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