]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmalloc.c | |
3 | * | |
4 | * Copyright (C) 1993 Linus Torvalds | |
5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
6 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
7 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
9 | */ |
10 | ||
db64fe02 | 11 | #include <linux/vmalloc.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> | |
14 | #include <linux/highmem.h> | |
c3edc401 | 15 | #include <linux/sched/signal.h> |
1da177e4 LT |
16 | #include <linux/slab.h> |
17 | #include <linux/spinlock.h> | |
18 | #include <linux/interrupt.h> | |
5f6a6a9c | 19 | #include <linux/proc_fs.h> |
a10aa579 | 20 | #include <linux/seq_file.h> |
3ac7fe5a | 21 | #include <linux/debugobjects.h> |
23016969 | 22 | #include <linux/kallsyms.h> |
db64fe02 | 23 | #include <linux/list.h> |
4da56b99 | 24 | #include <linux/notifier.h> |
db64fe02 NP |
25 | #include <linux/rbtree.h> |
26 | #include <linux/radix-tree.h> | |
27 | #include <linux/rcupdate.h> | |
f0aa6617 | 28 | #include <linux/pfn.h> |
89219d37 | 29 | #include <linux/kmemleak.h> |
60063497 | 30 | #include <linux/atomic.h> |
3b32123d | 31 | #include <linux/compiler.h> |
32fcfd40 | 32 | #include <linux/llist.h> |
0f616be1 | 33 | #include <linux/bitops.h> |
3b32123d | 34 | |
7c0f6ba6 | 35 | #include <linux/uaccess.h> |
1da177e4 | 36 | #include <asm/tlbflush.h> |
2dca6999 | 37 | #include <asm/shmparam.h> |
1da177e4 | 38 | |
dd56b046 MG |
39 | #include "internal.h" |
40 | ||
32fcfd40 AV |
41 | struct vfree_deferred { |
42 | struct llist_head list; | |
43 | struct work_struct wq; | |
44 | }; | |
45 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
46 | ||
47 | static void __vunmap(const void *, int); | |
48 | ||
49 | static void free_work(struct work_struct *w) | |
50 | { | |
51 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
52 | struct llist_node *llnode = llist_del_all(&p->list); | |
53 | while (llnode) { | |
54 | void *p = llnode; | |
55 | llnode = llist_next(llnode); | |
56 | __vunmap(p, 1); | |
57 | } | |
58 | } | |
59 | ||
db64fe02 | 60 | /*** Page table manipulation functions ***/ |
b221385b | 61 | |
1da177e4 LT |
62 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
63 | { | |
64 | pte_t *pte; | |
65 | ||
66 | pte = pte_offset_kernel(pmd, addr); | |
67 | do { | |
68 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
69 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
70 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
71 | } | |
72 | ||
db64fe02 | 73 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
74 | { |
75 | pmd_t *pmd; | |
76 | unsigned long next; | |
77 | ||
78 | pmd = pmd_offset(pud, addr); | |
79 | do { | |
80 | next = pmd_addr_end(addr, end); | |
b9820d8f TK |
81 | if (pmd_clear_huge(pmd)) |
82 | continue; | |
1da177e4 LT |
83 | if (pmd_none_or_clear_bad(pmd)) |
84 | continue; | |
85 | vunmap_pte_range(pmd, addr, next); | |
86 | } while (pmd++, addr = next, addr != end); | |
87 | } | |
88 | ||
c2febafc | 89 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end) |
1da177e4 LT |
90 | { |
91 | pud_t *pud; | |
92 | unsigned long next; | |
93 | ||
c2febafc | 94 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
95 | do { |
96 | next = pud_addr_end(addr, end); | |
b9820d8f TK |
97 | if (pud_clear_huge(pud)) |
98 | continue; | |
1da177e4 LT |
99 | if (pud_none_or_clear_bad(pud)) |
100 | continue; | |
101 | vunmap_pmd_range(pud, addr, next); | |
102 | } while (pud++, addr = next, addr != end); | |
103 | } | |
104 | ||
c2febafc KS |
105 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
106 | { | |
107 | p4d_t *p4d; | |
108 | unsigned long next; | |
109 | ||
110 | p4d = p4d_offset(pgd, addr); | |
111 | do { | |
112 | next = p4d_addr_end(addr, end); | |
113 | if (p4d_clear_huge(p4d)) | |
114 | continue; | |
115 | if (p4d_none_or_clear_bad(p4d)) | |
116 | continue; | |
117 | vunmap_pud_range(p4d, addr, next); | |
118 | } while (p4d++, addr = next, addr != end); | |
119 | } | |
120 | ||
db64fe02 | 121 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
122 | { |
123 | pgd_t *pgd; | |
124 | unsigned long next; | |
1da177e4 LT |
125 | |
126 | BUG_ON(addr >= end); | |
127 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
128 | do { |
129 | next = pgd_addr_end(addr, end); | |
130 | if (pgd_none_or_clear_bad(pgd)) | |
131 | continue; | |
c2febafc | 132 | vunmap_p4d_range(pgd, addr, next); |
1da177e4 | 133 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
134 | } |
135 | ||
136 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 137 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
138 | { |
139 | pte_t *pte; | |
140 | ||
db64fe02 NP |
141 | /* |
142 | * nr is a running index into the array which helps higher level | |
143 | * callers keep track of where we're up to. | |
144 | */ | |
145 | ||
872fec16 | 146 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
147 | if (!pte) |
148 | return -ENOMEM; | |
149 | do { | |
db64fe02 NP |
150 | struct page *page = pages[*nr]; |
151 | ||
152 | if (WARN_ON(!pte_none(*pte))) | |
153 | return -EBUSY; | |
154 | if (WARN_ON(!page)) | |
1da177e4 LT |
155 | return -ENOMEM; |
156 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 157 | (*nr)++; |
1da177e4 LT |
158 | } while (pte++, addr += PAGE_SIZE, addr != end); |
159 | return 0; | |
160 | } | |
161 | ||
db64fe02 NP |
162 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
163 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
164 | { |
165 | pmd_t *pmd; | |
166 | unsigned long next; | |
167 | ||
168 | pmd = pmd_alloc(&init_mm, pud, addr); | |
169 | if (!pmd) | |
170 | return -ENOMEM; | |
171 | do { | |
172 | next = pmd_addr_end(addr, end); | |
db64fe02 | 173 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
174 | return -ENOMEM; |
175 | } while (pmd++, addr = next, addr != end); | |
176 | return 0; | |
177 | } | |
178 | ||
c2febafc | 179 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, |
db64fe02 | 180 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
181 | { |
182 | pud_t *pud; | |
183 | unsigned long next; | |
184 | ||
c2febafc | 185 | pud = pud_alloc(&init_mm, p4d, addr); |
1da177e4 LT |
186 | if (!pud) |
187 | return -ENOMEM; | |
188 | do { | |
189 | next = pud_addr_end(addr, end); | |
db64fe02 | 190 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
191 | return -ENOMEM; |
192 | } while (pud++, addr = next, addr != end); | |
193 | return 0; | |
194 | } | |
195 | ||
c2febafc KS |
196 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, |
197 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
198 | { | |
199 | p4d_t *p4d; | |
200 | unsigned long next; | |
201 | ||
202 | p4d = p4d_alloc(&init_mm, pgd, addr); | |
203 | if (!p4d) | |
204 | return -ENOMEM; | |
205 | do { | |
206 | next = p4d_addr_end(addr, end); | |
207 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr)) | |
208 | return -ENOMEM; | |
209 | } while (p4d++, addr = next, addr != end); | |
210 | return 0; | |
211 | } | |
212 | ||
db64fe02 NP |
213 | /* |
214 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
215 | * will have pfns corresponding to the "pages" array. | |
216 | * | |
217 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
218 | */ | |
8fc48985 TH |
219 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
220 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
221 | { |
222 | pgd_t *pgd; | |
223 | unsigned long next; | |
2e4e27c7 | 224 | unsigned long addr = start; |
db64fe02 NP |
225 | int err = 0; |
226 | int nr = 0; | |
1da177e4 LT |
227 | |
228 | BUG_ON(addr >= end); | |
229 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
230 | do { |
231 | next = pgd_addr_end(addr, end); | |
c2febafc | 232 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 233 | if (err) |
bf88c8c8 | 234 | return err; |
1da177e4 | 235 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 236 | |
db64fe02 | 237 | return nr; |
1da177e4 LT |
238 | } |
239 | ||
8fc48985 TH |
240 | static int vmap_page_range(unsigned long start, unsigned long end, |
241 | pgprot_t prot, struct page **pages) | |
242 | { | |
243 | int ret; | |
244 | ||
245 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
246 | flush_cache_vmap(start, end); | |
247 | return ret; | |
248 | } | |
249 | ||
81ac3ad9 | 250 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
251 | { |
252 | /* | |
ab4f2ee1 | 253 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
254 | * and fall back on vmalloc() if that fails. Others |
255 | * just put it in the vmalloc space. | |
256 | */ | |
257 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
258 | unsigned long addr = (unsigned long)x; | |
259 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
260 | return 1; | |
261 | #endif | |
262 | return is_vmalloc_addr(x); | |
263 | } | |
264 | ||
48667e7a | 265 | /* |
add688fb | 266 | * Walk a vmap address to the struct page it maps. |
48667e7a | 267 | */ |
add688fb | 268 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
269 | { |
270 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 271 | struct page *page = NULL; |
48667e7a | 272 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
273 | p4d_t *p4d; |
274 | pud_t *pud; | |
275 | pmd_t *pmd; | |
276 | pte_t *ptep, pte; | |
48667e7a | 277 | |
7aa413de IM |
278 | /* |
279 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
280 | * architectures that do not vmalloc module space | |
281 | */ | |
73bdf0a6 | 282 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 283 | |
c2febafc KS |
284 | if (pgd_none(*pgd)) |
285 | return NULL; | |
286 | p4d = p4d_offset(pgd, addr); | |
287 | if (p4d_none(*p4d)) | |
288 | return NULL; | |
289 | pud = pud_offset(p4d, addr); | |
029c54b0 AB |
290 | |
291 | /* | |
292 | * Don't dereference bad PUD or PMD (below) entries. This will also | |
293 | * identify huge mappings, which we may encounter on architectures | |
294 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be | |
295 | * identified as vmalloc addresses by is_vmalloc_addr(), but are | |
296 | * not [unambiguously] associated with a struct page, so there is | |
297 | * no correct value to return for them. | |
298 | */ | |
299 | WARN_ON_ONCE(pud_bad(*pud)); | |
300 | if (pud_none(*pud) || pud_bad(*pud)) | |
c2febafc KS |
301 | return NULL; |
302 | pmd = pmd_offset(pud, addr); | |
029c54b0 AB |
303 | WARN_ON_ONCE(pmd_bad(*pmd)); |
304 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
c2febafc KS |
305 | return NULL; |
306 | ||
307 | ptep = pte_offset_map(pmd, addr); | |
308 | pte = *ptep; | |
309 | if (pte_present(pte)) | |
310 | page = pte_page(pte); | |
311 | pte_unmap(ptep); | |
add688fb | 312 | return page; |
48667e7a | 313 | } |
add688fb | 314 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
315 | |
316 | /* | |
add688fb | 317 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 318 | */ |
add688fb | 319 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 320 | { |
add688fb | 321 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 322 | } |
add688fb | 323 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 324 | |
db64fe02 NP |
325 | |
326 | /*** Global kva allocator ***/ | |
327 | ||
78c72746 | 328 | #define VM_LAZY_FREE 0x02 |
db64fe02 NP |
329 | #define VM_VM_AREA 0x04 |
330 | ||
db64fe02 | 331 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
332 | /* Export for kexec only */ |
333 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 334 | static LLIST_HEAD(vmap_purge_list); |
89699605 NP |
335 | static struct rb_root vmap_area_root = RB_ROOT; |
336 | ||
337 | /* The vmap cache globals are protected by vmap_area_lock */ | |
338 | static struct rb_node *free_vmap_cache; | |
339 | static unsigned long cached_hole_size; | |
340 | static unsigned long cached_vstart; | |
341 | static unsigned long cached_align; | |
342 | ||
ca23e405 | 343 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
344 | |
345 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 346 | { |
db64fe02 NP |
347 | struct rb_node *n = vmap_area_root.rb_node; |
348 | ||
349 | while (n) { | |
350 | struct vmap_area *va; | |
351 | ||
352 | va = rb_entry(n, struct vmap_area, rb_node); | |
353 | if (addr < va->va_start) | |
354 | n = n->rb_left; | |
cef2ac3f | 355 | else if (addr >= va->va_end) |
db64fe02 NP |
356 | n = n->rb_right; |
357 | else | |
358 | return va; | |
359 | } | |
360 | ||
361 | return NULL; | |
362 | } | |
363 | ||
364 | static void __insert_vmap_area(struct vmap_area *va) | |
365 | { | |
366 | struct rb_node **p = &vmap_area_root.rb_node; | |
367 | struct rb_node *parent = NULL; | |
368 | struct rb_node *tmp; | |
369 | ||
370 | while (*p) { | |
170168d0 | 371 | struct vmap_area *tmp_va; |
db64fe02 NP |
372 | |
373 | parent = *p; | |
170168d0 NK |
374 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
375 | if (va->va_start < tmp_va->va_end) | |
db64fe02 | 376 | p = &(*p)->rb_left; |
170168d0 | 377 | else if (va->va_end > tmp_va->va_start) |
db64fe02 NP |
378 | p = &(*p)->rb_right; |
379 | else | |
380 | BUG(); | |
381 | } | |
382 | ||
383 | rb_link_node(&va->rb_node, parent, p); | |
384 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
385 | ||
4341fa45 | 386 | /* address-sort this list */ |
db64fe02 NP |
387 | tmp = rb_prev(&va->rb_node); |
388 | if (tmp) { | |
389 | struct vmap_area *prev; | |
390 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
391 | list_add_rcu(&va->list, &prev->list); | |
392 | } else | |
393 | list_add_rcu(&va->list, &vmap_area_list); | |
394 | } | |
395 | ||
396 | static void purge_vmap_area_lazy(void); | |
397 | ||
4da56b99 CW |
398 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); |
399 | ||
db64fe02 NP |
400 | /* |
401 | * Allocate a region of KVA of the specified size and alignment, within the | |
402 | * vstart and vend. | |
403 | */ | |
404 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
405 | unsigned long align, | |
406 | unsigned long vstart, unsigned long vend, | |
407 | int node, gfp_t gfp_mask) | |
408 | { | |
409 | struct vmap_area *va; | |
410 | struct rb_node *n; | |
1da177e4 | 411 | unsigned long addr; |
db64fe02 | 412 | int purged = 0; |
89699605 | 413 | struct vmap_area *first; |
db64fe02 | 414 | |
7766970c | 415 | BUG_ON(!size); |
891c49ab | 416 | BUG_ON(offset_in_page(size)); |
89699605 | 417 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 418 | |
5803ed29 | 419 | might_sleep(); |
4da56b99 | 420 | |
db64fe02 NP |
421 | va = kmalloc_node(sizeof(struct vmap_area), |
422 | gfp_mask & GFP_RECLAIM_MASK, node); | |
423 | if (unlikely(!va)) | |
424 | return ERR_PTR(-ENOMEM); | |
425 | ||
7f88f88f CM |
426 | /* |
427 | * Only scan the relevant parts containing pointers to other objects | |
428 | * to avoid false negatives. | |
429 | */ | |
430 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK); | |
431 | ||
db64fe02 NP |
432 | retry: |
433 | spin_lock(&vmap_area_lock); | |
89699605 NP |
434 | /* |
435 | * Invalidate cache if we have more permissive parameters. | |
436 | * cached_hole_size notes the largest hole noticed _below_ | |
437 | * the vmap_area cached in free_vmap_cache: if size fits | |
438 | * into that hole, we want to scan from vstart to reuse | |
439 | * the hole instead of allocating above free_vmap_cache. | |
440 | * Note that __free_vmap_area may update free_vmap_cache | |
441 | * without updating cached_hole_size or cached_align. | |
442 | */ | |
443 | if (!free_vmap_cache || | |
444 | size < cached_hole_size || | |
445 | vstart < cached_vstart || | |
446 | align < cached_align) { | |
447 | nocache: | |
448 | cached_hole_size = 0; | |
449 | free_vmap_cache = NULL; | |
450 | } | |
451 | /* record if we encounter less permissive parameters */ | |
452 | cached_vstart = vstart; | |
453 | cached_align = align; | |
454 | ||
455 | /* find starting point for our search */ | |
456 | if (free_vmap_cache) { | |
457 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
248ac0e1 | 458 | addr = ALIGN(first->va_end, align); |
89699605 NP |
459 | if (addr < vstart) |
460 | goto nocache; | |
bcb615a8 | 461 | if (addr + size < addr) |
89699605 NP |
462 | goto overflow; |
463 | ||
464 | } else { | |
465 | addr = ALIGN(vstart, align); | |
bcb615a8 | 466 | if (addr + size < addr) |
89699605 NP |
467 | goto overflow; |
468 | ||
469 | n = vmap_area_root.rb_node; | |
470 | first = NULL; | |
471 | ||
472 | while (n) { | |
db64fe02 NP |
473 | struct vmap_area *tmp; |
474 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
475 | if (tmp->va_end >= addr) { | |
db64fe02 | 476 | first = tmp; |
89699605 NP |
477 | if (tmp->va_start <= addr) |
478 | break; | |
479 | n = n->rb_left; | |
480 | } else | |
db64fe02 | 481 | n = n->rb_right; |
89699605 | 482 | } |
db64fe02 NP |
483 | |
484 | if (!first) | |
485 | goto found; | |
db64fe02 | 486 | } |
89699605 NP |
487 | |
488 | /* from the starting point, walk areas until a suitable hole is found */ | |
248ac0e1 | 489 | while (addr + size > first->va_start && addr + size <= vend) { |
89699605 NP |
490 | if (addr + cached_hole_size < first->va_start) |
491 | cached_hole_size = first->va_start - addr; | |
248ac0e1 | 492 | addr = ALIGN(first->va_end, align); |
bcb615a8 | 493 | if (addr + size < addr) |
89699605 NP |
494 | goto overflow; |
495 | ||
92ca922f | 496 | if (list_is_last(&first->list, &vmap_area_list)) |
89699605 | 497 | goto found; |
92ca922f | 498 | |
6219c2a2 | 499 | first = list_next_entry(first, list); |
db64fe02 NP |
500 | } |
501 | ||
89699605 NP |
502 | found: |
503 | if (addr + size > vend) | |
504 | goto overflow; | |
db64fe02 NP |
505 | |
506 | va->va_start = addr; | |
507 | va->va_end = addr + size; | |
508 | va->flags = 0; | |
509 | __insert_vmap_area(va); | |
89699605 | 510 | free_vmap_cache = &va->rb_node; |
db64fe02 NP |
511 | spin_unlock(&vmap_area_lock); |
512 | ||
61e16557 | 513 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
514 | BUG_ON(va->va_start < vstart); |
515 | BUG_ON(va->va_end > vend); | |
516 | ||
db64fe02 | 517 | return va; |
89699605 NP |
518 | |
519 | overflow: | |
520 | spin_unlock(&vmap_area_lock); | |
521 | if (!purged) { | |
522 | purge_vmap_area_lazy(); | |
523 | purged = 1; | |
524 | goto retry; | |
525 | } | |
4da56b99 CW |
526 | |
527 | if (gfpflags_allow_blocking(gfp_mask)) { | |
528 | unsigned long freed = 0; | |
529 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
530 | if (freed > 0) { | |
531 | purged = 0; | |
532 | goto retry; | |
533 | } | |
534 | } | |
535 | ||
03497d76 | 536 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
537 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
538 | size); | |
89699605 NP |
539 | kfree(va); |
540 | return ERR_PTR(-EBUSY); | |
db64fe02 NP |
541 | } |
542 | ||
4da56b99 CW |
543 | int register_vmap_purge_notifier(struct notifier_block *nb) |
544 | { | |
545 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
546 | } | |
547 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
548 | ||
549 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
550 | { | |
551 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
552 | } | |
553 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
554 | ||
db64fe02 NP |
555 | static void __free_vmap_area(struct vmap_area *va) |
556 | { | |
557 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
89699605 NP |
558 | |
559 | if (free_vmap_cache) { | |
560 | if (va->va_end < cached_vstart) { | |
561 | free_vmap_cache = NULL; | |
562 | } else { | |
563 | struct vmap_area *cache; | |
564 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
565 | if (va->va_start <= cache->va_start) { | |
566 | free_vmap_cache = rb_prev(&va->rb_node); | |
567 | /* | |
568 | * We don't try to update cached_hole_size or | |
569 | * cached_align, but it won't go very wrong. | |
570 | */ | |
571 | } | |
572 | } | |
573 | } | |
db64fe02 NP |
574 | rb_erase(&va->rb_node, &vmap_area_root); |
575 | RB_CLEAR_NODE(&va->rb_node); | |
576 | list_del_rcu(&va->list); | |
577 | ||
ca23e405 TH |
578 | /* |
579 | * Track the highest possible candidate for pcpu area | |
580 | * allocation. Areas outside of vmalloc area can be returned | |
581 | * here too, consider only end addresses which fall inside | |
582 | * vmalloc area proper. | |
583 | */ | |
584 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
585 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
586 | ||
14769de9 | 587 | kfree_rcu(va, rcu_head); |
db64fe02 NP |
588 | } |
589 | ||
590 | /* | |
591 | * Free a region of KVA allocated by alloc_vmap_area | |
592 | */ | |
593 | static void free_vmap_area(struct vmap_area *va) | |
594 | { | |
595 | spin_lock(&vmap_area_lock); | |
596 | __free_vmap_area(va); | |
597 | spin_unlock(&vmap_area_lock); | |
598 | } | |
599 | ||
600 | /* | |
601 | * Clear the pagetable entries of a given vmap_area | |
602 | */ | |
603 | static void unmap_vmap_area(struct vmap_area *va) | |
604 | { | |
605 | vunmap_page_range(va->va_start, va->va_end); | |
606 | } | |
607 | ||
cd52858c NP |
608 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
609 | { | |
610 | /* | |
f48d97f3 JK |
611 | * Unmap page tables and force a TLB flush immediately if pagealloc |
612 | * debugging is enabled. This catches use after free bugs similarly to | |
613 | * those in linear kernel virtual address space after a page has been | |
614 | * freed. | |
cd52858c | 615 | * |
f48d97f3 JK |
616 | * All the lazy freeing logic is still retained, in order to minimise |
617 | * intrusiveness of this debugging feature. | |
cd52858c | 618 | * |
f48d97f3 JK |
619 | * This is going to be *slow* (linear kernel virtual address debugging |
620 | * doesn't do a broadcast TLB flush so it is a lot faster). | |
cd52858c | 621 | */ |
f48d97f3 JK |
622 | if (debug_pagealloc_enabled()) { |
623 | vunmap_page_range(start, end); | |
624 | flush_tlb_kernel_range(start, end); | |
625 | } | |
cd52858c NP |
626 | } |
627 | ||
db64fe02 NP |
628 | /* |
629 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
630 | * before attempting to purge with a TLB flush. | |
631 | * | |
632 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
633 | * and take slightly longer to purge, but it will linearly reduce the number of | |
634 | * global TLB flushes that must be performed. It would seem natural to scale | |
635 | * this number up linearly with the number of CPUs (because vmapping activity | |
636 | * could also scale linearly with the number of CPUs), however it is likely | |
637 | * that in practice, workloads might be constrained in other ways that mean | |
638 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
639 | * conservative and not introduce a big latency on huge systems, so go with | |
640 | * a less aggressive log scale. It will still be an improvement over the old | |
641 | * code, and it will be simple to change the scale factor if we find that it | |
642 | * becomes a problem on bigger systems. | |
643 | */ | |
644 | static unsigned long lazy_max_pages(void) | |
645 | { | |
646 | unsigned int log; | |
647 | ||
648 | log = fls(num_online_cpus()); | |
649 | ||
650 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
651 | } | |
652 | ||
653 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
654 | ||
0574ecd1 CH |
655 | /* |
656 | * Serialize vmap purging. There is no actual criticial section protected | |
657 | * by this look, but we want to avoid concurrent calls for performance | |
658 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
659 | */ | |
f9e09977 | 660 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 661 | |
02b709df NP |
662 | /* for per-CPU blocks */ |
663 | static void purge_fragmented_blocks_allcpus(void); | |
664 | ||
3ee48b6a CW |
665 | /* |
666 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
667 | * immediately freed. | |
668 | */ | |
669 | void set_iounmap_nonlazy(void) | |
670 | { | |
671 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); | |
672 | } | |
673 | ||
db64fe02 NP |
674 | /* |
675 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 676 | */ |
0574ecd1 | 677 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 678 | { |
80c4bd7a | 679 | struct llist_node *valist; |
db64fe02 | 680 | struct vmap_area *va; |
cbb76676 | 681 | struct vmap_area *n_va; |
763b218d | 682 | bool do_free = false; |
db64fe02 | 683 | |
0574ecd1 | 684 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 685 | |
80c4bd7a CW |
686 | valist = llist_del_all(&vmap_purge_list); |
687 | llist_for_each_entry(va, valist, purge_list) { | |
0574ecd1 CH |
688 | if (va->va_start < start) |
689 | start = va->va_start; | |
690 | if (va->va_end > end) | |
691 | end = va->va_end; | |
763b218d | 692 | do_free = true; |
db64fe02 | 693 | } |
db64fe02 | 694 | |
763b218d | 695 | if (!do_free) |
0574ecd1 | 696 | return false; |
db64fe02 | 697 | |
0574ecd1 | 698 | flush_tlb_kernel_range(start, end); |
db64fe02 | 699 | |
0574ecd1 | 700 | spin_lock(&vmap_area_lock); |
763b218d JF |
701 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
702 | int nr = (va->va_end - va->va_start) >> PAGE_SHIFT; | |
703 | ||
0574ecd1 | 704 | __free_vmap_area(va); |
763b218d JF |
705 | atomic_sub(nr, &vmap_lazy_nr); |
706 | cond_resched_lock(&vmap_area_lock); | |
707 | } | |
0574ecd1 CH |
708 | spin_unlock(&vmap_area_lock); |
709 | return true; | |
db64fe02 NP |
710 | } |
711 | ||
496850e5 NP |
712 | /* |
713 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
714 | * is already purging. | |
715 | */ | |
716 | static void try_purge_vmap_area_lazy(void) | |
717 | { | |
f9e09977 | 718 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 719 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 720 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 721 | } |
496850e5 NP |
722 | } |
723 | ||
db64fe02 NP |
724 | /* |
725 | * Kick off a purge of the outstanding lazy areas. | |
726 | */ | |
727 | static void purge_vmap_area_lazy(void) | |
728 | { | |
f9e09977 | 729 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
730 | purge_fragmented_blocks_allcpus(); |
731 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 732 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
733 | } |
734 | ||
735 | /* | |
64141da5 JF |
736 | * Free a vmap area, caller ensuring that the area has been unmapped |
737 | * and flush_cache_vunmap had been called for the correct range | |
738 | * previously. | |
db64fe02 | 739 | */ |
64141da5 | 740 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 741 | { |
80c4bd7a CW |
742 | int nr_lazy; |
743 | ||
744 | nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT, | |
745 | &vmap_lazy_nr); | |
746 | ||
747 | /* After this point, we may free va at any time */ | |
748 | llist_add(&va->purge_list, &vmap_purge_list); | |
749 | ||
750 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 751 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
752 | } |
753 | ||
b29acbdc NP |
754 | /* |
755 | * Free and unmap a vmap area | |
756 | */ | |
757 | static void free_unmap_vmap_area(struct vmap_area *va) | |
758 | { | |
759 | flush_cache_vunmap(va->va_start, va->va_end); | |
c8eef01e CH |
760 | unmap_vmap_area(va); |
761 | free_vmap_area_noflush(va); | |
b29acbdc NP |
762 | } |
763 | ||
db64fe02 NP |
764 | static struct vmap_area *find_vmap_area(unsigned long addr) |
765 | { | |
766 | struct vmap_area *va; | |
767 | ||
768 | spin_lock(&vmap_area_lock); | |
769 | va = __find_vmap_area(addr); | |
770 | spin_unlock(&vmap_area_lock); | |
771 | ||
772 | return va; | |
773 | } | |
774 | ||
db64fe02 NP |
775 | /*** Per cpu kva allocator ***/ |
776 | ||
777 | /* | |
778 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
779 | * room for at least 16 percpu vmap blocks per CPU. | |
780 | */ | |
781 | /* | |
782 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
783 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
784 | * instead (we just need a rough idea) | |
785 | */ | |
786 | #if BITS_PER_LONG == 32 | |
787 | #define VMALLOC_SPACE (128UL*1024*1024) | |
788 | #else | |
789 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
790 | #endif | |
791 | ||
792 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
793 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
794 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
795 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
796 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
797 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
798 | #define VMAP_BBMAP_BITS \ |
799 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
800 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
801 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
802 | |
803 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
804 | ||
9b463334 JF |
805 | static bool vmap_initialized __read_mostly = false; |
806 | ||
db64fe02 NP |
807 | struct vmap_block_queue { |
808 | spinlock_t lock; | |
809 | struct list_head free; | |
db64fe02 NP |
810 | }; |
811 | ||
812 | struct vmap_block { | |
813 | spinlock_t lock; | |
814 | struct vmap_area *va; | |
db64fe02 | 815 | unsigned long free, dirty; |
7d61bfe8 | 816 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
817 | struct list_head free_list; |
818 | struct rcu_head rcu_head; | |
02b709df | 819 | struct list_head purge; |
db64fe02 NP |
820 | }; |
821 | ||
822 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
823 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
824 | ||
825 | /* | |
826 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
827 | * in the free path. Could get rid of this if we change the API to return a | |
828 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
829 | */ | |
830 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
831 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
832 | ||
833 | /* | |
834 | * We should probably have a fallback mechanism to allocate virtual memory | |
835 | * out of partially filled vmap blocks. However vmap block sizing should be | |
836 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
837 | * big problem. | |
838 | */ | |
839 | ||
840 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
841 | { | |
842 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
843 | addr /= VMAP_BLOCK_SIZE; | |
844 | return addr; | |
845 | } | |
846 | ||
cf725ce2 RP |
847 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
848 | { | |
849 | unsigned long addr; | |
850 | ||
851 | addr = va_start + (pages_off << PAGE_SHIFT); | |
852 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
853 | return (void *)addr; | |
854 | } | |
855 | ||
856 | /** | |
857 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
858 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
859 | * @order: how many 2^order pages should be occupied in newly allocated block | |
860 | * @gfp_mask: flags for the page level allocator | |
861 | * | |
862 | * Returns: virtual address in a newly allocated block or ERR_PTR(-errno) | |
863 | */ | |
864 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
865 | { |
866 | struct vmap_block_queue *vbq; | |
867 | struct vmap_block *vb; | |
868 | struct vmap_area *va; | |
869 | unsigned long vb_idx; | |
870 | int node, err; | |
cf725ce2 | 871 | void *vaddr; |
db64fe02 NP |
872 | |
873 | node = numa_node_id(); | |
874 | ||
875 | vb = kmalloc_node(sizeof(struct vmap_block), | |
876 | gfp_mask & GFP_RECLAIM_MASK, node); | |
877 | if (unlikely(!vb)) | |
878 | return ERR_PTR(-ENOMEM); | |
879 | ||
880 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
881 | VMALLOC_START, VMALLOC_END, | |
882 | node, gfp_mask); | |
ddf9c6d4 | 883 | if (IS_ERR(va)) { |
db64fe02 | 884 | kfree(vb); |
e7d86340 | 885 | return ERR_CAST(va); |
db64fe02 NP |
886 | } |
887 | ||
888 | err = radix_tree_preload(gfp_mask); | |
889 | if (unlikely(err)) { | |
890 | kfree(vb); | |
891 | free_vmap_area(va); | |
892 | return ERR_PTR(err); | |
893 | } | |
894 | ||
cf725ce2 | 895 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
896 | spin_lock_init(&vb->lock); |
897 | vb->va = va; | |
cf725ce2 RP |
898 | /* At least something should be left free */ |
899 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
900 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 901 | vb->dirty = 0; |
7d61bfe8 RP |
902 | vb->dirty_min = VMAP_BBMAP_BITS; |
903 | vb->dirty_max = 0; | |
db64fe02 | 904 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
905 | |
906 | vb_idx = addr_to_vb_idx(va->va_start); | |
907 | spin_lock(&vmap_block_tree_lock); | |
908 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
909 | spin_unlock(&vmap_block_tree_lock); | |
910 | BUG_ON(err); | |
911 | radix_tree_preload_end(); | |
912 | ||
913 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 914 | spin_lock(&vbq->lock); |
68ac546f | 915 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 916 | spin_unlock(&vbq->lock); |
3f04ba85 | 917 | put_cpu_var(vmap_block_queue); |
db64fe02 | 918 | |
cf725ce2 | 919 | return vaddr; |
db64fe02 NP |
920 | } |
921 | ||
db64fe02 NP |
922 | static void free_vmap_block(struct vmap_block *vb) |
923 | { | |
924 | struct vmap_block *tmp; | |
925 | unsigned long vb_idx; | |
926 | ||
db64fe02 NP |
927 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
928 | spin_lock(&vmap_block_tree_lock); | |
929 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
930 | spin_unlock(&vmap_block_tree_lock); | |
931 | BUG_ON(tmp != vb); | |
932 | ||
64141da5 | 933 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 934 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
935 | } |
936 | ||
02b709df NP |
937 | static void purge_fragmented_blocks(int cpu) |
938 | { | |
939 | LIST_HEAD(purge); | |
940 | struct vmap_block *vb; | |
941 | struct vmap_block *n_vb; | |
942 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
943 | ||
944 | rcu_read_lock(); | |
945 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
946 | ||
947 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
948 | continue; | |
949 | ||
950 | spin_lock(&vb->lock); | |
951 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
952 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
953 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
954 | vb->dirty_min = 0; |
955 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
956 | spin_lock(&vbq->lock); |
957 | list_del_rcu(&vb->free_list); | |
958 | spin_unlock(&vbq->lock); | |
959 | spin_unlock(&vb->lock); | |
960 | list_add_tail(&vb->purge, &purge); | |
961 | } else | |
962 | spin_unlock(&vb->lock); | |
963 | } | |
964 | rcu_read_unlock(); | |
965 | ||
966 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
967 | list_del(&vb->purge); | |
968 | free_vmap_block(vb); | |
969 | } | |
970 | } | |
971 | ||
02b709df NP |
972 | static void purge_fragmented_blocks_allcpus(void) |
973 | { | |
974 | int cpu; | |
975 | ||
976 | for_each_possible_cpu(cpu) | |
977 | purge_fragmented_blocks(cpu); | |
978 | } | |
979 | ||
db64fe02 NP |
980 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
981 | { | |
982 | struct vmap_block_queue *vbq; | |
983 | struct vmap_block *vb; | |
cf725ce2 | 984 | void *vaddr = NULL; |
db64fe02 NP |
985 | unsigned int order; |
986 | ||
891c49ab | 987 | BUG_ON(offset_in_page(size)); |
db64fe02 | 988 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
989 | if (WARN_ON(size == 0)) { |
990 | /* | |
991 | * Allocating 0 bytes isn't what caller wants since | |
992 | * get_order(0) returns funny result. Just warn and terminate | |
993 | * early. | |
994 | */ | |
995 | return NULL; | |
996 | } | |
db64fe02 NP |
997 | order = get_order(size); |
998 | ||
db64fe02 NP |
999 | rcu_read_lock(); |
1000 | vbq = &get_cpu_var(vmap_block_queue); | |
1001 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1002 | unsigned long pages_off; |
db64fe02 NP |
1003 | |
1004 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1005 | if (vb->free < (1UL << order)) { |
1006 | spin_unlock(&vb->lock); | |
1007 | continue; | |
1008 | } | |
02b709df | 1009 | |
cf725ce2 RP |
1010 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1011 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1012 | vb->free -= 1UL << order; |
1013 | if (vb->free == 0) { | |
1014 | spin_lock(&vbq->lock); | |
1015 | list_del_rcu(&vb->free_list); | |
1016 | spin_unlock(&vbq->lock); | |
1017 | } | |
cf725ce2 | 1018 | |
02b709df NP |
1019 | spin_unlock(&vb->lock); |
1020 | break; | |
db64fe02 | 1021 | } |
02b709df | 1022 | |
3f04ba85 | 1023 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1024 | rcu_read_unlock(); |
1025 | ||
cf725ce2 RP |
1026 | /* Allocate new block if nothing was found */ |
1027 | if (!vaddr) | |
1028 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1029 | |
cf725ce2 | 1030 | return vaddr; |
db64fe02 NP |
1031 | } |
1032 | ||
1033 | static void vb_free(const void *addr, unsigned long size) | |
1034 | { | |
1035 | unsigned long offset; | |
1036 | unsigned long vb_idx; | |
1037 | unsigned int order; | |
1038 | struct vmap_block *vb; | |
1039 | ||
891c49ab | 1040 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1041 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc NP |
1042 | |
1043 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
1044 | ||
db64fe02 NP |
1045 | order = get_order(size); |
1046 | ||
1047 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
7d61bfe8 | 1048 | offset >>= PAGE_SHIFT; |
db64fe02 NP |
1049 | |
1050 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
1051 | rcu_read_lock(); | |
1052 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1053 | rcu_read_unlock(); | |
1054 | BUG_ON(!vb); | |
1055 | ||
64141da5 JF |
1056 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
1057 | ||
db64fe02 | 1058 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1059 | |
1060 | /* Expand dirty range */ | |
1061 | vb->dirty_min = min(vb->dirty_min, offset); | |
1062 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1063 | |
db64fe02 NP |
1064 | vb->dirty += 1UL << order; |
1065 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1066 | BUG_ON(vb->free); |
db64fe02 NP |
1067 | spin_unlock(&vb->lock); |
1068 | free_vmap_block(vb); | |
1069 | } else | |
1070 | spin_unlock(&vb->lock); | |
1071 | } | |
1072 | ||
1073 | /** | |
1074 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1075 | * | |
1076 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1077 | * to amortize TLB flushing overheads. What this means is that any page you | |
1078 | * have now, may, in a former life, have been mapped into kernel virtual | |
1079 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1080 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1081 | * | |
1082 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1083 | * be sure that none of the pages we have control over will have any aliases | |
1084 | * from the vmap layer. | |
1085 | */ | |
1086 | void vm_unmap_aliases(void) | |
1087 | { | |
1088 | unsigned long start = ULONG_MAX, end = 0; | |
1089 | int cpu; | |
1090 | int flush = 0; | |
1091 | ||
9b463334 JF |
1092 | if (unlikely(!vmap_initialized)) |
1093 | return; | |
1094 | ||
5803ed29 CH |
1095 | might_sleep(); |
1096 | ||
db64fe02 NP |
1097 | for_each_possible_cpu(cpu) { |
1098 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1099 | struct vmap_block *vb; | |
1100 | ||
1101 | rcu_read_lock(); | |
1102 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1103 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1104 | if (vb->dirty) { |
1105 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1106 | unsigned long s, e; |
b136be5e | 1107 | |
7d61bfe8 RP |
1108 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1109 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1110 | |
7d61bfe8 RP |
1111 | start = min(s, start); |
1112 | end = max(e, end); | |
db64fe02 | 1113 | |
7d61bfe8 | 1114 | flush = 1; |
db64fe02 NP |
1115 | } |
1116 | spin_unlock(&vb->lock); | |
1117 | } | |
1118 | rcu_read_unlock(); | |
1119 | } | |
1120 | ||
f9e09977 | 1121 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1122 | purge_fragmented_blocks_allcpus(); |
1123 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1124 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1125 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1126 | } |
1127 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
1128 | ||
1129 | /** | |
1130 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1131 | * @mem: the pointer returned by vm_map_ram | |
1132 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1133 | */ | |
1134 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1135 | { | |
65ee03c4 | 1136 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1137 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1138 | struct vmap_area *va; |
db64fe02 | 1139 | |
5803ed29 | 1140 | might_sleep(); |
db64fe02 NP |
1141 | BUG_ON(!addr); |
1142 | BUG_ON(addr < VMALLOC_START); | |
1143 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1144 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 NP |
1145 | |
1146 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 1147 | vmap_debug_free_range(addr, addr+size); |
db64fe02 | 1148 | |
9c3acf60 | 1149 | if (likely(count <= VMAP_MAX_ALLOC)) { |
db64fe02 | 1150 | vb_free(mem, size); |
9c3acf60 CH |
1151 | return; |
1152 | } | |
1153 | ||
1154 | va = find_vmap_area(addr); | |
1155 | BUG_ON(!va); | |
1156 | free_unmap_vmap_area(va); | |
db64fe02 NP |
1157 | } |
1158 | EXPORT_SYMBOL(vm_unmap_ram); | |
1159 | ||
1160 | /** | |
1161 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1162 | * @pages: an array of pointers to the pages to be mapped | |
1163 | * @count: number of pages | |
1164 | * @node: prefer to allocate data structures on this node | |
1165 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1166 | * |
36437638 GK |
1167 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1168 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1169 | * objects with vm_map_ram(), it could consume lots of address space through | |
1170 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1171 | * the end. Please use this function for short-lived objects. | |
1172 | * | |
e99c97ad | 1173 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1174 | */ |
1175 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1176 | { | |
65ee03c4 | 1177 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1178 | unsigned long addr; |
1179 | void *mem; | |
1180 | ||
1181 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1182 | mem = vb_alloc(size, GFP_KERNEL); | |
1183 | if (IS_ERR(mem)) | |
1184 | return NULL; | |
1185 | addr = (unsigned long)mem; | |
1186 | } else { | |
1187 | struct vmap_area *va; | |
1188 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1189 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1190 | if (IS_ERR(va)) | |
1191 | return NULL; | |
1192 | ||
1193 | addr = va->va_start; | |
1194 | mem = (void *)addr; | |
1195 | } | |
1196 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1197 | vm_unmap_ram(mem, count); | |
1198 | return NULL; | |
1199 | } | |
1200 | return mem; | |
1201 | } | |
1202 | EXPORT_SYMBOL(vm_map_ram); | |
1203 | ||
4341fa45 | 1204 | static struct vm_struct *vmlist __initdata; |
be9b7335 NP |
1205 | /** |
1206 | * vm_area_add_early - add vmap area early during boot | |
1207 | * @vm: vm_struct to add | |
1208 | * | |
1209 | * This function is used to add fixed kernel vm area to vmlist before | |
1210 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1211 | * should contain proper values and the other fields should be zero. | |
1212 | * | |
1213 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1214 | */ | |
1215 | void __init vm_area_add_early(struct vm_struct *vm) | |
1216 | { | |
1217 | struct vm_struct *tmp, **p; | |
1218 | ||
1219 | BUG_ON(vmap_initialized); | |
1220 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1221 | if (tmp->addr >= vm->addr) { | |
1222 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1223 | break; | |
1224 | } else | |
1225 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1226 | } | |
1227 | vm->next = *p; | |
1228 | *p = vm; | |
1229 | } | |
1230 | ||
f0aa6617 TH |
1231 | /** |
1232 | * vm_area_register_early - register vmap area early during boot | |
1233 | * @vm: vm_struct to register | |
c0c0a293 | 1234 | * @align: requested alignment |
f0aa6617 TH |
1235 | * |
1236 | * This function is used to register kernel vm area before | |
1237 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1238 | * proper values on entry and other fields should be zero. On return, | |
1239 | * vm->addr contains the allocated address. | |
1240 | * | |
1241 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1242 | */ | |
c0c0a293 | 1243 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1244 | { |
1245 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1246 | unsigned long addr; |
1247 | ||
1248 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1249 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1250 | |
c0c0a293 | 1251 | vm->addr = (void *)addr; |
f0aa6617 | 1252 | |
be9b7335 | 1253 | vm_area_add_early(vm); |
f0aa6617 TH |
1254 | } |
1255 | ||
db64fe02 NP |
1256 | void __init vmalloc_init(void) |
1257 | { | |
822c18f2 IK |
1258 | struct vmap_area *va; |
1259 | struct vm_struct *tmp; | |
db64fe02 NP |
1260 | int i; |
1261 | ||
1262 | for_each_possible_cpu(i) { | |
1263 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1264 | struct vfree_deferred *p; |
db64fe02 NP |
1265 | |
1266 | vbq = &per_cpu(vmap_block_queue, i); | |
1267 | spin_lock_init(&vbq->lock); | |
1268 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1269 | p = &per_cpu(vfree_deferred, i); |
1270 | init_llist_head(&p->list); | |
1271 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1272 | } |
9b463334 | 1273 | |
822c18f2 IK |
1274 | /* Import existing vmlist entries. */ |
1275 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1276 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
dbda591d | 1277 | va->flags = VM_VM_AREA; |
822c18f2 IK |
1278 | va->va_start = (unsigned long)tmp->addr; |
1279 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1280 | va->vm = tmp; |
822c18f2 IK |
1281 | __insert_vmap_area(va); |
1282 | } | |
ca23e405 TH |
1283 | |
1284 | vmap_area_pcpu_hole = VMALLOC_END; | |
1285 | ||
9b463334 | 1286 | vmap_initialized = true; |
db64fe02 NP |
1287 | } |
1288 | ||
8fc48985 TH |
1289 | /** |
1290 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1291 | * @addr: start of the VM area to map | |
1292 | * @size: size of the VM area to map | |
1293 | * @prot: page protection flags to use | |
1294 | * @pages: pages to map | |
1295 | * | |
1296 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1297 | * specify should have been allocated using get_vm_area() and its | |
1298 | * friends. | |
1299 | * | |
1300 | * NOTE: | |
1301 | * This function does NOT do any cache flushing. The caller is | |
1302 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1303 | * before calling this function. | |
1304 | * | |
1305 | * RETURNS: | |
1306 | * The number of pages mapped on success, -errno on failure. | |
1307 | */ | |
1308 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1309 | pgprot_t prot, struct page **pages) | |
1310 | { | |
1311 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1312 | } | |
1313 | ||
1314 | /** | |
1315 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1316 | * @addr: start of the VM area to unmap | |
1317 | * @size: size of the VM area to unmap | |
1318 | * | |
1319 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1320 | * specify should have been allocated using get_vm_area() and its | |
1321 | * friends. | |
1322 | * | |
1323 | * NOTE: | |
1324 | * This function does NOT do any cache flushing. The caller is | |
1325 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1326 | * before calling this function and flush_tlb_kernel_range() after. | |
1327 | */ | |
1328 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1329 | { | |
1330 | vunmap_page_range(addr, addr + size); | |
1331 | } | |
81e88fdc | 1332 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1333 | |
1334 | /** | |
1335 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1336 | * @addr: start of the VM area to unmap | |
1337 | * @size: size of the VM area to unmap | |
1338 | * | |
1339 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1340 | * the unmapping and tlb after. | |
1341 | */ | |
db64fe02 NP |
1342 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1343 | { | |
1344 | unsigned long end = addr + size; | |
f6fcba70 TH |
1345 | |
1346 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1347 | vunmap_page_range(addr, end); |
1348 | flush_tlb_kernel_range(addr, end); | |
1349 | } | |
93ef6d6c | 1350 | EXPORT_SYMBOL_GPL(unmap_kernel_range); |
db64fe02 | 1351 | |
f6f8ed47 | 1352 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
1353 | { |
1354 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 1355 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
1356 | int err; |
1357 | ||
f6f8ed47 | 1358 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 1359 | |
f6f8ed47 | 1360 | return err > 0 ? 0 : err; |
db64fe02 NP |
1361 | } |
1362 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1363 | ||
f5252e00 | 1364 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1365 | unsigned long flags, const void *caller) |
cf88c790 | 1366 | { |
c69480ad | 1367 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
1368 | vm->flags = flags; |
1369 | vm->addr = (void *)va->va_start; | |
1370 | vm->size = va->va_end - va->va_start; | |
1371 | vm->caller = caller; | |
db1aecaf | 1372 | va->vm = vm; |
cf88c790 | 1373 | va->flags |= VM_VM_AREA; |
c69480ad | 1374 | spin_unlock(&vmap_area_lock); |
f5252e00 | 1375 | } |
cf88c790 | 1376 | |
20fc02b4 | 1377 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 1378 | { |
d4033afd | 1379 | /* |
20fc02b4 | 1380 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
1381 | * we should make sure that vm has proper values. |
1382 | * Pair with smp_rmb() in show_numa_info(). | |
1383 | */ | |
1384 | smp_wmb(); | |
20fc02b4 | 1385 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
1386 | } |
1387 | ||
db64fe02 | 1388 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 1389 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 1390 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 1391 | { |
0006526d | 1392 | struct vmap_area *va; |
db64fe02 | 1393 | struct vm_struct *area; |
1da177e4 | 1394 | |
52fd24ca | 1395 | BUG_ON(in_interrupt()); |
1da177e4 | 1396 | size = PAGE_ALIGN(size); |
31be8309 OH |
1397 | if (unlikely(!size)) |
1398 | return NULL; | |
1da177e4 | 1399 | |
252e5c6e | 1400 | if (flags & VM_IOREMAP) |
1401 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
1402 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
1403 | ||
cf88c790 | 1404 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1405 | if (unlikely(!area)) |
1406 | return NULL; | |
1407 | ||
71394fe5 AR |
1408 | if (!(flags & VM_NO_GUARD)) |
1409 | size += PAGE_SIZE; | |
1da177e4 | 1410 | |
db64fe02 NP |
1411 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1412 | if (IS_ERR(va)) { | |
1413 | kfree(area); | |
1414 | return NULL; | |
1da177e4 | 1415 | } |
1da177e4 | 1416 | |
d82b1d85 | 1417 | setup_vmalloc_vm(area, va, flags, caller); |
f5252e00 | 1418 | |
1da177e4 | 1419 | return area; |
1da177e4 LT |
1420 | } |
1421 | ||
930fc45a CL |
1422 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1423 | unsigned long start, unsigned long end) | |
1424 | { | |
00ef2d2f DR |
1425 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1426 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 1427 | } |
5992b6da | 1428 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1429 | |
c2968612 BH |
1430 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1431 | unsigned long start, unsigned long end, | |
5e6cafc8 | 1432 | const void *caller) |
c2968612 | 1433 | { |
00ef2d2f DR |
1434 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1435 | GFP_KERNEL, caller); | |
c2968612 BH |
1436 | } |
1437 | ||
1da177e4 | 1438 | /** |
183ff22b | 1439 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1440 | * @size: size of the area |
1441 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1442 | * | |
1443 | * Search an area of @size in the kernel virtual mapping area, | |
1444 | * and reserved it for out purposes. Returns the area descriptor | |
1445 | * on success or %NULL on failure. | |
1446 | */ | |
1447 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1448 | { | |
2dca6999 | 1449 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
1450 | NUMA_NO_NODE, GFP_KERNEL, |
1451 | __builtin_return_address(0)); | |
23016969 CL |
1452 | } |
1453 | ||
1454 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 1455 | const void *caller) |
23016969 | 1456 | { |
2dca6999 | 1457 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 1458 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
1459 | } |
1460 | ||
e9da6e99 MS |
1461 | /** |
1462 | * find_vm_area - find a continuous kernel virtual area | |
1463 | * @addr: base address | |
1464 | * | |
1465 | * Search for the kernel VM area starting at @addr, and return it. | |
1466 | * It is up to the caller to do all required locking to keep the returned | |
1467 | * pointer valid. | |
1468 | */ | |
1469 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 1470 | { |
db64fe02 | 1471 | struct vmap_area *va; |
83342314 | 1472 | |
db64fe02 NP |
1473 | va = find_vmap_area((unsigned long)addr); |
1474 | if (va && va->flags & VM_VM_AREA) | |
db1aecaf | 1475 | return va->vm; |
1da177e4 | 1476 | |
1da177e4 | 1477 | return NULL; |
1da177e4 LT |
1478 | } |
1479 | ||
7856dfeb | 1480 | /** |
183ff22b | 1481 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1482 | * @addr: base address |
1483 | * | |
1484 | * Search for the kernel VM area starting at @addr, and remove it. | |
1485 | * This function returns the found VM area, but using it is NOT safe | |
1486 | * on SMP machines, except for its size or flags. | |
1487 | */ | |
b3bdda02 | 1488 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1489 | { |
db64fe02 NP |
1490 | struct vmap_area *va; |
1491 | ||
5803ed29 CH |
1492 | might_sleep(); |
1493 | ||
db64fe02 NP |
1494 | va = find_vmap_area((unsigned long)addr); |
1495 | if (va && va->flags & VM_VM_AREA) { | |
db1aecaf | 1496 | struct vm_struct *vm = va->vm; |
f5252e00 | 1497 | |
c69480ad JK |
1498 | spin_lock(&vmap_area_lock); |
1499 | va->vm = NULL; | |
1500 | va->flags &= ~VM_VM_AREA; | |
78c72746 | 1501 | va->flags |= VM_LAZY_FREE; |
c69480ad JK |
1502 | spin_unlock(&vmap_area_lock); |
1503 | ||
dd32c279 | 1504 | vmap_debug_free_range(va->va_start, va->va_end); |
a5af5aa8 | 1505 | kasan_free_shadow(vm); |
dd32c279 | 1506 | free_unmap_vmap_area(va); |
dd32c279 | 1507 | |
db64fe02 NP |
1508 | return vm; |
1509 | } | |
1510 | return NULL; | |
7856dfeb AK |
1511 | } |
1512 | ||
b3bdda02 | 1513 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1514 | { |
1515 | struct vm_struct *area; | |
1516 | ||
1517 | if (!addr) | |
1518 | return; | |
1519 | ||
e69e9d4a | 1520 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 1521 | addr)) |
1da177e4 | 1522 | return; |
1da177e4 LT |
1523 | |
1524 | area = remove_vm_area(addr); | |
1525 | if (unlikely(!area)) { | |
4c8573e2 | 1526 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1527 | addr); |
1da177e4 LT |
1528 | return; |
1529 | } | |
1530 | ||
7511c3ed JM |
1531 | debug_check_no_locks_freed(addr, get_vm_area_size(area)); |
1532 | debug_check_no_obj_freed(addr, get_vm_area_size(area)); | |
9a11b49a | 1533 | |
1da177e4 LT |
1534 | if (deallocate_pages) { |
1535 | int i; | |
1536 | ||
1537 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1538 | struct page *page = area->pages[i]; |
1539 | ||
1540 | BUG_ON(!page); | |
4949148a | 1541 | __free_pages(page, 0); |
1da177e4 LT |
1542 | } |
1543 | ||
244d63ee | 1544 | kvfree(area->pages); |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | kfree(area); | |
1548 | return; | |
1549 | } | |
bf22e37a AR |
1550 | |
1551 | static inline void __vfree_deferred(const void *addr) | |
1552 | { | |
1553 | /* | |
1554 | * Use raw_cpu_ptr() because this can be called from preemptible | |
1555 | * context. Preemption is absolutely fine here, because the llist_add() | |
1556 | * implementation is lockless, so it works even if we are adding to | |
1557 | * nother cpu's list. schedule_work() should be fine with this too. | |
1558 | */ | |
1559 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
1560 | ||
1561 | if (llist_add((struct llist_node *)addr, &p->list)) | |
1562 | schedule_work(&p->wq); | |
1563 | } | |
1564 | ||
1565 | /** | |
1566 | * vfree_atomic - release memory allocated by vmalloc() | |
1567 | * @addr: memory base address | |
1568 | * | |
1569 | * This one is just like vfree() but can be called in any atomic context | |
1570 | * except NMIs. | |
1571 | */ | |
1572 | void vfree_atomic(const void *addr) | |
1573 | { | |
1574 | BUG_ON(in_nmi()); | |
1575 | ||
1576 | kmemleak_free(addr); | |
1577 | ||
1578 | if (!addr) | |
1579 | return; | |
1580 | __vfree_deferred(addr); | |
1581 | } | |
1582 | ||
1da177e4 LT |
1583 | /** |
1584 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1585 | * @addr: memory base address |
1586 | * | |
183ff22b | 1587 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1588 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1589 | * NULL, no operation is performed. | |
1da177e4 | 1590 | * |
32fcfd40 AV |
1591 | * Must not be called in NMI context (strictly speaking, only if we don't |
1592 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
1593 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 1594 | * |
0e056eb5 | 1595 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 1596 | */ |
b3bdda02 | 1597 | void vfree(const void *addr) |
1da177e4 | 1598 | { |
32fcfd40 | 1599 | BUG_ON(in_nmi()); |
89219d37 CM |
1600 | |
1601 | kmemleak_free(addr); | |
1602 | ||
32fcfd40 AV |
1603 | if (!addr) |
1604 | return; | |
bf22e37a AR |
1605 | if (unlikely(in_interrupt())) |
1606 | __vfree_deferred(addr); | |
1607 | else | |
32fcfd40 | 1608 | __vunmap(addr, 1); |
1da177e4 | 1609 | } |
1da177e4 LT |
1610 | EXPORT_SYMBOL(vfree); |
1611 | ||
1612 | /** | |
1613 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1614 | * @addr: memory base address |
1615 | * | |
1616 | * Free the virtually contiguous memory area starting at @addr, | |
1617 | * which was created from the page array passed to vmap(). | |
1618 | * | |
80e93eff | 1619 | * Must not be called in interrupt context. |
1da177e4 | 1620 | */ |
b3bdda02 | 1621 | void vunmap(const void *addr) |
1da177e4 LT |
1622 | { |
1623 | BUG_ON(in_interrupt()); | |
34754b69 | 1624 | might_sleep(); |
32fcfd40 AV |
1625 | if (addr) |
1626 | __vunmap(addr, 0); | |
1da177e4 | 1627 | } |
1da177e4 LT |
1628 | EXPORT_SYMBOL(vunmap); |
1629 | ||
1630 | /** | |
1631 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1632 | * @pages: array of page pointers |
1633 | * @count: number of pages to map | |
1634 | * @flags: vm_area->flags | |
1635 | * @prot: page protection for the mapping | |
1636 | * | |
1637 | * Maps @count pages from @pages into contiguous kernel virtual | |
1638 | * space. | |
1639 | */ | |
1640 | void *vmap(struct page **pages, unsigned int count, | |
1641 | unsigned long flags, pgprot_t prot) | |
1642 | { | |
1643 | struct vm_struct *area; | |
65ee03c4 | 1644 | unsigned long size; /* In bytes */ |
1da177e4 | 1645 | |
34754b69 PZ |
1646 | might_sleep(); |
1647 | ||
4481374c | 1648 | if (count > totalram_pages) |
1da177e4 LT |
1649 | return NULL; |
1650 | ||
65ee03c4 GJM |
1651 | size = (unsigned long)count << PAGE_SHIFT; |
1652 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
1653 | if (!area) |
1654 | return NULL; | |
23016969 | 1655 | |
f6f8ed47 | 1656 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
1657 | vunmap(area->addr); |
1658 | return NULL; | |
1659 | } | |
1660 | ||
1661 | return area->addr; | |
1662 | } | |
1da177e4 LT |
1663 | EXPORT_SYMBOL(vmap); |
1664 | ||
8594a21c MH |
1665 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
1666 | gfp_t gfp_mask, pgprot_t prot, | |
1667 | int node, const void *caller); | |
e31d9eb5 | 1668 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 1669 | pgprot_t prot, int node) |
1da177e4 LT |
1670 | { |
1671 | struct page **pages; | |
1672 | unsigned int nr_pages, array_size, i; | |
930f036b | 1673 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
19809c2d | 1674 | const gfp_t alloc_mask = gfp_mask | __GFP_HIGHMEM | __GFP_NOWARN; |
1da177e4 | 1675 | |
762216ab | 1676 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
1677 | array_size = (nr_pages * sizeof(struct page *)); |
1678 | ||
1679 | area->nr_pages = nr_pages; | |
1680 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1681 | if (array_size > PAGE_SIZE) { |
976d6dfb | 1682 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
3722e13c | 1683 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 1684 | } else { |
976d6dfb | 1685 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 1686 | } |
1da177e4 LT |
1687 | area->pages = pages; |
1688 | if (!area->pages) { | |
1689 | remove_vm_area(area->addr); | |
1690 | kfree(area); | |
1691 | return NULL; | |
1692 | } | |
1da177e4 LT |
1693 | |
1694 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1695 | struct page *page; |
1696 | ||
5d17a73a MH |
1697 | if (fatal_signal_pending(current)) { |
1698 | area->nr_pages = i; | |
171012f5 | 1699 | goto fail_no_warn; |
5d17a73a MH |
1700 | } |
1701 | ||
4b90951c | 1702 | if (node == NUMA_NO_NODE) |
7877cdcc | 1703 | page = alloc_page(alloc_mask); |
930fc45a | 1704 | else |
7877cdcc | 1705 | page = alloc_pages_node(node, alloc_mask, 0); |
bf53d6f8 CL |
1706 | |
1707 | if (unlikely(!page)) { | |
1da177e4 LT |
1708 | /* Successfully allocated i pages, free them in __vunmap() */ |
1709 | area->nr_pages = i; | |
1710 | goto fail; | |
1711 | } | |
bf53d6f8 | 1712 | area->pages[i] = page; |
d0164adc | 1713 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 1714 | cond_resched(); |
1da177e4 LT |
1715 | } |
1716 | ||
f6f8ed47 | 1717 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
1718 | goto fail; |
1719 | return area->addr; | |
1720 | ||
1721 | fail: | |
a8e99259 | 1722 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 1723 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 1724 | (area->nr_pages*PAGE_SIZE), area->size); |
171012f5 | 1725 | fail_no_warn: |
1da177e4 LT |
1726 | vfree(area->addr); |
1727 | return NULL; | |
1728 | } | |
1729 | ||
1730 | /** | |
d0a21265 | 1731 | * __vmalloc_node_range - allocate virtually contiguous memory |
1da177e4 | 1732 | * @size: allocation size |
2dca6999 | 1733 | * @align: desired alignment |
d0a21265 DR |
1734 | * @start: vm area range start |
1735 | * @end: vm area range end | |
1da177e4 LT |
1736 | * @gfp_mask: flags for the page level allocator |
1737 | * @prot: protection mask for the allocated pages | |
cb9e3c29 | 1738 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) |
00ef2d2f | 1739 | * @node: node to use for allocation or NUMA_NO_NODE |
c85d194b | 1740 | * @caller: caller's return address |
1da177e4 LT |
1741 | * |
1742 | * Allocate enough pages to cover @size from the page level | |
1743 | * allocator with @gfp_mask flags. Map them into contiguous | |
1744 | * kernel virtual space, using a pagetable protection of @prot. | |
1745 | */ | |
d0a21265 DR |
1746 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
1747 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
1748 | pgprot_t prot, unsigned long vm_flags, int node, |
1749 | const void *caller) | |
1da177e4 LT |
1750 | { |
1751 | struct vm_struct *area; | |
89219d37 CM |
1752 | void *addr; |
1753 | unsigned long real_size = size; | |
1da177e4 LT |
1754 | |
1755 | size = PAGE_ALIGN(size); | |
4481374c | 1756 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
de7d2b56 | 1757 | goto fail; |
1da177e4 | 1758 | |
cb9e3c29 AR |
1759 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | |
1760 | vm_flags, start, end, node, gfp_mask, caller); | |
1da177e4 | 1761 | if (!area) |
de7d2b56 | 1762 | goto fail; |
1da177e4 | 1763 | |
3722e13c | 1764 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 1765 | if (!addr) |
b82225f3 | 1766 | return NULL; |
89219d37 | 1767 | |
f5252e00 | 1768 | /* |
20fc02b4 ZY |
1769 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
1770 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 1771 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 1772 | */ |
20fc02b4 | 1773 | clear_vm_uninitialized_flag(area); |
f5252e00 | 1774 | |
94f4a161 | 1775 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
1776 | |
1777 | return addr; | |
de7d2b56 JP |
1778 | |
1779 | fail: | |
a8e99259 | 1780 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 1781 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 1782 | return NULL; |
1da177e4 LT |
1783 | } |
1784 | ||
d0a21265 DR |
1785 | /** |
1786 | * __vmalloc_node - allocate virtually contiguous memory | |
1787 | * @size: allocation size | |
1788 | * @align: desired alignment | |
1789 | * @gfp_mask: flags for the page level allocator | |
1790 | * @prot: protection mask for the allocated pages | |
00ef2d2f | 1791 | * @node: node to use for allocation or NUMA_NO_NODE |
d0a21265 DR |
1792 | * @caller: caller's return address |
1793 | * | |
1794 | * Allocate enough pages to cover @size from the page level | |
1795 | * allocator with @gfp_mask flags. Map them into contiguous | |
1796 | * kernel virtual space, using a pagetable protection of @prot. | |
a7c3e901 | 1797 | * |
dcda9b04 | 1798 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
a7c3e901 MH |
1799 | * and __GFP_NOFAIL are not supported |
1800 | * | |
1801 | * Any use of gfp flags outside of GFP_KERNEL should be consulted | |
1802 | * with mm people. | |
1803 | * | |
d0a21265 | 1804 | */ |
8594a21c | 1805 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
d0a21265 | 1806 | gfp_t gfp_mask, pgprot_t prot, |
5e6cafc8 | 1807 | int node, const void *caller) |
d0a21265 DR |
1808 | { |
1809 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 1810 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
1811 | } |
1812 | ||
930fc45a CL |
1813 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1814 | { | |
00ef2d2f | 1815 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 1816 | __builtin_return_address(0)); |
930fc45a | 1817 | } |
1da177e4 LT |
1818 | EXPORT_SYMBOL(__vmalloc); |
1819 | ||
8594a21c MH |
1820 | static inline void *__vmalloc_node_flags(unsigned long size, |
1821 | int node, gfp_t flags) | |
1822 | { | |
1823 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
1824 | node, __builtin_return_address(0)); | |
1825 | } | |
1826 | ||
1827 | ||
1828 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags, | |
1829 | void *caller) | |
1830 | { | |
1831 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller); | |
1832 | } | |
1833 | ||
1da177e4 LT |
1834 | /** |
1835 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1836 | * @size: allocation size |
1da177e4 LT |
1837 | * Allocate enough pages to cover @size from the page level |
1838 | * allocator and map them into contiguous kernel virtual space. | |
1839 | * | |
c1c8897f | 1840 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1841 | * use __vmalloc() instead. |
1842 | */ | |
1843 | void *vmalloc(unsigned long size) | |
1844 | { | |
00ef2d2f | 1845 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 1846 | GFP_KERNEL); |
1da177e4 | 1847 | } |
1da177e4 LT |
1848 | EXPORT_SYMBOL(vmalloc); |
1849 | ||
e1ca7788 DY |
1850 | /** |
1851 | * vzalloc - allocate virtually contiguous memory with zero fill | |
1852 | * @size: allocation size | |
1853 | * Allocate enough pages to cover @size from the page level | |
1854 | * allocator and map them into contiguous kernel virtual space. | |
1855 | * The memory allocated is set to zero. | |
1856 | * | |
1857 | * For tight control over page level allocator and protection flags | |
1858 | * use __vmalloc() instead. | |
1859 | */ | |
1860 | void *vzalloc(unsigned long size) | |
1861 | { | |
00ef2d2f | 1862 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 1863 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
1864 | } |
1865 | EXPORT_SYMBOL(vzalloc); | |
1866 | ||
83342314 | 1867 | /** |
ead04089 REB |
1868 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1869 | * @size: allocation size | |
83342314 | 1870 | * |
ead04089 REB |
1871 | * The resulting memory area is zeroed so it can be mapped to userspace |
1872 | * without leaking data. | |
83342314 NP |
1873 | */ |
1874 | void *vmalloc_user(unsigned long size) | |
1875 | { | |
1876 | struct vm_struct *area; | |
1877 | void *ret; | |
1878 | ||
2dca6999 | 1879 | ret = __vmalloc_node(size, SHMLBA, |
19809c2d | 1880 | GFP_KERNEL | __GFP_ZERO, |
00ef2d2f DR |
1881 | PAGE_KERNEL, NUMA_NO_NODE, |
1882 | __builtin_return_address(0)); | |
2b4ac44e | 1883 | if (ret) { |
db64fe02 | 1884 | area = find_vm_area(ret); |
2b4ac44e | 1885 | area->flags |= VM_USERMAP; |
2b4ac44e | 1886 | } |
83342314 NP |
1887 | return ret; |
1888 | } | |
1889 | EXPORT_SYMBOL(vmalloc_user); | |
1890 | ||
930fc45a CL |
1891 | /** |
1892 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1893 | * @size: allocation size |
d44e0780 | 1894 | * @node: numa node |
930fc45a CL |
1895 | * |
1896 | * Allocate enough pages to cover @size from the page level | |
1897 | * allocator and map them into contiguous kernel virtual space. | |
1898 | * | |
c1c8897f | 1899 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1900 | * use __vmalloc() instead. |
1901 | */ | |
1902 | void *vmalloc_node(unsigned long size, int node) | |
1903 | { | |
19809c2d | 1904 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL, |
23016969 | 1905 | node, __builtin_return_address(0)); |
930fc45a CL |
1906 | } |
1907 | EXPORT_SYMBOL(vmalloc_node); | |
1908 | ||
e1ca7788 DY |
1909 | /** |
1910 | * vzalloc_node - allocate memory on a specific node with zero fill | |
1911 | * @size: allocation size | |
1912 | * @node: numa node | |
1913 | * | |
1914 | * Allocate enough pages to cover @size from the page level | |
1915 | * allocator and map them into contiguous kernel virtual space. | |
1916 | * The memory allocated is set to zero. | |
1917 | * | |
1918 | * For tight control over page level allocator and protection flags | |
1919 | * use __vmalloc_node() instead. | |
1920 | */ | |
1921 | void *vzalloc_node(unsigned long size, int node) | |
1922 | { | |
1923 | return __vmalloc_node_flags(size, node, | |
19809c2d | 1924 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
1925 | } |
1926 | EXPORT_SYMBOL(vzalloc_node); | |
1927 | ||
4dc3b16b PP |
1928 | #ifndef PAGE_KERNEL_EXEC |
1929 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1930 | #endif | |
1931 | ||
1da177e4 LT |
1932 | /** |
1933 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1934 | * @size: allocation size |
1935 | * | |
1936 | * Kernel-internal function to allocate enough pages to cover @size | |
1937 | * the page level allocator and map them into contiguous and | |
1938 | * executable kernel virtual space. | |
1939 | * | |
c1c8897f | 1940 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1941 | * use __vmalloc() instead. |
1942 | */ | |
1943 | ||
1da177e4 LT |
1944 | void *vmalloc_exec(unsigned long size) |
1945 | { | |
19809c2d | 1946 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC, |
00ef2d2f | 1947 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 LT |
1948 | } |
1949 | ||
0d08e0d3 | 1950 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1951 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1952 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1953 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1954 | #else |
1955 | #define GFP_VMALLOC32 GFP_KERNEL | |
1956 | #endif | |
1957 | ||
1da177e4 LT |
1958 | /** |
1959 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1960 | * @size: allocation size |
1961 | * | |
1962 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1963 | * page level allocator and map them into contiguous kernel virtual space. | |
1964 | */ | |
1965 | void *vmalloc_32(unsigned long size) | |
1966 | { | |
2dca6999 | 1967 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 1968 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 1969 | } |
1da177e4 LT |
1970 | EXPORT_SYMBOL(vmalloc_32); |
1971 | ||
83342314 | 1972 | /** |
ead04089 | 1973 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1974 | * @size: allocation size |
ead04089 REB |
1975 | * |
1976 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1977 | * mapped to userspace without leaking data. | |
83342314 NP |
1978 | */ |
1979 | void *vmalloc_32_user(unsigned long size) | |
1980 | { | |
1981 | struct vm_struct *area; | |
1982 | void *ret; | |
1983 | ||
2dca6999 | 1984 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
00ef2d2f | 1985 | NUMA_NO_NODE, __builtin_return_address(0)); |
2b4ac44e | 1986 | if (ret) { |
db64fe02 | 1987 | area = find_vm_area(ret); |
2b4ac44e | 1988 | area->flags |= VM_USERMAP; |
2b4ac44e | 1989 | } |
83342314 NP |
1990 | return ret; |
1991 | } | |
1992 | EXPORT_SYMBOL(vmalloc_32_user); | |
1993 | ||
d0107eb0 KH |
1994 | /* |
1995 | * small helper routine , copy contents to buf from addr. | |
1996 | * If the page is not present, fill zero. | |
1997 | */ | |
1998 | ||
1999 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2000 | { | |
2001 | struct page *p; | |
2002 | int copied = 0; | |
2003 | ||
2004 | while (count) { | |
2005 | unsigned long offset, length; | |
2006 | ||
891c49ab | 2007 | offset = offset_in_page(addr); |
d0107eb0 KH |
2008 | length = PAGE_SIZE - offset; |
2009 | if (length > count) | |
2010 | length = count; | |
2011 | p = vmalloc_to_page(addr); | |
2012 | /* | |
2013 | * To do safe access to this _mapped_ area, we need | |
2014 | * lock. But adding lock here means that we need to add | |
2015 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2016 | * interface, rarely used. Instead of that, we'll use | |
2017 | * kmap() and get small overhead in this access function. | |
2018 | */ | |
2019 | if (p) { | |
2020 | /* | |
2021 | * we can expect USER0 is not used (see vread/vwrite's | |
2022 | * function description) | |
2023 | */ | |
9b04c5fe | 2024 | void *map = kmap_atomic(p); |
d0107eb0 | 2025 | memcpy(buf, map + offset, length); |
9b04c5fe | 2026 | kunmap_atomic(map); |
d0107eb0 KH |
2027 | } else |
2028 | memset(buf, 0, length); | |
2029 | ||
2030 | addr += length; | |
2031 | buf += length; | |
2032 | copied += length; | |
2033 | count -= length; | |
2034 | } | |
2035 | return copied; | |
2036 | } | |
2037 | ||
2038 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2039 | { | |
2040 | struct page *p; | |
2041 | int copied = 0; | |
2042 | ||
2043 | while (count) { | |
2044 | unsigned long offset, length; | |
2045 | ||
891c49ab | 2046 | offset = offset_in_page(addr); |
d0107eb0 KH |
2047 | length = PAGE_SIZE - offset; |
2048 | if (length > count) | |
2049 | length = count; | |
2050 | p = vmalloc_to_page(addr); | |
2051 | /* | |
2052 | * To do safe access to this _mapped_ area, we need | |
2053 | * lock. But adding lock here means that we need to add | |
2054 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2055 | * interface, rarely used. Instead of that, we'll use | |
2056 | * kmap() and get small overhead in this access function. | |
2057 | */ | |
2058 | if (p) { | |
2059 | /* | |
2060 | * we can expect USER0 is not used (see vread/vwrite's | |
2061 | * function description) | |
2062 | */ | |
9b04c5fe | 2063 | void *map = kmap_atomic(p); |
d0107eb0 | 2064 | memcpy(map + offset, buf, length); |
9b04c5fe | 2065 | kunmap_atomic(map); |
d0107eb0 KH |
2066 | } |
2067 | addr += length; | |
2068 | buf += length; | |
2069 | copied += length; | |
2070 | count -= length; | |
2071 | } | |
2072 | return copied; | |
2073 | } | |
2074 | ||
2075 | /** | |
2076 | * vread() - read vmalloc area in a safe way. | |
2077 | * @buf: buffer for reading data | |
2078 | * @addr: vm address. | |
2079 | * @count: number of bytes to be read. | |
2080 | * | |
2081 | * Returns # of bytes which addr and buf should be increased. | |
2082 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't | |
2083 | * includes any intersect with alive vmalloc area. | |
2084 | * | |
2085 | * This function checks that addr is a valid vmalloc'ed area, and | |
2086 | * copy data from that area to a given buffer. If the given memory range | |
2087 | * of [addr...addr+count) includes some valid address, data is copied to | |
2088 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2089 | * IOREMAP area is treated as memory hole and no copy is done. | |
2090 | * | |
2091 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 2092 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
2093 | * |
2094 | * Note: In usual ops, vread() is never necessary because the caller | |
2095 | * should know vmalloc() area is valid and can use memcpy(). | |
2096 | * This is for routines which have to access vmalloc area without | |
2097 | * any informaion, as /dev/kmem. | |
2098 | * | |
2099 | */ | |
2100 | ||
1da177e4 LT |
2101 | long vread(char *buf, char *addr, unsigned long count) |
2102 | { | |
e81ce85f JK |
2103 | struct vmap_area *va; |
2104 | struct vm_struct *vm; | |
1da177e4 | 2105 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2106 | unsigned long buflen = count; |
1da177e4 LT |
2107 | unsigned long n; |
2108 | ||
2109 | /* Don't allow overflow */ | |
2110 | if ((unsigned long) addr + count < count) | |
2111 | count = -(unsigned long) addr; | |
2112 | ||
e81ce85f JK |
2113 | spin_lock(&vmap_area_lock); |
2114 | list_for_each_entry(va, &vmap_area_list, list) { | |
2115 | if (!count) | |
2116 | break; | |
2117 | ||
2118 | if (!(va->flags & VM_VM_AREA)) | |
2119 | continue; | |
2120 | ||
2121 | vm = va->vm; | |
2122 | vaddr = (char *) vm->addr; | |
762216ab | 2123 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2124 | continue; |
2125 | while (addr < vaddr) { | |
2126 | if (count == 0) | |
2127 | goto finished; | |
2128 | *buf = '\0'; | |
2129 | buf++; | |
2130 | addr++; | |
2131 | count--; | |
2132 | } | |
762216ab | 2133 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2134 | if (n > count) |
2135 | n = count; | |
e81ce85f | 2136 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2137 | aligned_vread(buf, addr, n); |
2138 | else /* IOREMAP area is treated as memory hole */ | |
2139 | memset(buf, 0, n); | |
2140 | buf += n; | |
2141 | addr += n; | |
2142 | count -= n; | |
1da177e4 LT |
2143 | } |
2144 | finished: | |
e81ce85f | 2145 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2146 | |
2147 | if (buf == buf_start) | |
2148 | return 0; | |
2149 | /* zero-fill memory holes */ | |
2150 | if (buf != buf_start + buflen) | |
2151 | memset(buf, 0, buflen - (buf - buf_start)); | |
2152 | ||
2153 | return buflen; | |
1da177e4 LT |
2154 | } |
2155 | ||
d0107eb0 KH |
2156 | /** |
2157 | * vwrite() - write vmalloc area in a safe way. | |
2158 | * @buf: buffer for source data | |
2159 | * @addr: vm address. | |
2160 | * @count: number of bytes to be read. | |
2161 | * | |
2162 | * Returns # of bytes which addr and buf should be incresed. | |
2163 | * (same number to @count). | |
2164 | * If [addr...addr+count) doesn't includes any intersect with valid | |
2165 | * vmalloc area, returns 0. | |
2166 | * | |
2167 | * This function checks that addr is a valid vmalloc'ed area, and | |
2168 | * copy data from a buffer to the given addr. If specified range of | |
2169 | * [addr...addr+count) includes some valid address, data is copied from | |
2170 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2171 | * IOREMAP area is treated as memory hole and no copy is done. | |
2172 | * | |
2173 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 2174 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
2175 | * |
2176 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2177 | * should know vmalloc() area is valid and can use memcpy(). | |
2178 | * This is for routines which have to access vmalloc area without | |
2179 | * any informaion, as /dev/kmem. | |
d0107eb0 KH |
2180 | */ |
2181 | ||
1da177e4 LT |
2182 | long vwrite(char *buf, char *addr, unsigned long count) |
2183 | { | |
e81ce85f JK |
2184 | struct vmap_area *va; |
2185 | struct vm_struct *vm; | |
d0107eb0 KH |
2186 | char *vaddr; |
2187 | unsigned long n, buflen; | |
2188 | int copied = 0; | |
1da177e4 LT |
2189 | |
2190 | /* Don't allow overflow */ | |
2191 | if ((unsigned long) addr + count < count) | |
2192 | count = -(unsigned long) addr; | |
d0107eb0 | 2193 | buflen = count; |
1da177e4 | 2194 | |
e81ce85f JK |
2195 | spin_lock(&vmap_area_lock); |
2196 | list_for_each_entry(va, &vmap_area_list, list) { | |
2197 | if (!count) | |
2198 | break; | |
2199 | ||
2200 | if (!(va->flags & VM_VM_AREA)) | |
2201 | continue; | |
2202 | ||
2203 | vm = va->vm; | |
2204 | vaddr = (char *) vm->addr; | |
762216ab | 2205 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2206 | continue; |
2207 | while (addr < vaddr) { | |
2208 | if (count == 0) | |
2209 | goto finished; | |
2210 | buf++; | |
2211 | addr++; | |
2212 | count--; | |
2213 | } | |
762216ab | 2214 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2215 | if (n > count) |
2216 | n = count; | |
e81ce85f | 2217 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2218 | aligned_vwrite(buf, addr, n); |
2219 | copied++; | |
2220 | } | |
2221 | buf += n; | |
2222 | addr += n; | |
2223 | count -= n; | |
1da177e4 LT |
2224 | } |
2225 | finished: | |
e81ce85f | 2226 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2227 | if (!copied) |
2228 | return 0; | |
2229 | return buflen; | |
1da177e4 | 2230 | } |
83342314 NP |
2231 | |
2232 | /** | |
e69e9d4a HD |
2233 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
2234 | * @vma: vma to cover | |
2235 | * @uaddr: target user address to start at | |
2236 | * @kaddr: virtual address of vmalloc kernel memory | |
2237 | * @size: size of map area | |
7682486b RD |
2238 | * |
2239 | * Returns: 0 for success, -Exxx on failure | |
83342314 | 2240 | * |
e69e9d4a HD |
2241 | * This function checks that @kaddr is a valid vmalloc'ed area, |
2242 | * and that it is big enough to cover the range starting at | |
2243 | * @uaddr in @vma. Will return failure if that criteria isn't | |
2244 | * met. | |
83342314 | 2245 | * |
72fd4a35 | 2246 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 2247 | */ |
e69e9d4a HD |
2248 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
2249 | void *kaddr, unsigned long size) | |
83342314 NP |
2250 | { |
2251 | struct vm_struct *area; | |
83342314 | 2252 | |
e69e9d4a HD |
2253 | size = PAGE_ALIGN(size); |
2254 | ||
2255 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
2256 | return -EINVAL; |
2257 | ||
e69e9d4a | 2258 | area = find_vm_area(kaddr); |
83342314 | 2259 | if (!area) |
db64fe02 | 2260 | return -EINVAL; |
83342314 NP |
2261 | |
2262 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2263 | return -EINVAL; |
83342314 | 2264 | |
e69e9d4a | 2265 | if (kaddr + size > area->addr + area->size) |
db64fe02 | 2266 | return -EINVAL; |
83342314 | 2267 | |
83342314 | 2268 | do { |
e69e9d4a | 2269 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
2270 | int ret; |
2271 | ||
83342314 NP |
2272 | ret = vm_insert_page(vma, uaddr, page); |
2273 | if (ret) | |
2274 | return ret; | |
2275 | ||
2276 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
2277 | kaddr += PAGE_SIZE; |
2278 | size -= PAGE_SIZE; | |
2279 | } while (size > 0); | |
83342314 | 2280 | |
314e51b9 | 2281 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 2282 | |
db64fe02 | 2283 | return 0; |
83342314 | 2284 | } |
e69e9d4a HD |
2285 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
2286 | ||
2287 | /** | |
2288 | * remap_vmalloc_range - map vmalloc pages to userspace | |
2289 | * @vma: vma to cover (map full range of vma) | |
2290 | * @addr: vmalloc memory | |
2291 | * @pgoff: number of pages into addr before first page to map | |
2292 | * | |
2293 | * Returns: 0 for success, -Exxx on failure | |
2294 | * | |
2295 | * This function checks that addr is a valid vmalloc'ed area, and | |
2296 | * that it is big enough to cover the vma. Will return failure if | |
2297 | * that criteria isn't met. | |
2298 | * | |
2299 | * Similar to remap_pfn_range() (see mm/memory.c) | |
2300 | */ | |
2301 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
2302 | unsigned long pgoff) | |
2303 | { | |
2304 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
2305 | addr + (pgoff << PAGE_SHIFT), | |
2306 | vma->vm_end - vma->vm_start); | |
2307 | } | |
83342314 NP |
2308 | EXPORT_SYMBOL(remap_vmalloc_range); |
2309 | ||
1eeb66a1 CH |
2310 | /* |
2311 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
2312 | * have one. | |
2313 | */ | |
3b32123d | 2314 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
2315 | { |
2316 | } | |
5f4352fb JF |
2317 | |
2318 | ||
2f569afd | 2319 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb | 2320 | { |
cd12909c DV |
2321 | pte_t ***p = data; |
2322 | ||
2323 | if (p) { | |
2324 | *(*p) = pte; | |
2325 | (*p)++; | |
2326 | } | |
5f4352fb JF |
2327 | return 0; |
2328 | } | |
2329 | ||
2330 | /** | |
2331 | * alloc_vm_area - allocate a range of kernel address space | |
2332 | * @size: size of the area | |
cd12909c | 2333 | * @ptes: returns the PTEs for the address space |
7682486b RD |
2334 | * |
2335 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
2336 | * |
2337 | * This function reserves a range of kernel address space, and | |
2338 | * allocates pagetables to map that range. No actual mappings | |
cd12909c DV |
2339 | * are created. |
2340 | * | |
2341 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) | |
2342 | * allocated for the VM area are returned. | |
5f4352fb | 2343 | */ |
cd12909c | 2344 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
2345 | { |
2346 | struct vm_struct *area; | |
2347 | ||
23016969 CL |
2348 | area = get_vm_area_caller(size, VM_IOREMAP, |
2349 | __builtin_return_address(0)); | |
5f4352fb JF |
2350 | if (area == NULL) |
2351 | return NULL; | |
2352 | ||
2353 | /* | |
2354 | * This ensures that page tables are constructed for this region | |
2355 | * of kernel virtual address space and mapped into init_mm. | |
2356 | */ | |
2357 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 2358 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
2359 | free_vm_area(area); |
2360 | return NULL; | |
2361 | } | |
2362 | ||
5f4352fb JF |
2363 | return area; |
2364 | } | |
2365 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
2366 | ||
2367 | void free_vm_area(struct vm_struct *area) | |
2368 | { | |
2369 | struct vm_struct *ret; | |
2370 | ret = remove_vm_area(area->addr); | |
2371 | BUG_ON(ret != area); | |
2372 | kfree(area); | |
2373 | } | |
2374 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 2375 | |
4f8b02b4 | 2376 | #ifdef CONFIG_SMP |
ca23e405 TH |
2377 | static struct vmap_area *node_to_va(struct rb_node *n) |
2378 | { | |
4583e773 | 2379 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
2380 | } |
2381 | ||
2382 | /** | |
2383 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
2384 | * @end: target address | |
2385 | * @pnext: out arg for the next vmap_area | |
2386 | * @pprev: out arg for the previous vmap_area | |
2387 | * | |
2388 | * Returns: %true if either or both of next and prev are found, | |
2389 | * %false if no vmap_area exists | |
2390 | * | |
2391 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2392 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2393 | */ | |
2394 | static bool pvm_find_next_prev(unsigned long end, | |
2395 | struct vmap_area **pnext, | |
2396 | struct vmap_area **pprev) | |
2397 | { | |
2398 | struct rb_node *n = vmap_area_root.rb_node; | |
2399 | struct vmap_area *va = NULL; | |
2400 | ||
2401 | while (n) { | |
2402 | va = rb_entry(n, struct vmap_area, rb_node); | |
2403 | if (end < va->va_end) | |
2404 | n = n->rb_left; | |
2405 | else if (end > va->va_end) | |
2406 | n = n->rb_right; | |
2407 | else | |
2408 | break; | |
2409 | } | |
2410 | ||
2411 | if (!va) | |
2412 | return false; | |
2413 | ||
2414 | if (va->va_end > end) { | |
2415 | *pnext = va; | |
2416 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2417 | } else { | |
2418 | *pprev = va; | |
2419 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2420 | } | |
2421 | return true; | |
2422 | } | |
2423 | ||
2424 | /** | |
2425 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2426 | * @pnext: in/out arg for the next vmap_area | |
2427 | * @pprev: in/out arg for the previous vmap_area | |
2428 | * @align: alignment | |
2429 | * | |
2430 | * Returns: determined end address | |
2431 | * | |
2432 | * Find the highest aligned address between *@pnext and *@pprev below | |
2433 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2434 | * down address is between the end addresses of the two vmap_areas. | |
2435 | * | |
2436 | * Please note that the address returned by this function may fall | |
2437 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2438 | * that. | |
2439 | */ | |
2440 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2441 | struct vmap_area **pprev, | |
2442 | unsigned long align) | |
2443 | { | |
2444 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2445 | unsigned long addr; | |
2446 | ||
2447 | if (*pnext) | |
2448 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2449 | else | |
2450 | addr = vmalloc_end; | |
2451 | ||
2452 | while (*pprev && (*pprev)->va_end > addr) { | |
2453 | *pnext = *pprev; | |
2454 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2455 | } | |
2456 | ||
2457 | return addr; | |
2458 | } | |
2459 | ||
2460 | /** | |
2461 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2462 | * @offsets: array containing offset of each area | |
2463 | * @sizes: array containing size of each area | |
2464 | * @nr_vms: the number of areas to allocate | |
2465 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
2466 | * |
2467 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2468 | * vm_structs on success, %NULL on failure | |
2469 | * | |
2470 | * Percpu allocator wants to use congruent vm areas so that it can | |
2471 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
2472 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
2473 | * be scattered pretty far, distance between two areas easily going up | |
2474 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
2475 | * areas are allocated from top. | |
ca23e405 TH |
2476 | * |
2477 | * Despite its complicated look, this allocator is rather simple. It | |
2478 | * does everything top-down and scans areas from the end looking for | |
2479 | * matching slot. While scanning, if any of the areas overlaps with | |
2480 | * existing vmap_area, the base address is pulled down to fit the | |
2481 | * area. Scanning is repeated till all the areas fit and then all | |
2482 | * necessary data structres are inserted and the result is returned. | |
2483 | */ | |
2484 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2485 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 2486 | size_t align) |
ca23e405 TH |
2487 | { |
2488 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2489 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2490 | struct vmap_area **vas, *prev, *next; | |
2491 | struct vm_struct **vms; | |
2492 | int area, area2, last_area, term_area; | |
2493 | unsigned long base, start, end, last_end; | |
2494 | bool purged = false; | |
2495 | ||
ca23e405 | 2496 | /* verify parameters and allocate data structures */ |
891c49ab | 2497 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
2498 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
2499 | start = offsets[area]; | |
2500 | end = start + sizes[area]; | |
2501 | ||
2502 | /* is everything aligned properly? */ | |
2503 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2504 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2505 | ||
2506 | /* detect the area with the highest address */ | |
2507 | if (start > offsets[last_area]) | |
2508 | last_area = area; | |
2509 | ||
2510 | for (area2 = 0; area2 < nr_vms; area2++) { | |
2511 | unsigned long start2 = offsets[area2]; | |
2512 | unsigned long end2 = start2 + sizes[area2]; | |
2513 | ||
2514 | if (area2 == area) | |
2515 | continue; | |
2516 | ||
2517 | BUG_ON(start2 >= start && start2 < end); | |
2518 | BUG_ON(end2 <= end && end2 > start); | |
2519 | } | |
2520 | } | |
2521 | last_end = offsets[last_area] + sizes[last_area]; | |
2522 | ||
2523 | if (vmalloc_end - vmalloc_start < last_end) { | |
2524 | WARN_ON(true); | |
2525 | return NULL; | |
2526 | } | |
2527 | ||
4d67d860 TM |
2528 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
2529 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 2530 | if (!vas || !vms) |
f1db7afd | 2531 | goto err_free2; |
ca23e405 TH |
2532 | |
2533 | for (area = 0; area < nr_vms; area++) { | |
ec3f64fc DR |
2534 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
2535 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); | |
ca23e405 TH |
2536 | if (!vas[area] || !vms[area]) |
2537 | goto err_free; | |
2538 | } | |
2539 | retry: | |
2540 | spin_lock(&vmap_area_lock); | |
2541 | ||
2542 | /* start scanning - we scan from the top, begin with the last area */ | |
2543 | area = term_area = last_area; | |
2544 | start = offsets[area]; | |
2545 | end = start + sizes[area]; | |
2546 | ||
2547 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2548 | base = vmalloc_end - last_end; | |
2549 | goto found; | |
2550 | } | |
2551 | base = pvm_determine_end(&next, &prev, align) - end; | |
2552 | ||
2553 | while (true) { | |
2554 | BUG_ON(next && next->va_end <= base + end); | |
2555 | BUG_ON(prev && prev->va_end > base + end); | |
2556 | ||
2557 | /* | |
2558 | * base might have underflowed, add last_end before | |
2559 | * comparing. | |
2560 | */ | |
2561 | if (base + last_end < vmalloc_start + last_end) { | |
2562 | spin_unlock(&vmap_area_lock); | |
2563 | if (!purged) { | |
2564 | purge_vmap_area_lazy(); | |
2565 | purged = true; | |
2566 | goto retry; | |
2567 | } | |
2568 | goto err_free; | |
2569 | } | |
2570 | ||
2571 | /* | |
2572 | * If next overlaps, move base downwards so that it's | |
2573 | * right below next and then recheck. | |
2574 | */ | |
2575 | if (next && next->va_start < base + end) { | |
2576 | base = pvm_determine_end(&next, &prev, align) - end; | |
2577 | term_area = area; | |
2578 | continue; | |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * If prev overlaps, shift down next and prev and move | |
2583 | * base so that it's right below new next and then | |
2584 | * recheck. | |
2585 | */ | |
2586 | if (prev && prev->va_end > base + start) { | |
2587 | next = prev; | |
2588 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2589 | base = pvm_determine_end(&next, &prev, align) - end; | |
2590 | term_area = area; | |
2591 | continue; | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * This area fits, move on to the previous one. If | |
2596 | * the previous one is the terminal one, we're done. | |
2597 | */ | |
2598 | area = (area + nr_vms - 1) % nr_vms; | |
2599 | if (area == term_area) | |
2600 | break; | |
2601 | start = offsets[area]; | |
2602 | end = start + sizes[area]; | |
2603 | pvm_find_next_prev(base + end, &next, &prev); | |
2604 | } | |
2605 | found: | |
2606 | /* we've found a fitting base, insert all va's */ | |
2607 | for (area = 0; area < nr_vms; area++) { | |
2608 | struct vmap_area *va = vas[area]; | |
2609 | ||
2610 | va->va_start = base + offsets[area]; | |
2611 | va->va_end = va->va_start + sizes[area]; | |
2612 | __insert_vmap_area(va); | |
2613 | } | |
2614 | ||
2615 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2616 | ||
2617 | spin_unlock(&vmap_area_lock); | |
2618 | ||
2619 | /* insert all vm's */ | |
2620 | for (area = 0; area < nr_vms; area++) | |
3645cb4a ZY |
2621 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
2622 | pcpu_get_vm_areas); | |
ca23e405 TH |
2623 | |
2624 | kfree(vas); | |
2625 | return vms; | |
2626 | ||
2627 | err_free: | |
2628 | for (area = 0; area < nr_vms; area++) { | |
f1db7afd KC |
2629 | kfree(vas[area]); |
2630 | kfree(vms[area]); | |
ca23e405 | 2631 | } |
f1db7afd | 2632 | err_free2: |
ca23e405 TH |
2633 | kfree(vas); |
2634 | kfree(vms); | |
2635 | return NULL; | |
2636 | } | |
2637 | ||
2638 | /** | |
2639 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2640 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2641 | * @nr_vms: the number of allocated areas | |
2642 | * | |
2643 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2644 | */ | |
2645 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2646 | { | |
2647 | int i; | |
2648 | ||
2649 | for (i = 0; i < nr_vms; i++) | |
2650 | free_vm_area(vms[i]); | |
2651 | kfree(vms); | |
2652 | } | |
4f8b02b4 | 2653 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
2654 | |
2655 | #ifdef CONFIG_PROC_FS | |
2656 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 2657 | __acquires(&vmap_area_lock) |
a10aa579 | 2658 | { |
d4033afd | 2659 | spin_lock(&vmap_area_lock); |
3f500069 | 2660 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
2661 | } |
2662 | ||
2663 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2664 | { | |
3f500069 | 2665 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
2666 | } |
2667 | ||
2668 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 2669 | __releases(&vmap_area_lock) |
a10aa579 | 2670 | { |
d4033afd | 2671 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
2672 | } |
2673 | ||
a47a126a ED |
2674 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2675 | { | |
e5adfffc | 2676 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
2677 | unsigned int nr, *counters = m->private; |
2678 | ||
2679 | if (!counters) | |
2680 | return; | |
2681 | ||
af12346c WL |
2682 | if (v->flags & VM_UNINITIALIZED) |
2683 | return; | |
7e5b528b DV |
2684 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
2685 | smp_rmb(); | |
af12346c | 2686 | |
a47a126a ED |
2687 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
2688 | ||
2689 | for (nr = 0; nr < v->nr_pages; nr++) | |
2690 | counters[page_to_nid(v->pages[nr])]++; | |
2691 | ||
2692 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2693 | if (counters[nr]) | |
2694 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2695 | } | |
2696 | } | |
2697 | ||
a10aa579 CL |
2698 | static int s_show(struct seq_file *m, void *p) |
2699 | { | |
3f500069 | 2700 | struct vmap_area *va; |
d4033afd JK |
2701 | struct vm_struct *v; |
2702 | ||
3f500069 | 2703 | va = list_entry(p, struct vmap_area, list); |
2704 | ||
c2ce8c14 WL |
2705 | /* |
2706 | * s_show can encounter race with remove_vm_area, !VM_VM_AREA on | |
2707 | * behalf of vmap area is being tear down or vm_map_ram allocation. | |
2708 | */ | |
78c72746 YX |
2709 | if (!(va->flags & VM_VM_AREA)) { |
2710 | seq_printf(m, "0x%pK-0x%pK %7ld %s\n", | |
2711 | (void *)va->va_start, (void *)va->va_end, | |
2712 | va->va_end - va->va_start, | |
2713 | va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram"); | |
2714 | ||
d4033afd | 2715 | return 0; |
78c72746 | 2716 | } |
d4033afd JK |
2717 | |
2718 | v = va->vm; | |
a10aa579 | 2719 | |
45ec1690 | 2720 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
2721 | v->addr, v->addr + v->size, v->size); |
2722 | ||
62c70bce JP |
2723 | if (v->caller) |
2724 | seq_printf(m, " %pS", v->caller); | |
23016969 | 2725 | |
a10aa579 CL |
2726 | if (v->nr_pages) |
2727 | seq_printf(m, " pages=%d", v->nr_pages); | |
2728 | ||
2729 | if (v->phys_addr) | |
199eaa05 | 2730 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
2731 | |
2732 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 2733 | seq_puts(m, " ioremap"); |
a10aa579 CL |
2734 | |
2735 | if (v->flags & VM_ALLOC) | |
f4527c90 | 2736 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
2737 | |
2738 | if (v->flags & VM_MAP) | |
f4527c90 | 2739 | seq_puts(m, " vmap"); |
a10aa579 CL |
2740 | |
2741 | if (v->flags & VM_USERMAP) | |
f4527c90 | 2742 | seq_puts(m, " user"); |
a10aa579 | 2743 | |
244d63ee | 2744 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 2745 | seq_puts(m, " vpages"); |
a10aa579 | 2746 | |
a47a126a | 2747 | show_numa_info(m, v); |
a10aa579 CL |
2748 | seq_putc(m, '\n'); |
2749 | return 0; | |
2750 | } | |
2751 | ||
5f6a6a9c | 2752 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2753 | .start = s_start, |
2754 | .next = s_next, | |
2755 | .stop = s_stop, | |
2756 | .show = s_show, | |
2757 | }; | |
5f6a6a9c AD |
2758 | |
2759 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2760 | { | |
703394c1 RJ |
2761 | if (IS_ENABLED(CONFIG_NUMA)) |
2762 | return seq_open_private(file, &vmalloc_op, | |
2763 | nr_node_ids * sizeof(unsigned int)); | |
2764 | else | |
2765 | return seq_open(file, &vmalloc_op); | |
5f6a6a9c AD |
2766 | } |
2767 | ||
2768 | static const struct file_operations proc_vmalloc_operations = { | |
2769 | .open = vmalloc_open, | |
2770 | .read = seq_read, | |
2771 | .llseek = seq_lseek, | |
2772 | .release = seq_release_private, | |
2773 | }; | |
2774 | ||
2775 | static int __init proc_vmalloc_init(void) | |
2776 | { | |
2777 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2778 | return 0; | |
2779 | } | |
2780 | module_init(proc_vmalloc_init); | |
db3808c1 | 2781 | |
a10aa579 CL |
2782 | #endif |
2783 |