]>
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> |
5da96bdd | 28 | #include <linux/io.h> |
db64fe02 | 29 | #include <linux/rcupdate.h> |
f0aa6617 | 30 | #include <linux/pfn.h> |
89219d37 | 31 | #include <linux/kmemleak.h> |
60063497 | 32 | #include <linux/atomic.h> |
3b32123d | 33 | #include <linux/compiler.h> |
4e5aa1f4 | 34 | #include <linux/memcontrol.h> |
32fcfd40 | 35 | #include <linux/llist.h> |
0f616be1 | 36 | #include <linux/bitops.h> |
68ad4a33 | 37 | #include <linux/rbtree_augmented.h> |
bdebd6a2 | 38 | #include <linux/overflow.h> |
c0eb315a | 39 | #include <linux/pgtable.h> |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
f7ee1f13 | 41 | #include <linux/hugetlb.h> |
451769eb | 42 | #include <linux/sched/mm.h> |
1da177e4 | 43 | #include <asm/tlbflush.h> |
2dca6999 | 44 | #include <asm/shmparam.h> |
1da177e4 | 45 | |
dd56b046 | 46 | #include "internal.h" |
2a681cfa | 47 | #include "pgalloc-track.h" |
dd56b046 | 48 | |
82a70ce0 CH |
49 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP |
50 | static unsigned int __ro_after_init ioremap_max_page_shift = BITS_PER_LONG - 1; | |
51 | ||
52 | static int __init set_nohugeiomap(char *str) | |
53 | { | |
54 | ioremap_max_page_shift = PAGE_SHIFT; | |
55 | return 0; | |
56 | } | |
57 | early_param("nohugeiomap", set_nohugeiomap); | |
58 | #else /* CONFIG_HAVE_ARCH_HUGE_VMAP */ | |
59 | static const unsigned int ioremap_max_page_shift = PAGE_SHIFT; | |
60 | #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */ | |
61 | ||
121e6f32 NP |
62 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC |
63 | static bool __ro_after_init vmap_allow_huge = true; | |
64 | ||
65 | static int __init set_nohugevmalloc(char *str) | |
66 | { | |
67 | vmap_allow_huge = false; | |
68 | return 0; | |
69 | } | |
70 | early_param("nohugevmalloc", set_nohugevmalloc); | |
71 | #else /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */ | |
72 | static const bool vmap_allow_huge = false; | |
73 | #endif /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */ | |
74 | ||
186525bd IM |
75 | bool is_vmalloc_addr(const void *x) |
76 | { | |
4aff1dc4 | 77 | unsigned long addr = (unsigned long)kasan_reset_tag(x); |
186525bd IM |
78 | |
79 | return addr >= VMALLOC_START && addr < VMALLOC_END; | |
80 | } | |
81 | EXPORT_SYMBOL(is_vmalloc_addr); | |
82 | ||
32fcfd40 AV |
83 | struct vfree_deferred { |
84 | struct llist_head list; | |
85 | struct work_struct wq; | |
86 | }; | |
87 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
88 | ||
89 | static void __vunmap(const void *, int); | |
90 | ||
91 | static void free_work(struct work_struct *w) | |
92 | { | |
93 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
94 | struct llist_node *t, *llnode; |
95 | ||
96 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
97 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
98 | } |
99 | ||
db64fe02 | 100 | /*** Page table manipulation functions ***/ |
5e9e3d77 NP |
101 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
102 | phys_addr_t phys_addr, pgprot_t prot, | |
f7ee1f13 | 103 | unsigned int max_page_shift, pgtbl_mod_mask *mask) |
5e9e3d77 NP |
104 | { |
105 | pte_t *pte; | |
106 | u64 pfn; | |
f7ee1f13 | 107 | unsigned long size = PAGE_SIZE; |
5e9e3d77 NP |
108 | |
109 | pfn = phys_addr >> PAGE_SHIFT; | |
110 | pte = pte_alloc_kernel_track(pmd, addr, mask); | |
111 | if (!pte) | |
112 | return -ENOMEM; | |
113 | do { | |
114 | BUG_ON(!pte_none(*pte)); | |
f7ee1f13 CL |
115 | |
116 | #ifdef CONFIG_HUGETLB_PAGE | |
117 | size = arch_vmap_pte_range_map_size(addr, end, pfn, max_page_shift); | |
118 | if (size != PAGE_SIZE) { | |
119 | pte_t entry = pfn_pte(pfn, prot); | |
120 | ||
f7ee1f13 CL |
121 | entry = arch_make_huge_pte(entry, ilog2(size), 0); |
122 | set_huge_pte_at(&init_mm, addr, pte, entry); | |
123 | pfn += PFN_DOWN(size); | |
124 | continue; | |
125 | } | |
126 | #endif | |
5e9e3d77 NP |
127 | set_pte_at(&init_mm, addr, pte, pfn_pte(pfn, prot)); |
128 | pfn++; | |
f7ee1f13 | 129 | } while (pte += PFN_DOWN(size), addr += size, addr != end); |
5e9e3d77 NP |
130 | *mask |= PGTBL_PTE_MODIFIED; |
131 | return 0; | |
132 | } | |
133 | ||
134 | static int vmap_try_huge_pmd(pmd_t *pmd, unsigned long addr, unsigned long end, | |
135 | phys_addr_t phys_addr, pgprot_t prot, | |
136 | unsigned int max_page_shift) | |
137 | { | |
138 | if (max_page_shift < PMD_SHIFT) | |
139 | return 0; | |
140 | ||
141 | if (!arch_vmap_pmd_supported(prot)) | |
142 | return 0; | |
143 | ||
144 | if ((end - addr) != PMD_SIZE) | |
145 | return 0; | |
146 | ||
147 | if (!IS_ALIGNED(addr, PMD_SIZE)) | |
148 | return 0; | |
149 | ||
150 | if (!IS_ALIGNED(phys_addr, PMD_SIZE)) | |
151 | return 0; | |
152 | ||
153 | if (pmd_present(*pmd) && !pmd_free_pte_page(pmd, addr)) | |
154 | return 0; | |
155 | ||
156 | return pmd_set_huge(pmd, phys_addr, prot); | |
157 | } | |
158 | ||
159 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, | |
160 | phys_addr_t phys_addr, pgprot_t prot, | |
161 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
162 | { | |
163 | pmd_t *pmd; | |
164 | unsigned long next; | |
165 | ||
166 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); | |
167 | if (!pmd) | |
168 | return -ENOMEM; | |
169 | do { | |
170 | next = pmd_addr_end(addr, end); | |
171 | ||
172 | if (vmap_try_huge_pmd(pmd, addr, next, phys_addr, prot, | |
173 | max_page_shift)) { | |
174 | *mask |= PGTBL_PMD_MODIFIED; | |
175 | continue; | |
176 | } | |
177 | ||
f7ee1f13 | 178 | if (vmap_pte_range(pmd, addr, next, phys_addr, prot, max_page_shift, mask)) |
5e9e3d77 NP |
179 | return -ENOMEM; |
180 | } while (pmd++, phys_addr += (next - addr), addr = next, addr != end); | |
181 | return 0; | |
182 | } | |
183 | ||
184 | static int vmap_try_huge_pud(pud_t *pud, unsigned long addr, unsigned long end, | |
185 | phys_addr_t phys_addr, pgprot_t prot, | |
186 | unsigned int max_page_shift) | |
187 | { | |
188 | if (max_page_shift < PUD_SHIFT) | |
189 | return 0; | |
190 | ||
191 | if (!arch_vmap_pud_supported(prot)) | |
192 | return 0; | |
193 | ||
194 | if ((end - addr) != PUD_SIZE) | |
195 | return 0; | |
196 | ||
197 | if (!IS_ALIGNED(addr, PUD_SIZE)) | |
198 | return 0; | |
199 | ||
200 | if (!IS_ALIGNED(phys_addr, PUD_SIZE)) | |
201 | return 0; | |
202 | ||
203 | if (pud_present(*pud) && !pud_free_pmd_page(pud, addr)) | |
204 | return 0; | |
205 | ||
206 | return pud_set_huge(pud, phys_addr, prot); | |
207 | } | |
208 | ||
209 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, | |
210 | phys_addr_t phys_addr, pgprot_t prot, | |
211 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
212 | { | |
213 | pud_t *pud; | |
214 | unsigned long next; | |
215 | ||
216 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); | |
217 | if (!pud) | |
218 | return -ENOMEM; | |
219 | do { | |
220 | next = pud_addr_end(addr, end); | |
221 | ||
222 | if (vmap_try_huge_pud(pud, addr, next, phys_addr, prot, | |
223 | max_page_shift)) { | |
224 | *mask |= PGTBL_PUD_MODIFIED; | |
225 | continue; | |
226 | } | |
227 | ||
228 | if (vmap_pmd_range(pud, addr, next, phys_addr, prot, | |
229 | max_page_shift, mask)) | |
230 | return -ENOMEM; | |
231 | } while (pud++, phys_addr += (next - addr), addr = next, addr != end); | |
232 | return 0; | |
233 | } | |
234 | ||
235 | static int vmap_try_huge_p4d(p4d_t *p4d, unsigned long addr, unsigned long end, | |
236 | phys_addr_t phys_addr, pgprot_t prot, | |
237 | unsigned int max_page_shift) | |
238 | { | |
239 | if (max_page_shift < P4D_SHIFT) | |
240 | return 0; | |
241 | ||
242 | if (!arch_vmap_p4d_supported(prot)) | |
243 | return 0; | |
244 | ||
245 | if ((end - addr) != P4D_SIZE) | |
246 | return 0; | |
247 | ||
248 | if (!IS_ALIGNED(addr, P4D_SIZE)) | |
249 | return 0; | |
250 | ||
251 | if (!IS_ALIGNED(phys_addr, P4D_SIZE)) | |
252 | return 0; | |
253 | ||
254 | if (p4d_present(*p4d) && !p4d_free_pud_page(p4d, addr)) | |
255 | return 0; | |
256 | ||
257 | return p4d_set_huge(p4d, phys_addr, prot); | |
258 | } | |
259 | ||
260 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, | |
261 | phys_addr_t phys_addr, pgprot_t prot, | |
262 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
263 | { | |
264 | p4d_t *p4d; | |
265 | unsigned long next; | |
266 | ||
267 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); | |
268 | if (!p4d) | |
269 | return -ENOMEM; | |
270 | do { | |
271 | next = p4d_addr_end(addr, end); | |
272 | ||
273 | if (vmap_try_huge_p4d(p4d, addr, next, phys_addr, prot, | |
274 | max_page_shift)) { | |
275 | *mask |= PGTBL_P4D_MODIFIED; | |
276 | continue; | |
277 | } | |
278 | ||
279 | if (vmap_pud_range(p4d, addr, next, phys_addr, prot, | |
280 | max_page_shift, mask)) | |
281 | return -ENOMEM; | |
282 | } while (p4d++, phys_addr += (next - addr), addr = next, addr != end); | |
283 | return 0; | |
284 | } | |
285 | ||
5d87510d | 286 | static int vmap_range_noflush(unsigned long addr, unsigned long end, |
5e9e3d77 NP |
287 | phys_addr_t phys_addr, pgprot_t prot, |
288 | unsigned int max_page_shift) | |
289 | { | |
290 | pgd_t *pgd; | |
291 | unsigned long start; | |
292 | unsigned long next; | |
293 | int err; | |
294 | pgtbl_mod_mask mask = 0; | |
295 | ||
296 | might_sleep(); | |
297 | BUG_ON(addr >= end); | |
298 | ||
299 | start = addr; | |
300 | pgd = pgd_offset_k(addr); | |
301 | do { | |
302 | next = pgd_addr_end(addr, end); | |
303 | err = vmap_p4d_range(pgd, addr, next, phys_addr, prot, | |
304 | max_page_shift, &mask); | |
305 | if (err) | |
306 | break; | |
307 | } while (pgd++, phys_addr += (next - addr), addr = next, addr != end); | |
308 | ||
5e9e3d77 NP |
309 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
310 | arch_sync_kernel_mappings(start, end); | |
311 | ||
312 | return err; | |
313 | } | |
b221385b | 314 | |
82a70ce0 CH |
315 | int ioremap_page_range(unsigned long addr, unsigned long end, |
316 | phys_addr_t phys_addr, pgprot_t prot) | |
5d87510d NP |
317 | { |
318 | int err; | |
319 | ||
8491502f | 320 | err = vmap_range_noflush(addr, end, phys_addr, pgprot_nx(prot), |
82a70ce0 | 321 | ioremap_max_page_shift); |
5d87510d | 322 | flush_cache_vmap(addr, end); |
5d87510d NP |
323 | return err; |
324 | } | |
325 | ||
2ba3e694 JR |
326 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
327 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
328 | { |
329 | pte_t *pte; | |
330 | ||
331 | pte = pte_offset_kernel(pmd, addr); | |
332 | do { | |
333 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
334 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
335 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
2ba3e694 | 336 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
337 | } |
338 | ||
2ba3e694 JR |
339 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, |
340 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
341 | { |
342 | pmd_t *pmd; | |
343 | unsigned long next; | |
2ba3e694 | 344 | int cleared; |
1da177e4 LT |
345 | |
346 | pmd = pmd_offset(pud, addr); | |
347 | do { | |
348 | next = pmd_addr_end(addr, end); | |
2ba3e694 JR |
349 | |
350 | cleared = pmd_clear_huge(pmd); | |
351 | if (cleared || pmd_bad(*pmd)) | |
352 | *mask |= PGTBL_PMD_MODIFIED; | |
353 | ||
354 | if (cleared) | |
b9820d8f | 355 | continue; |
1da177e4 LT |
356 | if (pmd_none_or_clear_bad(pmd)) |
357 | continue; | |
2ba3e694 | 358 | vunmap_pte_range(pmd, addr, next, mask); |
e47110e9 AK |
359 | |
360 | cond_resched(); | |
1da177e4 LT |
361 | } while (pmd++, addr = next, addr != end); |
362 | } | |
363 | ||
2ba3e694 JR |
364 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, |
365 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
366 | { |
367 | pud_t *pud; | |
368 | unsigned long next; | |
2ba3e694 | 369 | int cleared; |
1da177e4 | 370 | |
c2febafc | 371 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
372 | do { |
373 | next = pud_addr_end(addr, end); | |
2ba3e694 JR |
374 | |
375 | cleared = pud_clear_huge(pud); | |
376 | if (cleared || pud_bad(*pud)) | |
377 | *mask |= PGTBL_PUD_MODIFIED; | |
378 | ||
379 | if (cleared) | |
b9820d8f | 380 | continue; |
1da177e4 LT |
381 | if (pud_none_or_clear_bad(pud)) |
382 | continue; | |
2ba3e694 | 383 | vunmap_pmd_range(pud, addr, next, mask); |
1da177e4 LT |
384 | } while (pud++, addr = next, addr != end); |
385 | } | |
386 | ||
2ba3e694 JR |
387 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, |
388 | pgtbl_mod_mask *mask) | |
c2febafc KS |
389 | { |
390 | p4d_t *p4d; | |
391 | unsigned long next; | |
392 | ||
393 | p4d = p4d_offset(pgd, addr); | |
394 | do { | |
395 | next = p4d_addr_end(addr, end); | |
2ba3e694 | 396 | |
c8db8c26 L |
397 | p4d_clear_huge(p4d); |
398 | if (p4d_bad(*p4d)) | |
2ba3e694 JR |
399 | *mask |= PGTBL_P4D_MODIFIED; |
400 | ||
c2febafc KS |
401 | if (p4d_none_or_clear_bad(p4d)) |
402 | continue; | |
2ba3e694 | 403 | vunmap_pud_range(p4d, addr, next, mask); |
c2febafc KS |
404 | } while (p4d++, addr = next, addr != end); |
405 | } | |
406 | ||
4ad0ae8c NP |
407 | /* |
408 | * vunmap_range_noflush is similar to vunmap_range, but does not | |
409 | * flush caches or TLBs. | |
b521c43f | 410 | * |
4ad0ae8c NP |
411 | * The caller is responsible for calling flush_cache_vmap() before calling |
412 | * this function, and flush_tlb_kernel_range after it has returned | |
413 | * successfully (and before the addresses are expected to cause a page fault | |
414 | * or be re-mapped for something else, if TLB flushes are being delayed or | |
415 | * coalesced). | |
b521c43f | 416 | * |
4ad0ae8c | 417 | * This is an internal function only. Do not use outside mm/. |
b521c43f | 418 | */ |
4ad0ae8c | 419 | void vunmap_range_noflush(unsigned long start, unsigned long end) |
1da177e4 | 420 | { |
1da177e4 | 421 | unsigned long next; |
b521c43f | 422 | pgd_t *pgd; |
2ba3e694 JR |
423 | unsigned long addr = start; |
424 | pgtbl_mod_mask mask = 0; | |
1da177e4 LT |
425 | |
426 | BUG_ON(addr >= end); | |
427 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
428 | do { |
429 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
430 | if (pgd_bad(*pgd)) |
431 | mask |= PGTBL_PGD_MODIFIED; | |
1da177e4 LT |
432 | if (pgd_none_or_clear_bad(pgd)) |
433 | continue; | |
2ba3e694 | 434 | vunmap_p4d_range(pgd, addr, next, &mask); |
1da177e4 | 435 | } while (pgd++, addr = next, addr != end); |
2ba3e694 JR |
436 | |
437 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) | |
438 | arch_sync_kernel_mappings(start, end); | |
1da177e4 LT |
439 | } |
440 | ||
4ad0ae8c NP |
441 | /** |
442 | * vunmap_range - unmap kernel virtual addresses | |
443 | * @addr: start of the VM area to unmap | |
444 | * @end: end of the VM area to unmap (non-inclusive) | |
445 | * | |
446 | * Clears any present PTEs in the virtual address range, flushes TLBs and | |
447 | * caches. Any subsequent access to the address before it has been re-mapped | |
448 | * is a kernel bug. | |
449 | */ | |
450 | void vunmap_range(unsigned long addr, unsigned long end) | |
451 | { | |
452 | flush_cache_vunmap(addr, end); | |
453 | vunmap_range_noflush(addr, end); | |
454 | flush_tlb_kernel_range(addr, end); | |
455 | } | |
456 | ||
0a264884 | 457 | static int vmap_pages_pte_range(pmd_t *pmd, unsigned long addr, |
2ba3e694 JR |
458 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
459 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
460 | { |
461 | pte_t *pte; | |
462 | ||
db64fe02 NP |
463 | /* |
464 | * nr is a running index into the array which helps higher level | |
465 | * callers keep track of where we're up to. | |
466 | */ | |
467 | ||
2ba3e694 | 468 | pte = pte_alloc_kernel_track(pmd, addr, mask); |
1da177e4 LT |
469 | if (!pte) |
470 | return -ENOMEM; | |
471 | do { | |
db64fe02 NP |
472 | struct page *page = pages[*nr]; |
473 | ||
474 | if (WARN_ON(!pte_none(*pte))) | |
475 | return -EBUSY; | |
476 | if (WARN_ON(!page)) | |
1da177e4 | 477 | return -ENOMEM; |
4fcdcc12 YN |
478 | if (WARN_ON(!pfn_valid(page_to_pfn(page)))) |
479 | return -EINVAL; | |
480 | ||
1da177e4 | 481 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); |
db64fe02 | 482 | (*nr)++; |
1da177e4 | 483 | } while (pte++, addr += PAGE_SIZE, addr != end); |
2ba3e694 | 484 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
485 | return 0; |
486 | } | |
487 | ||
0a264884 | 488 | static int vmap_pages_pmd_range(pud_t *pud, unsigned long addr, |
2ba3e694 JR |
489 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
490 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
491 | { |
492 | pmd_t *pmd; | |
493 | unsigned long next; | |
494 | ||
2ba3e694 | 495 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); |
1da177e4 LT |
496 | if (!pmd) |
497 | return -ENOMEM; | |
498 | do { | |
499 | next = pmd_addr_end(addr, end); | |
0a264884 | 500 | if (vmap_pages_pte_range(pmd, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
501 | return -ENOMEM; |
502 | } while (pmd++, addr = next, addr != end); | |
503 | return 0; | |
504 | } | |
505 | ||
0a264884 | 506 | static int vmap_pages_pud_range(p4d_t *p4d, unsigned long addr, |
2ba3e694 JR |
507 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
508 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
509 | { |
510 | pud_t *pud; | |
511 | unsigned long next; | |
512 | ||
2ba3e694 | 513 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); |
1da177e4 LT |
514 | if (!pud) |
515 | return -ENOMEM; | |
516 | do { | |
517 | next = pud_addr_end(addr, end); | |
0a264884 | 518 | if (vmap_pages_pmd_range(pud, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
519 | return -ENOMEM; |
520 | } while (pud++, addr = next, addr != end); | |
521 | return 0; | |
522 | } | |
523 | ||
0a264884 | 524 | static int vmap_pages_p4d_range(pgd_t *pgd, unsigned long addr, |
2ba3e694 JR |
525 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
526 | pgtbl_mod_mask *mask) | |
c2febafc KS |
527 | { |
528 | p4d_t *p4d; | |
529 | unsigned long next; | |
530 | ||
2ba3e694 | 531 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); |
c2febafc KS |
532 | if (!p4d) |
533 | return -ENOMEM; | |
534 | do { | |
535 | next = p4d_addr_end(addr, end); | |
0a264884 | 536 | if (vmap_pages_pud_range(p4d, addr, next, prot, pages, nr, mask)) |
c2febafc KS |
537 | return -ENOMEM; |
538 | } while (p4d++, addr = next, addr != end); | |
539 | return 0; | |
540 | } | |
541 | ||
121e6f32 NP |
542 | static int vmap_small_pages_range_noflush(unsigned long addr, unsigned long end, |
543 | pgprot_t prot, struct page **pages) | |
1da177e4 | 544 | { |
2ba3e694 | 545 | unsigned long start = addr; |
b521c43f | 546 | pgd_t *pgd; |
121e6f32 | 547 | unsigned long next; |
db64fe02 NP |
548 | int err = 0; |
549 | int nr = 0; | |
2ba3e694 | 550 | pgtbl_mod_mask mask = 0; |
1da177e4 LT |
551 | |
552 | BUG_ON(addr >= end); | |
553 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
554 | do { |
555 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
556 | if (pgd_bad(*pgd)) |
557 | mask |= PGTBL_PGD_MODIFIED; | |
0a264884 | 558 | err = vmap_pages_p4d_range(pgd, addr, next, prot, pages, &nr, &mask); |
1da177e4 | 559 | if (err) |
bf88c8c8 | 560 | return err; |
1da177e4 | 561 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 562 | |
2ba3e694 JR |
563 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
564 | arch_sync_kernel_mappings(start, end); | |
565 | ||
60bb4465 | 566 | return 0; |
1da177e4 LT |
567 | } |
568 | ||
b67177ec NP |
569 | /* |
570 | * vmap_pages_range_noflush is similar to vmap_pages_range, but does not | |
571 | * flush caches. | |
572 | * | |
573 | * The caller is responsible for calling flush_cache_vmap() after this | |
574 | * function returns successfully and before the addresses are accessed. | |
575 | * | |
576 | * This is an internal function only. Do not use outside mm/. | |
577 | */ | |
578 | int vmap_pages_range_noflush(unsigned long addr, unsigned long end, | |
121e6f32 NP |
579 | pgprot_t prot, struct page **pages, unsigned int page_shift) |
580 | { | |
581 | unsigned int i, nr = (end - addr) >> PAGE_SHIFT; | |
582 | ||
583 | WARN_ON(page_shift < PAGE_SHIFT); | |
584 | ||
585 | if (!IS_ENABLED(CONFIG_HAVE_ARCH_HUGE_VMALLOC) || | |
586 | page_shift == PAGE_SHIFT) | |
587 | return vmap_small_pages_range_noflush(addr, end, prot, pages); | |
588 | ||
589 | for (i = 0; i < nr; i += 1U << (page_shift - PAGE_SHIFT)) { | |
590 | int err; | |
591 | ||
592 | err = vmap_range_noflush(addr, addr + (1UL << page_shift), | |
593 | __pa(page_address(pages[i])), prot, | |
594 | page_shift); | |
595 | if (err) | |
596 | return err; | |
597 | ||
598 | addr += 1UL << page_shift; | |
599 | } | |
600 | ||
601 | return 0; | |
602 | } | |
603 | ||
121e6f32 | 604 | /** |
b67177ec | 605 | * vmap_pages_range - map pages to a kernel virtual address |
121e6f32 | 606 | * @addr: start of the VM area to map |
b67177ec | 607 | * @end: end of the VM area to map (non-inclusive) |
121e6f32 | 608 | * @prot: page protection flags to use |
b67177ec NP |
609 | * @pages: pages to map (always PAGE_SIZE pages) |
610 | * @page_shift: maximum shift that the pages may be mapped with, @pages must | |
611 | * be aligned and contiguous up to at least this shift. | |
121e6f32 NP |
612 | * |
613 | * RETURNS: | |
614 | * 0 on success, -errno on failure. | |
615 | */ | |
b67177ec NP |
616 | static int vmap_pages_range(unsigned long addr, unsigned long end, |
617 | pgprot_t prot, struct page **pages, unsigned int page_shift) | |
8fc48985 | 618 | { |
b67177ec | 619 | int err; |
8fc48985 | 620 | |
b67177ec NP |
621 | err = vmap_pages_range_noflush(addr, end, prot, pages, page_shift); |
622 | flush_cache_vmap(addr, end); | |
623 | return err; | |
8fc48985 TH |
624 | } |
625 | ||
81ac3ad9 | 626 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
627 | { |
628 | /* | |
ab4f2ee1 | 629 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
630 | * and fall back on vmalloc() if that fails. Others |
631 | * just put it in the vmalloc space. | |
632 | */ | |
633 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
4aff1dc4 | 634 | unsigned long addr = (unsigned long)kasan_reset_tag(x); |
73bdf0a6 LT |
635 | if (addr >= MODULES_VADDR && addr < MODULES_END) |
636 | return 1; | |
637 | #endif | |
638 | return is_vmalloc_addr(x); | |
639 | } | |
640 | ||
48667e7a | 641 | /* |
c0eb315a NP |
642 | * Walk a vmap address to the struct page it maps. Huge vmap mappings will |
643 | * return the tail page that corresponds to the base page address, which | |
644 | * matches small vmap mappings. | |
48667e7a | 645 | */ |
add688fb | 646 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
647 | { |
648 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 649 | struct page *page = NULL; |
48667e7a | 650 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
651 | p4d_t *p4d; |
652 | pud_t *pud; | |
653 | pmd_t *pmd; | |
654 | pte_t *ptep, pte; | |
48667e7a | 655 | |
7aa413de IM |
656 | /* |
657 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
658 | * architectures that do not vmalloc module space | |
659 | */ | |
73bdf0a6 | 660 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 661 | |
c2febafc KS |
662 | if (pgd_none(*pgd)) |
663 | return NULL; | |
c0eb315a NP |
664 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
665 | return NULL; /* XXX: no allowance for huge pgd */ | |
666 | if (WARN_ON_ONCE(pgd_bad(*pgd))) | |
667 | return NULL; | |
668 | ||
c2febafc KS |
669 | p4d = p4d_offset(pgd, addr); |
670 | if (p4d_none(*p4d)) | |
671 | return NULL; | |
c0eb315a NP |
672 | if (p4d_leaf(*p4d)) |
673 | return p4d_page(*p4d) + ((addr & ~P4D_MASK) >> PAGE_SHIFT); | |
674 | if (WARN_ON_ONCE(p4d_bad(*p4d))) | |
675 | return NULL; | |
029c54b0 | 676 | |
c0eb315a NP |
677 | pud = pud_offset(p4d, addr); |
678 | if (pud_none(*pud)) | |
679 | return NULL; | |
680 | if (pud_leaf(*pud)) | |
681 | return pud_page(*pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
682 | if (WARN_ON_ONCE(pud_bad(*pud))) | |
c2febafc | 683 | return NULL; |
c0eb315a | 684 | |
c2febafc | 685 | pmd = pmd_offset(pud, addr); |
c0eb315a NP |
686 | if (pmd_none(*pmd)) |
687 | return NULL; | |
688 | if (pmd_leaf(*pmd)) | |
689 | return pmd_page(*pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
690 | if (WARN_ON_ONCE(pmd_bad(*pmd))) | |
c2febafc KS |
691 | return NULL; |
692 | ||
693 | ptep = pte_offset_map(pmd, addr); | |
694 | pte = *ptep; | |
695 | if (pte_present(pte)) | |
696 | page = pte_page(pte); | |
697 | pte_unmap(ptep); | |
c0eb315a | 698 | |
add688fb | 699 | return page; |
48667e7a | 700 | } |
add688fb | 701 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
702 | |
703 | /* | |
add688fb | 704 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 705 | */ |
add688fb | 706 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 707 | { |
add688fb | 708 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 709 | } |
add688fb | 710 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 711 | |
db64fe02 NP |
712 | |
713 | /*** Global kva allocator ***/ | |
714 | ||
bb850f4d | 715 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 716 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 717 | |
db64fe02 | 718 | |
db64fe02 | 719 | static DEFINE_SPINLOCK(vmap_area_lock); |
e36176be | 720 | static DEFINE_SPINLOCK(free_vmap_area_lock); |
f1c4069e JK |
721 | /* Export for kexec only */ |
722 | LIST_HEAD(vmap_area_list); | |
89699605 | 723 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 724 | static bool vmap_initialized __read_mostly; |
89699605 | 725 | |
96e2db45 URS |
726 | static struct rb_root purge_vmap_area_root = RB_ROOT; |
727 | static LIST_HEAD(purge_vmap_area_list); | |
728 | static DEFINE_SPINLOCK(purge_vmap_area_lock); | |
729 | ||
68ad4a33 URS |
730 | /* |
731 | * This kmem_cache is used for vmap_area objects. Instead of | |
732 | * allocating from slab we reuse an object from this cache to | |
733 | * make things faster. Especially in "no edge" splitting of | |
734 | * free block. | |
735 | */ | |
736 | static struct kmem_cache *vmap_area_cachep; | |
737 | ||
738 | /* | |
739 | * This linked list is used in pair with free_vmap_area_root. | |
740 | * It gives O(1) access to prev/next to perform fast coalescing. | |
741 | */ | |
742 | static LIST_HEAD(free_vmap_area_list); | |
743 | ||
744 | /* | |
745 | * This augment red-black tree represents the free vmap space. | |
746 | * All vmap_area objects in this tree are sorted by va->va_start | |
747 | * address. It is used for allocation and merging when a vmap | |
748 | * object is released. | |
749 | * | |
750 | * Each vmap_area node contains a maximum available free block | |
751 | * of its sub-tree, right or left. Therefore it is possible to | |
752 | * find a lowest match of free area. | |
753 | */ | |
754 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
755 | ||
82dd23e8 URS |
756 | /* |
757 | * Preload a CPU with one object for "no edge" split case. The | |
758 | * aim is to get rid of allocations from the atomic context, thus | |
759 | * to use more permissive allocation masks. | |
760 | */ | |
761 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
762 | ||
68ad4a33 URS |
763 | static __always_inline unsigned long |
764 | va_size(struct vmap_area *va) | |
765 | { | |
766 | return (va->va_end - va->va_start); | |
767 | } | |
768 | ||
769 | static __always_inline unsigned long | |
770 | get_subtree_max_size(struct rb_node *node) | |
771 | { | |
772 | struct vmap_area *va; | |
773 | ||
774 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
775 | return va ? va->subtree_max_size : 0; | |
776 | } | |
89699605 | 777 | |
315cc066 ML |
778 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
779 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 URS |
780 | |
781 | static void purge_vmap_area_lazy(void); | |
782 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
690467c8 URS |
783 | static void drain_vmap_area_work(struct work_struct *work); |
784 | static DECLARE_WORK(drain_vmap_work, drain_vmap_area_work); | |
db64fe02 | 785 | |
97105f0a RG |
786 | static atomic_long_t nr_vmalloc_pages; |
787 | ||
788 | unsigned long vmalloc_nr_pages(void) | |
789 | { | |
790 | return atomic_long_read(&nr_vmalloc_pages); | |
791 | } | |
792 | ||
f181234a CW |
793 | static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr) |
794 | { | |
795 | struct vmap_area *va = NULL; | |
796 | struct rb_node *n = vmap_area_root.rb_node; | |
797 | ||
4aff1dc4 AK |
798 | addr = (unsigned long)kasan_reset_tag((void *)addr); |
799 | ||
f181234a CW |
800 | while (n) { |
801 | struct vmap_area *tmp; | |
802 | ||
803 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
804 | if (tmp->va_end > addr) { | |
805 | va = tmp; | |
806 | if (tmp->va_start <= addr) | |
807 | break; | |
808 | ||
809 | n = n->rb_left; | |
810 | } else | |
811 | n = n->rb_right; | |
812 | } | |
813 | ||
814 | return va; | |
815 | } | |
816 | ||
db64fe02 | 817 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 818 | { |
db64fe02 NP |
819 | struct rb_node *n = vmap_area_root.rb_node; |
820 | ||
4aff1dc4 AK |
821 | addr = (unsigned long)kasan_reset_tag((void *)addr); |
822 | ||
db64fe02 NP |
823 | while (n) { |
824 | struct vmap_area *va; | |
825 | ||
826 | va = rb_entry(n, struct vmap_area, rb_node); | |
827 | if (addr < va->va_start) | |
828 | n = n->rb_left; | |
cef2ac3f | 829 | else if (addr >= va->va_end) |
db64fe02 NP |
830 | n = n->rb_right; |
831 | else | |
832 | return va; | |
833 | } | |
834 | ||
835 | return NULL; | |
836 | } | |
837 | ||
68ad4a33 URS |
838 | /* |
839 | * This function returns back addresses of parent node | |
840 | * and its left or right link for further processing. | |
9c801f61 URS |
841 | * |
842 | * Otherwise NULL is returned. In that case all further | |
843 | * steps regarding inserting of conflicting overlap range | |
844 | * have to be declined and actually considered as a bug. | |
68ad4a33 URS |
845 | */ |
846 | static __always_inline struct rb_node ** | |
847 | find_va_links(struct vmap_area *va, | |
848 | struct rb_root *root, struct rb_node *from, | |
849 | struct rb_node **parent) | |
850 | { | |
851 | struct vmap_area *tmp_va; | |
852 | struct rb_node **link; | |
853 | ||
854 | if (root) { | |
855 | link = &root->rb_node; | |
856 | if (unlikely(!*link)) { | |
857 | *parent = NULL; | |
858 | return link; | |
859 | } | |
860 | } else { | |
861 | link = &from; | |
862 | } | |
db64fe02 | 863 | |
68ad4a33 URS |
864 | /* |
865 | * Go to the bottom of the tree. When we hit the last point | |
866 | * we end up with parent rb_node and correct direction, i name | |
867 | * it link, where the new va->rb_node will be attached to. | |
868 | */ | |
869 | do { | |
870 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 871 | |
68ad4a33 URS |
872 | /* |
873 | * During the traversal we also do some sanity check. | |
874 | * Trigger the BUG() if there are sides(left/right) | |
875 | * or full overlaps. | |
876 | */ | |
877 | if (va->va_start < tmp_va->va_end && | |
878 | va->va_end <= tmp_va->va_start) | |
879 | link = &(*link)->rb_left; | |
880 | else if (va->va_end > tmp_va->va_start && | |
881 | va->va_start >= tmp_va->va_end) | |
882 | link = &(*link)->rb_right; | |
9c801f61 URS |
883 | else { |
884 | WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n", | |
885 | va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end); | |
886 | ||
887 | return NULL; | |
888 | } | |
68ad4a33 URS |
889 | } while (*link); |
890 | ||
891 | *parent = &tmp_va->rb_node; | |
892 | return link; | |
893 | } | |
894 | ||
895 | static __always_inline struct list_head * | |
896 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
897 | { | |
898 | struct list_head *list; | |
899 | ||
900 | if (unlikely(!parent)) | |
901 | /* | |
902 | * The red-black tree where we try to find VA neighbors | |
903 | * before merging or inserting is empty, i.e. it means | |
904 | * there is no free vmap space. Normally it does not | |
905 | * happen but we handle this case anyway. | |
906 | */ | |
907 | return NULL; | |
908 | ||
909 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
910 | return (&parent->rb_right == link ? list->next : list); | |
911 | } | |
912 | ||
913 | static __always_inline void | |
914 | link_va(struct vmap_area *va, struct rb_root *root, | |
915 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
916 | { | |
917 | /* | |
918 | * VA is still not in the list, but we can | |
919 | * identify its future previous list_head node. | |
920 | */ | |
921 | if (likely(parent)) { | |
922 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
923 | if (&parent->rb_right != link) | |
924 | head = head->prev; | |
db64fe02 NP |
925 | } |
926 | ||
68ad4a33 URS |
927 | /* Insert to the rb-tree */ |
928 | rb_link_node(&va->rb_node, parent, link); | |
929 | if (root == &free_vmap_area_root) { | |
930 | /* | |
931 | * Some explanation here. Just perform simple insertion | |
932 | * to the tree. We do not set va->subtree_max_size to | |
933 | * its current size before calling rb_insert_augmented(). | |
934 | * It is because of we populate the tree from the bottom | |
935 | * to parent levels when the node _is_ in the tree. | |
936 | * | |
937 | * Therefore we set subtree_max_size to zero after insertion, | |
938 | * to let __augment_tree_propagate_from() puts everything to | |
939 | * the correct order later on. | |
940 | */ | |
941 | rb_insert_augmented(&va->rb_node, | |
942 | root, &free_vmap_area_rb_augment_cb); | |
943 | va->subtree_max_size = 0; | |
944 | } else { | |
945 | rb_insert_color(&va->rb_node, root); | |
946 | } | |
db64fe02 | 947 | |
68ad4a33 URS |
948 | /* Address-sort this list */ |
949 | list_add(&va->list, head); | |
db64fe02 NP |
950 | } |
951 | ||
68ad4a33 URS |
952 | static __always_inline void |
953 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
954 | { | |
460e42d1 URS |
955 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
956 | return; | |
db64fe02 | 957 | |
460e42d1 URS |
958 | if (root == &free_vmap_area_root) |
959 | rb_erase_augmented(&va->rb_node, | |
960 | root, &free_vmap_area_rb_augment_cb); | |
961 | else | |
962 | rb_erase(&va->rb_node, root); | |
963 | ||
964 | list_del(&va->list); | |
965 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
966 | } |
967 | ||
bb850f4d | 968 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
c3385e84 JC |
969 | /* |
970 | * Gets called when remove the node and rotate. | |
971 | */ | |
972 | static __always_inline unsigned long | |
973 | compute_subtree_max_size(struct vmap_area *va) | |
974 | { | |
975 | return max3(va_size(va), | |
976 | get_subtree_max_size(va->rb_node.rb_left), | |
977 | get_subtree_max_size(va->rb_node.rb_right)); | |
978 | } | |
979 | ||
bb850f4d | 980 | static void |
da27c9ed | 981 | augment_tree_propagate_check(void) |
bb850f4d URS |
982 | { |
983 | struct vmap_area *va; | |
da27c9ed | 984 | unsigned long computed_size; |
bb850f4d | 985 | |
da27c9ed URS |
986 | list_for_each_entry(va, &free_vmap_area_list, list) { |
987 | computed_size = compute_subtree_max_size(va); | |
988 | if (computed_size != va->subtree_max_size) | |
989 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
990 | va_size(va), va->subtree_max_size); | |
bb850f4d | 991 | } |
bb850f4d URS |
992 | } |
993 | #endif | |
994 | ||
68ad4a33 URS |
995 | /* |
996 | * This function populates subtree_max_size from bottom to upper | |
997 | * levels starting from VA point. The propagation must be done | |
998 | * when VA size is modified by changing its va_start/va_end. Or | |
999 | * in case of newly inserting of VA to the tree. | |
1000 | * | |
1001 | * It means that __augment_tree_propagate_from() must be called: | |
1002 | * - After VA has been inserted to the tree(free path); | |
1003 | * - After VA has been shrunk(allocation path); | |
1004 | * - After VA has been increased(merging path). | |
1005 | * | |
1006 | * Please note that, it does not mean that upper parent nodes | |
1007 | * and their subtree_max_size are recalculated all the time up | |
1008 | * to the root node. | |
1009 | * | |
1010 | * 4--8 | |
1011 | * /\ | |
1012 | * / \ | |
1013 | * / \ | |
1014 | * 2--2 8--8 | |
1015 | * | |
1016 | * For example if we modify the node 4, shrinking it to 2, then | |
1017 | * no any modification is required. If we shrink the node 2 to 1 | |
1018 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
1019 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
1020 | * node becomes 4--6. | |
1021 | */ | |
1022 | static __always_inline void | |
1023 | augment_tree_propagate_from(struct vmap_area *va) | |
1024 | { | |
15ae144f URS |
1025 | /* |
1026 | * Populate the tree from bottom towards the root until | |
1027 | * the calculated maximum available size of checked node | |
1028 | * is equal to its current one. | |
1029 | */ | |
1030 | free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL); | |
bb850f4d URS |
1031 | |
1032 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
da27c9ed | 1033 | augment_tree_propagate_check(); |
bb850f4d | 1034 | #endif |
68ad4a33 URS |
1035 | } |
1036 | ||
1037 | static void | |
1038 | insert_vmap_area(struct vmap_area *va, | |
1039 | struct rb_root *root, struct list_head *head) | |
1040 | { | |
1041 | struct rb_node **link; | |
1042 | struct rb_node *parent; | |
1043 | ||
1044 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
1045 | if (link) |
1046 | link_va(va, root, parent, link, head); | |
68ad4a33 URS |
1047 | } |
1048 | ||
1049 | static void | |
1050 | insert_vmap_area_augment(struct vmap_area *va, | |
1051 | struct rb_node *from, struct rb_root *root, | |
1052 | struct list_head *head) | |
1053 | { | |
1054 | struct rb_node **link; | |
1055 | struct rb_node *parent; | |
1056 | ||
1057 | if (from) | |
1058 | link = find_va_links(va, NULL, from, &parent); | |
1059 | else | |
1060 | link = find_va_links(va, root, NULL, &parent); | |
1061 | ||
9c801f61 URS |
1062 | if (link) { |
1063 | link_va(va, root, parent, link, head); | |
1064 | augment_tree_propagate_from(va); | |
1065 | } | |
68ad4a33 URS |
1066 | } |
1067 | ||
1068 | /* | |
1069 | * Merge de-allocated chunk of VA memory with previous | |
1070 | * and next free blocks. If coalesce is not done a new | |
1071 | * free area is inserted. If VA has been merged, it is | |
1072 | * freed. | |
9c801f61 URS |
1073 | * |
1074 | * Please note, it can return NULL in case of overlap | |
1075 | * ranges, followed by WARN() report. Despite it is a | |
1076 | * buggy behaviour, a system can be alive and keep | |
1077 | * ongoing. | |
68ad4a33 | 1078 | */ |
3c5c3cfb | 1079 | static __always_inline struct vmap_area * |
68ad4a33 URS |
1080 | merge_or_add_vmap_area(struct vmap_area *va, |
1081 | struct rb_root *root, struct list_head *head) | |
1082 | { | |
1083 | struct vmap_area *sibling; | |
1084 | struct list_head *next; | |
1085 | struct rb_node **link; | |
1086 | struct rb_node *parent; | |
1087 | bool merged = false; | |
1088 | ||
1089 | /* | |
1090 | * Find a place in the tree where VA potentially will be | |
1091 | * inserted, unless it is merged with its sibling/siblings. | |
1092 | */ | |
1093 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
1094 | if (!link) |
1095 | return NULL; | |
68ad4a33 URS |
1096 | |
1097 | /* | |
1098 | * Get next node of VA to check if merging can be done. | |
1099 | */ | |
1100 | next = get_va_next_sibling(parent, link); | |
1101 | if (unlikely(next == NULL)) | |
1102 | goto insert; | |
1103 | ||
1104 | /* | |
1105 | * start end | |
1106 | * | | | |
1107 | * |<------VA------>|<-----Next----->| | |
1108 | * | | | |
1109 | * start end | |
1110 | */ | |
1111 | if (next != head) { | |
1112 | sibling = list_entry(next, struct vmap_area, list); | |
1113 | if (sibling->va_start == va->va_end) { | |
1114 | sibling->va_start = va->va_start; | |
1115 | ||
68ad4a33 URS |
1116 | /* Free vmap_area object. */ |
1117 | kmem_cache_free(vmap_area_cachep, va); | |
1118 | ||
1119 | /* Point to the new merged area. */ | |
1120 | va = sibling; | |
1121 | merged = true; | |
1122 | } | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * start end | |
1127 | * | | | |
1128 | * |<-----Prev----->|<------VA------>| | |
1129 | * | | | |
1130 | * start end | |
1131 | */ | |
1132 | if (next->prev != head) { | |
1133 | sibling = list_entry(next->prev, struct vmap_area, list); | |
1134 | if (sibling->va_end == va->va_start) { | |
5dd78640 URS |
1135 | /* |
1136 | * If both neighbors are coalesced, it is important | |
1137 | * to unlink the "next" node first, followed by merging | |
1138 | * with "previous" one. Otherwise the tree might not be | |
1139 | * fully populated if a sibling's augmented value is | |
1140 | * "normalized" because of rotation operations. | |
1141 | */ | |
54f63d9d URS |
1142 | if (merged) |
1143 | unlink_va(va, root); | |
68ad4a33 | 1144 | |
5dd78640 URS |
1145 | sibling->va_end = va->va_end; |
1146 | ||
68ad4a33 URS |
1147 | /* Free vmap_area object. */ |
1148 | kmem_cache_free(vmap_area_cachep, va); | |
3c5c3cfb DA |
1149 | |
1150 | /* Point to the new merged area. */ | |
1151 | va = sibling; | |
1152 | merged = true; | |
68ad4a33 URS |
1153 | } |
1154 | } | |
1155 | ||
1156 | insert: | |
5dd78640 | 1157 | if (!merged) |
68ad4a33 | 1158 | link_va(va, root, parent, link, head); |
3c5c3cfb | 1159 | |
96e2db45 URS |
1160 | return va; |
1161 | } | |
1162 | ||
1163 | static __always_inline struct vmap_area * | |
1164 | merge_or_add_vmap_area_augment(struct vmap_area *va, | |
1165 | struct rb_root *root, struct list_head *head) | |
1166 | { | |
1167 | va = merge_or_add_vmap_area(va, root, head); | |
1168 | if (va) | |
1169 | augment_tree_propagate_from(va); | |
1170 | ||
3c5c3cfb | 1171 | return va; |
68ad4a33 URS |
1172 | } |
1173 | ||
1174 | static __always_inline bool | |
1175 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
1176 | unsigned long align, unsigned long vstart) | |
1177 | { | |
1178 | unsigned long nva_start_addr; | |
1179 | ||
1180 | if (va->va_start > vstart) | |
1181 | nva_start_addr = ALIGN(va->va_start, align); | |
1182 | else | |
1183 | nva_start_addr = ALIGN(vstart, align); | |
1184 | ||
1185 | /* Can be overflowed due to big size or alignment. */ | |
1186 | if (nva_start_addr + size < nva_start_addr || | |
1187 | nva_start_addr < vstart) | |
1188 | return false; | |
1189 | ||
1190 | return (nva_start_addr + size <= va->va_end); | |
1191 | } | |
1192 | ||
1193 | /* | |
1194 | * Find the first free block(lowest start address) in the tree, | |
1195 | * that will accomplish the request corresponding to passing | |
9333fe98 UR |
1196 | * parameters. Please note, with an alignment bigger than PAGE_SIZE, |
1197 | * a search length is adjusted to account for worst case alignment | |
1198 | * overhead. | |
68ad4a33 URS |
1199 | */ |
1200 | static __always_inline struct vmap_area * | |
9333fe98 UR |
1201 | find_vmap_lowest_match(unsigned long size, unsigned long align, |
1202 | unsigned long vstart, bool adjust_search_size) | |
68ad4a33 URS |
1203 | { |
1204 | struct vmap_area *va; | |
1205 | struct rb_node *node; | |
9333fe98 | 1206 | unsigned long length; |
68ad4a33 URS |
1207 | |
1208 | /* Start from the root. */ | |
1209 | node = free_vmap_area_root.rb_node; | |
1210 | ||
9333fe98 UR |
1211 | /* Adjust the search size for alignment overhead. */ |
1212 | length = adjust_search_size ? size + align - 1 : size; | |
1213 | ||
68ad4a33 URS |
1214 | while (node) { |
1215 | va = rb_entry(node, struct vmap_area, rb_node); | |
1216 | ||
9333fe98 | 1217 | if (get_subtree_max_size(node->rb_left) >= length && |
68ad4a33 URS |
1218 | vstart < va->va_start) { |
1219 | node = node->rb_left; | |
1220 | } else { | |
1221 | if (is_within_this_va(va, size, align, vstart)) | |
1222 | return va; | |
1223 | ||
1224 | /* | |
1225 | * Does not make sense to go deeper towards the right | |
1226 | * sub-tree if it does not have a free block that is | |
9333fe98 | 1227 | * equal or bigger to the requested search length. |
68ad4a33 | 1228 | */ |
9333fe98 | 1229 | if (get_subtree_max_size(node->rb_right) >= length) { |
68ad4a33 URS |
1230 | node = node->rb_right; |
1231 | continue; | |
1232 | } | |
1233 | ||
1234 | /* | |
3806b041 | 1235 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 | 1236 | * that will satisfy the search criteria. It can happen |
9f531973 URS |
1237 | * due to "vstart" restriction or an alignment overhead |
1238 | * that is bigger then PAGE_SIZE. | |
68ad4a33 URS |
1239 | */ |
1240 | while ((node = rb_parent(node))) { | |
1241 | va = rb_entry(node, struct vmap_area, rb_node); | |
1242 | if (is_within_this_va(va, size, align, vstart)) | |
1243 | return va; | |
1244 | ||
9333fe98 | 1245 | if (get_subtree_max_size(node->rb_right) >= length && |
68ad4a33 | 1246 | vstart <= va->va_start) { |
9f531973 URS |
1247 | /* |
1248 | * Shift the vstart forward. Please note, we update it with | |
1249 | * parent's start address adding "1" because we do not want | |
1250 | * to enter same sub-tree after it has already been checked | |
1251 | * and no suitable free block found there. | |
1252 | */ | |
1253 | vstart = va->va_start + 1; | |
68ad4a33 URS |
1254 | node = node->rb_right; |
1255 | break; | |
1256 | } | |
1257 | } | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | return NULL; | |
1262 | } | |
1263 | ||
a6cf4e0f URS |
1264 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1265 | #include <linux/random.h> | |
1266 | ||
1267 | static struct vmap_area * | |
1268 | find_vmap_lowest_linear_match(unsigned long size, | |
1269 | unsigned long align, unsigned long vstart) | |
1270 | { | |
1271 | struct vmap_area *va; | |
1272 | ||
1273 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
1274 | if (!is_within_this_va(va, size, align, vstart)) | |
1275 | continue; | |
1276 | ||
1277 | return va; | |
1278 | } | |
1279 | ||
1280 | return NULL; | |
1281 | } | |
1282 | ||
1283 | static void | |
066fed59 | 1284 | find_vmap_lowest_match_check(unsigned long size, unsigned long align) |
a6cf4e0f URS |
1285 | { |
1286 | struct vmap_area *va_1, *va_2; | |
1287 | unsigned long vstart; | |
1288 | unsigned int rnd; | |
1289 | ||
1290 | get_random_bytes(&rnd, sizeof(rnd)); | |
1291 | vstart = VMALLOC_START + rnd; | |
1292 | ||
9333fe98 | 1293 | va_1 = find_vmap_lowest_match(size, align, vstart, false); |
066fed59 | 1294 | va_2 = find_vmap_lowest_linear_match(size, align, vstart); |
a6cf4e0f URS |
1295 | |
1296 | if (va_1 != va_2) | |
1297 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
1298 | va_1, va_2, vstart); | |
1299 | } | |
1300 | #endif | |
1301 | ||
68ad4a33 URS |
1302 | enum fit_type { |
1303 | NOTHING_FIT = 0, | |
1304 | FL_FIT_TYPE = 1, /* full fit */ | |
1305 | LE_FIT_TYPE = 2, /* left edge fit */ | |
1306 | RE_FIT_TYPE = 3, /* right edge fit */ | |
1307 | NE_FIT_TYPE = 4 /* no edge fit */ | |
1308 | }; | |
1309 | ||
1310 | static __always_inline enum fit_type | |
1311 | classify_va_fit_type(struct vmap_area *va, | |
1312 | unsigned long nva_start_addr, unsigned long size) | |
1313 | { | |
1314 | enum fit_type type; | |
1315 | ||
1316 | /* Check if it is within VA. */ | |
1317 | if (nva_start_addr < va->va_start || | |
1318 | nva_start_addr + size > va->va_end) | |
1319 | return NOTHING_FIT; | |
1320 | ||
1321 | /* Now classify. */ | |
1322 | if (va->va_start == nva_start_addr) { | |
1323 | if (va->va_end == nva_start_addr + size) | |
1324 | type = FL_FIT_TYPE; | |
1325 | else | |
1326 | type = LE_FIT_TYPE; | |
1327 | } else if (va->va_end == nva_start_addr + size) { | |
1328 | type = RE_FIT_TYPE; | |
1329 | } else { | |
1330 | type = NE_FIT_TYPE; | |
1331 | } | |
1332 | ||
1333 | return type; | |
1334 | } | |
1335 | ||
1336 | static __always_inline int | |
1337 | adjust_va_to_fit_type(struct vmap_area *va, | |
1b23ff80 | 1338 | unsigned long nva_start_addr, unsigned long size) |
68ad4a33 | 1339 | { |
2c929233 | 1340 | struct vmap_area *lva = NULL; |
1b23ff80 | 1341 | enum fit_type type = classify_va_fit_type(va, nva_start_addr, size); |
68ad4a33 URS |
1342 | |
1343 | if (type == FL_FIT_TYPE) { | |
1344 | /* | |
1345 | * No need to split VA, it fully fits. | |
1346 | * | |
1347 | * | | | |
1348 | * V NVA V | |
1349 | * |---------------| | |
1350 | */ | |
1351 | unlink_va(va, &free_vmap_area_root); | |
1352 | kmem_cache_free(vmap_area_cachep, va); | |
1353 | } else if (type == LE_FIT_TYPE) { | |
1354 | /* | |
1355 | * Split left edge of fit VA. | |
1356 | * | |
1357 | * | | | |
1358 | * V NVA V R | |
1359 | * |-------|-------| | |
1360 | */ | |
1361 | va->va_start += size; | |
1362 | } else if (type == RE_FIT_TYPE) { | |
1363 | /* | |
1364 | * Split right edge of fit VA. | |
1365 | * | |
1366 | * | | | |
1367 | * L V NVA V | |
1368 | * |-------|-------| | |
1369 | */ | |
1370 | va->va_end = nva_start_addr; | |
1371 | } else if (type == NE_FIT_TYPE) { | |
1372 | /* | |
1373 | * Split no edge of fit VA. | |
1374 | * | |
1375 | * | | | |
1376 | * L V NVA V R | |
1377 | * |---|-------|---| | |
1378 | */ | |
82dd23e8 URS |
1379 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
1380 | if (unlikely(!lva)) { | |
1381 | /* | |
1382 | * For percpu allocator we do not do any pre-allocation | |
1383 | * and leave it as it is. The reason is it most likely | |
1384 | * never ends up with NE_FIT_TYPE splitting. In case of | |
1385 | * percpu allocations offsets and sizes are aligned to | |
1386 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
1387 | * are its main fitting cases. | |
1388 | * | |
1389 | * There are a few exceptions though, as an example it is | |
1390 | * a first allocation (early boot up) when we have "one" | |
1391 | * big free space that has to be split. | |
060650a2 URS |
1392 | * |
1393 | * Also we can hit this path in case of regular "vmap" | |
1394 | * allocations, if "this" current CPU was not preloaded. | |
1395 | * See the comment in alloc_vmap_area() why. If so, then | |
1396 | * GFP_NOWAIT is used instead to get an extra object for | |
1397 | * split purpose. That is rare and most time does not | |
1398 | * occur. | |
1399 | * | |
1400 | * What happens if an allocation gets failed. Basically, | |
1401 | * an "overflow" path is triggered to purge lazily freed | |
1402 | * areas to free some memory, then, the "retry" path is | |
1403 | * triggered to repeat one more time. See more details | |
1404 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
1405 | */ |
1406 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
1407 | if (!lva) | |
1408 | return -1; | |
1409 | } | |
68ad4a33 URS |
1410 | |
1411 | /* | |
1412 | * Build the remainder. | |
1413 | */ | |
1414 | lva->va_start = va->va_start; | |
1415 | lva->va_end = nva_start_addr; | |
1416 | ||
1417 | /* | |
1418 | * Shrink this VA to remaining size. | |
1419 | */ | |
1420 | va->va_start = nva_start_addr + size; | |
1421 | } else { | |
1422 | return -1; | |
1423 | } | |
1424 | ||
1425 | if (type != FL_FIT_TYPE) { | |
1426 | augment_tree_propagate_from(va); | |
1427 | ||
2c929233 | 1428 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
1429 | insert_vmap_area_augment(lva, &va->rb_node, |
1430 | &free_vmap_area_root, &free_vmap_area_list); | |
1431 | } | |
1432 | ||
1433 | return 0; | |
1434 | } | |
1435 | ||
1436 | /* | |
1437 | * Returns a start address of the newly allocated area, if success. | |
1438 | * Otherwise a vend is returned that indicates failure. | |
1439 | */ | |
1440 | static __always_inline unsigned long | |
1441 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1442 | unsigned long vstart, unsigned long vend) |
68ad4a33 | 1443 | { |
9333fe98 | 1444 | bool adjust_search_size = true; |
68ad4a33 URS |
1445 | unsigned long nva_start_addr; |
1446 | struct vmap_area *va; | |
68ad4a33 URS |
1447 | int ret; |
1448 | ||
9333fe98 UR |
1449 | /* |
1450 | * Do not adjust when: | |
1451 | * a) align <= PAGE_SIZE, because it does not make any sense. | |
1452 | * All blocks(their start addresses) are at least PAGE_SIZE | |
1453 | * aligned anyway; | |
1454 | * b) a short range where a requested size corresponds to exactly | |
1455 | * specified [vstart:vend] interval and an alignment > PAGE_SIZE. | |
1456 | * With adjusted search length an allocation would not succeed. | |
1457 | */ | |
1458 | if (align <= PAGE_SIZE || (align > PAGE_SIZE && (vend - vstart) == size)) | |
1459 | adjust_search_size = false; | |
1460 | ||
1461 | va = find_vmap_lowest_match(size, align, vstart, adjust_search_size); | |
68ad4a33 URS |
1462 | if (unlikely(!va)) |
1463 | return vend; | |
1464 | ||
1465 | if (va->va_start > vstart) | |
1466 | nva_start_addr = ALIGN(va->va_start, align); | |
1467 | else | |
1468 | nva_start_addr = ALIGN(vstart, align); | |
1469 | ||
1470 | /* Check the "vend" restriction. */ | |
1471 | if (nva_start_addr + size > vend) | |
1472 | return vend; | |
1473 | ||
68ad4a33 | 1474 | /* Update the free vmap_area. */ |
1b23ff80 BH |
1475 | ret = adjust_va_to_fit_type(va, nva_start_addr, size); |
1476 | if (WARN_ON_ONCE(ret)) | |
68ad4a33 URS |
1477 | return vend; |
1478 | ||
a6cf4e0f | 1479 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
066fed59 | 1480 | find_vmap_lowest_match_check(size, align); |
a6cf4e0f URS |
1481 | #endif |
1482 | ||
68ad4a33 URS |
1483 | return nva_start_addr; |
1484 | } | |
4da56b99 | 1485 | |
d98c9e83 AR |
1486 | /* |
1487 | * Free a region of KVA allocated by alloc_vmap_area | |
1488 | */ | |
1489 | static void free_vmap_area(struct vmap_area *va) | |
1490 | { | |
1491 | /* | |
1492 | * Remove from the busy tree/list. | |
1493 | */ | |
1494 | spin_lock(&vmap_area_lock); | |
1495 | unlink_va(va, &vmap_area_root); | |
1496 | spin_unlock(&vmap_area_lock); | |
1497 | ||
1498 | /* | |
1499 | * Insert/Merge it back to the free tree/list. | |
1500 | */ | |
1501 | spin_lock(&free_vmap_area_lock); | |
96e2db45 | 1502 | merge_or_add_vmap_area_augment(va, &free_vmap_area_root, &free_vmap_area_list); |
d98c9e83 AR |
1503 | spin_unlock(&free_vmap_area_lock); |
1504 | } | |
1505 | ||
187f8cc4 URS |
1506 | static inline void |
1507 | preload_this_cpu_lock(spinlock_t *lock, gfp_t gfp_mask, int node) | |
1508 | { | |
1509 | struct vmap_area *va = NULL; | |
1510 | ||
1511 | /* | |
1512 | * Preload this CPU with one extra vmap_area object. It is used | |
1513 | * when fit type of free area is NE_FIT_TYPE. It guarantees that | |
1514 | * a CPU that does an allocation is preloaded. | |
1515 | * | |
1516 | * We do it in non-atomic context, thus it allows us to use more | |
1517 | * permissive allocation masks to be more stable under low memory | |
1518 | * condition and high memory pressure. | |
1519 | */ | |
1520 | if (!this_cpu_read(ne_fit_preload_node)) | |
1521 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); | |
1522 | ||
1523 | spin_lock(lock); | |
1524 | ||
1525 | if (va && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, va)) | |
1526 | kmem_cache_free(vmap_area_cachep, va); | |
1527 | } | |
1528 | ||
db64fe02 NP |
1529 | /* |
1530 | * Allocate a region of KVA of the specified size and alignment, within the | |
1531 | * vstart and vend. | |
1532 | */ | |
1533 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1534 | unsigned long align, | |
1535 | unsigned long vstart, unsigned long vend, | |
1536 | int node, gfp_t gfp_mask) | |
1537 | { | |
187f8cc4 | 1538 | struct vmap_area *va; |
12e376a6 | 1539 | unsigned long freed; |
1da177e4 | 1540 | unsigned long addr; |
db64fe02 | 1541 | int purged = 0; |
d98c9e83 | 1542 | int ret; |
db64fe02 | 1543 | |
7766970c | 1544 | BUG_ON(!size); |
891c49ab | 1545 | BUG_ON(offset_in_page(size)); |
89699605 | 1546 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1547 | |
68ad4a33 URS |
1548 | if (unlikely(!vmap_initialized)) |
1549 | return ERR_PTR(-EBUSY); | |
1550 | ||
5803ed29 | 1551 | might_sleep(); |
f07116d7 | 1552 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1553 | |
f07116d7 | 1554 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1555 | if (unlikely(!va)) |
1556 | return ERR_PTR(-ENOMEM); | |
1557 | ||
7f88f88f CM |
1558 | /* |
1559 | * Only scan the relevant parts containing pointers to other objects | |
1560 | * to avoid false negatives. | |
1561 | */ | |
f07116d7 | 1562 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1563 | |
db64fe02 | 1564 | retry: |
187f8cc4 URS |
1565 | preload_this_cpu_lock(&free_vmap_area_lock, gfp_mask, node); |
1566 | addr = __alloc_vmap_area(size, align, vstart, vend); | |
1567 | spin_unlock(&free_vmap_area_lock); | |
89699605 | 1568 | |
afd07389 | 1569 | /* |
68ad4a33 URS |
1570 | * If an allocation fails, the "vend" address is |
1571 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1572 | */ |
68ad4a33 | 1573 | if (unlikely(addr == vend)) |
89699605 | 1574 | goto overflow; |
db64fe02 NP |
1575 | |
1576 | va->va_start = addr; | |
1577 | va->va_end = addr + size; | |
688fcbfc | 1578 | va->vm = NULL; |
68ad4a33 | 1579 | |
e36176be URS |
1580 | spin_lock(&vmap_area_lock); |
1581 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
db64fe02 NP |
1582 | spin_unlock(&vmap_area_lock); |
1583 | ||
61e16557 | 1584 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1585 | BUG_ON(va->va_start < vstart); |
1586 | BUG_ON(va->va_end > vend); | |
1587 | ||
d98c9e83 AR |
1588 | ret = kasan_populate_vmalloc(addr, size); |
1589 | if (ret) { | |
1590 | free_vmap_area(va); | |
1591 | return ERR_PTR(ret); | |
1592 | } | |
1593 | ||
db64fe02 | 1594 | return va; |
89699605 NP |
1595 | |
1596 | overflow: | |
89699605 NP |
1597 | if (!purged) { |
1598 | purge_vmap_area_lazy(); | |
1599 | purged = 1; | |
1600 | goto retry; | |
1601 | } | |
4da56b99 | 1602 | |
12e376a6 URS |
1603 | freed = 0; |
1604 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1605 | ||
1606 | if (freed > 0) { | |
1607 | purged = 0; | |
1608 | goto retry; | |
4da56b99 CW |
1609 | } |
1610 | ||
03497d76 | 1611 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1612 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1613 | size); | |
68ad4a33 URS |
1614 | |
1615 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1616 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1617 | } |
1618 | ||
4da56b99 CW |
1619 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1620 | { | |
1621 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1622 | } | |
1623 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1624 | ||
1625 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1626 | { | |
1627 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1628 | } | |
1629 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1630 | ||
db64fe02 NP |
1631 | /* |
1632 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1633 | * before attempting to purge with a TLB flush. | |
1634 | * | |
1635 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1636 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1637 | * global TLB flushes that must be performed. It would seem natural to scale | |
1638 | * this number up linearly with the number of CPUs (because vmapping activity | |
1639 | * could also scale linearly with the number of CPUs), however it is likely | |
1640 | * that in practice, workloads might be constrained in other ways that mean | |
1641 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1642 | * conservative and not introduce a big latency on huge systems, so go with | |
1643 | * a less aggressive log scale. It will still be an improvement over the old | |
1644 | * code, and it will be simple to change the scale factor if we find that it | |
1645 | * becomes a problem on bigger systems. | |
1646 | */ | |
1647 | static unsigned long lazy_max_pages(void) | |
1648 | { | |
1649 | unsigned int log; | |
1650 | ||
1651 | log = fls(num_online_cpus()); | |
1652 | ||
1653 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1654 | } | |
1655 | ||
4d36e6f8 | 1656 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1657 | |
0574ecd1 | 1658 | /* |
f0953a1b | 1659 | * Serialize vmap purging. There is no actual critical section protected |
0574ecd1 CH |
1660 | * by this look, but we want to avoid concurrent calls for performance |
1661 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1662 | */ | |
f9e09977 | 1663 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1664 | |
02b709df NP |
1665 | /* for per-CPU blocks */ |
1666 | static void purge_fragmented_blocks_allcpus(void); | |
1667 | ||
db64fe02 NP |
1668 | /* |
1669 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1670 | */ |
0574ecd1 | 1671 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1672 | { |
4d36e6f8 | 1673 | unsigned long resched_threshold; |
96e2db45 URS |
1674 | struct list_head local_pure_list; |
1675 | struct vmap_area *va, *n_va; | |
db64fe02 | 1676 | |
0574ecd1 | 1677 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1678 | |
96e2db45 URS |
1679 | spin_lock(&purge_vmap_area_lock); |
1680 | purge_vmap_area_root = RB_ROOT; | |
1681 | list_replace_init(&purge_vmap_area_list, &local_pure_list); | |
1682 | spin_unlock(&purge_vmap_area_lock); | |
1683 | ||
1684 | if (unlikely(list_empty(&local_pure_list))) | |
68571be9 URS |
1685 | return false; |
1686 | ||
96e2db45 URS |
1687 | start = min(start, |
1688 | list_first_entry(&local_pure_list, | |
1689 | struct vmap_area, list)->va_start); | |
1690 | ||
1691 | end = max(end, | |
1692 | list_last_entry(&local_pure_list, | |
1693 | struct vmap_area, list)->va_end); | |
db64fe02 | 1694 | |
0574ecd1 | 1695 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1696 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1697 | |
e36176be | 1698 | spin_lock(&free_vmap_area_lock); |
96e2db45 | 1699 | list_for_each_entry_safe(va, n_va, &local_pure_list, list) { |
4d36e6f8 | 1700 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
3c5c3cfb DA |
1701 | unsigned long orig_start = va->va_start; |
1702 | unsigned long orig_end = va->va_end; | |
763b218d | 1703 | |
dd3b8353 URS |
1704 | /* |
1705 | * Finally insert or merge lazily-freed area. It is | |
1706 | * detached and there is no need to "unlink" it from | |
1707 | * anything. | |
1708 | */ | |
96e2db45 URS |
1709 | va = merge_or_add_vmap_area_augment(va, &free_vmap_area_root, |
1710 | &free_vmap_area_list); | |
3c5c3cfb | 1711 | |
9c801f61 URS |
1712 | if (!va) |
1713 | continue; | |
1714 | ||
3c5c3cfb DA |
1715 | if (is_vmalloc_or_module_addr((void *)orig_start)) |
1716 | kasan_release_vmalloc(orig_start, orig_end, | |
1717 | va->va_start, va->va_end); | |
dd3b8353 | 1718 | |
4d36e6f8 | 1719 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1720 | |
4d36e6f8 | 1721 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
e36176be | 1722 | cond_resched_lock(&free_vmap_area_lock); |
763b218d | 1723 | } |
e36176be | 1724 | spin_unlock(&free_vmap_area_lock); |
0574ecd1 | 1725 | return true; |
db64fe02 NP |
1726 | } |
1727 | ||
1728 | /* | |
1729 | * Kick off a purge of the outstanding lazy areas. | |
1730 | */ | |
1731 | static void purge_vmap_area_lazy(void) | |
1732 | { | |
f9e09977 | 1733 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1734 | purge_fragmented_blocks_allcpus(); |
1735 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1736 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1737 | } |
1738 | ||
690467c8 URS |
1739 | static void drain_vmap_area_work(struct work_struct *work) |
1740 | { | |
1741 | unsigned long nr_lazy; | |
1742 | ||
1743 | do { | |
1744 | mutex_lock(&vmap_purge_lock); | |
1745 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
1746 | mutex_unlock(&vmap_purge_lock); | |
1747 | ||
1748 | /* Recheck if further work is required. */ | |
1749 | nr_lazy = atomic_long_read(&vmap_lazy_nr); | |
1750 | } while (nr_lazy > lazy_max_pages()); | |
1751 | } | |
1752 | ||
db64fe02 | 1753 | /* |
64141da5 JF |
1754 | * Free a vmap area, caller ensuring that the area has been unmapped |
1755 | * and flush_cache_vunmap had been called for the correct range | |
1756 | * previously. | |
db64fe02 | 1757 | */ |
64141da5 | 1758 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1759 | { |
4d36e6f8 | 1760 | unsigned long nr_lazy; |
80c4bd7a | 1761 | |
dd3b8353 URS |
1762 | spin_lock(&vmap_area_lock); |
1763 | unlink_va(va, &vmap_area_root); | |
1764 | spin_unlock(&vmap_area_lock); | |
1765 | ||
4d36e6f8 URS |
1766 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1767 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a | 1768 | |
96e2db45 URS |
1769 | /* |
1770 | * Merge or place it to the purge tree/list. | |
1771 | */ | |
1772 | spin_lock(&purge_vmap_area_lock); | |
1773 | merge_or_add_vmap_area(va, | |
1774 | &purge_vmap_area_root, &purge_vmap_area_list); | |
1775 | spin_unlock(&purge_vmap_area_lock); | |
80c4bd7a | 1776 | |
96e2db45 | 1777 | /* After this point, we may free va at any time */ |
80c4bd7a | 1778 | if (unlikely(nr_lazy > lazy_max_pages())) |
690467c8 | 1779 | schedule_work(&drain_vmap_work); |
db64fe02 NP |
1780 | } |
1781 | ||
b29acbdc NP |
1782 | /* |
1783 | * Free and unmap a vmap area | |
1784 | */ | |
1785 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1786 | { | |
1787 | flush_cache_vunmap(va->va_start, va->va_end); | |
4ad0ae8c | 1788 | vunmap_range_noflush(va->va_start, va->va_end); |
8e57f8ac | 1789 | if (debug_pagealloc_enabled_static()) |
82a2e924 CP |
1790 | flush_tlb_kernel_range(va->va_start, va->va_end); |
1791 | ||
c8eef01e | 1792 | free_vmap_area_noflush(va); |
b29acbdc NP |
1793 | } |
1794 | ||
db64fe02 NP |
1795 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1796 | { | |
1797 | struct vmap_area *va; | |
1798 | ||
1799 | spin_lock(&vmap_area_lock); | |
1800 | va = __find_vmap_area(addr); | |
1801 | spin_unlock(&vmap_area_lock); | |
1802 | ||
1803 | return va; | |
1804 | } | |
1805 | ||
db64fe02 NP |
1806 | /*** Per cpu kva allocator ***/ |
1807 | ||
1808 | /* | |
1809 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1810 | * room for at least 16 percpu vmap blocks per CPU. | |
1811 | */ | |
1812 | /* | |
1813 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1814 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1815 | * instead (we just need a rough idea) | |
1816 | */ | |
1817 | #if BITS_PER_LONG == 32 | |
1818 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1819 | #else | |
1820 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1821 | #endif | |
1822 | ||
1823 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1824 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1825 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1826 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1827 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1828 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1829 | #define VMAP_BBMAP_BITS \ |
1830 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1831 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1832 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1833 | |
1834 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1835 | ||
1836 | struct vmap_block_queue { | |
1837 | spinlock_t lock; | |
1838 | struct list_head free; | |
db64fe02 NP |
1839 | }; |
1840 | ||
1841 | struct vmap_block { | |
1842 | spinlock_t lock; | |
1843 | struct vmap_area *va; | |
db64fe02 | 1844 | unsigned long free, dirty; |
7d61bfe8 | 1845 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1846 | struct list_head free_list; |
1847 | struct rcu_head rcu_head; | |
02b709df | 1848 | struct list_head purge; |
db64fe02 NP |
1849 | }; |
1850 | ||
1851 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1852 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1853 | ||
1854 | /* | |
0f14599c | 1855 | * XArray of vmap blocks, indexed by address, to quickly find a vmap block |
db64fe02 NP |
1856 | * in the free path. Could get rid of this if we change the API to return a |
1857 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1858 | */ | |
0f14599c | 1859 | static DEFINE_XARRAY(vmap_blocks); |
db64fe02 NP |
1860 | |
1861 | /* | |
1862 | * We should probably have a fallback mechanism to allocate virtual memory | |
1863 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1864 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1865 | * big problem. | |
1866 | */ | |
1867 | ||
1868 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1869 | { | |
1870 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1871 | addr /= VMAP_BLOCK_SIZE; | |
1872 | return addr; | |
1873 | } | |
1874 | ||
cf725ce2 RP |
1875 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1876 | { | |
1877 | unsigned long addr; | |
1878 | ||
1879 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1880 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1881 | return (void *)addr; | |
1882 | } | |
1883 | ||
1884 | /** | |
1885 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1886 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1887 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1888 | * @gfp_mask: flags for the page level allocator | |
1889 | * | |
a862f68a | 1890 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1891 | */ |
1892 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1893 | { |
1894 | struct vmap_block_queue *vbq; | |
1895 | struct vmap_block *vb; | |
1896 | struct vmap_area *va; | |
1897 | unsigned long vb_idx; | |
1898 | int node, err; | |
cf725ce2 | 1899 | void *vaddr; |
db64fe02 NP |
1900 | |
1901 | node = numa_node_id(); | |
1902 | ||
1903 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1904 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1905 | if (unlikely(!vb)) | |
1906 | return ERR_PTR(-ENOMEM); | |
1907 | ||
1908 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1909 | VMALLOC_START, VMALLOC_END, | |
1910 | node, gfp_mask); | |
ddf9c6d4 | 1911 | if (IS_ERR(va)) { |
db64fe02 | 1912 | kfree(vb); |
e7d86340 | 1913 | return ERR_CAST(va); |
db64fe02 NP |
1914 | } |
1915 | ||
cf725ce2 | 1916 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1917 | spin_lock_init(&vb->lock); |
1918 | vb->va = va; | |
cf725ce2 RP |
1919 | /* At least something should be left free */ |
1920 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1921 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1922 | vb->dirty = 0; |
7d61bfe8 RP |
1923 | vb->dirty_min = VMAP_BBMAP_BITS; |
1924 | vb->dirty_max = 0; | |
db64fe02 | 1925 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1926 | |
1927 | vb_idx = addr_to_vb_idx(va->va_start); | |
0f14599c MWO |
1928 | err = xa_insert(&vmap_blocks, vb_idx, vb, gfp_mask); |
1929 | if (err) { | |
1930 | kfree(vb); | |
1931 | free_vmap_area(va); | |
1932 | return ERR_PTR(err); | |
1933 | } | |
db64fe02 | 1934 | |
3f804920 | 1935 | vbq = raw_cpu_ptr(&vmap_block_queue); |
db64fe02 | 1936 | spin_lock(&vbq->lock); |
68ac546f | 1937 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1938 | spin_unlock(&vbq->lock); |
db64fe02 | 1939 | |
cf725ce2 | 1940 | return vaddr; |
db64fe02 NP |
1941 | } |
1942 | ||
db64fe02 NP |
1943 | static void free_vmap_block(struct vmap_block *vb) |
1944 | { | |
1945 | struct vmap_block *tmp; | |
db64fe02 | 1946 | |
0f14599c | 1947 | tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start)); |
db64fe02 NP |
1948 | BUG_ON(tmp != vb); |
1949 | ||
64141da5 | 1950 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1951 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1952 | } |
1953 | ||
02b709df NP |
1954 | static void purge_fragmented_blocks(int cpu) |
1955 | { | |
1956 | LIST_HEAD(purge); | |
1957 | struct vmap_block *vb; | |
1958 | struct vmap_block *n_vb; | |
1959 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1960 | ||
1961 | rcu_read_lock(); | |
1962 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1963 | ||
1964 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1965 | continue; | |
1966 | ||
1967 | spin_lock(&vb->lock); | |
1968 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1969 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1970 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1971 | vb->dirty_min = 0; |
1972 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1973 | spin_lock(&vbq->lock); |
1974 | list_del_rcu(&vb->free_list); | |
1975 | spin_unlock(&vbq->lock); | |
1976 | spin_unlock(&vb->lock); | |
1977 | list_add_tail(&vb->purge, &purge); | |
1978 | } else | |
1979 | spin_unlock(&vb->lock); | |
1980 | } | |
1981 | rcu_read_unlock(); | |
1982 | ||
1983 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1984 | list_del(&vb->purge); | |
1985 | free_vmap_block(vb); | |
1986 | } | |
1987 | } | |
1988 | ||
02b709df NP |
1989 | static void purge_fragmented_blocks_allcpus(void) |
1990 | { | |
1991 | int cpu; | |
1992 | ||
1993 | for_each_possible_cpu(cpu) | |
1994 | purge_fragmented_blocks(cpu); | |
1995 | } | |
1996 | ||
db64fe02 NP |
1997 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1998 | { | |
1999 | struct vmap_block_queue *vbq; | |
2000 | struct vmap_block *vb; | |
cf725ce2 | 2001 | void *vaddr = NULL; |
db64fe02 NP |
2002 | unsigned int order; |
2003 | ||
891c49ab | 2004 | BUG_ON(offset_in_page(size)); |
db64fe02 | 2005 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
2006 | if (WARN_ON(size == 0)) { |
2007 | /* | |
2008 | * Allocating 0 bytes isn't what caller wants since | |
2009 | * get_order(0) returns funny result. Just warn and terminate | |
2010 | * early. | |
2011 | */ | |
2012 | return NULL; | |
2013 | } | |
db64fe02 NP |
2014 | order = get_order(size); |
2015 | ||
db64fe02 | 2016 | rcu_read_lock(); |
3f804920 | 2017 | vbq = raw_cpu_ptr(&vmap_block_queue); |
db64fe02 | 2018 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
cf725ce2 | 2019 | unsigned long pages_off; |
db64fe02 NP |
2020 | |
2021 | spin_lock(&vb->lock); | |
cf725ce2 RP |
2022 | if (vb->free < (1UL << order)) { |
2023 | spin_unlock(&vb->lock); | |
2024 | continue; | |
2025 | } | |
02b709df | 2026 | |
cf725ce2 RP |
2027 | pages_off = VMAP_BBMAP_BITS - vb->free; |
2028 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
2029 | vb->free -= 1UL << order; |
2030 | if (vb->free == 0) { | |
2031 | spin_lock(&vbq->lock); | |
2032 | list_del_rcu(&vb->free_list); | |
2033 | spin_unlock(&vbq->lock); | |
2034 | } | |
cf725ce2 | 2035 | |
02b709df NP |
2036 | spin_unlock(&vb->lock); |
2037 | break; | |
db64fe02 | 2038 | } |
02b709df | 2039 | |
db64fe02 NP |
2040 | rcu_read_unlock(); |
2041 | ||
cf725ce2 RP |
2042 | /* Allocate new block if nothing was found */ |
2043 | if (!vaddr) | |
2044 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 2045 | |
cf725ce2 | 2046 | return vaddr; |
db64fe02 NP |
2047 | } |
2048 | ||
78a0e8c4 | 2049 | static void vb_free(unsigned long addr, unsigned long size) |
db64fe02 NP |
2050 | { |
2051 | unsigned long offset; | |
db64fe02 NP |
2052 | unsigned int order; |
2053 | struct vmap_block *vb; | |
2054 | ||
891c49ab | 2055 | BUG_ON(offset_in_page(size)); |
db64fe02 | 2056 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc | 2057 | |
78a0e8c4 | 2058 | flush_cache_vunmap(addr, addr + size); |
b29acbdc | 2059 | |
db64fe02 | 2060 | order = get_order(size); |
78a0e8c4 | 2061 | offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT; |
0f14599c | 2062 | vb = xa_load(&vmap_blocks, addr_to_vb_idx(addr)); |
db64fe02 | 2063 | |
4ad0ae8c | 2064 | vunmap_range_noflush(addr, addr + size); |
64141da5 | 2065 | |
8e57f8ac | 2066 | if (debug_pagealloc_enabled_static()) |
78a0e8c4 | 2067 | flush_tlb_kernel_range(addr, addr + size); |
82a2e924 | 2068 | |
db64fe02 | 2069 | spin_lock(&vb->lock); |
7d61bfe8 RP |
2070 | |
2071 | /* Expand dirty range */ | |
2072 | vb->dirty_min = min(vb->dirty_min, offset); | |
2073 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 2074 | |
db64fe02 NP |
2075 | vb->dirty += 1UL << order; |
2076 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 2077 | BUG_ON(vb->free); |
db64fe02 NP |
2078 | spin_unlock(&vb->lock); |
2079 | free_vmap_block(vb); | |
2080 | } else | |
2081 | spin_unlock(&vb->lock); | |
2082 | } | |
2083 | ||
868b104d | 2084 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 2085 | { |
db64fe02 | 2086 | int cpu; |
db64fe02 | 2087 | |
9b463334 JF |
2088 | if (unlikely(!vmap_initialized)) |
2089 | return; | |
2090 | ||
5803ed29 CH |
2091 | might_sleep(); |
2092 | ||
db64fe02 NP |
2093 | for_each_possible_cpu(cpu) { |
2094 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
2095 | struct vmap_block *vb; | |
2096 | ||
2097 | rcu_read_lock(); | |
2098 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 2099 | spin_lock(&vb->lock); |
ad216c03 | 2100 | if (vb->dirty && vb->dirty != VMAP_BBMAP_BITS) { |
7d61bfe8 | 2101 | unsigned long va_start = vb->va->va_start; |
db64fe02 | 2102 | unsigned long s, e; |
b136be5e | 2103 | |
7d61bfe8 RP |
2104 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
2105 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 2106 | |
7d61bfe8 RP |
2107 | start = min(s, start); |
2108 | end = max(e, end); | |
db64fe02 | 2109 | |
7d61bfe8 | 2110 | flush = 1; |
db64fe02 NP |
2111 | } |
2112 | spin_unlock(&vb->lock); | |
2113 | } | |
2114 | rcu_read_unlock(); | |
2115 | } | |
2116 | ||
f9e09977 | 2117 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
2118 | purge_fragmented_blocks_allcpus(); |
2119 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
2120 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 2121 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 2122 | } |
868b104d RE |
2123 | |
2124 | /** | |
2125 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
2126 | * | |
2127 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
2128 | * to amortize TLB flushing overheads. What this means is that any page you | |
2129 | * have now, may, in a former life, have been mapped into kernel virtual | |
2130 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
2131 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
2132 | * | |
2133 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
2134 | * be sure that none of the pages we have control over will have any aliases | |
2135 | * from the vmap layer. | |
2136 | */ | |
2137 | void vm_unmap_aliases(void) | |
2138 | { | |
2139 | unsigned long start = ULONG_MAX, end = 0; | |
2140 | int flush = 0; | |
2141 | ||
2142 | _vm_unmap_aliases(start, end, flush); | |
2143 | } | |
db64fe02 NP |
2144 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
2145 | ||
2146 | /** | |
2147 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
2148 | * @mem: the pointer returned by vm_map_ram | |
2149 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
2150 | */ | |
2151 | void vm_unmap_ram(const void *mem, unsigned int count) | |
2152 | { | |
65ee03c4 | 2153 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
4aff1dc4 | 2154 | unsigned long addr = (unsigned long)kasan_reset_tag(mem); |
9c3acf60 | 2155 | struct vmap_area *va; |
db64fe02 | 2156 | |
5803ed29 | 2157 | might_sleep(); |
db64fe02 NP |
2158 | BUG_ON(!addr); |
2159 | BUG_ON(addr < VMALLOC_START); | |
2160 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 2161 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 2162 | |
d98c9e83 AR |
2163 | kasan_poison_vmalloc(mem, size); |
2164 | ||
9c3acf60 | 2165 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 2166 | debug_check_no_locks_freed(mem, size); |
78a0e8c4 | 2167 | vb_free(addr, size); |
9c3acf60 CH |
2168 | return; |
2169 | } | |
2170 | ||
2171 | va = find_vmap_area(addr); | |
2172 | BUG_ON(!va); | |
05e3ff95 CP |
2173 | debug_check_no_locks_freed((void *)va->va_start, |
2174 | (va->va_end - va->va_start)); | |
9c3acf60 | 2175 | free_unmap_vmap_area(va); |
db64fe02 NP |
2176 | } |
2177 | EXPORT_SYMBOL(vm_unmap_ram); | |
2178 | ||
2179 | /** | |
2180 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
2181 | * @pages: an array of pointers to the pages to be mapped | |
2182 | * @count: number of pages | |
2183 | * @node: prefer to allocate data structures on this node | |
e99c97ad | 2184 | * |
36437638 GK |
2185 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
2186 | * faster than vmap so it's good. But if you mix long-life and short-life | |
2187 | * objects with vm_map_ram(), it could consume lots of address space through | |
2188 | * fragmentation (especially on a 32bit machine). You could see failures in | |
2189 | * the end. Please use this function for short-lived objects. | |
2190 | * | |
e99c97ad | 2191 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 | 2192 | */ |
d4efd79a | 2193 | void *vm_map_ram(struct page **pages, unsigned int count, int node) |
db64fe02 | 2194 | { |
65ee03c4 | 2195 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
2196 | unsigned long addr; |
2197 | void *mem; | |
2198 | ||
2199 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
2200 | mem = vb_alloc(size, GFP_KERNEL); | |
2201 | if (IS_ERR(mem)) | |
2202 | return NULL; | |
2203 | addr = (unsigned long)mem; | |
2204 | } else { | |
2205 | struct vmap_area *va; | |
2206 | va = alloc_vmap_area(size, PAGE_SIZE, | |
2207 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
2208 | if (IS_ERR(va)) | |
2209 | return NULL; | |
2210 | ||
2211 | addr = va->va_start; | |
2212 | mem = (void *)addr; | |
2213 | } | |
d98c9e83 | 2214 | |
b67177ec NP |
2215 | if (vmap_pages_range(addr, addr + size, PAGE_KERNEL, |
2216 | pages, PAGE_SHIFT) < 0) { | |
db64fe02 NP |
2217 | vm_unmap_ram(mem, count); |
2218 | return NULL; | |
2219 | } | |
b67177ec | 2220 | |
23689e91 AK |
2221 | /* |
2222 | * Mark the pages as accessible, now that they are mapped. | |
2223 | * With hardware tag-based KASAN, marking is skipped for | |
2224 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). | |
2225 | */ | |
f6e39794 | 2226 | mem = kasan_unpoison_vmalloc(mem, size, KASAN_VMALLOC_PROT_NORMAL); |
19f1c3ac | 2227 | |
db64fe02 NP |
2228 | return mem; |
2229 | } | |
2230 | EXPORT_SYMBOL(vm_map_ram); | |
2231 | ||
4341fa45 | 2232 | static struct vm_struct *vmlist __initdata; |
92eac168 | 2233 | |
121e6f32 NP |
2234 | static inline unsigned int vm_area_page_order(struct vm_struct *vm) |
2235 | { | |
2236 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC | |
2237 | return vm->page_order; | |
2238 | #else | |
2239 | return 0; | |
2240 | #endif | |
2241 | } | |
2242 | ||
2243 | static inline void set_vm_area_page_order(struct vm_struct *vm, unsigned int order) | |
2244 | { | |
2245 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC | |
2246 | vm->page_order = order; | |
2247 | #else | |
2248 | BUG_ON(order != 0); | |
2249 | #endif | |
2250 | } | |
2251 | ||
be9b7335 NP |
2252 | /** |
2253 | * vm_area_add_early - add vmap area early during boot | |
2254 | * @vm: vm_struct to add | |
2255 | * | |
2256 | * This function is used to add fixed kernel vm area to vmlist before | |
2257 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
2258 | * should contain proper values and the other fields should be zero. | |
2259 | * | |
2260 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2261 | */ | |
2262 | void __init vm_area_add_early(struct vm_struct *vm) | |
2263 | { | |
2264 | struct vm_struct *tmp, **p; | |
2265 | ||
2266 | BUG_ON(vmap_initialized); | |
2267 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
2268 | if (tmp->addr >= vm->addr) { | |
2269 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
2270 | break; | |
2271 | } else | |
2272 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
2273 | } | |
2274 | vm->next = *p; | |
2275 | *p = vm; | |
2276 | } | |
2277 | ||
f0aa6617 TH |
2278 | /** |
2279 | * vm_area_register_early - register vmap area early during boot | |
2280 | * @vm: vm_struct to register | |
c0c0a293 | 2281 | * @align: requested alignment |
f0aa6617 TH |
2282 | * |
2283 | * This function is used to register kernel vm area before | |
2284 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
2285 | * proper values on entry and other fields should be zero. On return, | |
2286 | * vm->addr contains the allocated address. | |
2287 | * | |
2288 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2289 | */ | |
c0c0a293 | 2290 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 | 2291 | { |
0eb68437 KW |
2292 | unsigned long addr = ALIGN(VMALLOC_START, align); |
2293 | struct vm_struct *cur, **p; | |
c0c0a293 | 2294 | |
0eb68437 | 2295 | BUG_ON(vmap_initialized); |
f0aa6617 | 2296 | |
0eb68437 KW |
2297 | for (p = &vmlist; (cur = *p) != NULL; p = &cur->next) { |
2298 | if ((unsigned long)cur->addr - addr >= vm->size) | |
2299 | break; | |
2300 | addr = ALIGN((unsigned long)cur->addr + cur->size, align); | |
2301 | } | |
f0aa6617 | 2302 | |
0eb68437 KW |
2303 | BUG_ON(addr > VMALLOC_END - vm->size); |
2304 | vm->addr = (void *)addr; | |
2305 | vm->next = *p; | |
2306 | *p = vm; | |
3252b1d8 | 2307 | kasan_populate_early_vm_area_shadow(vm->addr, vm->size); |
f0aa6617 TH |
2308 | } |
2309 | ||
68ad4a33 URS |
2310 | static void vmap_init_free_space(void) |
2311 | { | |
2312 | unsigned long vmap_start = 1; | |
2313 | const unsigned long vmap_end = ULONG_MAX; | |
2314 | struct vmap_area *busy, *free; | |
2315 | ||
2316 | /* | |
2317 | * B F B B B F | |
2318 | * -|-----|.....|-----|-----|-----|.....|- | |
2319 | * | The KVA space | | |
2320 | * |<--------------------------------->| | |
2321 | */ | |
2322 | list_for_each_entry(busy, &vmap_area_list, list) { | |
2323 | if (busy->va_start - vmap_start > 0) { | |
2324 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2325 | if (!WARN_ON_ONCE(!free)) { | |
2326 | free->va_start = vmap_start; | |
2327 | free->va_end = busy->va_start; | |
2328 | ||
2329 | insert_vmap_area_augment(free, NULL, | |
2330 | &free_vmap_area_root, | |
2331 | &free_vmap_area_list); | |
2332 | } | |
2333 | } | |
2334 | ||
2335 | vmap_start = busy->va_end; | |
2336 | } | |
2337 | ||
2338 | if (vmap_end - vmap_start > 0) { | |
2339 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2340 | if (!WARN_ON_ONCE(!free)) { | |
2341 | free->va_start = vmap_start; | |
2342 | free->va_end = vmap_end; | |
2343 | ||
2344 | insert_vmap_area_augment(free, NULL, | |
2345 | &free_vmap_area_root, | |
2346 | &free_vmap_area_list); | |
2347 | } | |
2348 | } | |
2349 | } | |
2350 | ||
db64fe02 NP |
2351 | void __init vmalloc_init(void) |
2352 | { | |
822c18f2 IK |
2353 | struct vmap_area *va; |
2354 | struct vm_struct *tmp; | |
db64fe02 NP |
2355 | int i; |
2356 | ||
68ad4a33 URS |
2357 | /* |
2358 | * Create the cache for vmap_area objects. | |
2359 | */ | |
2360 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
2361 | ||
db64fe02 NP |
2362 | for_each_possible_cpu(i) { |
2363 | struct vmap_block_queue *vbq; | |
32fcfd40 | 2364 | struct vfree_deferred *p; |
db64fe02 NP |
2365 | |
2366 | vbq = &per_cpu(vmap_block_queue, i); | |
2367 | spin_lock_init(&vbq->lock); | |
2368 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
2369 | p = &per_cpu(vfree_deferred, i); |
2370 | init_llist_head(&p->list); | |
2371 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 2372 | } |
9b463334 | 2373 | |
822c18f2 IK |
2374 | /* Import existing vmlist entries. */ |
2375 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
2376 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
2377 | if (WARN_ON_ONCE(!va)) | |
2378 | continue; | |
2379 | ||
822c18f2 IK |
2380 | va->va_start = (unsigned long)tmp->addr; |
2381 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 2382 | va->vm = tmp; |
68ad4a33 | 2383 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 2384 | } |
ca23e405 | 2385 | |
68ad4a33 URS |
2386 | /* |
2387 | * Now we can initialize a free vmap space. | |
2388 | */ | |
2389 | vmap_init_free_space(); | |
9b463334 | 2390 | vmap_initialized = true; |
db64fe02 NP |
2391 | } |
2392 | ||
e36176be URS |
2393 | static inline void setup_vmalloc_vm_locked(struct vm_struct *vm, |
2394 | struct vmap_area *va, unsigned long flags, const void *caller) | |
cf88c790 | 2395 | { |
cf88c790 TH |
2396 | vm->flags = flags; |
2397 | vm->addr = (void *)va->va_start; | |
2398 | vm->size = va->va_end - va->va_start; | |
2399 | vm->caller = caller; | |
db1aecaf | 2400 | va->vm = vm; |
e36176be URS |
2401 | } |
2402 | ||
2403 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, | |
2404 | unsigned long flags, const void *caller) | |
2405 | { | |
2406 | spin_lock(&vmap_area_lock); | |
2407 | setup_vmalloc_vm_locked(vm, va, flags, caller); | |
c69480ad | 2408 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2409 | } |
cf88c790 | 2410 | |
20fc02b4 | 2411 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2412 | { |
d4033afd | 2413 | /* |
20fc02b4 | 2414 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2415 | * we should make sure that vm has proper values. |
2416 | * Pair with smp_rmb() in show_numa_info(). | |
2417 | */ | |
2418 | smp_wmb(); | |
20fc02b4 | 2419 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2420 | } |
2421 | ||
db64fe02 | 2422 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
7ca3027b DA |
2423 | unsigned long align, unsigned long shift, unsigned long flags, |
2424 | unsigned long start, unsigned long end, int node, | |
2425 | gfp_t gfp_mask, const void *caller) | |
db64fe02 | 2426 | { |
0006526d | 2427 | struct vmap_area *va; |
db64fe02 | 2428 | struct vm_struct *area; |
d98c9e83 | 2429 | unsigned long requested_size = size; |
1da177e4 | 2430 | |
52fd24ca | 2431 | BUG_ON(in_interrupt()); |
7ca3027b | 2432 | size = ALIGN(size, 1ul << shift); |
31be8309 OH |
2433 | if (unlikely(!size)) |
2434 | return NULL; | |
1da177e4 | 2435 | |
252e5c6e | 2436 | if (flags & VM_IOREMAP) |
2437 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2438 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2439 | ||
cf88c790 | 2440 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2441 | if (unlikely(!area)) |
2442 | return NULL; | |
2443 | ||
71394fe5 AR |
2444 | if (!(flags & VM_NO_GUARD)) |
2445 | size += PAGE_SIZE; | |
1da177e4 | 2446 | |
db64fe02 NP |
2447 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2448 | if (IS_ERR(va)) { | |
2449 | kfree(area); | |
2450 | return NULL; | |
1da177e4 | 2451 | } |
1da177e4 | 2452 | |
d98c9e83 | 2453 | setup_vmalloc_vm(area, va, flags, caller); |
3c5c3cfb | 2454 | |
19f1c3ac AK |
2455 | /* |
2456 | * Mark pages for non-VM_ALLOC mappings as accessible. Do it now as a | |
2457 | * best-effort approach, as they can be mapped outside of vmalloc code. | |
2458 | * For VM_ALLOC mappings, the pages are marked as accessible after | |
2459 | * getting mapped in __vmalloc_node_range(). | |
23689e91 AK |
2460 | * With hardware tag-based KASAN, marking is skipped for |
2461 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). | |
19f1c3ac AK |
2462 | */ |
2463 | if (!(flags & VM_ALLOC)) | |
23689e91 | 2464 | area->addr = kasan_unpoison_vmalloc(area->addr, requested_size, |
f6e39794 | 2465 | KASAN_VMALLOC_PROT_NORMAL); |
1d96320f | 2466 | |
1da177e4 | 2467 | return area; |
1da177e4 LT |
2468 | } |
2469 | ||
c2968612 BH |
2470 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2471 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2472 | const void *caller) |
c2968612 | 2473 | { |
7ca3027b DA |
2474 | return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, start, end, |
2475 | NUMA_NO_NODE, GFP_KERNEL, caller); | |
c2968612 BH |
2476 | } |
2477 | ||
1da177e4 | 2478 | /** |
92eac168 MR |
2479 | * get_vm_area - reserve a contiguous kernel virtual area |
2480 | * @size: size of the area | |
2481 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2482 | * |
92eac168 MR |
2483 | * Search an area of @size in the kernel virtual mapping area, |
2484 | * and reserved it for out purposes. Returns the area descriptor | |
2485 | * on success or %NULL on failure. | |
a862f68a MR |
2486 | * |
2487 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2488 | */ |
2489 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2490 | { | |
7ca3027b DA |
2491 | return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, |
2492 | VMALLOC_START, VMALLOC_END, | |
00ef2d2f DR |
2493 | NUMA_NO_NODE, GFP_KERNEL, |
2494 | __builtin_return_address(0)); | |
23016969 CL |
2495 | } |
2496 | ||
2497 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2498 | const void *caller) |
23016969 | 2499 | { |
7ca3027b DA |
2500 | return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, |
2501 | VMALLOC_START, VMALLOC_END, | |
00ef2d2f | 2502 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2503 | } |
2504 | ||
e9da6e99 | 2505 | /** |
92eac168 MR |
2506 | * find_vm_area - find a continuous kernel virtual area |
2507 | * @addr: base address | |
e9da6e99 | 2508 | * |
92eac168 MR |
2509 | * Search for the kernel VM area starting at @addr, and return it. |
2510 | * It is up to the caller to do all required locking to keep the returned | |
2511 | * pointer valid. | |
a862f68a | 2512 | * |
74640617 | 2513 | * Return: the area descriptor on success or %NULL on failure. |
e9da6e99 MS |
2514 | */ |
2515 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2516 | { |
db64fe02 | 2517 | struct vmap_area *va; |
83342314 | 2518 | |
db64fe02 | 2519 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2520 | if (!va) |
2521 | return NULL; | |
1da177e4 | 2522 | |
688fcbfc | 2523 | return va->vm; |
1da177e4 LT |
2524 | } |
2525 | ||
7856dfeb | 2526 | /** |
92eac168 MR |
2527 | * remove_vm_area - find and remove a continuous kernel virtual area |
2528 | * @addr: base address | |
7856dfeb | 2529 | * |
92eac168 MR |
2530 | * Search for the kernel VM area starting at @addr, and remove it. |
2531 | * This function returns the found VM area, but using it is NOT safe | |
2532 | * on SMP machines, except for its size or flags. | |
a862f68a | 2533 | * |
74640617 | 2534 | * Return: the area descriptor on success or %NULL on failure. |
7856dfeb | 2535 | */ |
b3bdda02 | 2536 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2537 | { |
db64fe02 NP |
2538 | struct vmap_area *va; |
2539 | ||
5803ed29 CH |
2540 | might_sleep(); |
2541 | ||
dd3b8353 URS |
2542 | spin_lock(&vmap_area_lock); |
2543 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2544 | if (va && va->vm) { |
db1aecaf | 2545 | struct vm_struct *vm = va->vm; |
f5252e00 | 2546 | |
c69480ad | 2547 | va->vm = NULL; |
c69480ad JK |
2548 | spin_unlock(&vmap_area_lock); |
2549 | ||
63840de2 | 2550 | kasan_free_module_shadow(vm); |
dd32c279 | 2551 | free_unmap_vmap_area(va); |
dd32c279 | 2552 | |
db64fe02 NP |
2553 | return vm; |
2554 | } | |
dd3b8353 URS |
2555 | |
2556 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2557 | return NULL; |
7856dfeb AK |
2558 | } |
2559 | ||
868b104d RE |
2560 | static inline void set_area_direct_map(const struct vm_struct *area, |
2561 | int (*set_direct_map)(struct page *page)) | |
2562 | { | |
2563 | int i; | |
2564 | ||
121e6f32 | 2565 | /* HUGE_VMALLOC passes small pages to set_direct_map */ |
868b104d RE |
2566 | for (i = 0; i < area->nr_pages; i++) |
2567 | if (page_address(area->pages[i])) | |
2568 | set_direct_map(area->pages[i]); | |
2569 | } | |
2570 | ||
2571 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2572 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2573 | { | |
868b104d | 2574 | unsigned long start = ULONG_MAX, end = 0; |
121e6f32 | 2575 | unsigned int page_order = vm_area_page_order(area); |
868b104d | 2576 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; |
31e67340 | 2577 | int flush_dmap = 0; |
868b104d RE |
2578 | int i; |
2579 | ||
868b104d RE |
2580 | remove_vm_area(area->addr); |
2581 | ||
2582 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2583 | if (!flush_reset) | |
2584 | return; | |
2585 | ||
2586 | /* | |
2587 | * If not deallocating pages, just do the flush of the VM area and | |
2588 | * return. | |
2589 | */ | |
2590 | if (!deallocate_pages) { | |
2591 | vm_unmap_aliases(); | |
2592 | return; | |
2593 | } | |
2594 | ||
2595 | /* | |
2596 | * If execution gets here, flush the vm mapping and reset the direct | |
2597 | * map. Find the start and end range of the direct mappings to make sure | |
2598 | * the vm_unmap_aliases() flush includes the direct map. | |
2599 | */ | |
121e6f32 | 2600 | for (i = 0; i < area->nr_pages; i += 1U << page_order) { |
8e41f872 RE |
2601 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2602 | if (addr) { | |
121e6f32 NP |
2603 | unsigned long page_size; |
2604 | ||
2605 | page_size = PAGE_SIZE << page_order; | |
868b104d | 2606 | start = min(addr, start); |
121e6f32 | 2607 | end = max(addr + page_size, end); |
31e67340 | 2608 | flush_dmap = 1; |
868b104d RE |
2609 | } |
2610 | } | |
2611 | ||
2612 | /* | |
2613 | * Set direct map to something invalid so that it won't be cached if | |
2614 | * there are any accesses after the TLB flush, then flush the TLB and | |
2615 | * reset the direct map permissions to the default. | |
2616 | */ | |
2617 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2618 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2619 | set_area_direct_map(area, set_direct_map_default_noflush); |
2620 | } | |
2621 | ||
b3bdda02 | 2622 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2623 | { |
2624 | struct vm_struct *area; | |
2625 | ||
2626 | if (!addr) | |
2627 | return; | |
2628 | ||
e69e9d4a | 2629 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2630 | addr)) |
1da177e4 | 2631 | return; |
1da177e4 | 2632 | |
6ade2032 | 2633 | area = find_vm_area(addr); |
1da177e4 | 2634 | if (unlikely(!area)) { |
4c8573e2 | 2635 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2636 | addr); |
1da177e4 LT |
2637 | return; |
2638 | } | |
2639 | ||
05e3ff95 CP |
2640 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2641 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2642 | |
c041098c | 2643 | kasan_poison_vmalloc(area->addr, get_vm_area_size(area)); |
3c5c3cfb | 2644 | |
868b104d RE |
2645 | vm_remove_mappings(area, deallocate_pages); |
2646 | ||
1da177e4 | 2647 | if (deallocate_pages) { |
3b8000ae | 2648 | int i; |
1da177e4 | 2649 | |
3b8000ae | 2650 | for (i = 0; i < area->nr_pages; i++) { |
bf53d6f8 CL |
2651 | struct page *page = area->pages[i]; |
2652 | ||
2653 | BUG_ON(!page); | |
3b8000ae NP |
2654 | mod_memcg_page_state(page, MEMCG_VMALLOC, -1); |
2655 | /* | |
2656 | * High-order allocs for huge vmallocs are split, so | |
2657 | * can be freed as an array of order-0 allocations | |
2658 | */ | |
2659 | __free_pages(page, 0); | |
a850e932 | 2660 | cond_resched(); |
1da177e4 | 2661 | } |
97105f0a | 2662 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2663 | |
244d63ee | 2664 | kvfree(area->pages); |
1da177e4 LT |
2665 | } |
2666 | ||
2667 | kfree(area); | |
1da177e4 | 2668 | } |
bf22e37a AR |
2669 | |
2670 | static inline void __vfree_deferred(const void *addr) | |
2671 | { | |
2672 | /* | |
2673 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2674 | * context. Preemption is absolutely fine here, because the llist_add() | |
2675 | * implementation is lockless, so it works even if we are adding to | |
73221d88 | 2676 | * another cpu's list. schedule_work() should be fine with this too. |
bf22e37a AR |
2677 | */ |
2678 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2679 | ||
2680 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2681 | schedule_work(&p->wq); | |
2682 | } | |
2683 | ||
2684 | /** | |
92eac168 MR |
2685 | * vfree_atomic - release memory allocated by vmalloc() |
2686 | * @addr: memory base address | |
bf22e37a | 2687 | * |
92eac168 MR |
2688 | * This one is just like vfree() but can be called in any atomic context |
2689 | * except NMIs. | |
bf22e37a AR |
2690 | */ |
2691 | void vfree_atomic(const void *addr) | |
2692 | { | |
2693 | BUG_ON(in_nmi()); | |
2694 | ||
2695 | kmemleak_free(addr); | |
2696 | ||
2697 | if (!addr) | |
2698 | return; | |
2699 | __vfree_deferred(addr); | |
2700 | } | |
2701 | ||
c67dc624 RP |
2702 | static void __vfree(const void *addr) |
2703 | { | |
2704 | if (unlikely(in_interrupt())) | |
2705 | __vfree_deferred(addr); | |
2706 | else | |
2707 | __vunmap(addr, 1); | |
2708 | } | |
2709 | ||
1da177e4 | 2710 | /** |
fa307474 MWO |
2711 | * vfree - Release memory allocated by vmalloc() |
2712 | * @addr: Memory base address | |
1da177e4 | 2713 | * |
fa307474 MWO |
2714 | * Free the virtually continuous memory area starting at @addr, as obtained |
2715 | * from one of the vmalloc() family of APIs. This will usually also free the | |
2716 | * physical memory underlying the virtual allocation, but that memory is | |
2717 | * reference counted, so it will not be freed until the last user goes away. | |
1da177e4 | 2718 | * |
fa307474 | 2719 | * If @addr is NULL, no operation is performed. |
c9fcee51 | 2720 | * |
fa307474 | 2721 | * Context: |
92eac168 | 2722 | * May sleep if called *not* from interrupt context. |
fa307474 MWO |
2723 | * Must not be called in NMI context (strictly speaking, it could be |
2724 | * if we have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
f0953a1b | 2725 | * conventions for vfree() arch-dependent would be a really bad idea). |
1da177e4 | 2726 | */ |
b3bdda02 | 2727 | void vfree(const void *addr) |
1da177e4 | 2728 | { |
32fcfd40 | 2729 | BUG_ON(in_nmi()); |
89219d37 CM |
2730 | |
2731 | kmemleak_free(addr); | |
2732 | ||
a8dda165 AR |
2733 | might_sleep_if(!in_interrupt()); |
2734 | ||
32fcfd40 AV |
2735 | if (!addr) |
2736 | return; | |
c67dc624 RP |
2737 | |
2738 | __vfree(addr); | |
1da177e4 | 2739 | } |
1da177e4 LT |
2740 | EXPORT_SYMBOL(vfree); |
2741 | ||
2742 | /** | |
92eac168 MR |
2743 | * vunmap - release virtual mapping obtained by vmap() |
2744 | * @addr: memory base address | |
1da177e4 | 2745 | * |
92eac168 MR |
2746 | * Free the virtually contiguous memory area starting at @addr, |
2747 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2748 | * |
92eac168 | 2749 | * Must not be called in interrupt context. |
1da177e4 | 2750 | */ |
b3bdda02 | 2751 | void vunmap(const void *addr) |
1da177e4 LT |
2752 | { |
2753 | BUG_ON(in_interrupt()); | |
34754b69 | 2754 | might_sleep(); |
32fcfd40 AV |
2755 | if (addr) |
2756 | __vunmap(addr, 0); | |
1da177e4 | 2757 | } |
1da177e4 LT |
2758 | EXPORT_SYMBOL(vunmap); |
2759 | ||
2760 | /** | |
92eac168 MR |
2761 | * vmap - map an array of pages into virtually contiguous space |
2762 | * @pages: array of page pointers | |
2763 | * @count: number of pages to map | |
2764 | * @flags: vm_area->flags | |
2765 | * @prot: page protection for the mapping | |
2766 | * | |
b944afc9 CH |
2767 | * Maps @count pages from @pages into contiguous kernel virtual space. |
2768 | * If @flags contains %VM_MAP_PUT_PAGES the ownership of the pages array itself | |
2769 | * (which must be kmalloc or vmalloc memory) and one reference per pages in it | |
2770 | * are transferred from the caller to vmap(), and will be freed / dropped when | |
2771 | * vfree() is called on the return value. | |
a862f68a MR |
2772 | * |
2773 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2774 | */ |
2775 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2776 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2777 | { |
2778 | struct vm_struct *area; | |
b67177ec | 2779 | unsigned long addr; |
65ee03c4 | 2780 | unsigned long size; /* In bytes */ |
1da177e4 | 2781 | |
34754b69 PZ |
2782 | might_sleep(); |
2783 | ||
bd1a8fb2 PZ |
2784 | /* |
2785 | * Your top guard is someone else's bottom guard. Not having a top | |
2786 | * guard compromises someone else's mappings too. | |
2787 | */ | |
2788 | if (WARN_ON_ONCE(flags & VM_NO_GUARD)) | |
2789 | flags &= ~VM_NO_GUARD; | |
2790 | ||
ca79b0c2 | 2791 | if (count > totalram_pages()) |
1da177e4 LT |
2792 | return NULL; |
2793 | ||
65ee03c4 GJM |
2794 | size = (unsigned long)count << PAGE_SHIFT; |
2795 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2796 | if (!area) |
2797 | return NULL; | |
23016969 | 2798 | |
b67177ec NP |
2799 | addr = (unsigned long)area->addr; |
2800 | if (vmap_pages_range(addr, addr + size, pgprot_nx(prot), | |
2801 | pages, PAGE_SHIFT) < 0) { | |
1da177e4 LT |
2802 | vunmap(area->addr); |
2803 | return NULL; | |
2804 | } | |
2805 | ||
c22ee528 | 2806 | if (flags & VM_MAP_PUT_PAGES) { |
b944afc9 | 2807 | area->pages = pages; |
c22ee528 ML |
2808 | area->nr_pages = count; |
2809 | } | |
1da177e4 LT |
2810 | return area->addr; |
2811 | } | |
1da177e4 LT |
2812 | EXPORT_SYMBOL(vmap); |
2813 | ||
3e9a9e25 CH |
2814 | #ifdef CONFIG_VMAP_PFN |
2815 | struct vmap_pfn_data { | |
2816 | unsigned long *pfns; | |
2817 | pgprot_t prot; | |
2818 | unsigned int idx; | |
2819 | }; | |
2820 | ||
2821 | static int vmap_pfn_apply(pte_t *pte, unsigned long addr, void *private) | |
2822 | { | |
2823 | struct vmap_pfn_data *data = private; | |
2824 | ||
2825 | if (WARN_ON_ONCE(pfn_valid(data->pfns[data->idx]))) | |
2826 | return -EINVAL; | |
2827 | *pte = pte_mkspecial(pfn_pte(data->pfns[data->idx++], data->prot)); | |
2828 | return 0; | |
2829 | } | |
2830 | ||
2831 | /** | |
2832 | * vmap_pfn - map an array of PFNs into virtually contiguous space | |
2833 | * @pfns: array of PFNs | |
2834 | * @count: number of pages to map | |
2835 | * @prot: page protection for the mapping | |
2836 | * | |
2837 | * Maps @count PFNs from @pfns into contiguous kernel virtual space and returns | |
2838 | * the start address of the mapping. | |
2839 | */ | |
2840 | void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot) | |
2841 | { | |
2842 | struct vmap_pfn_data data = { .pfns = pfns, .prot = pgprot_nx(prot) }; | |
2843 | struct vm_struct *area; | |
2844 | ||
2845 | area = get_vm_area_caller(count * PAGE_SIZE, VM_IOREMAP, | |
2846 | __builtin_return_address(0)); | |
2847 | if (!area) | |
2848 | return NULL; | |
2849 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
2850 | count * PAGE_SIZE, vmap_pfn_apply, &data)) { | |
2851 | free_vm_area(area); | |
2852 | return NULL; | |
2853 | } | |
2854 | return area->addr; | |
2855 | } | |
2856 | EXPORT_SYMBOL_GPL(vmap_pfn); | |
2857 | #endif /* CONFIG_VMAP_PFN */ | |
2858 | ||
12b9f873 UR |
2859 | static inline unsigned int |
2860 | vm_area_alloc_pages(gfp_t gfp, int nid, | |
343ab817 | 2861 | unsigned int order, unsigned int nr_pages, struct page **pages) |
12b9f873 UR |
2862 | { |
2863 | unsigned int nr_allocated = 0; | |
ffb29b1c CW |
2864 | struct page *page; |
2865 | int i; | |
12b9f873 UR |
2866 | |
2867 | /* | |
2868 | * For order-0 pages we make use of bulk allocator, if | |
2869 | * the page array is partly or not at all populated due | |
2870 | * to fails, fallback to a single page allocator that is | |
2871 | * more permissive. | |
2872 | */ | |
c00b6b96 | 2873 | if (!order) { |
9376130c MH |
2874 | gfp_t bulk_gfp = gfp & ~__GFP_NOFAIL; |
2875 | ||
343ab817 URS |
2876 | while (nr_allocated < nr_pages) { |
2877 | unsigned int nr, nr_pages_request; | |
2878 | ||
2879 | /* | |
2880 | * A maximum allowed request is hard-coded and is 100 | |
2881 | * pages per call. That is done in order to prevent a | |
2882 | * long preemption off scenario in the bulk-allocator | |
2883 | * so the range is [1:100]. | |
2884 | */ | |
2885 | nr_pages_request = min(100U, nr_pages - nr_allocated); | |
2886 | ||
c00b6b96 CW |
2887 | /* memory allocation should consider mempolicy, we can't |
2888 | * wrongly use nearest node when nid == NUMA_NO_NODE, | |
2889 | * otherwise memory may be allocated in only one node, | |
98af39d5 | 2890 | * but mempolicy wants to alloc memory by interleaving. |
c00b6b96 CW |
2891 | */ |
2892 | if (IS_ENABLED(CONFIG_NUMA) && nid == NUMA_NO_NODE) | |
9376130c | 2893 | nr = alloc_pages_bulk_array_mempolicy(bulk_gfp, |
c00b6b96 CW |
2894 | nr_pages_request, |
2895 | pages + nr_allocated); | |
2896 | ||
2897 | else | |
9376130c | 2898 | nr = alloc_pages_bulk_array_node(bulk_gfp, nid, |
c00b6b96 CW |
2899 | nr_pages_request, |
2900 | pages + nr_allocated); | |
343ab817 URS |
2901 | |
2902 | nr_allocated += nr; | |
2903 | cond_resched(); | |
2904 | ||
2905 | /* | |
2906 | * If zero or pages were obtained partly, | |
2907 | * fallback to a single page allocator. | |
2908 | */ | |
2909 | if (nr != nr_pages_request) | |
2910 | break; | |
2911 | } | |
3b8000ae | 2912 | } |
12b9f873 UR |
2913 | |
2914 | /* High-order pages or fallback path if "bulk" fails. */ | |
12b9f873 | 2915 | |
ffb29b1c | 2916 | while (nr_allocated < nr_pages) { |
dd544141 VA |
2917 | if (fatal_signal_pending(current)) |
2918 | break; | |
2919 | ||
ffb29b1c CW |
2920 | if (nid == NUMA_NO_NODE) |
2921 | page = alloc_pages(gfp, order); | |
2922 | else | |
2923 | page = alloc_pages_node(nid, gfp, order); | |
12b9f873 UR |
2924 | if (unlikely(!page)) |
2925 | break; | |
3b8000ae NP |
2926 | /* |
2927 | * Higher order allocations must be able to be treated as | |
2928 | * indepdenent small pages by callers (as they can with | |
2929 | * small-page vmallocs). Some drivers do their own refcounting | |
2930 | * on vmalloc_to_page() pages, some use page->mapping, | |
2931 | * page->lru, etc. | |
2932 | */ | |
2933 | if (order) | |
2934 | split_page(page, order); | |
12b9f873 UR |
2935 | |
2936 | /* | |
2937 | * Careful, we allocate and map page-order pages, but | |
2938 | * tracking is done per PAGE_SIZE page so as to keep the | |
2939 | * vm_struct APIs independent of the physical/mapped size. | |
2940 | */ | |
2941 | for (i = 0; i < (1U << order); i++) | |
2942 | pages[nr_allocated + i] = page + i; | |
2943 | ||
12e376a6 | 2944 | cond_resched(); |
12b9f873 UR |
2945 | nr_allocated += 1U << order; |
2946 | } | |
2947 | ||
2948 | return nr_allocated; | |
2949 | } | |
2950 | ||
e31d9eb5 | 2951 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
121e6f32 NP |
2952 | pgprot_t prot, unsigned int page_shift, |
2953 | int node) | |
1da177e4 | 2954 | { |
930f036b | 2955 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
9376130c | 2956 | bool nofail = gfp_mask & __GFP_NOFAIL; |
121e6f32 NP |
2957 | unsigned long addr = (unsigned long)area->addr; |
2958 | unsigned long size = get_vm_area_size(area); | |
34fe6537 | 2959 | unsigned long array_size; |
121e6f32 NP |
2960 | unsigned int nr_small_pages = size >> PAGE_SHIFT; |
2961 | unsigned int page_order; | |
451769eb MH |
2962 | unsigned int flags; |
2963 | int ret; | |
1da177e4 | 2964 | |
121e6f32 | 2965 | array_size = (unsigned long)nr_small_pages * sizeof(struct page *); |
f255935b CH |
2966 | gfp_mask |= __GFP_NOWARN; |
2967 | if (!(gfp_mask & (GFP_DMA | GFP_DMA32))) | |
2968 | gfp_mask |= __GFP_HIGHMEM; | |
1da177e4 | 2969 | |
1da177e4 | 2970 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 2971 | if (array_size > PAGE_SIZE) { |
5c1f4e69 | 2972 | area->pages = __vmalloc_node(array_size, 1, nested_gfp, node, |
f255935b | 2973 | area->caller); |
286e1ea3 | 2974 | } else { |
5c1f4e69 | 2975 | area->pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2976 | } |
7ea36242 | 2977 | |
5c1f4e69 | 2978 | if (!area->pages) { |
c3d77172 | 2979 | warn_alloc(gfp_mask, NULL, |
f4bdfeaf URS |
2980 | "vmalloc error: size %lu, failed to allocated page array size %lu", |
2981 | nr_small_pages * PAGE_SIZE, array_size); | |
cd61413b | 2982 | free_vm_area(area); |
1da177e4 LT |
2983 | return NULL; |
2984 | } | |
1da177e4 | 2985 | |
121e6f32 | 2986 | set_vm_area_page_order(area, page_shift - PAGE_SHIFT); |
121e6f32 | 2987 | page_order = vm_area_page_order(area); |
bf53d6f8 | 2988 | |
c3d77172 URS |
2989 | area->nr_pages = vm_area_alloc_pages(gfp_mask | __GFP_NOWARN, |
2990 | node, page_order, nr_small_pages, area->pages); | |
5c1f4e69 | 2991 | |
97105f0a | 2992 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
4e5aa1f4 | 2993 | if (gfp_mask & __GFP_ACCOUNT) { |
3b8000ae | 2994 | int i; |
4e5aa1f4 | 2995 | |
3b8000ae NP |
2996 | for (i = 0; i < area->nr_pages; i++) |
2997 | mod_memcg_page_state(area->pages[i], MEMCG_VMALLOC, 1); | |
4e5aa1f4 | 2998 | } |
1da177e4 | 2999 | |
5c1f4e69 URS |
3000 | /* |
3001 | * If not enough pages were obtained to accomplish an | |
3002 | * allocation request, free them via __vfree() if any. | |
3003 | */ | |
3004 | if (area->nr_pages != nr_small_pages) { | |
c3d77172 | 3005 | warn_alloc(gfp_mask, NULL, |
f4bdfeaf | 3006 | "vmalloc error: size %lu, page order %u, failed to allocate pages", |
5c1f4e69 URS |
3007 | area->nr_pages * PAGE_SIZE, page_order); |
3008 | goto fail; | |
3009 | } | |
3010 | ||
451769eb MH |
3011 | /* |
3012 | * page tables allocations ignore external gfp mask, enforce it | |
3013 | * by the scope API | |
3014 | */ | |
3015 | if ((gfp_mask & (__GFP_FS | __GFP_IO)) == __GFP_IO) | |
3016 | flags = memalloc_nofs_save(); | |
3017 | else if ((gfp_mask & (__GFP_FS | __GFP_IO)) == 0) | |
3018 | flags = memalloc_noio_save(); | |
3019 | ||
9376130c MH |
3020 | do { |
3021 | ret = vmap_pages_range(addr, addr + size, prot, area->pages, | |
451769eb | 3022 | page_shift); |
9376130c MH |
3023 | if (nofail && (ret < 0)) |
3024 | schedule_timeout_uninterruptible(1); | |
3025 | } while (nofail && (ret < 0)); | |
451769eb MH |
3026 | |
3027 | if ((gfp_mask & (__GFP_FS | __GFP_IO)) == __GFP_IO) | |
3028 | memalloc_nofs_restore(flags); | |
3029 | else if ((gfp_mask & (__GFP_FS | __GFP_IO)) == 0) | |
3030 | memalloc_noio_restore(flags); | |
3031 | ||
3032 | if (ret < 0) { | |
c3d77172 | 3033 | warn_alloc(gfp_mask, NULL, |
f4bdfeaf URS |
3034 | "vmalloc error: size %lu, failed to map pages", |
3035 | area->nr_pages * PAGE_SIZE); | |
1da177e4 | 3036 | goto fail; |
d70bec8c | 3037 | } |
ed1f324c | 3038 | |
1da177e4 LT |
3039 | return area->addr; |
3040 | ||
3041 | fail: | |
c67dc624 | 3042 | __vfree(area->addr); |
1da177e4 LT |
3043 | return NULL; |
3044 | } | |
3045 | ||
3046 | /** | |
92eac168 MR |
3047 | * __vmalloc_node_range - allocate virtually contiguous memory |
3048 | * @size: allocation size | |
3049 | * @align: desired alignment | |
3050 | * @start: vm area range start | |
3051 | * @end: vm area range end | |
3052 | * @gfp_mask: flags for the page level allocator | |
3053 | * @prot: protection mask for the allocated pages | |
3054 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
3055 | * @node: node to use for allocation or NUMA_NO_NODE | |
3056 | * @caller: caller's return address | |
3057 | * | |
3058 | * Allocate enough pages to cover @size from the page level | |
b7d90e7a | 3059 | * allocator with @gfp_mask flags. Please note that the full set of gfp |
30d3f011 MH |
3060 | * flags are not supported. GFP_KERNEL, GFP_NOFS and GFP_NOIO are all |
3061 | * supported. | |
3062 | * Zone modifiers are not supported. From the reclaim modifiers | |
3063 | * __GFP_DIRECT_RECLAIM is required (aka GFP_NOWAIT is not supported) | |
3064 | * and only __GFP_NOFAIL is supported (i.e. __GFP_NORETRY and | |
3065 | * __GFP_RETRY_MAYFAIL are not supported). | |
3066 | * | |
3067 | * __GFP_NOWARN can be used to suppress failures messages. | |
b7d90e7a MH |
3068 | * |
3069 | * Map them into contiguous kernel virtual space, using a pagetable | |
3070 | * protection of @prot. | |
a862f68a MR |
3071 | * |
3072 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 3073 | */ |
d0a21265 DR |
3074 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
3075 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
3076 | pgprot_t prot, unsigned long vm_flags, int node, |
3077 | const void *caller) | |
1da177e4 LT |
3078 | { |
3079 | struct vm_struct *area; | |
19f1c3ac | 3080 | void *ret; |
f6e39794 | 3081 | kasan_vmalloc_flags_t kasan_flags = KASAN_VMALLOC_NONE; |
89219d37 | 3082 | unsigned long real_size = size; |
121e6f32 NP |
3083 | unsigned long real_align = align; |
3084 | unsigned int shift = PAGE_SHIFT; | |
1da177e4 | 3085 | |
d70bec8c NP |
3086 | if (WARN_ON_ONCE(!size)) |
3087 | return NULL; | |
3088 | ||
3089 | if ((size >> PAGE_SHIFT) > totalram_pages()) { | |
3090 | warn_alloc(gfp_mask, NULL, | |
f4bdfeaf URS |
3091 | "vmalloc error: size %lu, exceeds total pages", |
3092 | real_size); | |
d70bec8c | 3093 | return NULL; |
121e6f32 NP |
3094 | } |
3095 | ||
559089e0 | 3096 | if (vmap_allow_huge && (vm_flags & VM_ALLOW_HUGE_VMAP)) { |
121e6f32 | 3097 | unsigned long size_per_node; |
1da177e4 | 3098 | |
121e6f32 NP |
3099 | /* |
3100 | * Try huge pages. Only try for PAGE_KERNEL allocations, | |
3101 | * others like modules don't yet expect huge pages in | |
3102 | * their allocations due to apply_to_page_range not | |
3103 | * supporting them. | |
3104 | */ | |
3105 | ||
3106 | size_per_node = size; | |
3107 | if (node == NUMA_NO_NODE) | |
3108 | size_per_node /= num_online_nodes(); | |
3382bbee | 3109 | if (arch_vmap_pmd_supported(prot) && size_per_node >= PMD_SIZE) |
121e6f32 | 3110 | shift = PMD_SHIFT; |
3382bbee CL |
3111 | else |
3112 | shift = arch_vmap_pte_supported_shift(size_per_node); | |
3113 | ||
3114 | align = max(real_align, 1UL << shift); | |
3115 | size = ALIGN(real_size, 1UL << shift); | |
121e6f32 NP |
3116 | } |
3117 | ||
3118 | again: | |
7ca3027b DA |
3119 | area = __get_vm_area_node(real_size, align, shift, VM_ALLOC | |
3120 | VM_UNINITIALIZED | vm_flags, start, end, node, | |
3121 | gfp_mask, caller); | |
d70bec8c | 3122 | if (!area) { |
9376130c | 3123 | bool nofail = gfp_mask & __GFP_NOFAIL; |
d70bec8c | 3124 | warn_alloc(gfp_mask, NULL, |
9376130c MH |
3125 | "vmalloc error: size %lu, vm_struct allocation failed%s", |
3126 | real_size, (nofail) ? ". Retrying." : ""); | |
3127 | if (nofail) { | |
3128 | schedule_timeout_uninterruptible(1); | |
3129 | goto again; | |
3130 | } | |
de7d2b56 | 3131 | goto fail; |
d70bec8c | 3132 | } |
1da177e4 | 3133 | |
f6e39794 AK |
3134 | /* |
3135 | * Prepare arguments for __vmalloc_area_node() and | |
3136 | * kasan_unpoison_vmalloc(). | |
3137 | */ | |
3138 | if (pgprot_val(prot) == pgprot_val(PAGE_KERNEL)) { | |
3139 | if (kasan_hw_tags_enabled()) { | |
3140 | /* | |
3141 | * Modify protection bits to allow tagging. | |
3142 | * This must be done before mapping. | |
3143 | */ | |
3144 | prot = arch_vmap_pgprot_tagged(prot); | |
01d92c7f | 3145 | |
f6e39794 AK |
3146 | /* |
3147 | * Skip page_alloc poisoning and zeroing for physical | |
3148 | * pages backing VM_ALLOC mapping. Memory is instead | |
3149 | * poisoned and zeroed by kasan_unpoison_vmalloc(). | |
3150 | */ | |
3151 | gfp_mask |= __GFP_SKIP_KASAN_UNPOISON | __GFP_SKIP_ZERO; | |
3152 | } | |
3153 | ||
3154 | /* Take note that the mapping is PAGE_KERNEL. */ | |
3155 | kasan_flags |= KASAN_VMALLOC_PROT_NORMAL; | |
23689e91 AK |
3156 | } |
3157 | ||
01d92c7f | 3158 | /* Allocate physical pages and map them into vmalloc space. */ |
19f1c3ac AK |
3159 | ret = __vmalloc_area_node(area, gfp_mask, prot, shift, node); |
3160 | if (!ret) | |
121e6f32 | 3161 | goto fail; |
89219d37 | 3162 | |
23689e91 AK |
3163 | /* |
3164 | * Mark the pages as accessible, now that they are mapped. | |
3165 | * The init condition should match the one in post_alloc_hook() | |
3166 | * (except for the should_skip_init() check) to make sure that memory | |
3167 | * is initialized under the same conditions regardless of the enabled | |
3168 | * KASAN mode. | |
f6e39794 AK |
3169 | * Tag-based KASAN modes only assign tags to normal non-executable |
3170 | * allocations, see __kasan_unpoison_vmalloc(). | |
23689e91 | 3171 | */ |
f6e39794 | 3172 | kasan_flags |= KASAN_VMALLOC_VM_ALLOC; |
23689e91 AK |
3173 | if (!want_init_on_free() && want_init_on_alloc(gfp_mask)) |
3174 | kasan_flags |= KASAN_VMALLOC_INIT; | |
f6e39794 | 3175 | /* KASAN_VMALLOC_PROT_NORMAL already set if required. */ |
23689e91 | 3176 | area->addr = kasan_unpoison_vmalloc(area->addr, real_size, kasan_flags); |
19f1c3ac | 3177 | |
f5252e00 | 3178 | /* |
20fc02b4 ZY |
3179 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
3180 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 3181 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 3182 | */ |
20fc02b4 | 3183 | clear_vm_uninitialized_flag(area); |
f5252e00 | 3184 | |
7ca3027b | 3185 | size = PAGE_ALIGN(size); |
60115fa5 KW |
3186 | if (!(vm_flags & VM_DEFER_KMEMLEAK)) |
3187 | kmemleak_vmalloc(area, size, gfp_mask); | |
89219d37 | 3188 | |
19f1c3ac | 3189 | return area->addr; |
de7d2b56 JP |
3190 | |
3191 | fail: | |
121e6f32 NP |
3192 | if (shift > PAGE_SHIFT) { |
3193 | shift = PAGE_SHIFT; | |
3194 | align = real_align; | |
3195 | size = real_size; | |
3196 | goto again; | |
3197 | } | |
3198 | ||
de7d2b56 | 3199 | return NULL; |
1da177e4 LT |
3200 | } |
3201 | ||
d0a21265 | 3202 | /** |
92eac168 MR |
3203 | * __vmalloc_node - allocate virtually contiguous memory |
3204 | * @size: allocation size | |
3205 | * @align: desired alignment | |
3206 | * @gfp_mask: flags for the page level allocator | |
92eac168 MR |
3207 | * @node: node to use for allocation or NUMA_NO_NODE |
3208 | * @caller: caller's return address | |
a7c3e901 | 3209 | * |
f38fcb9c CH |
3210 | * Allocate enough pages to cover @size from the page level allocator with |
3211 | * @gfp_mask flags. Map them into contiguous kernel virtual space. | |
a7c3e901 | 3212 | * |
92eac168 MR |
3213 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
3214 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 3215 | * |
92eac168 MR |
3216 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
3217 | * with mm people. | |
a862f68a MR |
3218 | * |
3219 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 3220 | */ |
2b905948 | 3221 | void *__vmalloc_node(unsigned long size, unsigned long align, |
f38fcb9c | 3222 | gfp_t gfp_mask, int node, const void *caller) |
d0a21265 DR |
3223 | { |
3224 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
f38fcb9c | 3225 | gfp_mask, PAGE_KERNEL, 0, node, caller); |
d0a21265 | 3226 | } |
c3f896dc CH |
3227 | /* |
3228 | * This is only for performance analysis of vmalloc and stress purpose. | |
3229 | * It is required by vmalloc test module, therefore do not use it other | |
3230 | * than that. | |
3231 | */ | |
3232 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
3233 | EXPORT_SYMBOL_GPL(__vmalloc_node); | |
3234 | #endif | |
d0a21265 | 3235 | |
88dca4ca | 3236 | void *__vmalloc(unsigned long size, gfp_t gfp_mask) |
930fc45a | 3237 | { |
f38fcb9c | 3238 | return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE, |
23016969 | 3239 | __builtin_return_address(0)); |
930fc45a | 3240 | } |
1da177e4 LT |
3241 | EXPORT_SYMBOL(__vmalloc); |
3242 | ||
3243 | /** | |
92eac168 MR |
3244 | * vmalloc - allocate virtually contiguous memory |
3245 | * @size: allocation size | |
3246 | * | |
3247 | * Allocate enough pages to cover @size from the page level | |
3248 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 3249 | * |
92eac168 MR |
3250 | * For tight control over page level allocator and protection flags |
3251 | * use __vmalloc() instead. | |
a862f68a MR |
3252 | * |
3253 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
3254 | */ |
3255 | void *vmalloc(unsigned long size) | |
3256 | { | |
4d39d728 CH |
3257 | return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE, |
3258 | __builtin_return_address(0)); | |
1da177e4 | 3259 | } |
1da177e4 LT |
3260 | EXPORT_SYMBOL(vmalloc); |
3261 | ||
15a64f5a | 3262 | /** |
559089e0 SL |
3263 | * vmalloc_huge - allocate virtually contiguous memory, allow huge pages |
3264 | * @size: allocation size | |
3265 | * @gfp_mask: flags for the page level allocator | |
15a64f5a | 3266 | * |
559089e0 | 3267 | * Allocate enough pages to cover @size from the page level |
15a64f5a | 3268 | * allocator and map them into contiguous kernel virtual space. |
559089e0 SL |
3269 | * If @size is greater than or equal to PMD_SIZE, allow using |
3270 | * huge pages for the memory | |
15a64f5a CI |
3271 | * |
3272 | * Return: pointer to the allocated memory or %NULL on error | |
3273 | */ | |
559089e0 | 3274 | void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) |
15a64f5a CI |
3275 | { |
3276 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, | |
559089e0 | 3277 | gfp_mask, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP, |
15a64f5a CI |
3278 | NUMA_NO_NODE, __builtin_return_address(0)); |
3279 | } | |
559089e0 | 3280 | EXPORT_SYMBOL_GPL(vmalloc_huge); |
15a64f5a | 3281 | |
e1ca7788 | 3282 | /** |
92eac168 MR |
3283 | * vzalloc - allocate virtually contiguous memory with zero fill |
3284 | * @size: allocation size | |
3285 | * | |
3286 | * Allocate enough pages to cover @size from the page level | |
3287 | * allocator and map them into contiguous kernel virtual space. | |
3288 | * The memory allocated is set to zero. | |
3289 | * | |
3290 | * For tight control over page level allocator and protection flags | |
3291 | * use __vmalloc() instead. | |
a862f68a MR |
3292 | * |
3293 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
3294 | */ |
3295 | void *vzalloc(unsigned long size) | |
3296 | { | |
4d39d728 CH |
3297 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE, |
3298 | __builtin_return_address(0)); | |
e1ca7788 DY |
3299 | } |
3300 | EXPORT_SYMBOL(vzalloc); | |
3301 | ||
83342314 | 3302 | /** |
ead04089 REB |
3303 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
3304 | * @size: allocation size | |
83342314 | 3305 | * |
ead04089 REB |
3306 | * The resulting memory area is zeroed so it can be mapped to userspace |
3307 | * without leaking data. | |
a862f68a MR |
3308 | * |
3309 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
3310 | */ |
3311 | void *vmalloc_user(unsigned long size) | |
3312 | { | |
bc84c535 RP |
3313 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
3314 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
3315 | VM_USERMAP, NUMA_NO_NODE, | |
3316 | __builtin_return_address(0)); | |
83342314 NP |
3317 | } |
3318 | EXPORT_SYMBOL(vmalloc_user); | |
3319 | ||
930fc45a | 3320 | /** |
92eac168 MR |
3321 | * vmalloc_node - allocate memory on a specific node |
3322 | * @size: allocation size | |
3323 | * @node: numa node | |
930fc45a | 3324 | * |
92eac168 MR |
3325 | * Allocate enough pages to cover @size from the page level |
3326 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 3327 | * |
92eac168 MR |
3328 | * For tight control over page level allocator and protection flags |
3329 | * use __vmalloc() instead. | |
a862f68a MR |
3330 | * |
3331 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
3332 | */ |
3333 | void *vmalloc_node(unsigned long size, int node) | |
3334 | { | |
f38fcb9c CH |
3335 | return __vmalloc_node(size, 1, GFP_KERNEL, node, |
3336 | __builtin_return_address(0)); | |
930fc45a CL |
3337 | } |
3338 | EXPORT_SYMBOL(vmalloc_node); | |
3339 | ||
e1ca7788 DY |
3340 | /** |
3341 | * vzalloc_node - allocate memory on a specific node with zero fill | |
3342 | * @size: allocation size | |
3343 | * @node: numa node | |
3344 | * | |
3345 | * Allocate enough pages to cover @size from the page level | |
3346 | * allocator and map them into contiguous kernel virtual space. | |
3347 | * The memory allocated is set to zero. | |
3348 | * | |
a862f68a | 3349 | * Return: pointer to the allocated memory or %NULL on error |
e1ca7788 DY |
3350 | */ |
3351 | void *vzalloc_node(unsigned long size, int node) | |
3352 | { | |
4d39d728 CH |
3353 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node, |
3354 | __builtin_return_address(0)); | |
e1ca7788 DY |
3355 | } |
3356 | EXPORT_SYMBOL(vzalloc_node); | |
3357 | ||
0d08e0d3 | 3358 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 3359 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 3360 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 3361 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 3362 | #else |
698d0831 MH |
3363 | /* |
3364 | * 64b systems should always have either DMA or DMA32 zones. For others | |
3365 | * GFP_DMA32 should do the right thing and use the normal zone. | |
3366 | */ | |
68d68ff6 | 3367 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 AK |
3368 | #endif |
3369 | ||
1da177e4 | 3370 | /** |
92eac168 MR |
3371 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
3372 | * @size: allocation size | |
1da177e4 | 3373 | * |
92eac168 MR |
3374 | * Allocate enough 32bit PA addressable pages to cover @size from the |
3375 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
3376 | * |
3377 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
3378 | */ |
3379 | void *vmalloc_32(unsigned long size) | |
3380 | { | |
f38fcb9c CH |
3381 | return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE, |
3382 | __builtin_return_address(0)); | |
1da177e4 | 3383 | } |
1da177e4 LT |
3384 | EXPORT_SYMBOL(vmalloc_32); |
3385 | ||
83342314 | 3386 | /** |
ead04089 | 3387 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 3388 | * @size: allocation size |
ead04089 REB |
3389 | * |
3390 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
3391 | * mapped to userspace without leaking data. | |
a862f68a MR |
3392 | * |
3393 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
3394 | */ |
3395 | void *vmalloc_32_user(unsigned long size) | |
3396 | { | |
bc84c535 RP |
3397 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
3398 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
3399 | VM_USERMAP, NUMA_NO_NODE, | |
3400 | __builtin_return_address(0)); | |
83342314 NP |
3401 | } |
3402 | EXPORT_SYMBOL(vmalloc_32_user); | |
3403 | ||
d0107eb0 KH |
3404 | /* |
3405 | * small helper routine , copy contents to buf from addr. | |
3406 | * If the page is not present, fill zero. | |
3407 | */ | |
3408 | ||
3409 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
3410 | { | |
3411 | struct page *p; | |
3412 | int copied = 0; | |
3413 | ||
3414 | while (count) { | |
3415 | unsigned long offset, length; | |
3416 | ||
891c49ab | 3417 | offset = offset_in_page(addr); |
d0107eb0 KH |
3418 | length = PAGE_SIZE - offset; |
3419 | if (length > count) | |
3420 | length = count; | |
3421 | p = vmalloc_to_page(addr); | |
3422 | /* | |
3423 | * To do safe access to this _mapped_ area, we need | |
3424 | * lock. But adding lock here means that we need to add | |
f0953a1b | 3425 | * overhead of vmalloc()/vfree() calls for this _debug_ |
d0107eb0 KH |
3426 | * interface, rarely used. Instead of that, we'll use |
3427 | * kmap() and get small overhead in this access function. | |
3428 | */ | |
3429 | if (p) { | |
f7c8ce44 | 3430 | /* We can expect USER0 is not used -- see vread() */ |
9b04c5fe | 3431 | void *map = kmap_atomic(p); |
d0107eb0 | 3432 | memcpy(buf, map + offset, length); |
9b04c5fe | 3433 | kunmap_atomic(map); |
d0107eb0 KH |
3434 | } else |
3435 | memset(buf, 0, length); | |
3436 | ||
3437 | addr += length; | |
3438 | buf += length; | |
3439 | copied += length; | |
3440 | count -= length; | |
3441 | } | |
3442 | return copied; | |
3443 | } | |
3444 | ||
d0107eb0 | 3445 | /** |
92eac168 MR |
3446 | * vread() - read vmalloc area in a safe way. |
3447 | * @buf: buffer for reading data | |
3448 | * @addr: vm address. | |
3449 | * @count: number of bytes to be read. | |
3450 | * | |
92eac168 MR |
3451 | * This function checks that addr is a valid vmalloc'ed area, and |
3452 | * copy data from that area to a given buffer. If the given memory range | |
3453 | * of [addr...addr+count) includes some valid address, data is copied to | |
3454 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
3455 | * IOREMAP area is treated as memory hole and no copy is done. | |
3456 | * | |
3457 | * If [addr...addr+count) doesn't includes any intersects with alive | |
3458 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
3459 | * | |
3460 | * Note: In usual ops, vread() is never necessary because the caller | |
3461 | * should know vmalloc() area is valid and can use memcpy(). | |
3462 | * This is for routines which have to access vmalloc area without | |
bbcd53c9 | 3463 | * any information, as /proc/kcore. |
a862f68a MR |
3464 | * |
3465 | * Return: number of bytes for which addr and buf should be increased | |
3466 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
3467 | * include any intersection with valid vmalloc area | |
d0107eb0 | 3468 | */ |
1da177e4 LT |
3469 | long vread(char *buf, char *addr, unsigned long count) |
3470 | { | |
e81ce85f JK |
3471 | struct vmap_area *va; |
3472 | struct vm_struct *vm; | |
1da177e4 | 3473 | char *vaddr, *buf_start = buf; |
d0107eb0 | 3474 | unsigned long buflen = count; |
1da177e4 LT |
3475 | unsigned long n; |
3476 | ||
4aff1dc4 AK |
3477 | addr = kasan_reset_tag(addr); |
3478 | ||
1da177e4 LT |
3479 | /* Don't allow overflow */ |
3480 | if ((unsigned long) addr + count < count) | |
3481 | count = -(unsigned long) addr; | |
3482 | ||
e81ce85f | 3483 | spin_lock(&vmap_area_lock); |
f181234a | 3484 | va = find_vmap_area_exceed_addr((unsigned long)addr); |
f608788c SD |
3485 | if (!va) |
3486 | goto finished; | |
f181234a CW |
3487 | |
3488 | /* no intersects with alive vmap_area */ | |
3489 | if ((unsigned long)addr + count <= va->va_start) | |
3490 | goto finished; | |
3491 | ||
f608788c | 3492 | list_for_each_entry_from(va, &vmap_area_list, list) { |
e81ce85f JK |
3493 | if (!count) |
3494 | break; | |
3495 | ||
688fcbfc | 3496 | if (!va->vm) |
e81ce85f JK |
3497 | continue; |
3498 | ||
3499 | vm = va->vm; | |
3500 | vaddr = (char *) vm->addr; | |
762216ab | 3501 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
3502 | continue; |
3503 | while (addr < vaddr) { | |
3504 | if (count == 0) | |
3505 | goto finished; | |
3506 | *buf = '\0'; | |
3507 | buf++; | |
3508 | addr++; | |
3509 | count--; | |
3510 | } | |
762216ab | 3511 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
3512 | if (n > count) |
3513 | n = count; | |
e81ce85f | 3514 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
3515 | aligned_vread(buf, addr, n); |
3516 | else /* IOREMAP area is treated as memory hole */ | |
3517 | memset(buf, 0, n); | |
3518 | buf += n; | |
3519 | addr += n; | |
3520 | count -= n; | |
1da177e4 LT |
3521 | } |
3522 | finished: | |
e81ce85f | 3523 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3524 | |
3525 | if (buf == buf_start) | |
3526 | return 0; | |
3527 | /* zero-fill memory holes */ | |
3528 | if (buf != buf_start + buflen) | |
3529 | memset(buf, 0, buflen - (buf - buf_start)); | |
3530 | ||
3531 | return buflen; | |
1da177e4 LT |
3532 | } |
3533 | ||
83342314 | 3534 | /** |
92eac168 MR |
3535 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
3536 | * @vma: vma to cover | |
3537 | * @uaddr: target user address to start at | |
3538 | * @kaddr: virtual address of vmalloc kernel memory | |
bdebd6a2 | 3539 | * @pgoff: offset from @kaddr to start at |
92eac168 | 3540 | * @size: size of map area |
7682486b | 3541 | * |
92eac168 | 3542 | * Returns: 0 for success, -Exxx on failure |
83342314 | 3543 | * |
92eac168 MR |
3544 | * This function checks that @kaddr is a valid vmalloc'ed area, |
3545 | * and that it is big enough to cover the range starting at | |
3546 | * @uaddr in @vma. Will return failure if that criteria isn't | |
3547 | * met. | |
83342314 | 3548 | * |
92eac168 | 3549 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 3550 | */ |
e69e9d4a | 3551 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
bdebd6a2 JH |
3552 | void *kaddr, unsigned long pgoff, |
3553 | unsigned long size) | |
83342314 NP |
3554 | { |
3555 | struct vm_struct *area; | |
bdebd6a2 JH |
3556 | unsigned long off; |
3557 | unsigned long end_index; | |
3558 | ||
3559 | if (check_shl_overflow(pgoff, PAGE_SHIFT, &off)) | |
3560 | return -EINVAL; | |
83342314 | 3561 | |
e69e9d4a HD |
3562 | size = PAGE_ALIGN(size); |
3563 | ||
3564 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
3565 | return -EINVAL; |
3566 | ||
e69e9d4a | 3567 | area = find_vm_area(kaddr); |
83342314 | 3568 | if (!area) |
db64fe02 | 3569 | return -EINVAL; |
83342314 | 3570 | |
fe9041c2 | 3571 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3572 | return -EINVAL; |
83342314 | 3573 | |
bdebd6a2 JH |
3574 | if (check_add_overflow(size, off, &end_index) || |
3575 | end_index > get_vm_area_size(area)) | |
db64fe02 | 3576 | return -EINVAL; |
bdebd6a2 | 3577 | kaddr += off; |
83342314 | 3578 | |
83342314 | 3579 | do { |
e69e9d4a | 3580 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3581 | int ret; |
3582 | ||
83342314 NP |
3583 | ret = vm_insert_page(vma, uaddr, page); |
3584 | if (ret) | |
3585 | return ret; | |
3586 | ||
3587 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3588 | kaddr += PAGE_SIZE; |
3589 | size -= PAGE_SIZE; | |
3590 | } while (size > 0); | |
83342314 | 3591 | |
314e51b9 | 3592 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3593 | |
db64fe02 | 3594 | return 0; |
83342314 | 3595 | } |
e69e9d4a HD |
3596 | |
3597 | /** | |
92eac168 MR |
3598 | * remap_vmalloc_range - map vmalloc pages to userspace |
3599 | * @vma: vma to cover (map full range of vma) | |
3600 | * @addr: vmalloc memory | |
3601 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3602 | * |
92eac168 | 3603 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3604 | * |
92eac168 MR |
3605 | * This function checks that addr is a valid vmalloc'ed area, and |
3606 | * that it is big enough to cover the vma. Will return failure if | |
3607 | * that criteria isn't met. | |
e69e9d4a | 3608 | * |
92eac168 | 3609 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3610 | */ |
3611 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3612 | unsigned long pgoff) | |
3613 | { | |
3614 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
bdebd6a2 | 3615 | addr, pgoff, |
e69e9d4a HD |
3616 | vma->vm_end - vma->vm_start); |
3617 | } | |
83342314 NP |
3618 | EXPORT_SYMBOL(remap_vmalloc_range); |
3619 | ||
5f4352fb JF |
3620 | void free_vm_area(struct vm_struct *area) |
3621 | { | |
3622 | struct vm_struct *ret; | |
3623 | ret = remove_vm_area(area->addr); | |
3624 | BUG_ON(ret != area); | |
3625 | kfree(area); | |
3626 | } | |
3627 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3628 | |
4f8b02b4 | 3629 | #ifdef CONFIG_SMP |
ca23e405 TH |
3630 | static struct vmap_area *node_to_va(struct rb_node *n) |
3631 | { | |
4583e773 | 3632 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3633 | } |
3634 | ||
3635 | /** | |
68ad4a33 URS |
3636 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3637 | * @addr: target address | |
ca23e405 | 3638 | * |
68ad4a33 URS |
3639 | * Returns: vmap_area if it is found. If there is no such area |
3640 | * the first highest(reverse order) vmap_area is returned | |
3641 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3642 | * if there are no any areas before @addr. | |
ca23e405 | 3643 | */ |
68ad4a33 URS |
3644 | static struct vmap_area * |
3645 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3646 | { |
68ad4a33 URS |
3647 | struct vmap_area *va, *tmp; |
3648 | struct rb_node *n; | |
3649 | ||
3650 | n = free_vmap_area_root.rb_node; | |
3651 | va = NULL; | |
ca23e405 TH |
3652 | |
3653 | while (n) { | |
68ad4a33 URS |
3654 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3655 | if (tmp->va_start <= addr) { | |
3656 | va = tmp; | |
3657 | if (tmp->va_end >= addr) | |
3658 | break; | |
3659 | ||
ca23e405 | 3660 | n = n->rb_right; |
68ad4a33 URS |
3661 | } else { |
3662 | n = n->rb_left; | |
3663 | } | |
ca23e405 TH |
3664 | } |
3665 | ||
68ad4a33 | 3666 | return va; |
ca23e405 TH |
3667 | } |
3668 | ||
3669 | /** | |
68ad4a33 URS |
3670 | * pvm_determine_end_from_reverse - find the highest aligned address |
3671 | * of free block below VMALLOC_END | |
3672 | * @va: | |
3673 | * in - the VA we start the search(reverse order); | |
3674 | * out - the VA with the highest aligned end address. | |
799fa85d | 3675 | * @align: alignment for required highest address |
ca23e405 | 3676 | * |
68ad4a33 | 3677 | * Returns: determined end address within vmap_area |
ca23e405 | 3678 | */ |
68ad4a33 URS |
3679 | static unsigned long |
3680 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3681 | { |
68ad4a33 | 3682 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3683 | unsigned long addr; |
3684 | ||
68ad4a33 URS |
3685 | if (likely(*va)) { |
3686 | list_for_each_entry_from_reverse((*va), | |
3687 | &free_vmap_area_list, list) { | |
3688 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3689 | if ((*va)->va_start < addr) | |
3690 | return addr; | |
3691 | } | |
ca23e405 TH |
3692 | } |
3693 | ||
68ad4a33 | 3694 | return 0; |
ca23e405 TH |
3695 | } |
3696 | ||
3697 | /** | |
3698 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3699 | * @offsets: array containing offset of each area | |
3700 | * @sizes: array containing size of each area | |
3701 | * @nr_vms: the number of areas to allocate | |
3702 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3703 | * |
3704 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3705 | * vm_structs on success, %NULL on failure | |
3706 | * | |
3707 | * Percpu allocator wants to use congruent vm areas so that it can | |
3708 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3709 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3710 | * be scattered pretty far, distance between two areas easily going up | |
3711 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3712 | * areas are allocated from top. | |
ca23e405 | 3713 | * |
68ad4a33 URS |
3714 | * Despite its complicated look, this allocator is rather simple. It |
3715 | * does everything top-down and scans free blocks from the end looking | |
3716 | * for matching base. While scanning, if any of the areas do not fit the | |
3717 | * base address is pulled down to fit the area. Scanning is repeated till | |
3718 | * all the areas fit and then all necessary data structures are inserted | |
3719 | * and the result is returned. | |
ca23e405 TH |
3720 | */ |
3721 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3722 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3723 | size_t align) |
ca23e405 TH |
3724 | { |
3725 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3726 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3727 | struct vmap_area **vas, *va; |
ca23e405 TH |
3728 | struct vm_struct **vms; |
3729 | int area, area2, last_area, term_area; | |
253a496d | 3730 | unsigned long base, start, size, end, last_end, orig_start, orig_end; |
ca23e405 TH |
3731 | bool purged = false; |
3732 | ||
ca23e405 | 3733 | /* verify parameters and allocate data structures */ |
891c49ab | 3734 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3735 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3736 | start = offsets[area]; | |
3737 | end = start + sizes[area]; | |
3738 | ||
3739 | /* is everything aligned properly? */ | |
3740 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3741 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3742 | ||
3743 | /* detect the area with the highest address */ | |
3744 | if (start > offsets[last_area]) | |
3745 | last_area = area; | |
3746 | ||
c568da28 | 3747 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3748 | unsigned long start2 = offsets[area2]; |
3749 | unsigned long end2 = start2 + sizes[area2]; | |
3750 | ||
c568da28 | 3751 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3752 | } |
3753 | } | |
3754 | last_end = offsets[last_area] + sizes[last_area]; | |
3755 | ||
3756 | if (vmalloc_end - vmalloc_start < last_end) { | |
3757 | WARN_ON(true); | |
3758 | return NULL; | |
3759 | } | |
3760 | ||
4d67d860 TM |
3761 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3762 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3763 | if (!vas || !vms) |
f1db7afd | 3764 | goto err_free2; |
ca23e405 TH |
3765 | |
3766 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3767 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3768 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3769 | if (!vas[area] || !vms[area]) |
3770 | goto err_free; | |
3771 | } | |
3772 | retry: | |
e36176be | 3773 | spin_lock(&free_vmap_area_lock); |
ca23e405 TH |
3774 | |
3775 | /* start scanning - we scan from the top, begin with the last area */ | |
3776 | area = term_area = last_area; | |
3777 | start = offsets[area]; | |
3778 | end = start + sizes[area]; | |
3779 | ||
68ad4a33 URS |
3780 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3781 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3782 | |
3783 | while (true) { | |
ca23e405 TH |
3784 | /* |
3785 | * base might have underflowed, add last_end before | |
3786 | * comparing. | |
3787 | */ | |
68ad4a33 URS |
3788 | if (base + last_end < vmalloc_start + last_end) |
3789 | goto overflow; | |
ca23e405 TH |
3790 | |
3791 | /* | |
68ad4a33 | 3792 | * Fitting base has not been found. |
ca23e405 | 3793 | */ |
68ad4a33 URS |
3794 | if (va == NULL) |
3795 | goto overflow; | |
ca23e405 | 3796 | |
5336e52c | 3797 | /* |
d8cc323d | 3798 | * If required width exceeds current VA block, move |
5336e52c KS |
3799 | * base downwards and then recheck. |
3800 | */ | |
3801 | if (base + end > va->va_end) { | |
3802 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3803 | term_area = area; | |
3804 | continue; | |
3805 | } | |
3806 | ||
ca23e405 | 3807 | /* |
68ad4a33 | 3808 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3809 | */ |
5336e52c | 3810 | if (base + start < va->va_start) { |
68ad4a33 URS |
3811 | va = node_to_va(rb_prev(&va->rb_node)); |
3812 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3813 | term_area = area; |
3814 | continue; | |
3815 | } | |
3816 | ||
3817 | /* | |
3818 | * This area fits, move on to the previous one. If | |
3819 | * the previous one is the terminal one, we're done. | |
3820 | */ | |
3821 | area = (area + nr_vms - 1) % nr_vms; | |
3822 | if (area == term_area) | |
3823 | break; | |
68ad4a33 | 3824 | |
ca23e405 TH |
3825 | start = offsets[area]; |
3826 | end = start + sizes[area]; | |
68ad4a33 | 3827 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3828 | } |
68ad4a33 | 3829 | |
ca23e405 TH |
3830 | /* we've found a fitting base, insert all va's */ |
3831 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3832 | int ret; |
ca23e405 | 3833 | |
68ad4a33 URS |
3834 | start = base + offsets[area]; |
3835 | size = sizes[area]; | |
ca23e405 | 3836 | |
68ad4a33 URS |
3837 | va = pvm_find_va_enclose_addr(start); |
3838 | if (WARN_ON_ONCE(va == NULL)) | |
3839 | /* It is a BUG(), but trigger recovery instead. */ | |
3840 | goto recovery; | |
3841 | ||
1b23ff80 BH |
3842 | ret = adjust_va_to_fit_type(va, start, size); |
3843 | if (WARN_ON_ONCE(unlikely(ret))) | |
68ad4a33 URS |
3844 | /* It is a BUG(), but trigger recovery instead. */ |
3845 | goto recovery; | |
3846 | ||
68ad4a33 URS |
3847 | /* Allocated area. */ |
3848 | va = vas[area]; | |
3849 | va->va_start = start; | |
3850 | va->va_end = start + size; | |
68ad4a33 | 3851 | } |
ca23e405 | 3852 | |
e36176be | 3853 | spin_unlock(&free_vmap_area_lock); |
ca23e405 | 3854 | |
253a496d DA |
3855 | /* populate the kasan shadow space */ |
3856 | for (area = 0; area < nr_vms; area++) { | |
3857 | if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area])) | |
3858 | goto err_free_shadow; | |
253a496d DA |
3859 | } |
3860 | ||
ca23e405 | 3861 | /* insert all vm's */ |
e36176be URS |
3862 | spin_lock(&vmap_area_lock); |
3863 | for (area = 0; area < nr_vms; area++) { | |
3864 | insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list); | |
3865 | ||
3866 | setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC, | |
3645cb4a | 3867 | pcpu_get_vm_areas); |
e36176be URS |
3868 | } |
3869 | spin_unlock(&vmap_area_lock); | |
ca23e405 | 3870 | |
19f1c3ac AK |
3871 | /* |
3872 | * Mark allocated areas as accessible. Do it now as a best-effort | |
3873 | * approach, as they can be mapped outside of vmalloc code. | |
23689e91 AK |
3874 | * With hardware tag-based KASAN, marking is skipped for |
3875 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). | |
19f1c3ac | 3876 | */ |
1d96320f AK |
3877 | for (area = 0; area < nr_vms; area++) |
3878 | vms[area]->addr = kasan_unpoison_vmalloc(vms[area]->addr, | |
f6e39794 | 3879 | vms[area]->size, KASAN_VMALLOC_PROT_NORMAL); |
1d96320f | 3880 | |
ca23e405 TH |
3881 | kfree(vas); |
3882 | return vms; | |
3883 | ||
68ad4a33 | 3884 | recovery: |
e36176be URS |
3885 | /* |
3886 | * Remove previously allocated areas. There is no | |
3887 | * need in removing these areas from the busy tree, | |
3888 | * because they are inserted only on the final step | |
3889 | * and when pcpu_get_vm_areas() is success. | |
3890 | */ | |
68ad4a33 | 3891 | while (area--) { |
253a496d DA |
3892 | orig_start = vas[area]->va_start; |
3893 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
3894 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
3895 | &free_vmap_area_list); | |
9c801f61 URS |
3896 | if (va) |
3897 | kasan_release_vmalloc(orig_start, orig_end, | |
3898 | va->va_start, va->va_end); | |
68ad4a33 URS |
3899 | vas[area] = NULL; |
3900 | } | |
3901 | ||
3902 | overflow: | |
e36176be | 3903 | spin_unlock(&free_vmap_area_lock); |
68ad4a33 URS |
3904 | if (!purged) { |
3905 | purge_vmap_area_lazy(); | |
3906 | purged = true; | |
3907 | ||
3908 | /* Before "retry", check if we recover. */ | |
3909 | for (area = 0; area < nr_vms; area++) { | |
3910 | if (vas[area]) | |
3911 | continue; | |
3912 | ||
3913 | vas[area] = kmem_cache_zalloc( | |
3914 | vmap_area_cachep, GFP_KERNEL); | |
3915 | if (!vas[area]) | |
3916 | goto err_free; | |
3917 | } | |
3918 | ||
3919 | goto retry; | |
3920 | } | |
3921 | ||
ca23e405 TH |
3922 | err_free: |
3923 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3924 | if (vas[area]) |
3925 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3926 | ||
f1db7afd | 3927 | kfree(vms[area]); |
ca23e405 | 3928 | } |
f1db7afd | 3929 | err_free2: |
ca23e405 TH |
3930 | kfree(vas); |
3931 | kfree(vms); | |
3932 | return NULL; | |
253a496d DA |
3933 | |
3934 | err_free_shadow: | |
3935 | spin_lock(&free_vmap_area_lock); | |
3936 | /* | |
3937 | * We release all the vmalloc shadows, even the ones for regions that | |
3938 | * hadn't been successfully added. This relies on kasan_release_vmalloc | |
3939 | * being able to tolerate this case. | |
3940 | */ | |
3941 | for (area = 0; area < nr_vms; area++) { | |
3942 | orig_start = vas[area]->va_start; | |
3943 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
3944 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
3945 | &free_vmap_area_list); | |
9c801f61 URS |
3946 | if (va) |
3947 | kasan_release_vmalloc(orig_start, orig_end, | |
3948 | va->va_start, va->va_end); | |
253a496d DA |
3949 | vas[area] = NULL; |
3950 | kfree(vms[area]); | |
3951 | } | |
3952 | spin_unlock(&free_vmap_area_lock); | |
3953 | kfree(vas); | |
3954 | kfree(vms); | |
3955 | return NULL; | |
ca23e405 TH |
3956 | } |
3957 | ||
3958 | /** | |
3959 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3960 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3961 | * @nr_vms: the number of allocated areas | |
3962 | * | |
3963 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3964 | */ | |
3965 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3966 | { | |
3967 | int i; | |
3968 | ||
3969 | for (i = 0; i < nr_vms; i++) | |
3970 | free_vm_area(vms[i]); | |
3971 | kfree(vms); | |
3972 | } | |
4f8b02b4 | 3973 | #endif /* CONFIG_SMP */ |
a10aa579 | 3974 | |
5bb1bb35 | 3975 | #ifdef CONFIG_PRINTK |
98f18083 PM |
3976 | bool vmalloc_dump_obj(void *object) |
3977 | { | |
3978 | struct vm_struct *vm; | |
3979 | void *objp = (void *)PAGE_ALIGN((unsigned long)object); | |
3980 | ||
3981 | vm = find_vm_area(objp); | |
3982 | if (!vm) | |
3983 | return false; | |
bd34dcd4 PM |
3984 | pr_cont(" %u-page vmalloc region starting at %#lx allocated at %pS\n", |
3985 | vm->nr_pages, (unsigned long)vm->addr, vm->caller); | |
98f18083 PM |
3986 | return true; |
3987 | } | |
5bb1bb35 | 3988 | #endif |
98f18083 | 3989 | |
a10aa579 CL |
3990 | #ifdef CONFIG_PROC_FS |
3991 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e36176be | 3992 | __acquires(&vmap_purge_lock) |
d4033afd | 3993 | __acquires(&vmap_area_lock) |
a10aa579 | 3994 | { |
e36176be | 3995 | mutex_lock(&vmap_purge_lock); |
d4033afd | 3996 | spin_lock(&vmap_area_lock); |
e36176be | 3997 | |
3f500069 | 3998 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3999 | } |
4000 | ||
4001 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4002 | { | |
3f500069 | 4003 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
4004 | } |
4005 | ||
4006 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 4007 | __releases(&vmap_area_lock) |
0a7dd4e9 | 4008 | __releases(&vmap_purge_lock) |
a10aa579 | 4009 | { |
d4033afd | 4010 | spin_unlock(&vmap_area_lock); |
0a7dd4e9 | 4011 | mutex_unlock(&vmap_purge_lock); |
a10aa579 CL |
4012 | } |
4013 | ||
a47a126a ED |
4014 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
4015 | { | |
e5adfffc | 4016 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a | 4017 | unsigned int nr, *counters = m->private; |
51e50b3a | 4018 | unsigned int step = 1U << vm_area_page_order(v); |
a47a126a ED |
4019 | |
4020 | if (!counters) | |
4021 | return; | |
4022 | ||
af12346c WL |
4023 | if (v->flags & VM_UNINITIALIZED) |
4024 | return; | |
7e5b528b DV |
4025 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
4026 | smp_rmb(); | |
af12346c | 4027 | |
a47a126a ED |
4028 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
4029 | ||
51e50b3a ED |
4030 | for (nr = 0; nr < v->nr_pages; nr += step) |
4031 | counters[page_to_nid(v->pages[nr])] += step; | |
a47a126a ED |
4032 | for_each_node_state(nr, N_HIGH_MEMORY) |
4033 | if (counters[nr]) | |
4034 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
4035 | } | |
4036 | } | |
4037 | ||
dd3b8353 URS |
4038 | static void show_purge_info(struct seq_file *m) |
4039 | { | |
dd3b8353 URS |
4040 | struct vmap_area *va; |
4041 | ||
96e2db45 URS |
4042 | spin_lock(&purge_vmap_area_lock); |
4043 | list_for_each_entry(va, &purge_vmap_area_list, list) { | |
dd3b8353 URS |
4044 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", |
4045 | (void *)va->va_start, (void *)va->va_end, | |
4046 | va->va_end - va->va_start); | |
4047 | } | |
96e2db45 | 4048 | spin_unlock(&purge_vmap_area_lock); |
dd3b8353 URS |
4049 | } |
4050 | ||
a10aa579 CL |
4051 | static int s_show(struct seq_file *m, void *p) |
4052 | { | |
3f500069 | 4053 | struct vmap_area *va; |
d4033afd JK |
4054 | struct vm_struct *v; |
4055 | ||
3f500069 | 4056 | va = list_entry(p, struct vmap_area, list); |
4057 | ||
c2ce8c14 | 4058 | /* |
688fcbfc PL |
4059 | * s_show can encounter race with remove_vm_area, !vm on behalf |
4060 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 4061 | */ |
688fcbfc | 4062 | if (!va->vm) { |
dd3b8353 | 4063 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 4064 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 4065 | va->va_end - va->va_start); |
78c72746 | 4066 | |
7cc7913e | 4067 | goto final; |
78c72746 | 4068 | } |
d4033afd JK |
4069 | |
4070 | v = va->vm; | |
a10aa579 | 4071 | |
45ec1690 | 4072 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
4073 | v->addr, v->addr + v->size, v->size); |
4074 | ||
62c70bce JP |
4075 | if (v->caller) |
4076 | seq_printf(m, " %pS", v->caller); | |
23016969 | 4077 | |
a10aa579 CL |
4078 | if (v->nr_pages) |
4079 | seq_printf(m, " pages=%d", v->nr_pages); | |
4080 | ||
4081 | if (v->phys_addr) | |
199eaa05 | 4082 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
4083 | |
4084 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 4085 | seq_puts(m, " ioremap"); |
a10aa579 CL |
4086 | |
4087 | if (v->flags & VM_ALLOC) | |
f4527c90 | 4088 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
4089 | |
4090 | if (v->flags & VM_MAP) | |
f4527c90 | 4091 | seq_puts(m, " vmap"); |
a10aa579 CL |
4092 | |
4093 | if (v->flags & VM_USERMAP) | |
f4527c90 | 4094 | seq_puts(m, " user"); |
a10aa579 | 4095 | |
fe9041c2 CH |
4096 | if (v->flags & VM_DMA_COHERENT) |
4097 | seq_puts(m, " dma-coherent"); | |
4098 | ||
244d63ee | 4099 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 4100 | seq_puts(m, " vpages"); |
a10aa579 | 4101 | |
a47a126a | 4102 | show_numa_info(m, v); |
a10aa579 | 4103 | seq_putc(m, '\n'); |
dd3b8353 URS |
4104 | |
4105 | /* | |
96e2db45 | 4106 | * As a final step, dump "unpurged" areas. |
dd3b8353 | 4107 | */ |
7cc7913e | 4108 | final: |
dd3b8353 URS |
4109 | if (list_is_last(&va->list, &vmap_area_list)) |
4110 | show_purge_info(m); | |
4111 | ||
a10aa579 CL |
4112 | return 0; |
4113 | } | |
4114 | ||
5f6a6a9c | 4115 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
4116 | .start = s_start, |
4117 | .next = s_next, | |
4118 | .stop = s_stop, | |
4119 | .show = s_show, | |
4120 | }; | |
5f6a6a9c | 4121 | |
5f6a6a9c AD |
4122 | static int __init proc_vmalloc_init(void) |
4123 | { | |
fddda2b7 | 4124 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 4125 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
4126 | &vmalloc_op, |
4127 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 4128 | else |
0825a6f9 | 4129 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
4130 | return 0; |
4131 | } | |
4132 | module_init(proc_vmalloc_init); | |
db3808c1 | 4133 | |
a10aa579 | 4134 | #endif |