]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmalloc.c | |
3 | * | |
4 | * Copyright (C) 1993 Linus Torvalds | |
5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
6 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
7 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
1da177e4 LT |
9 | */ |
10 | ||
db64fe02 | 11 | #include <linux/vmalloc.h> |
1da177e4 LT |
12 | #include <linux/mm.h> |
13 | #include <linux/module.h> | |
14 | #include <linux/highmem.h> | |
d43c36dc | 15 | #include <linux/sched.h> |
1da177e4 LT |
16 | #include <linux/slab.h> |
17 | #include <linux/spinlock.h> | |
18 | #include <linux/interrupt.h> | |
5f6a6a9c | 19 | #include <linux/proc_fs.h> |
a10aa579 | 20 | #include <linux/seq_file.h> |
3ac7fe5a | 21 | #include <linux/debugobjects.h> |
23016969 | 22 | #include <linux/kallsyms.h> |
db64fe02 NP |
23 | #include <linux/list.h> |
24 | #include <linux/rbtree.h> | |
25 | #include <linux/radix-tree.h> | |
26 | #include <linux/rcupdate.h> | |
f0aa6617 | 27 | #include <linux/pfn.h> |
89219d37 | 28 | #include <linux/kmemleak.h> |
db64fe02 | 29 | #include <asm/atomic.h> |
1da177e4 LT |
30 | #include <asm/uaccess.h> |
31 | #include <asm/tlbflush.h> | |
2dca6999 | 32 | #include <asm/shmparam.h> |
1da177e4 | 33 | |
db64fe02 | 34 | /*** Page table manipulation functions ***/ |
b221385b | 35 | |
1da177e4 LT |
36 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
37 | { | |
38 | pte_t *pte; | |
39 | ||
40 | pte = pte_offset_kernel(pmd, addr); | |
41 | do { | |
42 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
43 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
44 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
45 | } | |
46 | ||
db64fe02 | 47 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
48 | { |
49 | pmd_t *pmd; | |
50 | unsigned long next; | |
51 | ||
52 | pmd = pmd_offset(pud, addr); | |
53 | do { | |
54 | next = pmd_addr_end(addr, end); | |
55 | if (pmd_none_or_clear_bad(pmd)) | |
56 | continue; | |
57 | vunmap_pte_range(pmd, addr, next); | |
58 | } while (pmd++, addr = next, addr != end); | |
59 | } | |
60 | ||
db64fe02 | 61 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
1da177e4 LT |
62 | { |
63 | pud_t *pud; | |
64 | unsigned long next; | |
65 | ||
66 | pud = pud_offset(pgd, addr); | |
67 | do { | |
68 | next = pud_addr_end(addr, end); | |
69 | if (pud_none_or_clear_bad(pud)) | |
70 | continue; | |
71 | vunmap_pmd_range(pud, addr, next); | |
72 | } while (pud++, addr = next, addr != end); | |
73 | } | |
74 | ||
db64fe02 | 75 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
1da177e4 LT |
76 | { |
77 | pgd_t *pgd; | |
78 | unsigned long next; | |
1da177e4 LT |
79 | |
80 | BUG_ON(addr >= end); | |
81 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
82 | do { |
83 | next = pgd_addr_end(addr, end); | |
84 | if (pgd_none_or_clear_bad(pgd)) | |
85 | continue; | |
86 | vunmap_pud_range(pgd, addr, next); | |
87 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
88 | } |
89 | ||
90 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 91 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
92 | { |
93 | pte_t *pte; | |
94 | ||
db64fe02 NP |
95 | /* |
96 | * nr is a running index into the array which helps higher level | |
97 | * callers keep track of where we're up to. | |
98 | */ | |
99 | ||
872fec16 | 100 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
101 | if (!pte) |
102 | return -ENOMEM; | |
103 | do { | |
db64fe02 NP |
104 | struct page *page = pages[*nr]; |
105 | ||
106 | if (WARN_ON(!pte_none(*pte))) | |
107 | return -EBUSY; | |
108 | if (WARN_ON(!page)) | |
1da177e4 LT |
109 | return -ENOMEM; |
110 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 111 | (*nr)++; |
1da177e4 LT |
112 | } while (pte++, addr += PAGE_SIZE, addr != end); |
113 | return 0; | |
114 | } | |
115 | ||
db64fe02 NP |
116 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
117 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
118 | { |
119 | pmd_t *pmd; | |
120 | unsigned long next; | |
121 | ||
122 | pmd = pmd_alloc(&init_mm, pud, addr); | |
123 | if (!pmd) | |
124 | return -ENOMEM; | |
125 | do { | |
126 | next = pmd_addr_end(addr, end); | |
db64fe02 | 127 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
128 | return -ENOMEM; |
129 | } while (pmd++, addr = next, addr != end); | |
130 | return 0; | |
131 | } | |
132 | ||
db64fe02 NP |
133 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
134 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
135 | { |
136 | pud_t *pud; | |
137 | unsigned long next; | |
138 | ||
139 | pud = pud_alloc(&init_mm, pgd, addr); | |
140 | if (!pud) | |
141 | return -ENOMEM; | |
142 | do { | |
143 | next = pud_addr_end(addr, end); | |
db64fe02 | 144 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
145 | return -ENOMEM; |
146 | } while (pud++, addr = next, addr != end); | |
147 | return 0; | |
148 | } | |
149 | ||
db64fe02 NP |
150 | /* |
151 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and | |
152 | * will have pfns corresponding to the "pages" array. | |
153 | * | |
154 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] | |
155 | */ | |
8fc48985 TH |
156 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
157 | pgprot_t prot, struct page **pages) | |
1da177e4 LT |
158 | { |
159 | pgd_t *pgd; | |
160 | unsigned long next; | |
2e4e27c7 | 161 | unsigned long addr = start; |
db64fe02 NP |
162 | int err = 0; |
163 | int nr = 0; | |
1da177e4 LT |
164 | |
165 | BUG_ON(addr >= end); | |
166 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
167 | do { |
168 | next = pgd_addr_end(addr, end); | |
db64fe02 | 169 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 170 | if (err) |
bf88c8c8 | 171 | return err; |
1da177e4 | 172 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 173 | |
db64fe02 | 174 | return nr; |
1da177e4 LT |
175 | } |
176 | ||
8fc48985 TH |
177 | static int vmap_page_range(unsigned long start, unsigned long end, |
178 | pgprot_t prot, struct page **pages) | |
179 | { | |
180 | int ret; | |
181 | ||
182 | ret = vmap_page_range_noflush(start, end, prot, pages); | |
183 | flush_cache_vmap(start, end); | |
184 | return ret; | |
185 | } | |
186 | ||
81ac3ad9 | 187 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
188 | { |
189 | /* | |
ab4f2ee1 | 190 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
191 | * and fall back on vmalloc() if that fails. Others |
192 | * just put it in the vmalloc space. | |
193 | */ | |
194 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
195 | unsigned long addr = (unsigned long)x; | |
196 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
197 | return 1; | |
198 | #endif | |
199 | return is_vmalloc_addr(x); | |
200 | } | |
201 | ||
48667e7a | 202 | /* |
db64fe02 | 203 | * Walk a vmap address to the struct page it maps. |
48667e7a | 204 | */ |
b3bdda02 | 205 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
206 | { |
207 | unsigned long addr = (unsigned long) vmalloc_addr; | |
208 | struct page *page = NULL; | |
209 | pgd_t *pgd = pgd_offset_k(addr); | |
48667e7a | 210 | |
7aa413de IM |
211 | /* |
212 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
213 | * architectures that do not vmalloc module space | |
214 | */ | |
73bdf0a6 | 215 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 216 | |
48667e7a | 217 | if (!pgd_none(*pgd)) { |
db64fe02 | 218 | pud_t *pud = pud_offset(pgd, addr); |
48667e7a | 219 | if (!pud_none(*pud)) { |
db64fe02 | 220 | pmd_t *pmd = pmd_offset(pud, addr); |
48667e7a | 221 | if (!pmd_none(*pmd)) { |
db64fe02 NP |
222 | pte_t *ptep, pte; |
223 | ||
48667e7a CL |
224 | ptep = pte_offset_map(pmd, addr); |
225 | pte = *ptep; | |
226 | if (pte_present(pte)) | |
227 | page = pte_page(pte); | |
228 | pte_unmap(ptep); | |
229 | } | |
230 | } | |
231 | } | |
232 | return page; | |
233 | } | |
234 | EXPORT_SYMBOL(vmalloc_to_page); | |
235 | ||
236 | /* | |
237 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
238 | */ | |
b3bdda02 | 239 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a CL |
240 | { |
241 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
242 | } | |
243 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
244 | ||
db64fe02 NP |
245 | |
246 | /*** Global kva allocator ***/ | |
247 | ||
248 | #define VM_LAZY_FREE 0x01 | |
249 | #define VM_LAZY_FREEING 0x02 | |
250 | #define VM_VM_AREA 0x04 | |
251 | ||
252 | struct vmap_area { | |
253 | unsigned long va_start; | |
254 | unsigned long va_end; | |
255 | unsigned long flags; | |
256 | struct rb_node rb_node; /* address sorted rbtree */ | |
257 | struct list_head list; /* address sorted list */ | |
258 | struct list_head purge_list; /* "lazy purge" list */ | |
259 | void *private; | |
260 | struct rcu_head rcu_head; | |
261 | }; | |
262 | ||
263 | static DEFINE_SPINLOCK(vmap_area_lock); | |
264 | static struct rb_root vmap_area_root = RB_ROOT; | |
265 | static LIST_HEAD(vmap_area_list); | |
ca23e405 | 266 | static unsigned long vmap_area_pcpu_hole; |
db64fe02 NP |
267 | |
268 | static struct vmap_area *__find_vmap_area(unsigned long addr) | |
1da177e4 | 269 | { |
db64fe02 NP |
270 | struct rb_node *n = vmap_area_root.rb_node; |
271 | ||
272 | while (n) { | |
273 | struct vmap_area *va; | |
274 | ||
275 | va = rb_entry(n, struct vmap_area, rb_node); | |
276 | if (addr < va->va_start) | |
277 | n = n->rb_left; | |
278 | else if (addr > va->va_start) | |
279 | n = n->rb_right; | |
280 | else | |
281 | return va; | |
282 | } | |
283 | ||
284 | return NULL; | |
285 | } | |
286 | ||
287 | static void __insert_vmap_area(struct vmap_area *va) | |
288 | { | |
289 | struct rb_node **p = &vmap_area_root.rb_node; | |
290 | struct rb_node *parent = NULL; | |
291 | struct rb_node *tmp; | |
292 | ||
293 | while (*p) { | |
170168d0 | 294 | struct vmap_area *tmp_va; |
db64fe02 NP |
295 | |
296 | parent = *p; | |
170168d0 NK |
297 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
298 | if (va->va_start < tmp_va->va_end) | |
db64fe02 | 299 | p = &(*p)->rb_left; |
170168d0 | 300 | else if (va->va_end > tmp_va->va_start) |
db64fe02 NP |
301 | p = &(*p)->rb_right; |
302 | else | |
303 | BUG(); | |
304 | } | |
305 | ||
306 | rb_link_node(&va->rb_node, parent, p); | |
307 | rb_insert_color(&va->rb_node, &vmap_area_root); | |
308 | ||
309 | /* address-sort this list so it is usable like the vmlist */ | |
310 | tmp = rb_prev(&va->rb_node); | |
311 | if (tmp) { | |
312 | struct vmap_area *prev; | |
313 | prev = rb_entry(tmp, struct vmap_area, rb_node); | |
314 | list_add_rcu(&va->list, &prev->list); | |
315 | } else | |
316 | list_add_rcu(&va->list, &vmap_area_list); | |
317 | } | |
318 | ||
319 | static void purge_vmap_area_lazy(void); | |
320 | ||
321 | /* | |
322 | * Allocate a region of KVA of the specified size and alignment, within the | |
323 | * vstart and vend. | |
324 | */ | |
325 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
326 | unsigned long align, | |
327 | unsigned long vstart, unsigned long vend, | |
328 | int node, gfp_t gfp_mask) | |
329 | { | |
330 | struct vmap_area *va; | |
331 | struct rb_node *n; | |
1da177e4 | 332 | unsigned long addr; |
db64fe02 NP |
333 | int purged = 0; |
334 | ||
7766970c | 335 | BUG_ON(!size); |
db64fe02 NP |
336 | BUG_ON(size & ~PAGE_MASK); |
337 | ||
db64fe02 NP |
338 | va = kmalloc_node(sizeof(struct vmap_area), |
339 | gfp_mask & GFP_RECLAIM_MASK, node); | |
340 | if (unlikely(!va)) | |
341 | return ERR_PTR(-ENOMEM); | |
342 | ||
343 | retry: | |
0ae15132 GC |
344 | addr = ALIGN(vstart, align); |
345 | ||
db64fe02 | 346 | spin_lock(&vmap_area_lock); |
7766970c NP |
347 | if (addr + size - 1 < addr) |
348 | goto overflow; | |
349 | ||
db64fe02 NP |
350 | /* XXX: could have a last_hole cache */ |
351 | n = vmap_area_root.rb_node; | |
352 | if (n) { | |
353 | struct vmap_area *first = NULL; | |
354 | ||
355 | do { | |
356 | struct vmap_area *tmp; | |
357 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
358 | if (tmp->va_end >= addr) { | |
359 | if (!first && tmp->va_start < addr + size) | |
360 | first = tmp; | |
361 | n = n->rb_left; | |
362 | } else { | |
363 | first = tmp; | |
364 | n = n->rb_right; | |
365 | } | |
366 | } while (n); | |
367 | ||
368 | if (!first) | |
369 | goto found; | |
370 | ||
371 | if (first->va_end < addr) { | |
372 | n = rb_next(&first->rb_node); | |
373 | if (n) | |
374 | first = rb_entry(n, struct vmap_area, rb_node); | |
375 | else | |
376 | goto found; | |
377 | } | |
378 | ||
f011c2da | 379 | while (addr + size > first->va_start && addr + size <= vend) { |
db64fe02 | 380 | addr = ALIGN(first->va_end + PAGE_SIZE, align); |
7766970c NP |
381 | if (addr + size - 1 < addr) |
382 | goto overflow; | |
db64fe02 NP |
383 | |
384 | n = rb_next(&first->rb_node); | |
385 | if (n) | |
386 | first = rb_entry(n, struct vmap_area, rb_node); | |
387 | else | |
388 | goto found; | |
389 | } | |
390 | } | |
391 | found: | |
392 | if (addr + size > vend) { | |
7766970c | 393 | overflow: |
db64fe02 NP |
394 | spin_unlock(&vmap_area_lock); |
395 | if (!purged) { | |
396 | purge_vmap_area_lazy(); | |
397 | purged = 1; | |
398 | goto retry; | |
399 | } | |
400 | if (printk_ratelimit()) | |
c1279c4e GC |
401 | printk(KERN_WARNING |
402 | "vmap allocation for size %lu failed: " | |
403 | "use vmalloc=<size> to increase size.\n", size); | |
2498ce42 | 404 | kfree(va); |
db64fe02 NP |
405 | return ERR_PTR(-EBUSY); |
406 | } | |
407 | ||
408 | BUG_ON(addr & (align-1)); | |
409 | ||
410 | va->va_start = addr; | |
411 | va->va_end = addr + size; | |
412 | va->flags = 0; | |
413 | __insert_vmap_area(va); | |
414 | spin_unlock(&vmap_area_lock); | |
415 | ||
416 | return va; | |
417 | } | |
418 | ||
419 | static void rcu_free_va(struct rcu_head *head) | |
420 | { | |
421 | struct vmap_area *va = container_of(head, struct vmap_area, rcu_head); | |
422 | ||
423 | kfree(va); | |
424 | } | |
425 | ||
426 | static void __free_vmap_area(struct vmap_area *va) | |
427 | { | |
428 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); | |
429 | rb_erase(&va->rb_node, &vmap_area_root); | |
430 | RB_CLEAR_NODE(&va->rb_node); | |
431 | list_del_rcu(&va->list); | |
432 | ||
ca23e405 TH |
433 | /* |
434 | * Track the highest possible candidate for pcpu area | |
435 | * allocation. Areas outside of vmalloc area can be returned | |
436 | * here too, consider only end addresses which fall inside | |
437 | * vmalloc area proper. | |
438 | */ | |
439 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) | |
440 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); | |
441 | ||
db64fe02 NP |
442 | call_rcu(&va->rcu_head, rcu_free_va); |
443 | } | |
444 | ||
445 | /* | |
446 | * Free a region of KVA allocated by alloc_vmap_area | |
447 | */ | |
448 | static void free_vmap_area(struct vmap_area *va) | |
449 | { | |
450 | spin_lock(&vmap_area_lock); | |
451 | __free_vmap_area(va); | |
452 | spin_unlock(&vmap_area_lock); | |
453 | } | |
454 | ||
455 | /* | |
456 | * Clear the pagetable entries of a given vmap_area | |
457 | */ | |
458 | static void unmap_vmap_area(struct vmap_area *va) | |
459 | { | |
460 | vunmap_page_range(va->va_start, va->va_end); | |
461 | } | |
462 | ||
cd52858c NP |
463 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
464 | { | |
465 | /* | |
466 | * Unmap page tables and force a TLB flush immediately if | |
467 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free | |
468 | * bugs similarly to those in linear kernel virtual address | |
469 | * space after a page has been freed. | |
470 | * | |
471 | * All the lazy freeing logic is still retained, in order to | |
472 | * minimise intrusiveness of this debugging feature. | |
473 | * | |
474 | * This is going to be *slow* (linear kernel virtual address | |
475 | * debugging doesn't do a broadcast TLB flush so it is a lot | |
476 | * faster). | |
477 | */ | |
478 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
479 | vunmap_page_range(start, end); | |
480 | flush_tlb_kernel_range(start, end); | |
481 | #endif | |
482 | } | |
483 | ||
db64fe02 NP |
484 | /* |
485 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
486 | * before attempting to purge with a TLB flush. | |
487 | * | |
488 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
489 | * and take slightly longer to purge, but it will linearly reduce the number of | |
490 | * global TLB flushes that must be performed. It would seem natural to scale | |
491 | * this number up linearly with the number of CPUs (because vmapping activity | |
492 | * could also scale linearly with the number of CPUs), however it is likely | |
493 | * that in practice, workloads might be constrained in other ways that mean | |
494 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
495 | * conservative and not introduce a big latency on huge systems, so go with | |
496 | * a less aggressive log scale. It will still be an improvement over the old | |
497 | * code, and it will be simple to change the scale factor if we find that it | |
498 | * becomes a problem on bigger systems. | |
499 | */ | |
500 | static unsigned long lazy_max_pages(void) | |
501 | { | |
502 | unsigned int log; | |
503 | ||
504 | log = fls(num_online_cpus()); | |
505 | ||
506 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
507 | } | |
508 | ||
509 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); | |
510 | ||
02b709df NP |
511 | /* for per-CPU blocks */ |
512 | static void purge_fragmented_blocks_allcpus(void); | |
513 | ||
3ee48b6a CW |
514 | /* |
515 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
516 | * immediately freed. | |
517 | */ | |
518 | void set_iounmap_nonlazy(void) | |
519 | { | |
520 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); | |
521 | } | |
522 | ||
db64fe02 NP |
523 | /* |
524 | * Purges all lazily-freed vmap areas. | |
525 | * | |
526 | * If sync is 0 then don't purge if there is already a purge in progress. | |
527 | * If force_flush is 1, then flush kernel TLBs between *start and *end even | |
528 | * if we found no lazy vmap areas to unmap (callers can use this to optimise | |
529 | * their own TLB flushing). | |
530 | * Returns with *start = min(*start, lowest purged address) | |
531 | * *end = max(*end, highest purged address) | |
532 | */ | |
533 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, | |
534 | int sync, int force_flush) | |
535 | { | |
46666d8a | 536 | static DEFINE_SPINLOCK(purge_lock); |
db64fe02 NP |
537 | LIST_HEAD(valist); |
538 | struct vmap_area *va; | |
cbb76676 | 539 | struct vmap_area *n_va; |
db64fe02 NP |
540 | int nr = 0; |
541 | ||
542 | /* | |
543 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers | |
544 | * should not expect such behaviour. This just simplifies locking for | |
545 | * the case that isn't actually used at the moment anyway. | |
546 | */ | |
547 | if (!sync && !force_flush) { | |
46666d8a | 548 | if (!spin_trylock(&purge_lock)) |
db64fe02 NP |
549 | return; |
550 | } else | |
46666d8a | 551 | spin_lock(&purge_lock); |
db64fe02 | 552 | |
02b709df NP |
553 | if (sync) |
554 | purge_fragmented_blocks_allcpus(); | |
555 | ||
db64fe02 NP |
556 | rcu_read_lock(); |
557 | list_for_each_entry_rcu(va, &vmap_area_list, list) { | |
558 | if (va->flags & VM_LAZY_FREE) { | |
559 | if (va->va_start < *start) | |
560 | *start = va->va_start; | |
561 | if (va->va_end > *end) | |
562 | *end = va->va_end; | |
563 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; | |
db64fe02 NP |
564 | list_add_tail(&va->purge_list, &valist); |
565 | va->flags |= VM_LAZY_FREEING; | |
566 | va->flags &= ~VM_LAZY_FREE; | |
567 | } | |
568 | } | |
569 | rcu_read_unlock(); | |
570 | ||
88f50044 | 571 | if (nr) |
db64fe02 | 572 | atomic_sub(nr, &vmap_lazy_nr); |
db64fe02 NP |
573 | |
574 | if (nr || force_flush) | |
575 | flush_tlb_kernel_range(*start, *end); | |
576 | ||
577 | if (nr) { | |
578 | spin_lock(&vmap_area_lock); | |
cbb76676 | 579 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
db64fe02 NP |
580 | __free_vmap_area(va); |
581 | spin_unlock(&vmap_area_lock); | |
582 | } | |
46666d8a | 583 | spin_unlock(&purge_lock); |
db64fe02 NP |
584 | } |
585 | ||
496850e5 NP |
586 | /* |
587 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
588 | * is already purging. | |
589 | */ | |
590 | static void try_purge_vmap_area_lazy(void) | |
591 | { | |
592 | unsigned long start = ULONG_MAX, end = 0; | |
593 | ||
594 | __purge_vmap_area_lazy(&start, &end, 0, 0); | |
595 | } | |
596 | ||
db64fe02 NP |
597 | /* |
598 | * Kick off a purge of the outstanding lazy areas. | |
599 | */ | |
600 | static void purge_vmap_area_lazy(void) | |
601 | { | |
602 | unsigned long start = ULONG_MAX, end = 0; | |
603 | ||
496850e5 | 604 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
db64fe02 NP |
605 | } |
606 | ||
607 | /* | |
64141da5 JF |
608 | * Free a vmap area, caller ensuring that the area has been unmapped |
609 | * and flush_cache_vunmap had been called for the correct range | |
610 | * previously. | |
db64fe02 | 611 | */ |
64141da5 | 612 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 NP |
613 | { |
614 | va->flags |= VM_LAZY_FREE; | |
615 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); | |
616 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) | |
496850e5 | 617 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
618 | } |
619 | ||
64141da5 JF |
620 | /* |
621 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been | |
622 | * called for the correct range previously. | |
623 | */ | |
624 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) | |
625 | { | |
626 | unmap_vmap_area(va); | |
627 | free_vmap_area_noflush(va); | |
628 | } | |
629 | ||
b29acbdc NP |
630 | /* |
631 | * Free and unmap a vmap area | |
632 | */ | |
633 | static void free_unmap_vmap_area(struct vmap_area *va) | |
634 | { | |
635 | flush_cache_vunmap(va->va_start, va->va_end); | |
636 | free_unmap_vmap_area_noflush(va); | |
637 | } | |
638 | ||
db64fe02 NP |
639 | static struct vmap_area *find_vmap_area(unsigned long addr) |
640 | { | |
641 | struct vmap_area *va; | |
642 | ||
643 | spin_lock(&vmap_area_lock); | |
644 | va = __find_vmap_area(addr); | |
645 | spin_unlock(&vmap_area_lock); | |
646 | ||
647 | return va; | |
648 | } | |
649 | ||
650 | static void free_unmap_vmap_area_addr(unsigned long addr) | |
651 | { | |
652 | struct vmap_area *va; | |
653 | ||
654 | va = find_vmap_area(addr); | |
655 | BUG_ON(!va); | |
656 | free_unmap_vmap_area(va); | |
657 | } | |
658 | ||
659 | ||
660 | /*** Per cpu kva allocator ***/ | |
661 | ||
662 | /* | |
663 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
664 | * room for at least 16 percpu vmap blocks per CPU. | |
665 | */ | |
666 | /* | |
667 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
668 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
669 | * instead (we just need a rough idea) | |
670 | */ | |
671 | #if BITS_PER_LONG == 32 | |
672 | #define VMALLOC_SPACE (128UL*1024*1024) | |
673 | #else | |
674 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
675 | #endif | |
676 | ||
677 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
678 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
679 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
680 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
681 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
682 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
683 | #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
684 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
685 | VMALLOC_PAGES / NR_CPUS / 16)) | |
686 | ||
687 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
688 | ||
9b463334 JF |
689 | static bool vmap_initialized __read_mostly = false; |
690 | ||
db64fe02 NP |
691 | struct vmap_block_queue { |
692 | spinlock_t lock; | |
693 | struct list_head free; | |
db64fe02 NP |
694 | }; |
695 | ||
696 | struct vmap_block { | |
697 | spinlock_t lock; | |
698 | struct vmap_area *va; | |
699 | struct vmap_block_queue *vbq; | |
700 | unsigned long free, dirty; | |
701 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); | |
702 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); | |
de560423 NP |
703 | struct list_head free_list; |
704 | struct rcu_head rcu_head; | |
02b709df | 705 | struct list_head purge; |
db64fe02 NP |
706 | }; |
707 | ||
708 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
709 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
710 | ||
711 | /* | |
712 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
713 | * in the free path. Could get rid of this if we change the API to return a | |
714 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
715 | */ | |
716 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
717 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
718 | ||
719 | /* | |
720 | * We should probably have a fallback mechanism to allocate virtual memory | |
721 | * out of partially filled vmap blocks. However vmap block sizing should be | |
722 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
723 | * big problem. | |
724 | */ | |
725 | ||
726 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
727 | { | |
728 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
729 | addr /= VMAP_BLOCK_SIZE; | |
730 | return addr; | |
731 | } | |
732 | ||
733 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) | |
734 | { | |
735 | struct vmap_block_queue *vbq; | |
736 | struct vmap_block *vb; | |
737 | struct vmap_area *va; | |
738 | unsigned long vb_idx; | |
739 | int node, err; | |
740 | ||
741 | node = numa_node_id(); | |
742 | ||
743 | vb = kmalloc_node(sizeof(struct vmap_block), | |
744 | gfp_mask & GFP_RECLAIM_MASK, node); | |
745 | if (unlikely(!vb)) | |
746 | return ERR_PTR(-ENOMEM); | |
747 | ||
748 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
749 | VMALLOC_START, VMALLOC_END, | |
750 | node, gfp_mask); | |
ddf9c6d4 | 751 | if (IS_ERR(va)) { |
db64fe02 | 752 | kfree(vb); |
e7d86340 | 753 | return ERR_CAST(va); |
db64fe02 NP |
754 | } |
755 | ||
756 | err = radix_tree_preload(gfp_mask); | |
757 | if (unlikely(err)) { | |
758 | kfree(vb); | |
759 | free_vmap_area(va); | |
760 | return ERR_PTR(err); | |
761 | } | |
762 | ||
763 | spin_lock_init(&vb->lock); | |
764 | vb->va = va; | |
765 | vb->free = VMAP_BBMAP_BITS; | |
766 | vb->dirty = 0; | |
767 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); | |
768 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); | |
769 | INIT_LIST_HEAD(&vb->free_list); | |
db64fe02 NP |
770 | |
771 | vb_idx = addr_to_vb_idx(va->va_start); | |
772 | spin_lock(&vmap_block_tree_lock); | |
773 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
774 | spin_unlock(&vmap_block_tree_lock); | |
775 | BUG_ON(err); | |
776 | radix_tree_preload_end(); | |
777 | ||
778 | vbq = &get_cpu_var(vmap_block_queue); | |
779 | vb->vbq = vbq; | |
780 | spin_lock(&vbq->lock); | |
de560423 | 781 | list_add_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 782 | spin_unlock(&vbq->lock); |
3f04ba85 | 783 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
784 | |
785 | return vb; | |
786 | } | |
787 | ||
788 | static void rcu_free_vb(struct rcu_head *head) | |
789 | { | |
790 | struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head); | |
791 | ||
792 | kfree(vb); | |
793 | } | |
794 | ||
795 | static void free_vmap_block(struct vmap_block *vb) | |
796 | { | |
797 | struct vmap_block *tmp; | |
798 | unsigned long vb_idx; | |
799 | ||
db64fe02 NP |
800 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
801 | spin_lock(&vmap_block_tree_lock); | |
802 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
803 | spin_unlock(&vmap_block_tree_lock); | |
804 | BUG_ON(tmp != vb); | |
805 | ||
64141da5 | 806 | free_vmap_area_noflush(vb->va); |
db64fe02 NP |
807 | call_rcu(&vb->rcu_head, rcu_free_vb); |
808 | } | |
809 | ||
02b709df NP |
810 | static void purge_fragmented_blocks(int cpu) |
811 | { | |
812 | LIST_HEAD(purge); | |
813 | struct vmap_block *vb; | |
814 | struct vmap_block *n_vb; | |
815 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
816 | ||
817 | rcu_read_lock(); | |
818 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
819 | ||
820 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
821 | continue; | |
822 | ||
823 | spin_lock(&vb->lock); | |
824 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
825 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
826 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
827 | bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS); | |
828 | bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS); | |
829 | spin_lock(&vbq->lock); | |
830 | list_del_rcu(&vb->free_list); | |
831 | spin_unlock(&vbq->lock); | |
832 | spin_unlock(&vb->lock); | |
833 | list_add_tail(&vb->purge, &purge); | |
834 | } else | |
835 | spin_unlock(&vb->lock); | |
836 | } | |
837 | rcu_read_unlock(); | |
838 | ||
839 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
840 | list_del(&vb->purge); | |
841 | free_vmap_block(vb); | |
842 | } | |
843 | } | |
844 | ||
845 | static void purge_fragmented_blocks_thiscpu(void) | |
846 | { | |
847 | purge_fragmented_blocks(smp_processor_id()); | |
848 | } | |
849 | ||
850 | static void purge_fragmented_blocks_allcpus(void) | |
851 | { | |
852 | int cpu; | |
853 | ||
854 | for_each_possible_cpu(cpu) | |
855 | purge_fragmented_blocks(cpu); | |
856 | } | |
857 | ||
db64fe02 NP |
858 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
859 | { | |
860 | struct vmap_block_queue *vbq; | |
861 | struct vmap_block *vb; | |
862 | unsigned long addr = 0; | |
863 | unsigned int order; | |
02b709df | 864 | int purge = 0; |
db64fe02 NP |
865 | |
866 | BUG_ON(size & ~PAGE_MASK); | |
867 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
868 | order = get_order(size); | |
869 | ||
870 | again: | |
871 | rcu_read_lock(); | |
872 | vbq = &get_cpu_var(vmap_block_queue); | |
873 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
874 | int i; | |
875 | ||
876 | spin_lock(&vb->lock); | |
02b709df NP |
877 | if (vb->free < 1UL << order) |
878 | goto next; | |
879 | ||
db64fe02 NP |
880 | i = bitmap_find_free_region(vb->alloc_map, |
881 | VMAP_BBMAP_BITS, order); | |
882 | ||
02b709df NP |
883 | if (i < 0) { |
884 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS) { | |
885 | /* fragmented and no outstanding allocations */ | |
886 | BUG_ON(vb->dirty != VMAP_BBMAP_BITS); | |
887 | purge = 1; | |
db64fe02 | 888 | } |
02b709df | 889 | goto next; |
db64fe02 | 890 | } |
02b709df NP |
891 | addr = vb->va->va_start + (i << PAGE_SHIFT); |
892 | BUG_ON(addr_to_vb_idx(addr) != | |
893 | addr_to_vb_idx(vb->va->va_start)); | |
894 | vb->free -= 1UL << order; | |
895 | if (vb->free == 0) { | |
896 | spin_lock(&vbq->lock); | |
897 | list_del_rcu(&vb->free_list); | |
898 | spin_unlock(&vbq->lock); | |
899 | } | |
900 | spin_unlock(&vb->lock); | |
901 | break; | |
902 | next: | |
db64fe02 NP |
903 | spin_unlock(&vb->lock); |
904 | } | |
02b709df NP |
905 | |
906 | if (purge) | |
907 | purge_fragmented_blocks_thiscpu(); | |
908 | ||
3f04ba85 | 909 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
910 | rcu_read_unlock(); |
911 | ||
912 | if (!addr) { | |
913 | vb = new_vmap_block(gfp_mask); | |
914 | if (IS_ERR(vb)) | |
915 | return vb; | |
916 | goto again; | |
917 | } | |
918 | ||
919 | return (void *)addr; | |
920 | } | |
921 | ||
922 | static void vb_free(const void *addr, unsigned long size) | |
923 | { | |
924 | unsigned long offset; | |
925 | unsigned long vb_idx; | |
926 | unsigned int order; | |
927 | struct vmap_block *vb; | |
928 | ||
929 | BUG_ON(size & ~PAGE_MASK); | |
930 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); | |
b29acbdc NP |
931 | |
932 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); | |
933 | ||
db64fe02 NP |
934 | order = get_order(size); |
935 | ||
936 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); | |
937 | ||
938 | vb_idx = addr_to_vb_idx((unsigned long)addr); | |
939 | rcu_read_lock(); | |
940 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
941 | rcu_read_unlock(); | |
942 | BUG_ON(!vb); | |
943 | ||
64141da5 JF |
944 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
945 | ||
db64fe02 | 946 | spin_lock(&vb->lock); |
de560423 | 947 | BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order)); |
d086817d | 948 | |
db64fe02 NP |
949 | vb->dirty += 1UL << order; |
950 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 951 | BUG_ON(vb->free); |
db64fe02 NP |
952 | spin_unlock(&vb->lock); |
953 | free_vmap_block(vb); | |
954 | } else | |
955 | spin_unlock(&vb->lock); | |
956 | } | |
957 | ||
958 | /** | |
959 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
960 | * | |
961 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
962 | * to amortize TLB flushing overheads. What this means is that any page you | |
963 | * have now, may, in a former life, have been mapped into kernel virtual | |
964 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
965 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
966 | * | |
967 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
968 | * be sure that none of the pages we have control over will have any aliases | |
969 | * from the vmap layer. | |
970 | */ | |
971 | void vm_unmap_aliases(void) | |
972 | { | |
973 | unsigned long start = ULONG_MAX, end = 0; | |
974 | int cpu; | |
975 | int flush = 0; | |
976 | ||
9b463334 JF |
977 | if (unlikely(!vmap_initialized)) |
978 | return; | |
979 | ||
db64fe02 NP |
980 | for_each_possible_cpu(cpu) { |
981 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
982 | struct vmap_block *vb; | |
983 | ||
984 | rcu_read_lock(); | |
985 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
986 | int i; | |
987 | ||
988 | spin_lock(&vb->lock); | |
989 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); | |
990 | while (i < VMAP_BBMAP_BITS) { | |
991 | unsigned long s, e; | |
992 | int j; | |
993 | j = find_next_zero_bit(vb->dirty_map, | |
994 | VMAP_BBMAP_BITS, i); | |
995 | ||
996 | s = vb->va->va_start + (i << PAGE_SHIFT); | |
997 | e = vb->va->va_start + (j << PAGE_SHIFT); | |
db64fe02 NP |
998 | flush = 1; |
999 | ||
1000 | if (s < start) | |
1001 | start = s; | |
1002 | if (e > end) | |
1003 | end = e; | |
1004 | ||
1005 | i = j; | |
1006 | i = find_next_bit(vb->dirty_map, | |
1007 | VMAP_BBMAP_BITS, i); | |
1008 | } | |
1009 | spin_unlock(&vb->lock); | |
1010 | } | |
1011 | rcu_read_unlock(); | |
1012 | } | |
1013 | ||
1014 | __purge_vmap_area_lazy(&start, &end, 1, flush); | |
1015 | } | |
1016 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
1017 | ||
1018 | /** | |
1019 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1020 | * @mem: the pointer returned by vm_map_ram | |
1021 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1022 | */ | |
1023 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1024 | { | |
1025 | unsigned long size = count << PAGE_SHIFT; | |
1026 | unsigned long addr = (unsigned long)mem; | |
1027 | ||
1028 | BUG_ON(!addr); | |
1029 | BUG_ON(addr < VMALLOC_START); | |
1030 | BUG_ON(addr > VMALLOC_END); | |
1031 | BUG_ON(addr & (PAGE_SIZE-1)); | |
1032 | ||
1033 | debug_check_no_locks_freed(mem, size); | |
cd52858c | 1034 | vmap_debug_free_range(addr, addr+size); |
db64fe02 NP |
1035 | |
1036 | if (likely(count <= VMAP_MAX_ALLOC)) | |
1037 | vb_free(mem, size); | |
1038 | else | |
1039 | free_unmap_vmap_area_addr(addr); | |
1040 | } | |
1041 | EXPORT_SYMBOL(vm_unmap_ram); | |
1042 | ||
1043 | /** | |
1044 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1045 | * @pages: an array of pointers to the pages to be mapped | |
1046 | * @count: number of pages | |
1047 | * @node: prefer to allocate data structures on this node | |
1048 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad RD |
1049 | * |
1050 | * Returns: a pointer to the address that has been mapped, or %NULL on failure | |
db64fe02 NP |
1051 | */ |
1052 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1053 | { | |
1054 | unsigned long size = count << PAGE_SHIFT; | |
1055 | unsigned long addr; | |
1056 | void *mem; | |
1057 | ||
1058 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1059 | mem = vb_alloc(size, GFP_KERNEL); | |
1060 | if (IS_ERR(mem)) | |
1061 | return NULL; | |
1062 | addr = (unsigned long)mem; | |
1063 | } else { | |
1064 | struct vmap_area *va; | |
1065 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1066 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1067 | if (IS_ERR(va)) | |
1068 | return NULL; | |
1069 | ||
1070 | addr = va->va_start; | |
1071 | mem = (void *)addr; | |
1072 | } | |
1073 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { | |
1074 | vm_unmap_ram(mem, count); | |
1075 | return NULL; | |
1076 | } | |
1077 | return mem; | |
1078 | } | |
1079 | EXPORT_SYMBOL(vm_map_ram); | |
1080 | ||
f0aa6617 TH |
1081 | /** |
1082 | * vm_area_register_early - register vmap area early during boot | |
1083 | * @vm: vm_struct to register | |
c0c0a293 | 1084 | * @align: requested alignment |
f0aa6617 TH |
1085 | * |
1086 | * This function is used to register kernel vm area before | |
1087 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1088 | * proper values on entry and other fields should be zero. On return, | |
1089 | * vm->addr contains the allocated address. | |
1090 | * | |
1091 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1092 | */ | |
c0c0a293 | 1093 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1094 | { |
1095 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1096 | unsigned long addr; |
1097 | ||
1098 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1099 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1100 | |
c0c0a293 | 1101 | vm->addr = (void *)addr; |
f0aa6617 TH |
1102 | |
1103 | vm->next = vmlist; | |
1104 | vmlist = vm; | |
1105 | } | |
1106 | ||
db64fe02 NP |
1107 | void __init vmalloc_init(void) |
1108 | { | |
822c18f2 IK |
1109 | struct vmap_area *va; |
1110 | struct vm_struct *tmp; | |
db64fe02 NP |
1111 | int i; |
1112 | ||
1113 | for_each_possible_cpu(i) { | |
1114 | struct vmap_block_queue *vbq; | |
1115 | ||
1116 | vbq = &per_cpu(vmap_block_queue, i); | |
1117 | spin_lock_init(&vbq->lock); | |
1118 | INIT_LIST_HEAD(&vbq->free); | |
db64fe02 | 1119 | } |
9b463334 | 1120 | |
822c18f2 IK |
1121 | /* Import existing vmlist entries. */ |
1122 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
43ebdac4 | 1123 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
822c18f2 IK |
1124 | va->flags = tmp->flags | VM_VM_AREA; |
1125 | va->va_start = (unsigned long)tmp->addr; | |
1126 | va->va_end = va->va_start + tmp->size; | |
1127 | __insert_vmap_area(va); | |
1128 | } | |
ca23e405 TH |
1129 | |
1130 | vmap_area_pcpu_hole = VMALLOC_END; | |
1131 | ||
9b463334 | 1132 | vmap_initialized = true; |
db64fe02 NP |
1133 | } |
1134 | ||
8fc48985 TH |
1135 | /** |
1136 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
1137 | * @addr: start of the VM area to map | |
1138 | * @size: size of the VM area to map | |
1139 | * @prot: page protection flags to use | |
1140 | * @pages: pages to map | |
1141 | * | |
1142 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1143 | * specify should have been allocated using get_vm_area() and its | |
1144 | * friends. | |
1145 | * | |
1146 | * NOTE: | |
1147 | * This function does NOT do any cache flushing. The caller is | |
1148 | * responsible for calling flush_cache_vmap() on to-be-mapped areas | |
1149 | * before calling this function. | |
1150 | * | |
1151 | * RETURNS: | |
1152 | * The number of pages mapped on success, -errno on failure. | |
1153 | */ | |
1154 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, | |
1155 | pgprot_t prot, struct page **pages) | |
1156 | { | |
1157 | return vmap_page_range_noflush(addr, addr + size, prot, pages); | |
1158 | } | |
1159 | ||
1160 | /** | |
1161 | * unmap_kernel_range_noflush - unmap kernel VM area | |
1162 | * @addr: start of the VM area to unmap | |
1163 | * @size: size of the VM area to unmap | |
1164 | * | |
1165 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size | |
1166 | * specify should have been allocated using get_vm_area() and its | |
1167 | * friends. | |
1168 | * | |
1169 | * NOTE: | |
1170 | * This function does NOT do any cache flushing. The caller is | |
1171 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas | |
1172 | * before calling this function and flush_tlb_kernel_range() after. | |
1173 | */ | |
1174 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1175 | { | |
1176 | vunmap_page_range(addr, addr + size); | |
1177 | } | |
81e88fdc | 1178 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
8fc48985 TH |
1179 | |
1180 | /** | |
1181 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
1182 | * @addr: start of the VM area to unmap | |
1183 | * @size: size of the VM area to unmap | |
1184 | * | |
1185 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
1186 | * the unmapping and tlb after. | |
1187 | */ | |
db64fe02 NP |
1188 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
1189 | { | |
1190 | unsigned long end = addr + size; | |
f6fcba70 TH |
1191 | |
1192 | flush_cache_vunmap(addr, end); | |
db64fe02 NP |
1193 | vunmap_page_range(addr, end); |
1194 | flush_tlb_kernel_range(addr, end); | |
1195 | } | |
1196 | ||
1197 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) | |
1198 | { | |
1199 | unsigned long addr = (unsigned long)area->addr; | |
1200 | unsigned long end = addr + area->size - PAGE_SIZE; | |
1201 | int err; | |
1202 | ||
1203 | err = vmap_page_range(addr, end, prot, *pages); | |
1204 | if (err > 0) { | |
1205 | *pages += err; | |
1206 | err = 0; | |
1207 | } | |
1208 | ||
1209 | return err; | |
1210 | } | |
1211 | EXPORT_SYMBOL_GPL(map_vm_area); | |
1212 | ||
1213 | /*** Old vmalloc interfaces ***/ | |
1214 | DEFINE_RWLOCK(vmlist_lock); | |
1215 | struct vm_struct *vmlist; | |
1216 | ||
cf88c790 TH |
1217 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
1218 | unsigned long flags, void *caller) | |
1219 | { | |
1220 | struct vm_struct *tmp, **p; | |
1221 | ||
1222 | vm->flags = flags; | |
1223 | vm->addr = (void *)va->va_start; | |
1224 | vm->size = va->va_end - va->va_start; | |
1225 | vm->caller = caller; | |
1226 | va->private = vm; | |
1227 | va->flags |= VM_VM_AREA; | |
1228 | ||
1229 | write_lock(&vmlist_lock); | |
1230 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1231 | if (tmp->addr >= vm->addr) | |
1232 | break; | |
1233 | } | |
1234 | vm->next = *p; | |
1235 | *p = vm; | |
1236 | write_unlock(&vmlist_lock); | |
1237 | } | |
1238 | ||
db64fe02 | 1239 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 DM |
1240 | unsigned long align, unsigned long flags, unsigned long start, |
1241 | unsigned long end, int node, gfp_t gfp_mask, void *caller) | |
db64fe02 NP |
1242 | { |
1243 | static struct vmap_area *va; | |
1244 | struct vm_struct *area; | |
1da177e4 | 1245 | |
52fd24ca | 1246 | BUG_ON(in_interrupt()); |
1da177e4 LT |
1247 | if (flags & VM_IOREMAP) { |
1248 | int bit = fls(size); | |
1249 | ||
1250 | if (bit > IOREMAP_MAX_ORDER) | |
1251 | bit = IOREMAP_MAX_ORDER; | |
1252 | else if (bit < PAGE_SHIFT) | |
1253 | bit = PAGE_SHIFT; | |
1254 | ||
1255 | align = 1ul << bit; | |
1256 | } | |
db64fe02 | 1257 | |
1da177e4 | 1258 | size = PAGE_ALIGN(size); |
31be8309 OH |
1259 | if (unlikely(!size)) |
1260 | return NULL; | |
1da177e4 | 1261 | |
cf88c790 | 1262 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
1263 | if (unlikely(!area)) |
1264 | return NULL; | |
1265 | ||
1da177e4 LT |
1266 | /* |
1267 | * We always allocate a guard page. | |
1268 | */ | |
1269 | size += PAGE_SIZE; | |
1270 | ||
db64fe02 NP |
1271 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
1272 | if (IS_ERR(va)) { | |
1273 | kfree(area); | |
1274 | return NULL; | |
1da177e4 | 1275 | } |
1da177e4 | 1276 | |
cf88c790 | 1277 | insert_vmalloc_vm(area, va, flags, caller); |
1da177e4 | 1278 | return area; |
1da177e4 LT |
1279 | } |
1280 | ||
930fc45a CL |
1281 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
1282 | unsigned long start, unsigned long end) | |
1283 | { | |
2dca6999 | 1284 | return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL, |
23016969 | 1285 | __builtin_return_address(0)); |
930fc45a | 1286 | } |
5992b6da | 1287 | EXPORT_SYMBOL_GPL(__get_vm_area); |
930fc45a | 1288 | |
c2968612 BH |
1289 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
1290 | unsigned long start, unsigned long end, | |
1291 | void *caller) | |
1292 | { | |
2dca6999 | 1293 | return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL, |
c2968612 BH |
1294 | caller); |
1295 | } | |
1296 | ||
1da177e4 | 1297 | /** |
183ff22b | 1298 | * get_vm_area - reserve a contiguous kernel virtual area |
1da177e4 LT |
1299 | * @size: size of the area |
1300 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1301 | * | |
1302 | * Search an area of @size in the kernel virtual mapping area, | |
1303 | * and reserved it for out purposes. Returns the area descriptor | |
1304 | * on success or %NULL on failure. | |
1305 | */ | |
1306 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
1307 | { | |
2dca6999 | 1308 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
23016969 CL |
1309 | -1, GFP_KERNEL, __builtin_return_address(0)); |
1310 | } | |
1311 | ||
1312 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
1313 | void *caller) | |
1314 | { | |
2dca6999 | 1315 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
23016969 | 1316 | -1, GFP_KERNEL, caller); |
1da177e4 LT |
1317 | } |
1318 | ||
db64fe02 | 1319 | static struct vm_struct *find_vm_area(const void *addr) |
83342314 | 1320 | { |
db64fe02 | 1321 | struct vmap_area *va; |
83342314 | 1322 | |
db64fe02 NP |
1323 | va = find_vmap_area((unsigned long)addr); |
1324 | if (va && va->flags & VM_VM_AREA) | |
1325 | return va->private; | |
1da177e4 | 1326 | |
1da177e4 | 1327 | return NULL; |
1da177e4 LT |
1328 | } |
1329 | ||
7856dfeb | 1330 | /** |
183ff22b | 1331 | * remove_vm_area - find and remove a continuous kernel virtual area |
7856dfeb AK |
1332 | * @addr: base address |
1333 | * | |
1334 | * Search for the kernel VM area starting at @addr, and remove it. | |
1335 | * This function returns the found VM area, but using it is NOT safe | |
1336 | * on SMP machines, except for its size or flags. | |
1337 | */ | |
b3bdda02 | 1338 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 1339 | { |
db64fe02 NP |
1340 | struct vmap_area *va; |
1341 | ||
1342 | va = find_vmap_area((unsigned long)addr); | |
1343 | if (va && va->flags & VM_VM_AREA) { | |
1344 | struct vm_struct *vm = va->private; | |
1345 | struct vm_struct *tmp, **p; | |
dd32c279 KH |
1346 | /* |
1347 | * remove from list and disallow access to this vm_struct | |
1348 | * before unmap. (address range confliction is maintained by | |
1349 | * vmap.) | |
1350 | */ | |
db64fe02 NP |
1351 | write_lock(&vmlist_lock); |
1352 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) | |
1353 | ; | |
1354 | *p = tmp->next; | |
1355 | write_unlock(&vmlist_lock); | |
1356 | ||
dd32c279 KH |
1357 | vmap_debug_free_range(va->va_start, va->va_end); |
1358 | free_unmap_vmap_area(va); | |
1359 | vm->size -= PAGE_SIZE; | |
1360 | ||
db64fe02 NP |
1361 | return vm; |
1362 | } | |
1363 | return NULL; | |
7856dfeb AK |
1364 | } |
1365 | ||
b3bdda02 | 1366 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
1367 | { |
1368 | struct vm_struct *area; | |
1369 | ||
1370 | if (!addr) | |
1371 | return; | |
1372 | ||
1373 | if ((PAGE_SIZE-1) & (unsigned long)addr) { | |
4c8573e2 | 1374 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
1da177e4 LT |
1375 | return; |
1376 | } | |
1377 | ||
1378 | area = remove_vm_area(addr); | |
1379 | if (unlikely(!area)) { | |
4c8573e2 | 1380 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 1381 | addr); |
1da177e4 LT |
1382 | return; |
1383 | } | |
1384 | ||
9a11b49a | 1385 | debug_check_no_locks_freed(addr, area->size); |
3ac7fe5a | 1386 | debug_check_no_obj_freed(addr, area->size); |
9a11b49a | 1387 | |
1da177e4 LT |
1388 | if (deallocate_pages) { |
1389 | int i; | |
1390 | ||
1391 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1392 | struct page *page = area->pages[i]; |
1393 | ||
1394 | BUG_ON(!page); | |
1395 | __free_page(page); | |
1da177e4 LT |
1396 | } |
1397 | ||
8757d5fa | 1398 | if (area->flags & VM_VPAGES) |
1da177e4 LT |
1399 | vfree(area->pages); |
1400 | else | |
1401 | kfree(area->pages); | |
1402 | } | |
1403 | ||
1404 | kfree(area); | |
1405 | return; | |
1406 | } | |
1407 | ||
1408 | /** | |
1409 | * vfree - release memory allocated by vmalloc() | |
1da177e4 LT |
1410 | * @addr: memory base address |
1411 | * | |
183ff22b | 1412 | * Free the virtually continuous memory area starting at @addr, as |
80e93eff PE |
1413 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
1414 | * NULL, no operation is performed. | |
1da177e4 | 1415 | * |
80e93eff | 1416 | * Must not be called in interrupt context. |
1da177e4 | 1417 | */ |
b3bdda02 | 1418 | void vfree(const void *addr) |
1da177e4 LT |
1419 | { |
1420 | BUG_ON(in_interrupt()); | |
89219d37 CM |
1421 | |
1422 | kmemleak_free(addr); | |
1423 | ||
1da177e4 LT |
1424 | __vunmap(addr, 1); |
1425 | } | |
1da177e4 LT |
1426 | EXPORT_SYMBOL(vfree); |
1427 | ||
1428 | /** | |
1429 | * vunmap - release virtual mapping obtained by vmap() | |
1da177e4 LT |
1430 | * @addr: memory base address |
1431 | * | |
1432 | * Free the virtually contiguous memory area starting at @addr, | |
1433 | * which was created from the page array passed to vmap(). | |
1434 | * | |
80e93eff | 1435 | * Must not be called in interrupt context. |
1da177e4 | 1436 | */ |
b3bdda02 | 1437 | void vunmap(const void *addr) |
1da177e4 LT |
1438 | { |
1439 | BUG_ON(in_interrupt()); | |
34754b69 | 1440 | might_sleep(); |
1da177e4 LT |
1441 | __vunmap(addr, 0); |
1442 | } | |
1da177e4 LT |
1443 | EXPORT_SYMBOL(vunmap); |
1444 | ||
1445 | /** | |
1446 | * vmap - map an array of pages into virtually contiguous space | |
1da177e4 LT |
1447 | * @pages: array of page pointers |
1448 | * @count: number of pages to map | |
1449 | * @flags: vm_area->flags | |
1450 | * @prot: page protection for the mapping | |
1451 | * | |
1452 | * Maps @count pages from @pages into contiguous kernel virtual | |
1453 | * space. | |
1454 | */ | |
1455 | void *vmap(struct page **pages, unsigned int count, | |
1456 | unsigned long flags, pgprot_t prot) | |
1457 | { | |
1458 | struct vm_struct *area; | |
1459 | ||
34754b69 PZ |
1460 | might_sleep(); |
1461 | ||
4481374c | 1462 | if (count > totalram_pages) |
1da177e4 LT |
1463 | return NULL; |
1464 | ||
23016969 CL |
1465 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
1466 | __builtin_return_address(0)); | |
1da177e4 LT |
1467 | if (!area) |
1468 | return NULL; | |
23016969 | 1469 | |
1da177e4 LT |
1470 | if (map_vm_area(area, prot, &pages)) { |
1471 | vunmap(area->addr); | |
1472 | return NULL; | |
1473 | } | |
1474 | ||
1475 | return area->addr; | |
1476 | } | |
1da177e4 LT |
1477 | EXPORT_SYMBOL(vmap); |
1478 | ||
2dca6999 DM |
1479 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
1480 | gfp_t gfp_mask, pgprot_t prot, | |
db64fe02 | 1481 | int node, void *caller); |
e31d9eb5 | 1482 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
23016969 | 1483 | pgprot_t prot, int node, void *caller) |
1da177e4 LT |
1484 | { |
1485 | struct page **pages; | |
1486 | unsigned int nr_pages, array_size, i; | |
976d6dfb | 1487 | gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
1da177e4 LT |
1488 | |
1489 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; | |
1490 | array_size = (nr_pages * sizeof(struct page *)); | |
1491 | ||
1492 | area->nr_pages = nr_pages; | |
1493 | /* Please note that the recursion is strictly bounded. */ | |
8757d5fa | 1494 | if (array_size > PAGE_SIZE) { |
976d6dfb | 1495 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
23016969 | 1496 | PAGE_KERNEL, node, caller); |
8757d5fa | 1497 | area->flags |= VM_VPAGES; |
286e1ea3 | 1498 | } else { |
976d6dfb | 1499 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 1500 | } |
1da177e4 | 1501 | area->pages = pages; |
23016969 | 1502 | area->caller = caller; |
1da177e4 LT |
1503 | if (!area->pages) { |
1504 | remove_vm_area(area->addr); | |
1505 | kfree(area); | |
1506 | return NULL; | |
1507 | } | |
1da177e4 LT |
1508 | |
1509 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
1510 | struct page *page; |
1511 | ||
930fc45a | 1512 | if (node < 0) |
bf53d6f8 | 1513 | page = alloc_page(gfp_mask); |
930fc45a | 1514 | else |
bf53d6f8 CL |
1515 | page = alloc_pages_node(node, gfp_mask, 0); |
1516 | ||
1517 | if (unlikely(!page)) { | |
1da177e4 LT |
1518 | /* Successfully allocated i pages, free them in __vunmap() */ |
1519 | area->nr_pages = i; | |
1520 | goto fail; | |
1521 | } | |
bf53d6f8 | 1522 | area->pages[i] = page; |
1da177e4 LT |
1523 | } |
1524 | ||
1525 | if (map_vm_area(area, prot, &pages)) | |
1526 | goto fail; | |
1527 | return area->addr; | |
1528 | ||
1529 | fail: | |
1530 | vfree(area->addr); | |
1531 | return NULL; | |
1532 | } | |
1533 | ||
1534 | /** | |
d0a21265 | 1535 | * __vmalloc_node_range - allocate virtually contiguous memory |
1da177e4 | 1536 | * @size: allocation size |
2dca6999 | 1537 | * @align: desired alignment |
d0a21265 DR |
1538 | * @start: vm area range start |
1539 | * @end: vm area range end | |
1da177e4 LT |
1540 | * @gfp_mask: flags for the page level allocator |
1541 | * @prot: protection mask for the allocated pages | |
d44e0780 | 1542 | * @node: node to use for allocation or -1 |
c85d194b | 1543 | * @caller: caller's return address |
1da177e4 LT |
1544 | * |
1545 | * Allocate enough pages to cover @size from the page level | |
1546 | * allocator with @gfp_mask flags. Map them into contiguous | |
1547 | * kernel virtual space, using a pagetable protection of @prot. | |
1548 | */ | |
d0a21265 DR |
1549 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
1550 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
1551 | pgprot_t prot, int node, void *caller) | |
1da177e4 LT |
1552 | { |
1553 | struct vm_struct *area; | |
89219d37 CM |
1554 | void *addr; |
1555 | unsigned long real_size = size; | |
1da177e4 LT |
1556 | |
1557 | size = PAGE_ALIGN(size); | |
4481374c | 1558 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
1da177e4 LT |
1559 | return NULL; |
1560 | ||
d0a21265 DR |
1561 | area = __get_vm_area_node(size, align, VM_ALLOC, start, end, node, |
1562 | gfp_mask, caller); | |
23016969 | 1563 | |
1da177e4 LT |
1564 | if (!area) |
1565 | return NULL; | |
1566 | ||
89219d37 CM |
1567 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
1568 | ||
1569 | /* | |
1570 | * A ref_count = 3 is needed because the vm_struct and vmap_area | |
1571 | * structures allocated in the __get_vm_area_node() function contain | |
1572 | * references to the virtual address of the vmalloc'ed block. | |
1573 | */ | |
1574 | kmemleak_alloc(addr, real_size, 3, gfp_mask); | |
1575 | ||
1576 | return addr; | |
1da177e4 LT |
1577 | } |
1578 | ||
d0a21265 DR |
1579 | /** |
1580 | * __vmalloc_node - allocate virtually contiguous memory | |
1581 | * @size: allocation size | |
1582 | * @align: desired alignment | |
1583 | * @gfp_mask: flags for the page level allocator | |
1584 | * @prot: protection mask for the allocated pages | |
1585 | * @node: node to use for allocation or -1 | |
1586 | * @caller: caller's return address | |
1587 | * | |
1588 | * Allocate enough pages to cover @size from the page level | |
1589 | * allocator with @gfp_mask flags. Map them into contiguous | |
1590 | * kernel virtual space, using a pagetable protection of @prot. | |
1591 | */ | |
1592 | static void *__vmalloc_node(unsigned long size, unsigned long align, | |
1593 | gfp_t gfp_mask, pgprot_t prot, | |
1594 | int node, void *caller) | |
1595 | { | |
1596 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
1597 | gfp_mask, prot, node, caller); | |
1598 | } | |
1599 | ||
930fc45a CL |
1600 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1601 | { | |
2dca6999 | 1602 | return __vmalloc_node(size, 1, gfp_mask, prot, -1, |
23016969 | 1603 | __builtin_return_address(0)); |
930fc45a | 1604 | } |
1da177e4 LT |
1605 | EXPORT_SYMBOL(__vmalloc); |
1606 | ||
e1ca7788 DY |
1607 | static inline void *__vmalloc_node_flags(unsigned long size, |
1608 | int node, gfp_t flags) | |
1609 | { | |
1610 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
1611 | node, __builtin_return_address(0)); | |
1612 | } | |
1613 | ||
1da177e4 LT |
1614 | /** |
1615 | * vmalloc - allocate virtually contiguous memory | |
1da177e4 | 1616 | * @size: allocation size |
1da177e4 LT |
1617 | * Allocate enough pages to cover @size from the page level |
1618 | * allocator and map them into contiguous kernel virtual space. | |
1619 | * | |
c1c8897f | 1620 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1621 | * use __vmalloc() instead. |
1622 | */ | |
1623 | void *vmalloc(unsigned long size) | |
1624 | { | |
e1ca7788 | 1625 | return __vmalloc_node_flags(size, -1, GFP_KERNEL | __GFP_HIGHMEM); |
1da177e4 | 1626 | } |
1da177e4 LT |
1627 | EXPORT_SYMBOL(vmalloc); |
1628 | ||
e1ca7788 DY |
1629 | /** |
1630 | * vzalloc - allocate virtually contiguous memory with zero fill | |
1631 | * @size: allocation size | |
1632 | * Allocate enough pages to cover @size from the page level | |
1633 | * allocator and map them into contiguous kernel virtual space. | |
1634 | * The memory allocated is set to zero. | |
1635 | * | |
1636 | * For tight control over page level allocator and protection flags | |
1637 | * use __vmalloc() instead. | |
1638 | */ | |
1639 | void *vzalloc(unsigned long size) | |
1640 | { | |
1641 | return __vmalloc_node_flags(size, -1, | |
1642 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); | |
1643 | } | |
1644 | EXPORT_SYMBOL(vzalloc); | |
1645 | ||
83342314 | 1646 | /** |
ead04089 REB |
1647 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
1648 | * @size: allocation size | |
83342314 | 1649 | * |
ead04089 REB |
1650 | * The resulting memory area is zeroed so it can be mapped to userspace |
1651 | * without leaking data. | |
83342314 NP |
1652 | */ |
1653 | void *vmalloc_user(unsigned long size) | |
1654 | { | |
1655 | struct vm_struct *area; | |
1656 | void *ret; | |
1657 | ||
2dca6999 DM |
1658 | ret = __vmalloc_node(size, SHMLBA, |
1659 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
84877848 | 1660 | PAGE_KERNEL, -1, __builtin_return_address(0)); |
2b4ac44e | 1661 | if (ret) { |
db64fe02 | 1662 | area = find_vm_area(ret); |
2b4ac44e | 1663 | area->flags |= VM_USERMAP; |
2b4ac44e | 1664 | } |
83342314 NP |
1665 | return ret; |
1666 | } | |
1667 | EXPORT_SYMBOL(vmalloc_user); | |
1668 | ||
930fc45a CL |
1669 | /** |
1670 | * vmalloc_node - allocate memory on a specific node | |
930fc45a | 1671 | * @size: allocation size |
d44e0780 | 1672 | * @node: numa node |
930fc45a CL |
1673 | * |
1674 | * Allocate enough pages to cover @size from the page level | |
1675 | * allocator and map them into contiguous kernel virtual space. | |
1676 | * | |
c1c8897f | 1677 | * For tight control over page level allocator and protection flags |
930fc45a CL |
1678 | * use __vmalloc() instead. |
1679 | */ | |
1680 | void *vmalloc_node(unsigned long size, int node) | |
1681 | { | |
2dca6999 | 1682 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
23016969 | 1683 | node, __builtin_return_address(0)); |
930fc45a CL |
1684 | } |
1685 | EXPORT_SYMBOL(vmalloc_node); | |
1686 | ||
e1ca7788 DY |
1687 | /** |
1688 | * vzalloc_node - allocate memory on a specific node with zero fill | |
1689 | * @size: allocation size | |
1690 | * @node: numa node | |
1691 | * | |
1692 | * Allocate enough pages to cover @size from the page level | |
1693 | * allocator and map them into contiguous kernel virtual space. | |
1694 | * The memory allocated is set to zero. | |
1695 | * | |
1696 | * For tight control over page level allocator and protection flags | |
1697 | * use __vmalloc_node() instead. | |
1698 | */ | |
1699 | void *vzalloc_node(unsigned long size, int node) | |
1700 | { | |
1701 | return __vmalloc_node_flags(size, node, | |
1702 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); | |
1703 | } | |
1704 | EXPORT_SYMBOL(vzalloc_node); | |
1705 | ||
4dc3b16b PP |
1706 | #ifndef PAGE_KERNEL_EXEC |
1707 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
1708 | #endif | |
1709 | ||
1da177e4 LT |
1710 | /** |
1711 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
1da177e4 LT |
1712 | * @size: allocation size |
1713 | * | |
1714 | * Kernel-internal function to allocate enough pages to cover @size | |
1715 | * the page level allocator and map them into contiguous and | |
1716 | * executable kernel virtual space. | |
1717 | * | |
c1c8897f | 1718 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
1719 | * use __vmalloc() instead. |
1720 | */ | |
1721 | ||
1da177e4 LT |
1722 | void *vmalloc_exec(unsigned long size) |
1723 | { | |
2dca6999 | 1724 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
84877848 | 1725 | -1, __builtin_return_address(0)); |
1da177e4 LT |
1726 | } |
1727 | ||
0d08e0d3 | 1728 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
7ac674f5 | 1729 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
0d08e0d3 | 1730 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
7ac674f5 | 1731 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
0d08e0d3 AK |
1732 | #else |
1733 | #define GFP_VMALLOC32 GFP_KERNEL | |
1734 | #endif | |
1735 | ||
1da177e4 LT |
1736 | /** |
1737 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
1738 | * @size: allocation size |
1739 | * | |
1740 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
1741 | * page level allocator and map them into contiguous kernel virtual space. | |
1742 | */ | |
1743 | void *vmalloc_32(unsigned long size) | |
1744 | { | |
2dca6999 | 1745 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
84877848 | 1746 | -1, __builtin_return_address(0)); |
1da177e4 | 1747 | } |
1da177e4 LT |
1748 | EXPORT_SYMBOL(vmalloc_32); |
1749 | ||
83342314 | 1750 | /** |
ead04089 | 1751 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
83342314 | 1752 | * @size: allocation size |
ead04089 REB |
1753 | * |
1754 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
1755 | * mapped to userspace without leaking data. | |
83342314 NP |
1756 | */ |
1757 | void *vmalloc_32_user(unsigned long size) | |
1758 | { | |
1759 | struct vm_struct *area; | |
1760 | void *ret; | |
1761 | ||
2dca6999 | 1762 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
84877848 | 1763 | -1, __builtin_return_address(0)); |
2b4ac44e | 1764 | if (ret) { |
db64fe02 | 1765 | area = find_vm_area(ret); |
2b4ac44e | 1766 | area->flags |= VM_USERMAP; |
2b4ac44e | 1767 | } |
83342314 NP |
1768 | return ret; |
1769 | } | |
1770 | EXPORT_SYMBOL(vmalloc_32_user); | |
1771 | ||
d0107eb0 KH |
1772 | /* |
1773 | * small helper routine , copy contents to buf from addr. | |
1774 | * If the page is not present, fill zero. | |
1775 | */ | |
1776 | ||
1777 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
1778 | { | |
1779 | struct page *p; | |
1780 | int copied = 0; | |
1781 | ||
1782 | while (count) { | |
1783 | unsigned long offset, length; | |
1784 | ||
1785 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1786 | length = PAGE_SIZE - offset; | |
1787 | if (length > count) | |
1788 | length = count; | |
1789 | p = vmalloc_to_page(addr); | |
1790 | /* | |
1791 | * To do safe access to this _mapped_ area, we need | |
1792 | * lock. But adding lock here means that we need to add | |
1793 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1794 | * interface, rarely used. Instead of that, we'll use | |
1795 | * kmap() and get small overhead in this access function. | |
1796 | */ | |
1797 | if (p) { | |
1798 | /* | |
1799 | * we can expect USER0 is not used (see vread/vwrite's | |
1800 | * function description) | |
1801 | */ | |
1802 | void *map = kmap_atomic(p, KM_USER0); | |
1803 | memcpy(buf, map + offset, length); | |
1804 | kunmap_atomic(map, KM_USER0); | |
1805 | } else | |
1806 | memset(buf, 0, length); | |
1807 | ||
1808 | addr += length; | |
1809 | buf += length; | |
1810 | copied += length; | |
1811 | count -= length; | |
1812 | } | |
1813 | return copied; | |
1814 | } | |
1815 | ||
1816 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
1817 | { | |
1818 | struct page *p; | |
1819 | int copied = 0; | |
1820 | ||
1821 | while (count) { | |
1822 | unsigned long offset, length; | |
1823 | ||
1824 | offset = (unsigned long)addr & ~PAGE_MASK; | |
1825 | length = PAGE_SIZE - offset; | |
1826 | if (length > count) | |
1827 | length = count; | |
1828 | p = vmalloc_to_page(addr); | |
1829 | /* | |
1830 | * To do safe access to this _mapped_ area, we need | |
1831 | * lock. But adding lock here means that we need to add | |
1832 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
1833 | * interface, rarely used. Instead of that, we'll use | |
1834 | * kmap() and get small overhead in this access function. | |
1835 | */ | |
1836 | if (p) { | |
1837 | /* | |
1838 | * we can expect USER0 is not used (see vread/vwrite's | |
1839 | * function description) | |
1840 | */ | |
1841 | void *map = kmap_atomic(p, KM_USER0); | |
1842 | memcpy(map + offset, buf, length); | |
1843 | kunmap_atomic(map, KM_USER0); | |
1844 | } | |
1845 | addr += length; | |
1846 | buf += length; | |
1847 | copied += length; | |
1848 | count -= length; | |
1849 | } | |
1850 | return copied; | |
1851 | } | |
1852 | ||
1853 | /** | |
1854 | * vread() - read vmalloc area in a safe way. | |
1855 | * @buf: buffer for reading data | |
1856 | * @addr: vm address. | |
1857 | * @count: number of bytes to be read. | |
1858 | * | |
1859 | * Returns # of bytes which addr and buf should be increased. | |
1860 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't | |
1861 | * includes any intersect with alive vmalloc area. | |
1862 | * | |
1863 | * This function checks that addr is a valid vmalloc'ed area, and | |
1864 | * copy data from that area to a given buffer. If the given memory range | |
1865 | * of [addr...addr+count) includes some valid address, data is copied to | |
1866 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
1867 | * IOREMAP area is treated as memory hole and no copy is done. | |
1868 | * | |
1869 | * If [addr...addr+count) doesn't includes any intersects with alive | |
1870 | * vm_struct area, returns 0. | |
1871 | * @buf should be kernel's buffer. Because this function uses KM_USER0, | |
1872 | * the caller should guarantee KM_USER0 is not used. | |
1873 | * | |
1874 | * Note: In usual ops, vread() is never necessary because the caller | |
1875 | * should know vmalloc() area is valid and can use memcpy(). | |
1876 | * This is for routines which have to access vmalloc area without | |
1877 | * any informaion, as /dev/kmem. | |
1878 | * | |
1879 | */ | |
1880 | ||
1da177e4 LT |
1881 | long vread(char *buf, char *addr, unsigned long count) |
1882 | { | |
1883 | struct vm_struct *tmp; | |
1884 | char *vaddr, *buf_start = buf; | |
d0107eb0 | 1885 | unsigned long buflen = count; |
1da177e4 LT |
1886 | unsigned long n; |
1887 | ||
1888 | /* Don't allow overflow */ | |
1889 | if ((unsigned long) addr + count < count) | |
1890 | count = -(unsigned long) addr; | |
1891 | ||
1892 | read_lock(&vmlist_lock); | |
d0107eb0 | 1893 | for (tmp = vmlist; count && tmp; tmp = tmp->next) { |
1da177e4 LT |
1894 | vaddr = (char *) tmp->addr; |
1895 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1896 | continue; | |
1897 | while (addr < vaddr) { | |
1898 | if (count == 0) | |
1899 | goto finished; | |
1900 | *buf = '\0'; | |
1901 | buf++; | |
1902 | addr++; | |
1903 | count--; | |
1904 | } | |
1905 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
d0107eb0 KH |
1906 | if (n > count) |
1907 | n = count; | |
1908 | if (!(tmp->flags & VM_IOREMAP)) | |
1909 | aligned_vread(buf, addr, n); | |
1910 | else /* IOREMAP area is treated as memory hole */ | |
1911 | memset(buf, 0, n); | |
1912 | buf += n; | |
1913 | addr += n; | |
1914 | count -= n; | |
1da177e4 LT |
1915 | } |
1916 | finished: | |
1917 | read_unlock(&vmlist_lock); | |
d0107eb0 KH |
1918 | |
1919 | if (buf == buf_start) | |
1920 | return 0; | |
1921 | /* zero-fill memory holes */ | |
1922 | if (buf != buf_start + buflen) | |
1923 | memset(buf, 0, buflen - (buf - buf_start)); | |
1924 | ||
1925 | return buflen; | |
1da177e4 LT |
1926 | } |
1927 | ||
d0107eb0 KH |
1928 | /** |
1929 | * vwrite() - write vmalloc area in a safe way. | |
1930 | * @buf: buffer for source data | |
1931 | * @addr: vm address. | |
1932 | * @count: number of bytes to be read. | |
1933 | * | |
1934 | * Returns # of bytes which addr and buf should be incresed. | |
1935 | * (same number to @count). | |
1936 | * If [addr...addr+count) doesn't includes any intersect with valid | |
1937 | * vmalloc area, returns 0. | |
1938 | * | |
1939 | * This function checks that addr is a valid vmalloc'ed area, and | |
1940 | * copy data from a buffer to the given addr. If specified range of | |
1941 | * [addr...addr+count) includes some valid address, data is copied from | |
1942 | * proper area of @buf. If there are memory holes, no copy to hole. | |
1943 | * IOREMAP area is treated as memory hole and no copy is done. | |
1944 | * | |
1945 | * If [addr...addr+count) doesn't includes any intersects with alive | |
1946 | * vm_struct area, returns 0. | |
1947 | * @buf should be kernel's buffer. Because this function uses KM_USER0, | |
1948 | * the caller should guarantee KM_USER0 is not used. | |
1949 | * | |
1950 | * Note: In usual ops, vwrite() is never necessary because the caller | |
1951 | * should know vmalloc() area is valid and can use memcpy(). | |
1952 | * This is for routines which have to access vmalloc area without | |
1953 | * any informaion, as /dev/kmem. | |
1954 | * | |
1955 | * The caller should guarantee KM_USER1 is not used. | |
1956 | */ | |
1957 | ||
1da177e4 LT |
1958 | long vwrite(char *buf, char *addr, unsigned long count) |
1959 | { | |
1960 | struct vm_struct *tmp; | |
d0107eb0 KH |
1961 | char *vaddr; |
1962 | unsigned long n, buflen; | |
1963 | int copied = 0; | |
1da177e4 LT |
1964 | |
1965 | /* Don't allow overflow */ | |
1966 | if ((unsigned long) addr + count < count) | |
1967 | count = -(unsigned long) addr; | |
d0107eb0 | 1968 | buflen = count; |
1da177e4 LT |
1969 | |
1970 | read_lock(&vmlist_lock); | |
d0107eb0 | 1971 | for (tmp = vmlist; count && tmp; tmp = tmp->next) { |
1da177e4 LT |
1972 | vaddr = (char *) tmp->addr; |
1973 | if (addr >= vaddr + tmp->size - PAGE_SIZE) | |
1974 | continue; | |
1975 | while (addr < vaddr) { | |
1976 | if (count == 0) | |
1977 | goto finished; | |
1978 | buf++; | |
1979 | addr++; | |
1980 | count--; | |
1981 | } | |
1982 | n = vaddr + tmp->size - PAGE_SIZE - addr; | |
d0107eb0 KH |
1983 | if (n > count) |
1984 | n = count; | |
1985 | if (!(tmp->flags & VM_IOREMAP)) { | |
1986 | aligned_vwrite(buf, addr, n); | |
1987 | copied++; | |
1988 | } | |
1989 | buf += n; | |
1990 | addr += n; | |
1991 | count -= n; | |
1da177e4 LT |
1992 | } |
1993 | finished: | |
1994 | read_unlock(&vmlist_lock); | |
d0107eb0 KH |
1995 | if (!copied) |
1996 | return 0; | |
1997 | return buflen; | |
1da177e4 | 1998 | } |
83342314 NP |
1999 | |
2000 | /** | |
2001 | * remap_vmalloc_range - map vmalloc pages to userspace | |
83342314 NP |
2002 | * @vma: vma to cover (map full range of vma) |
2003 | * @addr: vmalloc memory | |
2004 | * @pgoff: number of pages into addr before first page to map | |
7682486b RD |
2005 | * |
2006 | * Returns: 0 for success, -Exxx on failure | |
83342314 NP |
2007 | * |
2008 | * This function checks that addr is a valid vmalloc'ed area, and | |
2009 | * that it is big enough to cover the vma. Will return failure if | |
2010 | * that criteria isn't met. | |
2011 | * | |
72fd4a35 | 2012 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 NP |
2013 | */ |
2014 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
2015 | unsigned long pgoff) | |
2016 | { | |
2017 | struct vm_struct *area; | |
2018 | unsigned long uaddr = vma->vm_start; | |
2019 | unsigned long usize = vma->vm_end - vma->vm_start; | |
83342314 NP |
2020 | |
2021 | if ((PAGE_SIZE-1) & (unsigned long)addr) | |
2022 | return -EINVAL; | |
2023 | ||
db64fe02 | 2024 | area = find_vm_area(addr); |
83342314 | 2025 | if (!area) |
db64fe02 | 2026 | return -EINVAL; |
83342314 NP |
2027 | |
2028 | if (!(area->flags & VM_USERMAP)) | |
db64fe02 | 2029 | return -EINVAL; |
83342314 NP |
2030 | |
2031 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) | |
db64fe02 | 2032 | return -EINVAL; |
83342314 NP |
2033 | |
2034 | addr += pgoff << PAGE_SHIFT; | |
2035 | do { | |
2036 | struct page *page = vmalloc_to_page(addr); | |
db64fe02 NP |
2037 | int ret; |
2038 | ||
83342314 NP |
2039 | ret = vm_insert_page(vma, uaddr, page); |
2040 | if (ret) | |
2041 | return ret; | |
2042 | ||
2043 | uaddr += PAGE_SIZE; | |
2044 | addr += PAGE_SIZE; | |
2045 | usize -= PAGE_SIZE; | |
2046 | } while (usize > 0); | |
2047 | ||
2048 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ | |
2049 | vma->vm_flags |= VM_RESERVED; | |
2050 | ||
db64fe02 | 2051 | return 0; |
83342314 NP |
2052 | } |
2053 | EXPORT_SYMBOL(remap_vmalloc_range); | |
2054 | ||
1eeb66a1 CH |
2055 | /* |
2056 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
2057 | * have one. | |
2058 | */ | |
2059 | void __attribute__((weak)) vmalloc_sync_all(void) | |
2060 | { | |
2061 | } | |
5f4352fb JF |
2062 | |
2063 | ||
2f569afd | 2064 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
5f4352fb JF |
2065 | { |
2066 | /* apply_to_page_range() does all the hard work. */ | |
2067 | return 0; | |
2068 | } | |
2069 | ||
2070 | /** | |
2071 | * alloc_vm_area - allocate a range of kernel address space | |
2072 | * @size: size of the area | |
7682486b RD |
2073 | * |
2074 | * Returns: NULL on failure, vm_struct on success | |
5f4352fb JF |
2075 | * |
2076 | * This function reserves a range of kernel address space, and | |
2077 | * allocates pagetables to map that range. No actual mappings | |
2078 | * are created. If the kernel address space is not shared | |
2079 | * between processes, it syncs the pagetable across all | |
2080 | * processes. | |
2081 | */ | |
2082 | struct vm_struct *alloc_vm_area(size_t size) | |
2083 | { | |
2084 | struct vm_struct *area; | |
2085 | ||
23016969 CL |
2086 | area = get_vm_area_caller(size, VM_IOREMAP, |
2087 | __builtin_return_address(0)); | |
5f4352fb JF |
2088 | if (area == NULL) |
2089 | return NULL; | |
2090 | ||
2091 | /* | |
2092 | * This ensures that page tables are constructed for this region | |
2093 | * of kernel virtual address space and mapped into init_mm. | |
2094 | */ | |
2095 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
2096 | area->size, f, NULL)) { | |
2097 | free_vm_area(area); | |
2098 | return NULL; | |
2099 | } | |
2100 | ||
2101 | /* Make sure the pagetables are constructed in process kernel | |
2102 | mappings */ | |
2103 | vmalloc_sync_all(); | |
2104 | ||
2105 | return area; | |
2106 | } | |
2107 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
2108 | ||
2109 | void free_vm_area(struct vm_struct *area) | |
2110 | { | |
2111 | struct vm_struct *ret; | |
2112 | ret = remove_vm_area(area->addr); | |
2113 | BUG_ON(ret != area); | |
2114 | kfree(area); | |
2115 | } | |
2116 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 2117 | |
4f8b02b4 | 2118 | #ifdef CONFIG_SMP |
ca23e405 TH |
2119 | static struct vmap_area *node_to_va(struct rb_node *n) |
2120 | { | |
2121 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; | |
2122 | } | |
2123 | ||
2124 | /** | |
2125 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end | |
2126 | * @end: target address | |
2127 | * @pnext: out arg for the next vmap_area | |
2128 | * @pprev: out arg for the previous vmap_area | |
2129 | * | |
2130 | * Returns: %true if either or both of next and prev are found, | |
2131 | * %false if no vmap_area exists | |
2132 | * | |
2133 | * Find vmap_areas end addresses of which enclose @end. ie. if not | |
2134 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. | |
2135 | */ | |
2136 | static bool pvm_find_next_prev(unsigned long end, | |
2137 | struct vmap_area **pnext, | |
2138 | struct vmap_area **pprev) | |
2139 | { | |
2140 | struct rb_node *n = vmap_area_root.rb_node; | |
2141 | struct vmap_area *va = NULL; | |
2142 | ||
2143 | while (n) { | |
2144 | va = rb_entry(n, struct vmap_area, rb_node); | |
2145 | if (end < va->va_end) | |
2146 | n = n->rb_left; | |
2147 | else if (end > va->va_end) | |
2148 | n = n->rb_right; | |
2149 | else | |
2150 | break; | |
2151 | } | |
2152 | ||
2153 | if (!va) | |
2154 | return false; | |
2155 | ||
2156 | if (va->va_end > end) { | |
2157 | *pnext = va; | |
2158 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2159 | } else { | |
2160 | *pprev = va; | |
2161 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); | |
2162 | } | |
2163 | return true; | |
2164 | } | |
2165 | ||
2166 | /** | |
2167 | * pvm_determine_end - find the highest aligned address between two vmap_areas | |
2168 | * @pnext: in/out arg for the next vmap_area | |
2169 | * @pprev: in/out arg for the previous vmap_area | |
2170 | * @align: alignment | |
2171 | * | |
2172 | * Returns: determined end address | |
2173 | * | |
2174 | * Find the highest aligned address between *@pnext and *@pprev below | |
2175 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned | |
2176 | * down address is between the end addresses of the two vmap_areas. | |
2177 | * | |
2178 | * Please note that the address returned by this function may fall | |
2179 | * inside *@pnext vmap_area. The caller is responsible for checking | |
2180 | * that. | |
2181 | */ | |
2182 | static unsigned long pvm_determine_end(struct vmap_area **pnext, | |
2183 | struct vmap_area **pprev, | |
2184 | unsigned long align) | |
2185 | { | |
2186 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2187 | unsigned long addr; | |
2188 | ||
2189 | if (*pnext) | |
2190 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); | |
2191 | else | |
2192 | addr = vmalloc_end; | |
2193 | ||
2194 | while (*pprev && (*pprev)->va_end > addr) { | |
2195 | *pnext = *pprev; | |
2196 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); | |
2197 | } | |
2198 | ||
2199 | return addr; | |
2200 | } | |
2201 | ||
2202 | /** | |
2203 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
2204 | * @offsets: array containing offset of each area | |
2205 | * @sizes: array containing size of each area | |
2206 | * @nr_vms: the number of areas to allocate | |
2207 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
2208 | * |
2209 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
2210 | * vm_structs on success, %NULL on failure | |
2211 | * | |
2212 | * Percpu allocator wants to use congruent vm areas so that it can | |
2213 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
2214 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
2215 | * be scattered pretty far, distance between two areas easily going up | |
2216 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
2217 | * areas are allocated from top. | |
ca23e405 TH |
2218 | * |
2219 | * Despite its complicated look, this allocator is rather simple. It | |
2220 | * does everything top-down and scans areas from the end looking for | |
2221 | * matching slot. While scanning, if any of the areas overlaps with | |
2222 | * existing vmap_area, the base address is pulled down to fit the | |
2223 | * area. Scanning is repeated till all the areas fit and then all | |
2224 | * necessary data structres are inserted and the result is returned. | |
2225 | */ | |
2226 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
2227 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 2228 | size_t align) |
ca23e405 TH |
2229 | { |
2230 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
2231 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
2232 | struct vmap_area **vas, *prev, *next; | |
2233 | struct vm_struct **vms; | |
2234 | int area, area2, last_area, term_area; | |
2235 | unsigned long base, start, end, last_end; | |
2236 | bool purged = false; | |
2237 | ||
ca23e405 TH |
2238 | /* verify parameters and allocate data structures */ |
2239 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); | |
2240 | for (last_area = 0, area = 0; area < nr_vms; area++) { | |
2241 | start = offsets[area]; | |
2242 | end = start + sizes[area]; | |
2243 | ||
2244 | /* is everything aligned properly? */ | |
2245 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
2246 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
2247 | ||
2248 | /* detect the area with the highest address */ | |
2249 | if (start > offsets[last_area]) | |
2250 | last_area = area; | |
2251 | ||
2252 | for (area2 = 0; area2 < nr_vms; area2++) { | |
2253 | unsigned long start2 = offsets[area2]; | |
2254 | unsigned long end2 = start2 + sizes[area2]; | |
2255 | ||
2256 | if (area2 == area) | |
2257 | continue; | |
2258 | ||
2259 | BUG_ON(start2 >= start && start2 < end); | |
2260 | BUG_ON(end2 <= end && end2 > start); | |
2261 | } | |
2262 | } | |
2263 | last_end = offsets[last_area] + sizes[last_area]; | |
2264 | ||
2265 | if (vmalloc_end - vmalloc_start < last_end) { | |
2266 | WARN_ON(true); | |
2267 | return NULL; | |
2268 | } | |
2269 | ||
ec3f64fc DR |
2270 | vms = kzalloc(sizeof(vms[0]) * nr_vms, GFP_KERNEL); |
2271 | vas = kzalloc(sizeof(vas[0]) * nr_vms, GFP_KERNEL); | |
ca23e405 TH |
2272 | if (!vas || !vms) |
2273 | goto err_free; | |
2274 | ||
2275 | for (area = 0; area < nr_vms; area++) { | |
ec3f64fc DR |
2276 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
2277 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); | |
ca23e405 TH |
2278 | if (!vas[area] || !vms[area]) |
2279 | goto err_free; | |
2280 | } | |
2281 | retry: | |
2282 | spin_lock(&vmap_area_lock); | |
2283 | ||
2284 | /* start scanning - we scan from the top, begin with the last area */ | |
2285 | area = term_area = last_area; | |
2286 | start = offsets[area]; | |
2287 | end = start + sizes[area]; | |
2288 | ||
2289 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { | |
2290 | base = vmalloc_end - last_end; | |
2291 | goto found; | |
2292 | } | |
2293 | base = pvm_determine_end(&next, &prev, align) - end; | |
2294 | ||
2295 | while (true) { | |
2296 | BUG_ON(next && next->va_end <= base + end); | |
2297 | BUG_ON(prev && prev->va_end > base + end); | |
2298 | ||
2299 | /* | |
2300 | * base might have underflowed, add last_end before | |
2301 | * comparing. | |
2302 | */ | |
2303 | if (base + last_end < vmalloc_start + last_end) { | |
2304 | spin_unlock(&vmap_area_lock); | |
2305 | if (!purged) { | |
2306 | purge_vmap_area_lazy(); | |
2307 | purged = true; | |
2308 | goto retry; | |
2309 | } | |
2310 | goto err_free; | |
2311 | } | |
2312 | ||
2313 | /* | |
2314 | * If next overlaps, move base downwards so that it's | |
2315 | * right below next and then recheck. | |
2316 | */ | |
2317 | if (next && next->va_start < base + end) { | |
2318 | base = pvm_determine_end(&next, &prev, align) - end; | |
2319 | term_area = area; | |
2320 | continue; | |
2321 | } | |
2322 | ||
2323 | /* | |
2324 | * If prev overlaps, shift down next and prev and move | |
2325 | * base so that it's right below new next and then | |
2326 | * recheck. | |
2327 | */ | |
2328 | if (prev && prev->va_end > base + start) { | |
2329 | next = prev; | |
2330 | prev = node_to_va(rb_prev(&next->rb_node)); | |
2331 | base = pvm_determine_end(&next, &prev, align) - end; | |
2332 | term_area = area; | |
2333 | continue; | |
2334 | } | |
2335 | ||
2336 | /* | |
2337 | * This area fits, move on to the previous one. If | |
2338 | * the previous one is the terminal one, we're done. | |
2339 | */ | |
2340 | area = (area + nr_vms - 1) % nr_vms; | |
2341 | if (area == term_area) | |
2342 | break; | |
2343 | start = offsets[area]; | |
2344 | end = start + sizes[area]; | |
2345 | pvm_find_next_prev(base + end, &next, &prev); | |
2346 | } | |
2347 | found: | |
2348 | /* we've found a fitting base, insert all va's */ | |
2349 | for (area = 0; area < nr_vms; area++) { | |
2350 | struct vmap_area *va = vas[area]; | |
2351 | ||
2352 | va->va_start = base + offsets[area]; | |
2353 | va->va_end = va->va_start + sizes[area]; | |
2354 | __insert_vmap_area(va); | |
2355 | } | |
2356 | ||
2357 | vmap_area_pcpu_hole = base + offsets[last_area]; | |
2358 | ||
2359 | spin_unlock(&vmap_area_lock); | |
2360 | ||
2361 | /* insert all vm's */ | |
2362 | for (area = 0; area < nr_vms; area++) | |
2363 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, | |
2364 | pcpu_get_vm_areas); | |
2365 | ||
2366 | kfree(vas); | |
2367 | return vms; | |
2368 | ||
2369 | err_free: | |
2370 | for (area = 0; area < nr_vms; area++) { | |
2371 | if (vas) | |
2372 | kfree(vas[area]); | |
2373 | if (vms) | |
2374 | kfree(vms[area]); | |
2375 | } | |
2376 | kfree(vas); | |
2377 | kfree(vms); | |
2378 | return NULL; | |
2379 | } | |
2380 | ||
2381 | /** | |
2382 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
2383 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
2384 | * @nr_vms: the number of allocated areas | |
2385 | * | |
2386 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
2387 | */ | |
2388 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
2389 | { | |
2390 | int i; | |
2391 | ||
2392 | for (i = 0; i < nr_vms; i++) | |
2393 | free_vm_area(vms[i]); | |
2394 | kfree(vms); | |
2395 | } | |
4f8b02b4 | 2396 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
2397 | |
2398 | #ifdef CONFIG_PROC_FS | |
2399 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e199b5d1 | 2400 | __acquires(&vmlist_lock) |
a10aa579 CL |
2401 | { |
2402 | loff_t n = *pos; | |
2403 | struct vm_struct *v; | |
2404 | ||
2405 | read_lock(&vmlist_lock); | |
2406 | v = vmlist; | |
2407 | while (n > 0 && v) { | |
2408 | n--; | |
2409 | v = v->next; | |
2410 | } | |
2411 | if (!n) | |
2412 | return v; | |
2413 | ||
2414 | return NULL; | |
2415 | ||
2416 | } | |
2417 | ||
2418 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
2419 | { | |
2420 | struct vm_struct *v = p; | |
2421 | ||
2422 | ++*pos; | |
2423 | return v->next; | |
2424 | } | |
2425 | ||
2426 | static void s_stop(struct seq_file *m, void *p) | |
e199b5d1 | 2427 | __releases(&vmlist_lock) |
a10aa579 CL |
2428 | { |
2429 | read_unlock(&vmlist_lock); | |
2430 | } | |
2431 | ||
a47a126a ED |
2432 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
2433 | { | |
2434 | if (NUMA_BUILD) { | |
2435 | unsigned int nr, *counters = m->private; | |
2436 | ||
2437 | if (!counters) | |
2438 | return; | |
2439 | ||
2440 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); | |
2441 | ||
2442 | for (nr = 0; nr < v->nr_pages; nr++) | |
2443 | counters[page_to_nid(v->pages[nr])]++; | |
2444 | ||
2445 | for_each_node_state(nr, N_HIGH_MEMORY) | |
2446 | if (counters[nr]) | |
2447 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
2448 | } | |
2449 | } | |
2450 | ||
a10aa579 CL |
2451 | static int s_show(struct seq_file *m, void *p) |
2452 | { | |
2453 | struct vm_struct *v = p; | |
2454 | ||
2455 | seq_printf(m, "0x%p-0x%p %7ld", | |
2456 | v->addr, v->addr + v->size, v->size); | |
2457 | ||
62c70bce JP |
2458 | if (v->caller) |
2459 | seq_printf(m, " %pS", v->caller); | |
23016969 | 2460 | |
a10aa579 CL |
2461 | if (v->nr_pages) |
2462 | seq_printf(m, " pages=%d", v->nr_pages); | |
2463 | ||
2464 | if (v->phys_addr) | |
ffa71f33 | 2465 | seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr); |
a10aa579 CL |
2466 | |
2467 | if (v->flags & VM_IOREMAP) | |
2468 | seq_printf(m, " ioremap"); | |
2469 | ||
2470 | if (v->flags & VM_ALLOC) | |
2471 | seq_printf(m, " vmalloc"); | |
2472 | ||
2473 | if (v->flags & VM_MAP) | |
2474 | seq_printf(m, " vmap"); | |
2475 | ||
2476 | if (v->flags & VM_USERMAP) | |
2477 | seq_printf(m, " user"); | |
2478 | ||
2479 | if (v->flags & VM_VPAGES) | |
2480 | seq_printf(m, " vpages"); | |
2481 | ||
a47a126a | 2482 | show_numa_info(m, v); |
a10aa579 CL |
2483 | seq_putc(m, '\n'); |
2484 | return 0; | |
2485 | } | |
2486 | ||
5f6a6a9c | 2487 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
2488 | .start = s_start, |
2489 | .next = s_next, | |
2490 | .stop = s_stop, | |
2491 | .show = s_show, | |
2492 | }; | |
5f6a6a9c AD |
2493 | |
2494 | static int vmalloc_open(struct inode *inode, struct file *file) | |
2495 | { | |
2496 | unsigned int *ptr = NULL; | |
2497 | int ret; | |
2498 | ||
51980ac9 | 2499 | if (NUMA_BUILD) { |
5f6a6a9c | 2500 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); |
51980ac9 KV |
2501 | if (ptr == NULL) |
2502 | return -ENOMEM; | |
2503 | } | |
5f6a6a9c AD |
2504 | ret = seq_open(file, &vmalloc_op); |
2505 | if (!ret) { | |
2506 | struct seq_file *m = file->private_data; | |
2507 | m->private = ptr; | |
2508 | } else | |
2509 | kfree(ptr); | |
2510 | return ret; | |
2511 | } | |
2512 | ||
2513 | static const struct file_operations proc_vmalloc_operations = { | |
2514 | .open = vmalloc_open, | |
2515 | .read = seq_read, | |
2516 | .llseek = seq_lseek, | |
2517 | .release = seq_release_private, | |
2518 | }; | |
2519 | ||
2520 | static int __init proc_vmalloc_init(void) | |
2521 | { | |
2522 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); | |
2523 | return 0; | |
2524 | } | |
2525 | module_init(proc_vmalloc_init); | |
a10aa579 CL |
2526 | #endif |
2527 |