]>
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 | |
121e6f32 NP |
45 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC |
46 | static bool __ro_after_init vmap_allow_huge = true; | |
47 | ||
48 | static int __init set_nohugevmalloc(char *str) | |
49 | { | |
50 | vmap_allow_huge = false; | |
51 | return 0; | |
52 | } | |
53 | early_param("nohugevmalloc", set_nohugevmalloc); | |
54 | #else /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */ | |
55 | static const bool vmap_allow_huge = false; | |
56 | #endif /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */ | |
57 | ||
186525bd IM |
58 | bool is_vmalloc_addr(const void *x) |
59 | { | |
60 | unsigned long addr = (unsigned long)x; | |
61 | ||
62 | return addr >= VMALLOC_START && addr < VMALLOC_END; | |
63 | } | |
64 | EXPORT_SYMBOL(is_vmalloc_addr); | |
65 | ||
32fcfd40 AV |
66 | struct vfree_deferred { |
67 | struct llist_head list; | |
68 | struct work_struct wq; | |
69 | }; | |
70 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
71 | ||
72 | static void __vunmap(const void *, int); | |
73 | ||
74 | static void free_work(struct work_struct *w) | |
75 | { | |
76 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
77 | struct llist_node *t, *llnode; |
78 | ||
79 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
80 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
81 | } |
82 | ||
db64fe02 | 83 | /*** Page table manipulation functions ***/ |
5e9e3d77 NP |
84 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
85 | phys_addr_t phys_addr, pgprot_t prot, | |
86 | pgtbl_mod_mask *mask) | |
87 | { | |
88 | pte_t *pte; | |
89 | u64 pfn; | |
90 | ||
91 | pfn = phys_addr >> PAGE_SHIFT; | |
92 | pte = pte_alloc_kernel_track(pmd, addr, mask); | |
93 | if (!pte) | |
94 | return -ENOMEM; | |
95 | do { | |
96 | BUG_ON(!pte_none(*pte)); | |
97 | set_pte_at(&init_mm, addr, pte, pfn_pte(pfn, prot)); | |
98 | pfn++; | |
99 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
100 | *mask |= PGTBL_PTE_MODIFIED; | |
101 | return 0; | |
102 | } | |
103 | ||
104 | static int vmap_try_huge_pmd(pmd_t *pmd, unsigned long addr, unsigned long end, | |
105 | phys_addr_t phys_addr, pgprot_t prot, | |
106 | unsigned int max_page_shift) | |
107 | { | |
108 | if (max_page_shift < PMD_SHIFT) | |
109 | return 0; | |
110 | ||
111 | if (!arch_vmap_pmd_supported(prot)) | |
112 | return 0; | |
113 | ||
114 | if ((end - addr) != PMD_SIZE) | |
115 | return 0; | |
116 | ||
117 | if (!IS_ALIGNED(addr, PMD_SIZE)) | |
118 | return 0; | |
119 | ||
120 | if (!IS_ALIGNED(phys_addr, PMD_SIZE)) | |
121 | return 0; | |
122 | ||
123 | if (pmd_present(*pmd) && !pmd_free_pte_page(pmd, addr)) | |
124 | return 0; | |
125 | ||
126 | return pmd_set_huge(pmd, phys_addr, prot); | |
127 | } | |
128 | ||
129 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, | |
130 | phys_addr_t phys_addr, pgprot_t prot, | |
131 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
132 | { | |
133 | pmd_t *pmd; | |
134 | unsigned long next; | |
135 | ||
136 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); | |
137 | if (!pmd) | |
138 | return -ENOMEM; | |
139 | do { | |
140 | next = pmd_addr_end(addr, end); | |
141 | ||
142 | if (vmap_try_huge_pmd(pmd, addr, next, phys_addr, prot, | |
143 | max_page_shift)) { | |
144 | *mask |= PGTBL_PMD_MODIFIED; | |
145 | continue; | |
146 | } | |
147 | ||
148 | if (vmap_pte_range(pmd, addr, next, phys_addr, prot, mask)) | |
149 | return -ENOMEM; | |
150 | } while (pmd++, phys_addr += (next - addr), addr = next, addr != end); | |
151 | return 0; | |
152 | } | |
153 | ||
154 | static int vmap_try_huge_pud(pud_t *pud, unsigned long addr, unsigned long end, | |
155 | phys_addr_t phys_addr, pgprot_t prot, | |
156 | unsigned int max_page_shift) | |
157 | { | |
158 | if (max_page_shift < PUD_SHIFT) | |
159 | return 0; | |
160 | ||
161 | if (!arch_vmap_pud_supported(prot)) | |
162 | return 0; | |
163 | ||
164 | if ((end - addr) != PUD_SIZE) | |
165 | return 0; | |
166 | ||
167 | if (!IS_ALIGNED(addr, PUD_SIZE)) | |
168 | return 0; | |
169 | ||
170 | if (!IS_ALIGNED(phys_addr, PUD_SIZE)) | |
171 | return 0; | |
172 | ||
173 | if (pud_present(*pud) && !pud_free_pmd_page(pud, addr)) | |
174 | return 0; | |
175 | ||
176 | return pud_set_huge(pud, phys_addr, prot); | |
177 | } | |
178 | ||
179 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, | |
180 | phys_addr_t phys_addr, pgprot_t prot, | |
181 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
182 | { | |
183 | pud_t *pud; | |
184 | unsigned long next; | |
185 | ||
186 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); | |
187 | if (!pud) | |
188 | return -ENOMEM; | |
189 | do { | |
190 | next = pud_addr_end(addr, end); | |
191 | ||
192 | if (vmap_try_huge_pud(pud, addr, next, phys_addr, prot, | |
193 | max_page_shift)) { | |
194 | *mask |= PGTBL_PUD_MODIFIED; | |
195 | continue; | |
196 | } | |
197 | ||
198 | if (vmap_pmd_range(pud, addr, next, phys_addr, prot, | |
199 | max_page_shift, mask)) | |
200 | return -ENOMEM; | |
201 | } while (pud++, phys_addr += (next - addr), addr = next, addr != end); | |
202 | return 0; | |
203 | } | |
204 | ||
205 | static int vmap_try_huge_p4d(p4d_t *p4d, unsigned long addr, unsigned long end, | |
206 | phys_addr_t phys_addr, pgprot_t prot, | |
207 | unsigned int max_page_shift) | |
208 | { | |
209 | if (max_page_shift < P4D_SHIFT) | |
210 | return 0; | |
211 | ||
212 | if (!arch_vmap_p4d_supported(prot)) | |
213 | return 0; | |
214 | ||
215 | if ((end - addr) != P4D_SIZE) | |
216 | return 0; | |
217 | ||
218 | if (!IS_ALIGNED(addr, P4D_SIZE)) | |
219 | return 0; | |
220 | ||
221 | if (!IS_ALIGNED(phys_addr, P4D_SIZE)) | |
222 | return 0; | |
223 | ||
224 | if (p4d_present(*p4d) && !p4d_free_pud_page(p4d, addr)) | |
225 | return 0; | |
226 | ||
227 | return p4d_set_huge(p4d, phys_addr, prot); | |
228 | } | |
229 | ||
230 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, | |
231 | phys_addr_t phys_addr, pgprot_t prot, | |
232 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
233 | { | |
234 | p4d_t *p4d; | |
235 | unsigned long next; | |
236 | ||
237 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); | |
238 | if (!p4d) | |
239 | return -ENOMEM; | |
240 | do { | |
241 | next = p4d_addr_end(addr, end); | |
242 | ||
243 | if (vmap_try_huge_p4d(p4d, addr, next, phys_addr, prot, | |
244 | max_page_shift)) { | |
245 | *mask |= PGTBL_P4D_MODIFIED; | |
246 | continue; | |
247 | } | |
248 | ||
249 | if (vmap_pud_range(p4d, addr, next, phys_addr, prot, | |
250 | max_page_shift, mask)) | |
251 | return -ENOMEM; | |
252 | } while (p4d++, phys_addr += (next - addr), addr = next, addr != end); | |
253 | return 0; | |
254 | } | |
255 | ||
5d87510d | 256 | static int vmap_range_noflush(unsigned long addr, unsigned long end, |
5e9e3d77 NP |
257 | phys_addr_t phys_addr, pgprot_t prot, |
258 | unsigned int max_page_shift) | |
259 | { | |
260 | pgd_t *pgd; | |
261 | unsigned long start; | |
262 | unsigned long next; | |
263 | int err; | |
264 | pgtbl_mod_mask mask = 0; | |
265 | ||
266 | might_sleep(); | |
267 | BUG_ON(addr >= end); | |
268 | ||
269 | start = addr; | |
270 | pgd = pgd_offset_k(addr); | |
271 | do { | |
272 | next = pgd_addr_end(addr, end); | |
273 | err = vmap_p4d_range(pgd, addr, next, phys_addr, prot, | |
274 | max_page_shift, &mask); | |
275 | if (err) | |
276 | break; | |
277 | } while (pgd++, phys_addr += (next - addr), addr = next, addr != end); | |
278 | ||
5e9e3d77 NP |
279 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
280 | arch_sync_kernel_mappings(start, end); | |
281 | ||
282 | return err; | |
283 | } | |
b221385b | 284 | |
5d87510d NP |
285 | int vmap_range(unsigned long addr, unsigned long end, |
286 | phys_addr_t phys_addr, pgprot_t prot, | |
287 | unsigned int max_page_shift) | |
288 | { | |
289 | int err; | |
290 | ||
291 | err = vmap_range_noflush(addr, end, phys_addr, prot, max_page_shift); | |
292 | flush_cache_vmap(addr, end); | |
293 | ||
294 | return err; | |
295 | } | |
296 | ||
2ba3e694 JR |
297 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
298 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
299 | { |
300 | pte_t *pte; | |
301 | ||
302 | pte = pte_offset_kernel(pmd, addr); | |
303 | do { | |
304 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
305 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
306 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
2ba3e694 | 307 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
308 | } |
309 | ||
2ba3e694 JR |
310 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, |
311 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
312 | { |
313 | pmd_t *pmd; | |
314 | unsigned long next; | |
2ba3e694 | 315 | int cleared; |
1da177e4 LT |
316 | |
317 | pmd = pmd_offset(pud, addr); | |
318 | do { | |
319 | next = pmd_addr_end(addr, end); | |
2ba3e694 JR |
320 | |
321 | cleared = pmd_clear_huge(pmd); | |
322 | if (cleared || pmd_bad(*pmd)) | |
323 | *mask |= PGTBL_PMD_MODIFIED; | |
324 | ||
325 | if (cleared) | |
b9820d8f | 326 | continue; |
1da177e4 LT |
327 | if (pmd_none_or_clear_bad(pmd)) |
328 | continue; | |
2ba3e694 | 329 | vunmap_pte_range(pmd, addr, next, mask); |
e47110e9 AK |
330 | |
331 | cond_resched(); | |
1da177e4 LT |
332 | } while (pmd++, addr = next, addr != end); |
333 | } | |
334 | ||
2ba3e694 JR |
335 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, |
336 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
337 | { |
338 | pud_t *pud; | |
339 | unsigned long next; | |
2ba3e694 | 340 | int cleared; |
1da177e4 | 341 | |
c2febafc | 342 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
343 | do { |
344 | next = pud_addr_end(addr, end); | |
2ba3e694 JR |
345 | |
346 | cleared = pud_clear_huge(pud); | |
347 | if (cleared || pud_bad(*pud)) | |
348 | *mask |= PGTBL_PUD_MODIFIED; | |
349 | ||
350 | if (cleared) | |
b9820d8f | 351 | continue; |
1da177e4 LT |
352 | if (pud_none_or_clear_bad(pud)) |
353 | continue; | |
2ba3e694 | 354 | vunmap_pmd_range(pud, addr, next, mask); |
1da177e4 LT |
355 | } while (pud++, addr = next, addr != end); |
356 | } | |
357 | ||
2ba3e694 JR |
358 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, |
359 | pgtbl_mod_mask *mask) | |
c2febafc KS |
360 | { |
361 | p4d_t *p4d; | |
362 | unsigned long next; | |
2ba3e694 | 363 | int cleared; |
c2febafc KS |
364 | |
365 | p4d = p4d_offset(pgd, addr); | |
366 | do { | |
367 | next = p4d_addr_end(addr, end); | |
2ba3e694 JR |
368 | |
369 | cleared = p4d_clear_huge(p4d); | |
370 | if (cleared || p4d_bad(*p4d)) | |
371 | *mask |= PGTBL_P4D_MODIFIED; | |
372 | ||
373 | if (cleared) | |
c2febafc KS |
374 | continue; |
375 | if (p4d_none_or_clear_bad(p4d)) | |
376 | continue; | |
2ba3e694 | 377 | vunmap_pud_range(p4d, addr, next, mask); |
c2febafc KS |
378 | } while (p4d++, addr = next, addr != end); |
379 | } | |
380 | ||
4ad0ae8c NP |
381 | /* |
382 | * vunmap_range_noflush is similar to vunmap_range, but does not | |
383 | * flush caches or TLBs. | |
b521c43f | 384 | * |
4ad0ae8c NP |
385 | * The caller is responsible for calling flush_cache_vmap() before calling |
386 | * this function, and flush_tlb_kernel_range after it has returned | |
387 | * successfully (and before the addresses are expected to cause a page fault | |
388 | * or be re-mapped for something else, if TLB flushes are being delayed or | |
389 | * coalesced). | |
b521c43f | 390 | * |
4ad0ae8c | 391 | * This is an internal function only. Do not use outside mm/. |
b521c43f | 392 | */ |
4ad0ae8c | 393 | void vunmap_range_noflush(unsigned long start, unsigned long end) |
1da177e4 | 394 | { |
1da177e4 | 395 | unsigned long next; |
b521c43f | 396 | pgd_t *pgd; |
2ba3e694 JR |
397 | unsigned long addr = start; |
398 | pgtbl_mod_mask mask = 0; | |
1da177e4 LT |
399 | |
400 | BUG_ON(addr >= end); | |
401 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
402 | do { |
403 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
404 | if (pgd_bad(*pgd)) |
405 | mask |= PGTBL_PGD_MODIFIED; | |
1da177e4 LT |
406 | if (pgd_none_or_clear_bad(pgd)) |
407 | continue; | |
2ba3e694 | 408 | vunmap_p4d_range(pgd, addr, next, &mask); |
1da177e4 | 409 | } while (pgd++, addr = next, addr != end); |
2ba3e694 JR |
410 | |
411 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) | |
412 | arch_sync_kernel_mappings(start, end); | |
1da177e4 LT |
413 | } |
414 | ||
4ad0ae8c NP |
415 | /** |
416 | * vunmap_range - unmap kernel virtual addresses | |
417 | * @addr: start of the VM area to unmap | |
418 | * @end: end of the VM area to unmap (non-inclusive) | |
419 | * | |
420 | * Clears any present PTEs in the virtual address range, flushes TLBs and | |
421 | * caches. Any subsequent access to the address before it has been re-mapped | |
422 | * is a kernel bug. | |
423 | */ | |
424 | void vunmap_range(unsigned long addr, unsigned long end) | |
425 | { | |
426 | flush_cache_vunmap(addr, end); | |
427 | vunmap_range_noflush(addr, end); | |
428 | flush_tlb_kernel_range(addr, end); | |
429 | } | |
430 | ||
0a264884 | 431 | static int vmap_pages_pte_range(pmd_t *pmd, unsigned long addr, |
2ba3e694 JR |
432 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
433 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
434 | { |
435 | pte_t *pte; | |
436 | ||
db64fe02 NP |
437 | /* |
438 | * nr is a running index into the array which helps higher level | |
439 | * callers keep track of where we're up to. | |
440 | */ | |
441 | ||
2ba3e694 | 442 | pte = pte_alloc_kernel_track(pmd, addr, mask); |
1da177e4 LT |
443 | if (!pte) |
444 | return -ENOMEM; | |
445 | do { | |
db64fe02 NP |
446 | struct page *page = pages[*nr]; |
447 | ||
448 | if (WARN_ON(!pte_none(*pte))) | |
449 | return -EBUSY; | |
450 | if (WARN_ON(!page)) | |
1da177e4 LT |
451 | return -ENOMEM; |
452 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 453 | (*nr)++; |
1da177e4 | 454 | } while (pte++, addr += PAGE_SIZE, addr != end); |
2ba3e694 | 455 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
456 | return 0; |
457 | } | |
458 | ||
0a264884 | 459 | static int vmap_pages_pmd_range(pud_t *pud, unsigned long addr, |
2ba3e694 JR |
460 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
461 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
462 | { |
463 | pmd_t *pmd; | |
464 | unsigned long next; | |
465 | ||
2ba3e694 | 466 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); |
1da177e4 LT |
467 | if (!pmd) |
468 | return -ENOMEM; | |
469 | do { | |
470 | next = pmd_addr_end(addr, end); | |
0a264884 | 471 | if (vmap_pages_pte_range(pmd, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
472 | return -ENOMEM; |
473 | } while (pmd++, addr = next, addr != end); | |
474 | return 0; | |
475 | } | |
476 | ||
0a264884 | 477 | static int vmap_pages_pud_range(p4d_t *p4d, unsigned long addr, |
2ba3e694 JR |
478 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
479 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
480 | { |
481 | pud_t *pud; | |
482 | unsigned long next; | |
483 | ||
2ba3e694 | 484 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); |
1da177e4 LT |
485 | if (!pud) |
486 | return -ENOMEM; | |
487 | do { | |
488 | next = pud_addr_end(addr, end); | |
0a264884 | 489 | if (vmap_pages_pmd_range(pud, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
490 | return -ENOMEM; |
491 | } while (pud++, addr = next, addr != end); | |
492 | return 0; | |
493 | } | |
494 | ||
0a264884 | 495 | static int vmap_pages_p4d_range(pgd_t *pgd, unsigned long addr, |
2ba3e694 JR |
496 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
497 | pgtbl_mod_mask *mask) | |
c2febafc KS |
498 | { |
499 | p4d_t *p4d; | |
500 | unsigned long next; | |
501 | ||
2ba3e694 | 502 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); |
c2febafc KS |
503 | if (!p4d) |
504 | return -ENOMEM; | |
505 | do { | |
506 | next = p4d_addr_end(addr, end); | |
0a264884 | 507 | if (vmap_pages_pud_range(p4d, addr, next, prot, pages, nr, mask)) |
c2febafc KS |
508 | return -ENOMEM; |
509 | } while (p4d++, addr = next, addr != end); | |
510 | return 0; | |
511 | } | |
512 | ||
121e6f32 NP |
513 | static int vmap_small_pages_range_noflush(unsigned long addr, unsigned long end, |
514 | pgprot_t prot, struct page **pages) | |
1da177e4 | 515 | { |
2ba3e694 | 516 | unsigned long start = addr; |
b521c43f | 517 | pgd_t *pgd; |
121e6f32 | 518 | unsigned long next; |
db64fe02 NP |
519 | int err = 0; |
520 | int nr = 0; | |
2ba3e694 | 521 | pgtbl_mod_mask mask = 0; |
1da177e4 LT |
522 | |
523 | BUG_ON(addr >= end); | |
524 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
525 | do { |
526 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
527 | if (pgd_bad(*pgd)) |
528 | mask |= PGTBL_PGD_MODIFIED; | |
0a264884 | 529 | err = vmap_pages_p4d_range(pgd, addr, next, prot, pages, &nr, &mask); |
1da177e4 | 530 | if (err) |
bf88c8c8 | 531 | return err; |
1da177e4 | 532 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 533 | |
2ba3e694 JR |
534 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
535 | arch_sync_kernel_mappings(start, end); | |
536 | ||
60bb4465 | 537 | return 0; |
1da177e4 LT |
538 | } |
539 | ||
b67177ec NP |
540 | /* |
541 | * vmap_pages_range_noflush is similar to vmap_pages_range, but does not | |
542 | * flush caches. | |
543 | * | |
544 | * The caller is responsible for calling flush_cache_vmap() after this | |
545 | * function returns successfully and before the addresses are accessed. | |
546 | * | |
547 | * This is an internal function only. Do not use outside mm/. | |
548 | */ | |
549 | int vmap_pages_range_noflush(unsigned long addr, unsigned long end, | |
121e6f32 NP |
550 | pgprot_t prot, struct page **pages, unsigned int page_shift) |
551 | { | |
552 | unsigned int i, nr = (end - addr) >> PAGE_SHIFT; | |
553 | ||
554 | WARN_ON(page_shift < PAGE_SHIFT); | |
555 | ||
556 | if (!IS_ENABLED(CONFIG_HAVE_ARCH_HUGE_VMALLOC) || | |
557 | page_shift == PAGE_SHIFT) | |
558 | return vmap_small_pages_range_noflush(addr, end, prot, pages); | |
559 | ||
560 | for (i = 0; i < nr; i += 1U << (page_shift - PAGE_SHIFT)) { | |
561 | int err; | |
562 | ||
563 | err = vmap_range_noflush(addr, addr + (1UL << page_shift), | |
564 | __pa(page_address(pages[i])), prot, | |
565 | page_shift); | |
566 | if (err) | |
567 | return err; | |
568 | ||
569 | addr += 1UL << page_shift; | |
570 | } | |
571 | ||
572 | return 0; | |
573 | } | |
574 | ||
121e6f32 | 575 | /** |
b67177ec | 576 | * vmap_pages_range - map pages to a kernel virtual address |
121e6f32 | 577 | * @addr: start of the VM area to map |
b67177ec | 578 | * @end: end of the VM area to map (non-inclusive) |
121e6f32 | 579 | * @prot: page protection flags to use |
b67177ec NP |
580 | * @pages: pages to map (always PAGE_SIZE pages) |
581 | * @page_shift: maximum shift that the pages may be mapped with, @pages must | |
582 | * be aligned and contiguous up to at least this shift. | |
121e6f32 NP |
583 | * |
584 | * RETURNS: | |
585 | * 0 on success, -errno on failure. | |
586 | */ | |
b67177ec NP |
587 | static int vmap_pages_range(unsigned long addr, unsigned long end, |
588 | pgprot_t prot, struct page **pages, unsigned int page_shift) | |
8fc48985 | 589 | { |
b67177ec | 590 | int err; |
8fc48985 | 591 | |
b67177ec NP |
592 | err = vmap_pages_range_noflush(addr, end, prot, pages, page_shift); |
593 | flush_cache_vmap(addr, end); | |
594 | return err; | |
8fc48985 TH |
595 | } |
596 | ||
81ac3ad9 | 597 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
598 | { |
599 | /* | |
ab4f2ee1 | 600 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
601 | * and fall back on vmalloc() if that fails. Others |
602 | * just put it in the vmalloc space. | |
603 | */ | |
604 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
605 | unsigned long addr = (unsigned long)x; | |
606 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
607 | return 1; | |
608 | #endif | |
609 | return is_vmalloc_addr(x); | |
610 | } | |
611 | ||
48667e7a | 612 | /* |
c0eb315a NP |
613 | * Walk a vmap address to the struct page it maps. Huge vmap mappings will |
614 | * return the tail page that corresponds to the base page address, which | |
615 | * matches small vmap mappings. | |
48667e7a | 616 | */ |
add688fb | 617 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
618 | { |
619 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 620 | struct page *page = NULL; |
48667e7a | 621 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
622 | p4d_t *p4d; |
623 | pud_t *pud; | |
624 | pmd_t *pmd; | |
625 | pte_t *ptep, pte; | |
48667e7a | 626 | |
7aa413de IM |
627 | /* |
628 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
629 | * architectures that do not vmalloc module space | |
630 | */ | |
73bdf0a6 | 631 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 632 | |
c2febafc KS |
633 | if (pgd_none(*pgd)) |
634 | return NULL; | |
c0eb315a NP |
635 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
636 | return NULL; /* XXX: no allowance for huge pgd */ | |
637 | if (WARN_ON_ONCE(pgd_bad(*pgd))) | |
638 | return NULL; | |
639 | ||
c2febafc KS |
640 | p4d = p4d_offset(pgd, addr); |
641 | if (p4d_none(*p4d)) | |
642 | return NULL; | |
c0eb315a NP |
643 | if (p4d_leaf(*p4d)) |
644 | return p4d_page(*p4d) + ((addr & ~P4D_MASK) >> PAGE_SHIFT); | |
645 | if (WARN_ON_ONCE(p4d_bad(*p4d))) | |
646 | return NULL; | |
029c54b0 | 647 | |
c0eb315a NP |
648 | pud = pud_offset(p4d, addr); |
649 | if (pud_none(*pud)) | |
650 | return NULL; | |
651 | if (pud_leaf(*pud)) | |
652 | return pud_page(*pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
653 | if (WARN_ON_ONCE(pud_bad(*pud))) | |
c2febafc | 654 | return NULL; |
c0eb315a | 655 | |
c2febafc | 656 | pmd = pmd_offset(pud, addr); |
c0eb315a NP |
657 | if (pmd_none(*pmd)) |
658 | return NULL; | |
659 | if (pmd_leaf(*pmd)) | |
660 | return pmd_page(*pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
661 | if (WARN_ON_ONCE(pmd_bad(*pmd))) | |
c2febafc KS |
662 | return NULL; |
663 | ||
664 | ptep = pte_offset_map(pmd, addr); | |
665 | pte = *ptep; | |
666 | if (pte_present(pte)) | |
667 | page = pte_page(pte); | |
668 | pte_unmap(ptep); | |
c0eb315a | 669 | |
add688fb | 670 | return page; |
48667e7a | 671 | } |
add688fb | 672 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
673 | |
674 | /* | |
add688fb | 675 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 676 | */ |
add688fb | 677 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 678 | { |
add688fb | 679 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 680 | } |
add688fb | 681 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 682 | |
db64fe02 NP |
683 | |
684 | /*** Global kva allocator ***/ | |
685 | ||
bb850f4d | 686 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 687 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 688 | |
db64fe02 | 689 | |
db64fe02 | 690 | static DEFINE_SPINLOCK(vmap_area_lock); |
e36176be | 691 | static DEFINE_SPINLOCK(free_vmap_area_lock); |
f1c4069e JK |
692 | /* Export for kexec only */ |
693 | LIST_HEAD(vmap_area_list); | |
89699605 | 694 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 695 | static bool vmap_initialized __read_mostly; |
89699605 | 696 | |
96e2db45 URS |
697 | static struct rb_root purge_vmap_area_root = RB_ROOT; |
698 | static LIST_HEAD(purge_vmap_area_list); | |
699 | static DEFINE_SPINLOCK(purge_vmap_area_lock); | |
700 | ||
68ad4a33 URS |
701 | /* |
702 | * This kmem_cache is used for vmap_area objects. Instead of | |
703 | * allocating from slab we reuse an object from this cache to | |
704 | * make things faster. Especially in "no edge" splitting of | |
705 | * free block. | |
706 | */ | |
707 | static struct kmem_cache *vmap_area_cachep; | |
708 | ||
709 | /* | |
710 | * This linked list is used in pair with free_vmap_area_root. | |
711 | * It gives O(1) access to prev/next to perform fast coalescing. | |
712 | */ | |
713 | static LIST_HEAD(free_vmap_area_list); | |
714 | ||
715 | /* | |
716 | * This augment red-black tree represents the free vmap space. | |
717 | * All vmap_area objects in this tree are sorted by va->va_start | |
718 | * address. It is used for allocation and merging when a vmap | |
719 | * object is released. | |
720 | * | |
721 | * Each vmap_area node contains a maximum available free block | |
722 | * of its sub-tree, right or left. Therefore it is possible to | |
723 | * find a lowest match of free area. | |
724 | */ | |
725 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
726 | ||
82dd23e8 URS |
727 | /* |
728 | * Preload a CPU with one object for "no edge" split case. The | |
729 | * aim is to get rid of allocations from the atomic context, thus | |
730 | * to use more permissive allocation masks. | |
731 | */ | |
732 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
733 | ||
68ad4a33 URS |
734 | static __always_inline unsigned long |
735 | va_size(struct vmap_area *va) | |
736 | { | |
737 | return (va->va_end - va->va_start); | |
738 | } | |
739 | ||
740 | static __always_inline unsigned long | |
741 | get_subtree_max_size(struct rb_node *node) | |
742 | { | |
743 | struct vmap_area *va; | |
744 | ||
745 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
746 | return va ? va->subtree_max_size : 0; | |
747 | } | |
89699605 | 748 | |
68ad4a33 URS |
749 | /* |
750 | * Gets called when remove the node and rotate. | |
751 | */ | |
752 | static __always_inline unsigned long | |
753 | compute_subtree_max_size(struct vmap_area *va) | |
754 | { | |
755 | return max3(va_size(va), | |
756 | get_subtree_max_size(va->rb_node.rb_left), | |
757 | get_subtree_max_size(va->rb_node.rb_right)); | |
758 | } | |
759 | ||
315cc066 ML |
760 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
761 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 URS |
762 | |
763 | static void purge_vmap_area_lazy(void); | |
764 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
765 | static unsigned long lazy_max_pages(void); | |
db64fe02 | 766 | |
97105f0a RG |
767 | static atomic_long_t nr_vmalloc_pages; |
768 | ||
769 | unsigned long vmalloc_nr_pages(void) | |
770 | { | |
771 | return atomic_long_read(&nr_vmalloc_pages); | |
772 | } | |
773 | ||
db64fe02 | 774 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 775 | { |
db64fe02 NP |
776 | struct rb_node *n = vmap_area_root.rb_node; |
777 | ||
778 | while (n) { | |
779 | struct vmap_area *va; | |
780 | ||
781 | va = rb_entry(n, struct vmap_area, rb_node); | |
782 | if (addr < va->va_start) | |
783 | n = n->rb_left; | |
cef2ac3f | 784 | else if (addr >= va->va_end) |
db64fe02 NP |
785 | n = n->rb_right; |
786 | else | |
787 | return va; | |
788 | } | |
789 | ||
790 | return NULL; | |
791 | } | |
792 | ||
68ad4a33 URS |
793 | /* |
794 | * This function returns back addresses of parent node | |
795 | * and its left or right link for further processing. | |
9c801f61 URS |
796 | * |
797 | * Otherwise NULL is returned. In that case all further | |
798 | * steps regarding inserting of conflicting overlap range | |
799 | * have to be declined and actually considered as a bug. | |
68ad4a33 URS |
800 | */ |
801 | static __always_inline struct rb_node ** | |
802 | find_va_links(struct vmap_area *va, | |
803 | struct rb_root *root, struct rb_node *from, | |
804 | struct rb_node **parent) | |
805 | { | |
806 | struct vmap_area *tmp_va; | |
807 | struct rb_node **link; | |
808 | ||
809 | if (root) { | |
810 | link = &root->rb_node; | |
811 | if (unlikely(!*link)) { | |
812 | *parent = NULL; | |
813 | return link; | |
814 | } | |
815 | } else { | |
816 | link = &from; | |
817 | } | |
db64fe02 | 818 | |
68ad4a33 URS |
819 | /* |
820 | * Go to the bottom of the tree. When we hit the last point | |
821 | * we end up with parent rb_node and correct direction, i name | |
822 | * it link, where the new va->rb_node will be attached to. | |
823 | */ | |
824 | do { | |
825 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 826 | |
68ad4a33 URS |
827 | /* |
828 | * During the traversal we also do some sanity check. | |
829 | * Trigger the BUG() if there are sides(left/right) | |
830 | * or full overlaps. | |
831 | */ | |
832 | if (va->va_start < tmp_va->va_end && | |
833 | va->va_end <= tmp_va->va_start) | |
834 | link = &(*link)->rb_left; | |
835 | else if (va->va_end > tmp_va->va_start && | |
836 | va->va_start >= tmp_va->va_end) | |
837 | link = &(*link)->rb_right; | |
9c801f61 URS |
838 | else { |
839 | WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n", | |
840 | va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end); | |
841 | ||
842 | return NULL; | |
843 | } | |
68ad4a33 URS |
844 | } while (*link); |
845 | ||
846 | *parent = &tmp_va->rb_node; | |
847 | return link; | |
848 | } | |
849 | ||
850 | static __always_inline struct list_head * | |
851 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
852 | { | |
853 | struct list_head *list; | |
854 | ||
855 | if (unlikely(!parent)) | |
856 | /* | |
857 | * The red-black tree where we try to find VA neighbors | |
858 | * before merging or inserting is empty, i.e. it means | |
859 | * there is no free vmap space. Normally it does not | |
860 | * happen but we handle this case anyway. | |
861 | */ | |
862 | return NULL; | |
863 | ||
864 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
865 | return (&parent->rb_right == link ? list->next : list); | |
866 | } | |
867 | ||
868 | static __always_inline void | |
869 | link_va(struct vmap_area *va, struct rb_root *root, | |
870 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
871 | { | |
872 | /* | |
873 | * VA is still not in the list, but we can | |
874 | * identify its future previous list_head node. | |
875 | */ | |
876 | if (likely(parent)) { | |
877 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
878 | if (&parent->rb_right != link) | |
879 | head = head->prev; | |
db64fe02 NP |
880 | } |
881 | ||
68ad4a33 URS |
882 | /* Insert to the rb-tree */ |
883 | rb_link_node(&va->rb_node, parent, link); | |
884 | if (root == &free_vmap_area_root) { | |
885 | /* | |
886 | * Some explanation here. Just perform simple insertion | |
887 | * to the tree. We do not set va->subtree_max_size to | |
888 | * its current size before calling rb_insert_augmented(). | |
889 | * It is because of we populate the tree from the bottom | |
890 | * to parent levels when the node _is_ in the tree. | |
891 | * | |
892 | * Therefore we set subtree_max_size to zero after insertion, | |
893 | * to let __augment_tree_propagate_from() puts everything to | |
894 | * the correct order later on. | |
895 | */ | |
896 | rb_insert_augmented(&va->rb_node, | |
897 | root, &free_vmap_area_rb_augment_cb); | |
898 | va->subtree_max_size = 0; | |
899 | } else { | |
900 | rb_insert_color(&va->rb_node, root); | |
901 | } | |
db64fe02 | 902 | |
68ad4a33 URS |
903 | /* Address-sort this list */ |
904 | list_add(&va->list, head); | |
db64fe02 NP |
905 | } |
906 | ||
68ad4a33 URS |
907 | static __always_inline void |
908 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
909 | { | |
460e42d1 URS |
910 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
911 | return; | |
db64fe02 | 912 | |
460e42d1 URS |
913 | if (root == &free_vmap_area_root) |
914 | rb_erase_augmented(&va->rb_node, | |
915 | root, &free_vmap_area_rb_augment_cb); | |
916 | else | |
917 | rb_erase(&va->rb_node, root); | |
918 | ||
919 | list_del(&va->list); | |
920 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
921 | } |
922 | ||
bb850f4d URS |
923 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
924 | static void | |
da27c9ed | 925 | augment_tree_propagate_check(void) |
bb850f4d URS |
926 | { |
927 | struct vmap_area *va; | |
da27c9ed | 928 | unsigned long computed_size; |
bb850f4d | 929 | |
da27c9ed URS |
930 | list_for_each_entry(va, &free_vmap_area_list, list) { |
931 | computed_size = compute_subtree_max_size(va); | |
932 | if (computed_size != va->subtree_max_size) | |
933 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
934 | va_size(va), va->subtree_max_size); | |
bb850f4d | 935 | } |
bb850f4d URS |
936 | } |
937 | #endif | |
938 | ||
68ad4a33 URS |
939 | /* |
940 | * This function populates subtree_max_size from bottom to upper | |
941 | * levels starting from VA point. The propagation must be done | |
942 | * when VA size is modified by changing its va_start/va_end. Or | |
943 | * in case of newly inserting of VA to the tree. | |
944 | * | |
945 | * It means that __augment_tree_propagate_from() must be called: | |
946 | * - After VA has been inserted to the tree(free path); | |
947 | * - After VA has been shrunk(allocation path); | |
948 | * - After VA has been increased(merging path). | |
949 | * | |
950 | * Please note that, it does not mean that upper parent nodes | |
951 | * and their subtree_max_size are recalculated all the time up | |
952 | * to the root node. | |
953 | * | |
954 | * 4--8 | |
955 | * /\ | |
956 | * / \ | |
957 | * / \ | |
958 | * 2--2 8--8 | |
959 | * | |
960 | * For example if we modify the node 4, shrinking it to 2, then | |
961 | * no any modification is required. If we shrink the node 2 to 1 | |
962 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
963 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
964 | * node becomes 4--6. | |
965 | */ | |
966 | static __always_inline void | |
967 | augment_tree_propagate_from(struct vmap_area *va) | |
968 | { | |
15ae144f URS |
969 | /* |
970 | * Populate the tree from bottom towards the root until | |
971 | * the calculated maximum available size of checked node | |
972 | * is equal to its current one. | |
973 | */ | |
974 | free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL); | |
bb850f4d URS |
975 | |
976 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
da27c9ed | 977 | augment_tree_propagate_check(); |
bb850f4d | 978 | #endif |
68ad4a33 URS |
979 | } |
980 | ||
981 | static void | |
982 | insert_vmap_area(struct vmap_area *va, | |
983 | struct rb_root *root, struct list_head *head) | |
984 | { | |
985 | struct rb_node **link; | |
986 | struct rb_node *parent; | |
987 | ||
988 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
989 | if (link) |
990 | link_va(va, root, parent, link, head); | |
68ad4a33 URS |
991 | } |
992 | ||
993 | static void | |
994 | insert_vmap_area_augment(struct vmap_area *va, | |
995 | struct rb_node *from, struct rb_root *root, | |
996 | struct list_head *head) | |
997 | { | |
998 | struct rb_node **link; | |
999 | struct rb_node *parent; | |
1000 | ||
1001 | if (from) | |
1002 | link = find_va_links(va, NULL, from, &parent); | |
1003 | else | |
1004 | link = find_va_links(va, root, NULL, &parent); | |
1005 | ||
9c801f61 URS |
1006 | if (link) { |
1007 | link_va(va, root, parent, link, head); | |
1008 | augment_tree_propagate_from(va); | |
1009 | } | |
68ad4a33 URS |
1010 | } |
1011 | ||
1012 | /* | |
1013 | * Merge de-allocated chunk of VA memory with previous | |
1014 | * and next free blocks. If coalesce is not done a new | |
1015 | * free area is inserted. If VA has been merged, it is | |
1016 | * freed. | |
9c801f61 URS |
1017 | * |
1018 | * Please note, it can return NULL in case of overlap | |
1019 | * ranges, followed by WARN() report. Despite it is a | |
1020 | * buggy behaviour, a system can be alive and keep | |
1021 | * ongoing. | |
68ad4a33 | 1022 | */ |
3c5c3cfb | 1023 | static __always_inline struct vmap_area * |
68ad4a33 URS |
1024 | merge_or_add_vmap_area(struct vmap_area *va, |
1025 | struct rb_root *root, struct list_head *head) | |
1026 | { | |
1027 | struct vmap_area *sibling; | |
1028 | struct list_head *next; | |
1029 | struct rb_node **link; | |
1030 | struct rb_node *parent; | |
1031 | bool merged = false; | |
1032 | ||
1033 | /* | |
1034 | * Find a place in the tree where VA potentially will be | |
1035 | * inserted, unless it is merged with its sibling/siblings. | |
1036 | */ | |
1037 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
1038 | if (!link) |
1039 | return NULL; | |
68ad4a33 URS |
1040 | |
1041 | /* | |
1042 | * Get next node of VA to check if merging can be done. | |
1043 | */ | |
1044 | next = get_va_next_sibling(parent, link); | |
1045 | if (unlikely(next == NULL)) | |
1046 | goto insert; | |
1047 | ||
1048 | /* | |
1049 | * start end | |
1050 | * | | | |
1051 | * |<------VA------>|<-----Next----->| | |
1052 | * | | | |
1053 | * start end | |
1054 | */ | |
1055 | if (next != head) { | |
1056 | sibling = list_entry(next, struct vmap_area, list); | |
1057 | if (sibling->va_start == va->va_end) { | |
1058 | sibling->va_start = va->va_start; | |
1059 | ||
68ad4a33 URS |
1060 | /* Free vmap_area object. */ |
1061 | kmem_cache_free(vmap_area_cachep, va); | |
1062 | ||
1063 | /* Point to the new merged area. */ | |
1064 | va = sibling; | |
1065 | merged = true; | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * start end | |
1071 | * | | | |
1072 | * |<-----Prev----->|<------VA------>| | |
1073 | * | | | |
1074 | * start end | |
1075 | */ | |
1076 | if (next->prev != head) { | |
1077 | sibling = list_entry(next->prev, struct vmap_area, list); | |
1078 | if (sibling->va_end == va->va_start) { | |
5dd78640 URS |
1079 | /* |
1080 | * If both neighbors are coalesced, it is important | |
1081 | * to unlink the "next" node first, followed by merging | |
1082 | * with "previous" one. Otherwise the tree might not be | |
1083 | * fully populated if a sibling's augmented value is | |
1084 | * "normalized" because of rotation operations. | |
1085 | */ | |
54f63d9d URS |
1086 | if (merged) |
1087 | unlink_va(va, root); | |
68ad4a33 | 1088 | |
5dd78640 URS |
1089 | sibling->va_end = va->va_end; |
1090 | ||
68ad4a33 URS |
1091 | /* Free vmap_area object. */ |
1092 | kmem_cache_free(vmap_area_cachep, va); | |
3c5c3cfb DA |
1093 | |
1094 | /* Point to the new merged area. */ | |
1095 | va = sibling; | |
1096 | merged = true; | |
68ad4a33 URS |
1097 | } |
1098 | } | |
1099 | ||
1100 | insert: | |
5dd78640 | 1101 | if (!merged) |
68ad4a33 | 1102 | link_va(va, root, parent, link, head); |
3c5c3cfb | 1103 | |
96e2db45 URS |
1104 | return va; |
1105 | } | |
1106 | ||
1107 | static __always_inline struct vmap_area * | |
1108 | merge_or_add_vmap_area_augment(struct vmap_area *va, | |
1109 | struct rb_root *root, struct list_head *head) | |
1110 | { | |
1111 | va = merge_or_add_vmap_area(va, root, head); | |
1112 | if (va) | |
1113 | augment_tree_propagate_from(va); | |
1114 | ||
3c5c3cfb | 1115 | return va; |
68ad4a33 URS |
1116 | } |
1117 | ||
1118 | static __always_inline bool | |
1119 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
1120 | unsigned long align, unsigned long vstart) | |
1121 | { | |
1122 | unsigned long nva_start_addr; | |
1123 | ||
1124 | if (va->va_start > vstart) | |
1125 | nva_start_addr = ALIGN(va->va_start, align); | |
1126 | else | |
1127 | nva_start_addr = ALIGN(vstart, align); | |
1128 | ||
1129 | /* Can be overflowed due to big size or alignment. */ | |
1130 | if (nva_start_addr + size < nva_start_addr || | |
1131 | nva_start_addr < vstart) | |
1132 | return false; | |
1133 | ||
1134 | return (nva_start_addr + size <= va->va_end); | |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * Find the first free block(lowest start address) in the tree, | |
1139 | * that will accomplish the request corresponding to passing | |
1140 | * parameters. | |
1141 | */ | |
1142 | static __always_inline struct vmap_area * | |
1143 | find_vmap_lowest_match(unsigned long size, | |
1144 | unsigned long align, unsigned long vstart) | |
1145 | { | |
1146 | struct vmap_area *va; | |
1147 | struct rb_node *node; | |
1148 | unsigned long length; | |
1149 | ||
1150 | /* Start from the root. */ | |
1151 | node = free_vmap_area_root.rb_node; | |
1152 | ||
1153 | /* Adjust the search size for alignment overhead. */ | |
1154 | length = size + align - 1; | |
1155 | ||
1156 | while (node) { | |
1157 | va = rb_entry(node, struct vmap_area, rb_node); | |
1158 | ||
1159 | if (get_subtree_max_size(node->rb_left) >= length && | |
1160 | vstart < va->va_start) { | |
1161 | node = node->rb_left; | |
1162 | } else { | |
1163 | if (is_within_this_va(va, size, align, vstart)) | |
1164 | return va; | |
1165 | ||
1166 | /* | |
1167 | * Does not make sense to go deeper towards the right | |
1168 | * sub-tree if it does not have a free block that is | |
1169 | * equal or bigger to the requested search length. | |
1170 | */ | |
1171 | if (get_subtree_max_size(node->rb_right) >= length) { | |
1172 | node = node->rb_right; | |
1173 | continue; | |
1174 | } | |
1175 | ||
1176 | /* | |
3806b041 | 1177 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
1178 | * that will satisfy the search criteria. It can happen |
1179 | * only once due to "vstart" restriction. | |
1180 | */ | |
1181 | while ((node = rb_parent(node))) { | |
1182 | va = rb_entry(node, struct vmap_area, rb_node); | |
1183 | if (is_within_this_va(va, size, align, vstart)) | |
1184 | return va; | |
1185 | ||
1186 | if (get_subtree_max_size(node->rb_right) >= length && | |
1187 | vstart <= va->va_start) { | |
1188 | node = node->rb_right; | |
1189 | break; | |
1190 | } | |
1191 | } | |
1192 | } | |
1193 | } | |
1194 | ||
1195 | return NULL; | |
1196 | } | |
1197 | ||
a6cf4e0f URS |
1198 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1199 | #include <linux/random.h> | |
1200 | ||
1201 | static struct vmap_area * | |
1202 | find_vmap_lowest_linear_match(unsigned long size, | |
1203 | unsigned long align, unsigned long vstart) | |
1204 | { | |
1205 | struct vmap_area *va; | |
1206 | ||
1207 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
1208 | if (!is_within_this_va(va, size, align, vstart)) | |
1209 | continue; | |
1210 | ||
1211 | return va; | |
1212 | } | |
1213 | ||
1214 | return NULL; | |
1215 | } | |
1216 | ||
1217 | static void | |
1218 | find_vmap_lowest_match_check(unsigned long size) | |
1219 | { | |
1220 | struct vmap_area *va_1, *va_2; | |
1221 | unsigned long vstart; | |
1222 | unsigned int rnd; | |
1223 | ||
1224 | get_random_bytes(&rnd, sizeof(rnd)); | |
1225 | vstart = VMALLOC_START + rnd; | |
1226 | ||
1227 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
1228 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
1229 | ||
1230 | if (va_1 != va_2) | |
1231 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
1232 | va_1, va_2, vstart); | |
1233 | } | |
1234 | #endif | |
1235 | ||
68ad4a33 URS |
1236 | enum fit_type { |
1237 | NOTHING_FIT = 0, | |
1238 | FL_FIT_TYPE = 1, /* full fit */ | |
1239 | LE_FIT_TYPE = 2, /* left edge fit */ | |
1240 | RE_FIT_TYPE = 3, /* right edge fit */ | |
1241 | NE_FIT_TYPE = 4 /* no edge fit */ | |
1242 | }; | |
1243 | ||
1244 | static __always_inline enum fit_type | |
1245 | classify_va_fit_type(struct vmap_area *va, | |
1246 | unsigned long nva_start_addr, unsigned long size) | |
1247 | { | |
1248 | enum fit_type type; | |
1249 | ||
1250 | /* Check if it is within VA. */ | |
1251 | if (nva_start_addr < va->va_start || | |
1252 | nva_start_addr + size > va->va_end) | |
1253 | return NOTHING_FIT; | |
1254 | ||
1255 | /* Now classify. */ | |
1256 | if (va->va_start == nva_start_addr) { | |
1257 | if (va->va_end == nva_start_addr + size) | |
1258 | type = FL_FIT_TYPE; | |
1259 | else | |
1260 | type = LE_FIT_TYPE; | |
1261 | } else if (va->va_end == nva_start_addr + size) { | |
1262 | type = RE_FIT_TYPE; | |
1263 | } else { | |
1264 | type = NE_FIT_TYPE; | |
1265 | } | |
1266 | ||
1267 | return type; | |
1268 | } | |
1269 | ||
1270 | static __always_inline int | |
1271 | adjust_va_to_fit_type(struct vmap_area *va, | |
1272 | unsigned long nva_start_addr, unsigned long size, | |
1273 | enum fit_type type) | |
1274 | { | |
2c929233 | 1275 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
1276 | |
1277 | if (type == FL_FIT_TYPE) { | |
1278 | /* | |
1279 | * No need to split VA, it fully fits. | |
1280 | * | |
1281 | * | | | |
1282 | * V NVA V | |
1283 | * |---------------| | |
1284 | */ | |
1285 | unlink_va(va, &free_vmap_area_root); | |
1286 | kmem_cache_free(vmap_area_cachep, va); | |
1287 | } else if (type == LE_FIT_TYPE) { | |
1288 | /* | |
1289 | * Split left edge of fit VA. | |
1290 | * | |
1291 | * | | | |
1292 | * V NVA V R | |
1293 | * |-------|-------| | |
1294 | */ | |
1295 | va->va_start += size; | |
1296 | } else if (type == RE_FIT_TYPE) { | |
1297 | /* | |
1298 | * Split right edge of fit VA. | |
1299 | * | |
1300 | * | | | |
1301 | * L V NVA V | |
1302 | * |-------|-------| | |
1303 | */ | |
1304 | va->va_end = nva_start_addr; | |
1305 | } else if (type == NE_FIT_TYPE) { | |
1306 | /* | |
1307 | * Split no edge of fit VA. | |
1308 | * | |
1309 | * | | | |
1310 | * L V NVA V R | |
1311 | * |---|-------|---| | |
1312 | */ | |
82dd23e8 URS |
1313 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
1314 | if (unlikely(!lva)) { | |
1315 | /* | |
1316 | * For percpu allocator we do not do any pre-allocation | |
1317 | * and leave it as it is. The reason is it most likely | |
1318 | * never ends up with NE_FIT_TYPE splitting. In case of | |
1319 | * percpu allocations offsets and sizes are aligned to | |
1320 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
1321 | * are its main fitting cases. | |
1322 | * | |
1323 | * There are a few exceptions though, as an example it is | |
1324 | * a first allocation (early boot up) when we have "one" | |
1325 | * big free space that has to be split. | |
060650a2 URS |
1326 | * |
1327 | * Also we can hit this path in case of regular "vmap" | |
1328 | * allocations, if "this" current CPU was not preloaded. | |
1329 | * See the comment in alloc_vmap_area() why. If so, then | |
1330 | * GFP_NOWAIT is used instead to get an extra object for | |
1331 | * split purpose. That is rare and most time does not | |
1332 | * occur. | |
1333 | * | |
1334 | * What happens if an allocation gets failed. Basically, | |
1335 | * an "overflow" path is triggered to purge lazily freed | |
1336 | * areas to free some memory, then, the "retry" path is | |
1337 | * triggered to repeat one more time. See more details | |
1338 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
1339 | */ |
1340 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
1341 | if (!lva) | |
1342 | return -1; | |
1343 | } | |
68ad4a33 URS |
1344 | |
1345 | /* | |
1346 | * Build the remainder. | |
1347 | */ | |
1348 | lva->va_start = va->va_start; | |
1349 | lva->va_end = nva_start_addr; | |
1350 | ||
1351 | /* | |
1352 | * Shrink this VA to remaining size. | |
1353 | */ | |
1354 | va->va_start = nva_start_addr + size; | |
1355 | } else { | |
1356 | return -1; | |
1357 | } | |
1358 | ||
1359 | if (type != FL_FIT_TYPE) { | |
1360 | augment_tree_propagate_from(va); | |
1361 | ||
2c929233 | 1362 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
1363 | insert_vmap_area_augment(lva, &va->rb_node, |
1364 | &free_vmap_area_root, &free_vmap_area_list); | |
1365 | } | |
1366 | ||
1367 | return 0; | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Returns a start address of the newly allocated area, if success. | |
1372 | * Otherwise a vend is returned that indicates failure. | |
1373 | */ | |
1374 | static __always_inline unsigned long | |
1375 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1376 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1377 | { |
1378 | unsigned long nva_start_addr; | |
1379 | struct vmap_area *va; | |
1380 | enum fit_type type; | |
1381 | int ret; | |
1382 | ||
1383 | va = find_vmap_lowest_match(size, align, vstart); | |
1384 | if (unlikely(!va)) | |
1385 | return vend; | |
1386 | ||
1387 | if (va->va_start > vstart) | |
1388 | nva_start_addr = ALIGN(va->va_start, align); | |
1389 | else | |
1390 | nva_start_addr = ALIGN(vstart, align); | |
1391 | ||
1392 | /* Check the "vend" restriction. */ | |
1393 | if (nva_start_addr + size > vend) | |
1394 | return vend; | |
1395 | ||
1396 | /* Classify what we have found. */ | |
1397 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1398 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1399 | return vend; | |
1400 | ||
1401 | /* Update the free vmap_area. */ | |
1402 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1403 | if (ret) | |
1404 | return vend; | |
1405 | ||
a6cf4e0f URS |
1406 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1407 | find_vmap_lowest_match_check(size); | |
1408 | #endif | |
1409 | ||
68ad4a33 URS |
1410 | return nva_start_addr; |
1411 | } | |
4da56b99 | 1412 | |
d98c9e83 AR |
1413 | /* |
1414 | * Free a region of KVA allocated by alloc_vmap_area | |
1415 | */ | |
1416 | static void free_vmap_area(struct vmap_area *va) | |
1417 | { | |
1418 | /* | |
1419 | * Remove from the busy tree/list. | |
1420 | */ | |
1421 | spin_lock(&vmap_area_lock); | |
1422 | unlink_va(va, &vmap_area_root); | |
1423 | spin_unlock(&vmap_area_lock); | |
1424 | ||
1425 | /* | |
1426 | * Insert/Merge it back to the free tree/list. | |
1427 | */ | |
1428 | spin_lock(&free_vmap_area_lock); | |
96e2db45 | 1429 | merge_or_add_vmap_area_augment(va, &free_vmap_area_root, &free_vmap_area_list); |
d98c9e83 AR |
1430 | spin_unlock(&free_vmap_area_lock); |
1431 | } | |
1432 | ||
db64fe02 NP |
1433 | /* |
1434 | * Allocate a region of KVA of the specified size and alignment, within the | |
1435 | * vstart and vend. | |
1436 | */ | |
1437 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1438 | unsigned long align, | |
1439 | unsigned long vstart, unsigned long vend, | |
1440 | int node, gfp_t gfp_mask) | |
1441 | { | |
82dd23e8 | 1442 | struct vmap_area *va, *pva; |
1da177e4 | 1443 | unsigned long addr; |
db64fe02 | 1444 | int purged = 0; |
d98c9e83 | 1445 | int ret; |
db64fe02 | 1446 | |
7766970c | 1447 | BUG_ON(!size); |
891c49ab | 1448 | BUG_ON(offset_in_page(size)); |
89699605 | 1449 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1450 | |
68ad4a33 URS |
1451 | if (unlikely(!vmap_initialized)) |
1452 | return ERR_PTR(-EBUSY); | |
1453 | ||
5803ed29 | 1454 | might_sleep(); |
f07116d7 | 1455 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1456 | |
f07116d7 | 1457 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1458 | if (unlikely(!va)) |
1459 | return ERR_PTR(-ENOMEM); | |
1460 | ||
7f88f88f CM |
1461 | /* |
1462 | * Only scan the relevant parts containing pointers to other objects | |
1463 | * to avoid false negatives. | |
1464 | */ | |
f07116d7 | 1465 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1466 | |
db64fe02 | 1467 | retry: |
82dd23e8 | 1468 | /* |
81f1ba58 URS |
1469 | * Preload this CPU with one extra vmap_area object. It is used |
1470 | * when fit type of free area is NE_FIT_TYPE. Please note, it | |
1471 | * does not guarantee that an allocation occurs on a CPU that | |
1472 | * is preloaded, instead we minimize the case when it is not. | |
1473 | * It can happen because of cpu migration, because there is a | |
1474 | * race until the below spinlock is taken. | |
82dd23e8 URS |
1475 | * |
1476 | * The preload is done in non-atomic context, thus it allows us | |
1477 | * to use more permissive allocation masks to be more stable under | |
81f1ba58 URS |
1478 | * low memory condition and high memory pressure. In rare case, |
1479 | * if not preloaded, GFP_NOWAIT is used. | |
82dd23e8 | 1480 | * |
81f1ba58 | 1481 | * Set "pva" to NULL here, because of "retry" path. |
82dd23e8 | 1482 | */ |
81f1ba58 | 1483 | pva = NULL; |
82dd23e8 | 1484 | |
81f1ba58 URS |
1485 | if (!this_cpu_read(ne_fit_preload_node)) |
1486 | /* | |
1487 | * Even if it fails we do not really care about that. | |
1488 | * Just proceed as it is. If needed "overflow" path | |
1489 | * will refill the cache we allocate from. | |
1490 | */ | |
f07116d7 | 1491 | pva = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
82dd23e8 | 1492 | |
e36176be | 1493 | spin_lock(&free_vmap_area_lock); |
81f1ba58 URS |
1494 | |
1495 | if (pva && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) | |
1496 | kmem_cache_free(vmap_area_cachep, pva); | |
89699605 | 1497 | |
afd07389 | 1498 | /* |
68ad4a33 URS |
1499 | * If an allocation fails, the "vend" address is |
1500 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1501 | */ |
cacca6ba | 1502 | addr = __alloc_vmap_area(size, align, vstart, vend); |
e36176be URS |
1503 | spin_unlock(&free_vmap_area_lock); |
1504 | ||
68ad4a33 | 1505 | if (unlikely(addr == vend)) |
89699605 | 1506 | goto overflow; |
db64fe02 NP |
1507 | |
1508 | va->va_start = addr; | |
1509 | va->va_end = addr + size; | |
688fcbfc | 1510 | va->vm = NULL; |
68ad4a33 | 1511 | |
d98c9e83 | 1512 | |
e36176be URS |
1513 | spin_lock(&vmap_area_lock); |
1514 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
db64fe02 NP |
1515 | spin_unlock(&vmap_area_lock); |
1516 | ||
61e16557 | 1517 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1518 | BUG_ON(va->va_start < vstart); |
1519 | BUG_ON(va->va_end > vend); | |
1520 | ||
d98c9e83 AR |
1521 | ret = kasan_populate_vmalloc(addr, size); |
1522 | if (ret) { | |
1523 | free_vmap_area(va); | |
1524 | return ERR_PTR(ret); | |
1525 | } | |
1526 | ||
db64fe02 | 1527 | return va; |
89699605 NP |
1528 | |
1529 | overflow: | |
89699605 NP |
1530 | if (!purged) { |
1531 | purge_vmap_area_lazy(); | |
1532 | purged = 1; | |
1533 | goto retry; | |
1534 | } | |
4da56b99 CW |
1535 | |
1536 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1537 | unsigned long freed = 0; | |
1538 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1539 | if (freed > 0) { | |
1540 | purged = 0; | |
1541 | goto retry; | |
1542 | } | |
1543 | } | |
1544 | ||
03497d76 | 1545 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1546 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1547 | size); | |
68ad4a33 URS |
1548 | |
1549 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1550 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1551 | } |
1552 | ||
4da56b99 CW |
1553 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1554 | { | |
1555 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1556 | } | |
1557 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1558 | ||
1559 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1560 | { | |
1561 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1562 | } | |
1563 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1564 | ||
db64fe02 NP |
1565 | /* |
1566 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1567 | * before attempting to purge with a TLB flush. | |
1568 | * | |
1569 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1570 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1571 | * global TLB flushes that must be performed. It would seem natural to scale | |
1572 | * this number up linearly with the number of CPUs (because vmapping activity | |
1573 | * could also scale linearly with the number of CPUs), however it is likely | |
1574 | * that in practice, workloads might be constrained in other ways that mean | |
1575 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1576 | * conservative and not introduce a big latency on huge systems, so go with | |
1577 | * a less aggressive log scale. It will still be an improvement over the old | |
1578 | * code, and it will be simple to change the scale factor if we find that it | |
1579 | * becomes a problem on bigger systems. | |
1580 | */ | |
1581 | static unsigned long lazy_max_pages(void) | |
1582 | { | |
1583 | unsigned int log; | |
1584 | ||
1585 | log = fls(num_online_cpus()); | |
1586 | ||
1587 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1588 | } | |
1589 | ||
4d36e6f8 | 1590 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1591 | |
0574ecd1 CH |
1592 | /* |
1593 | * Serialize vmap purging. There is no actual criticial section protected | |
1594 | * by this look, but we want to avoid concurrent calls for performance | |
1595 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1596 | */ | |
f9e09977 | 1597 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1598 | |
02b709df NP |
1599 | /* for per-CPU blocks */ |
1600 | static void purge_fragmented_blocks_allcpus(void); | |
1601 | ||
3ee48b6a CW |
1602 | /* |
1603 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1604 | * immediately freed. | |
1605 | */ | |
1606 | void set_iounmap_nonlazy(void) | |
1607 | { | |
4d36e6f8 | 1608 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1609 | } |
1610 | ||
db64fe02 NP |
1611 | /* |
1612 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1613 | */ |
0574ecd1 | 1614 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1615 | { |
4d36e6f8 | 1616 | unsigned long resched_threshold; |
96e2db45 URS |
1617 | struct list_head local_pure_list; |
1618 | struct vmap_area *va, *n_va; | |
db64fe02 | 1619 | |
0574ecd1 | 1620 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1621 | |
96e2db45 URS |
1622 | spin_lock(&purge_vmap_area_lock); |
1623 | purge_vmap_area_root = RB_ROOT; | |
1624 | list_replace_init(&purge_vmap_area_list, &local_pure_list); | |
1625 | spin_unlock(&purge_vmap_area_lock); | |
1626 | ||
1627 | if (unlikely(list_empty(&local_pure_list))) | |
68571be9 URS |
1628 | return false; |
1629 | ||
96e2db45 URS |
1630 | start = min(start, |
1631 | list_first_entry(&local_pure_list, | |
1632 | struct vmap_area, list)->va_start); | |
1633 | ||
1634 | end = max(end, | |
1635 | list_last_entry(&local_pure_list, | |
1636 | struct vmap_area, list)->va_end); | |
db64fe02 | 1637 | |
0574ecd1 | 1638 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1639 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1640 | |
e36176be | 1641 | spin_lock(&free_vmap_area_lock); |
96e2db45 | 1642 | list_for_each_entry_safe(va, n_va, &local_pure_list, list) { |
4d36e6f8 | 1643 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
3c5c3cfb DA |
1644 | unsigned long orig_start = va->va_start; |
1645 | unsigned long orig_end = va->va_end; | |
763b218d | 1646 | |
dd3b8353 URS |
1647 | /* |
1648 | * Finally insert or merge lazily-freed area. It is | |
1649 | * detached and there is no need to "unlink" it from | |
1650 | * anything. | |
1651 | */ | |
96e2db45 URS |
1652 | va = merge_or_add_vmap_area_augment(va, &free_vmap_area_root, |
1653 | &free_vmap_area_list); | |
3c5c3cfb | 1654 | |
9c801f61 URS |
1655 | if (!va) |
1656 | continue; | |
1657 | ||
3c5c3cfb DA |
1658 | if (is_vmalloc_or_module_addr((void *)orig_start)) |
1659 | kasan_release_vmalloc(orig_start, orig_end, | |
1660 | va->va_start, va->va_end); | |
dd3b8353 | 1661 | |
4d36e6f8 | 1662 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1663 | |
4d36e6f8 | 1664 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
e36176be | 1665 | cond_resched_lock(&free_vmap_area_lock); |
763b218d | 1666 | } |
e36176be | 1667 | spin_unlock(&free_vmap_area_lock); |
0574ecd1 | 1668 | return true; |
db64fe02 NP |
1669 | } |
1670 | ||
496850e5 NP |
1671 | /* |
1672 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1673 | * is already purging. | |
1674 | */ | |
1675 | static void try_purge_vmap_area_lazy(void) | |
1676 | { | |
f9e09977 | 1677 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1678 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1679 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1680 | } |
496850e5 NP |
1681 | } |
1682 | ||
db64fe02 NP |
1683 | /* |
1684 | * Kick off a purge of the outstanding lazy areas. | |
1685 | */ | |
1686 | static void purge_vmap_area_lazy(void) | |
1687 | { | |
f9e09977 | 1688 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1689 | purge_fragmented_blocks_allcpus(); |
1690 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1691 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1692 | } |
1693 | ||
1694 | /* | |
64141da5 JF |
1695 | * Free a vmap area, caller ensuring that the area has been unmapped |
1696 | * and flush_cache_vunmap had been called for the correct range | |
1697 | * previously. | |
db64fe02 | 1698 | */ |
64141da5 | 1699 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1700 | { |
4d36e6f8 | 1701 | unsigned long nr_lazy; |
80c4bd7a | 1702 | |
dd3b8353 URS |
1703 | spin_lock(&vmap_area_lock); |
1704 | unlink_va(va, &vmap_area_root); | |
1705 | spin_unlock(&vmap_area_lock); | |
1706 | ||
4d36e6f8 URS |
1707 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1708 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a | 1709 | |
96e2db45 URS |
1710 | /* |
1711 | * Merge or place it to the purge tree/list. | |
1712 | */ | |
1713 | spin_lock(&purge_vmap_area_lock); | |
1714 | merge_or_add_vmap_area(va, | |
1715 | &purge_vmap_area_root, &purge_vmap_area_list); | |
1716 | spin_unlock(&purge_vmap_area_lock); | |
80c4bd7a | 1717 | |
96e2db45 | 1718 | /* After this point, we may free va at any time */ |
80c4bd7a | 1719 | if (unlikely(nr_lazy > lazy_max_pages())) |
496850e5 | 1720 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1721 | } |
1722 | ||
b29acbdc NP |
1723 | /* |
1724 | * Free and unmap a vmap area | |
1725 | */ | |
1726 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1727 | { | |
1728 | flush_cache_vunmap(va->va_start, va->va_end); | |
4ad0ae8c | 1729 | vunmap_range_noflush(va->va_start, va->va_end); |
8e57f8ac | 1730 | if (debug_pagealloc_enabled_static()) |
82a2e924 CP |
1731 | flush_tlb_kernel_range(va->va_start, va->va_end); |
1732 | ||
c8eef01e | 1733 | free_vmap_area_noflush(va); |
b29acbdc NP |
1734 | } |
1735 | ||
db64fe02 NP |
1736 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1737 | { | |
1738 | struct vmap_area *va; | |
1739 | ||
1740 | spin_lock(&vmap_area_lock); | |
1741 | va = __find_vmap_area(addr); | |
1742 | spin_unlock(&vmap_area_lock); | |
1743 | ||
1744 | return va; | |
1745 | } | |
1746 | ||
db64fe02 NP |
1747 | /*** Per cpu kva allocator ***/ |
1748 | ||
1749 | /* | |
1750 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1751 | * room for at least 16 percpu vmap blocks per CPU. | |
1752 | */ | |
1753 | /* | |
1754 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1755 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1756 | * instead (we just need a rough idea) | |
1757 | */ | |
1758 | #if BITS_PER_LONG == 32 | |
1759 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1760 | #else | |
1761 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1762 | #endif | |
1763 | ||
1764 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1765 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1766 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1767 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1768 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1769 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1770 | #define VMAP_BBMAP_BITS \ |
1771 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1772 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1773 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1774 | |
1775 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1776 | ||
1777 | struct vmap_block_queue { | |
1778 | spinlock_t lock; | |
1779 | struct list_head free; | |
db64fe02 NP |
1780 | }; |
1781 | ||
1782 | struct vmap_block { | |
1783 | spinlock_t lock; | |
1784 | struct vmap_area *va; | |
db64fe02 | 1785 | unsigned long free, dirty; |
7d61bfe8 | 1786 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1787 | struct list_head free_list; |
1788 | struct rcu_head rcu_head; | |
02b709df | 1789 | struct list_head purge; |
db64fe02 NP |
1790 | }; |
1791 | ||
1792 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1793 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1794 | ||
1795 | /* | |
0f14599c | 1796 | * XArray of vmap blocks, indexed by address, to quickly find a vmap block |
db64fe02 NP |
1797 | * in the free path. Could get rid of this if we change the API to return a |
1798 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1799 | */ | |
0f14599c | 1800 | static DEFINE_XARRAY(vmap_blocks); |
db64fe02 NP |
1801 | |
1802 | /* | |
1803 | * We should probably have a fallback mechanism to allocate virtual memory | |
1804 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1805 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1806 | * big problem. | |
1807 | */ | |
1808 | ||
1809 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1810 | { | |
1811 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1812 | addr /= VMAP_BLOCK_SIZE; | |
1813 | return addr; | |
1814 | } | |
1815 | ||
cf725ce2 RP |
1816 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1817 | { | |
1818 | unsigned long addr; | |
1819 | ||
1820 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1821 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1822 | return (void *)addr; | |
1823 | } | |
1824 | ||
1825 | /** | |
1826 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1827 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1828 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1829 | * @gfp_mask: flags for the page level allocator | |
1830 | * | |
a862f68a | 1831 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1832 | */ |
1833 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1834 | { |
1835 | struct vmap_block_queue *vbq; | |
1836 | struct vmap_block *vb; | |
1837 | struct vmap_area *va; | |
1838 | unsigned long vb_idx; | |
1839 | int node, err; | |
cf725ce2 | 1840 | void *vaddr; |
db64fe02 NP |
1841 | |
1842 | node = numa_node_id(); | |
1843 | ||
1844 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1845 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1846 | if (unlikely(!vb)) | |
1847 | return ERR_PTR(-ENOMEM); | |
1848 | ||
1849 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1850 | VMALLOC_START, VMALLOC_END, | |
1851 | node, gfp_mask); | |
ddf9c6d4 | 1852 | if (IS_ERR(va)) { |
db64fe02 | 1853 | kfree(vb); |
e7d86340 | 1854 | return ERR_CAST(va); |
db64fe02 NP |
1855 | } |
1856 | ||
cf725ce2 | 1857 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1858 | spin_lock_init(&vb->lock); |
1859 | vb->va = va; | |
cf725ce2 RP |
1860 | /* At least something should be left free */ |
1861 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1862 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1863 | vb->dirty = 0; |
7d61bfe8 RP |
1864 | vb->dirty_min = VMAP_BBMAP_BITS; |
1865 | vb->dirty_max = 0; | |
db64fe02 | 1866 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1867 | |
1868 | vb_idx = addr_to_vb_idx(va->va_start); | |
0f14599c MWO |
1869 | err = xa_insert(&vmap_blocks, vb_idx, vb, gfp_mask); |
1870 | if (err) { | |
1871 | kfree(vb); | |
1872 | free_vmap_area(va); | |
1873 | return ERR_PTR(err); | |
1874 | } | |
db64fe02 NP |
1875 | |
1876 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1877 | spin_lock(&vbq->lock); |
68ac546f | 1878 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1879 | spin_unlock(&vbq->lock); |
3f04ba85 | 1880 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1881 | |
cf725ce2 | 1882 | return vaddr; |
db64fe02 NP |
1883 | } |
1884 | ||
db64fe02 NP |
1885 | static void free_vmap_block(struct vmap_block *vb) |
1886 | { | |
1887 | struct vmap_block *tmp; | |
db64fe02 | 1888 | |
0f14599c | 1889 | tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start)); |
db64fe02 NP |
1890 | BUG_ON(tmp != vb); |
1891 | ||
64141da5 | 1892 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1893 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1894 | } |
1895 | ||
02b709df NP |
1896 | static void purge_fragmented_blocks(int cpu) |
1897 | { | |
1898 | LIST_HEAD(purge); | |
1899 | struct vmap_block *vb; | |
1900 | struct vmap_block *n_vb; | |
1901 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1902 | ||
1903 | rcu_read_lock(); | |
1904 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1905 | ||
1906 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1907 | continue; | |
1908 | ||
1909 | spin_lock(&vb->lock); | |
1910 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1911 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1912 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1913 | vb->dirty_min = 0; |
1914 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1915 | spin_lock(&vbq->lock); |
1916 | list_del_rcu(&vb->free_list); | |
1917 | spin_unlock(&vbq->lock); | |
1918 | spin_unlock(&vb->lock); | |
1919 | list_add_tail(&vb->purge, &purge); | |
1920 | } else | |
1921 | spin_unlock(&vb->lock); | |
1922 | } | |
1923 | rcu_read_unlock(); | |
1924 | ||
1925 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1926 | list_del(&vb->purge); | |
1927 | free_vmap_block(vb); | |
1928 | } | |
1929 | } | |
1930 | ||
02b709df NP |
1931 | static void purge_fragmented_blocks_allcpus(void) |
1932 | { | |
1933 | int cpu; | |
1934 | ||
1935 | for_each_possible_cpu(cpu) | |
1936 | purge_fragmented_blocks(cpu); | |
1937 | } | |
1938 | ||
db64fe02 NP |
1939 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1940 | { | |
1941 | struct vmap_block_queue *vbq; | |
1942 | struct vmap_block *vb; | |
cf725ce2 | 1943 | void *vaddr = NULL; |
db64fe02 NP |
1944 | unsigned int order; |
1945 | ||
891c49ab | 1946 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1947 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1948 | if (WARN_ON(size == 0)) { |
1949 | /* | |
1950 | * Allocating 0 bytes isn't what caller wants since | |
1951 | * get_order(0) returns funny result. Just warn and terminate | |
1952 | * early. | |
1953 | */ | |
1954 | return NULL; | |
1955 | } | |
db64fe02 NP |
1956 | order = get_order(size); |
1957 | ||
db64fe02 NP |
1958 | rcu_read_lock(); |
1959 | vbq = &get_cpu_var(vmap_block_queue); | |
1960 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1961 | unsigned long pages_off; |
db64fe02 NP |
1962 | |
1963 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1964 | if (vb->free < (1UL << order)) { |
1965 | spin_unlock(&vb->lock); | |
1966 | continue; | |
1967 | } | |
02b709df | 1968 | |
cf725ce2 RP |
1969 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1970 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1971 | vb->free -= 1UL << order; |
1972 | if (vb->free == 0) { | |
1973 | spin_lock(&vbq->lock); | |
1974 | list_del_rcu(&vb->free_list); | |
1975 | spin_unlock(&vbq->lock); | |
1976 | } | |
cf725ce2 | 1977 | |
02b709df NP |
1978 | spin_unlock(&vb->lock); |
1979 | break; | |
db64fe02 | 1980 | } |
02b709df | 1981 | |
3f04ba85 | 1982 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1983 | rcu_read_unlock(); |
1984 | ||
cf725ce2 RP |
1985 | /* Allocate new block if nothing was found */ |
1986 | if (!vaddr) | |
1987 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1988 | |
cf725ce2 | 1989 | return vaddr; |
db64fe02 NP |
1990 | } |
1991 | ||
78a0e8c4 | 1992 | static void vb_free(unsigned long addr, unsigned long size) |
db64fe02 NP |
1993 | { |
1994 | unsigned long offset; | |
db64fe02 NP |
1995 | unsigned int order; |
1996 | struct vmap_block *vb; | |
1997 | ||
891c49ab | 1998 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1999 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc | 2000 | |
78a0e8c4 | 2001 | flush_cache_vunmap(addr, addr + size); |
b29acbdc | 2002 | |
db64fe02 | 2003 | order = get_order(size); |
78a0e8c4 | 2004 | offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT; |
0f14599c | 2005 | vb = xa_load(&vmap_blocks, addr_to_vb_idx(addr)); |
db64fe02 | 2006 | |
4ad0ae8c | 2007 | vunmap_range_noflush(addr, addr + size); |
64141da5 | 2008 | |
8e57f8ac | 2009 | if (debug_pagealloc_enabled_static()) |
78a0e8c4 | 2010 | flush_tlb_kernel_range(addr, addr + size); |
82a2e924 | 2011 | |
db64fe02 | 2012 | spin_lock(&vb->lock); |
7d61bfe8 RP |
2013 | |
2014 | /* Expand dirty range */ | |
2015 | vb->dirty_min = min(vb->dirty_min, offset); | |
2016 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 2017 | |
db64fe02 NP |
2018 | vb->dirty += 1UL << order; |
2019 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 2020 | BUG_ON(vb->free); |
db64fe02 NP |
2021 | spin_unlock(&vb->lock); |
2022 | free_vmap_block(vb); | |
2023 | } else | |
2024 | spin_unlock(&vb->lock); | |
2025 | } | |
2026 | ||
868b104d | 2027 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 2028 | { |
db64fe02 | 2029 | int cpu; |
db64fe02 | 2030 | |
9b463334 JF |
2031 | if (unlikely(!vmap_initialized)) |
2032 | return; | |
2033 | ||
5803ed29 CH |
2034 | might_sleep(); |
2035 | ||
db64fe02 NP |
2036 | for_each_possible_cpu(cpu) { |
2037 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
2038 | struct vmap_block *vb; | |
2039 | ||
2040 | rcu_read_lock(); | |
2041 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 2042 | spin_lock(&vb->lock); |
7d61bfe8 RP |
2043 | if (vb->dirty) { |
2044 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 2045 | unsigned long s, e; |
b136be5e | 2046 | |
7d61bfe8 RP |
2047 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
2048 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 2049 | |
7d61bfe8 RP |
2050 | start = min(s, start); |
2051 | end = max(e, end); | |
db64fe02 | 2052 | |
7d61bfe8 | 2053 | flush = 1; |
db64fe02 NP |
2054 | } |
2055 | spin_unlock(&vb->lock); | |
2056 | } | |
2057 | rcu_read_unlock(); | |
2058 | } | |
2059 | ||
f9e09977 | 2060 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
2061 | purge_fragmented_blocks_allcpus(); |
2062 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
2063 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 2064 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 2065 | } |
868b104d RE |
2066 | |
2067 | /** | |
2068 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
2069 | * | |
2070 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
2071 | * to amortize TLB flushing overheads. What this means is that any page you | |
2072 | * have now, may, in a former life, have been mapped into kernel virtual | |
2073 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
2074 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
2075 | * | |
2076 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
2077 | * be sure that none of the pages we have control over will have any aliases | |
2078 | * from the vmap layer. | |
2079 | */ | |
2080 | void vm_unmap_aliases(void) | |
2081 | { | |
2082 | unsigned long start = ULONG_MAX, end = 0; | |
2083 | int flush = 0; | |
2084 | ||
2085 | _vm_unmap_aliases(start, end, flush); | |
2086 | } | |
db64fe02 NP |
2087 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
2088 | ||
2089 | /** | |
2090 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
2091 | * @mem: the pointer returned by vm_map_ram | |
2092 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
2093 | */ | |
2094 | void vm_unmap_ram(const void *mem, unsigned int count) | |
2095 | { | |
65ee03c4 | 2096 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 2097 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 2098 | struct vmap_area *va; |
db64fe02 | 2099 | |
5803ed29 | 2100 | might_sleep(); |
db64fe02 NP |
2101 | BUG_ON(!addr); |
2102 | BUG_ON(addr < VMALLOC_START); | |
2103 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 2104 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 2105 | |
d98c9e83 AR |
2106 | kasan_poison_vmalloc(mem, size); |
2107 | ||
9c3acf60 | 2108 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 2109 | debug_check_no_locks_freed(mem, size); |
78a0e8c4 | 2110 | vb_free(addr, size); |
9c3acf60 CH |
2111 | return; |
2112 | } | |
2113 | ||
2114 | va = find_vmap_area(addr); | |
2115 | BUG_ON(!va); | |
05e3ff95 CP |
2116 | debug_check_no_locks_freed((void *)va->va_start, |
2117 | (va->va_end - va->va_start)); | |
9c3acf60 | 2118 | free_unmap_vmap_area(va); |
db64fe02 NP |
2119 | } |
2120 | EXPORT_SYMBOL(vm_unmap_ram); | |
2121 | ||
2122 | /** | |
2123 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
2124 | * @pages: an array of pointers to the pages to be mapped | |
2125 | * @count: number of pages | |
2126 | * @node: prefer to allocate data structures on this node | |
e99c97ad | 2127 | * |
36437638 GK |
2128 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
2129 | * faster than vmap so it's good. But if you mix long-life and short-life | |
2130 | * objects with vm_map_ram(), it could consume lots of address space through | |
2131 | * fragmentation (especially on a 32bit machine). You could see failures in | |
2132 | * the end. Please use this function for short-lived objects. | |
2133 | * | |
e99c97ad | 2134 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 | 2135 | */ |
d4efd79a | 2136 | void *vm_map_ram(struct page **pages, unsigned int count, int node) |
db64fe02 | 2137 | { |
65ee03c4 | 2138 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
2139 | unsigned long addr; |
2140 | void *mem; | |
2141 | ||
2142 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
2143 | mem = vb_alloc(size, GFP_KERNEL); | |
2144 | if (IS_ERR(mem)) | |
2145 | return NULL; | |
2146 | addr = (unsigned long)mem; | |
2147 | } else { | |
2148 | struct vmap_area *va; | |
2149 | va = alloc_vmap_area(size, PAGE_SIZE, | |
2150 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
2151 | if (IS_ERR(va)) | |
2152 | return NULL; | |
2153 | ||
2154 | addr = va->va_start; | |
2155 | mem = (void *)addr; | |
2156 | } | |
d98c9e83 AR |
2157 | |
2158 | kasan_unpoison_vmalloc(mem, size); | |
2159 | ||
b67177ec NP |
2160 | if (vmap_pages_range(addr, addr + size, PAGE_KERNEL, |
2161 | pages, PAGE_SHIFT) < 0) { | |
db64fe02 NP |
2162 | vm_unmap_ram(mem, count); |
2163 | return NULL; | |
2164 | } | |
b67177ec | 2165 | |
db64fe02 NP |
2166 | return mem; |
2167 | } | |
2168 | EXPORT_SYMBOL(vm_map_ram); | |
2169 | ||
4341fa45 | 2170 | static struct vm_struct *vmlist __initdata; |
92eac168 | 2171 | |
121e6f32 NP |
2172 | static inline unsigned int vm_area_page_order(struct vm_struct *vm) |
2173 | { | |
2174 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC | |
2175 | return vm->page_order; | |
2176 | #else | |
2177 | return 0; | |
2178 | #endif | |
2179 | } | |
2180 | ||
2181 | static inline void set_vm_area_page_order(struct vm_struct *vm, unsigned int order) | |
2182 | { | |
2183 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC | |
2184 | vm->page_order = order; | |
2185 | #else | |
2186 | BUG_ON(order != 0); | |
2187 | #endif | |
2188 | } | |
2189 | ||
be9b7335 NP |
2190 | /** |
2191 | * vm_area_add_early - add vmap area early during boot | |
2192 | * @vm: vm_struct to add | |
2193 | * | |
2194 | * This function is used to add fixed kernel vm area to vmlist before | |
2195 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
2196 | * should contain proper values and the other fields should be zero. | |
2197 | * | |
2198 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2199 | */ | |
2200 | void __init vm_area_add_early(struct vm_struct *vm) | |
2201 | { | |
2202 | struct vm_struct *tmp, **p; | |
2203 | ||
2204 | BUG_ON(vmap_initialized); | |
2205 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
2206 | if (tmp->addr >= vm->addr) { | |
2207 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
2208 | break; | |
2209 | } else | |
2210 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
2211 | } | |
2212 | vm->next = *p; | |
2213 | *p = vm; | |
2214 | } | |
2215 | ||
f0aa6617 TH |
2216 | /** |
2217 | * vm_area_register_early - register vmap area early during boot | |
2218 | * @vm: vm_struct to register | |
c0c0a293 | 2219 | * @align: requested alignment |
f0aa6617 TH |
2220 | * |
2221 | * This function is used to register kernel vm area before | |
2222 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
2223 | * proper values on entry and other fields should be zero. On return, | |
2224 | * vm->addr contains the allocated address. | |
2225 | * | |
2226 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2227 | */ | |
c0c0a293 | 2228 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
2229 | { |
2230 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
2231 | unsigned long addr; |
2232 | ||
2233 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
2234 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 2235 | |
c0c0a293 | 2236 | vm->addr = (void *)addr; |
f0aa6617 | 2237 | |
be9b7335 | 2238 | vm_area_add_early(vm); |
f0aa6617 TH |
2239 | } |
2240 | ||
68ad4a33 URS |
2241 | static void vmap_init_free_space(void) |
2242 | { | |
2243 | unsigned long vmap_start = 1; | |
2244 | const unsigned long vmap_end = ULONG_MAX; | |
2245 | struct vmap_area *busy, *free; | |
2246 | ||
2247 | /* | |
2248 | * B F B B B F | |
2249 | * -|-----|.....|-----|-----|-----|.....|- | |
2250 | * | The KVA space | | |
2251 | * |<--------------------------------->| | |
2252 | */ | |
2253 | list_for_each_entry(busy, &vmap_area_list, list) { | |
2254 | if (busy->va_start - vmap_start > 0) { | |
2255 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2256 | if (!WARN_ON_ONCE(!free)) { | |
2257 | free->va_start = vmap_start; | |
2258 | free->va_end = busy->va_start; | |
2259 | ||
2260 | insert_vmap_area_augment(free, NULL, | |
2261 | &free_vmap_area_root, | |
2262 | &free_vmap_area_list); | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | vmap_start = busy->va_end; | |
2267 | } | |
2268 | ||
2269 | if (vmap_end - vmap_start > 0) { | |
2270 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2271 | if (!WARN_ON_ONCE(!free)) { | |
2272 | free->va_start = vmap_start; | |
2273 | free->va_end = vmap_end; | |
2274 | ||
2275 | insert_vmap_area_augment(free, NULL, | |
2276 | &free_vmap_area_root, | |
2277 | &free_vmap_area_list); | |
2278 | } | |
2279 | } | |
2280 | } | |
2281 | ||
db64fe02 NP |
2282 | void __init vmalloc_init(void) |
2283 | { | |
822c18f2 IK |
2284 | struct vmap_area *va; |
2285 | struct vm_struct *tmp; | |
db64fe02 NP |
2286 | int i; |
2287 | ||
68ad4a33 URS |
2288 | /* |
2289 | * Create the cache for vmap_area objects. | |
2290 | */ | |
2291 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
2292 | ||
db64fe02 NP |
2293 | for_each_possible_cpu(i) { |
2294 | struct vmap_block_queue *vbq; | |
32fcfd40 | 2295 | struct vfree_deferred *p; |
db64fe02 NP |
2296 | |
2297 | vbq = &per_cpu(vmap_block_queue, i); | |
2298 | spin_lock_init(&vbq->lock); | |
2299 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
2300 | p = &per_cpu(vfree_deferred, i); |
2301 | init_llist_head(&p->list); | |
2302 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 2303 | } |
9b463334 | 2304 | |
822c18f2 IK |
2305 | /* Import existing vmlist entries. */ |
2306 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
2307 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
2308 | if (WARN_ON_ONCE(!va)) | |
2309 | continue; | |
2310 | ||
822c18f2 IK |
2311 | va->va_start = (unsigned long)tmp->addr; |
2312 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 2313 | va->vm = tmp; |
68ad4a33 | 2314 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 2315 | } |
ca23e405 | 2316 | |
68ad4a33 URS |
2317 | /* |
2318 | * Now we can initialize a free vmap space. | |
2319 | */ | |
2320 | vmap_init_free_space(); | |
9b463334 | 2321 | vmap_initialized = true; |
db64fe02 NP |
2322 | } |
2323 | ||
e36176be URS |
2324 | static inline void setup_vmalloc_vm_locked(struct vm_struct *vm, |
2325 | struct vmap_area *va, unsigned long flags, const void *caller) | |
cf88c790 | 2326 | { |
cf88c790 TH |
2327 | vm->flags = flags; |
2328 | vm->addr = (void *)va->va_start; | |
2329 | vm->size = va->va_end - va->va_start; | |
2330 | vm->caller = caller; | |
db1aecaf | 2331 | va->vm = vm; |
e36176be URS |
2332 | } |
2333 | ||
2334 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, | |
2335 | unsigned long flags, const void *caller) | |
2336 | { | |
2337 | spin_lock(&vmap_area_lock); | |
2338 | setup_vmalloc_vm_locked(vm, va, flags, caller); | |
c69480ad | 2339 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2340 | } |
cf88c790 | 2341 | |
20fc02b4 | 2342 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2343 | { |
d4033afd | 2344 | /* |
20fc02b4 | 2345 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2346 | * we should make sure that vm has proper values. |
2347 | * Pair with smp_rmb() in show_numa_info(). | |
2348 | */ | |
2349 | smp_wmb(); | |
20fc02b4 | 2350 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2351 | } |
2352 | ||
db64fe02 | 2353 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2354 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2355 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2356 | { |
0006526d | 2357 | struct vmap_area *va; |
db64fe02 | 2358 | struct vm_struct *area; |
d98c9e83 | 2359 | unsigned long requested_size = size; |
1da177e4 | 2360 | |
52fd24ca | 2361 | BUG_ON(in_interrupt()); |
1da177e4 | 2362 | size = PAGE_ALIGN(size); |
31be8309 OH |
2363 | if (unlikely(!size)) |
2364 | return NULL; | |
1da177e4 | 2365 | |
252e5c6e | 2366 | if (flags & VM_IOREMAP) |
2367 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2368 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2369 | ||
cf88c790 | 2370 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2371 | if (unlikely(!area)) |
2372 | return NULL; | |
2373 | ||
71394fe5 AR |
2374 | if (!(flags & VM_NO_GUARD)) |
2375 | size += PAGE_SIZE; | |
1da177e4 | 2376 | |
db64fe02 NP |
2377 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2378 | if (IS_ERR(va)) { | |
2379 | kfree(area); | |
2380 | return NULL; | |
1da177e4 | 2381 | } |
1da177e4 | 2382 | |
d98c9e83 | 2383 | kasan_unpoison_vmalloc((void *)va->va_start, requested_size); |
f5252e00 | 2384 | |
d98c9e83 | 2385 | setup_vmalloc_vm(area, va, flags, caller); |
3c5c3cfb | 2386 | |
1da177e4 | 2387 | return area; |
1da177e4 LT |
2388 | } |
2389 | ||
c2968612 BH |
2390 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2391 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2392 | const void *caller) |
c2968612 | 2393 | { |
00ef2d2f DR |
2394 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2395 | GFP_KERNEL, caller); | |
c2968612 BH |
2396 | } |
2397 | ||
1da177e4 | 2398 | /** |
92eac168 MR |
2399 | * get_vm_area - reserve a contiguous kernel virtual area |
2400 | * @size: size of the area | |
2401 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2402 | * |
92eac168 MR |
2403 | * Search an area of @size in the kernel virtual mapping area, |
2404 | * and reserved it for out purposes. Returns the area descriptor | |
2405 | * on success or %NULL on failure. | |
a862f68a MR |
2406 | * |
2407 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2408 | */ |
2409 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2410 | { | |
2dca6999 | 2411 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2412 | NUMA_NO_NODE, GFP_KERNEL, |
2413 | __builtin_return_address(0)); | |
23016969 CL |
2414 | } |
2415 | ||
2416 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2417 | const void *caller) |
23016969 | 2418 | { |
2dca6999 | 2419 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2420 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2421 | } |
2422 | ||
e9da6e99 | 2423 | /** |
92eac168 MR |
2424 | * find_vm_area - find a continuous kernel virtual area |
2425 | * @addr: base address | |
e9da6e99 | 2426 | * |
92eac168 MR |
2427 | * Search for the kernel VM area starting at @addr, and return it. |
2428 | * It is up to the caller to do all required locking to keep the returned | |
2429 | * pointer valid. | |
a862f68a | 2430 | * |
74640617 | 2431 | * Return: the area descriptor on success or %NULL on failure. |
e9da6e99 MS |
2432 | */ |
2433 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2434 | { |
db64fe02 | 2435 | struct vmap_area *va; |
83342314 | 2436 | |
db64fe02 | 2437 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2438 | if (!va) |
2439 | return NULL; | |
1da177e4 | 2440 | |
688fcbfc | 2441 | return va->vm; |
1da177e4 LT |
2442 | } |
2443 | ||
7856dfeb | 2444 | /** |
92eac168 MR |
2445 | * remove_vm_area - find and remove a continuous kernel virtual area |
2446 | * @addr: base address | |
7856dfeb | 2447 | * |
92eac168 MR |
2448 | * Search for the kernel VM area starting at @addr, and remove it. |
2449 | * This function returns the found VM area, but using it is NOT safe | |
2450 | * on SMP machines, except for its size or flags. | |
a862f68a | 2451 | * |
74640617 | 2452 | * Return: the area descriptor on success or %NULL on failure. |
7856dfeb | 2453 | */ |
b3bdda02 | 2454 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2455 | { |
db64fe02 NP |
2456 | struct vmap_area *va; |
2457 | ||
5803ed29 CH |
2458 | might_sleep(); |
2459 | ||
dd3b8353 URS |
2460 | spin_lock(&vmap_area_lock); |
2461 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2462 | if (va && va->vm) { |
db1aecaf | 2463 | struct vm_struct *vm = va->vm; |
f5252e00 | 2464 | |
c69480ad | 2465 | va->vm = NULL; |
c69480ad JK |
2466 | spin_unlock(&vmap_area_lock); |
2467 | ||
a5af5aa8 | 2468 | kasan_free_shadow(vm); |
dd32c279 | 2469 | free_unmap_vmap_area(va); |
dd32c279 | 2470 | |
db64fe02 NP |
2471 | return vm; |
2472 | } | |
dd3b8353 URS |
2473 | |
2474 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2475 | return NULL; |
7856dfeb AK |
2476 | } |
2477 | ||
868b104d RE |
2478 | static inline void set_area_direct_map(const struct vm_struct *area, |
2479 | int (*set_direct_map)(struct page *page)) | |
2480 | { | |
2481 | int i; | |
2482 | ||
121e6f32 | 2483 | /* HUGE_VMALLOC passes small pages to set_direct_map */ |
868b104d RE |
2484 | for (i = 0; i < area->nr_pages; i++) |
2485 | if (page_address(area->pages[i])) | |
2486 | set_direct_map(area->pages[i]); | |
2487 | } | |
2488 | ||
2489 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2490 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2491 | { | |
868b104d | 2492 | unsigned long start = ULONG_MAX, end = 0; |
121e6f32 | 2493 | unsigned int page_order = vm_area_page_order(area); |
868b104d | 2494 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; |
31e67340 | 2495 | int flush_dmap = 0; |
868b104d RE |
2496 | int i; |
2497 | ||
868b104d RE |
2498 | remove_vm_area(area->addr); |
2499 | ||
2500 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2501 | if (!flush_reset) | |
2502 | return; | |
2503 | ||
2504 | /* | |
2505 | * If not deallocating pages, just do the flush of the VM area and | |
2506 | * return. | |
2507 | */ | |
2508 | if (!deallocate_pages) { | |
2509 | vm_unmap_aliases(); | |
2510 | return; | |
2511 | } | |
2512 | ||
2513 | /* | |
2514 | * If execution gets here, flush the vm mapping and reset the direct | |
2515 | * map. Find the start and end range of the direct mappings to make sure | |
2516 | * the vm_unmap_aliases() flush includes the direct map. | |
2517 | */ | |
121e6f32 | 2518 | for (i = 0; i < area->nr_pages; i += 1U << page_order) { |
8e41f872 RE |
2519 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2520 | if (addr) { | |
121e6f32 NP |
2521 | unsigned long page_size; |
2522 | ||
2523 | page_size = PAGE_SIZE << page_order; | |
868b104d | 2524 | start = min(addr, start); |
121e6f32 | 2525 | end = max(addr + page_size, end); |
31e67340 | 2526 | flush_dmap = 1; |
868b104d RE |
2527 | } |
2528 | } | |
2529 | ||
2530 | /* | |
2531 | * Set direct map to something invalid so that it won't be cached if | |
2532 | * there are any accesses after the TLB flush, then flush the TLB and | |
2533 | * reset the direct map permissions to the default. | |
2534 | */ | |
2535 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2536 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2537 | set_area_direct_map(area, set_direct_map_default_noflush); |
2538 | } | |
2539 | ||
b3bdda02 | 2540 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2541 | { |
2542 | struct vm_struct *area; | |
2543 | ||
2544 | if (!addr) | |
2545 | return; | |
2546 | ||
e69e9d4a | 2547 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2548 | addr)) |
1da177e4 | 2549 | return; |
1da177e4 | 2550 | |
6ade2032 | 2551 | area = find_vm_area(addr); |
1da177e4 | 2552 | if (unlikely(!area)) { |
4c8573e2 | 2553 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2554 | addr); |
1da177e4 LT |
2555 | return; |
2556 | } | |
2557 | ||
05e3ff95 CP |
2558 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2559 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2560 | |
c041098c | 2561 | kasan_poison_vmalloc(area->addr, get_vm_area_size(area)); |
3c5c3cfb | 2562 | |
868b104d RE |
2563 | vm_remove_mappings(area, deallocate_pages); |
2564 | ||
1da177e4 | 2565 | if (deallocate_pages) { |
121e6f32 | 2566 | unsigned int page_order = vm_area_page_order(area); |
1da177e4 LT |
2567 | int i; |
2568 | ||
121e6f32 | 2569 | for (i = 0; i < area->nr_pages; i += 1U << page_order) { |
bf53d6f8 CL |
2570 | struct page *page = area->pages[i]; |
2571 | ||
2572 | BUG_ON(!page); | |
121e6f32 | 2573 | __free_pages(page, page_order); |
1da177e4 | 2574 | } |
97105f0a | 2575 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2576 | |
244d63ee | 2577 | kvfree(area->pages); |
1da177e4 LT |
2578 | } |
2579 | ||
2580 | kfree(area); | |
1da177e4 | 2581 | } |
bf22e37a AR |
2582 | |
2583 | static inline void __vfree_deferred(const void *addr) | |
2584 | { | |
2585 | /* | |
2586 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2587 | * context. Preemption is absolutely fine here, because the llist_add() | |
2588 | * implementation is lockless, so it works even if we are adding to | |
73221d88 | 2589 | * another cpu's list. schedule_work() should be fine with this too. |
bf22e37a AR |
2590 | */ |
2591 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2592 | ||
2593 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2594 | schedule_work(&p->wq); | |
2595 | } | |
2596 | ||
2597 | /** | |
92eac168 MR |
2598 | * vfree_atomic - release memory allocated by vmalloc() |
2599 | * @addr: memory base address | |
bf22e37a | 2600 | * |
92eac168 MR |
2601 | * This one is just like vfree() but can be called in any atomic context |
2602 | * except NMIs. | |
bf22e37a AR |
2603 | */ |
2604 | void vfree_atomic(const void *addr) | |
2605 | { | |
2606 | BUG_ON(in_nmi()); | |
2607 | ||
2608 | kmemleak_free(addr); | |
2609 | ||
2610 | if (!addr) | |
2611 | return; | |
2612 | __vfree_deferred(addr); | |
2613 | } | |
2614 | ||
c67dc624 RP |
2615 | static void __vfree(const void *addr) |
2616 | { | |
2617 | if (unlikely(in_interrupt())) | |
2618 | __vfree_deferred(addr); | |
2619 | else | |
2620 | __vunmap(addr, 1); | |
2621 | } | |
2622 | ||
1da177e4 | 2623 | /** |
fa307474 MWO |
2624 | * vfree - Release memory allocated by vmalloc() |
2625 | * @addr: Memory base address | |
1da177e4 | 2626 | * |
fa307474 MWO |
2627 | * Free the virtually continuous memory area starting at @addr, as obtained |
2628 | * from one of the vmalloc() family of APIs. This will usually also free the | |
2629 | * physical memory underlying the virtual allocation, but that memory is | |
2630 | * reference counted, so it will not be freed until the last user goes away. | |
1da177e4 | 2631 | * |
fa307474 | 2632 | * If @addr is NULL, no operation is performed. |
c9fcee51 | 2633 | * |
fa307474 | 2634 | * Context: |
92eac168 | 2635 | * May sleep if called *not* from interrupt context. |
fa307474 MWO |
2636 | * Must not be called in NMI context (strictly speaking, it could be |
2637 | * if we have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2638 | * conventions for vfree() arch-depenedent would be a really bad idea). | |
1da177e4 | 2639 | */ |
b3bdda02 | 2640 | void vfree(const void *addr) |
1da177e4 | 2641 | { |
32fcfd40 | 2642 | BUG_ON(in_nmi()); |
89219d37 CM |
2643 | |
2644 | kmemleak_free(addr); | |
2645 | ||
a8dda165 AR |
2646 | might_sleep_if(!in_interrupt()); |
2647 | ||
32fcfd40 AV |
2648 | if (!addr) |
2649 | return; | |
c67dc624 RP |
2650 | |
2651 | __vfree(addr); | |
1da177e4 | 2652 | } |
1da177e4 LT |
2653 | EXPORT_SYMBOL(vfree); |
2654 | ||
2655 | /** | |
92eac168 MR |
2656 | * vunmap - release virtual mapping obtained by vmap() |
2657 | * @addr: memory base address | |
1da177e4 | 2658 | * |
92eac168 MR |
2659 | * Free the virtually contiguous memory area starting at @addr, |
2660 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2661 | * |
92eac168 | 2662 | * Must not be called in interrupt context. |
1da177e4 | 2663 | */ |
b3bdda02 | 2664 | void vunmap(const void *addr) |
1da177e4 LT |
2665 | { |
2666 | BUG_ON(in_interrupt()); | |
34754b69 | 2667 | might_sleep(); |
32fcfd40 AV |
2668 | if (addr) |
2669 | __vunmap(addr, 0); | |
1da177e4 | 2670 | } |
1da177e4 LT |
2671 | EXPORT_SYMBOL(vunmap); |
2672 | ||
2673 | /** | |
92eac168 MR |
2674 | * vmap - map an array of pages into virtually contiguous space |
2675 | * @pages: array of page pointers | |
2676 | * @count: number of pages to map | |
2677 | * @flags: vm_area->flags | |
2678 | * @prot: page protection for the mapping | |
2679 | * | |
b944afc9 CH |
2680 | * Maps @count pages from @pages into contiguous kernel virtual space. |
2681 | * If @flags contains %VM_MAP_PUT_PAGES the ownership of the pages array itself | |
2682 | * (which must be kmalloc or vmalloc memory) and one reference per pages in it | |
2683 | * are transferred from the caller to vmap(), and will be freed / dropped when | |
2684 | * vfree() is called on the return value. | |
a862f68a MR |
2685 | * |
2686 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2687 | */ |
2688 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2689 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2690 | { |
2691 | struct vm_struct *area; | |
b67177ec | 2692 | unsigned long addr; |
65ee03c4 | 2693 | unsigned long size; /* In bytes */ |
1da177e4 | 2694 | |
34754b69 PZ |
2695 | might_sleep(); |
2696 | ||
ca79b0c2 | 2697 | if (count > totalram_pages()) |
1da177e4 LT |
2698 | return NULL; |
2699 | ||
65ee03c4 GJM |
2700 | size = (unsigned long)count << PAGE_SHIFT; |
2701 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2702 | if (!area) |
2703 | return NULL; | |
23016969 | 2704 | |
b67177ec NP |
2705 | addr = (unsigned long)area->addr; |
2706 | if (vmap_pages_range(addr, addr + size, pgprot_nx(prot), | |
2707 | pages, PAGE_SHIFT) < 0) { | |
1da177e4 LT |
2708 | vunmap(area->addr); |
2709 | return NULL; | |
2710 | } | |
2711 | ||
c22ee528 | 2712 | if (flags & VM_MAP_PUT_PAGES) { |
b944afc9 | 2713 | area->pages = pages; |
c22ee528 ML |
2714 | area->nr_pages = count; |
2715 | } | |
1da177e4 LT |
2716 | return area->addr; |
2717 | } | |
1da177e4 LT |
2718 | EXPORT_SYMBOL(vmap); |
2719 | ||
3e9a9e25 CH |
2720 | #ifdef CONFIG_VMAP_PFN |
2721 | struct vmap_pfn_data { | |
2722 | unsigned long *pfns; | |
2723 | pgprot_t prot; | |
2724 | unsigned int idx; | |
2725 | }; | |
2726 | ||
2727 | static int vmap_pfn_apply(pte_t *pte, unsigned long addr, void *private) | |
2728 | { | |
2729 | struct vmap_pfn_data *data = private; | |
2730 | ||
2731 | if (WARN_ON_ONCE(pfn_valid(data->pfns[data->idx]))) | |
2732 | return -EINVAL; | |
2733 | *pte = pte_mkspecial(pfn_pte(data->pfns[data->idx++], data->prot)); | |
2734 | return 0; | |
2735 | } | |
2736 | ||
2737 | /** | |
2738 | * vmap_pfn - map an array of PFNs into virtually contiguous space | |
2739 | * @pfns: array of PFNs | |
2740 | * @count: number of pages to map | |
2741 | * @prot: page protection for the mapping | |
2742 | * | |
2743 | * Maps @count PFNs from @pfns into contiguous kernel virtual space and returns | |
2744 | * the start address of the mapping. | |
2745 | */ | |
2746 | void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot) | |
2747 | { | |
2748 | struct vmap_pfn_data data = { .pfns = pfns, .prot = pgprot_nx(prot) }; | |
2749 | struct vm_struct *area; | |
2750 | ||
2751 | area = get_vm_area_caller(count * PAGE_SIZE, VM_IOREMAP, | |
2752 | __builtin_return_address(0)); | |
2753 | if (!area) | |
2754 | return NULL; | |
2755 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
2756 | count * PAGE_SIZE, vmap_pfn_apply, &data)) { | |
2757 | free_vm_area(area); | |
2758 | return NULL; | |
2759 | } | |
2760 | return area->addr; | |
2761 | } | |
2762 | EXPORT_SYMBOL_GPL(vmap_pfn); | |
2763 | #endif /* CONFIG_VMAP_PFN */ | |
2764 | ||
e31d9eb5 | 2765 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
121e6f32 NP |
2766 | pgprot_t prot, unsigned int page_shift, |
2767 | int node) | |
1da177e4 | 2768 | { |
930f036b | 2769 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
121e6f32 NP |
2770 | unsigned long addr = (unsigned long)area->addr; |
2771 | unsigned long size = get_vm_area_size(area); | |
34fe6537 | 2772 | unsigned long array_size; |
121e6f32 NP |
2773 | unsigned int nr_small_pages = size >> PAGE_SHIFT; |
2774 | unsigned int page_order; | |
f255935b | 2775 | struct page **pages; |
121e6f32 | 2776 | unsigned int i; |
1da177e4 | 2777 | |
121e6f32 | 2778 | array_size = (unsigned long)nr_small_pages * sizeof(struct page *); |
f255935b CH |
2779 | gfp_mask |= __GFP_NOWARN; |
2780 | if (!(gfp_mask & (GFP_DMA | GFP_DMA32))) | |
2781 | gfp_mask |= __GFP_HIGHMEM; | |
1da177e4 | 2782 | |
1da177e4 | 2783 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 2784 | if (array_size > PAGE_SIZE) { |
f255935b CH |
2785 | pages = __vmalloc_node(array_size, 1, nested_gfp, node, |
2786 | area->caller); | |
286e1ea3 | 2787 | } else { |
976d6dfb | 2788 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2789 | } |
7ea36242 AK |
2790 | |
2791 | if (!pages) { | |
8945a723 | 2792 | free_vm_area(area); |
1da177e4 LT |
2793 | return NULL; |
2794 | } | |
1da177e4 | 2795 | |
7ea36242 | 2796 | area->pages = pages; |
121e6f32 NP |
2797 | area->nr_pages = nr_small_pages; |
2798 | set_vm_area_page_order(area, page_shift - PAGE_SHIFT); | |
7ea36242 | 2799 | |
121e6f32 | 2800 | page_order = vm_area_page_order(area); |
bf53d6f8 | 2801 | |
121e6f32 NP |
2802 | /* |
2803 | * Careful, we allocate and map page_order pages, but tracking is done | |
2804 | * per PAGE_SIZE page so as to keep the vm_struct APIs independent of | |
2805 | * the physical/mapped size. | |
2806 | */ | |
2807 | for (i = 0; i < area->nr_pages; i += 1U << page_order) { | |
2808 | struct page *page; | |
2809 | int p; | |
bf53d6f8 | 2810 | |
121e6f32 NP |
2811 | /* Compound pages required for remap_vmalloc_page */ |
2812 | page = alloc_pages_node(node, gfp_mask | __GFP_COMP, page_order); | |
bf53d6f8 | 2813 | if (unlikely(!page)) { |
82afbc32 | 2814 | /* Successfully allocated i pages, free them in __vfree() */ |
1da177e4 | 2815 | area->nr_pages = i; |
97105f0a | 2816 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 LT |
2817 | goto fail; |
2818 | } | |
121e6f32 NP |
2819 | |
2820 | for (p = 0; p < (1U << page_order); p++) | |
2821 | area->pages[i + p] = page + p; | |
2822 | ||
dcf61ff0 | 2823 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 2824 | cond_resched(); |
1da177e4 | 2825 | } |
97105f0a | 2826 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2827 | |
121e6f32 | 2828 | if (vmap_pages_range(addr, addr + size, prot, pages, page_shift) < 0) |
1da177e4 | 2829 | goto fail; |
ed1f324c | 2830 | |
1da177e4 LT |
2831 | return area->addr; |
2832 | ||
2833 | fail: | |
a8e99259 | 2834 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2835 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
121e6f32 | 2836 | (area->nr_pages*PAGE_SIZE), size); |
c67dc624 | 2837 | __vfree(area->addr); |
1da177e4 LT |
2838 | return NULL; |
2839 | } | |
2840 | ||
2841 | /** | |
92eac168 MR |
2842 | * __vmalloc_node_range - allocate virtually contiguous memory |
2843 | * @size: allocation size | |
2844 | * @align: desired alignment | |
2845 | * @start: vm area range start | |
2846 | * @end: vm area range end | |
2847 | * @gfp_mask: flags for the page level allocator | |
2848 | * @prot: protection mask for the allocated pages | |
2849 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2850 | * @node: node to use for allocation or NUMA_NO_NODE | |
2851 | * @caller: caller's return address | |
2852 | * | |
2853 | * Allocate enough pages to cover @size from the page level | |
2854 | * allocator with @gfp_mask flags. Map them into contiguous | |
2855 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2856 | * |
2857 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2858 | */ |
d0a21265 DR |
2859 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2860 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2861 | pgprot_t prot, unsigned long vm_flags, int node, |
2862 | const void *caller) | |
1da177e4 LT |
2863 | { |
2864 | struct vm_struct *area; | |
89219d37 CM |
2865 | void *addr; |
2866 | unsigned long real_size = size; | |
121e6f32 NP |
2867 | unsigned long real_align = align; |
2868 | unsigned int shift = PAGE_SHIFT; | |
1da177e4 | 2869 | |
121e6f32 NP |
2870 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) { |
2871 | area = NULL; | |
de7d2b56 | 2872 | goto fail; |
121e6f32 NP |
2873 | } |
2874 | ||
2875 | if (vmap_allow_huge && !(vm_flags & VM_NO_HUGE_VMAP) && | |
2876 | arch_vmap_pmd_supported(prot)) { | |
2877 | unsigned long size_per_node; | |
1da177e4 | 2878 | |
121e6f32 NP |
2879 | /* |
2880 | * Try huge pages. Only try for PAGE_KERNEL allocations, | |
2881 | * others like modules don't yet expect huge pages in | |
2882 | * their allocations due to apply_to_page_range not | |
2883 | * supporting them. | |
2884 | */ | |
2885 | ||
2886 | size_per_node = size; | |
2887 | if (node == NUMA_NO_NODE) | |
2888 | size_per_node /= num_online_nodes(); | |
2889 | if (size_per_node >= PMD_SIZE) { | |
2890 | shift = PMD_SHIFT; | |
2891 | align = max(real_align, 1UL << shift); | |
2892 | size = ALIGN(real_size, 1UL << shift); | |
2893 | } | |
2894 | } | |
2895 | ||
2896 | again: | |
2897 | size = PAGE_ALIGN(size); | |
2898 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED | | |
cb9e3c29 | 2899 | vm_flags, start, end, node, gfp_mask, caller); |
1da177e4 | 2900 | if (!area) |
de7d2b56 | 2901 | goto fail; |
1da177e4 | 2902 | |
121e6f32 | 2903 | addr = __vmalloc_area_node(area, gfp_mask, prot, shift, node); |
1368edf0 | 2904 | if (!addr) |
121e6f32 | 2905 | goto fail; |
89219d37 | 2906 | |
f5252e00 | 2907 | /* |
20fc02b4 ZY |
2908 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2909 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2910 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2911 | */ |
20fc02b4 | 2912 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2913 | |
94f4a161 | 2914 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2915 | |
2916 | return addr; | |
de7d2b56 JP |
2917 | |
2918 | fail: | |
121e6f32 NP |
2919 | if (shift > PAGE_SHIFT) { |
2920 | shift = PAGE_SHIFT; | |
2921 | align = real_align; | |
2922 | size = real_size; | |
2923 | goto again; | |
2924 | } | |
2925 | ||
2926 | if (!area) { | |
2927 | /* Warn for area allocation, page allocations already warn */ | |
2928 | warn_alloc(gfp_mask, NULL, | |
7877cdcc | 2929 | "vmalloc: allocation failure: %lu bytes", real_size); |
121e6f32 | 2930 | } |
de7d2b56 | 2931 | return NULL; |
1da177e4 LT |
2932 | } |
2933 | ||
d0a21265 | 2934 | /** |
92eac168 MR |
2935 | * __vmalloc_node - allocate virtually contiguous memory |
2936 | * @size: allocation size | |
2937 | * @align: desired alignment | |
2938 | * @gfp_mask: flags for the page level allocator | |
92eac168 MR |
2939 | * @node: node to use for allocation or NUMA_NO_NODE |
2940 | * @caller: caller's return address | |
a7c3e901 | 2941 | * |
f38fcb9c CH |
2942 | * Allocate enough pages to cover @size from the page level allocator with |
2943 | * @gfp_mask flags. Map them into contiguous kernel virtual space. | |
a7c3e901 | 2944 | * |
92eac168 MR |
2945 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2946 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2947 | * |
92eac168 MR |
2948 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2949 | * with mm people. | |
a862f68a MR |
2950 | * |
2951 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2952 | */ |
2b905948 | 2953 | void *__vmalloc_node(unsigned long size, unsigned long align, |
f38fcb9c | 2954 | gfp_t gfp_mask, int node, const void *caller) |
d0a21265 DR |
2955 | { |
2956 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
f38fcb9c | 2957 | gfp_mask, PAGE_KERNEL, 0, node, caller); |
d0a21265 | 2958 | } |
c3f896dc CH |
2959 | /* |
2960 | * This is only for performance analysis of vmalloc and stress purpose. | |
2961 | * It is required by vmalloc test module, therefore do not use it other | |
2962 | * than that. | |
2963 | */ | |
2964 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2965 | EXPORT_SYMBOL_GPL(__vmalloc_node); | |
2966 | #endif | |
d0a21265 | 2967 | |
88dca4ca | 2968 | void *__vmalloc(unsigned long size, gfp_t gfp_mask) |
930fc45a | 2969 | { |
f38fcb9c | 2970 | return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE, |
23016969 | 2971 | __builtin_return_address(0)); |
930fc45a | 2972 | } |
1da177e4 LT |
2973 | EXPORT_SYMBOL(__vmalloc); |
2974 | ||
2975 | /** | |
92eac168 MR |
2976 | * vmalloc - allocate virtually contiguous memory |
2977 | * @size: allocation size | |
2978 | * | |
2979 | * Allocate enough pages to cover @size from the page level | |
2980 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2981 | * |
92eac168 MR |
2982 | * For tight control over page level allocator and protection flags |
2983 | * use __vmalloc() instead. | |
a862f68a MR |
2984 | * |
2985 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2986 | */ |
2987 | void *vmalloc(unsigned long size) | |
2988 | { | |
4d39d728 CH |
2989 | return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE, |
2990 | __builtin_return_address(0)); | |
1da177e4 | 2991 | } |
1da177e4 LT |
2992 | EXPORT_SYMBOL(vmalloc); |
2993 | ||
e1ca7788 | 2994 | /** |
92eac168 MR |
2995 | * vzalloc - allocate virtually contiguous memory with zero fill |
2996 | * @size: allocation size | |
2997 | * | |
2998 | * Allocate enough pages to cover @size from the page level | |
2999 | * allocator and map them into contiguous kernel virtual space. | |
3000 | * The memory allocated is set to zero. | |
3001 | * | |
3002 | * For tight control over page level allocator and protection flags | |
3003 | * use __vmalloc() instead. | |
a862f68a MR |
3004 | * |
3005 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
3006 | */ |
3007 | void *vzalloc(unsigned long size) | |
3008 | { | |
4d39d728 CH |
3009 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE, |
3010 | __builtin_return_address(0)); | |
e1ca7788 DY |
3011 | } |
3012 | EXPORT_SYMBOL(vzalloc); | |
3013 | ||
83342314 | 3014 | /** |
ead04089 REB |
3015 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
3016 | * @size: allocation size | |
83342314 | 3017 | * |
ead04089 REB |
3018 | * The resulting memory area is zeroed so it can be mapped to userspace |
3019 | * without leaking data. | |
a862f68a MR |
3020 | * |
3021 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
3022 | */ |
3023 | void *vmalloc_user(unsigned long size) | |
3024 | { | |
bc84c535 RP |
3025 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
3026 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
3027 | VM_USERMAP, NUMA_NO_NODE, | |
3028 | __builtin_return_address(0)); | |
83342314 NP |
3029 | } |
3030 | EXPORT_SYMBOL(vmalloc_user); | |
3031 | ||
930fc45a | 3032 | /** |
92eac168 MR |
3033 | * vmalloc_node - allocate memory on a specific node |
3034 | * @size: allocation size | |
3035 | * @node: numa node | |
930fc45a | 3036 | * |
92eac168 MR |
3037 | * Allocate enough pages to cover @size from the page level |
3038 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 3039 | * |
92eac168 MR |
3040 | * For tight control over page level allocator and protection flags |
3041 | * use __vmalloc() instead. | |
a862f68a MR |
3042 | * |
3043 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
3044 | */ |
3045 | void *vmalloc_node(unsigned long size, int node) | |
3046 | { | |
f38fcb9c CH |
3047 | return __vmalloc_node(size, 1, GFP_KERNEL, node, |
3048 | __builtin_return_address(0)); | |
930fc45a CL |
3049 | } |
3050 | EXPORT_SYMBOL(vmalloc_node); | |
3051 | ||
e1ca7788 DY |
3052 | /** |
3053 | * vzalloc_node - allocate memory on a specific node with zero fill | |
3054 | * @size: allocation size | |
3055 | * @node: numa node | |
3056 | * | |
3057 | * Allocate enough pages to cover @size from the page level | |
3058 | * allocator and map them into contiguous kernel virtual space. | |
3059 | * The memory allocated is set to zero. | |
3060 | * | |
a862f68a | 3061 | * Return: pointer to the allocated memory or %NULL on error |
e1ca7788 DY |
3062 | */ |
3063 | void *vzalloc_node(unsigned long size, int node) | |
3064 | { | |
4d39d728 CH |
3065 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node, |
3066 | __builtin_return_address(0)); | |
e1ca7788 DY |
3067 | } |
3068 | EXPORT_SYMBOL(vzalloc_node); | |
3069 | ||
0d08e0d3 | 3070 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 3071 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 3072 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 3073 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 3074 | #else |
698d0831 MH |
3075 | /* |
3076 | * 64b systems should always have either DMA or DMA32 zones. For others | |
3077 | * GFP_DMA32 should do the right thing and use the normal zone. | |
3078 | */ | |
3079 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
3080 | #endif |
3081 | ||
1da177e4 | 3082 | /** |
92eac168 MR |
3083 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
3084 | * @size: allocation size | |
1da177e4 | 3085 | * |
92eac168 MR |
3086 | * Allocate enough 32bit PA addressable pages to cover @size from the |
3087 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
3088 | * |
3089 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
3090 | */ |
3091 | void *vmalloc_32(unsigned long size) | |
3092 | { | |
f38fcb9c CH |
3093 | return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE, |
3094 | __builtin_return_address(0)); | |
1da177e4 | 3095 | } |
1da177e4 LT |
3096 | EXPORT_SYMBOL(vmalloc_32); |
3097 | ||
83342314 | 3098 | /** |
ead04089 | 3099 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 3100 | * @size: allocation size |
ead04089 REB |
3101 | * |
3102 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
3103 | * mapped to userspace without leaking data. | |
a862f68a MR |
3104 | * |
3105 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
3106 | */ |
3107 | void *vmalloc_32_user(unsigned long size) | |
3108 | { | |
bc84c535 RP |
3109 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
3110 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
3111 | VM_USERMAP, NUMA_NO_NODE, | |
3112 | __builtin_return_address(0)); | |
83342314 NP |
3113 | } |
3114 | EXPORT_SYMBOL(vmalloc_32_user); | |
3115 | ||
d0107eb0 KH |
3116 | /* |
3117 | * small helper routine , copy contents to buf from addr. | |
3118 | * If the page is not present, fill zero. | |
3119 | */ | |
3120 | ||
3121 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
3122 | { | |
3123 | struct page *p; | |
3124 | int copied = 0; | |
3125 | ||
3126 | while (count) { | |
3127 | unsigned long offset, length; | |
3128 | ||
891c49ab | 3129 | offset = offset_in_page(addr); |
d0107eb0 KH |
3130 | length = PAGE_SIZE - offset; |
3131 | if (length > count) | |
3132 | length = count; | |
3133 | p = vmalloc_to_page(addr); | |
3134 | /* | |
3135 | * To do safe access to this _mapped_ area, we need | |
3136 | * lock. But adding lock here means that we need to add | |
3137 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
3138 | * interface, rarely used. Instead of that, we'll use | |
3139 | * kmap() and get small overhead in this access function. | |
3140 | */ | |
3141 | if (p) { | |
3142 | /* | |
3143 | * we can expect USER0 is not used (see vread/vwrite's | |
3144 | * function description) | |
3145 | */ | |
9b04c5fe | 3146 | void *map = kmap_atomic(p); |
d0107eb0 | 3147 | memcpy(buf, map + offset, length); |
9b04c5fe | 3148 | kunmap_atomic(map); |
d0107eb0 KH |
3149 | } else |
3150 | memset(buf, 0, length); | |
3151 | ||
3152 | addr += length; | |
3153 | buf += length; | |
3154 | copied += length; | |
3155 | count -= length; | |
3156 | } | |
3157 | return copied; | |
3158 | } | |
3159 | ||
3160 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
3161 | { | |
3162 | struct page *p; | |
3163 | int copied = 0; | |
3164 | ||
3165 | while (count) { | |
3166 | unsigned long offset, length; | |
3167 | ||
891c49ab | 3168 | offset = offset_in_page(addr); |
d0107eb0 KH |
3169 | length = PAGE_SIZE - offset; |
3170 | if (length > count) | |
3171 | length = count; | |
3172 | p = vmalloc_to_page(addr); | |
3173 | /* | |
3174 | * To do safe access to this _mapped_ area, we need | |
3175 | * lock. But adding lock here means that we need to add | |
3176 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
3177 | * interface, rarely used. Instead of that, we'll use | |
3178 | * kmap() and get small overhead in this access function. | |
3179 | */ | |
3180 | if (p) { | |
3181 | /* | |
3182 | * we can expect USER0 is not used (see vread/vwrite's | |
3183 | * function description) | |
3184 | */ | |
9b04c5fe | 3185 | void *map = kmap_atomic(p); |
d0107eb0 | 3186 | memcpy(map + offset, buf, length); |
9b04c5fe | 3187 | kunmap_atomic(map); |
d0107eb0 KH |
3188 | } |
3189 | addr += length; | |
3190 | buf += length; | |
3191 | copied += length; | |
3192 | count -= length; | |
3193 | } | |
3194 | return copied; | |
3195 | } | |
3196 | ||
3197 | /** | |
92eac168 MR |
3198 | * vread() - read vmalloc area in a safe way. |
3199 | * @buf: buffer for reading data | |
3200 | * @addr: vm address. | |
3201 | * @count: number of bytes to be read. | |
3202 | * | |
92eac168 MR |
3203 | * This function checks that addr is a valid vmalloc'ed area, and |
3204 | * copy data from that area to a given buffer. If the given memory range | |
3205 | * of [addr...addr+count) includes some valid address, data is copied to | |
3206 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
3207 | * IOREMAP area is treated as memory hole and no copy is done. | |
3208 | * | |
3209 | * If [addr...addr+count) doesn't includes any intersects with alive | |
3210 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
3211 | * | |
3212 | * Note: In usual ops, vread() is never necessary because the caller | |
3213 | * should know vmalloc() area is valid and can use memcpy(). | |
3214 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 3215 | * any information, as /dev/kmem. |
a862f68a MR |
3216 | * |
3217 | * Return: number of bytes for which addr and buf should be increased | |
3218 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
3219 | * include any intersection with valid vmalloc area | |
d0107eb0 | 3220 | */ |
1da177e4 LT |
3221 | long vread(char *buf, char *addr, unsigned long count) |
3222 | { | |
e81ce85f JK |
3223 | struct vmap_area *va; |
3224 | struct vm_struct *vm; | |
1da177e4 | 3225 | char *vaddr, *buf_start = buf; |
d0107eb0 | 3226 | unsigned long buflen = count; |
1da177e4 LT |
3227 | unsigned long n; |
3228 | ||
3229 | /* Don't allow overflow */ | |
3230 | if ((unsigned long) addr + count < count) | |
3231 | count = -(unsigned long) addr; | |
3232 | ||
e81ce85f | 3233 | spin_lock(&vmap_area_lock); |
f608788c SD |
3234 | va = __find_vmap_area((unsigned long)addr); |
3235 | if (!va) | |
3236 | goto finished; | |
3237 | list_for_each_entry_from(va, &vmap_area_list, list) { | |
e81ce85f JK |
3238 | if (!count) |
3239 | break; | |
3240 | ||
688fcbfc | 3241 | if (!va->vm) |
e81ce85f JK |
3242 | continue; |
3243 | ||
3244 | vm = va->vm; | |
3245 | vaddr = (char *) vm->addr; | |
762216ab | 3246 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
3247 | continue; |
3248 | while (addr < vaddr) { | |
3249 | if (count == 0) | |
3250 | goto finished; | |
3251 | *buf = '\0'; | |
3252 | buf++; | |
3253 | addr++; | |
3254 | count--; | |
3255 | } | |
762216ab | 3256 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
3257 | if (n > count) |
3258 | n = count; | |
e81ce85f | 3259 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
3260 | aligned_vread(buf, addr, n); |
3261 | else /* IOREMAP area is treated as memory hole */ | |
3262 | memset(buf, 0, n); | |
3263 | buf += n; | |
3264 | addr += n; | |
3265 | count -= n; | |
1da177e4 LT |
3266 | } |
3267 | finished: | |
e81ce85f | 3268 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3269 | |
3270 | if (buf == buf_start) | |
3271 | return 0; | |
3272 | /* zero-fill memory holes */ | |
3273 | if (buf != buf_start + buflen) | |
3274 | memset(buf, 0, buflen - (buf - buf_start)); | |
3275 | ||
3276 | return buflen; | |
1da177e4 LT |
3277 | } |
3278 | ||
d0107eb0 | 3279 | /** |
92eac168 MR |
3280 | * vwrite() - write vmalloc area in a safe way. |
3281 | * @buf: buffer for source data | |
3282 | * @addr: vm address. | |
3283 | * @count: number of bytes to be read. | |
3284 | * | |
92eac168 MR |
3285 | * This function checks that addr is a valid vmalloc'ed area, and |
3286 | * copy data from a buffer to the given addr. If specified range of | |
3287 | * [addr...addr+count) includes some valid address, data is copied from | |
3288 | * proper area of @buf. If there are memory holes, no copy to hole. | |
3289 | * IOREMAP area is treated as memory hole and no copy is done. | |
3290 | * | |
3291 | * If [addr...addr+count) doesn't includes any intersects with alive | |
3292 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
3293 | * | |
3294 | * Note: In usual ops, vwrite() is never necessary because the caller | |
3295 | * should know vmalloc() area is valid and can use memcpy(). | |
3296 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 3297 | * any information, as /dev/kmem. |
a862f68a MR |
3298 | * |
3299 | * Return: number of bytes for which addr and buf should be | |
3300 | * increased (same number as @count) or %0 if [addr...addr+count) | |
3301 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 3302 | */ |
1da177e4 LT |
3303 | long vwrite(char *buf, char *addr, unsigned long count) |
3304 | { | |
e81ce85f JK |
3305 | struct vmap_area *va; |
3306 | struct vm_struct *vm; | |
d0107eb0 KH |
3307 | char *vaddr; |
3308 | unsigned long n, buflen; | |
3309 | int copied = 0; | |
1da177e4 LT |
3310 | |
3311 | /* Don't allow overflow */ | |
3312 | if ((unsigned long) addr + count < count) | |
3313 | count = -(unsigned long) addr; | |
d0107eb0 | 3314 | buflen = count; |
1da177e4 | 3315 | |
e81ce85f JK |
3316 | spin_lock(&vmap_area_lock); |
3317 | list_for_each_entry(va, &vmap_area_list, list) { | |
3318 | if (!count) | |
3319 | break; | |
3320 | ||
688fcbfc | 3321 | if (!va->vm) |
e81ce85f JK |
3322 | continue; |
3323 | ||
3324 | vm = va->vm; | |
3325 | vaddr = (char *) vm->addr; | |
762216ab | 3326 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
3327 | continue; |
3328 | while (addr < vaddr) { | |
3329 | if (count == 0) | |
3330 | goto finished; | |
3331 | buf++; | |
3332 | addr++; | |
3333 | count--; | |
3334 | } | |
762216ab | 3335 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
3336 | if (n > count) |
3337 | n = count; | |
e81ce85f | 3338 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
3339 | aligned_vwrite(buf, addr, n); |
3340 | copied++; | |
3341 | } | |
3342 | buf += n; | |
3343 | addr += n; | |
3344 | count -= n; | |
1da177e4 LT |
3345 | } |
3346 | finished: | |
e81ce85f | 3347 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3348 | if (!copied) |
3349 | return 0; | |
3350 | return buflen; | |
1da177e4 | 3351 | } |
83342314 NP |
3352 | |
3353 | /** | |
92eac168 MR |
3354 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
3355 | * @vma: vma to cover | |
3356 | * @uaddr: target user address to start at | |
3357 | * @kaddr: virtual address of vmalloc kernel memory | |
bdebd6a2 | 3358 | * @pgoff: offset from @kaddr to start at |
92eac168 | 3359 | * @size: size of map area |
7682486b | 3360 | * |
92eac168 | 3361 | * Returns: 0 for success, -Exxx on failure |
83342314 | 3362 | * |
92eac168 MR |
3363 | * This function checks that @kaddr is a valid vmalloc'ed area, |
3364 | * and that it is big enough to cover the range starting at | |
3365 | * @uaddr in @vma. Will return failure if that criteria isn't | |
3366 | * met. | |
83342314 | 3367 | * |
92eac168 | 3368 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 3369 | */ |
e69e9d4a | 3370 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
bdebd6a2 JH |
3371 | void *kaddr, unsigned long pgoff, |
3372 | unsigned long size) | |
83342314 NP |
3373 | { |
3374 | struct vm_struct *area; | |
bdebd6a2 JH |
3375 | unsigned long off; |
3376 | unsigned long end_index; | |
3377 | ||
3378 | if (check_shl_overflow(pgoff, PAGE_SHIFT, &off)) | |
3379 | return -EINVAL; | |
83342314 | 3380 | |
e69e9d4a HD |
3381 | size = PAGE_ALIGN(size); |
3382 | ||
3383 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
3384 | return -EINVAL; |
3385 | ||
e69e9d4a | 3386 | area = find_vm_area(kaddr); |
83342314 | 3387 | if (!area) |
db64fe02 | 3388 | return -EINVAL; |
83342314 | 3389 | |
fe9041c2 | 3390 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3391 | return -EINVAL; |
83342314 | 3392 | |
bdebd6a2 JH |
3393 | if (check_add_overflow(size, off, &end_index) || |
3394 | end_index > get_vm_area_size(area)) | |
db64fe02 | 3395 | return -EINVAL; |
bdebd6a2 | 3396 | kaddr += off; |
83342314 | 3397 | |
83342314 | 3398 | do { |
e69e9d4a | 3399 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3400 | int ret; |
3401 | ||
83342314 NP |
3402 | ret = vm_insert_page(vma, uaddr, page); |
3403 | if (ret) | |
3404 | return ret; | |
3405 | ||
3406 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3407 | kaddr += PAGE_SIZE; |
3408 | size -= PAGE_SIZE; | |
3409 | } while (size > 0); | |
83342314 | 3410 | |
314e51b9 | 3411 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3412 | |
db64fe02 | 3413 | return 0; |
83342314 | 3414 | } |
e69e9d4a HD |
3415 | |
3416 | /** | |
92eac168 MR |
3417 | * remap_vmalloc_range - map vmalloc pages to userspace |
3418 | * @vma: vma to cover (map full range of vma) | |
3419 | * @addr: vmalloc memory | |
3420 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3421 | * |
92eac168 | 3422 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3423 | * |
92eac168 MR |
3424 | * This function checks that addr is a valid vmalloc'ed area, and |
3425 | * that it is big enough to cover the vma. Will return failure if | |
3426 | * that criteria isn't met. | |
e69e9d4a | 3427 | * |
92eac168 | 3428 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3429 | */ |
3430 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3431 | unsigned long pgoff) | |
3432 | { | |
3433 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
bdebd6a2 | 3434 | addr, pgoff, |
e69e9d4a HD |
3435 | vma->vm_end - vma->vm_start); |
3436 | } | |
83342314 NP |
3437 | EXPORT_SYMBOL(remap_vmalloc_range); |
3438 | ||
5f4352fb JF |
3439 | void free_vm_area(struct vm_struct *area) |
3440 | { | |
3441 | struct vm_struct *ret; | |
3442 | ret = remove_vm_area(area->addr); | |
3443 | BUG_ON(ret != area); | |
3444 | kfree(area); | |
3445 | } | |
3446 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3447 | |
4f8b02b4 | 3448 | #ifdef CONFIG_SMP |
ca23e405 TH |
3449 | static struct vmap_area *node_to_va(struct rb_node *n) |
3450 | { | |
4583e773 | 3451 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3452 | } |
3453 | ||
3454 | /** | |
68ad4a33 URS |
3455 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3456 | * @addr: target address | |
ca23e405 | 3457 | * |
68ad4a33 URS |
3458 | * Returns: vmap_area if it is found. If there is no such area |
3459 | * the first highest(reverse order) vmap_area is returned | |
3460 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3461 | * if there are no any areas before @addr. | |
ca23e405 | 3462 | */ |
68ad4a33 URS |
3463 | static struct vmap_area * |
3464 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3465 | { |
68ad4a33 URS |
3466 | struct vmap_area *va, *tmp; |
3467 | struct rb_node *n; | |
3468 | ||
3469 | n = free_vmap_area_root.rb_node; | |
3470 | va = NULL; | |
ca23e405 TH |
3471 | |
3472 | while (n) { | |
68ad4a33 URS |
3473 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3474 | if (tmp->va_start <= addr) { | |
3475 | va = tmp; | |
3476 | if (tmp->va_end >= addr) | |
3477 | break; | |
3478 | ||
ca23e405 | 3479 | n = n->rb_right; |
68ad4a33 URS |
3480 | } else { |
3481 | n = n->rb_left; | |
3482 | } | |
ca23e405 TH |
3483 | } |
3484 | ||
68ad4a33 | 3485 | return va; |
ca23e405 TH |
3486 | } |
3487 | ||
3488 | /** | |
68ad4a33 URS |
3489 | * pvm_determine_end_from_reverse - find the highest aligned address |
3490 | * of free block below VMALLOC_END | |
3491 | * @va: | |
3492 | * in - the VA we start the search(reverse order); | |
3493 | * out - the VA with the highest aligned end address. | |
799fa85d | 3494 | * @align: alignment for required highest address |
ca23e405 | 3495 | * |
68ad4a33 | 3496 | * Returns: determined end address within vmap_area |
ca23e405 | 3497 | */ |
68ad4a33 URS |
3498 | static unsigned long |
3499 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3500 | { |
68ad4a33 | 3501 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3502 | unsigned long addr; |
3503 | ||
68ad4a33 URS |
3504 | if (likely(*va)) { |
3505 | list_for_each_entry_from_reverse((*va), | |
3506 | &free_vmap_area_list, list) { | |
3507 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3508 | if ((*va)->va_start < addr) | |
3509 | return addr; | |
3510 | } | |
ca23e405 TH |
3511 | } |
3512 | ||
68ad4a33 | 3513 | return 0; |
ca23e405 TH |
3514 | } |
3515 | ||
3516 | /** | |
3517 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3518 | * @offsets: array containing offset of each area | |
3519 | * @sizes: array containing size of each area | |
3520 | * @nr_vms: the number of areas to allocate | |
3521 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3522 | * |
3523 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3524 | * vm_structs on success, %NULL on failure | |
3525 | * | |
3526 | * Percpu allocator wants to use congruent vm areas so that it can | |
3527 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3528 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3529 | * be scattered pretty far, distance between two areas easily going up | |
3530 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3531 | * areas are allocated from top. | |
ca23e405 | 3532 | * |
68ad4a33 URS |
3533 | * Despite its complicated look, this allocator is rather simple. It |
3534 | * does everything top-down and scans free blocks from the end looking | |
3535 | * for matching base. While scanning, if any of the areas do not fit the | |
3536 | * base address is pulled down to fit the area. Scanning is repeated till | |
3537 | * all the areas fit and then all necessary data structures are inserted | |
3538 | * and the result is returned. | |
ca23e405 TH |
3539 | */ |
3540 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3541 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3542 | size_t align) |
ca23e405 TH |
3543 | { |
3544 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3545 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3546 | struct vmap_area **vas, *va; |
ca23e405 TH |
3547 | struct vm_struct **vms; |
3548 | int area, area2, last_area, term_area; | |
253a496d | 3549 | unsigned long base, start, size, end, last_end, orig_start, orig_end; |
ca23e405 | 3550 | bool purged = false; |
68ad4a33 | 3551 | enum fit_type type; |
ca23e405 | 3552 | |
ca23e405 | 3553 | /* verify parameters and allocate data structures */ |
891c49ab | 3554 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3555 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3556 | start = offsets[area]; | |
3557 | end = start + sizes[area]; | |
3558 | ||
3559 | /* is everything aligned properly? */ | |
3560 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3561 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3562 | ||
3563 | /* detect the area with the highest address */ | |
3564 | if (start > offsets[last_area]) | |
3565 | last_area = area; | |
3566 | ||
c568da28 | 3567 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3568 | unsigned long start2 = offsets[area2]; |
3569 | unsigned long end2 = start2 + sizes[area2]; | |
3570 | ||
c568da28 | 3571 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3572 | } |
3573 | } | |
3574 | last_end = offsets[last_area] + sizes[last_area]; | |
3575 | ||
3576 | if (vmalloc_end - vmalloc_start < last_end) { | |
3577 | WARN_ON(true); | |
3578 | return NULL; | |
3579 | } | |
3580 | ||
4d67d860 TM |
3581 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3582 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3583 | if (!vas || !vms) |
f1db7afd | 3584 | goto err_free2; |
ca23e405 TH |
3585 | |
3586 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3587 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3588 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3589 | if (!vas[area] || !vms[area]) |
3590 | goto err_free; | |
3591 | } | |
3592 | retry: | |
e36176be | 3593 | spin_lock(&free_vmap_area_lock); |
ca23e405 TH |
3594 | |
3595 | /* start scanning - we scan from the top, begin with the last area */ | |
3596 | area = term_area = last_area; | |
3597 | start = offsets[area]; | |
3598 | end = start + sizes[area]; | |
3599 | ||
68ad4a33 URS |
3600 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3601 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3602 | |
3603 | while (true) { | |
ca23e405 TH |
3604 | /* |
3605 | * base might have underflowed, add last_end before | |
3606 | * comparing. | |
3607 | */ | |
68ad4a33 URS |
3608 | if (base + last_end < vmalloc_start + last_end) |
3609 | goto overflow; | |
ca23e405 TH |
3610 | |
3611 | /* | |
68ad4a33 | 3612 | * Fitting base has not been found. |
ca23e405 | 3613 | */ |
68ad4a33 URS |
3614 | if (va == NULL) |
3615 | goto overflow; | |
ca23e405 | 3616 | |
5336e52c | 3617 | /* |
d8cc323d | 3618 | * If required width exceeds current VA block, move |
5336e52c KS |
3619 | * base downwards and then recheck. |
3620 | */ | |
3621 | if (base + end > va->va_end) { | |
3622 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3623 | term_area = area; | |
3624 | continue; | |
3625 | } | |
3626 | ||
ca23e405 | 3627 | /* |
68ad4a33 | 3628 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3629 | */ |
5336e52c | 3630 | if (base + start < va->va_start) { |
68ad4a33 URS |
3631 | va = node_to_va(rb_prev(&va->rb_node)); |
3632 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3633 | term_area = area; |
3634 | continue; | |
3635 | } | |
3636 | ||
3637 | /* | |
3638 | * This area fits, move on to the previous one. If | |
3639 | * the previous one is the terminal one, we're done. | |
3640 | */ | |
3641 | area = (area + nr_vms - 1) % nr_vms; | |
3642 | if (area == term_area) | |
3643 | break; | |
68ad4a33 | 3644 | |
ca23e405 TH |
3645 | start = offsets[area]; |
3646 | end = start + sizes[area]; | |
68ad4a33 | 3647 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3648 | } |
68ad4a33 | 3649 | |
ca23e405 TH |
3650 | /* we've found a fitting base, insert all va's */ |
3651 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3652 | int ret; |
ca23e405 | 3653 | |
68ad4a33 URS |
3654 | start = base + offsets[area]; |
3655 | size = sizes[area]; | |
ca23e405 | 3656 | |
68ad4a33 URS |
3657 | va = pvm_find_va_enclose_addr(start); |
3658 | if (WARN_ON_ONCE(va == NULL)) | |
3659 | /* It is a BUG(), but trigger recovery instead. */ | |
3660 | goto recovery; | |
3661 | ||
3662 | type = classify_va_fit_type(va, start, size); | |
3663 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3664 | /* It is a BUG(), but trigger recovery instead. */ | |
3665 | goto recovery; | |
3666 | ||
3667 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3668 | if (unlikely(ret)) | |
3669 | goto recovery; | |
3670 | ||
3671 | /* Allocated area. */ | |
3672 | va = vas[area]; | |
3673 | va->va_start = start; | |
3674 | va->va_end = start + size; | |
68ad4a33 | 3675 | } |
ca23e405 | 3676 | |
e36176be | 3677 | spin_unlock(&free_vmap_area_lock); |
ca23e405 | 3678 | |
253a496d DA |
3679 | /* populate the kasan shadow space */ |
3680 | for (area = 0; area < nr_vms; area++) { | |
3681 | if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area])) | |
3682 | goto err_free_shadow; | |
3683 | ||
3684 | kasan_unpoison_vmalloc((void *)vas[area]->va_start, | |
3685 | sizes[area]); | |
3686 | } | |
3687 | ||
ca23e405 | 3688 | /* insert all vm's */ |
e36176be URS |
3689 | spin_lock(&vmap_area_lock); |
3690 | for (area = 0; area < nr_vms; area++) { | |
3691 | insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list); | |
3692 | ||
3693 | setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC, | |
3645cb4a | 3694 | pcpu_get_vm_areas); |
e36176be URS |
3695 | } |
3696 | spin_unlock(&vmap_area_lock); | |
ca23e405 TH |
3697 | |
3698 | kfree(vas); | |
3699 | return vms; | |
3700 | ||
68ad4a33 | 3701 | recovery: |
e36176be URS |
3702 | /* |
3703 | * Remove previously allocated areas. There is no | |
3704 | * need in removing these areas from the busy tree, | |
3705 | * because they are inserted only on the final step | |
3706 | * and when pcpu_get_vm_areas() is success. | |
3707 | */ | |
68ad4a33 | 3708 | while (area--) { |
253a496d DA |
3709 | orig_start = vas[area]->va_start; |
3710 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
3711 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
3712 | &free_vmap_area_list); | |
9c801f61 URS |
3713 | if (va) |
3714 | kasan_release_vmalloc(orig_start, orig_end, | |
3715 | va->va_start, va->va_end); | |
68ad4a33 URS |
3716 | vas[area] = NULL; |
3717 | } | |
3718 | ||
3719 | overflow: | |
e36176be | 3720 | spin_unlock(&free_vmap_area_lock); |
68ad4a33 URS |
3721 | if (!purged) { |
3722 | purge_vmap_area_lazy(); | |
3723 | purged = true; | |
3724 | ||
3725 | /* Before "retry", check if we recover. */ | |
3726 | for (area = 0; area < nr_vms; area++) { | |
3727 | if (vas[area]) | |
3728 | continue; | |
3729 | ||
3730 | vas[area] = kmem_cache_zalloc( | |
3731 | vmap_area_cachep, GFP_KERNEL); | |
3732 | if (!vas[area]) | |
3733 | goto err_free; | |
3734 | } | |
3735 | ||
3736 | goto retry; | |
3737 | } | |
3738 | ||
ca23e405 TH |
3739 | err_free: |
3740 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3741 | if (vas[area]) |
3742 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3743 | ||
f1db7afd | 3744 | kfree(vms[area]); |
ca23e405 | 3745 | } |
f1db7afd | 3746 | err_free2: |
ca23e405 TH |
3747 | kfree(vas); |
3748 | kfree(vms); | |
3749 | return NULL; | |
253a496d DA |
3750 | |
3751 | err_free_shadow: | |
3752 | spin_lock(&free_vmap_area_lock); | |
3753 | /* | |
3754 | * We release all the vmalloc shadows, even the ones for regions that | |
3755 | * hadn't been successfully added. This relies on kasan_release_vmalloc | |
3756 | * being able to tolerate this case. | |
3757 | */ | |
3758 | for (area = 0; area < nr_vms; area++) { | |
3759 | orig_start = vas[area]->va_start; | |
3760 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
3761 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
3762 | &free_vmap_area_list); | |
9c801f61 URS |
3763 | if (va) |
3764 | kasan_release_vmalloc(orig_start, orig_end, | |
3765 | va->va_start, va->va_end); | |
253a496d DA |
3766 | vas[area] = NULL; |
3767 | kfree(vms[area]); | |
3768 | } | |
3769 | spin_unlock(&free_vmap_area_lock); | |
3770 | kfree(vas); | |
3771 | kfree(vms); | |
3772 | return NULL; | |
ca23e405 TH |
3773 | } |
3774 | ||
3775 | /** | |
3776 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3777 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3778 | * @nr_vms: the number of allocated areas | |
3779 | * | |
3780 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3781 | */ | |
3782 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3783 | { | |
3784 | int i; | |
3785 | ||
3786 | for (i = 0; i < nr_vms; i++) | |
3787 | free_vm_area(vms[i]); | |
3788 | kfree(vms); | |
3789 | } | |
4f8b02b4 | 3790 | #endif /* CONFIG_SMP */ |
a10aa579 | 3791 | |
5bb1bb35 | 3792 | #ifdef CONFIG_PRINTK |
98f18083 PM |
3793 | bool vmalloc_dump_obj(void *object) |
3794 | { | |
3795 | struct vm_struct *vm; | |
3796 | void *objp = (void *)PAGE_ALIGN((unsigned long)object); | |
3797 | ||
3798 | vm = find_vm_area(objp); | |
3799 | if (!vm) | |
3800 | return false; | |
bd34dcd4 PM |
3801 | pr_cont(" %u-page vmalloc region starting at %#lx allocated at %pS\n", |
3802 | vm->nr_pages, (unsigned long)vm->addr, vm->caller); | |
98f18083 PM |
3803 | return true; |
3804 | } | |
5bb1bb35 | 3805 | #endif |
98f18083 | 3806 | |
a10aa579 CL |
3807 | #ifdef CONFIG_PROC_FS |
3808 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e36176be | 3809 | __acquires(&vmap_purge_lock) |
d4033afd | 3810 | __acquires(&vmap_area_lock) |
a10aa579 | 3811 | { |
e36176be | 3812 | mutex_lock(&vmap_purge_lock); |
d4033afd | 3813 | spin_lock(&vmap_area_lock); |
e36176be | 3814 | |
3f500069 | 3815 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3816 | } |
3817 | ||
3818 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3819 | { | |
3f500069 | 3820 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3821 | } |
3822 | ||
3823 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 3824 | __releases(&vmap_area_lock) |
0a7dd4e9 | 3825 | __releases(&vmap_purge_lock) |
a10aa579 | 3826 | { |
d4033afd | 3827 | spin_unlock(&vmap_area_lock); |
0a7dd4e9 | 3828 | mutex_unlock(&vmap_purge_lock); |
a10aa579 CL |
3829 | } |
3830 | ||
a47a126a ED |
3831 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3832 | { | |
e5adfffc | 3833 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3834 | unsigned int nr, *counters = m->private; |
3835 | ||
3836 | if (!counters) | |
3837 | return; | |
3838 | ||
af12346c WL |
3839 | if (v->flags & VM_UNINITIALIZED) |
3840 | return; | |
7e5b528b DV |
3841 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3842 | smp_rmb(); | |
af12346c | 3843 | |
a47a126a ED |
3844 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3845 | ||
3846 | for (nr = 0; nr < v->nr_pages; nr++) | |
3847 | counters[page_to_nid(v->pages[nr])]++; | |
3848 | ||
3849 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3850 | if (counters[nr]) | |
3851 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3852 | } | |
3853 | } | |
3854 | ||
dd3b8353 URS |
3855 | static void show_purge_info(struct seq_file *m) |
3856 | { | |
dd3b8353 URS |
3857 | struct vmap_area *va; |
3858 | ||
96e2db45 URS |
3859 | spin_lock(&purge_vmap_area_lock); |
3860 | list_for_each_entry(va, &purge_vmap_area_list, list) { | |
dd3b8353 URS |
3861 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", |
3862 | (void *)va->va_start, (void *)va->va_end, | |
3863 | va->va_end - va->va_start); | |
3864 | } | |
96e2db45 | 3865 | spin_unlock(&purge_vmap_area_lock); |
dd3b8353 URS |
3866 | } |
3867 | ||
a10aa579 CL |
3868 | static int s_show(struct seq_file *m, void *p) |
3869 | { | |
3f500069 | 3870 | struct vmap_area *va; |
d4033afd JK |
3871 | struct vm_struct *v; |
3872 | ||
3f500069 | 3873 | va = list_entry(p, struct vmap_area, list); |
3874 | ||
c2ce8c14 | 3875 | /* |
688fcbfc PL |
3876 | * s_show can encounter race with remove_vm_area, !vm on behalf |
3877 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 3878 | */ |
688fcbfc | 3879 | if (!va->vm) { |
dd3b8353 | 3880 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 3881 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 3882 | va->va_end - va->va_start); |
78c72746 | 3883 | |
d4033afd | 3884 | return 0; |
78c72746 | 3885 | } |
d4033afd JK |
3886 | |
3887 | v = va->vm; | |
a10aa579 | 3888 | |
45ec1690 | 3889 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3890 | v->addr, v->addr + v->size, v->size); |
3891 | ||
62c70bce JP |
3892 | if (v->caller) |
3893 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3894 | |
a10aa579 CL |
3895 | if (v->nr_pages) |
3896 | seq_printf(m, " pages=%d", v->nr_pages); | |
3897 | ||
3898 | if (v->phys_addr) | |
199eaa05 | 3899 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3900 | |
3901 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3902 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3903 | |
3904 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3905 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3906 | |
3907 | if (v->flags & VM_MAP) | |
f4527c90 | 3908 | seq_puts(m, " vmap"); |
a10aa579 CL |
3909 | |
3910 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3911 | seq_puts(m, " user"); |
a10aa579 | 3912 | |
fe9041c2 CH |
3913 | if (v->flags & VM_DMA_COHERENT) |
3914 | seq_puts(m, " dma-coherent"); | |
3915 | ||
244d63ee | 3916 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3917 | seq_puts(m, " vpages"); |
a10aa579 | 3918 | |
a47a126a | 3919 | show_numa_info(m, v); |
a10aa579 | 3920 | seq_putc(m, '\n'); |
dd3b8353 URS |
3921 | |
3922 | /* | |
96e2db45 | 3923 | * As a final step, dump "unpurged" areas. |
dd3b8353 URS |
3924 | */ |
3925 | if (list_is_last(&va->list, &vmap_area_list)) | |
3926 | show_purge_info(m); | |
3927 | ||
a10aa579 CL |
3928 | return 0; |
3929 | } | |
3930 | ||
5f6a6a9c | 3931 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3932 | .start = s_start, |
3933 | .next = s_next, | |
3934 | .stop = s_stop, | |
3935 | .show = s_show, | |
3936 | }; | |
5f6a6a9c | 3937 | |
5f6a6a9c AD |
3938 | static int __init proc_vmalloc_init(void) |
3939 | { | |
fddda2b7 | 3940 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3941 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3942 | &vmalloc_op, |
3943 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3944 | else |
0825a6f9 | 3945 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3946 | return 0; |
3947 | } | |
3948 | module_init(proc_vmalloc_init); | |
db3808c1 | 3949 | |
a10aa579 | 3950 | #endif |