Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmalloc.c | |
3 | * | |
4 | * Copyright (C) 1993 Linus Torvalds | |
5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
6 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
7 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
9 | */ |
10 | ||
db64fe02 | 11 | #include <linux/vmalloc.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> | |
14 | #include <linux/highmem.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/spinlock.h> | |
17 | #include <linux/interrupt.h> | |
5f6a6a9c | 18 | #include <linux/proc_fs.h> |
a10aa579 | 19 | #include <linux/seq_file.h> |
3ac7fe5a | 20 | #include <linux/debugobjects.h> |
23016969 | 21 | #include <linux/kallsyms.h> |
db64fe02 NP |
22 | #include <linux/list.h> |
23 | #include <linux/rbtree.h> | |
24 | #include <linux/radix-tree.h> | |
25 | #include <linux/rcupdate.h> | |
f0aa6617 | 26 | #include <linux/pfn.h> |
89219d37 | 27 | #include <linux/kmemleak.h> |
db64fe02 | 28 | #include <asm/atomic.h> |
1da177e4 LT |
29 | #include <asm/uaccess.h> |
30 | #include <asm/tlbflush.h> | |
31 | ||
32 | ||
db64fe02 | 33 | /*** Page table manipulation functions ***/ |
b221385b | 34 | |
1da177e4 LT |
35 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
36 | { | |
37 | pte_t *pte; | |
38 | ||
39 | pte = pte_offset_kernel(pmd, addr); | |
40 | do { | |
41 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
42 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
43 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
44 | } | |
45 | ||
db64fe02 | 46 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
47 | { |
48 | pmd_t *pmd; | |
49 | unsigned long next; | |
50 | ||
51 | pmd = pmd_offset(pud, addr); | |
52 | do { | |
53 | next = pmd_addr_end(addr, end); | |
54 | if (pmd_none_or_clear_bad(pmd)) | |
55 | continue; | |
56 | vunmap_pte_range(pmd, addr, next); | |
57 | } while (pmd++, addr = next, addr != end); | |
58 | } | |
59 | ||
db64fe02 | 60 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
1da177e4 LT |
61 | { |
62 | pud_t *pud; | |
63 | unsigned long next; | |
64 | ||
65 | pud = pud_offset(pgd, addr); | |
66 | do { | |
67 | next = pud_addr_end(addr, end); | |
68 | if (pud_none_or_clear_bad(pud)) | |
69 | continue; | |
70 | vunmap_pmd_range(pud, addr, next); | |
71 | } while (pud++, addr = next, addr != end); | |
72 | } | |
73 | ||
db64fe02 | 74 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
75 | { |
76 | pgd_t *pgd; | |
77 | unsigned long next; | |
1da177e4 LT |
78 | |
79 | BUG_ON(addr >= end); | |
80 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
81 | do { |
82 | next = pgd_addr_end(addr, end); | |
83 | if (pgd_none_or_clear_bad(pgd)) | |
84 | continue; | |
85 | vunmap_pud_range(pgd, addr, next); | |
86 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
87 | } |
88 | ||
89 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 90 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
91 | { |
92 | pte_t *pte; | |
93 | ||
db64fe02 NP |
94 | /* |
95 | * nr is a running index into the array which helps higher level | |
96 | * callers keep track of where we're up to. | |
97 | */ | |
98 | ||
872fec16 | 99 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
100 | if (!pte) |
101 | return -ENOMEM; | |
102 | do { | |
db64fe02 NP |
103 | struct page *page = pages[*nr]; |
104 | ||
105 | if (WARN_ON(!pte_none(*pte))) | |
106 | return -EBUSY; | |
107 | if (WARN_ON(!page)) | |
1da177e4 LT |
108 | return -ENOMEM; |
109 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 110 | (*nr)++; |
1da177e4 LT |
111 | } while (pte++, addr += PAGE_SIZE, addr != end); |
112 | return 0; | |
113 | } | |
114 | ||
db64fe02 NP |
115 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
116 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
117 | { |
118 | pmd_t *pmd; | |
119 | unsigned long next; | |
120 | ||
121 | pmd = pmd_alloc(&init_mm, pud, addr); | |
122 | if (!pmd) | |
123 | return -ENOMEM; | |
124 | do { | |
125 | next = pmd_addr_end(addr, end); | |
db64fe02 | 126 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
127 | return -ENOMEM; |
128 | } while (pmd++, addr = next, addr != end); | |
129 | return 0; | |
130 | } | |
131 | ||
db64fe02 NP |
132 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
133 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
134 | { |
135 | pud_t *pud; | |
136 | unsigned long next; | |
137 | ||
138 | pud = pud_alloc(&init_mm, pgd, addr); | |
139 | if (!pud) | |
140 | return -ENOMEM; | |
141 | do { | |
142 | next = pud_addr_end(addr, end); | |
db64fe02 | 143 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
144 | return -ENOMEM; |
145 | } while (pud++, addr = next, addr != end); | |
146 | return 0; | |
147 | } | |
148 | ||
db64fe02 NP |
149 | /* |
150 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
151 | * will have pfns corresponding to the "pages" array. | |
152 | * | |
153 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
154 | */ | |
8fc48985 TH |
155 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
156 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
157 | { |
158 | pgd_t *pgd; | |
159 | unsigned long next; | |
2e4e27c7 | 160 | unsigned long addr = start; |
db64fe02 NP |
161 | int err = 0; |
162 | int nr = 0; | |
1da177e4 LT |
163 | |
164 | BUG_ON(addr >= end); | |
165 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
166 | do { |
167 | next = pgd_addr_end(addr, end); | |
db64fe02 | 168 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 169 | if (err) |
bf88c8c8 | 170 | return err; |
1da177e4 | 171 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 172 | |
db64fe02 | 173 | return nr; |
1da177e4 LT |
174 | } |
175 | ||
8fc48985 TH |
176 | static int vmap_page_range(unsigned long start, unsigned long end, |
177 | pgprot_t prot, struct page **pages) | |
178 | { | |
179 | int ret; | |
180 | ||
181 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
182 | flush_cache_vmap(start, end); | |
183 | return ret; | |
184 | } | |
185 | ||
81ac3ad9 | 186 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
187 | { |
188 | /* | |
ab4f2ee1 | 189 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
190 | * and fall back on vmalloc() if that fails. Others |
191 | * just put it in the vmalloc space. | |
192 | */ | |
193 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
194 | unsigned long addr = (unsigned long)x; | |
195 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
196 | return 1; | |
197 | #endif | |
198 | return is_vmalloc_addr(x); | |
199 | } | |
200 | ||
48667e7a | 201 | /* |
db64fe02 | 202 | * Walk a vmap address to the struct page it maps. |
48667e7a | 203 | */ |
b3bdda02 | 204 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
205 | { |
206 | unsigned long addr = (unsigned long) vmalloc_addr; | |
207 | struct page *page = NULL; | |
208 | pgd_t *pgd = pgd_offset_k(addr); | |
48667e7a | 209 | |
7aa413de IM |
210 | /* |
211 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
212 | * architectures that do not vmalloc module space | |
213 | */ | |
73bdf0a6 | 214 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 215 | |
48667e7a | 216 | if (!pgd_none(*pgd)) { |
db64fe02 | 217 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 218 | if (!pud_none(*pud)) { |
db64fe02 | 219 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 220 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
221 | pte_t *ptep, pte; |
222 | ||
48667e7a CL |
223 | ptep = pte_offset_map(pmd, addr); |
224 | pte = *ptep; | |
225 | if (pte_present(pte)) | |
226 | page = pte_page(pte); | |
227 | pte_unmap(ptep); | |
228 | } | |
229 | } | |
230 | } | |
231 | return page; | |
232 | } | |
233 | EXPORT_SYMBOL(vmalloc_to_page); | |
234 | ||
235 | /* | |
236 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
237 | */ | |
b3bdda02 | 238 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
239 | { |
240 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
241 | } | |
242 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
243 | ||
db64fe02 NP |
244 | |
245 | /*** Global kva allocator ***/ | |
246 | ||
247 | #define VM_LAZY_FREE 0x01 | |
248 | #define VM_LAZY_FREEING 0x02 | |
249 | #define VM_VM_AREA 0x04 | |
250 | ||
251 | struct vmap_area { | |
252 | unsigned long va_start; | |
253 | unsigned long va_end; | |
254 | unsigned long flags; | |
255 | struct rb_node rb_node; /* address sorted rbtree */ | |
256 | struct list_head list; /* address sorted list */ | |
257 | struct list_head purge_list; /* "lazy purge" list */ | |
258 | void *private; | |
259 | struct rcu_head rcu_head; | |
260 | }; | |
261 | ||
262 | static DEFINE_SPINLOCK(vmap_area_lock); | |
263 | static struct rb_root vmap_area_root = RB_ROOT; | |
264 | static LIST_HEAD(vmap_area_list); | |
ca23e405 | 265 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
266 | |
267 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 268 | { |
db64fe02 NP |
269 | struct rb_node *n = vmap_area_root.rb_node; |
270 | ||
271 | while (n) { | |
272 | struct vmap_area *va; | |
273 | ||
274 | va = rb_entry(n, struct vmap_area, rb_node); | |
275 | if (addr < va->va_start) | |
276 | n = n->rb_left; | |
277 | else if (addr > va->va_start) | |
278 | n = n->rb_right; | |
279 | else | |
280 | return va; | |
281 | } | |
282 | ||
283 | return NULL; | |
284 | } | |
285 | ||
286 | static void __insert_vmap_area(struct vmap_area *va) | |
287 | { | |
288 | struct rb_node **p = &vmap_area_root.rb_node; | |
289 | struct rb_node *parent = NULL; | |
290 | struct rb_node *tmp; | |
291 | ||
292 | while (*p) { | |
293 | struct vmap_area *tmp; | |
294 | ||
295 | parent = *p; | |
296 | tmp = rb_entry(parent, struct vmap_area, rb_node); | |
297 | if (va->va_start < tmp->va_end) | |
298 | p = &(*p)->rb_left; | |
299 | else if (va->va_end > tmp->va_start) | |
300 | p = &(*p)->rb_right; | |
301 | else | |
302 | BUG(); | |
303 | } | |
304 | ||
305 | rb_link_node(&va->rb_node, parent, p); | |
306 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
307 | ||
308 | /* address-sort this list so it is usable like the vmlist */ | |
309 | tmp = rb_prev(&va->rb_node); | |
310 | if (tmp) { | |
311 | struct vmap_area *prev; | |
312 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
313 | list_add_rcu(&va->list, &prev->list); | |
314 | } else | |
315 | list_add_rcu(&va->list, &vmap_area_list); | |
316 | } | |
317 | ||
318 | static void purge_vmap_area_lazy(void); | |
319 | ||
320 | /* | |
321 | * Allocate a region of KVA of the specified size and alignment, within the | |
322 | * vstart and vend. | |
323 | */ | |
324 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
325 | unsigned long align, | |
326 | unsigned long vstart, unsigned long vend, | |
327 | int node, gfp_t gfp_mask) | |
328 | { | |
329 | struct vmap_area *va; | |
330 | struct rb_node *n; | |
1da177e4 | 331 | unsigned long addr; |
db64fe02 NP |
332 | int purged = 0; |
333 | ||
7766970c | 334 | BUG_ON(!size); |
db64fe02 NP |
335 | BUG_ON(size & ~PAGE_MASK); |
336 | ||
db64fe02 NP |
337 | va = kmalloc_node(sizeof(struct vmap_area), |
338 | gfp_mask & GFP_RECLAIM_MASK, node); | |
339 | if (unlikely(!va)) | |
340 | return ERR_PTR(-ENOMEM); | |
341 | ||
342 | retry: | |
0ae15132 GC |
343 | addr = ALIGN(vstart, align); |
344 | ||
db64fe02 | 345 | spin_lock(&vmap_area_lock); |
7766970c NP |
346 | if (addr + size - 1 < addr) |
347 | goto overflow; | |
348 | ||
db64fe02 NP |
349 | /* XXX: could have a last_hole cache */ |
350 | n = vmap_area_root.rb_node; | |
351 | if (n) { | |
352 | struct vmap_area *first = NULL; | |
353 | ||
354 | do { | |
355 | struct vmap_area *tmp; | |
356 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
357 | if (tmp->va_end >= addr) { | |
358 | if (!first && tmp->va_start < addr + size) | |
359 | first = tmp; | |
360 | n = n->rb_left; | |
361 | } else { | |
362 | first = tmp; | |
363 | n = n->rb_right; | |
364 | } | |
365 | } while (n); | |
366 | ||
367 | if (!first) | |
368 | goto found; | |
369 | ||
370 | if (first->va_end < addr) { | |
371 | n = rb_next(&first->rb_node); | |
372 | if (n) | |
373 | first = rb_entry(n, struct vmap_area, rb_node); | |
374 | else | |
375 | goto found; | |
376 | } | |
377 | ||
f011c2da | 378 | while (addr + size > first->va_start && addr + size <= vend) { |
db64fe02 | 379 | addr = ALIGN(first->va_end + PAGE_SIZE, align); |
7766970c NP |
380 | if (addr + size - 1 < addr) |
381 | goto overflow; | |
db64fe02 NP |
382 | |
383 | n = rb_next(&first->rb_node); | |
384 | if (n) | |
385 | first = rb_entry(n, struct vmap_area, rb_node); | |
386 | else | |
387 | goto found; | |
388 | } | |
389 | } | |
390 | found: | |
391 | if (addr + size > vend) { | |
7766970c | 392 | overflow: |
db64fe02 NP |
393 | spin_unlock(&vmap_area_lock); |
394 | if (!purged) { | |
395 | purge_vmap_area_lazy(); | |
396 | purged = 1; | |
397 | goto retry; | |
398 | } | |
399 | if (printk_ratelimit()) | |
c1279c4e GC |
400 | printk(KERN_WARNING |
401 | "vmap allocation for size %lu failed: " | |
402 | "use vmalloc=<size> to increase size.\n", size); | |
2498ce42 | 403 | kfree(va); |
db64fe02 NP |
404 | return ERR_PTR(-EBUSY); |
405 | } | |
406 | ||
407 | BUG_ON(addr & (align-1)); | |
408 | ||
409 | va->va_start = addr; | |
410 | va->va_end = addr + size; | |
411 | va->flags = 0; | |
412 | __insert_vmap_area(va); | |
413 | spin_unlock(&vmap_area_lock); | |
414 | ||
415 | return va; | |
416 | } | |
417 | ||
418 | static void rcu_free_va(struct rcu_head *head) | |
419 | { | |
420 | struct vmap_area *va = container_of(head, struct vmap_area, rcu_head); | |
421 | ||
422 | kfree(va); | |
423 | } | |
424 | ||
425 | static void __free_vmap_area(struct vmap_area *va) | |
426 | { | |
427 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
428 | rb_erase(&va->rb_node, &vmap_area_root); | |
429 | RB_CLEAR_NODE(&va->rb_node); | |
430 | list_del_rcu(&va->list); | |
431 | ||
ca23e405 TH |
432 | /* |
433 | * Track the highest possible candidate for pcpu area | |
434 | * allocation. Areas outside of vmalloc area can be returned | |
435 | * here too, consider only end addresses which fall inside | |
436 | * vmalloc area proper. | |
437 | */ | |
438 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
439 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
440 | ||
db64fe02 NP |
441 | call_rcu(&va->rcu_head, rcu_free_va); |
442 | } | |
443 | ||
444 | /* | |
445 | * Free a region of KVA allocated by alloc_vmap_area | |
446 | */ | |
447 | static void free_vmap_area(struct vmap_area *va) | |
448 | { | |
449 | spin_lock(&vmap_area_lock); | |
450 | __free_vmap_area(va); | |
451 | spin_unlock(&vmap_area_lock); | |
452 | } | |
453 | ||
454 | /* | |
455 | * Clear the pagetable entries of a given vmap_area | |
456 | */ | |
457 | static void unmap_vmap_area(struct vmap_area *va) | |
458 | { | |
459 | vunmap_page_range(va->va_start, va->va_end); | |
460 | } | |
461 | ||
cd52858c NP |
462 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
463 | { | |
464 | /* | |
465 | * Unmap page tables and force a TLB flush immediately if | |
466 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free | |
467 | * bugs similarly to those in linear kernel virtual address | |
468 | * space after a page has been freed. | |
469 | * | |
470 | * All the lazy freeing logic is still retained, in order to | |
471 | * minimise intrusiveness of this debugging feature. | |
472 | * | |
473 | * This is going to be *slow* (linear kernel virtual address | |
474 | * debugging doesn't do a broadcast TLB flush so it is a lot | |
475 | * faster). | |
476 | */ | |
477 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
478 | vunmap_page_range(start, end); | |
479 | flush_tlb_kernel_range(start, end); | |
480 | #endif | |
481 | } | |
482 | ||
db64fe02 NP |
483 | /* |
484 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
485 | * before attempting to purge with a TLB flush. | |
486 | * | |
487 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
488 | * and take slightly longer to purge, but it will linearly reduce the number of | |
489 | * global TLB flushes that must be performed. It would seem natural to scale | |
490 | * this number up linearly with the number of CPUs (because vmapping activity | |
491 | * could also scale linearly with the number of CPUs), however it is likely | |
492 | * that in practice, workloads might be constrained in other ways that mean | |
493 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
494 | * conservative and not introduce a big latency on huge systems, so go with | |
495 | * a less aggressive log scale. It will still be an improvement over the old | |
496 | * code, and it will be simple to change the scale factor if we find that it | |
497 | * becomes a problem on bigger systems. | |
498 | */ | |
499 | static unsigned long lazy_max_pages(void) | |
500 | { | |
501 | unsigned int log; | |
502 | ||
503 | log = fls(num_online_cpus()); | |
504 | ||
505 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
506 | } | |
507 | ||
508 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
509 | ||
510 | /* | |
511 | * Purges all lazily-freed vmap areas. | |
512 | * | |
513 | * If sync is 0 then don't purge if there is already a purge in progress. | |
514 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
515 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
516 | * their own TLB flushing). | |
517 | * Returns with *start = min(*start, lowest purged address) | |
518 | * *end = max(*end, highest purged address) | |
519 | */ | |
520 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
521 | int sync, int force_flush) | |
522 | { | |
46666d8a | 523 | static DEFINE_SPINLOCK(purge_lock); |
db64fe02 NP |
524 | LIST_HEAD(valist); |
525 | struct vmap_area *va; | |
cbb76676 | 526 | struct vmap_area *n_va; |
db64fe02 NP |
527 | int nr = 0; |
528 | ||
529 | /* | |
530 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
531 | * should not expect such behaviour. This just simplifies locking for | |
532 | * the case that isn't actually used at the moment anyway. | |
533 | */ | |
534 | if (!sync && !force_flush) { | |
46666d8a | 535 | if (!spin_trylock(&purge_lock)) |
db64fe02 NP |
536 | return; |
537 | } else | |
46666d8a | 538 | spin_lock(&purge_lock); |
db64fe02 NP |
539 | |
540 | rcu_read_lock(); | |
541 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
542 | if (va->flags & VM_LAZY_FREE) { | |
543 | if (va->va_start < *start) | |
544 | *start = va->va_start; | |
545 | if (va->va_end > *end) | |
546 | *end = va->va_end; | |
547 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
548 | unmap_vmap_area(va); | |
549 | list_add_tail(&va->purge_list, &valist); | |
550 | va->flags |= VM_LAZY_FREEING; | |
551 | va->flags &= ~VM_LAZY_FREE; | |
552 | } | |
553 | } | |
554 | rcu_read_unlock(); | |
555 | ||
556 | if (nr) { | |
557 | BUG_ON(nr > atomic_read(&vmap_lazy_nr)); | |
558 | atomic_sub(nr, &vmap_lazy_nr); | |
559 | } | |
560 | ||
561 | if (nr || force_flush) | |
562 | flush_tlb_kernel_range(*start, *end); | |
563 | ||
564 | if (nr) { | |
565 | spin_lock(&vmap_area_lock); | |
cbb76676 | 566 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
db64fe02 NP |
567 | __free_vmap_area(va); |
568 | spin_unlock(&vmap_area_lock); | |
569 | } | |
46666d8a | 570 | spin_unlock(&purge_lock); |
db64fe02 NP |
571 | } |
572 | ||
496850e5 NP |
573 | /* |
574 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
575 | * is already purging. | |
576 | */ | |
577 | static void try_purge_vmap_area_lazy(void) | |
578 | { | |
579 | unsigned long start = ULONG_MAX, end = 0; | |
580 | ||
581 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
582 | } | |
583 | ||
db64fe02 NP |
584 | /* |
585 | * Kick off a purge of the outstanding lazy areas. | |
586 | */ | |
587 | static void purge_vmap_area_lazy(void) | |
588 | { | |
589 | unsigned long start = ULONG_MAX, end = 0; | |
590 | ||
496850e5 | 591 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
db64fe02 NP |
592 | } |
593 | ||
594 | /* | |
b29acbdc NP |
595 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been |
596 | * called for the correct range previously. | |
db64fe02 | 597 | */ |
b29acbdc | 598 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) |
db64fe02 NP |
599 | { |
600 | va->flags |= VM_LAZY_FREE; | |
601 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
602 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
496850e5 | 603 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
604 | } |
605 | ||
b29acbdc NP |
606 | /* |
607 | * Free and unmap a vmap area | |
608 | */ | |
609 | static void free_unmap_vmap_area(struct vmap_area *va) | |
610 | { | |
611 | flush_cache_vunmap(va->va_start, va->va_end); | |
612 | free_unmap_vmap_area_noflush(va); | |
613 | } | |
614 | ||
db64fe02 NP |
615 | static struct vmap_area *find_vmap_area(unsigned long addr) |
616 | { | |
617 | struct vmap_area *va; | |
618 | ||
619 | spin_lock(&vmap_area_lock); | |
620 | va = __find_vmap_area(addr); | |
621 | spin_unlock(&vmap_area_lock); | |
622 | ||
623 | return va; | |
624 | } | |
625 | ||
626 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
627 | { | |
628 | struct vmap_area *va; | |
629 | ||
630 | va = find_vmap_area(addr); | |
631 | BUG_ON(!va); | |
632 | free_unmap_vmap_area(va); | |
633 | } | |
634 | ||
635 | ||
636 | /*** Per cpu kva allocator ***/ | |
637 | ||
638 | /* | |
639 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
640 | * room for at least 16 percpu vmap blocks per CPU. | |
641 | */ | |
642 | /* | |
643 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
644 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
645 | * instead (we just need a rough idea) | |
646 | */ | |
647 | #if BITS_PER_LONG == 32 | |
648 | #define VMALLOC_SPACE (128UL*1024*1024) | |
649 | #else | |
650 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
651 | #endif | |
652 | ||
653 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
654 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
655 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
656 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
657 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
658 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
659 | #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
660 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
661 | VMALLOC_PAGES / NR_CPUS / 16)) | |
662 | ||
663 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
664 | ||
9b463334 JF |
665 | static bool vmap_initialized __read_mostly = false; |
666 | ||
db64fe02 NP |
667 | struct vmap_block_queue { |
668 | spinlock_t lock; | |
669 | struct list_head free; | |
670 | struct list_head dirty; | |
671 | unsigned int nr_dirty; | |
672 | }; | |
673 | ||
674 | struct vmap_block { | |
675 | spinlock_t lock; | |
676 | struct vmap_area *va; | |
677 | struct vmap_block_queue *vbq; | |
678 | unsigned long free, dirty; | |
679 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
680 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
681 | union { | |
d086817d | 682 | struct list_head free_list; |
db64fe02 NP |
683 | struct rcu_head rcu_head; |
684 | }; | |
685 | }; | |
686 | ||
687 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
688 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
689 | ||
690 | /* | |
691 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
692 | * in the free path. Could get rid of this if we change the API to return a | |
693 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
694 | */ | |
695 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
696 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
697 | ||
698 | /* | |
699 | * We should probably have a fallback mechanism to allocate virtual memory | |
700 | * out of partially filled vmap blocks. However vmap block sizing should be | |
701 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
702 | * big problem. | |
703 | */ | |
704 | ||
705 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
706 | { | |
707 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
708 | addr /= VMAP_BLOCK_SIZE; | |
709 | return addr; | |
710 | } | |
711 | ||
712 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
713 | { | |
714 | struct vmap_block_queue *vbq; | |
715 | struct vmap_block *vb; | |
716 | struct vmap_area *va; | |
717 | unsigned long vb_idx; | |
718 | int node, err; | |
719 | ||
720 | node = numa_node_id(); | |
721 | ||
722 | vb = kmalloc_node(sizeof(struct vmap_block), | |
723 | gfp_mask & GFP_RECLAIM_MASK, node); | |
724 | if (unlikely(!vb)) | |
725 | return ERR_PTR(-ENOMEM); | |
726 | ||
727 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
728 | VMALLOC_START, VMALLOC_END, | |
729 | node, gfp_mask); | |
730 | if (unlikely(IS_ERR(va))) { | |
731 | kfree(vb); | |
732 | return ERR_PTR(PTR_ERR(va)); | |
733 | } | |
734 | ||
735 | err = radix_tree_preload(gfp_mask); | |
736 | if (unlikely(err)) { | |
737 | kfree(vb); | |
738 | free_vmap_area(va); | |
739 | return ERR_PTR(err); | |
740 | } | |
741 | ||
742 | spin_lock_init(&vb->lock); | |
743 | vb->va = va; | |
744 | vb->free = VMAP_BBMAP_BITS; | |
745 | vb->dirty = 0; | |
746 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
747 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
748 | INIT_LIST_HEAD(&vb->free_list); | |
db64fe02 NP |
749 | |
750 | vb_idx = addr_to_vb_idx(va->va_start); | |
751 | spin_lock(&vmap_block_tree_lock); | |
752 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
753 | spin_unlock(&vmap_block_tree_lock); | |
754 | BUG_ON(err); | |
755 | radix_tree_preload_end(); | |
756 | ||
757 | vbq = &get_cpu_var(vmap_block_queue); | |
758 | vb->vbq = vbq; | |
759 | spin_lock(&vbq->lock); | |
760 | list_add(&vb->free_list, &vbq->free); | |
761 | spin_unlock(&vbq->lock); | |
762 | put_cpu_var(vmap_cpu_blocks); | |
763 | ||
764 | return vb; | |
765 | } | |
766 | ||
767 | static void rcu_free_vb(struct rcu_head *head) | |
768 | { | |
769 | struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head); | |
770 | ||
771 | kfree(vb); | |
772 | } | |
773 | ||
774 | static void free_vmap_block(struct vmap_block *vb) | |
775 | { | |
776 | struct vmap_block *tmp; | |
777 | unsigned long vb_idx; | |
778 | ||
d086817d | 779 | BUG_ON(!list_empty(&vb->free_list)); |
db64fe02 NP |
780 | |
781 | vb_idx = addr_to_vb_idx(vb->va->va_start); | |
782 | spin_lock(&vmap_block_tree_lock); | |
783 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
784 | spin_unlock(&vmap_block_tree_lock); | |
785 | BUG_ON(tmp != vb); | |
786 | ||
b29acbdc | 787 | free_unmap_vmap_area_noflush(vb->va); |
db64fe02 NP |
788 | call_rcu(&vb->rcu_head, rcu_free_vb); |
789 | } | |
790 | ||
791 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) | |
792 | { | |
793 | struct vmap_block_queue *vbq; | |
794 | struct vmap_block *vb; | |
795 | unsigned long addr = 0; | |
796 | unsigned int order; | |
797 | ||
798 | BUG_ON(size & ~PAGE_MASK); | |
799 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
800 | order = get_order(size); | |
801 | ||
802 | again: | |
803 | rcu_read_lock(); | |
804 | vbq = &get_cpu_var(vmap_block_queue); | |
805 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
806 | int i; | |
807 | ||
808 | spin_lock(&vb->lock); | |
809 | i = bitmap_find_free_region(vb->alloc_map, | |
810 | VMAP_BBMAP_BITS, order); | |
811 | ||
812 | if (i >= 0) { | |
813 | addr = vb->va->va_start + (i << PAGE_SHIFT); | |
814 | BUG_ON(addr_to_vb_idx(addr) != | |
815 | addr_to_vb_idx(vb->va->va_start)); | |
816 | vb->free -= 1UL << order; | |
817 | if (vb->free == 0) { | |
818 | spin_lock(&vbq->lock); | |
819 | list_del_init(&vb->free_list); | |
820 | spin_unlock(&vbq->lock); | |
821 | } | |
822 | spin_unlock(&vb->lock); | |
823 | break; | |
824 | } | |
825 | spin_unlock(&vb->lock); | |
826 | } | |
827 | put_cpu_var(vmap_cpu_blocks); | |
828 | rcu_read_unlock(); | |
829 | ||
830 | if (!addr) { | |
831 | vb = new_vmap_block(gfp_mask); | |
832 | if (IS_ERR(vb)) | |
833 | return vb; | |
834 | goto again; | |
835 | } | |
836 | ||
837 | return (void *)addr; | |
838 | } | |
839 | ||
840 | static void vb_free(const void *addr, unsigned long size) | |
841 | { | |
842 | unsigned long offset; | |
843 | unsigned long vb_idx; | |
844 | unsigned int order; | |
845 | struct vmap_block *vb; | |
846 | ||
847 | BUG_ON(size & ~PAGE_MASK); | |
848 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
b29acbdc NP |
849 | |
850 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
851 | ||
db64fe02 NP |
852 | order = get_order(size); |
853 | ||
854 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
855 | ||
856 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
857 | rcu_read_lock(); | |
858 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
859 | rcu_read_unlock(); | |
860 | BUG_ON(!vb); | |
861 | ||
862 | spin_lock(&vb->lock); | |
863 | bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order); | |
d086817d | 864 | |
db64fe02 NP |
865 | vb->dirty += 1UL << order; |
866 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
867 | BUG_ON(vb->free || !list_empty(&vb->free_list)); | |
868 | spin_unlock(&vb->lock); | |
869 | free_vmap_block(vb); | |
870 | } else | |
871 | spin_unlock(&vb->lock); | |
872 | } | |
873 | ||
874 | /** | |
875 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
876 | * | |
877 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
878 | * to amortize TLB flushing overheads. What this means is that any page you | |
879 | * have now, may, in a former life, have been mapped into kernel virtual | |
880 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
881 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
882 | * | |
883 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
884 | * be sure that none of the pages we have control over will have any aliases | |
885 | * from the vmap layer. | |
886 | */ | |
887 | void vm_unmap_aliases(void) | |
888 | { | |
889 | unsigned long start = ULONG_MAX, end = 0; | |
890 | int cpu; | |
891 | int flush = 0; | |
892 | ||
9b463334 JF |
893 | if (unlikely(!vmap_initialized)) |
894 | return; | |
895 | ||
db64fe02 NP |
896 | for_each_possible_cpu(cpu) { |
897 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
898 | struct vmap_block *vb; | |
899 | ||
900 | rcu_read_lock(); | |
901 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
902 | int i; | |
903 | ||
904 | spin_lock(&vb->lock); | |
905 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
906 | while (i < VMAP_BBMAP_BITS) { | |
907 | unsigned long s, e; | |
908 | int j; | |
909 | j = find_next_zero_bit(vb->dirty_map, | |
910 | VMAP_BBMAP_BITS, i); | |
911 | ||
912 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
913 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
914 | vunmap_page_range(s, e); | |
915 | flush = 1; | |
916 | ||
917 | if (s < start) | |
918 | start = s; | |
919 | if (e > end) | |
920 | end = e; | |
921 | ||
922 | i = j; | |
923 | i = find_next_bit(vb->dirty_map, | |
924 | VMAP_BBMAP_BITS, i); | |
925 | } | |
926 | spin_unlock(&vb->lock); | |
927 | } | |
928 | rcu_read_unlock(); | |
929 | } | |
930 | ||
931 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
932 | } | |
933 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
934 | ||
935 | /** | |
936 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
937 | * @mem: the pointer returned by vm_map_ram | |
938 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
939 | */ | |
940 | void vm_unmap_ram(const void *mem, unsigned int count) | |
941 | { | |
942 | unsigned long size = count << PAGE_SHIFT; | |
943 | unsigned long addr = (unsigned long)mem; | |
944 | ||
945 | BUG_ON(!addr); | |
946 | BUG_ON(addr < VMALLOC_START); | |
947 | BUG_ON(addr > VMALLOC_END); | |
948 | BUG_ON(addr & (PAGE_SIZE-1)); | |
949 | ||
950 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 951 | vmap_debug_free_range(addr, addr+size); |
db64fe02 NP |
952 | |
953 | if (likely(count <= VMAP_MAX_ALLOC)) | |
954 | vb_free(mem, size); | |
955 | else | |
956 | free_unmap_vmap_area_addr(addr); | |
957 | } | |
958 | EXPORT_SYMBOL(vm_unmap_ram); | |
959 | ||
960 | /** | |
961 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
962 | * @pages: an array of pointers to the pages to be mapped | |
963 | * @count: number of pages | |
964 | * @node: prefer to allocate data structures on this node | |
965 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad RD |
966 | * |
967 | * Returns: a pointer to the address that has been mapped, or %NULL on failure | |
db64fe02 NP |
968 | */ |
969 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
970 | { | |
971 | unsigned long size = count << PAGE_SHIFT; | |
972 | unsigned long addr; | |
973 | void *mem; | |
974 | ||
975 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
976 | mem = vb_alloc(size, GFP_KERNEL); | |
977 | if (IS_ERR(mem)) | |
978 | return NULL; | |
979 | addr = (unsigned long)mem; | |
980 | } else { | |
981 | struct vmap_area *va; | |
982 | va = alloc_vmap_area(size, PAGE_SIZE, | |
983 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
984 | if (IS_ERR(va)) | |
985 | return NULL; | |
986 | ||
987 | addr = va->va_start; | |
988 | mem = (void *)addr; | |
989 | } | |
990 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
991 | vm_unmap_ram(mem, count); | |
992 | return NULL; | |
993 | } | |
994 | return mem; | |
995 | } | |
996 | EXPORT_SYMBOL(vm_map_ram); | |
997 | ||
f0aa6617 TH |
998 | /** |
999 | * vm_area_register_early - register vmap area early during boot | |
1000 | * @vm: vm_struct to register | |
c0c0a293 | 1001 | * @align: requested alignment |
f0aa6617 TH |
1002 | * |
1003 | * This function is used to register kernel vm area before | |
1004 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1005 | * proper values on entry and other fields should be zero. On return, | |
1006 | * vm->addr contains the allocated address. | |
1007 | * | |
1008 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1009 | */ | |
c0c0a293 | 1010 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1011 | { |
1012 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1013 | unsigned long addr; |
1014 | ||
1015 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1016 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1017 | |
c0c0a293 | 1018 | vm->addr = (void *)addr; |
f0aa6617 TH |
1019 | |
1020 | vm->next = vmlist; | |
1021 | vmlist = vm; | |
1022 | } | |
1023 | ||
db64fe02 NP |
1024 | void __init vmalloc_init(void) |
1025 | { | |
822c18f2 IK |
1026 | struct vmap_area *va; |
1027 | struct vm_struct *tmp; | |
db64fe02 NP |
1028 | int i; |
1029 | ||
1030 | for_each_possible_cpu(i) { | |
1031 | struct vmap_block_queue *vbq; | |
1032 | ||
1033 | vbq = &per_cpu(vmap_block_queue, i); | |
1034 | spin_lock_init(&vbq->lock); | |
1035 | INIT_LIST_HEAD(&vbq->free); | |
1036 | INIT_LIST_HEAD(&vbq->dirty); | |
1037 | vbq->nr_dirty = 0; | |
1038 | } | |
9b463334 | 1039 | |
822c18f2 IK |
1040 | /* Import existing vmlist entries. */ |
1041 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1042 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
822c18f2 IK |
1043 | va->flags = tmp->flags | VM_VM_AREA; |
1044 | va->va_start = (unsigned long)tmp->addr; | |
1045 | va->va_end = va->va_start + tmp->size; | |
1046 | __insert_vmap_area(va); | |
1047 | } | |
ca23e405 TH |
1048 | |
1049 | vmap_area_pcpu_hole = VMALLOC_END; | |
1050 | ||
9b463334 | 1051 | vmap_initialized = true; |
db64fe02 NP |
1052 | } |
1053 | ||
8fc48985 TH |
1054 | /** |
1055 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1056 | * @addr: start of the VM area to map | |
1057 | * @size: size of the VM area to map | |
1058 | * @prot: page protection flags to use | |
1059 | * @pages: pages to map | |
1060 | * | |
1061 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1062 | * specify should have been allocated using get_vm_area() and its | |
1063 | * friends. | |
1064 | * | |
1065 | * NOTE: | |
1066 | * This function does NOT do any cache flushing. The caller is | |
1067 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1068 | * before calling this function. | |
1069 | * | |
1070 | * RETURNS: | |
1071 | * The number of pages mapped on success, -errno on failure. | |
1072 | */ | |
1073 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1074 | pgprot_t prot, struct page **pages) | |
1075 | { | |
1076 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1077 | } | |
1078 | ||
1079 | /** | |
1080 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1081 | * @addr: start of the VM area to unmap | |
1082 | * @size: size of the VM area to unmap | |
1083 | * | |
1084 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1085 | * specify should have been allocated using get_vm_area() and its | |
1086 | * friends. | |
1087 | * | |
1088 | * NOTE: | |
1089 | * This function does NOT do any cache flushing. The caller is | |
1090 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1091 | * before calling this function and flush_tlb_kernel_range() after. | |
1092 | */ | |
1093 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1094 | { | |
1095 | vunmap_page_range(addr, addr + size); | |
1096 | } | |
1097 | ||
1098 | /** | |
1099 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1100 | * @addr: start of the VM area to unmap | |
1101 | * @size: size of the VM area to unmap | |
1102 | * | |
1103 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1104 | * the unmapping and tlb after. | |
1105 | */ | |
db64fe02 NP |
1106 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1107 | { | |
1108 | unsigned long end = addr + size; | |
f6fcba70 TH |
1109 | |
1110 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1111 | vunmap_page_range(addr, end); |
1112 | flush_tlb_kernel_range(addr, end); | |
1113 | } | |
1114 | ||
1115 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
1116 | { | |
1117 | unsigned long addr = (unsigned long)area->addr; | |
1118 | unsigned long end = addr + area->size - PAGE_SIZE; | |
1119 | int err; | |
1120 | ||
1121 | err = vmap_page_range(addr, end, prot, *pages); | |
1122 | if (err > 0) { | |
1123 | *pages += err; | |
1124 | err = 0; | |
1125 | } | |
1126 | ||
1127 | return err; | |
1128 | } | |
1129 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1130 | ||
1131 | /*** Old vmalloc interfaces ***/ | |
1132 | DEFINE_RWLOCK(vmlist_lock); | |
1133 | struct vm_struct *vmlist; | |
1134 | ||
cf88c790 TH |
1135 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
1136 | unsigned long flags, void *caller) | |
1137 | { | |
1138 | struct vm_struct *tmp, **p; | |
1139 | ||
1140 | vm->flags = flags; | |
1141 | vm->addr = (void *)va->va_start; | |
1142 | vm->size = va->va_end - va->va_start; | |
1143 | vm->caller = caller; | |
1144 | va->private = vm; | |
1145 | va->flags |= VM_VM_AREA; | |
1146 | ||
1147 | write_lock(&vmlist_lock); | |
1148 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1149 | if (tmp->addr >= vm->addr) | |
1150 | break; | |
1151 | } | |
1152 | vm->next = *p; | |
1153 | *p = vm; | |
1154 | write_unlock(&vmlist_lock); | |
1155 | } | |
1156 | ||
db64fe02 NP |
1157 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
1158 | unsigned long flags, unsigned long start, unsigned long end, | |
1159 | int node, gfp_t gfp_mask, void *caller) | |
1160 | { | |
1161 | static struct vmap_area *va; | |
1162 | struct vm_struct *area; | |
db64fe02 | 1163 | unsigned long align = 1; |
1da177e4 | 1164 | |
52fd24ca | 1165 | BUG_ON(in_interrupt()); |
1da177e4 LT |
1166 | if (flags & VM_IOREMAP) { |
1167 | int bit = fls(size); | |
1168 | ||
1169 | if (bit > IOREMAP_MAX_ORDER) | |
1170 | bit = IOREMAP_MAX_ORDER; | |
1171 | else if (bit < PAGE_SHIFT) | |
1172 | bit = PAGE_SHIFT; | |
1173 | ||
1174 | align = 1ul << bit; | |
1175 | } | |
db64fe02 | 1176 | |
1da177e4 | 1177 | size = PAGE_ALIGN(size); |
31be8309 OH |
1178 | if (unlikely(!size)) |
1179 | return NULL; | |
1da177e4 | 1180 | |
cf88c790 | 1181 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1182 | if (unlikely(!area)) |
1183 | return NULL; | |
1184 | ||
1da177e4 LT |
1185 | /* |
1186 | * We always allocate a guard page. | |
1187 | */ | |
1188 | size += PAGE_SIZE; | |
1189 | ||
db64fe02 NP |
1190 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1191 | if (IS_ERR(va)) { | |
1192 | kfree(area); | |
1193 | return NULL; | |
1da177e4 | 1194 | } |
1da177e4 | 1195 | |
cf88c790 | 1196 | insert_vmalloc_vm(area, va, flags, caller); |
1da177e4 | 1197 | return area; |
1da177e4 LT |
1198 | } |
1199 | ||
930fc45a CL |
1200 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1201 | unsigned long start, unsigned long end) | |
1202 | { | |
23016969 CL |
1203 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, |
1204 | __builtin_return_address(0)); | |
930fc45a | 1205 | } |
5992b6da | 1206 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1207 | |
c2968612 BH |
1208 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1209 | unsigned long start, unsigned long end, | |
1210 | void *caller) | |
1211 | { | |
1212 | return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL, | |
1213 | caller); | |
1214 | } | |
1215 | ||
1da177e4 | 1216 | /** |
183ff22b | 1217 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1218 | * @size: size of the area |
1219 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1220 | * | |
1221 | * Search an area of @size in the kernel virtual mapping area, | |
1222 | * and reserved it for out purposes. Returns the area descriptor | |
1223 | * on success or %NULL on failure. | |
1224 | */ | |
1225 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1226 | { | |
23016969 CL |
1227 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, |
1228 | -1, GFP_KERNEL, __builtin_return_address(0)); | |
1229 | } | |
1230 | ||
1231 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
1232 | void *caller) | |
1233 | { | |
1234 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, | |
1235 | -1, GFP_KERNEL, caller); | |
1da177e4 LT |
1236 | } |
1237 | ||
52fd24ca GP |
1238 | struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, |
1239 | int node, gfp_t gfp_mask) | |
930fc45a | 1240 | { |
52fd24ca | 1241 | return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node, |
23016969 | 1242 | gfp_mask, __builtin_return_address(0)); |
930fc45a CL |
1243 | } |
1244 | ||
db64fe02 | 1245 | static struct vm_struct *find_vm_area(const void *addr) |
83342314 | 1246 | { |
db64fe02 | 1247 | struct vmap_area *va; |
83342314 | 1248 | |
db64fe02 NP |
1249 | va = find_vmap_area((unsigned long)addr); |
1250 | if (va && va->flags & VM_VM_AREA) | |
1251 | return va->private; | |
1da177e4 | 1252 | |
1da177e4 | 1253 | return NULL; |
1da177e4 LT |
1254 | } |
1255 | ||
7856dfeb | 1256 | /** |
183ff22b | 1257 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1258 | * @addr: base address |
1259 | * | |
1260 | * Search for the kernel VM area starting at @addr, and remove it. | |
1261 | * This function returns the found VM area, but using it is NOT safe | |
1262 | * on SMP machines, except for its size or flags. | |
1263 | */ | |
b3bdda02 | 1264 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1265 | { |
db64fe02 NP |
1266 | struct vmap_area *va; |
1267 | ||
1268 | va = find_vmap_area((unsigned long)addr); | |
1269 | if (va && va->flags & VM_VM_AREA) { | |
1270 | struct vm_struct *vm = va->private; | |
1271 | struct vm_struct *tmp, **p; | |
dd32c279 KH |
1272 | /* |
1273 | * remove from list and disallow access to this vm_struct | |
1274 | * before unmap. (address range confliction is maintained by | |
1275 | * vmap.) | |
1276 | */ | |
db64fe02 NP |
1277 | write_lock(&vmlist_lock); |
1278 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) | |
1279 | ; | |
1280 | *p = tmp->next; | |
1281 | write_unlock(&vmlist_lock); | |
1282 | ||
dd32c279 KH |
1283 | vmap_debug_free_range(va->va_start, va->va_end); |
1284 | free_unmap_vmap_area(va); | |
1285 | vm->size -= PAGE_SIZE; | |
1286 | ||
db64fe02 NP |
1287 | return vm; |
1288 | } | |
1289 | return NULL; | |
7856dfeb AK |
1290 | } |
1291 | ||
b3bdda02 | 1292 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1293 | { |
1294 | struct vm_struct *area; | |
1295 | ||
1296 | if (!addr) | |
1297 | return; | |
1298 | ||
1299 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1300 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1301 | return; |
1302 | } | |
1303 | ||
1304 | area = remove_vm_area(addr); | |
1305 | if (unlikely(!area)) { | |
4c8573e2 | 1306 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1307 | addr); |
1da177e4 LT |
1308 | return; |
1309 | } | |
1310 | ||
9a11b49a | 1311 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1312 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1313 | |
1da177e4 LT |
1314 | if (deallocate_pages) { |
1315 | int i; | |
1316 | ||
1317 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1318 | struct page *page = area->pages[i]; |
1319 | ||
1320 | BUG_ON(!page); | |
1321 | __free_page(page); | |
1da177e4 LT |
1322 | } |
1323 | ||
8757d5fa | 1324 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1325 | vfree(area->pages); |
1326 | else | |
1327 | kfree(area->pages); | |
1328 | } | |
1329 | ||
1330 | kfree(area); | |
1331 | return; | |
1332 | } | |
1333 | ||
1334 | /** | |
1335 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1336 | * @addr: memory base address |
1337 | * | |
183ff22b | 1338 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1339 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1340 | * NULL, no operation is performed. | |
1da177e4 | 1341 | * |
80e93eff | 1342 | * Must not be called in interrupt context. |
1da177e4 | 1343 | */ |
b3bdda02 | 1344 | void vfree(const void *addr) |
1da177e4 LT |
1345 | { |
1346 | BUG_ON(in_interrupt()); | |
89219d37 CM |
1347 | |
1348 | kmemleak_free(addr); | |
1349 | ||
1da177e4 LT |
1350 | __vunmap(addr, 1); |
1351 | } | |
1da177e4 LT |
1352 | EXPORT_SYMBOL(vfree); |
1353 | ||
1354 | /** | |
1355 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1356 | * @addr: memory base address |
1357 | * | |
1358 | * Free the virtually contiguous memory area starting at @addr, | |
1359 | * which was created from the page array passed to vmap(). | |
1360 | * | |
80e93eff | 1361 | * Must not be called in interrupt context. |
1da177e4 | 1362 | */ |
b3bdda02 | 1363 | void vunmap(const void *addr) |
1da177e4 LT |
1364 | { |
1365 | BUG_ON(in_interrupt()); | |
34754b69 | 1366 | might_sleep(); |
1da177e4 LT |
1367 | __vunmap(addr, 0); |
1368 | } | |
1da177e4 LT |
1369 | EXPORT_SYMBOL(vunmap); |
1370 | ||
1371 | /** | |
1372 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1373 | * @pages: array of page pointers |
1374 | * @count: number of pages to map | |
1375 | * @flags: vm_area->flags | |
1376 | * @prot: page protection for the mapping | |
1377 | * | |
1378 | * Maps @count pages from @pages into contiguous kernel virtual | |
1379 | * space. | |
1380 | */ | |
1381 | void *vmap(struct page **pages, unsigned int count, | |
1382 | unsigned long flags, pgprot_t prot) | |
1383 | { | |
1384 | struct vm_struct *area; | |
1385 | ||
34754b69 PZ |
1386 | might_sleep(); |
1387 | ||
4481374c | 1388 | if (count > totalram_pages) |
1da177e4 LT |
1389 | return NULL; |
1390 | ||
23016969 CL |
1391 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1392 | __builtin_return_address(0)); | |
1da177e4 LT |
1393 | if (!area) |
1394 | return NULL; | |
23016969 | 1395 | |
1da177e4 LT |
1396 | if (map_vm_area(area, prot, &pages)) { |
1397 | vunmap(area->addr); | |
1398 | return NULL; | |
1399 | } | |
1400 | ||
1401 | return area->addr; | |
1402 | } | |
1da177e4 LT |
1403 | EXPORT_SYMBOL(vmap); |
1404 | ||
db64fe02 NP |
1405 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
1406 | int node, void *caller); | |
e31d9eb5 | 1407 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 1408 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
1409 | { |
1410 | struct page **pages; | |
1411 | unsigned int nr_pages, array_size, i; | |
1412 | ||
1413 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1414 | array_size = (nr_pages * sizeof(struct page *)); | |
1415 | ||
1416 | area->nr_pages = nr_pages; | |
1417 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1418 | if (array_size > PAGE_SIZE) { |
94f6030c | 1419 | pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO, |
23016969 | 1420 | PAGE_KERNEL, node, caller); |
8757d5fa | 1421 | area->flags |= VM_VPAGES; |
286e1ea3 AM |
1422 | } else { |
1423 | pages = kmalloc_node(array_size, | |
6cb06229 | 1424 | (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO, |
286e1ea3 AM |
1425 | node); |
1426 | } | |
1da177e4 | 1427 | area->pages = pages; |
23016969 | 1428 | area->caller = caller; |
1da177e4 LT |
1429 | if (!area->pages) { |
1430 | remove_vm_area(area->addr); | |
1431 | kfree(area); | |
1432 | return NULL; | |
1433 | } | |
1da177e4 LT |
1434 | |
1435 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1436 | struct page *page; |
1437 | ||
930fc45a | 1438 | if (node < 0) |
bf53d6f8 | 1439 | page = alloc_page(gfp_mask); |
930fc45a | 1440 | else |
bf53d6f8 CL |
1441 | page = alloc_pages_node(node, gfp_mask, 0); |
1442 | ||
1443 | if (unlikely(!page)) { | |
1da177e4 LT |
1444 | /* Successfully allocated i pages, free them in __vunmap() */ |
1445 | area->nr_pages = i; | |
1446 | goto fail; | |
1447 | } | |
bf53d6f8 | 1448 | area->pages[i] = page; |
1da177e4 LT |
1449 | } |
1450 | ||
1451 | if (map_vm_area(area, prot, &pages)) | |
1452 | goto fail; | |
1453 | return area->addr; | |
1454 | ||
1455 | fail: | |
1456 | vfree(area->addr); | |
1457 | return NULL; | |
1458 | } | |
1459 | ||
930fc45a CL |
1460 | void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot) |
1461 | { | |
89219d37 CM |
1462 | void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1, |
1463 | __builtin_return_address(0)); | |
1464 | ||
1465 | /* | |
1466 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1467 | * structures allocated in the __get_vm_area_node() function contain | |
1468 | * references to the virtual address of the vmalloc'ed block. | |
1469 | */ | |
1470 | kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask); | |
1471 | ||
1472 | return addr; | |
930fc45a CL |
1473 | } |
1474 | ||
1da177e4 | 1475 | /** |
930fc45a | 1476 | * __vmalloc_node - allocate virtually contiguous memory |
1da177e4 LT |
1477 | * @size: allocation size |
1478 | * @gfp_mask: flags for the page level allocator | |
1479 | * @prot: protection mask for the allocated pages | |
d44e0780 | 1480 | * @node: node to use for allocation or -1 |
c85d194b | 1481 | * @caller: caller's return address |
1da177e4 LT |
1482 | * |
1483 | * Allocate enough pages to cover @size from the page level | |
1484 | * allocator with @gfp_mask flags. Map them into contiguous | |
1485 | * kernel virtual space, using a pagetable protection of @prot. | |
1486 | */ | |
b221385b | 1487 | static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot, |
23016969 | 1488 | int node, void *caller) |
1da177e4 LT |
1489 | { |
1490 | struct vm_struct *area; | |
89219d37 CM |
1491 | void *addr; |
1492 | unsigned long real_size = size; | |
1da177e4 LT |
1493 | |
1494 | size = PAGE_ALIGN(size); | |
4481374c | 1495 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
1da177e4 LT |
1496 | return NULL; |
1497 | ||
23016969 CL |
1498 | area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END, |
1499 | node, gfp_mask, caller); | |
1500 | ||
1da177e4 LT |
1501 | if (!area) |
1502 | return NULL; | |
1503 | ||
89219d37 CM |
1504 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1505 | ||
1506 | /* | |
1507 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1508 | * structures allocated in the __get_vm_area_node() function contain | |
1509 | * references to the virtual address of the vmalloc'ed block. | |
1510 | */ | |
1511 | kmemleak_alloc(addr, real_size, 3, gfp_mask); | |
1512 | ||
1513 | return addr; | |
1da177e4 LT |
1514 | } |
1515 | ||
930fc45a CL |
1516 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1517 | { | |
23016969 CL |
1518 | return __vmalloc_node(size, gfp_mask, prot, -1, |
1519 | __builtin_return_address(0)); | |
930fc45a | 1520 | } |
1da177e4 LT |
1521 | EXPORT_SYMBOL(__vmalloc); |
1522 | ||
1523 | /** | |
1524 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1525 | * @size: allocation size |
1da177e4 LT |
1526 | * Allocate enough pages to cover @size from the page level |
1527 | * allocator and map them into contiguous kernel virtual space. | |
1528 | * | |
c1c8897f | 1529 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1530 | * use __vmalloc() instead. |
1531 | */ | |
1532 | void *vmalloc(unsigned long size) | |
1533 | { | |
23016969 CL |
1534 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1535 | -1, __builtin_return_address(0)); | |
1da177e4 | 1536 | } |
1da177e4 LT |
1537 | EXPORT_SYMBOL(vmalloc); |
1538 | ||
83342314 | 1539 | /** |
ead04089 REB |
1540 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1541 | * @size: allocation size | |
83342314 | 1542 | * |
ead04089 REB |
1543 | * The resulting memory area is zeroed so it can be mapped to userspace |
1544 | * without leaking data. | |
83342314 NP |
1545 | */ |
1546 | void *vmalloc_user(unsigned long size) | |
1547 | { | |
1548 | struct vm_struct *area; | |
1549 | void *ret; | |
1550 | ||
84877848 GC |
1551 | ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
1552 | PAGE_KERNEL, -1, __builtin_return_address(0)); | |
2b4ac44e | 1553 | if (ret) { |
db64fe02 | 1554 | area = find_vm_area(ret); |
2b4ac44e | 1555 | area->flags |= VM_USERMAP; |
2b4ac44e | 1556 | } |
83342314 NP |
1557 | return ret; |
1558 | } | |
1559 | EXPORT_SYMBOL(vmalloc_user); | |
1560 | ||
930fc45a CL |
1561 | /** |
1562 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1563 | * @size: allocation size |
d44e0780 | 1564 | * @node: numa node |
930fc45a CL |
1565 | * |
1566 | * Allocate enough pages to cover @size from the page level | |
1567 | * allocator and map them into contiguous kernel virtual space. | |
1568 | * | |
c1c8897f | 1569 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1570 | * use __vmalloc() instead. |
1571 | */ | |
1572 | void *vmalloc_node(unsigned long size, int node) | |
1573 | { | |
23016969 CL |
1574 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
1575 | node, __builtin_return_address(0)); | |
930fc45a CL |
1576 | } |
1577 | EXPORT_SYMBOL(vmalloc_node); | |
1578 | ||
4dc3b16b PP |
1579 | #ifndef PAGE_KERNEL_EXEC |
1580 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1581 | #endif | |
1582 | ||
1da177e4 LT |
1583 | /** |
1584 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1585 | * @size: allocation size |
1586 | * | |
1587 | * Kernel-internal function to allocate enough pages to cover @size | |
1588 | * the page level allocator and map them into contiguous and | |
1589 | * executable kernel virtual space. | |
1590 | * | |
c1c8897f | 1591 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1592 | * use __vmalloc() instead. |
1593 | */ | |
1594 | ||
1da177e4 LT |
1595 | void *vmalloc_exec(unsigned long size) |
1596 | { | |
84877848 GC |
1597 | return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
1598 | -1, __builtin_return_address(0)); | |
1da177e4 LT |
1599 | } |
1600 | ||
0d08e0d3 | 1601 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1602 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1603 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1604 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1605 | #else |
1606 | #define GFP_VMALLOC32 GFP_KERNEL | |
1607 | #endif | |
1608 | ||
1da177e4 LT |
1609 | /** |
1610 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1611 | * @size: allocation size |
1612 | * | |
1613 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1614 | * page level allocator and map them into contiguous kernel virtual space. | |
1615 | */ | |
1616 | void *vmalloc_32(unsigned long size) | |
1617 | { | |
84877848 GC |
1618 | return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL, |
1619 | -1, __builtin_return_address(0)); | |
1da177e4 | 1620 | } |
1da177e4 LT |
1621 | EXPORT_SYMBOL(vmalloc_32); |
1622 | ||
83342314 | 1623 | /** |
ead04089 | 1624 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1625 | * @size: allocation size |
ead04089 REB |
1626 | * |
1627 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1628 | * mapped to userspace without leaking data. | |
83342314 NP |
1629 | */ |
1630 | void *vmalloc_32_user(unsigned long size) | |
1631 | { | |
1632 | struct vm_struct *area; | |
1633 | void *ret; | |
1634 | ||
84877848 GC |
1635 | ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
1636 | -1, __builtin_return_address(0)); | |
2b4ac44e | 1637 | if (ret) { |
db64fe02 | 1638 | area = find_vm_area(ret); |
2b4ac44e | 1639 | area->flags |= VM_USERMAP; |
2b4ac44e | 1640 | } |
83342314 NP |
1641 | return ret; |
1642 | } | |
1643 | EXPORT_SYMBOL(vmalloc_32_user); | |
1644 | ||
d0107eb0 KH |
1645 | /* |
1646 | * small helper routine , copy contents to buf from addr. | |
1647 | * If the page is not present, fill zero. | |
1648 | */ | |
1649 | ||
1650 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
1651 | { | |
1652 | struct page *p; | |
1653 | int copied = 0; | |
1654 | ||
1655 | while (count) { | |
1656 | unsigned long offset, length; | |
1657 | ||
1658 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1659 | length = PAGE_SIZE - offset; | |
1660 | if (length > count) | |
1661 | length = count; | |
1662 | p = vmalloc_to_page(addr); | |
1663 | /* | |
1664 | * To do safe access to this _mapped_ area, we need | |
1665 | * lock. But adding lock here means that we need to add | |
1666 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1667 | * interface, rarely used. Instead of that, we'll use | |
1668 | * kmap() and get small overhead in this access function. | |
1669 | */ | |
1670 | if (p) { | |
1671 | /* | |
1672 | * we can expect USER0 is not used (see vread/vwrite's | |
1673 | * function description) | |
1674 | */ | |
1675 | void *map = kmap_atomic(p, KM_USER0); | |
1676 | memcpy(buf, map + offset, length); | |
1677 | kunmap_atomic(map, KM_USER0); | |
1678 | } else | |
1679 | memset(buf, 0, length); | |
1680 | ||
1681 | addr += length; | |
1682 | buf += length; | |
1683 | copied += length; | |
1684 | count -= length; | |
1685 | } | |
1686 | return copied; | |
1687 | } | |
1688 | ||
1689 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
1690 | { | |
1691 | struct page *p; | |
1692 | int copied = 0; | |
1693 | ||
1694 | while (count) { | |
1695 | unsigned long offset, length; | |
1696 | ||
1697 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1698 | length = PAGE_SIZE - offset; | |
1699 | if (length > count) | |
1700 | length = count; | |
1701 | p = vmalloc_to_page(addr); | |
1702 | /* | |
1703 | * To do safe access to this _mapped_ area, we need | |
1704 | * lock. But adding lock here means that we need to add | |
1705 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1706 | * interface, rarely used. Instead of that, we'll use | |
1707 | * kmap() and get small overhead in this access function. | |
1708 | */ | |
1709 | if (p) { | |
1710 | /* | |
1711 | * we can expect USER0 is not used (see vread/vwrite's | |
1712 | * function description) | |
1713 | */ | |
1714 | void *map = kmap_atomic(p, KM_USER0); | |
1715 | memcpy(map + offset, buf, length); | |
1716 | kunmap_atomic(map, KM_USER0); | |
1717 | } | |
1718 | addr += length; | |
1719 | buf += length; | |
1720 | copied += length; | |
1721 | count -= length; | |
1722 | } | |
1723 | return copied; | |
1724 | } | |
1725 | ||
1726 | /** | |
1727 | * vread() - read vmalloc area in a safe way. | |
1728 | * @buf: buffer for reading data | |
1729 | * @addr: vm address. | |
1730 | * @count: number of bytes to be read. | |
1731 | * | |
1732 | * Returns # of bytes which addr and buf should be increased. | |
1733 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't | |
1734 | * includes any intersect with alive vmalloc area. | |
1735 | * | |
1736 | * This function checks that addr is a valid vmalloc'ed area, and | |
1737 | * copy data from that area to a given buffer. If the given memory range | |
1738 | * of [addr...addr+count) includes some valid address, data is copied to | |
1739 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
1740 | * IOREMAP area is treated as memory hole and no copy is done. | |
1741 | * | |
1742 | * If [addr...addr+count) doesn't includes any intersects with alive | |
1743 | * vm_struct area, returns 0. | |
1744 | * @buf should be kernel's buffer. Because this function uses KM_USER0, | |
1745 | * the caller should guarantee KM_USER0 is not used. | |
1746 | * | |
1747 | * Note: In usual ops, vread() is never necessary because the caller | |
1748 | * should know vmalloc() area is valid and can use memcpy(). | |
1749 | * This is for routines which have to access vmalloc area without | |
1750 | * any informaion, as /dev/kmem. | |
1751 | * | |
1752 | */ | |
1753 | ||
1da177e4 LT |
1754 | long vread(char *buf, char *addr, unsigned long count) |
1755 | { | |
1756 | struct vm_struct *tmp; | |
1757 | char *vaddr, *buf_start = buf; | |
d0107eb0 | 1758 | unsigned long buflen = count; |
1da177e4 LT |
1759 | unsigned long n; |
1760 | ||
1761 | /* Don't allow overflow */ | |
1762 | if ((unsigned long) addr + count < count) | |
1763 | count = -(unsigned long) addr; | |
1764 | ||
1765 | read_lock(&vmlist_lock); | |
d0107eb0 | 1766 | for (tmp = vmlist; count && tmp; tmp = tmp->next) { |
1da177e4 LT |
1767 | vaddr = (char *) tmp->addr; |
1768 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1769 | continue; | |
1770 | while (addr < vaddr) { | |
1771 | if (count == 0) | |
1772 | goto finished; | |
1773 | *buf = '\0'; | |
1774 | buf++; | |
1775 | addr++; | |
1776 | count--; | |
1777 | } | |
1778 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
d0107eb0 KH |
1779 | if (n > count) |
1780 | n = count; | |
1781 | if (!(tmp->flags & VM_IOREMAP)) | |
1782 | aligned_vread(buf, addr, n); | |
1783 | else /* IOREMAP area is treated as memory hole */ | |
1784 | memset(buf, 0, n); | |
1785 | buf += n; | |
1786 | addr += n; | |
1787 | count -= n; | |
1da177e4 LT |
1788 | } |
1789 | finished: | |
1790 | read_unlock(&vmlist_lock); | |
d0107eb0 KH |
1791 | |
1792 | if (buf == buf_start) | |
1793 | return 0; | |
1794 | /* zero-fill memory holes */ | |
1795 | if (buf != buf_start + buflen) | |
1796 | memset(buf, 0, buflen - (buf - buf_start)); | |
1797 | ||
1798 | return buflen; | |
1da177e4 LT |
1799 | } |
1800 | ||
d0107eb0 KH |
1801 | /** |
1802 | * vwrite() - write vmalloc area in a safe way. | |
1803 | * @buf: buffer for source data | |
1804 | * @addr: vm address. | |
1805 | * @count: number of bytes to be read. | |
1806 | * | |
1807 | * Returns # of bytes which addr and buf should be incresed. | |
1808 | * (same number to @count). | |
1809 | * If [addr...addr+count) doesn't includes any intersect with valid | |
1810 | * vmalloc area, returns 0. | |
1811 | * | |
1812 | * This function checks that addr is a valid vmalloc'ed area, and | |
1813 | * copy data from a buffer to the given addr. If specified range of | |
1814 | * [addr...addr+count) includes some valid address, data is copied from | |
1815 | * proper area of @buf. If there are memory holes, no copy to hole. | |
1816 | * IOREMAP area is treated as memory hole and no copy is done. | |
1817 | * | |
1818 | * If [addr...addr+count) doesn't includes any intersects with alive | |
1819 | * vm_struct area, returns 0. | |
1820 | * @buf should be kernel's buffer. Because this function uses KM_USER0, | |
1821 | * the caller should guarantee KM_USER0 is not used. | |
1822 | * | |
1823 | * Note: In usual ops, vwrite() is never necessary because the caller | |
1824 | * should know vmalloc() area is valid and can use memcpy(). | |
1825 | * This is for routines which have to access vmalloc area without | |
1826 | * any informaion, as /dev/kmem. | |
1827 | * | |
1828 | * The caller should guarantee KM_USER1 is not used. | |
1829 | */ | |
1830 | ||
1da177e4 LT |
1831 | long vwrite(char *buf, char *addr, unsigned long count) |
1832 | { | |
1833 | struct vm_struct *tmp; | |
d0107eb0 KH |
1834 | char *vaddr; |
1835 | unsigned long n, buflen; | |
1836 | int copied = 0; | |
1da177e4 LT |
1837 | |
1838 | /* Don't allow overflow */ | |
1839 | if ((unsigned long) addr + count < count) | |
1840 | count = -(unsigned long) addr; | |
d0107eb0 | 1841 | buflen = count; |
1da177e4 LT |
1842 | |
1843 | read_lock(&vmlist_lock); | |
d0107eb0 | 1844 | for (tmp = vmlist; count && tmp; tmp = tmp->next) { |
1da177e4 LT |
1845 | vaddr = (char *) tmp->addr; |
1846 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1847 | continue; | |
1848 | while (addr < vaddr) { | |
1849 | if (count == 0) | |
1850 | goto finished; | |
1851 | buf++; | |
1852 | addr++; | |
1853 | count--; | |
1854 | } | |
1855 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
d0107eb0 KH |
1856 | if (n > count) |
1857 | n = count; | |
1858 | if (!(tmp->flags & VM_IOREMAP)) { | |
1859 | aligned_vwrite(buf, addr, n); | |
1860 | copied++; | |
1861 | } | |
1862 | buf += n; | |
1863 | addr += n; | |
1864 | count -= n; | |
1da177e4 LT |
1865 | } |
1866 | finished: | |
1867 | read_unlock(&vmlist_lock); | |
d0107eb0 KH |
1868 | if (!copied) |
1869 | return 0; | |
1870 | return buflen; | |
1da177e4 | 1871 | } |
83342314 NP |
1872 | |
1873 | /** | |
1874 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
1875 | * @vma: vma to cover (map full range of vma) |
1876 | * @addr: vmalloc memory | |
1877 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
1878 | * |
1879 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
1880 | * |
1881 | * This function checks that addr is a valid vmalloc'ed area, and | |
1882 | * that it is big enough to cover the vma. Will return failure if | |
1883 | * that criteria isn't met. | |
1884 | * | |
72fd4a35 | 1885 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
1886 | */ |
1887 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
1888 | unsigned long pgoff) | |
1889 | { | |
1890 | struct vm_struct *area; | |
1891 | unsigned long uaddr = vma->vm_start; | |
1892 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
1893 | |
1894 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
1895 | return -EINVAL; | |
1896 | ||
db64fe02 | 1897 | area = find_vm_area(addr); |
83342314 | 1898 | if (!area) |
db64fe02 | 1899 | return -EINVAL; |
83342314 NP |
1900 | |
1901 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 1902 | return -EINVAL; |
83342314 NP |
1903 | |
1904 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 1905 | return -EINVAL; |
83342314 NP |
1906 | |
1907 | addr += pgoff << PAGE_SHIFT; | |
1908 | do { | |
1909 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
1910 | int ret; |
1911 | ||
83342314 NP |
1912 | ret = vm_insert_page(vma, uaddr, page); |
1913 | if (ret) | |
1914 | return ret; | |
1915 | ||
1916 | uaddr += PAGE_SIZE; | |
1917 | addr += PAGE_SIZE; | |
1918 | usize -= PAGE_SIZE; | |
1919 | } while (usize > 0); | |
1920 | ||
1921 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
1922 | vma->vm_flags |= VM_RESERVED; | |
1923 | ||
db64fe02 | 1924 | return 0; |
83342314 NP |
1925 | } |
1926 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1927 | ||
1eeb66a1 CH |
1928 | /* |
1929 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
1930 | * have one. | |
1931 | */ | |
1932 | void __attribute__((weak)) vmalloc_sync_all(void) | |
1933 | { | |
1934 | } | |
5f4352fb JF |
1935 | |
1936 | ||
2f569afd | 1937 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
1938 | { |
1939 | /* apply_to_page_range() does all the hard work. */ | |
1940 | return 0; | |
1941 | } | |
1942 | ||
1943 | /** | |
1944 | * alloc_vm_area - allocate a range of kernel address space | |
1945 | * @size: size of the area | |
7682486b RD |
1946 | * |
1947 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
1948 | * |
1949 | * This function reserves a range of kernel address space, and | |
1950 | * allocates pagetables to map that range. No actual mappings | |
1951 | * are created. If the kernel address space is not shared | |
1952 | * between processes, it syncs the pagetable across all | |
1953 | * processes. | |
1954 | */ | |
1955 | struct vm_struct *alloc_vm_area(size_t size) | |
1956 | { | |
1957 | struct vm_struct *area; | |
1958 | ||
23016969 CL |
1959 | area = get_vm_area_caller(size, VM_IOREMAP, |
1960 | __builtin_return_address(0)); | |
5f4352fb JF |
1961 | if (area == NULL) |
1962 | return NULL; | |
1963 | ||
1964 | /* | |
1965 | * This ensures that page tables are constructed for this region | |
1966 | * of kernel virtual address space and mapped into init_mm. | |
1967 | */ | |
1968 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
1969 | area->size, f, NULL)) { | |
1970 | free_vm_area(area); | |
1971 | return NULL; | |
1972 | } | |
1973 | ||
1974 | /* Make sure the pagetables are constructed in process kernel | |
1975 | mappings */ | |
1976 | vmalloc_sync_all(); | |
1977 | ||
1978 | return area; | |
1979 | } | |
1980 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
1981 | ||
1982 | void free_vm_area(struct vm_struct *area) | |
1983 | { | |
1984 | struct vm_struct *ret; | |
1985 | ret = remove_vm_area(area->addr); | |
1986 | BUG_ON(ret != area); | |
1987 | kfree(area); | |
1988 | } | |
1989 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 1990 | |
ca23e405 TH |
1991 | static struct vmap_area *node_to_va(struct rb_node *n) |
1992 | { | |
1993 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; | |
1994 | } | |
1995 | ||
1996 | /** | |
1997 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
1998 | * @end: target address | |
1999 | * @pnext: out arg for the next vmap_area | |
2000 | * @pprev: out arg for the previous vmap_area | |
2001 | * | |
2002 | * Returns: %true if either or both of next and prev are found, | |
2003 | * %false if no vmap_area exists | |
2004 | * | |
2005 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2006 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2007 | */ | |
2008 | static bool pvm_find_next_prev(unsigned long end, | |
2009 | struct vmap_area **pnext, | |
2010 | struct vmap_area **pprev) | |
2011 | { | |
2012 | struct rb_node *n = vmap_area_root.rb_node; | |
2013 | struct vmap_area *va = NULL; | |
2014 | ||
2015 | while (n) { | |
2016 | va = rb_entry(n, struct vmap_area, rb_node); | |
2017 | if (end < va->va_end) | |
2018 | n = n->rb_left; | |
2019 | else if (end > va->va_end) | |
2020 | n = n->rb_right; | |
2021 | else | |
2022 | break; | |
2023 | } | |
2024 | ||
2025 | if (!va) | |
2026 | return false; | |
2027 | ||
2028 | if (va->va_end > end) { | |
2029 | *pnext = va; | |
2030 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2031 | } else { | |
2032 | *pprev = va; | |
2033 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2034 | } | |
2035 | return true; | |
2036 | } | |
2037 | ||
2038 | /** | |
2039 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2040 | * @pnext: in/out arg for the next vmap_area | |
2041 | * @pprev: in/out arg for the previous vmap_area | |
2042 | * @align: alignment | |
2043 | * | |
2044 | * Returns: determined end address | |
2045 | * | |
2046 | * Find the highest aligned address between *@pnext and *@pprev below | |
2047 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2048 | * down address is between the end addresses of the two vmap_areas. | |
2049 | * | |
2050 | * Please note that the address returned by this function may fall | |
2051 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2052 | * that. | |
2053 | */ | |
2054 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2055 | struct vmap_area **pprev, | |
2056 | unsigned long align) | |
2057 | { | |
2058 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2059 | unsigned long addr; | |
2060 | ||
2061 | if (*pnext) | |
2062 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2063 | else | |
2064 | addr = vmalloc_end; | |
2065 | ||
2066 | while (*pprev && (*pprev)->va_end > addr) { | |
2067 | *pnext = *pprev; | |
2068 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2069 | } | |
2070 | ||
2071 | return addr; | |
2072 | } | |
2073 | ||
2074 | /** | |
2075 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2076 | * @offsets: array containing offset of each area | |
2077 | * @sizes: array containing size of each area | |
2078 | * @nr_vms: the number of areas to allocate | |
2079 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
2080 | * @gfp_mask: allocation mask | |
2081 | * | |
2082 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2083 | * vm_structs on success, %NULL on failure | |
2084 | * | |
2085 | * Percpu allocator wants to use congruent vm areas so that it can | |
2086 | * maintain the offsets among percpu areas. This function allocates | |
2087 | * congruent vmalloc areas for it. These areas tend to be scattered | |
2088 | * pretty far, distance between two areas easily going up to | |
2089 | * gigabytes. To avoid interacting with regular vmallocs, these areas | |
2090 | * are allocated from top. | |
2091 | * | |
2092 | * Despite its complicated look, this allocator is rather simple. It | |
2093 | * does everything top-down and scans areas from the end looking for | |
2094 | * matching slot. While scanning, if any of the areas overlaps with | |
2095 | * existing vmap_area, the base address is pulled down to fit the | |
2096 | * area. Scanning is repeated till all the areas fit and then all | |
2097 | * necessary data structres are inserted and the result is returned. | |
2098 | */ | |
2099 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2100 | const size_t *sizes, int nr_vms, | |
2101 | size_t align, gfp_t gfp_mask) | |
2102 | { | |
2103 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2104 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2105 | struct vmap_area **vas, *prev, *next; | |
2106 | struct vm_struct **vms; | |
2107 | int area, area2, last_area, term_area; | |
2108 | unsigned long base, start, end, last_end; | |
2109 | bool purged = false; | |
2110 | ||
2111 | gfp_mask &= GFP_RECLAIM_MASK; | |
2112 | ||
2113 | /* verify parameters and allocate data structures */ | |
2114 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); | |
2115 | for (last_area = 0, area = 0; area < nr_vms; area++) { | |
2116 | start = offsets[area]; | |
2117 | end = start + sizes[area]; | |
2118 | ||
2119 | /* is everything aligned properly? */ | |
2120 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2121 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2122 | ||
2123 | /* detect the area with the highest address */ | |
2124 | if (start > offsets[last_area]) | |
2125 | last_area = area; | |
2126 | ||
2127 | for (area2 = 0; area2 < nr_vms; area2++) { | |
2128 | unsigned long start2 = offsets[area2]; | |
2129 | unsigned long end2 = start2 + sizes[area2]; | |
2130 | ||
2131 | if (area2 == area) | |
2132 | continue; | |
2133 | ||
2134 | BUG_ON(start2 >= start && start2 < end); | |
2135 | BUG_ON(end2 <= end && end2 > start); | |
2136 | } | |
2137 | } | |
2138 | last_end = offsets[last_area] + sizes[last_area]; | |
2139 | ||
2140 | if (vmalloc_end - vmalloc_start < last_end) { | |
2141 | WARN_ON(true); | |
2142 | return NULL; | |
2143 | } | |
2144 | ||
2145 | vms = kzalloc(sizeof(vms[0]) * nr_vms, gfp_mask); | |
2146 | vas = kzalloc(sizeof(vas[0]) * nr_vms, gfp_mask); | |
2147 | if (!vas || !vms) | |
2148 | goto err_free; | |
2149 | ||
2150 | for (area = 0; area < nr_vms; area++) { | |
2151 | vas[area] = kzalloc(sizeof(struct vmap_area), gfp_mask); | |
2152 | vms[area] = kzalloc(sizeof(struct vm_struct), gfp_mask); | |
2153 | if (!vas[area] || !vms[area]) | |
2154 | goto err_free; | |
2155 | } | |
2156 | retry: | |
2157 | spin_lock(&vmap_area_lock); | |
2158 | ||
2159 | /* start scanning - we scan from the top, begin with the last area */ | |
2160 | area = term_area = last_area; | |
2161 | start = offsets[area]; | |
2162 | end = start + sizes[area]; | |
2163 | ||
2164 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2165 | base = vmalloc_end - last_end; | |
2166 | goto found; | |
2167 | } | |
2168 | base = pvm_determine_end(&next, &prev, align) - end; | |
2169 | ||
2170 | while (true) { | |
2171 | BUG_ON(next && next->va_end <= base + end); | |
2172 | BUG_ON(prev && prev->va_end > base + end); | |
2173 | ||
2174 | /* | |
2175 | * base might have underflowed, add last_end before | |
2176 | * comparing. | |
2177 | */ | |
2178 | if (base + last_end < vmalloc_start + last_end) { | |
2179 | spin_unlock(&vmap_area_lock); | |
2180 | if (!purged) { | |
2181 | purge_vmap_area_lazy(); | |
2182 | purged = true; | |
2183 | goto retry; | |
2184 | } | |
2185 | goto err_free; | |
2186 | } | |
2187 | ||
2188 | /* | |
2189 | * If next overlaps, move base downwards so that it's | |
2190 | * right below next and then recheck. | |
2191 | */ | |
2192 | if (next && next->va_start < base + end) { | |
2193 | base = pvm_determine_end(&next, &prev, align) - end; | |
2194 | term_area = area; | |
2195 | continue; | |
2196 | } | |
2197 | ||
2198 | /* | |
2199 | * If prev overlaps, shift down next and prev and move | |
2200 | * base so that it's right below new next and then | |
2201 | * recheck. | |
2202 | */ | |
2203 | if (prev && prev->va_end > base + start) { | |
2204 | next = prev; | |
2205 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2206 | base = pvm_determine_end(&next, &prev, align) - end; | |
2207 | term_area = area; | |
2208 | continue; | |
2209 | } | |
2210 | ||
2211 | /* | |
2212 | * This area fits, move on to the previous one. If | |
2213 | * the previous one is the terminal one, we're done. | |
2214 | */ | |
2215 | area = (area + nr_vms - 1) % nr_vms; | |
2216 | if (area == term_area) | |
2217 | break; | |
2218 | start = offsets[area]; | |
2219 | end = start + sizes[area]; | |
2220 | pvm_find_next_prev(base + end, &next, &prev); | |
2221 | } | |
2222 | found: | |
2223 | /* we've found a fitting base, insert all va's */ | |
2224 | for (area = 0; area < nr_vms; area++) { | |
2225 | struct vmap_area *va = vas[area]; | |
2226 | ||
2227 | va->va_start = base + offsets[area]; | |
2228 | va->va_end = va->va_start + sizes[area]; | |
2229 | __insert_vmap_area(va); | |
2230 | } | |
2231 | ||
2232 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2233 | ||
2234 | spin_unlock(&vmap_area_lock); | |
2235 | ||
2236 | /* insert all vm's */ | |
2237 | for (area = 0; area < nr_vms; area++) | |
2238 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, | |
2239 | pcpu_get_vm_areas); | |
2240 | ||
2241 | kfree(vas); | |
2242 | return vms; | |
2243 | ||
2244 | err_free: | |
2245 | for (area = 0; area < nr_vms; area++) { | |
2246 | if (vas) | |
2247 | kfree(vas[area]); | |
2248 | if (vms) | |
2249 | kfree(vms[area]); | |
2250 | } | |
2251 | kfree(vas); | |
2252 | kfree(vms); | |
2253 | return NULL; | |
2254 | } | |
2255 | ||
2256 | /** | |
2257 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2258 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2259 | * @nr_vms: the number of allocated areas | |
2260 | * | |
2261 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2262 | */ | |
2263 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2264 | { | |
2265 | int i; | |
2266 | ||
2267 | for (i = 0; i < nr_vms; i++) | |
2268 | free_vm_area(vms[i]); | |
2269 | kfree(vms); | |
2270 | } | |
a10aa579 CL |
2271 | |
2272 | #ifdef CONFIG_PROC_FS | |
2273 | static void *s_start(struct seq_file *m, loff_t *pos) | |
2274 | { | |
2275 | loff_t n = *pos; | |
2276 | struct vm_struct *v; | |
2277 | ||
2278 | read_lock(&vmlist_lock); | |
2279 | v = vmlist; | |
2280 | while (n > 0 && v) { | |
2281 | n--; | |
2282 | v = v->next; | |
2283 | } | |
2284 | if (!n) | |
2285 | return v; | |
2286 | ||
2287 | return NULL; | |
2288 | ||
2289 | } | |
2290 | ||
2291 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2292 | { | |
2293 | struct vm_struct *v = p; | |
2294 | ||
2295 | ++*pos; | |
2296 | return v->next; | |
2297 | } | |
2298 | ||
2299 | static void s_stop(struct seq_file *m, void *p) | |
2300 | { | |
2301 | read_unlock(&vmlist_lock); | |
2302 | } | |
2303 | ||
a47a126a ED |
2304 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2305 | { | |
2306 | if (NUMA_BUILD) { | |
2307 | unsigned int nr, *counters = m->private; | |
2308 | ||
2309 | if (!counters) | |
2310 | return; | |
2311 | ||
2312 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); | |
2313 | ||
2314 | for (nr = 0; nr < v->nr_pages; nr++) | |
2315 | counters[page_to_nid(v->pages[nr])]++; | |
2316 | ||
2317 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2318 | if (counters[nr]) | |
2319 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2320 | } | |
2321 | } | |
2322 | ||
a10aa579 CL |
2323 | static int s_show(struct seq_file *m, void *p) |
2324 | { | |
2325 | struct vm_struct *v = p; | |
2326 | ||
2327 | seq_printf(m, "0x%p-0x%p %7ld", | |
2328 | v->addr, v->addr + v->size, v->size); | |
2329 | ||
23016969 | 2330 | if (v->caller) { |
9c246247 | 2331 | char buff[KSYM_SYMBOL_LEN]; |
23016969 CL |
2332 | |
2333 | seq_putc(m, ' '); | |
2334 | sprint_symbol(buff, (unsigned long)v->caller); | |
2335 | seq_puts(m, buff); | |
2336 | } | |
2337 | ||
a10aa579 CL |
2338 | if (v->nr_pages) |
2339 | seq_printf(m, " pages=%d", v->nr_pages); | |
2340 | ||
2341 | if (v->phys_addr) | |
2342 | seq_printf(m, " phys=%lx", v->phys_addr); | |
2343 | ||
2344 | if (v->flags & VM_IOREMAP) | |
2345 | seq_printf(m, " ioremap"); | |
2346 | ||
2347 | if (v->flags & VM_ALLOC) | |
2348 | seq_printf(m, " vmalloc"); | |
2349 | ||
2350 | if (v->flags & VM_MAP) | |
2351 | seq_printf(m, " vmap"); | |
2352 | ||
2353 | if (v->flags & VM_USERMAP) | |
2354 | seq_printf(m, " user"); | |
2355 | ||
2356 | if (v->flags & VM_VPAGES) | |
2357 | seq_printf(m, " vpages"); | |
2358 | ||
a47a126a | 2359 | show_numa_info(m, v); |
a10aa579 CL |
2360 | seq_putc(m, '\n'); |
2361 | return 0; | |
2362 | } | |
2363 | ||
5f6a6a9c | 2364 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2365 | .start = s_start, |
2366 | .next = s_next, | |
2367 | .stop = s_stop, | |
2368 | .show = s_show, | |
2369 | }; | |
5f6a6a9c AD |
2370 | |
2371 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2372 | { | |
2373 | unsigned int *ptr = NULL; | |
2374 | int ret; | |
2375 | ||
2376 | if (NUMA_BUILD) | |
2377 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); | |
2378 | ret = seq_open(file, &vmalloc_op); | |
2379 | if (!ret) { | |
2380 | struct seq_file *m = file->private_data; | |
2381 | m->private = ptr; | |
2382 | } else | |
2383 | kfree(ptr); | |
2384 | return ret; | |
2385 | } | |
2386 | ||
2387 | static const struct file_operations proc_vmalloc_operations = { | |
2388 | .open = vmalloc_open, | |
2389 | .read = seq_read, | |
2390 | .llseek = seq_lseek, | |
2391 | .release = seq_release_private, | |
2392 | }; | |
2393 | ||
2394 | static int __init proc_vmalloc_init(void) | |
2395 | { | |
2396 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2397 | return 0; | |
2398 | } | |
2399 | module_init(proc_vmalloc_init); | |
a10aa579 CL |
2400 | #endif |
2401 |