]>
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> | |
d43c36dc | 15 | #include <linux/sched.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> |
3ac7fe5a | 21 | #include <linux/debugobjects.h> |
23016969 | 22 | #include <linux/kallsyms.h> |
db64fe02 NP |
23 | #include <linux/list.h> |
24 | #include <linux/rbtree.h> | |
25 | #include <linux/radix-tree.h> | |
26 | #include <linux/rcupdate.h> | |
f0aa6617 | 27 | #include <linux/pfn.h> |
89219d37 | 28 | #include <linux/kmemleak.h> |
60063497 | 29 | #include <linux/atomic.h> |
3b32123d | 30 | #include <linux/compiler.h> |
32fcfd40 | 31 | #include <linux/llist.h> |
0f616be1 | 32 | #include <linux/bitops.h> |
3b32123d | 33 | |
1da177e4 LT |
34 | #include <asm/uaccess.h> |
35 | #include <asm/tlbflush.h> | |
2dca6999 | 36 | #include <asm/shmparam.h> |
1da177e4 | 37 | |
dd56b046 MG |
38 | #include "internal.h" |
39 | ||
32fcfd40 AV |
40 | struct vfree_deferred { |
41 | struct llist_head list; | |
42 | struct work_struct wq; | |
43 | }; | |
44 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
45 | ||
46 | static void __vunmap(const void *, int); | |
47 | ||
48 | static void free_work(struct work_struct *w) | |
49 | { | |
50 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
51 | struct llist_node *llnode = llist_del_all(&p->list); | |
52 | while (llnode) { | |
53 | void *p = llnode; | |
54 | llnode = llist_next(llnode); | |
55 | __vunmap(p, 1); | |
56 | } | |
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 | ||
db64fe02 | 88 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
1da177e4 LT |
89 | { |
90 | pud_t *pud; | |
91 | unsigned long next; | |
92 | ||
93 | pud = pud_offset(pgd, addr); | |
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 | ||
db64fe02 | 104 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
105 | { |
106 | pgd_t *pgd; | |
107 | unsigned long next; | |
1da177e4 LT |
108 | |
109 | BUG_ON(addr >= end); | |
110 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
111 | do { |
112 | next = pgd_addr_end(addr, end); | |
113 | if (pgd_none_or_clear_bad(pgd)) | |
114 | continue; | |
115 | vunmap_pud_range(pgd, addr, next); | |
116 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
117 | } |
118 | ||
119 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 120 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
121 | { |
122 | pte_t *pte; | |
123 | ||
db64fe02 NP |
124 | /* |
125 | * nr is a running index into the array which helps higher level | |
126 | * callers keep track of where we're up to. | |
127 | */ | |
128 | ||
872fec16 | 129 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
130 | if (!pte) |
131 | return -ENOMEM; | |
132 | do { | |
db64fe02 NP |
133 | struct page *page = pages[*nr]; |
134 | ||
135 | if (WARN_ON(!pte_none(*pte))) | |
136 | return -EBUSY; | |
137 | if (WARN_ON(!page)) | |
1da177e4 LT |
138 | return -ENOMEM; |
139 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 140 | (*nr)++; |
1da177e4 LT |
141 | } while (pte++, addr += PAGE_SIZE, addr != end); |
142 | return 0; | |
143 | } | |
144 | ||
db64fe02 NP |
145 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
146 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
147 | { |
148 | pmd_t *pmd; | |
149 | unsigned long next; | |
150 | ||
151 | pmd = pmd_alloc(&init_mm, pud, addr); | |
152 | if (!pmd) | |
153 | return -ENOMEM; | |
154 | do { | |
155 | next = pmd_addr_end(addr, end); | |
db64fe02 | 156 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
157 | return -ENOMEM; |
158 | } while (pmd++, addr = next, addr != end); | |
159 | return 0; | |
160 | } | |
161 | ||
db64fe02 NP |
162 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
163 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
164 | { |
165 | pud_t *pud; | |
166 | unsigned long next; | |
167 | ||
168 | pud = pud_alloc(&init_mm, pgd, addr); | |
169 | if (!pud) | |
170 | return -ENOMEM; | |
171 | do { | |
172 | next = pud_addr_end(addr, end); | |
db64fe02 | 173 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
174 | return -ENOMEM; |
175 | } while (pud++, addr = next, addr != end); | |
176 | return 0; | |
177 | } | |
178 | ||
db64fe02 NP |
179 | /* |
180 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
181 | * will have pfns corresponding to the "pages" array. | |
182 | * | |
183 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
184 | */ | |
8fc48985 TH |
185 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
186 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
187 | { |
188 | pgd_t *pgd; | |
189 | unsigned long next; | |
2e4e27c7 | 190 | unsigned long addr = start; |
db64fe02 NP |
191 | int err = 0; |
192 | int nr = 0; | |
1da177e4 LT |
193 | |
194 | BUG_ON(addr >= end); | |
195 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
196 | do { |
197 | next = pgd_addr_end(addr, end); | |
db64fe02 | 198 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 199 | if (err) |
bf88c8c8 | 200 | return err; |
1da177e4 | 201 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 202 | |
db64fe02 | 203 | return nr; |
1da177e4 LT |
204 | } |
205 | ||
8fc48985 TH |
206 | static int vmap_page_range(unsigned long start, unsigned long end, |
207 | pgprot_t prot, struct page **pages) | |
208 | { | |
209 | int ret; | |
210 | ||
211 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
212 | flush_cache_vmap(start, end); | |
213 | return ret; | |
214 | } | |
215 | ||
81ac3ad9 | 216 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
217 | { |
218 | /* | |
ab4f2ee1 | 219 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
220 | * and fall back on vmalloc() if that fails. Others |
221 | * just put it in the vmalloc space. | |
222 | */ | |
223 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
224 | unsigned long addr = (unsigned long)x; | |
225 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
226 | return 1; | |
227 | #endif | |
228 | return is_vmalloc_addr(x); | |
229 | } | |
230 | ||
48667e7a | 231 | /* |
add688fb | 232 | * Walk a vmap address to the struct page it maps. |
48667e7a | 233 | */ |
add688fb | 234 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
235 | { |
236 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 237 | struct page *page = NULL; |
48667e7a | 238 | pgd_t *pgd = pgd_offset_k(addr); |
48667e7a | 239 | |
7aa413de IM |
240 | /* |
241 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
242 | * architectures that do not vmalloc module space | |
243 | */ | |
73bdf0a6 | 244 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 245 | |
48667e7a | 246 | if (!pgd_none(*pgd)) { |
db64fe02 | 247 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 248 | if (!pud_none(*pud)) { |
db64fe02 | 249 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 250 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
251 | pte_t *ptep, pte; |
252 | ||
48667e7a CL |
253 | ptep = pte_offset_map(pmd, addr); |
254 | pte = *ptep; | |
255 | if (pte_present(pte)) | |
add688fb | 256 | page = pte_page(pte); |
48667e7a CL |
257 | pte_unmap(ptep); |
258 | } | |
259 | } | |
260 | } | |
add688fb | 261 | return page; |
48667e7a | 262 | } |
add688fb | 263 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
264 | |
265 | /* | |
add688fb | 266 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 267 | */ |
add688fb | 268 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 269 | { |
add688fb | 270 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 271 | } |
add688fb | 272 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 273 | |
db64fe02 NP |
274 | |
275 | /*** Global kva allocator ***/ | |
276 | ||
277 | #define VM_LAZY_FREE 0x01 | |
278 | #define VM_LAZY_FREEING 0x02 | |
279 | #define VM_VM_AREA 0x04 | |
280 | ||
db64fe02 | 281 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
282 | /* Export for kexec only */ |
283 | LIST_HEAD(vmap_area_list); | |
89699605 NP |
284 | static struct rb_root vmap_area_root = RB_ROOT; |
285 | ||
286 | /* The vmap cache globals are protected by vmap_area_lock */ | |
287 | static struct rb_node *free_vmap_cache; | |
288 | static unsigned long cached_hole_size; | |
289 | static unsigned long cached_vstart; | |
290 | static unsigned long cached_align; | |
291 | ||
ca23e405 | 292 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
293 | |
294 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 295 | { |
db64fe02 NP |
296 | struct rb_node *n = vmap_area_root.rb_node; |
297 | ||
298 | while (n) { | |
299 | struct vmap_area *va; | |
300 | ||
301 | va = rb_entry(n, struct vmap_area, rb_node); | |
302 | if (addr < va->va_start) | |
303 | n = n->rb_left; | |
cef2ac3f | 304 | else if (addr >= va->va_end) |
db64fe02 NP |
305 | n = n->rb_right; |
306 | else | |
307 | return va; | |
308 | } | |
309 | ||
310 | return NULL; | |
311 | } | |
312 | ||
313 | static void __insert_vmap_area(struct vmap_area *va) | |
314 | { | |
315 | struct rb_node **p = &vmap_area_root.rb_node; | |
316 | struct rb_node *parent = NULL; | |
317 | struct rb_node *tmp; | |
318 | ||
319 | while (*p) { | |
170168d0 | 320 | struct vmap_area *tmp_va; |
db64fe02 NP |
321 | |
322 | parent = *p; | |
170168d0 NK |
323 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
324 | if (va->va_start < tmp_va->va_end) | |
db64fe02 | 325 | p = &(*p)->rb_left; |
170168d0 | 326 | else if (va->va_end > tmp_va->va_start) |
db64fe02 NP |
327 | p = &(*p)->rb_right; |
328 | else | |
329 | BUG(); | |
330 | } | |
331 | ||
332 | rb_link_node(&va->rb_node, parent, p); | |
333 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
334 | ||
4341fa45 | 335 | /* address-sort this list */ |
db64fe02 NP |
336 | tmp = rb_prev(&va->rb_node); |
337 | if (tmp) { | |
338 | struct vmap_area *prev; | |
339 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
340 | list_add_rcu(&va->list, &prev->list); | |
341 | } else | |
342 | list_add_rcu(&va->list, &vmap_area_list); | |
343 | } | |
344 | ||
345 | static void purge_vmap_area_lazy(void); | |
346 | ||
347 | /* | |
348 | * Allocate a region of KVA of the specified size and alignment, within the | |
349 | * vstart and vend. | |
350 | */ | |
351 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
352 | unsigned long align, | |
353 | unsigned long vstart, unsigned long vend, | |
354 | int node, gfp_t gfp_mask) | |
355 | { | |
356 | struct vmap_area *va; | |
357 | struct rb_node *n; | |
1da177e4 | 358 | unsigned long addr; |
db64fe02 | 359 | int purged = 0; |
89699605 | 360 | struct vmap_area *first; |
db64fe02 | 361 | |
7766970c | 362 | BUG_ON(!size); |
891c49ab | 363 | BUG_ON(offset_in_page(size)); |
89699605 | 364 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 365 | |
db64fe02 NP |
366 | va = kmalloc_node(sizeof(struct vmap_area), |
367 | gfp_mask & GFP_RECLAIM_MASK, node); | |
368 | if (unlikely(!va)) | |
369 | return ERR_PTR(-ENOMEM); | |
370 | ||
7f88f88f CM |
371 | /* |
372 | * Only scan the relevant parts containing pointers to other objects | |
373 | * to avoid false negatives. | |
374 | */ | |
375 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK); | |
376 | ||
db64fe02 NP |
377 | retry: |
378 | spin_lock(&vmap_area_lock); | |
89699605 NP |
379 | /* |
380 | * Invalidate cache if we have more permissive parameters. | |
381 | * cached_hole_size notes the largest hole noticed _below_ | |
382 | * the vmap_area cached in free_vmap_cache: if size fits | |
383 | * into that hole, we want to scan from vstart to reuse | |
384 | * the hole instead of allocating above free_vmap_cache. | |
385 | * Note that __free_vmap_area may update free_vmap_cache | |
386 | * without updating cached_hole_size or cached_align. | |
387 | */ | |
388 | if (!free_vmap_cache || | |
389 | size < cached_hole_size || | |
390 | vstart < cached_vstart || | |
391 | align < cached_align) { | |
392 | nocache: | |
393 | cached_hole_size = 0; | |
394 | free_vmap_cache = NULL; | |
395 | } | |
396 | /* record if we encounter less permissive parameters */ | |
397 | cached_vstart = vstart; | |
398 | cached_align = align; | |
399 | ||
400 | /* find starting point for our search */ | |
401 | if (free_vmap_cache) { | |
402 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
248ac0e1 | 403 | addr = ALIGN(first->va_end, align); |
89699605 NP |
404 | if (addr < vstart) |
405 | goto nocache; | |
bcb615a8 | 406 | if (addr + size < addr) |
89699605 NP |
407 | goto overflow; |
408 | ||
409 | } else { | |
410 | addr = ALIGN(vstart, align); | |
bcb615a8 | 411 | if (addr + size < addr) |
89699605 NP |
412 | goto overflow; |
413 | ||
414 | n = vmap_area_root.rb_node; | |
415 | first = NULL; | |
416 | ||
417 | while (n) { | |
db64fe02 NP |
418 | struct vmap_area *tmp; |
419 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
420 | if (tmp->va_end >= addr) { | |
db64fe02 | 421 | first = tmp; |
89699605 NP |
422 | if (tmp->va_start <= addr) |
423 | break; | |
424 | n = n->rb_left; | |
425 | } else | |
db64fe02 | 426 | n = n->rb_right; |
89699605 | 427 | } |
db64fe02 NP |
428 | |
429 | if (!first) | |
430 | goto found; | |
db64fe02 | 431 | } |
89699605 NP |
432 | |
433 | /* from the starting point, walk areas until a suitable hole is found */ | |
248ac0e1 | 434 | while (addr + size > first->va_start && addr + size <= vend) { |
89699605 NP |
435 | if (addr + cached_hole_size < first->va_start) |
436 | cached_hole_size = first->va_start - addr; | |
248ac0e1 | 437 | addr = ALIGN(first->va_end, align); |
bcb615a8 | 438 | if (addr + size < addr) |
89699605 NP |
439 | goto overflow; |
440 | ||
92ca922f | 441 | if (list_is_last(&first->list, &vmap_area_list)) |
89699605 | 442 | goto found; |
92ca922f | 443 | |
6219c2a2 | 444 | first = list_next_entry(first, list); |
db64fe02 NP |
445 | } |
446 | ||
89699605 NP |
447 | found: |
448 | if (addr + size > vend) | |
449 | goto overflow; | |
db64fe02 NP |
450 | |
451 | va->va_start = addr; | |
452 | va->va_end = addr + size; | |
453 | va->flags = 0; | |
454 | __insert_vmap_area(va); | |
89699605 | 455 | free_vmap_cache = &va->rb_node; |
db64fe02 NP |
456 | spin_unlock(&vmap_area_lock); |
457 | ||
61e16557 | 458 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
459 | BUG_ON(va->va_start < vstart); |
460 | BUG_ON(va->va_end > vend); | |
461 | ||
db64fe02 | 462 | return va; |
89699605 NP |
463 | |
464 | overflow: | |
465 | spin_unlock(&vmap_area_lock); | |
466 | if (!purged) { | |
467 | purge_vmap_area_lazy(); | |
468 | purged = 1; | |
469 | goto retry; | |
470 | } | |
471 | if (printk_ratelimit()) | |
756a025f JP |
472 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
473 | size); | |
89699605 NP |
474 | kfree(va); |
475 | return ERR_PTR(-EBUSY); | |
db64fe02 NP |
476 | } |
477 | ||
db64fe02 NP |
478 | static void __free_vmap_area(struct vmap_area *va) |
479 | { | |
480 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
89699605 NP |
481 | |
482 | if (free_vmap_cache) { | |
483 | if (va->va_end < cached_vstart) { | |
484 | free_vmap_cache = NULL; | |
485 | } else { | |
486 | struct vmap_area *cache; | |
487 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
488 | if (va->va_start <= cache->va_start) { | |
489 | free_vmap_cache = rb_prev(&va->rb_node); | |
490 | /* | |
491 | * We don't try to update cached_hole_size or | |
492 | * cached_align, but it won't go very wrong. | |
493 | */ | |
494 | } | |
495 | } | |
496 | } | |
db64fe02 NP |
497 | rb_erase(&va->rb_node, &vmap_area_root); |
498 | RB_CLEAR_NODE(&va->rb_node); | |
499 | list_del_rcu(&va->list); | |
500 | ||
ca23e405 TH |
501 | /* |
502 | * Track the highest possible candidate for pcpu area | |
503 | * allocation. Areas outside of vmalloc area can be returned | |
504 | * here too, consider only end addresses which fall inside | |
505 | * vmalloc area proper. | |
506 | */ | |
507 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
508 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
509 | ||
14769de9 | 510 | kfree_rcu(va, rcu_head); |
db64fe02 NP |
511 | } |
512 | ||
513 | /* | |
514 | * Free a region of KVA allocated by alloc_vmap_area | |
515 | */ | |
516 | static void free_vmap_area(struct vmap_area *va) | |
517 | { | |
518 | spin_lock(&vmap_area_lock); | |
519 | __free_vmap_area(va); | |
520 | spin_unlock(&vmap_area_lock); | |
521 | } | |
522 | ||
523 | /* | |
524 | * Clear the pagetable entries of a given vmap_area | |
525 | */ | |
526 | static void unmap_vmap_area(struct vmap_area *va) | |
527 | { | |
528 | vunmap_page_range(va->va_start, va->va_end); | |
529 | } | |
530 | ||
cd52858c NP |
531 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
532 | { | |
533 | /* | |
f48d97f3 JK |
534 | * Unmap page tables and force a TLB flush immediately if pagealloc |
535 | * debugging is enabled. This catches use after free bugs similarly to | |
536 | * those in linear kernel virtual address space after a page has been | |
537 | * freed. | |
cd52858c | 538 | * |
f48d97f3 JK |
539 | * All the lazy freeing logic is still retained, in order to minimise |
540 | * intrusiveness of this debugging feature. | |
cd52858c | 541 | * |
f48d97f3 JK |
542 | * This is going to be *slow* (linear kernel virtual address debugging |
543 | * doesn't do a broadcast TLB flush so it is a lot faster). | |
cd52858c | 544 | */ |
f48d97f3 JK |
545 | if (debug_pagealloc_enabled()) { |
546 | vunmap_page_range(start, end); | |
547 | flush_tlb_kernel_range(start, end); | |
548 | } | |
cd52858c NP |
549 | } |
550 | ||
db64fe02 NP |
551 | /* |
552 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
553 | * before attempting to purge with a TLB flush. | |
554 | * | |
555 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
556 | * and take slightly longer to purge, but it will linearly reduce the number of | |
557 | * global TLB flushes that must be performed. It would seem natural to scale | |
558 | * this number up linearly with the number of CPUs (because vmapping activity | |
559 | * could also scale linearly with the number of CPUs), however it is likely | |
560 | * that in practice, workloads might be constrained in other ways that mean | |
561 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
562 | * conservative and not introduce a big latency on huge systems, so go with | |
563 | * a less aggressive log scale. It will still be an improvement over the old | |
564 | * code, and it will be simple to change the scale factor if we find that it | |
565 | * becomes a problem on bigger systems. | |
566 | */ | |
567 | static unsigned long lazy_max_pages(void) | |
568 | { | |
569 | unsigned int log; | |
570 | ||
571 | log = fls(num_online_cpus()); | |
572 | ||
573 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
574 | } | |
575 | ||
576 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
577 | ||
02b709df NP |
578 | /* for per-CPU blocks */ |
579 | static void purge_fragmented_blocks_allcpus(void); | |
580 | ||
3ee48b6a CW |
581 | /* |
582 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
583 | * immediately freed. | |
584 | */ | |
585 | void set_iounmap_nonlazy(void) | |
586 | { | |
587 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); | |
588 | } | |
589 | ||
db64fe02 NP |
590 | /* |
591 | * Purges all lazily-freed vmap areas. | |
592 | * | |
593 | * If sync is 0 then don't purge if there is already a purge in progress. | |
594 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
595 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
596 | * their own TLB flushing). | |
597 | * Returns with *start = min(*start, lowest purged address) | |
598 | * *end = max(*end, highest purged address) | |
599 | */ | |
600 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
601 | int sync, int force_flush) | |
602 | { | |
46666d8a | 603 | static DEFINE_SPINLOCK(purge_lock); |
db64fe02 NP |
604 | LIST_HEAD(valist); |
605 | struct vmap_area *va; | |
cbb76676 | 606 | struct vmap_area *n_va; |
db64fe02 NP |
607 | int nr = 0; |
608 | ||
609 | /* | |
610 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
611 | * should not expect such behaviour. This just simplifies locking for | |
612 | * the case that isn't actually used at the moment anyway. | |
613 | */ | |
614 | if (!sync && !force_flush) { | |
46666d8a | 615 | if (!spin_trylock(&purge_lock)) |
db64fe02 NP |
616 | return; |
617 | } else | |
46666d8a | 618 | spin_lock(&purge_lock); |
db64fe02 | 619 | |
02b709df NP |
620 | if (sync) |
621 | purge_fragmented_blocks_allcpus(); | |
622 | ||
db64fe02 NP |
623 | rcu_read_lock(); |
624 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
625 | if (va->flags & VM_LAZY_FREE) { | |
626 | if (va->va_start < *start) | |
627 | *start = va->va_start; | |
628 | if (va->va_end > *end) | |
629 | *end = va->va_end; | |
630 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
db64fe02 NP |
631 | list_add_tail(&va->purge_list, &valist); |
632 | va->flags |= VM_LAZY_FREEING; | |
633 | va->flags &= ~VM_LAZY_FREE; | |
634 | } | |
635 | } | |
636 | rcu_read_unlock(); | |
637 | ||
88f50044 | 638 | if (nr) |
db64fe02 | 639 | atomic_sub(nr, &vmap_lazy_nr); |
db64fe02 NP |
640 | |
641 | if (nr || force_flush) | |
642 | flush_tlb_kernel_range(*start, *end); | |
643 | ||
644 | if (nr) { | |
645 | spin_lock(&vmap_area_lock); | |
cbb76676 | 646 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
db64fe02 NP |
647 | __free_vmap_area(va); |
648 | spin_unlock(&vmap_area_lock); | |
649 | } | |
46666d8a | 650 | spin_unlock(&purge_lock); |
db64fe02 NP |
651 | } |
652 | ||
496850e5 NP |
653 | /* |
654 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
655 | * is already purging. | |
656 | */ | |
657 | static void try_purge_vmap_area_lazy(void) | |
658 | { | |
659 | unsigned long start = ULONG_MAX, end = 0; | |
660 | ||
661 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
662 | } | |
663 | ||
db64fe02 NP |
664 | /* |
665 | * Kick off a purge of the outstanding lazy areas. | |
666 | */ | |
667 | static void purge_vmap_area_lazy(void) | |
668 | { | |
669 | unsigned long start = ULONG_MAX, end = 0; | |
670 | ||
496850e5 | 671 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
db64fe02 NP |
672 | } |
673 | ||
674 | /* | |
64141da5 JF |
675 | * Free a vmap area, caller ensuring that the area has been unmapped |
676 | * and flush_cache_vunmap had been called for the correct range | |
677 | * previously. | |
db64fe02 | 678 | */ |
64141da5 | 679 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 NP |
680 | { |
681 | va->flags |= VM_LAZY_FREE; | |
682 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
683 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
496850e5 | 684 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
685 | } |
686 | ||
64141da5 JF |
687 | /* |
688 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been | |
689 | * called for the correct range previously. | |
690 | */ | |
691 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) | |
692 | { | |
693 | unmap_vmap_area(va); | |
694 | free_vmap_area_noflush(va); | |
695 | } | |
696 | ||
b29acbdc NP |
697 | /* |
698 | * Free and unmap a vmap area | |
699 | */ | |
700 | static void free_unmap_vmap_area(struct vmap_area *va) | |
701 | { | |
702 | flush_cache_vunmap(va->va_start, va->va_end); | |
703 | free_unmap_vmap_area_noflush(va); | |
704 | } | |
705 | ||
db64fe02 NP |
706 | static struct vmap_area *find_vmap_area(unsigned long addr) |
707 | { | |
708 | struct vmap_area *va; | |
709 | ||
710 | spin_lock(&vmap_area_lock); | |
711 | va = __find_vmap_area(addr); | |
712 | spin_unlock(&vmap_area_lock); | |
713 | ||
714 | return va; | |
715 | } | |
716 | ||
717 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
718 | { | |
719 | struct vmap_area *va; | |
720 | ||
721 | va = find_vmap_area(addr); | |
722 | BUG_ON(!va); | |
723 | free_unmap_vmap_area(va); | |
724 | } | |
725 | ||
726 | ||
727 | /*** Per cpu kva allocator ***/ | |
728 | ||
729 | /* | |
730 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
731 | * room for at least 16 percpu vmap blocks per CPU. | |
732 | */ | |
733 | /* | |
734 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
735 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
736 | * instead (we just need a rough idea) | |
737 | */ | |
738 | #if BITS_PER_LONG == 32 | |
739 | #define VMALLOC_SPACE (128UL*1024*1024) | |
740 | #else | |
741 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
742 | #endif | |
743 | ||
744 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
745 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
746 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
747 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
748 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
749 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
750 | #define VMAP_BBMAP_BITS \ |
751 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
752 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
753 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
754 | |
755 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
756 | ||
9b463334 JF |
757 | static bool vmap_initialized __read_mostly = false; |
758 | ||
db64fe02 NP |
759 | struct vmap_block_queue { |
760 | spinlock_t lock; | |
761 | struct list_head free; | |
db64fe02 NP |
762 | }; |
763 | ||
764 | struct vmap_block { | |
765 | spinlock_t lock; | |
766 | struct vmap_area *va; | |
db64fe02 | 767 | unsigned long free, dirty; |
7d61bfe8 | 768 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
769 | struct list_head free_list; |
770 | struct rcu_head rcu_head; | |
02b709df | 771 | struct list_head purge; |
db64fe02 NP |
772 | }; |
773 | ||
774 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
775 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
776 | ||
777 | /* | |
778 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
779 | * in the free path. Could get rid of this if we change the API to return a | |
780 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
781 | */ | |
782 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
783 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
784 | ||
785 | /* | |
786 | * We should probably have a fallback mechanism to allocate virtual memory | |
787 | * out of partially filled vmap blocks. However vmap block sizing should be | |
788 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
789 | * big problem. | |
790 | */ | |
791 | ||
792 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
793 | { | |
794 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
795 | addr /= VMAP_BLOCK_SIZE; | |
796 | return addr; | |
797 | } | |
798 | ||
cf725ce2 RP |
799 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
800 | { | |
801 | unsigned long addr; | |
802 | ||
803 | addr = va_start + (pages_off << PAGE_SHIFT); | |
804 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
805 | return (void *)addr; | |
806 | } | |
807 | ||
808 | /** | |
809 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
810 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
811 | * @order: how many 2^order pages should be occupied in newly allocated block | |
812 | * @gfp_mask: flags for the page level allocator | |
813 | * | |
814 | * Returns: virtual address in a newly allocated block or ERR_PTR(-errno) | |
815 | */ | |
816 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
817 | { |
818 | struct vmap_block_queue *vbq; | |
819 | struct vmap_block *vb; | |
820 | struct vmap_area *va; | |
821 | unsigned long vb_idx; | |
822 | int node, err; | |
cf725ce2 | 823 | void *vaddr; |
db64fe02 NP |
824 | |
825 | node = numa_node_id(); | |
826 | ||
827 | vb = kmalloc_node(sizeof(struct vmap_block), | |
828 | gfp_mask & GFP_RECLAIM_MASK, node); | |
829 | if (unlikely(!vb)) | |
830 | return ERR_PTR(-ENOMEM); | |
831 | ||
832 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
833 | VMALLOC_START, VMALLOC_END, | |
834 | node, gfp_mask); | |
ddf9c6d4 | 835 | if (IS_ERR(va)) { |
db64fe02 | 836 | kfree(vb); |
e7d86340 | 837 | return ERR_CAST(va); |
db64fe02 NP |
838 | } |
839 | ||
840 | err = radix_tree_preload(gfp_mask); | |
841 | if (unlikely(err)) { | |
842 | kfree(vb); | |
843 | free_vmap_area(va); | |
844 | return ERR_PTR(err); | |
845 | } | |
846 | ||
cf725ce2 | 847 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
848 | spin_lock_init(&vb->lock); |
849 | vb->va = va; | |
cf725ce2 RP |
850 | /* At least something should be left free */ |
851 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
852 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 853 | vb->dirty = 0; |
7d61bfe8 RP |
854 | vb->dirty_min = VMAP_BBMAP_BITS; |
855 | vb->dirty_max = 0; | |
db64fe02 | 856 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
857 | |
858 | vb_idx = addr_to_vb_idx(va->va_start); | |
859 | spin_lock(&vmap_block_tree_lock); | |
860 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
861 | spin_unlock(&vmap_block_tree_lock); | |
862 | BUG_ON(err); | |
863 | radix_tree_preload_end(); | |
864 | ||
865 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 866 | spin_lock(&vbq->lock); |
68ac546f | 867 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 868 | spin_unlock(&vbq->lock); |
3f04ba85 | 869 | put_cpu_var(vmap_block_queue); |
db64fe02 | 870 | |
cf725ce2 | 871 | return vaddr; |
db64fe02 NP |
872 | } |
873 | ||
db64fe02 NP |
874 | static void free_vmap_block(struct vmap_block *vb) |
875 | { | |
876 | struct vmap_block *tmp; | |
877 | unsigned long vb_idx; | |
878 | ||
db64fe02 NP |
879 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
880 | spin_lock(&vmap_block_tree_lock); | |
881 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
882 | spin_unlock(&vmap_block_tree_lock); | |
883 | BUG_ON(tmp != vb); | |
884 | ||
64141da5 | 885 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 886 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
887 | } |
888 | ||
02b709df NP |
889 | static void purge_fragmented_blocks(int cpu) |
890 | { | |
891 | LIST_HEAD(purge); | |
892 | struct vmap_block *vb; | |
893 | struct vmap_block *n_vb; | |
894 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
895 | ||
896 | rcu_read_lock(); | |
897 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
898 | ||
899 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
900 | continue; | |
901 | ||
902 | spin_lock(&vb->lock); | |
903 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
904 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
905 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
906 | vb->dirty_min = 0; |
907 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
908 | spin_lock(&vbq->lock); |
909 | list_del_rcu(&vb->free_list); | |
910 | spin_unlock(&vbq->lock); | |
911 | spin_unlock(&vb->lock); | |
912 | list_add_tail(&vb->purge, &purge); | |
913 | } else | |
914 | spin_unlock(&vb->lock); | |
915 | } | |
916 | rcu_read_unlock(); | |
917 | ||
918 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
919 | list_del(&vb->purge); | |
920 | free_vmap_block(vb); | |
921 | } | |
922 | } | |
923 | ||
02b709df NP |
924 | static void purge_fragmented_blocks_allcpus(void) |
925 | { | |
926 | int cpu; | |
927 | ||
928 | for_each_possible_cpu(cpu) | |
929 | purge_fragmented_blocks(cpu); | |
930 | } | |
931 | ||
db64fe02 NP |
932 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
933 | { | |
934 | struct vmap_block_queue *vbq; | |
935 | struct vmap_block *vb; | |
cf725ce2 | 936 | void *vaddr = NULL; |
db64fe02 NP |
937 | unsigned int order; |
938 | ||
891c49ab | 939 | BUG_ON(offset_in_page(size)); |
db64fe02 | 940 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
941 | if (WARN_ON(size == 0)) { |
942 | /* | |
943 | * Allocating 0 bytes isn't what caller wants since | |
944 | * get_order(0) returns funny result. Just warn and terminate | |
945 | * early. | |
946 | */ | |
947 | return NULL; | |
948 | } | |
db64fe02 NP |
949 | order = get_order(size); |
950 | ||
db64fe02 NP |
951 | rcu_read_lock(); |
952 | vbq = &get_cpu_var(vmap_block_queue); | |
953 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 954 | unsigned long pages_off; |
db64fe02 NP |
955 | |
956 | spin_lock(&vb->lock); | |
cf725ce2 RP |
957 | if (vb->free < (1UL << order)) { |
958 | spin_unlock(&vb->lock); | |
959 | continue; | |
960 | } | |
02b709df | 961 | |
cf725ce2 RP |
962 | pages_off = VMAP_BBMAP_BITS - vb->free; |
963 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
964 | vb->free -= 1UL << order; |
965 | if (vb->free == 0) { | |
966 | spin_lock(&vbq->lock); | |
967 | list_del_rcu(&vb->free_list); | |
968 | spin_unlock(&vbq->lock); | |
969 | } | |
cf725ce2 | 970 | |
02b709df NP |
971 | spin_unlock(&vb->lock); |
972 | break; | |
db64fe02 | 973 | } |
02b709df | 974 | |
3f04ba85 | 975 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
976 | rcu_read_unlock(); |
977 | ||
cf725ce2 RP |
978 | /* Allocate new block if nothing was found */ |
979 | if (!vaddr) | |
980 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 981 | |
cf725ce2 | 982 | return vaddr; |
db64fe02 NP |
983 | } |
984 | ||
985 | static void vb_free(const void *addr, unsigned long size) | |
986 | { | |
987 | unsigned long offset; | |
988 | unsigned long vb_idx; | |
989 | unsigned int order; | |
990 | struct vmap_block *vb; | |
991 | ||
891c49ab | 992 | BUG_ON(offset_in_page(size)); |
db64fe02 | 993 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc NP |
994 | |
995 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
996 | ||
db64fe02 NP |
997 | order = get_order(size); |
998 | ||
999 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
7d61bfe8 | 1000 | offset >>= PAGE_SHIFT; |
db64fe02 NP |
1001 | |
1002 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
1003 | rcu_read_lock(); | |
1004 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1005 | rcu_read_unlock(); | |
1006 | BUG_ON(!vb); | |
1007 | ||
64141da5 JF |
1008 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
1009 | ||
db64fe02 | 1010 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1011 | |
1012 | /* Expand dirty range */ | |
1013 | vb->dirty_min = min(vb->dirty_min, offset); | |
1014 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1015 | |
db64fe02 NP |
1016 | vb->dirty += 1UL << order; |
1017 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1018 | BUG_ON(vb->free); |
db64fe02 NP |
1019 | spin_unlock(&vb->lock); |
1020 | free_vmap_block(vb); | |
1021 | } else | |
1022 | spin_unlock(&vb->lock); | |
1023 | } | |
1024 | ||
1025 | /** | |
1026 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1027 | * | |
1028 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1029 | * to amortize TLB flushing overheads. What this means is that any page you | |
1030 | * have now, may, in a former life, have been mapped into kernel virtual | |
1031 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1032 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1033 | * | |
1034 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1035 | * be sure that none of the pages we have control over will have any aliases | |
1036 | * from the vmap layer. | |
1037 | */ | |
1038 | void vm_unmap_aliases(void) | |
1039 | { | |
1040 | unsigned long start = ULONG_MAX, end = 0; | |
1041 | int cpu; | |
1042 | int flush = 0; | |
1043 | ||
9b463334 JF |
1044 | if (unlikely(!vmap_initialized)) |
1045 | return; | |
1046 | ||
db64fe02 NP |
1047 | for_each_possible_cpu(cpu) { |
1048 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1049 | struct vmap_block *vb; | |
1050 | ||
1051 | rcu_read_lock(); | |
1052 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1053 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1054 | if (vb->dirty) { |
1055 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1056 | unsigned long s, e; |
b136be5e | 1057 | |
7d61bfe8 RP |
1058 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1059 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1060 | |
7d61bfe8 RP |
1061 | start = min(s, start); |
1062 | end = max(e, end); | |
db64fe02 | 1063 | |
7d61bfe8 | 1064 | flush = 1; |
db64fe02 NP |
1065 | } |
1066 | spin_unlock(&vb->lock); | |
1067 | } | |
1068 | rcu_read_unlock(); | |
1069 | } | |
1070 | ||
1071 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
1072 | } | |
1073 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
1074 | ||
1075 | /** | |
1076 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1077 | * @mem: the pointer returned by vm_map_ram | |
1078 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1079 | */ | |
1080 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1081 | { | |
1082 | unsigned long size = count << PAGE_SHIFT; | |
1083 | unsigned long addr = (unsigned long)mem; | |
1084 | ||
1085 | BUG_ON(!addr); | |
1086 | BUG_ON(addr < VMALLOC_START); | |
1087 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1088 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 NP |
1089 | |
1090 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 1091 | vmap_debug_free_range(addr, addr+size); |
db64fe02 NP |
1092 | |
1093 | if (likely(count <= VMAP_MAX_ALLOC)) | |
1094 | vb_free(mem, size); | |
1095 | else | |
1096 | free_unmap_vmap_area_addr(addr); | |
1097 | } | |
1098 | EXPORT_SYMBOL(vm_unmap_ram); | |
1099 | ||
1100 | /** | |
1101 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1102 | * @pages: an array of pointers to the pages to be mapped | |
1103 | * @count: number of pages | |
1104 | * @node: prefer to allocate data structures on this node | |
1105 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1106 | * |
36437638 GK |
1107 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1108 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1109 | * objects with vm_map_ram(), it could consume lots of address space through | |
1110 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1111 | * the end. Please use this function for short-lived objects. | |
1112 | * | |
e99c97ad | 1113 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1114 | */ |
1115 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1116 | { | |
1117 | unsigned long size = count << PAGE_SHIFT; | |
1118 | unsigned long addr; | |
1119 | void *mem; | |
1120 | ||
1121 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1122 | mem = vb_alloc(size, GFP_KERNEL); | |
1123 | if (IS_ERR(mem)) | |
1124 | return NULL; | |
1125 | addr = (unsigned long)mem; | |
1126 | } else { | |
1127 | struct vmap_area *va; | |
1128 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1129 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1130 | if (IS_ERR(va)) | |
1131 | return NULL; | |
1132 | ||
1133 | addr = va->va_start; | |
1134 | mem = (void *)addr; | |
1135 | } | |
1136 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1137 | vm_unmap_ram(mem, count); | |
1138 | return NULL; | |
1139 | } | |
1140 | return mem; | |
1141 | } | |
1142 | EXPORT_SYMBOL(vm_map_ram); | |
1143 | ||
4341fa45 | 1144 | static struct vm_struct *vmlist __initdata; |
be9b7335 NP |
1145 | /** |
1146 | * vm_area_add_early - add vmap area early during boot | |
1147 | * @vm: vm_struct to add | |
1148 | * | |
1149 | * This function is used to add fixed kernel vm area to vmlist before | |
1150 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1151 | * should contain proper values and the other fields should be zero. | |
1152 | * | |
1153 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1154 | */ | |
1155 | void __init vm_area_add_early(struct vm_struct *vm) | |
1156 | { | |
1157 | struct vm_struct *tmp, **p; | |
1158 | ||
1159 | BUG_ON(vmap_initialized); | |
1160 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1161 | if (tmp->addr >= vm->addr) { | |
1162 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1163 | break; | |
1164 | } else | |
1165 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1166 | } | |
1167 | vm->next = *p; | |
1168 | *p = vm; | |
1169 | } | |
1170 | ||
f0aa6617 TH |
1171 | /** |
1172 | * vm_area_register_early - register vmap area early during boot | |
1173 | * @vm: vm_struct to register | |
c0c0a293 | 1174 | * @align: requested alignment |
f0aa6617 TH |
1175 | * |
1176 | * This function is used to register kernel vm area before | |
1177 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1178 | * proper values on entry and other fields should be zero. On return, | |
1179 | * vm->addr contains the allocated address. | |
1180 | * | |
1181 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1182 | */ | |
c0c0a293 | 1183 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1184 | { |
1185 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1186 | unsigned long addr; |
1187 | ||
1188 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1189 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1190 | |
c0c0a293 | 1191 | vm->addr = (void *)addr; |
f0aa6617 | 1192 | |
be9b7335 | 1193 | vm_area_add_early(vm); |
f0aa6617 TH |
1194 | } |
1195 | ||
db64fe02 NP |
1196 | void __init vmalloc_init(void) |
1197 | { | |
822c18f2 IK |
1198 | struct vmap_area *va; |
1199 | struct vm_struct *tmp; | |
db64fe02 NP |
1200 | int i; |
1201 | ||
1202 | for_each_possible_cpu(i) { | |
1203 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1204 | struct vfree_deferred *p; |
db64fe02 NP |
1205 | |
1206 | vbq = &per_cpu(vmap_block_queue, i); | |
1207 | spin_lock_init(&vbq->lock); | |
1208 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1209 | p = &per_cpu(vfree_deferred, i); |
1210 | init_llist_head(&p->list); | |
1211 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1212 | } |
9b463334 | 1213 | |
822c18f2 IK |
1214 | /* Import existing vmlist entries. */ |
1215 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1216 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
dbda591d | 1217 | va->flags = VM_VM_AREA; |
822c18f2 IK |
1218 | va->va_start = (unsigned long)tmp->addr; |
1219 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1220 | va->vm = tmp; |
822c18f2 IK |
1221 | __insert_vmap_area(va); |
1222 | } | |
ca23e405 TH |
1223 | |
1224 | vmap_area_pcpu_hole = VMALLOC_END; | |
1225 | ||
9b463334 | 1226 | vmap_initialized = true; |
db64fe02 NP |
1227 | } |
1228 | ||
8fc48985 TH |
1229 | /** |
1230 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1231 | * @addr: start of the VM area to map | |
1232 | * @size: size of the VM area to map | |
1233 | * @prot: page protection flags to use | |
1234 | * @pages: pages to map | |
1235 | * | |
1236 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1237 | * specify should have been allocated using get_vm_area() and its | |
1238 | * friends. | |
1239 | * | |
1240 | * NOTE: | |
1241 | * This function does NOT do any cache flushing. The caller is | |
1242 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1243 | * before calling this function. | |
1244 | * | |
1245 | * RETURNS: | |
1246 | * The number of pages mapped on success, -errno on failure. | |
1247 | */ | |
1248 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1249 | pgprot_t prot, struct page **pages) | |
1250 | { | |
1251 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1252 | } | |
1253 | ||
1254 | /** | |
1255 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1256 | * @addr: start of the VM area to unmap | |
1257 | * @size: size of the VM area to unmap | |
1258 | * | |
1259 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1260 | * specify should have been allocated using get_vm_area() and its | |
1261 | * friends. | |
1262 | * | |
1263 | * NOTE: | |
1264 | * This function does NOT do any cache flushing. The caller is | |
1265 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1266 | * before calling this function and flush_tlb_kernel_range() after. | |
1267 | */ | |
1268 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1269 | { | |
1270 | vunmap_page_range(addr, addr + size); | |
1271 | } | |
81e88fdc | 1272 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1273 | |
1274 | /** | |
1275 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1276 | * @addr: start of the VM area to unmap | |
1277 | * @size: size of the VM area to unmap | |
1278 | * | |
1279 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1280 | * the unmapping and tlb after. | |
1281 | */ | |
db64fe02 NP |
1282 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1283 | { | |
1284 | unsigned long end = addr + size; | |
f6fcba70 TH |
1285 | |
1286 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1287 | vunmap_page_range(addr, end); |
1288 | flush_tlb_kernel_range(addr, end); | |
1289 | } | |
93ef6d6c | 1290 | EXPORT_SYMBOL_GPL(unmap_kernel_range); |
db64fe02 | 1291 | |
f6f8ed47 | 1292 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
1293 | { |
1294 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 1295 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
1296 | int err; |
1297 | ||
f6f8ed47 | 1298 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 1299 | |
f6f8ed47 | 1300 | return err > 0 ? 0 : err; |
db64fe02 NP |
1301 | } |
1302 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1303 | ||
f5252e00 | 1304 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1305 | unsigned long flags, const void *caller) |
cf88c790 | 1306 | { |
c69480ad | 1307 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
1308 | vm->flags = flags; |
1309 | vm->addr = (void *)va->va_start; | |
1310 | vm->size = va->va_end - va->va_start; | |
1311 | vm->caller = caller; | |
db1aecaf | 1312 | va->vm = vm; |
cf88c790 | 1313 | va->flags |= VM_VM_AREA; |
c69480ad | 1314 | spin_unlock(&vmap_area_lock); |
f5252e00 | 1315 | } |
cf88c790 | 1316 | |
20fc02b4 | 1317 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 1318 | { |
d4033afd | 1319 | /* |
20fc02b4 | 1320 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
1321 | * we should make sure that vm has proper values. |
1322 | * Pair with smp_rmb() in show_numa_info(). | |
1323 | */ | |
1324 | smp_wmb(); | |
20fc02b4 | 1325 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
1326 | } |
1327 | ||
db64fe02 | 1328 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 1329 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 1330 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 1331 | { |
0006526d | 1332 | struct vmap_area *va; |
db64fe02 | 1333 | struct vm_struct *area; |
1da177e4 | 1334 | |
52fd24ca | 1335 | BUG_ON(in_interrupt()); |
0f2d4a8e | 1336 | if (flags & VM_IOREMAP) |
0f616be1 TK |
1337 | align = 1ul << clamp_t(int, fls_long(size), |
1338 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
db64fe02 | 1339 | |
1da177e4 | 1340 | size = PAGE_ALIGN(size); |
31be8309 OH |
1341 | if (unlikely(!size)) |
1342 | return NULL; | |
1da177e4 | 1343 | |
cf88c790 | 1344 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1345 | if (unlikely(!area)) |
1346 | return NULL; | |
1347 | ||
71394fe5 AR |
1348 | if (!(flags & VM_NO_GUARD)) |
1349 | size += PAGE_SIZE; | |
1da177e4 | 1350 | |
db64fe02 NP |
1351 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1352 | if (IS_ERR(va)) { | |
1353 | kfree(area); | |
1354 | return NULL; | |
1da177e4 | 1355 | } |
1da177e4 | 1356 | |
d82b1d85 | 1357 | setup_vmalloc_vm(area, va, flags, caller); |
f5252e00 | 1358 | |
1da177e4 | 1359 | return area; |
1da177e4 LT |
1360 | } |
1361 | ||
930fc45a CL |
1362 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1363 | unsigned long start, unsigned long end) | |
1364 | { | |
00ef2d2f DR |
1365 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1366 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 1367 | } |
5992b6da | 1368 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1369 | |
c2968612 BH |
1370 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1371 | unsigned long start, unsigned long end, | |
5e6cafc8 | 1372 | const void *caller) |
c2968612 | 1373 | { |
00ef2d2f DR |
1374 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1375 | GFP_KERNEL, caller); | |
c2968612 BH |
1376 | } |
1377 | ||
1da177e4 | 1378 | /** |
183ff22b | 1379 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1380 | * @size: size of the area |
1381 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1382 | * | |
1383 | * Search an area of @size in the kernel virtual mapping area, | |
1384 | * and reserved it for out purposes. Returns the area descriptor | |
1385 | * on success or %NULL on failure. | |
1386 | */ | |
1387 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1388 | { | |
2dca6999 | 1389 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
1390 | NUMA_NO_NODE, GFP_KERNEL, |
1391 | __builtin_return_address(0)); | |
23016969 CL |
1392 | } |
1393 | ||
1394 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 1395 | const void *caller) |
23016969 | 1396 | { |
2dca6999 | 1397 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 1398 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
1399 | } |
1400 | ||
e9da6e99 MS |
1401 | /** |
1402 | * find_vm_area - find a continuous kernel virtual area | |
1403 | * @addr: base address | |
1404 | * | |
1405 | * Search for the kernel VM area starting at @addr, and return it. | |
1406 | * It is up to the caller to do all required locking to keep the returned | |
1407 | * pointer valid. | |
1408 | */ | |
1409 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 1410 | { |
db64fe02 | 1411 | struct vmap_area *va; |
83342314 | 1412 | |
db64fe02 NP |
1413 | va = find_vmap_area((unsigned long)addr); |
1414 | if (va && va->flags & VM_VM_AREA) | |
db1aecaf | 1415 | return va->vm; |
1da177e4 | 1416 | |
1da177e4 | 1417 | return NULL; |
1da177e4 LT |
1418 | } |
1419 | ||
7856dfeb | 1420 | /** |
183ff22b | 1421 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1422 | * @addr: base address |
1423 | * | |
1424 | * Search for the kernel VM area starting at @addr, and remove it. | |
1425 | * This function returns the found VM area, but using it is NOT safe | |
1426 | * on SMP machines, except for its size or flags. | |
1427 | */ | |
b3bdda02 | 1428 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1429 | { |
db64fe02 NP |
1430 | struct vmap_area *va; |
1431 | ||
1432 | va = find_vmap_area((unsigned long)addr); | |
1433 | if (va && va->flags & VM_VM_AREA) { | |
db1aecaf | 1434 | struct vm_struct *vm = va->vm; |
f5252e00 | 1435 | |
c69480ad JK |
1436 | spin_lock(&vmap_area_lock); |
1437 | va->vm = NULL; | |
1438 | va->flags &= ~VM_VM_AREA; | |
1439 | spin_unlock(&vmap_area_lock); | |
1440 | ||
dd32c279 | 1441 | vmap_debug_free_range(va->va_start, va->va_end); |
a5af5aa8 | 1442 | kasan_free_shadow(vm); |
dd32c279 | 1443 | free_unmap_vmap_area(va); |
dd32c279 | 1444 | |
db64fe02 NP |
1445 | return vm; |
1446 | } | |
1447 | return NULL; | |
7856dfeb AK |
1448 | } |
1449 | ||
b3bdda02 | 1450 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1451 | { |
1452 | struct vm_struct *area; | |
1453 | ||
1454 | if (!addr) | |
1455 | return; | |
1456 | ||
e69e9d4a | 1457 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 1458 | addr)) |
1da177e4 | 1459 | return; |
1da177e4 LT |
1460 | |
1461 | area = remove_vm_area(addr); | |
1462 | if (unlikely(!area)) { | |
4c8573e2 | 1463 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1464 | addr); |
1da177e4 LT |
1465 | return; |
1466 | } | |
1467 | ||
7511c3ed JM |
1468 | debug_check_no_locks_freed(addr, get_vm_area_size(area)); |
1469 | debug_check_no_obj_freed(addr, get_vm_area_size(area)); | |
9a11b49a | 1470 | |
1da177e4 LT |
1471 | if (deallocate_pages) { |
1472 | int i; | |
1473 | ||
1474 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1475 | struct page *page = area->pages[i]; |
1476 | ||
1477 | BUG_ON(!page); | |
37f08dda | 1478 | __free_kmem_pages(page, 0); |
1da177e4 LT |
1479 | } |
1480 | ||
244d63ee | 1481 | kvfree(area->pages); |
1da177e4 LT |
1482 | } |
1483 | ||
1484 | kfree(area); | |
1485 | return; | |
1486 | } | |
32fcfd40 | 1487 | |
1da177e4 LT |
1488 | /** |
1489 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1490 | * @addr: memory base address |
1491 | * | |
183ff22b | 1492 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1493 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1494 | * NULL, no operation is performed. | |
1da177e4 | 1495 | * |
32fcfd40 AV |
1496 | * Must not be called in NMI context (strictly speaking, only if we don't |
1497 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
1498 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 AM |
1499 | * |
1500 | * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node) | |
1da177e4 | 1501 | */ |
b3bdda02 | 1502 | void vfree(const void *addr) |
1da177e4 | 1503 | { |
32fcfd40 | 1504 | BUG_ON(in_nmi()); |
89219d37 CM |
1505 | |
1506 | kmemleak_free(addr); | |
1507 | ||
32fcfd40 AV |
1508 | if (!addr) |
1509 | return; | |
1510 | if (unlikely(in_interrupt())) { | |
7c8e0181 | 1511 | struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred); |
59d3132f ON |
1512 | if (llist_add((struct llist_node *)addr, &p->list)) |
1513 | schedule_work(&p->wq); | |
32fcfd40 AV |
1514 | } else |
1515 | __vunmap(addr, 1); | |
1da177e4 | 1516 | } |
1da177e4 LT |
1517 | EXPORT_SYMBOL(vfree); |
1518 | ||
1519 | /** | |
1520 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1521 | * @addr: memory base address |
1522 | * | |
1523 | * Free the virtually contiguous memory area starting at @addr, | |
1524 | * which was created from the page array passed to vmap(). | |
1525 | * | |
80e93eff | 1526 | * Must not be called in interrupt context. |
1da177e4 | 1527 | */ |
b3bdda02 | 1528 | void vunmap(const void *addr) |
1da177e4 LT |
1529 | { |
1530 | BUG_ON(in_interrupt()); | |
34754b69 | 1531 | might_sleep(); |
32fcfd40 AV |
1532 | if (addr) |
1533 | __vunmap(addr, 0); | |
1da177e4 | 1534 | } |
1da177e4 LT |
1535 | EXPORT_SYMBOL(vunmap); |
1536 | ||
1537 | /** | |
1538 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1539 | * @pages: array of page pointers |
1540 | * @count: number of pages to map | |
1541 | * @flags: vm_area->flags | |
1542 | * @prot: page protection for the mapping | |
1543 | * | |
1544 | * Maps @count pages from @pages into contiguous kernel virtual | |
1545 | * space. | |
1546 | */ | |
1547 | void *vmap(struct page **pages, unsigned int count, | |
1548 | unsigned long flags, pgprot_t prot) | |
1549 | { | |
1550 | struct vm_struct *area; | |
1551 | ||
34754b69 PZ |
1552 | might_sleep(); |
1553 | ||
4481374c | 1554 | if (count > totalram_pages) |
1da177e4 LT |
1555 | return NULL; |
1556 | ||
23016969 CL |
1557 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1558 | __builtin_return_address(0)); | |
1da177e4 LT |
1559 | if (!area) |
1560 | return NULL; | |
23016969 | 1561 | |
f6f8ed47 | 1562 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
1563 | vunmap(area->addr); |
1564 | return NULL; | |
1565 | } | |
1566 | ||
1567 | return area->addr; | |
1568 | } | |
1da177e4 LT |
1569 | EXPORT_SYMBOL(vmap); |
1570 | ||
2dca6999 DM |
1571 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
1572 | gfp_t gfp_mask, pgprot_t prot, | |
5e6cafc8 | 1573 | int node, const void *caller); |
e31d9eb5 | 1574 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 1575 | pgprot_t prot, int node) |
1da177e4 | 1576 | { |
22943ab1 | 1577 | const int order = 0; |
1da177e4 LT |
1578 | struct page **pages; |
1579 | unsigned int nr_pages, array_size, i; | |
930f036b DR |
1580 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
1581 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; | |
1da177e4 | 1582 | |
762216ab | 1583 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
1584 | array_size = (nr_pages * sizeof(struct page *)); |
1585 | ||
1586 | area->nr_pages = nr_pages; | |
1587 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1588 | if (array_size > PAGE_SIZE) { |
976d6dfb | 1589 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
3722e13c | 1590 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 1591 | } else { |
976d6dfb | 1592 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 1593 | } |
1da177e4 LT |
1594 | area->pages = pages; |
1595 | if (!area->pages) { | |
1596 | remove_vm_area(area->addr); | |
1597 | kfree(area); | |
1598 | return NULL; | |
1599 | } | |
1da177e4 LT |
1600 | |
1601 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1602 | struct page *page; |
1603 | ||
4b90951c | 1604 | if (node == NUMA_NO_NODE) |
37f08dda | 1605 | page = alloc_kmem_pages(alloc_mask, order); |
930fc45a | 1606 | else |
37f08dda | 1607 | page = alloc_kmem_pages_node(node, alloc_mask, order); |
bf53d6f8 CL |
1608 | |
1609 | if (unlikely(!page)) { | |
1da177e4 LT |
1610 | /* Successfully allocated i pages, free them in __vunmap() */ |
1611 | area->nr_pages = i; | |
1612 | goto fail; | |
1613 | } | |
bf53d6f8 | 1614 | area->pages[i] = page; |
d0164adc | 1615 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 1616 | cond_resched(); |
1da177e4 LT |
1617 | } |
1618 | ||
f6f8ed47 | 1619 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
1620 | goto fail; |
1621 | return area->addr; | |
1622 | ||
1623 | fail: | |
3ee9a4f0 JP |
1624 | warn_alloc_failed(gfp_mask, order, |
1625 | "vmalloc: allocation failure, allocated %ld of %ld bytes\n", | |
22943ab1 | 1626 | (area->nr_pages*PAGE_SIZE), area->size); |
1da177e4 LT |
1627 | vfree(area->addr); |
1628 | return NULL; | |
1629 | } | |
1630 | ||
1631 | /** | |
d0a21265 | 1632 | * __vmalloc_node_range - allocate virtually contiguous memory |
1da177e4 | 1633 | * @size: allocation size |
2dca6999 | 1634 | * @align: desired alignment |
d0a21265 DR |
1635 | * @start: vm area range start |
1636 | * @end: vm area range end | |
1da177e4 LT |
1637 | * @gfp_mask: flags for the page level allocator |
1638 | * @prot: protection mask for the allocated pages | |
cb9e3c29 | 1639 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) |
00ef2d2f | 1640 | * @node: node to use for allocation or NUMA_NO_NODE |
c85d194b | 1641 | * @caller: caller's return address |
1da177e4 LT |
1642 | * |
1643 | * Allocate enough pages to cover @size from the page level | |
1644 | * allocator with @gfp_mask flags. Map them into contiguous | |
1645 | * kernel virtual space, using a pagetable protection of @prot. | |
1646 | */ | |
d0a21265 DR |
1647 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
1648 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
1649 | pgprot_t prot, unsigned long vm_flags, int node, |
1650 | const void *caller) | |
1da177e4 LT |
1651 | { |
1652 | struct vm_struct *area; | |
89219d37 CM |
1653 | void *addr; |
1654 | unsigned long real_size = size; | |
1da177e4 LT |
1655 | |
1656 | size = PAGE_ALIGN(size); | |
4481374c | 1657 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
de7d2b56 | 1658 | goto fail; |
1da177e4 | 1659 | |
cb9e3c29 AR |
1660 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | |
1661 | vm_flags, start, end, node, gfp_mask, caller); | |
1da177e4 | 1662 | if (!area) |
de7d2b56 | 1663 | goto fail; |
1da177e4 | 1664 | |
3722e13c | 1665 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 1666 | if (!addr) |
b82225f3 | 1667 | return NULL; |
89219d37 | 1668 | |
f5252e00 | 1669 | /* |
20fc02b4 ZY |
1670 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
1671 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 1672 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 1673 | */ |
20fc02b4 | 1674 | clear_vm_uninitialized_flag(area); |
f5252e00 | 1675 | |
89219d37 | 1676 | /* |
7f88f88f CM |
1677 | * A ref_count = 2 is needed because vm_struct allocated in |
1678 | * __get_vm_area_node() contains a reference to the virtual address of | |
1679 | * the vmalloc'ed block. | |
89219d37 | 1680 | */ |
7f88f88f | 1681 | kmemleak_alloc(addr, real_size, 2, gfp_mask); |
89219d37 CM |
1682 | |
1683 | return addr; | |
de7d2b56 JP |
1684 | |
1685 | fail: | |
1686 | warn_alloc_failed(gfp_mask, 0, | |
1687 | "vmalloc: allocation failure: %lu bytes\n", | |
1688 | real_size); | |
1689 | return NULL; | |
1da177e4 LT |
1690 | } |
1691 | ||
d0a21265 DR |
1692 | /** |
1693 | * __vmalloc_node - allocate virtually contiguous memory | |
1694 | * @size: allocation size | |
1695 | * @align: desired alignment | |
1696 | * @gfp_mask: flags for the page level allocator | |
1697 | * @prot: protection mask for the allocated pages | |
00ef2d2f | 1698 | * @node: node to use for allocation or NUMA_NO_NODE |
d0a21265 DR |
1699 | * @caller: caller's return address |
1700 | * | |
1701 | * Allocate enough pages to cover @size from the page level | |
1702 | * allocator with @gfp_mask flags. Map them into contiguous | |
1703 | * kernel virtual space, using a pagetable protection of @prot. | |
1704 | */ | |
1705 | static void *__vmalloc_node(unsigned long size, unsigned long align, | |
1706 | gfp_t gfp_mask, pgprot_t prot, | |
5e6cafc8 | 1707 | int node, const void *caller) |
d0a21265 DR |
1708 | { |
1709 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 1710 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
1711 | } |
1712 | ||
930fc45a CL |
1713 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1714 | { | |
00ef2d2f | 1715 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 1716 | __builtin_return_address(0)); |
930fc45a | 1717 | } |
1da177e4 LT |
1718 | EXPORT_SYMBOL(__vmalloc); |
1719 | ||
e1ca7788 DY |
1720 | static inline void *__vmalloc_node_flags(unsigned long size, |
1721 | int node, gfp_t flags) | |
1722 | { | |
1723 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
1724 | node, __builtin_return_address(0)); | |
1725 | } | |
1726 | ||
1da177e4 LT |
1727 | /** |
1728 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1729 | * @size: allocation size |
1da177e4 LT |
1730 | * Allocate enough pages to cover @size from the page level |
1731 | * allocator and map them into contiguous kernel virtual space. | |
1732 | * | |
c1c8897f | 1733 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1734 | * use __vmalloc() instead. |
1735 | */ | |
1736 | void *vmalloc(unsigned long size) | |
1737 | { | |
00ef2d2f DR |
1738 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
1739 | GFP_KERNEL | __GFP_HIGHMEM); | |
1da177e4 | 1740 | } |
1da177e4 LT |
1741 | EXPORT_SYMBOL(vmalloc); |
1742 | ||
e1ca7788 DY |
1743 | /** |
1744 | * vzalloc - allocate virtually contiguous memory with zero fill | |
1745 | * @size: allocation size | |
1746 | * Allocate enough pages to cover @size from the page level | |
1747 | * allocator and map them into contiguous kernel virtual space. | |
1748 | * The memory allocated is set to zero. | |
1749 | * | |
1750 | * For tight control over page level allocator and protection flags | |
1751 | * use __vmalloc() instead. | |
1752 | */ | |
1753 | void *vzalloc(unsigned long size) | |
1754 | { | |
00ef2d2f | 1755 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
e1ca7788 DY |
1756 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
1757 | } | |
1758 | EXPORT_SYMBOL(vzalloc); | |
1759 | ||
83342314 | 1760 | /** |
ead04089 REB |
1761 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1762 | * @size: allocation size | |
83342314 | 1763 | * |
ead04089 REB |
1764 | * The resulting memory area is zeroed so it can be mapped to userspace |
1765 | * without leaking data. | |
83342314 NP |
1766 | */ |
1767 | void *vmalloc_user(unsigned long size) | |
1768 | { | |
1769 | struct vm_struct *area; | |
1770 | void *ret; | |
1771 | ||
2dca6999 DM |
1772 | ret = __vmalloc_node(size, SHMLBA, |
1773 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
00ef2d2f DR |
1774 | PAGE_KERNEL, NUMA_NO_NODE, |
1775 | __builtin_return_address(0)); | |
2b4ac44e | 1776 | if (ret) { |
db64fe02 | 1777 | area = find_vm_area(ret); |
2b4ac44e | 1778 | area->flags |= VM_USERMAP; |
2b4ac44e | 1779 | } |
83342314 NP |
1780 | return ret; |
1781 | } | |
1782 | EXPORT_SYMBOL(vmalloc_user); | |
1783 | ||
930fc45a CL |
1784 | /** |
1785 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1786 | * @size: allocation size |
d44e0780 | 1787 | * @node: numa node |
930fc45a CL |
1788 | * |
1789 | * Allocate enough pages to cover @size from the page level | |
1790 | * allocator and map them into contiguous kernel virtual space. | |
1791 | * | |
c1c8897f | 1792 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1793 | * use __vmalloc() instead. |
1794 | */ | |
1795 | void *vmalloc_node(unsigned long size, int node) | |
1796 | { | |
2dca6999 | 1797 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
23016969 | 1798 | node, __builtin_return_address(0)); |
930fc45a CL |
1799 | } |
1800 | EXPORT_SYMBOL(vmalloc_node); | |
1801 | ||
e1ca7788 DY |
1802 | /** |
1803 | * vzalloc_node - allocate memory on a specific node with zero fill | |
1804 | * @size: allocation size | |
1805 | * @node: numa node | |
1806 | * | |
1807 | * Allocate enough pages to cover @size from the page level | |
1808 | * allocator and map them into contiguous kernel virtual space. | |
1809 | * The memory allocated is set to zero. | |
1810 | * | |
1811 | * For tight control over page level allocator and protection flags | |
1812 | * use __vmalloc_node() instead. | |
1813 | */ | |
1814 | void *vzalloc_node(unsigned long size, int node) | |
1815 | { | |
1816 | return __vmalloc_node_flags(size, node, | |
1817 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); | |
1818 | } | |
1819 | EXPORT_SYMBOL(vzalloc_node); | |
1820 | ||
4dc3b16b PP |
1821 | #ifndef PAGE_KERNEL_EXEC |
1822 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1823 | #endif | |
1824 | ||
1da177e4 LT |
1825 | /** |
1826 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1827 | * @size: allocation size |
1828 | * | |
1829 | * Kernel-internal function to allocate enough pages to cover @size | |
1830 | * the page level allocator and map them into contiguous and | |
1831 | * executable kernel virtual space. | |
1832 | * | |
c1c8897f | 1833 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1834 | * use __vmalloc() instead. |
1835 | */ | |
1836 | ||
1da177e4 LT |
1837 | void *vmalloc_exec(unsigned long size) |
1838 | { | |
2dca6999 | 1839 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
00ef2d2f | 1840 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 LT |
1841 | } |
1842 | ||
0d08e0d3 | 1843 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1844 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1845 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1846 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1847 | #else |
1848 | #define GFP_VMALLOC32 GFP_KERNEL | |
1849 | #endif | |
1850 | ||
1da177e4 LT |
1851 | /** |
1852 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1853 | * @size: allocation size |
1854 | * | |
1855 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1856 | * page level allocator and map them into contiguous kernel virtual space. | |
1857 | */ | |
1858 | void *vmalloc_32(unsigned long size) | |
1859 | { | |
2dca6999 | 1860 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 1861 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 1862 | } |
1da177e4 LT |
1863 | EXPORT_SYMBOL(vmalloc_32); |
1864 | ||
83342314 | 1865 | /** |
ead04089 | 1866 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1867 | * @size: allocation size |
ead04089 REB |
1868 | * |
1869 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1870 | * mapped to userspace without leaking data. | |
83342314 NP |
1871 | */ |
1872 | void *vmalloc_32_user(unsigned long size) | |
1873 | { | |
1874 | struct vm_struct *area; | |
1875 | void *ret; | |
1876 | ||
2dca6999 | 1877 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
00ef2d2f | 1878 | NUMA_NO_NODE, __builtin_return_address(0)); |
2b4ac44e | 1879 | if (ret) { |
db64fe02 | 1880 | area = find_vm_area(ret); |
2b4ac44e | 1881 | area->flags |= VM_USERMAP; |
2b4ac44e | 1882 | } |
83342314 NP |
1883 | return ret; |
1884 | } | |
1885 | EXPORT_SYMBOL(vmalloc_32_user); | |
1886 | ||
d0107eb0 KH |
1887 | /* |
1888 | * small helper routine , copy contents to buf from addr. | |
1889 | * If the page is not present, fill zero. | |
1890 | */ | |
1891 | ||
1892 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
1893 | { | |
1894 | struct page *p; | |
1895 | int copied = 0; | |
1896 | ||
1897 | while (count) { | |
1898 | unsigned long offset, length; | |
1899 | ||
891c49ab | 1900 | offset = offset_in_page(addr); |
d0107eb0 KH |
1901 | length = PAGE_SIZE - offset; |
1902 | if (length > count) | |
1903 | length = count; | |
1904 | p = vmalloc_to_page(addr); | |
1905 | /* | |
1906 | * To do safe access to this _mapped_ area, we need | |
1907 | * lock. But adding lock here means that we need to add | |
1908 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1909 | * interface, rarely used. Instead of that, we'll use | |
1910 | * kmap() and get small overhead in this access function. | |
1911 | */ | |
1912 | if (p) { | |
1913 | /* | |
1914 | * we can expect USER0 is not used (see vread/vwrite's | |
1915 | * function description) | |
1916 | */ | |
9b04c5fe | 1917 | void *map = kmap_atomic(p); |
d0107eb0 | 1918 | memcpy(buf, map + offset, length); |
9b04c5fe | 1919 | kunmap_atomic(map); |
d0107eb0 KH |
1920 | } else |
1921 | memset(buf, 0, length); | |
1922 | ||
1923 | addr += length; | |
1924 | buf += length; | |
1925 | copied += length; | |
1926 | count -= length; | |
1927 | } | |
1928 | return copied; | |
1929 | } | |
1930 | ||
1931 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
1932 | { | |
1933 | struct page *p; | |
1934 | int copied = 0; | |
1935 | ||
1936 | while (count) { | |
1937 | unsigned long offset, length; | |
1938 | ||
891c49ab | 1939 | offset = offset_in_page(addr); |
d0107eb0 KH |
1940 | length = PAGE_SIZE - offset; |
1941 | if (length > count) | |
1942 | length = count; | |
1943 | p = vmalloc_to_page(addr); | |
1944 | /* | |
1945 | * To do safe access to this _mapped_ area, we need | |
1946 | * lock. But adding lock here means that we need to add | |
1947 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1948 | * interface, rarely used. Instead of that, we'll use | |
1949 | * kmap() and get small overhead in this access function. | |
1950 | */ | |
1951 | if (p) { | |
1952 | /* | |
1953 | * we can expect USER0 is not used (see vread/vwrite's | |
1954 | * function description) | |
1955 | */ | |
9b04c5fe | 1956 | void *map = kmap_atomic(p); |
d0107eb0 | 1957 | memcpy(map + offset, buf, length); |
9b04c5fe | 1958 | kunmap_atomic(map); |
d0107eb0 KH |
1959 | } |
1960 | addr += length; | |
1961 | buf += length; | |
1962 | copied += length; | |
1963 | count -= length; | |
1964 | } | |
1965 | return copied; | |
1966 | } | |
1967 | ||
1968 | /** | |
1969 | * vread() - read vmalloc area in a safe way. | |
1970 | * @buf: buffer for reading data | |
1971 | * @addr: vm address. | |
1972 | * @count: number of bytes to be read. | |
1973 | * | |
1974 | * Returns # of bytes which addr and buf should be increased. | |
1975 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't | |
1976 | * includes any intersect with alive vmalloc area. | |
1977 | * | |
1978 | * This function checks that addr is a valid vmalloc'ed area, and | |
1979 | * copy data from that area to a given buffer. If the given memory range | |
1980 | * of [addr...addr+count) includes some valid address, data is copied to | |
1981 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
1982 | * IOREMAP area is treated as memory hole and no copy is done. | |
1983 | * | |
1984 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 1985 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
1986 | * |
1987 | * Note: In usual ops, vread() is never necessary because the caller | |
1988 | * should know vmalloc() area is valid and can use memcpy(). | |
1989 | * This is for routines which have to access vmalloc area without | |
1990 | * any informaion, as /dev/kmem. | |
1991 | * | |
1992 | */ | |
1993 | ||
1da177e4 LT |
1994 | long vread(char *buf, char *addr, unsigned long count) |
1995 | { | |
e81ce85f JK |
1996 | struct vmap_area *va; |
1997 | struct vm_struct *vm; | |
1da177e4 | 1998 | char *vaddr, *buf_start = buf; |
d0107eb0 | 1999 | unsigned long buflen = count; |
1da177e4 LT |
2000 | unsigned long n; |
2001 | ||
2002 | /* Don't allow overflow */ | |
2003 | if ((unsigned long) addr + count < count) | |
2004 | count = -(unsigned long) addr; | |
2005 | ||
e81ce85f JK |
2006 | spin_lock(&vmap_area_lock); |
2007 | list_for_each_entry(va, &vmap_area_list, list) { | |
2008 | if (!count) | |
2009 | break; | |
2010 | ||
2011 | if (!(va->flags & VM_VM_AREA)) | |
2012 | continue; | |
2013 | ||
2014 | vm = va->vm; | |
2015 | vaddr = (char *) vm->addr; | |
762216ab | 2016 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2017 | continue; |
2018 | while (addr < vaddr) { | |
2019 | if (count == 0) | |
2020 | goto finished; | |
2021 | *buf = '\0'; | |
2022 | buf++; | |
2023 | addr++; | |
2024 | count--; | |
2025 | } | |
762216ab | 2026 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2027 | if (n > count) |
2028 | n = count; | |
e81ce85f | 2029 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2030 | aligned_vread(buf, addr, n); |
2031 | else /* IOREMAP area is treated as memory hole */ | |
2032 | memset(buf, 0, n); | |
2033 | buf += n; | |
2034 | addr += n; | |
2035 | count -= n; | |
1da177e4 LT |
2036 | } |
2037 | finished: | |
e81ce85f | 2038 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2039 | |
2040 | if (buf == buf_start) | |
2041 | return 0; | |
2042 | /* zero-fill memory holes */ | |
2043 | if (buf != buf_start + buflen) | |
2044 | memset(buf, 0, buflen - (buf - buf_start)); | |
2045 | ||
2046 | return buflen; | |
1da177e4 LT |
2047 | } |
2048 | ||
d0107eb0 KH |
2049 | /** |
2050 | * vwrite() - write vmalloc area in a safe way. | |
2051 | * @buf: buffer for source data | |
2052 | * @addr: vm address. | |
2053 | * @count: number of bytes to be read. | |
2054 | * | |
2055 | * Returns # of bytes which addr and buf should be incresed. | |
2056 | * (same number to @count). | |
2057 | * If [addr...addr+count) doesn't includes any intersect with valid | |
2058 | * vmalloc area, returns 0. | |
2059 | * | |
2060 | * This function checks that addr is a valid vmalloc'ed area, and | |
2061 | * copy data from a buffer to the given addr. If specified range of | |
2062 | * [addr...addr+count) includes some valid address, data is copied from | |
2063 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2064 | * IOREMAP area is treated as memory hole and no copy is done. | |
2065 | * | |
2066 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 2067 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
2068 | * |
2069 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2070 | * should know vmalloc() area is valid and can use memcpy(). | |
2071 | * This is for routines which have to access vmalloc area without | |
2072 | * any informaion, as /dev/kmem. | |
d0107eb0 KH |
2073 | */ |
2074 | ||
1da177e4 LT |
2075 | long vwrite(char *buf, char *addr, unsigned long count) |
2076 | { | |
e81ce85f JK |
2077 | struct vmap_area *va; |
2078 | struct vm_struct *vm; | |
d0107eb0 KH |
2079 | char *vaddr; |
2080 | unsigned long n, buflen; | |
2081 | int copied = 0; | |
1da177e4 LT |
2082 | |
2083 | /* Don't allow overflow */ | |
2084 | if ((unsigned long) addr + count < count) | |
2085 | count = -(unsigned long) addr; | |
d0107eb0 | 2086 | buflen = count; |
1da177e4 | 2087 | |
e81ce85f JK |
2088 | spin_lock(&vmap_area_lock); |
2089 | list_for_each_entry(va, &vmap_area_list, list) { | |
2090 | if (!count) | |
2091 | break; | |
2092 | ||
2093 | if (!(va->flags & VM_VM_AREA)) | |
2094 | continue; | |
2095 | ||
2096 | vm = va->vm; | |
2097 | vaddr = (char *) vm->addr; | |
762216ab | 2098 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2099 | continue; |
2100 | while (addr < vaddr) { | |
2101 | if (count == 0) | |
2102 | goto finished; | |
2103 | buf++; | |
2104 | addr++; | |
2105 | count--; | |
2106 | } | |
762216ab | 2107 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2108 | if (n > count) |
2109 | n = count; | |
e81ce85f | 2110 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2111 | aligned_vwrite(buf, addr, n); |
2112 | copied++; | |
2113 | } | |
2114 | buf += n; | |
2115 | addr += n; | |
2116 | count -= n; | |
1da177e4 LT |
2117 | } |
2118 | finished: | |
e81ce85f | 2119 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2120 | if (!copied) |
2121 | return 0; | |
2122 | return buflen; | |
1da177e4 | 2123 | } |
83342314 NP |
2124 | |
2125 | /** | |
e69e9d4a HD |
2126 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
2127 | * @vma: vma to cover | |
2128 | * @uaddr: target user address to start at | |
2129 | * @kaddr: virtual address of vmalloc kernel memory | |
2130 | * @size: size of map area | |
7682486b RD |
2131 | * |
2132 | * Returns: 0 for success, -Exxx on failure | |
83342314 | 2133 | * |
e69e9d4a HD |
2134 | * This function checks that @kaddr is a valid vmalloc'ed area, |
2135 | * and that it is big enough to cover the range starting at | |
2136 | * @uaddr in @vma. Will return failure if that criteria isn't | |
2137 | * met. | |
83342314 | 2138 | * |
72fd4a35 | 2139 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 2140 | */ |
e69e9d4a HD |
2141 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
2142 | void *kaddr, unsigned long size) | |
83342314 NP |
2143 | { |
2144 | struct vm_struct *area; | |
83342314 | 2145 | |
e69e9d4a HD |
2146 | size = PAGE_ALIGN(size); |
2147 | ||
2148 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
2149 | return -EINVAL; |
2150 | ||
e69e9d4a | 2151 | area = find_vm_area(kaddr); |
83342314 | 2152 | if (!area) |
db64fe02 | 2153 | return -EINVAL; |
83342314 NP |
2154 | |
2155 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2156 | return -EINVAL; |
83342314 | 2157 | |
e69e9d4a | 2158 | if (kaddr + size > area->addr + area->size) |
db64fe02 | 2159 | return -EINVAL; |
83342314 | 2160 | |
83342314 | 2161 | do { |
e69e9d4a | 2162 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
2163 | int ret; |
2164 | ||
83342314 NP |
2165 | ret = vm_insert_page(vma, uaddr, page); |
2166 | if (ret) | |
2167 | return ret; | |
2168 | ||
2169 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
2170 | kaddr += PAGE_SIZE; |
2171 | size -= PAGE_SIZE; | |
2172 | } while (size > 0); | |
83342314 | 2173 | |
314e51b9 | 2174 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 2175 | |
db64fe02 | 2176 | return 0; |
83342314 | 2177 | } |
e69e9d4a HD |
2178 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
2179 | ||
2180 | /** | |
2181 | * remap_vmalloc_range - map vmalloc pages to userspace | |
2182 | * @vma: vma to cover (map full range of vma) | |
2183 | * @addr: vmalloc memory | |
2184 | * @pgoff: number of pages into addr before first page to map | |
2185 | * | |
2186 | * Returns: 0 for success, -Exxx on failure | |
2187 | * | |
2188 | * This function checks that addr is a valid vmalloc'ed area, and | |
2189 | * that it is big enough to cover the vma. Will return failure if | |
2190 | * that criteria isn't met. | |
2191 | * | |
2192 | * Similar to remap_pfn_range() (see mm/memory.c) | |
2193 | */ | |
2194 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
2195 | unsigned long pgoff) | |
2196 | { | |
2197 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
2198 | addr + (pgoff << PAGE_SHIFT), | |
2199 | vma->vm_end - vma->vm_start); | |
2200 | } | |
83342314 NP |
2201 | EXPORT_SYMBOL(remap_vmalloc_range); |
2202 | ||
1eeb66a1 CH |
2203 | /* |
2204 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
2205 | * have one. | |
2206 | */ | |
3b32123d | 2207 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
2208 | { |
2209 | } | |
5f4352fb JF |
2210 | |
2211 | ||
2f569afd | 2212 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb | 2213 | { |
cd12909c DV |
2214 | pte_t ***p = data; |
2215 | ||
2216 | if (p) { | |
2217 | *(*p) = pte; | |
2218 | (*p)++; | |
2219 | } | |
5f4352fb JF |
2220 | return 0; |
2221 | } | |
2222 | ||
2223 | /** | |
2224 | * alloc_vm_area - allocate a range of kernel address space | |
2225 | * @size: size of the area | |
cd12909c | 2226 | * @ptes: returns the PTEs for the address space |
7682486b RD |
2227 | * |
2228 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
2229 | * |
2230 | * This function reserves a range of kernel address space, and | |
2231 | * allocates pagetables to map that range. No actual mappings | |
cd12909c DV |
2232 | * are created. |
2233 | * | |
2234 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) | |
2235 | * allocated for the VM area are returned. | |
5f4352fb | 2236 | */ |
cd12909c | 2237 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
2238 | { |
2239 | struct vm_struct *area; | |
2240 | ||
23016969 CL |
2241 | area = get_vm_area_caller(size, VM_IOREMAP, |
2242 | __builtin_return_address(0)); | |
5f4352fb JF |
2243 | if (area == NULL) |
2244 | return NULL; | |
2245 | ||
2246 | /* | |
2247 | * This ensures that page tables are constructed for this region | |
2248 | * of kernel virtual address space and mapped into init_mm. | |
2249 | */ | |
2250 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 2251 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
2252 | free_vm_area(area); |
2253 | return NULL; | |
2254 | } | |
2255 | ||
5f4352fb JF |
2256 | return area; |
2257 | } | |
2258 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
2259 | ||
2260 | void free_vm_area(struct vm_struct *area) | |
2261 | { | |
2262 | struct vm_struct *ret; | |
2263 | ret = remove_vm_area(area->addr); | |
2264 | BUG_ON(ret != area); | |
2265 | kfree(area); | |
2266 | } | |
2267 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 2268 | |
4f8b02b4 | 2269 | #ifdef CONFIG_SMP |
ca23e405 TH |
2270 | static struct vmap_area *node_to_va(struct rb_node *n) |
2271 | { | |
2272 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; | |
2273 | } | |
2274 | ||
2275 | /** | |
2276 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
2277 | * @end: target address | |
2278 | * @pnext: out arg for the next vmap_area | |
2279 | * @pprev: out arg for the previous vmap_area | |
2280 | * | |
2281 | * Returns: %true if either or both of next and prev are found, | |
2282 | * %false if no vmap_area exists | |
2283 | * | |
2284 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2285 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2286 | */ | |
2287 | static bool pvm_find_next_prev(unsigned long end, | |
2288 | struct vmap_area **pnext, | |
2289 | struct vmap_area **pprev) | |
2290 | { | |
2291 | struct rb_node *n = vmap_area_root.rb_node; | |
2292 | struct vmap_area *va = NULL; | |
2293 | ||
2294 | while (n) { | |
2295 | va = rb_entry(n, struct vmap_area, rb_node); | |
2296 | if (end < va->va_end) | |
2297 | n = n->rb_left; | |
2298 | else if (end > va->va_end) | |
2299 | n = n->rb_right; | |
2300 | else | |
2301 | break; | |
2302 | } | |
2303 | ||
2304 | if (!va) | |
2305 | return false; | |
2306 | ||
2307 | if (va->va_end > end) { | |
2308 | *pnext = va; | |
2309 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2310 | } else { | |
2311 | *pprev = va; | |
2312 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2313 | } | |
2314 | return true; | |
2315 | } | |
2316 | ||
2317 | /** | |
2318 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2319 | * @pnext: in/out arg for the next vmap_area | |
2320 | * @pprev: in/out arg for the previous vmap_area | |
2321 | * @align: alignment | |
2322 | * | |
2323 | * Returns: determined end address | |
2324 | * | |
2325 | * Find the highest aligned address between *@pnext and *@pprev below | |
2326 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2327 | * down address is between the end addresses of the two vmap_areas. | |
2328 | * | |
2329 | * Please note that the address returned by this function may fall | |
2330 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2331 | * that. | |
2332 | */ | |
2333 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2334 | struct vmap_area **pprev, | |
2335 | unsigned long align) | |
2336 | { | |
2337 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2338 | unsigned long addr; | |
2339 | ||
2340 | if (*pnext) | |
2341 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2342 | else | |
2343 | addr = vmalloc_end; | |
2344 | ||
2345 | while (*pprev && (*pprev)->va_end > addr) { | |
2346 | *pnext = *pprev; | |
2347 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2348 | } | |
2349 | ||
2350 | return addr; | |
2351 | } | |
2352 | ||
2353 | /** | |
2354 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2355 | * @offsets: array containing offset of each area | |
2356 | * @sizes: array containing size of each area | |
2357 | * @nr_vms: the number of areas to allocate | |
2358 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
2359 | * |
2360 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2361 | * vm_structs on success, %NULL on failure | |
2362 | * | |
2363 | * Percpu allocator wants to use congruent vm areas so that it can | |
2364 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
2365 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
2366 | * be scattered pretty far, distance between two areas easily going up | |
2367 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
2368 | * areas are allocated from top. | |
ca23e405 TH |
2369 | * |
2370 | * Despite its complicated look, this allocator is rather simple. It | |
2371 | * does everything top-down and scans areas from the end looking for | |
2372 | * matching slot. While scanning, if any of the areas overlaps with | |
2373 | * existing vmap_area, the base address is pulled down to fit the | |
2374 | * area. Scanning is repeated till all the areas fit and then all | |
2375 | * necessary data structres are inserted and the result is returned. | |
2376 | */ | |
2377 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2378 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 2379 | size_t align) |
ca23e405 TH |
2380 | { |
2381 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2382 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2383 | struct vmap_area **vas, *prev, *next; | |
2384 | struct vm_struct **vms; | |
2385 | int area, area2, last_area, term_area; | |
2386 | unsigned long base, start, end, last_end; | |
2387 | bool purged = false; | |
2388 | ||
ca23e405 | 2389 | /* verify parameters and allocate data structures */ |
891c49ab | 2390 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
2391 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
2392 | start = offsets[area]; | |
2393 | end = start + sizes[area]; | |
2394 | ||
2395 | /* is everything aligned properly? */ | |
2396 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2397 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2398 | ||
2399 | /* detect the area with the highest address */ | |
2400 | if (start > offsets[last_area]) | |
2401 | last_area = area; | |
2402 | ||
2403 | for (area2 = 0; area2 < nr_vms; area2++) { | |
2404 | unsigned long start2 = offsets[area2]; | |
2405 | unsigned long end2 = start2 + sizes[area2]; | |
2406 | ||
2407 | if (area2 == area) | |
2408 | continue; | |
2409 | ||
2410 | BUG_ON(start2 >= start && start2 < end); | |
2411 | BUG_ON(end2 <= end && end2 > start); | |
2412 | } | |
2413 | } | |
2414 | last_end = offsets[last_area] + sizes[last_area]; | |
2415 | ||
2416 | if (vmalloc_end - vmalloc_start < last_end) { | |
2417 | WARN_ON(true); | |
2418 | return NULL; | |
2419 | } | |
2420 | ||
4d67d860 TM |
2421 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
2422 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 2423 | if (!vas || !vms) |
f1db7afd | 2424 | goto err_free2; |
ca23e405 TH |
2425 | |
2426 | for (area = 0; area < nr_vms; area++) { | |
ec3f64fc DR |
2427 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
2428 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); | |
ca23e405 TH |
2429 | if (!vas[area] || !vms[area]) |
2430 | goto err_free; | |
2431 | } | |
2432 | retry: | |
2433 | spin_lock(&vmap_area_lock); | |
2434 | ||
2435 | /* start scanning - we scan from the top, begin with the last area */ | |
2436 | area = term_area = last_area; | |
2437 | start = offsets[area]; | |
2438 | end = start + sizes[area]; | |
2439 | ||
2440 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2441 | base = vmalloc_end - last_end; | |
2442 | goto found; | |
2443 | } | |
2444 | base = pvm_determine_end(&next, &prev, align) - end; | |
2445 | ||
2446 | while (true) { | |
2447 | BUG_ON(next && next->va_end <= base + end); | |
2448 | BUG_ON(prev && prev->va_end > base + end); | |
2449 | ||
2450 | /* | |
2451 | * base might have underflowed, add last_end before | |
2452 | * comparing. | |
2453 | */ | |
2454 | if (base + last_end < vmalloc_start + last_end) { | |
2455 | spin_unlock(&vmap_area_lock); | |
2456 | if (!purged) { | |
2457 | purge_vmap_area_lazy(); | |
2458 | purged = true; | |
2459 | goto retry; | |
2460 | } | |
2461 | goto err_free; | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * If next overlaps, move base downwards so that it's | |
2466 | * right below next and then recheck. | |
2467 | */ | |
2468 | if (next && next->va_start < base + end) { | |
2469 | base = pvm_determine_end(&next, &prev, align) - end; | |
2470 | term_area = area; | |
2471 | continue; | |
2472 | } | |
2473 | ||
2474 | /* | |
2475 | * If prev overlaps, shift down next and prev and move | |
2476 | * base so that it's right below new next and then | |
2477 | * recheck. | |
2478 | */ | |
2479 | if (prev && prev->va_end > base + start) { | |
2480 | next = prev; | |
2481 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2482 | base = pvm_determine_end(&next, &prev, align) - end; | |
2483 | term_area = area; | |
2484 | continue; | |
2485 | } | |
2486 | ||
2487 | /* | |
2488 | * This area fits, move on to the previous one. If | |
2489 | * the previous one is the terminal one, we're done. | |
2490 | */ | |
2491 | area = (area + nr_vms - 1) % nr_vms; | |
2492 | if (area == term_area) | |
2493 | break; | |
2494 | start = offsets[area]; | |
2495 | end = start + sizes[area]; | |
2496 | pvm_find_next_prev(base + end, &next, &prev); | |
2497 | } | |
2498 | found: | |
2499 | /* we've found a fitting base, insert all va's */ | |
2500 | for (area = 0; area < nr_vms; area++) { | |
2501 | struct vmap_area *va = vas[area]; | |
2502 | ||
2503 | va->va_start = base + offsets[area]; | |
2504 | va->va_end = va->va_start + sizes[area]; | |
2505 | __insert_vmap_area(va); | |
2506 | } | |
2507 | ||
2508 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2509 | ||
2510 | spin_unlock(&vmap_area_lock); | |
2511 | ||
2512 | /* insert all vm's */ | |
2513 | for (area = 0; area < nr_vms; area++) | |
3645cb4a ZY |
2514 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
2515 | pcpu_get_vm_areas); | |
ca23e405 TH |
2516 | |
2517 | kfree(vas); | |
2518 | return vms; | |
2519 | ||
2520 | err_free: | |
2521 | for (area = 0; area < nr_vms; area++) { | |
f1db7afd KC |
2522 | kfree(vas[area]); |
2523 | kfree(vms[area]); | |
ca23e405 | 2524 | } |
f1db7afd | 2525 | err_free2: |
ca23e405 TH |
2526 | kfree(vas); |
2527 | kfree(vms); | |
2528 | return NULL; | |
2529 | } | |
2530 | ||
2531 | /** | |
2532 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2533 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2534 | * @nr_vms: the number of allocated areas | |
2535 | * | |
2536 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2537 | */ | |
2538 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2539 | { | |
2540 | int i; | |
2541 | ||
2542 | for (i = 0; i < nr_vms; i++) | |
2543 | free_vm_area(vms[i]); | |
2544 | kfree(vms); | |
2545 | } | |
4f8b02b4 | 2546 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
2547 | |
2548 | #ifdef CONFIG_PROC_FS | |
2549 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 2550 | __acquires(&vmap_area_lock) |
a10aa579 CL |
2551 | { |
2552 | loff_t n = *pos; | |
d4033afd | 2553 | struct vmap_area *va; |
a10aa579 | 2554 | |
d4033afd | 2555 | spin_lock(&vmap_area_lock); |
6219c2a2 | 2556 | va = list_first_entry(&vmap_area_list, typeof(*va), list); |
d4033afd | 2557 | while (n > 0 && &va->list != &vmap_area_list) { |
a10aa579 | 2558 | n--; |
6219c2a2 | 2559 | va = list_next_entry(va, list); |
a10aa579 | 2560 | } |
d4033afd JK |
2561 | if (!n && &va->list != &vmap_area_list) |
2562 | return va; | |
a10aa579 CL |
2563 | |
2564 | return NULL; | |
2565 | ||
2566 | } | |
2567 | ||
2568 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2569 | { | |
d4033afd | 2570 | struct vmap_area *va = p, *next; |
a10aa579 CL |
2571 | |
2572 | ++*pos; | |
6219c2a2 | 2573 | next = list_next_entry(va, list); |
d4033afd JK |
2574 | if (&next->list != &vmap_area_list) |
2575 | return next; | |
2576 | ||
2577 | return NULL; | |
a10aa579 CL |
2578 | } |
2579 | ||
2580 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 2581 | __releases(&vmap_area_lock) |
a10aa579 | 2582 | { |
d4033afd | 2583 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
2584 | } |
2585 | ||
a47a126a ED |
2586 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2587 | { | |
e5adfffc | 2588 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
2589 | unsigned int nr, *counters = m->private; |
2590 | ||
2591 | if (!counters) | |
2592 | return; | |
2593 | ||
af12346c WL |
2594 | if (v->flags & VM_UNINITIALIZED) |
2595 | return; | |
7e5b528b DV |
2596 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
2597 | smp_rmb(); | |
af12346c | 2598 | |
a47a126a ED |
2599 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
2600 | ||
2601 | for (nr = 0; nr < v->nr_pages; nr++) | |
2602 | counters[page_to_nid(v->pages[nr])]++; | |
2603 | ||
2604 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2605 | if (counters[nr]) | |
2606 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2607 | } | |
2608 | } | |
2609 | ||
a10aa579 CL |
2610 | static int s_show(struct seq_file *m, void *p) |
2611 | { | |
d4033afd JK |
2612 | struct vmap_area *va = p; |
2613 | struct vm_struct *v; | |
2614 | ||
c2ce8c14 WL |
2615 | /* |
2616 | * s_show can encounter race with remove_vm_area, !VM_VM_AREA on | |
2617 | * behalf of vmap area is being tear down or vm_map_ram allocation. | |
2618 | */ | |
2619 | if (!(va->flags & VM_VM_AREA)) | |
d4033afd | 2620 | return 0; |
d4033afd JK |
2621 | |
2622 | v = va->vm; | |
a10aa579 | 2623 | |
45ec1690 | 2624 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
2625 | v->addr, v->addr + v->size, v->size); |
2626 | ||
62c70bce JP |
2627 | if (v->caller) |
2628 | seq_printf(m, " %pS", v->caller); | |
23016969 | 2629 | |
a10aa579 CL |
2630 | if (v->nr_pages) |
2631 | seq_printf(m, " pages=%d", v->nr_pages); | |
2632 | ||
2633 | if (v->phys_addr) | |
ffa71f33 | 2634 | seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr); |
a10aa579 CL |
2635 | |
2636 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 2637 | seq_puts(m, " ioremap"); |
a10aa579 CL |
2638 | |
2639 | if (v->flags & VM_ALLOC) | |
f4527c90 | 2640 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
2641 | |
2642 | if (v->flags & VM_MAP) | |
f4527c90 | 2643 | seq_puts(m, " vmap"); |
a10aa579 CL |
2644 | |
2645 | if (v->flags & VM_USERMAP) | |
f4527c90 | 2646 | seq_puts(m, " user"); |
a10aa579 | 2647 | |
244d63ee | 2648 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 2649 | seq_puts(m, " vpages"); |
a10aa579 | 2650 | |
a47a126a | 2651 | show_numa_info(m, v); |
a10aa579 CL |
2652 | seq_putc(m, '\n'); |
2653 | return 0; | |
2654 | } | |
2655 | ||
5f6a6a9c | 2656 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2657 | .start = s_start, |
2658 | .next = s_next, | |
2659 | .stop = s_stop, | |
2660 | .show = s_show, | |
2661 | }; | |
5f6a6a9c AD |
2662 | |
2663 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2664 | { | |
703394c1 RJ |
2665 | if (IS_ENABLED(CONFIG_NUMA)) |
2666 | return seq_open_private(file, &vmalloc_op, | |
2667 | nr_node_ids * sizeof(unsigned int)); | |
2668 | else | |
2669 | return seq_open(file, &vmalloc_op); | |
5f6a6a9c AD |
2670 | } |
2671 | ||
2672 | static const struct file_operations proc_vmalloc_operations = { | |
2673 | .open = vmalloc_open, | |
2674 | .read = seq_read, | |
2675 | .llseek = seq_lseek, | |
2676 | .release = seq_release_private, | |
2677 | }; | |
2678 | ||
2679 | static int __init proc_vmalloc_init(void) | |
2680 | { | |
2681 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2682 | return 0; | |
2683 | } | |
2684 | module_init(proc_vmalloc_init); | |
db3808c1 | 2685 | |
a10aa579 CL |
2686 | #endif |
2687 |