]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmalloc.c | |
3 | * | |
4 | * Copyright (C) 1993 Linus Torvalds | |
5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
6 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
7 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
9 | */ |
10 | ||
db64fe02 | 11 | #include <linux/vmalloc.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> | |
14 | #include <linux/highmem.h> | |
c3edc401 | 15 | #include <linux/sched/signal.h> |
1da177e4 LT |
16 | #include <linux/slab.h> |
17 | #include <linux/spinlock.h> | |
18 | #include <linux/interrupt.h> | |
5f6a6a9c | 19 | #include <linux/proc_fs.h> |
a10aa579 | 20 | #include <linux/seq_file.h> |
868b104d | 21 | #include <linux/set_memory.h> |
3ac7fe5a | 22 | #include <linux/debugobjects.h> |
23016969 | 23 | #include <linux/kallsyms.h> |
db64fe02 | 24 | #include <linux/list.h> |
4da56b99 | 25 | #include <linux/notifier.h> |
db64fe02 NP |
26 | #include <linux/rbtree.h> |
27 | #include <linux/radix-tree.h> | |
28 | #include <linux/rcupdate.h> | |
f0aa6617 | 29 | #include <linux/pfn.h> |
89219d37 | 30 | #include <linux/kmemleak.h> |
60063497 | 31 | #include <linux/atomic.h> |
3b32123d | 32 | #include <linux/compiler.h> |
32fcfd40 | 33 | #include <linux/llist.h> |
0f616be1 | 34 | #include <linux/bitops.h> |
3b32123d | 35 | |
7c0f6ba6 | 36 | #include <linux/uaccess.h> |
1da177e4 | 37 | #include <asm/tlbflush.h> |
2dca6999 | 38 | #include <asm/shmparam.h> |
1da177e4 | 39 | |
dd56b046 MG |
40 | #include "internal.h" |
41 | ||
32fcfd40 AV |
42 | struct vfree_deferred { |
43 | struct llist_head list; | |
44 | struct work_struct wq; | |
45 | }; | |
46 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
47 | ||
48 | static void __vunmap(const void *, int); | |
49 | ||
50 | static void free_work(struct work_struct *w) | |
51 | { | |
52 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
53 | struct llist_node *t, *llnode; |
54 | ||
55 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
56 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
57 | } |
58 | ||
db64fe02 | 59 | /*** Page table manipulation functions ***/ |
b221385b | 60 | |
1da177e4 LT |
61 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
62 | { | |
63 | pte_t *pte; | |
64 | ||
65 | pte = pte_offset_kernel(pmd, addr); | |
66 | do { | |
67 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
68 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
69 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
70 | } | |
71 | ||
db64fe02 | 72 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
73 | { |
74 | pmd_t *pmd; | |
75 | unsigned long next; | |
76 | ||
77 | pmd = pmd_offset(pud, addr); | |
78 | do { | |
79 | next = pmd_addr_end(addr, end); | |
b9820d8f TK |
80 | if (pmd_clear_huge(pmd)) |
81 | continue; | |
1da177e4 LT |
82 | if (pmd_none_or_clear_bad(pmd)) |
83 | continue; | |
84 | vunmap_pte_range(pmd, addr, next); | |
85 | } while (pmd++, addr = next, addr != end); | |
86 | } | |
87 | ||
c2febafc | 88 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end) |
1da177e4 LT |
89 | { |
90 | pud_t *pud; | |
91 | unsigned long next; | |
92 | ||
c2febafc | 93 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
94 | do { |
95 | next = pud_addr_end(addr, end); | |
b9820d8f TK |
96 | if (pud_clear_huge(pud)) |
97 | continue; | |
1da177e4 LT |
98 | if (pud_none_or_clear_bad(pud)) |
99 | continue; | |
100 | vunmap_pmd_range(pud, addr, next); | |
101 | } while (pud++, addr = next, addr != end); | |
102 | } | |
103 | ||
c2febafc KS |
104 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
105 | { | |
106 | p4d_t *p4d; | |
107 | unsigned long next; | |
108 | ||
109 | p4d = p4d_offset(pgd, addr); | |
110 | do { | |
111 | next = p4d_addr_end(addr, end); | |
112 | if (p4d_clear_huge(p4d)) | |
113 | continue; | |
114 | if (p4d_none_or_clear_bad(p4d)) | |
115 | continue; | |
116 | vunmap_pud_range(p4d, addr, next); | |
117 | } while (p4d++, addr = next, addr != end); | |
118 | } | |
119 | ||
db64fe02 | 120 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
121 | { |
122 | pgd_t *pgd; | |
123 | unsigned long next; | |
1da177e4 LT |
124 | |
125 | BUG_ON(addr >= end); | |
126 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
127 | do { |
128 | next = pgd_addr_end(addr, end); | |
129 | if (pgd_none_or_clear_bad(pgd)) | |
130 | continue; | |
c2febafc | 131 | vunmap_p4d_range(pgd, addr, next); |
1da177e4 | 132 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
133 | } |
134 | ||
135 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 136 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
137 | { |
138 | pte_t *pte; | |
139 | ||
db64fe02 NP |
140 | /* |
141 | * nr is a running index into the array which helps higher level | |
142 | * callers keep track of where we're up to. | |
143 | */ | |
144 | ||
872fec16 | 145 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
146 | if (!pte) |
147 | return -ENOMEM; | |
148 | do { | |
db64fe02 NP |
149 | struct page *page = pages[*nr]; |
150 | ||
151 | if (WARN_ON(!pte_none(*pte))) | |
152 | return -EBUSY; | |
153 | if (WARN_ON(!page)) | |
1da177e4 LT |
154 | return -ENOMEM; |
155 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 156 | (*nr)++; |
1da177e4 LT |
157 | } while (pte++, addr += PAGE_SIZE, addr != end); |
158 | return 0; | |
159 | } | |
160 | ||
db64fe02 NP |
161 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
162 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
163 | { |
164 | pmd_t *pmd; | |
165 | unsigned long next; | |
166 | ||
167 | pmd = pmd_alloc(&init_mm, pud, addr); | |
168 | if (!pmd) | |
169 | return -ENOMEM; | |
170 | do { | |
171 | next = pmd_addr_end(addr, end); | |
db64fe02 | 172 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
173 | return -ENOMEM; |
174 | } while (pmd++, addr = next, addr != end); | |
175 | return 0; | |
176 | } | |
177 | ||
c2febafc | 178 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, |
db64fe02 | 179 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
180 | { |
181 | pud_t *pud; | |
182 | unsigned long next; | |
183 | ||
c2febafc | 184 | pud = pud_alloc(&init_mm, p4d, addr); |
1da177e4 LT |
185 | if (!pud) |
186 | return -ENOMEM; | |
187 | do { | |
188 | next = pud_addr_end(addr, end); | |
db64fe02 | 189 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
190 | return -ENOMEM; |
191 | } while (pud++, addr = next, addr != end); | |
192 | return 0; | |
193 | } | |
194 | ||
c2febafc KS |
195 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, |
196 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
197 | { | |
198 | p4d_t *p4d; | |
199 | unsigned long next; | |
200 | ||
201 | p4d = p4d_alloc(&init_mm, pgd, addr); | |
202 | if (!p4d) | |
203 | return -ENOMEM; | |
204 | do { | |
205 | next = p4d_addr_end(addr, end); | |
206 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr)) | |
207 | return -ENOMEM; | |
208 | } while (p4d++, addr = next, addr != end); | |
209 | return 0; | |
210 | } | |
211 | ||
db64fe02 NP |
212 | /* |
213 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
214 | * will have pfns corresponding to the "pages" array. | |
215 | * | |
216 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
217 | */ | |
8fc48985 TH |
218 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
219 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
220 | { |
221 | pgd_t *pgd; | |
222 | unsigned long next; | |
2e4e27c7 | 223 | unsigned long addr = start; |
db64fe02 NP |
224 | int err = 0; |
225 | int nr = 0; | |
1da177e4 LT |
226 | |
227 | BUG_ON(addr >= end); | |
228 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
229 | do { |
230 | next = pgd_addr_end(addr, end); | |
c2febafc | 231 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 232 | if (err) |
bf88c8c8 | 233 | return err; |
1da177e4 | 234 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 235 | |
db64fe02 | 236 | return nr; |
1da177e4 LT |
237 | } |
238 | ||
8fc48985 TH |
239 | static int vmap_page_range(unsigned long start, unsigned long end, |
240 | pgprot_t prot, struct page **pages) | |
241 | { | |
242 | int ret; | |
243 | ||
244 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
245 | flush_cache_vmap(start, end); | |
246 | return ret; | |
247 | } | |
248 | ||
81ac3ad9 | 249 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
250 | { |
251 | /* | |
ab4f2ee1 | 252 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
253 | * and fall back on vmalloc() if that fails. Others |
254 | * just put it in the vmalloc space. | |
255 | */ | |
256 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
257 | unsigned long addr = (unsigned long)x; | |
258 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
259 | return 1; | |
260 | #endif | |
261 | return is_vmalloc_addr(x); | |
262 | } | |
263 | ||
48667e7a | 264 | /* |
add688fb | 265 | * Walk a vmap address to the struct page it maps. |
48667e7a | 266 | */ |
add688fb | 267 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
268 | { |
269 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 270 | struct page *page = NULL; |
48667e7a | 271 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
272 | p4d_t *p4d; |
273 | pud_t *pud; | |
274 | pmd_t *pmd; | |
275 | pte_t *ptep, pte; | |
48667e7a | 276 | |
7aa413de IM |
277 | /* |
278 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
279 | * architectures that do not vmalloc module space | |
280 | */ | |
73bdf0a6 | 281 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 282 | |
c2febafc KS |
283 | if (pgd_none(*pgd)) |
284 | return NULL; | |
285 | p4d = p4d_offset(pgd, addr); | |
286 | if (p4d_none(*p4d)) | |
287 | return NULL; | |
288 | pud = pud_offset(p4d, addr); | |
029c54b0 AB |
289 | |
290 | /* | |
291 | * Don't dereference bad PUD or PMD (below) entries. This will also | |
292 | * identify huge mappings, which we may encounter on architectures | |
293 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be | |
294 | * identified as vmalloc addresses by is_vmalloc_addr(), but are | |
295 | * not [unambiguously] associated with a struct page, so there is | |
296 | * no correct value to return for them. | |
297 | */ | |
298 | WARN_ON_ONCE(pud_bad(*pud)); | |
299 | if (pud_none(*pud) || pud_bad(*pud)) | |
c2febafc KS |
300 | return NULL; |
301 | pmd = pmd_offset(pud, addr); | |
029c54b0 AB |
302 | WARN_ON_ONCE(pmd_bad(*pmd)); |
303 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
c2febafc KS |
304 | return NULL; |
305 | ||
306 | ptep = pte_offset_map(pmd, addr); | |
307 | pte = *ptep; | |
308 | if (pte_present(pte)) | |
309 | page = pte_page(pte); | |
310 | pte_unmap(ptep); | |
add688fb | 311 | return page; |
48667e7a | 312 | } |
add688fb | 313 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
314 | |
315 | /* | |
add688fb | 316 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 317 | */ |
add688fb | 318 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 319 | { |
add688fb | 320 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 321 | } |
add688fb | 322 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 323 | |
db64fe02 NP |
324 | |
325 | /*** Global kva allocator ***/ | |
326 | ||
78c72746 | 327 | #define VM_LAZY_FREE 0x02 |
db64fe02 NP |
328 | #define VM_VM_AREA 0x04 |
329 | ||
db64fe02 | 330 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
331 | /* Export for kexec only */ |
332 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 333 | static LLIST_HEAD(vmap_purge_list); |
89699605 NP |
334 | static struct rb_root vmap_area_root = RB_ROOT; |
335 | ||
336 | /* The vmap cache globals are protected by vmap_area_lock */ | |
337 | static struct rb_node *free_vmap_cache; | |
338 | static unsigned long cached_hole_size; | |
339 | static unsigned long cached_vstart; | |
340 | static unsigned long cached_align; | |
341 | ||
ca23e405 | 342 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
343 | |
344 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 345 | { |
db64fe02 NP |
346 | struct rb_node *n = vmap_area_root.rb_node; |
347 | ||
348 | while (n) { | |
349 | struct vmap_area *va; | |
350 | ||
351 | va = rb_entry(n, struct vmap_area, rb_node); | |
352 | if (addr < va->va_start) | |
353 | n = n->rb_left; | |
cef2ac3f | 354 | else if (addr >= va->va_end) |
db64fe02 NP |
355 | n = n->rb_right; |
356 | else | |
357 | return va; | |
358 | } | |
359 | ||
360 | return NULL; | |
361 | } | |
362 | ||
363 | static void __insert_vmap_area(struct vmap_area *va) | |
364 | { | |
365 | struct rb_node **p = &vmap_area_root.rb_node; | |
366 | struct rb_node *parent = NULL; | |
367 | struct rb_node *tmp; | |
368 | ||
369 | while (*p) { | |
170168d0 | 370 | struct vmap_area *tmp_va; |
db64fe02 NP |
371 | |
372 | parent = *p; | |
170168d0 NK |
373 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
374 | if (va->va_start < tmp_va->va_end) | |
db64fe02 | 375 | p = &(*p)->rb_left; |
170168d0 | 376 | else if (va->va_end > tmp_va->va_start) |
db64fe02 NP |
377 | p = &(*p)->rb_right; |
378 | else | |
379 | BUG(); | |
380 | } | |
381 | ||
382 | rb_link_node(&va->rb_node, parent, p); | |
383 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
384 | ||
4341fa45 | 385 | /* address-sort this list */ |
db64fe02 NP |
386 | tmp = rb_prev(&va->rb_node); |
387 | if (tmp) { | |
388 | struct vmap_area *prev; | |
389 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
390 | list_add_rcu(&va->list, &prev->list); | |
391 | } else | |
392 | list_add_rcu(&va->list, &vmap_area_list); | |
393 | } | |
394 | ||
395 | static void purge_vmap_area_lazy(void); | |
396 | ||
4da56b99 CW |
397 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); |
398 | ||
db64fe02 NP |
399 | /* |
400 | * Allocate a region of KVA of the specified size and alignment, within the | |
401 | * vstart and vend. | |
402 | */ | |
403 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
404 | unsigned long align, | |
405 | unsigned long vstart, unsigned long vend, | |
406 | int node, gfp_t gfp_mask) | |
407 | { | |
408 | struct vmap_area *va; | |
409 | struct rb_node *n; | |
1da177e4 | 410 | unsigned long addr; |
db64fe02 | 411 | int purged = 0; |
89699605 | 412 | struct vmap_area *first; |
db64fe02 | 413 | |
7766970c | 414 | BUG_ON(!size); |
891c49ab | 415 | BUG_ON(offset_in_page(size)); |
89699605 | 416 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 417 | |
5803ed29 | 418 | might_sleep(); |
4da56b99 | 419 | |
db64fe02 NP |
420 | va = kmalloc_node(sizeof(struct vmap_area), |
421 | gfp_mask & GFP_RECLAIM_MASK, node); | |
422 | if (unlikely(!va)) | |
423 | return ERR_PTR(-ENOMEM); | |
424 | ||
7f88f88f CM |
425 | /* |
426 | * Only scan the relevant parts containing pointers to other objects | |
427 | * to avoid false negatives. | |
428 | */ | |
429 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK); | |
430 | ||
db64fe02 NP |
431 | retry: |
432 | spin_lock(&vmap_area_lock); | |
89699605 NP |
433 | /* |
434 | * Invalidate cache if we have more permissive parameters. | |
435 | * cached_hole_size notes the largest hole noticed _below_ | |
436 | * the vmap_area cached in free_vmap_cache: if size fits | |
437 | * into that hole, we want to scan from vstart to reuse | |
438 | * the hole instead of allocating above free_vmap_cache. | |
439 | * Note that __free_vmap_area may update free_vmap_cache | |
440 | * without updating cached_hole_size or cached_align. | |
441 | */ | |
442 | if (!free_vmap_cache || | |
443 | size < cached_hole_size || | |
444 | vstart < cached_vstart || | |
445 | align < cached_align) { | |
446 | nocache: | |
447 | cached_hole_size = 0; | |
448 | free_vmap_cache = NULL; | |
449 | } | |
450 | /* record if we encounter less permissive parameters */ | |
451 | cached_vstart = vstart; | |
452 | cached_align = align; | |
453 | ||
454 | /* find starting point for our search */ | |
455 | if (free_vmap_cache) { | |
456 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
248ac0e1 | 457 | addr = ALIGN(first->va_end, align); |
89699605 NP |
458 | if (addr < vstart) |
459 | goto nocache; | |
bcb615a8 | 460 | if (addr + size < addr) |
89699605 NP |
461 | goto overflow; |
462 | ||
463 | } else { | |
464 | addr = ALIGN(vstart, align); | |
bcb615a8 | 465 | if (addr + size < addr) |
89699605 NP |
466 | goto overflow; |
467 | ||
468 | n = vmap_area_root.rb_node; | |
469 | first = NULL; | |
470 | ||
471 | while (n) { | |
db64fe02 NP |
472 | struct vmap_area *tmp; |
473 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
474 | if (tmp->va_end >= addr) { | |
db64fe02 | 475 | first = tmp; |
89699605 NP |
476 | if (tmp->va_start <= addr) |
477 | break; | |
478 | n = n->rb_left; | |
479 | } else | |
db64fe02 | 480 | n = n->rb_right; |
89699605 | 481 | } |
db64fe02 NP |
482 | |
483 | if (!first) | |
484 | goto found; | |
db64fe02 | 485 | } |
89699605 NP |
486 | |
487 | /* from the starting point, walk areas until a suitable hole is found */ | |
248ac0e1 | 488 | while (addr + size > first->va_start && addr + size <= vend) { |
89699605 NP |
489 | if (addr + cached_hole_size < first->va_start) |
490 | cached_hole_size = first->va_start - addr; | |
248ac0e1 | 491 | addr = ALIGN(first->va_end, align); |
bcb615a8 | 492 | if (addr + size < addr) |
89699605 NP |
493 | goto overflow; |
494 | ||
92ca922f | 495 | if (list_is_last(&first->list, &vmap_area_list)) |
89699605 | 496 | goto found; |
92ca922f | 497 | |
6219c2a2 | 498 | first = list_next_entry(first, list); |
db64fe02 NP |
499 | } |
500 | ||
89699605 | 501 | found: |
afd07389 URS |
502 | /* |
503 | * Check also calculated address against the vstart, | |
504 | * because it can be 0 because of big align request. | |
505 | */ | |
506 | if (addr + size > vend || addr < vstart) | |
89699605 | 507 | goto overflow; |
db64fe02 NP |
508 | |
509 | va->va_start = addr; | |
510 | va->va_end = addr + size; | |
511 | va->flags = 0; | |
512 | __insert_vmap_area(va); | |
89699605 | 513 | free_vmap_cache = &va->rb_node; |
db64fe02 NP |
514 | spin_unlock(&vmap_area_lock); |
515 | ||
61e16557 | 516 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
517 | BUG_ON(va->va_start < vstart); |
518 | BUG_ON(va->va_end > vend); | |
519 | ||
db64fe02 | 520 | return va; |
89699605 NP |
521 | |
522 | overflow: | |
523 | spin_unlock(&vmap_area_lock); | |
524 | if (!purged) { | |
525 | purge_vmap_area_lazy(); | |
526 | purged = 1; | |
527 | goto retry; | |
528 | } | |
4da56b99 CW |
529 | |
530 | if (gfpflags_allow_blocking(gfp_mask)) { | |
531 | unsigned long freed = 0; | |
532 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
533 | if (freed > 0) { | |
534 | purged = 0; | |
535 | goto retry; | |
536 | } | |
537 | } | |
538 | ||
03497d76 | 539 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
540 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
541 | size); | |
89699605 NP |
542 | kfree(va); |
543 | return ERR_PTR(-EBUSY); | |
db64fe02 NP |
544 | } |
545 | ||
4da56b99 CW |
546 | int register_vmap_purge_notifier(struct notifier_block *nb) |
547 | { | |
548 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
549 | } | |
550 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
551 | ||
552 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
553 | { | |
554 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
555 | } | |
556 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
557 | ||
db64fe02 NP |
558 | static void __free_vmap_area(struct vmap_area *va) |
559 | { | |
560 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
89699605 NP |
561 | |
562 | if (free_vmap_cache) { | |
563 | if (va->va_end < cached_vstart) { | |
564 | free_vmap_cache = NULL; | |
565 | } else { | |
566 | struct vmap_area *cache; | |
567 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
568 | if (va->va_start <= cache->va_start) { | |
569 | free_vmap_cache = rb_prev(&va->rb_node); | |
570 | /* | |
571 | * We don't try to update cached_hole_size or | |
572 | * cached_align, but it won't go very wrong. | |
573 | */ | |
574 | } | |
575 | } | |
576 | } | |
db64fe02 NP |
577 | rb_erase(&va->rb_node, &vmap_area_root); |
578 | RB_CLEAR_NODE(&va->rb_node); | |
579 | list_del_rcu(&va->list); | |
580 | ||
ca23e405 TH |
581 | /* |
582 | * Track the highest possible candidate for pcpu area | |
583 | * allocation. Areas outside of vmalloc area can be returned | |
584 | * here too, consider only end addresses which fall inside | |
585 | * vmalloc area proper. | |
586 | */ | |
587 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
588 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
589 | ||
14769de9 | 590 | kfree_rcu(va, rcu_head); |
db64fe02 NP |
591 | } |
592 | ||
593 | /* | |
594 | * Free a region of KVA allocated by alloc_vmap_area | |
595 | */ | |
596 | static void free_vmap_area(struct vmap_area *va) | |
597 | { | |
598 | spin_lock(&vmap_area_lock); | |
599 | __free_vmap_area(va); | |
600 | spin_unlock(&vmap_area_lock); | |
601 | } | |
602 | ||
603 | /* | |
604 | * Clear the pagetable entries of a given vmap_area | |
605 | */ | |
606 | static void unmap_vmap_area(struct vmap_area *va) | |
607 | { | |
608 | vunmap_page_range(va->va_start, va->va_end); | |
609 | } | |
610 | ||
611 | /* | |
612 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
613 | * before attempting to purge with a TLB flush. | |
614 | * | |
615 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
616 | * and take slightly longer to purge, but it will linearly reduce the number of | |
617 | * global TLB flushes that must be performed. It would seem natural to scale | |
618 | * this number up linearly with the number of CPUs (because vmapping activity | |
619 | * could also scale linearly with the number of CPUs), however it is likely | |
620 | * that in practice, workloads might be constrained in other ways that mean | |
621 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
622 | * conservative and not introduce a big latency on huge systems, so go with | |
623 | * a less aggressive log scale. It will still be an improvement over the old | |
624 | * code, and it will be simple to change the scale factor if we find that it | |
625 | * becomes a problem on bigger systems. | |
626 | */ | |
627 | static unsigned long lazy_max_pages(void) | |
628 | { | |
629 | unsigned int log; | |
630 | ||
631 | log = fls(num_online_cpus()); | |
632 | ||
633 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
634 | } | |
635 | ||
636 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
637 | ||
0574ecd1 CH |
638 | /* |
639 | * Serialize vmap purging. There is no actual criticial section protected | |
640 | * by this look, but we want to avoid concurrent calls for performance | |
641 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
642 | */ | |
f9e09977 | 643 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 644 | |
02b709df NP |
645 | /* for per-CPU blocks */ |
646 | static void purge_fragmented_blocks_allcpus(void); | |
647 | ||
3ee48b6a CW |
648 | /* |
649 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
650 | * immediately freed. | |
651 | */ | |
652 | void set_iounmap_nonlazy(void) | |
653 | { | |
654 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); | |
655 | } | |
656 | ||
db64fe02 NP |
657 | /* |
658 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 659 | */ |
0574ecd1 | 660 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 661 | { |
80c4bd7a | 662 | struct llist_node *valist; |
db64fe02 | 663 | struct vmap_area *va; |
cbb76676 | 664 | struct vmap_area *n_va; |
68571be9 | 665 | int resched_threshold; |
db64fe02 | 666 | |
0574ecd1 | 667 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 668 | |
80c4bd7a | 669 | valist = llist_del_all(&vmap_purge_list); |
68571be9 URS |
670 | if (unlikely(valist == NULL)) |
671 | return false; | |
672 | ||
673 | /* | |
674 | * TODO: to calculate a flush range without looping. | |
675 | * The list can be up to lazy_max_pages() elements. | |
676 | */ | |
80c4bd7a | 677 | llist_for_each_entry(va, valist, purge_list) { |
0574ecd1 CH |
678 | if (va->va_start < start) |
679 | start = va->va_start; | |
680 | if (va->va_end > end) | |
681 | end = va->va_end; | |
db64fe02 | 682 | } |
db64fe02 | 683 | |
0574ecd1 | 684 | flush_tlb_kernel_range(start, end); |
68571be9 | 685 | resched_threshold = (int) lazy_max_pages() << 1; |
db64fe02 | 686 | |
0574ecd1 | 687 | spin_lock(&vmap_area_lock); |
763b218d JF |
688 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
689 | int nr = (va->va_end - va->va_start) >> PAGE_SHIFT; | |
690 | ||
0574ecd1 | 691 | __free_vmap_area(va); |
763b218d | 692 | atomic_sub(nr, &vmap_lazy_nr); |
68571be9 URS |
693 | |
694 | if (atomic_read(&vmap_lazy_nr) < resched_threshold) | |
695 | cond_resched_lock(&vmap_area_lock); | |
763b218d | 696 | } |
0574ecd1 CH |
697 | spin_unlock(&vmap_area_lock); |
698 | return true; | |
db64fe02 NP |
699 | } |
700 | ||
496850e5 NP |
701 | /* |
702 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
703 | * is already purging. | |
704 | */ | |
705 | static void try_purge_vmap_area_lazy(void) | |
706 | { | |
f9e09977 | 707 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 708 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 709 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 710 | } |
496850e5 NP |
711 | } |
712 | ||
db64fe02 NP |
713 | /* |
714 | * Kick off a purge of the outstanding lazy areas. | |
715 | */ | |
716 | static void purge_vmap_area_lazy(void) | |
717 | { | |
f9e09977 | 718 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
719 | purge_fragmented_blocks_allcpus(); |
720 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 721 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
722 | } |
723 | ||
724 | /* | |
64141da5 JF |
725 | * Free a vmap area, caller ensuring that the area has been unmapped |
726 | * and flush_cache_vunmap had been called for the correct range | |
727 | * previously. | |
db64fe02 | 728 | */ |
64141da5 | 729 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 730 | { |
80c4bd7a CW |
731 | int nr_lazy; |
732 | ||
733 | nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT, | |
734 | &vmap_lazy_nr); | |
735 | ||
736 | /* After this point, we may free va at any time */ | |
737 | llist_add(&va->purge_list, &vmap_purge_list); | |
738 | ||
739 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 740 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
741 | } |
742 | ||
b29acbdc NP |
743 | /* |
744 | * Free and unmap a vmap area | |
745 | */ | |
746 | static void free_unmap_vmap_area(struct vmap_area *va) | |
747 | { | |
748 | flush_cache_vunmap(va->va_start, va->va_end); | |
c8eef01e | 749 | unmap_vmap_area(va); |
82a2e924 CP |
750 | if (debug_pagealloc_enabled()) |
751 | flush_tlb_kernel_range(va->va_start, va->va_end); | |
752 | ||
c8eef01e | 753 | free_vmap_area_noflush(va); |
b29acbdc NP |
754 | } |
755 | ||
db64fe02 NP |
756 | static struct vmap_area *find_vmap_area(unsigned long addr) |
757 | { | |
758 | struct vmap_area *va; | |
759 | ||
760 | spin_lock(&vmap_area_lock); | |
761 | va = __find_vmap_area(addr); | |
762 | spin_unlock(&vmap_area_lock); | |
763 | ||
764 | return va; | |
765 | } | |
766 | ||
db64fe02 NP |
767 | /*** Per cpu kva allocator ***/ |
768 | ||
769 | /* | |
770 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
771 | * room for at least 16 percpu vmap blocks per CPU. | |
772 | */ | |
773 | /* | |
774 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
775 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
776 | * instead (we just need a rough idea) | |
777 | */ | |
778 | #if BITS_PER_LONG == 32 | |
779 | #define VMALLOC_SPACE (128UL*1024*1024) | |
780 | #else | |
781 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
782 | #endif | |
783 | ||
784 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
785 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
786 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
787 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
788 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
789 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
790 | #define VMAP_BBMAP_BITS \ |
791 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
792 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
793 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
794 | |
795 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
796 | ||
9b463334 JF |
797 | static bool vmap_initialized __read_mostly = false; |
798 | ||
db64fe02 NP |
799 | struct vmap_block_queue { |
800 | spinlock_t lock; | |
801 | struct list_head free; | |
db64fe02 NP |
802 | }; |
803 | ||
804 | struct vmap_block { | |
805 | spinlock_t lock; | |
806 | struct vmap_area *va; | |
db64fe02 | 807 | unsigned long free, dirty; |
7d61bfe8 | 808 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
809 | struct list_head free_list; |
810 | struct rcu_head rcu_head; | |
02b709df | 811 | struct list_head purge; |
db64fe02 NP |
812 | }; |
813 | ||
814 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
815 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
816 | ||
817 | /* | |
818 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
819 | * in the free path. Could get rid of this if we change the API to return a | |
820 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
821 | */ | |
822 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
823 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
824 | ||
825 | /* | |
826 | * We should probably have a fallback mechanism to allocate virtual memory | |
827 | * out of partially filled vmap blocks. However vmap block sizing should be | |
828 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
829 | * big problem. | |
830 | */ | |
831 | ||
832 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
833 | { | |
834 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
835 | addr /= VMAP_BLOCK_SIZE; | |
836 | return addr; | |
837 | } | |
838 | ||
cf725ce2 RP |
839 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
840 | { | |
841 | unsigned long addr; | |
842 | ||
843 | addr = va_start + (pages_off << PAGE_SHIFT); | |
844 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
845 | return (void *)addr; | |
846 | } | |
847 | ||
848 | /** | |
849 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
850 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
851 | * @order: how many 2^order pages should be occupied in newly allocated block | |
852 | * @gfp_mask: flags for the page level allocator | |
853 | * | |
a862f68a | 854 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
855 | */ |
856 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
857 | { |
858 | struct vmap_block_queue *vbq; | |
859 | struct vmap_block *vb; | |
860 | struct vmap_area *va; | |
861 | unsigned long vb_idx; | |
862 | int node, err; | |
cf725ce2 | 863 | void *vaddr; |
db64fe02 NP |
864 | |
865 | node = numa_node_id(); | |
866 | ||
867 | vb = kmalloc_node(sizeof(struct vmap_block), | |
868 | gfp_mask & GFP_RECLAIM_MASK, node); | |
869 | if (unlikely(!vb)) | |
870 | return ERR_PTR(-ENOMEM); | |
871 | ||
872 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
873 | VMALLOC_START, VMALLOC_END, | |
874 | node, gfp_mask); | |
ddf9c6d4 | 875 | if (IS_ERR(va)) { |
db64fe02 | 876 | kfree(vb); |
e7d86340 | 877 | return ERR_CAST(va); |
db64fe02 NP |
878 | } |
879 | ||
880 | err = radix_tree_preload(gfp_mask); | |
881 | if (unlikely(err)) { | |
882 | kfree(vb); | |
883 | free_vmap_area(va); | |
884 | return ERR_PTR(err); | |
885 | } | |
886 | ||
cf725ce2 | 887 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
888 | spin_lock_init(&vb->lock); |
889 | vb->va = va; | |
cf725ce2 RP |
890 | /* At least something should be left free */ |
891 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
892 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 893 | vb->dirty = 0; |
7d61bfe8 RP |
894 | vb->dirty_min = VMAP_BBMAP_BITS; |
895 | vb->dirty_max = 0; | |
db64fe02 | 896 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
897 | |
898 | vb_idx = addr_to_vb_idx(va->va_start); | |
899 | spin_lock(&vmap_block_tree_lock); | |
900 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
901 | spin_unlock(&vmap_block_tree_lock); | |
902 | BUG_ON(err); | |
903 | radix_tree_preload_end(); | |
904 | ||
905 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 906 | spin_lock(&vbq->lock); |
68ac546f | 907 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 908 | spin_unlock(&vbq->lock); |
3f04ba85 | 909 | put_cpu_var(vmap_block_queue); |
db64fe02 | 910 | |
cf725ce2 | 911 | return vaddr; |
db64fe02 NP |
912 | } |
913 | ||
db64fe02 NP |
914 | static void free_vmap_block(struct vmap_block *vb) |
915 | { | |
916 | struct vmap_block *tmp; | |
917 | unsigned long vb_idx; | |
918 | ||
db64fe02 NP |
919 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
920 | spin_lock(&vmap_block_tree_lock); | |
921 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
922 | spin_unlock(&vmap_block_tree_lock); | |
923 | BUG_ON(tmp != vb); | |
924 | ||
64141da5 | 925 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 926 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
927 | } |
928 | ||
02b709df NP |
929 | static void purge_fragmented_blocks(int cpu) |
930 | { | |
931 | LIST_HEAD(purge); | |
932 | struct vmap_block *vb; | |
933 | struct vmap_block *n_vb; | |
934 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
935 | ||
936 | rcu_read_lock(); | |
937 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
938 | ||
939 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
940 | continue; | |
941 | ||
942 | spin_lock(&vb->lock); | |
943 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
944 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
945 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
946 | vb->dirty_min = 0; |
947 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
948 | spin_lock(&vbq->lock); |
949 | list_del_rcu(&vb->free_list); | |
950 | spin_unlock(&vbq->lock); | |
951 | spin_unlock(&vb->lock); | |
952 | list_add_tail(&vb->purge, &purge); | |
953 | } else | |
954 | spin_unlock(&vb->lock); | |
955 | } | |
956 | rcu_read_unlock(); | |
957 | ||
958 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
959 | list_del(&vb->purge); | |
960 | free_vmap_block(vb); | |
961 | } | |
962 | } | |
963 | ||
02b709df NP |
964 | static void purge_fragmented_blocks_allcpus(void) |
965 | { | |
966 | int cpu; | |
967 | ||
968 | for_each_possible_cpu(cpu) | |
969 | purge_fragmented_blocks(cpu); | |
970 | } | |
971 | ||
db64fe02 NP |
972 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
973 | { | |
974 | struct vmap_block_queue *vbq; | |
975 | struct vmap_block *vb; | |
cf725ce2 | 976 | void *vaddr = NULL; |
db64fe02 NP |
977 | unsigned int order; |
978 | ||
891c49ab | 979 | BUG_ON(offset_in_page(size)); |
db64fe02 | 980 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
981 | if (WARN_ON(size == 0)) { |
982 | /* | |
983 | * Allocating 0 bytes isn't what caller wants since | |
984 | * get_order(0) returns funny result. Just warn and terminate | |
985 | * early. | |
986 | */ | |
987 | return NULL; | |
988 | } | |
db64fe02 NP |
989 | order = get_order(size); |
990 | ||
db64fe02 NP |
991 | rcu_read_lock(); |
992 | vbq = &get_cpu_var(vmap_block_queue); | |
993 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 994 | unsigned long pages_off; |
db64fe02 NP |
995 | |
996 | spin_lock(&vb->lock); | |
cf725ce2 RP |
997 | if (vb->free < (1UL << order)) { |
998 | spin_unlock(&vb->lock); | |
999 | continue; | |
1000 | } | |
02b709df | 1001 | |
cf725ce2 RP |
1002 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1003 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1004 | vb->free -= 1UL << order; |
1005 | if (vb->free == 0) { | |
1006 | spin_lock(&vbq->lock); | |
1007 | list_del_rcu(&vb->free_list); | |
1008 | spin_unlock(&vbq->lock); | |
1009 | } | |
cf725ce2 | 1010 | |
02b709df NP |
1011 | spin_unlock(&vb->lock); |
1012 | break; | |
db64fe02 | 1013 | } |
02b709df | 1014 | |
3f04ba85 | 1015 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1016 | rcu_read_unlock(); |
1017 | ||
cf725ce2 RP |
1018 | /* Allocate new block if nothing was found */ |
1019 | if (!vaddr) | |
1020 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1021 | |
cf725ce2 | 1022 | return vaddr; |
db64fe02 NP |
1023 | } |
1024 | ||
1025 | static void vb_free(const void *addr, unsigned long size) | |
1026 | { | |
1027 | unsigned long offset; | |
1028 | unsigned long vb_idx; | |
1029 | unsigned int order; | |
1030 | struct vmap_block *vb; | |
1031 | ||
891c49ab | 1032 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1033 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc NP |
1034 | |
1035 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
1036 | ||
db64fe02 NP |
1037 | order = get_order(size); |
1038 | ||
1039 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
7d61bfe8 | 1040 | offset >>= PAGE_SHIFT; |
db64fe02 NP |
1041 | |
1042 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
1043 | rcu_read_lock(); | |
1044 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1045 | rcu_read_unlock(); | |
1046 | BUG_ON(!vb); | |
1047 | ||
64141da5 JF |
1048 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
1049 | ||
82a2e924 CP |
1050 | if (debug_pagealloc_enabled()) |
1051 | flush_tlb_kernel_range((unsigned long)addr, | |
1052 | (unsigned long)addr + size); | |
1053 | ||
db64fe02 | 1054 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1055 | |
1056 | /* Expand dirty range */ | |
1057 | vb->dirty_min = min(vb->dirty_min, offset); | |
1058 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1059 | |
db64fe02 NP |
1060 | vb->dirty += 1UL << order; |
1061 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1062 | BUG_ON(vb->free); |
db64fe02 NP |
1063 | spin_unlock(&vb->lock); |
1064 | free_vmap_block(vb); | |
1065 | } else | |
1066 | spin_unlock(&vb->lock); | |
1067 | } | |
1068 | ||
868b104d | 1069 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1070 | { |
db64fe02 | 1071 | int cpu; |
db64fe02 | 1072 | |
9b463334 JF |
1073 | if (unlikely(!vmap_initialized)) |
1074 | return; | |
1075 | ||
5803ed29 CH |
1076 | might_sleep(); |
1077 | ||
db64fe02 NP |
1078 | for_each_possible_cpu(cpu) { |
1079 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1080 | struct vmap_block *vb; | |
1081 | ||
1082 | rcu_read_lock(); | |
1083 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1084 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1085 | if (vb->dirty) { |
1086 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1087 | unsigned long s, e; |
b136be5e | 1088 | |
7d61bfe8 RP |
1089 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1090 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1091 | |
7d61bfe8 RP |
1092 | start = min(s, start); |
1093 | end = max(e, end); | |
db64fe02 | 1094 | |
7d61bfe8 | 1095 | flush = 1; |
db64fe02 NP |
1096 | } |
1097 | spin_unlock(&vb->lock); | |
1098 | } | |
1099 | rcu_read_unlock(); | |
1100 | } | |
1101 | ||
f9e09977 | 1102 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1103 | purge_fragmented_blocks_allcpus(); |
1104 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1105 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1106 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 1107 | } |
868b104d RE |
1108 | |
1109 | /** | |
1110 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1111 | * | |
1112 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1113 | * to amortize TLB flushing overheads. What this means is that any page you | |
1114 | * have now, may, in a former life, have been mapped into kernel virtual | |
1115 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1116 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1117 | * | |
1118 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1119 | * be sure that none of the pages we have control over will have any aliases | |
1120 | * from the vmap layer. | |
1121 | */ | |
1122 | void vm_unmap_aliases(void) | |
1123 | { | |
1124 | unsigned long start = ULONG_MAX, end = 0; | |
1125 | int flush = 0; | |
1126 | ||
1127 | _vm_unmap_aliases(start, end, flush); | |
1128 | } | |
db64fe02 NP |
1129 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
1130 | ||
1131 | /** | |
1132 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1133 | * @mem: the pointer returned by vm_map_ram | |
1134 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1135 | */ | |
1136 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1137 | { | |
65ee03c4 | 1138 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1139 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1140 | struct vmap_area *va; |
db64fe02 | 1141 | |
5803ed29 | 1142 | might_sleep(); |
db64fe02 NP |
1143 | BUG_ON(!addr); |
1144 | BUG_ON(addr < VMALLOC_START); | |
1145 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1146 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 1147 | |
9c3acf60 | 1148 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 1149 | debug_check_no_locks_freed(mem, size); |
db64fe02 | 1150 | vb_free(mem, size); |
9c3acf60 CH |
1151 | return; |
1152 | } | |
1153 | ||
1154 | va = find_vmap_area(addr); | |
1155 | BUG_ON(!va); | |
05e3ff95 CP |
1156 | debug_check_no_locks_freed((void *)va->va_start, |
1157 | (va->va_end - va->va_start)); | |
9c3acf60 | 1158 | free_unmap_vmap_area(va); |
db64fe02 NP |
1159 | } |
1160 | EXPORT_SYMBOL(vm_unmap_ram); | |
1161 | ||
1162 | /** | |
1163 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1164 | * @pages: an array of pointers to the pages to be mapped | |
1165 | * @count: number of pages | |
1166 | * @node: prefer to allocate data structures on this node | |
1167 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1168 | * |
36437638 GK |
1169 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1170 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1171 | * objects with vm_map_ram(), it could consume lots of address space through | |
1172 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1173 | * the end. Please use this function for short-lived objects. | |
1174 | * | |
e99c97ad | 1175 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1176 | */ |
1177 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1178 | { | |
65ee03c4 | 1179 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1180 | unsigned long addr; |
1181 | void *mem; | |
1182 | ||
1183 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1184 | mem = vb_alloc(size, GFP_KERNEL); | |
1185 | if (IS_ERR(mem)) | |
1186 | return NULL; | |
1187 | addr = (unsigned long)mem; | |
1188 | } else { | |
1189 | struct vmap_area *va; | |
1190 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1191 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1192 | if (IS_ERR(va)) | |
1193 | return NULL; | |
1194 | ||
1195 | addr = va->va_start; | |
1196 | mem = (void *)addr; | |
1197 | } | |
1198 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1199 | vm_unmap_ram(mem, count); | |
1200 | return NULL; | |
1201 | } | |
1202 | return mem; | |
1203 | } | |
1204 | EXPORT_SYMBOL(vm_map_ram); | |
1205 | ||
4341fa45 | 1206 | static struct vm_struct *vmlist __initdata; |
92eac168 | 1207 | |
be9b7335 NP |
1208 | /** |
1209 | * vm_area_add_early - add vmap area early during boot | |
1210 | * @vm: vm_struct to add | |
1211 | * | |
1212 | * This function is used to add fixed kernel vm area to vmlist before | |
1213 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1214 | * should contain proper values and the other fields should be zero. | |
1215 | * | |
1216 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1217 | */ | |
1218 | void __init vm_area_add_early(struct vm_struct *vm) | |
1219 | { | |
1220 | struct vm_struct *tmp, **p; | |
1221 | ||
1222 | BUG_ON(vmap_initialized); | |
1223 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1224 | if (tmp->addr >= vm->addr) { | |
1225 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1226 | break; | |
1227 | } else | |
1228 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1229 | } | |
1230 | vm->next = *p; | |
1231 | *p = vm; | |
1232 | } | |
1233 | ||
f0aa6617 TH |
1234 | /** |
1235 | * vm_area_register_early - register vmap area early during boot | |
1236 | * @vm: vm_struct to register | |
c0c0a293 | 1237 | * @align: requested alignment |
f0aa6617 TH |
1238 | * |
1239 | * This function is used to register kernel vm area before | |
1240 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1241 | * proper values on entry and other fields should be zero. On return, | |
1242 | * vm->addr contains the allocated address. | |
1243 | * | |
1244 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1245 | */ | |
c0c0a293 | 1246 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1247 | { |
1248 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1249 | unsigned long addr; |
1250 | ||
1251 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1252 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1253 | |
c0c0a293 | 1254 | vm->addr = (void *)addr; |
f0aa6617 | 1255 | |
be9b7335 | 1256 | vm_area_add_early(vm); |
f0aa6617 TH |
1257 | } |
1258 | ||
db64fe02 NP |
1259 | void __init vmalloc_init(void) |
1260 | { | |
822c18f2 IK |
1261 | struct vmap_area *va; |
1262 | struct vm_struct *tmp; | |
db64fe02 NP |
1263 | int i; |
1264 | ||
1265 | for_each_possible_cpu(i) { | |
1266 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1267 | struct vfree_deferred *p; |
db64fe02 NP |
1268 | |
1269 | vbq = &per_cpu(vmap_block_queue, i); | |
1270 | spin_lock_init(&vbq->lock); | |
1271 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1272 | p = &per_cpu(vfree_deferred, i); |
1273 | init_llist_head(&p->list); | |
1274 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1275 | } |
9b463334 | 1276 | |
822c18f2 IK |
1277 | /* Import existing vmlist entries. */ |
1278 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1279 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
dbda591d | 1280 | va->flags = VM_VM_AREA; |
822c18f2 IK |
1281 | va->va_start = (unsigned long)tmp->addr; |
1282 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1283 | va->vm = tmp; |
822c18f2 IK |
1284 | __insert_vmap_area(va); |
1285 | } | |
ca23e405 TH |
1286 | |
1287 | vmap_area_pcpu_hole = VMALLOC_END; | |
1288 | ||
9b463334 | 1289 | vmap_initialized = true; |
db64fe02 NP |
1290 | } |
1291 | ||
8fc48985 TH |
1292 | /** |
1293 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1294 | * @addr: start of the VM area to map | |
1295 | * @size: size of the VM area to map | |
1296 | * @prot: page protection flags to use | |
1297 | * @pages: pages to map | |
1298 | * | |
1299 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1300 | * specify should have been allocated using get_vm_area() and its | |
1301 | * friends. | |
1302 | * | |
1303 | * NOTE: | |
1304 | * This function does NOT do any cache flushing. The caller is | |
1305 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1306 | * before calling this function. | |
1307 | * | |
1308 | * RETURNS: | |
1309 | * The number of pages mapped on success, -errno on failure. | |
1310 | */ | |
1311 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1312 | pgprot_t prot, struct page **pages) | |
1313 | { | |
1314 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1315 | } | |
1316 | ||
1317 | /** | |
1318 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1319 | * @addr: start of the VM area to unmap | |
1320 | * @size: size of the VM area to unmap | |
1321 | * | |
1322 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1323 | * specify should have been allocated using get_vm_area() and its | |
1324 | * friends. | |
1325 | * | |
1326 | * NOTE: | |
1327 | * This function does NOT do any cache flushing. The caller is | |
1328 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1329 | * before calling this function and flush_tlb_kernel_range() after. | |
1330 | */ | |
1331 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1332 | { | |
1333 | vunmap_page_range(addr, addr + size); | |
1334 | } | |
81e88fdc | 1335 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1336 | |
1337 | /** | |
1338 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1339 | * @addr: start of the VM area to unmap | |
1340 | * @size: size of the VM area to unmap | |
1341 | * | |
1342 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1343 | * the unmapping and tlb after. | |
1344 | */ | |
db64fe02 NP |
1345 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1346 | { | |
1347 | unsigned long end = addr + size; | |
f6fcba70 TH |
1348 | |
1349 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1350 | vunmap_page_range(addr, end); |
1351 | flush_tlb_kernel_range(addr, end); | |
1352 | } | |
93ef6d6c | 1353 | EXPORT_SYMBOL_GPL(unmap_kernel_range); |
db64fe02 | 1354 | |
f6f8ed47 | 1355 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
1356 | { |
1357 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 1358 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
1359 | int err; |
1360 | ||
f6f8ed47 | 1361 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 1362 | |
f6f8ed47 | 1363 | return err > 0 ? 0 : err; |
db64fe02 NP |
1364 | } |
1365 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1366 | ||
f5252e00 | 1367 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1368 | unsigned long flags, const void *caller) |
cf88c790 | 1369 | { |
c69480ad | 1370 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
1371 | vm->flags = flags; |
1372 | vm->addr = (void *)va->va_start; | |
1373 | vm->size = va->va_end - va->va_start; | |
1374 | vm->caller = caller; | |
db1aecaf | 1375 | va->vm = vm; |
cf88c790 | 1376 | va->flags |= VM_VM_AREA; |
c69480ad | 1377 | spin_unlock(&vmap_area_lock); |
f5252e00 | 1378 | } |
cf88c790 | 1379 | |
20fc02b4 | 1380 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 1381 | { |
d4033afd | 1382 | /* |
20fc02b4 | 1383 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
1384 | * we should make sure that vm has proper values. |
1385 | * Pair with smp_rmb() in show_numa_info(). | |
1386 | */ | |
1387 | smp_wmb(); | |
20fc02b4 | 1388 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
1389 | } |
1390 | ||
db64fe02 | 1391 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 1392 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 1393 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 1394 | { |
0006526d | 1395 | struct vmap_area *va; |
db64fe02 | 1396 | struct vm_struct *area; |
1da177e4 | 1397 | |
52fd24ca | 1398 | BUG_ON(in_interrupt()); |
1da177e4 | 1399 | size = PAGE_ALIGN(size); |
31be8309 OH |
1400 | if (unlikely(!size)) |
1401 | return NULL; | |
1da177e4 | 1402 | |
252e5c6e | 1403 | if (flags & VM_IOREMAP) |
1404 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
1405 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
1406 | ||
cf88c790 | 1407 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1408 | if (unlikely(!area)) |
1409 | return NULL; | |
1410 | ||
71394fe5 AR |
1411 | if (!(flags & VM_NO_GUARD)) |
1412 | size += PAGE_SIZE; | |
1da177e4 | 1413 | |
db64fe02 NP |
1414 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1415 | if (IS_ERR(va)) { | |
1416 | kfree(area); | |
1417 | return NULL; | |
1da177e4 | 1418 | } |
1da177e4 | 1419 | |
d82b1d85 | 1420 | setup_vmalloc_vm(area, va, flags, caller); |
f5252e00 | 1421 | |
1da177e4 | 1422 | return area; |
1da177e4 LT |
1423 | } |
1424 | ||
930fc45a CL |
1425 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1426 | unsigned long start, unsigned long end) | |
1427 | { | |
00ef2d2f DR |
1428 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1429 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 1430 | } |
5992b6da | 1431 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1432 | |
c2968612 BH |
1433 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1434 | unsigned long start, unsigned long end, | |
5e6cafc8 | 1435 | const void *caller) |
c2968612 | 1436 | { |
00ef2d2f DR |
1437 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1438 | GFP_KERNEL, caller); | |
c2968612 BH |
1439 | } |
1440 | ||
1da177e4 | 1441 | /** |
92eac168 MR |
1442 | * get_vm_area - reserve a contiguous kernel virtual area |
1443 | * @size: size of the area | |
1444 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 1445 | * |
92eac168 MR |
1446 | * Search an area of @size in the kernel virtual mapping area, |
1447 | * and reserved it for out purposes. Returns the area descriptor | |
1448 | * on success or %NULL on failure. | |
a862f68a MR |
1449 | * |
1450 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
1451 | */ |
1452 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1453 | { | |
2dca6999 | 1454 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
1455 | NUMA_NO_NODE, GFP_KERNEL, |
1456 | __builtin_return_address(0)); | |
23016969 CL |
1457 | } |
1458 | ||
1459 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 1460 | const void *caller) |
23016969 | 1461 | { |
2dca6999 | 1462 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 1463 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
1464 | } |
1465 | ||
e9da6e99 | 1466 | /** |
92eac168 MR |
1467 | * find_vm_area - find a continuous kernel virtual area |
1468 | * @addr: base address | |
e9da6e99 | 1469 | * |
92eac168 MR |
1470 | * Search for the kernel VM area starting at @addr, and return it. |
1471 | * It is up to the caller to do all required locking to keep the returned | |
1472 | * pointer valid. | |
a862f68a MR |
1473 | * |
1474 | * Return: pointer to the found area or %NULL on faulure | |
e9da6e99 MS |
1475 | */ |
1476 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 1477 | { |
db64fe02 | 1478 | struct vmap_area *va; |
83342314 | 1479 | |
db64fe02 NP |
1480 | va = find_vmap_area((unsigned long)addr); |
1481 | if (va && va->flags & VM_VM_AREA) | |
db1aecaf | 1482 | return va->vm; |
1da177e4 | 1483 | |
1da177e4 | 1484 | return NULL; |
1da177e4 LT |
1485 | } |
1486 | ||
7856dfeb | 1487 | /** |
92eac168 MR |
1488 | * remove_vm_area - find and remove a continuous kernel virtual area |
1489 | * @addr: base address | |
7856dfeb | 1490 | * |
92eac168 MR |
1491 | * Search for the kernel VM area starting at @addr, and remove it. |
1492 | * This function returns the found VM area, but using it is NOT safe | |
1493 | * on SMP machines, except for its size or flags. | |
a862f68a MR |
1494 | * |
1495 | * Return: pointer to the found area or %NULL on faulure | |
7856dfeb | 1496 | */ |
b3bdda02 | 1497 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1498 | { |
db64fe02 NP |
1499 | struct vmap_area *va; |
1500 | ||
5803ed29 CH |
1501 | might_sleep(); |
1502 | ||
db64fe02 NP |
1503 | va = find_vmap_area((unsigned long)addr); |
1504 | if (va && va->flags & VM_VM_AREA) { | |
db1aecaf | 1505 | struct vm_struct *vm = va->vm; |
f5252e00 | 1506 | |
c69480ad JK |
1507 | spin_lock(&vmap_area_lock); |
1508 | va->vm = NULL; | |
1509 | va->flags &= ~VM_VM_AREA; | |
78c72746 | 1510 | va->flags |= VM_LAZY_FREE; |
c69480ad JK |
1511 | spin_unlock(&vmap_area_lock); |
1512 | ||
a5af5aa8 | 1513 | kasan_free_shadow(vm); |
dd32c279 | 1514 | free_unmap_vmap_area(va); |
dd32c279 | 1515 | |
db64fe02 NP |
1516 | return vm; |
1517 | } | |
1518 | return NULL; | |
7856dfeb AK |
1519 | } |
1520 | ||
868b104d RE |
1521 | static inline void set_area_direct_map(const struct vm_struct *area, |
1522 | int (*set_direct_map)(struct page *page)) | |
1523 | { | |
1524 | int i; | |
1525 | ||
1526 | for (i = 0; i < area->nr_pages; i++) | |
1527 | if (page_address(area->pages[i])) | |
1528 | set_direct_map(area->pages[i]); | |
1529 | } | |
1530 | ||
1531 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
1532 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
1533 | { | |
1534 | unsigned long addr = (unsigned long)area->addr; | |
1535 | unsigned long start = ULONG_MAX, end = 0; | |
1536 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
1537 | int i; | |
1538 | ||
1539 | /* | |
1540 | * The below block can be removed when all architectures that have | |
1541 | * direct map permissions also have set_direct_map_() implementations. | |
1542 | * This is concerned with resetting the direct map any an vm alias with | |
1543 | * execute permissions, without leaving a RW+X window. | |
1544 | */ | |
1545 | if (flush_reset && !IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) { | |
1546 | set_memory_nx(addr, area->nr_pages); | |
1547 | set_memory_rw(addr, area->nr_pages); | |
1548 | } | |
1549 | ||
1550 | remove_vm_area(area->addr); | |
1551 | ||
1552 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
1553 | if (!flush_reset) | |
1554 | return; | |
1555 | ||
1556 | /* | |
1557 | * If not deallocating pages, just do the flush of the VM area and | |
1558 | * return. | |
1559 | */ | |
1560 | if (!deallocate_pages) { | |
1561 | vm_unmap_aliases(); | |
1562 | return; | |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * If execution gets here, flush the vm mapping and reset the direct | |
1567 | * map. Find the start and end range of the direct mappings to make sure | |
1568 | * the vm_unmap_aliases() flush includes the direct map. | |
1569 | */ | |
1570 | for (i = 0; i < area->nr_pages; i++) { | |
1571 | if (page_address(area->pages[i])) { | |
1572 | start = min(addr, start); | |
1573 | end = max(addr, end); | |
1574 | } | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * Set direct map to something invalid so that it won't be cached if | |
1579 | * there are any accesses after the TLB flush, then flush the TLB and | |
1580 | * reset the direct map permissions to the default. | |
1581 | */ | |
1582 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
1583 | _vm_unmap_aliases(start, end, 1); | |
1584 | set_area_direct_map(area, set_direct_map_default_noflush); | |
1585 | } | |
1586 | ||
b3bdda02 | 1587 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1588 | { |
1589 | struct vm_struct *area; | |
1590 | ||
1591 | if (!addr) | |
1592 | return; | |
1593 | ||
e69e9d4a | 1594 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 1595 | addr)) |
1da177e4 | 1596 | return; |
1da177e4 | 1597 | |
6ade2032 | 1598 | area = find_vm_area(addr); |
1da177e4 | 1599 | if (unlikely(!area)) { |
4c8573e2 | 1600 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1601 | addr); |
1da177e4 LT |
1602 | return; |
1603 | } | |
1604 | ||
05e3ff95 CP |
1605 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
1606 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 1607 | |
868b104d RE |
1608 | vm_remove_mappings(area, deallocate_pages); |
1609 | ||
1da177e4 LT |
1610 | if (deallocate_pages) { |
1611 | int i; | |
1612 | ||
1613 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1614 | struct page *page = area->pages[i]; |
1615 | ||
1616 | BUG_ON(!page); | |
4949148a | 1617 | __free_pages(page, 0); |
1da177e4 LT |
1618 | } |
1619 | ||
244d63ee | 1620 | kvfree(area->pages); |
1da177e4 LT |
1621 | } |
1622 | ||
1623 | kfree(area); | |
1624 | return; | |
1625 | } | |
bf22e37a AR |
1626 | |
1627 | static inline void __vfree_deferred(const void *addr) | |
1628 | { | |
1629 | /* | |
1630 | * Use raw_cpu_ptr() because this can be called from preemptible | |
1631 | * context. Preemption is absolutely fine here, because the llist_add() | |
1632 | * implementation is lockless, so it works even if we are adding to | |
1633 | * nother cpu's list. schedule_work() should be fine with this too. | |
1634 | */ | |
1635 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
1636 | ||
1637 | if (llist_add((struct llist_node *)addr, &p->list)) | |
1638 | schedule_work(&p->wq); | |
1639 | } | |
1640 | ||
1641 | /** | |
92eac168 MR |
1642 | * vfree_atomic - release memory allocated by vmalloc() |
1643 | * @addr: memory base address | |
bf22e37a | 1644 | * |
92eac168 MR |
1645 | * This one is just like vfree() but can be called in any atomic context |
1646 | * except NMIs. | |
bf22e37a AR |
1647 | */ |
1648 | void vfree_atomic(const void *addr) | |
1649 | { | |
1650 | BUG_ON(in_nmi()); | |
1651 | ||
1652 | kmemleak_free(addr); | |
1653 | ||
1654 | if (!addr) | |
1655 | return; | |
1656 | __vfree_deferred(addr); | |
1657 | } | |
1658 | ||
c67dc624 RP |
1659 | static void __vfree(const void *addr) |
1660 | { | |
1661 | if (unlikely(in_interrupt())) | |
1662 | __vfree_deferred(addr); | |
1663 | else | |
1664 | __vunmap(addr, 1); | |
1665 | } | |
1666 | ||
1da177e4 | 1667 | /** |
92eac168 MR |
1668 | * vfree - release memory allocated by vmalloc() |
1669 | * @addr: memory base address | |
1da177e4 | 1670 | * |
92eac168 MR |
1671 | * Free the virtually continuous memory area starting at @addr, as |
1672 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is | |
1673 | * NULL, no operation is performed. | |
1da177e4 | 1674 | * |
92eac168 MR |
1675 | * Must not be called in NMI context (strictly speaking, only if we don't |
1676 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
1677 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 1678 | * |
92eac168 | 1679 | * May sleep if called *not* from interrupt context. |
3ca4ea3a | 1680 | * |
92eac168 | 1681 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 1682 | */ |
b3bdda02 | 1683 | void vfree(const void *addr) |
1da177e4 | 1684 | { |
32fcfd40 | 1685 | BUG_ON(in_nmi()); |
89219d37 CM |
1686 | |
1687 | kmemleak_free(addr); | |
1688 | ||
a8dda165 AR |
1689 | might_sleep_if(!in_interrupt()); |
1690 | ||
32fcfd40 AV |
1691 | if (!addr) |
1692 | return; | |
c67dc624 RP |
1693 | |
1694 | __vfree(addr); | |
1da177e4 | 1695 | } |
1da177e4 LT |
1696 | EXPORT_SYMBOL(vfree); |
1697 | ||
1698 | /** | |
92eac168 MR |
1699 | * vunmap - release virtual mapping obtained by vmap() |
1700 | * @addr: memory base address | |
1da177e4 | 1701 | * |
92eac168 MR |
1702 | * Free the virtually contiguous memory area starting at @addr, |
1703 | * which was created from the page array passed to vmap(). | |
1da177e4 | 1704 | * |
92eac168 | 1705 | * Must not be called in interrupt context. |
1da177e4 | 1706 | */ |
b3bdda02 | 1707 | void vunmap(const void *addr) |
1da177e4 LT |
1708 | { |
1709 | BUG_ON(in_interrupt()); | |
34754b69 | 1710 | might_sleep(); |
32fcfd40 AV |
1711 | if (addr) |
1712 | __vunmap(addr, 0); | |
1da177e4 | 1713 | } |
1da177e4 LT |
1714 | EXPORT_SYMBOL(vunmap); |
1715 | ||
1716 | /** | |
92eac168 MR |
1717 | * vmap - map an array of pages into virtually contiguous space |
1718 | * @pages: array of page pointers | |
1719 | * @count: number of pages to map | |
1720 | * @flags: vm_area->flags | |
1721 | * @prot: page protection for the mapping | |
1722 | * | |
1723 | * Maps @count pages from @pages into contiguous kernel virtual | |
1724 | * space. | |
a862f68a MR |
1725 | * |
1726 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
1727 | */ |
1728 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 1729 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
1730 | { |
1731 | struct vm_struct *area; | |
65ee03c4 | 1732 | unsigned long size; /* In bytes */ |
1da177e4 | 1733 | |
34754b69 PZ |
1734 | might_sleep(); |
1735 | ||
ca79b0c2 | 1736 | if (count > totalram_pages()) |
1da177e4 LT |
1737 | return NULL; |
1738 | ||
65ee03c4 GJM |
1739 | size = (unsigned long)count << PAGE_SHIFT; |
1740 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
1741 | if (!area) |
1742 | return NULL; | |
23016969 | 1743 | |
f6f8ed47 | 1744 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
1745 | vunmap(area->addr); |
1746 | return NULL; | |
1747 | } | |
1748 | ||
1749 | return area->addr; | |
1750 | } | |
1da177e4 LT |
1751 | EXPORT_SYMBOL(vmap); |
1752 | ||
8594a21c MH |
1753 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
1754 | gfp_t gfp_mask, pgprot_t prot, | |
1755 | int node, const void *caller); | |
e31d9eb5 | 1756 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 1757 | pgprot_t prot, int node) |
1da177e4 LT |
1758 | { |
1759 | struct page **pages; | |
1760 | unsigned int nr_pages, array_size, i; | |
930f036b | 1761 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
704b862f LA |
1762 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; |
1763 | const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ? | |
1764 | 0 : | |
1765 | __GFP_HIGHMEM; | |
1da177e4 | 1766 | |
762216ab | 1767 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
1768 | array_size = (nr_pages * sizeof(struct page *)); |
1769 | ||
1770 | area->nr_pages = nr_pages; | |
1771 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1772 | if (array_size > PAGE_SIZE) { |
704b862f | 1773 | pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask, |
3722e13c | 1774 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 1775 | } else { |
976d6dfb | 1776 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 1777 | } |
1da177e4 LT |
1778 | area->pages = pages; |
1779 | if (!area->pages) { | |
1780 | remove_vm_area(area->addr); | |
1781 | kfree(area); | |
1782 | return NULL; | |
1783 | } | |
1da177e4 LT |
1784 | |
1785 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1786 | struct page *page; |
1787 | ||
4b90951c | 1788 | if (node == NUMA_NO_NODE) |
704b862f | 1789 | page = alloc_page(alloc_mask|highmem_mask); |
930fc45a | 1790 | else |
704b862f | 1791 | page = alloc_pages_node(node, alloc_mask|highmem_mask, 0); |
bf53d6f8 CL |
1792 | |
1793 | if (unlikely(!page)) { | |
1da177e4 LT |
1794 | /* Successfully allocated i pages, free them in __vunmap() */ |
1795 | area->nr_pages = i; | |
1796 | goto fail; | |
1797 | } | |
bf53d6f8 | 1798 | area->pages[i] = page; |
704b862f | 1799 | if (gfpflags_allow_blocking(gfp_mask|highmem_mask)) |
660654f9 | 1800 | cond_resched(); |
1da177e4 LT |
1801 | } |
1802 | ||
f6f8ed47 | 1803 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
1804 | goto fail; |
1805 | return area->addr; | |
1806 | ||
1807 | fail: | |
a8e99259 | 1808 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 1809 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 1810 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 1811 | __vfree(area->addr); |
1da177e4 LT |
1812 | return NULL; |
1813 | } | |
1814 | ||
1815 | /** | |
92eac168 MR |
1816 | * __vmalloc_node_range - allocate virtually contiguous memory |
1817 | * @size: allocation size | |
1818 | * @align: desired alignment | |
1819 | * @start: vm area range start | |
1820 | * @end: vm area range end | |
1821 | * @gfp_mask: flags for the page level allocator | |
1822 | * @prot: protection mask for the allocated pages | |
1823 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
1824 | * @node: node to use for allocation or NUMA_NO_NODE | |
1825 | * @caller: caller's return address | |
1826 | * | |
1827 | * Allocate enough pages to cover @size from the page level | |
1828 | * allocator with @gfp_mask flags. Map them into contiguous | |
1829 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
1830 | * |
1831 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 1832 | */ |
d0a21265 DR |
1833 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
1834 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
1835 | pgprot_t prot, unsigned long vm_flags, int node, |
1836 | const void *caller) | |
1da177e4 LT |
1837 | { |
1838 | struct vm_struct *area; | |
89219d37 CM |
1839 | void *addr; |
1840 | unsigned long real_size = size; | |
1da177e4 LT |
1841 | |
1842 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 1843 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 1844 | goto fail; |
1da177e4 | 1845 | |
cb9e3c29 AR |
1846 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | |
1847 | vm_flags, start, end, node, gfp_mask, caller); | |
1da177e4 | 1848 | if (!area) |
de7d2b56 | 1849 | goto fail; |
1da177e4 | 1850 | |
3722e13c | 1851 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 1852 | if (!addr) |
b82225f3 | 1853 | return NULL; |
89219d37 | 1854 | |
f5252e00 | 1855 | /* |
20fc02b4 ZY |
1856 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
1857 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 1858 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 1859 | */ |
20fc02b4 | 1860 | clear_vm_uninitialized_flag(area); |
f5252e00 | 1861 | |
94f4a161 | 1862 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
1863 | |
1864 | return addr; | |
de7d2b56 JP |
1865 | |
1866 | fail: | |
a8e99259 | 1867 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 1868 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 1869 | return NULL; |
1da177e4 LT |
1870 | } |
1871 | ||
153178ed URS |
1872 | /* |
1873 | * This is only for performance analysis of vmalloc and stress purpose. | |
1874 | * It is required by vmalloc test module, therefore do not use it other | |
1875 | * than that. | |
1876 | */ | |
1877 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
1878 | EXPORT_SYMBOL_GPL(__vmalloc_node_range); | |
1879 | #endif | |
1880 | ||
d0a21265 | 1881 | /** |
92eac168 MR |
1882 | * __vmalloc_node - allocate virtually contiguous memory |
1883 | * @size: allocation size | |
1884 | * @align: desired alignment | |
1885 | * @gfp_mask: flags for the page level allocator | |
1886 | * @prot: protection mask for the allocated pages | |
1887 | * @node: node to use for allocation or NUMA_NO_NODE | |
1888 | * @caller: caller's return address | |
a7c3e901 | 1889 | * |
92eac168 MR |
1890 | * Allocate enough pages to cover @size from the page level |
1891 | * allocator with @gfp_mask flags. Map them into contiguous | |
1892 | * kernel virtual space, using a pagetable protection of @prot. | |
a7c3e901 | 1893 | * |
92eac168 MR |
1894 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
1895 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 1896 | * |
92eac168 MR |
1897 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
1898 | * with mm people. | |
a862f68a MR |
1899 | * |
1900 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 1901 | */ |
8594a21c | 1902 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
d0a21265 | 1903 | gfp_t gfp_mask, pgprot_t prot, |
5e6cafc8 | 1904 | int node, const void *caller) |
d0a21265 DR |
1905 | { |
1906 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 1907 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
1908 | } |
1909 | ||
930fc45a CL |
1910 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1911 | { | |
00ef2d2f | 1912 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 1913 | __builtin_return_address(0)); |
930fc45a | 1914 | } |
1da177e4 LT |
1915 | EXPORT_SYMBOL(__vmalloc); |
1916 | ||
8594a21c MH |
1917 | static inline void *__vmalloc_node_flags(unsigned long size, |
1918 | int node, gfp_t flags) | |
1919 | { | |
1920 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
1921 | node, __builtin_return_address(0)); | |
1922 | } | |
1923 | ||
1924 | ||
1925 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags, | |
1926 | void *caller) | |
1927 | { | |
1928 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller); | |
1929 | } | |
1930 | ||
1da177e4 | 1931 | /** |
92eac168 MR |
1932 | * vmalloc - allocate virtually contiguous memory |
1933 | * @size: allocation size | |
1934 | * | |
1935 | * Allocate enough pages to cover @size from the page level | |
1936 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 1937 | * |
92eac168 MR |
1938 | * For tight control over page level allocator and protection flags |
1939 | * use __vmalloc() instead. | |
a862f68a MR |
1940 | * |
1941 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
1942 | */ |
1943 | void *vmalloc(unsigned long size) | |
1944 | { | |
00ef2d2f | 1945 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 1946 | GFP_KERNEL); |
1da177e4 | 1947 | } |
1da177e4 LT |
1948 | EXPORT_SYMBOL(vmalloc); |
1949 | ||
e1ca7788 | 1950 | /** |
92eac168 MR |
1951 | * vzalloc - allocate virtually contiguous memory with zero fill |
1952 | * @size: allocation size | |
1953 | * | |
1954 | * Allocate enough pages to cover @size from the page level | |
1955 | * allocator and map them into contiguous kernel virtual space. | |
1956 | * The memory allocated is set to zero. | |
1957 | * | |
1958 | * For tight control over page level allocator and protection flags | |
1959 | * use __vmalloc() instead. | |
a862f68a MR |
1960 | * |
1961 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
1962 | */ |
1963 | void *vzalloc(unsigned long size) | |
1964 | { | |
00ef2d2f | 1965 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 1966 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
1967 | } |
1968 | EXPORT_SYMBOL(vzalloc); | |
1969 | ||
83342314 | 1970 | /** |
ead04089 REB |
1971 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1972 | * @size: allocation size | |
83342314 | 1973 | * |
ead04089 REB |
1974 | * The resulting memory area is zeroed so it can be mapped to userspace |
1975 | * without leaking data. | |
a862f68a MR |
1976 | * |
1977 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
1978 | */ |
1979 | void *vmalloc_user(unsigned long size) | |
1980 | { | |
bc84c535 RP |
1981 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
1982 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
1983 | VM_USERMAP, NUMA_NO_NODE, | |
1984 | __builtin_return_address(0)); | |
83342314 NP |
1985 | } |
1986 | EXPORT_SYMBOL(vmalloc_user); | |
1987 | ||
930fc45a | 1988 | /** |
92eac168 MR |
1989 | * vmalloc_node - allocate memory on a specific node |
1990 | * @size: allocation size | |
1991 | * @node: numa node | |
930fc45a | 1992 | * |
92eac168 MR |
1993 | * Allocate enough pages to cover @size from the page level |
1994 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 1995 | * |
92eac168 MR |
1996 | * For tight control over page level allocator and protection flags |
1997 | * use __vmalloc() instead. | |
a862f68a MR |
1998 | * |
1999 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2000 | */ |
2001 | void *vmalloc_node(unsigned long size, int node) | |
2002 | { | |
19809c2d | 2003 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL, |
23016969 | 2004 | node, __builtin_return_address(0)); |
930fc45a CL |
2005 | } |
2006 | EXPORT_SYMBOL(vmalloc_node); | |
2007 | ||
e1ca7788 DY |
2008 | /** |
2009 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2010 | * @size: allocation size | |
2011 | * @node: numa node | |
2012 | * | |
2013 | * Allocate enough pages to cover @size from the page level | |
2014 | * allocator and map them into contiguous kernel virtual space. | |
2015 | * The memory allocated is set to zero. | |
2016 | * | |
2017 | * For tight control over page level allocator and protection flags | |
2018 | * use __vmalloc_node() instead. | |
a862f68a MR |
2019 | * |
2020 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2021 | */ |
2022 | void *vzalloc_node(unsigned long size, int node) | |
2023 | { | |
2024 | return __vmalloc_node_flags(size, node, | |
19809c2d | 2025 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2026 | } |
2027 | EXPORT_SYMBOL(vzalloc_node); | |
2028 | ||
1da177e4 | 2029 | /** |
92eac168 MR |
2030 | * vmalloc_exec - allocate virtually contiguous, executable memory |
2031 | * @size: allocation size | |
1da177e4 | 2032 | * |
92eac168 MR |
2033 | * Kernel-internal function to allocate enough pages to cover @size |
2034 | * the page level allocator and map them into contiguous and | |
2035 | * executable kernel virtual space. | |
1da177e4 | 2036 | * |
92eac168 MR |
2037 | * For tight control over page level allocator and protection flags |
2038 | * use __vmalloc() instead. | |
a862f68a MR |
2039 | * |
2040 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 | 2041 | */ |
1da177e4 LT |
2042 | void *vmalloc_exec(unsigned long size) |
2043 | { | |
868b104d RE |
2044 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, |
2045 | GFP_KERNEL, PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS, | |
2046 | NUMA_NO_NODE, __builtin_return_address(0)); | |
1da177e4 LT |
2047 | } |
2048 | ||
0d08e0d3 | 2049 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2050 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2051 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2052 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2053 | #else |
698d0831 MH |
2054 | /* |
2055 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2056 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2057 | */ | |
2058 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2059 | #endif |
2060 | ||
1da177e4 | 2061 | /** |
92eac168 MR |
2062 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2063 | * @size: allocation size | |
1da177e4 | 2064 | * |
92eac168 MR |
2065 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2066 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2067 | * |
2068 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2069 | */ |
2070 | void *vmalloc_32(unsigned long size) | |
2071 | { | |
2dca6999 | 2072 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 2073 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 2074 | } |
1da177e4 LT |
2075 | EXPORT_SYMBOL(vmalloc_32); |
2076 | ||
83342314 | 2077 | /** |
ead04089 | 2078 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2079 | * @size: allocation size |
ead04089 REB |
2080 | * |
2081 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2082 | * mapped to userspace without leaking data. | |
a862f68a MR |
2083 | * |
2084 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2085 | */ |
2086 | void *vmalloc_32_user(unsigned long size) | |
2087 | { | |
bc84c535 RP |
2088 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2089 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2090 | VM_USERMAP, NUMA_NO_NODE, | |
2091 | __builtin_return_address(0)); | |
83342314 NP |
2092 | } |
2093 | EXPORT_SYMBOL(vmalloc_32_user); | |
2094 | ||
d0107eb0 KH |
2095 | /* |
2096 | * small helper routine , copy contents to buf from addr. | |
2097 | * If the page is not present, fill zero. | |
2098 | */ | |
2099 | ||
2100 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2101 | { | |
2102 | struct page *p; | |
2103 | int copied = 0; | |
2104 | ||
2105 | while (count) { | |
2106 | unsigned long offset, length; | |
2107 | ||
891c49ab | 2108 | offset = offset_in_page(addr); |
d0107eb0 KH |
2109 | length = PAGE_SIZE - offset; |
2110 | if (length > count) | |
2111 | length = count; | |
2112 | p = vmalloc_to_page(addr); | |
2113 | /* | |
2114 | * To do safe access to this _mapped_ area, we need | |
2115 | * lock. But adding lock here means that we need to add | |
2116 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2117 | * interface, rarely used. Instead of that, we'll use | |
2118 | * kmap() and get small overhead in this access function. | |
2119 | */ | |
2120 | if (p) { | |
2121 | /* | |
2122 | * we can expect USER0 is not used (see vread/vwrite's | |
2123 | * function description) | |
2124 | */ | |
9b04c5fe | 2125 | void *map = kmap_atomic(p); |
d0107eb0 | 2126 | memcpy(buf, map + offset, length); |
9b04c5fe | 2127 | kunmap_atomic(map); |
d0107eb0 KH |
2128 | } else |
2129 | memset(buf, 0, length); | |
2130 | ||
2131 | addr += length; | |
2132 | buf += length; | |
2133 | copied += length; | |
2134 | count -= length; | |
2135 | } | |
2136 | return copied; | |
2137 | } | |
2138 | ||
2139 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2140 | { | |
2141 | struct page *p; | |
2142 | int copied = 0; | |
2143 | ||
2144 | while (count) { | |
2145 | unsigned long offset, length; | |
2146 | ||
891c49ab | 2147 | offset = offset_in_page(addr); |
d0107eb0 KH |
2148 | length = PAGE_SIZE - offset; |
2149 | if (length > count) | |
2150 | length = count; | |
2151 | p = vmalloc_to_page(addr); | |
2152 | /* | |
2153 | * To do safe access to this _mapped_ area, we need | |
2154 | * lock. But adding lock here means that we need to add | |
2155 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2156 | * interface, rarely used. Instead of that, we'll use | |
2157 | * kmap() and get small overhead in this access function. | |
2158 | */ | |
2159 | if (p) { | |
2160 | /* | |
2161 | * we can expect USER0 is not used (see vread/vwrite's | |
2162 | * function description) | |
2163 | */ | |
9b04c5fe | 2164 | void *map = kmap_atomic(p); |
d0107eb0 | 2165 | memcpy(map + offset, buf, length); |
9b04c5fe | 2166 | kunmap_atomic(map); |
d0107eb0 KH |
2167 | } |
2168 | addr += length; | |
2169 | buf += length; | |
2170 | copied += length; | |
2171 | count -= length; | |
2172 | } | |
2173 | return copied; | |
2174 | } | |
2175 | ||
2176 | /** | |
92eac168 MR |
2177 | * vread() - read vmalloc area in a safe way. |
2178 | * @buf: buffer for reading data | |
2179 | * @addr: vm address. | |
2180 | * @count: number of bytes to be read. | |
2181 | * | |
92eac168 MR |
2182 | * This function checks that addr is a valid vmalloc'ed area, and |
2183 | * copy data from that area to a given buffer. If the given memory range | |
2184 | * of [addr...addr+count) includes some valid address, data is copied to | |
2185 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2186 | * IOREMAP area is treated as memory hole and no copy is done. | |
2187 | * | |
2188 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2189 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2190 | * | |
2191 | * Note: In usual ops, vread() is never necessary because the caller | |
2192 | * should know vmalloc() area is valid and can use memcpy(). | |
2193 | * This is for routines which have to access vmalloc area without | |
2194 | * any informaion, as /dev/kmem. | |
a862f68a MR |
2195 | * |
2196 | * Return: number of bytes for which addr and buf should be increased | |
2197 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
2198 | * include any intersection with valid vmalloc area | |
d0107eb0 | 2199 | */ |
1da177e4 LT |
2200 | long vread(char *buf, char *addr, unsigned long count) |
2201 | { | |
e81ce85f JK |
2202 | struct vmap_area *va; |
2203 | struct vm_struct *vm; | |
1da177e4 | 2204 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2205 | unsigned long buflen = count; |
1da177e4 LT |
2206 | unsigned long n; |
2207 | ||
2208 | /* Don't allow overflow */ | |
2209 | if ((unsigned long) addr + count < count) | |
2210 | count = -(unsigned long) addr; | |
2211 | ||
e81ce85f JK |
2212 | spin_lock(&vmap_area_lock); |
2213 | list_for_each_entry(va, &vmap_area_list, list) { | |
2214 | if (!count) | |
2215 | break; | |
2216 | ||
2217 | if (!(va->flags & VM_VM_AREA)) | |
2218 | continue; | |
2219 | ||
2220 | vm = va->vm; | |
2221 | vaddr = (char *) vm->addr; | |
762216ab | 2222 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2223 | continue; |
2224 | while (addr < vaddr) { | |
2225 | if (count == 0) | |
2226 | goto finished; | |
2227 | *buf = '\0'; | |
2228 | buf++; | |
2229 | addr++; | |
2230 | count--; | |
2231 | } | |
762216ab | 2232 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2233 | if (n > count) |
2234 | n = count; | |
e81ce85f | 2235 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2236 | aligned_vread(buf, addr, n); |
2237 | else /* IOREMAP area is treated as memory hole */ | |
2238 | memset(buf, 0, n); | |
2239 | buf += n; | |
2240 | addr += n; | |
2241 | count -= n; | |
1da177e4 LT |
2242 | } |
2243 | finished: | |
e81ce85f | 2244 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2245 | |
2246 | if (buf == buf_start) | |
2247 | return 0; | |
2248 | /* zero-fill memory holes */ | |
2249 | if (buf != buf_start + buflen) | |
2250 | memset(buf, 0, buflen - (buf - buf_start)); | |
2251 | ||
2252 | return buflen; | |
1da177e4 LT |
2253 | } |
2254 | ||
d0107eb0 | 2255 | /** |
92eac168 MR |
2256 | * vwrite() - write vmalloc area in a safe way. |
2257 | * @buf: buffer for source data | |
2258 | * @addr: vm address. | |
2259 | * @count: number of bytes to be read. | |
2260 | * | |
92eac168 MR |
2261 | * This function checks that addr is a valid vmalloc'ed area, and |
2262 | * copy data from a buffer to the given addr. If specified range of | |
2263 | * [addr...addr+count) includes some valid address, data is copied from | |
2264 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2265 | * IOREMAP area is treated as memory hole and no copy is done. | |
2266 | * | |
2267 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2268 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2269 | * | |
2270 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2271 | * should know vmalloc() area is valid and can use memcpy(). | |
2272 | * This is for routines which have to access vmalloc area without | |
2273 | * any informaion, as /dev/kmem. | |
a862f68a MR |
2274 | * |
2275 | * Return: number of bytes for which addr and buf should be | |
2276 | * increased (same number as @count) or %0 if [addr...addr+count) | |
2277 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 2278 | */ |
1da177e4 LT |
2279 | long vwrite(char *buf, char *addr, unsigned long count) |
2280 | { | |
e81ce85f JK |
2281 | struct vmap_area *va; |
2282 | struct vm_struct *vm; | |
d0107eb0 KH |
2283 | char *vaddr; |
2284 | unsigned long n, buflen; | |
2285 | int copied = 0; | |
1da177e4 LT |
2286 | |
2287 | /* Don't allow overflow */ | |
2288 | if ((unsigned long) addr + count < count) | |
2289 | count = -(unsigned long) addr; | |
d0107eb0 | 2290 | buflen = count; |
1da177e4 | 2291 | |
e81ce85f JK |
2292 | spin_lock(&vmap_area_lock); |
2293 | list_for_each_entry(va, &vmap_area_list, list) { | |
2294 | if (!count) | |
2295 | break; | |
2296 | ||
2297 | if (!(va->flags & VM_VM_AREA)) | |
2298 | continue; | |
2299 | ||
2300 | vm = va->vm; | |
2301 | vaddr = (char *) vm->addr; | |
762216ab | 2302 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2303 | continue; |
2304 | while (addr < vaddr) { | |
2305 | if (count == 0) | |
2306 | goto finished; | |
2307 | buf++; | |
2308 | addr++; | |
2309 | count--; | |
2310 | } | |
762216ab | 2311 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2312 | if (n > count) |
2313 | n = count; | |
e81ce85f | 2314 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2315 | aligned_vwrite(buf, addr, n); |
2316 | copied++; | |
2317 | } | |
2318 | buf += n; | |
2319 | addr += n; | |
2320 | count -= n; | |
1da177e4 LT |
2321 | } |
2322 | finished: | |
e81ce85f | 2323 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2324 | if (!copied) |
2325 | return 0; | |
2326 | return buflen; | |
1da177e4 | 2327 | } |
83342314 NP |
2328 | |
2329 | /** | |
92eac168 MR |
2330 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
2331 | * @vma: vma to cover | |
2332 | * @uaddr: target user address to start at | |
2333 | * @kaddr: virtual address of vmalloc kernel memory | |
2334 | * @size: size of map area | |
7682486b | 2335 | * |
92eac168 | 2336 | * Returns: 0 for success, -Exxx on failure |
83342314 | 2337 | * |
92eac168 MR |
2338 | * This function checks that @kaddr is a valid vmalloc'ed area, |
2339 | * and that it is big enough to cover the range starting at | |
2340 | * @uaddr in @vma. Will return failure if that criteria isn't | |
2341 | * met. | |
83342314 | 2342 | * |
92eac168 | 2343 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 2344 | */ |
e69e9d4a HD |
2345 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
2346 | void *kaddr, unsigned long size) | |
83342314 NP |
2347 | { |
2348 | struct vm_struct *area; | |
83342314 | 2349 | |
e69e9d4a HD |
2350 | size = PAGE_ALIGN(size); |
2351 | ||
2352 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
2353 | return -EINVAL; |
2354 | ||
e69e9d4a | 2355 | area = find_vm_area(kaddr); |
83342314 | 2356 | if (!area) |
db64fe02 | 2357 | return -EINVAL; |
83342314 NP |
2358 | |
2359 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2360 | return -EINVAL; |
83342314 | 2361 | |
401592d2 | 2362 | if (kaddr + size > area->addr + get_vm_area_size(area)) |
db64fe02 | 2363 | return -EINVAL; |
83342314 | 2364 | |
83342314 | 2365 | do { |
e69e9d4a | 2366 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
2367 | int ret; |
2368 | ||
83342314 NP |
2369 | ret = vm_insert_page(vma, uaddr, page); |
2370 | if (ret) | |
2371 | return ret; | |
2372 | ||
2373 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
2374 | kaddr += PAGE_SIZE; |
2375 | size -= PAGE_SIZE; | |
2376 | } while (size > 0); | |
83342314 | 2377 | |
314e51b9 | 2378 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 2379 | |
db64fe02 | 2380 | return 0; |
83342314 | 2381 | } |
e69e9d4a HD |
2382 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
2383 | ||
2384 | /** | |
92eac168 MR |
2385 | * remap_vmalloc_range - map vmalloc pages to userspace |
2386 | * @vma: vma to cover (map full range of vma) | |
2387 | * @addr: vmalloc memory | |
2388 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 2389 | * |
92eac168 | 2390 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 2391 | * |
92eac168 MR |
2392 | * This function checks that addr is a valid vmalloc'ed area, and |
2393 | * that it is big enough to cover the vma. Will return failure if | |
2394 | * that criteria isn't met. | |
e69e9d4a | 2395 | * |
92eac168 | 2396 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
2397 | */ |
2398 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
2399 | unsigned long pgoff) | |
2400 | { | |
2401 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
2402 | addr + (pgoff << PAGE_SHIFT), | |
2403 | vma->vm_end - vma->vm_start); | |
2404 | } | |
83342314 NP |
2405 | EXPORT_SYMBOL(remap_vmalloc_range); |
2406 | ||
1eeb66a1 CH |
2407 | /* |
2408 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
2409 | * have one. | |
2410 | */ | |
3b32123d | 2411 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
2412 | { |
2413 | } | |
5f4352fb JF |
2414 | |
2415 | ||
2f569afd | 2416 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb | 2417 | { |
cd12909c DV |
2418 | pte_t ***p = data; |
2419 | ||
2420 | if (p) { | |
2421 | *(*p) = pte; | |
2422 | (*p)++; | |
2423 | } | |
5f4352fb JF |
2424 | return 0; |
2425 | } | |
2426 | ||
2427 | /** | |
92eac168 MR |
2428 | * alloc_vm_area - allocate a range of kernel address space |
2429 | * @size: size of the area | |
2430 | * @ptes: returns the PTEs for the address space | |
7682486b | 2431 | * |
92eac168 | 2432 | * Returns: NULL on failure, vm_struct on success |
5f4352fb | 2433 | * |
92eac168 MR |
2434 | * This function reserves a range of kernel address space, and |
2435 | * allocates pagetables to map that range. No actual mappings | |
2436 | * are created. | |
cd12909c | 2437 | * |
92eac168 MR |
2438 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
2439 | * allocated for the VM area are returned. | |
5f4352fb | 2440 | */ |
cd12909c | 2441 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
2442 | { |
2443 | struct vm_struct *area; | |
2444 | ||
23016969 CL |
2445 | area = get_vm_area_caller(size, VM_IOREMAP, |
2446 | __builtin_return_address(0)); | |
5f4352fb JF |
2447 | if (area == NULL) |
2448 | return NULL; | |
2449 | ||
2450 | /* | |
2451 | * This ensures that page tables are constructed for this region | |
2452 | * of kernel virtual address space and mapped into init_mm. | |
2453 | */ | |
2454 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 2455 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
2456 | free_vm_area(area); |
2457 | return NULL; | |
2458 | } | |
2459 | ||
5f4352fb JF |
2460 | return area; |
2461 | } | |
2462 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
2463 | ||
2464 | void free_vm_area(struct vm_struct *area) | |
2465 | { | |
2466 | struct vm_struct *ret; | |
2467 | ret = remove_vm_area(area->addr); | |
2468 | BUG_ON(ret != area); | |
2469 | kfree(area); | |
2470 | } | |
2471 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 2472 | |
4f8b02b4 | 2473 | #ifdef CONFIG_SMP |
ca23e405 TH |
2474 | static struct vmap_area *node_to_va(struct rb_node *n) |
2475 | { | |
4583e773 | 2476 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
2477 | } |
2478 | ||
2479 | /** | |
2480 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
2481 | * @end: target address | |
2482 | * @pnext: out arg for the next vmap_area | |
2483 | * @pprev: out arg for the previous vmap_area | |
2484 | * | |
2485 | * Returns: %true if either or both of next and prev are found, | |
2486 | * %false if no vmap_area exists | |
2487 | * | |
2488 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2489 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2490 | */ | |
2491 | static bool pvm_find_next_prev(unsigned long end, | |
2492 | struct vmap_area **pnext, | |
2493 | struct vmap_area **pprev) | |
2494 | { | |
2495 | struct rb_node *n = vmap_area_root.rb_node; | |
2496 | struct vmap_area *va = NULL; | |
2497 | ||
2498 | while (n) { | |
2499 | va = rb_entry(n, struct vmap_area, rb_node); | |
2500 | if (end < va->va_end) | |
2501 | n = n->rb_left; | |
2502 | else if (end > va->va_end) | |
2503 | n = n->rb_right; | |
2504 | else | |
2505 | break; | |
2506 | } | |
2507 | ||
2508 | if (!va) | |
2509 | return false; | |
2510 | ||
2511 | if (va->va_end > end) { | |
2512 | *pnext = va; | |
2513 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2514 | } else { | |
2515 | *pprev = va; | |
2516 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2517 | } | |
2518 | return true; | |
2519 | } | |
2520 | ||
2521 | /** | |
2522 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2523 | * @pnext: in/out arg for the next vmap_area | |
2524 | * @pprev: in/out arg for the previous vmap_area | |
2525 | * @align: alignment | |
2526 | * | |
2527 | * Returns: determined end address | |
2528 | * | |
2529 | * Find the highest aligned address between *@pnext and *@pprev below | |
2530 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2531 | * down address is between the end addresses of the two vmap_areas. | |
2532 | * | |
2533 | * Please note that the address returned by this function may fall | |
2534 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2535 | * that. | |
2536 | */ | |
2537 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2538 | struct vmap_area **pprev, | |
2539 | unsigned long align) | |
2540 | { | |
2541 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2542 | unsigned long addr; | |
2543 | ||
2544 | if (*pnext) | |
2545 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2546 | else | |
2547 | addr = vmalloc_end; | |
2548 | ||
2549 | while (*pprev && (*pprev)->va_end > addr) { | |
2550 | *pnext = *pprev; | |
2551 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2552 | } | |
2553 | ||
2554 | return addr; | |
2555 | } | |
2556 | ||
2557 | /** | |
2558 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2559 | * @offsets: array containing offset of each area | |
2560 | * @sizes: array containing size of each area | |
2561 | * @nr_vms: the number of areas to allocate | |
2562 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
2563 | * |
2564 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2565 | * vm_structs on success, %NULL on failure | |
2566 | * | |
2567 | * Percpu allocator wants to use congruent vm areas so that it can | |
2568 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
2569 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
2570 | * be scattered pretty far, distance between two areas easily going up | |
2571 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
2572 | * areas are allocated from top. | |
ca23e405 TH |
2573 | * |
2574 | * Despite its complicated look, this allocator is rather simple. It | |
2575 | * does everything top-down and scans areas from the end looking for | |
2576 | * matching slot. While scanning, if any of the areas overlaps with | |
2577 | * existing vmap_area, the base address is pulled down to fit the | |
2578 | * area. Scanning is repeated till all the areas fit and then all | |
c568da28 | 2579 | * necessary data structures are inserted and the result is returned. |
ca23e405 TH |
2580 | */ |
2581 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2582 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 2583 | size_t align) |
ca23e405 TH |
2584 | { |
2585 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2586 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2587 | struct vmap_area **vas, *prev, *next; | |
2588 | struct vm_struct **vms; | |
2589 | int area, area2, last_area, term_area; | |
2590 | unsigned long base, start, end, last_end; | |
2591 | bool purged = false; | |
2592 | ||
ca23e405 | 2593 | /* verify parameters and allocate data structures */ |
891c49ab | 2594 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
2595 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
2596 | start = offsets[area]; | |
2597 | end = start + sizes[area]; | |
2598 | ||
2599 | /* is everything aligned properly? */ | |
2600 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2601 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2602 | ||
2603 | /* detect the area with the highest address */ | |
2604 | if (start > offsets[last_area]) | |
2605 | last_area = area; | |
2606 | ||
c568da28 | 2607 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
2608 | unsigned long start2 = offsets[area2]; |
2609 | unsigned long end2 = start2 + sizes[area2]; | |
2610 | ||
c568da28 | 2611 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
2612 | } |
2613 | } | |
2614 | last_end = offsets[last_area] + sizes[last_area]; | |
2615 | ||
2616 | if (vmalloc_end - vmalloc_start < last_end) { | |
2617 | WARN_ON(true); | |
2618 | return NULL; | |
2619 | } | |
2620 | ||
4d67d860 TM |
2621 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
2622 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 2623 | if (!vas || !vms) |
f1db7afd | 2624 | goto err_free2; |
ca23e405 TH |
2625 | |
2626 | for (area = 0; area < nr_vms; area++) { | |
ec3f64fc DR |
2627 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
2628 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); | |
ca23e405 TH |
2629 | if (!vas[area] || !vms[area]) |
2630 | goto err_free; | |
2631 | } | |
2632 | retry: | |
2633 | spin_lock(&vmap_area_lock); | |
2634 | ||
2635 | /* start scanning - we scan from the top, begin with the last area */ | |
2636 | area = term_area = last_area; | |
2637 | start = offsets[area]; | |
2638 | end = start + sizes[area]; | |
2639 | ||
2640 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2641 | base = vmalloc_end - last_end; | |
2642 | goto found; | |
2643 | } | |
2644 | base = pvm_determine_end(&next, &prev, align) - end; | |
2645 | ||
2646 | while (true) { | |
2647 | BUG_ON(next && next->va_end <= base + end); | |
2648 | BUG_ON(prev && prev->va_end > base + end); | |
2649 | ||
2650 | /* | |
2651 | * base might have underflowed, add last_end before | |
2652 | * comparing. | |
2653 | */ | |
2654 | if (base + last_end < vmalloc_start + last_end) { | |
2655 | spin_unlock(&vmap_area_lock); | |
2656 | if (!purged) { | |
2657 | purge_vmap_area_lazy(); | |
2658 | purged = true; | |
2659 | goto retry; | |
2660 | } | |
2661 | goto err_free; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * If next overlaps, move base downwards so that it's | |
2666 | * right below next and then recheck. | |
2667 | */ | |
2668 | if (next && next->va_start < base + end) { | |
2669 | base = pvm_determine_end(&next, &prev, align) - end; | |
2670 | term_area = area; | |
2671 | continue; | |
2672 | } | |
2673 | ||
2674 | /* | |
2675 | * If prev overlaps, shift down next and prev and move | |
2676 | * base so that it's right below new next and then | |
2677 | * recheck. | |
2678 | */ | |
2679 | if (prev && prev->va_end > base + start) { | |
2680 | next = prev; | |
2681 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2682 | base = pvm_determine_end(&next, &prev, align) - end; | |
2683 | term_area = area; | |
2684 | continue; | |
2685 | } | |
2686 | ||
2687 | /* | |
2688 | * This area fits, move on to the previous one. If | |
2689 | * the previous one is the terminal one, we're done. | |
2690 | */ | |
2691 | area = (area + nr_vms - 1) % nr_vms; | |
2692 | if (area == term_area) | |
2693 | break; | |
2694 | start = offsets[area]; | |
2695 | end = start + sizes[area]; | |
2696 | pvm_find_next_prev(base + end, &next, &prev); | |
2697 | } | |
2698 | found: | |
2699 | /* we've found a fitting base, insert all va's */ | |
2700 | for (area = 0; area < nr_vms; area++) { | |
2701 | struct vmap_area *va = vas[area]; | |
2702 | ||
2703 | va->va_start = base + offsets[area]; | |
2704 | va->va_end = va->va_start + sizes[area]; | |
2705 | __insert_vmap_area(va); | |
2706 | } | |
2707 | ||
2708 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2709 | ||
2710 | spin_unlock(&vmap_area_lock); | |
2711 | ||
2712 | /* insert all vm's */ | |
2713 | for (area = 0; area < nr_vms; area++) | |
3645cb4a ZY |
2714 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
2715 | pcpu_get_vm_areas); | |
ca23e405 TH |
2716 | |
2717 | kfree(vas); | |
2718 | return vms; | |
2719 | ||
2720 | err_free: | |
2721 | for (area = 0; area < nr_vms; area++) { | |
f1db7afd KC |
2722 | kfree(vas[area]); |
2723 | kfree(vms[area]); | |
ca23e405 | 2724 | } |
f1db7afd | 2725 | err_free2: |
ca23e405 TH |
2726 | kfree(vas); |
2727 | kfree(vms); | |
2728 | return NULL; | |
2729 | } | |
2730 | ||
2731 | /** | |
2732 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2733 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2734 | * @nr_vms: the number of allocated areas | |
2735 | * | |
2736 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2737 | */ | |
2738 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2739 | { | |
2740 | int i; | |
2741 | ||
2742 | for (i = 0; i < nr_vms; i++) | |
2743 | free_vm_area(vms[i]); | |
2744 | kfree(vms); | |
2745 | } | |
4f8b02b4 | 2746 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
2747 | |
2748 | #ifdef CONFIG_PROC_FS | |
2749 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 2750 | __acquires(&vmap_area_lock) |
a10aa579 | 2751 | { |
d4033afd | 2752 | spin_lock(&vmap_area_lock); |
3f500069 | 2753 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
2754 | } |
2755 | ||
2756 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2757 | { | |
3f500069 | 2758 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
2759 | } |
2760 | ||
2761 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 2762 | __releases(&vmap_area_lock) |
a10aa579 | 2763 | { |
d4033afd | 2764 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
2765 | } |
2766 | ||
a47a126a ED |
2767 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2768 | { | |
e5adfffc | 2769 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
2770 | unsigned int nr, *counters = m->private; |
2771 | ||
2772 | if (!counters) | |
2773 | return; | |
2774 | ||
af12346c WL |
2775 | if (v->flags & VM_UNINITIALIZED) |
2776 | return; | |
7e5b528b DV |
2777 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
2778 | smp_rmb(); | |
af12346c | 2779 | |
a47a126a ED |
2780 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
2781 | ||
2782 | for (nr = 0; nr < v->nr_pages; nr++) | |
2783 | counters[page_to_nid(v->pages[nr])]++; | |
2784 | ||
2785 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2786 | if (counters[nr]) | |
2787 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2788 | } | |
2789 | } | |
2790 | ||
a10aa579 CL |
2791 | static int s_show(struct seq_file *m, void *p) |
2792 | { | |
3f500069 | 2793 | struct vmap_area *va; |
d4033afd JK |
2794 | struct vm_struct *v; |
2795 | ||
3f500069 | 2796 | va = list_entry(p, struct vmap_area, list); |
2797 | ||
c2ce8c14 WL |
2798 | /* |
2799 | * s_show can encounter race with remove_vm_area, !VM_VM_AREA on | |
2800 | * behalf of vmap area is being tear down or vm_map_ram allocation. | |
2801 | */ | |
78c72746 YX |
2802 | if (!(va->flags & VM_VM_AREA)) { |
2803 | seq_printf(m, "0x%pK-0x%pK %7ld %s\n", | |
2804 | (void *)va->va_start, (void *)va->va_end, | |
2805 | va->va_end - va->va_start, | |
2806 | va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram"); | |
2807 | ||
d4033afd | 2808 | return 0; |
78c72746 | 2809 | } |
d4033afd JK |
2810 | |
2811 | v = va->vm; | |
a10aa579 | 2812 | |
45ec1690 | 2813 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
2814 | v->addr, v->addr + v->size, v->size); |
2815 | ||
62c70bce JP |
2816 | if (v->caller) |
2817 | seq_printf(m, " %pS", v->caller); | |
23016969 | 2818 | |
a10aa579 CL |
2819 | if (v->nr_pages) |
2820 | seq_printf(m, " pages=%d", v->nr_pages); | |
2821 | ||
2822 | if (v->phys_addr) | |
199eaa05 | 2823 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
2824 | |
2825 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 2826 | seq_puts(m, " ioremap"); |
a10aa579 CL |
2827 | |
2828 | if (v->flags & VM_ALLOC) | |
f4527c90 | 2829 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
2830 | |
2831 | if (v->flags & VM_MAP) | |
f4527c90 | 2832 | seq_puts(m, " vmap"); |
a10aa579 CL |
2833 | |
2834 | if (v->flags & VM_USERMAP) | |
f4527c90 | 2835 | seq_puts(m, " user"); |
a10aa579 | 2836 | |
244d63ee | 2837 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 2838 | seq_puts(m, " vpages"); |
a10aa579 | 2839 | |
a47a126a | 2840 | show_numa_info(m, v); |
a10aa579 CL |
2841 | seq_putc(m, '\n'); |
2842 | return 0; | |
2843 | } | |
2844 | ||
5f6a6a9c | 2845 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2846 | .start = s_start, |
2847 | .next = s_next, | |
2848 | .stop = s_stop, | |
2849 | .show = s_show, | |
2850 | }; | |
5f6a6a9c | 2851 | |
5f6a6a9c AD |
2852 | static int __init proc_vmalloc_init(void) |
2853 | { | |
fddda2b7 | 2854 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 2855 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
2856 | &vmalloc_op, |
2857 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 2858 | else |
0825a6f9 | 2859 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
2860 | return 0; |
2861 | } | |
2862 | module_init(proc_vmalloc_init); | |
db3808c1 | 2863 | |
a10aa579 | 2864 | #endif |