]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Copyright (C) 1993 Linus Torvalds |
4 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
5 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
6 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 7 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
d758ffe6 | 8 | * Improving global KVA allocator, Uladzislau Rezki, Sony, May 2019 |
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> | |
c3edc401 | 15 | #include <linux/sched/signal.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> |
868b104d | 21 | #include <linux/set_memory.h> |
3ac7fe5a | 22 | #include <linux/debugobjects.h> |
23016969 | 23 | #include <linux/kallsyms.h> |
db64fe02 | 24 | #include <linux/list.h> |
4da56b99 | 25 | #include <linux/notifier.h> |
db64fe02 | 26 | #include <linux/rbtree.h> |
0f14599c | 27 | #include <linux/xarray.h> |
db64fe02 | 28 | #include <linux/rcupdate.h> |
f0aa6617 | 29 | #include <linux/pfn.h> |
89219d37 | 30 | #include <linux/kmemleak.h> |
60063497 | 31 | #include <linux/atomic.h> |
3b32123d | 32 | #include <linux/compiler.h> |
32fcfd40 | 33 | #include <linux/llist.h> |
0f616be1 | 34 | #include <linux/bitops.h> |
68ad4a33 | 35 | #include <linux/rbtree_augmented.h> |
bdebd6a2 | 36 | #include <linux/overflow.h> |
c0eb315a | 37 | #include <linux/pgtable.h> |
7c0f6ba6 | 38 | #include <linux/uaccess.h> |
1da177e4 | 39 | #include <asm/tlbflush.h> |
2dca6999 | 40 | #include <asm/shmparam.h> |
1da177e4 | 41 | |
dd56b046 | 42 | #include "internal.h" |
2a681cfa | 43 | #include "pgalloc-track.h" |
dd56b046 | 44 | |
186525bd IM |
45 | bool is_vmalloc_addr(const void *x) |
46 | { | |
47 | unsigned long addr = (unsigned long)x; | |
48 | ||
49 | return addr >= VMALLOC_START && addr < VMALLOC_END; | |
50 | } | |
51 | EXPORT_SYMBOL(is_vmalloc_addr); | |
52 | ||
32fcfd40 AV |
53 | struct vfree_deferred { |
54 | struct llist_head list; | |
55 | struct work_struct wq; | |
56 | }; | |
57 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
58 | ||
59 | static void __vunmap(const void *, int); | |
60 | ||
61 | static void free_work(struct work_struct *w) | |
62 | { | |
63 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
64 | struct llist_node *t, *llnode; |
65 | ||
66 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
67 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
68 | } |
69 | ||
db64fe02 | 70 | /*** Page table manipulation functions ***/ |
5e9e3d77 NP |
71 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
72 | phys_addr_t phys_addr, pgprot_t prot, | |
73 | pgtbl_mod_mask *mask) | |
74 | { | |
75 | pte_t *pte; | |
76 | u64 pfn; | |
77 | ||
78 | pfn = phys_addr >> PAGE_SHIFT; | |
79 | pte = pte_alloc_kernel_track(pmd, addr, mask); | |
80 | if (!pte) | |
81 | return -ENOMEM; | |
82 | do { | |
83 | BUG_ON(!pte_none(*pte)); | |
84 | set_pte_at(&init_mm, addr, pte, pfn_pte(pfn, prot)); | |
85 | pfn++; | |
86 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
87 | *mask |= PGTBL_PTE_MODIFIED; | |
88 | return 0; | |
89 | } | |
90 | ||
91 | static int vmap_try_huge_pmd(pmd_t *pmd, unsigned long addr, unsigned long end, | |
92 | phys_addr_t phys_addr, pgprot_t prot, | |
93 | unsigned int max_page_shift) | |
94 | { | |
95 | if (max_page_shift < PMD_SHIFT) | |
96 | return 0; | |
97 | ||
98 | if (!arch_vmap_pmd_supported(prot)) | |
99 | return 0; | |
100 | ||
101 | if ((end - addr) != PMD_SIZE) | |
102 | return 0; | |
103 | ||
104 | if (!IS_ALIGNED(addr, PMD_SIZE)) | |
105 | return 0; | |
106 | ||
107 | if (!IS_ALIGNED(phys_addr, PMD_SIZE)) | |
108 | return 0; | |
109 | ||
110 | if (pmd_present(*pmd) && !pmd_free_pte_page(pmd, addr)) | |
111 | return 0; | |
112 | ||
113 | return pmd_set_huge(pmd, phys_addr, prot); | |
114 | } | |
115 | ||
116 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, | |
117 | phys_addr_t phys_addr, pgprot_t prot, | |
118 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
119 | { | |
120 | pmd_t *pmd; | |
121 | unsigned long next; | |
122 | ||
123 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); | |
124 | if (!pmd) | |
125 | return -ENOMEM; | |
126 | do { | |
127 | next = pmd_addr_end(addr, end); | |
128 | ||
129 | if (vmap_try_huge_pmd(pmd, addr, next, phys_addr, prot, | |
130 | max_page_shift)) { | |
131 | *mask |= PGTBL_PMD_MODIFIED; | |
132 | continue; | |
133 | } | |
134 | ||
135 | if (vmap_pte_range(pmd, addr, next, phys_addr, prot, mask)) | |
136 | return -ENOMEM; | |
137 | } while (pmd++, phys_addr += (next - addr), addr = next, addr != end); | |
138 | return 0; | |
139 | } | |
140 | ||
141 | static int vmap_try_huge_pud(pud_t *pud, unsigned long addr, unsigned long end, | |
142 | phys_addr_t phys_addr, pgprot_t prot, | |
143 | unsigned int max_page_shift) | |
144 | { | |
145 | if (max_page_shift < PUD_SHIFT) | |
146 | return 0; | |
147 | ||
148 | if (!arch_vmap_pud_supported(prot)) | |
149 | return 0; | |
150 | ||
151 | if ((end - addr) != PUD_SIZE) | |
152 | return 0; | |
153 | ||
154 | if (!IS_ALIGNED(addr, PUD_SIZE)) | |
155 | return 0; | |
156 | ||
157 | if (!IS_ALIGNED(phys_addr, PUD_SIZE)) | |
158 | return 0; | |
159 | ||
160 | if (pud_present(*pud) && !pud_free_pmd_page(pud, addr)) | |
161 | return 0; | |
162 | ||
163 | return pud_set_huge(pud, phys_addr, prot); | |
164 | } | |
165 | ||
166 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, | |
167 | phys_addr_t phys_addr, pgprot_t prot, | |
168 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
169 | { | |
170 | pud_t *pud; | |
171 | unsigned long next; | |
172 | ||
173 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); | |
174 | if (!pud) | |
175 | return -ENOMEM; | |
176 | do { | |
177 | next = pud_addr_end(addr, end); | |
178 | ||
179 | if (vmap_try_huge_pud(pud, addr, next, phys_addr, prot, | |
180 | max_page_shift)) { | |
181 | *mask |= PGTBL_PUD_MODIFIED; | |
182 | continue; | |
183 | } | |
184 | ||
185 | if (vmap_pmd_range(pud, addr, next, phys_addr, prot, | |
186 | max_page_shift, mask)) | |
187 | return -ENOMEM; | |
188 | } while (pud++, phys_addr += (next - addr), addr = next, addr != end); | |
189 | return 0; | |
190 | } | |
191 | ||
192 | static int vmap_try_huge_p4d(p4d_t *p4d, unsigned long addr, unsigned long end, | |
193 | phys_addr_t phys_addr, pgprot_t prot, | |
194 | unsigned int max_page_shift) | |
195 | { | |
196 | if (max_page_shift < P4D_SHIFT) | |
197 | return 0; | |
198 | ||
199 | if (!arch_vmap_p4d_supported(prot)) | |
200 | return 0; | |
201 | ||
202 | if ((end - addr) != P4D_SIZE) | |
203 | return 0; | |
204 | ||
205 | if (!IS_ALIGNED(addr, P4D_SIZE)) | |
206 | return 0; | |
207 | ||
208 | if (!IS_ALIGNED(phys_addr, P4D_SIZE)) | |
209 | return 0; | |
210 | ||
211 | if (p4d_present(*p4d) && !p4d_free_pud_page(p4d, addr)) | |
212 | return 0; | |
213 | ||
214 | return p4d_set_huge(p4d, phys_addr, prot); | |
215 | } | |
216 | ||
217 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, | |
218 | phys_addr_t phys_addr, pgprot_t prot, | |
219 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
220 | { | |
221 | p4d_t *p4d; | |
222 | unsigned long next; | |
223 | ||
224 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); | |
225 | if (!p4d) | |
226 | return -ENOMEM; | |
227 | do { | |
228 | next = p4d_addr_end(addr, end); | |
229 | ||
230 | if (vmap_try_huge_p4d(p4d, addr, next, phys_addr, prot, | |
231 | max_page_shift)) { | |
232 | *mask |= PGTBL_P4D_MODIFIED; | |
233 | continue; | |
234 | } | |
235 | ||
236 | if (vmap_pud_range(p4d, addr, next, phys_addr, prot, | |
237 | max_page_shift, mask)) | |
238 | return -ENOMEM; | |
239 | } while (p4d++, phys_addr += (next - addr), addr = next, addr != end); | |
240 | return 0; | |
241 | } | |
242 | ||
5d87510d | 243 | static int vmap_range_noflush(unsigned long addr, unsigned long end, |
5e9e3d77 NP |
244 | phys_addr_t phys_addr, pgprot_t prot, |
245 | unsigned int max_page_shift) | |
246 | { | |
247 | pgd_t *pgd; | |
248 | unsigned long start; | |
249 | unsigned long next; | |
250 | int err; | |
251 | pgtbl_mod_mask mask = 0; | |
252 | ||
253 | might_sleep(); | |
254 | BUG_ON(addr >= end); | |
255 | ||
256 | start = addr; | |
257 | pgd = pgd_offset_k(addr); | |
258 | do { | |
259 | next = pgd_addr_end(addr, end); | |
260 | err = vmap_p4d_range(pgd, addr, next, phys_addr, prot, | |
261 | max_page_shift, &mask); | |
262 | if (err) | |
263 | break; | |
264 | } while (pgd++, phys_addr += (next - addr), addr = next, addr != end); | |
265 | ||
5e9e3d77 NP |
266 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
267 | arch_sync_kernel_mappings(start, end); | |
268 | ||
269 | return err; | |
270 | } | |
b221385b | 271 | |
5d87510d NP |
272 | int vmap_range(unsigned long addr, unsigned long end, |
273 | phys_addr_t phys_addr, pgprot_t prot, | |
274 | unsigned int max_page_shift) | |
275 | { | |
276 | int err; | |
277 | ||
278 | err = vmap_range_noflush(addr, end, phys_addr, prot, max_page_shift); | |
279 | flush_cache_vmap(addr, end); | |
280 | ||
281 | return err; | |
282 | } | |
283 | ||
2ba3e694 JR |
284 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
285 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
286 | { |
287 | pte_t *pte; | |
288 | ||
289 | pte = pte_offset_kernel(pmd, addr); | |
290 | do { | |
291 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
292 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
293 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
2ba3e694 | 294 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
295 | } |
296 | ||
2ba3e694 JR |
297 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, |
298 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
299 | { |
300 | pmd_t *pmd; | |
301 | unsigned long next; | |
2ba3e694 | 302 | int cleared; |
1da177e4 LT |
303 | |
304 | pmd = pmd_offset(pud, addr); | |
305 | do { | |
306 | next = pmd_addr_end(addr, end); | |
2ba3e694 JR |
307 | |
308 | cleared = pmd_clear_huge(pmd); | |
309 | if (cleared || pmd_bad(*pmd)) | |
310 | *mask |= PGTBL_PMD_MODIFIED; | |
311 | ||
312 | if (cleared) | |
b9820d8f | 313 | continue; |
1da177e4 LT |
314 | if (pmd_none_or_clear_bad(pmd)) |
315 | continue; | |
2ba3e694 | 316 | vunmap_pte_range(pmd, addr, next, mask); |
e47110e9 AK |
317 | |
318 | cond_resched(); | |
1da177e4 LT |
319 | } while (pmd++, addr = next, addr != end); |
320 | } | |
321 | ||
2ba3e694 JR |
322 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, |
323 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
324 | { |
325 | pud_t *pud; | |
326 | unsigned long next; | |
2ba3e694 | 327 | int cleared; |
1da177e4 | 328 | |
c2febafc | 329 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
330 | do { |
331 | next = pud_addr_end(addr, end); | |
2ba3e694 JR |
332 | |
333 | cleared = pud_clear_huge(pud); | |
334 | if (cleared || pud_bad(*pud)) | |
335 | *mask |= PGTBL_PUD_MODIFIED; | |
336 | ||
337 | if (cleared) | |
b9820d8f | 338 | continue; |
1da177e4 LT |
339 | if (pud_none_or_clear_bad(pud)) |
340 | continue; | |
2ba3e694 | 341 | vunmap_pmd_range(pud, addr, next, mask); |
1da177e4 LT |
342 | } while (pud++, addr = next, addr != end); |
343 | } | |
344 | ||
2ba3e694 JR |
345 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, |
346 | pgtbl_mod_mask *mask) | |
c2febafc KS |
347 | { |
348 | p4d_t *p4d; | |
349 | unsigned long next; | |
2ba3e694 | 350 | int cleared; |
c2febafc KS |
351 | |
352 | p4d = p4d_offset(pgd, addr); | |
353 | do { | |
354 | next = p4d_addr_end(addr, end); | |
2ba3e694 JR |
355 | |
356 | cleared = p4d_clear_huge(p4d); | |
357 | if (cleared || p4d_bad(*p4d)) | |
358 | *mask |= PGTBL_P4D_MODIFIED; | |
359 | ||
360 | if (cleared) | |
c2febafc KS |
361 | continue; |
362 | if (p4d_none_or_clear_bad(p4d)) | |
363 | continue; | |
2ba3e694 | 364 | vunmap_pud_range(p4d, addr, next, mask); |
c2febafc KS |
365 | } while (p4d++, addr = next, addr != end); |
366 | } | |
367 | ||
b521c43f CH |
368 | /** |
369 | * unmap_kernel_range_noflush - unmap kernel VM area | |
2ba3e694 | 370 | * @start: start of the VM area to unmap |
b521c43f CH |
371 | * @size: size of the VM area to unmap |
372 | * | |
373 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size specify | |
374 | * should have been allocated using get_vm_area() and its friends. | |
375 | * | |
376 | * NOTE: | |
377 | * This function does NOT do any cache flushing. The caller is responsible | |
378 | * for calling flush_cache_vunmap() on to-be-mapped areas before calling this | |
379 | * function and flush_tlb_kernel_range() after. | |
380 | */ | |
2ba3e694 | 381 | void unmap_kernel_range_noflush(unsigned long start, unsigned long size) |
1da177e4 | 382 | { |
2ba3e694 | 383 | unsigned long end = start + size; |
1da177e4 | 384 | unsigned long next; |
b521c43f | 385 | pgd_t *pgd; |
2ba3e694 JR |
386 | unsigned long addr = start; |
387 | pgtbl_mod_mask mask = 0; | |
1da177e4 LT |
388 | |
389 | BUG_ON(addr >= end); | |
390 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
391 | do { |
392 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
393 | if (pgd_bad(*pgd)) |
394 | mask |= PGTBL_PGD_MODIFIED; | |
1da177e4 LT |
395 | if (pgd_none_or_clear_bad(pgd)) |
396 | continue; | |
2ba3e694 | 397 | vunmap_p4d_range(pgd, addr, next, &mask); |
1da177e4 | 398 | } while (pgd++, addr = next, addr != end); |
2ba3e694 JR |
399 | |
400 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) | |
401 | arch_sync_kernel_mappings(start, end); | |
1da177e4 LT |
402 | } |
403 | ||
0a264884 | 404 | static int vmap_pages_pte_range(pmd_t *pmd, unsigned long addr, |
2ba3e694 JR |
405 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
406 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
407 | { |
408 | pte_t *pte; | |
409 | ||
db64fe02 NP |
410 | /* |
411 | * nr is a running index into the array which helps higher level | |
412 | * callers keep track of where we're up to. | |
413 | */ | |
414 | ||
2ba3e694 | 415 | pte = pte_alloc_kernel_track(pmd, addr, mask); |
1da177e4 LT |
416 | if (!pte) |
417 | return -ENOMEM; | |
418 | do { | |
db64fe02 NP |
419 | struct page *page = pages[*nr]; |
420 | ||
421 | if (WARN_ON(!pte_none(*pte))) | |
422 | return -EBUSY; | |
423 | if (WARN_ON(!page)) | |
1da177e4 LT |
424 | return -ENOMEM; |
425 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 426 | (*nr)++; |
1da177e4 | 427 | } while (pte++, addr += PAGE_SIZE, addr != end); |
2ba3e694 | 428 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
429 | return 0; |
430 | } | |
431 | ||
0a264884 | 432 | static int vmap_pages_pmd_range(pud_t *pud, unsigned long addr, |
2ba3e694 JR |
433 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
434 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
435 | { |
436 | pmd_t *pmd; | |
437 | unsigned long next; | |
438 | ||
2ba3e694 | 439 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); |
1da177e4 LT |
440 | if (!pmd) |
441 | return -ENOMEM; | |
442 | do { | |
443 | next = pmd_addr_end(addr, end); | |
0a264884 | 444 | if (vmap_pages_pte_range(pmd, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
445 | return -ENOMEM; |
446 | } while (pmd++, addr = next, addr != end); | |
447 | return 0; | |
448 | } | |
449 | ||
0a264884 | 450 | static int vmap_pages_pud_range(p4d_t *p4d, unsigned long addr, |
2ba3e694 JR |
451 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
452 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
453 | { |
454 | pud_t *pud; | |
455 | unsigned long next; | |
456 | ||
2ba3e694 | 457 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); |
1da177e4 LT |
458 | if (!pud) |
459 | return -ENOMEM; | |
460 | do { | |
461 | next = pud_addr_end(addr, end); | |
0a264884 | 462 | if (vmap_pages_pmd_range(pud, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
463 | return -ENOMEM; |
464 | } while (pud++, addr = next, addr != end); | |
465 | return 0; | |
466 | } | |
467 | ||
0a264884 | 468 | static int vmap_pages_p4d_range(pgd_t *pgd, unsigned long addr, |
2ba3e694 JR |
469 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
470 | pgtbl_mod_mask *mask) | |
c2febafc KS |
471 | { |
472 | p4d_t *p4d; | |
473 | unsigned long next; | |
474 | ||
2ba3e694 | 475 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); |
c2febafc KS |
476 | if (!p4d) |
477 | return -ENOMEM; | |
478 | do { | |
479 | next = p4d_addr_end(addr, end); | |
0a264884 | 480 | if (vmap_pages_pud_range(p4d, addr, next, prot, pages, nr, mask)) |
c2febafc KS |
481 | return -ENOMEM; |
482 | } while (p4d++, addr = next, addr != end); | |
483 | return 0; | |
484 | } | |
485 | ||
b521c43f CH |
486 | /** |
487 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
488 | * @addr: start of the VM area to map | |
489 | * @size: size of the VM area to map | |
490 | * @prot: page protection flags to use | |
491 | * @pages: pages to map | |
db64fe02 | 492 | * |
b521c43f CH |
493 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size specify should |
494 | * have been allocated using get_vm_area() and its friends. | |
495 | * | |
496 | * NOTE: | |
497 | * This function does NOT do any cache flushing. The caller is responsible for | |
498 | * calling flush_cache_vmap() on to-be-mapped areas before calling this | |
499 | * function. | |
500 | * | |
501 | * RETURNS: | |
60bb4465 | 502 | * 0 on success, -errno on failure. |
db64fe02 | 503 | */ |
b521c43f CH |
504 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, |
505 | pgprot_t prot, struct page **pages) | |
1da177e4 | 506 | { |
2ba3e694 | 507 | unsigned long start = addr; |
b521c43f | 508 | unsigned long end = addr + size; |
1da177e4 | 509 | unsigned long next; |
b521c43f | 510 | pgd_t *pgd; |
db64fe02 NP |
511 | int err = 0; |
512 | int nr = 0; | |
2ba3e694 | 513 | pgtbl_mod_mask mask = 0; |
1da177e4 LT |
514 | |
515 | BUG_ON(addr >= end); | |
516 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
517 | do { |
518 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
519 | if (pgd_bad(*pgd)) |
520 | mask |= PGTBL_PGD_MODIFIED; | |
0a264884 | 521 | err = vmap_pages_p4d_range(pgd, addr, next, prot, pages, &nr, &mask); |
1da177e4 | 522 | if (err) |
bf88c8c8 | 523 | return err; |
1da177e4 | 524 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 525 | |
2ba3e694 JR |
526 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
527 | arch_sync_kernel_mappings(start, end); | |
528 | ||
60bb4465 | 529 | return 0; |
1da177e4 LT |
530 | } |
531 | ||
ed1f324c CH |
532 | int map_kernel_range(unsigned long start, unsigned long size, pgprot_t prot, |
533 | struct page **pages) | |
8fc48985 TH |
534 | { |
535 | int ret; | |
536 | ||
a29adb62 CH |
537 | ret = map_kernel_range_noflush(start, size, prot, pages); |
538 | flush_cache_vmap(start, start + size); | |
8fc48985 TH |
539 | return ret; |
540 | } | |
541 | ||
81ac3ad9 | 542 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
543 | { |
544 | /* | |
ab4f2ee1 | 545 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
546 | * and fall back on vmalloc() if that fails. Others |
547 | * just put it in the vmalloc space. | |
548 | */ | |
549 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
550 | unsigned long addr = (unsigned long)x; | |
551 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
552 | return 1; | |
553 | #endif | |
554 | return is_vmalloc_addr(x); | |
555 | } | |
556 | ||
48667e7a | 557 | /* |
c0eb315a NP |
558 | * Walk a vmap address to the struct page it maps. Huge vmap mappings will |
559 | * return the tail page that corresponds to the base page address, which | |
560 | * matches small vmap mappings. | |
48667e7a | 561 | */ |
add688fb | 562 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
563 | { |
564 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 565 | struct page *page = NULL; |
48667e7a | 566 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
567 | p4d_t *p4d; |
568 | pud_t *pud; | |
569 | pmd_t *pmd; | |
570 | pte_t *ptep, pte; | |
48667e7a | 571 | |
7aa413de IM |
572 | /* |
573 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
574 | * architectures that do not vmalloc module space | |
575 | */ | |
73bdf0a6 | 576 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 577 | |
c2febafc KS |
578 | if (pgd_none(*pgd)) |
579 | return NULL; | |
c0eb315a NP |
580 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
581 | return NULL; /* XXX: no allowance for huge pgd */ | |
582 | if (WARN_ON_ONCE(pgd_bad(*pgd))) | |
583 | return NULL; | |
584 | ||
c2febafc KS |
585 | p4d = p4d_offset(pgd, addr); |
586 | if (p4d_none(*p4d)) | |
587 | return NULL; | |
c0eb315a NP |
588 | if (p4d_leaf(*p4d)) |
589 | return p4d_page(*p4d) + ((addr & ~P4D_MASK) >> PAGE_SHIFT); | |
590 | if (WARN_ON_ONCE(p4d_bad(*p4d))) | |
591 | return NULL; | |
029c54b0 | 592 | |
c0eb315a NP |
593 | pud = pud_offset(p4d, addr); |
594 | if (pud_none(*pud)) | |
595 | return NULL; | |
596 | if (pud_leaf(*pud)) | |
597 | return pud_page(*pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
598 | if (WARN_ON_ONCE(pud_bad(*pud))) | |
c2febafc | 599 | return NULL; |
c0eb315a | 600 | |
c2febafc | 601 | pmd = pmd_offset(pud, addr); |
c0eb315a NP |
602 | if (pmd_none(*pmd)) |
603 | return NULL; | |
604 | if (pmd_leaf(*pmd)) | |
605 | return pmd_page(*pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
606 | if (WARN_ON_ONCE(pmd_bad(*pmd))) | |
c2febafc KS |
607 | return NULL; |
608 | ||
609 | ptep = pte_offset_map(pmd, addr); | |
610 | pte = *ptep; | |
611 | if (pte_present(pte)) | |
612 | page = pte_page(pte); | |
613 | pte_unmap(ptep); | |
c0eb315a | 614 | |
add688fb | 615 | return page; |
48667e7a | 616 | } |
add688fb | 617 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
618 | |
619 | /* | |
add688fb | 620 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 621 | */ |
add688fb | 622 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 623 | { |
add688fb | 624 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 625 | } |
add688fb | 626 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 627 | |
db64fe02 NP |
628 | |
629 | /*** Global kva allocator ***/ | |
630 | ||
bb850f4d | 631 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 632 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 633 | |
db64fe02 | 634 | |
db64fe02 | 635 | static DEFINE_SPINLOCK(vmap_area_lock); |
e36176be | 636 | static DEFINE_SPINLOCK(free_vmap_area_lock); |
f1c4069e JK |
637 | /* Export for kexec only */ |
638 | LIST_HEAD(vmap_area_list); | |
89699605 | 639 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 640 | static bool vmap_initialized __read_mostly; |
89699605 | 641 | |
96e2db45 URS |
642 | static struct rb_root purge_vmap_area_root = RB_ROOT; |
643 | static LIST_HEAD(purge_vmap_area_list); | |
644 | static DEFINE_SPINLOCK(purge_vmap_area_lock); | |
645 | ||
68ad4a33 URS |
646 | /* |
647 | * This kmem_cache is used for vmap_area objects. Instead of | |
648 | * allocating from slab we reuse an object from this cache to | |
649 | * make things faster. Especially in "no edge" splitting of | |
650 | * free block. | |
651 | */ | |
652 | static struct kmem_cache *vmap_area_cachep; | |
653 | ||
654 | /* | |
655 | * This linked list is used in pair with free_vmap_area_root. | |
656 | * It gives O(1) access to prev/next to perform fast coalescing. | |
657 | */ | |
658 | static LIST_HEAD(free_vmap_area_list); | |
659 | ||
660 | /* | |
661 | * This augment red-black tree represents the free vmap space. | |
662 | * All vmap_area objects in this tree are sorted by va->va_start | |
663 | * address. It is used for allocation and merging when a vmap | |
664 | * object is released. | |
665 | * | |
666 | * Each vmap_area node contains a maximum available free block | |
667 | * of its sub-tree, right or left. Therefore it is possible to | |
668 | * find a lowest match of free area. | |
669 | */ | |
670 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
671 | ||
82dd23e8 URS |
672 | /* |
673 | * Preload a CPU with one object for "no edge" split case. The | |
674 | * aim is to get rid of allocations from the atomic context, thus | |
675 | * to use more permissive allocation masks. | |
676 | */ | |
677 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
678 | ||
68ad4a33 URS |
679 | static __always_inline unsigned long |
680 | va_size(struct vmap_area *va) | |
681 | { | |
682 | return (va->va_end - va->va_start); | |
683 | } | |
684 | ||
685 | static __always_inline unsigned long | |
686 | get_subtree_max_size(struct rb_node *node) | |
687 | { | |
688 | struct vmap_area *va; | |
689 | ||
690 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
691 | return va ? va->subtree_max_size : 0; | |
692 | } | |
89699605 | 693 | |
68ad4a33 URS |
694 | /* |
695 | * Gets called when remove the node and rotate. | |
696 | */ | |
697 | static __always_inline unsigned long | |
698 | compute_subtree_max_size(struct vmap_area *va) | |
699 | { | |
700 | return max3(va_size(va), | |
701 | get_subtree_max_size(va->rb_node.rb_left), | |
702 | get_subtree_max_size(va->rb_node.rb_right)); | |
703 | } | |
704 | ||
315cc066 ML |
705 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
706 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 URS |
707 | |
708 | static void purge_vmap_area_lazy(void); | |
709 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
710 | static unsigned long lazy_max_pages(void); | |
db64fe02 | 711 | |
97105f0a RG |
712 | static atomic_long_t nr_vmalloc_pages; |
713 | ||
714 | unsigned long vmalloc_nr_pages(void) | |
715 | { | |
716 | return atomic_long_read(&nr_vmalloc_pages); | |
717 | } | |
718 | ||
db64fe02 | 719 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 720 | { |
db64fe02 NP |
721 | struct rb_node *n = vmap_area_root.rb_node; |
722 | ||
723 | while (n) { | |
724 | struct vmap_area *va; | |
725 | ||
726 | va = rb_entry(n, struct vmap_area, rb_node); | |
727 | if (addr < va->va_start) | |
728 | n = n->rb_left; | |
cef2ac3f | 729 | else if (addr >= va->va_end) |
db64fe02 NP |
730 | n = n->rb_right; |
731 | else | |
732 | return va; | |
733 | } | |
734 | ||
735 | return NULL; | |
736 | } | |
737 | ||
68ad4a33 URS |
738 | /* |
739 | * This function returns back addresses of parent node | |
740 | * and its left or right link for further processing. | |
9c801f61 URS |
741 | * |
742 | * Otherwise NULL is returned. In that case all further | |
743 | * steps regarding inserting of conflicting overlap range | |
744 | * have to be declined and actually considered as a bug. | |
68ad4a33 URS |
745 | */ |
746 | static __always_inline struct rb_node ** | |
747 | find_va_links(struct vmap_area *va, | |
748 | struct rb_root *root, struct rb_node *from, | |
749 | struct rb_node **parent) | |
750 | { | |
751 | struct vmap_area *tmp_va; | |
752 | struct rb_node **link; | |
753 | ||
754 | if (root) { | |
755 | link = &root->rb_node; | |
756 | if (unlikely(!*link)) { | |
757 | *parent = NULL; | |
758 | return link; | |
759 | } | |
760 | } else { | |
761 | link = &from; | |
762 | } | |
db64fe02 | 763 | |
68ad4a33 URS |
764 | /* |
765 | * Go to the bottom of the tree. When we hit the last point | |
766 | * we end up with parent rb_node and correct direction, i name | |
767 | * it link, where the new va->rb_node will be attached to. | |
768 | */ | |
769 | do { | |
770 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 771 | |
68ad4a33 URS |
772 | /* |
773 | * During the traversal we also do some sanity check. | |
774 | * Trigger the BUG() if there are sides(left/right) | |
775 | * or full overlaps. | |
776 | */ | |
777 | if (va->va_start < tmp_va->va_end && | |
778 | va->va_end <= tmp_va->va_start) | |
779 | link = &(*link)->rb_left; | |
780 | else if (va->va_end > tmp_va->va_start && | |
781 | va->va_start >= tmp_va->va_end) | |
782 | link = &(*link)->rb_right; | |
9c801f61 URS |
783 | else { |
784 | WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n", | |
785 | va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end); | |
786 | ||
787 | return NULL; | |
788 | } | |
68ad4a33 URS |
789 | } while (*link); |
790 | ||
791 | *parent = &tmp_va->rb_node; | |
792 | return link; | |
793 | } | |
794 | ||
795 | static __always_inline struct list_head * | |
796 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
797 | { | |
798 | struct list_head *list; | |
799 | ||
800 | if (unlikely(!parent)) | |
801 | /* | |
802 | * The red-black tree where we try to find VA neighbors | |
803 | * before merging or inserting is empty, i.e. it means | |
804 | * there is no free vmap space. Normally it does not | |
805 | * happen but we handle this case anyway. | |
806 | */ | |
807 | return NULL; | |
808 | ||
809 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
810 | return (&parent->rb_right == link ? list->next : list); | |
811 | } | |
812 | ||
813 | static __always_inline void | |
814 | link_va(struct vmap_area *va, struct rb_root *root, | |
815 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
816 | { | |
817 | /* | |
818 | * VA is still not in the list, but we can | |
819 | * identify its future previous list_head node. | |
820 | */ | |
821 | if (likely(parent)) { | |
822 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
823 | if (&parent->rb_right != link) | |
824 | head = head->prev; | |
db64fe02 NP |
825 | } |
826 | ||
68ad4a33 URS |
827 | /* Insert to the rb-tree */ |
828 | rb_link_node(&va->rb_node, parent, link); | |
829 | if (root == &free_vmap_area_root) { | |
830 | /* | |
831 | * Some explanation here. Just perform simple insertion | |
832 | * to the tree. We do not set va->subtree_max_size to | |
833 | * its current size before calling rb_insert_augmented(). | |
834 | * It is because of we populate the tree from the bottom | |
835 | * to parent levels when the node _is_ in the tree. | |
836 | * | |
837 | * Therefore we set subtree_max_size to zero after insertion, | |
838 | * to let __augment_tree_propagate_from() puts everything to | |
839 | * the correct order later on. | |
840 | */ | |
841 | rb_insert_augmented(&va->rb_node, | |
842 | root, &free_vmap_area_rb_augment_cb); | |
843 | va->subtree_max_size = 0; | |
844 | } else { | |
845 | rb_insert_color(&va->rb_node, root); | |
846 | } | |
db64fe02 | 847 | |
68ad4a33 URS |
848 | /* Address-sort this list */ |
849 | list_add(&va->list, head); | |
db64fe02 NP |
850 | } |
851 | ||
68ad4a33 URS |
852 | static __always_inline void |
853 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
854 | { | |
460e42d1 URS |
855 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
856 | return; | |
db64fe02 | 857 | |
460e42d1 URS |
858 | if (root == &free_vmap_area_root) |
859 | rb_erase_augmented(&va->rb_node, | |
860 | root, &free_vmap_area_rb_augment_cb); | |
861 | else | |
862 | rb_erase(&va->rb_node, root); | |
863 | ||
864 | list_del(&va->list); | |
865 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
866 | } |
867 | ||
bb850f4d URS |
868 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
869 | static void | |
da27c9ed | 870 | augment_tree_propagate_check(void) |
bb850f4d URS |
871 | { |
872 | struct vmap_area *va; | |
da27c9ed | 873 | unsigned long computed_size; |
bb850f4d | 874 | |
da27c9ed URS |
875 | list_for_each_entry(va, &free_vmap_area_list, list) { |
876 | computed_size = compute_subtree_max_size(va); | |
877 | if (computed_size != va->subtree_max_size) | |
878 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
879 | va_size(va), va->subtree_max_size); | |
bb850f4d | 880 | } |
bb850f4d URS |
881 | } |
882 | #endif | |
883 | ||
68ad4a33 URS |
884 | /* |
885 | * This function populates subtree_max_size from bottom to upper | |
886 | * levels starting from VA point. The propagation must be done | |
887 | * when VA size is modified by changing its va_start/va_end. Or | |
888 | * in case of newly inserting of VA to the tree. | |
889 | * | |
890 | * It means that __augment_tree_propagate_from() must be called: | |
891 | * - After VA has been inserted to the tree(free path); | |
892 | * - After VA has been shrunk(allocation path); | |
893 | * - After VA has been increased(merging path). | |
894 | * | |
895 | * Please note that, it does not mean that upper parent nodes | |
896 | * and their subtree_max_size are recalculated all the time up | |
897 | * to the root node. | |
898 | * | |
899 | * 4--8 | |
900 | * /\ | |
901 | * / \ | |
902 | * / \ | |
903 | * 2--2 8--8 | |
904 | * | |
905 | * For example if we modify the node 4, shrinking it to 2, then | |
906 | * no any modification is required. If we shrink the node 2 to 1 | |
907 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
908 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
909 | * node becomes 4--6. | |
910 | */ | |
911 | static __always_inline void | |
912 | augment_tree_propagate_from(struct vmap_area *va) | |
913 | { | |
15ae144f URS |
914 | /* |
915 | * Populate the tree from bottom towards the root until | |
916 | * the calculated maximum available size of checked node | |
917 | * is equal to its current one. | |
918 | */ | |
919 | free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL); | |
bb850f4d URS |
920 | |
921 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
da27c9ed | 922 | augment_tree_propagate_check(); |
bb850f4d | 923 | #endif |
68ad4a33 URS |
924 | } |
925 | ||
926 | static void | |
927 | insert_vmap_area(struct vmap_area *va, | |
928 | struct rb_root *root, struct list_head *head) | |
929 | { | |
930 | struct rb_node **link; | |
931 | struct rb_node *parent; | |
932 | ||
933 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
934 | if (link) |
935 | link_va(va, root, parent, link, head); | |
68ad4a33 URS |
936 | } |
937 | ||
938 | static void | |
939 | insert_vmap_area_augment(struct vmap_area *va, | |
940 | struct rb_node *from, struct rb_root *root, | |
941 | struct list_head *head) | |
942 | { | |
943 | struct rb_node **link; | |
944 | struct rb_node *parent; | |
945 | ||
946 | if (from) | |
947 | link = find_va_links(va, NULL, from, &parent); | |
948 | else | |
949 | link = find_va_links(va, root, NULL, &parent); | |
950 | ||
9c801f61 URS |
951 | if (link) { |
952 | link_va(va, root, parent, link, head); | |
953 | augment_tree_propagate_from(va); | |
954 | } | |
68ad4a33 URS |
955 | } |
956 | ||
957 | /* | |
958 | * Merge de-allocated chunk of VA memory with previous | |
959 | * and next free blocks. If coalesce is not done a new | |
960 | * free area is inserted. If VA has been merged, it is | |
961 | * freed. | |
9c801f61 URS |
962 | * |
963 | * Please note, it can return NULL in case of overlap | |
964 | * ranges, followed by WARN() report. Despite it is a | |
965 | * buggy behaviour, a system can be alive and keep | |
966 | * ongoing. | |
68ad4a33 | 967 | */ |
3c5c3cfb | 968 | static __always_inline struct vmap_area * |
68ad4a33 URS |
969 | merge_or_add_vmap_area(struct vmap_area *va, |
970 | struct rb_root *root, struct list_head *head) | |
971 | { | |
972 | struct vmap_area *sibling; | |
973 | struct list_head *next; | |
974 | struct rb_node **link; | |
975 | struct rb_node *parent; | |
976 | bool merged = false; | |
977 | ||
978 | /* | |
979 | * Find a place in the tree where VA potentially will be | |
980 | * inserted, unless it is merged with its sibling/siblings. | |
981 | */ | |
982 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
983 | if (!link) |
984 | return NULL; | |
68ad4a33 URS |
985 | |
986 | /* | |
987 | * Get next node of VA to check if merging can be done. | |
988 | */ | |
989 | next = get_va_next_sibling(parent, link); | |
990 | if (unlikely(next == NULL)) | |
991 | goto insert; | |
992 | ||
993 | /* | |
994 | * start end | |
995 | * | | | |
996 | * |<------VA------>|<-----Next----->| | |
997 | * | | | |
998 | * start end | |
999 | */ | |
1000 | if (next != head) { | |
1001 | sibling = list_entry(next, struct vmap_area, list); | |
1002 | if (sibling->va_start == va->va_end) { | |
1003 | sibling->va_start = va->va_start; | |
1004 | ||
68ad4a33 URS |
1005 | /* Free vmap_area object. */ |
1006 | kmem_cache_free(vmap_area_cachep, va); | |
1007 | ||
1008 | /* Point to the new merged area. */ | |
1009 | va = sibling; | |
1010 | merged = true; | |
1011 | } | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * start end | |
1016 | * | | | |
1017 | * |<-----Prev----->|<------VA------>| | |
1018 | * | | | |
1019 | * start end | |
1020 | */ | |
1021 | if (next->prev != head) { | |
1022 | sibling = list_entry(next->prev, struct vmap_area, list); | |
1023 | if (sibling->va_end == va->va_start) { | |
5dd78640 URS |
1024 | /* |
1025 | * If both neighbors are coalesced, it is important | |
1026 | * to unlink the "next" node first, followed by merging | |
1027 | * with "previous" one. Otherwise the tree might not be | |
1028 | * fully populated if a sibling's augmented value is | |
1029 | * "normalized" because of rotation operations. | |
1030 | */ | |
54f63d9d URS |
1031 | if (merged) |
1032 | unlink_va(va, root); | |
68ad4a33 | 1033 | |
5dd78640 URS |
1034 | sibling->va_end = va->va_end; |
1035 | ||
68ad4a33 URS |
1036 | /* Free vmap_area object. */ |
1037 | kmem_cache_free(vmap_area_cachep, va); | |
3c5c3cfb DA |
1038 | |
1039 | /* Point to the new merged area. */ | |
1040 | va = sibling; | |
1041 | merged = true; | |
68ad4a33 URS |
1042 | } |
1043 | } | |
1044 | ||
1045 | insert: | |
5dd78640 | 1046 | if (!merged) |
68ad4a33 | 1047 | link_va(va, root, parent, link, head); |
3c5c3cfb | 1048 | |
96e2db45 URS |
1049 | return va; |
1050 | } | |
1051 | ||
1052 | static __always_inline struct vmap_area * | |
1053 | merge_or_add_vmap_area_augment(struct vmap_area *va, | |
1054 | struct rb_root *root, struct list_head *head) | |
1055 | { | |
1056 | va = merge_or_add_vmap_area(va, root, head); | |
1057 | if (va) | |
1058 | augment_tree_propagate_from(va); | |
1059 | ||
3c5c3cfb | 1060 | return va; |
68ad4a33 URS |
1061 | } |
1062 | ||
1063 | static __always_inline bool | |
1064 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
1065 | unsigned long align, unsigned long vstart) | |
1066 | { | |
1067 | unsigned long nva_start_addr; | |
1068 | ||
1069 | if (va->va_start > vstart) | |
1070 | nva_start_addr = ALIGN(va->va_start, align); | |
1071 | else | |
1072 | nva_start_addr = ALIGN(vstart, align); | |
1073 | ||
1074 | /* Can be overflowed due to big size or alignment. */ | |
1075 | if (nva_start_addr + size < nva_start_addr || | |
1076 | nva_start_addr < vstart) | |
1077 | return false; | |
1078 | ||
1079 | return (nva_start_addr + size <= va->va_end); | |
1080 | } | |
1081 | ||
1082 | /* | |
1083 | * Find the first free block(lowest start address) in the tree, | |
1084 | * that will accomplish the request corresponding to passing | |
1085 | * parameters. | |
1086 | */ | |
1087 | static __always_inline struct vmap_area * | |
1088 | find_vmap_lowest_match(unsigned long size, | |
1089 | unsigned long align, unsigned long vstart) | |
1090 | { | |
1091 | struct vmap_area *va; | |
1092 | struct rb_node *node; | |
1093 | unsigned long length; | |
1094 | ||
1095 | /* Start from the root. */ | |
1096 | node = free_vmap_area_root.rb_node; | |
1097 | ||
1098 | /* Adjust the search size for alignment overhead. */ | |
1099 | length = size + align - 1; | |
1100 | ||
1101 | while (node) { | |
1102 | va = rb_entry(node, struct vmap_area, rb_node); | |
1103 | ||
1104 | if (get_subtree_max_size(node->rb_left) >= length && | |
1105 | vstart < va->va_start) { | |
1106 | node = node->rb_left; | |
1107 | } else { | |
1108 | if (is_within_this_va(va, size, align, vstart)) | |
1109 | return va; | |
1110 | ||
1111 | /* | |
1112 | * Does not make sense to go deeper towards the right | |
1113 | * sub-tree if it does not have a free block that is | |
1114 | * equal or bigger to the requested search length. | |
1115 | */ | |
1116 | if (get_subtree_max_size(node->rb_right) >= length) { | |
1117 | node = node->rb_right; | |
1118 | continue; | |
1119 | } | |
1120 | ||
1121 | /* | |
3806b041 | 1122 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
1123 | * that will satisfy the search criteria. It can happen |
1124 | * only once due to "vstart" restriction. | |
1125 | */ | |
1126 | while ((node = rb_parent(node))) { | |
1127 | va = rb_entry(node, struct vmap_area, rb_node); | |
1128 | if (is_within_this_va(va, size, align, vstart)) | |
1129 | return va; | |
1130 | ||
1131 | if (get_subtree_max_size(node->rb_right) >= length && | |
1132 | vstart <= va->va_start) { | |
1133 | node = node->rb_right; | |
1134 | break; | |
1135 | } | |
1136 | } | |
1137 | } | |
1138 | } | |
1139 | ||
1140 | return NULL; | |
1141 | } | |
1142 | ||
a6cf4e0f URS |
1143 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1144 | #include <linux/random.h> | |
1145 | ||
1146 | static struct vmap_area * | |
1147 | find_vmap_lowest_linear_match(unsigned long size, | |
1148 | unsigned long align, unsigned long vstart) | |
1149 | { | |
1150 | struct vmap_area *va; | |
1151 | ||
1152 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
1153 | if (!is_within_this_va(va, size, align, vstart)) | |
1154 | continue; | |
1155 | ||
1156 | return va; | |
1157 | } | |
1158 | ||
1159 | return NULL; | |
1160 | } | |
1161 | ||
1162 | static void | |
1163 | find_vmap_lowest_match_check(unsigned long size) | |
1164 | { | |
1165 | struct vmap_area *va_1, *va_2; | |
1166 | unsigned long vstart; | |
1167 | unsigned int rnd; | |
1168 | ||
1169 | get_random_bytes(&rnd, sizeof(rnd)); | |
1170 | vstart = VMALLOC_START + rnd; | |
1171 | ||
1172 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
1173 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
1174 | ||
1175 | if (va_1 != va_2) | |
1176 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
1177 | va_1, va_2, vstart); | |
1178 | } | |
1179 | #endif | |
1180 | ||
68ad4a33 URS |
1181 | enum fit_type { |
1182 | NOTHING_FIT = 0, | |
1183 | FL_FIT_TYPE = 1, /* full fit */ | |
1184 | LE_FIT_TYPE = 2, /* left edge fit */ | |
1185 | RE_FIT_TYPE = 3, /* right edge fit */ | |
1186 | NE_FIT_TYPE = 4 /* no edge fit */ | |
1187 | }; | |
1188 | ||
1189 | static __always_inline enum fit_type | |
1190 | classify_va_fit_type(struct vmap_area *va, | |
1191 | unsigned long nva_start_addr, unsigned long size) | |
1192 | { | |
1193 | enum fit_type type; | |
1194 | ||
1195 | /* Check if it is within VA. */ | |
1196 | if (nva_start_addr < va->va_start || | |
1197 | nva_start_addr + size > va->va_end) | |
1198 | return NOTHING_FIT; | |
1199 | ||
1200 | /* Now classify. */ | |
1201 | if (va->va_start == nva_start_addr) { | |
1202 | if (va->va_end == nva_start_addr + size) | |
1203 | type = FL_FIT_TYPE; | |
1204 | else | |
1205 | type = LE_FIT_TYPE; | |
1206 | } else if (va->va_end == nva_start_addr + size) { | |
1207 | type = RE_FIT_TYPE; | |
1208 | } else { | |
1209 | type = NE_FIT_TYPE; | |
1210 | } | |
1211 | ||
1212 | return type; | |
1213 | } | |
1214 | ||
1215 | static __always_inline int | |
1216 | adjust_va_to_fit_type(struct vmap_area *va, | |
1217 | unsigned long nva_start_addr, unsigned long size, | |
1218 | enum fit_type type) | |
1219 | { | |
2c929233 | 1220 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
1221 | |
1222 | if (type == FL_FIT_TYPE) { | |
1223 | /* | |
1224 | * No need to split VA, it fully fits. | |
1225 | * | |
1226 | * | | | |
1227 | * V NVA V | |
1228 | * |---------------| | |
1229 | */ | |
1230 | unlink_va(va, &free_vmap_area_root); | |
1231 | kmem_cache_free(vmap_area_cachep, va); | |
1232 | } else if (type == LE_FIT_TYPE) { | |
1233 | /* | |
1234 | * Split left edge of fit VA. | |
1235 | * | |
1236 | * | | | |
1237 | * V NVA V R | |
1238 | * |-------|-------| | |
1239 | */ | |
1240 | va->va_start += size; | |
1241 | } else if (type == RE_FIT_TYPE) { | |
1242 | /* | |
1243 | * Split right edge of fit VA. | |
1244 | * | |
1245 | * | | | |
1246 | * L V NVA V | |
1247 | * |-------|-------| | |
1248 | */ | |
1249 | va->va_end = nva_start_addr; | |
1250 | } else if (type == NE_FIT_TYPE) { | |
1251 | /* | |
1252 | * Split no edge of fit VA. | |
1253 | * | |
1254 | * | | | |
1255 | * L V NVA V R | |
1256 | * |---|-------|---| | |
1257 | */ | |
82dd23e8 URS |
1258 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
1259 | if (unlikely(!lva)) { | |
1260 | /* | |
1261 | * For percpu allocator we do not do any pre-allocation | |
1262 | * and leave it as it is. The reason is it most likely | |
1263 | * never ends up with NE_FIT_TYPE splitting. In case of | |
1264 | * percpu allocations offsets and sizes are aligned to | |
1265 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
1266 | * are its main fitting cases. | |
1267 | * | |
1268 | * There are a few exceptions though, as an example it is | |
1269 | * a first allocation (early boot up) when we have "one" | |
1270 | * big free space that has to be split. | |
060650a2 URS |
1271 | * |
1272 | * Also we can hit this path in case of regular "vmap" | |
1273 | * allocations, if "this" current CPU was not preloaded. | |
1274 | * See the comment in alloc_vmap_area() why. If so, then | |
1275 | * GFP_NOWAIT is used instead to get an extra object for | |
1276 | * split purpose. That is rare and most time does not | |
1277 | * occur. | |
1278 | * | |
1279 | * What happens if an allocation gets failed. Basically, | |
1280 | * an "overflow" path is triggered to purge lazily freed | |
1281 | * areas to free some memory, then, the "retry" path is | |
1282 | * triggered to repeat one more time. See more details | |
1283 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
1284 | */ |
1285 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
1286 | if (!lva) | |
1287 | return -1; | |
1288 | } | |
68ad4a33 URS |
1289 | |
1290 | /* | |
1291 | * Build the remainder. | |
1292 | */ | |
1293 | lva->va_start = va->va_start; | |
1294 | lva->va_end = nva_start_addr; | |
1295 | ||
1296 | /* | |
1297 | * Shrink this VA to remaining size. | |
1298 | */ | |
1299 | va->va_start = nva_start_addr + size; | |
1300 | } else { | |
1301 | return -1; | |
1302 | } | |
1303 | ||
1304 | if (type != FL_FIT_TYPE) { | |
1305 | augment_tree_propagate_from(va); | |
1306 | ||
2c929233 | 1307 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
1308 | insert_vmap_area_augment(lva, &va->rb_node, |
1309 | &free_vmap_area_root, &free_vmap_area_list); | |
1310 | } | |
1311 | ||
1312 | return 0; | |
1313 | } | |
1314 | ||
1315 | /* | |
1316 | * Returns a start address of the newly allocated area, if success. | |
1317 | * Otherwise a vend is returned that indicates failure. | |
1318 | */ | |
1319 | static __always_inline unsigned long | |
1320 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1321 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1322 | { |
1323 | unsigned long nva_start_addr; | |
1324 | struct vmap_area *va; | |
1325 | enum fit_type type; | |
1326 | int ret; | |
1327 | ||
1328 | va = find_vmap_lowest_match(size, align, vstart); | |
1329 | if (unlikely(!va)) | |
1330 | return vend; | |
1331 | ||
1332 | if (va->va_start > vstart) | |
1333 | nva_start_addr = ALIGN(va->va_start, align); | |
1334 | else | |
1335 | nva_start_addr = ALIGN(vstart, align); | |
1336 | ||
1337 | /* Check the "vend" restriction. */ | |
1338 | if (nva_start_addr + size > vend) | |
1339 | return vend; | |
1340 | ||
1341 | /* Classify what we have found. */ | |
1342 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1343 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1344 | return vend; | |
1345 | ||
1346 | /* Update the free vmap_area. */ | |
1347 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1348 | if (ret) | |
1349 | return vend; | |
1350 | ||
a6cf4e0f URS |
1351 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1352 | find_vmap_lowest_match_check(size); | |
1353 | #endif | |
1354 | ||
68ad4a33 URS |
1355 | return nva_start_addr; |
1356 | } | |
4da56b99 | 1357 | |
d98c9e83 AR |
1358 | /* |
1359 | * Free a region of KVA allocated by alloc_vmap_area | |
1360 | */ | |
1361 | static void free_vmap_area(struct vmap_area *va) | |
1362 | { | |
1363 | /* | |
1364 | * Remove from the busy tree/list. | |
1365 | */ | |
1366 | spin_lock(&vmap_area_lock); | |
1367 | unlink_va(va, &vmap_area_root); | |
1368 | spin_unlock(&vmap_area_lock); | |
1369 | ||
1370 | /* | |
1371 | * Insert/Merge it back to the free tree/list. | |
1372 | */ | |
1373 | spin_lock(&free_vmap_area_lock); | |
96e2db45 | 1374 | merge_or_add_vmap_area_augment(va, &free_vmap_area_root, &free_vmap_area_list); |
d98c9e83 AR |
1375 | spin_unlock(&free_vmap_area_lock); |
1376 | } | |
1377 | ||
db64fe02 NP |
1378 | /* |
1379 | * Allocate a region of KVA of the specified size and alignment, within the | |
1380 | * vstart and vend. | |
1381 | */ | |
1382 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1383 | unsigned long align, | |
1384 | unsigned long vstart, unsigned long vend, | |
1385 | int node, gfp_t gfp_mask) | |
1386 | { | |
82dd23e8 | 1387 | struct vmap_area *va, *pva; |
1da177e4 | 1388 | unsigned long addr; |
db64fe02 | 1389 | int purged = 0; |
d98c9e83 | 1390 | int ret; |
db64fe02 | 1391 | |
7766970c | 1392 | BUG_ON(!size); |
891c49ab | 1393 | BUG_ON(offset_in_page(size)); |
89699605 | 1394 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1395 | |
68ad4a33 URS |
1396 | if (unlikely(!vmap_initialized)) |
1397 | return ERR_PTR(-EBUSY); | |
1398 | ||
5803ed29 | 1399 | might_sleep(); |
f07116d7 | 1400 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1401 | |
f07116d7 | 1402 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1403 | if (unlikely(!va)) |
1404 | return ERR_PTR(-ENOMEM); | |
1405 | ||
7f88f88f CM |
1406 | /* |
1407 | * Only scan the relevant parts containing pointers to other objects | |
1408 | * to avoid false negatives. | |
1409 | */ | |
f07116d7 | 1410 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1411 | |
db64fe02 | 1412 | retry: |
82dd23e8 | 1413 | /* |
81f1ba58 URS |
1414 | * Preload this CPU with one extra vmap_area object. It is used |
1415 | * when fit type of free area is NE_FIT_TYPE. Please note, it | |
1416 | * does not guarantee that an allocation occurs on a CPU that | |
1417 | * is preloaded, instead we minimize the case when it is not. | |
1418 | * It can happen because of cpu migration, because there is a | |
1419 | * race until the below spinlock is taken. | |
82dd23e8 URS |
1420 | * |
1421 | * The preload is done in non-atomic context, thus it allows us | |
1422 | * to use more permissive allocation masks to be more stable under | |
81f1ba58 URS |
1423 | * low memory condition and high memory pressure. In rare case, |
1424 | * if not preloaded, GFP_NOWAIT is used. | |
82dd23e8 | 1425 | * |
81f1ba58 | 1426 | * Set "pva" to NULL here, because of "retry" path. |
82dd23e8 | 1427 | */ |
81f1ba58 | 1428 | pva = NULL; |
82dd23e8 | 1429 | |
81f1ba58 URS |
1430 | if (!this_cpu_read(ne_fit_preload_node)) |
1431 | /* | |
1432 | * Even if it fails we do not really care about that. | |
1433 | * Just proceed as it is. If needed "overflow" path | |
1434 | * will refill the cache we allocate from. | |
1435 | */ | |
f07116d7 | 1436 | pva = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
82dd23e8 | 1437 | |
e36176be | 1438 | spin_lock(&free_vmap_area_lock); |
81f1ba58 URS |
1439 | |
1440 | if (pva && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) | |
1441 | kmem_cache_free(vmap_area_cachep, pva); | |
89699605 | 1442 | |
afd07389 | 1443 | /* |
68ad4a33 URS |
1444 | * If an allocation fails, the "vend" address is |
1445 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1446 | */ |
cacca6ba | 1447 | addr = __alloc_vmap_area(size, align, vstart, vend); |
e36176be URS |
1448 | spin_unlock(&free_vmap_area_lock); |
1449 | ||
68ad4a33 | 1450 | if (unlikely(addr == vend)) |
89699605 | 1451 | goto overflow; |
db64fe02 NP |
1452 | |
1453 | va->va_start = addr; | |
1454 | va->va_end = addr + size; | |
688fcbfc | 1455 | va->vm = NULL; |
68ad4a33 | 1456 | |
d98c9e83 | 1457 | |
e36176be URS |
1458 | spin_lock(&vmap_area_lock); |
1459 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
db64fe02 NP |
1460 | spin_unlock(&vmap_area_lock); |
1461 | ||
61e16557 | 1462 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1463 | BUG_ON(va->va_start < vstart); |
1464 | BUG_ON(va->va_end > vend); | |
1465 | ||
d98c9e83 AR |
1466 | ret = kasan_populate_vmalloc(addr, size); |
1467 | if (ret) { | |
1468 | free_vmap_area(va); | |
1469 | return ERR_PTR(ret); | |
1470 | } | |
1471 | ||
db64fe02 | 1472 | return va; |
89699605 NP |
1473 | |
1474 | overflow: | |
89699605 NP |
1475 | if (!purged) { |
1476 | purge_vmap_area_lazy(); | |
1477 | purged = 1; | |
1478 | goto retry; | |
1479 | } | |
4da56b99 CW |
1480 | |
1481 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1482 | unsigned long freed = 0; | |
1483 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1484 | if (freed > 0) { | |
1485 | purged = 0; | |
1486 | goto retry; | |
1487 | } | |
1488 | } | |
1489 | ||
03497d76 | 1490 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1491 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1492 | size); | |
68ad4a33 URS |
1493 | |
1494 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1495 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1496 | } |
1497 | ||
4da56b99 CW |
1498 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1499 | { | |
1500 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1501 | } | |
1502 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1503 | ||
1504 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1505 | { | |
1506 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1507 | } | |
1508 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1509 | ||
db64fe02 NP |
1510 | /* |
1511 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1512 | * before attempting to purge with a TLB flush. | |
1513 | * | |
1514 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1515 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1516 | * global TLB flushes that must be performed. It would seem natural to scale | |
1517 | * this number up linearly with the number of CPUs (because vmapping activity | |
1518 | * could also scale linearly with the number of CPUs), however it is likely | |
1519 | * that in practice, workloads might be constrained in other ways that mean | |
1520 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1521 | * conservative and not introduce a big latency on huge systems, so go with | |
1522 | * a less aggressive log scale. It will still be an improvement over the old | |
1523 | * code, and it will be simple to change the scale factor if we find that it | |
1524 | * becomes a problem on bigger systems. | |
1525 | */ | |
1526 | static unsigned long lazy_max_pages(void) | |
1527 | { | |
1528 | unsigned int log; | |
1529 | ||
1530 | log = fls(num_online_cpus()); | |
1531 | ||
1532 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1533 | } | |
1534 | ||
4d36e6f8 | 1535 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1536 | |
0574ecd1 CH |
1537 | /* |
1538 | * Serialize vmap purging. There is no actual criticial section protected | |
1539 | * by this look, but we want to avoid concurrent calls for performance | |
1540 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1541 | */ | |
f9e09977 | 1542 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1543 | |
02b709df NP |
1544 | /* for per-CPU blocks */ |
1545 | static void purge_fragmented_blocks_allcpus(void); | |
1546 | ||
3ee48b6a CW |
1547 | /* |
1548 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1549 | * immediately freed. | |
1550 | */ | |
1551 | void set_iounmap_nonlazy(void) | |
1552 | { | |
4d36e6f8 | 1553 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1554 | } |
1555 | ||
db64fe02 NP |
1556 | /* |
1557 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1558 | */ |
0574ecd1 | 1559 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1560 | { |
4d36e6f8 | 1561 | unsigned long resched_threshold; |
96e2db45 URS |
1562 | struct list_head local_pure_list; |
1563 | struct vmap_area *va, *n_va; | |
db64fe02 | 1564 | |
0574ecd1 | 1565 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1566 | |
96e2db45 URS |
1567 | spin_lock(&purge_vmap_area_lock); |
1568 | purge_vmap_area_root = RB_ROOT; | |
1569 | list_replace_init(&purge_vmap_area_list, &local_pure_list); | |
1570 | spin_unlock(&purge_vmap_area_lock); | |
1571 | ||
1572 | if (unlikely(list_empty(&local_pure_list))) | |
68571be9 URS |
1573 | return false; |
1574 | ||
96e2db45 URS |
1575 | start = min(start, |
1576 | list_first_entry(&local_pure_list, | |
1577 | struct vmap_area, list)->va_start); | |
1578 | ||
1579 | end = max(end, | |
1580 | list_last_entry(&local_pure_list, | |
1581 | struct vmap_area, list)->va_end); | |
db64fe02 | 1582 | |
0574ecd1 | 1583 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1584 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1585 | |
e36176be | 1586 | spin_lock(&free_vmap_area_lock); |
96e2db45 | 1587 | list_for_each_entry_safe(va, n_va, &local_pure_list, list) { |
4d36e6f8 | 1588 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
3c5c3cfb DA |
1589 | unsigned long orig_start = va->va_start; |
1590 | unsigned long orig_end = va->va_end; | |
763b218d | 1591 | |
dd3b8353 URS |
1592 | /* |
1593 | * Finally insert or merge lazily-freed area. It is | |
1594 | * detached and there is no need to "unlink" it from | |
1595 | * anything. | |
1596 | */ | |
96e2db45 URS |
1597 | va = merge_or_add_vmap_area_augment(va, &free_vmap_area_root, |
1598 | &free_vmap_area_list); | |
3c5c3cfb | 1599 | |
9c801f61 URS |
1600 | if (!va) |
1601 | continue; | |
1602 | ||
3c5c3cfb DA |
1603 | if (is_vmalloc_or_module_addr((void *)orig_start)) |
1604 | kasan_release_vmalloc(orig_start, orig_end, | |
1605 | va->va_start, va->va_end); | |
dd3b8353 | 1606 | |
4d36e6f8 | 1607 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1608 | |
4d36e6f8 | 1609 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
e36176be | 1610 | cond_resched_lock(&free_vmap_area_lock); |
763b218d | 1611 | } |
e36176be | 1612 | spin_unlock(&free_vmap_area_lock); |
0574ecd1 | 1613 | return true; |
db64fe02 NP |
1614 | } |
1615 | ||
496850e5 NP |
1616 | /* |
1617 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1618 | * is already purging. | |
1619 | */ | |
1620 | static void try_purge_vmap_area_lazy(void) | |
1621 | { | |
f9e09977 | 1622 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1623 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1624 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1625 | } |
496850e5 NP |
1626 | } |
1627 | ||
db64fe02 NP |
1628 | /* |
1629 | * Kick off a purge of the outstanding lazy areas. | |
1630 | */ | |
1631 | static void purge_vmap_area_lazy(void) | |
1632 | { | |
f9e09977 | 1633 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1634 | purge_fragmented_blocks_allcpus(); |
1635 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1636 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1637 | } |
1638 | ||
1639 | /* | |
64141da5 JF |
1640 | * Free a vmap area, caller ensuring that the area has been unmapped |
1641 | * and flush_cache_vunmap had been called for the correct range | |
1642 | * previously. | |
db64fe02 | 1643 | */ |
64141da5 | 1644 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1645 | { |
4d36e6f8 | 1646 | unsigned long nr_lazy; |
80c4bd7a | 1647 | |
dd3b8353 URS |
1648 | spin_lock(&vmap_area_lock); |
1649 | unlink_va(va, &vmap_area_root); | |
1650 | spin_unlock(&vmap_area_lock); | |
1651 | ||
4d36e6f8 URS |
1652 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1653 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a | 1654 | |
96e2db45 URS |
1655 | /* |
1656 | * Merge or place it to the purge tree/list. | |
1657 | */ | |
1658 | spin_lock(&purge_vmap_area_lock); | |
1659 | merge_or_add_vmap_area(va, | |
1660 | &purge_vmap_area_root, &purge_vmap_area_list); | |
1661 | spin_unlock(&purge_vmap_area_lock); | |
80c4bd7a | 1662 | |
96e2db45 | 1663 | /* After this point, we may free va at any time */ |
80c4bd7a | 1664 | if (unlikely(nr_lazy > lazy_max_pages())) |
496850e5 | 1665 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1666 | } |
1667 | ||
b29acbdc NP |
1668 | /* |
1669 | * Free and unmap a vmap area | |
1670 | */ | |
1671 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1672 | { | |
1673 | flush_cache_vunmap(va->va_start, va->va_end); | |
855e57a1 | 1674 | unmap_kernel_range_noflush(va->va_start, va->va_end - va->va_start); |
8e57f8ac | 1675 | if (debug_pagealloc_enabled_static()) |
82a2e924 CP |
1676 | flush_tlb_kernel_range(va->va_start, va->va_end); |
1677 | ||
c8eef01e | 1678 | free_vmap_area_noflush(va); |
b29acbdc NP |
1679 | } |
1680 | ||
db64fe02 NP |
1681 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1682 | { | |
1683 | struct vmap_area *va; | |
1684 | ||
1685 | spin_lock(&vmap_area_lock); | |
1686 | va = __find_vmap_area(addr); | |
1687 | spin_unlock(&vmap_area_lock); | |
1688 | ||
1689 | return va; | |
1690 | } | |
1691 | ||
db64fe02 NP |
1692 | /*** Per cpu kva allocator ***/ |
1693 | ||
1694 | /* | |
1695 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1696 | * room for at least 16 percpu vmap blocks per CPU. | |
1697 | */ | |
1698 | /* | |
1699 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1700 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1701 | * instead (we just need a rough idea) | |
1702 | */ | |
1703 | #if BITS_PER_LONG == 32 | |
1704 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1705 | #else | |
1706 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1707 | #endif | |
1708 | ||
1709 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1710 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1711 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1712 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1713 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1714 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1715 | #define VMAP_BBMAP_BITS \ |
1716 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1717 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1718 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1719 | |
1720 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1721 | ||
1722 | struct vmap_block_queue { | |
1723 | spinlock_t lock; | |
1724 | struct list_head free; | |
db64fe02 NP |
1725 | }; |
1726 | ||
1727 | struct vmap_block { | |
1728 | spinlock_t lock; | |
1729 | struct vmap_area *va; | |
db64fe02 | 1730 | unsigned long free, dirty; |
7d61bfe8 | 1731 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1732 | struct list_head free_list; |
1733 | struct rcu_head rcu_head; | |
02b709df | 1734 | struct list_head purge; |
db64fe02 NP |
1735 | }; |
1736 | ||
1737 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1738 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1739 | ||
1740 | /* | |
0f14599c | 1741 | * XArray of vmap blocks, indexed by address, to quickly find a vmap block |
db64fe02 NP |
1742 | * in the free path. Could get rid of this if we change the API to return a |
1743 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1744 | */ | |
0f14599c | 1745 | static DEFINE_XARRAY(vmap_blocks); |
db64fe02 NP |
1746 | |
1747 | /* | |
1748 | * We should probably have a fallback mechanism to allocate virtual memory | |
1749 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1750 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1751 | * big problem. | |
1752 | */ | |
1753 | ||
1754 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1755 | { | |
1756 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1757 | addr /= VMAP_BLOCK_SIZE; | |
1758 | return addr; | |
1759 | } | |
1760 | ||
cf725ce2 RP |
1761 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1762 | { | |
1763 | unsigned long addr; | |
1764 | ||
1765 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1766 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1767 | return (void *)addr; | |
1768 | } | |
1769 | ||
1770 | /** | |
1771 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1772 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1773 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1774 | * @gfp_mask: flags for the page level allocator | |
1775 | * | |
a862f68a | 1776 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1777 | */ |
1778 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1779 | { |
1780 | struct vmap_block_queue *vbq; | |
1781 | struct vmap_block *vb; | |
1782 | struct vmap_area *va; | |
1783 | unsigned long vb_idx; | |
1784 | int node, err; | |
cf725ce2 | 1785 | void *vaddr; |
db64fe02 NP |
1786 | |
1787 | node = numa_node_id(); | |
1788 | ||
1789 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1790 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1791 | if (unlikely(!vb)) | |
1792 | return ERR_PTR(-ENOMEM); | |
1793 | ||
1794 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1795 | VMALLOC_START, VMALLOC_END, | |
1796 | node, gfp_mask); | |
ddf9c6d4 | 1797 | if (IS_ERR(va)) { |
db64fe02 | 1798 | kfree(vb); |
e7d86340 | 1799 | return ERR_CAST(va); |
db64fe02 NP |
1800 | } |
1801 | ||
cf725ce2 | 1802 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1803 | spin_lock_init(&vb->lock); |
1804 | vb->va = va; | |
cf725ce2 RP |
1805 | /* At least something should be left free */ |
1806 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1807 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1808 | vb->dirty = 0; |
7d61bfe8 RP |
1809 | vb->dirty_min = VMAP_BBMAP_BITS; |
1810 | vb->dirty_max = 0; | |
db64fe02 | 1811 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1812 | |
1813 | vb_idx = addr_to_vb_idx(va->va_start); | |
0f14599c MWO |
1814 | err = xa_insert(&vmap_blocks, vb_idx, vb, gfp_mask); |
1815 | if (err) { | |
1816 | kfree(vb); | |
1817 | free_vmap_area(va); | |
1818 | return ERR_PTR(err); | |
1819 | } | |
db64fe02 NP |
1820 | |
1821 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1822 | spin_lock(&vbq->lock); |
68ac546f | 1823 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1824 | spin_unlock(&vbq->lock); |
3f04ba85 | 1825 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1826 | |
cf725ce2 | 1827 | return vaddr; |
db64fe02 NP |
1828 | } |
1829 | ||
db64fe02 NP |
1830 | static void free_vmap_block(struct vmap_block *vb) |
1831 | { | |
1832 | struct vmap_block *tmp; | |
db64fe02 | 1833 | |
0f14599c | 1834 | tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start)); |
db64fe02 NP |
1835 | BUG_ON(tmp != vb); |
1836 | ||
64141da5 | 1837 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1838 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1839 | } |
1840 | ||
02b709df NP |
1841 | static void purge_fragmented_blocks(int cpu) |
1842 | { | |
1843 | LIST_HEAD(purge); | |
1844 | struct vmap_block *vb; | |
1845 | struct vmap_block *n_vb; | |
1846 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1847 | ||
1848 | rcu_read_lock(); | |
1849 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1850 | ||
1851 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1852 | continue; | |
1853 | ||
1854 | spin_lock(&vb->lock); | |
1855 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1856 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1857 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1858 | vb->dirty_min = 0; |
1859 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1860 | spin_lock(&vbq->lock); |
1861 | list_del_rcu(&vb->free_list); | |
1862 | spin_unlock(&vbq->lock); | |
1863 | spin_unlock(&vb->lock); | |
1864 | list_add_tail(&vb->purge, &purge); | |
1865 | } else | |
1866 | spin_unlock(&vb->lock); | |
1867 | } | |
1868 | rcu_read_unlock(); | |
1869 | ||
1870 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1871 | list_del(&vb->purge); | |
1872 | free_vmap_block(vb); | |
1873 | } | |
1874 | } | |
1875 | ||
02b709df NP |
1876 | static void purge_fragmented_blocks_allcpus(void) |
1877 | { | |
1878 | int cpu; | |
1879 | ||
1880 | for_each_possible_cpu(cpu) | |
1881 | purge_fragmented_blocks(cpu); | |
1882 | } | |
1883 | ||
db64fe02 NP |
1884 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1885 | { | |
1886 | struct vmap_block_queue *vbq; | |
1887 | struct vmap_block *vb; | |
cf725ce2 | 1888 | void *vaddr = NULL; |
db64fe02 NP |
1889 | unsigned int order; |
1890 | ||
891c49ab | 1891 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1892 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1893 | if (WARN_ON(size == 0)) { |
1894 | /* | |
1895 | * Allocating 0 bytes isn't what caller wants since | |
1896 | * get_order(0) returns funny result. Just warn and terminate | |
1897 | * early. | |
1898 | */ | |
1899 | return NULL; | |
1900 | } | |
db64fe02 NP |
1901 | order = get_order(size); |
1902 | ||
db64fe02 NP |
1903 | rcu_read_lock(); |
1904 | vbq = &get_cpu_var(vmap_block_queue); | |
1905 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1906 | unsigned long pages_off; |
db64fe02 NP |
1907 | |
1908 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1909 | if (vb->free < (1UL << order)) { |
1910 | spin_unlock(&vb->lock); | |
1911 | continue; | |
1912 | } | |
02b709df | 1913 | |
cf725ce2 RP |
1914 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1915 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1916 | vb->free -= 1UL << order; |
1917 | if (vb->free == 0) { | |
1918 | spin_lock(&vbq->lock); | |
1919 | list_del_rcu(&vb->free_list); | |
1920 | spin_unlock(&vbq->lock); | |
1921 | } | |
cf725ce2 | 1922 | |
02b709df NP |
1923 | spin_unlock(&vb->lock); |
1924 | break; | |
db64fe02 | 1925 | } |
02b709df | 1926 | |
3f04ba85 | 1927 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1928 | rcu_read_unlock(); |
1929 | ||
cf725ce2 RP |
1930 | /* Allocate new block if nothing was found */ |
1931 | if (!vaddr) | |
1932 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1933 | |
cf725ce2 | 1934 | return vaddr; |
db64fe02 NP |
1935 | } |
1936 | ||
78a0e8c4 | 1937 | static void vb_free(unsigned long addr, unsigned long size) |
db64fe02 NP |
1938 | { |
1939 | unsigned long offset; | |
db64fe02 NP |
1940 | unsigned int order; |
1941 | struct vmap_block *vb; | |
1942 | ||
891c49ab | 1943 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1944 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc | 1945 | |
78a0e8c4 | 1946 | flush_cache_vunmap(addr, addr + size); |
b29acbdc | 1947 | |
db64fe02 | 1948 | order = get_order(size); |
78a0e8c4 | 1949 | offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT; |
0f14599c | 1950 | vb = xa_load(&vmap_blocks, addr_to_vb_idx(addr)); |
db64fe02 | 1951 | |
b521c43f | 1952 | unmap_kernel_range_noflush(addr, size); |
64141da5 | 1953 | |
8e57f8ac | 1954 | if (debug_pagealloc_enabled_static()) |
78a0e8c4 | 1955 | flush_tlb_kernel_range(addr, addr + size); |
82a2e924 | 1956 | |
db64fe02 | 1957 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1958 | |
1959 | /* Expand dirty range */ | |
1960 | vb->dirty_min = min(vb->dirty_min, offset); | |
1961 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1962 | |
db64fe02 NP |
1963 | vb->dirty += 1UL << order; |
1964 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1965 | BUG_ON(vb->free); |
db64fe02 NP |
1966 | spin_unlock(&vb->lock); |
1967 | free_vmap_block(vb); | |
1968 | } else | |
1969 | spin_unlock(&vb->lock); | |
1970 | } | |
1971 | ||
868b104d | 1972 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1973 | { |
db64fe02 | 1974 | int cpu; |
db64fe02 | 1975 | |
9b463334 JF |
1976 | if (unlikely(!vmap_initialized)) |
1977 | return; | |
1978 | ||
5803ed29 CH |
1979 | might_sleep(); |
1980 | ||
db64fe02 NP |
1981 | for_each_possible_cpu(cpu) { |
1982 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1983 | struct vmap_block *vb; | |
1984 | ||
1985 | rcu_read_lock(); | |
1986 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1987 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1988 | if (vb->dirty) { |
1989 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1990 | unsigned long s, e; |
b136be5e | 1991 | |
7d61bfe8 RP |
1992 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1993 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1994 | |
7d61bfe8 RP |
1995 | start = min(s, start); |
1996 | end = max(e, end); | |
db64fe02 | 1997 | |
7d61bfe8 | 1998 | flush = 1; |
db64fe02 NP |
1999 | } |
2000 | spin_unlock(&vb->lock); | |
2001 | } | |
2002 | rcu_read_unlock(); | |
2003 | } | |
2004 | ||
f9e09977 | 2005 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
2006 | purge_fragmented_blocks_allcpus(); |
2007 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
2008 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 2009 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 2010 | } |
868b104d RE |
2011 | |
2012 | /** | |
2013 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
2014 | * | |
2015 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
2016 | * to amortize TLB flushing overheads. What this means is that any page you | |
2017 | * have now, may, in a former life, have been mapped into kernel virtual | |
2018 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
2019 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
2020 | * | |
2021 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
2022 | * be sure that none of the pages we have control over will have any aliases | |
2023 | * from the vmap layer. | |
2024 | */ | |
2025 | void vm_unmap_aliases(void) | |
2026 | { | |
2027 | unsigned long start = ULONG_MAX, end = 0; | |
2028 | int flush = 0; | |
2029 | ||
2030 | _vm_unmap_aliases(start, end, flush); | |
2031 | } | |
db64fe02 NP |
2032 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
2033 | ||
2034 | /** | |
2035 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
2036 | * @mem: the pointer returned by vm_map_ram | |
2037 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
2038 | */ | |
2039 | void vm_unmap_ram(const void *mem, unsigned int count) | |
2040 | { | |
65ee03c4 | 2041 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 2042 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 2043 | struct vmap_area *va; |
db64fe02 | 2044 | |
5803ed29 | 2045 | might_sleep(); |
db64fe02 NP |
2046 | BUG_ON(!addr); |
2047 | BUG_ON(addr < VMALLOC_START); | |
2048 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 2049 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 2050 | |
d98c9e83 AR |
2051 | kasan_poison_vmalloc(mem, size); |
2052 | ||
9c3acf60 | 2053 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 2054 | debug_check_no_locks_freed(mem, size); |
78a0e8c4 | 2055 | vb_free(addr, size); |
9c3acf60 CH |
2056 | return; |
2057 | } | |
2058 | ||
2059 | va = find_vmap_area(addr); | |
2060 | BUG_ON(!va); | |
05e3ff95 CP |
2061 | debug_check_no_locks_freed((void *)va->va_start, |
2062 | (va->va_end - va->va_start)); | |
9c3acf60 | 2063 | free_unmap_vmap_area(va); |
db64fe02 NP |
2064 | } |
2065 | EXPORT_SYMBOL(vm_unmap_ram); | |
2066 | ||
2067 | /** | |
2068 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
2069 | * @pages: an array of pointers to the pages to be mapped | |
2070 | * @count: number of pages | |
2071 | * @node: prefer to allocate data structures on this node | |
e99c97ad | 2072 | * |
36437638 GK |
2073 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
2074 | * faster than vmap so it's good. But if you mix long-life and short-life | |
2075 | * objects with vm_map_ram(), it could consume lots of address space through | |
2076 | * fragmentation (especially on a 32bit machine). You could see failures in | |
2077 | * the end. Please use this function for short-lived objects. | |
2078 | * | |
e99c97ad | 2079 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 | 2080 | */ |
d4efd79a | 2081 | void *vm_map_ram(struct page **pages, unsigned int count, int node) |
db64fe02 | 2082 | { |
65ee03c4 | 2083 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
2084 | unsigned long addr; |
2085 | void *mem; | |
2086 | ||
2087 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
2088 | mem = vb_alloc(size, GFP_KERNEL); | |
2089 | if (IS_ERR(mem)) | |
2090 | return NULL; | |
2091 | addr = (unsigned long)mem; | |
2092 | } else { | |
2093 | struct vmap_area *va; | |
2094 | va = alloc_vmap_area(size, PAGE_SIZE, | |
2095 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
2096 | if (IS_ERR(va)) | |
2097 | return NULL; | |
2098 | ||
2099 | addr = va->va_start; | |
2100 | mem = (void *)addr; | |
2101 | } | |
d98c9e83 AR |
2102 | |
2103 | kasan_unpoison_vmalloc(mem, size); | |
2104 | ||
d4efd79a | 2105 | if (map_kernel_range(addr, size, PAGE_KERNEL, pages) < 0) { |
db64fe02 NP |
2106 | vm_unmap_ram(mem, count); |
2107 | return NULL; | |
2108 | } | |
2109 | return mem; | |
2110 | } | |
2111 | EXPORT_SYMBOL(vm_map_ram); | |
2112 | ||
4341fa45 | 2113 | static struct vm_struct *vmlist __initdata; |
92eac168 | 2114 | |
be9b7335 NP |
2115 | /** |
2116 | * vm_area_add_early - add vmap area early during boot | |
2117 | * @vm: vm_struct to add | |
2118 | * | |
2119 | * This function is used to add fixed kernel vm area to vmlist before | |
2120 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
2121 | * should contain proper values and the other fields should be zero. | |
2122 | * | |
2123 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2124 | */ | |
2125 | void __init vm_area_add_early(struct vm_struct *vm) | |
2126 | { | |
2127 | struct vm_struct *tmp, **p; | |
2128 | ||
2129 | BUG_ON(vmap_initialized); | |
2130 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
2131 | if (tmp->addr >= vm->addr) { | |
2132 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
2133 | break; | |
2134 | } else | |
2135 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
2136 | } | |
2137 | vm->next = *p; | |
2138 | *p = vm; | |
2139 | } | |
2140 | ||
f0aa6617 TH |
2141 | /** |
2142 | * vm_area_register_early - register vmap area early during boot | |
2143 | * @vm: vm_struct to register | |
c0c0a293 | 2144 | * @align: requested alignment |
f0aa6617 TH |
2145 | * |
2146 | * This function is used to register kernel vm area before | |
2147 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
2148 | * proper values on entry and other fields should be zero. On return, | |
2149 | * vm->addr contains the allocated address. | |
2150 | * | |
2151 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2152 | */ | |
c0c0a293 | 2153 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
2154 | { |
2155 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
2156 | unsigned long addr; |
2157 | ||
2158 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
2159 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 2160 | |
c0c0a293 | 2161 | vm->addr = (void *)addr; |
f0aa6617 | 2162 | |
be9b7335 | 2163 | vm_area_add_early(vm); |
f0aa6617 TH |
2164 | } |
2165 | ||
68ad4a33 URS |
2166 | static void vmap_init_free_space(void) |
2167 | { | |
2168 | unsigned long vmap_start = 1; | |
2169 | const unsigned long vmap_end = ULONG_MAX; | |
2170 | struct vmap_area *busy, *free; | |
2171 | ||
2172 | /* | |
2173 | * B F B B B F | |
2174 | * -|-----|.....|-----|-----|-----|.....|- | |
2175 | * | The KVA space | | |
2176 | * |<--------------------------------->| | |
2177 | */ | |
2178 | list_for_each_entry(busy, &vmap_area_list, list) { | |
2179 | if (busy->va_start - vmap_start > 0) { | |
2180 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2181 | if (!WARN_ON_ONCE(!free)) { | |
2182 | free->va_start = vmap_start; | |
2183 | free->va_end = busy->va_start; | |
2184 | ||
2185 | insert_vmap_area_augment(free, NULL, | |
2186 | &free_vmap_area_root, | |
2187 | &free_vmap_area_list); | |
2188 | } | |
2189 | } | |
2190 | ||
2191 | vmap_start = busy->va_end; | |
2192 | } | |
2193 | ||
2194 | if (vmap_end - vmap_start > 0) { | |
2195 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2196 | if (!WARN_ON_ONCE(!free)) { | |
2197 | free->va_start = vmap_start; | |
2198 | free->va_end = vmap_end; | |
2199 | ||
2200 | insert_vmap_area_augment(free, NULL, | |
2201 | &free_vmap_area_root, | |
2202 | &free_vmap_area_list); | |
2203 | } | |
2204 | } | |
2205 | } | |
2206 | ||
db64fe02 NP |
2207 | void __init vmalloc_init(void) |
2208 | { | |
822c18f2 IK |
2209 | struct vmap_area *va; |
2210 | struct vm_struct *tmp; | |
db64fe02 NP |
2211 | int i; |
2212 | ||
68ad4a33 URS |
2213 | /* |
2214 | * Create the cache for vmap_area objects. | |
2215 | */ | |
2216 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
2217 | ||
db64fe02 NP |
2218 | for_each_possible_cpu(i) { |
2219 | struct vmap_block_queue *vbq; | |
32fcfd40 | 2220 | struct vfree_deferred *p; |
db64fe02 NP |
2221 | |
2222 | vbq = &per_cpu(vmap_block_queue, i); | |
2223 | spin_lock_init(&vbq->lock); | |
2224 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
2225 | p = &per_cpu(vfree_deferred, i); |
2226 | init_llist_head(&p->list); | |
2227 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 2228 | } |
9b463334 | 2229 | |
822c18f2 IK |
2230 | /* Import existing vmlist entries. */ |
2231 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
2232 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
2233 | if (WARN_ON_ONCE(!va)) | |
2234 | continue; | |
2235 | ||
822c18f2 IK |
2236 | va->va_start = (unsigned long)tmp->addr; |
2237 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 2238 | va->vm = tmp; |
68ad4a33 | 2239 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 2240 | } |
ca23e405 | 2241 | |
68ad4a33 URS |
2242 | /* |
2243 | * Now we can initialize a free vmap space. | |
2244 | */ | |
2245 | vmap_init_free_space(); | |
9b463334 | 2246 | vmap_initialized = true; |
db64fe02 NP |
2247 | } |
2248 | ||
8fc48985 TH |
2249 | /** |
2250 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
2251 | * @addr: start of the VM area to unmap | |
2252 | * @size: size of the VM area to unmap | |
2253 | * | |
2254 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
2255 | * the unmapping and tlb after. | |
2256 | */ | |
db64fe02 NP |
2257 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
2258 | { | |
2259 | unsigned long end = addr + size; | |
f6fcba70 TH |
2260 | |
2261 | flush_cache_vunmap(addr, end); | |
b521c43f | 2262 | unmap_kernel_range_noflush(addr, size); |
db64fe02 NP |
2263 | flush_tlb_kernel_range(addr, end); |
2264 | } | |
2265 | ||
e36176be URS |
2266 | static inline void setup_vmalloc_vm_locked(struct vm_struct *vm, |
2267 | struct vmap_area *va, unsigned long flags, const void *caller) | |
cf88c790 | 2268 | { |
cf88c790 TH |
2269 | vm->flags = flags; |
2270 | vm->addr = (void *)va->va_start; | |
2271 | vm->size = va->va_end - va->va_start; | |
2272 | vm->caller = caller; | |
db1aecaf | 2273 | va->vm = vm; |
e36176be URS |
2274 | } |
2275 | ||
2276 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, | |
2277 | unsigned long flags, const void *caller) | |
2278 | { | |
2279 | spin_lock(&vmap_area_lock); | |
2280 | setup_vmalloc_vm_locked(vm, va, flags, caller); | |
c69480ad | 2281 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2282 | } |
cf88c790 | 2283 | |
20fc02b4 | 2284 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2285 | { |
d4033afd | 2286 | /* |
20fc02b4 | 2287 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2288 | * we should make sure that vm has proper values. |
2289 | * Pair with smp_rmb() in show_numa_info(). | |
2290 | */ | |
2291 | smp_wmb(); | |
20fc02b4 | 2292 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2293 | } |
2294 | ||
db64fe02 | 2295 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2296 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2297 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2298 | { |
0006526d | 2299 | struct vmap_area *va; |
db64fe02 | 2300 | struct vm_struct *area; |
d98c9e83 | 2301 | unsigned long requested_size = size; |
1da177e4 | 2302 | |
52fd24ca | 2303 | BUG_ON(in_interrupt()); |
1da177e4 | 2304 | size = PAGE_ALIGN(size); |
31be8309 OH |
2305 | if (unlikely(!size)) |
2306 | return NULL; | |
1da177e4 | 2307 | |
252e5c6e | 2308 | if (flags & VM_IOREMAP) |
2309 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2310 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2311 | ||
cf88c790 | 2312 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2313 | if (unlikely(!area)) |
2314 | return NULL; | |
2315 | ||
71394fe5 AR |
2316 | if (!(flags & VM_NO_GUARD)) |
2317 | size += PAGE_SIZE; | |
1da177e4 | 2318 | |
db64fe02 NP |
2319 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2320 | if (IS_ERR(va)) { | |
2321 | kfree(area); | |
2322 | return NULL; | |
1da177e4 | 2323 | } |
1da177e4 | 2324 | |
d98c9e83 | 2325 | kasan_unpoison_vmalloc((void *)va->va_start, requested_size); |
f5252e00 | 2326 | |
d98c9e83 | 2327 | setup_vmalloc_vm(area, va, flags, caller); |
3c5c3cfb | 2328 | |
1da177e4 | 2329 | return area; |
1da177e4 LT |
2330 | } |
2331 | ||
c2968612 BH |
2332 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2333 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2334 | const void *caller) |
c2968612 | 2335 | { |
00ef2d2f DR |
2336 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2337 | GFP_KERNEL, caller); | |
c2968612 BH |
2338 | } |
2339 | ||
1da177e4 | 2340 | /** |
92eac168 MR |
2341 | * get_vm_area - reserve a contiguous kernel virtual area |
2342 | * @size: size of the area | |
2343 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2344 | * |
92eac168 MR |
2345 | * Search an area of @size in the kernel virtual mapping area, |
2346 | * and reserved it for out purposes. Returns the area descriptor | |
2347 | * on success or %NULL on failure. | |
a862f68a MR |
2348 | * |
2349 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2350 | */ |
2351 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2352 | { | |
2dca6999 | 2353 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2354 | NUMA_NO_NODE, GFP_KERNEL, |
2355 | __builtin_return_address(0)); | |
23016969 CL |
2356 | } |
2357 | ||
2358 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2359 | const void *caller) |
23016969 | 2360 | { |
2dca6999 | 2361 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2362 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2363 | } |
2364 | ||
e9da6e99 | 2365 | /** |
92eac168 MR |
2366 | * find_vm_area - find a continuous kernel virtual area |
2367 | * @addr: base address | |
e9da6e99 | 2368 | * |
92eac168 MR |
2369 | * Search for the kernel VM area starting at @addr, and return it. |
2370 | * It is up to the caller to do all required locking to keep the returned | |
2371 | * pointer valid. | |
a862f68a | 2372 | * |
74640617 | 2373 | * Return: the area descriptor on success or %NULL on failure. |
e9da6e99 MS |
2374 | */ |
2375 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2376 | { |
db64fe02 | 2377 | struct vmap_area *va; |
83342314 | 2378 | |
db64fe02 | 2379 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2380 | if (!va) |
2381 | return NULL; | |
1da177e4 | 2382 | |
688fcbfc | 2383 | return va->vm; |
1da177e4 LT |
2384 | } |
2385 | ||
7856dfeb | 2386 | /** |
92eac168 MR |
2387 | * remove_vm_area - find and remove a continuous kernel virtual area |
2388 | * @addr: base address | |
7856dfeb | 2389 | * |
92eac168 MR |
2390 | * Search for the kernel VM area starting at @addr, and remove it. |
2391 | * This function returns the found VM area, but using it is NOT safe | |
2392 | * on SMP machines, except for its size or flags. | |
a862f68a | 2393 | * |
74640617 | 2394 | * Return: the area descriptor on success or %NULL on failure. |
7856dfeb | 2395 | */ |
b3bdda02 | 2396 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2397 | { |
db64fe02 NP |
2398 | struct vmap_area *va; |
2399 | ||
5803ed29 CH |
2400 | might_sleep(); |
2401 | ||
dd3b8353 URS |
2402 | spin_lock(&vmap_area_lock); |
2403 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2404 | if (va && va->vm) { |
db1aecaf | 2405 | struct vm_struct *vm = va->vm; |
f5252e00 | 2406 | |
c69480ad | 2407 | va->vm = NULL; |
c69480ad JK |
2408 | spin_unlock(&vmap_area_lock); |
2409 | ||
a5af5aa8 | 2410 | kasan_free_shadow(vm); |
dd32c279 | 2411 | free_unmap_vmap_area(va); |
dd32c279 | 2412 | |
db64fe02 NP |
2413 | return vm; |
2414 | } | |
dd3b8353 URS |
2415 | |
2416 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2417 | return NULL; |
7856dfeb AK |
2418 | } |
2419 | ||
868b104d RE |
2420 | static inline void set_area_direct_map(const struct vm_struct *area, |
2421 | int (*set_direct_map)(struct page *page)) | |
2422 | { | |
2423 | int i; | |
2424 | ||
2425 | for (i = 0; i < area->nr_pages; i++) | |
2426 | if (page_address(area->pages[i])) | |
2427 | set_direct_map(area->pages[i]); | |
2428 | } | |
2429 | ||
2430 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2431 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2432 | { | |
868b104d RE |
2433 | unsigned long start = ULONG_MAX, end = 0; |
2434 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
31e67340 | 2435 | int flush_dmap = 0; |
868b104d RE |
2436 | int i; |
2437 | ||
868b104d RE |
2438 | remove_vm_area(area->addr); |
2439 | ||
2440 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2441 | if (!flush_reset) | |
2442 | return; | |
2443 | ||
2444 | /* | |
2445 | * If not deallocating pages, just do the flush of the VM area and | |
2446 | * return. | |
2447 | */ | |
2448 | if (!deallocate_pages) { | |
2449 | vm_unmap_aliases(); | |
2450 | return; | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * If execution gets here, flush the vm mapping and reset the direct | |
2455 | * map. Find the start and end range of the direct mappings to make sure | |
2456 | * the vm_unmap_aliases() flush includes the direct map. | |
2457 | */ | |
2458 | for (i = 0; i < area->nr_pages; i++) { | |
8e41f872 RE |
2459 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2460 | if (addr) { | |
868b104d | 2461 | start = min(addr, start); |
8e41f872 | 2462 | end = max(addr + PAGE_SIZE, end); |
31e67340 | 2463 | flush_dmap = 1; |
868b104d RE |
2464 | } |
2465 | } | |
2466 | ||
2467 | /* | |
2468 | * Set direct map to something invalid so that it won't be cached if | |
2469 | * there are any accesses after the TLB flush, then flush the TLB and | |
2470 | * reset the direct map permissions to the default. | |
2471 | */ | |
2472 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2473 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2474 | set_area_direct_map(area, set_direct_map_default_noflush); |
2475 | } | |
2476 | ||
b3bdda02 | 2477 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2478 | { |
2479 | struct vm_struct *area; | |
2480 | ||
2481 | if (!addr) | |
2482 | return; | |
2483 | ||
e69e9d4a | 2484 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2485 | addr)) |
1da177e4 | 2486 | return; |
1da177e4 | 2487 | |
6ade2032 | 2488 | area = find_vm_area(addr); |
1da177e4 | 2489 | if (unlikely(!area)) { |
4c8573e2 | 2490 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2491 | addr); |
1da177e4 LT |
2492 | return; |
2493 | } | |
2494 | ||
05e3ff95 CP |
2495 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2496 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2497 | |
c041098c | 2498 | kasan_poison_vmalloc(area->addr, get_vm_area_size(area)); |
3c5c3cfb | 2499 | |
868b104d RE |
2500 | vm_remove_mappings(area, deallocate_pages); |
2501 | ||
1da177e4 LT |
2502 | if (deallocate_pages) { |
2503 | int i; | |
2504 | ||
2505 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2506 | struct page *page = area->pages[i]; |
2507 | ||
2508 | BUG_ON(!page); | |
4949148a | 2509 | __free_pages(page, 0); |
1da177e4 | 2510 | } |
97105f0a | 2511 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2512 | |
244d63ee | 2513 | kvfree(area->pages); |
1da177e4 LT |
2514 | } |
2515 | ||
2516 | kfree(area); | |
1da177e4 | 2517 | } |
bf22e37a AR |
2518 | |
2519 | static inline void __vfree_deferred(const void *addr) | |
2520 | { | |
2521 | /* | |
2522 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2523 | * context. Preemption is absolutely fine here, because the llist_add() | |
2524 | * implementation is lockless, so it works even if we are adding to | |
73221d88 | 2525 | * another cpu's list. schedule_work() should be fine with this too. |
bf22e37a AR |
2526 | */ |
2527 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2528 | ||
2529 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2530 | schedule_work(&p->wq); | |
2531 | } | |
2532 | ||
2533 | /** | |
92eac168 MR |
2534 | * vfree_atomic - release memory allocated by vmalloc() |
2535 | * @addr: memory base address | |
bf22e37a | 2536 | * |
92eac168 MR |
2537 | * This one is just like vfree() but can be called in any atomic context |
2538 | * except NMIs. | |
bf22e37a AR |
2539 | */ |
2540 | void vfree_atomic(const void *addr) | |
2541 | { | |
2542 | BUG_ON(in_nmi()); | |
2543 | ||
2544 | kmemleak_free(addr); | |
2545 | ||
2546 | if (!addr) | |
2547 | return; | |
2548 | __vfree_deferred(addr); | |
2549 | } | |
2550 | ||
c67dc624 RP |
2551 | static void __vfree(const void *addr) |
2552 | { | |
2553 | if (unlikely(in_interrupt())) | |
2554 | __vfree_deferred(addr); | |
2555 | else | |
2556 | __vunmap(addr, 1); | |
2557 | } | |
2558 | ||
1da177e4 | 2559 | /** |
fa307474 MWO |
2560 | * vfree - Release memory allocated by vmalloc() |
2561 | * @addr: Memory base address | |
1da177e4 | 2562 | * |
fa307474 MWO |
2563 | * Free the virtually continuous memory area starting at @addr, as obtained |
2564 | * from one of the vmalloc() family of APIs. This will usually also free the | |
2565 | * physical memory underlying the virtual allocation, but that memory is | |
2566 | * reference counted, so it will not be freed until the last user goes away. | |
1da177e4 | 2567 | * |
fa307474 | 2568 | * If @addr is NULL, no operation is performed. |
c9fcee51 | 2569 | * |
fa307474 | 2570 | * Context: |
92eac168 | 2571 | * May sleep if called *not* from interrupt context. |
fa307474 MWO |
2572 | * Must not be called in NMI context (strictly speaking, it could be |
2573 | * if we have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2574 | * conventions for vfree() arch-depenedent would be a really bad idea). | |
1da177e4 | 2575 | */ |
b3bdda02 | 2576 | void vfree(const void *addr) |
1da177e4 | 2577 | { |
32fcfd40 | 2578 | BUG_ON(in_nmi()); |
89219d37 CM |
2579 | |
2580 | kmemleak_free(addr); | |
2581 | ||
a8dda165 AR |
2582 | might_sleep_if(!in_interrupt()); |
2583 | ||
32fcfd40 AV |
2584 | if (!addr) |
2585 | return; | |
c67dc624 RP |
2586 | |
2587 | __vfree(addr); | |
1da177e4 | 2588 | } |
1da177e4 LT |
2589 | EXPORT_SYMBOL(vfree); |
2590 | ||
2591 | /** | |
92eac168 MR |
2592 | * vunmap - release virtual mapping obtained by vmap() |
2593 | * @addr: memory base address | |
1da177e4 | 2594 | * |
92eac168 MR |
2595 | * Free the virtually contiguous memory area starting at @addr, |
2596 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2597 | * |
92eac168 | 2598 | * Must not be called in interrupt context. |
1da177e4 | 2599 | */ |
b3bdda02 | 2600 | void vunmap(const void *addr) |
1da177e4 LT |
2601 | { |
2602 | BUG_ON(in_interrupt()); | |
34754b69 | 2603 | might_sleep(); |
32fcfd40 AV |
2604 | if (addr) |
2605 | __vunmap(addr, 0); | |
1da177e4 | 2606 | } |
1da177e4 LT |
2607 | EXPORT_SYMBOL(vunmap); |
2608 | ||
2609 | /** | |
92eac168 MR |
2610 | * vmap - map an array of pages into virtually contiguous space |
2611 | * @pages: array of page pointers | |
2612 | * @count: number of pages to map | |
2613 | * @flags: vm_area->flags | |
2614 | * @prot: page protection for the mapping | |
2615 | * | |
b944afc9 CH |
2616 | * Maps @count pages from @pages into contiguous kernel virtual space. |
2617 | * If @flags contains %VM_MAP_PUT_PAGES the ownership of the pages array itself | |
2618 | * (which must be kmalloc or vmalloc memory) and one reference per pages in it | |
2619 | * are transferred from the caller to vmap(), and will be freed / dropped when | |
2620 | * vfree() is called on the return value. | |
a862f68a MR |
2621 | * |
2622 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2623 | */ |
2624 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2625 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2626 | { |
2627 | struct vm_struct *area; | |
65ee03c4 | 2628 | unsigned long size; /* In bytes */ |
1da177e4 | 2629 | |
34754b69 PZ |
2630 | might_sleep(); |
2631 | ||
ca79b0c2 | 2632 | if (count > totalram_pages()) |
1da177e4 LT |
2633 | return NULL; |
2634 | ||
65ee03c4 GJM |
2635 | size = (unsigned long)count << PAGE_SHIFT; |
2636 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2637 | if (!area) |
2638 | return NULL; | |
23016969 | 2639 | |
cca98e9f | 2640 | if (map_kernel_range((unsigned long)area->addr, size, pgprot_nx(prot), |
ed1f324c | 2641 | pages) < 0) { |
1da177e4 LT |
2642 | vunmap(area->addr); |
2643 | return NULL; | |
2644 | } | |
2645 | ||
c22ee528 | 2646 | if (flags & VM_MAP_PUT_PAGES) { |
b944afc9 | 2647 | area->pages = pages; |
c22ee528 ML |
2648 | area->nr_pages = count; |
2649 | } | |
1da177e4 LT |
2650 | return area->addr; |
2651 | } | |
1da177e4 LT |
2652 | EXPORT_SYMBOL(vmap); |
2653 | ||
3e9a9e25 CH |
2654 | #ifdef CONFIG_VMAP_PFN |
2655 | struct vmap_pfn_data { | |
2656 | unsigned long *pfns; | |
2657 | pgprot_t prot; | |
2658 | unsigned int idx; | |
2659 | }; | |
2660 | ||
2661 | static int vmap_pfn_apply(pte_t *pte, unsigned long addr, void *private) | |
2662 | { | |
2663 | struct vmap_pfn_data *data = private; | |
2664 | ||
2665 | if (WARN_ON_ONCE(pfn_valid(data->pfns[data->idx]))) | |
2666 | return -EINVAL; | |
2667 | *pte = pte_mkspecial(pfn_pte(data->pfns[data->idx++], data->prot)); | |
2668 | return 0; | |
2669 | } | |
2670 | ||
2671 | /** | |
2672 | * vmap_pfn - map an array of PFNs into virtually contiguous space | |
2673 | * @pfns: array of PFNs | |
2674 | * @count: number of pages to map | |
2675 | * @prot: page protection for the mapping | |
2676 | * | |
2677 | * Maps @count PFNs from @pfns into contiguous kernel virtual space and returns | |
2678 | * the start address of the mapping. | |
2679 | */ | |
2680 | void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot) | |
2681 | { | |
2682 | struct vmap_pfn_data data = { .pfns = pfns, .prot = pgprot_nx(prot) }; | |
2683 | struct vm_struct *area; | |
2684 | ||
2685 | area = get_vm_area_caller(count * PAGE_SIZE, VM_IOREMAP, | |
2686 | __builtin_return_address(0)); | |
2687 | if (!area) | |
2688 | return NULL; | |
2689 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
2690 | count * PAGE_SIZE, vmap_pfn_apply, &data)) { | |
2691 | free_vm_area(area); | |
2692 | return NULL; | |
2693 | } | |
2694 | return area->addr; | |
2695 | } | |
2696 | EXPORT_SYMBOL_GPL(vmap_pfn); | |
2697 | #endif /* CONFIG_VMAP_PFN */ | |
2698 | ||
e31d9eb5 | 2699 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 2700 | pgprot_t prot, int node) |
1da177e4 | 2701 | { |
930f036b | 2702 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
f255935b | 2703 | unsigned int nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
34fe6537 AM |
2704 | unsigned long array_size; |
2705 | unsigned int i; | |
f255935b | 2706 | struct page **pages; |
1da177e4 | 2707 | |
34fe6537 | 2708 | array_size = (unsigned long)nr_pages * sizeof(struct page *); |
f255935b CH |
2709 | gfp_mask |= __GFP_NOWARN; |
2710 | if (!(gfp_mask & (GFP_DMA | GFP_DMA32))) | |
2711 | gfp_mask |= __GFP_HIGHMEM; | |
1da177e4 | 2712 | |
1da177e4 | 2713 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 2714 | if (array_size > PAGE_SIZE) { |
f255935b CH |
2715 | pages = __vmalloc_node(array_size, 1, nested_gfp, node, |
2716 | area->caller); | |
286e1ea3 | 2717 | } else { |
976d6dfb | 2718 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2719 | } |
7ea36242 AK |
2720 | |
2721 | if (!pages) { | |
8945a723 | 2722 | free_vm_area(area); |
1da177e4 LT |
2723 | return NULL; |
2724 | } | |
1da177e4 | 2725 | |
7ea36242 AK |
2726 | area->pages = pages; |
2727 | area->nr_pages = nr_pages; | |
2728 | ||
1da177e4 | 2729 | for (i = 0; i < area->nr_pages; i++) { |
bf53d6f8 CL |
2730 | struct page *page; |
2731 | ||
4b90951c | 2732 | if (node == NUMA_NO_NODE) |
f255935b | 2733 | page = alloc_page(gfp_mask); |
930fc45a | 2734 | else |
f255935b | 2735 | page = alloc_pages_node(node, gfp_mask, 0); |
bf53d6f8 CL |
2736 | |
2737 | if (unlikely(!page)) { | |
82afbc32 | 2738 | /* Successfully allocated i pages, free them in __vfree() */ |
1da177e4 | 2739 | area->nr_pages = i; |
97105f0a | 2740 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 LT |
2741 | goto fail; |
2742 | } | |
bf53d6f8 | 2743 | area->pages[i] = page; |
dcf61ff0 | 2744 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 2745 | cond_resched(); |
1da177e4 | 2746 | } |
97105f0a | 2747 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2748 | |
ed1f324c CH |
2749 | if (map_kernel_range((unsigned long)area->addr, get_vm_area_size(area), |
2750 | prot, pages) < 0) | |
1da177e4 | 2751 | goto fail; |
ed1f324c | 2752 | |
1da177e4 LT |
2753 | return area->addr; |
2754 | ||
2755 | fail: | |
a8e99259 | 2756 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2757 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 2758 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 2759 | __vfree(area->addr); |
1da177e4 LT |
2760 | return NULL; |
2761 | } | |
2762 | ||
2763 | /** | |
92eac168 MR |
2764 | * __vmalloc_node_range - allocate virtually contiguous memory |
2765 | * @size: allocation size | |
2766 | * @align: desired alignment | |
2767 | * @start: vm area range start | |
2768 | * @end: vm area range end | |
2769 | * @gfp_mask: flags for the page level allocator | |
2770 | * @prot: protection mask for the allocated pages | |
2771 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2772 | * @node: node to use for allocation or NUMA_NO_NODE | |
2773 | * @caller: caller's return address | |
2774 | * | |
2775 | * Allocate enough pages to cover @size from the page level | |
2776 | * allocator with @gfp_mask flags. Map them into contiguous | |
2777 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2778 | * |
2779 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2780 | */ |
d0a21265 DR |
2781 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2782 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2783 | pgprot_t prot, unsigned long vm_flags, int node, |
2784 | const void *caller) | |
1da177e4 LT |
2785 | { |
2786 | struct vm_struct *area; | |
89219d37 CM |
2787 | void *addr; |
2788 | unsigned long real_size = size; | |
1da177e4 LT |
2789 | |
2790 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 2791 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 2792 | goto fail; |
1da177e4 | 2793 | |
d98c9e83 | 2794 | area = __get_vm_area_node(real_size, align, VM_ALLOC | VM_UNINITIALIZED | |
cb9e3c29 | 2795 | vm_flags, start, end, node, gfp_mask, caller); |
1da177e4 | 2796 | if (!area) |
de7d2b56 | 2797 | goto fail; |
1da177e4 | 2798 | |
3722e13c | 2799 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 2800 | if (!addr) |
b82225f3 | 2801 | return NULL; |
89219d37 | 2802 | |
f5252e00 | 2803 | /* |
20fc02b4 ZY |
2804 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2805 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2806 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2807 | */ |
20fc02b4 | 2808 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2809 | |
94f4a161 | 2810 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2811 | |
2812 | return addr; | |
de7d2b56 JP |
2813 | |
2814 | fail: | |
a8e99259 | 2815 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2816 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 2817 | return NULL; |
1da177e4 LT |
2818 | } |
2819 | ||
d0a21265 | 2820 | /** |
92eac168 MR |
2821 | * __vmalloc_node - allocate virtually contiguous memory |
2822 | * @size: allocation size | |
2823 | * @align: desired alignment | |
2824 | * @gfp_mask: flags for the page level allocator | |
92eac168 MR |
2825 | * @node: node to use for allocation or NUMA_NO_NODE |
2826 | * @caller: caller's return address | |
a7c3e901 | 2827 | * |
f38fcb9c CH |
2828 | * Allocate enough pages to cover @size from the page level allocator with |
2829 | * @gfp_mask flags. Map them into contiguous kernel virtual space. | |
a7c3e901 | 2830 | * |
92eac168 MR |
2831 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2832 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2833 | * |
92eac168 MR |
2834 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2835 | * with mm people. | |
a862f68a MR |
2836 | * |
2837 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2838 | */ |
2b905948 | 2839 | void *__vmalloc_node(unsigned long size, unsigned long align, |
f38fcb9c | 2840 | gfp_t gfp_mask, int node, const void *caller) |
d0a21265 DR |
2841 | { |
2842 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
f38fcb9c | 2843 | gfp_mask, PAGE_KERNEL, 0, node, caller); |
d0a21265 | 2844 | } |
c3f896dc CH |
2845 | /* |
2846 | * This is only for performance analysis of vmalloc and stress purpose. | |
2847 | * It is required by vmalloc test module, therefore do not use it other | |
2848 | * than that. | |
2849 | */ | |
2850 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2851 | EXPORT_SYMBOL_GPL(__vmalloc_node); | |
2852 | #endif | |
d0a21265 | 2853 | |
88dca4ca | 2854 | void *__vmalloc(unsigned long size, gfp_t gfp_mask) |
930fc45a | 2855 | { |
f38fcb9c | 2856 | return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE, |
23016969 | 2857 | __builtin_return_address(0)); |
930fc45a | 2858 | } |
1da177e4 LT |
2859 | EXPORT_SYMBOL(__vmalloc); |
2860 | ||
2861 | /** | |
92eac168 MR |
2862 | * vmalloc - allocate virtually contiguous memory |
2863 | * @size: allocation size | |
2864 | * | |
2865 | * Allocate enough pages to cover @size from the page level | |
2866 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2867 | * |
92eac168 MR |
2868 | * For tight control over page level allocator and protection flags |
2869 | * use __vmalloc() instead. | |
a862f68a MR |
2870 | * |
2871 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2872 | */ |
2873 | void *vmalloc(unsigned long size) | |
2874 | { | |
4d39d728 CH |
2875 | return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE, |
2876 | __builtin_return_address(0)); | |
1da177e4 | 2877 | } |
1da177e4 LT |
2878 | EXPORT_SYMBOL(vmalloc); |
2879 | ||
e1ca7788 | 2880 | /** |
92eac168 MR |
2881 | * vzalloc - allocate virtually contiguous memory with zero fill |
2882 | * @size: allocation size | |
2883 | * | |
2884 | * Allocate enough pages to cover @size from the page level | |
2885 | * allocator and map them into contiguous kernel virtual space. | |
2886 | * The memory allocated is set to zero. | |
2887 | * | |
2888 | * For tight control over page level allocator and protection flags | |
2889 | * use __vmalloc() instead. | |
a862f68a MR |
2890 | * |
2891 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2892 | */ |
2893 | void *vzalloc(unsigned long size) | |
2894 | { | |
4d39d728 CH |
2895 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE, |
2896 | __builtin_return_address(0)); | |
e1ca7788 DY |
2897 | } |
2898 | EXPORT_SYMBOL(vzalloc); | |
2899 | ||
83342314 | 2900 | /** |
ead04089 REB |
2901 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
2902 | * @size: allocation size | |
83342314 | 2903 | * |
ead04089 REB |
2904 | * The resulting memory area is zeroed so it can be mapped to userspace |
2905 | * without leaking data. | |
a862f68a MR |
2906 | * |
2907 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2908 | */ |
2909 | void *vmalloc_user(unsigned long size) | |
2910 | { | |
bc84c535 RP |
2911 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2912 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
2913 | VM_USERMAP, NUMA_NO_NODE, | |
2914 | __builtin_return_address(0)); | |
83342314 NP |
2915 | } |
2916 | EXPORT_SYMBOL(vmalloc_user); | |
2917 | ||
930fc45a | 2918 | /** |
92eac168 MR |
2919 | * vmalloc_node - allocate memory on a specific node |
2920 | * @size: allocation size | |
2921 | * @node: numa node | |
930fc45a | 2922 | * |
92eac168 MR |
2923 | * Allocate enough pages to cover @size from the page level |
2924 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 2925 | * |
92eac168 MR |
2926 | * For tight control over page level allocator and protection flags |
2927 | * use __vmalloc() instead. | |
a862f68a MR |
2928 | * |
2929 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2930 | */ |
2931 | void *vmalloc_node(unsigned long size, int node) | |
2932 | { | |
f38fcb9c CH |
2933 | return __vmalloc_node(size, 1, GFP_KERNEL, node, |
2934 | __builtin_return_address(0)); | |
930fc45a CL |
2935 | } |
2936 | EXPORT_SYMBOL(vmalloc_node); | |
2937 | ||
e1ca7788 DY |
2938 | /** |
2939 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2940 | * @size: allocation size | |
2941 | * @node: numa node | |
2942 | * | |
2943 | * Allocate enough pages to cover @size from the page level | |
2944 | * allocator and map them into contiguous kernel virtual space. | |
2945 | * The memory allocated is set to zero. | |
2946 | * | |
a862f68a | 2947 | * Return: pointer to the allocated memory or %NULL on error |
e1ca7788 DY |
2948 | */ |
2949 | void *vzalloc_node(unsigned long size, int node) | |
2950 | { | |
4d39d728 CH |
2951 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node, |
2952 | __builtin_return_address(0)); | |
e1ca7788 DY |
2953 | } |
2954 | EXPORT_SYMBOL(vzalloc_node); | |
2955 | ||
0d08e0d3 | 2956 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2957 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2958 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2959 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2960 | #else |
698d0831 MH |
2961 | /* |
2962 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2963 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2964 | */ | |
2965 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2966 | #endif |
2967 | ||
1da177e4 | 2968 | /** |
92eac168 MR |
2969 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2970 | * @size: allocation size | |
1da177e4 | 2971 | * |
92eac168 MR |
2972 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2973 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2974 | * |
2975 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2976 | */ |
2977 | void *vmalloc_32(unsigned long size) | |
2978 | { | |
f38fcb9c CH |
2979 | return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE, |
2980 | __builtin_return_address(0)); | |
1da177e4 | 2981 | } |
1da177e4 LT |
2982 | EXPORT_SYMBOL(vmalloc_32); |
2983 | ||
83342314 | 2984 | /** |
ead04089 | 2985 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2986 | * @size: allocation size |
ead04089 REB |
2987 | * |
2988 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2989 | * mapped to userspace without leaking data. | |
a862f68a MR |
2990 | * |
2991 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2992 | */ |
2993 | void *vmalloc_32_user(unsigned long size) | |
2994 | { | |
bc84c535 RP |
2995 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2996 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2997 | VM_USERMAP, NUMA_NO_NODE, | |
2998 | __builtin_return_address(0)); | |
83342314 NP |
2999 | } |
3000 | EXPORT_SYMBOL(vmalloc_32_user); | |
3001 | ||
d0107eb0 KH |
3002 | /* |
3003 | * small helper routine , copy contents to buf from addr. | |
3004 | * If the page is not present, fill zero. | |
3005 | */ | |
3006 | ||
3007 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
3008 | { | |
3009 | struct page *p; | |
3010 | int copied = 0; | |
3011 | ||
3012 | while (count) { | |
3013 | unsigned long offset, length; | |
3014 | ||
891c49ab | 3015 | offset = offset_in_page(addr); |
d0107eb0 KH |
3016 | length = PAGE_SIZE - offset; |
3017 | if (length > count) | |
3018 | length = count; | |
3019 | p = vmalloc_to_page(addr); | |
3020 | /* | |
3021 | * To do safe access to this _mapped_ area, we need | |
3022 | * lock. But adding lock here means that we need to add | |
3023 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
3024 | * interface, rarely used. Instead of that, we'll use | |
3025 | * kmap() and get small overhead in this access function. | |
3026 | */ | |
3027 | if (p) { | |
3028 | /* | |
3029 | * we can expect USER0 is not used (see vread/vwrite's | |
3030 | * function description) | |
3031 | */ | |
9b04c5fe | 3032 | void *map = kmap_atomic(p); |
d0107eb0 | 3033 | memcpy(buf, map + offset, length); |
9b04c5fe | 3034 | kunmap_atomic(map); |
d0107eb0 KH |
3035 | } else |
3036 | memset(buf, 0, length); | |
3037 | ||
3038 | addr += length; | |
3039 | buf += length; | |
3040 | copied += length; | |
3041 | count -= length; | |
3042 | } | |
3043 | return copied; | |
3044 | } | |
3045 | ||
3046 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
3047 | { | |
3048 | struct page *p; | |
3049 | int copied = 0; | |
3050 | ||
3051 | while (count) { | |
3052 | unsigned long offset, length; | |
3053 | ||
891c49ab | 3054 | offset = offset_in_page(addr); |
d0107eb0 KH |
3055 | length = PAGE_SIZE - offset; |
3056 | if (length > count) | |
3057 | length = count; | |
3058 | p = vmalloc_to_page(addr); | |
3059 | /* | |
3060 | * To do safe access to this _mapped_ area, we need | |
3061 | * lock. But adding lock here means that we need to add | |
3062 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
3063 | * interface, rarely used. Instead of that, we'll use | |
3064 | * kmap() and get small overhead in this access function. | |
3065 | */ | |
3066 | if (p) { | |
3067 | /* | |
3068 | * we can expect USER0 is not used (see vread/vwrite's | |
3069 | * function description) | |
3070 | */ | |
9b04c5fe | 3071 | void *map = kmap_atomic(p); |
d0107eb0 | 3072 | memcpy(map + offset, buf, length); |
9b04c5fe | 3073 | kunmap_atomic(map); |
d0107eb0 KH |
3074 | } |
3075 | addr += length; | |
3076 | buf += length; | |
3077 | copied += length; | |
3078 | count -= length; | |
3079 | } | |
3080 | return copied; | |
3081 | } | |
3082 | ||
3083 | /** | |
92eac168 MR |
3084 | * vread() - read vmalloc area in a safe way. |
3085 | * @buf: buffer for reading data | |
3086 | * @addr: vm address. | |
3087 | * @count: number of bytes to be read. | |
3088 | * | |
92eac168 MR |
3089 | * This function checks that addr is a valid vmalloc'ed area, and |
3090 | * copy data from that area to a given buffer. If the given memory range | |
3091 | * of [addr...addr+count) includes some valid address, data is copied to | |
3092 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
3093 | * IOREMAP area is treated as memory hole and no copy is done. | |
3094 | * | |
3095 | * If [addr...addr+count) doesn't includes any intersects with alive | |
3096 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
3097 | * | |
3098 | * Note: In usual ops, vread() is never necessary because the caller | |
3099 | * should know vmalloc() area is valid and can use memcpy(). | |
3100 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 3101 | * any information, as /dev/kmem. |
a862f68a MR |
3102 | * |
3103 | * Return: number of bytes for which addr and buf should be increased | |
3104 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
3105 | * include any intersection with valid vmalloc area | |
d0107eb0 | 3106 | */ |
1da177e4 LT |
3107 | long vread(char *buf, char *addr, unsigned long count) |
3108 | { | |
e81ce85f JK |
3109 | struct vmap_area *va; |
3110 | struct vm_struct *vm; | |
1da177e4 | 3111 | char *vaddr, *buf_start = buf; |
d0107eb0 | 3112 | unsigned long buflen = count; |
1da177e4 LT |
3113 | unsigned long n; |
3114 | ||
3115 | /* Don't allow overflow */ | |
3116 | if ((unsigned long) addr + count < count) | |
3117 | count = -(unsigned long) addr; | |
3118 | ||
e81ce85f | 3119 | spin_lock(&vmap_area_lock); |
f608788c SD |
3120 | va = __find_vmap_area((unsigned long)addr); |
3121 | if (!va) | |
3122 | goto finished; | |
3123 | list_for_each_entry_from(va, &vmap_area_list, list) { | |
e81ce85f JK |
3124 | if (!count) |
3125 | break; | |
3126 | ||
688fcbfc | 3127 | if (!va->vm) |
e81ce85f JK |
3128 | continue; |
3129 | ||
3130 | vm = va->vm; | |
3131 | vaddr = (char *) vm->addr; | |
762216ab | 3132 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
3133 | continue; |
3134 | while (addr < vaddr) { | |
3135 | if (count == 0) | |
3136 | goto finished; | |
3137 | *buf = '\0'; | |
3138 | buf++; | |
3139 | addr++; | |
3140 | count--; | |
3141 | } | |
762216ab | 3142 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
3143 | if (n > count) |
3144 | n = count; | |
e81ce85f | 3145 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
3146 | aligned_vread(buf, addr, n); |
3147 | else /* IOREMAP area is treated as memory hole */ | |
3148 | memset(buf, 0, n); | |
3149 | buf += n; | |
3150 | addr += n; | |
3151 | count -= n; | |
1da177e4 LT |
3152 | } |
3153 | finished: | |
e81ce85f | 3154 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3155 | |
3156 | if (buf == buf_start) | |
3157 | return 0; | |
3158 | /* zero-fill memory holes */ | |
3159 | if (buf != buf_start + buflen) | |
3160 | memset(buf, 0, buflen - (buf - buf_start)); | |
3161 | ||
3162 | return buflen; | |
1da177e4 LT |
3163 | } |
3164 | ||
d0107eb0 | 3165 | /** |
92eac168 MR |
3166 | * vwrite() - write vmalloc area in a safe way. |
3167 | * @buf: buffer for source data | |
3168 | * @addr: vm address. | |
3169 | * @count: number of bytes to be read. | |
3170 | * | |
92eac168 MR |
3171 | * This function checks that addr is a valid vmalloc'ed area, and |
3172 | * copy data from a buffer to the given addr. If specified range of | |
3173 | * [addr...addr+count) includes some valid address, data is copied from | |
3174 | * proper area of @buf. If there are memory holes, no copy to hole. | |
3175 | * IOREMAP area is treated as memory hole and no copy is done. | |
3176 | * | |
3177 | * If [addr...addr+count) doesn't includes any intersects with alive | |
3178 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
3179 | * | |
3180 | * Note: In usual ops, vwrite() is never necessary because the caller | |
3181 | * should know vmalloc() area is valid and can use memcpy(). | |
3182 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 3183 | * any information, as /dev/kmem. |
a862f68a MR |
3184 | * |
3185 | * Return: number of bytes for which addr and buf should be | |
3186 | * increased (same number as @count) or %0 if [addr...addr+count) | |
3187 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 3188 | */ |
1da177e4 LT |
3189 | long vwrite(char *buf, char *addr, unsigned long count) |
3190 | { | |
e81ce85f JK |
3191 | struct vmap_area *va; |
3192 | struct vm_struct *vm; | |
d0107eb0 KH |
3193 | char *vaddr; |
3194 | unsigned long n, buflen; | |
3195 | int copied = 0; | |
1da177e4 LT |
3196 | |
3197 | /* Don't allow overflow */ | |
3198 | if ((unsigned long) addr + count < count) | |
3199 | count = -(unsigned long) addr; | |
d0107eb0 | 3200 | buflen = count; |
1da177e4 | 3201 | |
e81ce85f JK |
3202 | spin_lock(&vmap_area_lock); |
3203 | list_for_each_entry(va, &vmap_area_list, list) { | |
3204 | if (!count) | |
3205 | break; | |
3206 | ||
688fcbfc | 3207 | if (!va->vm) |
e81ce85f JK |
3208 | continue; |
3209 | ||
3210 | vm = va->vm; | |
3211 | vaddr = (char *) vm->addr; | |
762216ab | 3212 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
3213 | continue; |
3214 | while (addr < vaddr) { | |
3215 | if (count == 0) | |
3216 | goto finished; | |
3217 | buf++; | |
3218 | addr++; | |
3219 | count--; | |
3220 | } | |
762216ab | 3221 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
3222 | if (n > count) |
3223 | n = count; | |
e81ce85f | 3224 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
3225 | aligned_vwrite(buf, addr, n); |
3226 | copied++; | |
3227 | } | |
3228 | buf += n; | |
3229 | addr += n; | |
3230 | count -= n; | |
1da177e4 LT |
3231 | } |
3232 | finished: | |
e81ce85f | 3233 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3234 | if (!copied) |
3235 | return 0; | |
3236 | return buflen; | |
1da177e4 | 3237 | } |
83342314 NP |
3238 | |
3239 | /** | |
92eac168 MR |
3240 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
3241 | * @vma: vma to cover | |
3242 | * @uaddr: target user address to start at | |
3243 | * @kaddr: virtual address of vmalloc kernel memory | |
bdebd6a2 | 3244 | * @pgoff: offset from @kaddr to start at |
92eac168 | 3245 | * @size: size of map area |
7682486b | 3246 | * |
92eac168 | 3247 | * Returns: 0 for success, -Exxx on failure |
83342314 | 3248 | * |
92eac168 MR |
3249 | * This function checks that @kaddr is a valid vmalloc'ed area, |
3250 | * and that it is big enough to cover the range starting at | |
3251 | * @uaddr in @vma. Will return failure if that criteria isn't | |
3252 | * met. | |
83342314 | 3253 | * |
92eac168 | 3254 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 3255 | */ |
e69e9d4a | 3256 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
bdebd6a2 JH |
3257 | void *kaddr, unsigned long pgoff, |
3258 | unsigned long size) | |
83342314 NP |
3259 | { |
3260 | struct vm_struct *area; | |
bdebd6a2 JH |
3261 | unsigned long off; |
3262 | unsigned long end_index; | |
3263 | ||
3264 | if (check_shl_overflow(pgoff, PAGE_SHIFT, &off)) | |
3265 | return -EINVAL; | |
83342314 | 3266 | |
e69e9d4a HD |
3267 | size = PAGE_ALIGN(size); |
3268 | ||
3269 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
3270 | return -EINVAL; |
3271 | ||
e69e9d4a | 3272 | area = find_vm_area(kaddr); |
83342314 | 3273 | if (!area) |
db64fe02 | 3274 | return -EINVAL; |
83342314 | 3275 | |
fe9041c2 | 3276 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3277 | return -EINVAL; |
83342314 | 3278 | |
bdebd6a2 JH |
3279 | if (check_add_overflow(size, off, &end_index) || |
3280 | end_index > get_vm_area_size(area)) | |
db64fe02 | 3281 | return -EINVAL; |
bdebd6a2 | 3282 | kaddr += off; |
83342314 | 3283 | |
83342314 | 3284 | do { |
e69e9d4a | 3285 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3286 | int ret; |
3287 | ||
83342314 NP |
3288 | ret = vm_insert_page(vma, uaddr, page); |
3289 | if (ret) | |
3290 | return ret; | |
3291 | ||
3292 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3293 | kaddr += PAGE_SIZE; |
3294 | size -= PAGE_SIZE; | |
3295 | } while (size > 0); | |
83342314 | 3296 | |
314e51b9 | 3297 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3298 | |
db64fe02 | 3299 | return 0; |
83342314 | 3300 | } |
e69e9d4a HD |
3301 | |
3302 | /** | |
92eac168 MR |
3303 | * remap_vmalloc_range - map vmalloc pages to userspace |
3304 | * @vma: vma to cover (map full range of vma) | |
3305 | * @addr: vmalloc memory | |
3306 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3307 | * |
92eac168 | 3308 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3309 | * |
92eac168 MR |
3310 | * This function checks that addr is a valid vmalloc'ed area, and |
3311 | * that it is big enough to cover the vma. Will return failure if | |
3312 | * that criteria isn't met. | |
e69e9d4a | 3313 | * |
92eac168 | 3314 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3315 | */ |
3316 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3317 | unsigned long pgoff) | |
3318 | { | |
3319 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
bdebd6a2 | 3320 | addr, pgoff, |
e69e9d4a HD |
3321 | vma->vm_end - vma->vm_start); |
3322 | } | |
83342314 NP |
3323 | EXPORT_SYMBOL(remap_vmalloc_range); |
3324 | ||
5f4352fb JF |
3325 | void free_vm_area(struct vm_struct *area) |
3326 | { | |
3327 | struct vm_struct *ret; | |
3328 | ret = remove_vm_area(area->addr); | |
3329 | BUG_ON(ret != area); | |
3330 | kfree(area); | |
3331 | } | |
3332 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3333 | |
4f8b02b4 | 3334 | #ifdef CONFIG_SMP |
ca23e405 TH |
3335 | static struct vmap_area *node_to_va(struct rb_node *n) |
3336 | { | |
4583e773 | 3337 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3338 | } |
3339 | ||
3340 | /** | |
68ad4a33 URS |
3341 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3342 | * @addr: target address | |
ca23e405 | 3343 | * |
68ad4a33 URS |
3344 | * Returns: vmap_area if it is found. If there is no such area |
3345 | * the first highest(reverse order) vmap_area is returned | |
3346 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3347 | * if there are no any areas before @addr. | |
ca23e405 | 3348 | */ |
68ad4a33 URS |
3349 | static struct vmap_area * |
3350 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3351 | { |
68ad4a33 URS |
3352 | struct vmap_area *va, *tmp; |
3353 | struct rb_node *n; | |
3354 | ||
3355 | n = free_vmap_area_root.rb_node; | |
3356 | va = NULL; | |
ca23e405 TH |
3357 | |
3358 | while (n) { | |
68ad4a33 URS |
3359 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3360 | if (tmp->va_start <= addr) { | |
3361 | va = tmp; | |
3362 | if (tmp->va_end >= addr) | |
3363 | break; | |
3364 | ||
ca23e405 | 3365 | n = n->rb_right; |
68ad4a33 URS |
3366 | } else { |
3367 | n = n->rb_left; | |
3368 | } | |
ca23e405 TH |
3369 | } |
3370 | ||
68ad4a33 | 3371 | return va; |
ca23e405 TH |
3372 | } |
3373 | ||
3374 | /** | |
68ad4a33 URS |
3375 | * pvm_determine_end_from_reverse - find the highest aligned address |
3376 | * of free block below VMALLOC_END | |
3377 | * @va: | |
3378 | * in - the VA we start the search(reverse order); | |
3379 | * out - the VA with the highest aligned end address. | |
799fa85d | 3380 | * @align: alignment for required highest address |
ca23e405 | 3381 | * |
68ad4a33 | 3382 | * Returns: determined end address within vmap_area |
ca23e405 | 3383 | */ |
68ad4a33 URS |
3384 | static unsigned long |
3385 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3386 | { |
68ad4a33 | 3387 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3388 | unsigned long addr; |
3389 | ||
68ad4a33 URS |
3390 | if (likely(*va)) { |
3391 | list_for_each_entry_from_reverse((*va), | |
3392 | &free_vmap_area_list, list) { | |
3393 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3394 | if ((*va)->va_start < addr) | |
3395 | return addr; | |
3396 | } | |
ca23e405 TH |
3397 | } |
3398 | ||
68ad4a33 | 3399 | return 0; |
ca23e405 TH |
3400 | } |
3401 | ||
3402 | /** | |
3403 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3404 | * @offsets: array containing offset of each area | |
3405 | * @sizes: array containing size of each area | |
3406 | * @nr_vms: the number of areas to allocate | |
3407 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3408 | * |
3409 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3410 | * vm_structs on success, %NULL on failure | |
3411 | * | |
3412 | * Percpu allocator wants to use congruent vm areas so that it can | |
3413 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3414 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3415 | * be scattered pretty far, distance between two areas easily going up | |
3416 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3417 | * areas are allocated from top. | |
ca23e405 | 3418 | * |
68ad4a33 URS |
3419 | * Despite its complicated look, this allocator is rather simple. It |
3420 | * does everything top-down and scans free blocks from the end looking | |
3421 | * for matching base. While scanning, if any of the areas do not fit the | |
3422 | * base address is pulled down to fit the area. Scanning is repeated till | |
3423 | * all the areas fit and then all necessary data structures are inserted | |
3424 | * and the result is returned. | |
ca23e405 TH |
3425 | */ |
3426 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3427 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3428 | size_t align) |
ca23e405 TH |
3429 | { |
3430 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3431 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3432 | struct vmap_area **vas, *va; |
ca23e405 TH |
3433 | struct vm_struct **vms; |
3434 | int area, area2, last_area, term_area; | |
253a496d | 3435 | unsigned long base, start, size, end, last_end, orig_start, orig_end; |
ca23e405 | 3436 | bool purged = false; |
68ad4a33 | 3437 | enum fit_type type; |
ca23e405 | 3438 | |
ca23e405 | 3439 | /* verify parameters and allocate data structures */ |
891c49ab | 3440 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3441 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3442 | start = offsets[area]; | |
3443 | end = start + sizes[area]; | |
3444 | ||
3445 | /* is everything aligned properly? */ | |
3446 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3447 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3448 | ||
3449 | /* detect the area with the highest address */ | |
3450 | if (start > offsets[last_area]) | |
3451 | last_area = area; | |
3452 | ||
c568da28 | 3453 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3454 | unsigned long start2 = offsets[area2]; |
3455 | unsigned long end2 = start2 + sizes[area2]; | |
3456 | ||
c568da28 | 3457 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3458 | } |
3459 | } | |
3460 | last_end = offsets[last_area] + sizes[last_area]; | |
3461 | ||
3462 | if (vmalloc_end - vmalloc_start < last_end) { | |
3463 | WARN_ON(true); | |
3464 | return NULL; | |
3465 | } | |
3466 | ||
4d67d860 TM |
3467 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3468 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3469 | if (!vas || !vms) |
f1db7afd | 3470 | goto err_free2; |
ca23e405 TH |
3471 | |
3472 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3473 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3474 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3475 | if (!vas[area] || !vms[area]) |
3476 | goto err_free; | |
3477 | } | |
3478 | retry: | |
e36176be | 3479 | spin_lock(&free_vmap_area_lock); |
ca23e405 TH |
3480 | |
3481 | /* start scanning - we scan from the top, begin with the last area */ | |
3482 | area = term_area = last_area; | |
3483 | start = offsets[area]; | |
3484 | end = start + sizes[area]; | |
3485 | ||
68ad4a33 URS |
3486 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3487 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3488 | |
3489 | while (true) { | |
ca23e405 TH |
3490 | /* |
3491 | * base might have underflowed, add last_end before | |
3492 | * comparing. | |
3493 | */ | |
68ad4a33 URS |
3494 | if (base + last_end < vmalloc_start + last_end) |
3495 | goto overflow; | |
ca23e405 TH |
3496 | |
3497 | /* | |
68ad4a33 | 3498 | * Fitting base has not been found. |
ca23e405 | 3499 | */ |
68ad4a33 URS |
3500 | if (va == NULL) |
3501 | goto overflow; | |
ca23e405 | 3502 | |
5336e52c | 3503 | /* |
d8cc323d | 3504 | * If required width exceeds current VA block, move |
5336e52c KS |
3505 | * base downwards and then recheck. |
3506 | */ | |
3507 | if (base + end > va->va_end) { | |
3508 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3509 | term_area = area; | |
3510 | continue; | |
3511 | } | |
3512 | ||
ca23e405 | 3513 | /* |
68ad4a33 | 3514 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3515 | */ |
5336e52c | 3516 | if (base + start < va->va_start) { |
68ad4a33 URS |
3517 | va = node_to_va(rb_prev(&va->rb_node)); |
3518 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3519 | term_area = area; |
3520 | continue; | |
3521 | } | |
3522 | ||
3523 | /* | |
3524 | * This area fits, move on to the previous one. If | |
3525 | * the previous one is the terminal one, we're done. | |
3526 | */ | |
3527 | area = (area + nr_vms - 1) % nr_vms; | |
3528 | if (area == term_area) | |
3529 | break; | |
68ad4a33 | 3530 | |
ca23e405 TH |
3531 | start = offsets[area]; |
3532 | end = start + sizes[area]; | |
68ad4a33 | 3533 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3534 | } |
68ad4a33 | 3535 | |
ca23e405 TH |
3536 | /* we've found a fitting base, insert all va's */ |
3537 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3538 | int ret; |
ca23e405 | 3539 | |
68ad4a33 URS |
3540 | start = base + offsets[area]; |
3541 | size = sizes[area]; | |
ca23e405 | 3542 | |
68ad4a33 URS |
3543 | va = pvm_find_va_enclose_addr(start); |
3544 | if (WARN_ON_ONCE(va == NULL)) | |
3545 | /* It is a BUG(), but trigger recovery instead. */ | |
3546 | goto recovery; | |
3547 | ||
3548 | type = classify_va_fit_type(va, start, size); | |
3549 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3550 | /* It is a BUG(), but trigger recovery instead. */ | |
3551 | goto recovery; | |
3552 | ||
3553 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3554 | if (unlikely(ret)) | |
3555 | goto recovery; | |
3556 | ||
3557 | /* Allocated area. */ | |
3558 | va = vas[area]; | |
3559 | va->va_start = start; | |
3560 | va->va_end = start + size; | |
68ad4a33 | 3561 | } |
ca23e405 | 3562 | |
e36176be | 3563 | spin_unlock(&free_vmap_area_lock); |
ca23e405 | 3564 | |
253a496d DA |
3565 | /* populate the kasan shadow space */ |
3566 | for (area = 0; area < nr_vms; area++) { | |
3567 | if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area])) | |
3568 | goto err_free_shadow; | |
3569 | ||
3570 | kasan_unpoison_vmalloc((void *)vas[area]->va_start, | |
3571 | sizes[area]); | |
3572 | } | |
3573 | ||
ca23e405 | 3574 | /* insert all vm's */ |
e36176be URS |
3575 | spin_lock(&vmap_area_lock); |
3576 | for (area = 0; area < nr_vms; area++) { | |
3577 | insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list); | |
3578 | ||
3579 | setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC, | |
3645cb4a | 3580 | pcpu_get_vm_areas); |
e36176be URS |
3581 | } |
3582 | spin_unlock(&vmap_area_lock); | |
ca23e405 TH |
3583 | |
3584 | kfree(vas); | |
3585 | return vms; | |
3586 | ||
68ad4a33 | 3587 | recovery: |
e36176be URS |
3588 | /* |
3589 | * Remove previously allocated areas. There is no | |
3590 | * need in removing these areas from the busy tree, | |
3591 | * because they are inserted only on the final step | |
3592 | * and when pcpu_get_vm_areas() is success. | |
3593 | */ | |
68ad4a33 | 3594 | while (area--) { |
253a496d DA |
3595 | orig_start = vas[area]->va_start; |
3596 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
3597 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
3598 | &free_vmap_area_list); | |
9c801f61 URS |
3599 | if (va) |
3600 | kasan_release_vmalloc(orig_start, orig_end, | |
3601 | va->va_start, va->va_end); | |
68ad4a33 URS |
3602 | vas[area] = NULL; |
3603 | } | |
3604 | ||
3605 | overflow: | |
e36176be | 3606 | spin_unlock(&free_vmap_area_lock); |
68ad4a33 URS |
3607 | if (!purged) { |
3608 | purge_vmap_area_lazy(); | |
3609 | purged = true; | |
3610 | ||
3611 | /* Before "retry", check if we recover. */ | |
3612 | for (area = 0; area < nr_vms; area++) { | |
3613 | if (vas[area]) | |
3614 | continue; | |
3615 | ||
3616 | vas[area] = kmem_cache_zalloc( | |
3617 | vmap_area_cachep, GFP_KERNEL); | |
3618 | if (!vas[area]) | |
3619 | goto err_free; | |
3620 | } | |
3621 | ||
3622 | goto retry; | |
3623 | } | |
3624 | ||
ca23e405 TH |
3625 | err_free: |
3626 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3627 | if (vas[area]) |
3628 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3629 | ||
f1db7afd | 3630 | kfree(vms[area]); |
ca23e405 | 3631 | } |
f1db7afd | 3632 | err_free2: |
ca23e405 TH |
3633 | kfree(vas); |
3634 | kfree(vms); | |
3635 | return NULL; | |
253a496d DA |
3636 | |
3637 | err_free_shadow: | |
3638 | spin_lock(&free_vmap_area_lock); | |
3639 | /* | |
3640 | * We release all the vmalloc shadows, even the ones for regions that | |
3641 | * hadn't been successfully added. This relies on kasan_release_vmalloc | |
3642 | * being able to tolerate this case. | |
3643 | */ | |
3644 | for (area = 0; area < nr_vms; area++) { | |
3645 | orig_start = vas[area]->va_start; | |
3646 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
3647 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
3648 | &free_vmap_area_list); | |
9c801f61 URS |
3649 | if (va) |
3650 | kasan_release_vmalloc(orig_start, orig_end, | |
3651 | va->va_start, va->va_end); | |
253a496d DA |
3652 | vas[area] = NULL; |
3653 | kfree(vms[area]); | |
3654 | } | |
3655 | spin_unlock(&free_vmap_area_lock); | |
3656 | kfree(vas); | |
3657 | kfree(vms); | |
3658 | return NULL; | |
ca23e405 TH |
3659 | } |
3660 | ||
3661 | /** | |
3662 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3663 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3664 | * @nr_vms: the number of allocated areas | |
3665 | * | |
3666 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3667 | */ | |
3668 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3669 | { | |
3670 | int i; | |
3671 | ||
3672 | for (i = 0; i < nr_vms; i++) | |
3673 | free_vm_area(vms[i]); | |
3674 | kfree(vms); | |
3675 | } | |
4f8b02b4 | 3676 | #endif /* CONFIG_SMP */ |
a10aa579 | 3677 | |
5bb1bb35 | 3678 | #ifdef CONFIG_PRINTK |
98f18083 PM |
3679 | bool vmalloc_dump_obj(void *object) |
3680 | { | |
3681 | struct vm_struct *vm; | |
3682 | void *objp = (void *)PAGE_ALIGN((unsigned long)object); | |
3683 | ||
3684 | vm = find_vm_area(objp); | |
3685 | if (!vm) | |
3686 | return false; | |
bd34dcd4 PM |
3687 | pr_cont(" %u-page vmalloc region starting at %#lx allocated at %pS\n", |
3688 | vm->nr_pages, (unsigned long)vm->addr, vm->caller); | |
98f18083 PM |
3689 | return true; |
3690 | } | |
5bb1bb35 | 3691 | #endif |
98f18083 | 3692 | |
a10aa579 CL |
3693 | #ifdef CONFIG_PROC_FS |
3694 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e36176be | 3695 | __acquires(&vmap_purge_lock) |
d4033afd | 3696 | __acquires(&vmap_area_lock) |
a10aa579 | 3697 | { |
e36176be | 3698 | mutex_lock(&vmap_purge_lock); |
d4033afd | 3699 | spin_lock(&vmap_area_lock); |
e36176be | 3700 | |
3f500069 | 3701 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3702 | } |
3703 | ||
3704 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3705 | { | |
3f500069 | 3706 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3707 | } |
3708 | ||
3709 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 3710 | __releases(&vmap_area_lock) |
0a7dd4e9 | 3711 | __releases(&vmap_purge_lock) |
a10aa579 | 3712 | { |
d4033afd | 3713 | spin_unlock(&vmap_area_lock); |
0a7dd4e9 | 3714 | mutex_unlock(&vmap_purge_lock); |
a10aa579 CL |
3715 | } |
3716 | ||
a47a126a ED |
3717 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3718 | { | |
e5adfffc | 3719 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3720 | unsigned int nr, *counters = m->private; |
3721 | ||
3722 | if (!counters) | |
3723 | return; | |
3724 | ||
af12346c WL |
3725 | if (v->flags & VM_UNINITIALIZED) |
3726 | return; | |
7e5b528b DV |
3727 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3728 | smp_rmb(); | |
af12346c | 3729 | |
a47a126a ED |
3730 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3731 | ||
3732 | for (nr = 0; nr < v->nr_pages; nr++) | |
3733 | counters[page_to_nid(v->pages[nr])]++; | |
3734 | ||
3735 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3736 | if (counters[nr]) | |
3737 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3738 | } | |
3739 | } | |
3740 | ||
dd3b8353 URS |
3741 | static void show_purge_info(struct seq_file *m) |
3742 | { | |
dd3b8353 URS |
3743 | struct vmap_area *va; |
3744 | ||
96e2db45 URS |
3745 | spin_lock(&purge_vmap_area_lock); |
3746 | list_for_each_entry(va, &purge_vmap_area_list, list) { | |
dd3b8353 URS |
3747 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", |
3748 | (void *)va->va_start, (void *)va->va_end, | |
3749 | va->va_end - va->va_start); | |
3750 | } | |
96e2db45 | 3751 | spin_unlock(&purge_vmap_area_lock); |
dd3b8353 URS |
3752 | } |
3753 | ||
a10aa579 CL |
3754 | static int s_show(struct seq_file *m, void *p) |
3755 | { | |
3f500069 | 3756 | struct vmap_area *va; |
d4033afd JK |
3757 | struct vm_struct *v; |
3758 | ||
3f500069 | 3759 | va = list_entry(p, struct vmap_area, list); |
3760 | ||
c2ce8c14 | 3761 | /* |
688fcbfc PL |
3762 | * s_show can encounter race with remove_vm_area, !vm on behalf |
3763 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 3764 | */ |
688fcbfc | 3765 | if (!va->vm) { |
dd3b8353 | 3766 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 3767 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 3768 | va->va_end - va->va_start); |
78c72746 | 3769 | |
d4033afd | 3770 | return 0; |
78c72746 | 3771 | } |
d4033afd JK |
3772 | |
3773 | v = va->vm; | |
a10aa579 | 3774 | |
45ec1690 | 3775 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3776 | v->addr, v->addr + v->size, v->size); |
3777 | ||
62c70bce JP |
3778 | if (v->caller) |
3779 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3780 | |
a10aa579 CL |
3781 | if (v->nr_pages) |
3782 | seq_printf(m, " pages=%d", v->nr_pages); | |
3783 | ||
3784 | if (v->phys_addr) | |
199eaa05 | 3785 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3786 | |
3787 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3788 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3789 | |
3790 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3791 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3792 | |
3793 | if (v->flags & VM_MAP) | |
f4527c90 | 3794 | seq_puts(m, " vmap"); |
a10aa579 CL |
3795 | |
3796 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3797 | seq_puts(m, " user"); |
a10aa579 | 3798 | |
fe9041c2 CH |
3799 | if (v->flags & VM_DMA_COHERENT) |
3800 | seq_puts(m, " dma-coherent"); | |
3801 | ||
244d63ee | 3802 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3803 | seq_puts(m, " vpages"); |
a10aa579 | 3804 | |
a47a126a | 3805 | show_numa_info(m, v); |
a10aa579 | 3806 | seq_putc(m, '\n'); |
dd3b8353 URS |
3807 | |
3808 | /* | |
96e2db45 | 3809 | * As a final step, dump "unpurged" areas. |
dd3b8353 URS |
3810 | */ |
3811 | if (list_is_last(&va->list, &vmap_area_list)) | |
3812 | show_purge_info(m); | |
3813 | ||
a10aa579 CL |
3814 | return 0; |
3815 | } | |
3816 | ||
5f6a6a9c | 3817 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3818 | .start = s_start, |
3819 | .next = s_next, | |
3820 | .stop = s_stop, | |
3821 | .show = s_show, | |
3822 | }; | |
5f6a6a9c | 3823 | |
5f6a6a9c AD |
3824 | static int __init proc_vmalloc_init(void) |
3825 | { | |
fddda2b7 | 3826 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3827 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3828 | &vmalloc_op, |
3829 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3830 | else |
0825a6f9 | 3831 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3832 | return 0; |
3833 | } | |
3834 | module_init(proc_vmalloc_init); | |
db3808c1 | 3835 | |
a10aa579 | 3836 | #endif |