]>
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> |
4c91c07c | 36 | #include <linux/uio.h> |
0f616be1 | 37 | #include <linux/bitops.h> |
68ad4a33 | 38 | #include <linux/rbtree_augmented.h> |
bdebd6a2 | 39 | #include <linux/overflow.h> |
c0eb315a | 40 | #include <linux/pgtable.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 | |
cf243da6 URS |
46 | #define CREATE_TRACE_POINTS |
47 | #include <trace/events/vmalloc.h> | |
48 | ||
dd56b046 | 49 | #include "internal.h" |
2a681cfa | 50 | #include "pgalloc-track.h" |
dd56b046 | 51 | |
82a70ce0 CH |
52 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP |
53 | static unsigned int __ro_after_init ioremap_max_page_shift = BITS_PER_LONG - 1; | |
54 | ||
55 | static int __init set_nohugeiomap(char *str) | |
56 | { | |
57 | ioremap_max_page_shift = PAGE_SHIFT; | |
58 | return 0; | |
59 | } | |
60 | early_param("nohugeiomap", set_nohugeiomap); | |
61 | #else /* CONFIG_HAVE_ARCH_HUGE_VMAP */ | |
62 | static const unsigned int ioremap_max_page_shift = PAGE_SHIFT; | |
63 | #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */ | |
64 | ||
121e6f32 NP |
65 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC |
66 | static bool __ro_after_init vmap_allow_huge = true; | |
67 | ||
68 | static int __init set_nohugevmalloc(char *str) | |
69 | { | |
70 | vmap_allow_huge = false; | |
71 | return 0; | |
72 | } | |
73 | early_param("nohugevmalloc", set_nohugevmalloc); | |
74 | #else /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */ | |
75 | static const bool vmap_allow_huge = false; | |
76 | #endif /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */ | |
77 | ||
186525bd IM |
78 | bool is_vmalloc_addr(const void *x) |
79 | { | |
4aff1dc4 | 80 | unsigned long addr = (unsigned long)kasan_reset_tag(x); |
186525bd IM |
81 | |
82 | return addr >= VMALLOC_START && addr < VMALLOC_END; | |
83 | } | |
84 | EXPORT_SYMBOL(is_vmalloc_addr); | |
85 | ||
32fcfd40 AV |
86 | struct vfree_deferred { |
87 | struct llist_head list; | |
88 | struct work_struct wq; | |
89 | }; | |
90 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
91 | ||
db64fe02 | 92 | /*** Page table manipulation functions ***/ |
5e9e3d77 NP |
93 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
94 | phys_addr_t phys_addr, pgprot_t prot, | |
f7ee1f13 | 95 | unsigned int max_page_shift, pgtbl_mod_mask *mask) |
5e9e3d77 NP |
96 | { |
97 | pte_t *pte; | |
98 | u64 pfn; | |
f7ee1f13 | 99 | unsigned long size = PAGE_SIZE; |
5e9e3d77 NP |
100 | |
101 | pfn = phys_addr >> PAGE_SHIFT; | |
102 | pte = pte_alloc_kernel_track(pmd, addr, mask); | |
103 | if (!pte) | |
104 | return -ENOMEM; | |
105 | do { | |
106 | BUG_ON(!pte_none(*pte)); | |
f7ee1f13 CL |
107 | |
108 | #ifdef CONFIG_HUGETLB_PAGE | |
109 | size = arch_vmap_pte_range_map_size(addr, end, pfn, max_page_shift); | |
110 | if (size != PAGE_SIZE) { | |
111 | pte_t entry = pfn_pte(pfn, prot); | |
112 | ||
f7ee1f13 CL |
113 | entry = arch_make_huge_pte(entry, ilog2(size), 0); |
114 | set_huge_pte_at(&init_mm, addr, pte, entry); | |
115 | pfn += PFN_DOWN(size); | |
116 | continue; | |
117 | } | |
118 | #endif | |
5e9e3d77 NP |
119 | set_pte_at(&init_mm, addr, pte, pfn_pte(pfn, prot)); |
120 | pfn++; | |
f7ee1f13 | 121 | } while (pte += PFN_DOWN(size), addr += size, addr != end); |
5e9e3d77 NP |
122 | *mask |= PGTBL_PTE_MODIFIED; |
123 | return 0; | |
124 | } | |
125 | ||
126 | static int vmap_try_huge_pmd(pmd_t *pmd, unsigned long addr, unsigned long end, | |
127 | phys_addr_t phys_addr, pgprot_t prot, | |
128 | unsigned int max_page_shift) | |
129 | { | |
130 | if (max_page_shift < PMD_SHIFT) | |
131 | return 0; | |
132 | ||
133 | if (!arch_vmap_pmd_supported(prot)) | |
134 | return 0; | |
135 | ||
136 | if ((end - addr) != PMD_SIZE) | |
137 | return 0; | |
138 | ||
139 | if (!IS_ALIGNED(addr, PMD_SIZE)) | |
140 | return 0; | |
141 | ||
142 | if (!IS_ALIGNED(phys_addr, PMD_SIZE)) | |
143 | return 0; | |
144 | ||
145 | if (pmd_present(*pmd) && !pmd_free_pte_page(pmd, addr)) | |
146 | return 0; | |
147 | ||
148 | return pmd_set_huge(pmd, phys_addr, prot); | |
149 | } | |
150 | ||
151 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, | |
152 | phys_addr_t phys_addr, pgprot_t prot, | |
153 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
154 | { | |
155 | pmd_t *pmd; | |
156 | unsigned long next; | |
157 | ||
158 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); | |
159 | if (!pmd) | |
160 | return -ENOMEM; | |
161 | do { | |
162 | next = pmd_addr_end(addr, end); | |
163 | ||
164 | if (vmap_try_huge_pmd(pmd, addr, next, phys_addr, prot, | |
165 | max_page_shift)) { | |
166 | *mask |= PGTBL_PMD_MODIFIED; | |
167 | continue; | |
168 | } | |
169 | ||
f7ee1f13 | 170 | if (vmap_pte_range(pmd, addr, next, phys_addr, prot, max_page_shift, mask)) |
5e9e3d77 NP |
171 | return -ENOMEM; |
172 | } while (pmd++, phys_addr += (next - addr), addr = next, addr != end); | |
173 | return 0; | |
174 | } | |
175 | ||
176 | static int vmap_try_huge_pud(pud_t *pud, unsigned long addr, unsigned long end, | |
177 | phys_addr_t phys_addr, pgprot_t prot, | |
178 | unsigned int max_page_shift) | |
179 | { | |
180 | if (max_page_shift < PUD_SHIFT) | |
181 | return 0; | |
182 | ||
183 | if (!arch_vmap_pud_supported(prot)) | |
184 | return 0; | |
185 | ||
186 | if ((end - addr) != PUD_SIZE) | |
187 | return 0; | |
188 | ||
189 | if (!IS_ALIGNED(addr, PUD_SIZE)) | |
190 | return 0; | |
191 | ||
192 | if (!IS_ALIGNED(phys_addr, PUD_SIZE)) | |
193 | return 0; | |
194 | ||
195 | if (pud_present(*pud) && !pud_free_pmd_page(pud, addr)) | |
196 | return 0; | |
197 | ||
198 | return pud_set_huge(pud, phys_addr, prot); | |
199 | } | |
200 | ||
201 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, | |
202 | phys_addr_t phys_addr, pgprot_t prot, | |
203 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
204 | { | |
205 | pud_t *pud; | |
206 | unsigned long next; | |
207 | ||
208 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); | |
209 | if (!pud) | |
210 | return -ENOMEM; | |
211 | do { | |
212 | next = pud_addr_end(addr, end); | |
213 | ||
214 | if (vmap_try_huge_pud(pud, addr, next, phys_addr, prot, | |
215 | max_page_shift)) { | |
216 | *mask |= PGTBL_PUD_MODIFIED; | |
217 | continue; | |
218 | } | |
219 | ||
220 | if (vmap_pmd_range(pud, addr, next, phys_addr, prot, | |
221 | max_page_shift, mask)) | |
222 | return -ENOMEM; | |
223 | } while (pud++, phys_addr += (next - addr), addr = next, addr != end); | |
224 | return 0; | |
225 | } | |
226 | ||
227 | static int vmap_try_huge_p4d(p4d_t *p4d, unsigned long addr, unsigned long end, | |
228 | phys_addr_t phys_addr, pgprot_t prot, | |
229 | unsigned int max_page_shift) | |
230 | { | |
231 | if (max_page_shift < P4D_SHIFT) | |
232 | return 0; | |
233 | ||
234 | if (!arch_vmap_p4d_supported(prot)) | |
235 | return 0; | |
236 | ||
237 | if ((end - addr) != P4D_SIZE) | |
238 | return 0; | |
239 | ||
240 | if (!IS_ALIGNED(addr, P4D_SIZE)) | |
241 | return 0; | |
242 | ||
243 | if (!IS_ALIGNED(phys_addr, P4D_SIZE)) | |
244 | return 0; | |
245 | ||
246 | if (p4d_present(*p4d) && !p4d_free_pud_page(p4d, addr)) | |
247 | return 0; | |
248 | ||
249 | return p4d_set_huge(p4d, phys_addr, prot); | |
250 | } | |
251 | ||
252 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, | |
253 | phys_addr_t phys_addr, pgprot_t prot, | |
254 | unsigned int max_page_shift, pgtbl_mod_mask *mask) | |
255 | { | |
256 | p4d_t *p4d; | |
257 | unsigned long next; | |
258 | ||
259 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); | |
260 | if (!p4d) | |
261 | return -ENOMEM; | |
262 | do { | |
263 | next = p4d_addr_end(addr, end); | |
264 | ||
265 | if (vmap_try_huge_p4d(p4d, addr, next, phys_addr, prot, | |
266 | max_page_shift)) { | |
267 | *mask |= PGTBL_P4D_MODIFIED; | |
268 | continue; | |
269 | } | |
270 | ||
271 | if (vmap_pud_range(p4d, addr, next, phys_addr, prot, | |
272 | max_page_shift, mask)) | |
273 | return -ENOMEM; | |
274 | } while (p4d++, phys_addr += (next - addr), addr = next, addr != end); | |
275 | return 0; | |
276 | } | |
277 | ||
5d87510d | 278 | static int vmap_range_noflush(unsigned long addr, unsigned long end, |
5e9e3d77 NP |
279 | phys_addr_t phys_addr, pgprot_t prot, |
280 | unsigned int max_page_shift) | |
281 | { | |
282 | pgd_t *pgd; | |
283 | unsigned long start; | |
284 | unsigned long next; | |
285 | int err; | |
286 | pgtbl_mod_mask mask = 0; | |
287 | ||
288 | might_sleep(); | |
289 | BUG_ON(addr >= end); | |
290 | ||
291 | start = addr; | |
292 | pgd = pgd_offset_k(addr); | |
293 | do { | |
294 | next = pgd_addr_end(addr, end); | |
295 | err = vmap_p4d_range(pgd, addr, next, phys_addr, prot, | |
296 | max_page_shift, &mask); | |
297 | if (err) | |
298 | break; | |
299 | } while (pgd++, phys_addr += (next - addr), addr = next, addr != end); | |
300 | ||
5e9e3d77 NP |
301 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
302 | arch_sync_kernel_mappings(start, end); | |
303 | ||
304 | return err; | |
305 | } | |
b221385b | 306 | |
82a70ce0 CH |
307 | int ioremap_page_range(unsigned long addr, unsigned long end, |
308 | phys_addr_t phys_addr, pgprot_t prot) | |
5d87510d NP |
309 | { |
310 | int err; | |
311 | ||
8491502f | 312 | err = vmap_range_noflush(addr, end, phys_addr, pgprot_nx(prot), |
82a70ce0 | 313 | ioremap_max_page_shift); |
5d87510d | 314 | flush_cache_vmap(addr, end); |
b073d7f8 | 315 | if (!err) |
fdea03e1 AP |
316 | err = kmsan_ioremap_page_range(addr, end, phys_addr, prot, |
317 | ioremap_max_page_shift); | |
5d87510d NP |
318 | return err; |
319 | } | |
320 | ||
2ba3e694 JR |
321 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
322 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
323 | { |
324 | pte_t *pte; | |
325 | ||
326 | pte = pte_offset_kernel(pmd, addr); | |
327 | do { | |
328 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
329 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
330 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
2ba3e694 | 331 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
332 | } |
333 | ||
2ba3e694 JR |
334 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, |
335 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
336 | { |
337 | pmd_t *pmd; | |
338 | unsigned long next; | |
2ba3e694 | 339 | int cleared; |
1da177e4 LT |
340 | |
341 | pmd = pmd_offset(pud, addr); | |
342 | do { | |
343 | next = pmd_addr_end(addr, end); | |
2ba3e694 JR |
344 | |
345 | cleared = pmd_clear_huge(pmd); | |
346 | if (cleared || pmd_bad(*pmd)) | |
347 | *mask |= PGTBL_PMD_MODIFIED; | |
348 | ||
349 | if (cleared) | |
b9820d8f | 350 | continue; |
1da177e4 LT |
351 | if (pmd_none_or_clear_bad(pmd)) |
352 | continue; | |
2ba3e694 | 353 | vunmap_pte_range(pmd, addr, next, mask); |
e47110e9 AK |
354 | |
355 | cond_resched(); | |
1da177e4 LT |
356 | } while (pmd++, addr = next, addr != end); |
357 | } | |
358 | ||
2ba3e694 JR |
359 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, |
360 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
361 | { |
362 | pud_t *pud; | |
363 | unsigned long next; | |
2ba3e694 | 364 | int cleared; |
1da177e4 | 365 | |
c2febafc | 366 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
367 | do { |
368 | next = pud_addr_end(addr, end); | |
2ba3e694 JR |
369 | |
370 | cleared = pud_clear_huge(pud); | |
371 | if (cleared || pud_bad(*pud)) | |
372 | *mask |= PGTBL_PUD_MODIFIED; | |
373 | ||
374 | if (cleared) | |
b9820d8f | 375 | continue; |
1da177e4 LT |
376 | if (pud_none_or_clear_bad(pud)) |
377 | continue; | |
2ba3e694 | 378 | vunmap_pmd_range(pud, addr, next, mask); |
1da177e4 LT |
379 | } while (pud++, addr = next, addr != end); |
380 | } | |
381 | ||
2ba3e694 JR |
382 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, |
383 | pgtbl_mod_mask *mask) | |
c2febafc KS |
384 | { |
385 | p4d_t *p4d; | |
386 | unsigned long next; | |
387 | ||
388 | p4d = p4d_offset(pgd, addr); | |
389 | do { | |
390 | next = p4d_addr_end(addr, end); | |
2ba3e694 | 391 | |
c8db8c26 L |
392 | p4d_clear_huge(p4d); |
393 | if (p4d_bad(*p4d)) | |
2ba3e694 JR |
394 | *mask |= PGTBL_P4D_MODIFIED; |
395 | ||
c2febafc KS |
396 | if (p4d_none_or_clear_bad(p4d)) |
397 | continue; | |
2ba3e694 | 398 | vunmap_pud_range(p4d, addr, next, mask); |
c2febafc KS |
399 | } while (p4d++, addr = next, addr != end); |
400 | } | |
401 | ||
4ad0ae8c NP |
402 | /* |
403 | * vunmap_range_noflush is similar to vunmap_range, but does not | |
404 | * flush caches or TLBs. | |
b521c43f | 405 | * |
4ad0ae8c NP |
406 | * The caller is responsible for calling flush_cache_vmap() before calling |
407 | * this function, and flush_tlb_kernel_range after it has returned | |
408 | * successfully (and before the addresses are expected to cause a page fault | |
409 | * or be re-mapped for something else, if TLB flushes are being delayed or | |
410 | * coalesced). | |
b521c43f | 411 | * |
4ad0ae8c | 412 | * This is an internal function only. Do not use outside mm/. |
b521c43f | 413 | */ |
b073d7f8 | 414 | void __vunmap_range_noflush(unsigned long start, unsigned long end) |
1da177e4 | 415 | { |
1da177e4 | 416 | unsigned long next; |
b521c43f | 417 | pgd_t *pgd; |
2ba3e694 JR |
418 | unsigned long addr = start; |
419 | pgtbl_mod_mask mask = 0; | |
1da177e4 LT |
420 | |
421 | BUG_ON(addr >= end); | |
422 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
423 | do { |
424 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
425 | if (pgd_bad(*pgd)) |
426 | mask |= PGTBL_PGD_MODIFIED; | |
1da177e4 LT |
427 | if (pgd_none_or_clear_bad(pgd)) |
428 | continue; | |
2ba3e694 | 429 | vunmap_p4d_range(pgd, addr, next, &mask); |
1da177e4 | 430 | } while (pgd++, addr = next, addr != end); |
2ba3e694 JR |
431 | |
432 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) | |
433 | arch_sync_kernel_mappings(start, end); | |
1da177e4 LT |
434 | } |
435 | ||
b073d7f8 AP |
436 | void vunmap_range_noflush(unsigned long start, unsigned long end) |
437 | { | |
438 | kmsan_vunmap_range_noflush(start, end); | |
439 | __vunmap_range_noflush(start, end); | |
440 | } | |
441 | ||
4ad0ae8c NP |
442 | /** |
443 | * vunmap_range - unmap kernel virtual addresses | |
444 | * @addr: start of the VM area to unmap | |
445 | * @end: end of the VM area to unmap (non-inclusive) | |
446 | * | |
447 | * Clears any present PTEs in the virtual address range, flushes TLBs and | |
448 | * caches. Any subsequent access to the address before it has been re-mapped | |
449 | * is a kernel bug. | |
450 | */ | |
451 | void vunmap_range(unsigned long addr, unsigned long end) | |
452 | { | |
453 | flush_cache_vunmap(addr, end); | |
454 | vunmap_range_noflush(addr, end); | |
455 | flush_tlb_kernel_range(addr, end); | |
456 | } | |
457 | ||
0a264884 | 458 | static int vmap_pages_pte_range(pmd_t *pmd, unsigned long addr, |
2ba3e694 JR |
459 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
460 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
461 | { |
462 | pte_t *pte; | |
463 | ||
db64fe02 NP |
464 | /* |
465 | * nr is a running index into the array which helps higher level | |
466 | * callers keep track of where we're up to. | |
467 | */ | |
468 | ||
2ba3e694 | 469 | pte = pte_alloc_kernel_track(pmd, addr, mask); |
1da177e4 LT |
470 | if (!pte) |
471 | return -ENOMEM; | |
472 | do { | |
db64fe02 NP |
473 | struct page *page = pages[*nr]; |
474 | ||
475 | if (WARN_ON(!pte_none(*pte))) | |
476 | return -EBUSY; | |
477 | if (WARN_ON(!page)) | |
1da177e4 | 478 | return -ENOMEM; |
4fcdcc12 YN |
479 | if (WARN_ON(!pfn_valid(page_to_pfn(page)))) |
480 | return -EINVAL; | |
481 | ||
1da177e4 | 482 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); |
db64fe02 | 483 | (*nr)++; |
1da177e4 | 484 | } while (pte++, addr += PAGE_SIZE, addr != end); |
2ba3e694 | 485 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
486 | return 0; |
487 | } | |
488 | ||
0a264884 | 489 | static int vmap_pages_pmd_range(pud_t *pud, unsigned long addr, |
2ba3e694 JR |
490 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
491 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
492 | { |
493 | pmd_t *pmd; | |
494 | unsigned long next; | |
495 | ||
2ba3e694 | 496 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); |
1da177e4 LT |
497 | if (!pmd) |
498 | return -ENOMEM; | |
499 | do { | |
500 | next = pmd_addr_end(addr, end); | |
0a264884 | 501 | if (vmap_pages_pte_range(pmd, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
502 | return -ENOMEM; |
503 | } while (pmd++, addr = next, addr != end); | |
504 | return 0; | |
505 | } | |
506 | ||
0a264884 | 507 | static int vmap_pages_pud_range(p4d_t *p4d, unsigned long addr, |
2ba3e694 JR |
508 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
509 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
510 | { |
511 | pud_t *pud; | |
512 | unsigned long next; | |
513 | ||
2ba3e694 | 514 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); |
1da177e4 LT |
515 | if (!pud) |
516 | return -ENOMEM; | |
517 | do { | |
518 | next = pud_addr_end(addr, end); | |
0a264884 | 519 | if (vmap_pages_pmd_range(pud, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
520 | return -ENOMEM; |
521 | } while (pud++, addr = next, addr != end); | |
522 | return 0; | |
523 | } | |
524 | ||
0a264884 | 525 | static int vmap_pages_p4d_range(pgd_t *pgd, unsigned long addr, |
2ba3e694 JR |
526 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
527 | pgtbl_mod_mask *mask) | |
c2febafc KS |
528 | { |
529 | p4d_t *p4d; | |
530 | unsigned long next; | |
531 | ||
2ba3e694 | 532 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); |
c2febafc KS |
533 | if (!p4d) |
534 | return -ENOMEM; | |
535 | do { | |
536 | next = p4d_addr_end(addr, end); | |
0a264884 | 537 | if (vmap_pages_pud_range(p4d, addr, next, prot, pages, nr, mask)) |
c2febafc KS |
538 | return -ENOMEM; |
539 | } while (p4d++, addr = next, addr != end); | |
540 | return 0; | |
541 | } | |
542 | ||
121e6f32 NP |
543 | static int vmap_small_pages_range_noflush(unsigned long addr, unsigned long end, |
544 | pgprot_t prot, struct page **pages) | |
1da177e4 | 545 | { |
2ba3e694 | 546 | unsigned long start = addr; |
b521c43f | 547 | pgd_t *pgd; |
121e6f32 | 548 | unsigned long next; |
db64fe02 NP |
549 | int err = 0; |
550 | int nr = 0; | |
2ba3e694 | 551 | pgtbl_mod_mask mask = 0; |
1da177e4 LT |
552 | |
553 | BUG_ON(addr >= end); | |
554 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
555 | do { |
556 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
557 | if (pgd_bad(*pgd)) |
558 | mask |= PGTBL_PGD_MODIFIED; | |
0a264884 | 559 | err = vmap_pages_p4d_range(pgd, addr, next, prot, pages, &nr, &mask); |
1da177e4 | 560 | if (err) |
bf88c8c8 | 561 | return err; |
1da177e4 | 562 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 563 | |
2ba3e694 JR |
564 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
565 | arch_sync_kernel_mappings(start, end); | |
566 | ||
60bb4465 | 567 | return 0; |
1da177e4 LT |
568 | } |
569 | ||
b67177ec NP |
570 | /* |
571 | * vmap_pages_range_noflush is similar to vmap_pages_range, but does not | |
572 | * flush caches. | |
573 | * | |
574 | * The caller is responsible for calling flush_cache_vmap() after this | |
575 | * function returns successfully and before the addresses are accessed. | |
576 | * | |
577 | * This is an internal function only. Do not use outside mm/. | |
578 | */ | |
b073d7f8 | 579 | int __vmap_pages_range_noflush(unsigned long addr, unsigned long end, |
121e6f32 NP |
580 | pgprot_t prot, struct page **pages, unsigned int page_shift) |
581 | { | |
582 | unsigned int i, nr = (end - addr) >> PAGE_SHIFT; | |
583 | ||
584 | WARN_ON(page_shift < PAGE_SHIFT); | |
585 | ||
586 | if (!IS_ENABLED(CONFIG_HAVE_ARCH_HUGE_VMALLOC) || | |
587 | page_shift == PAGE_SHIFT) | |
588 | return vmap_small_pages_range_noflush(addr, end, prot, pages); | |
589 | ||
590 | for (i = 0; i < nr; i += 1U << (page_shift - PAGE_SHIFT)) { | |
591 | int err; | |
592 | ||
593 | err = vmap_range_noflush(addr, addr + (1UL << page_shift), | |
08262ac5 | 594 | page_to_phys(pages[i]), prot, |
121e6f32 NP |
595 | page_shift); |
596 | if (err) | |
597 | return err; | |
598 | ||
599 | addr += 1UL << page_shift; | |
600 | } | |
601 | ||
602 | return 0; | |
603 | } | |
b073d7f8 AP |
604 | |
605 | int vmap_pages_range_noflush(unsigned long addr, unsigned long end, | |
606 | pgprot_t prot, struct page **pages, unsigned int page_shift) | |
607 | { | |
47ebd031 AP |
608 | int ret = kmsan_vmap_pages_range_noflush(addr, end, prot, pages, |
609 | page_shift); | |
610 | ||
611 | if (ret) | |
612 | return ret; | |
b073d7f8 AP |
613 | return __vmap_pages_range_noflush(addr, end, prot, pages, page_shift); |
614 | } | |
121e6f32 | 615 | |
121e6f32 | 616 | /** |
b67177ec | 617 | * vmap_pages_range - map pages to a kernel virtual address |
121e6f32 | 618 | * @addr: start of the VM area to map |
b67177ec | 619 | * @end: end of the VM area to map (non-inclusive) |
121e6f32 | 620 | * @prot: page protection flags to use |
b67177ec NP |
621 | * @pages: pages to map (always PAGE_SIZE pages) |
622 | * @page_shift: maximum shift that the pages may be mapped with, @pages must | |
623 | * be aligned and contiguous up to at least this shift. | |
121e6f32 NP |
624 | * |
625 | * RETURNS: | |
626 | * 0 on success, -errno on failure. | |
627 | */ | |
b67177ec NP |
628 | static int vmap_pages_range(unsigned long addr, unsigned long end, |
629 | pgprot_t prot, struct page **pages, unsigned int page_shift) | |
8fc48985 | 630 | { |
b67177ec | 631 | int err; |
8fc48985 | 632 | |
b67177ec NP |
633 | err = vmap_pages_range_noflush(addr, end, prot, pages, page_shift); |
634 | flush_cache_vmap(addr, end); | |
635 | return err; | |
8fc48985 TH |
636 | } |
637 | ||
81ac3ad9 | 638 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
639 | { |
640 | /* | |
ab4f2ee1 | 641 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
642 | * and fall back on vmalloc() if that fails. Others |
643 | * just put it in the vmalloc space. | |
644 | */ | |
645 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
4aff1dc4 | 646 | unsigned long addr = (unsigned long)kasan_reset_tag(x); |
73bdf0a6 LT |
647 | if (addr >= MODULES_VADDR && addr < MODULES_END) |
648 | return 1; | |
649 | #endif | |
650 | return is_vmalloc_addr(x); | |
651 | } | |
01858469 | 652 | EXPORT_SYMBOL_GPL(is_vmalloc_or_module_addr); |
73bdf0a6 | 653 | |
48667e7a | 654 | /* |
c0eb315a NP |
655 | * Walk a vmap address to the struct page it maps. Huge vmap mappings will |
656 | * return the tail page that corresponds to the base page address, which | |
657 | * matches small vmap mappings. | |
48667e7a | 658 | */ |
add688fb | 659 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
660 | { |
661 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 662 | struct page *page = NULL; |
48667e7a | 663 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
664 | p4d_t *p4d; |
665 | pud_t *pud; | |
666 | pmd_t *pmd; | |
667 | pte_t *ptep, pte; | |
48667e7a | 668 | |
7aa413de IM |
669 | /* |
670 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
671 | * architectures that do not vmalloc module space | |
672 | */ | |
73bdf0a6 | 673 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 674 | |
c2febafc KS |
675 | if (pgd_none(*pgd)) |
676 | return NULL; | |
c0eb315a NP |
677 | if (WARN_ON_ONCE(pgd_leaf(*pgd))) |
678 | return NULL; /* XXX: no allowance for huge pgd */ | |
679 | if (WARN_ON_ONCE(pgd_bad(*pgd))) | |
680 | return NULL; | |
681 | ||
c2febafc KS |
682 | p4d = p4d_offset(pgd, addr); |
683 | if (p4d_none(*p4d)) | |
684 | return NULL; | |
c0eb315a NP |
685 | if (p4d_leaf(*p4d)) |
686 | return p4d_page(*p4d) + ((addr & ~P4D_MASK) >> PAGE_SHIFT); | |
687 | if (WARN_ON_ONCE(p4d_bad(*p4d))) | |
688 | return NULL; | |
029c54b0 | 689 | |
c0eb315a NP |
690 | pud = pud_offset(p4d, addr); |
691 | if (pud_none(*pud)) | |
692 | return NULL; | |
693 | if (pud_leaf(*pud)) | |
694 | return pud_page(*pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
695 | if (WARN_ON_ONCE(pud_bad(*pud))) | |
c2febafc | 696 | return NULL; |
c0eb315a | 697 | |
c2febafc | 698 | pmd = pmd_offset(pud, addr); |
c0eb315a NP |
699 | if (pmd_none(*pmd)) |
700 | return NULL; | |
701 | if (pmd_leaf(*pmd)) | |
702 | return pmd_page(*pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
703 | if (WARN_ON_ONCE(pmd_bad(*pmd))) | |
c2febafc KS |
704 | return NULL; |
705 | ||
706 | ptep = pte_offset_map(pmd, addr); | |
707 | pte = *ptep; | |
708 | if (pte_present(pte)) | |
709 | page = pte_page(pte); | |
710 | pte_unmap(ptep); | |
c0eb315a | 711 | |
add688fb | 712 | return page; |
48667e7a | 713 | } |
add688fb | 714 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
715 | |
716 | /* | |
add688fb | 717 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 718 | */ |
add688fb | 719 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 720 | { |
add688fb | 721 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 722 | } |
add688fb | 723 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 724 | |
db64fe02 NP |
725 | |
726 | /*** Global kva allocator ***/ | |
727 | ||
bb850f4d | 728 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 729 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 730 | |
db64fe02 | 731 | |
db64fe02 | 732 | static DEFINE_SPINLOCK(vmap_area_lock); |
e36176be | 733 | static DEFINE_SPINLOCK(free_vmap_area_lock); |
f1c4069e JK |
734 | /* Export for kexec only */ |
735 | LIST_HEAD(vmap_area_list); | |
89699605 | 736 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 737 | static bool vmap_initialized __read_mostly; |
89699605 | 738 | |
96e2db45 URS |
739 | static struct rb_root purge_vmap_area_root = RB_ROOT; |
740 | static LIST_HEAD(purge_vmap_area_list); | |
741 | static DEFINE_SPINLOCK(purge_vmap_area_lock); | |
742 | ||
68ad4a33 URS |
743 | /* |
744 | * This kmem_cache is used for vmap_area objects. Instead of | |
745 | * allocating from slab we reuse an object from this cache to | |
746 | * make things faster. Especially in "no edge" splitting of | |
747 | * free block. | |
748 | */ | |
749 | static struct kmem_cache *vmap_area_cachep; | |
750 | ||
751 | /* | |
752 | * This linked list is used in pair with free_vmap_area_root. | |
753 | * It gives O(1) access to prev/next to perform fast coalescing. | |
754 | */ | |
755 | static LIST_HEAD(free_vmap_area_list); | |
756 | ||
757 | /* | |
758 | * This augment red-black tree represents the free vmap space. | |
759 | * All vmap_area objects in this tree are sorted by va->va_start | |
760 | * address. It is used for allocation and merging when a vmap | |
761 | * object is released. | |
762 | * | |
763 | * Each vmap_area node contains a maximum available free block | |
764 | * of its sub-tree, right or left. Therefore it is possible to | |
765 | * find a lowest match of free area. | |
766 | */ | |
767 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
768 | ||
82dd23e8 URS |
769 | /* |
770 | * Preload a CPU with one object for "no edge" split case. The | |
771 | * aim is to get rid of allocations from the atomic context, thus | |
772 | * to use more permissive allocation masks. | |
773 | */ | |
774 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
775 | ||
68ad4a33 URS |
776 | static __always_inline unsigned long |
777 | va_size(struct vmap_area *va) | |
778 | { | |
779 | return (va->va_end - va->va_start); | |
780 | } | |
781 | ||
782 | static __always_inline unsigned long | |
783 | get_subtree_max_size(struct rb_node *node) | |
784 | { | |
785 | struct vmap_area *va; | |
786 | ||
787 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
788 | return va ? va->subtree_max_size : 0; | |
789 | } | |
89699605 | 790 | |
315cc066 ML |
791 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
792 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 | 793 | |
77e50af0 | 794 | static void reclaim_and_purge_vmap_areas(void); |
68ad4a33 | 795 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); |
690467c8 URS |
796 | static void drain_vmap_area_work(struct work_struct *work); |
797 | static DECLARE_WORK(drain_vmap_work, drain_vmap_area_work); | |
db64fe02 | 798 | |
97105f0a RG |
799 | static atomic_long_t nr_vmalloc_pages; |
800 | ||
801 | unsigned long vmalloc_nr_pages(void) | |
802 | { | |
803 | return atomic_long_read(&nr_vmalloc_pages); | |
804 | } | |
805 | ||
153090f2 | 806 | /* Look up the first VA which satisfies addr < va_end, NULL if none. */ |
f181234a CW |
807 | static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr) |
808 | { | |
809 | struct vmap_area *va = NULL; | |
810 | struct rb_node *n = vmap_area_root.rb_node; | |
811 | ||
4aff1dc4 AK |
812 | addr = (unsigned long)kasan_reset_tag((void *)addr); |
813 | ||
f181234a CW |
814 | while (n) { |
815 | struct vmap_area *tmp; | |
816 | ||
817 | tmp = rb_entry(n, struct vmap_area, rb_node); | |
818 | if (tmp->va_end > addr) { | |
819 | va = tmp; | |
820 | if (tmp->va_start <= addr) | |
821 | break; | |
822 | ||
823 | n = n->rb_left; | |
824 | } else | |
825 | n = n->rb_right; | |
826 | } | |
827 | ||
828 | return va; | |
829 | } | |
830 | ||
899c6efe | 831 | static struct vmap_area *__find_vmap_area(unsigned long addr, struct rb_root *root) |
1da177e4 | 832 | { |
899c6efe | 833 | struct rb_node *n = root->rb_node; |
db64fe02 | 834 | |
4aff1dc4 AK |
835 | addr = (unsigned long)kasan_reset_tag((void *)addr); |
836 | ||
db64fe02 NP |
837 | while (n) { |
838 | struct vmap_area *va; | |
839 | ||
840 | va = rb_entry(n, struct vmap_area, rb_node); | |
841 | if (addr < va->va_start) | |
842 | n = n->rb_left; | |
cef2ac3f | 843 | else if (addr >= va->va_end) |
db64fe02 NP |
844 | n = n->rb_right; |
845 | else | |
846 | return va; | |
847 | } | |
848 | ||
849 | return NULL; | |
850 | } | |
851 | ||
68ad4a33 URS |
852 | /* |
853 | * This function returns back addresses of parent node | |
854 | * and its left or right link for further processing. | |
9c801f61 URS |
855 | * |
856 | * Otherwise NULL is returned. In that case all further | |
857 | * steps regarding inserting of conflicting overlap range | |
858 | * have to be declined and actually considered as a bug. | |
68ad4a33 URS |
859 | */ |
860 | static __always_inline struct rb_node ** | |
861 | find_va_links(struct vmap_area *va, | |
862 | struct rb_root *root, struct rb_node *from, | |
863 | struct rb_node **parent) | |
864 | { | |
865 | struct vmap_area *tmp_va; | |
866 | struct rb_node **link; | |
867 | ||
868 | if (root) { | |
869 | link = &root->rb_node; | |
870 | if (unlikely(!*link)) { | |
871 | *parent = NULL; | |
872 | return link; | |
873 | } | |
874 | } else { | |
875 | link = &from; | |
876 | } | |
db64fe02 | 877 | |
68ad4a33 URS |
878 | /* |
879 | * Go to the bottom of the tree. When we hit the last point | |
880 | * we end up with parent rb_node and correct direction, i name | |
881 | * it link, where the new va->rb_node will be attached to. | |
882 | */ | |
883 | do { | |
884 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 885 | |
68ad4a33 URS |
886 | /* |
887 | * During the traversal we also do some sanity check. | |
888 | * Trigger the BUG() if there are sides(left/right) | |
889 | * or full overlaps. | |
890 | */ | |
753df96b | 891 | if (va->va_end <= tmp_va->va_start) |
68ad4a33 | 892 | link = &(*link)->rb_left; |
753df96b | 893 | else if (va->va_start >= tmp_va->va_end) |
68ad4a33 | 894 | link = &(*link)->rb_right; |
9c801f61 URS |
895 | else { |
896 | WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n", | |
897 | va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end); | |
898 | ||
899 | return NULL; | |
900 | } | |
68ad4a33 URS |
901 | } while (*link); |
902 | ||
903 | *parent = &tmp_va->rb_node; | |
904 | return link; | |
905 | } | |
906 | ||
907 | static __always_inline struct list_head * | |
908 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
909 | { | |
910 | struct list_head *list; | |
911 | ||
912 | if (unlikely(!parent)) | |
913 | /* | |
914 | * The red-black tree where we try to find VA neighbors | |
915 | * before merging or inserting is empty, i.e. it means | |
916 | * there is no free vmap space. Normally it does not | |
917 | * happen but we handle this case anyway. | |
918 | */ | |
919 | return NULL; | |
920 | ||
921 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
922 | return (&parent->rb_right == link ? list->next : list); | |
923 | } | |
924 | ||
925 | static __always_inline void | |
8eb510db URS |
926 | __link_va(struct vmap_area *va, struct rb_root *root, |
927 | struct rb_node *parent, struct rb_node **link, | |
928 | struct list_head *head, bool augment) | |
68ad4a33 URS |
929 | { |
930 | /* | |
931 | * VA is still not in the list, but we can | |
932 | * identify its future previous list_head node. | |
933 | */ | |
934 | if (likely(parent)) { | |
935 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
936 | if (&parent->rb_right != link) | |
937 | head = head->prev; | |
db64fe02 NP |
938 | } |
939 | ||
68ad4a33 URS |
940 | /* Insert to the rb-tree */ |
941 | rb_link_node(&va->rb_node, parent, link); | |
8eb510db | 942 | if (augment) { |
68ad4a33 URS |
943 | /* |
944 | * Some explanation here. Just perform simple insertion | |
945 | * to the tree. We do not set va->subtree_max_size to | |
946 | * its current size before calling rb_insert_augmented(). | |
153090f2 | 947 | * It is because we populate the tree from the bottom |
68ad4a33 URS |
948 | * to parent levels when the node _is_ in the tree. |
949 | * | |
950 | * Therefore we set subtree_max_size to zero after insertion, | |
951 | * to let __augment_tree_propagate_from() puts everything to | |
952 | * the correct order later on. | |
953 | */ | |
954 | rb_insert_augmented(&va->rb_node, | |
955 | root, &free_vmap_area_rb_augment_cb); | |
956 | va->subtree_max_size = 0; | |
957 | } else { | |
958 | rb_insert_color(&va->rb_node, root); | |
959 | } | |
db64fe02 | 960 | |
68ad4a33 URS |
961 | /* Address-sort this list */ |
962 | list_add(&va->list, head); | |
db64fe02 NP |
963 | } |
964 | ||
68ad4a33 | 965 | static __always_inline void |
8eb510db URS |
966 | link_va(struct vmap_area *va, struct rb_root *root, |
967 | struct rb_node *parent, struct rb_node **link, | |
968 | struct list_head *head) | |
969 | { | |
970 | __link_va(va, root, parent, link, head, false); | |
971 | } | |
972 | ||
973 | static __always_inline void | |
974 | link_va_augment(struct vmap_area *va, struct rb_root *root, | |
975 | struct rb_node *parent, struct rb_node **link, | |
976 | struct list_head *head) | |
977 | { | |
978 | __link_va(va, root, parent, link, head, true); | |
979 | } | |
980 | ||
981 | static __always_inline void | |
982 | __unlink_va(struct vmap_area *va, struct rb_root *root, bool augment) | |
68ad4a33 | 983 | { |
460e42d1 URS |
984 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
985 | return; | |
db64fe02 | 986 | |
8eb510db | 987 | if (augment) |
460e42d1 URS |
988 | rb_erase_augmented(&va->rb_node, |
989 | root, &free_vmap_area_rb_augment_cb); | |
990 | else | |
991 | rb_erase(&va->rb_node, root); | |
992 | ||
5d7a7c54 | 993 | list_del_init(&va->list); |
460e42d1 | 994 | RB_CLEAR_NODE(&va->rb_node); |
68ad4a33 URS |
995 | } |
996 | ||
8eb510db URS |
997 | static __always_inline void |
998 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
999 | { | |
1000 | __unlink_va(va, root, false); | |
1001 | } | |
1002 | ||
1003 | static __always_inline void | |
1004 | unlink_va_augment(struct vmap_area *va, struct rb_root *root) | |
1005 | { | |
1006 | __unlink_va(va, root, true); | |
1007 | } | |
1008 | ||
bb850f4d | 1009 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
c3385e84 JC |
1010 | /* |
1011 | * Gets called when remove the node and rotate. | |
1012 | */ | |
1013 | static __always_inline unsigned long | |
1014 | compute_subtree_max_size(struct vmap_area *va) | |
1015 | { | |
1016 | return max3(va_size(va), | |
1017 | get_subtree_max_size(va->rb_node.rb_left), | |
1018 | get_subtree_max_size(va->rb_node.rb_right)); | |
1019 | } | |
1020 | ||
bb850f4d | 1021 | static void |
da27c9ed | 1022 | augment_tree_propagate_check(void) |
bb850f4d URS |
1023 | { |
1024 | struct vmap_area *va; | |
da27c9ed | 1025 | unsigned long computed_size; |
bb850f4d | 1026 | |
da27c9ed URS |
1027 | list_for_each_entry(va, &free_vmap_area_list, list) { |
1028 | computed_size = compute_subtree_max_size(va); | |
1029 | if (computed_size != va->subtree_max_size) | |
1030 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
1031 | va_size(va), va->subtree_max_size); | |
bb850f4d | 1032 | } |
bb850f4d URS |
1033 | } |
1034 | #endif | |
1035 | ||
68ad4a33 URS |
1036 | /* |
1037 | * This function populates subtree_max_size from bottom to upper | |
1038 | * levels starting from VA point. The propagation must be done | |
1039 | * when VA size is modified by changing its va_start/va_end. Or | |
1040 | * in case of newly inserting of VA to the tree. | |
1041 | * | |
1042 | * It means that __augment_tree_propagate_from() must be called: | |
1043 | * - After VA has been inserted to the tree(free path); | |
1044 | * - After VA has been shrunk(allocation path); | |
1045 | * - After VA has been increased(merging path). | |
1046 | * | |
1047 | * Please note that, it does not mean that upper parent nodes | |
1048 | * and their subtree_max_size are recalculated all the time up | |
1049 | * to the root node. | |
1050 | * | |
1051 | * 4--8 | |
1052 | * /\ | |
1053 | * / \ | |
1054 | * / \ | |
1055 | * 2--2 8--8 | |
1056 | * | |
1057 | * For example if we modify the node 4, shrinking it to 2, then | |
1058 | * no any modification is required. If we shrink the node 2 to 1 | |
1059 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
1060 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
1061 | * node becomes 4--6. | |
1062 | */ | |
1063 | static __always_inline void | |
1064 | augment_tree_propagate_from(struct vmap_area *va) | |
1065 | { | |
15ae144f URS |
1066 | /* |
1067 | * Populate the tree from bottom towards the root until | |
1068 | * the calculated maximum available size of checked node | |
1069 | * is equal to its current one. | |
1070 | */ | |
1071 | free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL); | |
bb850f4d URS |
1072 | |
1073 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
da27c9ed | 1074 | augment_tree_propagate_check(); |
bb850f4d | 1075 | #endif |
68ad4a33 URS |
1076 | } |
1077 | ||
1078 | static void | |
1079 | insert_vmap_area(struct vmap_area *va, | |
1080 | struct rb_root *root, struct list_head *head) | |
1081 | { | |
1082 | struct rb_node **link; | |
1083 | struct rb_node *parent; | |
1084 | ||
1085 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
1086 | if (link) |
1087 | link_va(va, root, parent, link, head); | |
68ad4a33 URS |
1088 | } |
1089 | ||
1090 | static void | |
1091 | insert_vmap_area_augment(struct vmap_area *va, | |
1092 | struct rb_node *from, struct rb_root *root, | |
1093 | struct list_head *head) | |
1094 | { | |
1095 | struct rb_node **link; | |
1096 | struct rb_node *parent; | |
1097 | ||
1098 | if (from) | |
1099 | link = find_va_links(va, NULL, from, &parent); | |
1100 | else | |
1101 | link = find_va_links(va, root, NULL, &parent); | |
1102 | ||
9c801f61 | 1103 | if (link) { |
8eb510db | 1104 | link_va_augment(va, root, parent, link, head); |
9c801f61 URS |
1105 | augment_tree_propagate_from(va); |
1106 | } | |
68ad4a33 URS |
1107 | } |
1108 | ||
1109 | /* | |
1110 | * Merge de-allocated chunk of VA memory with previous | |
1111 | * and next free blocks. If coalesce is not done a new | |
1112 | * free area is inserted. If VA has been merged, it is | |
1113 | * freed. | |
9c801f61 URS |
1114 | * |
1115 | * Please note, it can return NULL in case of overlap | |
1116 | * ranges, followed by WARN() report. Despite it is a | |
1117 | * buggy behaviour, a system can be alive and keep | |
1118 | * ongoing. | |
68ad4a33 | 1119 | */ |
3c5c3cfb | 1120 | static __always_inline struct vmap_area * |
8eb510db URS |
1121 | __merge_or_add_vmap_area(struct vmap_area *va, |
1122 | struct rb_root *root, struct list_head *head, bool augment) | |
68ad4a33 URS |
1123 | { |
1124 | struct vmap_area *sibling; | |
1125 | struct list_head *next; | |
1126 | struct rb_node **link; | |
1127 | struct rb_node *parent; | |
1128 | bool merged = false; | |
1129 | ||
1130 | /* | |
1131 | * Find a place in the tree where VA potentially will be | |
1132 | * inserted, unless it is merged with its sibling/siblings. | |
1133 | */ | |
1134 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
1135 | if (!link) |
1136 | return NULL; | |
68ad4a33 URS |
1137 | |
1138 | /* | |
1139 | * Get next node of VA to check if merging can be done. | |
1140 | */ | |
1141 | next = get_va_next_sibling(parent, link); | |
1142 | if (unlikely(next == NULL)) | |
1143 | goto insert; | |
1144 | ||
1145 | /* | |
1146 | * start end | |
1147 | * | | | |
1148 | * |<------VA------>|<-----Next----->| | |
1149 | * | | | |
1150 | * start end | |
1151 | */ | |
1152 | if (next != head) { | |
1153 | sibling = list_entry(next, struct vmap_area, list); | |
1154 | if (sibling->va_start == va->va_end) { | |
1155 | sibling->va_start = va->va_start; | |
1156 | ||
68ad4a33 URS |
1157 | /* Free vmap_area object. */ |
1158 | kmem_cache_free(vmap_area_cachep, va); | |
1159 | ||
1160 | /* Point to the new merged area. */ | |
1161 | va = sibling; | |
1162 | merged = true; | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * start end | |
1168 | * | | | |
1169 | * |<-----Prev----->|<------VA------>| | |
1170 | * | | | |
1171 | * start end | |
1172 | */ | |
1173 | if (next->prev != head) { | |
1174 | sibling = list_entry(next->prev, struct vmap_area, list); | |
1175 | if (sibling->va_end == va->va_start) { | |
5dd78640 URS |
1176 | /* |
1177 | * If both neighbors are coalesced, it is important | |
1178 | * to unlink the "next" node first, followed by merging | |
1179 | * with "previous" one. Otherwise the tree might not be | |
1180 | * fully populated if a sibling's augmented value is | |
1181 | * "normalized" because of rotation operations. | |
1182 | */ | |
54f63d9d | 1183 | if (merged) |
8eb510db | 1184 | __unlink_va(va, root, augment); |
68ad4a33 | 1185 | |
5dd78640 URS |
1186 | sibling->va_end = va->va_end; |
1187 | ||
68ad4a33 URS |
1188 | /* Free vmap_area object. */ |
1189 | kmem_cache_free(vmap_area_cachep, va); | |
3c5c3cfb DA |
1190 | |
1191 | /* Point to the new merged area. */ | |
1192 | va = sibling; | |
1193 | merged = true; | |
68ad4a33 URS |
1194 | } |
1195 | } | |
1196 | ||
1197 | insert: | |
5dd78640 | 1198 | if (!merged) |
8eb510db | 1199 | __link_va(va, root, parent, link, head, augment); |
3c5c3cfb | 1200 | |
96e2db45 URS |
1201 | return va; |
1202 | } | |
1203 | ||
8eb510db URS |
1204 | static __always_inline struct vmap_area * |
1205 | merge_or_add_vmap_area(struct vmap_area *va, | |
1206 | struct rb_root *root, struct list_head *head) | |
1207 | { | |
1208 | return __merge_or_add_vmap_area(va, root, head, false); | |
1209 | } | |
1210 | ||
96e2db45 URS |
1211 | static __always_inline struct vmap_area * |
1212 | merge_or_add_vmap_area_augment(struct vmap_area *va, | |
1213 | struct rb_root *root, struct list_head *head) | |
1214 | { | |
8eb510db | 1215 | va = __merge_or_add_vmap_area(va, root, head, true); |
96e2db45 URS |
1216 | if (va) |
1217 | augment_tree_propagate_from(va); | |
1218 | ||
3c5c3cfb | 1219 | return va; |
68ad4a33 URS |
1220 | } |
1221 | ||
1222 | static __always_inline bool | |
1223 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
1224 | unsigned long align, unsigned long vstart) | |
1225 | { | |
1226 | unsigned long nva_start_addr; | |
1227 | ||
1228 | if (va->va_start > vstart) | |
1229 | nva_start_addr = ALIGN(va->va_start, align); | |
1230 | else | |
1231 | nva_start_addr = ALIGN(vstart, align); | |
1232 | ||
1233 | /* Can be overflowed due to big size or alignment. */ | |
1234 | if (nva_start_addr + size < nva_start_addr || | |
1235 | nva_start_addr < vstart) | |
1236 | return false; | |
1237 | ||
1238 | return (nva_start_addr + size <= va->va_end); | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Find the first free block(lowest start address) in the tree, | |
1243 | * that will accomplish the request corresponding to passing | |
9333fe98 UR |
1244 | * parameters. Please note, with an alignment bigger than PAGE_SIZE, |
1245 | * a search length is adjusted to account for worst case alignment | |
1246 | * overhead. | |
68ad4a33 URS |
1247 | */ |
1248 | static __always_inline struct vmap_area * | |
f9863be4 URS |
1249 | find_vmap_lowest_match(struct rb_root *root, unsigned long size, |
1250 | unsigned long align, unsigned long vstart, bool adjust_search_size) | |
68ad4a33 URS |
1251 | { |
1252 | struct vmap_area *va; | |
1253 | struct rb_node *node; | |
9333fe98 | 1254 | unsigned long length; |
68ad4a33 URS |
1255 | |
1256 | /* Start from the root. */ | |
f9863be4 | 1257 | node = root->rb_node; |
68ad4a33 | 1258 | |
9333fe98 UR |
1259 | /* Adjust the search size for alignment overhead. */ |
1260 | length = adjust_search_size ? size + align - 1 : size; | |
1261 | ||
68ad4a33 URS |
1262 | while (node) { |
1263 | va = rb_entry(node, struct vmap_area, rb_node); | |
1264 | ||
9333fe98 | 1265 | if (get_subtree_max_size(node->rb_left) >= length && |
68ad4a33 URS |
1266 | vstart < va->va_start) { |
1267 | node = node->rb_left; | |
1268 | } else { | |
1269 | if (is_within_this_va(va, size, align, vstart)) | |
1270 | return va; | |
1271 | ||
1272 | /* | |
1273 | * Does not make sense to go deeper towards the right | |
1274 | * sub-tree if it does not have a free block that is | |
9333fe98 | 1275 | * equal or bigger to the requested search length. |
68ad4a33 | 1276 | */ |
9333fe98 | 1277 | if (get_subtree_max_size(node->rb_right) >= length) { |
68ad4a33 URS |
1278 | node = node->rb_right; |
1279 | continue; | |
1280 | } | |
1281 | ||
1282 | /* | |
3806b041 | 1283 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 | 1284 | * that will satisfy the search criteria. It can happen |
9f531973 URS |
1285 | * due to "vstart" restriction or an alignment overhead |
1286 | * that is bigger then PAGE_SIZE. | |
68ad4a33 URS |
1287 | */ |
1288 | while ((node = rb_parent(node))) { | |
1289 | va = rb_entry(node, struct vmap_area, rb_node); | |
1290 | if (is_within_this_va(va, size, align, vstart)) | |
1291 | return va; | |
1292 | ||
9333fe98 | 1293 | if (get_subtree_max_size(node->rb_right) >= length && |
68ad4a33 | 1294 | vstart <= va->va_start) { |
9f531973 URS |
1295 | /* |
1296 | * Shift the vstart forward. Please note, we update it with | |
1297 | * parent's start address adding "1" because we do not want | |
1298 | * to enter same sub-tree after it has already been checked | |
1299 | * and no suitable free block found there. | |
1300 | */ | |
1301 | vstart = va->va_start + 1; | |
68ad4a33 URS |
1302 | node = node->rb_right; |
1303 | break; | |
1304 | } | |
1305 | } | |
1306 | } | |
1307 | } | |
1308 | ||
1309 | return NULL; | |
1310 | } | |
1311 | ||
a6cf4e0f URS |
1312 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1313 | #include <linux/random.h> | |
1314 | ||
1315 | static struct vmap_area * | |
bd1264c3 | 1316 | find_vmap_lowest_linear_match(struct list_head *head, unsigned long size, |
a6cf4e0f URS |
1317 | unsigned long align, unsigned long vstart) |
1318 | { | |
1319 | struct vmap_area *va; | |
1320 | ||
bd1264c3 | 1321 | list_for_each_entry(va, head, list) { |
a6cf4e0f URS |
1322 | if (!is_within_this_va(va, size, align, vstart)) |
1323 | continue; | |
1324 | ||
1325 | return va; | |
1326 | } | |
1327 | ||
1328 | return NULL; | |
1329 | } | |
1330 | ||
1331 | static void | |
bd1264c3 SL |
1332 | find_vmap_lowest_match_check(struct rb_root *root, struct list_head *head, |
1333 | unsigned long size, unsigned long align) | |
a6cf4e0f URS |
1334 | { |
1335 | struct vmap_area *va_1, *va_2; | |
1336 | unsigned long vstart; | |
1337 | unsigned int rnd; | |
1338 | ||
1339 | get_random_bytes(&rnd, sizeof(rnd)); | |
1340 | vstart = VMALLOC_START + rnd; | |
1341 | ||
bd1264c3 SL |
1342 | va_1 = find_vmap_lowest_match(root, size, align, vstart, false); |
1343 | va_2 = find_vmap_lowest_linear_match(head, size, align, vstart); | |
a6cf4e0f URS |
1344 | |
1345 | if (va_1 != va_2) | |
1346 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
1347 | va_1, va_2, vstart); | |
1348 | } | |
1349 | #endif | |
1350 | ||
68ad4a33 URS |
1351 | enum fit_type { |
1352 | NOTHING_FIT = 0, | |
1353 | FL_FIT_TYPE = 1, /* full fit */ | |
1354 | LE_FIT_TYPE = 2, /* left edge fit */ | |
1355 | RE_FIT_TYPE = 3, /* right edge fit */ | |
1356 | NE_FIT_TYPE = 4 /* no edge fit */ | |
1357 | }; | |
1358 | ||
1359 | static __always_inline enum fit_type | |
1360 | classify_va_fit_type(struct vmap_area *va, | |
1361 | unsigned long nva_start_addr, unsigned long size) | |
1362 | { | |
1363 | enum fit_type type; | |
1364 | ||
1365 | /* Check if it is within VA. */ | |
1366 | if (nva_start_addr < va->va_start || | |
1367 | nva_start_addr + size > va->va_end) | |
1368 | return NOTHING_FIT; | |
1369 | ||
1370 | /* Now classify. */ | |
1371 | if (va->va_start == nva_start_addr) { | |
1372 | if (va->va_end == nva_start_addr + size) | |
1373 | type = FL_FIT_TYPE; | |
1374 | else | |
1375 | type = LE_FIT_TYPE; | |
1376 | } else if (va->va_end == nva_start_addr + size) { | |
1377 | type = RE_FIT_TYPE; | |
1378 | } else { | |
1379 | type = NE_FIT_TYPE; | |
1380 | } | |
1381 | ||
1382 | return type; | |
1383 | } | |
1384 | ||
1385 | static __always_inline int | |
f9863be4 URS |
1386 | adjust_va_to_fit_type(struct rb_root *root, struct list_head *head, |
1387 | struct vmap_area *va, unsigned long nva_start_addr, | |
1388 | unsigned long size) | |
68ad4a33 | 1389 | { |
2c929233 | 1390 | struct vmap_area *lva = NULL; |
1b23ff80 | 1391 | enum fit_type type = classify_va_fit_type(va, nva_start_addr, size); |
68ad4a33 URS |
1392 | |
1393 | if (type == FL_FIT_TYPE) { | |
1394 | /* | |
1395 | * No need to split VA, it fully fits. | |
1396 | * | |
1397 | * | | | |
1398 | * V NVA V | |
1399 | * |---------------| | |
1400 | */ | |
f9863be4 | 1401 | unlink_va_augment(va, root); |
68ad4a33 URS |
1402 | kmem_cache_free(vmap_area_cachep, va); |
1403 | } else if (type == LE_FIT_TYPE) { | |
1404 | /* | |
1405 | * Split left edge of fit VA. | |
1406 | * | |
1407 | * | | | |
1408 | * V NVA V R | |
1409 | * |-------|-------| | |
1410 | */ | |
1411 | va->va_start += size; | |
1412 | } else if (type == RE_FIT_TYPE) { | |
1413 | /* | |
1414 | * Split right edge of fit VA. | |
1415 | * | |
1416 | * | | | |
1417 | * L V NVA V | |
1418 | * |-------|-------| | |
1419 | */ | |
1420 | va->va_end = nva_start_addr; | |
1421 | } else if (type == NE_FIT_TYPE) { | |
1422 | /* | |
1423 | * Split no edge of fit VA. | |
1424 | * | |
1425 | * | | | |
1426 | * L V NVA V R | |
1427 | * |---|-------|---| | |
1428 | */ | |
82dd23e8 URS |
1429 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
1430 | if (unlikely(!lva)) { | |
1431 | /* | |
1432 | * For percpu allocator we do not do any pre-allocation | |
1433 | * and leave it as it is. The reason is it most likely | |
1434 | * never ends up with NE_FIT_TYPE splitting. In case of | |
1435 | * percpu allocations offsets and sizes are aligned to | |
1436 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
1437 | * are its main fitting cases. | |
1438 | * | |
1439 | * There are a few exceptions though, as an example it is | |
1440 | * a first allocation (early boot up) when we have "one" | |
1441 | * big free space that has to be split. | |
060650a2 URS |
1442 | * |
1443 | * Also we can hit this path in case of regular "vmap" | |
1444 | * allocations, if "this" current CPU was not preloaded. | |
1445 | * See the comment in alloc_vmap_area() why. If so, then | |
1446 | * GFP_NOWAIT is used instead to get an extra object for | |
1447 | * split purpose. That is rare and most time does not | |
1448 | * occur. | |
1449 | * | |
1450 | * What happens if an allocation gets failed. Basically, | |
1451 | * an "overflow" path is triggered to purge lazily freed | |
1452 | * areas to free some memory, then, the "retry" path is | |
1453 | * triggered to repeat one more time. See more details | |
1454 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
1455 | */ |
1456 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
1457 | if (!lva) | |
1458 | return -1; | |
1459 | } | |
68ad4a33 URS |
1460 | |
1461 | /* | |
1462 | * Build the remainder. | |
1463 | */ | |
1464 | lva->va_start = va->va_start; | |
1465 | lva->va_end = nva_start_addr; | |
1466 | ||
1467 | /* | |
1468 | * Shrink this VA to remaining size. | |
1469 | */ | |
1470 | va->va_start = nva_start_addr + size; | |
1471 | } else { | |
1472 | return -1; | |
1473 | } | |
1474 | ||
1475 | if (type != FL_FIT_TYPE) { | |
1476 | augment_tree_propagate_from(va); | |
1477 | ||
2c929233 | 1478 | if (lva) /* type == NE_FIT_TYPE */ |
f9863be4 | 1479 | insert_vmap_area_augment(lva, &va->rb_node, root, head); |
68ad4a33 URS |
1480 | } |
1481 | ||
1482 | return 0; | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * Returns a start address of the newly allocated area, if success. | |
1487 | * Otherwise a vend is returned that indicates failure. | |
1488 | */ | |
1489 | static __always_inline unsigned long | |
f9863be4 URS |
1490 | __alloc_vmap_area(struct rb_root *root, struct list_head *head, |
1491 | unsigned long size, unsigned long align, | |
cacca6ba | 1492 | unsigned long vstart, unsigned long vend) |
68ad4a33 | 1493 | { |
9333fe98 | 1494 | bool adjust_search_size = true; |
68ad4a33 URS |
1495 | unsigned long nva_start_addr; |
1496 | struct vmap_area *va; | |
68ad4a33 URS |
1497 | int ret; |
1498 | ||
9333fe98 UR |
1499 | /* |
1500 | * Do not adjust when: | |
1501 | * a) align <= PAGE_SIZE, because it does not make any sense. | |
1502 | * All blocks(their start addresses) are at least PAGE_SIZE | |
1503 | * aligned anyway; | |
1504 | * b) a short range where a requested size corresponds to exactly | |
1505 | * specified [vstart:vend] interval and an alignment > PAGE_SIZE. | |
1506 | * With adjusted search length an allocation would not succeed. | |
1507 | */ | |
1508 | if (align <= PAGE_SIZE || (align > PAGE_SIZE && (vend - vstart) == size)) | |
1509 | adjust_search_size = false; | |
1510 | ||
f9863be4 | 1511 | va = find_vmap_lowest_match(root, size, align, vstart, adjust_search_size); |
68ad4a33 URS |
1512 | if (unlikely(!va)) |
1513 | return vend; | |
1514 | ||
1515 | if (va->va_start > vstart) | |
1516 | nva_start_addr = ALIGN(va->va_start, align); | |
1517 | else | |
1518 | nva_start_addr = ALIGN(vstart, align); | |
1519 | ||
1520 | /* Check the "vend" restriction. */ | |
1521 | if (nva_start_addr + size > vend) | |
1522 | return vend; | |
1523 | ||
68ad4a33 | 1524 | /* Update the free vmap_area. */ |
f9863be4 | 1525 | ret = adjust_va_to_fit_type(root, head, va, nva_start_addr, size); |
1b23ff80 | 1526 | if (WARN_ON_ONCE(ret)) |
68ad4a33 URS |
1527 | return vend; |
1528 | ||
a6cf4e0f | 1529 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
bd1264c3 | 1530 | find_vmap_lowest_match_check(root, head, size, align); |
a6cf4e0f URS |
1531 | #endif |
1532 | ||
68ad4a33 URS |
1533 | return nva_start_addr; |
1534 | } | |
4da56b99 | 1535 | |
d98c9e83 AR |
1536 | /* |
1537 | * Free a region of KVA allocated by alloc_vmap_area | |
1538 | */ | |
1539 | static void free_vmap_area(struct vmap_area *va) | |
1540 | { | |
1541 | /* | |
1542 | * Remove from the busy tree/list. | |
1543 | */ | |
1544 | spin_lock(&vmap_area_lock); | |
1545 | unlink_va(va, &vmap_area_root); | |
1546 | spin_unlock(&vmap_area_lock); | |
1547 | ||
1548 | /* | |
1549 | * Insert/Merge it back to the free tree/list. | |
1550 | */ | |
1551 | spin_lock(&free_vmap_area_lock); | |
96e2db45 | 1552 | merge_or_add_vmap_area_augment(va, &free_vmap_area_root, &free_vmap_area_list); |
d98c9e83 AR |
1553 | spin_unlock(&free_vmap_area_lock); |
1554 | } | |
1555 | ||
187f8cc4 URS |
1556 | static inline void |
1557 | preload_this_cpu_lock(spinlock_t *lock, gfp_t gfp_mask, int node) | |
1558 | { | |
1559 | struct vmap_area *va = NULL; | |
1560 | ||
1561 | /* | |
1562 | * Preload this CPU with one extra vmap_area object. It is used | |
1563 | * when fit type of free area is NE_FIT_TYPE. It guarantees that | |
1564 | * a CPU that does an allocation is preloaded. | |
1565 | * | |
1566 | * We do it in non-atomic context, thus it allows us to use more | |
1567 | * permissive allocation masks to be more stable under low memory | |
1568 | * condition and high memory pressure. | |
1569 | */ | |
1570 | if (!this_cpu_read(ne_fit_preload_node)) | |
1571 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); | |
1572 | ||
1573 | spin_lock(lock); | |
1574 | ||
1575 | if (va && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, va)) | |
1576 | kmem_cache_free(vmap_area_cachep, va); | |
1577 | } | |
1578 | ||
db64fe02 NP |
1579 | /* |
1580 | * Allocate a region of KVA of the specified size and alignment, within the | |
1581 | * vstart and vend. | |
1582 | */ | |
1583 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1584 | unsigned long align, | |
1585 | unsigned long vstart, unsigned long vend, | |
869176a0 BH |
1586 | int node, gfp_t gfp_mask, |
1587 | unsigned long va_flags) | |
db64fe02 | 1588 | { |
187f8cc4 | 1589 | struct vmap_area *va; |
12e376a6 | 1590 | unsigned long freed; |
1da177e4 | 1591 | unsigned long addr; |
db64fe02 | 1592 | int purged = 0; |
d98c9e83 | 1593 | int ret; |
db64fe02 | 1594 | |
7e4a32c0 HL |
1595 | if (unlikely(!size || offset_in_page(size) || !is_power_of_2(align))) |
1596 | return ERR_PTR(-EINVAL); | |
db64fe02 | 1597 | |
68ad4a33 URS |
1598 | if (unlikely(!vmap_initialized)) |
1599 | return ERR_PTR(-EBUSY); | |
1600 | ||
5803ed29 | 1601 | might_sleep(); |
f07116d7 | 1602 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1603 | |
f07116d7 | 1604 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1605 | if (unlikely(!va)) |
1606 | return ERR_PTR(-ENOMEM); | |
1607 | ||
7f88f88f CM |
1608 | /* |
1609 | * Only scan the relevant parts containing pointers to other objects | |
1610 | * to avoid false negatives. | |
1611 | */ | |
f07116d7 | 1612 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1613 | |
db64fe02 | 1614 | retry: |
187f8cc4 | 1615 | preload_this_cpu_lock(&free_vmap_area_lock, gfp_mask, node); |
f9863be4 URS |
1616 | addr = __alloc_vmap_area(&free_vmap_area_root, &free_vmap_area_list, |
1617 | size, align, vstart, vend); | |
187f8cc4 | 1618 | spin_unlock(&free_vmap_area_lock); |
89699605 | 1619 | |
cf243da6 URS |
1620 | trace_alloc_vmap_area(addr, size, align, vstart, vend, addr == vend); |
1621 | ||
afd07389 | 1622 | /* |
68ad4a33 URS |
1623 | * If an allocation fails, the "vend" address is |
1624 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1625 | */ |
68ad4a33 | 1626 | if (unlikely(addr == vend)) |
89699605 | 1627 | goto overflow; |
db64fe02 NP |
1628 | |
1629 | va->va_start = addr; | |
1630 | va->va_end = addr + size; | |
688fcbfc | 1631 | va->vm = NULL; |
869176a0 | 1632 | va->flags = va_flags; |
68ad4a33 | 1633 | |
e36176be URS |
1634 | spin_lock(&vmap_area_lock); |
1635 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
db64fe02 NP |
1636 | spin_unlock(&vmap_area_lock); |
1637 | ||
61e16557 | 1638 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1639 | BUG_ON(va->va_start < vstart); |
1640 | BUG_ON(va->va_end > vend); | |
1641 | ||
d98c9e83 AR |
1642 | ret = kasan_populate_vmalloc(addr, size); |
1643 | if (ret) { | |
1644 | free_vmap_area(va); | |
1645 | return ERR_PTR(ret); | |
1646 | } | |
1647 | ||
db64fe02 | 1648 | return va; |
89699605 NP |
1649 | |
1650 | overflow: | |
89699605 | 1651 | if (!purged) { |
77e50af0 | 1652 | reclaim_and_purge_vmap_areas(); |
89699605 NP |
1653 | purged = 1; |
1654 | goto retry; | |
1655 | } | |
4da56b99 | 1656 | |
12e376a6 URS |
1657 | freed = 0; |
1658 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1659 | ||
1660 | if (freed > 0) { | |
1661 | purged = 0; | |
1662 | goto retry; | |
4da56b99 CW |
1663 | } |
1664 | ||
03497d76 | 1665 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1666 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1667 | size); | |
68ad4a33 URS |
1668 | |
1669 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1670 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1671 | } |
1672 | ||
4da56b99 CW |
1673 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1674 | { | |
1675 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1678 | ||
1679 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1680 | { | |
1681 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1682 | } | |
1683 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1684 | ||
db64fe02 NP |
1685 | /* |
1686 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1687 | * before attempting to purge with a TLB flush. | |
1688 | * | |
1689 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1690 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1691 | * global TLB flushes that must be performed. It would seem natural to scale | |
1692 | * this number up linearly with the number of CPUs (because vmapping activity | |
1693 | * could also scale linearly with the number of CPUs), however it is likely | |
1694 | * that in practice, workloads might be constrained in other ways that mean | |
1695 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1696 | * conservative and not introduce a big latency on huge systems, so go with | |
1697 | * a less aggressive log scale. It will still be an improvement over the old | |
1698 | * code, and it will be simple to change the scale factor if we find that it | |
1699 | * becomes a problem on bigger systems. | |
1700 | */ | |
1701 | static unsigned long lazy_max_pages(void) | |
1702 | { | |
1703 | unsigned int log; | |
1704 | ||
1705 | log = fls(num_online_cpus()); | |
1706 | ||
1707 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1708 | } | |
1709 | ||
4d36e6f8 | 1710 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1711 | |
0574ecd1 | 1712 | /* |
f0953a1b | 1713 | * Serialize vmap purging. There is no actual critical section protected |
153090f2 | 1714 | * by this lock, but we want to avoid concurrent calls for performance |
0574ecd1 CH |
1715 | * reasons and to make the pcpu_get_vm_areas more deterministic. |
1716 | */ | |
f9e09977 | 1717 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1718 | |
02b709df NP |
1719 | /* for per-CPU blocks */ |
1720 | static void purge_fragmented_blocks_allcpus(void); | |
1721 | ||
db64fe02 NP |
1722 | /* |
1723 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1724 | */ |
0574ecd1 | 1725 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1726 | { |
4d36e6f8 | 1727 | unsigned long resched_threshold; |
6030fd5f | 1728 | unsigned int num_purged_areas = 0; |
baa468a6 | 1729 | struct list_head local_purge_list; |
96e2db45 | 1730 | struct vmap_area *va, *n_va; |
db64fe02 | 1731 | |
0574ecd1 | 1732 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1733 | |
96e2db45 URS |
1734 | spin_lock(&purge_vmap_area_lock); |
1735 | purge_vmap_area_root = RB_ROOT; | |
baa468a6 | 1736 | list_replace_init(&purge_vmap_area_list, &local_purge_list); |
96e2db45 URS |
1737 | spin_unlock(&purge_vmap_area_lock); |
1738 | ||
baa468a6 | 1739 | if (unlikely(list_empty(&local_purge_list))) |
6030fd5f | 1740 | goto out; |
68571be9 | 1741 | |
96e2db45 | 1742 | start = min(start, |
baa468a6 | 1743 | list_first_entry(&local_purge_list, |
96e2db45 URS |
1744 | struct vmap_area, list)->va_start); |
1745 | ||
1746 | end = max(end, | |
baa468a6 | 1747 | list_last_entry(&local_purge_list, |
96e2db45 | 1748 | struct vmap_area, list)->va_end); |
db64fe02 | 1749 | |
0574ecd1 | 1750 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1751 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1752 | |
e36176be | 1753 | spin_lock(&free_vmap_area_lock); |
baa468a6 | 1754 | list_for_each_entry_safe(va, n_va, &local_purge_list, list) { |
4d36e6f8 | 1755 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
3c5c3cfb DA |
1756 | unsigned long orig_start = va->va_start; |
1757 | unsigned long orig_end = va->va_end; | |
763b218d | 1758 | |
dd3b8353 URS |
1759 | /* |
1760 | * Finally insert or merge lazily-freed area. It is | |
1761 | * detached and there is no need to "unlink" it from | |
1762 | * anything. | |
1763 | */ | |
96e2db45 URS |
1764 | va = merge_or_add_vmap_area_augment(va, &free_vmap_area_root, |
1765 | &free_vmap_area_list); | |
3c5c3cfb | 1766 | |
9c801f61 URS |
1767 | if (!va) |
1768 | continue; | |
1769 | ||
3c5c3cfb DA |
1770 | if (is_vmalloc_or_module_addr((void *)orig_start)) |
1771 | kasan_release_vmalloc(orig_start, orig_end, | |
1772 | va->va_start, va->va_end); | |
dd3b8353 | 1773 | |
4d36e6f8 | 1774 | atomic_long_sub(nr, &vmap_lazy_nr); |
6030fd5f | 1775 | num_purged_areas++; |
68571be9 | 1776 | |
4d36e6f8 | 1777 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
e36176be | 1778 | cond_resched_lock(&free_vmap_area_lock); |
763b218d | 1779 | } |
e36176be | 1780 | spin_unlock(&free_vmap_area_lock); |
6030fd5f URS |
1781 | |
1782 | out: | |
1783 | trace_purge_vmap_area_lazy(start, end, num_purged_areas); | |
1784 | return num_purged_areas > 0; | |
db64fe02 NP |
1785 | } |
1786 | ||
1787 | /* | |
77e50af0 | 1788 | * Reclaim vmap areas by purging fragmented blocks and purge_vmap_area_list. |
db64fe02 | 1789 | */ |
77e50af0 TG |
1790 | static void reclaim_and_purge_vmap_areas(void) |
1791 | ||
db64fe02 | 1792 | { |
f9e09977 | 1793 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1794 | purge_fragmented_blocks_allcpus(); |
1795 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1796 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1797 | } |
1798 | ||
690467c8 URS |
1799 | static void drain_vmap_area_work(struct work_struct *work) |
1800 | { | |
1801 | unsigned long nr_lazy; | |
1802 | ||
1803 | do { | |
1804 | mutex_lock(&vmap_purge_lock); | |
1805 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
1806 | mutex_unlock(&vmap_purge_lock); | |
1807 | ||
1808 | /* Recheck if further work is required. */ | |
1809 | nr_lazy = atomic_long_read(&vmap_lazy_nr); | |
1810 | } while (nr_lazy > lazy_max_pages()); | |
1811 | } | |
1812 | ||
db64fe02 | 1813 | /* |
edd89818 URS |
1814 | * Free a vmap area, caller ensuring that the area has been unmapped, |
1815 | * unlinked and flush_cache_vunmap had been called for the correct | |
1816 | * range previously. | |
db64fe02 | 1817 | */ |
64141da5 | 1818 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1819 | { |
8c4196fe URS |
1820 | unsigned long nr_lazy_max = lazy_max_pages(); |
1821 | unsigned long va_start = va->va_start; | |
4d36e6f8 | 1822 | unsigned long nr_lazy; |
80c4bd7a | 1823 | |
edd89818 URS |
1824 | if (WARN_ON_ONCE(!list_empty(&va->list))) |
1825 | return; | |
dd3b8353 | 1826 | |
4d36e6f8 URS |
1827 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1828 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a | 1829 | |
96e2db45 URS |
1830 | /* |
1831 | * Merge or place it to the purge tree/list. | |
1832 | */ | |
1833 | spin_lock(&purge_vmap_area_lock); | |
1834 | merge_or_add_vmap_area(va, | |
1835 | &purge_vmap_area_root, &purge_vmap_area_list); | |
1836 | spin_unlock(&purge_vmap_area_lock); | |
80c4bd7a | 1837 | |
8c4196fe URS |
1838 | trace_free_vmap_area_noflush(va_start, nr_lazy, nr_lazy_max); |
1839 | ||
96e2db45 | 1840 | /* After this point, we may free va at any time */ |
8c4196fe | 1841 | if (unlikely(nr_lazy > nr_lazy_max)) |
690467c8 | 1842 | schedule_work(&drain_vmap_work); |
db64fe02 NP |
1843 | } |
1844 | ||
b29acbdc NP |
1845 | /* |
1846 | * Free and unmap a vmap area | |
1847 | */ | |
1848 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1849 | { | |
1850 | flush_cache_vunmap(va->va_start, va->va_end); | |
4ad0ae8c | 1851 | vunmap_range_noflush(va->va_start, va->va_end); |
8e57f8ac | 1852 | if (debug_pagealloc_enabled_static()) |
82a2e924 CP |
1853 | flush_tlb_kernel_range(va->va_start, va->va_end); |
1854 | ||
c8eef01e | 1855 | free_vmap_area_noflush(va); |
b29acbdc NP |
1856 | } |
1857 | ||
993d0b28 | 1858 | struct vmap_area *find_vmap_area(unsigned long addr) |
db64fe02 NP |
1859 | { |
1860 | struct vmap_area *va; | |
1861 | ||
1862 | spin_lock(&vmap_area_lock); | |
899c6efe | 1863 | va = __find_vmap_area(addr, &vmap_area_root); |
db64fe02 NP |
1864 | spin_unlock(&vmap_area_lock); |
1865 | ||
1866 | return va; | |
1867 | } | |
1868 | ||
edd89818 URS |
1869 | static struct vmap_area *find_unlink_vmap_area(unsigned long addr) |
1870 | { | |
1871 | struct vmap_area *va; | |
1872 | ||
1873 | spin_lock(&vmap_area_lock); | |
1874 | va = __find_vmap_area(addr, &vmap_area_root); | |
1875 | if (va) | |
1876 | unlink_va(va, &vmap_area_root); | |
1877 | spin_unlock(&vmap_area_lock); | |
1878 | ||
1879 | return va; | |
1880 | } | |
1881 | ||
db64fe02 NP |
1882 | /*** Per cpu kva allocator ***/ |
1883 | ||
1884 | /* | |
1885 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1886 | * room for at least 16 percpu vmap blocks per CPU. | |
1887 | */ | |
1888 | /* | |
1889 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1890 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1891 | * instead (we just need a rough idea) | |
1892 | */ | |
1893 | #if BITS_PER_LONG == 32 | |
1894 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1895 | #else | |
1896 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1897 | #endif | |
1898 | ||
1899 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1900 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1901 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1902 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1903 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1904 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1905 | #define VMAP_BBMAP_BITS \ |
1906 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1907 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1908 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1909 | |
1910 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1911 | ||
77e50af0 TG |
1912 | /* |
1913 | * Purge threshold to prevent overeager purging of fragmented blocks for | |
1914 | * regular operations: Purge if vb->free is less than 1/4 of the capacity. | |
1915 | */ | |
1916 | #define VMAP_PURGE_THRESHOLD (VMAP_BBMAP_BITS / 4) | |
1917 | ||
869176a0 BH |
1918 | #define VMAP_RAM 0x1 /* indicates vm_map_ram area*/ |
1919 | #define VMAP_BLOCK 0x2 /* mark out the vmap_block sub-type*/ | |
1920 | #define VMAP_FLAGS_MASK 0x3 | |
1921 | ||
db64fe02 NP |
1922 | struct vmap_block_queue { |
1923 | spinlock_t lock; | |
1924 | struct list_head free; | |
062eacf5 URS |
1925 | |
1926 | /* | |
1927 | * An xarray requires an extra memory dynamically to | |
1928 | * be allocated. If it is an issue, we can use rb-tree | |
1929 | * instead. | |
1930 | */ | |
1931 | struct xarray vmap_blocks; | |
db64fe02 NP |
1932 | }; |
1933 | ||
1934 | struct vmap_block { | |
1935 | spinlock_t lock; | |
1936 | struct vmap_area *va; | |
db64fe02 | 1937 | unsigned long free, dirty; |
d76f9954 | 1938 | DECLARE_BITMAP(used_map, VMAP_BBMAP_BITS); |
7d61bfe8 | 1939 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1940 | struct list_head free_list; |
1941 | struct rcu_head rcu_head; | |
02b709df | 1942 | struct list_head purge; |
db64fe02 NP |
1943 | }; |
1944 | ||
1945 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1946 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1947 | ||
1948 | /* | |
062eacf5 URS |
1949 | * In order to fast access to any "vmap_block" associated with a |
1950 | * specific address, we use a hash. | |
1951 | * | |
1952 | * A per-cpu vmap_block_queue is used in both ways, to serialize | |
1953 | * an access to free block chains among CPUs(alloc path) and it | |
1954 | * also acts as a vmap_block hash(alloc/free paths). It means we | |
1955 | * overload it, since we already have the per-cpu array which is | |
1956 | * used as a hash table. When used as a hash a 'cpu' passed to | |
1957 | * per_cpu() is not actually a CPU but rather a hash index. | |
1958 | * | |
fa1c77c1 | 1959 | * A hash function is addr_to_vb_xa() which hashes any address |
062eacf5 URS |
1960 | * to a specific index(in a hash) it belongs to. This then uses a |
1961 | * per_cpu() macro to access an array with generated index. | |
1962 | * | |
1963 | * An example: | |
1964 | * | |
1965 | * CPU_1 CPU_2 CPU_0 | |
1966 | * | | | | |
1967 | * V V V | |
1968 | * 0 10 20 30 40 50 60 | |
1969 | * |------|------|------|------|------|------|...<vmap address space> | |
1970 | * CPU0 CPU1 CPU2 CPU0 CPU1 CPU2 | |
1971 | * | |
1972 | * - CPU_1 invokes vm_unmap_ram(6), 6 belongs to CPU0 zone, thus | |
1973 | * it access: CPU0/INDEX0 -> vmap_blocks -> xa_lock; | |
1974 | * | |
1975 | * - CPU_2 invokes vm_unmap_ram(11), 11 belongs to CPU1 zone, thus | |
1976 | * it access: CPU1/INDEX1 -> vmap_blocks -> xa_lock; | |
1977 | * | |
1978 | * - CPU_0 invokes vm_unmap_ram(20), 20 belongs to CPU2 zone, thus | |
1979 | * it access: CPU2/INDEX2 -> vmap_blocks -> xa_lock. | |
1980 | * | |
1981 | * This technique almost always avoids lock contention on insert/remove, | |
1982 | * however xarray spinlocks protect against any contention that remains. | |
db64fe02 | 1983 | */ |
062eacf5 | 1984 | static struct xarray * |
fa1c77c1 | 1985 | addr_to_vb_xa(unsigned long addr) |
062eacf5 URS |
1986 | { |
1987 | int index = (addr / VMAP_BLOCK_SIZE) % num_possible_cpus(); | |
1988 | ||
1989 | return &per_cpu(vmap_block_queue, index).vmap_blocks; | |
1990 | } | |
db64fe02 NP |
1991 | |
1992 | /* | |
1993 | * We should probably have a fallback mechanism to allocate virtual memory | |
1994 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1995 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1996 | * big problem. | |
1997 | */ | |
1998 | ||
1999 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
2000 | { | |
2001 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
2002 | addr /= VMAP_BLOCK_SIZE; | |
2003 | return addr; | |
2004 | } | |
2005 | ||
cf725ce2 RP |
2006 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
2007 | { | |
2008 | unsigned long addr; | |
2009 | ||
2010 | addr = va_start + (pages_off << PAGE_SHIFT); | |
2011 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
2012 | return (void *)addr; | |
2013 | } | |
2014 | ||
2015 | /** | |
2016 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
2017 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
2018 | * @order: how many 2^order pages should be occupied in newly allocated block | |
2019 | * @gfp_mask: flags for the page level allocator | |
2020 | * | |
a862f68a | 2021 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
2022 | */ |
2023 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
2024 | { |
2025 | struct vmap_block_queue *vbq; | |
2026 | struct vmap_block *vb; | |
2027 | struct vmap_area *va; | |
062eacf5 | 2028 | struct xarray *xa; |
db64fe02 NP |
2029 | unsigned long vb_idx; |
2030 | int node, err; | |
cf725ce2 | 2031 | void *vaddr; |
db64fe02 NP |
2032 | |
2033 | node = numa_node_id(); | |
2034 | ||
2035 | vb = kmalloc_node(sizeof(struct vmap_block), | |
2036 | gfp_mask & GFP_RECLAIM_MASK, node); | |
2037 | if (unlikely(!vb)) | |
2038 | return ERR_PTR(-ENOMEM); | |
2039 | ||
2040 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
2041 | VMALLOC_START, VMALLOC_END, | |
869176a0 BH |
2042 | node, gfp_mask, |
2043 | VMAP_RAM|VMAP_BLOCK); | |
ddf9c6d4 | 2044 | if (IS_ERR(va)) { |
db64fe02 | 2045 | kfree(vb); |
e7d86340 | 2046 | return ERR_CAST(va); |
db64fe02 NP |
2047 | } |
2048 | ||
cf725ce2 | 2049 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
2050 | spin_lock_init(&vb->lock); |
2051 | vb->va = va; | |
cf725ce2 RP |
2052 | /* At least something should be left free */ |
2053 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
d76f9954 | 2054 | bitmap_zero(vb->used_map, VMAP_BBMAP_BITS); |
cf725ce2 | 2055 | vb->free = VMAP_BBMAP_BITS - (1UL << order); |
db64fe02 | 2056 | vb->dirty = 0; |
7d61bfe8 RP |
2057 | vb->dirty_min = VMAP_BBMAP_BITS; |
2058 | vb->dirty_max = 0; | |
d76f9954 | 2059 | bitmap_set(vb->used_map, 0, (1UL << order)); |
db64fe02 | 2060 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 | 2061 | |
fa1c77c1 | 2062 | xa = addr_to_vb_xa(va->va_start); |
db64fe02 | 2063 | vb_idx = addr_to_vb_idx(va->va_start); |
062eacf5 | 2064 | err = xa_insert(xa, vb_idx, vb, gfp_mask); |
0f14599c MWO |
2065 | if (err) { |
2066 | kfree(vb); | |
2067 | free_vmap_area(va); | |
2068 | return ERR_PTR(err); | |
2069 | } | |
db64fe02 | 2070 | |
3f804920 | 2071 | vbq = raw_cpu_ptr(&vmap_block_queue); |
db64fe02 | 2072 | spin_lock(&vbq->lock); |
68ac546f | 2073 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 2074 | spin_unlock(&vbq->lock); |
db64fe02 | 2075 | |
cf725ce2 | 2076 | return vaddr; |
db64fe02 NP |
2077 | } |
2078 | ||
db64fe02 NP |
2079 | static void free_vmap_block(struct vmap_block *vb) |
2080 | { | |
2081 | struct vmap_block *tmp; | |
062eacf5 | 2082 | struct xarray *xa; |
db64fe02 | 2083 | |
fa1c77c1 | 2084 | xa = addr_to_vb_xa(vb->va->va_start); |
062eacf5 | 2085 | tmp = xa_erase(xa, addr_to_vb_idx(vb->va->va_start)); |
db64fe02 NP |
2086 | BUG_ON(tmp != vb); |
2087 | ||
edd89818 URS |
2088 | spin_lock(&vmap_area_lock); |
2089 | unlink_va(vb->va, &vmap_area_root); | |
2090 | spin_unlock(&vmap_area_lock); | |
2091 | ||
64141da5 | 2092 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 2093 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
2094 | } |
2095 | ||
ca5e46c3 | 2096 | static bool purge_fragmented_block(struct vmap_block *vb, |
77e50af0 TG |
2097 | struct vmap_block_queue *vbq, struct list_head *purge_list, |
2098 | bool force_purge) | |
ca5e46c3 TG |
2099 | { |
2100 | if (vb->free + vb->dirty != VMAP_BBMAP_BITS || | |
2101 | vb->dirty == VMAP_BBMAP_BITS) | |
2102 | return false; | |
2103 | ||
77e50af0 TG |
2104 | /* Don't overeagerly purge usable blocks unless requested */ |
2105 | if (!(force_purge || vb->free < VMAP_PURGE_THRESHOLD)) | |
2106 | return false; | |
2107 | ||
ca5e46c3 | 2108 | /* prevent further allocs after releasing lock */ |
7f48121e | 2109 | WRITE_ONCE(vb->free, 0); |
ca5e46c3 | 2110 | /* prevent purging it again */ |
7f48121e | 2111 | WRITE_ONCE(vb->dirty, VMAP_BBMAP_BITS); |
ca5e46c3 TG |
2112 | vb->dirty_min = 0; |
2113 | vb->dirty_max = VMAP_BBMAP_BITS; | |
2114 | spin_lock(&vbq->lock); | |
2115 | list_del_rcu(&vb->free_list); | |
2116 | spin_unlock(&vbq->lock); | |
2117 | list_add_tail(&vb->purge, purge_list); | |
2118 | return true; | |
2119 | } | |
2120 | ||
2121 | static void free_purged_blocks(struct list_head *purge_list) | |
2122 | { | |
2123 | struct vmap_block *vb, *n_vb; | |
2124 | ||
2125 | list_for_each_entry_safe(vb, n_vb, purge_list, purge) { | |
2126 | list_del(&vb->purge); | |
2127 | free_vmap_block(vb); | |
2128 | } | |
2129 | } | |
2130 | ||
02b709df NP |
2131 | static void purge_fragmented_blocks(int cpu) |
2132 | { | |
2133 | LIST_HEAD(purge); | |
2134 | struct vmap_block *vb; | |
02b709df NP |
2135 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); |
2136 | ||
2137 | rcu_read_lock(); | |
2138 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
7f48121e TG |
2139 | unsigned long free = READ_ONCE(vb->free); |
2140 | unsigned long dirty = READ_ONCE(vb->dirty); | |
2141 | ||
2142 | if (free + dirty != VMAP_BBMAP_BITS || | |
2143 | dirty == VMAP_BBMAP_BITS) | |
02b709df NP |
2144 | continue; |
2145 | ||
2146 | spin_lock(&vb->lock); | |
77e50af0 | 2147 | purge_fragmented_block(vb, vbq, &purge, true); |
ca5e46c3 | 2148 | spin_unlock(&vb->lock); |
02b709df NP |
2149 | } |
2150 | rcu_read_unlock(); | |
ca5e46c3 | 2151 | free_purged_blocks(&purge); |
02b709df NP |
2152 | } |
2153 | ||
02b709df NP |
2154 | static void purge_fragmented_blocks_allcpus(void) |
2155 | { | |
2156 | int cpu; | |
2157 | ||
2158 | for_each_possible_cpu(cpu) | |
2159 | purge_fragmented_blocks(cpu); | |
2160 | } | |
2161 | ||
db64fe02 NP |
2162 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
2163 | { | |
2164 | struct vmap_block_queue *vbq; | |
2165 | struct vmap_block *vb; | |
cf725ce2 | 2166 | void *vaddr = NULL; |
db64fe02 NP |
2167 | unsigned int order; |
2168 | ||
891c49ab | 2169 | BUG_ON(offset_in_page(size)); |
db64fe02 | 2170 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
2171 | if (WARN_ON(size == 0)) { |
2172 | /* | |
2173 | * Allocating 0 bytes isn't what caller wants since | |
2174 | * get_order(0) returns funny result. Just warn and terminate | |
2175 | * early. | |
2176 | */ | |
2177 | return NULL; | |
2178 | } | |
db64fe02 NP |
2179 | order = get_order(size); |
2180 | ||
db64fe02 | 2181 | rcu_read_lock(); |
3f804920 | 2182 | vbq = raw_cpu_ptr(&vmap_block_queue); |
db64fe02 | 2183 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
cf725ce2 | 2184 | unsigned long pages_off; |
db64fe02 | 2185 | |
43d76502 TG |
2186 | if (READ_ONCE(vb->free) < (1UL << order)) |
2187 | continue; | |
2188 | ||
db64fe02 | 2189 | spin_lock(&vb->lock); |
cf725ce2 RP |
2190 | if (vb->free < (1UL << order)) { |
2191 | spin_unlock(&vb->lock); | |
2192 | continue; | |
2193 | } | |
02b709df | 2194 | |
cf725ce2 RP |
2195 | pages_off = VMAP_BBMAP_BITS - vb->free; |
2196 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
43d76502 | 2197 | WRITE_ONCE(vb->free, vb->free - (1UL << order)); |
d76f9954 | 2198 | bitmap_set(vb->used_map, pages_off, (1UL << order)); |
02b709df NP |
2199 | if (vb->free == 0) { |
2200 | spin_lock(&vbq->lock); | |
2201 | list_del_rcu(&vb->free_list); | |
2202 | spin_unlock(&vbq->lock); | |
2203 | } | |
cf725ce2 | 2204 | |
02b709df NP |
2205 | spin_unlock(&vb->lock); |
2206 | break; | |
db64fe02 | 2207 | } |
02b709df | 2208 | |
db64fe02 NP |
2209 | rcu_read_unlock(); |
2210 | ||
cf725ce2 RP |
2211 | /* Allocate new block if nothing was found */ |
2212 | if (!vaddr) | |
2213 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 2214 | |
cf725ce2 | 2215 | return vaddr; |
db64fe02 NP |
2216 | } |
2217 | ||
78a0e8c4 | 2218 | static void vb_free(unsigned long addr, unsigned long size) |
db64fe02 NP |
2219 | { |
2220 | unsigned long offset; | |
db64fe02 NP |
2221 | unsigned int order; |
2222 | struct vmap_block *vb; | |
062eacf5 | 2223 | struct xarray *xa; |
db64fe02 | 2224 | |
891c49ab | 2225 | BUG_ON(offset_in_page(size)); |
db64fe02 | 2226 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc | 2227 | |
78a0e8c4 | 2228 | flush_cache_vunmap(addr, addr + size); |
b29acbdc | 2229 | |
db64fe02 | 2230 | order = get_order(size); |
78a0e8c4 | 2231 | offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT; |
062eacf5 | 2232 | |
fa1c77c1 | 2233 | xa = addr_to_vb_xa(addr); |
062eacf5 URS |
2234 | vb = xa_load(xa, addr_to_vb_idx(addr)); |
2235 | ||
d76f9954 BH |
2236 | spin_lock(&vb->lock); |
2237 | bitmap_clear(vb->used_map, offset, (1UL << order)); | |
2238 | spin_unlock(&vb->lock); | |
db64fe02 | 2239 | |
4ad0ae8c | 2240 | vunmap_range_noflush(addr, addr + size); |
64141da5 | 2241 | |
8e57f8ac | 2242 | if (debug_pagealloc_enabled_static()) |
78a0e8c4 | 2243 | flush_tlb_kernel_range(addr, addr + size); |
82a2e924 | 2244 | |
db64fe02 | 2245 | spin_lock(&vb->lock); |
7d61bfe8 | 2246 | |
a09fad96 | 2247 | /* Expand the not yet TLB flushed dirty range */ |
7d61bfe8 RP |
2248 | vb->dirty_min = min(vb->dirty_min, offset); |
2249 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 2250 | |
7f48121e | 2251 | WRITE_ONCE(vb->dirty, vb->dirty + (1UL << order)); |
db64fe02 | 2252 | if (vb->dirty == VMAP_BBMAP_BITS) { |
de560423 | 2253 | BUG_ON(vb->free); |
db64fe02 NP |
2254 | spin_unlock(&vb->lock); |
2255 | free_vmap_block(vb); | |
2256 | } else | |
2257 | spin_unlock(&vb->lock); | |
2258 | } | |
2259 | ||
868b104d | 2260 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 2261 | { |
ca5e46c3 | 2262 | LIST_HEAD(purge_list); |
db64fe02 | 2263 | int cpu; |
db64fe02 | 2264 | |
9b463334 JF |
2265 | if (unlikely(!vmap_initialized)) |
2266 | return; | |
2267 | ||
ca5e46c3 | 2268 | mutex_lock(&vmap_purge_lock); |
5803ed29 | 2269 | |
db64fe02 NP |
2270 | for_each_possible_cpu(cpu) { |
2271 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
2272 | struct vmap_block *vb; | |
fc1e0d98 | 2273 | unsigned long idx; |
db64fe02 NP |
2274 | |
2275 | rcu_read_lock(); | |
fc1e0d98 | 2276 | xa_for_each(&vbq->vmap_blocks, idx, vb) { |
db64fe02 | 2277 | spin_lock(&vb->lock); |
ca5e46c3 TG |
2278 | |
2279 | /* | |
2280 | * Try to purge a fragmented block first. If it's | |
2281 | * not purgeable, check whether there is dirty | |
2282 | * space to be flushed. | |
2283 | */ | |
77e50af0 | 2284 | if (!purge_fragmented_block(vb, vbq, &purge_list, false) && |
a09fad96 | 2285 | vb->dirty_max && vb->dirty != VMAP_BBMAP_BITS) { |
7d61bfe8 | 2286 | unsigned long va_start = vb->va->va_start; |
db64fe02 | 2287 | unsigned long s, e; |
b136be5e | 2288 | |
7d61bfe8 RP |
2289 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
2290 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 2291 | |
7d61bfe8 RP |
2292 | start = min(s, start); |
2293 | end = max(e, end); | |
db64fe02 | 2294 | |
a09fad96 TG |
2295 | /* Prevent that this is flushed again */ |
2296 | vb->dirty_min = VMAP_BBMAP_BITS; | |
2297 | vb->dirty_max = 0; | |
2298 | ||
7d61bfe8 | 2299 | flush = 1; |
db64fe02 NP |
2300 | } |
2301 | spin_unlock(&vb->lock); | |
2302 | } | |
2303 | rcu_read_unlock(); | |
2304 | } | |
ca5e46c3 | 2305 | free_purged_blocks(&purge_list); |
db64fe02 | 2306 | |
0574ecd1 CH |
2307 | if (!__purge_vmap_area_lazy(start, end) && flush) |
2308 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 2309 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 2310 | } |
868b104d RE |
2311 | |
2312 | /** | |
2313 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
2314 | * | |
2315 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
2316 | * to amortize TLB flushing overheads. What this means is that any page you | |
2317 | * have now, may, in a former life, have been mapped into kernel virtual | |
2318 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
2319 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
2320 | * | |
2321 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
2322 | * be sure that none of the pages we have control over will have any aliases | |
2323 | * from the vmap layer. | |
2324 | */ | |
2325 | void vm_unmap_aliases(void) | |
2326 | { | |
2327 | unsigned long start = ULONG_MAX, end = 0; | |
2328 | int flush = 0; | |
2329 | ||
2330 | _vm_unmap_aliases(start, end, flush); | |
2331 | } | |
db64fe02 NP |
2332 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
2333 | ||
2334 | /** | |
2335 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
2336 | * @mem: the pointer returned by vm_map_ram | |
2337 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
2338 | */ | |
2339 | void vm_unmap_ram(const void *mem, unsigned int count) | |
2340 | { | |
65ee03c4 | 2341 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
4aff1dc4 | 2342 | unsigned long addr = (unsigned long)kasan_reset_tag(mem); |
9c3acf60 | 2343 | struct vmap_area *va; |
db64fe02 | 2344 | |
5803ed29 | 2345 | might_sleep(); |
db64fe02 NP |
2346 | BUG_ON(!addr); |
2347 | BUG_ON(addr < VMALLOC_START); | |
2348 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 2349 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 2350 | |
d98c9e83 AR |
2351 | kasan_poison_vmalloc(mem, size); |
2352 | ||
9c3acf60 | 2353 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 2354 | debug_check_no_locks_freed(mem, size); |
78a0e8c4 | 2355 | vb_free(addr, size); |
9c3acf60 CH |
2356 | return; |
2357 | } | |
2358 | ||
edd89818 | 2359 | va = find_unlink_vmap_area(addr); |
14687619 URS |
2360 | if (WARN_ON_ONCE(!va)) |
2361 | return; | |
2362 | ||
05e3ff95 CP |
2363 | debug_check_no_locks_freed((void *)va->va_start, |
2364 | (va->va_end - va->va_start)); | |
9c3acf60 | 2365 | free_unmap_vmap_area(va); |
db64fe02 NP |
2366 | } |
2367 | EXPORT_SYMBOL(vm_unmap_ram); | |
2368 | ||
2369 | /** | |
2370 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
2371 | * @pages: an array of pointers to the pages to be mapped | |
2372 | * @count: number of pages | |
2373 | * @node: prefer to allocate data structures on this node | |
e99c97ad | 2374 | * |
36437638 GK |
2375 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
2376 | * faster than vmap so it's good. But if you mix long-life and short-life | |
2377 | * objects with vm_map_ram(), it could consume lots of address space through | |
2378 | * fragmentation (especially on a 32bit machine). You could see failures in | |
2379 | * the end. Please use this function for short-lived objects. | |
2380 | * | |
e99c97ad | 2381 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 | 2382 | */ |
d4efd79a | 2383 | void *vm_map_ram(struct page **pages, unsigned int count, int node) |
db64fe02 | 2384 | { |
65ee03c4 | 2385 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
2386 | unsigned long addr; |
2387 | void *mem; | |
2388 | ||
2389 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
2390 | mem = vb_alloc(size, GFP_KERNEL); | |
2391 | if (IS_ERR(mem)) | |
2392 | return NULL; | |
2393 | addr = (unsigned long)mem; | |
2394 | } else { | |
2395 | struct vmap_area *va; | |
2396 | va = alloc_vmap_area(size, PAGE_SIZE, | |
869176a0 BH |
2397 | VMALLOC_START, VMALLOC_END, |
2398 | node, GFP_KERNEL, VMAP_RAM); | |
db64fe02 NP |
2399 | if (IS_ERR(va)) |
2400 | return NULL; | |
2401 | ||
2402 | addr = va->va_start; | |
2403 | mem = (void *)addr; | |
2404 | } | |
d98c9e83 | 2405 | |
b67177ec NP |
2406 | if (vmap_pages_range(addr, addr + size, PAGE_KERNEL, |
2407 | pages, PAGE_SHIFT) < 0) { | |
db64fe02 NP |
2408 | vm_unmap_ram(mem, count); |
2409 | return NULL; | |
2410 | } | |
b67177ec | 2411 | |
23689e91 AK |
2412 | /* |
2413 | * Mark the pages as accessible, now that they are mapped. | |
2414 | * With hardware tag-based KASAN, marking is skipped for | |
2415 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). | |
2416 | */ | |
f6e39794 | 2417 | mem = kasan_unpoison_vmalloc(mem, size, KASAN_VMALLOC_PROT_NORMAL); |
19f1c3ac | 2418 | |
db64fe02 NP |
2419 | return mem; |
2420 | } | |
2421 | EXPORT_SYMBOL(vm_map_ram); | |
2422 | ||
4341fa45 | 2423 | static struct vm_struct *vmlist __initdata; |
92eac168 | 2424 | |
121e6f32 NP |
2425 | static inline unsigned int vm_area_page_order(struct vm_struct *vm) |
2426 | { | |
2427 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC | |
2428 | return vm->page_order; | |
2429 | #else | |
2430 | return 0; | |
2431 | #endif | |
2432 | } | |
2433 | ||
2434 | static inline void set_vm_area_page_order(struct vm_struct *vm, unsigned int order) | |
2435 | { | |
2436 | #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC | |
2437 | vm->page_order = order; | |
2438 | #else | |
2439 | BUG_ON(order != 0); | |
2440 | #endif | |
2441 | } | |
2442 | ||
be9b7335 NP |
2443 | /** |
2444 | * vm_area_add_early - add vmap area early during boot | |
2445 | * @vm: vm_struct to add | |
2446 | * | |
2447 | * This function is used to add fixed kernel vm area to vmlist before | |
2448 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
2449 | * should contain proper values and the other fields should be zero. | |
2450 | * | |
2451 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2452 | */ | |
2453 | void __init vm_area_add_early(struct vm_struct *vm) | |
2454 | { | |
2455 | struct vm_struct *tmp, **p; | |
2456 | ||
2457 | BUG_ON(vmap_initialized); | |
2458 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
2459 | if (tmp->addr >= vm->addr) { | |
2460 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
2461 | break; | |
2462 | } else | |
2463 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
2464 | } | |
2465 | vm->next = *p; | |
2466 | *p = vm; | |
2467 | } | |
2468 | ||
f0aa6617 TH |
2469 | /** |
2470 | * vm_area_register_early - register vmap area early during boot | |
2471 | * @vm: vm_struct to register | |
c0c0a293 | 2472 | * @align: requested alignment |
f0aa6617 TH |
2473 | * |
2474 | * This function is used to register kernel vm area before | |
2475 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
2476 | * proper values on entry and other fields should be zero. On return, | |
2477 | * vm->addr contains the allocated address. | |
2478 | * | |
2479 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
2480 | */ | |
c0c0a293 | 2481 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 | 2482 | { |
0eb68437 KW |
2483 | unsigned long addr = ALIGN(VMALLOC_START, align); |
2484 | struct vm_struct *cur, **p; | |
c0c0a293 | 2485 | |
0eb68437 | 2486 | BUG_ON(vmap_initialized); |
f0aa6617 | 2487 | |
0eb68437 KW |
2488 | for (p = &vmlist; (cur = *p) != NULL; p = &cur->next) { |
2489 | if ((unsigned long)cur->addr - addr >= vm->size) | |
2490 | break; | |
2491 | addr = ALIGN((unsigned long)cur->addr + cur->size, align); | |
2492 | } | |
f0aa6617 | 2493 | |
0eb68437 KW |
2494 | BUG_ON(addr > VMALLOC_END - vm->size); |
2495 | vm->addr = (void *)addr; | |
2496 | vm->next = *p; | |
2497 | *p = vm; | |
3252b1d8 | 2498 | kasan_populate_early_vm_area_shadow(vm->addr, vm->size); |
f0aa6617 TH |
2499 | } |
2500 | ||
68ad4a33 URS |
2501 | static void vmap_init_free_space(void) |
2502 | { | |
2503 | unsigned long vmap_start = 1; | |
2504 | const unsigned long vmap_end = ULONG_MAX; | |
2505 | struct vmap_area *busy, *free; | |
2506 | ||
2507 | /* | |
2508 | * B F B B B F | |
2509 | * -|-----|.....|-----|-----|-----|.....|- | |
2510 | * | The KVA space | | |
2511 | * |<--------------------------------->| | |
2512 | */ | |
2513 | list_for_each_entry(busy, &vmap_area_list, list) { | |
2514 | if (busy->va_start - vmap_start > 0) { | |
2515 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2516 | if (!WARN_ON_ONCE(!free)) { | |
2517 | free->va_start = vmap_start; | |
2518 | free->va_end = busy->va_start; | |
2519 | ||
2520 | insert_vmap_area_augment(free, NULL, | |
2521 | &free_vmap_area_root, | |
2522 | &free_vmap_area_list); | |
2523 | } | |
2524 | } | |
2525 | ||
2526 | vmap_start = busy->va_end; | |
2527 | } | |
2528 | ||
2529 | if (vmap_end - vmap_start > 0) { | |
2530 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
2531 | if (!WARN_ON_ONCE(!free)) { | |
2532 | free->va_start = vmap_start; | |
2533 | free->va_end = vmap_end; | |
2534 | ||
2535 | insert_vmap_area_augment(free, NULL, | |
2536 | &free_vmap_area_root, | |
2537 | &free_vmap_area_list); | |
2538 | } | |
2539 | } | |
2540 | } | |
2541 | ||
e36176be URS |
2542 | static inline void setup_vmalloc_vm_locked(struct vm_struct *vm, |
2543 | struct vmap_area *va, unsigned long flags, const void *caller) | |
cf88c790 | 2544 | { |
cf88c790 TH |
2545 | vm->flags = flags; |
2546 | vm->addr = (void *)va->va_start; | |
2547 | vm->size = va->va_end - va->va_start; | |
2548 | vm->caller = caller; | |
db1aecaf | 2549 | va->vm = vm; |
e36176be URS |
2550 | } |
2551 | ||
2552 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, | |
2553 | unsigned long flags, const void *caller) | |
2554 | { | |
2555 | spin_lock(&vmap_area_lock); | |
2556 | setup_vmalloc_vm_locked(vm, va, flags, caller); | |
c69480ad | 2557 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2558 | } |
cf88c790 | 2559 | |
20fc02b4 | 2560 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2561 | { |
d4033afd | 2562 | /* |
20fc02b4 | 2563 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2564 | * we should make sure that vm has proper values. |
2565 | * Pair with smp_rmb() in show_numa_info(). | |
2566 | */ | |
2567 | smp_wmb(); | |
20fc02b4 | 2568 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2569 | } |
2570 | ||
db64fe02 | 2571 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
7ca3027b DA |
2572 | unsigned long align, unsigned long shift, unsigned long flags, |
2573 | unsigned long start, unsigned long end, int node, | |
2574 | gfp_t gfp_mask, const void *caller) | |
db64fe02 | 2575 | { |
0006526d | 2576 | struct vmap_area *va; |
db64fe02 | 2577 | struct vm_struct *area; |
d98c9e83 | 2578 | unsigned long requested_size = size; |
1da177e4 | 2579 | |
52fd24ca | 2580 | BUG_ON(in_interrupt()); |
7ca3027b | 2581 | size = ALIGN(size, 1ul << shift); |
31be8309 OH |
2582 | if (unlikely(!size)) |
2583 | return NULL; | |
1da177e4 | 2584 | |
252e5c6e | 2585 | if (flags & VM_IOREMAP) |
2586 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2587 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2588 | ||
cf88c790 | 2589 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2590 | if (unlikely(!area)) |
2591 | return NULL; | |
2592 | ||
71394fe5 AR |
2593 | if (!(flags & VM_NO_GUARD)) |
2594 | size += PAGE_SIZE; | |
1da177e4 | 2595 | |
869176a0 | 2596 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask, 0); |
db64fe02 NP |
2597 | if (IS_ERR(va)) { |
2598 | kfree(area); | |
2599 | return NULL; | |
1da177e4 | 2600 | } |
1da177e4 | 2601 | |
d98c9e83 | 2602 | setup_vmalloc_vm(area, va, flags, caller); |
3c5c3cfb | 2603 | |
19f1c3ac AK |
2604 | /* |
2605 | * Mark pages for non-VM_ALLOC mappings as accessible. Do it now as a | |
2606 | * best-effort approach, as they can be mapped outside of vmalloc code. | |
2607 | * For VM_ALLOC mappings, the pages are marked as accessible after | |
2608 | * getting mapped in __vmalloc_node_range(). | |
23689e91 AK |
2609 | * With hardware tag-based KASAN, marking is skipped for |
2610 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). | |
19f1c3ac AK |
2611 | */ |
2612 | if (!(flags & VM_ALLOC)) | |
23689e91 | 2613 | area->addr = kasan_unpoison_vmalloc(area->addr, requested_size, |
f6e39794 | 2614 | KASAN_VMALLOC_PROT_NORMAL); |
1d96320f | 2615 | |
1da177e4 | 2616 | return area; |
1da177e4 LT |
2617 | } |
2618 | ||
c2968612 BH |
2619 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2620 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2621 | const void *caller) |
c2968612 | 2622 | { |
7ca3027b DA |
2623 | return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, start, end, |
2624 | NUMA_NO_NODE, GFP_KERNEL, caller); | |
c2968612 BH |
2625 | } |
2626 | ||
1da177e4 | 2627 | /** |
92eac168 MR |
2628 | * get_vm_area - reserve a contiguous kernel virtual area |
2629 | * @size: size of the area | |
2630 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2631 | * |
92eac168 MR |
2632 | * Search an area of @size in the kernel virtual mapping area, |
2633 | * and reserved it for out purposes. Returns the area descriptor | |
2634 | * on success or %NULL on failure. | |
a862f68a MR |
2635 | * |
2636 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2637 | */ |
2638 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2639 | { | |
7ca3027b DA |
2640 | return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, |
2641 | VMALLOC_START, VMALLOC_END, | |
00ef2d2f DR |
2642 | NUMA_NO_NODE, GFP_KERNEL, |
2643 | __builtin_return_address(0)); | |
23016969 CL |
2644 | } |
2645 | ||
2646 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2647 | const void *caller) |
23016969 | 2648 | { |
7ca3027b DA |
2649 | return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, |
2650 | VMALLOC_START, VMALLOC_END, | |
00ef2d2f | 2651 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2652 | } |
2653 | ||
e9da6e99 | 2654 | /** |
92eac168 MR |
2655 | * find_vm_area - find a continuous kernel virtual area |
2656 | * @addr: base address | |
e9da6e99 | 2657 | * |
92eac168 MR |
2658 | * Search for the kernel VM area starting at @addr, and return it. |
2659 | * It is up to the caller to do all required locking to keep the returned | |
2660 | * pointer valid. | |
a862f68a | 2661 | * |
74640617 | 2662 | * Return: the area descriptor on success or %NULL on failure. |
e9da6e99 MS |
2663 | */ |
2664 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2665 | { |
db64fe02 | 2666 | struct vmap_area *va; |
83342314 | 2667 | |
db64fe02 | 2668 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2669 | if (!va) |
2670 | return NULL; | |
1da177e4 | 2671 | |
688fcbfc | 2672 | return va->vm; |
1da177e4 LT |
2673 | } |
2674 | ||
7856dfeb | 2675 | /** |
92eac168 MR |
2676 | * remove_vm_area - find and remove a continuous kernel virtual area |
2677 | * @addr: base address | |
7856dfeb | 2678 | * |
92eac168 MR |
2679 | * Search for the kernel VM area starting at @addr, and remove it. |
2680 | * This function returns the found VM area, but using it is NOT safe | |
2681 | * on SMP machines, except for its size or flags. | |
a862f68a | 2682 | * |
74640617 | 2683 | * Return: the area descriptor on success or %NULL on failure. |
7856dfeb | 2684 | */ |
b3bdda02 | 2685 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2686 | { |
db64fe02 | 2687 | struct vmap_area *va; |
75c59ce7 | 2688 | struct vm_struct *vm; |
db64fe02 | 2689 | |
5803ed29 CH |
2690 | might_sleep(); |
2691 | ||
17d3ef43 CH |
2692 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
2693 | addr)) | |
2694 | return NULL; | |
c69480ad | 2695 | |
75c59ce7 CH |
2696 | va = find_unlink_vmap_area((unsigned long)addr); |
2697 | if (!va || !va->vm) | |
2698 | return NULL; | |
2699 | vm = va->vm; | |
dd32c279 | 2700 | |
17d3ef43 CH |
2701 | debug_check_no_locks_freed(vm->addr, get_vm_area_size(vm)); |
2702 | debug_check_no_obj_freed(vm->addr, get_vm_area_size(vm)); | |
75c59ce7 | 2703 | kasan_free_module_shadow(vm); |
17d3ef43 | 2704 | kasan_poison_vmalloc(vm->addr, get_vm_area_size(vm)); |
dd3b8353 | 2705 | |
75c59ce7 CH |
2706 | free_unmap_vmap_area(va); |
2707 | return vm; | |
7856dfeb AK |
2708 | } |
2709 | ||
868b104d RE |
2710 | static inline void set_area_direct_map(const struct vm_struct *area, |
2711 | int (*set_direct_map)(struct page *page)) | |
2712 | { | |
2713 | int i; | |
2714 | ||
121e6f32 | 2715 | /* HUGE_VMALLOC passes small pages to set_direct_map */ |
868b104d RE |
2716 | for (i = 0; i < area->nr_pages; i++) |
2717 | if (page_address(area->pages[i])) | |
2718 | set_direct_map(area->pages[i]); | |
2719 | } | |
2720 | ||
9e5fa0ae CH |
2721 | /* |
2722 | * Flush the vm mapping and reset the direct map. | |
2723 | */ | |
2724 | static void vm_reset_perms(struct vm_struct *area) | |
868b104d | 2725 | { |
868b104d | 2726 | unsigned long start = ULONG_MAX, end = 0; |
121e6f32 | 2727 | unsigned int page_order = vm_area_page_order(area); |
31e67340 | 2728 | int flush_dmap = 0; |
868b104d RE |
2729 | int i; |
2730 | ||
868b104d | 2731 | /* |
9e5fa0ae | 2732 | * Find the start and end range of the direct mappings to make sure that |
868b104d RE |
2733 | * the vm_unmap_aliases() flush includes the direct map. |
2734 | */ | |
121e6f32 | 2735 | for (i = 0; i < area->nr_pages; i += 1U << page_order) { |
8e41f872 | 2736 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
9e5fa0ae | 2737 | |
8e41f872 | 2738 | if (addr) { |
121e6f32 NP |
2739 | unsigned long page_size; |
2740 | ||
2741 | page_size = PAGE_SIZE << page_order; | |
868b104d | 2742 | start = min(addr, start); |
121e6f32 | 2743 | end = max(addr + page_size, end); |
31e67340 | 2744 | flush_dmap = 1; |
868b104d RE |
2745 | } |
2746 | } | |
2747 | ||
2748 | /* | |
2749 | * Set direct map to something invalid so that it won't be cached if | |
2750 | * there are any accesses after the TLB flush, then flush the TLB and | |
2751 | * reset the direct map permissions to the default. | |
2752 | */ | |
2753 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2754 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2755 | set_area_direct_map(area, set_direct_map_default_noflush); |
2756 | } | |
2757 | ||
208162f4 | 2758 | static void delayed_vfree_work(struct work_struct *w) |
1da177e4 | 2759 | { |
208162f4 CH |
2760 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); |
2761 | struct llist_node *t, *llnode; | |
bf22e37a | 2762 | |
208162f4 | 2763 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) |
5d3d31d6 | 2764 | vfree(llnode); |
bf22e37a AR |
2765 | } |
2766 | ||
2767 | /** | |
92eac168 MR |
2768 | * vfree_atomic - release memory allocated by vmalloc() |
2769 | * @addr: memory base address | |
bf22e37a | 2770 | * |
92eac168 MR |
2771 | * This one is just like vfree() but can be called in any atomic context |
2772 | * except NMIs. | |
bf22e37a AR |
2773 | */ |
2774 | void vfree_atomic(const void *addr) | |
2775 | { | |
01e2e839 | 2776 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); |
bf22e37a | 2777 | |
01e2e839 | 2778 | BUG_ON(in_nmi()); |
bf22e37a AR |
2779 | kmemleak_free(addr); |
2780 | ||
01e2e839 CH |
2781 | /* |
2782 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2783 | * context. Preemption is absolutely fine here, because the llist_add() | |
2784 | * implementation is lockless, so it works even if we are adding to | |
2785 | * another cpu's list. schedule_work() should be fine with this too. | |
2786 | */ | |
2787 | if (addr && llist_add((struct llist_node *)addr, &p->list)) | |
2788 | schedule_work(&p->wq); | |
c67dc624 RP |
2789 | } |
2790 | ||
1da177e4 | 2791 | /** |
fa307474 MWO |
2792 | * vfree - Release memory allocated by vmalloc() |
2793 | * @addr: Memory base address | |
1da177e4 | 2794 | * |
fa307474 MWO |
2795 | * Free the virtually continuous memory area starting at @addr, as obtained |
2796 | * from one of the vmalloc() family of APIs. This will usually also free the | |
2797 | * physical memory underlying the virtual allocation, but that memory is | |
2798 | * reference counted, so it will not be freed until the last user goes away. | |
1da177e4 | 2799 | * |
fa307474 | 2800 | * If @addr is NULL, no operation is performed. |
c9fcee51 | 2801 | * |
fa307474 | 2802 | * Context: |
92eac168 | 2803 | * May sleep if called *not* from interrupt context. |
fa307474 MWO |
2804 | * Must not be called in NMI context (strictly speaking, it could be |
2805 | * if we have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
f0953a1b | 2806 | * conventions for vfree() arch-dependent would be a really bad idea). |
1da177e4 | 2807 | */ |
b3bdda02 | 2808 | void vfree(const void *addr) |
1da177e4 | 2809 | { |
79311c1f CH |
2810 | struct vm_struct *vm; |
2811 | int i; | |
89219d37 | 2812 | |
01e2e839 CH |
2813 | if (unlikely(in_interrupt())) { |
2814 | vfree_atomic(addr); | |
2815 | return; | |
2816 | } | |
89219d37 | 2817 | |
01e2e839 | 2818 | BUG_ON(in_nmi()); |
89219d37 | 2819 | kmemleak_free(addr); |
01e2e839 | 2820 | might_sleep(); |
a8dda165 | 2821 | |
32fcfd40 AV |
2822 | if (!addr) |
2823 | return; | |
c67dc624 | 2824 | |
79311c1f CH |
2825 | vm = remove_vm_area(addr); |
2826 | if (unlikely(!vm)) { | |
2827 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", | |
2828 | addr); | |
2829 | return; | |
2830 | } | |
2831 | ||
9e5fa0ae CH |
2832 | if (unlikely(vm->flags & VM_FLUSH_RESET_PERMS)) |
2833 | vm_reset_perms(vm); | |
79311c1f CH |
2834 | for (i = 0; i < vm->nr_pages; i++) { |
2835 | struct page *page = vm->pages[i]; | |
2836 | ||
2837 | BUG_ON(!page); | |
2838 | mod_memcg_page_state(page, MEMCG_VMALLOC, -1); | |
2839 | /* | |
2840 | * High-order allocs for huge vmallocs are split, so | |
2841 | * can be freed as an array of order-0 allocations | |
2842 | */ | |
dcc1be11 | 2843 | __free_page(page); |
79311c1f CH |
2844 | cond_resched(); |
2845 | } | |
2846 | atomic_long_sub(vm->nr_pages, &nr_vmalloc_pages); | |
2847 | kvfree(vm->pages); | |
2848 | kfree(vm); | |
1da177e4 | 2849 | } |
1da177e4 LT |
2850 | EXPORT_SYMBOL(vfree); |
2851 | ||
2852 | /** | |
92eac168 MR |
2853 | * vunmap - release virtual mapping obtained by vmap() |
2854 | * @addr: memory base address | |
1da177e4 | 2855 | * |
92eac168 MR |
2856 | * Free the virtually contiguous memory area starting at @addr, |
2857 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2858 | * |
92eac168 | 2859 | * Must not be called in interrupt context. |
1da177e4 | 2860 | */ |
b3bdda02 | 2861 | void vunmap(const void *addr) |
1da177e4 | 2862 | { |
79311c1f CH |
2863 | struct vm_struct *vm; |
2864 | ||
1da177e4 | 2865 | BUG_ON(in_interrupt()); |
34754b69 | 2866 | might_sleep(); |
79311c1f CH |
2867 | |
2868 | if (!addr) | |
2869 | return; | |
2870 | vm = remove_vm_area(addr); | |
2871 | if (unlikely(!vm)) { | |
2872 | WARN(1, KERN_ERR "Trying to vunmap() nonexistent vm area (%p)\n", | |
2873 | addr); | |
2874 | return; | |
2875 | } | |
2876 | kfree(vm); | |
1da177e4 | 2877 | } |
1da177e4 LT |
2878 | EXPORT_SYMBOL(vunmap); |
2879 | ||
2880 | /** | |
92eac168 MR |
2881 | * vmap - map an array of pages into virtually contiguous space |
2882 | * @pages: array of page pointers | |
2883 | * @count: number of pages to map | |
2884 | * @flags: vm_area->flags | |
2885 | * @prot: page protection for the mapping | |
2886 | * | |
b944afc9 CH |
2887 | * Maps @count pages from @pages into contiguous kernel virtual space. |
2888 | * If @flags contains %VM_MAP_PUT_PAGES the ownership of the pages array itself | |
2889 | * (which must be kmalloc or vmalloc memory) and one reference per pages in it | |
2890 | * are transferred from the caller to vmap(), and will be freed / dropped when | |
2891 | * vfree() is called on the return value. | |
a862f68a MR |
2892 | * |
2893 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2894 | */ |
2895 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2896 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2897 | { |
2898 | struct vm_struct *area; | |
b67177ec | 2899 | unsigned long addr; |
65ee03c4 | 2900 | unsigned long size; /* In bytes */ |
1da177e4 | 2901 | |
34754b69 PZ |
2902 | might_sleep(); |
2903 | ||
37f3605e CH |
2904 | if (WARN_ON_ONCE(flags & VM_FLUSH_RESET_PERMS)) |
2905 | return NULL; | |
2906 | ||
bd1a8fb2 PZ |
2907 | /* |
2908 | * Your top guard is someone else's bottom guard. Not having a top | |
2909 | * guard compromises someone else's mappings too. | |
2910 | */ | |
2911 | if (WARN_ON_ONCE(flags & VM_NO_GUARD)) | |
2912 | flags &= ~VM_NO_GUARD; | |
2913 | ||
ca79b0c2 | 2914 | if (count > totalram_pages()) |
1da177e4 LT |
2915 | return NULL; |
2916 | ||
65ee03c4 GJM |
2917 | size = (unsigned long)count << PAGE_SHIFT; |
2918 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2919 | if (!area) |
2920 | return NULL; | |
23016969 | 2921 | |
b67177ec NP |
2922 | addr = (unsigned long)area->addr; |
2923 | if (vmap_pages_range(addr, addr + size, pgprot_nx(prot), | |
2924 | pages, PAGE_SHIFT) < 0) { | |
1da177e4 LT |
2925 | vunmap(area->addr); |
2926 | return NULL; | |
2927 | } | |
2928 | ||
c22ee528 | 2929 | if (flags & VM_MAP_PUT_PAGES) { |
b944afc9 | 2930 | area->pages = pages; |
c22ee528 ML |
2931 | area->nr_pages = count; |
2932 | } | |
1da177e4 LT |
2933 | return area->addr; |
2934 | } | |
1da177e4 LT |
2935 | EXPORT_SYMBOL(vmap); |
2936 | ||
3e9a9e25 CH |
2937 | #ifdef CONFIG_VMAP_PFN |
2938 | struct vmap_pfn_data { | |
2939 | unsigned long *pfns; | |
2940 | pgprot_t prot; | |
2941 | unsigned int idx; | |
2942 | }; | |
2943 | ||
2944 | static int vmap_pfn_apply(pte_t *pte, unsigned long addr, void *private) | |
2945 | { | |
2946 | struct vmap_pfn_data *data = private; | |
2947 | ||
2948 | if (WARN_ON_ONCE(pfn_valid(data->pfns[data->idx]))) | |
2949 | return -EINVAL; | |
2950 | *pte = pte_mkspecial(pfn_pte(data->pfns[data->idx++], data->prot)); | |
2951 | return 0; | |
2952 | } | |
2953 | ||
2954 | /** | |
2955 | * vmap_pfn - map an array of PFNs into virtually contiguous space | |
2956 | * @pfns: array of PFNs | |
2957 | * @count: number of pages to map | |
2958 | * @prot: page protection for the mapping | |
2959 | * | |
2960 | * Maps @count PFNs from @pfns into contiguous kernel virtual space and returns | |
2961 | * the start address of the mapping. | |
2962 | */ | |
2963 | void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot) | |
2964 | { | |
2965 | struct vmap_pfn_data data = { .pfns = pfns, .prot = pgprot_nx(prot) }; | |
2966 | struct vm_struct *area; | |
2967 | ||
2968 | area = get_vm_area_caller(count * PAGE_SIZE, VM_IOREMAP, | |
2969 | __builtin_return_address(0)); | |
2970 | if (!area) | |
2971 | return NULL; | |
2972 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
2973 | count * PAGE_SIZE, vmap_pfn_apply, &data)) { | |
2974 | free_vm_area(area); | |
2975 | return NULL; | |
2976 | } | |
2977 | return area->addr; | |
2978 | } | |
2979 | EXPORT_SYMBOL_GPL(vmap_pfn); | |
2980 | #endif /* CONFIG_VMAP_PFN */ | |
2981 | ||
12b9f873 UR |
2982 | static inline unsigned int |
2983 | vm_area_alloc_pages(gfp_t gfp, int nid, | |
343ab817 | 2984 | unsigned int order, unsigned int nr_pages, struct page **pages) |
12b9f873 UR |
2985 | { |
2986 | unsigned int nr_allocated = 0; | |
e9c3cda4 MH |
2987 | gfp_t alloc_gfp = gfp; |
2988 | bool nofail = false; | |
ffb29b1c CW |
2989 | struct page *page; |
2990 | int i; | |
12b9f873 UR |
2991 | |
2992 | /* | |
2993 | * For order-0 pages we make use of bulk allocator, if | |
2994 | * the page array is partly or not at all populated due | |
2995 | * to fails, fallback to a single page allocator that is | |
2996 | * more permissive. | |
2997 | */ | |
c00b6b96 | 2998 | if (!order) { |
e9c3cda4 | 2999 | /* bulk allocator doesn't support nofail req. officially */ |
9376130c MH |
3000 | gfp_t bulk_gfp = gfp & ~__GFP_NOFAIL; |
3001 | ||
343ab817 URS |
3002 | while (nr_allocated < nr_pages) { |
3003 | unsigned int nr, nr_pages_request; | |
3004 | ||
3005 | /* | |
3006 | * A maximum allowed request is hard-coded and is 100 | |
3007 | * pages per call. That is done in order to prevent a | |
3008 | * long preemption off scenario in the bulk-allocator | |
3009 | * so the range is [1:100]. | |
3010 | */ | |
3011 | nr_pages_request = min(100U, nr_pages - nr_allocated); | |
3012 | ||
c00b6b96 CW |
3013 | /* memory allocation should consider mempolicy, we can't |
3014 | * wrongly use nearest node when nid == NUMA_NO_NODE, | |
3015 | * otherwise memory may be allocated in only one node, | |
98af39d5 | 3016 | * but mempolicy wants to alloc memory by interleaving. |
c00b6b96 CW |
3017 | */ |
3018 | if (IS_ENABLED(CONFIG_NUMA) && nid == NUMA_NO_NODE) | |
9376130c | 3019 | nr = alloc_pages_bulk_array_mempolicy(bulk_gfp, |
c00b6b96 CW |
3020 | nr_pages_request, |
3021 | pages + nr_allocated); | |
3022 | ||
3023 | else | |
9376130c | 3024 | nr = alloc_pages_bulk_array_node(bulk_gfp, nid, |
c00b6b96 CW |
3025 | nr_pages_request, |
3026 | pages + nr_allocated); | |
343ab817 URS |
3027 | |
3028 | nr_allocated += nr; | |
3029 | cond_resched(); | |
3030 | ||
3031 | /* | |
3032 | * If zero or pages were obtained partly, | |
3033 | * fallback to a single page allocator. | |
3034 | */ | |
3035 | if (nr != nr_pages_request) | |
3036 | break; | |
3037 | } | |
e9c3cda4 MH |
3038 | } else if (gfp & __GFP_NOFAIL) { |
3039 | /* | |
3040 | * Higher order nofail allocations are really expensive and | |
3041 | * potentially dangerous (pre-mature OOM, disruptive reclaim | |
3042 | * and compaction etc. | |
3043 | */ | |
3044 | alloc_gfp &= ~__GFP_NOFAIL; | |
3045 | nofail = true; | |
3b8000ae | 3046 | } |
12b9f873 UR |
3047 | |
3048 | /* High-order pages or fallback path if "bulk" fails. */ | |
ffb29b1c | 3049 | while (nr_allocated < nr_pages) { |
dd544141 VA |
3050 | if (fatal_signal_pending(current)) |
3051 | break; | |
3052 | ||
ffb29b1c | 3053 | if (nid == NUMA_NO_NODE) |
e9c3cda4 | 3054 | page = alloc_pages(alloc_gfp, order); |
ffb29b1c | 3055 | else |
e9c3cda4 MH |
3056 | page = alloc_pages_node(nid, alloc_gfp, order); |
3057 | if (unlikely(!page)) { | |
3058 | if (!nofail) | |
3059 | break; | |
3060 | ||
3061 | /* fall back to the zero order allocations */ | |
3062 | alloc_gfp |= __GFP_NOFAIL; | |
3063 | order = 0; | |
3064 | continue; | |
3065 | } | |
3066 | ||
3b8000ae NP |
3067 | /* |
3068 | * Higher order allocations must be able to be treated as | |
3069 | * indepdenent small pages by callers (as they can with | |
3070 | * small-page vmallocs). Some drivers do their own refcounting | |
3071 | * on vmalloc_to_page() pages, some use page->mapping, | |
3072 | * page->lru, etc. | |
3073 | */ | |
3074 | if (order) | |
3075 | split_page(page, order); | |
12b9f873 UR |
3076 | |
3077 | /* | |
3078 | * Careful, we allocate and map page-order pages, but | |
3079 | * tracking is done per PAGE_SIZE page so as to keep the | |
3080 | * vm_struct APIs independent of the physical/mapped size. | |
3081 | */ | |
3082 | for (i = 0; i < (1U << order); i++) | |
3083 | pages[nr_allocated + i] = page + i; | |
3084 | ||
12e376a6 | 3085 | cond_resched(); |
12b9f873 UR |
3086 | nr_allocated += 1U << order; |
3087 | } | |
3088 | ||
3089 | return nr_allocated; | |
3090 | } | |
3091 | ||
e31d9eb5 | 3092 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
121e6f32 NP |
3093 | pgprot_t prot, unsigned int page_shift, |
3094 | int node) | |
1da177e4 | 3095 | { |
930f036b | 3096 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
9376130c | 3097 | bool nofail = gfp_mask & __GFP_NOFAIL; |
121e6f32 NP |
3098 | unsigned long addr = (unsigned long)area->addr; |
3099 | unsigned long size = get_vm_area_size(area); | |
34fe6537 | 3100 | unsigned long array_size; |
121e6f32 NP |
3101 | unsigned int nr_small_pages = size >> PAGE_SHIFT; |
3102 | unsigned int page_order; | |
451769eb MH |
3103 | unsigned int flags; |
3104 | int ret; | |
1da177e4 | 3105 | |
121e6f32 | 3106 | array_size = (unsigned long)nr_small_pages * sizeof(struct page *); |
80b1d8fd | 3107 | |
f255935b CH |
3108 | if (!(gfp_mask & (GFP_DMA | GFP_DMA32))) |
3109 | gfp_mask |= __GFP_HIGHMEM; | |
1da177e4 | 3110 | |
1da177e4 | 3111 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 3112 | if (array_size > PAGE_SIZE) { |
5c1f4e69 | 3113 | area->pages = __vmalloc_node(array_size, 1, nested_gfp, node, |
f255935b | 3114 | area->caller); |
286e1ea3 | 3115 | } else { |
5c1f4e69 | 3116 | area->pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 3117 | } |
7ea36242 | 3118 | |
5c1f4e69 | 3119 | if (!area->pages) { |
c3d77172 | 3120 | warn_alloc(gfp_mask, NULL, |
f4bdfeaf URS |
3121 | "vmalloc error: size %lu, failed to allocated page array size %lu", |
3122 | nr_small_pages * PAGE_SIZE, array_size); | |
cd61413b | 3123 | free_vm_area(area); |
1da177e4 LT |
3124 | return NULL; |
3125 | } | |
1da177e4 | 3126 | |
121e6f32 | 3127 | set_vm_area_page_order(area, page_shift - PAGE_SHIFT); |
121e6f32 | 3128 | page_order = vm_area_page_order(area); |
bf53d6f8 | 3129 | |
c3d77172 URS |
3130 | area->nr_pages = vm_area_alloc_pages(gfp_mask | __GFP_NOWARN, |
3131 | node, page_order, nr_small_pages, area->pages); | |
5c1f4e69 | 3132 | |
97105f0a | 3133 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
4e5aa1f4 | 3134 | if (gfp_mask & __GFP_ACCOUNT) { |
3b8000ae | 3135 | int i; |
4e5aa1f4 | 3136 | |
3b8000ae NP |
3137 | for (i = 0; i < area->nr_pages; i++) |
3138 | mod_memcg_page_state(area->pages[i], MEMCG_VMALLOC, 1); | |
4e5aa1f4 | 3139 | } |
1da177e4 | 3140 | |
5c1f4e69 URS |
3141 | /* |
3142 | * If not enough pages were obtained to accomplish an | |
f41f036b | 3143 | * allocation request, free them via vfree() if any. |
5c1f4e69 URS |
3144 | */ |
3145 | if (area->nr_pages != nr_small_pages) { | |
f349b15e YS |
3146 | /* vm_area_alloc_pages() can also fail due to a fatal signal */ |
3147 | if (!fatal_signal_pending(current)) | |
3148 | warn_alloc(gfp_mask, NULL, | |
3149 | "vmalloc error: size %lu, page order %u, failed to allocate pages", | |
3150 | area->nr_pages * PAGE_SIZE, page_order); | |
5c1f4e69 URS |
3151 | goto fail; |
3152 | } | |
3153 | ||
451769eb MH |
3154 | /* |
3155 | * page tables allocations ignore external gfp mask, enforce it | |
3156 | * by the scope API | |
3157 | */ | |
3158 | if ((gfp_mask & (__GFP_FS | __GFP_IO)) == __GFP_IO) | |
3159 | flags = memalloc_nofs_save(); | |
3160 | else if ((gfp_mask & (__GFP_FS | __GFP_IO)) == 0) | |
3161 | flags = memalloc_noio_save(); | |
3162 | ||
9376130c MH |
3163 | do { |
3164 | ret = vmap_pages_range(addr, addr + size, prot, area->pages, | |
451769eb | 3165 | page_shift); |
9376130c MH |
3166 | if (nofail && (ret < 0)) |
3167 | schedule_timeout_uninterruptible(1); | |
3168 | } while (nofail && (ret < 0)); | |
451769eb MH |
3169 | |
3170 | if ((gfp_mask & (__GFP_FS | __GFP_IO)) == __GFP_IO) | |
3171 | memalloc_nofs_restore(flags); | |
3172 | else if ((gfp_mask & (__GFP_FS | __GFP_IO)) == 0) | |
3173 | memalloc_noio_restore(flags); | |
3174 | ||
3175 | if (ret < 0) { | |
c3d77172 | 3176 | warn_alloc(gfp_mask, NULL, |
f4bdfeaf URS |
3177 | "vmalloc error: size %lu, failed to map pages", |
3178 | area->nr_pages * PAGE_SIZE); | |
1da177e4 | 3179 | goto fail; |
d70bec8c | 3180 | } |
ed1f324c | 3181 | |
1da177e4 LT |
3182 | return area->addr; |
3183 | ||
3184 | fail: | |
f41f036b | 3185 | vfree(area->addr); |
1da177e4 LT |
3186 | return NULL; |
3187 | } | |
3188 | ||
3189 | /** | |
92eac168 MR |
3190 | * __vmalloc_node_range - allocate virtually contiguous memory |
3191 | * @size: allocation size | |
3192 | * @align: desired alignment | |
3193 | * @start: vm area range start | |
3194 | * @end: vm area range end | |
3195 | * @gfp_mask: flags for the page level allocator | |
3196 | * @prot: protection mask for the allocated pages | |
3197 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
3198 | * @node: node to use for allocation or NUMA_NO_NODE | |
3199 | * @caller: caller's return address | |
3200 | * | |
3201 | * Allocate enough pages to cover @size from the page level | |
b7d90e7a | 3202 | * allocator with @gfp_mask flags. Please note that the full set of gfp |
30d3f011 MH |
3203 | * flags are not supported. GFP_KERNEL, GFP_NOFS and GFP_NOIO are all |
3204 | * supported. | |
3205 | * Zone modifiers are not supported. From the reclaim modifiers | |
3206 | * __GFP_DIRECT_RECLAIM is required (aka GFP_NOWAIT is not supported) | |
3207 | * and only __GFP_NOFAIL is supported (i.e. __GFP_NORETRY and | |
3208 | * __GFP_RETRY_MAYFAIL are not supported). | |
3209 | * | |
3210 | * __GFP_NOWARN can be used to suppress failures messages. | |
b7d90e7a MH |
3211 | * |
3212 | * Map them into contiguous kernel virtual space, using a pagetable | |
3213 | * protection of @prot. | |
a862f68a MR |
3214 | * |
3215 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 3216 | */ |
d0a21265 DR |
3217 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
3218 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
3219 | pgprot_t prot, unsigned long vm_flags, int node, |
3220 | const void *caller) | |
1da177e4 LT |
3221 | { |
3222 | struct vm_struct *area; | |
19f1c3ac | 3223 | void *ret; |
f6e39794 | 3224 | kasan_vmalloc_flags_t kasan_flags = KASAN_VMALLOC_NONE; |
89219d37 | 3225 | unsigned long real_size = size; |
121e6f32 NP |
3226 | unsigned long real_align = align; |
3227 | unsigned int shift = PAGE_SHIFT; | |
1da177e4 | 3228 | |
d70bec8c NP |
3229 | if (WARN_ON_ONCE(!size)) |
3230 | return NULL; | |
3231 | ||
3232 | if ((size >> PAGE_SHIFT) > totalram_pages()) { | |
3233 | warn_alloc(gfp_mask, NULL, | |
f4bdfeaf URS |
3234 | "vmalloc error: size %lu, exceeds total pages", |
3235 | real_size); | |
d70bec8c | 3236 | return NULL; |
121e6f32 NP |
3237 | } |
3238 | ||
559089e0 | 3239 | if (vmap_allow_huge && (vm_flags & VM_ALLOW_HUGE_VMAP)) { |
121e6f32 | 3240 | unsigned long size_per_node; |
1da177e4 | 3241 | |
121e6f32 NP |
3242 | /* |
3243 | * Try huge pages. Only try for PAGE_KERNEL allocations, | |
3244 | * others like modules don't yet expect huge pages in | |
3245 | * their allocations due to apply_to_page_range not | |
3246 | * supporting them. | |
3247 | */ | |
3248 | ||
3249 | size_per_node = size; | |
3250 | if (node == NUMA_NO_NODE) | |
3251 | size_per_node /= num_online_nodes(); | |
3382bbee | 3252 | if (arch_vmap_pmd_supported(prot) && size_per_node >= PMD_SIZE) |
121e6f32 | 3253 | shift = PMD_SHIFT; |
3382bbee CL |
3254 | else |
3255 | shift = arch_vmap_pte_supported_shift(size_per_node); | |
3256 | ||
3257 | align = max(real_align, 1UL << shift); | |
3258 | size = ALIGN(real_size, 1UL << shift); | |
121e6f32 NP |
3259 | } |
3260 | ||
3261 | again: | |
7ca3027b DA |
3262 | area = __get_vm_area_node(real_size, align, shift, VM_ALLOC | |
3263 | VM_UNINITIALIZED | vm_flags, start, end, node, | |
3264 | gfp_mask, caller); | |
d70bec8c | 3265 | if (!area) { |
9376130c | 3266 | bool nofail = gfp_mask & __GFP_NOFAIL; |
d70bec8c | 3267 | warn_alloc(gfp_mask, NULL, |
9376130c MH |
3268 | "vmalloc error: size %lu, vm_struct allocation failed%s", |
3269 | real_size, (nofail) ? ". Retrying." : ""); | |
3270 | if (nofail) { | |
3271 | schedule_timeout_uninterruptible(1); | |
3272 | goto again; | |
3273 | } | |
de7d2b56 | 3274 | goto fail; |
d70bec8c | 3275 | } |
1da177e4 | 3276 | |
f6e39794 AK |
3277 | /* |
3278 | * Prepare arguments for __vmalloc_area_node() and | |
3279 | * kasan_unpoison_vmalloc(). | |
3280 | */ | |
3281 | if (pgprot_val(prot) == pgprot_val(PAGE_KERNEL)) { | |
3282 | if (kasan_hw_tags_enabled()) { | |
3283 | /* | |
3284 | * Modify protection bits to allow tagging. | |
3285 | * This must be done before mapping. | |
3286 | */ | |
3287 | prot = arch_vmap_pgprot_tagged(prot); | |
01d92c7f | 3288 | |
f6e39794 AK |
3289 | /* |
3290 | * Skip page_alloc poisoning and zeroing for physical | |
3291 | * pages backing VM_ALLOC mapping. Memory is instead | |
3292 | * poisoned and zeroed by kasan_unpoison_vmalloc(). | |
3293 | */ | |
0a54864f | 3294 | gfp_mask |= __GFP_SKIP_KASAN | __GFP_SKIP_ZERO; |
f6e39794 AK |
3295 | } |
3296 | ||
3297 | /* Take note that the mapping is PAGE_KERNEL. */ | |
3298 | kasan_flags |= KASAN_VMALLOC_PROT_NORMAL; | |
23689e91 AK |
3299 | } |
3300 | ||
01d92c7f | 3301 | /* Allocate physical pages and map them into vmalloc space. */ |
19f1c3ac AK |
3302 | ret = __vmalloc_area_node(area, gfp_mask, prot, shift, node); |
3303 | if (!ret) | |
121e6f32 | 3304 | goto fail; |
89219d37 | 3305 | |
23689e91 AK |
3306 | /* |
3307 | * Mark the pages as accessible, now that they are mapped. | |
6c2f761d AK |
3308 | * The condition for setting KASAN_VMALLOC_INIT should complement the |
3309 | * one in post_alloc_hook() with regards to the __GFP_SKIP_ZERO check | |
3310 | * to make sure that memory is initialized under the same conditions. | |
f6e39794 AK |
3311 | * Tag-based KASAN modes only assign tags to normal non-executable |
3312 | * allocations, see __kasan_unpoison_vmalloc(). | |
23689e91 | 3313 | */ |
f6e39794 | 3314 | kasan_flags |= KASAN_VMALLOC_VM_ALLOC; |
6c2f761d AK |
3315 | if (!want_init_on_free() && want_init_on_alloc(gfp_mask) && |
3316 | (gfp_mask & __GFP_SKIP_ZERO)) | |
23689e91 | 3317 | kasan_flags |= KASAN_VMALLOC_INIT; |
f6e39794 | 3318 | /* KASAN_VMALLOC_PROT_NORMAL already set if required. */ |
23689e91 | 3319 | area->addr = kasan_unpoison_vmalloc(area->addr, real_size, kasan_flags); |
19f1c3ac | 3320 | |
f5252e00 | 3321 | /* |
20fc02b4 ZY |
3322 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
3323 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 3324 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 3325 | */ |
20fc02b4 | 3326 | clear_vm_uninitialized_flag(area); |
f5252e00 | 3327 | |
7ca3027b | 3328 | size = PAGE_ALIGN(size); |
60115fa5 KW |
3329 | if (!(vm_flags & VM_DEFER_KMEMLEAK)) |
3330 | kmemleak_vmalloc(area, size, gfp_mask); | |
89219d37 | 3331 | |
19f1c3ac | 3332 | return area->addr; |
de7d2b56 JP |
3333 | |
3334 | fail: | |
121e6f32 NP |
3335 | if (shift > PAGE_SHIFT) { |
3336 | shift = PAGE_SHIFT; | |
3337 | align = real_align; | |
3338 | size = real_size; | |
3339 | goto again; | |
3340 | } | |
3341 | ||
de7d2b56 | 3342 | return NULL; |
1da177e4 LT |
3343 | } |
3344 | ||
d0a21265 | 3345 | /** |
92eac168 MR |
3346 | * __vmalloc_node - allocate virtually contiguous memory |
3347 | * @size: allocation size | |
3348 | * @align: desired alignment | |
3349 | * @gfp_mask: flags for the page level allocator | |
92eac168 MR |
3350 | * @node: node to use for allocation or NUMA_NO_NODE |
3351 | * @caller: caller's return address | |
a7c3e901 | 3352 | * |
f38fcb9c CH |
3353 | * Allocate enough pages to cover @size from the page level allocator with |
3354 | * @gfp_mask flags. Map them into contiguous kernel virtual space. | |
a7c3e901 | 3355 | * |
92eac168 MR |
3356 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
3357 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 3358 | * |
92eac168 MR |
3359 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
3360 | * with mm people. | |
a862f68a MR |
3361 | * |
3362 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 3363 | */ |
2b905948 | 3364 | void *__vmalloc_node(unsigned long size, unsigned long align, |
f38fcb9c | 3365 | gfp_t gfp_mask, int node, const void *caller) |
d0a21265 DR |
3366 | { |
3367 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
f38fcb9c | 3368 | gfp_mask, PAGE_KERNEL, 0, node, caller); |
d0a21265 | 3369 | } |
c3f896dc CH |
3370 | /* |
3371 | * This is only for performance analysis of vmalloc and stress purpose. | |
3372 | * It is required by vmalloc test module, therefore do not use it other | |
3373 | * than that. | |
3374 | */ | |
3375 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
3376 | EXPORT_SYMBOL_GPL(__vmalloc_node); | |
3377 | #endif | |
d0a21265 | 3378 | |
88dca4ca | 3379 | void *__vmalloc(unsigned long size, gfp_t gfp_mask) |
930fc45a | 3380 | { |
f38fcb9c | 3381 | return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE, |
23016969 | 3382 | __builtin_return_address(0)); |
930fc45a | 3383 | } |
1da177e4 LT |
3384 | EXPORT_SYMBOL(__vmalloc); |
3385 | ||
3386 | /** | |
92eac168 MR |
3387 | * vmalloc - allocate virtually contiguous memory |
3388 | * @size: allocation size | |
3389 | * | |
3390 | * Allocate enough pages to cover @size from the page level | |
3391 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 3392 | * |
92eac168 MR |
3393 | * For tight control over page level allocator and protection flags |
3394 | * use __vmalloc() instead. | |
a862f68a MR |
3395 | * |
3396 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
3397 | */ |
3398 | void *vmalloc(unsigned long size) | |
3399 | { | |
4d39d728 CH |
3400 | return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE, |
3401 | __builtin_return_address(0)); | |
1da177e4 | 3402 | } |
1da177e4 LT |
3403 | EXPORT_SYMBOL(vmalloc); |
3404 | ||
15a64f5a | 3405 | /** |
559089e0 SL |
3406 | * vmalloc_huge - allocate virtually contiguous memory, allow huge pages |
3407 | * @size: allocation size | |
3408 | * @gfp_mask: flags for the page level allocator | |
15a64f5a | 3409 | * |
559089e0 | 3410 | * Allocate enough pages to cover @size from the page level |
15a64f5a | 3411 | * allocator and map them into contiguous kernel virtual space. |
559089e0 SL |
3412 | * If @size is greater than or equal to PMD_SIZE, allow using |
3413 | * huge pages for the memory | |
15a64f5a CI |
3414 | * |
3415 | * Return: pointer to the allocated memory or %NULL on error | |
3416 | */ | |
559089e0 | 3417 | void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) |
15a64f5a CI |
3418 | { |
3419 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, | |
559089e0 | 3420 | gfp_mask, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP, |
15a64f5a CI |
3421 | NUMA_NO_NODE, __builtin_return_address(0)); |
3422 | } | |
559089e0 | 3423 | EXPORT_SYMBOL_GPL(vmalloc_huge); |
15a64f5a | 3424 | |
e1ca7788 | 3425 | /** |
92eac168 MR |
3426 | * vzalloc - allocate virtually contiguous memory with zero fill |
3427 | * @size: allocation size | |
3428 | * | |
3429 | * Allocate enough pages to cover @size from the page level | |
3430 | * allocator and map them into contiguous kernel virtual space. | |
3431 | * The memory allocated is set to zero. | |
3432 | * | |
3433 | * For tight control over page level allocator and protection flags | |
3434 | * use __vmalloc() instead. | |
a862f68a MR |
3435 | * |
3436 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
3437 | */ |
3438 | void *vzalloc(unsigned long size) | |
3439 | { | |
4d39d728 CH |
3440 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE, |
3441 | __builtin_return_address(0)); | |
e1ca7788 DY |
3442 | } |
3443 | EXPORT_SYMBOL(vzalloc); | |
3444 | ||
83342314 | 3445 | /** |
ead04089 REB |
3446 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
3447 | * @size: allocation size | |
83342314 | 3448 | * |
ead04089 REB |
3449 | * The resulting memory area is zeroed so it can be mapped to userspace |
3450 | * without leaking data. | |
a862f68a MR |
3451 | * |
3452 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
3453 | */ |
3454 | void *vmalloc_user(unsigned long size) | |
3455 | { | |
bc84c535 RP |
3456 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
3457 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
3458 | VM_USERMAP, NUMA_NO_NODE, | |
3459 | __builtin_return_address(0)); | |
83342314 NP |
3460 | } |
3461 | EXPORT_SYMBOL(vmalloc_user); | |
3462 | ||
930fc45a | 3463 | /** |
92eac168 MR |
3464 | * vmalloc_node - allocate memory on a specific node |
3465 | * @size: allocation size | |
3466 | * @node: numa node | |
930fc45a | 3467 | * |
92eac168 MR |
3468 | * Allocate enough pages to cover @size from the page level |
3469 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 3470 | * |
92eac168 MR |
3471 | * For tight control over page level allocator and protection flags |
3472 | * use __vmalloc() instead. | |
a862f68a MR |
3473 | * |
3474 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
3475 | */ |
3476 | void *vmalloc_node(unsigned long size, int node) | |
3477 | { | |
f38fcb9c CH |
3478 | return __vmalloc_node(size, 1, GFP_KERNEL, node, |
3479 | __builtin_return_address(0)); | |
930fc45a CL |
3480 | } |
3481 | EXPORT_SYMBOL(vmalloc_node); | |
3482 | ||
e1ca7788 DY |
3483 | /** |
3484 | * vzalloc_node - allocate memory on a specific node with zero fill | |
3485 | * @size: allocation size | |
3486 | * @node: numa node | |
3487 | * | |
3488 | * Allocate enough pages to cover @size from the page level | |
3489 | * allocator and map them into contiguous kernel virtual space. | |
3490 | * The memory allocated is set to zero. | |
3491 | * | |
a862f68a | 3492 | * Return: pointer to the allocated memory or %NULL on error |
e1ca7788 DY |
3493 | */ |
3494 | void *vzalloc_node(unsigned long size, int node) | |
3495 | { | |
4d39d728 CH |
3496 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node, |
3497 | __builtin_return_address(0)); | |
e1ca7788 DY |
3498 | } |
3499 | EXPORT_SYMBOL(vzalloc_node); | |
3500 | ||
0d08e0d3 | 3501 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 3502 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 3503 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 3504 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 3505 | #else |
698d0831 MH |
3506 | /* |
3507 | * 64b systems should always have either DMA or DMA32 zones. For others | |
3508 | * GFP_DMA32 should do the right thing and use the normal zone. | |
3509 | */ | |
68d68ff6 | 3510 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 AK |
3511 | #endif |
3512 | ||
1da177e4 | 3513 | /** |
92eac168 MR |
3514 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
3515 | * @size: allocation size | |
1da177e4 | 3516 | * |
92eac168 MR |
3517 | * Allocate enough 32bit PA addressable pages to cover @size from the |
3518 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
3519 | * |
3520 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
3521 | */ |
3522 | void *vmalloc_32(unsigned long size) | |
3523 | { | |
f38fcb9c CH |
3524 | return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE, |
3525 | __builtin_return_address(0)); | |
1da177e4 | 3526 | } |
1da177e4 LT |
3527 | EXPORT_SYMBOL(vmalloc_32); |
3528 | ||
83342314 | 3529 | /** |
ead04089 | 3530 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 3531 | * @size: allocation size |
ead04089 REB |
3532 | * |
3533 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
3534 | * mapped to userspace without leaking data. | |
a862f68a MR |
3535 | * |
3536 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
3537 | */ |
3538 | void *vmalloc_32_user(unsigned long size) | |
3539 | { | |
bc84c535 RP |
3540 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
3541 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
3542 | VM_USERMAP, NUMA_NO_NODE, | |
3543 | __builtin_return_address(0)); | |
83342314 NP |
3544 | } |
3545 | EXPORT_SYMBOL(vmalloc_32_user); | |
3546 | ||
d0107eb0 | 3547 | /* |
4c91c07c LS |
3548 | * Atomically zero bytes in the iterator. |
3549 | * | |
3550 | * Returns the number of zeroed bytes. | |
d0107eb0 | 3551 | */ |
4c91c07c LS |
3552 | static size_t zero_iter(struct iov_iter *iter, size_t count) |
3553 | { | |
3554 | size_t remains = count; | |
3555 | ||
3556 | while (remains > 0) { | |
3557 | size_t num, copied; | |
3558 | ||
3559 | num = remains < PAGE_SIZE ? remains : PAGE_SIZE; | |
3560 | copied = copy_page_to_iter_nofault(ZERO_PAGE(0), 0, num, iter); | |
3561 | remains -= copied; | |
3562 | ||
3563 | if (copied < num) | |
3564 | break; | |
3565 | } | |
d0107eb0 | 3566 | |
4c91c07c LS |
3567 | return count - remains; |
3568 | } | |
3569 | ||
3570 | /* | |
3571 | * small helper routine, copy contents to iter from addr. | |
3572 | * If the page is not present, fill zero. | |
3573 | * | |
3574 | * Returns the number of copied bytes. | |
3575 | */ | |
3576 | static size_t aligned_vread_iter(struct iov_iter *iter, | |
3577 | const char *addr, size_t count) | |
d0107eb0 | 3578 | { |
4c91c07c LS |
3579 | size_t remains = count; |
3580 | struct page *page; | |
d0107eb0 | 3581 | |
4c91c07c | 3582 | while (remains > 0) { |
d0107eb0 | 3583 | unsigned long offset, length; |
4c91c07c | 3584 | size_t copied = 0; |
d0107eb0 | 3585 | |
891c49ab | 3586 | offset = offset_in_page(addr); |
d0107eb0 | 3587 | length = PAGE_SIZE - offset; |
4c91c07c LS |
3588 | if (length > remains) |
3589 | length = remains; | |
3590 | page = vmalloc_to_page(addr); | |
d0107eb0 | 3591 | /* |
4c91c07c LS |
3592 | * To do safe access to this _mapped_ area, we need lock. But |
3593 | * adding lock here means that we need to add overhead of | |
3594 | * vmalloc()/vfree() calls for this _debug_ interface, rarely | |
3595 | * used. Instead of that, we'll use an local mapping via | |
3596 | * copy_page_to_iter_nofault() and accept a small overhead in | |
3597 | * this access function. | |
d0107eb0 | 3598 | */ |
4c91c07c LS |
3599 | if (page) |
3600 | copied = copy_page_to_iter_nofault(page, offset, | |
3601 | length, iter); | |
3602 | else | |
3603 | copied = zero_iter(iter, length); | |
d0107eb0 | 3604 | |
4c91c07c LS |
3605 | addr += copied; |
3606 | remains -= copied; | |
3607 | ||
3608 | if (copied != length) | |
3609 | break; | |
d0107eb0 | 3610 | } |
4c91c07c LS |
3611 | |
3612 | return count - remains; | |
d0107eb0 KH |
3613 | } |
3614 | ||
4c91c07c LS |
3615 | /* |
3616 | * Read from a vm_map_ram region of memory. | |
3617 | * | |
3618 | * Returns the number of copied bytes. | |
3619 | */ | |
3620 | static size_t vmap_ram_vread_iter(struct iov_iter *iter, const char *addr, | |
3621 | size_t count, unsigned long flags) | |
06c89946 BH |
3622 | { |
3623 | char *start; | |
3624 | struct vmap_block *vb; | |
062eacf5 | 3625 | struct xarray *xa; |
06c89946 | 3626 | unsigned long offset; |
4c91c07c LS |
3627 | unsigned int rs, re; |
3628 | size_t remains, n; | |
06c89946 BH |
3629 | |
3630 | /* | |
3631 | * If it's area created by vm_map_ram() interface directly, but | |
3632 | * not further subdividing and delegating management to vmap_block, | |
3633 | * handle it here. | |
3634 | */ | |
4c91c07c LS |
3635 | if (!(flags & VMAP_BLOCK)) |
3636 | return aligned_vread_iter(iter, addr, count); | |
3637 | ||
3638 | remains = count; | |
06c89946 BH |
3639 | |
3640 | /* | |
3641 | * Area is split into regions and tracked with vmap_block, read out | |
3642 | * each region and zero fill the hole between regions. | |
3643 | */ | |
fa1c77c1 | 3644 | xa = addr_to_vb_xa((unsigned long) addr); |
062eacf5 | 3645 | vb = xa_load(xa, addr_to_vb_idx((unsigned long)addr)); |
06c89946 | 3646 | if (!vb) |
4c91c07c | 3647 | goto finished_zero; |
06c89946 BH |
3648 | |
3649 | spin_lock(&vb->lock); | |
3650 | if (bitmap_empty(vb->used_map, VMAP_BBMAP_BITS)) { | |
3651 | spin_unlock(&vb->lock); | |
4c91c07c | 3652 | goto finished_zero; |
06c89946 | 3653 | } |
4c91c07c | 3654 | |
06c89946 | 3655 | for_each_set_bitrange(rs, re, vb->used_map, VMAP_BBMAP_BITS) { |
4c91c07c LS |
3656 | size_t copied; |
3657 | ||
3658 | if (remains == 0) | |
3659 | goto finished; | |
3660 | ||
06c89946 | 3661 | start = vmap_block_vaddr(vb->va->va_start, rs); |
4c91c07c LS |
3662 | |
3663 | if (addr < start) { | |
3664 | size_t to_zero = min_t(size_t, start - addr, remains); | |
3665 | size_t zeroed = zero_iter(iter, to_zero); | |
3666 | ||
3667 | addr += zeroed; | |
3668 | remains -= zeroed; | |
3669 | ||
3670 | if (remains == 0 || zeroed != to_zero) | |
3671 | goto finished; | |
06c89946 | 3672 | } |
4c91c07c | 3673 | |
06c89946 BH |
3674 | /*it could start reading from the middle of used region*/ |
3675 | offset = offset_in_page(addr); | |
3676 | n = ((re - rs + 1) << PAGE_SHIFT) - offset; | |
4c91c07c LS |
3677 | if (n > remains) |
3678 | n = remains; | |
3679 | ||
3680 | copied = aligned_vread_iter(iter, start + offset, n); | |
06c89946 | 3681 | |
4c91c07c LS |
3682 | addr += copied; |
3683 | remains -= copied; | |
3684 | ||
3685 | if (copied != n) | |
3686 | goto finished; | |
06c89946 | 3687 | } |
4c91c07c | 3688 | |
06c89946 BH |
3689 | spin_unlock(&vb->lock); |
3690 | ||
4c91c07c | 3691 | finished_zero: |
06c89946 | 3692 | /* zero-fill the left dirty or free regions */ |
4c91c07c LS |
3693 | return count - remains + zero_iter(iter, remains); |
3694 | finished: | |
3695 | /* We couldn't copy/zero everything */ | |
3696 | spin_unlock(&vb->lock); | |
3697 | return count - remains; | |
06c89946 BH |
3698 | } |
3699 | ||
d0107eb0 | 3700 | /** |
4c91c07c LS |
3701 | * vread_iter() - read vmalloc area in a safe way to an iterator. |
3702 | * @iter: the iterator to which data should be written. | |
3703 | * @addr: vm address. | |
3704 | * @count: number of bytes to be read. | |
92eac168 | 3705 | * |
92eac168 MR |
3706 | * This function checks that addr is a valid vmalloc'ed area, and |
3707 | * copy data from that area to a given buffer. If the given memory range | |
3708 | * of [addr...addr+count) includes some valid address, data is copied to | |
3709 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
3710 | * IOREMAP area is treated as memory hole and no copy is done. | |
3711 | * | |
3712 | * If [addr...addr+count) doesn't includes any intersects with alive | |
3713 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
3714 | * | |
3715 | * Note: In usual ops, vread() is never necessary because the caller | |
3716 | * should know vmalloc() area is valid and can use memcpy(). | |
3717 | * This is for routines which have to access vmalloc area without | |
bbcd53c9 | 3718 | * any information, as /proc/kcore. |
a862f68a MR |
3719 | * |
3720 | * Return: number of bytes for which addr and buf should be increased | |
3721 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
3722 | * include any intersection with valid vmalloc area | |
d0107eb0 | 3723 | */ |
4c91c07c | 3724 | long vread_iter(struct iov_iter *iter, const char *addr, size_t count) |
1da177e4 | 3725 | { |
e81ce85f JK |
3726 | struct vmap_area *va; |
3727 | struct vm_struct *vm; | |
4c91c07c LS |
3728 | char *vaddr; |
3729 | size_t n, size, flags, remains; | |
1da177e4 | 3730 | |
4aff1dc4 AK |
3731 | addr = kasan_reset_tag(addr); |
3732 | ||
1da177e4 LT |
3733 | /* Don't allow overflow */ |
3734 | if ((unsigned long) addr + count < count) | |
3735 | count = -(unsigned long) addr; | |
3736 | ||
4c91c07c LS |
3737 | remains = count; |
3738 | ||
e81ce85f | 3739 | spin_lock(&vmap_area_lock); |
f181234a | 3740 | va = find_vmap_area_exceed_addr((unsigned long)addr); |
f608788c | 3741 | if (!va) |
4c91c07c | 3742 | goto finished_zero; |
f181234a CW |
3743 | |
3744 | /* no intersects with alive vmap_area */ | |
4c91c07c LS |
3745 | if ((unsigned long)addr + remains <= va->va_start) |
3746 | goto finished_zero; | |
f181234a | 3747 | |
f608788c | 3748 | list_for_each_entry_from(va, &vmap_area_list, list) { |
4c91c07c LS |
3749 | size_t copied; |
3750 | ||
3751 | if (remains == 0) | |
3752 | goto finished; | |
e81ce85f | 3753 | |
06c89946 BH |
3754 | vm = va->vm; |
3755 | flags = va->flags & VMAP_FLAGS_MASK; | |
3756 | /* | |
3757 | * VMAP_BLOCK indicates a sub-type of vm_map_ram area, need | |
3758 | * be set together with VMAP_RAM. | |
3759 | */ | |
3760 | WARN_ON(flags == VMAP_BLOCK); | |
3761 | ||
3762 | if (!vm && !flags) | |
e81ce85f JK |
3763 | continue; |
3764 | ||
30a7a9b1 BH |
3765 | if (vm && (vm->flags & VM_UNINITIALIZED)) |
3766 | continue; | |
4c91c07c | 3767 | |
30a7a9b1 BH |
3768 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3769 | smp_rmb(); | |
3770 | ||
06c89946 BH |
3771 | vaddr = (char *) va->va_start; |
3772 | size = vm ? get_vm_area_size(vm) : va_size(va); | |
3773 | ||
3774 | if (addr >= vaddr + size) | |
1da177e4 | 3775 | continue; |
4c91c07c LS |
3776 | |
3777 | if (addr < vaddr) { | |
3778 | size_t to_zero = min_t(size_t, vaddr - addr, remains); | |
3779 | size_t zeroed = zero_iter(iter, to_zero); | |
3780 | ||
3781 | addr += zeroed; | |
3782 | remains -= zeroed; | |
3783 | ||
3784 | if (remains == 0 || zeroed != to_zero) | |
1da177e4 | 3785 | goto finished; |
1da177e4 | 3786 | } |
4c91c07c | 3787 | |
06c89946 | 3788 | n = vaddr + size - addr; |
4c91c07c LS |
3789 | if (n > remains) |
3790 | n = remains; | |
06c89946 BH |
3791 | |
3792 | if (flags & VMAP_RAM) | |
4c91c07c | 3793 | copied = vmap_ram_vread_iter(iter, addr, n, flags); |
06c89946 | 3794 | else if (!(vm->flags & VM_IOREMAP)) |
4c91c07c | 3795 | copied = aligned_vread_iter(iter, addr, n); |
d0107eb0 | 3796 | else /* IOREMAP area is treated as memory hole */ |
4c91c07c LS |
3797 | copied = zero_iter(iter, n); |
3798 | ||
3799 | addr += copied; | |
3800 | remains -= copied; | |
3801 | ||
3802 | if (copied != n) | |
3803 | goto finished; | |
1da177e4 | 3804 | } |
d0107eb0 | 3805 | |
4c91c07c LS |
3806 | finished_zero: |
3807 | spin_unlock(&vmap_area_lock); | |
d0107eb0 | 3808 | /* zero-fill memory holes */ |
4c91c07c LS |
3809 | return count - remains + zero_iter(iter, remains); |
3810 | finished: | |
3811 | /* Nothing remains, or We couldn't copy/zero everything. */ | |
3812 | spin_unlock(&vmap_area_lock); | |
d0107eb0 | 3813 | |
4c91c07c | 3814 | return count - remains; |
1da177e4 LT |
3815 | } |
3816 | ||
83342314 | 3817 | /** |
92eac168 MR |
3818 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
3819 | * @vma: vma to cover | |
3820 | * @uaddr: target user address to start at | |
3821 | * @kaddr: virtual address of vmalloc kernel memory | |
bdebd6a2 | 3822 | * @pgoff: offset from @kaddr to start at |
92eac168 | 3823 | * @size: size of map area |
7682486b | 3824 | * |
92eac168 | 3825 | * Returns: 0 for success, -Exxx on failure |
83342314 | 3826 | * |
92eac168 MR |
3827 | * This function checks that @kaddr is a valid vmalloc'ed area, |
3828 | * and that it is big enough to cover the range starting at | |
3829 | * @uaddr in @vma. Will return failure if that criteria isn't | |
3830 | * met. | |
83342314 | 3831 | * |
92eac168 | 3832 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 3833 | */ |
e69e9d4a | 3834 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
bdebd6a2 JH |
3835 | void *kaddr, unsigned long pgoff, |
3836 | unsigned long size) | |
83342314 NP |
3837 | { |
3838 | struct vm_struct *area; | |
bdebd6a2 JH |
3839 | unsigned long off; |
3840 | unsigned long end_index; | |
3841 | ||
3842 | if (check_shl_overflow(pgoff, PAGE_SHIFT, &off)) | |
3843 | return -EINVAL; | |
83342314 | 3844 | |
e69e9d4a HD |
3845 | size = PAGE_ALIGN(size); |
3846 | ||
3847 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
3848 | return -EINVAL; |
3849 | ||
e69e9d4a | 3850 | area = find_vm_area(kaddr); |
83342314 | 3851 | if (!area) |
db64fe02 | 3852 | return -EINVAL; |
83342314 | 3853 | |
fe9041c2 | 3854 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3855 | return -EINVAL; |
83342314 | 3856 | |
bdebd6a2 JH |
3857 | if (check_add_overflow(size, off, &end_index) || |
3858 | end_index > get_vm_area_size(area)) | |
db64fe02 | 3859 | return -EINVAL; |
bdebd6a2 | 3860 | kaddr += off; |
83342314 | 3861 | |
83342314 | 3862 | do { |
e69e9d4a | 3863 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3864 | int ret; |
3865 | ||
83342314 NP |
3866 | ret = vm_insert_page(vma, uaddr, page); |
3867 | if (ret) | |
3868 | return ret; | |
3869 | ||
3870 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3871 | kaddr += PAGE_SIZE; |
3872 | size -= PAGE_SIZE; | |
3873 | } while (size > 0); | |
83342314 | 3874 | |
1c71222e | 3875 | vm_flags_set(vma, VM_DONTEXPAND | VM_DONTDUMP); |
83342314 | 3876 | |
db64fe02 | 3877 | return 0; |
83342314 | 3878 | } |
e69e9d4a HD |
3879 | |
3880 | /** | |
92eac168 MR |
3881 | * remap_vmalloc_range - map vmalloc pages to userspace |
3882 | * @vma: vma to cover (map full range of vma) | |
3883 | * @addr: vmalloc memory | |
3884 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3885 | * |
92eac168 | 3886 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3887 | * |
92eac168 MR |
3888 | * This function checks that addr is a valid vmalloc'ed area, and |
3889 | * that it is big enough to cover the vma. Will return failure if | |
3890 | * that criteria isn't met. | |
e69e9d4a | 3891 | * |
92eac168 | 3892 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3893 | */ |
3894 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3895 | unsigned long pgoff) | |
3896 | { | |
3897 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
bdebd6a2 | 3898 | addr, pgoff, |
e69e9d4a HD |
3899 | vma->vm_end - vma->vm_start); |
3900 | } | |
83342314 NP |
3901 | EXPORT_SYMBOL(remap_vmalloc_range); |
3902 | ||
5f4352fb JF |
3903 | void free_vm_area(struct vm_struct *area) |
3904 | { | |
3905 | struct vm_struct *ret; | |
3906 | ret = remove_vm_area(area->addr); | |
3907 | BUG_ON(ret != area); | |
3908 | kfree(area); | |
3909 | } | |
3910 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3911 | |
4f8b02b4 | 3912 | #ifdef CONFIG_SMP |
ca23e405 TH |
3913 | static struct vmap_area *node_to_va(struct rb_node *n) |
3914 | { | |
4583e773 | 3915 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3916 | } |
3917 | ||
3918 | /** | |
68ad4a33 URS |
3919 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3920 | * @addr: target address | |
ca23e405 | 3921 | * |
68ad4a33 URS |
3922 | * Returns: vmap_area if it is found. If there is no such area |
3923 | * the first highest(reverse order) vmap_area is returned | |
3924 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3925 | * if there are no any areas before @addr. | |
ca23e405 | 3926 | */ |
68ad4a33 URS |
3927 | static struct vmap_area * |
3928 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3929 | { |
68ad4a33 URS |
3930 | struct vmap_area *va, *tmp; |
3931 | struct rb_node *n; | |
3932 | ||
3933 | n = free_vmap_area_root.rb_node; | |
3934 | va = NULL; | |
ca23e405 TH |
3935 | |
3936 | while (n) { | |
68ad4a33 URS |
3937 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3938 | if (tmp->va_start <= addr) { | |
3939 | va = tmp; | |
3940 | if (tmp->va_end >= addr) | |
3941 | break; | |
3942 | ||
ca23e405 | 3943 | n = n->rb_right; |
68ad4a33 URS |
3944 | } else { |
3945 | n = n->rb_left; | |
3946 | } | |
ca23e405 TH |
3947 | } |
3948 | ||
68ad4a33 | 3949 | return va; |
ca23e405 TH |
3950 | } |
3951 | ||
3952 | /** | |
68ad4a33 URS |
3953 | * pvm_determine_end_from_reverse - find the highest aligned address |
3954 | * of free block below VMALLOC_END | |
3955 | * @va: | |
3956 | * in - the VA we start the search(reverse order); | |
3957 | * out - the VA with the highest aligned end address. | |
799fa85d | 3958 | * @align: alignment for required highest address |
ca23e405 | 3959 | * |
68ad4a33 | 3960 | * Returns: determined end address within vmap_area |
ca23e405 | 3961 | */ |
68ad4a33 URS |
3962 | static unsigned long |
3963 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3964 | { |
68ad4a33 | 3965 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3966 | unsigned long addr; |
3967 | ||
68ad4a33 URS |
3968 | if (likely(*va)) { |
3969 | list_for_each_entry_from_reverse((*va), | |
3970 | &free_vmap_area_list, list) { | |
3971 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3972 | if ((*va)->va_start < addr) | |
3973 | return addr; | |
3974 | } | |
ca23e405 TH |
3975 | } |
3976 | ||
68ad4a33 | 3977 | return 0; |
ca23e405 TH |
3978 | } |
3979 | ||
3980 | /** | |
3981 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3982 | * @offsets: array containing offset of each area | |
3983 | * @sizes: array containing size of each area | |
3984 | * @nr_vms: the number of areas to allocate | |
3985 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3986 | * |
3987 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3988 | * vm_structs on success, %NULL on failure | |
3989 | * | |
3990 | * Percpu allocator wants to use congruent vm areas so that it can | |
3991 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3992 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3993 | * be scattered pretty far, distance between two areas easily going up | |
3994 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3995 | * areas are allocated from top. | |
ca23e405 | 3996 | * |
68ad4a33 URS |
3997 | * Despite its complicated look, this allocator is rather simple. It |
3998 | * does everything top-down and scans free blocks from the end looking | |
3999 | * for matching base. While scanning, if any of the areas do not fit the | |
4000 | * base address is pulled down to fit the area. Scanning is repeated till | |
4001 | * all the areas fit and then all necessary data structures are inserted | |
4002 | * and the result is returned. | |
ca23e405 TH |
4003 | */ |
4004 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
4005 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 4006 | size_t align) |
ca23e405 TH |
4007 | { |
4008 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
4009 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 4010 | struct vmap_area **vas, *va; |
ca23e405 TH |
4011 | struct vm_struct **vms; |
4012 | int area, area2, last_area, term_area; | |
253a496d | 4013 | unsigned long base, start, size, end, last_end, orig_start, orig_end; |
ca23e405 TH |
4014 | bool purged = false; |
4015 | ||
ca23e405 | 4016 | /* verify parameters and allocate data structures */ |
891c49ab | 4017 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
4018 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
4019 | start = offsets[area]; | |
4020 | end = start + sizes[area]; | |
4021 | ||
4022 | /* is everything aligned properly? */ | |
4023 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
4024 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
4025 | ||
4026 | /* detect the area with the highest address */ | |
4027 | if (start > offsets[last_area]) | |
4028 | last_area = area; | |
4029 | ||
c568da28 | 4030 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
4031 | unsigned long start2 = offsets[area2]; |
4032 | unsigned long end2 = start2 + sizes[area2]; | |
4033 | ||
c568da28 | 4034 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
4035 | } |
4036 | } | |
4037 | last_end = offsets[last_area] + sizes[last_area]; | |
4038 | ||
4039 | if (vmalloc_end - vmalloc_start < last_end) { | |
4040 | WARN_ON(true); | |
4041 | return NULL; | |
4042 | } | |
4043 | ||
4d67d860 TM |
4044 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
4045 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 4046 | if (!vas || !vms) |
f1db7afd | 4047 | goto err_free2; |
ca23e405 TH |
4048 | |
4049 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 4050 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 4051 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
4052 | if (!vas[area] || !vms[area]) |
4053 | goto err_free; | |
4054 | } | |
4055 | retry: | |
e36176be | 4056 | spin_lock(&free_vmap_area_lock); |
ca23e405 TH |
4057 | |
4058 | /* start scanning - we scan from the top, begin with the last area */ | |
4059 | area = term_area = last_area; | |
4060 | start = offsets[area]; | |
4061 | end = start + sizes[area]; | |
4062 | ||
68ad4a33 URS |
4063 | va = pvm_find_va_enclose_addr(vmalloc_end); |
4064 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
4065 | |
4066 | while (true) { | |
ca23e405 TH |
4067 | /* |
4068 | * base might have underflowed, add last_end before | |
4069 | * comparing. | |
4070 | */ | |
68ad4a33 URS |
4071 | if (base + last_end < vmalloc_start + last_end) |
4072 | goto overflow; | |
ca23e405 TH |
4073 | |
4074 | /* | |
68ad4a33 | 4075 | * Fitting base has not been found. |
ca23e405 | 4076 | */ |
68ad4a33 URS |
4077 | if (va == NULL) |
4078 | goto overflow; | |
ca23e405 | 4079 | |
5336e52c | 4080 | /* |
d8cc323d | 4081 | * If required width exceeds current VA block, move |
5336e52c KS |
4082 | * base downwards and then recheck. |
4083 | */ | |
4084 | if (base + end > va->va_end) { | |
4085 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
4086 | term_area = area; | |
4087 | continue; | |
4088 | } | |
4089 | ||
ca23e405 | 4090 | /* |
68ad4a33 | 4091 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 4092 | */ |
5336e52c | 4093 | if (base + start < va->va_start) { |
68ad4a33 URS |
4094 | va = node_to_va(rb_prev(&va->rb_node)); |
4095 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
4096 | term_area = area; |
4097 | continue; | |
4098 | } | |
4099 | ||
4100 | /* | |
4101 | * This area fits, move on to the previous one. If | |
4102 | * the previous one is the terminal one, we're done. | |
4103 | */ | |
4104 | area = (area + nr_vms - 1) % nr_vms; | |
4105 | if (area == term_area) | |
4106 | break; | |
68ad4a33 | 4107 | |
ca23e405 TH |
4108 | start = offsets[area]; |
4109 | end = start + sizes[area]; | |
68ad4a33 | 4110 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 4111 | } |
68ad4a33 | 4112 | |
ca23e405 TH |
4113 | /* we've found a fitting base, insert all va's */ |
4114 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 4115 | int ret; |
ca23e405 | 4116 | |
68ad4a33 URS |
4117 | start = base + offsets[area]; |
4118 | size = sizes[area]; | |
ca23e405 | 4119 | |
68ad4a33 URS |
4120 | va = pvm_find_va_enclose_addr(start); |
4121 | if (WARN_ON_ONCE(va == NULL)) | |
4122 | /* It is a BUG(), but trigger recovery instead. */ | |
4123 | goto recovery; | |
4124 | ||
f9863be4 URS |
4125 | ret = adjust_va_to_fit_type(&free_vmap_area_root, |
4126 | &free_vmap_area_list, | |
4127 | va, start, size); | |
1b23ff80 | 4128 | if (WARN_ON_ONCE(unlikely(ret))) |
68ad4a33 URS |
4129 | /* It is a BUG(), but trigger recovery instead. */ |
4130 | goto recovery; | |
4131 | ||
68ad4a33 URS |
4132 | /* Allocated area. */ |
4133 | va = vas[area]; | |
4134 | va->va_start = start; | |
4135 | va->va_end = start + size; | |
68ad4a33 | 4136 | } |
ca23e405 | 4137 | |
e36176be | 4138 | spin_unlock(&free_vmap_area_lock); |
ca23e405 | 4139 | |
253a496d DA |
4140 | /* populate the kasan shadow space */ |
4141 | for (area = 0; area < nr_vms; area++) { | |
4142 | if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area])) | |
4143 | goto err_free_shadow; | |
253a496d DA |
4144 | } |
4145 | ||
ca23e405 | 4146 | /* insert all vm's */ |
e36176be URS |
4147 | spin_lock(&vmap_area_lock); |
4148 | for (area = 0; area < nr_vms; area++) { | |
4149 | insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list); | |
4150 | ||
4151 | setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC, | |
3645cb4a | 4152 | pcpu_get_vm_areas); |
e36176be URS |
4153 | } |
4154 | spin_unlock(&vmap_area_lock); | |
ca23e405 | 4155 | |
19f1c3ac AK |
4156 | /* |
4157 | * Mark allocated areas as accessible. Do it now as a best-effort | |
4158 | * approach, as they can be mapped outside of vmalloc code. | |
23689e91 AK |
4159 | * With hardware tag-based KASAN, marking is skipped for |
4160 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). | |
19f1c3ac | 4161 | */ |
1d96320f AK |
4162 | for (area = 0; area < nr_vms; area++) |
4163 | vms[area]->addr = kasan_unpoison_vmalloc(vms[area]->addr, | |
f6e39794 | 4164 | vms[area]->size, KASAN_VMALLOC_PROT_NORMAL); |
1d96320f | 4165 | |
ca23e405 TH |
4166 | kfree(vas); |
4167 | return vms; | |
4168 | ||
68ad4a33 | 4169 | recovery: |
e36176be URS |
4170 | /* |
4171 | * Remove previously allocated areas. There is no | |
4172 | * need in removing these areas from the busy tree, | |
4173 | * because they are inserted only on the final step | |
4174 | * and when pcpu_get_vm_areas() is success. | |
4175 | */ | |
68ad4a33 | 4176 | while (area--) { |
253a496d DA |
4177 | orig_start = vas[area]->va_start; |
4178 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
4179 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
4180 | &free_vmap_area_list); | |
9c801f61 URS |
4181 | if (va) |
4182 | kasan_release_vmalloc(orig_start, orig_end, | |
4183 | va->va_start, va->va_end); | |
68ad4a33 URS |
4184 | vas[area] = NULL; |
4185 | } | |
4186 | ||
4187 | overflow: | |
e36176be | 4188 | spin_unlock(&free_vmap_area_lock); |
68ad4a33 | 4189 | if (!purged) { |
77e50af0 | 4190 | reclaim_and_purge_vmap_areas(); |
68ad4a33 URS |
4191 | purged = true; |
4192 | ||
4193 | /* Before "retry", check if we recover. */ | |
4194 | for (area = 0; area < nr_vms; area++) { | |
4195 | if (vas[area]) | |
4196 | continue; | |
4197 | ||
4198 | vas[area] = kmem_cache_zalloc( | |
4199 | vmap_area_cachep, GFP_KERNEL); | |
4200 | if (!vas[area]) | |
4201 | goto err_free; | |
4202 | } | |
4203 | ||
4204 | goto retry; | |
4205 | } | |
4206 | ||
ca23e405 TH |
4207 | err_free: |
4208 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
4209 | if (vas[area]) |
4210 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
4211 | ||
f1db7afd | 4212 | kfree(vms[area]); |
ca23e405 | 4213 | } |
f1db7afd | 4214 | err_free2: |
ca23e405 TH |
4215 | kfree(vas); |
4216 | kfree(vms); | |
4217 | return NULL; | |
253a496d DA |
4218 | |
4219 | err_free_shadow: | |
4220 | spin_lock(&free_vmap_area_lock); | |
4221 | /* | |
4222 | * We release all the vmalloc shadows, even the ones for regions that | |
4223 | * hadn't been successfully added. This relies on kasan_release_vmalloc | |
4224 | * being able to tolerate this case. | |
4225 | */ | |
4226 | for (area = 0; area < nr_vms; area++) { | |
4227 | orig_start = vas[area]->va_start; | |
4228 | orig_end = vas[area]->va_end; | |
96e2db45 URS |
4229 | va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root, |
4230 | &free_vmap_area_list); | |
9c801f61 URS |
4231 | if (va) |
4232 | kasan_release_vmalloc(orig_start, orig_end, | |
4233 | va->va_start, va->va_end); | |
253a496d DA |
4234 | vas[area] = NULL; |
4235 | kfree(vms[area]); | |
4236 | } | |
4237 | spin_unlock(&free_vmap_area_lock); | |
4238 | kfree(vas); | |
4239 | kfree(vms); | |
4240 | return NULL; | |
ca23e405 TH |
4241 | } |
4242 | ||
4243 | /** | |
4244 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
4245 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
4246 | * @nr_vms: the number of allocated areas | |
4247 | * | |
4248 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
4249 | */ | |
4250 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
4251 | { | |
4252 | int i; | |
4253 | ||
4254 | for (i = 0; i < nr_vms; i++) | |
4255 | free_vm_area(vms[i]); | |
4256 | kfree(vms); | |
4257 | } | |
4f8b02b4 | 4258 | #endif /* CONFIG_SMP */ |
a10aa579 | 4259 | |
5bb1bb35 | 4260 | #ifdef CONFIG_PRINTK |
98f18083 PM |
4261 | bool vmalloc_dump_obj(void *object) |
4262 | { | |
4263 | struct vm_struct *vm; | |
4264 | void *objp = (void *)PAGE_ALIGN((unsigned long)object); | |
4265 | ||
4266 | vm = find_vm_area(objp); | |
4267 | if (!vm) | |
4268 | return false; | |
bd34dcd4 PM |
4269 | pr_cont(" %u-page vmalloc region starting at %#lx allocated at %pS\n", |
4270 | vm->nr_pages, (unsigned long)vm->addr, vm->caller); | |
98f18083 PM |
4271 | return true; |
4272 | } | |
5bb1bb35 | 4273 | #endif |
98f18083 | 4274 | |
a10aa579 CL |
4275 | #ifdef CONFIG_PROC_FS |
4276 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e36176be | 4277 | __acquires(&vmap_purge_lock) |
d4033afd | 4278 | __acquires(&vmap_area_lock) |
a10aa579 | 4279 | { |
e36176be | 4280 | mutex_lock(&vmap_purge_lock); |
d4033afd | 4281 | spin_lock(&vmap_area_lock); |
e36176be | 4282 | |
3f500069 | 4283 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
4284 | } |
4285 | ||
4286 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4287 | { | |
3f500069 | 4288 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
4289 | } |
4290 | ||
4291 | static void s_stop(struct seq_file *m, void *p) | |
d4033afd | 4292 | __releases(&vmap_area_lock) |
0a7dd4e9 | 4293 | __releases(&vmap_purge_lock) |
a10aa579 | 4294 | { |
d4033afd | 4295 | spin_unlock(&vmap_area_lock); |
0a7dd4e9 | 4296 | mutex_unlock(&vmap_purge_lock); |
a10aa579 CL |
4297 | } |
4298 | ||
a47a126a ED |
4299 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
4300 | { | |
e5adfffc | 4301 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a | 4302 | unsigned int nr, *counters = m->private; |
51e50b3a | 4303 | unsigned int step = 1U << vm_area_page_order(v); |
a47a126a ED |
4304 | |
4305 | if (!counters) | |
4306 | return; | |
4307 | ||
af12346c WL |
4308 | if (v->flags & VM_UNINITIALIZED) |
4309 | return; | |
7e5b528b DV |
4310 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
4311 | smp_rmb(); | |
af12346c | 4312 | |
a47a126a ED |
4313 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
4314 | ||
51e50b3a ED |
4315 | for (nr = 0; nr < v->nr_pages; nr += step) |
4316 | counters[page_to_nid(v->pages[nr])] += step; | |
a47a126a ED |
4317 | for_each_node_state(nr, N_HIGH_MEMORY) |
4318 | if (counters[nr]) | |
4319 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
4320 | } | |
4321 | } | |
4322 | ||
dd3b8353 URS |
4323 | static void show_purge_info(struct seq_file *m) |
4324 | { | |
dd3b8353 URS |
4325 | struct vmap_area *va; |
4326 | ||
96e2db45 URS |
4327 | spin_lock(&purge_vmap_area_lock); |
4328 | list_for_each_entry(va, &purge_vmap_area_list, list) { | |
dd3b8353 URS |
4329 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", |
4330 | (void *)va->va_start, (void *)va->va_end, | |
4331 | va->va_end - va->va_start); | |
4332 | } | |
96e2db45 | 4333 | spin_unlock(&purge_vmap_area_lock); |
dd3b8353 URS |
4334 | } |
4335 | ||
a10aa579 CL |
4336 | static int s_show(struct seq_file *m, void *p) |
4337 | { | |
3f500069 | 4338 | struct vmap_area *va; |
d4033afd JK |
4339 | struct vm_struct *v; |
4340 | ||
3f500069 | 4341 | va = list_entry(p, struct vmap_area, list); |
4342 | ||
688fcbfc | 4343 | if (!va->vm) { |
bba9697b BH |
4344 | if (va->flags & VMAP_RAM) |
4345 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", | |
4346 | (void *)va->va_start, (void *)va->va_end, | |
4347 | va->va_end - va->va_start); | |
78c72746 | 4348 | |
7cc7913e | 4349 | goto final; |
78c72746 | 4350 | } |
d4033afd JK |
4351 | |
4352 | v = va->vm; | |
a10aa579 | 4353 | |
45ec1690 | 4354 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
4355 | v->addr, v->addr + v->size, v->size); |
4356 | ||
62c70bce JP |
4357 | if (v->caller) |
4358 | seq_printf(m, " %pS", v->caller); | |
23016969 | 4359 | |
a10aa579 CL |
4360 | if (v->nr_pages) |
4361 | seq_printf(m, " pages=%d", v->nr_pages); | |
4362 | ||
4363 | if (v->phys_addr) | |
199eaa05 | 4364 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
4365 | |
4366 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 4367 | seq_puts(m, " ioremap"); |
a10aa579 CL |
4368 | |
4369 | if (v->flags & VM_ALLOC) | |
f4527c90 | 4370 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
4371 | |
4372 | if (v->flags & VM_MAP) | |
f4527c90 | 4373 | seq_puts(m, " vmap"); |
a10aa579 CL |
4374 | |
4375 | if (v->flags & VM_USERMAP) | |
f4527c90 | 4376 | seq_puts(m, " user"); |
a10aa579 | 4377 | |
fe9041c2 CH |
4378 | if (v->flags & VM_DMA_COHERENT) |
4379 | seq_puts(m, " dma-coherent"); | |
4380 | ||
244d63ee | 4381 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 4382 | seq_puts(m, " vpages"); |
a10aa579 | 4383 | |
a47a126a | 4384 | show_numa_info(m, v); |
a10aa579 | 4385 | seq_putc(m, '\n'); |
dd3b8353 URS |
4386 | |
4387 | /* | |
96e2db45 | 4388 | * As a final step, dump "unpurged" areas. |
dd3b8353 | 4389 | */ |
7cc7913e | 4390 | final: |
dd3b8353 URS |
4391 | if (list_is_last(&va->list, &vmap_area_list)) |
4392 | show_purge_info(m); | |
4393 | ||
a10aa579 CL |
4394 | return 0; |
4395 | } | |
4396 | ||
5f6a6a9c | 4397 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
4398 | .start = s_start, |
4399 | .next = s_next, | |
4400 | .stop = s_stop, | |
4401 | .show = s_show, | |
4402 | }; | |
5f6a6a9c | 4403 | |
5f6a6a9c AD |
4404 | static int __init proc_vmalloc_init(void) |
4405 | { | |
fddda2b7 | 4406 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 4407 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
4408 | &vmalloc_op, |
4409 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 4410 | else |
0825a6f9 | 4411 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
4412 | return 0; |
4413 | } | |
4414 | module_init(proc_vmalloc_init); | |
db3808c1 | 4415 | |
a10aa579 | 4416 | #endif |
208162f4 CH |
4417 | |
4418 | void __init vmalloc_init(void) | |
4419 | { | |
4420 | struct vmap_area *va; | |
4421 | struct vm_struct *tmp; | |
4422 | int i; | |
4423 | ||
4424 | /* | |
4425 | * Create the cache for vmap_area objects. | |
4426 | */ | |
4427 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
4428 | ||
4429 | for_each_possible_cpu(i) { | |
4430 | struct vmap_block_queue *vbq; | |
4431 | struct vfree_deferred *p; | |
4432 | ||
4433 | vbq = &per_cpu(vmap_block_queue, i); | |
4434 | spin_lock_init(&vbq->lock); | |
4435 | INIT_LIST_HEAD(&vbq->free); | |
4436 | p = &per_cpu(vfree_deferred, i); | |
4437 | init_llist_head(&p->list); | |
4438 | INIT_WORK(&p->wq, delayed_vfree_work); | |
062eacf5 | 4439 | xa_init(&vbq->vmap_blocks); |
208162f4 CH |
4440 | } |
4441 | ||
4442 | /* Import existing vmlist entries. */ | |
4443 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
4444 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
4445 | if (WARN_ON_ONCE(!va)) | |
4446 | continue; | |
4447 | ||
4448 | va->va_start = (unsigned long)tmp->addr; | |
4449 | va->va_end = va->va_start + tmp->size; | |
4450 | va->vm = tmp; | |
4451 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
4452 | } | |
4453 | ||
4454 | /* | |
4455 | * Now we can initialize a free vmap space. | |
4456 | */ | |
4457 | vmap_init_free_space(); | |
4458 | vmap_initialized = true; | |
4459 | } |