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> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/spinlock.h> | |
17 | #include <linux/interrupt.h> | |
5f6a6a9c | 18 | #include <linux/proc_fs.h> |
a10aa579 | 19 | #include <linux/seq_file.h> |
3ac7fe5a | 20 | #include <linux/debugobjects.h> |
23016969 | 21 | #include <linux/kallsyms.h> |
db64fe02 NP |
22 | #include <linux/list.h> |
23 | #include <linux/rbtree.h> | |
24 | #include <linux/radix-tree.h> | |
25 | #include <linux/rcupdate.h> | |
f0aa6617 | 26 | #include <linux/pfn.h> |
89219d37 | 27 | #include <linux/kmemleak.h> |
1da177e4 | 28 | |
db64fe02 | 29 | #include <asm/atomic.h> |
1da177e4 LT |
30 | #include <asm/uaccess.h> |
31 | #include <asm/tlbflush.h> | |
32 | ||
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 | ||
73bdf0a6 LT |
187 | static inline int is_vmalloc_or_module_addr(const void *x) |
188 | { | |
189 | /* | |
ab4f2ee1 | 190 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
191 | * and fall back on vmalloc() if that fails. Others |
192 | * just put it in the vmalloc space. | |
193 | */ | |
194 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
195 | unsigned long addr = (unsigned long)x; | |
196 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
197 | return 1; | |
198 | #endif | |
199 | return is_vmalloc_addr(x); | |
200 | } | |
201 | ||
48667e7a | 202 | /* |
db64fe02 | 203 | * Walk a vmap address to the struct page it maps. |
48667e7a | 204 | */ |
b3bdda02 | 205 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
206 | { |
207 | unsigned long addr = (unsigned long) vmalloc_addr; | |
208 | struct page *page = NULL; | |
209 | pgd_t *pgd = pgd_offset_k(addr); | |
48667e7a | 210 | |
7aa413de IM |
211 | /* |
212 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
213 | * architectures that do not vmalloc module space | |
214 | */ | |
73bdf0a6 | 215 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 216 | |
48667e7a | 217 | if (!pgd_none(*pgd)) { |
db64fe02 | 218 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 219 | if (!pud_none(*pud)) { |
db64fe02 | 220 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 221 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
222 | pte_t *ptep, pte; |
223 | ||
48667e7a CL |
224 | ptep = pte_offset_map(pmd, addr); |
225 | pte = *ptep; | |
226 | if (pte_present(pte)) | |
227 | page = pte_page(pte); | |
228 | pte_unmap(ptep); | |
229 | } | |
230 | } | |
231 | } | |
232 | return page; | |
233 | } | |
234 | EXPORT_SYMBOL(vmalloc_to_page); | |
235 | ||
236 | /* | |
237 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
238 | */ | |
b3bdda02 | 239 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
240 | { |
241 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
242 | } | |
243 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
244 | ||
db64fe02 NP |
245 | |
246 | /*** Global kva allocator ***/ | |
247 | ||
248 | #define VM_LAZY_FREE 0x01 | |
249 | #define VM_LAZY_FREEING 0x02 | |
250 | #define VM_VM_AREA 0x04 | |
251 | ||
252 | struct vmap_area { | |
253 | unsigned long va_start; | |
254 | unsigned long va_end; | |
255 | unsigned long flags; | |
256 | struct rb_node rb_node; /* address sorted rbtree */ | |
257 | struct list_head list; /* address sorted list */ | |
258 | struct list_head purge_list; /* "lazy purge" list */ | |
259 | void *private; | |
260 | struct rcu_head rcu_head; | |
261 | }; | |
262 | ||
263 | static DEFINE_SPINLOCK(vmap_area_lock); | |
264 | static struct rb_root vmap_area_root = RB_ROOT; | |
265 | static LIST_HEAD(vmap_area_list); | |
ca23e405 | 266 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
267 | |
268 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 269 | { |
db64fe02 NP |
270 | struct rb_node *n = vmap_area_root.rb_node; |
271 | ||
272 | while (n) { | |
273 | struct vmap_area *va; | |
274 | ||
275 | va = rb_entry(n, struct vmap_area, rb_node); | |
276 | if (addr < va->va_start) | |
277 | n = n->rb_left; | |
278 | else if (addr > va->va_start) | |
279 | n = n->rb_right; | |
280 | else | |
281 | return va; | |
282 | } | |
283 | ||
284 | return NULL; | |
285 | } | |
286 | ||
287 | static void __insert_vmap_area(struct vmap_area *va) | |
288 | { | |
289 | struct rb_node **p = &vmap_area_root.rb_node; | |
290 | struct rb_node *parent = NULL; | |
291 | struct rb_node *tmp; | |
292 | ||
293 | while (*p) { | |
294 | struct vmap_area *tmp; | |
295 | ||
296 | parent = *p; | |
297 | tmp = rb_entry(parent, struct vmap_area, rb_node); | |
298 | if (va->va_start < tmp->va_end) | |
299 | p = &(*p)->rb_left; | |
300 | else if (va->va_end > tmp->va_start) | |
301 | p = &(*p)->rb_right; | |
302 | else | |
303 | BUG(); | |
304 | } | |
305 | ||
306 | rb_link_node(&va->rb_node, parent, p); | |
307 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
308 | ||
309 | /* address-sort this list so it is usable like the vmlist */ | |
310 | tmp = rb_prev(&va->rb_node); | |
311 | if (tmp) { | |
312 | struct vmap_area *prev; | |
313 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
314 | list_add_rcu(&va->list, &prev->list); | |
315 | } else | |
316 | list_add_rcu(&va->list, &vmap_area_list); | |
317 | } | |
318 | ||
319 | static void purge_vmap_area_lazy(void); | |
320 | ||
321 | /* | |
322 | * Allocate a region of KVA of the specified size and alignment, within the | |
323 | * vstart and vend. | |
324 | */ | |
325 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
326 | unsigned long align, | |
327 | unsigned long vstart, unsigned long vend, | |
328 | int node, gfp_t gfp_mask) | |
329 | { | |
330 | struct vmap_area *va; | |
331 | struct rb_node *n; | |
1da177e4 | 332 | unsigned long addr; |
db64fe02 NP |
333 | int purged = 0; |
334 | ||
7766970c | 335 | BUG_ON(!size); |
db64fe02 NP |
336 | BUG_ON(size & ~PAGE_MASK); |
337 | ||
db64fe02 NP |
338 | va = kmalloc_node(sizeof(struct vmap_area), |
339 | gfp_mask & GFP_RECLAIM_MASK, node); | |
340 | if (unlikely(!va)) | |
341 | return ERR_PTR(-ENOMEM); | |
342 | ||
343 | retry: | |
0ae15132 GC |
344 | addr = ALIGN(vstart, align); |
345 | ||
db64fe02 | 346 | spin_lock(&vmap_area_lock); |
7766970c NP |
347 | if (addr + size - 1 < addr) |
348 | goto overflow; | |
349 | ||
db64fe02 NP |
350 | /* XXX: could have a last_hole cache */ |
351 | n = vmap_area_root.rb_node; | |
352 | if (n) { | |
353 | struct vmap_area *first = NULL; | |
354 | ||
355 | do { | |
356 | struct vmap_area *tmp; | |
357 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
358 | if (tmp->va_end >= addr) { | |
359 | if (!first && tmp->va_start < addr + size) | |
360 | first = tmp; | |
361 | n = n->rb_left; | |
362 | } else { | |
363 | first = tmp; | |
364 | n = n->rb_right; | |
365 | } | |
366 | } while (n); | |
367 | ||
368 | if (!first) | |
369 | goto found; | |
370 | ||
371 | if (first->va_end < addr) { | |
372 | n = rb_next(&first->rb_node); | |
373 | if (n) | |
374 | first = rb_entry(n, struct vmap_area, rb_node); | |
375 | else | |
376 | goto found; | |
377 | } | |
378 | ||
f011c2da | 379 | while (addr + size > first->va_start && addr + size <= vend) { |
db64fe02 | 380 | addr = ALIGN(first->va_end + PAGE_SIZE, align); |
7766970c NP |
381 | if (addr + size - 1 < addr) |
382 | goto overflow; | |
db64fe02 NP |
383 | |
384 | n = rb_next(&first->rb_node); | |
385 | if (n) | |
386 | first = rb_entry(n, struct vmap_area, rb_node); | |
387 | else | |
388 | goto found; | |
389 | } | |
390 | } | |
391 | found: | |
392 | if (addr + size > vend) { | |
7766970c | 393 | overflow: |
db64fe02 NP |
394 | spin_unlock(&vmap_area_lock); |
395 | if (!purged) { | |
396 | purge_vmap_area_lazy(); | |
397 | purged = 1; | |
398 | goto retry; | |
399 | } | |
400 | if (printk_ratelimit()) | |
c1279c4e GC |
401 | printk(KERN_WARNING |
402 | "vmap allocation for size %lu failed: " | |
403 | "use vmalloc=<size> to increase size.\n", size); | |
2498ce42 | 404 | kfree(va); |
db64fe02 NP |
405 | return ERR_PTR(-EBUSY); |
406 | } | |
407 | ||
408 | BUG_ON(addr & (align-1)); | |
409 | ||
410 | va->va_start = addr; | |
411 | va->va_end = addr + size; | |
412 | va->flags = 0; | |
413 | __insert_vmap_area(va); | |
414 | spin_unlock(&vmap_area_lock); | |
415 | ||
416 | return va; | |
417 | } | |
418 | ||
419 | static void rcu_free_va(struct rcu_head *head) | |
420 | { | |
421 | struct vmap_area *va = container_of(head, struct vmap_area, rcu_head); | |
422 | ||
423 | kfree(va); | |
424 | } | |
425 | ||
426 | static void __free_vmap_area(struct vmap_area *va) | |
427 | { | |
428 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
429 | rb_erase(&va->rb_node, &vmap_area_root); | |
430 | RB_CLEAR_NODE(&va->rb_node); | |
431 | list_del_rcu(&va->list); | |
432 | ||
ca23e405 TH |
433 | /* |
434 | * Track the highest possible candidate for pcpu area | |
435 | * allocation. Areas outside of vmalloc area can be returned | |
436 | * here too, consider only end addresses which fall inside | |
437 | * vmalloc area proper. | |
438 | */ | |
439 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
440 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
441 | ||
db64fe02 NP |
442 | call_rcu(&va->rcu_head, rcu_free_va); |
443 | } | |
444 | ||
445 | /* | |
446 | * Free a region of KVA allocated by alloc_vmap_area | |
447 | */ | |
448 | static void free_vmap_area(struct vmap_area *va) | |
449 | { | |
450 | spin_lock(&vmap_area_lock); | |
451 | __free_vmap_area(va); | |
452 | spin_unlock(&vmap_area_lock); | |
453 | } | |
454 | ||
455 | /* | |
456 | * Clear the pagetable entries of a given vmap_area | |
457 | */ | |
458 | static void unmap_vmap_area(struct vmap_area *va) | |
459 | { | |
460 | vunmap_page_range(va->va_start, va->va_end); | |
461 | } | |
462 | ||
cd52858c NP |
463 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
464 | { | |
465 | /* | |
466 | * Unmap page tables and force a TLB flush immediately if | |
467 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free | |
468 | * bugs similarly to those in linear kernel virtual address | |
469 | * space after a page has been freed. | |
470 | * | |
471 | * All the lazy freeing logic is still retained, in order to | |
472 | * minimise intrusiveness of this debugging feature. | |
473 | * | |
474 | * This is going to be *slow* (linear kernel virtual address | |
475 | * debugging doesn't do a broadcast TLB flush so it is a lot | |
476 | * faster). | |
477 | */ | |
478 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
479 | vunmap_page_range(start, end); | |
480 | flush_tlb_kernel_range(start, end); | |
481 | #endif | |
482 | } | |
483 | ||
db64fe02 NP |
484 | /* |
485 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
486 | * before attempting to purge with a TLB flush. | |
487 | * | |
488 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
489 | * and take slightly longer to purge, but it will linearly reduce the number of | |
490 | * global TLB flushes that must be performed. It would seem natural to scale | |
491 | * this number up linearly with the number of CPUs (because vmapping activity | |
492 | * could also scale linearly with the number of CPUs), however it is likely | |
493 | * that in practice, workloads might be constrained in other ways that mean | |
494 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
495 | * conservative and not introduce a big latency on huge systems, so go with | |
496 | * a less aggressive log scale. It will still be an improvement over the old | |
497 | * code, and it will be simple to change the scale factor if we find that it | |
498 | * becomes a problem on bigger systems. | |
499 | */ | |
500 | static unsigned long lazy_max_pages(void) | |
501 | { | |
502 | unsigned int log; | |
503 | ||
504 | log = fls(num_online_cpus()); | |
505 | ||
506 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
507 | } | |
508 | ||
509 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
510 | ||
511 | /* | |
512 | * Purges all lazily-freed vmap areas. | |
513 | * | |
514 | * If sync is 0 then don't purge if there is already a purge in progress. | |
515 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
516 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
517 | * their own TLB flushing). | |
518 | * Returns with *start = min(*start, lowest purged address) | |
519 | * *end = max(*end, highest purged address) | |
520 | */ | |
521 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
522 | int sync, int force_flush) | |
523 | { | |
46666d8a | 524 | static DEFINE_SPINLOCK(purge_lock); |
db64fe02 NP |
525 | LIST_HEAD(valist); |
526 | struct vmap_area *va; | |
cbb76676 | 527 | struct vmap_area *n_va; |
db64fe02 NP |
528 | int nr = 0; |
529 | ||
530 | /* | |
531 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
532 | * should not expect such behaviour. This just simplifies locking for | |
533 | * the case that isn't actually used at the moment anyway. | |
534 | */ | |
535 | if (!sync && !force_flush) { | |
46666d8a | 536 | if (!spin_trylock(&purge_lock)) |
db64fe02 NP |
537 | return; |
538 | } else | |
46666d8a | 539 | spin_lock(&purge_lock); |
db64fe02 NP |
540 | |
541 | rcu_read_lock(); | |
542 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
543 | if (va->flags & VM_LAZY_FREE) { | |
544 | if (va->va_start < *start) | |
545 | *start = va->va_start; | |
546 | if (va->va_end > *end) | |
547 | *end = va->va_end; | |
548 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
549 | unmap_vmap_area(va); | |
550 | list_add_tail(&va->purge_list, &valist); | |
551 | va->flags |= VM_LAZY_FREEING; | |
552 | va->flags &= ~VM_LAZY_FREE; | |
553 | } | |
554 | } | |
555 | rcu_read_unlock(); | |
556 | ||
557 | if (nr) { | |
558 | BUG_ON(nr > atomic_read(&vmap_lazy_nr)); | |
559 | atomic_sub(nr, &vmap_lazy_nr); | |
560 | } | |
561 | ||
562 | if (nr || force_flush) | |
563 | flush_tlb_kernel_range(*start, *end); | |
564 | ||
565 | if (nr) { | |
566 | spin_lock(&vmap_area_lock); | |
cbb76676 | 567 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
db64fe02 NP |
568 | __free_vmap_area(va); |
569 | spin_unlock(&vmap_area_lock); | |
570 | } | |
46666d8a | 571 | spin_unlock(&purge_lock); |
db64fe02 NP |
572 | } |
573 | ||
496850e5 NP |
574 | /* |
575 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
576 | * is already purging. | |
577 | */ | |
578 | static void try_purge_vmap_area_lazy(void) | |
579 | { | |
580 | unsigned long start = ULONG_MAX, end = 0; | |
581 | ||
582 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
583 | } | |
584 | ||
db64fe02 NP |
585 | /* |
586 | * Kick off a purge of the outstanding lazy areas. | |
587 | */ | |
588 | static void purge_vmap_area_lazy(void) | |
589 | { | |
590 | unsigned long start = ULONG_MAX, end = 0; | |
591 | ||
496850e5 | 592 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
db64fe02 NP |
593 | } |
594 | ||
595 | /* | |
b29acbdc NP |
596 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been |
597 | * called for the correct range previously. | |
db64fe02 | 598 | */ |
b29acbdc | 599 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) |
db64fe02 NP |
600 | { |
601 | va->flags |= VM_LAZY_FREE; | |
602 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
603 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
496850e5 | 604 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
605 | } |
606 | ||
b29acbdc NP |
607 | /* |
608 | * Free and unmap a vmap area | |
609 | */ | |
610 | static void free_unmap_vmap_area(struct vmap_area *va) | |
611 | { | |
612 | flush_cache_vunmap(va->va_start, va->va_end); | |
613 | free_unmap_vmap_area_noflush(va); | |
614 | } | |
615 | ||
db64fe02 NP |
616 | static struct vmap_area *find_vmap_area(unsigned long addr) |
617 | { | |
618 | struct vmap_area *va; | |
619 | ||
620 | spin_lock(&vmap_area_lock); | |
621 | va = __find_vmap_area(addr); | |
622 | spin_unlock(&vmap_area_lock); | |
623 | ||
624 | return va; | |
625 | } | |
626 | ||
627 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
628 | { | |
629 | struct vmap_area *va; | |
630 | ||
631 | va = find_vmap_area(addr); | |
632 | BUG_ON(!va); | |
633 | free_unmap_vmap_area(va); | |
634 | } | |
635 | ||
636 | ||
637 | /*** Per cpu kva allocator ***/ | |
638 | ||
639 | /* | |
640 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
641 | * room for at least 16 percpu vmap blocks per CPU. | |
642 | */ | |
643 | /* | |
644 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
645 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
646 | * instead (we just need a rough idea) | |
647 | */ | |
648 | #if BITS_PER_LONG == 32 | |
649 | #define VMALLOC_SPACE (128UL*1024*1024) | |
650 | #else | |
651 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
652 | #endif | |
653 | ||
654 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
655 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
656 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
657 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
658 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
659 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
660 | #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
661 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
662 | VMALLOC_PAGES / NR_CPUS / 16)) | |
663 | ||
664 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
665 | ||
9b463334 JF |
666 | static bool vmap_initialized __read_mostly = false; |
667 | ||
db64fe02 NP |
668 | struct vmap_block_queue { |
669 | spinlock_t lock; | |
670 | struct list_head free; | |
671 | struct list_head dirty; | |
672 | unsigned int nr_dirty; | |
673 | }; | |
674 | ||
675 | struct vmap_block { | |
676 | spinlock_t lock; | |
677 | struct vmap_area *va; | |
678 | struct vmap_block_queue *vbq; | |
679 | unsigned long free, dirty; | |
680 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
681 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
682 | union { | |
d086817d | 683 | struct list_head free_list; |
db64fe02 NP |
684 | struct rcu_head rcu_head; |
685 | }; | |
686 | }; | |
687 | ||
688 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
689 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
690 | ||
691 | /* | |
692 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
693 | * in the free path. Could get rid of this if we change the API to return a | |
694 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
695 | */ | |
696 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
697 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
698 | ||
699 | /* | |
700 | * We should probably have a fallback mechanism to allocate virtual memory | |
701 | * out of partially filled vmap blocks. However vmap block sizing should be | |
702 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
703 | * big problem. | |
704 | */ | |
705 | ||
706 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
707 | { | |
708 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
709 | addr /= VMAP_BLOCK_SIZE; | |
710 | return addr; | |
711 | } | |
712 | ||
713 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
714 | { | |
715 | struct vmap_block_queue *vbq; | |
716 | struct vmap_block *vb; | |
717 | struct vmap_area *va; | |
718 | unsigned long vb_idx; | |
719 | int node, err; | |
720 | ||
721 | node = numa_node_id(); | |
722 | ||
723 | vb = kmalloc_node(sizeof(struct vmap_block), | |
724 | gfp_mask & GFP_RECLAIM_MASK, node); | |
725 | if (unlikely(!vb)) | |
726 | return ERR_PTR(-ENOMEM); | |
727 | ||
728 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
729 | VMALLOC_START, VMALLOC_END, | |
730 | node, gfp_mask); | |
731 | if (unlikely(IS_ERR(va))) { | |
732 | kfree(vb); | |
733 | return ERR_PTR(PTR_ERR(va)); | |
734 | } | |
735 | ||
736 | err = radix_tree_preload(gfp_mask); | |
737 | if (unlikely(err)) { | |
738 | kfree(vb); | |
739 | free_vmap_area(va); | |
740 | return ERR_PTR(err); | |
741 | } | |
742 | ||
743 | spin_lock_init(&vb->lock); | |
744 | vb->va = va; | |
745 | vb->free = VMAP_BBMAP_BITS; | |
746 | vb->dirty = 0; | |
747 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
748 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
749 | INIT_LIST_HEAD(&vb->free_list); | |
db64fe02 NP |
750 | |
751 | vb_idx = addr_to_vb_idx(va->va_start); | |
752 | spin_lock(&vmap_block_tree_lock); | |
753 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
754 | spin_unlock(&vmap_block_tree_lock); | |
755 | BUG_ON(err); | |
756 | radix_tree_preload_end(); | |
757 | ||
758 | vbq = &get_cpu_var(vmap_block_queue); | |
759 | vb->vbq = vbq; | |
760 | spin_lock(&vbq->lock); | |
761 | list_add(&vb->free_list, &vbq->free); | |
762 | spin_unlock(&vbq->lock); | |
763 | put_cpu_var(vmap_cpu_blocks); | |
764 | ||
765 | return vb; | |
766 | } | |
767 | ||
768 | static void rcu_free_vb(struct rcu_head *head) | |
769 | { | |
770 | struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head); | |
771 | ||
772 | kfree(vb); | |
773 | } | |
774 | ||
775 | static void free_vmap_block(struct vmap_block *vb) | |
776 | { | |
777 | struct vmap_block *tmp; | |
778 | unsigned long vb_idx; | |
779 | ||
d086817d | 780 | BUG_ON(!list_empty(&vb->free_list)); |
db64fe02 NP |
781 | |
782 | vb_idx = addr_to_vb_idx(vb->va->va_start); | |
783 | spin_lock(&vmap_block_tree_lock); | |
784 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
785 | spin_unlock(&vmap_block_tree_lock); | |
786 | BUG_ON(tmp != vb); | |
787 | ||
b29acbdc | 788 | free_unmap_vmap_area_noflush(vb->va); |
db64fe02 NP |
789 | call_rcu(&vb->rcu_head, rcu_free_vb); |
790 | } | |
791 | ||
792 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) | |
793 | { | |
794 | struct vmap_block_queue *vbq; | |
795 | struct vmap_block *vb; | |
796 | unsigned long addr = 0; | |
797 | unsigned int order; | |
798 | ||
799 | BUG_ON(size & ~PAGE_MASK); | |
800 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
801 | order = get_order(size); | |
802 | ||
803 | again: | |
804 | rcu_read_lock(); | |
805 | vbq = &get_cpu_var(vmap_block_queue); | |
806 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
807 | int i; | |
808 | ||
809 | spin_lock(&vb->lock); | |
810 | i = bitmap_find_free_region(vb->alloc_map, | |
811 | VMAP_BBMAP_BITS, order); | |
812 | ||
813 | if (i >= 0) { | |
814 | addr = vb->va->va_start + (i << PAGE_SHIFT); | |
815 | BUG_ON(addr_to_vb_idx(addr) != | |
816 | addr_to_vb_idx(vb->va->va_start)); | |
817 | vb->free -= 1UL << order; | |
818 | if (vb->free == 0) { | |
819 | spin_lock(&vbq->lock); | |
820 | list_del_init(&vb->free_list); | |
821 | spin_unlock(&vbq->lock); | |
822 | } | |
823 | spin_unlock(&vb->lock); | |
824 | break; | |
825 | } | |
826 | spin_unlock(&vb->lock); | |
827 | } | |
828 | put_cpu_var(vmap_cpu_blocks); | |
829 | rcu_read_unlock(); | |
830 | ||
831 | if (!addr) { | |
832 | vb = new_vmap_block(gfp_mask); | |
833 | if (IS_ERR(vb)) | |
834 | return vb; | |
835 | goto again; | |
836 | } | |
837 | ||
838 | return (void *)addr; | |
839 | } | |
840 | ||
841 | static void vb_free(const void *addr, unsigned long size) | |
842 | { | |
843 | unsigned long offset; | |
844 | unsigned long vb_idx; | |
845 | unsigned int order; | |
846 | struct vmap_block *vb; | |
847 | ||
848 | BUG_ON(size & ~PAGE_MASK); | |
849 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
b29acbdc NP |
850 | |
851 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
852 | ||
db64fe02 NP |
853 | order = get_order(size); |
854 | ||
855 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
856 | ||
857 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
858 | rcu_read_lock(); | |
859 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
860 | rcu_read_unlock(); | |
861 | BUG_ON(!vb); | |
862 | ||
863 | spin_lock(&vb->lock); | |
864 | bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order); | |
d086817d | 865 | |
db64fe02 NP |
866 | vb->dirty += 1UL << order; |
867 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
868 | BUG_ON(vb->free || !list_empty(&vb->free_list)); | |
869 | spin_unlock(&vb->lock); | |
870 | free_vmap_block(vb); | |
871 | } else | |
872 | spin_unlock(&vb->lock); | |
873 | } | |
874 | ||
875 | /** | |
876 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
877 | * | |
878 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
879 | * to amortize TLB flushing overheads. What this means is that any page you | |
880 | * have now, may, in a former life, have been mapped into kernel virtual | |
881 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
882 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
883 | * | |
884 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
885 | * be sure that none of the pages we have control over will have any aliases | |
886 | * from the vmap layer. | |
887 | */ | |
888 | void vm_unmap_aliases(void) | |
889 | { | |
890 | unsigned long start = ULONG_MAX, end = 0; | |
891 | int cpu; | |
892 | int flush = 0; | |
893 | ||
9b463334 JF |
894 | if (unlikely(!vmap_initialized)) |
895 | return; | |
896 | ||
db64fe02 NP |
897 | for_each_possible_cpu(cpu) { |
898 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
899 | struct vmap_block *vb; | |
900 | ||
901 | rcu_read_lock(); | |
902 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
903 | int i; | |
904 | ||
905 | spin_lock(&vb->lock); | |
906 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
907 | while (i < VMAP_BBMAP_BITS) { | |
908 | unsigned long s, e; | |
909 | int j; | |
910 | j = find_next_zero_bit(vb->dirty_map, | |
911 | VMAP_BBMAP_BITS, i); | |
912 | ||
913 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
914 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
915 | vunmap_page_range(s, e); | |
916 | flush = 1; | |
917 | ||
918 | if (s < start) | |
919 | start = s; | |
920 | if (e > end) | |
921 | end = e; | |
922 | ||
923 | i = j; | |
924 | i = find_next_bit(vb->dirty_map, | |
925 | VMAP_BBMAP_BITS, i); | |
926 | } | |
927 | spin_unlock(&vb->lock); | |
928 | } | |
929 | rcu_read_unlock(); | |
930 | } | |
931 | ||
932 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
933 | } | |
934 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
935 | ||
936 | /** | |
937 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
938 | * @mem: the pointer returned by vm_map_ram | |
939 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
940 | */ | |
941 | void vm_unmap_ram(const void *mem, unsigned int count) | |
942 | { | |
943 | unsigned long size = count << PAGE_SHIFT; | |
944 | unsigned long addr = (unsigned long)mem; | |
945 | ||
946 | BUG_ON(!addr); | |
947 | BUG_ON(addr < VMALLOC_START); | |
948 | BUG_ON(addr > VMALLOC_END); | |
949 | BUG_ON(addr & (PAGE_SIZE-1)); | |
950 | ||
951 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 952 | vmap_debug_free_range(addr, addr+size); |
db64fe02 NP |
953 | |
954 | if (likely(count <= VMAP_MAX_ALLOC)) | |
955 | vb_free(mem, size); | |
956 | else | |
957 | free_unmap_vmap_area_addr(addr); | |
958 | } | |
959 | EXPORT_SYMBOL(vm_unmap_ram); | |
960 | ||
961 | /** | |
962 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
963 | * @pages: an array of pointers to the pages to be mapped | |
964 | * @count: number of pages | |
965 | * @node: prefer to allocate data structures on this node | |
966 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad RD |
967 | * |
968 | * Returns: a pointer to the address that has been mapped, or %NULL on failure | |
db64fe02 NP |
969 | */ |
970 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
971 | { | |
972 | unsigned long size = count << PAGE_SHIFT; | |
973 | unsigned long addr; | |
974 | void *mem; | |
975 | ||
976 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
977 | mem = vb_alloc(size, GFP_KERNEL); | |
978 | if (IS_ERR(mem)) | |
979 | return NULL; | |
980 | addr = (unsigned long)mem; | |
981 | } else { | |
982 | struct vmap_area *va; | |
983 | va = alloc_vmap_area(size, PAGE_SIZE, | |
984 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
985 | if (IS_ERR(va)) | |
986 | return NULL; | |
987 | ||
988 | addr = va->va_start; | |
989 | mem = (void *)addr; | |
990 | } | |
991 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
992 | vm_unmap_ram(mem, count); | |
993 | return NULL; | |
994 | } | |
995 | return mem; | |
996 | } | |
997 | EXPORT_SYMBOL(vm_map_ram); | |
998 | ||
f0aa6617 TH |
999 | /** |
1000 | * vm_area_register_early - register vmap area early during boot | |
1001 | * @vm: vm_struct to register | |
c0c0a293 | 1002 | * @align: requested alignment |
f0aa6617 TH |
1003 | * |
1004 | * This function is used to register kernel vm area before | |
1005 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1006 | * proper values on entry and other fields should be zero. On return, | |
1007 | * vm->addr contains the allocated address. | |
1008 | * | |
1009 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1010 | */ | |
c0c0a293 | 1011 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1012 | { |
1013 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1014 | unsigned long addr; |
1015 | ||
1016 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1017 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1018 | |
c0c0a293 | 1019 | vm->addr = (void *)addr; |
f0aa6617 TH |
1020 | |
1021 | vm->next = vmlist; | |
1022 | vmlist = vm; | |
1023 | } | |
1024 | ||
db64fe02 NP |
1025 | void __init vmalloc_init(void) |
1026 | { | |
822c18f2 IK |
1027 | struct vmap_area *va; |
1028 | struct vm_struct *tmp; | |
db64fe02 NP |
1029 | int i; |
1030 | ||
1031 | for_each_possible_cpu(i) { | |
1032 | struct vmap_block_queue *vbq; | |
1033 | ||
1034 | vbq = &per_cpu(vmap_block_queue, i); | |
1035 | spin_lock_init(&vbq->lock); | |
1036 | INIT_LIST_HEAD(&vbq->free); | |
1037 | INIT_LIST_HEAD(&vbq->dirty); | |
1038 | vbq->nr_dirty = 0; | |
1039 | } | |
9b463334 | 1040 | |
822c18f2 IK |
1041 | /* Import existing vmlist entries. */ |
1042 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1043 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
822c18f2 IK |
1044 | va->flags = tmp->flags | VM_VM_AREA; |
1045 | va->va_start = (unsigned long)tmp->addr; | |
1046 | va->va_end = va->va_start + tmp->size; | |
1047 | __insert_vmap_area(va); | |
1048 | } | |
ca23e405 TH |
1049 | |
1050 | vmap_area_pcpu_hole = VMALLOC_END; | |
1051 | ||
9b463334 | 1052 | vmap_initialized = true; |
db64fe02 NP |
1053 | } |
1054 | ||
8fc48985 TH |
1055 | /** |
1056 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1057 | * @addr: start of the VM area to map | |
1058 | * @size: size of the VM area to map | |
1059 | * @prot: page protection flags to use | |
1060 | * @pages: pages to map | |
1061 | * | |
1062 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1063 | * specify should have been allocated using get_vm_area() and its | |
1064 | * friends. | |
1065 | * | |
1066 | * NOTE: | |
1067 | * This function does NOT do any cache flushing. The caller is | |
1068 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1069 | * before calling this function. | |
1070 | * | |
1071 | * RETURNS: | |
1072 | * The number of pages mapped on success, -errno on failure. | |
1073 | */ | |
1074 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1075 | pgprot_t prot, struct page **pages) | |
1076 | { | |
1077 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1078 | } | |
1079 | ||
1080 | /** | |
1081 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1082 | * @addr: start of the VM area to unmap | |
1083 | * @size: size of the VM area to unmap | |
1084 | * | |
1085 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1086 | * specify should have been allocated using get_vm_area() and its | |
1087 | * friends. | |
1088 | * | |
1089 | * NOTE: | |
1090 | * This function does NOT do any cache flushing. The caller is | |
1091 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1092 | * before calling this function and flush_tlb_kernel_range() after. | |
1093 | */ | |
1094 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1095 | { | |
1096 | vunmap_page_range(addr, addr + size); | |
1097 | } | |
1098 | ||
1099 | /** | |
1100 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1101 | * @addr: start of the VM area to unmap | |
1102 | * @size: size of the VM area to unmap | |
1103 | * | |
1104 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1105 | * the unmapping and tlb after. | |
1106 | */ | |
db64fe02 NP |
1107 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1108 | { | |
1109 | unsigned long end = addr + size; | |
f6fcba70 TH |
1110 | |
1111 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1112 | vunmap_page_range(addr, end); |
1113 | flush_tlb_kernel_range(addr, end); | |
1114 | } | |
1115 | ||
1116 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
1117 | { | |
1118 | unsigned long addr = (unsigned long)area->addr; | |
1119 | unsigned long end = addr + area->size - PAGE_SIZE; | |
1120 | int err; | |
1121 | ||
1122 | err = vmap_page_range(addr, end, prot, *pages); | |
1123 | if (err > 0) { | |
1124 | *pages += err; | |
1125 | err = 0; | |
1126 | } | |
1127 | ||
1128 | return err; | |
1129 | } | |
1130 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1131 | ||
1132 | /*** Old vmalloc interfaces ***/ | |
1133 | DEFINE_RWLOCK(vmlist_lock); | |
1134 | struct vm_struct *vmlist; | |
1135 | ||
cf88c790 TH |
1136 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
1137 | unsigned long flags, void *caller) | |
1138 | { | |
1139 | struct vm_struct *tmp, **p; | |
1140 | ||
1141 | vm->flags = flags; | |
1142 | vm->addr = (void *)va->va_start; | |
1143 | vm->size = va->va_end - va->va_start; | |
1144 | vm->caller = caller; | |
1145 | va->private = vm; | |
1146 | va->flags |= VM_VM_AREA; | |
1147 | ||
1148 | write_lock(&vmlist_lock); | |
1149 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1150 | if (tmp->addr >= vm->addr) | |
1151 | break; | |
1152 | } | |
1153 | vm->next = *p; | |
1154 | *p = vm; | |
1155 | write_unlock(&vmlist_lock); | |
1156 | } | |
1157 | ||
db64fe02 NP |
1158 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
1159 | unsigned long flags, unsigned long start, unsigned long end, | |
1160 | int node, gfp_t gfp_mask, void *caller) | |
1161 | { | |
1162 | static struct vmap_area *va; | |
1163 | struct vm_struct *area; | |
db64fe02 | 1164 | unsigned long align = 1; |
1da177e4 | 1165 | |
52fd24ca | 1166 | BUG_ON(in_interrupt()); |
1da177e4 LT |
1167 | if (flags & VM_IOREMAP) { |
1168 | int bit = fls(size); | |
1169 | ||
1170 | if (bit > IOREMAP_MAX_ORDER) | |
1171 | bit = IOREMAP_MAX_ORDER; | |
1172 | else if (bit < PAGE_SHIFT) | |
1173 | bit = PAGE_SHIFT; | |
1174 | ||
1175 | align = 1ul << bit; | |
1176 | } | |
db64fe02 | 1177 | |
1da177e4 | 1178 | size = PAGE_ALIGN(size); |
31be8309 OH |
1179 | if (unlikely(!size)) |
1180 | return NULL; | |
1da177e4 | 1181 | |
cf88c790 | 1182 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1183 | if (unlikely(!area)) |
1184 | return NULL; | |
1185 | ||
1da177e4 LT |
1186 | /* |
1187 | * We always allocate a guard page. | |
1188 | */ | |
1189 | size += PAGE_SIZE; | |
1190 | ||
db64fe02 NP |
1191 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1192 | if (IS_ERR(va)) { | |
1193 | kfree(area); | |
1194 | return NULL; | |
1da177e4 | 1195 | } |
1da177e4 | 1196 | |
cf88c790 | 1197 | insert_vmalloc_vm(area, va, flags, caller); |
1da177e4 | 1198 | return area; |
1da177e4 LT |
1199 | } |
1200 | ||
930fc45a CL |
1201 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1202 | unsigned long start, unsigned long end) | |
1203 | { | |
23016969 CL |
1204 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, |
1205 | __builtin_return_address(0)); | |
930fc45a | 1206 | } |
5992b6da | 1207 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1208 | |
c2968612 BH |
1209 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1210 | unsigned long start, unsigned long end, | |
1211 | void *caller) | |
1212 | { | |
1213 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, | |
1214 | caller); | |
1215 | } | |
1216 | ||
1da177e4 | 1217 | /** |
183ff22b | 1218 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1219 | * @size: size of the area |
1220 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1221 | * | |
1222 | * Search an area of @size in the kernel virtual mapping area, | |
1223 | * and reserved it for out purposes. Returns the area descriptor | |
1224 | * on success or %NULL on failure. | |
1225 | */ | |
1226 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1227 | { | |
23016969 CL |
1228 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, |
1229 | -1, GFP_KERNEL, __builtin_return_address(0)); | |
1230 | } | |
1231 | ||
1232 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
1233 | void *caller) | |
1234 | { | |
1235 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, | |
1236 | -1, GFP_KERNEL, caller); | |
1da177e4 LT |
1237 | } |
1238 | ||
52fd24ca GP |
1239 | struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, |
1240 | int node, gfp_t gfp_mask) | |
930fc45a | 1241 | { |
52fd24ca | 1242 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node, |
23016969 | 1243 | gfp_mask, __builtin_return_address(0)); |
930fc45a CL |
1244 | } |
1245 | ||
db64fe02 | 1246 | static struct vm_struct *find_vm_area(const void *addr) |
83342314 | 1247 | { |
db64fe02 | 1248 | struct vmap_area *va; |
83342314 | 1249 | |
db64fe02 NP |
1250 | va = find_vmap_area((unsigned long)addr); |
1251 | if (va && va->flags & VM_VM_AREA) | |
1252 | return va->private; | |
1da177e4 | 1253 | |
1da177e4 | 1254 | return NULL; |
1da177e4 LT |
1255 | } |
1256 | ||
7856dfeb | 1257 | /** |
183ff22b | 1258 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1259 | * @addr: base address |
1260 | * | |
1261 | * Search for the kernel VM area starting at @addr, and remove it. | |
1262 | * This function returns the found VM area, but using it is NOT safe | |
1263 | * on SMP machines, except for its size or flags. | |
1264 | */ | |
b3bdda02 | 1265 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1266 | { |
db64fe02 NP |
1267 | struct vmap_area *va; |
1268 | ||
1269 | va = find_vmap_area((unsigned long)addr); | |
1270 | if (va && va->flags & VM_VM_AREA) { | |
1271 | struct vm_struct *vm = va->private; | |
1272 | struct vm_struct *tmp, **p; | |
cd52858c NP |
1273 | |
1274 | vmap_debug_free_range(va->va_start, va->va_end); | |
db64fe02 NP |
1275 | free_unmap_vmap_area(va); |
1276 | vm->size -= PAGE_SIZE; | |
1277 | ||
1278 | write_lock(&vmlist_lock); | |
1279 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) | |
1280 | ; | |
1281 | *p = tmp->next; | |
1282 | write_unlock(&vmlist_lock); | |
1283 | ||
1284 | return vm; | |
1285 | } | |
1286 | return NULL; | |
7856dfeb AK |
1287 | } |
1288 | ||
b3bdda02 | 1289 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1290 | { |
1291 | struct vm_struct *area; | |
1292 | ||
1293 | if (!addr) | |
1294 | return; | |
1295 | ||
1296 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1297 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1298 | return; |
1299 | } | |
1300 | ||
1301 | area = remove_vm_area(addr); | |
1302 | if (unlikely(!area)) { | |
4c8573e2 | 1303 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1304 | addr); |
1da177e4 LT |
1305 | return; |
1306 | } | |
1307 | ||
9a11b49a | 1308 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1309 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1310 | |
1da177e4 LT |
1311 | if (deallocate_pages) { |
1312 | int i; | |
1313 | ||
1314 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1315 | struct page *page = area->pages[i]; |
1316 | ||
1317 | BUG_ON(!page); | |
1318 | __free_page(page); | |
1da177e4 LT |
1319 | } |
1320 | ||
8757d5fa | 1321 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1322 | vfree(area->pages); |
1323 | else | |
1324 | kfree(area->pages); | |
1325 | } | |
1326 | ||
1327 | kfree(area); | |
1328 | return; | |
1329 | } | |
1330 | ||
1331 | /** | |
1332 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1333 | * @addr: memory base address |
1334 | * | |
183ff22b | 1335 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1336 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1337 | * NULL, no operation is performed. | |
1da177e4 | 1338 | * |
80e93eff | 1339 | * Must not be called in interrupt context. |
1da177e4 | 1340 | */ |
b3bdda02 | 1341 | void vfree(const void *addr) |
1da177e4 LT |
1342 | { |
1343 | BUG_ON(in_interrupt()); | |
89219d37 CM |
1344 | |
1345 | kmemleak_free(addr); | |
1346 | ||
1da177e4 LT |
1347 | __vunmap(addr, 1); |
1348 | } | |
1da177e4 LT |
1349 | EXPORT_SYMBOL(vfree); |
1350 | ||
1351 | /** | |
1352 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1353 | * @addr: memory base address |
1354 | * | |
1355 | * Free the virtually contiguous memory area starting at @addr, | |
1356 | * which was created from the page array passed to vmap(). | |
1357 | * | |
80e93eff | 1358 | * Must not be called in interrupt context. |
1da177e4 | 1359 | */ |
b3bdda02 | 1360 | void vunmap(const void *addr) |
1da177e4 LT |
1361 | { |
1362 | BUG_ON(in_interrupt()); | |
34754b69 | 1363 | might_sleep(); |
1da177e4 LT |
1364 | __vunmap(addr, 0); |
1365 | } | |
1da177e4 LT |
1366 | EXPORT_SYMBOL(vunmap); |
1367 | ||
1368 | /** | |
1369 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1370 | * @pages: array of page pointers |
1371 | * @count: number of pages to map | |
1372 | * @flags: vm_area->flags | |
1373 | * @prot: page protection for the mapping | |
1374 | * | |
1375 | * Maps @count pages from @pages into contiguous kernel virtual | |
1376 | * space. | |
1377 | */ | |
1378 | void *vmap(struct page **pages, unsigned int count, | |
1379 | unsigned long flags, pgprot_t prot) | |
1380 | { | |
1381 | struct vm_struct *area; | |
1382 | ||
34754b69 PZ |
1383 | might_sleep(); |
1384 | ||
1da177e4 LT |
1385 | if (count > num_physpages) |
1386 | return NULL; | |
1387 | ||
23016969 CL |
1388 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1389 | __builtin_return_address(0)); | |
1da177e4 LT |
1390 | if (!area) |
1391 | return NULL; | |
23016969 | 1392 | |
1da177e4 LT |
1393 | if (map_vm_area(area, prot, &pages)) { |
1394 | vunmap(area->addr); | |
1395 | return NULL; | |
1396 | } | |
1397 | ||
1398 | return area->addr; | |
1399 | } | |
1da177e4 LT |
1400 | EXPORT_SYMBOL(vmap); |
1401 | ||
db64fe02 NP |
1402 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
1403 | int node, void *caller); | |
e31d9eb5 | 1404 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 1405 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
1406 | { |
1407 | struct page **pages; | |
1408 | unsigned int nr_pages, array_size, i; | |
1409 | ||
1410 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1411 | array_size = (nr_pages * sizeof(struct page *)); | |
1412 | ||
1413 | area->nr_pages = nr_pages; | |
1414 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1415 | if (array_size > PAGE_SIZE) { |
94f6030c | 1416 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
23016969 | 1417 | PAGE_KERNEL, node, caller); |
8757d5fa | 1418 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
1419 | } else { |
1420 | pages = kmalloc_node(array_size, | |
6cb06229 | 1421 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
1422 | node); |
1423 | } | |
1da177e4 | 1424 | area->pages = pages; |
23016969 | 1425 | area->caller = caller; |
1da177e4 LT |
1426 | if (!area->pages) { |
1427 | remove_vm_area(area->addr); | |
1428 | kfree(area); | |
1429 | return NULL; | |
1430 | } | |
1da177e4 LT |
1431 | |
1432 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1433 | struct page *page; |
1434 | ||
930fc45a | 1435 | if (node < 0) |
bf53d6f8 | 1436 | page = alloc_page(gfp_mask); |
930fc45a | 1437 | else |
bf53d6f8 CL |
1438 | page = alloc_pages_node(node, gfp_mask, 0); |
1439 | ||
1440 | if (unlikely(!page)) { | |
1da177e4 LT |
1441 | /* Successfully allocated i pages, free them in __vunmap() */ |
1442 | area->nr_pages = i; | |
1443 | goto fail; | |
1444 | } | |
bf53d6f8 | 1445 | area->pages[i] = page; |
1da177e4 LT |
1446 | } |
1447 | ||
1448 | if (map_vm_area(area, prot, &pages)) | |
1449 | goto fail; | |
1450 | return area->addr; | |
1451 | ||
1452 | fail: | |
1453 | vfree(area->addr); | |
1454 | return NULL; | |
1455 | } | |
1456 | ||
930fc45a CL |
1457 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
1458 | { | |
89219d37 CM |
1459 | void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1, |
1460 | __builtin_return_address(0)); | |
1461 | ||
1462 | /* | |
1463 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1464 | * structures allocated in the __get_vm_area_node() function contain | |
1465 | * references to the virtual address of the vmalloc'ed block. | |
1466 | */ | |
1467 | kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask); | |
1468 | ||
1469 | return addr; | |
930fc45a CL |
1470 | } |
1471 | ||
1da177e4 | 1472 | /** |
930fc45a | 1473 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
1474 | * @size: allocation size |
1475 | * @gfp_mask: flags for the page level allocator | |
1476 | * @prot: protection mask for the allocated pages | |
d44e0780 | 1477 | * @node: node to use for allocation or -1 |
c85d194b | 1478 | * @caller: caller's return address |
1da177e4 LT |
1479 | * |
1480 | * Allocate enough pages to cover @size from the page level | |
1481 | * allocator with @gfp_mask flags. Map them into contiguous | |
1482 | * kernel virtual space, using a pagetable protection of @prot. | |
1483 | */ | |
b221385b | 1484 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 1485 | int node, void *caller) |
1da177e4 LT |
1486 | { |
1487 | struct vm_struct *area; | |
89219d37 CM |
1488 | void *addr; |
1489 | unsigned long real_size = size; | |
1da177e4 LT |
1490 | |
1491 | size = PAGE_ALIGN(size); | |
1492 | if (!size || (size >> PAGE_SHIFT) > num_physpages) | |
1493 | return NULL; | |
1494 | ||
23016969 CL |
1495 | area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END, |
1496 | node, gfp_mask, caller); | |
1497 | ||
1da177e4 LT |
1498 | if (!area) |
1499 | return NULL; | |
1500 | ||
89219d37 CM |
1501 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1502 | ||
1503 | /* | |
1504 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1505 | * structures allocated in the __get_vm_area_node() function contain | |
1506 | * references to the virtual address of the vmalloc'ed block. | |
1507 | */ | |
1508 | kmemleak_alloc(addr, real_size, 3, gfp_mask); | |
1509 | ||
1510 | return addr; | |
1da177e4 LT |
1511 | } |
1512 | ||
930fc45a CL |
1513 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1514 | { | |
23016969 CL |
1515 | return __vmalloc_node(size, gfp_mask, prot, -1, |
1516 | __builtin_return_address(0)); | |
930fc45a | 1517 | } |
1da177e4 LT |
1518 | EXPORT_SYMBOL(__vmalloc); |
1519 | ||
1520 | /** | |
1521 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1522 | * @size: allocation size |
1da177e4 LT |
1523 | * Allocate enough pages to cover @size from the page level |
1524 | * allocator and map them into contiguous kernel virtual space. | |
1525 | * | |
c1c8897f | 1526 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1527 | * use __vmalloc() instead. |
1528 | */ | |
1529 | void *vmalloc(unsigned long size) | |
1530 | { | |
23016969 CL |
1531 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1532 | -1, __builtin_return_address(0)); | |
1da177e4 | 1533 | } |
1da177e4 LT |
1534 | EXPORT_SYMBOL(vmalloc); |
1535 | ||
83342314 | 1536 | /** |
ead04089 REB |
1537 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1538 | * @size: allocation size | |
83342314 | 1539 | * |
ead04089 REB |
1540 | * The resulting memory area is zeroed so it can be mapped to userspace |
1541 | * without leaking data. | |
83342314 NP |
1542 | */ |
1543 | void *vmalloc_user(unsigned long size) | |
1544 | { | |
1545 | struct vm_struct *area; | |
1546 | void *ret; | |
1547 | ||
84877848 GC |
1548 | ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
1549 | PAGE_KERNEL, -1, __builtin_return_address(0)); | |
2b4ac44e | 1550 | if (ret) { |
db64fe02 | 1551 | area = find_vm_area(ret); |
2b4ac44e | 1552 | area->flags |= VM_USERMAP; |
2b4ac44e | 1553 | } |
83342314 NP |
1554 | return ret; |
1555 | } | |
1556 | EXPORT_SYMBOL(vmalloc_user); | |
1557 | ||
930fc45a CL |
1558 | /** |
1559 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1560 | * @size: allocation size |
d44e0780 | 1561 | * @node: numa node |
930fc45a CL |
1562 | * |
1563 | * Allocate enough pages to cover @size from the page level | |
1564 | * allocator and map them into contiguous kernel virtual space. | |
1565 | * | |
c1c8897f | 1566 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1567 | * use __vmalloc() instead. |
1568 | */ | |
1569 | void *vmalloc_node(unsigned long size, int node) | |
1570 | { | |
23016969 CL |
1571 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1572 | node, __builtin_return_address(0)); | |
930fc45a CL |
1573 | } |
1574 | EXPORT_SYMBOL(vmalloc_node); | |
1575 | ||
4dc3b16b PP |
1576 | #ifndef PAGE_KERNEL_EXEC |
1577 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1578 | #endif | |
1579 | ||
1da177e4 LT |
1580 | /** |
1581 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1582 | * @size: allocation size |
1583 | * | |
1584 | * Kernel-internal function to allocate enough pages to cover @size | |
1585 | * the page level allocator and map them into contiguous and | |
1586 | * executable kernel virtual space. | |
1587 | * | |
c1c8897f | 1588 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1589 | * use __vmalloc() instead. |
1590 | */ | |
1591 | ||
1da177e4 LT |
1592 | void *vmalloc_exec(unsigned long size) |
1593 | { | |
84877848 GC |
1594 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
1595 | -1, __builtin_return_address(0)); | |
1da177e4 LT |
1596 | } |
1597 | ||
0d08e0d3 | 1598 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1599 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1600 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1601 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1602 | #else |
1603 | #define GFP_VMALLOC32 GFP_KERNEL | |
1604 | #endif | |
1605 | ||
1da177e4 LT |
1606 | /** |
1607 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1608 | * @size: allocation size |
1609 | * | |
1610 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1611 | * page level allocator and map them into contiguous kernel virtual space. | |
1612 | */ | |
1613 | void *vmalloc_32(unsigned long size) | |
1614 | { | |
84877848 GC |
1615 | return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL, |
1616 | -1, __builtin_return_address(0)); | |
1da177e4 | 1617 | } |
1da177e4 LT |
1618 | EXPORT_SYMBOL(vmalloc_32); |
1619 | ||
83342314 | 1620 | /** |
ead04089 | 1621 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1622 | * @size: allocation size |
ead04089 REB |
1623 | * |
1624 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1625 | * mapped to userspace without leaking data. | |
83342314 NP |
1626 | */ |
1627 | void *vmalloc_32_user(unsigned long size) | |
1628 | { | |
1629 | struct vm_struct *area; | |
1630 | void *ret; | |
1631 | ||
84877848 GC |
1632 | ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
1633 | -1, __builtin_return_address(0)); | |
2b4ac44e | 1634 | if (ret) { |
db64fe02 | 1635 | area = find_vm_area(ret); |
2b4ac44e | 1636 | area->flags |= VM_USERMAP; |
2b4ac44e | 1637 | } |
83342314 NP |
1638 | return ret; |
1639 | } | |
1640 | EXPORT_SYMBOL(vmalloc_32_user); | |
1641 | ||
1da177e4 LT |
1642 | long vread(char *buf, char *addr, unsigned long count) |
1643 | { | |
1644 | struct vm_struct *tmp; | |
1645 | char *vaddr, *buf_start = buf; | |
1646 | unsigned long n; | |
1647 | ||
1648 | /* Don't allow overflow */ | |
1649 | if ((unsigned long) addr + count < count) | |
1650 | count = -(unsigned long) addr; | |
1651 | ||
1652 | read_lock(&vmlist_lock); | |
1653 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
1654 | vaddr = (char *) tmp->addr; | |
1655 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1656 | continue; | |
1657 | while (addr < vaddr) { | |
1658 | if (count == 0) | |
1659 | goto finished; | |
1660 | *buf = '\0'; | |
1661 | buf++; | |
1662 | addr++; | |
1663 | count--; | |
1664 | } | |
1665 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
1666 | do { | |
1667 | if (count == 0) | |
1668 | goto finished; | |
1669 | *buf = *addr; | |
1670 | buf++; | |
1671 | addr++; | |
1672 | count--; | |
1673 | } while (--n > 0); | |
1674 | } | |
1675 | finished: | |
1676 | read_unlock(&vmlist_lock); | |
1677 | return buf - buf_start; | |
1678 | } | |
1679 | ||
1680 | long vwrite(char *buf, char *addr, unsigned long count) | |
1681 | { | |
1682 | struct vm_struct *tmp; | |
1683 | char *vaddr, *buf_start = buf; | |
1684 | unsigned long n; | |
1685 | ||
1686 | /* Don't allow overflow */ | |
1687 | if ((unsigned long) addr + count < count) | |
1688 | count = -(unsigned long) addr; | |
1689 | ||
1690 | read_lock(&vmlist_lock); | |
1691 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
1692 | vaddr = (char *) tmp->addr; | |
1693 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1694 | continue; | |
1695 | while (addr < vaddr) { | |
1696 | if (count == 0) | |
1697 | goto finished; | |
1698 | buf++; | |
1699 | addr++; | |
1700 | count--; | |
1701 | } | |
1702 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
1703 | do { | |
1704 | if (count == 0) | |
1705 | goto finished; | |
1706 | *addr = *buf; | |
1707 | buf++; | |
1708 | addr++; | |
1709 | count--; | |
1710 | } while (--n > 0); | |
1711 | } | |
1712 | finished: | |
1713 | read_unlock(&vmlist_lock); | |
1714 | return buf - buf_start; | |
1715 | } | |
83342314 NP |
1716 | |
1717 | /** | |
1718 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
1719 | * @vma: vma to cover (map full range of vma) |
1720 | * @addr: vmalloc memory | |
1721 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
1722 | * |
1723 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
1724 | * |
1725 | * This function checks that addr is a valid vmalloc'ed area, and | |
1726 | * that it is big enough to cover the vma. Will return failure if | |
1727 | * that criteria isn't met. | |
1728 | * | |
72fd4a35 | 1729 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
1730 | */ |
1731 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
1732 | unsigned long pgoff) | |
1733 | { | |
1734 | struct vm_struct *area; | |
1735 | unsigned long uaddr = vma->vm_start; | |
1736 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
1737 | |
1738 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
1739 | return -EINVAL; | |
1740 | ||
db64fe02 | 1741 | area = find_vm_area(addr); |
83342314 | 1742 | if (!area) |
db64fe02 | 1743 | return -EINVAL; |
83342314 NP |
1744 | |
1745 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 1746 | return -EINVAL; |
83342314 NP |
1747 | |
1748 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 1749 | return -EINVAL; |
83342314 NP |
1750 | |
1751 | addr += pgoff << PAGE_SHIFT; | |
1752 | do { | |
1753 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
1754 | int ret; |
1755 | ||
83342314 NP |
1756 | ret = vm_insert_page(vma, uaddr, page); |
1757 | if (ret) | |
1758 | return ret; | |
1759 | ||
1760 | uaddr += PAGE_SIZE; | |
1761 | addr += PAGE_SIZE; | |
1762 | usize -= PAGE_SIZE; | |
1763 | } while (usize > 0); | |
1764 | ||
1765 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
1766 | vma->vm_flags |= VM_RESERVED; | |
1767 | ||
db64fe02 | 1768 | return 0; |
83342314 NP |
1769 | } |
1770 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1771 | ||
1eeb66a1 CH |
1772 | /* |
1773 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
1774 | * have one. | |
1775 | */ | |
1776 | void __attribute__((weak)) vmalloc_sync_all(void) | |
1777 | { | |
1778 | } | |
5f4352fb JF |
1779 | |
1780 | ||
2f569afd | 1781 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
1782 | { |
1783 | /* apply_to_page_range() does all the hard work. */ | |
1784 | return 0; | |
1785 | } | |
1786 | ||
1787 | /** | |
1788 | * alloc_vm_area - allocate a range of kernel address space | |
1789 | * @size: size of the area | |
7682486b RD |
1790 | * |
1791 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
1792 | * |
1793 | * This function reserves a range of kernel address space, and | |
1794 | * allocates pagetables to map that range. No actual mappings | |
1795 | * are created. If the kernel address space is not shared | |
1796 | * between processes, it syncs the pagetable across all | |
1797 | * processes. | |
1798 | */ | |
1799 | struct vm_struct *alloc_vm_area(size_t size) | |
1800 | { | |
1801 | struct vm_struct *area; | |
1802 | ||
23016969 CL |
1803 | area = get_vm_area_caller(size, VM_IOREMAP, |
1804 | __builtin_return_address(0)); | |
5f4352fb JF |
1805 | if (area == NULL) |
1806 | return NULL; | |
1807 | ||
1808 | /* | |
1809 | * This ensures that page tables are constructed for this region | |
1810 | * of kernel virtual address space and mapped into init_mm. | |
1811 | */ | |
1812 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
1813 | area->size, f, NULL)) { | |
1814 | free_vm_area(area); | |
1815 | return NULL; | |
1816 | } | |
1817 | ||
1818 | /* Make sure the pagetables are constructed in process kernel | |
1819 | mappings */ | |
1820 | vmalloc_sync_all(); | |
1821 | ||
1822 | return area; | |
1823 | } | |
1824 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
1825 | ||
1826 | void free_vm_area(struct vm_struct *area) | |
1827 | { | |
1828 | struct vm_struct *ret; | |
1829 | ret = remove_vm_area(area->addr); | |
1830 | BUG_ON(ret != area); | |
1831 | kfree(area); | |
1832 | } | |
1833 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 1834 | |
ca23e405 TH |
1835 | static struct vmap_area *node_to_va(struct rb_node *n) |
1836 | { | |
1837 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; | |
1838 | } | |
1839 | ||
1840 | /** | |
1841 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
1842 | * @end: target address | |
1843 | * @pnext: out arg for the next vmap_area | |
1844 | * @pprev: out arg for the previous vmap_area | |
1845 | * | |
1846 | * Returns: %true if either or both of next and prev are found, | |
1847 | * %false if no vmap_area exists | |
1848 | * | |
1849 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
1850 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
1851 | */ | |
1852 | static bool pvm_find_next_prev(unsigned long end, | |
1853 | struct vmap_area **pnext, | |
1854 | struct vmap_area **pprev) | |
1855 | { | |
1856 | struct rb_node *n = vmap_area_root.rb_node; | |
1857 | struct vmap_area *va = NULL; | |
1858 | ||
1859 | while (n) { | |
1860 | va = rb_entry(n, struct vmap_area, rb_node); | |
1861 | if (end < va->va_end) | |
1862 | n = n->rb_left; | |
1863 | else if (end > va->va_end) | |
1864 | n = n->rb_right; | |
1865 | else | |
1866 | break; | |
1867 | } | |
1868 | ||
1869 | if (!va) | |
1870 | return false; | |
1871 | ||
1872 | if (va->va_end > end) { | |
1873 | *pnext = va; | |
1874 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
1875 | } else { | |
1876 | *pprev = va; | |
1877 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
1878 | } | |
1879 | return true; | |
1880 | } | |
1881 | ||
1882 | /** | |
1883 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
1884 | * @pnext: in/out arg for the next vmap_area | |
1885 | * @pprev: in/out arg for the previous vmap_area | |
1886 | * @align: alignment | |
1887 | * | |
1888 | * Returns: determined end address | |
1889 | * | |
1890 | * Find the highest aligned address between *@pnext and *@pprev below | |
1891 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
1892 | * down address is between the end addresses of the two vmap_areas. | |
1893 | * | |
1894 | * Please note that the address returned by this function may fall | |
1895 | * inside *@pnext vmap_area. The caller is responsible for checking | |
1896 | * that. | |
1897 | */ | |
1898 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
1899 | struct vmap_area **pprev, | |
1900 | unsigned long align) | |
1901 | { | |
1902 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
1903 | unsigned long addr; | |
1904 | ||
1905 | if (*pnext) | |
1906 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
1907 | else | |
1908 | addr = vmalloc_end; | |
1909 | ||
1910 | while (*pprev && (*pprev)->va_end > addr) { | |
1911 | *pnext = *pprev; | |
1912 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
1913 | } | |
1914 | ||
1915 | return addr; | |
1916 | } | |
1917 | ||
1918 | /** | |
1919 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
1920 | * @offsets: array containing offset of each area | |
1921 | * @sizes: array containing size of each area | |
1922 | * @nr_vms: the number of areas to allocate | |
1923 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
1924 | * @gfp_mask: allocation mask | |
1925 | * | |
1926 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
1927 | * vm_structs on success, %NULL on failure | |
1928 | * | |
1929 | * Percpu allocator wants to use congruent vm areas so that it can | |
1930 | * maintain the offsets among percpu areas. This function allocates | |
1931 | * congruent vmalloc areas for it. These areas tend to be scattered | |
1932 | * pretty far, distance between two areas easily going up to | |
1933 | * gigabytes. To avoid interacting with regular vmallocs, these areas | |
1934 | * are allocated from top. | |
1935 | * | |
1936 | * Despite its complicated look, this allocator is rather simple. It | |
1937 | * does everything top-down and scans areas from the end looking for | |
1938 | * matching slot. While scanning, if any of the areas overlaps with | |
1939 | * existing vmap_area, the base address is pulled down to fit the | |
1940 | * area. Scanning is repeated till all the areas fit and then all | |
1941 | * necessary data structres are inserted and the result is returned. | |
1942 | */ | |
1943 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
1944 | const size_t *sizes, int nr_vms, | |
1945 | size_t align, gfp_t gfp_mask) | |
1946 | { | |
1947 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
1948 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
1949 | struct vmap_area **vas, *prev, *next; | |
1950 | struct vm_struct **vms; | |
1951 | int area, area2, last_area, term_area; | |
1952 | unsigned long base, start, end, last_end; | |
1953 | bool purged = false; | |
1954 | ||
1955 | gfp_mask &= GFP_RECLAIM_MASK; | |
1956 | ||
1957 | /* verify parameters and allocate data structures */ | |
1958 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); | |
1959 | for (last_area = 0, area = 0; area < nr_vms; area++) { | |
1960 | start = offsets[area]; | |
1961 | end = start + sizes[area]; | |
1962 | ||
1963 | /* is everything aligned properly? */ | |
1964 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
1965 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
1966 | ||
1967 | /* detect the area with the highest address */ | |
1968 | if (start > offsets[last_area]) | |
1969 | last_area = area; | |
1970 | ||
1971 | for (area2 = 0; area2 < nr_vms; area2++) { | |
1972 | unsigned long start2 = offsets[area2]; | |
1973 | unsigned long end2 = start2 + sizes[area2]; | |
1974 | ||
1975 | if (area2 == area) | |
1976 | continue; | |
1977 | ||
1978 | BUG_ON(start2 >= start && start2 < end); | |
1979 | BUG_ON(end2 <= end && end2 > start); | |
1980 | } | |
1981 | } | |
1982 | last_end = offsets[last_area] + sizes[last_area]; | |
1983 | ||
1984 | if (vmalloc_end - vmalloc_start < last_end) { | |
1985 | WARN_ON(true); | |
1986 | return NULL; | |
1987 | } | |
1988 | ||
1989 | vms = kzalloc(sizeof(vms[0]) * nr_vms, gfp_mask); | |
1990 | vas = kzalloc(sizeof(vas[0]) * nr_vms, gfp_mask); | |
1991 | if (!vas || !vms) | |
1992 | goto err_free; | |
1993 | ||
1994 | for (area = 0; area < nr_vms; area++) { | |
1995 | vas[area] = kzalloc(sizeof(struct vmap_area), gfp_mask); | |
1996 | vms[area] = kzalloc(sizeof(struct vm_struct), gfp_mask); | |
1997 | if (!vas[area] || !vms[area]) | |
1998 | goto err_free; | |
1999 | } | |
2000 | retry: | |
2001 | spin_lock(&vmap_area_lock); | |
2002 | ||
2003 | /* start scanning - we scan from the top, begin with the last area */ | |
2004 | area = term_area = last_area; | |
2005 | start = offsets[area]; | |
2006 | end = start + sizes[area]; | |
2007 | ||
2008 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2009 | base = vmalloc_end - last_end; | |
2010 | goto found; | |
2011 | } | |
2012 | base = pvm_determine_end(&next, &prev, align) - end; | |
2013 | ||
2014 | while (true) { | |
2015 | BUG_ON(next && next->va_end <= base + end); | |
2016 | BUG_ON(prev && prev->va_end > base + end); | |
2017 | ||
2018 | /* | |
2019 | * base might have underflowed, add last_end before | |
2020 | * comparing. | |
2021 | */ | |
2022 | if (base + last_end < vmalloc_start + last_end) { | |
2023 | spin_unlock(&vmap_area_lock); | |
2024 | if (!purged) { | |
2025 | purge_vmap_area_lazy(); | |
2026 | purged = true; | |
2027 | goto retry; | |
2028 | } | |
2029 | goto err_free; | |
2030 | } | |
2031 | ||
2032 | /* | |
2033 | * If next overlaps, move base downwards so that it's | |
2034 | * right below next and then recheck. | |
2035 | */ | |
2036 | if (next && next->va_start < base + end) { | |
2037 | base = pvm_determine_end(&next, &prev, align) - end; | |
2038 | term_area = area; | |
2039 | continue; | |
2040 | } | |
2041 | ||
2042 | /* | |
2043 | * If prev overlaps, shift down next and prev and move | |
2044 | * base so that it's right below new next and then | |
2045 | * recheck. | |
2046 | */ | |
2047 | if (prev && prev->va_end > base + start) { | |
2048 | next = prev; | |
2049 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2050 | base = pvm_determine_end(&next, &prev, align) - end; | |
2051 | term_area = area; | |
2052 | continue; | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * This area fits, move on to the previous one. If | |
2057 | * the previous one is the terminal one, we're done. | |
2058 | */ | |
2059 | area = (area + nr_vms - 1) % nr_vms; | |
2060 | if (area == term_area) | |
2061 | break; | |
2062 | start = offsets[area]; | |
2063 | end = start + sizes[area]; | |
2064 | pvm_find_next_prev(base + end, &next, &prev); | |
2065 | } | |
2066 | found: | |
2067 | /* we've found a fitting base, insert all va's */ | |
2068 | for (area = 0; area < nr_vms; area++) { | |
2069 | struct vmap_area *va = vas[area]; | |
2070 | ||
2071 | va->va_start = base + offsets[area]; | |
2072 | va->va_end = va->va_start + sizes[area]; | |
2073 | __insert_vmap_area(va); | |
2074 | } | |
2075 | ||
2076 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2077 | ||
2078 | spin_unlock(&vmap_area_lock); | |
2079 | ||
2080 | /* insert all vm's */ | |
2081 | for (area = 0; area < nr_vms; area++) | |
2082 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, | |
2083 | pcpu_get_vm_areas); | |
2084 | ||
2085 | kfree(vas); | |
2086 | return vms; | |
2087 | ||
2088 | err_free: | |
2089 | for (area = 0; area < nr_vms; area++) { | |
2090 | if (vas) | |
2091 | kfree(vas[area]); | |
2092 | if (vms) | |
2093 | kfree(vms[area]); | |
2094 | } | |
2095 | kfree(vas); | |
2096 | kfree(vms); | |
2097 | return NULL; | |
2098 | } | |
2099 | ||
2100 | /** | |
2101 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2102 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2103 | * @nr_vms: the number of allocated areas | |
2104 | * | |
2105 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2106 | */ | |
2107 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2108 | { | |
2109 | int i; | |
2110 | ||
2111 | for (i = 0; i < nr_vms; i++) | |
2112 | free_vm_area(vms[i]); | |
2113 | kfree(vms); | |
2114 | } | |
a10aa579 CL |
2115 | |
2116 | #ifdef CONFIG_PROC_FS | |
2117 | static void *s_start(struct seq_file *m, loff_t *pos) | |
2118 | { | |
2119 | loff_t n = *pos; | |
2120 | struct vm_struct *v; | |
2121 | ||
2122 | read_lock(&vmlist_lock); | |
2123 | v = vmlist; | |
2124 | while (n > 0 && v) { | |
2125 | n--; | |
2126 | v = v->next; | |
2127 | } | |
2128 | if (!n) | |
2129 | return v; | |
2130 | ||
2131 | return NULL; | |
2132 | ||
2133 | } | |
2134 | ||
2135 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2136 | { | |
2137 | struct vm_struct *v = p; | |
2138 | ||
2139 | ++*pos; | |
2140 | return v->next; | |
2141 | } | |
2142 | ||
2143 | static void s_stop(struct seq_file *m, void *p) | |
2144 | { | |
2145 | read_unlock(&vmlist_lock); | |
2146 | } | |
2147 | ||
a47a126a ED |
2148 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2149 | { | |
2150 | if (NUMA_BUILD) { | |
2151 | unsigned int nr, *counters = m->private; | |
2152 | ||
2153 | if (!counters) | |
2154 | return; | |
2155 | ||
2156 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); | |
2157 | ||
2158 | for (nr = 0; nr < v->nr_pages; nr++) | |
2159 | counters[page_to_nid(v->pages[nr])]++; | |
2160 | ||
2161 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2162 | if (counters[nr]) | |
2163 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2164 | } | |
2165 | } | |
2166 | ||
a10aa579 CL |
2167 | static int s_show(struct seq_file *m, void *p) |
2168 | { | |
2169 | struct vm_struct *v = p; | |
2170 | ||
2171 | seq_printf(m, "0x%p-0x%p %7ld", | |
2172 | v->addr, v->addr + v->size, v->size); | |
2173 | ||
23016969 | 2174 | if (v->caller) { |
9c246247 | 2175 | char buff[KSYM_SYMBOL_LEN]; |
23016969 CL |
2176 | |
2177 | seq_putc(m, ' '); | |
2178 | sprint_symbol(buff, (unsigned long)v->caller); | |
2179 | seq_puts(m, buff); | |
2180 | } | |
2181 | ||
a10aa579 CL |
2182 | if (v->nr_pages) |
2183 | seq_printf(m, " pages=%d", v->nr_pages); | |
2184 | ||
2185 | if (v->phys_addr) | |
2186 | seq_printf(m, " phys=%lx", v->phys_addr); | |
2187 | ||
2188 | if (v->flags & VM_IOREMAP) | |
2189 | seq_printf(m, " ioremap"); | |
2190 | ||
2191 | if (v->flags & VM_ALLOC) | |
2192 | seq_printf(m, " vmalloc"); | |
2193 | ||
2194 | if (v->flags & VM_MAP) | |
2195 | seq_printf(m, " vmap"); | |
2196 | ||
2197 | if (v->flags & VM_USERMAP) | |
2198 | seq_printf(m, " user"); | |
2199 | ||
2200 | if (v->flags & VM_VPAGES) | |
2201 | seq_printf(m, " vpages"); | |
2202 | ||
a47a126a | 2203 | show_numa_info(m, v); |
a10aa579 CL |
2204 | seq_putc(m, '\n'); |
2205 | return 0; | |
2206 | } | |
2207 | ||
5f6a6a9c | 2208 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2209 | .start = s_start, |
2210 | .next = s_next, | |
2211 | .stop = s_stop, | |
2212 | .show = s_show, | |
2213 | }; | |
5f6a6a9c AD |
2214 | |
2215 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2216 | { | |
2217 | unsigned int *ptr = NULL; | |
2218 | int ret; | |
2219 | ||
2220 | if (NUMA_BUILD) | |
2221 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); | |
2222 | ret = seq_open(file, &vmalloc_op); | |
2223 | if (!ret) { | |
2224 | struct seq_file *m = file->private_data; | |
2225 | m->private = ptr; | |
2226 | } else | |
2227 | kfree(ptr); | |
2228 | return ret; | |
2229 | } | |
2230 | ||
2231 | static const struct file_operations proc_vmalloc_operations = { | |
2232 | .open = vmalloc_open, | |
2233 | .read = seq_read, | |
2234 | .llseek = seq_lseek, | |
2235 | .release = seq_release_private, | |
2236 | }; | |
2237 | ||
2238 | static int __init proc_vmalloc_init(void) | |
2239 | { | |
2240 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2241 | return 0; | |
2242 | } | |
2243 | module_init(proc_vmalloc_init); | |
a10aa579 CL |
2244 | #endif |
2245 |