]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmalloc.c | |
3 | * | |
4 | * Copyright (C) 1993 Linus Torvalds | |
5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
6 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
7 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
9 | */ |
10 | ||
db64fe02 | 11 | #include <linux/vmalloc.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> | |
14 | #include <linux/highmem.h> | |
d43c36dc | 15 | #include <linux/sched.h> |
1da177e4 LT |
16 | #include <linux/slab.h> |
17 | #include <linux/spinlock.h> | |
18 | #include <linux/interrupt.h> | |
5f6a6a9c | 19 | #include <linux/proc_fs.h> |
a10aa579 | 20 | #include <linux/seq_file.h> |
3ac7fe5a | 21 | #include <linux/debugobjects.h> |
23016969 | 22 | #include <linux/kallsyms.h> |
db64fe02 NP |
23 | #include <linux/list.h> |
24 | #include <linux/rbtree.h> | |
25 | #include <linux/radix-tree.h> | |
26 | #include <linux/rcupdate.h> | |
f0aa6617 | 27 | #include <linux/pfn.h> |
89219d37 | 28 | #include <linux/kmemleak.h> |
60063497 | 29 | #include <linux/atomic.h> |
1da177e4 LT |
30 | #include <asm/uaccess.h> |
31 | #include <asm/tlbflush.h> | |
2dca6999 | 32 | #include <asm/shmparam.h> |
1da177e4 | 33 | |
db64fe02 | 34 | /*** Page table manipulation functions ***/ |
b221385b | 35 | |
1da177e4 LT |
36 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
37 | { | |
38 | pte_t *pte; | |
39 | ||
40 | pte = pte_offset_kernel(pmd, addr); | |
41 | do { | |
42 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
43 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
44 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
45 | } | |
46 | ||
db64fe02 | 47 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
48 | { |
49 | pmd_t *pmd; | |
50 | unsigned long next; | |
51 | ||
52 | pmd = pmd_offset(pud, addr); | |
53 | do { | |
54 | next = pmd_addr_end(addr, end); | |
55 | if (pmd_none_or_clear_bad(pmd)) | |
56 | continue; | |
57 | vunmap_pte_range(pmd, addr, next); | |
58 | } while (pmd++, addr = next, addr != end); | |
59 | } | |
60 | ||
db64fe02 | 61 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
1da177e4 LT |
62 | { |
63 | pud_t *pud; | |
64 | unsigned long next; | |
65 | ||
66 | pud = pud_offset(pgd, addr); | |
67 | do { | |
68 | next = pud_addr_end(addr, end); | |
69 | if (pud_none_or_clear_bad(pud)) | |
70 | continue; | |
71 | vunmap_pmd_range(pud, addr, next); | |
72 | } while (pud++, addr = next, addr != end); | |
73 | } | |
74 | ||
db64fe02 | 75 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
76 | { |
77 | pgd_t *pgd; | |
78 | unsigned long next; | |
1da177e4 LT |
79 | |
80 | BUG_ON(addr >= end); | |
81 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
82 | do { |
83 | next = pgd_addr_end(addr, end); | |
84 | if (pgd_none_or_clear_bad(pgd)) | |
85 | continue; | |
86 | vunmap_pud_range(pgd, addr, next); | |
87 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
88 | } |
89 | ||
90 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 91 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
92 | { |
93 | pte_t *pte; | |
94 | ||
db64fe02 NP |
95 | /* |
96 | * nr is a running index into the array which helps higher level | |
97 | * callers keep track of where we're up to. | |
98 | */ | |
99 | ||
872fec16 | 100 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
101 | if (!pte) |
102 | return -ENOMEM; | |
103 | do { | |
db64fe02 NP |
104 | struct page *page = pages[*nr]; |
105 | ||
106 | if (WARN_ON(!pte_none(*pte))) | |
107 | return -EBUSY; | |
108 | if (WARN_ON(!page)) | |
1da177e4 LT |
109 | return -ENOMEM; |
110 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 111 | (*nr)++; |
1da177e4 LT |
112 | } while (pte++, addr += PAGE_SIZE, addr != end); |
113 | return 0; | |
114 | } | |
115 | ||
db64fe02 NP |
116 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
117 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
118 | { |
119 | pmd_t *pmd; | |
120 | unsigned long next; | |
121 | ||
122 | pmd = pmd_alloc(&init_mm, pud, addr); | |
123 | if (!pmd) | |
124 | return -ENOMEM; | |
125 | do { | |
126 | next = pmd_addr_end(addr, end); | |
db64fe02 | 127 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
128 | return -ENOMEM; |
129 | } while (pmd++, addr = next, addr != end); | |
130 | return 0; | |
131 | } | |
132 | ||
db64fe02 NP |
133 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
134 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
135 | { |
136 | pud_t *pud; | |
137 | unsigned long next; | |
138 | ||
139 | pud = pud_alloc(&init_mm, pgd, addr); | |
140 | if (!pud) | |
141 | return -ENOMEM; | |
142 | do { | |
143 | next = pud_addr_end(addr, end); | |
db64fe02 | 144 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
145 | return -ENOMEM; |
146 | } while (pud++, addr = next, addr != end); | |
147 | return 0; | |
148 | } | |
149 | ||
db64fe02 NP |
150 | /* |
151 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
152 | * will have pfns corresponding to the "pages" array. | |
153 | * | |
154 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
155 | */ | |
8fc48985 TH |
156 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
157 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
158 | { |
159 | pgd_t *pgd; | |
160 | unsigned long next; | |
2e4e27c7 | 161 | unsigned long addr = start; |
db64fe02 NP |
162 | int err = 0; |
163 | int nr = 0; | |
1da177e4 LT |
164 | |
165 | BUG_ON(addr >= end); | |
166 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
167 | do { |
168 | next = pgd_addr_end(addr, end); | |
db64fe02 | 169 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 170 | if (err) |
bf88c8c8 | 171 | return err; |
1da177e4 | 172 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 173 | |
db64fe02 | 174 | return nr; |
1da177e4 LT |
175 | } |
176 | ||
8fc48985 TH |
177 | static int vmap_page_range(unsigned long start, unsigned long end, |
178 | pgprot_t prot, struct page **pages) | |
179 | { | |
180 | int ret; | |
181 | ||
182 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
183 | flush_cache_vmap(start, end); | |
184 | return ret; | |
185 | } | |
186 | ||
81ac3ad9 | 187 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
188 | { |
189 | /* | |
ab4f2ee1 | 190 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
191 | * and fall back on vmalloc() if that fails. Others |
192 | * just put it in the vmalloc space. | |
193 | */ | |
194 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
195 | unsigned long addr = (unsigned long)x; | |
196 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
197 | return 1; | |
198 | #endif | |
199 | return is_vmalloc_addr(x); | |
200 | } | |
201 | ||
48667e7a | 202 | /* |
db64fe02 | 203 | * Walk a vmap address to the struct page it maps. |
48667e7a | 204 | */ |
b3bdda02 | 205 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
206 | { |
207 | unsigned long addr = (unsigned long) vmalloc_addr; | |
208 | struct page *page = NULL; | |
209 | pgd_t *pgd = pgd_offset_k(addr); | |
48667e7a | 210 | |
7aa413de IM |
211 | /* |
212 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
213 | * architectures that do not vmalloc module space | |
214 | */ | |
73bdf0a6 | 215 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 216 | |
48667e7a | 217 | if (!pgd_none(*pgd)) { |
db64fe02 | 218 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 219 | if (!pud_none(*pud)) { |
db64fe02 | 220 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 221 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
222 | pte_t *ptep, pte; |
223 | ||
48667e7a CL |
224 | ptep = pte_offset_map(pmd, addr); |
225 | pte = *ptep; | |
226 | if (pte_present(pte)) | |
227 | page = pte_page(pte); | |
228 | pte_unmap(ptep); | |
229 | } | |
230 | } | |
231 | } | |
232 | return page; | |
233 | } | |
234 | EXPORT_SYMBOL(vmalloc_to_page); | |
235 | ||
236 | /* | |
237 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
238 | */ | |
b3bdda02 | 239 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
240 | { |
241 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
242 | } | |
243 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
244 | ||
db64fe02 NP |
245 | |
246 | /*** Global kva allocator ***/ | |
247 | ||
248 | #define VM_LAZY_FREE 0x01 | |
249 | #define VM_LAZY_FREEING 0x02 | |
250 | #define VM_VM_AREA 0x04 | |
251 | ||
252 | struct vmap_area { | |
253 | unsigned long va_start; | |
254 | unsigned long va_end; | |
255 | unsigned long flags; | |
256 | struct rb_node rb_node; /* address sorted rbtree */ | |
257 | struct list_head list; /* address sorted list */ | |
258 | struct list_head purge_list; /* "lazy purge" list */ | |
db1aecaf | 259 | struct vm_struct *vm; |
db64fe02 NP |
260 | struct rcu_head rcu_head; |
261 | }; | |
262 | ||
263 | static DEFINE_SPINLOCK(vmap_area_lock); | |
f1c4069e JK |
264 | /* Export for kexec only */ |
265 | LIST_HEAD(vmap_area_list); | |
89699605 NP |
266 | static struct rb_root vmap_area_root = RB_ROOT; |
267 | ||
268 | /* The vmap cache globals are protected by vmap_area_lock */ | |
269 | static struct rb_node *free_vmap_cache; | |
270 | static unsigned long cached_hole_size; | |
271 | static unsigned long cached_vstart; | |
272 | static unsigned long cached_align; | |
273 | ||
ca23e405 | 274 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 | 275 | |
f1c4069e JK |
276 | /*** Old vmalloc interfaces ***/ |
277 | static DEFINE_RWLOCK(vmlist_lock); | |
278 | static struct vm_struct *vmlist; | |
279 | ||
db64fe02 | 280 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 281 | { |
db64fe02 NP |
282 | struct rb_node *n = vmap_area_root.rb_node; |
283 | ||
284 | while (n) { | |
285 | struct vmap_area *va; | |
286 | ||
287 | va = rb_entry(n, struct vmap_area, rb_node); | |
288 | if (addr < va->va_start) | |
289 | n = n->rb_left; | |
290 | else if (addr > va->va_start) | |
291 | n = n->rb_right; | |
292 | else | |
293 | return va; | |
294 | } | |
295 | ||
296 | return NULL; | |
297 | } | |
298 | ||
299 | static void __insert_vmap_area(struct vmap_area *va) | |
300 | { | |
301 | struct rb_node **p = &vmap_area_root.rb_node; | |
302 | struct rb_node *parent = NULL; | |
303 | struct rb_node *tmp; | |
304 | ||
305 | while (*p) { | |
170168d0 | 306 | struct vmap_area *tmp_va; |
db64fe02 NP |
307 | |
308 | parent = *p; | |
170168d0 NK |
309 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
310 | if (va->va_start < tmp_va->va_end) | |
db64fe02 | 311 | p = &(*p)->rb_left; |
170168d0 | 312 | else if (va->va_end > tmp_va->va_start) |
db64fe02 NP |
313 | p = &(*p)->rb_right; |
314 | else | |
315 | BUG(); | |
316 | } | |
317 | ||
318 | rb_link_node(&va->rb_node, parent, p); | |
319 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
320 | ||
321 | /* address-sort this list so it is usable like the vmlist */ | |
322 | tmp = rb_prev(&va->rb_node); | |
323 | if (tmp) { | |
324 | struct vmap_area *prev; | |
325 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
326 | list_add_rcu(&va->list, &prev->list); | |
327 | } else | |
328 | list_add_rcu(&va->list, &vmap_area_list); | |
329 | } | |
330 | ||
331 | static void purge_vmap_area_lazy(void); | |
332 | ||
333 | /* | |
334 | * Allocate a region of KVA of the specified size and alignment, within the | |
335 | * vstart and vend. | |
336 | */ | |
337 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
338 | unsigned long align, | |
339 | unsigned long vstart, unsigned long vend, | |
340 | int node, gfp_t gfp_mask) | |
341 | { | |
342 | struct vmap_area *va; | |
343 | struct rb_node *n; | |
1da177e4 | 344 | unsigned long addr; |
db64fe02 | 345 | int purged = 0; |
89699605 | 346 | struct vmap_area *first; |
db64fe02 | 347 | |
7766970c | 348 | BUG_ON(!size); |
db64fe02 | 349 | BUG_ON(size & ~PAGE_MASK); |
89699605 | 350 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 351 | |
db64fe02 NP |
352 | va = kmalloc_node(sizeof(struct vmap_area), |
353 | gfp_mask & GFP_RECLAIM_MASK, node); | |
354 | if (unlikely(!va)) | |
355 | return ERR_PTR(-ENOMEM); | |
356 | ||
357 | retry: | |
358 | spin_lock(&vmap_area_lock); | |
89699605 NP |
359 | /* |
360 | * Invalidate cache if we have more permissive parameters. | |
361 | * cached_hole_size notes the largest hole noticed _below_ | |
362 | * the vmap_area cached in free_vmap_cache: if size fits | |
363 | * into that hole, we want to scan from vstart to reuse | |
364 | * the hole instead of allocating above free_vmap_cache. | |
365 | * Note that __free_vmap_area may update free_vmap_cache | |
366 | * without updating cached_hole_size or cached_align. | |
367 | */ | |
368 | if (!free_vmap_cache || | |
369 | size < cached_hole_size || | |
370 | vstart < cached_vstart || | |
371 | align < cached_align) { | |
372 | nocache: | |
373 | cached_hole_size = 0; | |
374 | free_vmap_cache = NULL; | |
375 | } | |
376 | /* record if we encounter less permissive parameters */ | |
377 | cached_vstart = vstart; | |
378 | cached_align = align; | |
379 | ||
380 | /* find starting point for our search */ | |
381 | if (free_vmap_cache) { | |
382 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
248ac0e1 | 383 | addr = ALIGN(first->va_end, align); |
89699605 NP |
384 | if (addr < vstart) |
385 | goto nocache; | |
386 | if (addr + size - 1 < addr) | |
387 | goto overflow; | |
388 | ||
389 | } else { | |
390 | addr = ALIGN(vstart, align); | |
391 | if (addr + size - 1 < addr) | |
392 | goto overflow; | |
393 | ||
394 | n = vmap_area_root.rb_node; | |
395 | first = NULL; | |
396 | ||
397 | while (n) { | |
db64fe02 NP |
398 | struct vmap_area *tmp; |
399 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
400 | if (tmp->va_end >= addr) { | |
db64fe02 | 401 | first = tmp; |
89699605 NP |
402 | if (tmp->va_start <= addr) |
403 | break; | |
404 | n = n->rb_left; | |
405 | } else | |
db64fe02 | 406 | n = n->rb_right; |
89699605 | 407 | } |
db64fe02 NP |
408 | |
409 | if (!first) | |
410 | goto found; | |
db64fe02 | 411 | } |
89699605 NP |
412 | |
413 | /* from the starting point, walk areas until a suitable hole is found */ | |
248ac0e1 | 414 | while (addr + size > first->va_start && addr + size <= vend) { |
89699605 NP |
415 | if (addr + cached_hole_size < first->va_start) |
416 | cached_hole_size = first->va_start - addr; | |
248ac0e1 | 417 | addr = ALIGN(first->va_end, align); |
89699605 NP |
418 | if (addr + size - 1 < addr) |
419 | goto overflow; | |
420 | ||
92ca922f | 421 | if (list_is_last(&first->list, &vmap_area_list)) |
89699605 | 422 | goto found; |
92ca922f H |
423 | |
424 | first = list_entry(first->list.next, | |
425 | struct vmap_area, list); | |
db64fe02 NP |
426 | } |
427 | ||
89699605 NP |
428 | found: |
429 | if (addr + size > vend) | |
430 | goto overflow; | |
db64fe02 NP |
431 | |
432 | va->va_start = addr; | |
433 | va->va_end = addr + size; | |
434 | va->flags = 0; | |
435 | __insert_vmap_area(va); | |
89699605 | 436 | free_vmap_cache = &va->rb_node; |
db64fe02 NP |
437 | spin_unlock(&vmap_area_lock); |
438 | ||
89699605 NP |
439 | BUG_ON(va->va_start & (align-1)); |
440 | BUG_ON(va->va_start < vstart); | |
441 | BUG_ON(va->va_end > vend); | |
442 | ||
db64fe02 | 443 | return va; |
89699605 NP |
444 | |
445 | overflow: | |
446 | spin_unlock(&vmap_area_lock); | |
447 | if (!purged) { | |
448 | purge_vmap_area_lazy(); | |
449 | purged = 1; | |
450 | goto retry; | |
451 | } | |
452 | if (printk_ratelimit()) | |
453 | printk(KERN_WARNING | |
454 | "vmap allocation for size %lu failed: " | |
455 | "use vmalloc=<size> to increase size.\n", size); | |
456 | kfree(va); | |
457 | return ERR_PTR(-EBUSY); | |
db64fe02 NP |
458 | } |
459 | ||
db64fe02 NP |
460 | static void __free_vmap_area(struct vmap_area *va) |
461 | { | |
462 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
89699605 NP |
463 | |
464 | if (free_vmap_cache) { | |
465 | if (va->va_end < cached_vstart) { | |
466 | free_vmap_cache = NULL; | |
467 | } else { | |
468 | struct vmap_area *cache; | |
469 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
470 | if (va->va_start <= cache->va_start) { | |
471 | free_vmap_cache = rb_prev(&va->rb_node); | |
472 | /* | |
473 | * We don't try to update cached_hole_size or | |
474 | * cached_align, but it won't go very wrong. | |
475 | */ | |
476 | } | |
477 | } | |
478 | } | |
db64fe02 NP |
479 | rb_erase(&va->rb_node, &vmap_area_root); |
480 | RB_CLEAR_NODE(&va->rb_node); | |
481 | list_del_rcu(&va->list); | |
482 | ||
ca23e405 TH |
483 | /* |
484 | * Track the highest possible candidate for pcpu area | |
485 | * allocation. Areas outside of vmalloc area can be returned | |
486 | * here too, consider only end addresses which fall inside | |
487 | * vmalloc area proper. | |
488 | */ | |
489 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
490 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
491 | ||
14769de9 | 492 | kfree_rcu(va, rcu_head); |
db64fe02 NP |
493 | } |
494 | ||
495 | /* | |
496 | * Free a region of KVA allocated by alloc_vmap_area | |
497 | */ | |
498 | static void free_vmap_area(struct vmap_area *va) | |
499 | { | |
500 | spin_lock(&vmap_area_lock); | |
501 | __free_vmap_area(va); | |
502 | spin_unlock(&vmap_area_lock); | |
503 | } | |
504 | ||
505 | /* | |
506 | * Clear the pagetable entries of a given vmap_area | |
507 | */ | |
508 | static void unmap_vmap_area(struct vmap_area *va) | |
509 | { | |
510 | vunmap_page_range(va->va_start, va->va_end); | |
511 | } | |
512 | ||
cd52858c NP |
513 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
514 | { | |
515 | /* | |
516 | * Unmap page tables and force a TLB flush immediately if | |
517 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free | |
518 | * bugs similarly to those in linear kernel virtual address | |
519 | * space after a page has been freed. | |
520 | * | |
521 | * All the lazy freeing logic is still retained, in order to | |
522 | * minimise intrusiveness of this debugging feature. | |
523 | * | |
524 | * This is going to be *slow* (linear kernel virtual address | |
525 | * debugging doesn't do a broadcast TLB flush so it is a lot | |
526 | * faster). | |
527 | */ | |
528 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
529 | vunmap_page_range(start, end); | |
530 | flush_tlb_kernel_range(start, end); | |
531 | #endif | |
532 | } | |
533 | ||
db64fe02 NP |
534 | /* |
535 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
536 | * before attempting to purge with a TLB flush. | |
537 | * | |
538 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
539 | * and take slightly longer to purge, but it will linearly reduce the number of | |
540 | * global TLB flushes that must be performed. It would seem natural to scale | |
541 | * this number up linearly with the number of CPUs (because vmapping activity | |
542 | * could also scale linearly with the number of CPUs), however it is likely | |
543 | * that in practice, workloads might be constrained in other ways that mean | |
544 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
545 | * conservative and not introduce a big latency on huge systems, so go with | |
546 | * a less aggressive log scale. It will still be an improvement over the old | |
547 | * code, and it will be simple to change the scale factor if we find that it | |
548 | * becomes a problem on bigger systems. | |
549 | */ | |
550 | static unsigned long lazy_max_pages(void) | |
551 | { | |
552 | unsigned int log; | |
553 | ||
554 | log = fls(num_online_cpus()); | |
555 | ||
556 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
557 | } | |
558 | ||
559 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
560 | ||
02b709df NP |
561 | /* for per-CPU blocks */ |
562 | static void purge_fragmented_blocks_allcpus(void); | |
563 | ||
3ee48b6a CW |
564 | /* |
565 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
566 | * immediately freed. | |
567 | */ | |
568 | void set_iounmap_nonlazy(void) | |
569 | { | |
570 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); | |
571 | } | |
572 | ||
db64fe02 NP |
573 | /* |
574 | * Purges all lazily-freed vmap areas. | |
575 | * | |
576 | * If sync is 0 then don't purge if there is already a purge in progress. | |
577 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
578 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
579 | * their own TLB flushing). | |
580 | * Returns with *start = min(*start, lowest purged address) | |
581 | * *end = max(*end, highest purged address) | |
582 | */ | |
583 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
584 | int sync, int force_flush) | |
585 | { | |
46666d8a | 586 | static DEFINE_SPINLOCK(purge_lock); |
db64fe02 NP |
587 | LIST_HEAD(valist); |
588 | struct vmap_area *va; | |
cbb76676 | 589 | struct vmap_area *n_va; |
db64fe02 NP |
590 | int nr = 0; |
591 | ||
592 | /* | |
593 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
594 | * should not expect such behaviour. This just simplifies locking for | |
595 | * the case that isn't actually used at the moment anyway. | |
596 | */ | |
597 | if (!sync && !force_flush) { | |
46666d8a | 598 | if (!spin_trylock(&purge_lock)) |
db64fe02 NP |
599 | return; |
600 | } else | |
46666d8a | 601 | spin_lock(&purge_lock); |
db64fe02 | 602 | |
02b709df NP |
603 | if (sync) |
604 | purge_fragmented_blocks_allcpus(); | |
605 | ||
db64fe02 NP |
606 | rcu_read_lock(); |
607 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
608 | if (va->flags & VM_LAZY_FREE) { | |
609 | if (va->va_start < *start) | |
610 | *start = va->va_start; | |
611 | if (va->va_end > *end) | |
612 | *end = va->va_end; | |
613 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
db64fe02 NP |
614 | list_add_tail(&va->purge_list, &valist); |
615 | va->flags |= VM_LAZY_FREEING; | |
616 | va->flags &= ~VM_LAZY_FREE; | |
617 | } | |
618 | } | |
619 | rcu_read_unlock(); | |
620 | ||
88f50044 | 621 | if (nr) |
db64fe02 | 622 | atomic_sub(nr, &vmap_lazy_nr); |
db64fe02 NP |
623 | |
624 | if (nr || force_flush) | |
625 | flush_tlb_kernel_range(*start, *end); | |
626 | ||
627 | if (nr) { | |
628 | spin_lock(&vmap_area_lock); | |
cbb76676 | 629 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
db64fe02 NP |
630 | __free_vmap_area(va); |
631 | spin_unlock(&vmap_area_lock); | |
632 | } | |
46666d8a | 633 | spin_unlock(&purge_lock); |
db64fe02 NP |
634 | } |
635 | ||
496850e5 NP |
636 | /* |
637 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
638 | * is already purging. | |
639 | */ | |
640 | static void try_purge_vmap_area_lazy(void) | |
641 | { | |
642 | unsigned long start = ULONG_MAX, end = 0; | |
643 | ||
644 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
645 | } | |
646 | ||
db64fe02 NP |
647 | /* |
648 | * Kick off a purge of the outstanding lazy areas. | |
649 | */ | |
650 | static void purge_vmap_area_lazy(void) | |
651 | { | |
652 | unsigned long start = ULONG_MAX, end = 0; | |
653 | ||
496850e5 | 654 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
db64fe02 NP |
655 | } |
656 | ||
657 | /* | |
64141da5 JF |
658 | * Free a vmap area, caller ensuring that the area has been unmapped |
659 | * and flush_cache_vunmap had been called for the correct range | |
660 | * previously. | |
db64fe02 | 661 | */ |
64141da5 | 662 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 NP |
663 | { |
664 | va->flags |= VM_LAZY_FREE; | |
665 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
666 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
496850e5 | 667 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
668 | } |
669 | ||
64141da5 JF |
670 | /* |
671 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been | |
672 | * called for the correct range previously. | |
673 | */ | |
674 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) | |
675 | { | |
676 | unmap_vmap_area(va); | |
677 | free_vmap_area_noflush(va); | |
678 | } | |
679 | ||
b29acbdc NP |
680 | /* |
681 | * Free and unmap a vmap area | |
682 | */ | |
683 | static void free_unmap_vmap_area(struct vmap_area *va) | |
684 | { | |
685 | flush_cache_vunmap(va->va_start, va->va_end); | |
686 | free_unmap_vmap_area_noflush(va); | |
687 | } | |
688 | ||
db64fe02 NP |
689 | static struct vmap_area *find_vmap_area(unsigned long addr) |
690 | { | |
691 | struct vmap_area *va; | |
692 | ||
693 | spin_lock(&vmap_area_lock); | |
694 | va = __find_vmap_area(addr); | |
695 | spin_unlock(&vmap_area_lock); | |
696 | ||
697 | return va; | |
698 | } | |
699 | ||
700 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
701 | { | |
702 | struct vmap_area *va; | |
703 | ||
704 | va = find_vmap_area(addr); | |
705 | BUG_ON(!va); | |
706 | free_unmap_vmap_area(va); | |
707 | } | |
708 | ||
709 | ||
710 | /*** Per cpu kva allocator ***/ | |
711 | ||
712 | /* | |
713 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
714 | * room for at least 16 percpu vmap blocks per CPU. | |
715 | */ | |
716 | /* | |
717 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
718 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
719 | * instead (we just need a rough idea) | |
720 | */ | |
721 | #if BITS_PER_LONG == 32 | |
722 | #define VMALLOC_SPACE (128UL*1024*1024) | |
723 | #else | |
724 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
725 | #endif | |
726 | ||
727 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
728 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
729 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
730 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
731 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
732 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
733 | #define VMAP_BBMAP_BITS \ |
734 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
735 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
736 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
737 | |
738 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
739 | ||
9b463334 JF |
740 | static bool vmap_initialized __read_mostly = false; |
741 | ||
db64fe02 NP |
742 | struct vmap_block_queue { |
743 | spinlock_t lock; | |
744 | struct list_head free; | |
db64fe02 NP |
745 | }; |
746 | ||
747 | struct vmap_block { | |
748 | spinlock_t lock; | |
749 | struct vmap_area *va; | |
750 | struct vmap_block_queue *vbq; | |
751 | unsigned long free, dirty; | |
752 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
753 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
de560423 NP |
754 | struct list_head free_list; |
755 | struct rcu_head rcu_head; | |
02b709df | 756 | struct list_head purge; |
db64fe02 NP |
757 | }; |
758 | ||
759 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
760 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
761 | ||
762 | /* | |
763 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
764 | * in the free path. Could get rid of this if we change the API to return a | |
765 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
766 | */ | |
767 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
768 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
769 | ||
770 | /* | |
771 | * We should probably have a fallback mechanism to allocate virtual memory | |
772 | * out of partially filled vmap blocks. However vmap block sizing should be | |
773 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
774 | * big problem. | |
775 | */ | |
776 | ||
777 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
778 | { | |
779 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
780 | addr /= VMAP_BLOCK_SIZE; | |
781 | return addr; | |
782 | } | |
783 | ||
784 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
785 | { | |
786 | struct vmap_block_queue *vbq; | |
787 | struct vmap_block *vb; | |
788 | struct vmap_area *va; | |
789 | unsigned long vb_idx; | |
790 | int node, err; | |
791 | ||
792 | node = numa_node_id(); | |
793 | ||
794 | vb = kmalloc_node(sizeof(struct vmap_block), | |
795 | gfp_mask & GFP_RECLAIM_MASK, node); | |
796 | if (unlikely(!vb)) | |
797 | return ERR_PTR(-ENOMEM); | |
798 | ||
799 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
800 | VMALLOC_START, VMALLOC_END, | |
801 | node, gfp_mask); | |
ddf9c6d4 | 802 | if (IS_ERR(va)) { |
db64fe02 | 803 | kfree(vb); |
e7d86340 | 804 | return ERR_CAST(va); |
db64fe02 NP |
805 | } |
806 | ||
807 | err = radix_tree_preload(gfp_mask); | |
808 | if (unlikely(err)) { | |
809 | kfree(vb); | |
810 | free_vmap_area(va); | |
811 | return ERR_PTR(err); | |
812 | } | |
813 | ||
814 | spin_lock_init(&vb->lock); | |
815 | vb->va = va; | |
816 | vb->free = VMAP_BBMAP_BITS; | |
817 | vb->dirty = 0; | |
818 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
819 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
820 | INIT_LIST_HEAD(&vb->free_list); | |
db64fe02 NP |
821 | |
822 | vb_idx = addr_to_vb_idx(va->va_start); | |
823 | spin_lock(&vmap_block_tree_lock); | |
824 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
825 | spin_unlock(&vmap_block_tree_lock); | |
826 | BUG_ON(err); | |
827 | radix_tree_preload_end(); | |
828 | ||
829 | vbq = &get_cpu_var(vmap_block_queue); | |
830 | vb->vbq = vbq; | |
831 | spin_lock(&vbq->lock); | |
de560423 | 832 | list_add_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 833 | spin_unlock(&vbq->lock); |
3f04ba85 | 834 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
835 | |
836 | return vb; | |
837 | } | |
838 | ||
db64fe02 NP |
839 | static void free_vmap_block(struct vmap_block *vb) |
840 | { | |
841 | struct vmap_block *tmp; | |
842 | unsigned long vb_idx; | |
843 | ||
db64fe02 NP |
844 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
845 | spin_lock(&vmap_block_tree_lock); | |
846 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
847 | spin_unlock(&vmap_block_tree_lock); | |
848 | BUG_ON(tmp != vb); | |
849 | ||
64141da5 | 850 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 851 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
852 | } |
853 | ||
02b709df NP |
854 | static void purge_fragmented_blocks(int cpu) |
855 | { | |
856 | LIST_HEAD(purge); | |
857 | struct vmap_block *vb; | |
858 | struct vmap_block *n_vb; | |
859 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
860 | ||
861 | rcu_read_lock(); | |
862 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
863 | ||
864 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
865 | continue; | |
866 | ||
867 | spin_lock(&vb->lock); | |
868 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
869 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
870 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
871 | bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS); | |
872 | bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS); | |
873 | spin_lock(&vbq->lock); | |
874 | list_del_rcu(&vb->free_list); | |
875 | spin_unlock(&vbq->lock); | |
876 | spin_unlock(&vb->lock); | |
877 | list_add_tail(&vb->purge, &purge); | |
878 | } else | |
879 | spin_unlock(&vb->lock); | |
880 | } | |
881 | rcu_read_unlock(); | |
882 | ||
883 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
884 | list_del(&vb->purge); | |
885 | free_vmap_block(vb); | |
886 | } | |
887 | } | |
888 | ||
889 | static void purge_fragmented_blocks_thiscpu(void) | |
890 | { | |
891 | purge_fragmented_blocks(smp_processor_id()); | |
892 | } | |
893 | ||
894 | static void purge_fragmented_blocks_allcpus(void) | |
895 | { | |
896 | int cpu; | |
897 | ||
898 | for_each_possible_cpu(cpu) | |
899 | purge_fragmented_blocks(cpu); | |
900 | } | |
901 | ||
db64fe02 NP |
902 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
903 | { | |
904 | struct vmap_block_queue *vbq; | |
905 | struct vmap_block *vb; | |
906 | unsigned long addr = 0; | |
907 | unsigned int order; | |
02b709df | 908 | int purge = 0; |
db64fe02 NP |
909 | |
910 | BUG_ON(size & ~PAGE_MASK); | |
911 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
aa91c4d8 JK |
912 | if (WARN_ON(size == 0)) { |
913 | /* | |
914 | * Allocating 0 bytes isn't what caller wants since | |
915 | * get_order(0) returns funny result. Just warn and terminate | |
916 | * early. | |
917 | */ | |
918 | return NULL; | |
919 | } | |
db64fe02 NP |
920 | order = get_order(size); |
921 | ||
922 | again: | |
923 | rcu_read_lock(); | |
924 | vbq = &get_cpu_var(vmap_block_queue); | |
925 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
926 | int i; | |
927 | ||
928 | spin_lock(&vb->lock); | |
02b709df NP |
929 | if (vb->free < 1UL << order) |
930 | goto next; | |
931 | ||
db64fe02 NP |
932 | i = bitmap_find_free_region(vb->alloc_map, |
933 | VMAP_BBMAP_BITS, order); | |
934 | ||
02b709df NP |
935 | if (i < 0) { |
936 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS) { | |
937 | /* fragmented and no outstanding allocations */ | |
938 | BUG_ON(vb->dirty != VMAP_BBMAP_BITS); | |
939 | purge = 1; | |
db64fe02 | 940 | } |
02b709df | 941 | goto next; |
db64fe02 | 942 | } |
02b709df NP |
943 | addr = vb->va->va_start + (i << PAGE_SHIFT); |
944 | BUG_ON(addr_to_vb_idx(addr) != | |
945 | addr_to_vb_idx(vb->va->va_start)); | |
946 | vb->free -= 1UL << order; | |
947 | if (vb->free == 0) { | |
948 | spin_lock(&vbq->lock); | |
949 | list_del_rcu(&vb->free_list); | |
950 | spin_unlock(&vbq->lock); | |
951 | } | |
952 | spin_unlock(&vb->lock); | |
953 | break; | |
954 | next: | |
db64fe02 NP |
955 | spin_unlock(&vb->lock); |
956 | } | |
02b709df NP |
957 | |
958 | if (purge) | |
959 | purge_fragmented_blocks_thiscpu(); | |
960 | ||
3f04ba85 | 961 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
962 | rcu_read_unlock(); |
963 | ||
964 | if (!addr) { | |
965 | vb = new_vmap_block(gfp_mask); | |
966 | if (IS_ERR(vb)) | |
967 | return vb; | |
968 | goto again; | |
969 | } | |
970 | ||
971 | return (void *)addr; | |
972 | } | |
973 | ||
974 | static void vb_free(const void *addr, unsigned long size) | |
975 | { | |
976 | unsigned long offset; | |
977 | unsigned long vb_idx; | |
978 | unsigned int order; | |
979 | struct vmap_block *vb; | |
980 | ||
981 | BUG_ON(size & ~PAGE_MASK); | |
982 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
b29acbdc NP |
983 | |
984 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
985 | ||
db64fe02 NP |
986 | order = get_order(size); |
987 | ||
988 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
989 | ||
990 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
991 | rcu_read_lock(); | |
992 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
993 | rcu_read_unlock(); | |
994 | BUG_ON(!vb); | |
995 | ||
64141da5 JF |
996 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
997 | ||
db64fe02 | 998 | spin_lock(&vb->lock); |
de560423 | 999 | BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order)); |
d086817d | 1000 | |
db64fe02 NP |
1001 | vb->dirty += 1UL << order; |
1002 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1003 | BUG_ON(vb->free); |
db64fe02 NP |
1004 | spin_unlock(&vb->lock); |
1005 | free_vmap_block(vb); | |
1006 | } else | |
1007 | spin_unlock(&vb->lock); | |
1008 | } | |
1009 | ||
1010 | /** | |
1011 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1012 | * | |
1013 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1014 | * to amortize TLB flushing overheads. What this means is that any page you | |
1015 | * have now, may, in a former life, have been mapped into kernel virtual | |
1016 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1017 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1018 | * | |
1019 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1020 | * be sure that none of the pages we have control over will have any aliases | |
1021 | * from the vmap layer. | |
1022 | */ | |
1023 | void vm_unmap_aliases(void) | |
1024 | { | |
1025 | unsigned long start = ULONG_MAX, end = 0; | |
1026 | int cpu; | |
1027 | int flush = 0; | |
1028 | ||
9b463334 JF |
1029 | if (unlikely(!vmap_initialized)) |
1030 | return; | |
1031 | ||
db64fe02 NP |
1032 | for_each_possible_cpu(cpu) { |
1033 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1034 | struct vmap_block *vb; | |
1035 | ||
1036 | rcu_read_lock(); | |
1037 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1038 | int i; | |
1039 | ||
1040 | spin_lock(&vb->lock); | |
1041 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
1042 | while (i < VMAP_BBMAP_BITS) { | |
1043 | unsigned long s, e; | |
1044 | int j; | |
1045 | j = find_next_zero_bit(vb->dirty_map, | |
1046 | VMAP_BBMAP_BITS, i); | |
1047 | ||
1048 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
1049 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
db64fe02 NP |
1050 | flush = 1; |
1051 | ||
1052 | if (s < start) | |
1053 | start = s; | |
1054 | if (e > end) | |
1055 | end = e; | |
1056 | ||
1057 | i = j; | |
1058 | i = find_next_bit(vb->dirty_map, | |
1059 | VMAP_BBMAP_BITS, i); | |
1060 | } | |
1061 | spin_unlock(&vb->lock); | |
1062 | } | |
1063 | rcu_read_unlock(); | |
1064 | } | |
1065 | ||
1066 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
1067 | } | |
1068 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
1069 | ||
1070 | /** | |
1071 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1072 | * @mem: the pointer returned by vm_map_ram | |
1073 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1074 | */ | |
1075 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1076 | { | |
1077 | unsigned long size = count << PAGE_SHIFT; | |
1078 | unsigned long addr = (unsigned long)mem; | |
1079 | ||
1080 | BUG_ON(!addr); | |
1081 | BUG_ON(addr < VMALLOC_START); | |
1082 | BUG_ON(addr > VMALLOC_END); | |
1083 | BUG_ON(addr & (PAGE_SIZE-1)); | |
1084 | ||
1085 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 1086 | vmap_debug_free_range(addr, addr+size); |
db64fe02 NP |
1087 | |
1088 | if (likely(count <= VMAP_MAX_ALLOC)) | |
1089 | vb_free(mem, size); | |
1090 | else | |
1091 | free_unmap_vmap_area_addr(addr); | |
1092 | } | |
1093 | EXPORT_SYMBOL(vm_unmap_ram); | |
1094 | ||
1095 | /** | |
1096 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1097 | * @pages: an array of pointers to the pages to be mapped | |
1098 | * @count: number of pages | |
1099 | * @node: prefer to allocate data structures on this node | |
1100 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad RD |
1101 | * |
1102 | * Returns: a pointer to the address that has been mapped, or %NULL on failure | |
db64fe02 NP |
1103 | */ |
1104 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1105 | { | |
1106 | unsigned long size = count << PAGE_SHIFT; | |
1107 | unsigned long addr; | |
1108 | void *mem; | |
1109 | ||
1110 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1111 | mem = vb_alloc(size, GFP_KERNEL); | |
1112 | if (IS_ERR(mem)) | |
1113 | return NULL; | |
1114 | addr = (unsigned long)mem; | |
1115 | } else { | |
1116 | struct vmap_area *va; | |
1117 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1118 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1119 | if (IS_ERR(va)) | |
1120 | return NULL; | |
1121 | ||
1122 | addr = va->va_start; | |
1123 | mem = (void *)addr; | |
1124 | } | |
1125 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1126 | vm_unmap_ram(mem, count); | |
1127 | return NULL; | |
1128 | } | |
1129 | return mem; | |
1130 | } | |
1131 | EXPORT_SYMBOL(vm_map_ram); | |
1132 | ||
be9b7335 NP |
1133 | /** |
1134 | * vm_area_add_early - add vmap area early during boot | |
1135 | * @vm: vm_struct to add | |
1136 | * | |
1137 | * This function is used to add fixed kernel vm area to vmlist before | |
1138 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1139 | * should contain proper values and the other fields should be zero. | |
1140 | * | |
1141 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1142 | */ | |
1143 | void __init vm_area_add_early(struct vm_struct *vm) | |
1144 | { | |
1145 | struct vm_struct *tmp, **p; | |
1146 | ||
1147 | BUG_ON(vmap_initialized); | |
1148 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1149 | if (tmp->addr >= vm->addr) { | |
1150 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1151 | break; | |
1152 | } else | |
1153 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1154 | } | |
1155 | vm->next = *p; | |
1156 | *p = vm; | |
1157 | } | |
1158 | ||
f0aa6617 TH |
1159 | /** |
1160 | * vm_area_register_early - register vmap area early during boot | |
1161 | * @vm: vm_struct to register | |
c0c0a293 | 1162 | * @align: requested alignment |
f0aa6617 TH |
1163 | * |
1164 | * This function is used to register kernel vm area before | |
1165 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1166 | * proper values on entry and other fields should be zero. On return, | |
1167 | * vm->addr contains the allocated address. | |
1168 | * | |
1169 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1170 | */ | |
c0c0a293 | 1171 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1172 | { |
1173 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1174 | unsigned long addr; |
1175 | ||
1176 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1177 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1178 | |
c0c0a293 | 1179 | vm->addr = (void *)addr; |
f0aa6617 | 1180 | |
be9b7335 | 1181 | vm_area_add_early(vm); |
f0aa6617 TH |
1182 | } |
1183 | ||
db64fe02 NP |
1184 | void __init vmalloc_init(void) |
1185 | { | |
822c18f2 IK |
1186 | struct vmap_area *va; |
1187 | struct vm_struct *tmp; | |
db64fe02 NP |
1188 | int i; |
1189 | ||
1190 | for_each_possible_cpu(i) { | |
1191 | struct vmap_block_queue *vbq; | |
1192 | ||
1193 | vbq = &per_cpu(vmap_block_queue, i); | |
1194 | spin_lock_init(&vbq->lock); | |
1195 | INIT_LIST_HEAD(&vbq->free); | |
db64fe02 | 1196 | } |
9b463334 | 1197 | |
822c18f2 IK |
1198 | /* Import existing vmlist entries. */ |
1199 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1200 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
dbda591d | 1201 | va->flags = VM_VM_AREA; |
822c18f2 IK |
1202 | va->va_start = (unsigned long)tmp->addr; |
1203 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1204 | va->vm = tmp; |
822c18f2 IK |
1205 | __insert_vmap_area(va); |
1206 | } | |
ca23e405 TH |
1207 | |
1208 | vmap_area_pcpu_hole = VMALLOC_END; | |
1209 | ||
9b463334 | 1210 | vmap_initialized = true; |
db64fe02 NP |
1211 | } |
1212 | ||
8fc48985 TH |
1213 | /** |
1214 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1215 | * @addr: start of the VM area to map | |
1216 | * @size: size of the VM area to map | |
1217 | * @prot: page protection flags to use | |
1218 | * @pages: pages to map | |
1219 | * | |
1220 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1221 | * specify should have been allocated using get_vm_area() and its | |
1222 | * friends. | |
1223 | * | |
1224 | * NOTE: | |
1225 | * This function does NOT do any cache flushing. The caller is | |
1226 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1227 | * before calling this function. | |
1228 | * | |
1229 | * RETURNS: | |
1230 | * The number of pages mapped on success, -errno on failure. | |
1231 | */ | |
1232 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1233 | pgprot_t prot, struct page **pages) | |
1234 | { | |
1235 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1236 | } | |
1237 | ||
1238 | /** | |
1239 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1240 | * @addr: start of the VM area to unmap | |
1241 | * @size: size of the VM area to unmap | |
1242 | * | |
1243 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1244 | * specify should have been allocated using get_vm_area() and its | |
1245 | * friends. | |
1246 | * | |
1247 | * NOTE: | |
1248 | * This function does NOT do any cache flushing. The caller is | |
1249 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1250 | * before calling this function and flush_tlb_kernel_range() after. | |
1251 | */ | |
1252 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1253 | { | |
1254 | vunmap_page_range(addr, addr + size); | |
1255 | } | |
81e88fdc | 1256 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1257 | |
1258 | /** | |
1259 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1260 | * @addr: start of the VM area to unmap | |
1261 | * @size: size of the VM area to unmap | |
1262 | * | |
1263 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1264 | * the unmapping and tlb after. | |
1265 | */ | |
db64fe02 NP |
1266 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1267 | { | |
1268 | unsigned long end = addr + size; | |
f6fcba70 TH |
1269 | |
1270 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1271 | vunmap_page_range(addr, end); |
1272 | flush_tlb_kernel_range(addr, end); | |
1273 | } | |
1274 | ||
1275 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
1276 | { | |
1277 | unsigned long addr = (unsigned long)area->addr; | |
1278 | unsigned long end = addr + area->size - PAGE_SIZE; | |
1279 | int err; | |
1280 | ||
1281 | err = vmap_page_range(addr, end, prot, *pages); | |
1282 | if (err > 0) { | |
1283 | *pages += err; | |
1284 | err = 0; | |
1285 | } | |
1286 | ||
1287 | return err; | |
1288 | } | |
1289 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1290 | ||
f5252e00 | 1291 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1292 | unsigned long flags, const void *caller) |
cf88c790 | 1293 | { |
c69480ad | 1294 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
1295 | vm->flags = flags; |
1296 | vm->addr = (void *)va->va_start; | |
1297 | vm->size = va->va_end - va->va_start; | |
1298 | vm->caller = caller; | |
db1aecaf | 1299 | va->vm = vm; |
cf88c790 | 1300 | va->flags |= VM_VM_AREA; |
c69480ad | 1301 | spin_unlock(&vmap_area_lock); |
f5252e00 | 1302 | } |
cf88c790 | 1303 | |
f5252e00 MH |
1304 | static void insert_vmalloc_vmlist(struct vm_struct *vm) |
1305 | { | |
1306 | struct vm_struct *tmp, **p; | |
1307 | ||
d4033afd JK |
1308 | /* |
1309 | * Before removing VM_UNLIST, | |
1310 | * we should make sure that vm has proper values. | |
1311 | * Pair with smp_rmb() in show_numa_info(). | |
1312 | */ | |
1313 | smp_wmb(); | |
f5252e00 | 1314 | vm->flags &= ~VM_UNLIST; |
d4033afd | 1315 | |
cf88c790 TH |
1316 | write_lock(&vmlist_lock); |
1317 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1318 | if (tmp->addr >= vm->addr) | |
1319 | break; | |
1320 | } | |
1321 | vm->next = *p; | |
1322 | *p = vm; | |
1323 | write_unlock(&vmlist_lock); | |
1324 | } | |
1325 | ||
f5252e00 | 1326 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1327 | unsigned long flags, const void *caller) |
f5252e00 MH |
1328 | { |
1329 | setup_vmalloc_vm(vm, va, flags, caller); | |
1330 | insert_vmalloc_vmlist(vm); | |
1331 | } | |
1332 | ||
db64fe02 | 1333 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 1334 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 1335 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 1336 | { |
0006526d | 1337 | struct vmap_area *va; |
db64fe02 | 1338 | struct vm_struct *area; |
1da177e4 | 1339 | |
52fd24ca | 1340 | BUG_ON(in_interrupt()); |
1da177e4 LT |
1341 | if (flags & VM_IOREMAP) { |
1342 | int bit = fls(size); | |
1343 | ||
1344 | if (bit > IOREMAP_MAX_ORDER) | |
1345 | bit = IOREMAP_MAX_ORDER; | |
1346 | else if (bit < PAGE_SHIFT) | |
1347 | bit = PAGE_SHIFT; | |
1348 | ||
1349 | align = 1ul << bit; | |
1350 | } | |
db64fe02 | 1351 | |
1da177e4 | 1352 | size = PAGE_ALIGN(size); |
31be8309 OH |
1353 | if (unlikely(!size)) |
1354 | return NULL; | |
1da177e4 | 1355 | |
cf88c790 | 1356 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1357 | if (unlikely(!area)) |
1358 | return NULL; | |
1359 | ||
1da177e4 LT |
1360 | /* |
1361 | * We always allocate a guard page. | |
1362 | */ | |
1363 | size += PAGE_SIZE; | |
1364 | ||
db64fe02 NP |
1365 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1366 | if (IS_ERR(va)) { | |
1367 | kfree(area); | |
1368 | return NULL; | |
1da177e4 | 1369 | } |
1da177e4 | 1370 | |
f5252e00 MH |
1371 | /* |
1372 | * When this function is called from __vmalloc_node_range, | |
1373 | * we do not add vm_struct to vmlist here to avoid | |
1374 | * accessing uninitialized members of vm_struct such as | |
1375 | * pages and nr_pages fields. They will be set later. | |
1376 | * To distinguish it from others, we use a VM_UNLIST flag. | |
1377 | */ | |
1378 | if (flags & VM_UNLIST) | |
1379 | setup_vmalloc_vm(area, va, flags, caller); | |
1380 | else | |
1381 | insert_vmalloc_vm(area, va, flags, caller); | |
1382 | ||
1da177e4 | 1383 | return area; |
1da177e4 LT |
1384 | } |
1385 | ||
930fc45a CL |
1386 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1387 | unsigned long start, unsigned long end) | |
1388 | { | |
00ef2d2f DR |
1389 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1390 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 1391 | } |
5992b6da | 1392 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1393 | |
c2968612 BH |
1394 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1395 | unsigned long start, unsigned long end, | |
5e6cafc8 | 1396 | const void *caller) |
c2968612 | 1397 | { |
00ef2d2f DR |
1398 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1399 | GFP_KERNEL, caller); | |
c2968612 BH |
1400 | } |
1401 | ||
1da177e4 | 1402 | /** |
183ff22b | 1403 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1404 | * @size: size of the area |
1405 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1406 | * | |
1407 | * Search an area of @size in the kernel virtual mapping area, | |
1408 | * and reserved it for out purposes. Returns the area descriptor | |
1409 | * on success or %NULL on failure. | |
1410 | */ | |
1411 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1412 | { | |
2dca6999 | 1413 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
1414 | NUMA_NO_NODE, GFP_KERNEL, |
1415 | __builtin_return_address(0)); | |
23016969 CL |
1416 | } |
1417 | ||
1418 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 1419 | const void *caller) |
23016969 | 1420 | { |
2dca6999 | 1421 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 1422 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
1423 | } |
1424 | ||
e9da6e99 MS |
1425 | /** |
1426 | * find_vm_area - find a continuous kernel virtual area | |
1427 | * @addr: base address | |
1428 | * | |
1429 | * Search for the kernel VM area starting at @addr, and return it. | |
1430 | * It is up to the caller to do all required locking to keep the returned | |
1431 | * pointer valid. | |
1432 | */ | |
1433 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 1434 | { |
db64fe02 | 1435 | struct vmap_area *va; |
83342314 | 1436 | |
db64fe02 NP |
1437 | va = find_vmap_area((unsigned long)addr); |
1438 | if (va && va->flags & VM_VM_AREA) | |
db1aecaf | 1439 | return va->vm; |
1da177e4 | 1440 | |
1da177e4 | 1441 | return NULL; |
1da177e4 LT |
1442 | } |
1443 | ||
7856dfeb | 1444 | /** |
183ff22b | 1445 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1446 | * @addr: base address |
1447 | * | |
1448 | * Search for the kernel VM area starting at @addr, and remove it. | |
1449 | * This function returns the found VM area, but using it is NOT safe | |
1450 | * on SMP machines, except for its size or flags. | |
1451 | */ | |
b3bdda02 | 1452 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1453 | { |
db64fe02 NP |
1454 | struct vmap_area *va; |
1455 | ||
1456 | va = find_vmap_area((unsigned long)addr); | |
1457 | if (va && va->flags & VM_VM_AREA) { | |
db1aecaf | 1458 | struct vm_struct *vm = va->vm; |
f5252e00 | 1459 | |
c69480ad JK |
1460 | spin_lock(&vmap_area_lock); |
1461 | va->vm = NULL; | |
1462 | va->flags &= ~VM_VM_AREA; | |
1463 | spin_unlock(&vmap_area_lock); | |
1464 | ||
f5252e00 MH |
1465 | if (!(vm->flags & VM_UNLIST)) { |
1466 | struct vm_struct *tmp, **p; | |
1467 | /* | |
1468 | * remove from list and disallow access to | |
1469 | * this vm_struct before unmap. (address range | |
1470 | * confliction is maintained by vmap.) | |
1471 | */ | |
1472 | write_lock(&vmlist_lock); | |
1473 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) | |
1474 | ; | |
1475 | *p = tmp->next; | |
1476 | write_unlock(&vmlist_lock); | |
1477 | } | |
db64fe02 | 1478 | |
dd32c279 KH |
1479 | vmap_debug_free_range(va->va_start, va->va_end); |
1480 | free_unmap_vmap_area(va); | |
1481 | vm->size -= PAGE_SIZE; | |
1482 | ||
db64fe02 NP |
1483 | return vm; |
1484 | } | |
1485 | return NULL; | |
7856dfeb AK |
1486 | } |
1487 | ||
b3bdda02 | 1488 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1489 | { |
1490 | struct vm_struct *area; | |
1491 | ||
1492 | if (!addr) | |
1493 | return; | |
1494 | ||
1495 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1496 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1497 | return; |
1498 | } | |
1499 | ||
1500 | area = remove_vm_area(addr); | |
1501 | if (unlikely(!area)) { | |
4c8573e2 | 1502 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1503 | addr); |
1da177e4 LT |
1504 | return; |
1505 | } | |
1506 | ||
9a11b49a | 1507 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1508 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1509 | |
1da177e4 LT |
1510 | if (deallocate_pages) { |
1511 | int i; | |
1512 | ||
1513 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1514 | struct page *page = area->pages[i]; |
1515 | ||
1516 | BUG_ON(!page); | |
1517 | __free_page(page); | |
1da177e4 LT |
1518 | } |
1519 | ||
8757d5fa | 1520 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1521 | vfree(area->pages); |
1522 | else | |
1523 | kfree(area->pages); | |
1524 | } | |
1525 | ||
1526 | kfree(area); | |
1527 | return; | |
1528 | } | |
1529 | ||
1530 | /** | |
1531 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1532 | * @addr: memory base address |
1533 | * | |
183ff22b | 1534 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1535 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1536 | * NULL, no operation is performed. | |
1da177e4 | 1537 | * |
80e93eff | 1538 | * Must not be called in interrupt context. |
1da177e4 | 1539 | */ |
b3bdda02 | 1540 | void vfree(const void *addr) |
1da177e4 LT |
1541 | { |
1542 | BUG_ON(in_interrupt()); | |
89219d37 CM |
1543 | |
1544 | kmemleak_free(addr); | |
1545 | ||
1da177e4 LT |
1546 | __vunmap(addr, 1); |
1547 | } | |
1da177e4 LT |
1548 | EXPORT_SYMBOL(vfree); |
1549 | ||
1550 | /** | |
1551 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1552 | * @addr: memory base address |
1553 | * | |
1554 | * Free the virtually contiguous memory area starting at @addr, | |
1555 | * which was created from the page array passed to vmap(). | |
1556 | * | |
80e93eff | 1557 | * Must not be called in interrupt context. |
1da177e4 | 1558 | */ |
b3bdda02 | 1559 | void vunmap(const void *addr) |
1da177e4 LT |
1560 | { |
1561 | BUG_ON(in_interrupt()); | |
34754b69 | 1562 | might_sleep(); |
1da177e4 LT |
1563 | __vunmap(addr, 0); |
1564 | } | |
1da177e4 LT |
1565 | EXPORT_SYMBOL(vunmap); |
1566 | ||
1567 | /** | |
1568 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1569 | * @pages: array of page pointers |
1570 | * @count: number of pages to map | |
1571 | * @flags: vm_area->flags | |
1572 | * @prot: page protection for the mapping | |
1573 | * | |
1574 | * Maps @count pages from @pages into contiguous kernel virtual | |
1575 | * space. | |
1576 | */ | |
1577 | void *vmap(struct page **pages, unsigned int count, | |
1578 | unsigned long flags, pgprot_t prot) | |
1579 | { | |
1580 | struct vm_struct *area; | |
1581 | ||
34754b69 PZ |
1582 | might_sleep(); |
1583 | ||
4481374c | 1584 | if (count > totalram_pages) |
1da177e4 LT |
1585 | return NULL; |
1586 | ||
23016969 CL |
1587 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1588 | __builtin_return_address(0)); | |
1da177e4 LT |
1589 | if (!area) |
1590 | return NULL; | |
23016969 | 1591 | |
1da177e4 LT |
1592 | if (map_vm_area(area, prot, &pages)) { |
1593 | vunmap(area->addr); | |
1594 | return NULL; | |
1595 | } | |
1596 | ||
1597 | return area->addr; | |
1598 | } | |
1da177e4 LT |
1599 | EXPORT_SYMBOL(vmap); |
1600 | ||
2dca6999 DM |
1601 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
1602 | gfp_t gfp_mask, pgprot_t prot, | |
5e6cafc8 | 1603 | int node, const void *caller); |
e31d9eb5 | 1604 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
5e6cafc8 | 1605 | pgprot_t prot, int node, const void *caller) |
1da177e4 | 1606 | { |
22943ab1 | 1607 | const int order = 0; |
1da177e4 LT |
1608 | struct page **pages; |
1609 | unsigned int nr_pages, array_size, i; | |
976d6dfb | 1610 | gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
1da177e4 LT |
1611 | |
1612 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1613 | array_size = (nr_pages * sizeof(struct page *)); | |
1614 | ||
1615 | area->nr_pages = nr_pages; | |
1616 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1617 | if (array_size > PAGE_SIZE) { |
976d6dfb | 1618 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
23016969 | 1619 | PAGE_KERNEL, node, caller); |
8757d5fa | 1620 | area->flags |= VM_VPAGES; |
286e1ea3 | 1621 | } else { |
976d6dfb | 1622 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 1623 | } |
1da177e4 | 1624 | area->pages = pages; |
23016969 | 1625 | area->caller = caller; |
1da177e4 LT |
1626 | if (!area->pages) { |
1627 | remove_vm_area(area->addr); | |
1628 | kfree(area); | |
1629 | return NULL; | |
1630 | } | |
1da177e4 LT |
1631 | |
1632 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 | 1633 | struct page *page; |
22943ab1 | 1634 | gfp_t tmp_mask = gfp_mask | __GFP_NOWARN; |
bf53d6f8 | 1635 | |
930fc45a | 1636 | if (node < 0) |
22943ab1 | 1637 | page = alloc_page(tmp_mask); |
930fc45a | 1638 | else |
22943ab1 | 1639 | page = alloc_pages_node(node, tmp_mask, order); |
bf53d6f8 CL |
1640 | |
1641 | if (unlikely(!page)) { | |
1da177e4 LT |
1642 | /* Successfully allocated i pages, free them in __vunmap() */ |
1643 | area->nr_pages = i; | |
1644 | goto fail; | |
1645 | } | |
bf53d6f8 | 1646 | area->pages[i] = page; |
1da177e4 LT |
1647 | } |
1648 | ||
1649 | if (map_vm_area(area, prot, &pages)) | |
1650 | goto fail; | |
1651 | return area->addr; | |
1652 | ||
1653 | fail: | |
3ee9a4f0 JP |
1654 | warn_alloc_failed(gfp_mask, order, |
1655 | "vmalloc: allocation failure, allocated %ld of %ld bytes\n", | |
22943ab1 | 1656 | (area->nr_pages*PAGE_SIZE), area->size); |
1da177e4 LT |
1657 | vfree(area->addr); |
1658 | return NULL; | |
1659 | } | |
1660 | ||
1661 | /** | |
d0a21265 | 1662 | * __vmalloc_node_range - allocate virtually contiguous memory |
1da177e4 | 1663 | * @size: allocation size |
2dca6999 | 1664 | * @align: desired alignment |
d0a21265 DR |
1665 | * @start: vm area range start |
1666 | * @end: vm area range end | |
1da177e4 LT |
1667 | * @gfp_mask: flags for the page level allocator |
1668 | * @prot: protection mask for the allocated pages | |
00ef2d2f | 1669 | * @node: node to use for allocation or NUMA_NO_NODE |
c85d194b | 1670 | * @caller: caller's return address |
1da177e4 LT |
1671 | * |
1672 | * Allocate enough pages to cover @size from the page level | |
1673 | * allocator with @gfp_mask flags. Map them into contiguous | |
1674 | * kernel virtual space, using a pagetable protection of @prot. | |
1675 | */ | |
d0a21265 DR |
1676 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
1677 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
5e6cafc8 | 1678 | pgprot_t prot, int node, const void *caller) |
1da177e4 LT |
1679 | { |
1680 | struct vm_struct *area; | |
89219d37 CM |
1681 | void *addr; |
1682 | unsigned long real_size = size; | |
1da177e4 LT |
1683 | |
1684 | size = PAGE_ALIGN(size); | |
4481374c | 1685 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
de7d2b56 | 1686 | goto fail; |
1da177e4 | 1687 | |
f5252e00 MH |
1688 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST, |
1689 | start, end, node, gfp_mask, caller); | |
1da177e4 | 1690 | if (!area) |
de7d2b56 | 1691 | goto fail; |
1da177e4 | 1692 | |
89219d37 | 1693 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1368edf0 MG |
1694 | if (!addr) |
1695 | return NULL; | |
89219d37 | 1696 | |
f5252e00 MH |
1697 | /* |
1698 | * In this function, newly allocated vm_struct is not added | |
1699 | * to vmlist at __get_vm_area_node(). so, it is added here. | |
1700 | */ | |
1701 | insert_vmalloc_vmlist(area); | |
1702 | ||
89219d37 CM |
1703 | /* |
1704 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1705 | * structures allocated in the __get_vm_area_node() function contain | |
1706 | * references to the virtual address of the vmalloc'ed block. | |
1707 | */ | |
1708 | kmemleak_alloc(addr, real_size, 3, gfp_mask); | |
1709 | ||
1710 | return addr; | |
de7d2b56 JP |
1711 | |
1712 | fail: | |
1713 | warn_alloc_failed(gfp_mask, 0, | |
1714 | "vmalloc: allocation failure: %lu bytes\n", | |
1715 | real_size); | |
1716 | return NULL; | |
1da177e4 LT |
1717 | } |
1718 | ||
d0a21265 DR |
1719 | /** |
1720 | * __vmalloc_node - allocate virtually contiguous memory | |
1721 | * @size: allocation size | |
1722 | * @align: desired alignment | |
1723 | * @gfp_mask: flags for the page level allocator | |
1724 | * @prot: protection mask for the allocated pages | |
00ef2d2f | 1725 | * @node: node to use for allocation or NUMA_NO_NODE |
d0a21265 DR |
1726 | * @caller: caller's return address |
1727 | * | |
1728 | * Allocate enough pages to cover @size from the page level | |
1729 | * allocator with @gfp_mask flags. Map them into contiguous | |
1730 | * kernel virtual space, using a pagetable protection of @prot. | |
1731 | */ | |
1732 | static void *__vmalloc_node(unsigned long size, unsigned long align, | |
1733 | gfp_t gfp_mask, pgprot_t prot, | |
5e6cafc8 | 1734 | int node, const void *caller) |
d0a21265 DR |
1735 | { |
1736 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
1737 | gfp_mask, prot, node, caller); | |
1738 | } | |
1739 | ||
930fc45a CL |
1740 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1741 | { | |
00ef2d2f | 1742 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 1743 | __builtin_return_address(0)); |
930fc45a | 1744 | } |
1da177e4 LT |
1745 | EXPORT_SYMBOL(__vmalloc); |
1746 | ||
e1ca7788 DY |
1747 | static inline void *__vmalloc_node_flags(unsigned long size, |
1748 | int node, gfp_t flags) | |
1749 | { | |
1750 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
1751 | node, __builtin_return_address(0)); | |
1752 | } | |
1753 | ||
1da177e4 LT |
1754 | /** |
1755 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1756 | * @size: allocation size |
1da177e4 LT |
1757 | * Allocate enough pages to cover @size from the page level |
1758 | * allocator and map them into contiguous kernel virtual space. | |
1759 | * | |
c1c8897f | 1760 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1761 | * use __vmalloc() instead. |
1762 | */ | |
1763 | void *vmalloc(unsigned long size) | |
1764 | { | |
00ef2d2f DR |
1765 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
1766 | GFP_KERNEL | __GFP_HIGHMEM); | |
1da177e4 | 1767 | } |
1da177e4 LT |
1768 | EXPORT_SYMBOL(vmalloc); |
1769 | ||
e1ca7788 DY |
1770 | /** |
1771 | * vzalloc - allocate virtually contiguous memory with zero fill | |
1772 | * @size: allocation size | |
1773 | * Allocate enough pages to cover @size from the page level | |
1774 | * allocator and map them into contiguous kernel virtual space. | |
1775 | * The memory allocated is set to zero. | |
1776 | * | |
1777 | * For tight control over page level allocator and protection flags | |
1778 | * use __vmalloc() instead. | |
1779 | */ | |
1780 | void *vzalloc(unsigned long size) | |
1781 | { | |
00ef2d2f | 1782 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
e1ca7788 DY |
1783 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
1784 | } | |
1785 | EXPORT_SYMBOL(vzalloc); | |
1786 | ||
83342314 | 1787 | /** |
ead04089 REB |
1788 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1789 | * @size: allocation size | |
83342314 | 1790 | * |
ead04089 REB |
1791 | * The resulting memory area is zeroed so it can be mapped to userspace |
1792 | * without leaking data. | |
83342314 NP |
1793 | */ |
1794 | void *vmalloc_user(unsigned long size) | |
1795 | { | |
1796 | struct vm_struct *area; | |
1797 | void *ret; | |
1798 | ||
2dca6999 DM |
1799 | ret = __vmalloc_node(size, SHMLBA, |
1800 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
00ef2d2f DR |
1801 | PAGE_KERNEL, NUMA_NO_NODE, |
1802 | __builtin_return_address(0)); | |
2b4ac44e | 1803 | if (ret) { |
db64fe02 | 1804 | area = find_vm_area(ret); |
2b4ac44e | 1805 | area->flags |= VM_USERMAP; |
2b4ac44e | 1806 | } |
83342314 NP |
1807 | return ret; |
1808 | } | |
1809 | EXPORT_SYMBOL(vmalloc_user); | |
1810 | ||
930fc45a CL |
1811 | /** |
1812 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1813 | * @size: allocation size |
d44e0780 | 1814 | * @node: numa node |
930fc45a CL |
1815 | * |
1816 | * Allocate enough pages to cover @size from the page level | |
1817 | * allocator and map them into contiguous kernel virtual space. | |
1818 | * | |
c1c8897f | 1819 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1820 | * use __vmalloc() instead. |
1821 | */ | |
1822 | void *vmalloc_node(unsigned long size, int node) | |
1823 | { | |
2dca6999 | 1824 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
23016969 | 1825 | node, __builtin_return_address(0)); |
930fc45a CL |
1826 | } |
1827 | EXPORT_SYMBOL(vmalloc_node); | |
1828 | ||
e1ca7788 DY |
1829 | /** |
1830 | * vzalloc_node - allocate memory on a specific node with zero fill | |
1831 | * @size: allocation size | |
1832 | * @node: numa node | |
1833 | * | |
1834 | * Allocate enough pages to cover @size from the page level | |
1835 | * allocator and map them into contiguous kernel virtual space. | |
1836 | * The memory allocated is set to zero. | |
1837 | * | |
1838 | * For tight control over page level allocator and protection flags | |
1839 | * use __vmalloc_node() instead. | |
1840 | */ | |
1841 | void *vzalloc_node(unsigned long size, int node) | |
1842 | { | |
1843 | return __vmalloc_node_flags(size, node, | |
1844 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); | |
1845 | } | |
1846 | EXPORT_SYMBOL(vzalloc_node); | |
1847 | ||
4dc3b16b PP |
1848 | #ifndef PAGE_KERNEL_EXEC |
1849 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1850 | #endif | |
1851 | ||
1da177e4 LT |
1852 | /** |
1853 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1854 | * @size: allocation size |
1855 | * | |
1856 | * Kernel-internal function to allocate enough pages to cover @size | |
1857 | * the page level allocator and map them into contiguous and | |
1858 | * executable kernel virtual space. | |
1859 | * | |
c1c8897f | 1860 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1861 | * use __vmalloc() instead. |
1862 | */ | |
1863 | ||
1da177e4 LT |
1864 | void *vmalloc_exec(unsigned long size) |
1865 | { | |
2dca6999 | 1866 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
00ef2d2f | 1867 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 LT |
1868 | } |
1869 | ||
0d08e0d3 | 1870 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1871 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1872 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1873 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1874 | #else |
1875 | #define GFP_VMALLOC32 GFP_KERNEL | |
1876 | #endif | |
1877 | ||
1da177e4 LT |
1878 | /** |
1879 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1880 | * @size: allocation size |
1881 | * | |
1882 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1883 | * page level allocator and map them into contiguous kernel virtual space. | |
1884 | */ | |
1885 | void *vmalloc_32(unsigned long size) | |
1886 | { | |
2dca6999 | 1887 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 1888 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 1889 | } |
1da177e4 LT |
1890 | EXPORT_SYMBOL(vmalloc_32); |
1891 | ||
83342314 | 1892 | /** |
ead04089 | 1893 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1894 | * @size: allocation size |
ead04089 REB |
1895 | * |
1896 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1897 | * mapped to userspace without leaking data. | |
83342314 NP |
1898 | */ |
1899 | void *vmalloc_32_user(unsigned long size) | |
1900 | { | |
1901 | struct vm_struct *area; | |
1902 | void *ret; | |
1903 | ||
2dca6999 | 1904 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
00ef2d2f | 1905 | NUMA_NO_NODE, __builtin_return_address(0)); |
2b4ac44e | 1906 | if (ret) { |
db64fe02 | 1907 | area = find_vm_area(ret); |
2b4ac44e | 1908 | area->flags |= VM_USERMAP; |
2b4ac44e | 1909 | } |
83342314 NP |
1910 | return ret; |
1911 | } | |
1912 | EXPORT_SYMBOL(vmalloc_32_user); | |
1913 | ||
d0107eb0 KH |
1914 | /* |
1915 | * small helper routine , copy contents to buf from addr. | |
1916 | * If the page is not present, fill zero. | |
1917 | */ | |
1918 | ||
1919 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
1920 | { | |
1921 | struct page *p; | |
1922 | int copied = 0; | |
1923 | ||
1924 | while (count) { | |
1925 | unsigned long offset, length; | |
1926 | ||
1927 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1928 | length = PAGE_SIZE - offset; | |
1929 | if (length > count) | |
1930 | length = count; | |
1931 | p = vmalloc_to_page(addr); | |
1932 | /* | |
1933 | * To do safe access to this _mapped_ area, we need | |
1934 | * lock. But adding lock here means that we need to add | |
1935 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1936 | * interface, rarely used. Instead of that, we'll use | |
1937 | * kmap() and get small overhead in this access function. | |
1938 | */ | |
1939 | if (p) { | |
1940 | /* | |
1941 | * we can expect USER0 is not used (see vread/vwrite's | |
1942 | * function description) | |
1943 | */ | |
9b04c5fe | 1944 | void *map = kmap_atomic(p); |
d0107eb0 | 1945 | memcpy(buf, map + offset, length); |
9b04c5fe | 1946 | kunmap_atomic(map); |
d0107eb0 KH |
1947 | } else |
1948 | memset(buf, 0, length); | |
1949 | ||
1950 | addr += length; | |
1951 | buf += length; | |
1952 | copied += length; | |
1953 | count -= length; | |
1954 | } | |
1955 | return copied; | |
1956 | } | |
1957 | ||
1958 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
1959 | { | |
1960 | struct page *p; | |
1961 | int copied = 0; | |
1962 | ||
1963 | while (count) { | |
1964 | unsigned long offset, length; | |
1965 | ||
1966 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1967 | length = PAGE_SIZE - offset; | |
1968 | if (length > count) | |
1969 | length = count; | |
1970 | p = vmalloc_to_page(addr); | |
1971 | /* | |
1972 | * To do safe access to this _mapped_ area, we need | |
1973 | * lock. But adding lock here means that we need to add | |
1974 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1975 | * interface, rarely used. Instead of that, we'll use | |
1976 | * kmap() and get small overhead in this access function. | |
1977 | */ | |
1978 | if (p) { | |
1979 | /* | |
1980 | * we can expect USER0 is not used (see vread/vwrite's | |
1981 | * function description) | |
1982 | */ | |
9b04c5fe | 1983 | void *map = kmap_atomic(p); |
d0107eb0 | 1984 | memcpy(map + offset, buf, length); |
9b04c5fe | 1985 | kunmap_atomic(map); |
d0107eb0 KH |
1986 | } |
1987 | addr += length; | |
1988 | buf += length; | |
1989 | copied += length; | |
1990 | count -= length; | |
1991 | } | |
1992 | return copied; | |
1993 | } | |
1994 | ||
1995 | /** | |
1996 | * vread() - read vmalloc area in a safe way. | |
1997 | * @buf: buffer for reading data | |
1998 | * @addr: vm address. | |
1999 | * @count: number of bytes to be read. | |
2000 | * | |
2001 | * Returns # of bytes which addr and buf should be increased. | |
2002 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't | |
2003 | * includes any intersect with alive vmalloc area. | |
2004 | * | |
2005 | * This function checks that addr is a valid vmalloc'ed area, and | |
2006 | * copy data from that area to a given buffer. If the given memory range | |
2007 | * of [addr...addr+count) includes some valid address, data is copied to | |
2008 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2009 | * IOREMAP area is treated as memory hole and no copy is done. | |
2010 | * | |
2011 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 2012 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
2013 | * |
2014 | * Note: In usual ops, vread() is never necessary because the caller | |
2015 | * should know vmalloc() area is valid and can use memcpy(). | |
2016 | * This is for routines which have to access vmalloc area without | |
2017 | * any informaion, as /dev/kmem. | |
2018 | * | |
2019 | */ | |
2020 | ||
1da177e4 LT |
2021 | long vread(char *buf, char *addr, unsigned long count) |
2022 | { | |
e81ce85f JK |
2023 | struct vmap_area *va; |
2024 | struct vm_struct *vm; | |
1da177e4 | 2025 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2026 | unsigned long buflen = count; |
1da177e4 LT |
2027 | unsigned long n; |
2028 | ||
2029 | /* Don't allow overflow */ | |
2030 | if ((unsigned long) addr + count < count) | |
2031 | count = -(unsigned long) addr; | |
2032 | ||
e81ce85f JK |
2033 | spin_lock(&vmap_area_lock); |
2034 | list_for_each_entry(va, &vmap_area_list, list) { | |
2035 | if (!count) | |
2036 | break; | |
2037 | ||
2038 | if (!(va->flags & VM_VM_AREA)) | |
2039 | continue; | |
2040 | ||
2041 | vm = va->vm; | |
2042 | vaddr = (char *) vm->addr; | |
2043 | if (addr >= vaddr + vm->size - PAGE_SIZE) | |
1da177e4 LT |
2044 | continue; |
2045 | while (addr < vaddr) { | |
2046 | if (count == 0) | |
2047 | goto finished; | |
2048 | *buf = '\0'; | |
2049 | buf++; | |
2050 | addr++; | |
2051 | count--; | |
2052 | } | |
e81ce85f | 2053 | n = vaddr + vm->size - PAGE_SIZE - addr; |
d0107eb0 KH |
2054 | if (n > count) |
2055 | n = count; | |
e81ce85f | 2056 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2057 | aligned_vread(buf, addr, n); |
2058 | else /* IOREMAP area is treated as memory hole */ | |
2059 | memset(buf, 0, n); | |
2060 | buf += n; | |
2061 | addr += n; | |
2062 | count -= n; | |
1da177e4 LT |
2063 | } |
2064 | finished: | |
e81ce85f | 2065 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2066 | |
2067 | if (buf == buf_start) | |
2068 | return 0; | |
2069 | /* zero-fill memory holes */ | |
2070 | if (buf != buf_start + buflen) | |
2071 | memset(buf, 0, buflen - (buf - buf_start)); | |
2072 | ||
2073 | return buflen; | |
1da177e4 LT |
2074 | } |
2075 | ||
d0107eb0 KH |
2076 | /** |
2077 | * vwrite() - write vmalloc area in a safe way. | |
2078 | * @buf: buffer for source data | |
2079 | * @addr: vm address. | |
2080 | * @count: number of bytes to be read. | |
2081 | * | |
2082 | * Returns # of bytes which addr and buf should be incresed. | |
2083 | * (same number to @count). | |
2084 | * If [addr...addr+count) doesn't includes any intersect with valid | |
2085 | * vmalloc area, returns 0. | |
2086 | * | |
2087 | * This function checks that addr is a valid vmalloc'ed area, and | |
2088 | * copy data from a buffer to the given addr. If specified range of | |
2089 | * [addr...addr+count) includes some valid address, data is copied from | |
2090 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2091 | * IOREMAP area is treated as memory hole and no copy is done. | |
2092 | * | |
2093 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 2094 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
2095 | * |
2096 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2097 | * should know vmalloc() area is valid and can use memcpy(). | |
2098 | * This is for routines which have to access vmalloc area without | |
2099 | * any informaion, as /dev/kmem. | |
d0107eb0 KH |
2100 | */ |
2101 | ||
1da177e4 LT |
2102 | long vwrite(char *buf, char *addr, unsigned long count) |
2103 | { | |
e81ce85f JK |
2104 | struct vmap_area *va; |
2105 | struct vm_struct *vm; | |
d0107eb0 KH |
2106 | char *vaddr; |
2107 | unsigned long n, buflen; | |
2108 | int copied = 0; | |
1da177e4 LT |
2109 | |
2110 | /* Don't allow overflow */ | |
2111 | if ((unsigned long) addr + count < count) | |
2112 | count = -(unsigned long) addr; | |
d0107eb0 | 2113 | buflen = count; |
1da177e4 | 2114 | |
e81ce85f JK |
2115 | spin_lock(&vmap_area_lock); |
2116 | list_for_each_entry(va, &vmap_area_list, list) { | |
2117 | if (!count) | |
2118 | break; | |
2119 | ||
2120 | if (!(va->flags & VM_VM_AREA)) | |
2121 | continue; | |
2122 | ||
2123 | vm = va->vm; | |
2124 | vaddr = (char *) vm->addr; | |
2125 | if (addr >= vaddr + vm->size - PAGE_SIZE) | |
1da177e4 LT |
2126 | continue; |
2127 | while (addr < vaddr) { | |
2128 | if (count == 0) | |
2129 | goto finished; | |
2130 | buf++; | |
2131 | addr++; | |
2132 | count--; | |
2133 | } | |
e81ce85f | 2134 | n = vaddr + vm->size - PAGE_SIZE - addr; |
d0107eb0 KH |
2135 | if (n > count) |
2136 | n = count; | |
e81ce85f | 2137 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2138 | aligned_vwrite(buf, addr, n); |
2139 | copied++; | |
2140 | } | |
2141 | buf += n; | |
2142 | addr += n; | |
2143 | count -= n; | |
1da177e4 LT |
2144 | } |
2145 | finished: | |
e81ce85f | 2146 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2147 | if (!copied) |
2148 | return 0; | |
2149 | return buflen; | |
1da177e4 | 2150 | } |
83342314 NP |
2151 | |
2152 | /** | |
2153 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
2154 | * @vma: vma to cover (map full range of vma) |
2155 | * @addr: vmalloc memory | |
2156 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
2157 | * |
2158 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
2159 | * |
2160 | * This function checks that addr is a valid vmalloc'ed area, and | |
2161 | * that it is big enough to cover the vma. Will return failure if | |
2162 | * that criteria isn't met. | |
2163 | * | |
72fd4a35 | 2164 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
2165 | */ |
2166 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
2167 | unsigned long pgoff) | |
2168 | { | |
2169 | struct vm_struct *area; | |
2170 | unsigned long uaddr = vma->vm_start; | |
2171 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
2172 | |
2173 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
2174 | return -EINVAL; | |
2175 | ||
db64fe02 | 2176 | area = find_vm_area(addr); |
83342314 | 2177 | if (!area) |
db64fe02 | 2178 | return -EINVAL; |
83342314 NP |
2179 | |
2180 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2181 | return -EINVAL; |
83342314 NP |
2182 | |
2183 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 2184 | return -EINVAL; |
83342314 NP |
2185 | |
2186 | addr += pgoff << PAGE_SHIFT; | |
2187 | do { | |
2188 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
2189 | int ret; |
2190 | ||
83342314 NP |
2191 | ret = vm_insert_page(vma, uaddr, page); |
2192 | if (ret) | |
2193 | return ret; | |
2194 | ||
2195 | uaddr += PAGE_SIZE; | |
2196 | addr += PAGE_SIZE; | |
2197 | usize -= PAGE_SIZE; | |
2198 | } while (usize > 0); | |
2199 | ||
314e51b9 | 2200 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 2201 | |
db64fe02 | 2202 | return 0; |
83342314 NP |
2203 | } |
2204 | EXPORT_SYMBOL(remap_vmalloc_range); | |
2205 | ||
1eeb66a1 CH |
2206 | /* |
2207 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
2208 | * have one. | |
2209 | */ | |
2210 | void __attribute__((weak)) vmalloc_sync_all(void) | |
2211 | { | |
2212 | } | |
5f4352fb JF |
2213 | |
2214 | ||
2f569afd | 2215 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb | 2216 | { |
cd12909c DV |
2217 | pte_t ***p = data; |
2218 | ||
2219 | if (p) { | |
2220 | *(*p) = pte; | |
2221 | (*p)++; | |
2222 | } | |
5f4352fb JF |
2223 | return 0; |
2224 | } | |
2225 | ||
2226 | /** | |
2227 | * alloc_vm_area - allocate a range of kernel address space | |
2228 | * @size: size of the area | |
cd12909c | 2229 | * @ptes: returns the PTEs for the address space |
7682486b RD |
2230 | * |
2231 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
2232 | * |
2233 | * This function reserves a range of kernel address space, and | |
2234 | * allocates pagetables to map that range. No actual mappings | |
cd12909c DV |
2235 | * are created. |
2236 | * | |
2237 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) | |
2238 | * allocated for the VM area are returned. | |
5f4352fb | 2239 | */ |
cd12909c | 2240 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
2241 | { |
2242 | struct vm_struct *area; | |
2243 | ||
23016969 CL |
2244 | area = get_vm_area_caller(size, VM_IOREMAP, |
2245 | __builtin_return_address(0)); | |
5f4352fb JF |
2246 | if (area == NULL) |
2247 | return NULL; | |
2248 | ||
2249 | /* | |
2250 | * This ensures that page tables are constructed for this region | |
2251 | * of kernel virtual address space and mapped into init_mm. | |
2252 | */ | |
2253 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 2254 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
2255 | free_vm_area(area); |
2256 | return NULL; | |
2257 | } | |
2258 | ||
5f4352fb JF |
2259 | return area; |
2260 | } | |
2261 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
2262 | ||
2263 | void free_vm_area(struct vm_struct *area) | |
2264 | { | |
2265 | struct vm_struct *ret; | |
2266 | ret = remove_vm_area(area->addr); | |
2267 | BUG_ON(ret != area); | |
2268 | kfree(area); | |
2269 | } | |
2270 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 2271 | |
4f8b02b4 | 2272 | #ifdef CONFIG_SMP |
ca23e405 TH |
2273 | static struct vmap_area *node_to_va(struct rb_node *n) |
2274 | { | |
2275 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; | |
2276 | } | |
2277 | ||
2278 | /** | |
2279 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
2280 | * @end: target address | |
2281 | * @pnext: out arg for the next vmap_area | |
2282 | * @pprev: out arg for the previous vmap_area | |
2283 | * | |
2284 | * Returns: %true if either or both of next and prev are found, | |
2285 | * %false if no vmap_area exists | |
2286 | * | |
2287 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2288 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2289 | */ | |
2290 | static bool pvm_find_next_prev(unsigned long end, | |
2291 | struct vmap_area **pnext, | |
2292 | struct vmap_area **pprev) | |
2293 | { | |
2294 | struct rb_node *n = vmap_area_root.rb_node; | |
2295 | struct vmap_area *va = NULL; | |
2296 | ||
2297 | while (n) { | |
2298 | va = rb_entry(n, struct vmap_area, rb_node); | |
2299 | if (end < va->va_end) | |
2300 | n = n->rb_left; | |
2301 | else if (end > va->va_end) | |
2302 | n = n->rb_right; | |
2303 | else | |
2304 | break; | |
2305 | } | |
2306 | ||
2307 | if (!va) | |
2308 | return false; | |
2309 | ||
2310 | if (va->va_end > end) { | |
2311 | *pnext = va; | |
2312 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2313 | } else { | |
2314 | *pprev = va; | |
2315 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2316 | } | |
2317 | return true; | |
2318 | } | |
2319 | ||
2320 | /** | |
2321 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2322 | * @pnext: in/out arg for the next vmap_area | |
2323 | * @pprev: in/out arg for the previous vmap_area | |
2324 | * @align: alignment | |
2325 | * | |
2326 | * Returns: determined end address | |
2327 | * | |
2328 | * Find the highest aligned address between *@pnext and *@pprev below | |
2329 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2330 | * down address is between the end addresses of the two vmap_areas. | |
2331 | * | |
2332 | * Please note that the address returned by this function may fall | |
2333 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2334 | * that. | |
2335 | */ | |
2336 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2337 | struct vmap_area **pprev, | |
2338 | unsigned long align) | |
2339 | { | |
2340 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2341 | unsigned long addr; | |
2342 | ||
2343 | if (*pnext) | |
2344 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2345 | else | |
2346 | addr = vmalloc_end; | |
2347 | ||
2348 | while (*pprev && (*pprev)->va_end > addr) { | |
2349 | *pnext = *pprev; | |
2350 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2351 | } | |
2352 | ||
2353 | return addr; | |
2354 | } | |
2355 | ||
2356 | /** | |
2357 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2358 | * @offsets: array containing offset of each area | |
2359 | * @sizes: array containing size of each area | |
2360 | * @nr_vms: the number of areas to allocate | |
2361 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
2362 | * |
2363 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2364 | * vm_structs on success, %NULL on failure | |
2365 | * | |
2366 | * Percpu allocator wants to use congruent vm areas so that it can | |
2367 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
2368 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
2369 | * be scattered pretty far, distance between two areas easily going up | |
2370 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
2371 | * areas are allocated from top. | |
ca23e405 TH |
2372 | * |
2373 | * Despite its complicated look, this allocator is rather simple. It | |
2374 | * does everything top-down and scans areas from the end looking for | |
2375 | * matching slot. While scanning, if any of the areas overlaps with | |
2376 | * existing vmap_area, the base address is pulled down to fit the | |
2377 | * area. Scanning is repeated till all the areas fit and then all | |
2378 | * necessary data structres are inserted and the result is returned. | |
2379 | */ | |
2380 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2381 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 2382 | size_t align) |
ca23e405 TH |
2383 | { |
2384 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2385 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2386 | struct vmap_area **vas, *prev, *next; | |
2387 | struct vm_struct **vms; | |
2388 | int area, area2, last_area, term_area; | |
2389 | unsigned long base, start, end, last_end; | |
2390 | bool purged = false; | |
2391 | ||
ca23e405 TH |
2392 | /* verify parameters and allocate data structures */ |
2393 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); | |
2394 | for (last_area = 0, area = 0; area < nr_vms; area++) { | |
2395 | start = offsets[area]; | |
2396 | end = start + sizes[area]; | |
2397 | ||
2398 | /* is everything aligned properly? */ | |
2399 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2400 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2401 | ||
2402 | /* detect the area with the highest address */ | |
2403 | if (start > offsets[last_area]) | |
2404 | last_area = area; | |
2405 | ||
2406 | for (area2 = 0; area2 < nr_vms; area2++) { | |
2407 | unsigned long start2 = offsets[area2]; | |
2408 | unsigned long end2 = start2 + sizes[area2]; | |
2409 | ||
2410 | if (area2 == area) | |
2411 | continue; | |
2412 | ||
2413 | BUG_ON(start2 >= start && start2 < end); | |
2414 | BUG_ON(end2 <= end && end2 > start); | |
2415 | } | |
2416 | } | |
2417 | last_end = offsets[last_area] + sizes[last_area]; | |
2418 | ||
2419 | if (vmalloc_end - vmalloc_start < last_end) { | |
2420 | WARN_ON(true); | |
2421 | return NULL; | |
2422 | } | |
2423 | ||
4d67d860 TM |
2424 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
2425 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 2426 | if (!vas || !vms) |
f1db7afd | 2427 | goto err_free2; |
ca23e405 TH |
2428 | |
2429 | for (area = 0; area < nr_vms; area++) { | |
ec3f64fc DR |
2430 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
2431 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); | |
ca23e405 TH |
2432 | if (!vas[area] || !vms[area]) |
2433 | goto err_free; | |
2434 | } | |
2435 | retry: | |
2436 | spin_lock(&vmap_area_lock); | |
2437 | ||
2438 | /* start scanning - we scan from the top, begin with the last area */ | |
2439 | area = term_area = last_area; | |
2440 | start = offsets[area]; | |
2441 | end = start + sizes[area]; | |
2442 | ||
2443 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2444 | base = vmalloc_end - last_end; | |
2445 | goto found; | |
2446 | } | |
2447 | base = pvm_determine_end(&next, &prev, align) - end; | |
2448 | ||
2449 | while (true) { | |
2450 | BUG_ON(next && next->va_end <= base + end); | |
2451 | BUG_ON(prev && prev->va_end > base + end); | |
2452 | ||
2453 | /* | |
2454 | * base might have underflowed, add last_end before | |
2455 | * comparing. | |
2456 | */ | |
2457 | if (base + last_end < vmalloc_start + last_end) { | |
2458 | spin_unlock(&vmap_area_lock); | |
2459 | if (!purged) { | |
2460 | purge_vmap_area_lazy(); | |
2461 | purged = true; | |
2462 | goto retry; | |
2463 | } | |
2464 | goto err_free; | |
2465 | } | |
2466 | ||
2467 | /* | |
2468 | * If next overlaps, move base downwards so that it's | |
2469 | * right below next and then recheck. | |
2470 | */ | |
2471 | if (next && next->va_start < base + end) { | |
2472 | base = pvm_determine_end(&next, &prev, align) - end; | |
2473 | term_area = area; | |
2474 | continue; | |
2475 | } | |
2476 | ||
2477 | /* | |
2478 | * If prev overlaps, shift down next and prev and move | |
2479 | * base so that it's right below new next and then | |
2480 | * recheck. | |
2481 | */ | |
2482 | if (prev && prev->va_end > base + start) { | |
2483 | next = prev; | |
2484 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2485 | base = pvm_determine_end(&next, &prev, align) - end; | |
2486 | term_area = area; | |
2487 | continue; | |
2488 | } | |
2489 | ||
2490 | /* | |
2491 | * This area fits, move on to the previous one. If | |
2492 | * the previous one is the terminal one, we're done. | |
2493 | */ | |
2494 | area = (area + nr_vms - 1) % nr_vms; | |
2495 | if (area == term_area) | |
2496 | break; | |
2497 | start = offsets[area]; | |
2498 | end = start + sizes[area]; | |
2499 | pvm_find_next_prev(base + end, &next, &prev); | |
2500 | } | |
2501 | found: | |
2502 | /* we've found a fitting base, insert all va's */ | |
2503 | for (area = 0; area < nr_vms; area++) { | |
2504 | struct vmap_area *va = vas[area]; | |
2505 | ||
2506 | va->va_start = base + offsets[area]; | |
2507 | va->va_end = va->va_start + sizes[area]; | |
2508 | __insert_vmap_area(va); | |
2509 | } | |
2510 | ||
2511 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2512 | ||
2513 | spin_unlock(&vmap_area_lock); | |
2514 | ||
2515 | /* insert all vm's */ | |
2516 | for (area = 0; area < nr_vms; area++) | |
2517 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, | |
2518 | pcpu_get_vm_areas); | |
2519 | ||
2520 | kfree(vas); | |
2521 | return vms; | |
2522 | ||
2523 | err_free: | |
2524 | for (area = 0; area < nr_vms; area++) { | |
f1db7afd KC |
2525 | kfree(vas[area]); |
2526 | kfree(vms[area]); | |
ca23e405 | 2527 | } |
f1db7afd | 2528 | err_free2: |
ca23e405 TH |
2529 | kfree(vas); |
2530 | kfree(vms); | |
2531 | return NULL; | |
2532 | } | |
2533 | ||
2534 | /** | |
2535 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2536 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2537 | * @nr_vms: the number of allocated areas | |
2538 | * | |
2539 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2540 | */ | |
2541 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2542 | { | |
2543 | int i; | |
2544 | ||
2545 | for (i = 0; i < nr_vms; i++) | |
2546 | free_vm_area(vms[i]); | |
2547 | kfree(vms); | |
2548 | } | |
4f8b02b4 | 2549 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
2550 | |
2551 | #ifdef CONFIG_PROC_FS | |
2552 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 2553 | __acquires(&vmap_area_lock) |
a10aa579 CL |
2554 | { |
2555 | loff_t n = *pos; | |
d4033afd | 2556 | struct vmap_area *va; |
a10aa579 | 2557 | |
d4033afd JK |
2558 | spin_lock(&vmap_area_lock); |
2559 | va = list_entry((&vmap_area_list)->next, typeof(*va), list); | |
2560 | while (n > 0 && &va->list != &vmap_area_list) { | |
a10aa579 | 2561 | n--; |
d4033afd | 2562 | va = list_entry(va->list.next, typeof(*va), list); |
a10aa579 | 2563 | } |
d4033afd JK |
2564 | if (!n && &va->list != &vmap_area_list) |
2565 | return va; | |
a10aa579 CL |
2566 | |
2567 | return NULL; | |
2568 | ||
2569 | } | |
2570 | ||
2571 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2572 | { | |
d4033afd | 2573 | struct vmap_area *va = p, *next; |
a10aa579 CL |
2574 | |
2575 | ++*pos; | |
d4033afd JK |
2576 | next = list_entry(va->list.next, typeof(*va), list); |
2577 | if (&next->list != &vmap_area_list) | |
2578 | return next; | |
2579 | ||
2580 | return NULL; | |
a10aa579 CL |
2581 | } |
2582 | ||
2583 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 2584 | __releases(&vmap_area_lock) |
a10aa579 | 2585 | { |
d4033afd | 2586 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
2587 | } |
2588 | ||
a47a126a ED |
2589 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2590 | { | |
e5adfffc | 2591 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
2592 | unsigned int nr, *counters = m->private; |
2593 | ||
2594 | if (!counters) | |
2595 | return; | |
2596 | ||
d4033afd JK |
2597 | /* Pair with smp_wmb() in insert_vmalloc_vmlist() */ |
2598 | smp_rmb(); | |
2599 | if (v->flags & VM_UNLIST) | |
2600 | return; | |
2601 | ||
a47a126a ED |
2602 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
2603 | ||
2604 | for (nr = 0; nr < v->nr_pages; nr++) | |
2605 | counters[page_to_nid(v->pages[nr])]++; | |
2606 | ||
2607 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2608 | if (counters[nr]) | |
2609 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2610 | } | |
2611 | } | |
2612 | ||
a10aa579 CL |
2613 | static int s_show(struct seq_file *m, void *p) |
2614 | { | |
d4033afd JK |
2615 | struct vmap_area *va = p; |
2616 | struct vm_struct *v; | |
2617 | ||
2618 | if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING)) | |
2619 | return 0; | |
2620 | ||
2621 | if (!(va->flags & VM_VM_AREA)) { | |
2622 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", | |
2623 | (void *)va->va_start, (void *)va->va_end, | |
2624 | va->va_end - va->va_start); | |
2625 | return 0; | |
2626 | } | |
2627 | ||
2628 | v = va->vm; | |
a10aa579 | 2629 | |
45ec1690 | 2630 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
2631 | v->addr, v->addr + v->size, v->size); |
2632 | ||
62c70bce JP |
2633 | if (v->caller) |
2634 | seq_printf(m, " %pS", v->caller); | |
23016969 | 2635 | |
a10aa579 CL |
2636 | if (v->nr_pages) |
2637 | seq_printf(m, " pages=%d", v->nr_pages); | |
2638 | ||
2639 | if (v->phys_addr) | |
ffa71f33 | 2640 | seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr); |
a10aa579 CL |
2641 | |
2642 | if (v->flags & VM_IOREMAP) | |
2643 | seq_printf(m, " ioremap"); | |
2644 | ||
2645 | if (v->flags & VM_ALLOC) | |
2646 | seq_printf(m, " vmalloc"); | |
2647 | ||
2648 | if (v->flags & VM_MAP) | |
2649 | seq_printf(m, " vmap"); | |
2650 | ||
2651 | if (v->flags & VM_USERMAP) | |
2652 | seq_printf(m, " user"); | |
2653 | ||
2654 | if (v->flags & VM_VPAGES) | |
2655 | seq_printf(m, " vpages"); | |
2656 | ||
a47a126a | 2657 | show_numa_info(m, v); |
a10aa579 CL |
2658 | seq_putc(m, '\n'); |
2659 | return 0; | |
2660 | } | |
2661 | ||
5f6a6a9c | 2662 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2663 | .start = s_start, |
2664 | .next = s_next, | |
2665 | .stop = s_stop, | |
2666 | .show = s_show, | |
2667 | }; | |
5f6a6a9c AD |
2668 | |
2669 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2670 | { | |
2671 | unsigned int *ptr = NULL; | |
2672 | int ret; | |
2673 | ||
e5adfffc | 2674 | if (IS_ENABLED(CONFIG_NUMA)) { |
5f6a6a9c | 2675 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); |
51980ac9 KV |
2676 | if (ptr == NULL) |
2677 | return -ENOMEM; | |
2678 | } | |
5f6a6a9c AD |
2679 | ret = seq_open(file, &vmalloc_op); |
2680 | if (!ret) { | |
2681 | struct seq_file *m = file->private_data; | |
2682 | m->private = ptr; | |
2683 | } else | |
2684 | kfree(ptr); | |
2685 | return ret; | |
2686 | } | |
2687 | ||
2688 | static const struct file_operations proc_vmalloc_operations = { | |
2689 | .open = vmalloc_open, | |
2690 | .read = seq_read, | |
2691 | .llseek = seq_lseek, | |
2692 | .release = seq_release_private, | |
2693 | }; | |
2694 | ||
2695 | static int __init proc_vmalloc_init(void) | |
2696 | { | |
2697 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2698 | return 0; | |
2699 | } | |
2700 | module_init(proc_vmalloc_init); | |
db3808c1 JK |
2701 | |
2702 | void get_vmalloc_info(struct vmalloc_info *vmi) | |
2703 | { | |
f98782dd | 2704 | struct vmap_area *va; |
db3808c1 JK |
2705 | unsigned long free_area_size; |
2706 | unsigned long prev_end; | |
2707 | ||
2708 | vmi->used = 0; | |
f98782dd | 2709 | vmi->largest_chunk = 0; |
db3808c1 | 2710 | |
f98782dd | 2711 | prev_end = VMALLOC_START; |
db3808c1 | 2712 | |
f98782dd | 2713 | spin_lock(&vmap_area_lock); |
db3808c1 | 2714 | |
f98782dd JK |
2715 | if (list_empty(&vmap_area_list)) { |
2716 | vmi->largest_chunk = VMALLOC_TOTAL; | |
2717 | goto out; | |
2718 | } | |
db3808c1 | 2719 | |
f98782dd JK |
2720 | list_for_each_entry(va, &vmap_area_list, list) { |
2721 | unsigned long addr = va->va_start; | |
db3808c1 | 2722 | |
f98782dd JK |
2723 | /* |
2724 | * Some archs keep another range for modules in vmalloc space | |
2725 | */ | |
2726 | if (addr < VMALLOC_START) | |
2727 | continue; | |
2728 | if (addr >= VMALLOC_END) | |
2729 | break; | |
db3808c1 | 2730 | |
f98782dd JK |
2731 | if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING)) |
2732 | continue; | |
db3808c1 | 2733 | |
f98782dd | 2734 | vmi->used += (va->va_end - va->va_start); |
db3808c1 | 2735 | |
f98782dd JK |
2736 | free_area_size = addr - prev_end; |
2737 | if (vmi->largest_chunk < free_area_size) | |
2738 | vmi->largest_chunk = free_area_size; | |
db3808c1 | 2739 | |
f98782dd | 2740 | prev_end = va->va_end; |
db3808c1 | 2741 | } |
f98782dd JK |
2742 | |
2743 | if (VMALLOC_END - prev_end > vmi->largest_chunk) | |
2744 | vmi->largest_chunk = VMALLOC_END - prev_end; | |
2745 | ||
2746 | out: | |
2747 | spin_unlock(&vmap_area_lock); | |
db3808c1 | 2748 | } |
a10aa579 CL |
2749 | #endif |
2750 |