URS |
971 | * |
972 | * Also we can hit this path in case of regular "vmap" | |
973 | * allocations, if "this" current CPU was not preloaded. | |
974 | * See the comment in alloc_vmap_area() why. If so, then | |
975 | * GFP_NOWAIT is used instead to get an extra object for | |
976 | * split purpose. That is rare and most time does not | |
977 | * occur. | |
978 | * | |
979 | * What happens if an allocation gets failed. Basically, | |
980 | * an "overflow" path is triggered to purge lazily freed | |
981 | * areas to free some memory, then, the "retry" path is | |
982 | * triggered to repeat one more time. See more details | |
983 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
984 | */ |
985 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
986 | if (!lva) | |
987 | return -1; | |
988 | } | |
68ad4a33 URS |
989 | |
990 | /* | |
991 | * Build the remainder. | |
992 | */ | |
993 | lva->va_start = va->va_start; | |
994 | lva->va_end = nva_start_addr; | |
995 | ||
996 | /* | |
997 | * Shrink this VA to remaining size. | |
998 | */ | |
999 | va->va_start = nva_start_addr + size; | |
1000 | } else { | |
1001 | return -1; | |
1002 | } | |
1003 | ||
1004 | if (type != FL_FIT_TYPE) { | |
1005 | augment_tree_propagate_from(va); | |
1006 | ||
2c929233 | 1007 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
1008 | insert_vmap_area_augment(lva, &va->rb_node, |
1009 | &free_vmap_area_root, &free_vmap_area_list); | |
1010 | } | |
1011 | ||
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Returns a start address of the newly allocated area, if success. | |
1017 | * Otherwise a vend is returned that indicates failure. | |
1018 | */ | |
1019 | static __always_inline unsigned long | |
1020 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1021 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1022 | { |
1023 | unsigned long nva_start_addr; | |
1024 | struct vmap_area *va; | |
1025 | enum fit_type type; | |
1026 | int ret; | |
1027 | ||
1028 | va = find_vmap_lowest_match(size, align, vstart); | |
1029 | if (unlikely(!va)) | |
1030 | return vend; | |
1031 | ||
1032 | if (va->va_start > vstart) | |
1033 | nva_start_addr = ALIGN(va->va_start, align); | |
1034 | else | |
1035 | nva_start_addr = ALIGN(vstart, align); | |
1036 | ||
1037 | /* Check the "vend" restriction. */ | |
1038 | if (nva_start_addr + size > vend) | |
1039 | return vend; | |
1040 | ||
1041 | /* Classify what we have found. */ | |
1042 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1043 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1044 | return vend; | |
1045 | ||
1046 | /* Update the free vmap_area. */ | |
1047 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1048 | if (ret) | |
1049 | return vend; | |
1050 | ||
a6cf4e0f URS |
1051 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1052 | find_vmap_lowest_match_check(size); | |
1053 | #endif | |
1054 | ||
68ad4a33 URS |
1055 | return nva_start_addr; |
1056 | } | |
4da56b99 | 1057 | |
db64fe02 NP |
1058 | /* |
1059 | * Allocate a region of KVA of the specified size and alignment, within the | |
1060 | * vstart and vend. | |
1061 | */ | |
1062 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1063 | unsigned long align, | |
1064 | unsigned long vstart, unsigned long vend, | |
1065 | int node, gfp_t gfp_mask) | |
1066 | { | |
82dd23e8 | 1067 | struct vmap_area *va, *pva; |
1da177e4 | 1068 | unsigned long addr; |
db64fe02 NP |
1069 | int purged = 0; |
1070 | ||
7766970c | 1071 | BUG_ON(!size); |
891c49ab | 1072 | BUG_ON(offset_in_page(size)); |
89699605 | 1073 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1074 | |
68ad4a33 URS |
1075 | if (unlikely(!vmap_initialized)) |
1076 | return ERR_PTR(-EBUSY); | |
1077 | ||
5803ed29 | 1078 | might_sleep(); |
f07116d7 | 1079 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1080 | |
f07116d7 | 1081 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1082 | if (unlikely(!va)) |
1083 | return ERR_PTR(-ENOMEM); | |
1084 | ||
7f88f88f CM |
1085 | /* |
1086 | * Only scan the relevant parts containing pointers to other objects | |
1087 | * to avoid false negatives. | |
1088 | */ | |
f07116d7 | 1089 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1090 | |
db64fe02 | 1091 | retry: |
82dd23e8 | 1092 | /* |
81f1ba58 URS |
1093 | * Preload this CPU with one extra vmap_area object. It is used |
1094 | * when fit type of free area is NE_FIT_TYPE. Please note, it | |
1095 | * does not guarantee that an allocation occurs on a CPU that | |
1096 | * is preloaded, instead we minimize the case when it is not. | |
1097 | * It can happen because of cpu migration, because there is a | |
1098 | * race until the below spinlock is taken. | |
82dd23e8 URS |
1099 | * |
1100 | * The preload is done in non-atomic context, thus it allows us | |
1101 | * to use more permissive allocation masks to be more stable under | |
81f1ba58 URS |
1102 | * low memory condition and high memory pressure. In rare case, |
1103 | * if not preloaded, GFP_NOWAIT is used. | |
82dd23e8 | 1104 | * |
81f1ba58 | 1105 | * Set "pva" to NULL here, because of "retry" path. |
82dd23e8 | 1106 | */ |
81f1ba58 | 1107 | pva = NULL; |
82dd23e8 | 1108 | |
81f1ba58 URS |
1109 | if (!this_cpu_read(ne_fit_preload_node)) |
1110 | /* | |
1111 | * Even if it fails we do not really care about that. | |
1112 | * Just proceed as it is. If needed "overflow" path | |
1113 | * will refill the cache we allocate from. | |
1114 | */ | |
f07116d7 | 1115 | pva = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
82dd23e8 | 1116 | |
db64fe02 | 1117 | spin_lock(&vmap_area_lock); |
81f1ba58 URS |
1118 | |
1119 | if (pva && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) | |
1120 | kmem_cache_free(vmap_area_cachep, pva); | |
89699605 | 1121 | |
afd07389 | 1122 | /* |
68ad4a33 URS |
1123 | * If an allocation fails, the "vend" address is |
1124 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1125 | */ |
cacca6ba | 1126 | addr = __alloc_vmap_area(size, align, vstart, vend); |
68ad4a33 | 1127 | if (unlikely(addr == vend)) |
89699605 | 1128 | goto overflow; |
db64fe02 NP |
1129 | |
1130 | va->va_start = addr; | |
1131 | va->va_end = addr + size; | |
688fcbfc | 1132 | va->vm = NULL; |
68ad4a33 URS |
1133 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
1134 | ||
db64fe02 NP |
1135 | spin_unlock(&vmap_area_lock); |
1136 | ||
61e16557 | 1137 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1138 | BUG_ON(va->va_start < vstart); |
1139 | BUG_ON(va->va_end > vend); | |
1140 | ||
db64fe02 | 1141 | return va; |
89699605 NP |
1142 | |
1143 | overflow: | |
1144 | spin_unlock(&vmap_area_lock); | |
1145 | if (!purged) { | |
1146 | purge_vmap_area_lazy(); | |
1147 | purged = 1; | |
1148 | goto retry; | |
1149 | } | |
4da56b99 CW |
1150 | |
1151 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1152 | unsigned long freed = 0; | |
1153 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1154 | if (freed > 0) { | |
1155 | purged = 0; | |
1156 | goto retry; | |
1157 | } | |
1158 | } | |
1159 | ||
03497d76 | 1160 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1161 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1162 | size); | |
68ad4a33 URS |
1163 | |
1164 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1165 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1166 | } |
1167 | ||
4da56b99 CW |
1168 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1169 | { | |
1170 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1171 | } | |
1172 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1173 | ||
1174 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1175 | { | |
1176 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1177 | } | |
1178 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1179 | ||
db64fe02 NP |
1180 | static void __free_vmap_area(struct vmap_area *va) |
1181 | { | |
ca23e405 | 1182 | /* |
68ad4a33 | 1183 | * Remove from the busy tree/list. |
ca23e405 | 1184 | */ |
68ad4a33 | 1185 | unlink_va(va, &vmap_area_root); |
ca23e405 | 1186 | |
68ad4a33 URS |
1187 | /* |
1188 | * Merge VA with its neighbors, otherwise just add it. | |
1189 | */ | |
1190 | merge_or_add_vmap_area(va, | |
1191 | &free_vmap_area_root, &free_vmap_area_list); | |
db64fe02 NP |
1192 | } |
1193 | ||
1194 | /* | |
1195 | * Free a region of KVA allocated by alloc_vmap_area | |
1196 | */ | |
1197 | static void free_vmap_area(struct vmap_area *va) | |
1198 | { | |
1199 | spin_lock(&vmap_area_lock); | |
1200 | __free_vmap_area(va); | |
1201 | spin_unlock(&vmap_area_lock); | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Clear the pagetable entries of a given vmap_area | |
1206 | */ | |
1207 | static void unmap_vmap_area(struct vmap_area *va) | |
1208 | { | |
1209 | vunmap_page_range(va->va_start, va->va_end); | |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1214 | * before attempting to purge with a TLB flush. | |
1215 | * | |
1216 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1217 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1218 | * global TLB flushes that must be performed. It would seem natural to scale | |
1219 | * this number up linearly with the number of CPUs (because vmapping activity | |
1220 | * could also scale linearly with the number of CPUs), however it is likely | |
1221 | * that in practice, workloads might be constrained in other ways that mean | |
1222 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1223 | * conservative and not introduce a big latency on huge systems, so go with | |
1224 | * a less aggressive log scale. It will still be an improvement over the old | |
1225 | * code, and it will be simple to change the scale factor if we find that it | |
1226 | * becomes a problem on bigger systems. | |
1227 | */ | |
1228 | static unsigned long lazy_max_pages(void) | |
1229 | { | |
1230 | unsigned int log; | |
1231 | ||
1232 | log = fls(num_online_cpus()); | |
1233 | ||
1234 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1235 | } | |
1236 | ||
4d36e6f8 | 1237 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1238 | |
0574ecd1 CH |
1239 | /* |
1240 | * Serialize vmap purging. There is no actual criticial section protected | |
1241 | * by this look, but we want to avoid concurrent calls for performance | |
1242 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1243 | */ | |
f9e09977 | 1244 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1245 | |
02b709df NP |
1246 | /* for per-CPU blocks */ |
1247 | static void purge_fragmented_blocks_allcpus(void); | |
1248 | ||
3ee48b6a CW |
1249 | /* |
1250 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1251 | * immediately freed. | |
1252 | */ | |
1253 | void set_iounmap_nonlazy(void) | |
1254 | { | |
4d36e6f8 | 1255 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1256 | } |
1257 | ||
db64fe02 NP |
1258 | /* |
1259 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1260 | */ |
0574ecd1 | 1261 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1262 | { |
4d36e6f8 | 1263 | unsigned long resched_threshold; |
80c4bd7a | 1264 | struct llist_node *valist; |
db64fe02 | 1265 | struct vmap_area *va; |
cbb76676 | 1266 | struct vmap_area *n_va; |
db64fe02 | 1267 | |
0574ecd1 | 1268 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1269 | |
80c4bd7a | 1270 | valist = llist_del_all(&vmap_purge_list); |
68571be9 URS |
1271 | if (unlikely(valist == NULL)) |
1272 | return false; | |
1273 | ||
3f8fd02b JR |
1274 | /* |
1275 | * First make sure the mappings are removed from all page-tables | |
1276 | * before they are freed. | |
1277 | */ | |
1278 | vmalloc_sync_all(); | |
1279 | ||
68571be9 URS |
1280 | /* |
1281 | * TODO: to calculate a flush range without looping. | |
1282 | * The list can be up to lazy_max_pages() elements. | |
1283 | */ | |
80c4bd7a | 1284 | llist_for_each_entry(va, valist, purge_list) { |
0574ecd1 CH |
1285 | if (va->va_start < start) |
1286 | start = va->va_start; | |
1287 | if (va->va_end > end) | |
1288 | end = va->va_end; | |
db64fe02 | 1289 | } |
db64fe02 | 1290 | |
0574ecd1 | 1291 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1292 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1293 | |
0574ecd1 | 1294 | spin_lock(&vmap_area_lock); |
763b218d | 1295 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
4d36e6f8 | 1296 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
763b218d | 1297 | |
dd3b8353 URS |
1298 | /* |
1299 | * Finally insert or merge lazily-freed area. It is | |
1300 | * detached and there is no need to "unlink" it from | |
1301 | * anything. | |
1302 | */ | |
1303 | merge_or_add_vmap_area(va, | |
1304 | &free_vmap_area_root, &free_vmap_area_list); | |
1305 | ||
4d36e6f8 | 1306 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1307 | |
4d36e6f8 | 1308 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
68571be9 | 1309 | cond_resched_lock(&vmap_area_lock); |
763b218d | 1310 | } |
0574ecd1 CH |
1311 | spin_unlock(&vmap_area_lock); |
1312 | return true; | |
db64fe02 NP |
1313 | } |
1314 | ||
496850e5 NP |
1315 | /* |
1316 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1317 | * is already purging. | |
1318 | */ | |
1319 | static void try_purge_vmap_area_lazy(void) | |
1320 | { | |
f9e09977 | 1321 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1322 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1323 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1324 | } |
496850e5 NP |
1325 | } |
1326 | ||
db64fe02 NP |
1327 | /* |
1328 | * Kick off a purge of the outstanding lazy areas. | |
1329 | */ | |
1330 | static void purge_vmap_area_lazy(void) | |
1331 | { | |
f9e09977 | 1332 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1333 | purge_fragmented_blocks_allcpus(); |
1334 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1335 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1336 | } |
1337 | ||
1338 | /* | |
64141da5 JF |
1339 | * Free a vmap area, caller ensuring that the area has been unmapped |
1340 | * and flush_cache_vunmap had been called for the correct range | |
1341 | * previously. | |
db64fe02 | 1342 | */ |
64141da5 | 1343 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1344 | { |
4d36e6f8 | 1345 | unsigned long nr_lazy; |
80c4bd7a | 1346 | |
dd3b8353 URS |
1347 | spin_lock(&vmap_area_lock); |
1348 | unlink_va(va, &vmap_area_root); | |
1349 | spin_unlock(&vmap_area_lock); | |
1350 | ||
4d36e6f8 URS |
1351 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1352 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a CW |
1353 | |
1354 | /* After this point, we may free va at any time */ | |
1355 | llist_add(&va->purge_list, &vmap_purge_list); | |
1356 | ||
1357 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 1358 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1359 | } |
1360 | ||
b29acbdc NP |
1361 | /* |
1362 | * Free and unmap a vmap area | |
1363 | */ | |
1364 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1365 | { | |
1366 | flush_cache_vunmap(va->va_start, va->va_end); | |
c8eef01e | 1367 | unmap_vmap_area(va); |
82a2e924 CP |
1368 | if (debug_pagealloc_enabled()) |
1369 | flush_tlb_kernel_range(va->va_start, va->va_end); | |
1370 | ||
c8eef01e | 1371 | free_vmap_area_noflush(va); |
b29acbdc NP |
1372 | } |
1373 | ||
db64fe02 NP |
1374 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1375 | { | |
1376 | struct vmap_area *va; | |
1377 | ||
1378 | spin_lock(&vmap_area_lock); | |
1379 | va = __find_vmap_area(addr); | |
1380 | spin_unlock(&vmap_area_lock); | |
1381 | ||
1382 | return va; | |
1383 | } | |
1384 | ||
db64fe02 NP |
1385 | /*** Per cpu kva allocator ***/ |
1386 | ||
1387 | /* | |
1388 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1389 | * room for at least 16 percpu vmap blocks per CPU. | |
1390 | */ | |
1391 | /* | |
1392 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1393 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1394 | * instead (we just need a rough idea) | |
1395 | */ | |
1396 | #if BITS_PER_LONG == 32 | |
1397 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1398 | #else | |
1399 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1400 | #endif | |
1401 | ||
1402 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1403 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1404 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1405 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1406 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1407 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1408 | #define VMAP_BBMAP_BITS \ |
1409 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1410 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1411 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1412 | |
1413 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1414 | ||
1415 | struct vmap_block_queue { | |
1416 | spinlock_t lock; | |
1417 | struct list_head free; | |
db64fe02 NP |
1418 | }; |
1419 | ||
1420 | struct vmap_block { | |
1421 | spinlock_t lock; | |
1422 | struct vmap_area *va; | |
db64fe02 | 1423 | unsigned long free, dirty; |
7d61bfe8 | 1424 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1425 | struct list_head free_list; |
1426 | struct rcu_head rcu_head; | |
02b709df | 1427 | struct list_head purge; |
db64fe02 NP |
1428 | }; |
1429 | ||
1430 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1431 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1432 | ||
1433 | /* | |
1434 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
1435 | * in the free path. Could get rid of this if we change the API to return a | |
1436 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1437 | */ | |
1438 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
1439 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
1440 | ||
1441 | /* | |
1442 | * We should probably have a fallback mechanism to allocate virtual memory | |
1443 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1444 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1445 | * big problem. | |
1446 | */ | |
1447 | ||
1448 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1449 | { | |
1450 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1451 | addr /= VMAP_BLOCK_SIZE; | |
1452 | return addr; | |
1453 | } | |
1454 | ||
cf725ce2 RP |
1455 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1456 | { | |
1457 | unsigned long addr; | |
1458 | ||
1459 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1460 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1461 | return (void *)addr; | |
1462 | } | |
1463 | ||
1464 | /** | |
1465 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1466 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1467 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1468 | * @gfp_mask: flags for the page level allocator | |
1469 | * | |
a862f68a | 1470 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1471 | */ |
1472 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1473 | { |
1474 | struct vmap_block_queue *vbq; | |
1475 | struct vmap_block *vb; | |
1476 | struct vmap_area *va; | |
1477 | unsigned long vb_idx; | |
1478 | int node, err; | |
cf725ce2 | 1479 | void *vaddr; |
db64fe02 NP |
1480 | |
1481 | node = numa_node_id(); | |
1482 | ||
1483 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1484 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1485 | if (unlikely(!vb)) | |
1486 | return ERR_PTR(-ENOMEM); | |
1487 | ||
1488 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1489 | VMALLOC_START, VMALLOC_END, | |
1490 | node, gfp_mask); | |
ddf9c6d4 | 1491 | if (IS_ERR(va)) { |
db64fe02 | 1492 | kfree(vb); |
e7d86340 | 1493 | return ERR_CAST(va); |
db64fe02 NP |
1494 | } |
1495 | ||
1496 | err = radix_tree_preload(gfp_mask); | |
1497 | if (unlikely(err)) { | |
1498 | kfree(vb); | |
1499 | free_vmap_area(va); | |
1500 | return ERR_PTR(err); | |
1501 | } | |
1502 | ||
cf725ce2 | 1503 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1504 | spin_lock_init(&vb->lock); |
1505 | vb->va = va; | |
cf725ce2 RP |
1506 | /* At least something should be left free */ |
1507 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1508 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1509 | vb->dirty = 0; |
7d61bfe8 RP |
1510 | vb->dirty_min = VMAP_BBMAP_BITS; |
1511 | vb->dirty_max = 0; | |
db64fe02 | 1512 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1513 | |
1514 | vb_idx = addr_to_vb_idx(va->va_start); | |
1515 | spin_lock(&vmap_block_tree_lock); | |
1516 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
1517 | spin_unlock(&vmap_block_tree_lock); | |
1518 | BUG_ON(err); | |
1519 | radix_tree_preload_end(); | |
1520 | ||
1521 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1522 | spin_lock(&vbq->lock); |
68ac546f | 1523 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1524 | spin_unlock(&vbq->lock); |
3f04ba85 | 1525 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1526 | |
cf725ce2 | 1527 | return vaddr; |
db64fe02 NP |
1528 | } |
1529 | ||
db64fe02 NP |
1530 | static void free_vmap_block(struct vmap_block *vb) |
1531 | { | |
1532 | struct vmap_block *tmp; | |
1533 | unsigned long vb_idx; | |
1534 | ||
db64fe02 NP |
1535 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
1536 | spin_lock(&vmap_block_tree_lock); | |
1537 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
1538 | spin_unlock(&vmap_block_tree_lock); | |
1539 | BUG_ON(tmp != vb); | |
1540 | ||
64141da5 | 1541 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1542 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1543 | } |
1544 | ||
02b709df NP |
1545 | static void purge_fragmented_blocks(int cpu) |
1546 | { | |
1547 | LIST_HEAD(purge); | |
1548 | struct vmap_block *vb; | |
1549 | struct vmap_block *n_vb; | |
1550 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1551 | ||
1552 | rcu_read_lock(); | |
1553 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1554 | ||
1555 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1556 | continue; | |
1557 | ||
1558 | spin_lock(&vb->lock); | |
1559 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1560 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1561 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1562 | vb->dirty_min = 0; |
1563 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1564 | spin_lock(&vbq->lock); |
1565 | list_del_rcu(&vb->free_list); | |
1566 | spin_unlock(&vbq->lock); | |
1567 | spin_unlock(&vb->lock); | |
1568 | list_add_tail(&vb->purge, &purge); | |
1569 | } else | |
1570 | spin_unlock(&vb->lock); | |
1571 | } | |
1572 | rcu_read_unlock(); | |
1573 | ||
1574 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1575 | list_del(&vb->purge); | |
1576 | free_vmap_block(vb); | |
1577 | } | |
1578 | } | |
1579 | ||
02b709df NP |
1580 | static void purge_fragmented_blocks_allcpus(void) |
1581 | { | |
1582 | int cpu; | |
1583 | ||
1584 | for_each_possible_cpu(cpu) | |
1585 | purge_fragmented_blocks(cpu); | |
1586 | } | |
1587 | ||
db64fe02 NP |
1588 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1589 | { | |
1590 | struct vmap_block_queue *vbq; | |
1591 | struct vmap_block *vb; | |
cf725ce2 | 1592 | void *vaddr = NULL; |
db64fe02 NP |
1593 | unsigned int order; |
1594 | ||
891c49ab | 1595 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1596 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1597 | if (WARN_ON(size == 0)) { |
1598 | /* | |
1599 | * Allocating 0 bytes isn't what caller wants since | |
1600 | * get_order(0) returns funny result. Just warn and terminate | |
1601 | * early. | |
1602 | */ | |
1603 | return NULL; | |
1604 | } | |
db64fe02 NP |
1605 | order = get_order(size); |
1606 | ||
db64fe02 NP |
1607 | rcu_read_lock(); |
1608 | vbq = &get_cpu_var(vmap_block_queue); | |
1609 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1610 | unsigned long pages_off; |
db64fe02 NP |
1611 | |
1612 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1613 | if (vb->free < (1UL << order)) { |
1614 | spin_unlock(&vb->lock); | |
1615 | continue; | |
1616 | } | |
02b709df | 1617 | |
cf725ce2 RP |
1618 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1619 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1620 | vb->free -= 1UL << order; |
1621 | if (vb->free == 0) { | |
1622 | spin_lock(&vbq->lock); | |
1623 | list_del_rcu(&vb->free_list); | |
1624 | spin_unlock(&vbq->lock); | |
1625 | } | |
cf725ce2 | 1626 | |
02b709df NP |
1627 | spin_unlock(&vb->lock); |
1628 | break; | |
db64fe02 | 1629 | } |
02b709df | 1630 | |
3f04ba85 | 1631 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1632 | rcu_read_unlock(); |
1633 | ||
cf725ce2 RP |
1634 | /* Allocate new block if nothing was found */ |
1635 | if (!vaddr) | |
1636 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1637 | |
cf725ce2 | 1638 | return vaddr; |
db64fe02 NP |
1639 | } |
1640 | ||
1641 | static void vb_free(const void *addr, unsigned long size) | |
1642 | { | |
1643 | unsigned long offset; | |
1644 | unsigned long vb_idx; | |
1645 | unsigned int order; | |
1646 | struct vmap_block *vb; | |
1647 | ||
891c49ab | 1648 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1649 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc NP |
1650 | |
1651 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
1652 | ||
db64fe02 NP |
1653 | order = get_order(size); |
1654 | ||
1655 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
7d61bfe8 | 1656 | offset >>= PAGE_SHIFT; |
db64fe02 NP |
1657 | |
1658 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
1659 | rcu_read_lock(); | |
1660 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1661 | rcu_read_unlock(); | |
1662 | BUG_ON(!vb); | |
1663 | ||
64141da5 JF |
1664 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
1665 | ||
82a2e924 CP |
1666 | if (debug_pagealloc_enabled()) |
1667 | flush_tlb_kernel_range((unsigned long)addr, | |
1668 | (unsigned long)addr + size); | |
1669 | ||
db64fe02 | 1670 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1671 | |
1672 | /* Expand dirty range */ | |
1673 | vb->dirty_min = min(vb->dirty_min, offset); | |
1674 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1675 | |
db64fe02 NP |
1676 | vb->dirty += 1UL << order; |
1677 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1678 | BUG_ON(vb->free); |
db64fe02 NP |
1679 | spin_unlock(&vb->lock); |
1680 | free_vmap_block(vb); | |
1681 | } else | |
1682 | spin_unlock(&vb->lock); | |
1683 | } | |
1684 | ||
868b104d | 1685 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1686 | { |
db64fe02 | 1687 | int cpu; |
db64fe02 | 1688 | |
9b463334 JF |
1689 | if (unlikely(!vmap_initialized)) |
1690 | return; | |
1691 | ||
5803ed29 CH |
1692 | might_sleep(); |
1693 | ||
db64fe02 NP |
1694 | for_each_possible_cpu(cpu) { |
1695 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1696 | struct vmap_block *vb; | |
1697 | ||
1698 | rcu_read_lock(); | |
1699 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1700 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1701 | if (vb->dirty) { |
1702 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1703 | unsigned long s, e; |
b136be5e | 1704 | |
7d61bfe8 RP |
1705 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1706 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1707 | |
7d61bfe8 RP |
1708 | start = min(s, start); |
1709 | end = max(e, end); | |
db64fe02 | 1710 | |
7d61bfe8 | 1711 | flush = 1; |
db64fe02 NP |
1712 | } |
1713 | spin_unlock(&vb->lock); | |
1714 | } | |
1715 | rcu_read_unlock(); | |
1716 | } | |
1717 | ||
f9e09977 | 1718 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1719 | purge_fragmented_blocks_allcpus(); |
1720 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1721 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1722 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 1723 | } |
868b104d RE |
1724 | |
1725 | /** | |
1726 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1727 | * | |
1728 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1729 | * to amortize TLB flushing overheads. What this means is that any page you | |
1730 | * have now, may, in a former life, have been mapped into kernel virtual | |
1731 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1732 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1733 | * | |
1734 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1735 | * be sure that none of the pages we have control over will have any aliases | |
1736 | * from the vmap layer. | |
1737 | */ | |
1738 | void vm_unmap_aliases(void) | |
1739 | { | |
1740 | unsigned long start = ULONG_MAX, end = 0; | |
1741 | int flush = 0; | |
1742 | ||
1743 | _vm_unmap_aliases(start, end, flush); | |
1744 | } | |
db64fe02 NP |
1745 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
1746 | ||
1747 | /** | |
1748 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1749 | * @mem: the pointer returned by vm_map_ram | |
1750 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1751 | */ | |
1752 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1753 | { | |
65ee03c4 | 1754 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1755 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1756 | struct vmap_area *va; |
db64fe02 | 1757 | |
5803ed29 | 1758 | might_sleep(); |
db64fe02 NP |
1759 | BUG_ON(!addr); |
1760 | BUG_ON(addr < VMALLOC_START); | |
1761 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1762 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 1763 | |
9c3acf60 | 1764 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 1765 | debug_check_no_locks_freed(mem, size); |
db64fe02 | 1766 | vb_free(mem, size); |
9c3acf60 CH |
1767 | return; |
1768 | } | |
1769 | ||
1770 | va = find_vmap_area(addr); | |
1771 | BUG_ON(!va); | |
05e3ff95 CP |
1772 | debug_check_no_locks_freed((void *)va->va_start, |
1773 | (va->va_end - va->va_start)); | |
9c3acf60 | 1774 | free_unmap_vmap_area(va); |
db64fe02 NP |
1775 | } |
1776 | EXPORT_SYMBOL(vm_unmap_ram); | |
1777 | ||
1778 | /** | |
1779 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1780 | * @pages: an array of pointers to the pages to be mapped | |
1781 | * @count: number of pages | |
1782 | * @node: prefer to allocate data structures on this node | |
1783 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1784 | * |
36437638 GK |
1785 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1786 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1787 | * objects with vm_map_ram(), it could consume lots of address space through | |
1788 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1789 | * the end. Please use this function for short-lived objects. | |
1790 | * | |
e99c97ad | 1791 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1792 | */ |
1793 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1794 | { | |
65ee03c4 | 1795 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1796 | unsigned long addr; |
1797 | void *mem; | |
1798 | ||
1799 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1800 | mem = vb_alloc(size, GFP_KERNEL); | |
1801 | if (IS_ERR(mem)) | |
1802 | return NULL; | |
1803 | addr = (unsigned long)mem; | |
1804 | } else { | |
1805 | struct vmap_area *va; | |
1806 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1807 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1808 | if (IS_ERR(va)) | |
1809 | return NULL; | |
1810 | ||
1811 | addr = va->va_start; | |
1812 | mem = (void *)addr; | |
1813 | } | |
1814 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1815 | vm_unmap_ram(mem, count); | |
1816 | return NULL; | |
1817 | } | |
1818 | return mem; | |
1819 | } | |
1820 | EXPORT_SYMBOL(vm_map_ram); | |
1821 | ||
4341fa45 | 1822 | static struct vm_struct *vmlist __initdata; |
92eac168 | 1823 | |
be9b7335 NP |
1824 | /** |
1825 | * vm_area_add_early - add vmap area early during boot | |
1826 | * @vm: vm_struct to add | |
1827 | * | |
1828 | * This function is used to add fixed kernel vm area to vmlist before | |
1829 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1830 | * should contain proper values and the other fields should be zero. | |
1831 | * | |
1832 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1833 | */ | |
1834 | void __init vm_area_add_early(struct vm_struct *vm) | |
1835 | { | |
1836 | struct vm_struct *tmp, **p; | |
1837 | ||
1838 | BUG_ON(vmap_initialized); | |
1839 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1840 | if (tmp->addr >= vm->addr) { | |
1841 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1842 | break; | |
1843 | } else | |
1844 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1845 | } | |
1846 | vm->next = *p; | |
1847 | *p = vm; | |
1848 | } | |
1849 | ||
f0aa6617 TH |
1850 | /** |
1851 | * vm_area_register_early - register vmap area early during boot | |
1852 | * @vm: vm_struct to register | |
c0c0a293 | 1853 | * @align: requested alignment |
f0aa6617 TH |
1854 | * |
1855 | * This function is used to register kernel vm area before | |
1856 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1857 | * proper values on entry and other fields should be zero. On return, | |
1858 | * vm->addr contains the allocated address. | |
1859 | * | |
1860 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1861 | */ | |
c0c0a293 | 1862 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1863 | { |
1864 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1865 | unsigned long addr; |
1866 | ||
1867 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1868 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1869 | |
c0c0a293 | 1870 | vm->addr = (void *)addr; |
f0aa6617 | 1871 | |
be9b7335 | 1872 | vm_area_add_early(vm); |
f0aa6617 TH |
1873 | } |
1874 | ||
68ad4a33 URS |
1875 | static void vmap_init_free_space(void) |
1876 | { | |
1877 | unsigned long vmap_start = 1; | |
1878 | const unsigned long vmap_end = ULONG_MAX; | |
1879 | struct vmap_area *busy, *free; | |
1880 | ||
1881 | /* | |
1882 | * B F B B B F | |
1883 | * -|-----|.....|-----|-----|-----|.....|- | |
1884 | * | The KVA space | | |
1885 | * |<--------------------------------->| | |
1886 | */ | |
1887 | list_for_each_entry(busy, &vmap_area_list, list) { | |
1888 | if (busy->va_start - vmap_start > 0) { | |
1889 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1890 | if (!WARN_ON_ONCE(!free)) { | |
1891 | free->va_start = vmap_start; | |
1892 | free->va_end = busy->va_start; | |
1893 | ||
1894 | insert_vmap_area_augment(free, NULL, | |
1895 | &free_vmap_area_root, | |
1896 | &free_vmap_area_list); | |
1897 | } | |
1898 | } | |
1899 | ||
1900 | vmap_start = busy->va_end; | |
1901 | } | |
1902 | ||
1903 | if (vmap_end - vmap_start > 0) { | |
1904 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1905 | if (!WARN_ON_ONCE(!free)) { | |
1906 | free->va_start = vmap_start; | |
1907 | free->va_end = vmap_end; | |
1908 | ||
1909 | insert_vmap_area_augment(free, NULL, | |
1910 | &free_vmap_area_root, | |
1911 | &free_vmap_area_list); | |
1912 | } | |
1913 | } | |
1914 | } | |
1915 | ||
db64fe02 NP |
1916 | void __init vmalloc_init(void) |
1917 | { | |
822c18f2 IK |
1918 | struct vmap_area *va; |
1919 | struct vm_struct *tmp; | |
db64fe02 NP |
1920 | int i; |
1921 | ||
68ad4a33 URS |
1922 | /* |
1923 | * Create the cache for vmap_area objects. | |
1924 | */ | |
1925 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
1926 | ||
db64fe02 NP |
1927 | for_each_possible_cpu(i) { |
1928 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1929 | struct vfree_deferred *p; |
db64fe02 NP |
1930 | |
1931 | vbq = &per_cpu(vmap_block_queue, i); | |
1932 | spin_lock_init(&vbq->lock); | |
1933 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1934 | p = &per_cpu(vfree_deferred, i); |
1935 | init_llist_head(&p->list); | |
1936 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1937 | } |
9b463334 | 1938 | |
822c18f2 IK |
1939 | /* Import existing vmlist entries. */ |
1940 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
1941 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
1942 | if (WARN_ON_ONCE(!va)) | |
1943 | continue; | |
1944 | ||
822c18f2 IK |
1945 | va->va_start = (unsigned long)tmp->addr; |
1946 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1947 | va->vm = tmp; |
68ad4a33 | 1948 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 1949 | } |
ca23e405 | 1950 | |
68ad4a33 URS |
1951 | /* |
1952 | * Now we can initialize a free vmap space. | |
1953 | */ | |
1954 | vmap_init_free_space(); | |
9b463334 | 1955 | vmap_initialized = true; |
db64fe02 NP |
1956 | } |
1957 | ||
8fc48985 TH |
1958 | /** |
1959 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1960 | * @addr: start of the VM area to map | |
1961 | * @size: size of the VM area to map | |
1962 | * @prot: page protection flags to use | |
1963 | * @pages: pages to map | |
1964 | * | |
1965 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1966 | * specify should have been allocated using get_vm_area() and its | |
1967 | * friends. | |
1968 | * | |
1969 | * NOTE: | |
1970 | * This function does NOT do any cache flushing. The caller is | |
1971 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1972 | * before calling this function. | |
1973 | * | |
1974 | * RETURNS: | |
1975 | * The number of pages mapped on success, -errno on failure. | |
1976 | */ | |
1977 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1978 | pgprot_t prot, struct page **pages) | |
1979 | { | |
1980 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1981 | } | |
1982 | ||
1983 | /** | |
1984 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1985 | * @addr: start of the VM area to unmap | |
1986 | * @size: size of the VM area to unmap | |
1987 | * | |
1988 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1989 | * specify should have been allocated using get_vm_area() and its | |
1990 | * friends. | |
1991 | * | |
1992 | * NOTE: | |
1993 | * This function does NOT do any cache flushing. The caller is | |
1994 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1995 | * before calling this function and flush_tlb_kernel_range() after. | |
1996 | */ | |
1997 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1998 | { | |
1999 | vunmap_page_range(addr, addr + size); | |
2000 | } | |
81e88fdc | 2001 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
2002 | |
2003 | /** | |
2004 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
2005 | * @addr: start of the VM area to unmap | |
2006 | * @size: size of the VM area to unmap | |
2007 | * | |
2008 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
2009 | * the unmapping and tlb after. | |
2010 | */ | |
db64fe02 NP |
2011 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
2012 | { | |
2013 | unsigned long end = addr + size; | |
f6fcba70 TH |
2014 | |
2015 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
2016 | vunmap_page_range(addr, end); |
2017 | flush_tlb_kernel_range(addr, end); | |
2018 | } | |
93ef6d6c | 2019 | EXPORT_SYMBOL_GPL(unmap_kernel_range); |
db64fe02 | 2020 | |
f6f8ed47 | 2021 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
2022 | { |
2023 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 2024 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
2025 | int err; |
2026 | ||
f6f8ed47 | 2027 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 2028 | |
f6f8ed47 | 2029 | return err > 0 ? 0 : err; |
db64fe02 NP |
2030 | } |
2031 | EXPORT_SYMBOL_GPL(map_vm_area); | |
2032 | ||
f5252e00 | 2033 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 2034 | unsigned long flags, const void *caller) |
cf88c790 | 2035 | { |
c69480ad | 2036 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
2037 | vm->flags = flags; |
2038 | vm->addr = (void *)va->va_start; | |
2039 | vm->size = va->va_end - va->va_start; | |
2040 | vm->caller = caller; | |
db1aecaf | 2041 | va->vm = vm; |
c69480ad | 2042 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2043 | } |
cf88c790 | 2044 | |
20fc02b4 | 2045 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2046 | { |
d4033afd | 2047 | /* |
20fc02b4 | 2048 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2049 | * we should make sure that vm has proper values. |
2050 | * Pair with smp_rmb() in show_numa_info(). | |
2051 | */ | |
2052 | smp_wmb(); | |
20fc02b4 | 2053 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2054 | } |
2055 | ||
db64fe02 | 2056 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2057 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2058 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2059 | { |
0006526d | 2060 | struct vmap_area *va; |
db64fe02 | 2061 | struct vm_struct *area; |
1da177e4 | 2062 | |
52fd24ca | 2063 | BUG_ON(in_interrupt()); |
1da177e4 | 2064 | size = PAGE_ALIGN(size); |
31be8309 OH |
2065 | if (unlikely(!size)) |
2066 | return NULL; | |
1da177e4 | 2067 | |
252e5c6e | 2068 | if (flags & VM_IOREMAP) |
2069 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2070 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2071 | ||
cf88c790 | 2072 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2073 | if (unlikely(!area)) |
2074 | return NULL; | |
2075 | ||
71394fe5 AR |
2076 | if (!(flags & VM_NO_GUARD)) |
2077 | size += PAGE_SIZE; | |
1da177e4 | 2078 | |
db64fe02 NP |
2079 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2080 | if (IS_ERR(va)) { | |
2081 | kfree(area); | |
2082 | return NULL; | |
1da177e4 | 2083 | } |
1da177e4 | 2084 | |
d82b1d85 | 2085 | setup_vmalloc_vm(area, va, flags, caller); |
f5252e00 | 2086 | |
1da177e4 | 2087 | return area; |
1da177e4 LT |
2088 | } |
2089 | ||
930fc45a CL |
2090 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
2091 | unsigned long start, unsigned long end) | |
2092 | { | |
00ef2d2f DR |
2093 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2094 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 2095 | } |
5992b6da | 2096 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 2097 | |
c2968612 BH |
2098 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2099 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2100 | const void *caller) |
c2968612 | 2101 | { |
00ef2d2f DR |
2102 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2103 | GFP_KERNEL, caller); | |
c2968612 BH |
2104 | } |
2105 | ||
1da177e4 | 2106 | /** |
92eac168 MR |
2107 | * get_vm_area - reserve a contiguous kernel virtual area |
2108 | * @size: size of the area | |
2109 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2110 | * |
92eac168 MR |
2111 | * Search an area of @size in the kernel virtual mapping area, |
2112 | * and reserved it for out purposes. Returns the area descriptor | |
2113 | * on success or %NULL on failure. | |
a862f68a MR |
2114 | * |
2115 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2116 | */ |
2117 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2118 | { | |
2dca6999 | 2119 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2120 | NUMA_NO_NODE, GFP_KERNEL, |
2121 | __builtin_return_address(0)); | |
23016969 CL |
2122 | } |
2123 | ||
2124 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2125 | const void *caller) |
23016969 | 2126 | { |
2dca6999 | 2127 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2128 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2129 | } |
2130 | ||
e9da6e99 | 2131 | /** |
92eac168 MR |
2132 | * find_vm_area - find a continuous kernel virtual area |
2133 | * @addr: base address | |
e9da6e99 | 2134 | * |
92eac168 MR |
2135 | * Search for the kernel VM area starting at @addr, and return it. |
2136 | * It is up to the caller to do all required locking to keep the returned | |
2137 | * pointer valid. | |
a862f68a MR |
2138 | * |
2139 | * Return: pointer to the found area or %NULL on faulure | |
e9da6e99 MS |
2140 | */ |
2141 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2142 | { |
db64fe02 | 2143 | struct vmap_area *va; |
83342314 | 2144 | |
db64fe02 | 2145 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2146 | if (!va) |
2147 | return NULL; | |
1da177e4 | 2148 | |
688fcbfc | 2149 | return va->vm; |
1da177e4 LT |
2150 | } |
2151 | ||
7856dfeb | 2152 | /** |
92eac168 MR |
2153 | * remove_vm_area - find and remove a continuous kernel virtual area |
2154 | * @addr: base address | |
7856dfeb | 2155 | * |
92eac168 MR |
2156 | * Search for the kernel VM area starting at @addr, and remove it. |
2157 | * This function returns the found VM area, but using it is NOT safe | |
2158 | * on SMP machines, except for its size or flags. | |
a862f68a MR |
2159 | * |
2160 | * Return: pointer to the found area or %NULL on faulure | |
7856dfeb | 2161 | */ |
b3bdda02 | 2162 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2163 | { |
db64fe02 NP |
2164 | struct vmap_area *va; |
2165 | ||
5803ed29 CH |
2166 | might_sleep(); |
2167 | ||
dd3b8353 URS |
2168 | spin_lock(&vmap_area_lock); |
2169 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2170 | if (va && va->vm) { |
db1aecaf | 2171 | struct vm_struct *vm = va->vm; |
f5252e00 | 2172 | |
c69480ad | 2173 | va->vm = NULL; |
c69480ad JK |
2174 | spin_unlock(&vmap_area_lock); |
2175 | ||
a5af5aa8 | 2176 | kasan_free_shadow(vm); |
dd32c279 | 2177 | free_unmap_vmap_area(va); |
dd32c279 | 2178 | |
db64fe02 NP |
2179 | return vm; |
2180 | } | |
dd3b8353 URS |
2181 | |
2182 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2183 | return NULL; |
7856dfeb AK |
2184 | } |
2185 | ||
868b104d RE |
2186 | static inline void set_area_direct_map(const struct vm_struct *area, |
2187 | int (*set_direct_map)(struct page *page)) | |
2188 | { | |
2189 | int i; | |
2190 | ||
2191 | for (i = 0; i < area->nr_pages; i++) | |
2192 | if (page_address(area->pages[i])) | |
2193 | set_direct_map(area->pages[i]); | |
2194 | } | |
2195 | ||
2196 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2197 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2198 | { | |
868b104d RE |
2199 | unsigned long start = ULONG_MAX, end = 0; |
2200 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
31e67340 | 2201 | int flush_dmap = 0; |
868b104d RE |
2202 | int i; |
2203 | ||
868b104d RE |
2204 | remove_vm_area(area->addr); |
2205 | ||
2206 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2207 | if (!flush_reset) | |
2208 | return; | |
2209 | ||
2210 | /* | |
2211 | * If not deallocating pages, just do the flush of the VM area and | |
2212 | * return. | |
2213 | */ | |
2214 | if (!deallocate_pages) { | |
2215 | vm_unmap_aliases(); | |
2216 | return; | |
2217 | } | |
2218 | ||
2219 | /* | |
2220 | * If execution gets here, flush the vm mapping and reset the direct | |
2221 | * map. Find the start and end range of the direct mappings to make sure | |
2222 | * the vm_unmap_aliases() flush includes the direct map. | |
2223 | */ | |
2224 | for (i = 0; i < area->nr_pages; i++) { | |
8e41f872 RE |
2225 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2226 | if (addr) { | |
868b104d | 2227 | start = min(addr, start); |
8e41f872 | 2228 | end = max(addr + PAGE_SIZE, end); |
31e67340 | 2229 | flush_dmap = 1; |
868b104d RE |
2230 | } |
2231 | } | |
2232 | ||
2233 | /* | |
2234 | * Set direct map to something invalid so that it won't be cached if | |
2235 | * there are any accesses after the TLB flush, then flush the TLB and | |
2236 | * reset the direct map permissions to the default. | |
2237 | */ | |
2238 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2239 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2240 | set_area_direct_map(area, set_direct_map_default_noflush); |
2241 | } | |
2242 | ||
b3bdda02 | 2243 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2244 | { |
2245 | struct vm_struct *area; | |
2246 | ||
2247 | if (!addr) | |
2248 | return; | |
2249 | ||
e69e9d4a | 2250 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2251 | addr)) |
1da177e4 | 2252 | return; |
1da177e4 | 2253 | |
6ade2032 | 2254 | area = find_vm_area(addr); |
1da177e4 | 2255 | if (unlikely(!area)) { |
4c8573e2 | 2256 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2257 | addr); |
1da177e4 LT |
2258 | return; |
2259 | } | |
2260 | ||
05e3ff95 CP |
2261 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2262 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2263 | |
868b104d RE |
2264 | vm_remove_mappings(area, deallocate_pages); |
2265 | ||
1da177e4 LT |
2266 | if (deallocate_pages) { |
2267 | int i; | |
2268 | ||
2269 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2270 | struct page *page = area->pages[i]; |
2271 | ||
2272 | BUG_ON(!page); | |
4949148a | 2273 | __free_pages(page, 0); |
1da177e4 | 2274 | } |
97105f0a | 2275 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2276 | |
244d63ee | 2277 | kvfree(area->pages); |
1da177e4 LT |
2278 | } |
2279 | ||
2280 | kfree(area); | |
2281 | return; | |
2282 | } | |
bf22e37a AR |
2283 | |
2284 | static inline void __vfree_deferred(const void *addr) | |
2285 | { | |
2286 | /* | |
2287 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2288 | * context. Preemption is absolutely fine here, because the llist_add() | |
2289 | * implementation is lockless, so it works even if we are adding to | |
2290 | * nother cpu's list. schedule_work() should be fine with this too. | |
2291 | */ | |
2292 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2293 | ||
2294 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2295 | schedule_work(&p->wq); | |
2296 | } | |
2297 | ||
2298 | /** | |
92eac168 MR |
2299 | * vfree_atomic - release memory allocated by vmalloc() |
2300 | * @addr: memory base address | |
bf22e37a | 2301 | * |
92eac168 MR |
2302 | * This one is just like vfree() but can be called in any atomic context |
2303 | * except NMIs. | |
bf22e37a AR |
2304 | */ |
2305 | void vfree_atomic(const void *addr) | |
2306 | { | |
2307 | BUG_ON(in_nmi()); | |
2308 | ||
2309 | kmemleak_free(addr); | |
2310 | ||
2311 | if (!addr) | |
2312 | return; | |
2313 | __vfree_deferred(addr); | |
2314 | } | |
2315 | ||
c67dc624 RP |
2316 | static void __vfree(const void *addr) |
2317 | { | |
2318 | if (unlikely(in_interrupt())) | |
2319 | __vfree_deferred(addr); | |
2320 | else | |
2321 | __vunmap(addr, 1); | |
2322 | } | |
2323 | ||
1da177e4 | 2324 | /** |
92eac168 MR |
2325 | * vfree - release memory allocated by vmalloc() |
2326 | * @addr: memory base address | |
1da177e4 | 2327 | * |
92eac168 MR |
2328 | * Free the virtually continuous memory area starting at @addr, as |
2329 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is | |
2330 | * NULL, no operation is performed. | |
1da177e4 | 2331 | * |
92eac168 MR |
2332 | * Must not be called in NMI context (strictly speaking, only if we don't |
2333 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2334 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 2335 | * |
92eac168 | 2336 | * May sleep if called *not* from interrupt context. |
3ca4ea3a | 2337 | * |
92eac168 | 2338 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 2339 | */ |
b3bdda02 | 2340 | void vfree(const void *addr) |
1da177e4 | 2341 | { |
32fcfd40 | 2342 | BUG_ON(in_nmi()); |
89219d37 CM |
2343 | |
2344 | kmemleak_free(addr); | |
2345 | ||
a8dda165 AR |
2346 | might_sleep_if(!in_interrupt()); |
2347 | ||
32fcfd40 AV |
2348 | if (!addr) |
2349 | return; | |
c67dc624 RP |
2350 | |
2351 | __vfree(addr); | |
1da177e4 | 2352 | } |
1da177e4 LT |
2353 | EXPORT_SYMBOL(vfree); |
2354 | ||
2355 | /** | |
92eac168 MR |
2356 | * vunmap - release virtual mapping obtained by vmap() |
2357 | * @addr: memory base address | |
1da177e4 | 2358 | * |
92eac168 MR |
2359 | * Free the virtually contiguous memory area starting at @addr, |
2360 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2361 | * |
92eac168 | 2362 | * Must not be called in interrupt context. |
1da177e4 | 2363 | */ |
b3bdda02 | 2364 | void vunmap(const void *addr) |
1da177e4 LT |
2365 | { |
2366 | BUG_ON(in_interrupt()); | |
34754b69 | 2367 | might_sleep(); |
32fcfd40 AV |
2368 | if (addr) |
2369 | __vunmap(addr, 0); | |
1da177e4 | 2370 | } |
1da177e4 LT |
2371 | EXPORT_SYMBOL(vunmap); |
2372 | ||
2373 | /** | |
92eac168 MR |
2374 | * vmap - map an array of pages into virtually contiguous space |
2375 | * @pages: array of page pointers | |
2376 | * @count: number of pages to map | |
2377 | * @flags: vm_area->flags | |
2378 | * @prot: page protection for the mapping | |
2379 | * | |
2380 | * Maps @count pages from @pages into contiguous kernel virtual | |
2381 | * space. | |
a862f68a MR |
2382 | * |
2383 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2384 | */ |
2385 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2386 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2387 | { |
2388 | struct vm_struct *area; | |
65ee03c4 | 2389 | unsigned long size; /* In bytes */ |
1da177e4 | 2390 | |
34754b69 PZ |
2391 | might_sleep(); |
2392 | ||
ca79b0c2 | 2393 | if (count > totalram_pages()) |
1da177e4 LT |
2394 | return NULL; |
2395 | ||
65ee03c4 GJM |
2396 | size = (unsigned long)count << PAGE_SHIFT; |
2397 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2398 | if (!area) |
2399 | return NULL; | |
23016969 | 2400 | |
f6f8ed47 | 2401 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
2402 | vunmap(area->addr); |
2403 | return NULL; | |
2404 | } | |
2405 | ||
2406 | return area->addr; | |
2407 | } | |
1da177e4 LT |
2408 | EXPORT_SYMBOL(vmap); |
2409 | ||
8594a21c MH |
2410 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
2411 | gfp_t gfp_mask, pgprot_t prot, | |
2412 | int node, const void *caller); | |
e31d9eb5 | 2413 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 2414 | pgprot_t prot, int node) |
1da177e4 LT |
2415 | { |
2416 | struct page **pages; | |
2417 | unsigned int nr_pages, array_size, i; | |
930f036b | 2418 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
704b862f LA |
2419 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; |
2420 | const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ? | |
2421 | 0 : | |
2422 | __GFP_HIGHMEM; | |
1da177e4 | 2423 | |
762216ab | 2424 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
2425 | array_size = (nr_pages * sizeof(struct page *)); |
2426 | ||
1da177e4 | 2427 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 2428 | if (array_size > PAGE_SIZE) { |
704b862f | 2429 | pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask, |
3722e13c | 2430 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 2431 | } else { |
976d6dfb | 2432 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2433 | } |
7ea36242 AK |
2434 | |
2435 | if (!pages) { | |
1da177e4 LT |
2436 | remove_vm_area(area->addr); |
2437 | kfree(area); | |
2438 | return NULL; | |
2439 | } | |
1da177e4 | 2440 | |
7ea36242 AK |
2441 | area->pages = pages; |
2442 | area->nr_pages = nr_pages; | |
2443 | ||
1da177e4 | 2444 | for (i = 0; i < area->nr_pages; i++) { |
bf53d6f8 CL |
2445 | struct page *page; |
2446 | ||
4b90951c | 2447 | if (node == NUMA_NO_NODE) |
704b862f | 2448 | page = alloc_page(alloc_mask|highmem_mask); |
930fc45a | 2449 | else |
704b862f | 2450 | page = alloc_pages_node(node, alloc_mask|highmem_mask, 0); |
bf53d6f8 CL |
2451 | |
2452 | if (unlikely(!page)) { | |
1da177e4 LT |
2453 | /* Successfully allocated i pages, free them in __vunmap() */ |
2454 | area->nr_pages = i; | |
97105f0a | 2455 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 LT |
2456 | goto fail; |
2457 | } | |
bf53d6f8 | 2458 | area->pages[i] = page; |
dcf61ff0 | 2459 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 2460 | cond_resched(); |
1da177e4 | 2461 | } |
97105f0a | 2462 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2463 | |
f6f8ed47 | 2464 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
2465 | goto fail; |
2466 | return area->addr; | |
2467 | ||
2468 | fail: | |
a8e99259 | 2469 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2470 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 2471 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 2472 | __vfree(area->addr); |
1da177e4 LT |
2473 | return NULL; |
2474 | } | |
2475 | ||
2476 | /** | |
92eac168 MR |
2477 | * __vmalloc_node_range - allocate virtually contiguous memory |
2478 | * @size: allocation size | |
2479 | * @align: desired alignment | |
2480 | * @start: vm area range start | |
2481 | * @end: vm area range end | |
2482 | * @gfp_mask: flags for the page level allocator | |
2483 | * @prot: protection mask for the allocated pages | |
2484 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2485 | * @node: node to use for allocation or NUMA_NO_NODE | |
2486 | * @caller: caller's return address | |
2487 | * | |
2488 | * Allocate enough pages to cover @size from the page level | |
2489 | * allocator with @gfp_mask flags. Map them into contiguous | |
2490 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2491 | * |
2492 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2493 | */ |
d0a21265 DR |
2494 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2495 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2496 | pgprot_t prot, unsigned long vm_flags, int node, |
2497 | const void *caller) | |
1da177e4 LT |
2498 | { |
2499 | struct vm_struct *area; | |
89219d37 CM |
2500 | void *addr; |
2501 | unsigned long real_size = size; | |
1da177e4 LT |
2502 | |
2503 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 2504 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 2505 | goto fail; |
1da177e4 | 2506 | |
cb9e3c29 AR |
2507 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | |
2508 | vm_flags, start, end, node, gfp_mask, caller); | |
1da177e4 | 2509 | if (!area) |
de7d2b56 | 2510 | goto fail; |
1da177e4 | 2511 | |
3722e13c | 2512 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 2513 | if (!addr) |
b82225f3 | 2514 | return NULL; |
89219d37 | 2515 | |
f5252e00 | 2516 | /* |
20fc02b4 ZY |
2517 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2518 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2519 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2520 | */ |
20fc02b4 | 2521 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2522 | |
94f4a161 | 2523 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2524 | |
2525 | return addr; | |
de7d2b56 JP |
2526 | |
2527 | fail: | |
a8e99259 | 2528 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2529 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 2530 | return NULL; |
1da177e4 LT |
2531 | } |
2532 | ||
153178ed URS |
2533 | /* |
2534 | * This is only for performance analysis of vmalloc and stress purpose. | |
2535 | * It is required by vmalloc test module, therefore do not use it other | |
2536 | * than that. | |
2537 | */ | |
2538 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2539 | EXPORT_SYMBOL_GPL(__vmalloc_node_range); | |
2540 | #endif | |
2541 | ||
d0a21265 | 2542 | /** |
92eac168 MR |
2543 | * __vmalloc_node - allocate virtually contiguous memory |
2544 | * @size: allocation size | |
2545 | * @align: desired alignment | |
2546 | * @gfp_mask: flags for the page level allocator | |
2547 | * @prot: protection mask for the allocated pages | |
2548 | * @node: node to use for allocation or NUMA_NO_NODE | |
2549 | * @caller: caller's return address | |
a7c3e901 | 2550 | * |
92eac168 MR |
2551 | * Allocate enough pages to cover @size from the page level |
2552 | * allocator with @gfp_mask flags. Map them into contiguous | |
2553 | * kernel virtual space, using a pagetable protection of @prot. | |
a7c3e901 | 2554 | * |
92eac168 MR |
2555 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2556 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2557 | * |
92eac168 MR |
2558 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2559 | * with mm people. | |
a862f68a MR |
2560 | * |
2561 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2562 | */ |
8594a21c | 2563 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
d0a21265 | 2564 | gfp_t gfp_mask, pgprot_t prot, |
5e6cafc8 | 2565 | int node, const void *caller) |
d0a21265 DR |
2566 | { |
2567 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 2568 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
2569 | } |
2570 | ||
930fc45a CL |
2571 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
2572 | { | |
00ef2d2f | 2573 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 2574 | __builtin_return_address(0)); |
930fc45a | 2575 | } |
1da177e4 LT |
2576 | EXPORT_SYMBOL(__vmalloc); |
2577 | ||
8594a21c MH |
2578 | static inline void *__vmalloc_node_flags(unsigned long size, |
2579 | int node, gfp_t flags) | |
2580 | { | |
2581 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
2582 | node, __builtin_return_address(0)); | |
2583 | } | |
2584 | ||
2585 | ||
2586 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags, | |
2587 | void *caller) | |
2588 | { | |
2589 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller); | |
2590 | } | |
2591 | ||
1da177e4 | 2592 | /** |
92eac168 MR |
2593 | * vmalloc - allocate virtually contiguous memory |
2594 | * @size: allocation size | |
2595 | * | |
2596 | * Allocate enough pages to cover @size from the page level | |
2597 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2598 | * |
92eac168 MR |
2599 | * For tight control over page level allocator and protection flags |
2600 | * use __vmalloc() instead. | |
a862f68a MR |
2601 | * |
2602 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2603 | */ |
2604 | void *vmalloc(unsigned long size) | |
2605 | { | |
00ef2d2f | 2606 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2607 | GFP_KERNEL); |
1da177e4 | 2608 | } |
1da177e4 LT |
2609 | EXPORT_SYMBOL(vmalloc); |
2610 | ||
e1ca7788 | 2611 | /** |
92eac168 MR |
2612 | * vzalloc - allocate virtually contiguous memory with zero fill |
2613 | * @size: allocation size | |
2614 | * | |
2615 | * Allocate enough pages to cover @size from the page level | |
2616 | * allocator and map them into contiguous kernel virtual space. | |
2617 | * The memory allocated is set to zero. | |
2618 | * | |
2619 | * For tight control over page level allocator and protection flags | |
2620 | * use __vmalloc() instead. | |
a862f68a MR |
2621 | * |
2622 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2623 | */ |
2624 | void *vzalloc(unsigned long size) | |
2625 | { | |
00ef2d2f | 2626 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2627 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2628 | } |
2629 | EXPORT_SYMBOL(vzalloc); | |
2630 | ||
83342314 | 2631 | /** |
ead04089 REB |
2632 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
2633 | * @size: allocation size | |
83342314 | 2634 | * |
ead04089 REB |
2635 | * The resulting memory area is zeroed so it can be mapped to userspace |
2636 | * without leaking data. | |
a862f68a MR |
2637 | * |
2638 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2639 | */ |
2640 | void *vmalloc_user(unsigned long size) | |
2641 | { | |
bc84c535 RP |
2642 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2643 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
2644 | VM_USERMAP, NUMA_NO_NODE, | |
2645 | __builtin_return_address(0)); | |
83342314 NP |
2646 | } |
2647 | EXPORT_SYMBOL(vmalloc_user); | |
2648 | ||
930fc45a | 2649 | /** |
92eac168 MR |
2650 | * vmalloc_node - allocate memory on a specific node |
2651 | * @size: allocation size | |
2652 | * @node: numa node | |
930fc45a | 2653 | * |
92eac168 MR |
2654 | * Allocate enough pages to cover @size from the page level |
2655 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 2656 | * |
92eac168 MR |
2657 | * For tight control over page level allocator and protection flags |
2658 | * use __vmalloc() instead. | |
a862f68a MR |
2659 | * |
2660 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2661 | */ |
2662 | void *vmalloc_node(unsigned long size, int node) | |
2663 | { | |
19809c2d | 2664 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL, |
23016969 | 2665 | node, __builtin_return_address(0)); |
930fc45a CL |
2666 | } |
2667 | EXPORT_SYMBOL(vmalloc_node); | |
2668 | ||
e1ca7788 DY |
2669 | /** |
2670 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2671 | * @size: allocation size | |
2672 | * @node: numa node | |
2673 | * | |
2674 | * Allocate enough pages to cover @size from the page level | |
2675 | * allocator and map them into contiguous kernel virtual space. | |
2676 | * The memory allocated is set to zero. | |
2677 | * | |
2678 | * For tight control over page level allocator and protection flags | |
2679 | * use __vmalloc_node() instead. | |
a862f68a MR |
2680 | * |
2681 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2682 | */ |
2683 | void *vzalloc_node(unsigned long size, int node) | |
2684 | { | |
2685 | return __vmalloc_node_flags(size, node, | |
19809c2d | 2686 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2687 | } |
2688 | EXPORT_SYMBOL(vzalloc_node); | |
2689 | ||
fc970227 AN |
2690 | /** |
2691 | * vmalloc_user_node_flags - allocate memory for userspace on a specific node | |
2692 | * @size: allocation size | |
2693 | * @node: numa node | |
2694 | * @flags: flags for the page level allocator | |
2695 | * | |
2696 | * The resulting memory area is zeroed so it can be mapped to userspace | |
2697 | * without leaking data. | |
2698 | * | |
2699 | * Return: pointer to the allocated memory or %NULL on error | |
2700 | */ | |
2701 | void *vmalloc_user_node_flags(unsigned long size, int node, gfp_t flags) | |
2702 | { | |
2703 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, | |
2704 | flags | __GFP_ZERO, PAGE_KERNEL, | |
2705 | VM_USERMAP, node, | |
2706 | __builtin_return_address(0)); | |
2707 | } | |
2708 | EXPORT_SYMBOL(vmalloc_user_node_flags); | |
2709 | ||
1da177e4 | 2710 | /** |
92eac168 MR |
2711 | * vmalloc_exec - allocate virtually contiguous, executable memory |
2712 | * @size: allocation size | |
1da177e4 | 2713 | * |
92eac168 MR |
2714 | * Kernel-internal function to allocate enough pages to cover @size |
2715 | * the page level allocator and map them into contiguous and | |
2716 | * executable kernel virtual space. | |
1da177e4 | 2717 | * |
92eac168 MR |
2718 | * For tight control over page level allocator and protection flags |
2719 | * use __vmalloc() instead. | |
a862f68a MR |
2720 | * |
2721 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 | 2722 | */ |
1da177e4 LT |
2723 | void *vmalloc_exec(unsigned long size) |
2724 | { | |
868b104d RE |
2725 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, |
2726 | GFP_KERNEL, PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS, | |
2727 | NUMA_NO_NODE, __builtin_return_address(0)); | |
1da177e4 LT |
2728 | } |
2729 | ||
0d08e0d3 | 2730 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2731 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2732 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2733 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2734 | #else |
698d0831 MH |
2735 | /* |
2736 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2737 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2738 | */ | |
2739 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2740 | #endif |
2741 | ||
1da177e4 | 2742 | /** |
92eac168 MR |
2743 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2744 | * @size: allocation size | |
1da177e4 | 2745 | * |
92eac168 MR |
2746 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2747 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2748 | * |
2749 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2750 | */ |
2751 | void *vmalloc_32(unsigned long size) | |
2752 | { | |
2dca6999 | 2753 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 2754 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 2755 | } |
1da177e4 LT |
2756 | EXPORT_SYMBOL(vmalloc_32); |
2757 | ||
83342314 | 2758 | /** |
ead04089 | 2759 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2760 | * @size: allocation size |
ead04089 REB |
2761 | * |
2762 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2763 | * mapped to userspace without leaking data. | |
a862f68a MR |
2764 | * |
2765 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2766 | */ |
2767 | void *vmalloc_32_user(unsigned long size) | |
2768 | { | |
bc84c535 RP |
2769 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2770 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2771 | VM_USERMAP, NUMA_NO_NODE, | |
2772 | __builtin_return_address(0)); | |
83342314 NP |
2773 | } |
2774 | EXPORT_SYMBOL(vmalloc_32_user); | |
2775 | ||
d0107eb0 KH |
2776 | /* |
2777 | * small helper routine , copy contents to buf from addr. | |
2778 | * If the page is not present, fill zero. | |
2779 | */ | |
2780 | ||
2781 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2782 | { | |
2783 | struct page *p; | |
2784 | int copied = 0; | |
2785 | ||
2786 | while (count) { | |
2787 | unsigned long offset, length; | |
2788 | ||
891c49ab | 2789 | offset = offset_in_page(addr); |
d0107eb0 KH |
2790 | length = PAGE_SIZE - offset; |
2791 | if (length > count) | |
2792 | length = count; | |
2793 | p = vmalloc_to_page(addr); | |
2794 | /* | |
2795 | * To do safe access to this _mapped_ area, we need | |
2796 | * lock. But adding lock here means that we need to add | |
2797 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2798 | * interface, rarely used. Instead of that, we'll use | |
2799 | * kmap() and get small overhead in this access function. | |
2800 | */ | |
2801 | if (p) { | |
2802 | /* | |
2803 | * we can expect USER0 is not used (see vread/vwrite's | |
2804 | * function description) | |
2805 | */ | |
9b04c5fe | 2806 | void *map = kmap_atomic(p); |
d0107eb0 | 2807 | memcpy(buf, map + offset, length); |
9b04c5fe | 2808 | kunmap_atomic(map); |
d0107eb0 KH |
2809 | } else |
2810 | memset(buf, 0, length); | |
2811 | ||
2812 | addr += length; | |
2813 | buf += length; | |
2814 | copied += length; | |
2815 | count -= length; | |
2816 | } | |
2817 | return copied; | |
2818 | } | |
2819 | ||
2820 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2821 | { | |
2822 | struct page *p; | |
2823 | int copied = 0; | |
2824 | ||
2825 | while (count) { | |
2826 | unsigned long offset, length; | |
2827 | ||
891c49ab | 2828 | offset = offset_in_page(addr); |
d0107eb0 KH |
2829 | length = PAGE_SIZE - offset; |
2830 | if (length > count) | |
2831 | length = count; | |
2832 | p = vmalloc_to_page(addr); | |
2833 | /* | |
2834 | * To do safe access to this _mapped_ area, we need | |
2835 | * lock. But adding lock here means that we need to add | |
2836 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2837 | * interface, rarely used. Instead of that, we'll use | |
2838 | * kmap() and get small overhead in this access function. | |
2839 | */ | |
2840 | if (p) { | |
2841 | /* | |
2842 | * we can expect USER0 is not used (see vread/vwrite's | |
2843 | * function description) | |
2844 | */ | |
9b04c5fe | 2845 | void *map = kmap_atomic(p); |
d0107eb0 | 2846 | memcpy(map + offset, buf, length); |
9b04c5fe | 2847 | kunmap_atomic(map); |
d0107eb0 KH |
2848 | } |
2849 | addr += length; | |
2850 | buf += length; | |
2851 | copied += length; | |
2852 | count -= length; | |
2853 | } | |
2854 | return copied; | |
2855 | } | |
2856 | ||
2857 | /** | |
92eac168 MR |
2858 | * vread() - read vmalloc area in a safe way. |
2859 | * @buf: buffer for reading data | |
2860 | * @addr: vm address. | |
2861 | * @count: number of bytes to be read. | |
2862 | * | |
92eac168 MR |
2863 | * This function checks that addr is a valid vmalloc'ed area, and |
2864 | * copy data from that area to a given buffer. If the given memory range | |
2865 | * of [addr...addr+count) includes some valid address, data is copied to | |
2866 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2867 | * IOREMAP area is treated as memory hole and no copy is done. | |
2868 | * | |
2869 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2870 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2871 | * | |
2872 | * Note: In usual ops, vread() is never necessary because the caller | |
2873 | * should know vmalloc() area is valid and can use memcpy(). | |
2874 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2875 | * any information, as /dev/kmem. |
a862f68a MR |
2876 | * |
2877 | * Return: number of bytes for which addr and buf should be increased | |
2878 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
2879 | * include any intersection with valid vmalloc area | |
d0107eb0 | 2880 | */ |
1da177e4 LT |
2881 | long vread(char *buf, char *addr, unsigned long count) |
2882 | { | |
e81ce85f JK |
2883 | struct vmap_area *va; |
2884 | struct vm_struct *vm; | |
1da177e4 | 2885 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2886 | unsigned long buflen = count; |
1da177e4 LT |
2887 | unsigned long n; |
2888 | ||
2889 | /* Don't allow overflow */ | |
2890 | if ((unsigned long) addr + count < count) | |
2891 | count = -(unsigned long) addr; | |
2892 | ||
e81ce85f JK |
2893 | spin_lock(&vmap_area_lock); |
2894 | list_for_each_entry(va, &vmap_area_list, list) { | |
2895 | if (!count) | |
2896 | break; | |
2897 | ||
688fcbfc | 2898 | if (!va->vm) |
e81ce85f JK |
2899 | continue; |
2900 | ||
2901 | vm = va->vm; | |
2902 | vaddr = (char *) vm->addr; | |
762216ab | 2903 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2904 | continue; |
2905 | while (addr < vaddr) { | |
2906 | if (count == 0) | |
2907 | goto finished; | |
2908 | *buf = '\0'; | |
2909 | buf++; | |
2910 | addr++; | |
2911 | count--; | |
2912 | } | |
762216ab | 2913 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2914 | if (n > count) |
2915 | n = count; | |
e81ce85f | 2916 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2917 | aligned_vread(buf, addr, n); |
2918 | else /* IOREMAP area is treated as memory hole */ | |
2919 | memset(buf, 0, n); | |
2920 | buf += n; | |
2921 | addr += n; | |
2922 | count -= n; | |
1da177e4 LT |
2923 | } |
2924 | finished: | |
e81ce85f | 2925 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2926 | |
2927 | if (buf == buf_start) | |
2928 | return 0; | |
2929 | /* zero-fill memory holes */ | |
2930 | if (buf != buf_start + buflen) | |
2931 | memset(buf, 0, buflen - (buf - buf_start)); | |
2932 | ||
2933 | return buflen; | |
1da177e4 LT |
2934 | } |
2935 | ||
d0107eb0 | 2936 | /** |
92eac168 MR |
2937 | * vwrite() - write vmalloc area in a safe way. |
2938 | * @buf: buffer for source data | |
2939 | * @addr: vm address. | |
2940 | * @count: number of bytes to be read. | |
2941 | * | |
92eac168 MR |
2942 | * This function checks that addr is a valid vmalloc'ed area, and |
2943 | * copy data from a buffer to the given addr. If specified range of | |
2944 | * [addr...addr+count) includes some valid address, data is copied from | |
2945 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2946 | * IOREMAP area is treated as memory hole and no copy is done. | |
2947 | * | |
2948 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2949 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2950 | * | |
2951 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2952 | * should know vmalloc() area is valid and can use memcpy(). | |
2953 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2954 | * any information, as /dev/kmem. |
a862f68a MR |
2955 | * |
2956 | * Return: number of bytes for which addr and buf should be | |
2957 | * increased (same number as @count) or %0 if [addr...addr+count) | |
2958 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 2959 | */ |
1da177e4 LT |
2960 | long vwrite(char *buf, char *addr, unsigned long count) |
2961 | { | |
e81ce85f JK |
2962 | struct vmap_area *va; |
2963 | struct vm_struct *vm; | |
d0107eb0 KH |
2964 | char *vaddr; |
2965 | unsigned long n, buflen; | |
2966 | int copied = 0; | |
1da177e4 LT |
2967 | |
2968 | /* Don't allow overflow */ | |
2969 | if ((unsigned long) addr + count < count) | |
2970 | count = -(unsigned long) addr; | |
d0107eb0 | 2971 | buflen = count; |
1da177e4 | 2972 | |
e81ce85f JK |
2973 | spin_lock(&vmap_area_lock); |
2974 | list_for_each_entry(va, &vmap_area_list, list) { | |
2975 | if (!count) | |
2976 | break; | |
2977 | ||
688fcbfc | 2978 | if (!va->vm) |
e81ce85f JK |
2979 | continue; |
2980 | ||
2981 | vm = va->vm; | |
2982 | vaddr = (char *) vm->addr; | |
762216ab | 2983 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2984 | continue; |
2985 | while (addr < vaddr) { | |
2986 | if (count == 0) | |
2987 | goto finished; | |
2988 | buf++; | |
2989 | addr++; | |
2990 | count--; | |
2991 | } | |
762216ab | 2992 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2993 | if (n > count) |
2994 | n = count; | |
e81ce85f | 2995 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2996 | aligned_vwrite(buf, addr, n); |
2997 | copied++; | |
2998 | } | |
2999 | buf += n; | |
3000 | addr += n; | |
3001 | count -= n; | |
1da177e4 LT |
3002 | } |
3003 | finished: | |
e81ce85f | 3004 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3005 | if (!copied) |
3006 | return 0; | |
3007 | return buflen; | |
1da177e4 | 3008 | } |
83342314 NP |
3009 | |
3010 | /** | |
92eac168 MR |
3011 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
3012 | * @vma: vma to cover | |
3013 | * @uaddr: target user address to start at | |
3014 | * @kaddr: virtual address of vmalloc kernel memory | |
3015 | * @size: size of map area | |
7682486b | 3016 | * |
92eac168 | 3017 | * Returns: 0 for success, -Exxx on failure |
83342314 | 3018 | * |
92eac168 MR |
3019 | * This function checks that @kaddr is a valid vmalloc'ed area, |
3020 | * and that it is big enough to cover the range starting at | |
3021 | * @uaddr in @vma. Will return failure if that criteria isn't | |
3022 | * met. | |
83342314 | 3023 | * |
92eac168 | 3024 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 3025 | */ |
e69e9d4a HD |
3026 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
3027 | void *kaddr, unsigned long size) | |
83342314 NP |
3028 | { |
3029 | struct vm_struct *area; | |
83342314 | 3030 | |
e69e9d4a HD |
3031 | size = PAGE_ALIGN(size); |
3032 | ||
3033 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
3034 | return -EINVAL; |
3035 | ||
e69e9d4a | 3036 | area = find_vm_area(kaddr); |
83342314 | 3037 | if (!area) |
db64fe02 | 3038 | return -EINVAL; |
83342314 | 3039 | |
fe9041c2 | 3040 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3041 | return -EINVAL; |
83342314 | 3042 | |
401592d2 | 3043 | if (kaddr + size > area->addr + get_vm_area_size(area)) |
db64fe02 | 3044 | return -EINVAL; |
83342314 | 3045 | |
83342314 | 3046 | do { |
e69e9d4a | 3047 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3048 | int ret; |
3049 | ||
83342314 NP |
3050 | ret = vm_insert_page(vma, uaddr, page); |
3051 | if (ret) | |
3052 | return ret; | |
3053 | ||
3054 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3055 | kaddr += PAGE_SIZE; |
3056 | size -= PAGE_SIZE; | |
3057 | } while (size > 0); | |
83342314 | 3058 | |
314e51b9 | 3059 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3060 | |
db64fe02 | 3061 | return 0; |
83342314 | 3062 | } |
e69e9d4a HD |
3063 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
3064 | ||
3065 | /** | |
92eac168 MR |
3066 | * remap_vmalloc_range - map vmalloc pages to userspace |
3067 | * @vma: vma to cover (map full range of vma) | |
3068 | * @addr: vmalloc memory | |
3069 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3070 | * |
92eac168 | 3071 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3072 | * |
92eac168 MR |
3073 | * This function checks that addr is a valid vmalloc'ed area, and |
3074 | * that it is big enough to cover the vma. Will return failure if | |
3075 | * that criteria isn't met. | |
e69e9d4a | 3076 | * |
92eac168 | 3077 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3078 | */ |
3079 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3080 | unsigned long pgoff) | |
3081 | { | |
3082 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
3083 | addr + (pgoff << PAGE_SHIFT), | |
3084 | vma->vm_end - vma->vm_start); | |
3085 | } | |
83342314 NP |
3086 | EXPORT_SYMBOL(remap_vmalloc_range); |
3087 | ||
1eeb66a1 CH |
3088 | /* |
3089 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
3090 | * have one. | |
3f8fd02b JR |
3091 | * |
3092 | * The purpose of this function is to make sure the vmalloc area | |
3093 | * mappings are identical in all page-tables in the system. | |
1eeb66a1 | 3094 | */ |
3b32123d | 3095 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
3096 | { |
3097 | } | |
5f4352fb JF |
3098 | |
3099 | ||
8b1e0f81 | 3100 | static int f(pte_t *pte, unsigned long addr, void *data) |
5f4352fb | 3101 | { |
cd12909c DV |
3102 | pte_t ***p = data; |
3103 | ||
3104 | if (p) { | |
3105 | *(*p) = pte; | |
3106 | (*p)++; | |
3107 | } | |
5f4352fb JF |
3108 | return 0; |
3109 | } | |
3110 | ||
3111 | /** | |
92eac168 MR |
3112 | * alloc_vm_area - allocate a range of kernel address space |
3113 | * @size: size of the area | |
3114 | * @ptes: returns the PTEs for the address space | |
7682486b | 3115 | * |
92eac168 | 3116 | * Returns: NULL on failure, vm_struct on success |
5f4352fb | 3117 | * |
92eac168 MR |
3118 | * This function reserves a range of kernel address space, and |
3119 | * allocates pagetables to map that range. No actual mappings | |
3120 | * are created. | |
cd12909c | 3121 | * |
92eac168 MR |
3122 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
3123 | * allocated for the VM area are returned. | |
5f4352fb | 3124 | */ |
cd12909c | 3125 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
3126 | { |
3127 | struct vm_struct *area; | |
3128 | ||
23016969 CL |
3129 | area = get_vm_area_caller(size, VM_IOREMAP, |
3130 | __builtin_return_address(0)); | |
5f4352fb JF |
3131 | if (area == NULL) |
3132 | return NULL; | |
3133 | ||
3134 | /* | |
3135 | * This ensures that page tables are constructed for this region | |
3136 | * of kernel virtual address space and mapped into init_mm. | |
3137 | */ | |
3138 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 3139 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
3140 | free_vm_area(area); |
3141 | return NULL; | |
3142 | } | |
3143 | ||
5f4352fb JF |
3144 | return area; |
3145 | } | |
3146 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
3147 | ||
3148 | void free_vm_area(struct vm_struct *area) | |
3149 | { | |
3150 | struct vm_struct *ret; | |
3151 | ret = remove_vm_area(area->addr); | |
3152 | BUG_ON(ret != area); | |
3153 | kfree(area); | |
3154 | } | |
3155 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3156 | |
4f8b02b4 | 3157 | #ifdef CONFIG_SMP |
ca23e405 TH |
3158 | static struct vmap_area *node_to_va(struct rb_node *n) |
3159 | { | |
4583e773 | 3160 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3161 | } |
3162 | ||
3163 | /** | |
68ad4a33 URS |
3164 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3165 | * @addr: target address | |
ca23e405 | 3166 | * |
68ad4a33 URS |
3167 | * Returns: vmap_area if it is found. If there is no such area |
3168 | * the first highest(reverse order) vmap_area is returned | |
3169 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3170 | * if there are no any areas before @addr. | |
ca23e405 | 3171 | */ |
68ad4a33 URS |
3172 | static struct vmap_area * |
3173 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3174 | { |
68ad4a33 URS |
3175 | struct vmap_area *va, *tmp; |
3176 | struct rb_node *n; | |
3177 | ||
3178 | n = free_vmap_area_root.rb_node; | |
3179 | va = NULL; | |
ca23e405 TH |
3180 | |
3181 | while (n) { | |
68ad4a33 URS |
3182 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3183 | if (tmp->va_start <= addr) { | |
3184 | va = tmp; | |
3185 | if (tmp->va_end >= addr) | |
3186 | break; | |
3187 | ||
ca23e405 | 3188 | n = n->rb_right; |
68ad4a33 URS |
3189 | } else { |
3190 | n = n->rb_left; | |
3191 | } | |
ca23e405 TH |
3192 | } |
3193 | ||
68ad4a33 | 3194 | return va; |
ca23e405 TH |
3195 | } |
3196 | ||
3197 | /** | |
68ad4a33 URS |
3198 | * pvm_determine_end_from_reverse - find the highest aligned address |
3199 | * of free block below VMALLOC_END | |
3200 | * @va: | |
3201 | * in - the VA we start the search(reverse order); | |
3202 | * out - the VA with the highest aligned end address. | |
ca23e405 | 3203 | * |
68ad4a33 | 3204 | * Returns: determined end address within vmap_area |
ca23e405 | 3205 | */ |
68ad4a33 URS |
3206 | static unsigned long |
3207 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3208 | { |
68ad4a33 | 3209 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3210 | unsigned long addr; |
3211 | ||
68ad4a33 URS |
3212 | if (likely(*va)) { |
3213 | list_for_each_entry_from_reverse((*va), | |
3214 | &free_vmap_area_list, list) { | |
3215 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3216 | if ((*va)->va_start < addr) | |
3217 | return addr; | |
3218 | } | |
ca23e405 TH |
3219 | } |
3220 | ||
68ad4a33 | 3221 | return 0; |
ca23e405 TH |
3222 | } |
3223 | ||
3224 | /** | |
3225 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3226 | * @offsets: array containing offset of each area | |
3227 | * @sizes: array containing size of each area | |
3228 | * @nr_vms: the number of areas to allocate | |
3229 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3230 | * |
3231 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3232 | * vm_structs on success, %NULL on failure | |
3233 | * | |
3234 | * Percpu allocator wants to use congruent vm areas so that it can | |
3235 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3236 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3237 | * be scattered pretty far, distance between two areas easily going up | |
3238 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3239 | * areas are allocated from top. | |
ca23e405 | 3240 | * |
68ad4a33 URS |
3241 | * Despite its complicated look, this allocator is rather simple. It |
3242 | * does everything top-down and scans free blocks from the end looking | |
3243 | * for matching base. While scanning, if any of the areas do not fit the | |
3244 | * base address is pulled down to fit the area. Scanning is repeated till | |
3245 | * all the areas fit and then all necessary data structures are inserted | |
3246 | * and the result is returned. | |
ca23e405 TH |
3247 | */ |
3248 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3249 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3250 | size_t align) |
ca23e405 TH |
3251 | { |
3252 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3253 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3254 | struct vmap_area **vas, *va; |
ca23e405 TH |
3255 | struct vm_struct **vms; |
3256 | int area, area2, last_area, term_area; | |
68ad4a33 | 3257 | unsigned long base, start, size, end, last_end; |
ca23e405 | 3258 | bool purged = false; |
68ad4a33 | 3259 | enum fit_type type; |
ca23e405 | 3260 | |
ca23e405 | 3261 | /* verify parameters and allocate data structures */ |
891c49ab | 3262 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3263 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3264 | start = offsets[area]; | |
3265 | end = start + sizes[area]; | |
3266 | ||
3267 | /* is everything aligned properly? */ | |
3268 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3269 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3270 | ||
3271 | /* detect the area with the highest address */ | |
3272 | if (start > offsets[last_area]) | |
3273 | last_area = area; | |
3274 | ||
c568da28 | 3275 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3276 | unsigned long start2 = offsets[area2]; |
3277 | unsigned long end2 = start2 + sizes[area2]; | |
3278 | ||
c568da28 | 3279 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3280 | } |
3281 | } | |
3282 | last_end = offsets[last_area] + sizes[last_area]; | |
3283 | ||
3284 | if (vmalloc_end - vmalloc_start < last_end) { | |
3285 | WARN_ON(true); | |
3286 | return NULL; | |
3287 | } | |
3288 | ||
4d67d860 TM |
3289 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3290 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3291 | if (!vas || !vms) |
f1db7afd | 3292 | goto err_free2; |
ca23e405 TH |
3293 | |
3294 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3295 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3296 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3297 | if (!vas[area] || !vms[area]) |
3298 | goto err_free; | |
3299 | } | |
3300 | retry: | |
3301 | spin_lock(&vmap_area_lock); | |
3302 | ||
3303 | /* start scanning - we scan from the top, begin with the last area */ | |
3304 | area = term_area = last_area; | |
3305 | start = offsets[area]; | |
3306 | end = start + sizes[area]; | |
3307 | ||
68ad4a33 URS |
3308 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3309 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3310 | |
3311 | while (true) { | |
ca23e405 TH |
3312 | /* |
3313 | * base might have underflowed, add last_end before | |
3314 | * comparing. | |
3315 | */ | |
68ad4a33 URS |
3316 | if (base + last_end < vmalloc_start + last_end) |
3317 | goto overflow; | |
ca23e405 TH |
3318 | |
3319 | /* | |
68ad4a33 | 3320 | * Fitting base has not been found. |
ca23e405 | 3321 | */ |
68ad4a33 URS |
3322 | if (va == NULL) |
3323 | goto overflow; | |
ca23e405 | 3324 | |
5336e52c KS |
3325 | /* |
3326 | * If required width exeeds current VA block, move | |
3327 | * base downwards and then recheck. | |
3328 | */ | |
3329 | if (base + end > va->va_end) { | |
3330 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3331 | term_area = area; | |
3332 | continue; | |
3333 | } | |
3334 | ||
ca23e405 | 3335 | /* |
68ad4a33 | 3336 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3337 | */ |
5336e52c | 3338 | if (base + start < va->va_start) { |
68ad4a33 URS |
3339 | va = node_to_va(rb_prev(&va->rb_node)); |
3340 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3341 | term_area = area; |
3342 | continue; | |
3343 | } | |
3344 | ||
3345 | /* | |
3346 | * This area fits, move on to the previous one. If | |
3347 | * the previous one is the terminal one, we're done. | |
3348 | */ | |
3349 | area = (area + nr_vms - 1) % nr_vms; | |
3350 | if (area == term_area) | |
3351 | break; | |
68ad4a33 | 3352 | |
ca23e405 TH |
3353 | start = offsets[area]; |
3354 | end = start + sizes[area]; | |
68ad4a33 | 3355 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3356 | } |
68ad4a33 | 3357 | |
ca23e405 TH |
3358 | /* we've found a fitting base, insert all va's */ |
3359 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3360 | int ret; |
ca23e405 | 3361 | |
68ad4a33 URS |
3362 | start = base + offsets[area]; |
3363 | size = sizes[area]; | |
ca23e405 | 3364 | |
68ad4a33 URS |
3365 | va = pvm_find_va_enclose_addr(start); |
3366 | if (WARN_ON_ONCE(va == NULL)) | |
3367 | /* It is a BUG(), but trigger recovery instead. */ | |
3368 | goto recovery; | |
3369 | ||
3370 | type = classify_va_fit_type(va, start, size); | |
3371 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3372 | /* It is a BUG(), but trigger recovery instead. */ | |
3373 | goto recovery; | |
3374 | ||
3375 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3376 | if (unlikely(ret)) | |
3377 | goto recovery; | |
3378 | ||
3379 | /* Allocated area. */ | |
3380 | va = vas[area]; | |
3381 | va->va_start = start; | |
3382 | va->va_end = start + size; | |
3383 | ||
3384 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
3385 | } | |
ca23e405 TH |
3386 | |
3387 | spin_unlock(&vmap_area_lock); | |
3388 | ||
3389 | /* insert all vm's */ | |
3390 | for (area = 0; area < nr_vms; area++) | |
3645cb4a ZY |
3391 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
3392 | pcpu_get_vm_areas); | |
ca23e405 TH |
3393 | |
3394 | kfree(vas); | |
3395 | return vms; | |
3396 | ||
68ad4a33 URS |
3397 | recovery: |
3398 | /* Remove previously inserted areas. */ | |
3399 | while (area--) { | |
3400 | __free_vmap_area(vas[area]); | |
3401 | vas[area] = NULL; | |
3402 | } | |
3403 | ||
3404 | overflow: | |
3405 | spin_unlock(&vmap_area_lock); | |
3406 | if (!purged) { | |
3407 | purge_vmap_area_lazy(); | |
3408 | purged = true; | |
3409 | ||
3410 | /* Before "retry", check if we recover. */ | |
3411 | for (area = 0; area < nr_vms; area++) { | |
3412 | if (vas[area]) | |
3413 | continue; | |
3414 | ||
3415 | vas[area] = kmem_cache_zalloc( | |
3416 | vmap_area_cachep, GFP_KERNEL); | |
3417 | if (!vas[area]) | |
3418 | goto err_free; | |
3419 | } | |
3420 | ||
3421 | goto retry; | |
3422 | } | |
3423 | ||
ca23e405 TH |
3424 | err_free: |
3425 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3426 | if (vas[area]) |
3427 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3428 | ||
f1db7afd | 3429 | kfree(vms[area]); |
ca23e405 | 3430 | } |
f1db7afd | 3431 | err_free2: |
ca23e405 TH |
3432 | kfree(vas); |
3433 | kfree(vms); | |
3434 | return NULL; | |
3435 | } | |
3436 | ||
3437 | /** | |
3438 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3439 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3440 | * @nr_vms: the number of allocated areas | |
3441 | * | |
3442 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3443 | */ | |
3444 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3445 | { | |
3446 | int i; | |
3447 | ||
3448 | for (i = 0; i < nr_vms; i++) | |
3449 | free_vm_area(vms[i]); | |
3450 | kfree(vms); | |
3451 | } | |
4f8b02b4 | 3452 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
3453 | |
3454 | #ifdef CONFIG_PROC_FS | |
3455 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 3456 | __acquires(&vmap_area_lock) |
a10aa579 | 3457 | { |
d4033afd | 3458 | spin_lock(&vmap_area_lock); |
3f500069 | 3459 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3460 | } |
3461 | ||
3462 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3463 | { | |
3f500069 | 3464 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3465 | } |
3466 | ||
3467 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 3468 | __releases(&vmap_area_lock) |
a10aa579 | 3469 | { |
d4033afd | 3470 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
3471 | } |
3472 | ||
a47a126a ED |
3473 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3474 | { | |
e5adfffc | 3475 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3476 | unsigned int nr, *counters = m->private; |
3477 | ||
3478 | if (!counters) | |
3479 | return; | |
3480 | ||
af12346c WL |
3481 | if (v->flags & VM_UNINITIALIZED) |
3482 | return; | |
7e5b528b DV |
3483 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3484 | smp_rmb(); | |
af12346c | 3485 | |
a47a126a ED |
3486 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3487 | ||
3488 | for (nr = 0; nr < v->nr_pages; nr++) | |
3489 | counters[page_to_nid(v->pages[nr])]++; | |
3490 | ||
3491 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3492 | if (counters[nr]) | |
3493 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3494 | } | |
3495 | } | |
3496 | ||
dd3b8353 URS |
3497 | static void show_purge_info(struct seq_file *m) |
3498 | { | |
3499 | struct llist_node *head; | |
3500 | struct vmap_area *va; | |
3501 | ||
3502 | head = READ_ONCE(vmap_purge_list.first); | |
3503 | if (head == NULL) | |
3504 | return; | |
3505 | ||
3506 | llist_for_each_entry(va, head, purge_list) { | |
3507 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", | |
3508 | (void *)va->va_start, (void *)va->va_end, | |
3509 | va->va_end - va->va_start); | |
3510 | } | |
3511 | } | |
3512 | ||
a10aa579 CL |
3513 | static int s_show(struct seq_file *m, void *p) |
3514 | { | |
3f500069 | 3515 | struct vmap_area *va; |
d4033afd JK |
3516 | struct vm_struct *v; |
3517 | ||
3f500069 | 3518 | va = list_entry(p, struct vmap_area, list); |
3519 | ||
c2ce8c14 | 3520 | /* |
688fcbfc PL |
3521 | * s_show can encounter race with remove_vm_area, !vm on behalf |
3522 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 3523 | */ |
688fcbfc | 3524 | if (!va->vm) { |
dd3b8353 | 3525 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 3526 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 3527 | va->va_end - va->va_start); |
78c72746 | 3528 | |
d4033afd | 3529 | return 0; |
78c72746 | 3530 | } |
d4033afd JK |
3531 | |
3532 | v = va->vm; | |
a10aa579 | 3533 | |
45ec1690 | 3534 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3535 | v->addr, v->addr + v->size, v->size); |
3536 | ||
62c70bce JP |
3537 | if (v->caller) |
3538 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3539 | |
a10aa579 CL |
3540 | if (v->nr_pages) |
3541 | seq_printf(m, " pages=%d", v->nr_pages); | |
3542 | ||
3543 | if (v->phys_addr) | |
199eaa05 | 3544 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3545 | |
3546 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3547 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3548 | |
3549 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3550 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3551 | |
3552 | if (v->flags & VM_MAP) | |
f4527c90 | 3553 | seq_puts(m, " vmap"); |
a10aa579 CL |
3554 | |
3555 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3556 | seq_puts(m, " user"); |
a10aa579 | 3557 | |
fe9041c2 CH |
3558 | if (v->flags & VM_DMA_COHERENT) |
3559 | seq_puts(m, " dma-coherent"); | |
3560 | ||
244d63ee | 3561 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3562 | seq_puts(m, " vpages"); |
a10aa579 | 3563 | |
a47a126a | 3564 | show_numa_info(m, v); |
a10aa579 | 3565 | seq_putc(m, '\n'); |
dd3b8353 URS |
3566 | |
3567 | /* | |
3568 | * As a final step, dump "unpurged" areas. Note, | |
3569 | * that entire "/proc/vmallocinfo" output will not | |
3570 | * be address sorted, because the purge list is not | |
3571 | * sorted. | |
3572 | */ | |
3573 | if (list_is_last(&va->list, &vmap_area_list)) | |
3574 | show_purge_info(m); | |
3575 | ||
a10aa579 CL |
3576 | return 0; |
3577 | } | |
3578 | ||
5f6a6a9c | 3579 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3580 | .start = s_start, |
3581 | .next = s_next, | |
3582 | .stop = s_stop, | |
3583 | .show = s_show, | |
3584 | }; | |
5f6a6a9c | 3585 | |
5f6a6a9c AD |
3586 | static int __init proc_vmalloc_init(void) |
3587 | { | |
fddda2b7 | 3588 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3589 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3590 | &vmalloc_op, |
3591 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3592 | else |
0825a6f9 | 3593 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3594 | return 0; |
3595 | } | |
3596 | module_init(proc_vmalloc_init); | |
db3808c1 | 3597 | |
a10aa579 | 3598 | #endif |