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 | ||
db64fe02 | 252 | static DEFINE_SPINLOCK(vmap_area_lock); |
f1c4069e JK |
253 | /* Export for kexec only */ |
254 | LIST_HEAD(vmap_area_list); | |
89699605 NP |
255 | static struct rb_root vmap_area_root = RB_ROOT; |
256 | ||
257 | /* The vmap cache globals are protected by vmap_area_lock */ | |
258 | static struct rb_node *free_vmap_cache; | |
259 | static unsigned long cached_hole_size; | |
260 | static unsigned long cached_vstart; | |
261 | static unsigned long cached_align; | |
262 | ||
ca23e405 | 263 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
264 | |
265 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 266 | { |
db64fe02 NP |
267 | struct rb_node *n = vmap_area_root.rb_node; |
268 | ||
269 | while (n) { | |
270 | struct vmap_area *va; | |
271 | ||
272 | va = rb_entry(n, struct vmap_area, rb_node); | |
273 | if (addr < va->va_start) | |
274 | n = n->rb_left; | |
275 | else if (addr > va->va_start) | |
276 | n = n->rb_right; | |
277 | else | |
278 | return va; | |
279 | } | |
280 | ||
281 | return NULL; | |
282 | } | |
283 | ||
284 | static void __insert_vmap_area(struct vmap_area *va) | |
285 | { | |
286 | struct rb_node **p = &vmap_area_root.rb_node; | |
287 | struct rb_node *parent = NULL; | |
288 | struct rb_node *tmp; | |
289 | ||
290 | while (*p) { | |
170168d0 | 291 | struct vmap_area *tmp_va; |
db64fe02 NP |
292 | |
293 | parent = *p; | |
170168d0 NK |
294 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
295 | if (va->va_start < tmp_va->va_end) | |
db64fe02 | 296 | p = &(*p)->rb_left; |
170168d0 | 297 | else if (va->va_end > tmp_va->va_start) |
db64fe02 NP |
298 | p = &(*p)->rb_right; |
299 | else | |
300 | BUG(); | |
301 | } | |
302 | ||
303 | rb_link_node(&va->rb_node, parent, p); | |
304 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
305 | ||
4341fa45 | 306 | /* address-sort this list */ |
db64fe02 NP |
307 | tmp = rb_prev(&va->rb_node); |
308 | if (tmp) { | |
309 | struct vmap_area *prev; | |
310 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
311 | list_add_rcu(&va->list, &prev->list); | |
312 | } else | |
313 | list_add_rcu(&va->list, &vmap_area_list); | |
314 | } | |
315 | ||
316 | static void purge_vmap_area_lazy(void); | |
317 | ||
318 | /* | |
319 | * Allocate a region of KVA of the specified size and alignment, within the | |
320 | * vstart and vend. | |
321 | */ | |
322 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
323 | unsigned long align, | |
324 | unsigned long vstart, unsigned long vend, | |
325 | int node, gfp_t gfp_mask) | |
326 | { | |
327 | struct vmap_area *va; | |
328 | struct rb_node *n; | |
1da177e4 | 329 | unsigned long addr; |
db64fe02 | 330 | int purged = 0; |
89699605 | 331 | struct vmap_area *first; |
db64fe02 | 332 | |
7766970c | 333 | BUG_ON(!size); |
db64fe02 | 334 | BUG_ON(size & ~PAGE_MASK); |
89699605 | 335 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 336 | |
db64fe02 NP |
337 | va = kmalloc_node(sizeof(struct vmap_area), |
338 | gfp_mask & GFP_RECLAIM_MASK, node); | |
339 | if (unlikely(!va)) | |
340 | return ERR_PTR(-ENOMEM); | |
341 | ||
342 | retry: | |
343 | spin_lock(&vmap_area_lock); | |
89699605 NP |
344 | /* |
345 | * Invalidate cache if we have more permissive parameters. | |
346 | * cached_hole_size notes the largest hole noticed _below_ | |
347 | * the vmap_area cached in free_vmap_cache: if size fits | |
348 | * into that hole, we want to scan from vstart to reuse | |
349 | * the hole instead of allocating above free_vmap_cache. | |
350 | * Note that __free_vmap_area may update free_vmap_cache | |
351 | * without updating cached_hole_size or cached_align. | |
352 | */ | |
353 | if (!free_vmap_cache || | |
354 | size < cached_hole_size || | |
355 | vstart < cached_vstart || | |
356 | align < cached_align) { | |
357 | nocache: | |
358 | cached_hole_size = 0; | |
359 | free_vmap_cache = NULL; | |
360 | } | |
361 | /* record if we encounter less permissive parameters */ | |
362 | cached_vstart = vstart; | |
363 | cached_align = align; | |
364 | ||
365 | /* find starting point for our search */ | |
366 | if (free_vmap_cache) { | |
367 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
248ac0e1 | 368 | addr = ALIGN(first->va_end, align); |
89699605 NP |
369 | if (addr < vstart) |
370 | goto nocache; | |
371 | if (addr + size - 1 < addr) | |
372 | goto overflow; | |
373 | ||
374 | } else { | |
375 | addr = ALIGN(vstart, align); | |
376 | if (addr + size - 1 < addr) | |
377 | goto overflow; | |
378 | ||
379 | n = vmap_area_root.rb_node; | |
380 | first = NULL; | |
381 | ||
382 | while (n) { | |
db64fe02 NP |
383 | struct vmap_area *tmp; |
384 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
385 | if (tmp->va_end >= addr) { | |
db64fe02 | 386 | first = tmp; |
89699605 NP |
387 | if (tmp->va_start <= addr) |
388 | break; | |
389 | n = n->rb_left; | |
390 | } else | |
db64fe02 | 391 | n = n->rb_right; |
89699605 | 392 | } |
db64fe02 NP |
393 | |
394 | if (!first) | |
395 | goto found; | |
db64fe02 | 396 | } |
89699605 NP |
397 | |
398 | /* from the starting point, walk areas until a suitable hole is found */ | |
248ac0e1 | 399 | while (addr + size > first->va_start && addr + size <= vend) { |
89699605 NP |
400 | if (addr + cached_hole_size < first->va_start) |
401 | cached_hole_size = first->va_start - addr; | |
248ac0e1 | 402 | addr = ALIGN(first->va_end, align); |
89699605 NP |
403 | if (addr + size - 1 < addr) |
404 | goto overflow; | |
405 | ||
92ca922f | 406 | if (list_is_last(&first->list, &vmap_area_list)) |
89699605 | 407 | goto found; |
92ca922f H |
408 | |
409 | first = list_entry(first->list.next, | |
410 | struct vmap_area, list); | |
db64fe02 NP |
411 | } |
412 | ||
89699605 NP |
413 | found: |
414 | if (addr + size > vend) | |
415 | goto overflow; | |
db64fe02 NP |
416 | |
417 | va->va_start = addr; | |
418 | va->va_end = addr + size; | |
419 | va->flags = 0; | |
420 | __insert_vmap_area(va); | |
89699605 | 421 | free_vmap_cache = &va->rb_node; |
db64fe02 NP |
422 | spin_unlock(&vmap_area_lock); |
423 | ||
89699605 NP |
424 | BUG_ON(va->va_start & (align-1)); |
425 | BUG_ON(va->va_start < vstart); | |
426 | BUG_ON(va->va_end > vend); | |
427 | ||
db64fe02 | 428 | return va; |
89699605 NP |
429 | |
430 | overflow: | |
431 | spin_unlock(&vmap_area_lock); | |
432 | if (!purged) { | |
433 | purge_vmap_area_lazy(); | |
434 | purged = 1; | |
435 | goto retry; | |
436 | } | |
437 | if (printk_ratelimit()) | |
438 | printk(KERN_WARNING | |
439 | "vmap allocation for size %lu failed: " | |
440 | "use vmalloc=<size> to increase size.\n", size); | |
441 | kfree(va); | |
442 | return ERR_PTR(-EBUSY); | |
db64fe02 NP |
443 | } |
444 | ||
db64fe02 NP |
445 | static void __free_vmap_area(struct vmap_area *va) |
446 | { | |
447 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
89699605 NP |
448 | |
449 | if (free_vmap_cache) { | |
450 | if (va->va_end < cached_vstart) { | |
451 | free_vmap_cache = NULL; | |
452 | } else { | |
453 | struct vmap_area *cache; | |
454 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); | |
455 | if (va->va_start <= cache->va_start) { | |
456 | free_vmap_cache = rb_prev(&va->rb_node); | |
457 | /* | |
458 | * We don't try to update cached_hole_size or | |
459 | * cached_align, but it won't go very wrong. | |
460 | */ | |
461 | } | |
462 | } | |
463 | } | |
db64fe02 NP |
464 | rb_erase(&va->rb_node, &vmap_area_root); |
465 | RB_CLEAR_NODE(&va->rb_node); | |
466 | list_del_rcu(&va->list); | |
467 | ||
ca23e405 TH |
468 | /* |
469 | * Track the highest possible candidate for pcpu area | |
470 | * allocation. Areas outside of vmalloc area can be returned | |
471 | * here too, consider only end addresses which fall inside | |
472 | * vmalloc area proper. | |
473 | */ | |
474 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
475 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
476 | ||
14769de9 | 477 | kfree_rcu(va, rcu_head); |
db64fe02 NP |
478 | } |
479 | ||
480 | /* | |
481 | * Free a region of KVA allocated by alloc_vmap_area | |
482 | */ | |
483 | static void free_vmap_area(struct vmap_area *va) | |
484 | { | |
485 | spin_lock(&vmap_area_lock); | |
486 | __free_vmap_area(va); | |
487 | spin_unlock(&vmap_area_lock); | |
488 | } | |
489 | ||
490 | /* | |
491 | * Clear the pagetable entries of a given vmap_area | |
492 | */ | |
493 | static void unmap_vmap_area(struct vmap_area *va) | |
494 | { | |
495 | vunmap_page_range(va->va_start, va->va_end); | |
496 | } | |
497 | ||
cd52858c NP |
498 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
499 | { | |
500 | /* | |
501 | * Unmap page tables and force a TLB flush immediately if | |
502 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free | |
503 | * bugs similarly to those in linear kernel virtual address | |
504 | * space after a page has been freed. | |
505 | * | |
506 | * All the lazy freeing logic is still retained, in order to | |
507 | * minimise intrusiveness of this debugging feature. | |
508 | * | |
509 | * This is going to be *slow* (linear kernel virtual address | |
510 | * debugging doesn't do a broadcast TLB flush so it is a lot | |
511 | * faster). | |
512 | */ | |
513 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
514 | vunmap_page_range(start, end); | |
515 | flush_tlb_kernel_range(start, end); | |
516 | #endif | |
517 | } | |
518 | ||
db64fe02 NP |
519 | /* |
520 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
521 | * before attempting to purge with a TLB flush. | |
522 | * | |
523 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
524 | * and take slightly longer to purge, but it will linearly reduce the number of | |
525 | * global TLB flushes that must be performed. It would seem natural to scale | |
526 | * this number up linearly with the number of CPUs (because vmapping activity | |
527 | * could also scale linearly with the number of CPUs), however it is likely | |
528 | * that in practice, workloads might be constrained in other ways that mean | |
529 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
530 | * conservative and not introduce a big latency on huge systems, so go with | |
531 | * a less aggressive log scale. It will still be an improvement over the old | |
532 | * code, and it will be simple to change the scale factor if we find that it | |
533 | * becomes a problem on bigger systems. | |
534 | */ | |
535 | static unsigned long lazy_max_pages(void) | |
536 | { | |
537 | unsigned int log; | |
538 | ||
539 | log = fls(num_online_cpus()); | |
540 | ||
541 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
542 | } | |
543 | ||
544 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
545 | ||
02b709df NP |
546 | /* for per-CPU blocks */ |
547 | static void purge_fragmented_blocks_allcpus(void); | |
548 | ||
3ee48b6a CW |
549 | /* |
550 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
551 | * immediately freed. | |
552 | */ | |
553 | void set_iounmap_nonlazy(void) | |
554 | { | |
555 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); | |
556 | } | |
557 | ||
db64fe02 NP |
558 | /* |
559 | * Purges all lazily-freed vmap areas. | |
560 | * | |
561 | * If sync is 0 then don't purge if there is already a purge in progress. | |
562 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
563 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
564 | * their own TLB flushing). | |
565 | * Returns with *start = min(*start, lowest purged address) | |
566 | * *end = max(*end, highest purged address) | |
567 | */ | |
568 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
569 | int sync, int force_flush) | |
570 | { | |
46666d8a | 571 | static DEFINE_SPINLOCK(purge_lock); |
db64fe02 NP |
572 | LIST_HEAD(valist); |
573 | struct vmap_area *va; | |
cbb76676 | 574 | struct vmap_area *n_va; |
db64fe02 NP |
575 | int nr = 0; |
576 | ||
577 | /* | |
578 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
579 | * should not expect such behaviour. This just simplifies locking for | |
580 | * the case that isn't actually used at the moment anyway. | |
581 | */ | |
582 | if (!sync && !force_flush) { | |
46666d8a | 583 | if (!spin_trylock(&purge_lock)) |
db64fe02 NP |
584 | return; |
585 | } else | |
46666d8a | 586 | spin_lock(&purge_lock); |
db64fe02 | 587 | |
02b709df NP |
588 | if (sync) |
589 | purge_fragmented_blocks_allcpus(); | |
590 | ||
db64fe02 NP |
591 | rcu_read_lock(); |
592 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
593 | if (va->flags & VM_LAZY_FREE) { | |
594 | if (va->va_start < *start) | |
595 | *start = va->va_start; | |
596 | if (va->va_end > *end) | |
597 | *end = va->va_end; | |
598 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
db64fe02 NP |
599 | list_add_tail(&va->purge_list, &valist); |
600 | va->flags |= VM_LAZY_FREEING; | |
601 | va->flags &= ~VM_LAZY_FREE; | |
602 | } | |
603 | } | |
604 | rcu_read_unlock(); | |
605 | ||
88f50044 | 606 | if (nr) |
db64fe02 | 607 | atomic_sub(nr, &vmap_lazy_nr); |
db64fe02 NP |
608 | |
609 | if (nr || force_flush) | |
610 | flush_tlb_kernel_range(*start, *end); | |
611 | ||
612 | if (nr) { | |
613 | spin_lock(&vmap_area_lock); | |
cbb76676 | 614 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
db64fe02 NP |
615 | __free_vmap_area(va); |
616 | spin_unlock(&vmap_area_lock); | |
617 | } | |
46666d8a | 618 | spin_unlock(&purge_lock); |
db64fe02 NP |
619 | } |
620 | ||
496850e5 NP |
621 | /* |
622 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
623 | * is already purging. | |
624 | */ | |
625 | static void try_purge_vmap_area_lazy(void) | |
626 | { | |
627 | unsigned long start = ULONG_MAX, end = 0; | |
628 | ||
629 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
630 | } | |
631 | ||
db64fe02 NP |
632 | /* |
633 | * Kick off a purge of the outstanding lazy areas. | |
634 | */ | |
635 | static void purge_vmap_area_lazy(void) | |
636 | { | |
637 | unsigned long start = ULONG_MAX, end = 0; | |
638 | ||
496850e5 | 639 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
db64fe02 NP |
640 | } |
641 | ||
642 | /* | |
64141da5 JF |
643 | * Free a vmap area, caller ensuring that the area has been unmapped |
644 | * and flush_cache_vunmap had been called for the correct range | |
645 | * previously. | |
db64fe02 | 646 | */ |
64141da5 | 647 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 NP |
648 | { |
649 | va->flags |= VM_LAZY_FREE; | |
650 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
651 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
496850e5 | 652 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
653 | } |
654 | ||
64141da5 JF |
655 | /* |
656 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been | |
657 | * called for the correct range previously. | |
658 | */ | |
659 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) | |
660 | { | |
661 | unmap_vmap_area(va); | |
662 | free_vmap_area_noflush(va); | |
663 | } | |
664 | ||
b29acbdc NP |
665 | /* |
666 | * Free and unmap a vmap area | |
667 | */ | |
668 | static void free_unmap_vmap_area(struct vmap_area *va) | |
669 | { | |
670 | flush_cache_vunmap(va->va_start, va->va_end); | |
671 | free_unmap_vmap_area_noflush(va); | |
672 | } | |
673 | ||
db64fe02 NP |
674 | static struct vmap_area *find_vmap_area(unsigned long addr) |
675 | { | |
676 | struct vmap_area *va; | |
677 | ||
678 | spin_lock(&vmap_area_lock); | |
679 | va = __find_vmap_area(addr); | |
680 | spin_unlock(&vmap_area_lock); | |
681 | ||
682 | return va; | |
683 | } | |
684 | ||
685 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
686 | { | |
687 | struct vmap_area *va; | |
688 | ||
689 | va = find_vmap_area(addr); | |
690 | BUG_ON(!va); | |
691 | free_unmap_vmap_area(va); | |
692 | } | |
693 | ||
694 | ||
695 | /*** Per cpu kva allocator ***/ | |
696 | ||
697 | /* | |
698 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
699 | * room for at least 16 percpu vmap blocks per CPU. | |
700 | */ | |
701 | /* | |
702 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
703 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
704 | * instead (we just need a rough idea) | |
705 | */ | |
706 | #if BITS_PER_LONG == 32 | |
707 | #define VMALLOC_SPACE (128UL*1024*1024) | |
708 | #else | |
709 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
710 | #endif | |
711 | ||
712 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
713 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
714 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
715 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
716 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
717 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
718 | #define VMAP_BBMAP_BITS \ |
719 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
720 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
721 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
722 | |
723 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
724 | ||
9b463334 JF |
725 | static bool vmap_initialized __read_mostly = false; |
726 | ||
db64fe02 NP |
727 | struct vmap_block_queue { |
728 | spinlock_t lock; | |
729 | struct list_head free; | |
db64fe02 NP |
730 | }; |
731 | ||
732 | struct vmap_block { | |
733 | spinlock_t lock; | |
734 | struct vmap_area *va; | |
735 | struct vmap_block_queue *vbq; | |
736 | unsigned long free, dirty; | |
737 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
738 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
de560423 NP |
739 | struct list_head free_list; |
740 | struct rcu_head rcu_head; | |
02b709df | 741 | struct list_head purge; |
db64fe02 NP |
742 | }; |
743 | ||
744 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
745 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
746 | ||
747 | /* | |
748 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
749 | * in the free path. Could get rid of this if we change the API to return a | |
750 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
751 | */ | |
752 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
753 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
754 | ||
755 | /* | |
756 | * We should probably have a fallback mechanism to allocate virtual memory | |
757 | * out of partially filled vmap blocks. However vmap block sizing should be | |
758 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
759 | * big problem. | |
760 | */ | |
761 | ||
762 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
763 | { | |
764 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
765 | addr /= VMAP_BLOCK_SIZE; | |
766 | return addr; | |
767 | } | |
768 | ||
769 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
770 | { | |
771 | struct vmap_block_queue *vbq; | |
772 | struct vmap_block *vb; | |
773 | struct vmap_area *va; | |
774 | unsigned long vb_idx; | |
775 | int node, err; | |
776 | ||
777 | node = numa_node_id(); | |
778 | ||
779 | vb = kmalloc_node(sizeof(struct vmap_block), | |
780 | gfp_mask & GFP_RECLAIM_MASK, node); | |
781 | if (unlikely(!vb)) | |
782 | return ERR_PTR(-ENOMEM); | |
783 | ||
784 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
785 | VMALLOC_START, VMALLOC_END, | |
786 | node, gfp_mask); | |
ddf9c6d4 | 787 | if (IS_ERR(va)) { |
db64fe02 | 788 | kfree(vb); |
e7d86340 | 789 | return ERR_CAST(va); |
db64fe02 NP |
790 | } |
791 | ||
792 | err = radix_tree_preload(gfp_mask); | |
793 | if (unlikely(err)) { | |
794 | kfree(vb); | |
795 | free_vmap_area(va); | |
796 | return ERR_PTR(err); | |
797 | } | |
798 | ||
799 | spin_lock_init(&vb->lock); | |
800 | vb->va = va; | |
801 | vb->free = VMAP_BBMAP_BITS; | |
802 | vb->dirty = 0; | |
803 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
804 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
805 | INIT_LIST_HEAD(&vb->free_list); | |
db64fe02 NP |
806 | |
807 | vb_idx = addr_to_vb_idx(va->va_start); | |
808 | spin_lock(&vmap_block_tree_lock); | |
809 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
810 | spin_unlock(&vmap_block_tree_lock); | |
811 | BUG_ON(err); | |
812 | radix_tree_preload_end(); | |
813 | ||
814 | vbq = &get_cpu_var(vmap_block_queue); | |
815 | vb->vbq = vbq; | |
816 | spin_lock(&vbq->lock); | |
de560423 | 817 | list_add_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 818 | spin_unlock(&vbq->lock); |
3f04ba85 | 819 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
820 | |
821 | return vb; | |
822 | } | |
823 | ||
db64fe02 NP |
824 | static void free_vmap_block(struct vmap_block *vb) |
825 | { | |
826 | struct vmap_block *tmp; | |
827 | unsigned long vb_idx; | |
828 | ||
db64fe02 NP |
829 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
830 | spin_lock(&vmap_block_tree_lock); | |
831 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
832 | spin_unlock(&vmap_block_tree_lock); | |
833 | BUG_ON(tmp != vb); | |
834 | ||
64141da5 | 835 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 836 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
837 | } |
838 | ||
02b709df NP |
839 | static void purge_fragmented_blocks(int cpu) |
840 | { | |
841 | LIST_HEAD(purge); | |
842 | struct vmap_block *vb; | |
843 | struct vmap_block *n_vb; | |
844 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
845 | ||
846 | rcu_read_lock(); | |
847 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
848 | ||
849 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
850 | continue; | |
851 | ||
852 | spin_lock(&vb->lock); | |
853 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
854 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
855 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
856 | bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS); | |
857 | bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS); | |
858 | spin_lock(&vbq->lock); | |
859 | list_del_rcu(&vb->free_list); | |
860 | spin_unlock(&vbq->lock); | |
861 | spin_unlock(&vb->lock); | |
862 | list_add_tail(&vb->purge, &purge); | |
863 | } else | |
864 | spin_unlock(&vb->lock); | |
865 | } | |
866 | rcu_read_unlock(); | |
867 | ||
868 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
869 | list_del(&vb->purge); | |
870 | free_vmap_block(vb); | |
871 | } | |
872 | } | |
873 | ||
874 | static void purge_fragmented_blocks_thiscpu(void) | |
875 | { | |
876 | purge_fragmented_blocks(smp_processor_id()); | |
877 | } | |
878 | ||
879 | static void purge_fragmented_blocks_allcpus(void) | |
880 | { | |
881 | int cpu; | |
882 | ||
883 | for_each_possible_cpu(cpu) | |
884 | purge_fragmented_blocks(cpu); | |
885 | } | |
886 | ||
db64fe02 NP |
887 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
888 | { | |
889 | struct vmap_block_queue *vbq; | |
890 | struct vmap_block *vb; | |
891 | unsigned long addr = 0; | |
892 | unsigned int order; | |
02b709df | 893 | int purge = 0; |
db64fe02 NP |
894 | |
895 | BUG_ON(size & ~PAGE_MASK); | |
896 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
aa91c4d8 JK |
897 | if (WARN_ON(size == 0)) { |
898 | /* | |
899 | * Allocating 0 bytes isn't what caller wants since | |
900 | * get_order(0) returns funny result. Just warn and terminate | |
901 | * early. | |
902 | */ | |
903 | return NULL; | |
904 | } | |
db64fe02 NP |
905 | order = get_order(size); |
906 | ||
907 | again: | |
908 | rcu_read_lock(); | |
909 | vbq = &get_cpu_var(vmap_block_queue); | |
910 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
911 | int i; | |
912 | ||
913 | spin_lock(&vb->lock); | |
02b709df NP |
914 | if (vb->free < 1UL << order) |
915 | goto next; | |
916 | ||
db64fe02 NP |
917 | i = bitmap_find_free_region(vb->alloc_map, |
918 | VMAP_BBMAP_BITS, order); | |
919 | ||
02b709df NP |
920 | if (i < 0) { |
921 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS) { | |
922 | /* fragmented and no outstanding allocations */ | |
923 | BUG_ON(vb->dirty != VMAP_BBMAP_BITS); | |
924 | purge = 1; | |
db64fe02 | 925 | } |
02b709df | 926 | goto next; |
db64fe02 | 927 | } |
02b709df NP |
928 | addr = vb->va->va_start + (i << PAGE_SHIFT); |
929 | BUG_ON(addr_to_vb_idx(addr) != | |
930 | addr_to_vb_idx(vb->va->va_start)); | |
931 | vb->free -= 1UL << order; | |
932 | if (vb->free == 0) { | |
933 | spin_lock(&vbq->lock); | |
934 | list_del_rcu(&vb->free_list); | |
935 | spin_unlock(&vbq->lock); | |
936 | } | |
937 | spin_unlock(&vb->lock); | |
938 | break; | |
939 | next: | |
db64fe02 NP |
940 | spin_unlock(&vb->lock); |
941 | } | |
02b709df NP |
942 | |
943 | if (purge) | |
944 | purge_fragmented_blocks_thiscpu(); | |
945 | ||
3f04ba85 | 946 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
947 | rcu_read_unlock(); |
948 | ||
949 | if (!addr) { | |
950 | vb = new_vmap_block(gfp_mask); | |
951 | if (IS_ERR(vb)) | |
952 | return vb; | |
953 | goto again; | |
954 | } | |
955 | ||
956 | return (void *)addr; | |
957 | } | |
958 | ||
959 | static void vb_free(const void *addr, unsigned long size) | |
960 | { | |
961 | unsigned long offset; | |
962 | unsigned long vb_idx; | |
963 | unsigned int order; | |
964 | struct vmap_block *vb; | |
965 | ||
966 | BUG_ON(size & ~PAGE_MASK); | |
967 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
b29acbdc NP |
968 | |
969 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
970 | ||
db64fe02 NP |
971 | order = get_order(size); |
972 | ||
973 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
974 | ||
975 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
976 | rcu_read_lock(); | |
977 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
978 | rcu_read_unlock(); | |
979 | BUG_ON(!vb); | |
980 | ||
64141da5 JF |
981 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
982 | ||
db64fe02 | 983 | spin_lock(&vb->lock); |
de560423 | 984 | BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order)); |
d086817d | 985 | |
db64fe02 NP |
986 | vb->dirty += 1UL << order; |
987 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 988 | BUG_ON(vb->free); |
db64fe02 NP |
989 | spin_unlock(&vb->lock); |
990 | free_vmap_block(vb); | |
991 | } else | |
992 | spin_unlock(&vb->lock); | |
993 | } | |
994 | ||
995 | /** | |
996 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
997 | * | |
998 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
999 | * to amortize TLB flushing overheads. What this means is that any page you | |
1000 | * have now, may, in a former life, have been mapped into kernel virtual | |
1001 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1002 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1003 | * | |
1004 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1005 | * be sure that none of the pages we have control over will have any aliases | |
1006 | * from the vmap layer. | |
1007 | */ | |
1008 | void vm_unmap_aliases(void) | |
1009 | { | |
1010 | unsigned long start = ULONG_MAX, end = 0; | |
1011 | int cpu; | |
1012 | int flush = 0; | |
1013 | ||
9b463334 JF |
1014 | if (unlikely(!vmap_initialized)) |
1015 | return; | |
1016 | ||
db64fe02 NP |
1017 | for_each_possible_cpu(cpu) { |
1018 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1019 | struct vmap_block *vb; | |
1020 | ||
1021 | rcu_read_lock(); | |
1022 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1023 | int i; | |
1024 | ||
1025 | spin_lock(&vb->lock); | |
1026 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
1027 | while (i < VMAP_BBMAP_BITS) { | |
1028 | unsigned long s, e; | |
1029 | int j; | |
1030 | j = find_next_zero_bit(vb->dirty_map, | |
1031 | VMAP_BBMAP_BITS, i); | |
1032 | ||
1033 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
1034 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
db64fe02 NP |
1035 | flush = 1; |
1036 | ||
1037 | if (s < start) | |
1038 | start = s; | |
1039 | if (e > end) | |
1040 | end = e; | |
1041 | ||
1042 | i = j; | |
1043 | i = find_next_bit(vb->dirty_map, | |
1044 | VMAP_BBMAP_BITS, i); | |
1045 | } | |
1046 | spin_unlock(&vb->lock); | |
1047 | } | |
1048 | rcu_read_unlock(); | |
1049 | } | |
1050 | ||
1051 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
1052 | } | |
1053 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
1054 | ||
1055 | /** | |
1056 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1057 | * @mem: the pointer returned by vm_map_ram | |
1058 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1059 | */ | |
1060 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1061 | { | |
1062 | unsigned long size = count << PAGE_SHIFT; | |
1063 | unsigned long addr = (unsigned long)mem; | |
1064 | ||
1065 | BUG_ON(!addr); | |
1066 | BUG_ON(addr < VMALLOC_START); | |
1067 | BUG_ON(addr > VMALLOC_END); | |
1068 | BUG_ON(addr & (PAGE_SIZE-1)); | |
1069 | ||
1070 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 1071 | vmap_debug_free_range(addr, addr+size); |
db64fe02 NP |
1072 | |
1073 | if (likely(count <= VMAP_MAX_ALLOC)) | |
1074 | vb_free(mem, size); | |
1075 | else | |
1076 | free_unmap_vmap_area_addr(addr); | |
1077 | } | |
1078 | EXPORT_SYMBOL(vm_unmap_ram); | |
1079 | ||
1080 | /** | |
1081 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1082 | * @pages: an array of pointers to the pages to be mapped | |
1083 | * @count: number of pages | |
1084 | * @node: prefer to allocate data structures on this node | |
1085 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad RD |
1086 | * |
1087 | * Returns: a pointer to the address that has been mapped, or %NULL on failure | |
db64fe02 NP |
1088 | */ |
1089 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1090 | { | |
1091 | unsigned long size = count << PAGE_SHIFT; | |
1092 | unsigned long addr; | |
1093 | void *mem; | |
1094 | ||
1095 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1096 | mem = vb_alloc(size, GFP_KERNEL); | |
1097 | if (IS_ERR(mem)) | |
1098 | return NULL; | |
1099 | addr = (unsigned long)mem; | |
1100 | } else { | |
1101 | struct vmap_area *va; | |
1102 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1103 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1104 | if (IS_ERR(va)) | |
1105 | return NULL; | |
1106 | ||
1107 | addr = va->va_start; | |
1108 | mem = (void *)addr; | |
1109 | } | |
1110 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1111 | vm_unmap_ram(mem, count); | |
1112 | return NULL; | |
1113 | } | |
1114 | return mem; | |
1115 | } | |
1116 | EXPORT_SYMBOL(vm_map_ram); | |
1117 | ||
4341fa45 | 1118 | static struct vm_struct *vmlist __initdata; |
be9b7335 NP |
1119 | /** |
1120 | * vm_area_add_early - add vmap area early during boot | |
1121 | * @vm: vm_struct to add | |
1122 | * | |
1123 | * This function is used to add fixed kernel vm area to vmlist before | |
1124 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1125 | * should contain proper values and the other fields should be zero. | |
1126 | * | |
1127 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1128 | */ | |
1129 | void __init vm_area_add_early(struct vm_struct *vm) | |
1130 | { | |
1131 | struct vm_struct *tmp, **p; | |
1132 | ||
1133 | BUG_ON(vmap_initialized); | |
1134 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1135 | if (tmp->addr >= vm->addr) { | |
1136 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1137 | break; | |
1138 | } else | |
1139 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1140 | } | |
1141 | vm->next = *p; | |
1142 | *p = vm; | |
1143 | } | |
1144 | ||
f0aa6617 TH |
1145 | /** |
1146 | * vm_area_register_early - register vmap area early during boot | |
1147 | * @vm: vm_struct to register | |
c0c0a293 | 1148 | * @align: requested alignment |
f0aa6617 TH |
1149 | * |
1150 | * This function is used to register kernel vm area before | |
1151 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1152 | * proper values on entry and other fields should be zero. On return, | |
1153 | * vm->addr contains the allocated address. | |
1154 | * | |
1155 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1156 | */ | |
c0c0a293 | 1157 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1158 | { |
1159 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1160 | unsigned long addr; |
1161 | ||
1162 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1163 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1164 | |
c0c0a293 | 1165 | vm->addr = (void *)addr; |
f0aa6617 | 1166 | |
be9b7335 | 1167 | vm_area_add_early(vm); |
f0aa6617 TH |
1168 | } |
1169 | ||
db64fe02 NP |
1170 | void __init vmalloc_init(void) |
1171 | { | |
822c18f2 IK |
1172 | struct vmap_area *va; |
1173 | struct vm_struct *tmp; | |
db64fe02 NP |
1174 | int i; |
1175 | ||
1176 | for_each_possible_cpu(i) { | |
1177 | struct vmap_block_queue *vbq; | |
1178 | ||
1179 | vbq = &per_cpu(vmap_block_queue, i); | |
1180 | spin_lock_init(&vbq->lock); | |
1181 | INIT_LIST_HEAD(&vbq->free); | |
db64fe02 | 1182 | } |
9b463334 | 1183 | |
822c18f2 IK |
1184 | /* Import existing vmlist entries. */ |
1185 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1186 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
dbda591d | 1187 | va->flags = VM_VM_AREA; |
822c18f2 IK |
1188 | va->va_start = (unsigned long)tmp->addr; |
1189 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 1190 | va->vm = tmp; |
822c18f2 IK |
1191 | __insert_vmap_area(va); |
1192 | } | |
ca23e405 TH |
1193 | |
1194 | vmap_area_pcpu_hole = VMALLOC_END; | |
1195 | ||
9b463334 | 1196 | vmap_initialized = true; |
db64fe02 NP |
1197 | } |
1198 | ||
8fc48985 TH |
1199 | /** |
1200 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1201 | * @addr: start of the VM area to map | |
1202 | * @size: size of the VM area to map | |
1203 | * @prot: page protection flags to use | |
1204 | * @pages: pages to map | |
1205 | * | |
1206 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1207 | * specify should have been allocated using get_vm_area() and its | |
1208 | * friends. | |
1209 | * | |
1210 | * NOTE: | |
1211 | * This function does NOT do any cache flushing. The caller is | |
1212 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1213 | * before calling this function. | |
1214 | * | |
1215 | * RETURNS: | |
1216 | * The number of pages mapped on success, -errno on failure. | |
1217 | */ | |
1218 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1219 | pgprot_t prot, struct page **pages) | |
1220 | { | |
1221 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1222 | } | |
1223 | ||
1224 | /** | |
1225 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1226 | * @addr: start of the VM area to unmap | |
1227 | * @size: size of the VM area to unmap | |
1228 | * | |
1229 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1230 | * specify should have been allocated using get_vm_area() and its | |
1231 | * friends. | |
1232 | * | |
1233 | * NOTE: | |
1234 | * This function does NOT do any cache flushing. The caller is | |
1235 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1236 | * before calling this function and flush_tlb_kernel_range() after. | |
1237 | */ | |
1238 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1239 | { | |
1240 | vunmap_page_range(addr, addr + size); | |
1241 | } | |
81e88fdc | 1242 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1243 | |
1244 | /** | |
1245 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1246 | * @addr: start of the VM area to unmap | |
1247 | * @size: size of the VM area to unmap | |
1248 | * | |
1249 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1250 | * the unmapping and tlb after. | |
1251 | */ | |
db64fe02 NP |
1252 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1253 | { | |
1254 | unsigned long end = addr + size; | |
f6fcba70 TH |
1255 | |
1256 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1257 | vunmap_page_range(addr, end); |
1258 | flush_tlb_kernel_range(addr, end); | |
1259 | } | |
1260 | ||
1261 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
1262 | { | |
1263 | unsigned long addr = (unsigned long)area->addr; | |
1264 | unsigned long end = addr + area->size - PAGE_SIZE; | |
1265 | int err; | |
1266 | ||
1267 | err = vmap_page_range(addr, end, prot, *pages); | |
1268 | if (err > 0) { | |
1269 | *pages += err; | |
1270 | err = 0; | |
1271 | } | |
1272 | ||
1273 | return err; | |
1274 | } | |
1275 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1276 | ||
f5252e00 | 1277 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1278 | unsigned long flags, const void *caller) |
cf88c790 | 1279 | { |
c69480ad | 1280 | spin_lock(&vmap_area_lock); |
cf88c790 TH |
1281 | vm->flags = flags; |
1282 | vm->addr = (void *)va->va_start; | |
1283 | vm->size = va->va_end - va->va_start; | |
1284 | vm->caller = caller; | |
db1aecaf | 1285 | va->vm = vm; |
cf88c790 | 1286 | va->flags |= VM_VM_AREA; |
c69480ad | 1287 | spin_unlock(&vmap_area_lock); |
f5252e00 | 1288 | } |
cf88c790 | 1289 | |
4341fa45 | 1290 | static void clear_vm_unlist(struct vm_struct *vm) |
f5252e00 | 1291 | { |
d4033afd JK |
1292 | /* |
1293 | * Before removing VM_UNLIST, | |
1294 | * we should make sure that vm has proper values. | |
1295 | * Pair with smp_rmb() in show_numa_info(). | |
1296 | */ | |
1297 | smp_wmb(); | |
f5252e00 | 1298 | vm->flags &= ~VM_UNLIST; |
cf88c790 TH |
1299 | } |
1300 | ||
f5252e00 | 1301 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
5e6cafc8 | 1302 | unsigned long flags, const void *caller) |
f5252e00 MH |
1303 | { |
1304 | setup_vmalloc_vm(vm, va, flags, caller); | |
4341fa45 | 1305 | clear_vm_unlist(vm); |
f5252e00 MH |
1306 | } |
1307 | ||
db64fe02 | 1308 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 1309 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 1310 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 1311 | { |
0006526d | 1312 | struct vmap_area *va; |
db64fe02 | 1313 | struct vm_struct *area; |
1da177e4 | 1314 | |
52fd24ca | 1315 | BUG_ON(in_interrupt()); |
1da177e4 LT |
1316 | if (flags & VM_IOREMAP) { |
1317 | int bit = fls(size); | |
1318 | ||
1319 | if (bit > IOREMAP_MAX_ORDER) | |
1320 | bit = IOREMAP_MAX_ORDER; | |
1321 | else if (bit < PAGE_SHIFT) | |
1322 | bit = PAGE_SHIFT; | |
1323 | ||
1324 | align = 1ul << bit; | |
1325 | } | |
db64fe02 | 1326 | |
1da177e4 | 1327 | size = PAGE_ALIGN(size); |
31be8309 OH |
1328 | if (unlikely(!size)) |
1329 | return NULL; | |
1da177e4 | 1330 | |
cf88c790 | 1331 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1332 | if (unlikely(!area)) |
1333 | return NULL; | |
1334 | ||
1da177e4 LT |
1335 | /* |
1336 | * We always allocate a guard page. | |
1337 | */ | |
1338 | size += PAGE_SIZE; | |
1339 | ||
db64fe02 NP |
1340 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1341 | if (IS_ERR(va)) { | |
1342 | kfree(area); | |
1343 | return NULL; | |
1da177e4 | 1344 | } |
1da177e4 | 1345 | |
f5252e00 MH |
1346 | /* |
1347 | * When this function is called from __vmalloc_node_range, | |
4341fa45 JK |
1348 | * we add VM_UNLIST flag to avoid accessing uninitialized |
1349 | * members of vm_struct such as pages and nr_pages fields. | |
1350 | * They will be set later. | |
f5252e00 MH |
1351 | */ |
1352 | if (flags & VM_UNLIST) | |
1353 | setup_vmalloc_vm(area, va, flags, caller); | |
1354 | else | |
1355 | insert_vmalloc_vm(area, va, flags, caller); | |
1356 | ||
1da177e4 | 1357 | return area; |
1da177e4 LT |
1358 | } |
1359 | ||
930fc45a CL |
1360 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1361 | unsigned long start, unsigned long end) | |
1362 | { | |
00ef2d2f DR |
1363 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1364 | GFP_KERNEL, __builtin_return_address(0)); | |
930fc45a | 1365 | } |
5992b6da | 1366 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1367 | |
c2968612 BH |
1368 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1369 | unsigned long start, unsigned long end, | |
5e6cafc8 | 1370 | const void *caller) |
c2968612 | 1371 | { |
00ef2d2f DR |
1372 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
1373 | GFP_KERNEL, caller); | |
c2968612 BH |
1374 | } |
1375 | ||
1da177e4 | 1376 | /** |
183ff22b | 1377 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1378 | * @size: size of the area |
1379 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1380 | * | |
1381 | * Search an area of @size in the kernel virtual mapping area, | |
1382 | * and reserved it for out purposes. Returns the area descriptor | |
1383 | * on success or %NULL on failure. | |
1384 | */ | |
1385 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1386 | { | |
2dca6999 | 1387 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
1388 | NUMA_NO_NODE, GFP_KERNEL, |
1389 | __builtin_return_address(0)); | |
23016969 CL |
1390 | } |
1391 | ||
1392 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 1393 | const void *caller) |
23016969 | 1394 | { |
2dca6999 | 1395 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 1396 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
1397 | } |
1398 | ||
e9da6e99 MS |
1399 | /** |
1400 | * find_vm_area - find a continuous kernel virtual area | |
1401 | * @addr: base address | |
1402 | * | |
1403 | * Search for the kernel VM area starting at @addr, and return it. | |
1404 | * It is up to the caller to do all required locking to keep the returned | |
1405 | * pointer valid. | |
1406 | */ | |
1407 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 1408 | { |
db64fe02 | 1409 | struct vmap_area *va; |
83342314 | 1410 | |
db64fe02 NP |
1411 | va = find_vmap_area((unsigned long)addr); |
1412 | if (va && va->flags & VM_VM_AREA) | |
db1aecaf | 1413 | return va->vm; |
1da177e4 | 1414 | |
1da177e4 | 1415 | return NULL; |
1da177e4 LT |
1416 | } |
1417 | ||
7856dfeb | 1418 | /** |
183ff22b | 1419 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1420 | * @addr: base address |
1421 | * | |
1422 | * Search for the kernel VM area starting at @addr, and remove it. | |
1423 | * This function returns the found VM area, but using it is NOT safe | |
1424 | * on SMP machines, except for its size or flags. | |
1425 | */ | |
b3bdda02 | 1426 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1427 | { |
db64fe02 NP |
1428 | struct vmap_area *va; |
1429 | ||
1430 | va = find_vmap_area((unsigned long)addr); | |
1431 | if (va && va->flags & VM_VM_AREA) { | |
db1aecaf | 1432 | struct vm_struct *vm = va->vm; |
f5252e00 | 1433 | |
c69480ad JK |
1434 | spin_lock(&vmap_area_lock); |
1435 | va->vm = NULL; | |
1436 | va->flags &= ~VM_VM_AREA; | |
1437 | spin_unlock(&vmap_area_lock); | |
1438 | ||
dd32c279 KH |
1439 | vmap_debug_free_range(va->va_start, va->va_end); |
1440 | free_unmap_vmap_area(va); | |
1441 | vm->size -= PAGE_SIZE; | |
1442 | ||
db64fe02 NP |
1443 | return vm; |
1444 | } | |
1445 | return NULL; | |
7856dfeb AK |
1446 | } |
1447 | ||
b3bdda02 | 1448 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1449 | { |
1450 | struct vm_struct *area; | |
1451 | ||
1452 | if (!addr) | |
1453 | return; | |
1454 | ||
1455 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1456 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1457 | return; |
1458 | } | |
1459 | ||
1460 | area = remove_vm_area(addr); | |
1461 | if (unlikely(!area)) { | |
4c8573e2 | 1462 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1463 | addr); |
1da177e4 LT |
1464 | return; |
1465 | } | |
1466 | ||
9a11b49a | 1467 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1468 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1469 | |
1da177e4 LT |
1470 | if (deallocate_pages) { |
1471 | int i; | |
1472 | ||
1473 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1474 | struct page *page = area->pages[i]; |
1475 | ||
1476 | BUG_ON(!page); | |
1477 | __free_page(page); | |
1da177e4 LT |
1478 | } |
1479 | ||
8757d5fa | 1480 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1481 | vfree(area->pages); |
1482 | else | |
1483 | kfree(area->pages); | |
1484 | } | |
1485 | ||
1486 | kfree(area); | |
1487 | return; | |
1488 | } | |
1489 | ||
1490 | /** | |
1491 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1492 | * @addr: memory base address |
1493 | * | |
183ff22b | 1494 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1495 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1496 | * NULL, no operation is performed. | |
1da177e4 | 1497 | * |
80e93eff | 1498 | * Must not be called in interrupt context. |
1da177e4 | 1499 | */ |
b3bdda02 | 1500 | void vfree(const void *addr) |
1da177e4 LT |
1501 | { |
1502 | BUG_ON(in_interrupt()); | |
89219d37 CM |
1503 | |
1504 | kmemleak_free(addr); | |
1505 | ||
1da177e4 LT |
1506 | __vunmap(addr, 1); |
1507 | } | |
1da177e4 LT |
1508 | EXPORT_SYMBOL(vfree); |
1509 | ||
1510 | /** | |
1511 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1512 | * @addr: memory base address |
1513 | * | |
1514 | * Free the virtually contiguous memory area starting at @addr, | |
1515 | * which was created from the page array passed to vmap(). | |
1516 | * | |
80e93eff | 1517 | * Must not be called in interrupt context. |
1da177e4 | 1518 | */ |
b3bdda02 | 1519 | void vunmap(const void *addr) |
1da177e4 LT |
1520 | { |
1521 | BUG_ON(in_interrupt()); | |
34754b69 | 1522 | might_sleep(); |
1da177e4 LT |
1523 | __vunmap(addr, 0); |
1524 | } | |
1da177e4 LT |
1525 | EXPORT_SYMBOL(vunmap); |
1526 | ||
1527 | /** | |
1528 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1529 | * @pages: array of page pointers |
1530 | * @count: number of pages to map | |
1531 | * @flags: vm_area->flags | |
1532 | * @prot: page protection for the mapping | |
1533 | * | |
1534 | * Maps @count pages from @pages into contiguous kernel virtual | |
1535 | * space. | |
1536 | */ | |
1537 | void *vmap(struct page **pages, unsigned int count, | |
1538 | unsigned long flags, pgprot_t prot) | |
1539 | { | |
1540 | struct vm_struct *area; | |
1541 | ||
34754b69 PZ |
1542 | might_sleep(); |
1543 | ||
4481374c | 1544 | if (count > totalram_pages) |
1da177e4 LT |
1545 | return NULL; |
1546 | ||
23016969 CL |
1547 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1548 | __builtin_return_address(0)); | |
1da177e4 LT |
1549 | if (!area) |
1550 | return NULL; | |
23016969 | 1551 | |
1da177e4 LT |
1552 | if (map_vm_area(area, prot, &pages)) { |
1553 | vunmap(area->addr); | |
1554 | return NULL; | |
1555 | } | |
1556 | ||
1557 | return area->addr; | |
1558 | } | |
1da177e4 LT |
1559 | EXPORT_SYMBOL(vmap); |
1560 | ||
2dca6999 DM |
1561 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
1562 | gfp_t gfp_mask, pgprot_t prot, | |
5e6cafc8 | 1563 | int node, const void *caller); |
e31d9eb5 | 1564 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
5e6cafc8 | 1565 | pgprot_t prot, int node, const void *caller) |
1da177e4 | 1566 | { |
22943ab1 | 1567 | const int order = 0; |
1da177e4 LT |
1568 | struct page **pages; |
1569 | unsigned int nr_pages, array_size, i; | |
976d6dfb | 1570 | gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
1da177e4 LT |
1571 | |
1572 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1573 | array_size = (nr_pages * sizeof(struct page *)); | |
1574 | ||
1575 | area->nr_pages = nr_pages; | |
1576 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1577 | if (array_size > PAGE_SIZE) { |
976d6dfb | 1578 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
23016969 | 1579 | PAGE_KERNEL, node, caller); |
8757d5fa | 1580 | area->flags |= VM_VPAGES; |
286e1ea3 | 1581 | } else { |
976d6dfb | 1582 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 1583 | } |
1da177e4 | 1584 | area->pages = pages; |
23016969 | 1585 | area->caller = caller; |
1da177e4 LT |
1586 | if (!area->pages) { |
1587 | remove_vm_area(area->addr); | |
1588 | kfree(area); | |
1589 | return NULL; | |
1590 | } | |
1da177e4 LT |
1591 | |
1592 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 | 1593 | struct page *page; |
22943ab1 | 1594 | gfp_t tmp_mask = gfp_mask | __GFP_NOWARN; |
bf53d6f8 | 1595 | |
930fc45a | 1596 | if (node < 0) |
22943ab1 | 1597 | page = alloc_page(tmp_mask); |
930fc45a | 1598 | else |
22943ab1 | 1599 | page = alloc_pages_node(node, tmp_mask, order); |
bf53d6f8 CL |
1600 | |
1601 | if (unlikely(!page)) { | |
1da177e4 LT |
1602 | /* Successfully allocated i pages, free them in __vunmap() */ |
1603 | area->nr_pages = i; | |
1604 | goto fail; | |
1605 | } | |
bf53d6f8 | 1606 | area->pages[i] = page; |
1da177e4 LT |
1607 | } |
1608 | ||
1609 | if (map_vm_area(area, prot, &pages)) | |
1610 | goto fail; | |
1611 | return area->addr; | |
1612 | ||
1613 | fail: | |
3ee9a4f0 JP |
1614 | warn_alloc_failed(gfp_mask, order, |
1615 | "vmalloc: allocation failure, allocated %ld of %ld bytes\n", | |
22943ab1 | 1616 | (area->nr_pages*PAGE_SIZE), area->size); |
1da177e4 LT |
1617 | vfree(area->addr); |
1618 | return NULL; | |
1619 | } | |
1620 | ||
1621 | /** | |
d0a21265 | 1622 | * __vmalloc_node_range - allocate virtually contiguous memory |
1da177e4 | 1623 | * @size: allocation size |
2dca6999 | 1624 | * @align: desired alignment |
d0a21265 DR |
1625 | * @start: vm area range start |
1626 | * @end: vm area range end | |
1da177e4 LT |
1627 | * @gfp_mask: flags for the page level allocator |
1628 | * @prot: protection mask for the allocated pages | |
00ef2d2f | 1629 | * @node: node to use for allocation or NUMA_NO_NODE |
c85d194b | 1630 | * @caller: caller's return address |
1da177e4 LT |
1631 | * |
1632 | * Allocate enough pages to cover @size from the page level | |
1633 | * allocator with @gfp_mask flags. Map them into contiguous | |
1634 | * kernel virtual space, using a pagetable protection of @prot. | |
1635 | */ | |
d0a21265 DR |
1636 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
1637 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
5e6cafc8 | 1638 | pgprot_t prot, int node, const void *caller) |
1da177e4 LT |
1639 | { |
1640 | struct vm_struct *area; | |
89219d37 CM |
1641 | void *addr; |
1642 | unsigned long real_size = size; | |
1da177e4 LT |
1643 | |
1644 | size = PAGE_ALIGN(size); | |
4481374c | 1645 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
de7d2b56 | 1646 | goto fail; |
1da177e4 | 1647 | |
f5252e00 MH |
1648 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST, |
1649 | start, end, node, gfp_mask, caller); | |
1da177e4 | 1650 | if (!area) |
de7d2b56 | 1651 | goto fail; |
1da177e4 | 1652 | |
89219d37 | 1653 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1368edf0 MG |
1654 | if (!addr) |
1655 | return NULL; | |
89219d37 | 1656 | |
f5252e00 | 1657 | /* |
4341fa45 JK |
1658 | * In this function, newly allocated vm_struct has VM_UNLIST flag. |
1659 | * It means that vm_struct is not fully initialized. | |
1660 | * Now, it is fully initialized, so remove this flag here. | |
f5252e00 | 1661 | */ |
4341fa45 | 1662 | clear_vm_unlist(area); |
f5252e00 | 1663 | |
89219d37 CM |
1664 | /* |
1665 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1666 | * structures allocated in the __get_vm_area_node() function contain | |
1667 | * references to the virtual address of the vmalloc'ed block. | |
1668 | */ | |
1669 | kmemleak_alloc(addr, real_size, 3, gfp_mask); | |
1670 | ||
1671 | return addr; | |
de7d2b56 JP |
1672 | |
1673 | fail: | |
1674 | warn_alloc_failed(gfp_mask, 0, | |
1675 | "vmalloc: allocation failure: %lu bytes\n", | |
1676 | real_size); | |
1677 | return NULL; | |
1da177e4 LT |
1678 | } |
1679 | ||
d0a21265 DR |
1680 | /** |
1681 | * __vmalloc_node - allocate virtually contiguous memory | |
1682 | * @size: allocation size | |
1683 | * @align: desired alignment | |
1684 | * @gfp_mask: flags for the page level allocator | |
1685 | * @prot: protection mask for the allocated pages | |
00ef2d2f | 1686 | * @node: node to use for allocation or NUMA_NO_NODE |
d0a21265 DR |
1687 | * @caller: caller's return address |
1688 | * | |
1689 | * Allocate enough pages to cover @size from the page level | |
1690 | * allocator with @gfp_mask flags. Map them into contiguous | |
1691 | * kernel virtual space, using a pagetable protection of @prot. | |
1692 | */ | |
1693 | static void *__vmalloc_node(unsigned long size, unsigned long align, | |
1694 | gfp_t gfp_mask, pgprot_t prot, | |
5e6cafc8 | 1695 | int node, const void *caller) |
d0a21265 DR |
1696 | { |
1697 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
1698 | gfp_mask, prot, node, caller); | |
1699 | } | |
1700 | ||
930fc45a CL |
1701 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1702 | { | |
00ef2d2f | 1703 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 1704 | __builtin_return_address(0)); |
930fc45a | 1705 | } |
1da177e4 LT |
1706 | EXPORT_SYMBOL(__vmalloc); |
1707 | ||
e1ca7788 DY |
1708 | static inline void *__vmalloc_node_flags(unsigned long size, |
1709 | int node, gfp_t flags) | |
1710 | { | |
1711 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
1712 | node, __builtin_return_address(0)); | |
1713 | } | |
1714 | ||
1da177e4 LT |
1715 | /** |
1716 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1717 | * @size: allocation size |
1da177e4 LT |
1718 | * Allocate enough pages to cover @size from the page level |
1719 | * allocator and map them into contiguous kernel virtual space. | |
1720 | * | |
c1c8897f | 1721 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1722 | * use __vmalloc() instead. |
1723 | */ | |
1724 | void *vmalloc(unsigned long size) | |
1725 | { | |
00ef2d2f DR |
1726 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
1727 | GFP_KERNEL | __GFP_HIGHMEM); | |
1da177e4 | 1728 | } |
1da177e4 LT |
1729 | EXPORT_SYMBOL(vmalloc); |
1730 | ||
e1ca7788 DY |
1731 | /** |
1732 | * vzalloc - allocate virtually contiguous memory with zero fill | |
1733 | * @size: allocation size | |
1734 | * Allocate enough pages to cover @size from the page level | |
1735 | * allocator and map them into contiguous kernel virtual space. | |
1736 | * The memory allocated is set to zero. | |
1737 | * | |
1738 | * For tight control over page level allocator and protection flags | |
1739 | * use __vmalloc() instead. | |
1740 | */ | |
1741 | void *vzalloc(unsigned long size) | |
1742 | { | |
00ef2d2f | 1743 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
e1ca7788 DY |
1744 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
1745 | } | |
1746 | EXPORT_SYMBOL(vzalloc); | |
1747 | ||
83342314 | 1748 | /** |
ead04089 REB |
1749 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1750 | * @size: allocation size | |
83342314 | 1751 | * |
ead04089 REB |
1752 | * The resulting memory area is zeroed so it can be mapped to userspace |
1753 | * without leaking data. | |
83342314 NP |
1754 | */ |
1755 | void *vmalloc_user(unsigned long size) | |
1756 | { | |
1757 | struct vm_struct *area; | |
1758 | void *ret; | |
1759 | ||
2dca6999 DM |
1760 | ret = __vmalloc_node(size, SHMLBA, |
1761 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
00ef2d2f DR |
1762 | PAGE_KERNEL, NUMA_NO_NODE, |
1763 | __builtin_return_address(0)); | |
2b4ac44e | 1764 | if (ret) { |
db64fe02 | 1765 | area = find_vm_area(ret); |
2b4ac44e | 1766 | area->flags |= VM_USERMAP; |
2b4ac44e | 1767 | } |
83342314 NP |
1768 | return ret; |
1769 | } | |
1770 | EXPORT_SYMBOL(vmalloc_user); | |
1771 | ||
930fc45a CL |
1772 | /** |
1773 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1774 | * @size: allocation size |
d44e0780 | 1775 | * @node: numa node |
930fc45a CL |
1776 | * |
1777 | * Allocate enough pages to cover @size from the page level | |
1778 | * allocator and map them into contiguous kernel virtual space. | |
1779 | * | |
c1c8897f | 1780 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1781 | * use __vmalloc() instead. |
1782 | */ | |
1783 | void *vmalloc_node(unsigned long size, int node) | |
1784 | { | |
2dca6999 | 1785 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
23016969 | 1786 | node, __builtin_return_address(0)); |
930fc45a CL |
1787 | } |
1788 | EXPORT_SYMBOL(vmalloc_node); | |
1789 | ||
e1ca7788 DY |
1790 | /** |
1791 | * vzalloc_node - allocate memory on a specific node with zero fill | |
1792 | * @size: allocation size | |
1793 | * @node: numa node | |
1794 | * | |
1795 | * Allocate enough pages to cover @size from the page level | |
1796 | * allocator and map them into contiguous kernel virtual space. | |
1797 | * The memory allocated is set to zero. | |
1798 | * | |
1799 | * For tight control over page level allocator and protection flags | |
1800 | * use __vmalloc_node() instead. | |
1801 | */ | |
1802 | void *vzalloc_node(unsigned long size, int node) | |
1803 | { | |
1804 | return __vmalloc_node_flags(size, node, | |
1805 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); | |
1806 | } | |
1807 | EXPORT_SYMBOL(vzalloc_node); | |
1808 | ||
4dc3b16b PP |
1809 | #ifndef PAGE_KERNEL_EXEC |
1810 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1811 | #endif | |
1812 | ||
1da177e4 LT |
1813 | /** |
1814 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1815 | * @size: allocation size |
1816 | * | |
1817 | * Kernel-internal function to allocate enough pages to cover @size | |
1818 | * the page level allocator and map them into contiguous and | |
1819 | * executable kernel virtual space. | |
1820 | * | |
c1c8897f | 1821 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1822 | * use __vmalloc() instead. |
1823 | */ | |
1824 | ||
1da177e4 LT |
1825 | void *vmalloc_exec(unsigned long size) |
1826 | { | |
2dca6999 | 1827 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
00ef2d2f | 1828 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 LT |
1829 | } |
1830 | ||
0d08e0d3 | 1831 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1832 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1833 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1834 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1835 | #else |
1836 | #define GFP_VMALLOC32 GFP_KERNEL | |
1837 | #endif | |
1838 | ||
1da177e4 LT |
1839 | /** |
1840 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1841 | * @size: allocation size |
1842 | * | |
1843 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1844 | * page level allocator and map them into contiguous kernel virtual space. | |
1845 | */ | |
1846 | void *vmalloc_32(unsigned long size) | |
1847 | { | |
2dca6999 | 1848 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 1849 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 1850 | } |
1da177e4 LT |
1851 | EXPORT_SYMBOL(vmalloc_32); |
1852 | ||
83342314 | 1853 | /** |
ead04089 | 1854 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1855 | * @size: allocation size |
ead04089 REB |
1856 | * |
1857 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1858 | * mapped to userspace without leaking data. | |
83342314 NP |
1859 | */ |
1860 | void *vmalloc_32_user(unsigned long size) | |
1861 | { | |
1862 | struct vm_struct *area; | |
1863 | void *ret; | |
1864 | ||
2dca6999 | 1865 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
00ef2d2f | 1866 | NUMA_NO_NODE, __builtin_return_address(0)); |
2b4ac44e | 1867 | if (ret) { |
db64fe02 | 1868 | area = find_vm_area(ret); |
2b4ac44e | 1869 | area->flags |= VM_USERMAP; |
2b4ac44e | 1870 | } |
83342314 NP |
1871 | return ret; |
1872 | } | |
1873 | EXPORT_SYMBOL(vmalloc_32_user); | |
1874 | ||
d0107eb0 KH |
1875 | /* |
1876 | * small helper routine , copy contents to buf from addr. | |
1877 | * If the page is not present, fill zero. | |
1878 | */ | |
1879 | ||
1880 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
1881 | { | |
1882 | struct page *p; | |
1883 | int copied = 0; | |
1884 | ||
1885 | while (count) { | |
1886 | unsigned long offset, length; | |
1887 | ||
1888 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1889 | length = PAGE_SIZE - offset; | |
1890 | if (length > count) | |
1891 | length = count; | |
1892 | p = vmalloc_to_page(addr); | |
1893 | /* | |
1894 | * To do safe access to this _mapped_ area, we need | |
1895 | * lock. But adding lock here means that we need to add | |
1896 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1897 | * interface, rarely used. Instead of that, we'll use | |
1898 | * kmap() and get small overhead in this access function. | |
1899 | */ | |
1900 | if (p) { | |
1901 | /* | |
1902 | * we can expect USER0 is not used (see vread/vwrite's | |
1903 | * function description) | |
1904 | */ | |
9b04c5fe | 1905 | void *map = kmap_atomic(p); |
d0107eb0 | 1906 | memcpy(buf, map + offset, length); |
9b04c5fe | 1907 | kunmap_atomic(map); |
d0107eb0 KH |
1908 | } else |
1909 | memset(buf, 0, length); | |
1910 | ||
1911 | addr += length; | |
1912 | buf += length; | |
1913 | copied += length; | |
1914 | count -= length; | |
1915 | } | |
1916 | return copied; | |
1917 | } | |
1918 | ||
1919 | static int aligned_vwrite(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(map + offset, buf, length); |
9b04c5fe | 1946 | kunmap_atomic(map); |
d0107eb0 KH |
1947 | } |
1948 | addr += length; | |
1949 | buf += length; | |
1950 | copied += length; | |
1951 | count -= length; | |
1952 | } | |
1953 | return copied; | |
1954 | } | |
1955 | ||
1956 | /** | |
1957 | * vread() - read vmalloc area in a safe way. | |
1958 | * @buf: buffer for reading data | |
1959 | * @addr: vm address. | |
1960 | * @count: number of bytes to be read. | |
1961 | * | |
1962 | * Returns # of bytes which addr and buf should be increased. | |
1963 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't | |
1964 | * includes any intersect with alive vmalloc area. | |
1965 | * | |
1966 | * This function checks that addr is a valid vmalloc'ed area, and | |
1967 | * copy data from that area to a given buffer. If the given memory range | |
1968 | * of [addr...addr+count) includes some valid address, data is copied to | |
1969 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
1970 | * IOREMAP area is treated as memory hole and no copy is done. | |
1971 | * | |
1972 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 1973 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
1974 | * |
1975 | * Note: In usual ops, vread() is never necessary because the caller | |
1976 | * should know vmalloc() area is valid and can use memcpy(). | |
1977 | * This is for routines which have to access vmalloc area without | |
1978 | * any informaion, as /dev/kmem. | |
1979 | * | |
1980 | */ | |
1981 | ||
1da177e4 LT |
1982 | long vread(char *buf, char *addr, unsigned long count) |
1983 | { | |
e81ce85f JK |
1984 | struct vmap_area *va; |
1985 | struct vm_struct *vm; | |
1da177e4 | 1986 | char *vaddr, *buf_start = buf; |
d0107eb0 | 1987 | unsigned long buflen = count; |
1da177e4 LT |
1988 | unsigned long n; |
1989 | ||
1990 | /* Don't allow overflow */ | |
1991 | if ((unsigned long) addr + count < count) | |
1992 | count = -(unsigned long) addr; | |
1993 | ||
e81ce85f JK |
1994 | spin_lock(&vmap_area_lock); |
1995 | list_for_each_entry(va, &vmap_area_list, list) { | |
1996 | if (!count) | |
1997 | break; | |
1998 | ||
1999 | if (!(va->flags & VM_VM_AREA)) | |
2000 | continue; | |
2001 | ||
2002 | vm = va->vm; | |
2003 | vaddr = (char *) vm->addr; | |
2004 | if (addr >= vaddr + vm->size - PAGE_SIZE) | |
1da177e4 LT |
2005 | continue; |
2006 | while (addr < vaddr) { | |
2007 | if (count == 0) | |
2008 | goto finished; | |
2009 | *buf = '\0'; | |
2010 | buf++; | |
2011 | addr++; | |
2012 | count--; | |
2013 | } | |
e81ce85f | 2014 | n = vaddr + vm->size - PAGE_SIZE - addr; |
d0107eb0 KH |
2015 | if (n > count) |
2016 | n = count; | |
e81ce85f | 2017 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2018 | aligned_vread(buf, addr, n); |
2019 | else /* IOREMAP area is treated as memory hole */ | |
2020 | memset(buf, 0, n); | |
2021 | buf += n; | |
2022 | addr += n; | |
2023 | count -= n; | |
1da177e4 LT |
2024 | } |
2025 | finished: | |
e81ce85f | 2026 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2027 | |
2028 | if (buf == buf_start) | |
2029 | return 0; | |
2030 | /* zero-fill memory holes */ | |
2031 | if (buf != buf_start + buflen) | |
2032 | memset(buf, 0, buflen - (buf - buf_start)); | |
2033 | ||
2034 | return buflen; | |
1da177e4 LT |
2035 | } |
2036 | ||
d0107eb0 KH |
2037 | /** |
2038 | * vwrite() - write vmalloc area in a safe way. | |
2039 | * @buf: buffer for source data | |
2040 | * @addr: vm address. | |
2041 | * @count: number of bytes to be read. | |
2042 | * | |
2043 | * Returns # of bytes which addr and buf should be incresed. | |
2044 | * (same number to @count). | |
2045 | * If [addr...addr+count) doesn't includes any intersect with valid | |
2046 | * vmalloc area, returns 0. | |
2047 | * | |
2048 | * This function checks that addr is a valid vmalloc'ed area, and | |
2049 | * copy data from a buffer to the given addr. If specified range of | |
2050 | * [addr...addr+count) includes some valid address, data is copied from | |
2051 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2052 | * IOREMAP area is treated as memory hole and no copy is done. | |
2053 | * | |
2054 | * If [addr...addr+count) doesn't includes any intersects with alive | |
a8e5202d | 2055 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
d0107eb0 KH |
2056 | * |
2057 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2058 | * should know vmalloc() area is valid and can use memcpy(). | |
2059 | * This is for routines which have to access vmalloc area without | |
2060 | * any informaion, as /dev/kmem. | |
d0107eb0 KH |
2061 | */ |
2062 | ||
1da177e4 LT |
2063 | long vwrite(char *buf, char *addr, unsigned long count) |
2064 | { | |
e81ce85f JK |
2065 | struct vmap_area *va; |
2066 | struct vm_struct *vm; | |
d0107eb0 KH |
2067 | char *vaddr; |
2068 | unsigned long n, buflen; | |
2069 | int copied = 0; | |
1da177e4 LT |
2070 | |
2071 | /* Don't allow overflow */ | |
2072 | if ((unsigned long) addr + count < count) | |
2073 | count = -(unsigned long) addr; | |
d0107eb0 | 2074 | buflen = count; |
1da177e4 | 2075 | |
e81ce85f JK |
2076 | spin_lock(&vmap_area_lock); |
2077 | list_for_each_entry(va, &vmap_area_list, list) { | |
2078 | if (!count) | |
2079 | break; | |
2080 | ||
2081 | if (!(va->flags & VM_VM_AREA)) | |
2082 | continue; | |
2083 | ||
2084 | vm = va->vm; | |
2085 | vaddr = (char *) vm->addr; | |
2086 | if (addr >= vaddr + vm->size - PAGE_SIZE) | |
1da177e4 LT |
2087 | continue; |
2088 | while (addr < vaddr) { | |
2089 | if (count == 0) | |
2090 | goto finished; | |
2091 | buf++; | |
2092 | addr++; | |
2093 | count--; | |
2094 | } | |
e81ce85f | 2095 | n = vaddr + vm->size - PAGE_SIZE - addr; |
d0107eb0 KH |
2096 | if (n > count) |
2097 | n = count; | |
e81ce85f | 2098 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2099 | aligned_vwrite(buf, addr, n); |
2100 | copied++; | |
2101 | } | |
2102 | buf += n; | |
2103 | addr += n; | |
2104 | count -= n; | |
1da177e4 LT |
2105 | } |
2106 | finished: | |
e81ce85f | 2107 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2108 | if (!copied) |
2109 | return 0; | |
2110 | return buflen; | |
1da177e4 | 2111 | } |
83342314 NP |
2112 | |
2113 | /** | |
2114 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
2115 | * @vma: vma to cover (map full range of vma) |
2116 | * @addr: vmalloc memory | |
2117 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
2118 | * |
2119 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
2120 | * |
2121 | * This function checks that addr is a valid vmalloc'ed area, and | |
2122 | * that it is big enough to cover the vma. Will return failure if | |
2123 | * that criteria isn't met. | |
2124 | * | |
72fd4a35 | 2125 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
2126 | */ |
2127 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
2128 | unsigned long pgoff) | |
2129 | { | |
2130 | struct vm_struct *area; | |
2131 | unsigned long uaddr = vma->vm_start; | |
2132 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
2133 | |
2134 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
2135 | return -EINVAL; | |
2136 | ||
db64fe02 | 2137 | area = find_vm_area(addr); |
83342314 | 2138 | if (!area) |
db64fe02 | 2139 | return -EINVAL; |
83342314 NP |
2140 | |
2141 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2142 | return -EINVAL; |
83342314 NP |
2143 | |
2144 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 2145 | return -EINVAL; |
83342314 NP |
2146 | |
2147 | addr += pgoff << PAGE_SHIFT; | |
2148 | do { | |
2149 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
2150 | int ret; |
2151 | ||
83342314 NP |
2152 | ret = vm_insert_page(vma, uaddr, page); |
2153 | if (ret) | |
2154 | return ret; | |
2155 | ||
2156 | uaddr += PAGE_SIZE; | |
2157 | addr += PAGE_SIZE; | |
2158 | usize -= PAGE_SIZE; | |
2159 | } while (usize > 0); | |
2160 | ||
314e51b9 | 2161 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 2162 | |
db64fe02 | 2163 | return 0; |
83342314 NP |
2164 | } |
2165 | EXPORT_SYMBOL(remap_vmalloc_range); | |
2166 | ||
1eeb66a1 CH |
2167 | /* |
2168 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
2169 | * have one. | |
2170 | */ | |
2171 | void __attribute__((weak)) vmalloc_sync_all(void) | |
2172 | { | |
2173 | } | |
5f4352fb JF |
2174 | |
2175 | ||
2f569afd | 2176 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb | 2177 | { |
cd12909c DV |
2178 | pte_t ***p = data; |
2179 | ||
2180 | if (p) { | |
2181 | *(*p) = pte; | |
2182 | (*p)++; | |
2183 | } | |
5f4352fb JF |
2184 | return 0; |
2185 | } | |
2186 | ||
2187 | /** | |
2188 | * alloc_vm_area - allocate a range of kernel address space | |
2189 | * @size: size of the area | |
cd12909c | 2190 | * @ptes: returns the PTEs for the address space |
7682486b RD |
2191 | * |
2192 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
2193 | * |
2194 | * This function reserves a range of kernel address space, and | |
2195 | * allocates pagetables to map that range. No actual mappings | |
cd12909c DV |
2196 | * are created. |
2197 | * | |
2198 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) | |
2199 | * allocated for the VM area are returned. | |
5f4352fb | 2200 | */ |
cd12909c | 2201 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
2202 | { |
2203 | struct vm_struct *area; | |
2204 | ||
23016969 CL |
2205 | area = get_vm_area_caller(size, VM_IOREMAP, |
2206 | __builtin_return_address(0)); | |
5f4352fb JF |
2207 | if (area == NULL) |
2208 | return NULL; | |
2209 | ||
2210 | /* | |
2211 | * This ensures that page tables are constructed for this region | |
2212 | * of kernel virtual address space and mapped into init_mm. | |
2213 | */ | |
2214 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 2215 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
2216 | free_vm_area(area); |
2217 | return NULL; | |
2218 | } | |
2219 | ||
5f4352fb JF |
2220 | return area; |
2221 | } | |
2222 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
2223 | ||
2224 | void free_vm_area(struct vm_struct *area) | |
2225 | { | |
2226 | struct vm_struct *ret; | |
2227 | ret = remove_vm_area(area->addr); | |
2228 | BUG_ON(ret != area); | |
2229 | kfree(area); | |
2230 | } | |
2231 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 2232 | |
4f8b02b4 | 2233 | #ifdef CONFIG_SMP |
ca23e405 TH |
2234 | static struct vmap_area *node_to_va(struct rb_node *n) |
2235 | { | |
2236 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; | |
2237 | } | |
2238 | ||
2239 | /** | |
2240 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
2241 | * @end: target address | |
2242 | * @pnext: out arg for the next vmap_area | |
2243 | * @pprev: out arg for the previous vmap_area | |
2244 | * | |
2245 | * Returns: %true if either or both of next and prev are found, | |
2246 | * %false if no vmap_area exists | |
2247 | * | |
2248 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2249 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2250 | */ | |
2251 | static bool pvm_find_next_prev(unsigned long end, | |
2252 | struct vmap_area **pnext, | |
2253 | struct vmap_area **pprev) | |
2254 | { | |
2255 | struct rb_node *n = vmap_area_root.rb_node; | |
2256 | struct vmap_area *va = NULL; | |
2257 | ||
2258 | while (n) { | |
2259 | va = rb_entry(n, struct vmap_area, rb_node); | |
2260 | if (end < va->va_end) | |
2261 | n = n->rb_left; | |
2262 | else if (end > va->va_end) | |
2263 | n = n->rb_right; | |
2264 | else | |
2265 | break; | |
2266 | } | |
2267 | ||
2268 | if (!va) | |
2269 | return false; | |
2270 | ||
2271 | if (va->va_end > end) { | |
2272 | *pnext = va; | |
2273 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2274 | } else { | |
2275 | *pprev = va; | |
2276 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2277 | } | |
2278 | return true; | |
2279 | } | |
2280 | ||
2281 | /** | |
2282 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2283 | * @pnext: in/out arg for the next vmap_area | |
2284 | * @pprev: in/out arg for the previous vmap_area | |
2285 | * @align: alignment | |
2286 | * | |
2287 | * Returns: determined end address | |
2288 | * | |
2289 | * Find the highest aligned address between *@pnext and *@pprev below | |
2290 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2291 | * down address is between the end addresses of the two vmap_areas. | |
2292 | * | |
2293 | * Please note that the address returned by this function may fall | |
2294 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2295 | * that. | |
2296 | */ | |
2297 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2298 | struct vmap_area **pprev, | |
2299 | unsigned long align) | |
2300 | { | |
2301 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2302 | unsigned long addr; | |
2303 | ||
2304 | if (*pnext) | |
2305 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2306 | else | |
2307 | addr = vmalloc_end; | |
2308 | ||
2309 | while (*pprev && (*pprev)->va_end > addr) { | |
2310 | *pnext = *pprev; | |
2311 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2312 | } | |
2313 | ||
2314 | return addr; | |
2315 | } | |
2316 | ||
2317 | /** | |
2318 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2319 | * @offsets: array containing offset of each area | |
2320 | * @sizes: array containing size of each area | |
2321 | * @nr_vms: the number of areas to allocate | |
2322 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
2323 | * |
2324 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2325 | * vm_structs on success, %NULL on failure | |
2326 | * | |
2327 | * Percpu allocator wants to use congruent vm areas so that it can | |
2328 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
2329 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
2330 | * be scattered pretty far, distance between two areas easily going up | |
2331 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
2332 | * areas are allocated from top. | |
ca23e405 TH |
2333 | * |
2334 | * Despite its complicated look, this allocator is rather simple. It | |
2335 | * does everything top-down and scans areas from the end looking for | |
2336 | * matching slot. While scanning, if any of the areas overlaps with | |
2337 | * existing vmap_area, the base address is pulled down to fit the | |
2338 | * area. Scanning is repeated till all the areas fit and then all | |
2339 | * necessary data structres are inserted and the result is returned. | |
2340 | */ | |
2341 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2342 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 2343 | size_t align) |
ca23e405 TH |
2344 | { |
2345 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2346 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2347 | struct vmap_area **vas, *prev, *next; | |
2348 | struct vm_struct **vms; | |
2349 | int area, area2, last_area, term_area; | |
2350 | unsigned long base, start, end, last_end; | |
2351 | bool purged = false; | |
2352 | ||
ca23e405 TH |
2353 | /* verify parameters and allocate data structures */ |
2354 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); | |
2355 | for (last_area = 0, area = 0; area < nr_vms; area++) { | |
2356 | start = offsets[area]; | |
2357 | end = start + sizes[area]; | |
2358 | ||
2359 | /* is everything aligned properly? */ | |
2360 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2361 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2362 | ||
2363 | /* detect the area with the highest address */ | |
2364 | if (start > offsets[last_area]) | |
2365 | last_area = area; | |
2366 | ||
2367 | for (area2 = 0; area2 < nr_vms; area2++) { | |
2368 | unsigned long start2 = offsets[area2]; | |
2369 | unsigned long end2 = start2 + sizes[area2]; | |
2370 | ||
2371 | if (area2 == area) | |
2372 | continue; | |
2373 | ||
2374 | BUG_ON(start2 >= start && start2 < end); | |
2375 | BUG_ON(end2 <= end && end2 > start); | |
2376 | } | |
2377 | } | |
2378 | last_end = offsets[last_area] + sizes[last_area]; | |
2379 | ||
2380 | if (vmalloc_end - vmalloc_start < last_end) { | |
2381 | WARN_ON(true); | |
2382 | return NULL; | |
2383 | } | |
2384 | ||
4d67d860 TM |
2385 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
2386 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 2387 | if (!vas || !vms) |
f1db7afd | 2388 | goto err_free2; |
ca23e405 TH |
2389 | |
2390 | for (area = 0; area < nr_vms; area++) { | |
ec3f64fc DR |
2391 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
2392 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); | |
ca23e405 TH |
2393 | if (!vas[area] || !vms[area]) |
2394 | goto err_free; | |
2395 | } | |
2396 | retry: | |
2397 | spin_lock(&vmap_area_lock); | |
2398 | ||
2399 | /* start scanning - we scan from the top, begin with the last area */ | |
2400 | area = term_area = last_area; | |
2401 | start = offsets[area]; | |
2402 | end = start + sizes[area]; | |
2403 | ||
2404 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2405 | base = vmalloc_end - last_end; | |
2406 | goto found; | |
2407 | } | |
2408 | base = pvm_determine_end(&next, &prev, align) - end; | |
2409 | ||
2410 | while (true) { | |
2411 | BUG_ON(next && next->va_end <= base + end); | |
2412 | BUG_ON(prev && prev->va_end > base + end); | |
2413 | ||
2414 | /* | |
2415 | * base might have underflowed, add last_end before | |
2416 | * comparing. | |
2417 | */ | |
2418 | if (base + last_end < vmalloc_start + last_end) { | |
2419 | spin_unlock(&vmap_area_lock); | |
2420 | if (!purged) { | |
2421 | purge_vmap_area_lazy(); | |
2422 | purged = true; | |
2423 | goto retry; | |
2424 | } | |
2425 | goto err_free; | |
2426 | } | |
2427 | ||
2428 | /* | |
2429 | * If next overlaps, move base downwards so that it's | |
2430 | * right below next and then recheck. | |
2431 | */ | |
2432 | if (next && next->va_start < base + end) { | |
2433 | base = pvm_determine_end(&next, &prev, align) - end; | |
2434 | term_area = area; | |
2435 | continue; | |
2436 | } | |
2437 | ||
2438 | /* | |
2439 | * If prev overlaps, shift down next and prev and move | |
2440 | * base so that it's right below new next and then | |
2441 | * recheck. | |
2442 | */ | |
2443 | if (prev && prev->va_end > base + start) { | |
2444 | next = prev; | |
2445 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2446 | base = pvm_determine_end(&next, &prev, align) - end; | |
2447 | term_area = area; | |
2448 | continue; | |
2449 | } | |
2450 | ||
2451 | /* | |
2452 | * This area fits, move on to the previous one. If | |
2453 | * the previous one is the terminal one, we're done. | |
2454 | */ | |
2455 | area = (area + nr_vms - 1) % nr_vms; | |
2456 | if (area == term_area) | |
2457 | break; | |
2458 | start = offsets[area]; | |
2459 | end = start + sizes[area]; | |
2460 | pvm_find_next_prev(base + end, &next, &prev); | |
2461 | } | |
2462 | found: | |
2463 | /* we've found a fitting base, insert all va's */ | |
2464 | for (area = 0; area < nr_vms; area++) { | |
2465 | struct vmap_area *va = vas[area]; | |
2466 | ||
2467 | va->va_start = base + offsets[area]; | |
2468 | va->va_end = va->va_start + sizes[area]; | |
2469 | __insert_vmap_area(va); | |
2470 | } | |
2471 | ||
2472 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2473 | ||
2474 | spin_unlock(&vmap_area_lock); | |
2475 | ||
2476 | /* insert all vm's */ | |
2477 | for (area = 0; area < nr_vms; area++) | |
2478 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, | |
2479 | pcpu_get_vm_areas); | |
2480 | ||
2481 | kfree(vas); | |
2482 | return vms; | |
2483 | ||
2484 | err_free: | |
2485 | for (area = 0; area < nr_vms; area++) { | |
f1db7afd KC |
2486 | kfree(vas[area]); |
2487 | kfree(vms[area]); | |
ca23e405 | 2488 | } |
f1db7afd | 2489 | err_free2: |
ca23e405 TH |
2490 | kfree(vas); |
2491 | kfree(vms); | |
2492 | return NULL; | |
2493 | } | |
2494 | ||
2495 | /** | |
2496 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2497 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2498 | * @nr_vms: the number of allocated areas | |
2499 | * | |
2500 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2501 | */ | |
2502 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2503 | { | |
2504 | int i; | |
2505 | ||
2506 | for (i = 0; i < nr_vms; i++) | |
2507 | free_vm_area(vms[i]); | |
2508 | kfree(vms); | |
2509 | } | |
4f8b02b4 | 2510 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
2511 | |
2512 | #ifdef CONFIG_PROC_FS | |
2513 | static void *s_start(struct seq_file *m, loff_t *pos) | |
d4033afd | 2514 | __acquires(&vmap_area_lock) |
a10aa579 CL |
2515 | { |
2516 | loff_t n = *pos; | |
d4033afd | 2517 | struct vmap_area *va; |
a10aa579 | 2518 | |
d4033afd JK |
2519 | spin_lock(&vmap_area_lock); |
2520 | va = list_entry((&vmap_area_list)->next, typeof(*va), list); | |
2521 | while (n > 0 && &va->list != &vmap_area_list) { | |
a10aa579 | 2522 | n--; |
d4033afd | 2523 | va = list_entry(va->list.next, typeof(*va), list); |
a10aa579 | 2524 | } |
d4033afd JK |
2525 | if (!n && &va->list != &vmap_area_list) |
2526 | return va; | |
a10aa579 CL |
2527 | |
2528 | return NULL; | |
2529 | ||
2530 | } | |
2531 | ||
2532 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2533 | { | |
d4033afd | 2534 | struct vmap_area *va = p, *next; |
a10aa579 CL |
2535 | |
2536 | ++*pos; | |
d4033afd JK |
2537 | next = list_entry(va->list.next, typeof(*va), list); |
2538 | if (&next->list != &vmap_area_list) | |
2539 | return next; | |
2540 | ||
2541 | return NULL; | |
a10aa579 CL |
2542 | } |
2543 | ||
2544 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 2545 | __releases(&vmap_area_lock) |
a10aa579 | 2546 | { |
d4033afd | 2547 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
2548 | } |
2549 | ||
a47a126a ED |
2550 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2551 | { | |
e5adfffc | 2552 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
2553 | unsigned int nr, *counters = m->private; |
2554 | ||
2555 | if (!counters) | |
2556 | return; | |
2557 | ||
4341fa45 | 2558 | /* Pair with smp_wmb() in clear_vm_unlist() */ |
d4033afd JK |
2559 | smp_rmb(); |
2560 | if (v->flags & VM_UNLIST) | |
2561 | return; | |
2562 | ||
a47a126a ED |
2563 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
2564 | ||
2565 | for (nr = 0; nr < v->nr_pages; nr++) | |
2566 | counters[page_to_nid(v->pages[nr])]++; | |
2567 | ||
2568 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2569 | if (counters[nr]) | |
2570 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2571 | } | |
2572 | } | |
2573 | ||
a10aa579 CL |
2574 | static int s_show(struct seq_file *m, void *p) |
2575 | { | |
d4033afd JK |
2576 | struct vmap_area *va = p; |
2577 | struct vm_struct *v; | |
2578 | ||
2579 | if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING)) | |
2580 | return 0; | |
2581 | ||
2582 | if (!(va->flags & VM_VM_AREA)) { | |
2583 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", | |
2584 | (void *)va->va_start, (void *)va->va_end, | |
2585 | va->va_end - va->va_start); | |
2586 | return 0; | |
2587 | } | |
2588 | ||
2589 | v = va->vm; | |
a10aa579 | 2590 | |
45ec1690 | 2591 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
2592 | v->addr, v->addr + v->size, v->size); |
2593 | ||
62c70bce JP |
2594 | if (v->caller) |
2595 | seq_printf(m, " %pS", v->caller); | |
23016969 | 2596 | |
a10aa579 CL |
2597 | if (v->nr_pages) |
2598 | seq_printf(m, " pages=%d", v->nr_pages); | |
2599 | ||
2600 | if (v->phys_addr) | |
ffa71f33 | 2601 | seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr); |
a10aa579 CL |
2602 | |
2603 | if (v->flags & VM_IOREMAP) | |
2604 | seq_printf(m, " ioremap"); | |
2605 | ||
2606 | if (v->flags & VM_ALLOC) | |
2607 | seq_printf(m, " vmalloc"); | |
2608 | ||
2609 | if (v->flags & VM_MAP) | |
2610 | seq_printf(m, " vmap"); | |
2611 | ||
2612 | if (v->flags & VM_USERMAP) | |
2613 | seq_printf(m, " user"); | |
2614 | ||
2615 | if (v->flags & VM_VPAGES) | |
2616 | seq_printf(m, " vpages"); | |
2617 | ||
a47a126a | 2618 | show_numa_info(m, v); |
a10aa579 CL |
2619 | seq_putc(m, '\n'); |
2620 | return 0; | |
2621 | } | |
2622 | ||
5f6a6a9c | 2623 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2624 | .start = s_start, |
2625 | .next = s_next, | |
2626 | .stop = s_stop, | |
2627 | .show = s_show, | |
2628 | }; | |
5f6a6a9c AD |
2629 | |
2630 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2631 | { | |
2632 | unsigned int *ptr = NULL; | |
2633 | int ret; | |
2634 | ||
e5adfffc | 2635 | if (IS_ENABLED(CONFIG_NUMA)) { |
5f6a6a9c | 2636 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); |
51980ac9 KV |
2637 | if (ptr == NULL) |
2638 | return -ENOMEM; | |
2639 | } | |
5f6a6a9c AD |
2640 | ret = seq_open(file, &vmalloc_op); |
2641 | if (!ret) { | |
2642 | struct seq_file *m = file->private_data; | |
2643 | m->private = ptr; | |
2644 | } else | |
2645 | kfree(ptr); | |
2646 | return ret; | |
2647 | } | |
2648 | ||
2649 | static const struct file_operations proc_vmalloc_operations = { | |
2650 | .open = vmalloc_open, | |
2651 | .read = seq_read, | |
2652 | .llseek = seq_lseek, | |
2653 | .release = seq_release_private, | |
2654 | }; | |
2655 | ||
2656 | static int __init proc_vmalloc_init(void) | |
2657 | { | |
2658 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2659 | return 0; | |
2660 | } | |
2661 | module_init(proc_vmalloc_init); | |
db3808c1 JK |
2662 | |
2663 | void get_vmalloc_info(struct vmalloc_info *vmi) | |
2664 | { | |
f98782dd | 2665 | struct vmap_area *va; |
db3808c1 JK |
2666 | unsigned long free_area_size; |
2667 | unsigned long prev_end; | |
2668 | ||
2669 | vmi->used = 0; | |
f98782dd | 2670 | vmi->largest_chunk = 0; |
db3808c1 | 2671 | |
f98782dd | 2672 | prev_end = VMALLOC_START; |
db3808c1 | 2673 | |
f98782dd | 2674 | spin_lock(&vmap_area_lock); |
db3808c1 | 2675 | |
f98782dd JK |
2676 | if (list_empty(&vmap_area_list)) { |
2677 | vmi->largest_chunk = VMALLOC_TOTAL; | |
2678 | goto out; | |
2679 | } | |
db3808c1 | 2680 | |
f98782dd JK |
2681 | list_for_each_entry(va, &vmap_area_list, list) { |
2682 | unsigned long addr = va->va_start; | |
db3808c1 | 2683 | |
f98782dd JK |
2684 | /* |
2685 | * Some archs keep another range for modules in vmalloc space | |
2686 | */ | |
2687 | if (addr < VMALLOC_START) | |
2688 | continue; | |
2689 | if (addr >= VMALLOC_END) | |
2690 | break; | |
db3808c1 | 2691 | |
f98782dd JK |
2692 | if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING)) |
2693 | continue; | |
db3808c1 | 2694 | |
f98782dd | 2695 | vmi->used += (va->va_end - va->va_start); |
db3808c1 | 2696 | |
f98782dd JK |
2697 | free_area_size = addr - prev_end; |
2698 | if (vmi->largest_chunk < free_area_size) | |
2699 | vmi->largest_chunk = free_area_size; | |
db3808c1 | 2700 | |
f98782dd | 2701 | prev_end = va->va_end; |
db3808c1 | 2702 | } |
f98782dd JK |
2703 | |
2704 | if (VMALLOC_END - prev_end > vmi->largest_chunk) | |
2705 | vmi->largest_chunk = VMALLOC_END - prev_end; | |
2706 | ||
2707 | out: | |
2708 | spin_unlock(&vmap_area_lock); | |
db3808c1 | 2709 | } |
a10aa579 CL |
2710 | #endif |
2711 |