]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/vmalloc.c | |
4 | * | |
5 | * Copyright (C) 1993 Linus Torvalds | |
6 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
7 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 | |
8 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 | |
930fc45a | 9 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
d758ffe6 | 10 | * Improving global KVA allocator, Uladzislau Rezki, Sony, May 2019 |
1da177e4 LT |
11 | */ |
12 | ||
db64fe02 | 13 | #include <linux/vmalloc.h> |
1da177e4 LT |
14 | #include <linux/mm.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/highmem.h> | |
c3edc401 | 17 | #include <linux/sched/signal.h> |
1da177e4 LT |
18 | #include <linux/slab.h> |
19 | #include <linux/spinlock.h> | |
20 | #include <linux/interrupt.h> | |
5f6a6a9c | 21 | #include <linux/proc_fs.h> |
a10aa579 | 22 | #include <linux/seq_file.h> |
868b104d | 23 | #include <linux/set_memory.h> |
3ac7fe5a | 24 | #include <linux/debugobjects.h> |
23016969 | 25 | #include <linux/kallsyms.h> |
db64fe02 | 26 | #include <linux/list.h> |
4da56b99 | 27 | #include <linux/notifier.h> |
db64fe02 | 28 | #include <linux/rbtree.h> |
0f14599c | 29 | #include <linux/xarray.h> |
db64fe02 | 30 | #include <linux/rcupdate.h> |
f0aa6617 | 31 | #include <linux/pfn.h> |
89219d37 | 32 | #include <linux/kmemleak.h> |
60063497 | 33 | #include <linux/atomic.h> |
3b32123d | 34 | #include <linux/compiler.h> |
32fcfd40 | 35 | #include <linux/llist.h> |
0f616be1 | 36 | #include <linux/bitops.h> |
68ad4a33 | 37 | #include <linux/rbtree_augmented.h> |
bdebd6a2 | 38 | #include <linux/overflow.h> |
3b32123d | 39 | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 | 41 | #include <asm/tlbflush.h> |
2dca6999 | 42 | #include <asm/shmparam.h> |
1da177e4 | 43 | |
dd56b046 | 44 | #include "internal.h" |
2a681cfa | 45 | #include "pgalloc-track.h" |
dd56b046 | 46 | |
186525bd IM |
47 | bool is_vmalloc_addr(const void *x) |
48 | { | |
49 | unsigned long addr = (unsigned long)x; | |
50 | ||
51 | return addr >= VMALLOC_START && addr < VMALLOC_END; | |
52 | } | |
53 | EXPORT_SYMBOL(is_vmalloc_addr); | |
54 | ||
32fcfd40 AV |
55 | struct vfree_deferred { |
56 | struct llist_head list; | |
57 | struct work_struct wq; | |
58 | }; | |
59 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
60 | ||
61 | static void __vunmap(const void *, int); | |
62 | ||
63 | static void free_work(struct work_struct *w) | |
64 | { | |
65 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
66 | struct llist_node *t, *llnode; |
67 | ||
68 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
69 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
70 | } |
71 | ||
db64fe02 | 72 | /*** Page table manipulation functions ***/ |
b221385b | 73 | |
2ba3e694 JR |
74 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
75 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
76 | { |
77 | pte_t *pte; | |
78 | ||
79 | pte = pte_offset_kernel(pmd, addr); | |
80 | do { | |
81 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
82 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
83 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
2ba3e694 | 84 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
85 | } |
86 | ||
2ba3e694 JR |
87 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end, |
88 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
89 | { |
90 | pmd_t *pmd; | |
91 | unsigned long next; | |
2ba3e694 | 92 | int cleared; |
1da177e4 LT |
93 | |
94 | pmd = pmd_offset(pud, addr); | |
95 | do { | |
96 | next = pmd_addr_end(addr, end); | |
2ba3e694 JR |
97 | |
98 | cleared = pmd_clear_huge(pmd); | |
99 | if (cleared || pmd_bad(*pmd)) | |
100 | *mask |= PGTBL_PMD_MODIFIED; | |
101 | ||
102 | if (cleared) | |
b9820d8f | 103 | continue; |
1da177e4 LT |
104 | if (pmd_none_or_clear_bad(pmd)) |
105 | continue; | |
2ba3e694 | 106 | vunmap_pte_range(pmd, addr, next, mask); |
e47110e9 AK |
107 | |
108 | cond_resched(); | |
1da177e4 LT |
109 | } while (pmd++, addr = next, addr != end); |
110 | } | |
111 | ||
2ba3e694 JR |
112 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end, |
113 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
114 | { |
115 | pud_t *pud; | |
116 | unsigned long next; | |
2ba3e694 | 117 | int cleared; |
1da177e4 | 118 | |
c2febafc | 119 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
120 | do { |
121 | next = pud_addr_end(addr, end); | |
2ba3e694 JR |
122 | |
123 | cleared = pud_clear_huge(pud); | |
124 | if (cleared || pud_bad(*pud)) | |
125 | *mask |= PGTBL_PUD_MODIFIED; | |
126 | ||
127 | if (cleared) | |
b9820d8f | 128 | continue; |
1da177e4 LT |
129 | if (pud_none_or_clear_bad(pud)) |
130 | continue; | |
2ba3e694 | 131 | vunmap_pmd_range(pud, addr, next, mask); |
1da177e4 LT |
132 | } while (pud++, addr = next, addr != end); |
133 | } | |
134 | ||
2ba3e694 JR |
135 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end, |
136 | pgtbl_mod_mask *mask) | |
c2febafc KS |
137 | { |
138 | p4d_t *p4d; | |
139 | unsigned long next; | |
2ba3e694 | 140 | int cleared; |
c2febafc KS |
141 | |
142 | p4d = p4d_offset(pgd, addr); | |
143 | do { | |
144 | next = p4d_addr_end(addr, end); | |
2ba3e694 JR |
145 | |
146 | cleared = p4d_clear_huge(p4d); | |
147 | if (cleared || p4d_bad(*p4d)) | |
148 | *mask |= PGTBL_P4D_MODIFIED; | |
149 | ||
150 | if (cleared) | |
c2febafc KS |
151 | continue; |
152 | if (p4d_none_or_clear_bad(p4d)) | |
153 | continue; | |
2ba3e694 | 154 | vunmap_pud_range(p4d, addr, next, mask); |
c2febafc KS |
155 | } while (p4d++, addr = next, addr != end); |
156 | } | |
157 | ||
b521c43f CH |
158 | /** |
159 | * unmap_kernel_range_noflush - unmap kernel VM area | |
2ba3e694 | 160 | * @start: start of the VM area to unmap |
b521c43f CH |
161 | * @size: size of the VM area to unmap |
162 | * | |
163 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size specify | |
164 | * should have been allocated using get_vm_area() and its friends. | |
165 | * | |
166 | * NOTE: | |
167 | * This function does NOT do any cache flushing. The caller is responsible | |
168 | * for calling flush_cache_vunmap() on to-be-mapped areas before calling this | |
169 | * function and flush_tlb_kernel_range() after. | |
170 | */ | |
2ba3e694 | 171 | void unmap_kernel_range_noflush(unsigned long start, unsigned long size) |
1da177e4 | 172 | { |
2ba3e694 | 173 | unsigned long end = start + size; |
1da177e4 | 174 | unsigned long next; |
b521c43f | 175 | pgd_t *pgd; |
2ba3e694 JR |
176 | unsigned long addr = start; |
177 | pgtbl_mod_mask mask = 0; | |
1da177e4 LT |
178 | |
179 | BUG_ON(addr >= end); | |
180 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
181 | do { |
182 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
183 | if (pgd_bad(*pgd)) |
184 | mask |= PGTBL_PGD_MODIFIED; | |
1da177e4 LT |
185 | if (pgd_none_or_clear_bad(pgd)) |
186 | continue; | |
2ba3e694 | 187 | vunmap_p4d_range(pgd, addr, next, &mask); |
1da177e4 | 188 | } while (pgd++, addr = next, addr != end); |
2ba3e694 JR |
189 | |
190 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) | |
191 | arch_sync_kernel_mappings(start, end); | |
1da177e4 LT |
192 | } |
193 | ||
194 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
2ba3e694 JR |
195 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
196 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
197 | { |
198 | pte_t *pte; | |
199 | ||
db64fe02 NP |
200 | /* |
201 | * nr is a running index into the array which helps higher level | |
202 | * callers keep track of where we're up to. | |
203 | */ | |
204 | ||
2ba3e694 | 205 | pte = pte_alloc_kernel_track(pmd, addr, mask); |
1da177e4 LT |
206 | if (!pte) |
207 | return -ENOMEM; | |
208 | do { | |
db64fe02 NP |
209 | struct page *page = pages[*nr]; |
210 | ||
211 | if (WARN_ON(!pte_none(*pte))) | |
212 | return -EBUSY; | |
213 | if (WARN_ON(!page)) | |
1da177e4 LT |
214 | return -ENOMEM; |
215 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 216 | (*nr)++; |
1da177e4 | 217 | } while (pte++, addr += PAGE_SIZE, addr != end); |
2ba3e694 | 218 | *mask |= PGTBL_PTE_MODIFIED; |
1da177e4 LT |
219 | return 0; |
220 | } | |
221 | ||
db64fe02 | 222 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
2ba3e694 JR |
223 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
224 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
225 | { |
226 | pmd_t *pmd; | |
227 | unsigned long next; | |
228 | ||
2ba3e694 | 229 | pmd = pmd_alloc_track(&init_mm, pud, addr, mask); |
1da177e4 LT |
230 | if (!pmd) |
231 | return -ENOMEM; | |
232 | do { | |
233 | next = pmd_addr_end(addr, end); | |
2ba3e694 | 234 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
235 | return -ENOMEM; |
236 | } while (pmd++, addr = next, addr != end); | |
237 | return 0; | |
238 | } | |
239 | ||
c2febafc | 240 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, |
2ba3e694 JR |
241 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
242 | pgtbl_mod_mask *mask) | |
1da177e4 LT |
243 | { |
244 | pud_t *pud; | |
245 | unsigned long next; | |
246 | ||
2ba3e694 | 247 | pud = pud_alloc_track(&init_mm, p4d, addr, mask); |
1da177e4 LT |
248 | if (!pud) |
249 | return -ENOMEM; | |
250 | do { | |
251 | next = pud_addr_end(addr, end); | |
2ba3e694 | 252 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr, mask)) |
1da177e4 LT |
253 | return -ENOMEM; |
254 | } while (pud++, addr = next, addr != end); | |
255 | return 0; | |
256 | } | |
257 | ||
c2febafc | 258 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, |
2ba3e694 JR |
259 | unsigned long end, pgprot_t prot, struct page **pages, int *nr, |
260 | pgtbl_mod_mask *mask) | |
c2febafc KS |
261 | { |
262 | p4d_t *p4d; | |
263 | unsigned long next; | |
264 | ||
2ba3e694 | 265 | p4d = p4d_alloc_track(&init_mm, pgd, addr, mask); |
c2febafc KS |
266 | if (!p4d) |
267 | return -ENOMEM; | |
268 | do { | |
269 | next = p4d_addr_end(addr, end); | |
2ba3e694 | 270 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr, mask)) |
c2febafc KS |
271 | return -ENOMEM; |
272 | } while (p4d++, addr = next, addr != end); | |
273 | return 0; | |
274 | } | |
275 | ||
b521c43f CH |
276 | /** |
277 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
278 | * @addr: start of the VM area to map | |
279 | * @size: size of the VM area to map | |
280 | * @prot: page protection flags to use | |
281 | * @pages: pages to map | |
db64fe02 | 282 | * |
b521c43f CH |
283 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size specify should |
284 | * have been allocated using get_vm_area() and its friends. | |
285 | * | |
286 | * NOTE: | |
287 | * This function does NOT do any cache flushing. The caller is responsible for | |
288 | * calling flush_cache_vmap() on to-be-mapped areas before calling this | |
289 | * function. | |
290 | * | |
291 | * RETURNS: | |
60bb4465 | 292 | * 0 on success, -errno on failure. |
db64fe02 | 293 | */ |
b521c43f CH |
294 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, |
295 | pgprot_t prot, struct page **pages) | |
1da177e4 | 296 | { |
2ba3e694 | 297 | unsigned long start = addr; |
b521c43f | 298 | unsigned long end = addr + size; |
1da177e4 | 299 | unsigned long next; |
b521c43f | 300 | pgd_t *pgd; |
db64fe02 NP |
301 | int err = 0; |
302 | int nr = 0; | |
2ba3e694 | 303 | pgtbl_mod_mask mask = 0; |
1da177e4 LT |
304 | |
305 | BUG_ON(addr >= end); | |
306 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
307 | do { |
308 | next = pgd_addr_end(addr, end); | |
2ba3e694 JR |
309 | if (pgd_bad(*pgd)) |
310 | mask |= PGTBL_PGD_MODIFIED; | |
311 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr, &mask); | |
1da177e4 | 312 | if (err) |
bf88c8c8 | 313 | return err; |
1da177e4 | 314 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 315 | |
2ba3e694 JR |
316 | if (mask & ARCH_PAGE_TABLE_SYNC_MASK) |
317 | arch_sync_kernel_mappings(start, end); | |
318 | ||
60bb4465 | 319 | return 0; |
1da177e4 LT |
320 | } |
321 | ||
ed1f324c CH |
322 | int map_kernel_range(unsigned long start, unsigned long size, pgprot_t prot, |
323 | struct page **pages) | |
8fc48985 TH |
324 | { |
325 | int ret; | |
326 | ||
a29adb62 CH |
327 | ret = map_kernel_range_noflush(start, size, prot, pages); |
328 | flush_cache_vmap(start, start + size); | |
8fc48985 TH |
329 | return ret; |
330 | } | |
331 | ||
81ac3ad9 | 332 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
333 | { |
334 | /* | |
ab4f2ee1 | 335 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
336 | * and fall back on vmalloc() if that fails. Others |
337 | * just put it in the vmalloc space. | |
338 | */ | |
339 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
340 | unsigned long addr = (unsigned long)x; | |
341 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
342 | return 1; | |
343 | #endif | |
344 | return is_vmalloc_addr(x); | |
345 | } | |
346 | ||
48667e7a | 347 | /* |
add688fb | 348 | * Walk a vmap address to the struct page it maps. |
48667e7a | 349 | */ |
add688fb | 350 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
351 | { |
352 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 353 | struct page *page = NULL; |
48667e7a | 354 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
355 | p4d_t *p4d; |
356 | pud_t *pud; | |
357 | pmd_t *pmd; | |
358 | pte_t *ptep, pte; | |
48667e7a | 359 | |
7aa413de IM |
360 | /* |
361 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
362 | * architectures that do not vmalloc module space | |
363 | */ | |
73bdf0a6 | 364 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 365 | |
c2febafc KS |
366 | if (pgd_none(*pgd)) |
367 | return NULL; | |
368 | p4d = p4d_offset(pgd, addr); | |
369 | if (p4d_none(*p4d)) | |
370 | return NULL; | |
371 | pud = pud_offset(p4d, addr); | |
029c54b0 AB |
372 | |
373 | /* | |
374 | * Don't dereference bad PUD or PMD (below) entries. This will also | |
375 | * identify huge mappings, which we may encounter on architectures | |
376 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be | |
377 | * identified as vmalloc addresses by is_vmalloc_addr(), but are | |
378 | * not [unambiguously] associated with a struct page, so there is | |
379 | * no correct value to return for them. | |
380 | */ | |
381 | WARN_ON_ONCE(pud_bad(*pud)); | |
382 | if (pud_none(*pud) || pud_bad(*pud)) | |
c2febafc KS |
383 | return NULL; |
384 | pmd = pmd_offset(pud, addr); | |
029c54b0 AB |
385 | WARN_ON_ONCE(pmd_bad(*pmd)); |
386 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
c2febafc KS |
387 | return NULL; |
388 | ||
389 | ptep = pte_offset_map(pmd, addr); | |
390 | pte = *ptep; | |
391 | if (pte_present(pte)) | |
392 | page = pte_page(pte); | |
393 | pte_unmap(ptep); | |
add688fb | 394 | return page; |
48667e7a | 395 | } |
add688fb | 396 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
397 | |
398 | /* | |
add688fb | 399 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 400 | */ |
add688fb | 401 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 402 | { |
add688fb | 403 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 404 | } |
add688fb | 405 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 406 | |
db64fe02 NP |
407 | |
408 | /*** Global kva allocator ***/ | |
409 | ||
bb850f4d | 410 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 411 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 412 | |
db64fe02 | 413 | |
db64fe02 | 414 | static DEFINE_SPINLOCK(vmap_area_lock); |
e36176be | 415 | static DEFINE_SPINLOCK(free_vmap_area_lock); |
f1c4069e JK |
416 | /* Export for kexec only */ |
417 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 418 | static LLIST_HEAD(vmap_purge_list); |
89699605 | 419 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 420 | static bool vmap_initialized __read_mostly; |
89699605 | 421 | |
68ad4a33 URS |
422 | /* |
423 | * This kmem_cache is used for vmap_area objects. Instead of | |
424 | * allocating from slab we reuse an object from this cache to | |
425 | * make things faster. Especially in "no edge" splitting of | |
426 | * free block. | |
427 | */ | |
428 | static struct kmem_cache *vmap_area_cachep; | |
429 | ||
430 | /* | |
431 | * This linked list is used in pair with free_vmap_area_root. | |
432 | * It gives O(1) access to prev/next to perform fast coalescing. | |
433 | */ | |
434 | static LIST_HEAD(free_vmap_area_list); | |
435 | ||
436 | /* | |
437 | * This augment red-black tree represents the free vmap space. | |
438 | * All vmap_area objects in this tree are sorted by va->va_start | |
439 | * address. It is used for allocation and merging when a vmap | |
440 | * object is released. | |
441 | * | |
442 | * Each vmap_area node contains a maximum available free block | |
443 | * of its sub-tree, right or left. Therefore it is possible to | |
444 | * find a lowest match of free area. | |
445 | */ | |
446 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
447 | ||
82dd23e8 URS |
448 | /* |
449 | * Preload a CPU with one object for "no edge" split case. The | |
450 | * aim is to get rid of allocations from the atomic context, thus | |
451 | * to use more permissive allocation masks. | |
452 | */ | |
453 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
454 | ||
68ad4a33 URS |
455 | static __always_inline unsigned long |
456 | va_size(struct vmap_area *va) | |
457 | { | |
458 | return (va->va_end - va->va_start); | |
459 | } | |
460 | ||
461 | static __always_inline unsigned long | |
462 | get_subtree_max_size(struct rb_node *node) | |
463 | { | |
464 | struct vmap_area *va; | |
465 | ||
466 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
467 | return va ? va->subtree_max_size : 0; | |
468 | } | |
89699605 | 469 | |
68ad4a33 URS |
470 | /* |
471 | * Gets called when remove the node and rotate. | |
472 | */ | |
473 | static __always_inline unsigned long | |
474 | compute_subtree_max_size(struct vmap_area *va) | |
475 | { | |
476 | return max3(va_size(va), | |
477 | get_subtree_max_size(va->rb_node.rb_left), | |
478 | get_subtree_max_size(va->rb_node.rb_right)); | |
479 | } | |
480 | ||
315cc066 ML |
481 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
482 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 URS |
483 | |
484 | static void purge_vmap_area_lazy(void); | |
485 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
486 | static unsigned long lazy_max_pages(void); | |
db64fe02 | 487 | |
97105f0a RG |
488 | static atomic_long_t nr_vmalloc_pages; |
489 | ||
490 | unsigned long vmalloc_nr_pages(void) | |
491 | { | |
492 | return atomic_long_read(&nr_vmalloc_pages); | |
493 | } | |
494 | ||
db64fe02 | 495 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 496 | { |
db64fe02 NP |
497 | struct rb_node *n = vmap_area_root.rb_node; |
498 | ||
499 | while (n) { | |
500 | struct vmap_area *va; | |
501 | ||
502 | va = rb_entry(n, struct vmap_area, rb_node); | |
503 | if (addr < va->va_start) | |
504 | n = n->rb_left; | |
cef2ac3f | 505 | else if (addr >= va->va_end) |
db64fe02 NP |
506 | n = n->rb_right; |
507 | else | |
508 | return va; | |
509 | } | |
510 | ||
511 | return NULL; | |
512 | } | |
513 | ||
68ad4a33 URS |
514 | /* |
515 | * This function returns back addresses of parent node | |
516 | * and its left or right link for further processing. | |
9c801f61 URS |
517 | * |
518 | * Otherwise NULL is returned. In that case all further | |
519 | * steps regarding inserting of conflicting overlap range | |
520 | * have to be declined and actually considered as a bug. | |
68ad4a33 URS |
521 | */ |
522 | static __always_inline struct rb_node ** | |
523 | find_va_links(struct vmap_area *va, | |
524 | struct rb_root *root, struct rb_node *from, | |
525 | struct rb_node **parent) | |
526 | { | |
527 | struct vmap_area *tmp_va; | |
528 | struct rb_node **link; | |
529 | ||
530 | if (root) { | |
531 | link = &root->rb_node; | |
532 | if (unlikely(!*link)) { | |
533 | *parent = NULL; | |
534 | return link; | |
535 | } | |
536 | } else { | |
537 | link = &from; | |
538 | } | |
db64fe02 | 539 | |
68ad4a33 URS |
540 | /* |
541 | * Go to the bottom of the tree. When we hit the last point | |
542 | * we end up with parent rb_node and correct direction, i name | |
543 | * it link, where the new va->rb_node will be attached to. | |
544 | */ | |
545 | do { | |
546 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 547 | |
68ad4a33 URS |
548 | /* |
549 | * During the traversal we also do some sanity check. | |
550 | * Trigger the BUG() if there are sides(left/right) | |
551 | * or full overlaps. | |
552 | */ | |
553 | if (va->va_start < tmp_va->va_end && | |
554 | va->va_end <= tmp_va->va_start) | |
555 | link = &(*link)->rb_left; | |
556 | else if (va->va_end > tmp_va->va_start && | |
557 | va->va_start >= tmp_va->va_end) | |
558 | link = &(*link)->rb_right; | |
9c801f61 URS |
559 | else { |
560 | WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n", | |
561 | va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end); | |
562 | ||
563 | return NULL; | |
564 | } | |
68ad4a33 URS |
565 | } while (*link); |
566 | ||
567 | *parent = &tmp_va->rb_node; | |
568 | return link; | |
569 | } | |
570 | ||
571 | static __always_inline struct list_head * | |
572 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
573 | { | |
574 | struct list_head *list; | |
575 | ||
576 | if (unlikely(!parent)) | |
577 | /* | |
578 | * The red-black tree where we try to find VA neighbors | |
579 | * before merging or inserting is empty, i.e. it means | |
580 | * there is no free vmap space. Normally it does not | |
581 | * happen but we handle this case anyway. | |
582 | */ | |
583 | return NULL; | |
584 | ||
585 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
586 | return (&parent->rb_right == link ? list->next : list); | |
587 | } | |
588 | ||
589 | static __always_inline void | |
590 | link_va(struct vmap_area *va, struct rb_root *root, | |
591 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
592 | { | |
593 | /* | |
594 | * VA is still not in the list, but we can | |
595 | * identify its future previous list_head node. | |
596 | */ | |
597 | if (likely(parent)) { | |
598 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
599 | if (&parent->rb_right != link) | |
600 | head = head->prev; | |
db64fe02 NP |
601 | } |
602 | ||
68ad4a33 URS |
603 | /* Insert to the rb-tree */ |
604 | rb_link_node(&va->rb_node, parent, link); | |
605 | if (root == &free_vmap_area_root) { | |
606 | /* | |
607 | * Some explanation here. Just perform simple insertion | |
608 | * to the tree. We do not set va->subtree_max_size to | |
609 | * its current size before calling rb_insert_augmented(). | |
610 | * It is because of we populate the tree from the bottom | |
611 | * to parent levels when the node _is_ in the tree. | |
612 | * | |
613 | * Therefore we set subtree_max_size to zero after insertion, | |
614 | * to let __augment_tree_propagate_from() puts everything to | |
615 | * the correct order later on. | |
616 | */ | |
617 | rb_insert_augmented(&va->rb_node, | |
618 | root, &free_vmap_area_rb_augment_cb); | |
619 | va->subtree_max_size = 0; | |
620 | } else { | |
621 | rb_insert_color(&va->rb_node, root); | |
622 | } | |
db64fe02 | 623 | |
68ad4a33 URS |
624 | /* Address-sort this list */ |
625 | list_add(&va->list, head); | |
db64fe02 NP |
626 | } |
627 | ||
68ad4a33 URS |
628 | static __always_inline void |
629 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
630 | { | |
460e42d1 URS |
631 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
632 | return; | |
db64fe02 | 633 | |
460e42d1 URS |
634 | if (root == &free_vmap_area_root) |
635 | rb_erase_augmented(&va->rb_node, | |
636 | root, &free_vmap_area_rb_augment_cb); | |
637 | else | |
638 | rb_erase(&va->rb_node, root); | |
639 | ||
640 | list_del(&va->list); | |
641 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
642 | } |
643 | ||
bb850f4d URS |
644 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
645 | static void | |
da27c9ed | 646 | augment_tree_propagate_check(void) |
bb850f4d URS |
647 | { |
648 | struct vmap_area *va; | |
da27c9ed | 649 | unsigned long computed_size; |
bb850f4d | 650 | |
da27c9ed URS |
651 | list_for_each_entry(va, &free_vmap_area_list, list) { |
652 | computed_size = compute_subtree_max_size(va); | |
653 | if (computed_size != va->subtree_max_size) | |
654 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
655 | va_size(va), va->subtree_max_size); | |
bb850f4d | 656 | } |
bb850f4d URS |
657 | } |
658 | #endif | |
659 | ||
68ad4a33 URS |
660 | /* |
661 | * This function populates subtree_max_size from bottom to upper | |
662 | * levels starting from VA point. The propagation must be done | |
663 | * when VA size is modified by changing its va_start/va_end. Or | |
664 | * in case of newly inserting of VA to the tree. | |
665 | * | |
666 | * It means that __augment_tree_propagate_from() must be called: | |
667 | * - After VA has been inserted to the tree(free path); | |
668 | * - After VA has been shrunk(allocation path); | |
669 | * - After VA has been increased(merging path). | |
670 | * | |
671 | * Please note that, it does not mean that upper parent nodes | |
672 | * and their subtree_max_size are recalculated all the time up | |
673 | * to the root node. | |
674 | * | |
675 | * 4--8 | |
676 | * /\ | |
677 | * / \ | |
678 | * / \ | |
679 | * 2--2 8--8 | |
680 | * | |
681 | * For example if we modify the node 4, shrinking it to 2, then | |
682 | * no any modification is required. If we shrink the node 2 to 1 | |
683 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
684 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
685 | * node becomes 4--6. | |
686 | */ | |
687 | static __always_inline void | |
688 | augment_tree_propagate_from(struct vmap_area *va) | |
689 | { | |
15ae144f URS |
690 | /* |
691 | * Populate the tree from bottom towards the root until | |
692 | * the calculated maximum available size of checked node | |
693 | * is equal to its current one. | |
694 | */ | |
695 | free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL); | |
bb850f4d URS |
696 | |
697 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
da27c9ed | 698 | augment_tree_propagate_check(); |
bb850f4d | 699 | #endif |
68ad4a33 URS |
700 | } |
701 | ||
702 | static void | |
703 | insert_vmap_area(struct vmap_area *va, | |
704 | struct rb_root *root, struct list_head *head) | |
705 | { | |
706 | struct rb_node **link; | |
707 | struct rb_node *parent; | |
708 | ||
709 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
710 | if (link) |
711 | link_va(va, root, parent, link, head); | |
68ad4a33 URS |
712 | } |
713 | ||
714 | static void | |
715 | insert_vmap_area_augment(struct vmap_area *va, | |
716 | struct rb_node *from, struct rb_root *root, | |
717 | struct list_head *head) | |
718 | { | |
719 | struct rb_node **link; | |
720 | struct rb_node *parent; | |
721 | ||
722 | if (from) | |
723 | link = find_va_links(va, NULL, from, &parent); | |
724 | else | |
725 | link = find_va_links(va, root, NULL, &parent); | |
726 | ||
9c801f61 URS |
727 | if (link) { |
728 | link_va(va, root, parent, link, head); | |
729 | augment_tree_propagate_from(va); | |
730 | } | |
68ad4a33 URS |
731 | } |
732 | ||
733 | /* | |
734 | * Merge de-allocated chunk of VA memory with previous | |
735 | * and next free blocks. If coalesce is not done a new | |
736 | * free area is inserted. If VA has been merged, it is | |
737 | * freed. | |
9c801f61 URS |
738 | * |
739 | * Please note, it can return NULL in case of overlap | |
740 | * ranges, followed by WARN() report. Despite it is a | |
741 | * buggy behaviour, a system can be alive and keep | |
742 | * ongoing. | |
68ad4a33 | 743 | */ |
3c5c3cfb | 744 | static __always_inline struct vmap_area * |
68ad4a33 URS |
745 | merge_or_add_vmap_area(struct vmap_area *va, |
746 | struct rb_root *root, struct list_head *head) | |
747 | { | |
748 | struct vmap_area *sibling; | |
749 | struct list_head *next; | |
750 | struct rb_node **link; | |
751 | struct rb_node *parent; | |
752 | bool merged = false; | |
753 | ||
754 | /* | |
755 | * Find a place in the tree where VA potentially will be | |
756 | * inserted, unless it is merged with its sibling/siblings. | |
757 | */ | |
758 | link = find_va_links(va, root, NULL, &parent); | |
9c801f61 URS |
759 | if (!link) |
760 | return NULL; | |
68ad4a33 URS |
761 | |
762 | /* | |
763 | * Get next node of VA to check if merging can be done. | |
764 | */ | |
765 | next = get_va_next_sibling(parent, link); | |
766 | if (unlikely(next == NULL)) | |
767 | goto insert; | |
768 | ||
769 | /* | |
770 | * start end | |
771 | * | | | |
772 | * |<------VA------>|<-----Next----->| | |
773 | * | | | |
774 | * start end | |
775 | */ | |
776 | if (next != head) { | |
777 | sibling = list_entry(next, struct vmap_area, list); | |
778 | if (sibling->va_start == va->va_end) { | |
779 | sibling->va_start = va->va_start; | |
780 | ||
68ad4a33 URS |
781 | /* Free vmap_area object. */ |
782 | kmem_cache_free(vmap_area_cachep, va); | |
783 | ||
784 | /* Point to the new merged area. */ | |
785 | va = sibling; | |
786 | merged = true; | |
787 | } | |
788 | } | |
789 | ||
790 | /* | |
791 | * start end | |
792 | * | | | |
793 | * |<-----Prev----->|<------VA------>| | |
794 | * | | | |
795 | * start end | |
796 | */ | |
797 | if (next->prev != head) { | |
798 | sibling = list_entry(next->prev, struct vmap_area, list); | |
799 | if (sibling->va_end == va->va_start) { | |
5dd78640 URS |
800 | /* |
801 | * If both neighbors are coalesced, it is important | |
802 | * to unlink the "next" node first, followed by merging | |
803 | * with "previous" one. Otherwise the tree might not be | |
804 | * fully populated if a sibling's augmented value is | |
805 | * "normalized" because of rotation operations. | |
806 | */ | |
54f63d9d URS |
807 | if (merged) |
808 | unlink_va(va, root); | |
68ad4a33 | 809 | |
5dd78640 URS |
810 | sibling->va_end = va->va_end; |
811 | ||
68ad4a33 URS |
812 | /* Free vmap_area object. */ |
813 | kmem_cache_free(vmap_area_cachep, va); | |
3c5c3cfb DA |
814 | |
815 | /* Point to the new merged area. */ | |
816 | va = sibling; | |
817 | merged = true; | |
68ad4a33 URS |
818 | } |
819 | } | |
820 | ||
821 | insert: | |
5dd78640 | 822 | if (!merged) |
68ad4a33 | 823 | link_va(va, root, parent, link, head); |
3c5c3cfb | 824 | |
5dd78640 URS |
825 | /* |
826 | * Last step is to check and update the tree. | |
827 | */ | |
828 | augment_tree_propagate_from(va); | |
3c5c3cfb | 829 | return va; |
68ad4a33 URS |
830 | } |
831 | ||
832 | static __always_inline bool | |
833 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
834 | unsigned long align, unsigned long vstart) | |
835 | { | |
836 | unsigned long nva_start_addr; | |
837 | ||
838 | if (va->va_start > vstart) | |
839 | nva_start_addr = ALIGN(va->va_start, align); | |
840 | else | |
841 | nva_start_addr = ALIGN(vstart, align); | |
842 | ||
843 | /* Can be overflowed due to big size or alignment. */ | |
844 | if (nva_start_addr + size < nva_start_addr || | |
845 | nva_start_addr < vstart) | |
846 | return false; | |
847 | ||
848 | return (nva_start_addr + size <= va->va_end); | |
849 | } | |
850 | ||
851 | /* | |
852 | * Find the first free block(lowest start address) in the tree, | |
853 | * that will accomplish the request corresponding to passing | |
854 | * parameters. | |
855 | */ | |
856 | static __always_inline struct vmap_area * | |
857 | find_vmap_lowest_match(unsigned long size, | |
858 | unsigned long align, unsigned long vstart) | |
859 | { | |
860 | struct vmap_area *va; | |
861 | struct rb_node *node; | |
862 | unsigned long length; | |
863 | ||
864 | /* Start from the root. */ | |
865 | node = free_vmap_area_root.rb_node; | |
866 | ||
867 | /* Adjust the search size for alignment overhead. */ | |
868 | length = size + align - 1; | |
869 | ||
870 | while (node) { | |
871 | va = rb_entry(node, struct vmap_area, rb_node); | |
872 | ||
873 | if (get_subtree_max_size(node->rb_left) >= length && | |
874 | vstart < va->va_start) { | |
875 | node = node->rb_left; | |
876 | } else { | |
877 | if (is_within_this_va(va, size, align, vstart)) | |
878 | return va; | |
879 | ||
880 | /* | |
881 | * Does not make sense to go deeper towards the right | |
882 | * sub-tree if it does not have a free block that is | |
883 | * equal or bigger to the requested search length. | |
884 | */ | |
885 | if (get_subtree_max_size(node->rb_right) >= length) { | |
886 | node = node->rb_right; | |
887 | continue; | |
888 | } | |
889 | ||
890 | /* | |
3806b041 | 891 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
892 | * that will satisfy the search criteria. It can happen |
893 | * only once due to "vstart" restriction. | |
894 | */ | |
895 | while ((node = rb_parent(node))) { | |
896 | va = rb_entry(node, struct vmap_area, rb_node); | |
897 | if (is_within_this_va(va, size, align, vstart)) | |
898 | return va; | |
899 | ||
900 | if (get_subtree_max_size(node->rb_right) >= length && | |
901 | vstart <= va->va_start) { | |
902 | node = node->rb_right; | |
903 | break; | |
904 | } | |
905 | } | |
906 | } | |
907 | } | |
908 | ||
909 | return NULL; | |
910 | } | |
911 | ||
a6cf4e0f URS |
912 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
913 | #include <linux/random.h> | |
914 | ||
915 | static struct vmap_area * | |
916 | find_vmap_lowest_linear_match(unsigned long size, | |
917 | unsigned long align, unsigned long vstart) | |
918 | { | |
919 | struct vmap_area *va; | |
920 | ||
921 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
922 | if (!is_within_this_va(va, size, align, vstart)) | |
923 | continue; | |
924 | ||
925 | return va; | |
926 | } | |
927 | ||
928 | return NULL; | |
929 | } | |
930 | ||
931 | static void | |
932 | find_vmap_lowest_match_check(unsigned long size) | |
933 | { | |
934 | struct vmap_area *va_1, *va_2; | |
935 | unsigned long vstart; | |
936 | unsigned int rnd; | |
937 | ||
938 | get_random_bytes(&rnd, sizeof(rnd)); | |
939 | vstart = VMALLOC_START + rnd; | |
940 | ||
941 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
942 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
943 | ||
944 | if (va_1 != va_2) | |
945 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
946 | va_1, va_2, vstart); | |
947 | } | |
948 | #endif | |
949 | ||
68ad4a33 URS |
950 | enum fit_type { |
951 | NOTHING_FIT = 0, | |
952 | FL_FIT_TYPE = 1, /* full fit */ | |
953 | LE_FIT_TYPE = 2, /* left edge fit */ | |
954 | RE_FIT_TYPE = 3, /* right edge fit */ | |
955 | NE_FIT_TYPE = 4 /* no edge fit */ | |
956 | }; | |
957 | ||
958 | static __always_inline enum fit_type | |
959 | classify_va_fit_type(struct vmap_area *va, | |
960 | unsigned long nva_start_addr, unsigned long size) | |
961 | { | |
962 | enum fit_type type; | |
963 | ||
964 | /* Check if it is within VA. */ | |
965 | if (nva_start_addr < va->va_start || | |
966 | nva_start_addr + size > va->va_end) | |
967 | return NOTHING_FIT; | |
968 | ||
969 | /* Now classify. */ | |
970 | if (va->va_start == nva_start_addr) { | |
971 | if (va->va_end == nva_start_addr + size) | |
972 | type = FL_FIT_TYPE; | |
973 | else | |
974 | type = LE_FIT_TYPE; | |
975 | } else if (va->va_end == nva_start_addr + size) { | |
976 | type = RE_FIT_TYPE; | |
977 | } else { | |
978 | type = NE_FIT_TYPE; | |
979 | } | |
980 | ||
981 | return type; | |
982 | } | |
983 | ||
984 | static __always_inline int | |
985 | adjust_va_to_fit_type(struct vmap_area *va, | |
986 | unsigned long nva_start_addr, unsigned long size, | |
987 | enum fit_type type) | |
988 | { | |
2c929233 | 989 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
990 | |
991 | if (type == FL_FIT_TYPE) { | |
992 | /* | |
993 | * No need to split VA, it fully fits. | |
994 | * | |
995 | * | | | |
996 | * V NVA V | |
997 | * |---------------| | |
998 | */ | |
999 | unlink_va(va, &free_vmap_area_root); | |
1000 | kmem_cache_free(vmap_area_cachep, va); | |
1001 | } else if (type == LE_FIT_TYPE) { | |
1002 | /* | |
1003 | * Split left edge of fit VA. | |
1004 | * | |
1005 | * | | | |
1006 | * V NVA V R | |
1007 | * |-------|-------| | |
1008 | */ | |
1009 | va->va_start += size; | |
1010 | } else if (type == RE_FIT_TYPE) { | |
1011 | /* | |
1012 | * Split right edge of fit VA. | |
1013 | * | |
1014 | * | | | |
1015 | * L V NVA V | |
1016 | * |-------|-------| | |
1017 | */ | |
1018 | va->va_end = nva_start_addr; | |
1019 | } else if (type == NE_FIT_TYPE) { | |
1020 | /* | |
1021 | * Split no edge of fit VA. | |
1022 | * | |
1023 | * | | | |
1024 | * L V NVA V R | |
1025 | * |---|-------|---| | |
1026 | */ | |
82dd23e8 URS |
1027 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
1028 | if (unlikely(!lva)) { | |
1029 | /* | |
1030 | * For percpu allocator we do not do any pre-allocation | |
1031 | * and leave it as it is. The reason is it most likely | |
1032 | * never ends up with NE_FIT_TYPE splitting. In case of | |
1033 | * percpu allocations offsets and sizes are aligned to | |
1034 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
1035 | * are its main fitting cases. | |
1036 | * | |
1037 | * There are a few exceptions though, as an example it is | |
1038 | * a first allocation (early boot up) when we have "one" | |
1039 | * big free space that has to be split. | |
060650a2 URS |
1040 | * |
1041 | * Also we can hit this path in case of regular "vmap" | |
1042 | * allocations, if "this" current CPU was not preloaded. | |
1043 | * See the comment in alloc_vmap_area() why. If so, then | |
1044 | * GFP_NOWAIT is used instead to get an extra object for | |
1045 | * split purpose. That is rare and most time does not | |
1046 | * occur. | |
1047 | * | |
1048 | * What happens if an allocation gets failed. Basically, | |
1049 | * an "overflow" path is triggered to purge lazily freed | |
1050 | * areas to free some memory, then, the "retry" path is | |
1051 | * triggered to repeat one more time. See more details | |
1052 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
1053 | */ |
1054 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
1055 | if (!lva) | |
1056 | return -1; | |
1057 | } | |
68ad4a33 URS |
1058 | |
1059 | /* | |
1060 | * Build the remainder. | |
1061 | */ | |
1062 | lva->va_start = va->va_start; | |
1063 | lva->va_end = nva_start_addr; | |
1064 | ||
1065 | /* | |
1066 | * Shrink this VA to remaining size. | |
1067 | */ | |
1068 | va->va_start = nva_start_addr + size; | |
1069 | } else { | |
1070 | return -1; | |
1071 | } | |
1072 | ||
1073 | if (type != FL_FIT_TYPE) { | |
1074 | augment_tree_propagate_from(va); | |
1075 | ||
2c929233 | 1076 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
1077 | insert_vmap_area_augment(lva, &va->rb_node, |
1078 | &free_vmap_area_root, &free_vmap_area_list); | |
1079 | } | |
1080 | ||
1081 | return 0; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Returns a start address of the newly allocated area, if success. | |
1086 | * Otherwise a vend is returned that indicates failure. | |
1087 | */ | |
1088 | static __always_inline unsigned long | |
1089 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1090 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1091 | { |
1092 | unsigned long nva_start_addr; | |
1093 | struct vmap_area *va; | |
1094 | enum fit_type type; | |
1095 | int ret; | |
1096 | ||
1097 | va = find_vmap_lowest_match(size, align, vstart); | |
1098 | if (unlikely(!va)) | |
1099 | return vend; | |
1100 | ||
1101 | if (va->va_start > vstart) | |
1102 | nva_start_addr = ALIGN(va->va_start, align); | |
1103 | else | |
1104 | nva_start_addr = ALIGN(vstart, align); | |
1105 | ||
1106 | /* Check the "vend" restriction. */ | |
1107 | if (nva_start_addr + size > vend) | |
1108 | return vend; | |
1109 | ||
1110 | /* Classify what we have found. */ | |
1111 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1112 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1113 | return vend; | |
1114 | ||
1115 | /* Update the free vmap_area. */ | |
1116 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1117 | if (ret) | |
1118 | return vend; | |
1119 | ||
a6cf4e0f URS |
1120 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1121 | find_vmap_lowest_match_check(size); | |
1122 | #endif | |
1123 | ||
68ad4a33 URS |
1124 | return nva_start_addr; |
1125 | } | |
4da56b99 | 1126 | |
d98c9e83 AR |
1127 | /* |
1128 | * Free a region of KVA allocated by alloc_vmap_area | |
1129 | */ | |
1130 | static void free_vmap_area(struct vmap_area *va) | |
1131 | { | |
1132 | /* | |
1133 | * Remove from the busy tree/list. | |
1134 | */ | |
1135 | spin_lock(&vmap_area_lock); | |
1136 | unlink_va(va, &vmap_area_root); | |
1137 | spin_unlock(&vmap_area_lock); | |
1138 | ||
1139 | /* | |
1140 | * Insert/Merge it back to the free tree/list. | |
1141 | */ | |
1142 | spin_lock(&free_vmap_area_lock); | |
1143 | merge_or_add_vmap_area(va, &free_vmap_area_root, &free_vmap_area_list); | |
1144 | spin_unlock(&free_vmap_area_lock); | |
1145 | } | |
1146 | ||
db64fe02 NP |
1147 | /* |
1148 | * Allocate a region of KVA of the specified size and alignment, within the | |
1149 | * vstart and vend. | |
1150 | */ | |
1151 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1152 | unsigned long align, | |
1153 | unsigned long vstart, unsigned long vend, | |
1154 | int node, gfp_t gfp_mask) | |
1155 | { | |
82dd23e8 | 1156 | struct vmap_area *va, *pva; |
1da177e4 | 1157 | unsigned long addr; |
db64fe02 | 1158 | int purged = 0; |
d98c9e83 | 1159 | int ret; |
db64fe02 | 1160 | |
7766970c | 1161 | BUG_ON(!size); |
891c49ab | 1162 | BUG_ON(offset_in_page(size)); |
89699605 | 1163 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1164 | |
68ad4a33 URS |
1165 | if (unlikely(!vmap_initialized)) |
1166 | return ERR_PTR(-EBUSY); | |
1167 | ||
5803ed29 | 1168 | might_sleep(); |
f07116d7 | 1169 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1170 | |
f07116d7 | 1171 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1172 | if (unlikely(!va)) |
1173 | return ERR_PTR(-ENOMEM); | |
1174 | ||
7f88f88f CM |
1175 | /* |
1176 | * Only scan the relevant parts containing pointers to other objects | |
1177 | * to avoid false negatives. | |
1178 | */ | |
f07116d7 | 1179 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1180 | |
db64fe02 | 1181 | retry: |
82dd23e8 | 1182 | /* |
81f1ba58 URS |
1183 | * Preload this CPU with one extra vmap_area object. It is used |
1184 | * when fit type of free area is NE_FIT_TYPE. Please note, it | |
1185 | * does not guarantee that an allocation occurs on a CPU that | |
1186 | * is preloaded, instead we minimize the case when it is not. | |
1187 | * It can happen because of cpu migration, because there is a | |
1188 | * race until the below spinlock is taken. | |
82dd23e8 URS |
1189 | * |
1190 | * The preload is done in non-atomic context, thus it allows us | |
1191 | * to use more permissive allocation masks to be more stable under | |
81f1ba58 URS |
1192 | * low memory condition and high memory pressure. In rare case, |
1193 | * if not preloaded, GFP_NOWAIT is used. | |
82dd23e8 | 1194 | * |
81f1ba58 | 1195 | * Set "pva" to NULL here, because of "retry" path. |
82dd23e8 | 1196 | */ |
81f1ba58 | 1197 | pva = NULL; |
82dd23e8 | 1198 | |
81f1ba58 URS |
1199 | if (!this_cpu_read(ne_fit_preload_node)) |
1200 | /* | |
1201 | * Even if it fails we do not really care about that. | |
1202 | * Just proceed as it is. If needed "overflow" path | |
1203 | * will refill the cache we allocate from. | |
1204 | */ | |
f07116d7 | 1205 | pva = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
82dd23e8 | 1206 | |
e36176be | 1207 | spin_lock(&free_vmap_area_lock); |
81f1ba58 URS |
1208 | |
1209 | if (pva && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) | |
1210 | kmem_cache_free(vmap_area_cachep, pva); | |
89699605 | 1211 | |
afd07389 | 1212 | /* |
68ad4a33 URS |
1213 | * If an allocation fails, the "vend" address is |
1214 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1215 | */ |
cacca6ba | 1216 | addr = __alloc_vmap_area(size, align, vstart, vend); |
e36176be URS |
1217 | spin_unlock(&free_vmap_area_lock); |
1218 | ||
68ad4a33 | 1219 | if (unlikely(addr == vend)) |
89699605 | 1220 | goto overflow; |
db64fe02 NP |
1221 | |
1222 | va->va_start = addr; | |
1223 | va->va_end = addr + size; | |
688fcbfc | 1224 | va->vm = NULL; |
68ad4a33 | 1225 | |
d98c9e83 | 1226 | |
e36176be URS |
1227 | spin_lock(&vmap_area_lock); |
1228 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
db64fe02 NP |
1229 | spin_unlock(&vmap_area_lock); |
1230 | ||
61e16557 | 1231 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1232 | BUG_ON(va->va_start < vstart); |
1233 | BUG_ON(va->va_end > vend); | |
1234 | ||
d98c9e83 AR |
1235 | ret = kasan_populate_vmalloc(addr, size); |
1236 | if (ret) { | |
1237 | free_vmap_area(va); | |
1238 | return ERR_PTR(ret); | |
1239 | } | |
1240 | ||
db64fe02 | 1241 | return va; |
89699605 NP |
1242 | |
1243 | overflow: | |
89699605 NP |
1244 | if (!purged) { |
1245 | purge_vmap_area_lazy(); | |
1246 | purged = 1; | |
1247 | goto retry; | |
1248 | } | |
4da56b99 CW |
1249 | |
1250 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1251 | unsigned long freed = 0; | |
1252 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1253 | if (freed > 0) { | |
1254 | purged = 0; | |
1255 | goto retry; | |
1256 | } | |
1257 | } | |
1258 | ||
03497d76 | 1259 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1260 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1261 | size); | |
68ad4a33 URS |
1262 | |
1263 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1264 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1265 | } |
1266 | ||
4da56b99 CW |
1267 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1268 | { | |
1269 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1270 | } | |
1271 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1272 | ||
1273 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1274 | { | |
1275 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1276 | } | |
1277 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1278 | ||
db64fe02 NP |
1279 | /* |
1280 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1281 | * before attempting to purge with a TLB flush. | |
1282 | * | |
1283 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1284 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1285 | * global TLB flushes that must be performed. It would seem natural to scale | |
1286 | * this number up linearly with the number of CPUs (because vmapping activity | |
1287 | * could also scale linearly with the number of CPUs), however it is likely | |
1288 | * that in practice, workloads might be constrained in other ways that mean | |
1289 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1290 | * conservative and not introduce a big latency on huge systems, so go with | |
1291 | * a less aggressive log scale. It will still be an improvement over the old | |
1292 | * code, and it will be simple to change the scale factor if we find that it | |
1293 | * becomes a problem on bigger systems. | |
1294 | */ | |
1295 | static unsigned long lazy_max_pages(void) | |
1296 | { | |
1297 | unsigned int log; | |
1298 | ||
1299 | log = fls(num_online_cpus()); | |
1300 | ||
1301 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1302 | } | |
1303 | ||
4d36e6f8 | 1304 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1305 | |
0574ecd1 CH |
1306 | /* |
1307 | * Serialize vmap purging. There is no actual criticial section protected | |
1308 | * by this look, but we want to avoid concurrent calls for performance | |
1309 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1310 | */ | |
f9e09977 | 1311 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1312 | |
02b709df NP |
1313 | /* for per-CPU blocks */ |
1314 | static void purge_fragmented_blocks_allcpus(void); | |
1315 | ||
3ee48b6a CW |
1316 | /* |
1317 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1318 | * immediately freed. | |
1319 | */ | |
1320 | void set_iounmap_nonlazy(void) | |
1321 | { | |
4d36e6f8 | 1322 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1323 | } |
1324 | ||
db64fe02 NP |
1325 | /* |
1326 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1327 | */ |
0574ecd1 | 1328 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1329 | { |
4d36e6f8 | 1330 | unsigned long resched_threshold; |
80c4bd7a | 1331 | struct llist_node *valist; |
db64fe02 | 1332 | struct vmap_area *va; |
cbb76676 | 1333 | struct vmap_area *n_va; |
db64fe02 | 1334 | |
0574ecd1 | 1335 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1336 | |
80c4bd7a | 1337 | valist = llist_del_all(&vmap_purge_list); |
68571be9 URS |
1338 | if (unlikely(valist == NULL)) |
1339 | return false; | |
1340 | ||
1341 | /* | |
1342 | * TODO: to calculate a flush range without looping. | |
1343 | * The list can be up to lazy_max_pages() elements. | |
1344 | */ | |
80c4bd7a | 1345 | llist_for_each_entry(va, valist, purge_list) { |
0574ecd1 CH |
1346 | if (va->va_start < start) |
1347 | start = va->va_start; | |
1348 | if (va->va_end > end) | |
1349 | end = va->va_end; | |
db64fe02 | 1350 | } |
db64fe02 | 1351 | |
0574ecd1 | 1352 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1353 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1354 | |
e36176be | 1355 | spin_lock(&free_vmap_area_lock); |
763b218d | 1356 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
4d36e6f8 | 1357 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
3c5c3cfb DA |
1358 | unsigned long orig_start = va->va_start; |
1359 | unsigned long orig_end = va->va_end; | |
763b218d | 1360 | |
dd3b8353 URS |
1361 | /* |
1362 | * Finally insert or merge lazily-freed area. It is | |
1363 | * detached and there is no need to "unlink" it from | |
1364 | * anything. | |
1365 | */ | |
3c5c3cfb DA |
1366 | va = merge_or_add_vmap_area(va, &free_vmap_area_root, |
1367 | &free_vmap_area_list); | |
1368 | ||
9c801f61 URS |
1369 | if (!va) |
1370 | continue; | |
1371 | ||
3c5c3cfb DA |
1372 | if (is_vmalloc_or_module_addr((void *)orig_start)) |
1373 | kasan_release_vmalloc(orig_start, orig_end, | |
1374 | va->va_start, va->va_end); | |
dd3b8353 | 1375 | |
4d36e6f8 | 1376 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1377 | |
4d36e6f8 | 1378 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
e36176be | 1379 | cond_resched_lock(&free_vmap_area_lock); |
763b218d | 1380 | } |
e36176be | 1381 | spin_unlock(&free_vmap_area_lock); |
0574ecd1 | 1382 | return true; |
db64fe02 NP |
1383 | } |
1384 | ||
496850e5 NP |
1385 | /* |
1386 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1387 | * is already purging. | |
1388 | */ | |
1389 | static void try_purge_vmap_area_lazy(void) | |
1390 | { | |
f9e09977 | 1391 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1392 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1393 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1394 | } |
496850e5 NP |
1395 | } |
1396 | ||
db64fe02 NP |
1397 | /* |
1398 | * Kick off a purge of the outstanding lazy areas. | |
1399 | */ | |
1400 | static void purge_vmap_area_lazy(void) | |
1401 | { | |
f9e09977 | 1402 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1403 | purge_fragmented_blocks_allcpus(); |
1404 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1405 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1406 | } |
1407 | ||
1408 | /* | |
64141da5 JF |
1409 | * Free a vmap area, caller ensuring that the area has been unmapped |
1410 | * and flush_cache_vunmap had been called for the correct range | |
1411 | * previously. | |
db64fe02 | 1412 | */ |
64141da5 | 1413 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1414 | { |
4d36e6f8 | 1415 | unsigned long nr_lazy; |
80c4bd7a | 1416 | |
dd3b8353 URS |
1417 | spin_lock(&vmap_area_lock); |
1418 | unlink_va(va, &vmap_area_root); | |
1419 | spin_unlock(&vmap_area_lock); | |
1420 | ||
4d36e6f8 URS |
1421 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1422 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a CW |
1423 | |
1424 | /* After this point, we may free va at any time */ | |
1425 | llist_add(&va->purge_list, &vmap_purge_list); | |
1426 | ||
1427 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 1428 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1429 | } |
1430 | ||
b29acbdc NP |
1431 | /* |
1432 | * Free and unmap a vmap area | |
1433 | */ | |
1434 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1435 | { | |
1436 | flush_cache_vunmap(va->va_start, va->va_end); | |
855e57a1 | 1437 | unmap_kernel_range_noflush(va->va_start, va->va_end - va->va_start); |
8e57f8ac | 1438 | if (debug_pagealloc_enabled_static()) |
82a2e924 CP |
1439 | flush_tlb_kernel_range(va->va_start, va->va_end); |
1440 | ||
c8eef01e | 1441 | free_vmap_area_noflush(va); |
b29acbdc NP |
1442 | } |
1443 | ||
db64fe02 NP |
1444 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1445 | { | |
1446 | struct vmap_area *va; | |
1447 | ||
1448 | spin_lock(&vmap_area_lock); | |
1449 | va = __find_vmap_area(addr); | |
1450 | spin_unlock(&vmap_area_lock); | |
1451 | ||
1452 | return va; | |
1453 | } | |
1454 | ||
db64fe02 NP |
1455 | /*** Per cpu kva allocator ***/ |
1456 | ||
1457 | /* | |
1458 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1459 | * room for at least 16 percpu vmap blocks per CPU. | |
1460 | */ | |
1461 | /* | |
1462 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1463 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1464 | * instead (we just need a rough idea) | |
1465 | */ | |
1466 | #if BITS_PER_LONG == 32 | |
1467 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1468 | #else | |
1469 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1470 | #endif | |
1471 | ||
1472 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1473 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1474 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1475 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1476 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1477 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1478 | #define VMAP_BBMAP_BITS \ |
1479 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1480 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1481 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1482 | |
1483 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1484 | ||
1485 | struct vmap_block_queue { | |
1486 | spinlock_t lock; | |
1487 | struct list_head free; | |
db64fe02 NP |
1488 | }; |
1489 | ||
1490 | struct vmap_block { | |
1491 | spinlock_t lock; | |
1492 | struct vmap_area *va; | |
db64fe02 | 1493 | unsigned long free, dirty; |
7d61bfe8 | 1494 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1495 | struct list_head free_list; |
1496 | struct rcu_head rcu_head; | |
02b709df | 1497 | struct list_head purge; |
db64fe02 NP |
1498 | }; |
1499 | ||
1500 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1501 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1502 | ||
1503 | /* | |
0f14599c | 1504 | * XArray of vmap blocks, indexed by address, to quickly find a vmap block |
db64fe02 NP |
1505 | * in the free path. Could get rid of this if we change the API to return a |
1506 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1507 | */ | |
0f14599c | 1508 | static DEFINE_XARRAY(vmap_blocks); |
db64fe02 NP |
1509 | |
1510 | /* | |
1511 | * We should probably have a fallback mechanism to allocate virtual memory | |
1512 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1513 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1514 | * big problem. | |
1515 | */ | |
1516 | ||
1517 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1518 | { | |
1519 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1520 | addr /= VMAP_BLOCK_SIZE; | |
1521 | return addr; | |
1522 | } | |
1523 | ||
cf725ce2 RP |
1524 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1525 | { | |
1526 | unsigned long addr; | |
1527 | ||
1528 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1529 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1530 | return (void *)addr; | |
1531 | } | |
1532 | ||
1533 | /** | |
1534 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1535 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1536 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1537 | * @gfp_mask: flags for the page level allocator | |
1538 | * | |
a862f68a | 1539 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1540 | */ |
1541 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1542 | { |
1543 | struct vmap_block_queue *vbq; | |
1544 | struct vmap_block *vb; | |
1545 | struct vmap_area *va; | |
1546 | unsigned long vb_idx; | |
1547 | int node, err; | |
cf725ce2 | 1548 | void *vaddr; |
db64fe02 NP |
1549 | |
1550 | node = numa_node_id(); | |
1551 | ||
1552 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1553 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1554 | if (unlikely(!vb)) | |
1555 | return ERR_PTR(-ENOMEM); | |
1556 | ||
1557 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1558 | VMALLOC_START, VMALLOC_END, | |
1559 | node, gfp_mask); | |
ddf9c6d4 | 1560 | if (IS_ERR(va)) { |
db64fe02 | 1561 | kfree(vb); |
e7d86340 | 1562 | return ERR_CAST(va); |
db64fe02 NP |
1563 | } |
1564 | ||
cf725ce2 | 1565 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1566 | spin_lock_init(&vb->lock); |
1567 | vb->va = va; | |
cf725ce2 RP |
1568 | /* At least something should be left free */ |
1569 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1570 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1571 | vb->dirty = 0; |
7d61bfe8 RP |
1572 | vb->dirty_min = VMAP_BBMAP_BITS; |
1573 | vb->dirty_max = 0; | |
db64fe02 | 1574 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1575 | |
1576 | vb_idx = addr_to_vb_idx(va->va_start); | |
0f14599c MWO |
1577 | err = xa_insert(&vmap_blocks, vb_idx, vb, gfp_mask); |
1578 | if (err) { | |
1579 | kfree(vb); | |
1580 | free_vmap_area(va); | |
1581 | return ERR_PTR(err); | |
1582 | } | |
db64fe02 NP |
1583 | |
1584 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1585 | spin_lock(&vbq->lock); |
68ac546f | 1586 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1587 | spin_unlock(&vbq->lock); |
3f04ba85 | 1588 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1589 | |
cf725ce2 | 1590 | return vaddr; |
db64fe02 NP |
1591 | } |
1592 | ||
db64fe02 NP |
1593 | static void free_vmap_block(struct vmap_block *vb) |
1594 | { | |
1595 | struct vmap_block *tmp; | |
db64fe02 | 1596 | |
0f14599c | 1597 | tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start)); |
db64fe02 NP |
1598 | BUG_ON(tmp != vb); |
1599 | ||
64141da5 | 1600 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1601 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1602 | } |
1603 | ||
02b709df NP |
1604 | static void purge_fragmented_blocks(int cpu) |
1605 | { | |
1606 | LIST_HEAD(purge); | |
1607 | struct vmap_block *vb; | |
1608 | struct vmap_block *n_vb; | |
1609 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1610 | ||
1611 | rcu_read_lock(); | |
1612 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1613 | ||
1614 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1615 | continue; | |
1616 | ||
1617 | spin_lock(&vb->lock); | |
1618 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1619 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1620 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1621 | vb->dirty_min = 0; |
1622 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1623 | spin_lock(&vbq->lock); |
1624 | list_del_rcu(&vb->free_list); | |
1625 | spin_unlock(&vbq->lock); | |
1626 | spin_unlock(&vb->lock); | |
1627 | list_add_tail(&vb->purge, &purge); | |
1628 | } else | |
1629 | spin_unlock(&vb->lock); | |
1630 | } | |
1631 | rcu_read_unlock(); | |
1632 | ||
1633 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1634 | list_del(&vb->purge); | |
1635 | free_vmap_block(vb); | |
1636 | } | |
1637 | } | |
1638 | ||
02b709df NP |
1639 | static void purge_fragmented_blocks_allcpus(void) |
1640 | { | |
1641 | int cpu; | |
1642 | ||
1643 | for_each_possible_cpu(cpu) | |
1644 | purge_fragmented_blocks(cpu); | |
1645 | } | |
1646 | ||
db64fe02 NP |
1647 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1648 | { | |
1649 | struct vmap_block_queue *vbq; | |
1650 | struct vmap_block *vb; | |
cf725ce2 | 1651 | void *vaddr = NULL; |
db64fe02 NP |
1652 | unsigned int order; |
1653 | ||
891c49ab | 1654 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1655 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1656 | if (WARN_ON(size == 0)) { |
1657 | /* | |
1658 | * Allocating 0 bytes isn't what caller wants since | |
1659 | * get_order(0) returns funny result. Just warn and terminate | |
1660 | * early. | |
1661 | */ | |
1662 | return NULL; | |
1663 | } | |
db64fe02 NP |
1664 | order = get_order(size); |
1665 | ||
db64fe02 NP |
1666 | rcu_read_lock(); |
1667 | vbq = &get_cpu_var(vmap_block_queue); | |
1668 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1669 | unsigned long pages_off; |
db64fe02 NP |
1670 | |
1671 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1672 | if (vb->free < (1UL << order)) { |
1673 | spin_unlock(&vb->lock); | |
1674 | continue; | |
1675 | } | |
02b709df | 1676 | |
cf725ce2 RP |
1677 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1678 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1679 | vb->free -= 1UL << order; |
1680 | if (vb->free == 0) { | |
1681 | spin_lock(&vbq->lock); | |
1682 | list_del_rcu(&vb->free_list); | |
1683 | spin_unlock(&vbq->lock); | |
1684 | } | |
cf725ce2 | 1685 | |
02b709df NP |
1686 | spin_unlock(&vb->lock); |
1687 | break; | |
db64fe02 | 1688 | } |
02b709df | 1689 | |
3f04ba85 | 1690 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1691 | rcu_read_unlock(); |
1692 | ||
cf725ce2 RP |
1693 | /* Allocate new block if nothing was found */ |
1694 | if (!vaddr) | |
1695 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1696 | |
cf725ce2 | 1697 | return vaddr; |
db64fe02 NP |
1698 | } |
1699 | ||
78a0e8c4 | 1700 | static void vb_free(unsigned long addr, unsigned long size) |
db64fe02 NP |
1701 | { |
1702 | unsigned long offset; | |
db64fe02 NP |
1703 | unsigned int order; |
1704 | struct vmap_block *vb; | |
1705 | ||
891c49ab | 1706 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1707 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc | 1708 | |
78a0e8c4 | 1709 | flush_cache_vunmap(addr, addr + size); |
b29acbdc | 1710 | |
db64fe02 | 1711 | order = get_order(size); |
78a0e8c4 | 1712 | offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT; |
0f14599c | 1713 | vb = xa_load(&vmap_blocks, addr_to_vb_idx(addr)); |
db64fe02 | 1714 | |
b521c43f | 1715 | unmap_kernel_range_noflush(addr, size); |
64141da5 | 1716 | |
8e57f8ac | 1717 | if (debug_pagealloc_enabled_static()) |
78a0e8c4 | 1718 | flush_tlb_kernel_range(addr, addr + size); |
82a2e924 | 1719 | |
db64fe02 | 1720 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1721 | |
1722 | /* Expand dirty range */ | |
1723 | vb->dirty_min = min(vb->dirty_min, offset); | |
1724 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1725 | |
db64fe02 NP |
1726 | vb->dirty += 1UL << order; |
1727 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1728 | BUG_ON(vb->free); |
db64fe02 NP |
1729 | spin_unlock(&vb->lock); |
1730 | free_vmap_block(vb); | |
1731 | } else | |
1732 | spin_unlock(&vb->lock); | |
1733 | } | |
1734 | ||
868b104d | 1735 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1736 | { |
db64fe02 | 1737 | int cpu; |
db64fe02 | 1738 | |
9b463334 JF |
1739 | if (unlikely(!vmap_initialized)) |
1740 | return; | |
1741 | ||
5803ed29 CH |
1742 | might_sleep(); |
1743 | ||
db64fe02 NP |
1744 | for_each_possible_cpu(cpu) { |
1745 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1746 | struct vmap_block *vb; | |
1747 | ||
1748 | rcu_read_lock(); | |
1749 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1750 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1751 | if (vb->dirty) { |
1752 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1753 | unsigned long s, e; |
b136be5e | 1754 | |
7d61bfe8 RP |
1755 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1756 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1757 | |
7d61bfe8 RP |
1758 | start = min(s, start); |
1759 | end = max(e, end); | |
db64fe02 | 1760 | |
7d61bfe8 | 1761 | flush = 1; |
db64fe02 NP |
1762 | } |
1763 | spin_unlock(&vb->lock); | |
1764 | } | |
1765 | rcu_read_unlock(); | |
1766 | } | |
1767 | ||
f9e09977 | 1768 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1769 | purge_fragmented_blocks_allcpus(); |
1770 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1771 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1772 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 1773 | } |
868b104d RE |
1774 | |
1775 | /** | |
1776 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1777 | * | |
1778 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1779 | * to amortize TLB flushing overheads. What this means is that any page you | |
1780 | * have now, may, in a former life, have been mapped into kernel virtual | |
1781 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1782 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1783 | * | |
1784 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1785 | * be sure that none of the pages we have control over will have any aliases | |
1786 | * from the vmap layer. | |
1787 | */ | |
1788 | void vm_unmap_aliases(void) | |
1789 | { | |
1790 | unsigned long start = ULONG_MAX, end = 0; | |
1791 | int flush = 0; | |
1792 | ||
1793 | _vm_unmap_aliases(start, end, flush); | |
1794 | } | |
db64fe02 NP |
1795 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
1796 | ||
1797 | /** | |
1798 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1799 | * @mem: the pointer returned by vm_map_ram | |
1800 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1801 | */ | |
1802 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1803 | { | |
65ee03c4 | 1804 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1805 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1806 | struct vmap_area *va; |
db64fe02 | 1807 | |
5803ed29 | 1808 | might_sleep(); |
db64fe02 NP |
1809 | BUG_ON(!addr); |
1810 | BUG_ON(addr < VMALLOC_START); | |
1811 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1812 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 1813 | |
d98c9e83 AR |
1814 | kasan_poison_vmalloc(mem, size); |
1815 | ||
9c3acf60 | 1816 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 1817 | debug_check_no_locks_freed(mem, size); |
78a0e8c4 | 1818 | vb_free(addr, size); |
9c3acf60 CH |
1819 | return; |
1820 | } | |
1821 | ||
1822 | va = find_vmap_area(addr); | |
1823 | BUG_ON(!va); | |
05e3ff95 CP |
1824 | debug_check_no_locks_freed((void *)va->va_start, |
1825 | (va->va_end - va->va_start)); | |
9c3acf60 | 1826 | free_unmap_vmap_area(va); |
db64fe02 NP |
1827 | } |
1828 | EXPORT_SYMBOL(vm_unmap_ram); | |
1829 | ||
1830 | /** | |
1831 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1832 | * @pages: an array of pointers to the pages to be mapped | |
1833 | * @count: number of pages | |
1834 | * @node: prefer to allocate data structures on this node | |
e99c97ad | 1835 | * |
36437638 GK |
1836 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1837 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1838 | * objects with vm_map_ram(), it could consume lots of address space through | |
1839 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1840 | * the end. Please use this function for short-lived objects. | |
1841 | * | |
e99c97ad | 1842 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 | 1843 | */ |
d4efd79a | 1844 | void *vm_map_ram(struct page **pages, unsigned int count, int node) |
db64fe02 | 1845 | { |
65ee03c4 | 1846 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1847 | unsigned long addr; |
1848 | void *mem; | |
1849 | ||
1850 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1851 | mem = vb_alloc(size, GFP_KERNEL); | |
1852 | if (IS_ERR(mem)) | |
1853 | return NULL; | |
1854 | addr = (unsigned long)mem; | |
1855 | } else { | |
1856 | struct vmap_area *va; | |
1857 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1858 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1859 | if (IS_ERR(va)) | |
1860 | return NULL; | |
1861 | ||
1862 | addr = va->va_start; | |
1863 | mem = (void *)addr; | |
1864 | } | |
d98c9e83 AR |
1865 | |
1866 | kasan_unpoison_vmalloc(mem, size); | |
1867 | ||
d4efd79a | 1868 | if (map_kernel_range(addr, size, PAGE_KERNEL, pages) < 0) { |
db64fe02 NP |
1869 | vm_unmap_ram(mem, count); |
1870 | return NULL; | |
1871 | } | |
1872 | return mem; | |
1873 | } | |
1874 | EXPORT_SYMBOL(vm_map_ram); | |
1875 | ||
4341fa45 | 1876 | static struct vm_struct *vmlist __initdata; |
92eac168 | 1877 | |
be9b7335 NP |
1878 | /** |
1879 | * vm_area_add_early - add vmap area early during boot | |
1880 | * @vm: vm_struct to add | |
1881 | * | |
1882 | * This function is used to add fixed kernel vm area to vmlist before | |
1883 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1884 | * should contain proper values and the other fields should be zero. | |
1885 | * | |
1886 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1887 | */ | |
1888 | void __init vm_area_add_early(struct vm_struct *vm) | |
1889 | { | |
1890 | struct vm_struct *tmp, **p; | |
1891 | ||
1892 | BUG_ON(vmap_initialized); | |
1893 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1894 | if (tmp->addr >= vm->addr) { | |
1895 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1896 | break; | |
1897 | } else | |
1898 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1899 | } | |
1900 | vm->next = *p; | |
1901 | *p = vm; | |
1902 | } | |
1903 | ||
f0aa6617 TH |
1904 | /** |
1905 | * vm_area_register_early - register vmap area early during boot | |
1906 | * @vm: vm_struct to register | |
c0c0a293 | 1907 | * @align: requested alignment |
f0aa6617 TH |
1908 | * |
1909 | * This function is used to register kernel vm area before | |
1910 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1911 | * proper values on entry and other fields should be zero. On return, | |
1912 | * vm->addr contains the allocated address. | |
1913 | * | |
1914 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1915 | */ | |
c0c0a293 | 1916 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1917 | { |
1918 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1919 | unsigned long addr; |
1920 | ||
1921 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1922 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1923 | |
c0c0a293 | 1924 | vm->addr = (void *)addr; |
f0aa6617 | 1925 | |
be9b7335 | 1926 | vm_area_add_early(vm); |
f0aa6617 TH |
1927 | } |
1928 | ||
68ad4a33 URS |
1929 | static void vmap_init_free_space(void) |
1930 | { | |
1931 | unsigned long vmap_start = 1; | |
1932 | const unsigned long vmap_end = ULONG_MAX; | |
1933 | struct vmap_area *busy, *free; | |
1934 | ||
1935 | /* | |
1936 | * B F B B B F | |
1937 | * -|-----|.....|-----|-----|-----|.....|- | |
1938 | * | The KVA space | | |
1939 | * |<--------------------------------->| | |
1940 | */ | |
1941 | list_for_each_entry(busy, &vmap_area_list, list) { | |
1942 | if (busy->va_start - vmap_start > 0) { | |
1943 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1944 | if (!WARN_ON_ONCE(!free)) { | |
1945 | free->va_start = vmap_start; | |
1946 | free->va_end = busy->va_start; | |
1947 | ||
1948 | insert_vmap_area_augment(free, NULL, | |
1949 | &free_vmap_area_root, | |
1950 | &free_vmap_area_list); | |
1951 | } | |
1952 | } | |
1953 | ||
1954 | vmap_start = busy->va_end; | |
1955 | } | |
1956 | ||
1957 | if (vmap_end - vmap_start > 0) { | |
1958 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1959 | if (!WARN_ON_ONCE(!free)) { | |
1960 | free->va_start = vmap_start; | |
1961 | free->va_end = vmap_end; | |
1962 | ||
1963 | insert_vmap_area_augment(free, NULL, | |
1964 | &free_vmap_area_root, | |
1965 | &free_vmap_area_list); | |
1966 | } | |
1967 | } | |
1968 | } | |
1969 | ||
db64fe02 NP |
1970 | void __init vmalloc_init(void) |
1971 | { | |
822c18f2 IK |
1972 | struct vmap_area *va; |
1973 | struct vm_struct *tmp; | |
db64fe02 NP |
1974 | int i; |
1975 | ||
68ad4a33 URS |
1976 | /* |
1977 | * Create the cache for vmap_area objects. | |
1978 | */ | |
1979 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
1980 | ||
db64fe02 NP |
1981 | for_each_possible_cpu(i) { |
1982 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1983 | struct vfree_deferred *p; |
db64fe02 NP |
1984 | |
1985 | vbq = &per_cpu(vmap_block_queue, i); | |
1986 | spin_lock_init(&vbq->lock); | |
1987 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1988 | p = &per_cpu(vfree_deferred, i); |
1989 | init_llist_head(&p->list); | |
1990 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1991 | } |
9b463334 | 1992 | |
822c18f2 IK |
1993 | /* Import existing vmlist entries. */ |
1994 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
1995 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
1996 | if (WARN_ON_ONCE(!va)) | |
1997 | continue; | |
1998 | ||
822c18f2 IK |
1999 | va->va_start = (unsigned long)tmp->addr; |
2000 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 2001 | va->vm = tmp; |
68ad4a33 | 2002 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 2003 | } |
ca23e405 | 2004 | |
68ad4a33 URS |
2005 | /* |
2006 | * Now we can initialize a free vmap space. | |
2007 | */ | |
2008 | vmap_init_free_space(); | |
9b463334 | 2009 | vmap_initialized = true; |
db64fe02 NP |
2010 | } |
2011 | ||
8fc48985 TH |
2012 | /** |
2013 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
2014 | * @addr: start of the VM area to unmap | |
2015 | * @size: size of the VM area to unmap | |
2016 | * | |
2017 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
2018 | * the unmapping and tlb after. | |
2019 | */ | |
db64fe02 NP |
2020 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
2021 | { | |
2022 | unsigned long end = addr + size; | |
f6fcba70 TH |
2023 | |
2024 | flush_cache_vunmap(addr, end); | |
b521c43f | 2025 | unmap_kernel_range_noflush(addr, size); |
db64fe02 NP |
2026 | flush_tlb_kernel_range(addr, end); |
2027 | } | |
2028 | ||
e36176be URS |
2029 | static inline void setup_vmalloc_vm_locked(struct vm_struct *vm, |
2030 | struct vmap_area *va, unsigned long flags, const void *caller) | |
cf88c790 | 2031 | { |
cf88c790 TH |
2032 | vm->flags = flags; |
2033 | vm->addr = (void *)va->va_start; | |
2034 | vm->size = va->va_end - va->va_start; | |
2035 | vm->caller = caller; | |
db1aecaf | 2036 | va->vm = vm; |
e36176be URS |
2037 | } |
2038 | ||
2039 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, | |
2040 | unsigned long flags, const void *caller) | |
2041 | { | |
2042 | spin_lock(&vmap_area_lock); | |
2043 | setup_vmalloc_vm_locked(vm, va, flags, caller); | |
c69480ad | 2044 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2045 | } |
cf88c790 | 2046 | |
20fc02b4 | 2047 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2048 | { |
d4033afd | 2049 | /* |
20fc02b4 | 2050 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2051 | * we should make sure that vm has proper values. |
2052 | * Pair with smp_rmb() in show_numa_info(). | |
2053 | */ | |
2054 | smp_wmb(); | |
20fc02b4 | 2055 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2056 | } |
2057 | ||
db64fe02 | 2058 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2059 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2060 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2061 | { |
0006526d | 2062 | struct vmap_area *va; |
db64fe02 | 2063 | struct vm_struct *area; |
d98c9e83 | 2064 | unsigned long requested_size = size; |
1da177e4 | 2065 | |
52fd24ca | 2066 | BUG_ON(in_interrupt()); |
1da177e4 | 2067 | size = PAGE_ALIGN(size); |
31be8309 OH |
2068 | if (unlikely(!size)) |
2069 | return NULL; | |
1da177e4 | 2070 | |
252e5c6e | 2071 | if (flags & VM_IOREMAP) |
2072 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2073 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2074 | ||
cf88c790 | 2075 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2076 | if (unlikely(!area)) |
2077 | return NULL; | |
2078 | ||
71394fe5 AR |
2079 | if (!(flags & VM_NO_GUARD)) |
2080 | size += PAGE_SIZE; | |
1da177e4 | 2081 | |
db64fe02 NP |
2082 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2083 | if (IS_ERR(va)) { | |
2084 | kfree(area); | |
2085 | return NULL; | |
1da177e4 | 2086 | } |
1da177e4 | 2087 | |
d98c9e83 | 2088 | kasan_unpoison_vmalloc((void *)va->va_start, requested_size); |
f5252e00 | 2089 | |
d98c9e83 | 2090 | setup_vmalloc_vm(area, va, flags, caller); |
3c5c3cfb | 2091 | |
1da177e4 | 2092 | return area; |
1da177e4 LT |
2093 | } |
2094 | ||
c2968612 BH |
2095 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2096 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2097 | const void *caller) |
c2968612 | 2098 | { |
00ef2d2f DR |
2099 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2100 | GFP_KERNEL, caller); | |
c2968612 BH |
2101 | } |
2102 | ||
1da177e4 | 2103 | /** |
92eac168 MR |
2104 | * get_vm_area - reserve a contiguous kernel virtual area |
2105 | * @size: size of the area | |
2106 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2107 | * |
92eac168 MR |
2108 | * Search an area of @size in the kernel virtual mapping area, |
2109 | * and reserved it for out purposes. Returns the area descriptor | |
2110 | * on success or %NULL on failure. | |
a862f68a MR |
2111 | * |
2112 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2113 | */ |
2114 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2115 | { | |
2dca6999 | 2116 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2117 | NUMA_NO_NODE, GFP_KERNEL, |
2118 | __builtin_return_address(0)); | |
23016969 CL |
2119 | } |
2120 | ||
2121 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2122 | const void *caller) |
23016969 | 2123 | { |
2dca6999 | 2124 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2125 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2126 | } |
2127 | ||
e9da6e99 | 2128 | /** |
92eac168 MR |
2129 | * find_vm_area - find a continuous kernel virtual area |
2130 | * @addr: base address | |
e9da6e99 | 2131 | * |
92eac168 MR |
2132 | * Search for the kernel VM area starting at @addr, and return it. |
2133 | * It is up to the caller to do all required locking to keep the returned | |
2134 | * pointer valid. | |
a862f68a | 2135 | * |
74640617 | 2136 | * Return: the area descriptor on success or %NULL on failure. |
e9da6e99 MS |
2137 | */ |
2138 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2139 | { |
db64fe02 | 2140 | struct vmap_area *va; |
83342314 | 2141 | |
db64fe02 | 2142 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2143 | if (!va) |
2144 | return NULL; | |
1da177e4 | 2145 | |
688fcbfc | 2146 | return va->vm; |
1da177e4 LT |
2147 | } |
2148 | ||
7856dfeb | 2149 | /** |
92eac168 MR |
2150 | * remove_vm_area - find and remove a continuous kernel virtual area |
2151 | * @addr: base address | |
7856dfeb | 2152 | * |
92eac168 MR |
2153 | * Search for the kernel VM area starting at @addr, and remove it. |
2154 | * This function returns the found VM area, but using it is NOT safe | |
2155 | * on SMP machines, except for its size or flags. | |
a862f68a | 2156 | * |
74640617 | 2157 | * Return: the area descriptor on success or %NULL on failure. |
7856dfeb | 2158 | */ |
b3bdda02 | 2159 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2160 | { |
db64fe02 NP |
2161 | struct vmap_area *va; |
2162 | ||
5803ed29 CH |
2163 | might_sleep(); |
2164 | ||
dd3b8353 URS |
2165 | spin_lock(&vmap_area_lock); |
2166 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2167 | if (va && va->vm) { |
db1aecaf | 2168 | struct vm_struct *vm = va->vm; |
f5252e00 | 2169 | |
c69480ad | 2170 | va->vm = NULL; |
c69480ad JK |
2171 | spin_unlock(&vmap_area_lock); |
2172 | ||
a5af5aa8 | 2173 | kasan_free_shadow(vm); |
dd32c279 | 2174 | free_unmap_vmap_area(va); |
dd32c279 | 2175 | |
db64fe02 NP |
2176 | return vm; |
2177 | } | |
dd3b8353 URS |
2178 | |
2179 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2180 | return NULL; |
7856dfeb AK |
2181 | } |
2182 | ||
868b104d RE |
2183 | static inline void set_area_direct_map(const struct vm_struct *area, |
2184 | int (*set_direct_map)(struct page *page)) | |
2185 | { | |
2186 | int i; | |
2187 | ||
2188 | for (i = 0; i < area->nr_pages; i++) | |
2189 | if (page_address(area->pages[i])) | |
2190 | set_direct_map(area->pages[i]); | |
2191 | } | |
2192 | ||
2193 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2194 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2195 | { | |
868b104d RE |
2196 | unsigned long start = ULONG_MAX, end = 0; |
2197 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
31e67340 | 2198 | int flush_dmap = 0; |
868b104d RE |
2199 | int i; |
2200 | ||
868b104d RE |
2201 | remove_vm_area(area->addr); |
2202 | ||
2203 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2204 | if (!flush_reset) | |
2205 | return; | |
2206 | ||
2207 | /* | |
2208 | * If not deallocating pages, just do the flush of the VM area and | |
2209 | * return. | |
2210 | */ | |
2211 | if (!deallocate_pages) { | |
2212 | vm_unmap_aliases(); | |
2213 | return; | |
2214 | } | |
2215 | ||
2216 | /* | |
2217 | * If execution gets here, flush the vm mapping and reset the direct | |
2218 | * map. Find the start and end range of the direct mappings to make sure | |
2219 | * the vm_unmap_aliases() flush includes the direct map. | |
2220 | */ | |
2221 | for (i = 0; i < area->nr_pages; i++) { | |
8e41f872 RE |
2222 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2223 | if (addr) { | |
868b104d | 2224 | start = min(addr, start); |
8e41f872 | 2225 | end = max(addr + PAGE_SIZE, end); |
31e67340 | 2226 | flush_dmap = 1; |
868b104d RE |
2227 | } |
2228 | } | |
2229 | ||
2230 | /* | |
2231 | * Set direct map to something invalid so that it won't be cached if | |
2232 | * there are any accesses after the TLB flush, then flush the TLB and | |
2233 | * reset the direct map permissions to the default. | |
2234 | */ | |
2235 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2236 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2237 | set_area_direct_map(area, set_direct_map_default_noflush); |
2238 | } | |
2239 | ||
b3bdda02 | 2240 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2241 | { |
2242 | struct vm_struct *area; | |
2243 | ||
2244 | if (!addr) | |
2245 | return; | |
2246 | ||
e69e9d4a | 2247 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2248 | addr)) |
1da177e4 | 2249 | return; |
1da177e4 | 2250 | |
6ade2032 | 2251 | area = find_vm_area(addr); |
1da177e4 | 2252 | if (unlikely(!area)) { |
4c8573e2 | 2253 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2254 | addr); |
1da177e4 LT |
2255 | return; |
2256 | } | |
2257 | ||
05e3ff95 CP |
2258 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2259 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2260 | |
d98c9e83 | 2261 | kasan_poison_vmalloc(area->addr, area->size); |
3c5c3cfb | 2262 | |
868b104d RE |
2263 | vm_remove_mappings(area, deallocate_pages); |
2264 | ||
1da177e4 LT |
2265 | if (deallocate_pages) { |
2266 | int i; | |
2267 | ||
2268 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2269 | struct page *page = area->pages[i]; |
2270 | ||
2271 | BUG_ON(!page); | |
4949148a | 2272 | __free_pages(page, 0); |
1da177e4 | 2273 | } |
97105f0a | 2274 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2275 | |
244d63ee | 2276 | kvfree(area->pages); |
1da177e4 LT |
2277 | } |
2278 | ||
2279 | kfree(area); | |
2280 | return; | |
2281 | } | |
bf22e37a AR |
2282 | |
2283 | static inline void __vfree_deferred(const void *addr) | |
2284 | { | |
2285 | /* | |
2286 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2287 | * context. Preemption is absolutely fine here, because the llist_add() | |
2288 | * implementation is lockless, so it works even if we are adding to | |
73221d88 | 2289 | * another cpu's list. schedule_work() should be fine with this too. |
bf22e37a AR |
2290 | */ |
2291 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2292 | ||
2293 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2294 | schedule_work(&p->wq); | |
2295 | } | |
2296 | ||
2297 | /** | |
92eac168 MR |
2298 | * vfree_atomic - release memory allocated by vmalloc() |
2299 | * @addr: memory base address | |
bf22e37a | 2300 | * |
92eac168 MR |
2301 | * This one is just like vfree() but can be called in any atomic context |
2302 | * except NMIs. | |
bf22e37a AR |
2303 | */ |
2304 | void vfree_atomic(const void *addr) | |
2305 | { | |
2306 | BUG_ON(in_nmi()); | |
2307 | ||
2308 | kmemleak_free(addr); | |
2309 | ||
2310 | if (!addr) | |
2311 | return; | |
2312 | __vfree_deferred(addr); | |
2313 | } | |
2314 | ||
c67dc624 RP |
2315 | static void __vfree(const void *addr) |
2316 | { | |
2317 | if (unlikely(in_interrupt())) | |
2318 | __vfree_deferred(addr); | |
2319 | else | |
2320 | __vunmap(addr, 1); | |
2321 | } | |
2322 | ||
1da177e4 | 2323 | /** |
92eac168 MR |
2324 | * vfree - release memory allocated by vmalloc() |
2325 | * @addr: memory base address | |
1da177e4 | 2326 | * |
92eac168 MR |
2327 | * Free the virtually continuous memory area starting at @addr, as |
2328 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is | |
2329 | * NULL, no operation is performed. | |
1da177e4 | 2330 | * |
92eac168 MR |
2331 | * Must not be called in NMI context (strictly speaking, only if we don't |
2332 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2333 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 2334 | * |
92eac168 | 2335 | * May sleep if called *not* from interrupt context. |
3ca4ea3a | 2336 | * |
92eac168 | 2337 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 2338 | */ |
b3bdda02 | 2339 | void vfree(const void *addr) |
1da177e4 | 2340 | { |
32fcfd40 | 2341 | BUG_ON(in_nmi()); |
89219d37 CM |
2342 | |
2343 | kmemleak_free(addr); | |
2344 | ||
a8dda165 AR |
2345 | might_sleep_if(!in_interrupt()); |
2346 | ||
32fcfd40 AV |
2347 | if (!addr) |
2348 | return; | |
c67dc624 RP |
2349 | |
2350 | __vfree(addr); | |
1da177e4 | 2351 | } |
1da177e4 LT |
2352 | EXPORT_SYMBOL(vfree); |
2353 | ||
2354 | /** | |
92eac168 MR |
2355 | * vunmap - release virtual mapping obtained by vmap() |
2356 | * @addr: memory base address | |
1da177e4 | 2357 | * |
92eac168 MR |
2358 | * Free the virtually contiguous memory area starting at @addr, |
2359 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2360 | * |
92eac168 | 2361 | * Must not be called in interrupt context. |
1da177e4 | 2362 | */ |
b3bdda02 | 2363 | void vunmap(const void *addr) |
1da177e4 LT |
2364 | { |
2365 | BUG_ON(in_interrupt()); | |
34754b69 | 2366 | might_sleep(); |
32fcfd40 AV |
2367 | if (addr) |
2368 | __vunmap(addr, 0); | |
1da177e4 | 2369 | } |
1da177e4 LT |
2370 | EXPORT_SYMBOL(vunmap); |
2371 | ||
2372 | /** | |
92eac168 MR |
2373 | * vmap - map an array of pages into virtually contiguous space |
2374 | * @pages: array of page pointers | |
2375 | * @count: number of pages to map | |
2376 | * @flags: vm_area->flags | |
2377 | * @prot: page protection for the mapping | |
2378 | * | |
2379 | * Maps @count pages from @pages into contiguous kernel virtual | |
2380 | * space. | |
a862f68a MR |
2381 | * |
2382 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2383 | */ |
2384 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2385 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2386 | { |
2387 | struct vm_struct *area; | |
65ee03c4 | 2388 | unsigned long size; /* In bytes */ |
1da177e4 | 2389 | |
34754b69 PZ |
2390 | might_sleep(); |
2391 | ||
ca79b0c2 | 2392 | if (count > totalram_pages()) |
1da177e4 LT |
2393 | return NULL; |
2394 | ||
65ee03c4 GJM |
2395 | size = (unsigned long)count << PAGE_SHIFT; |
2396 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2397 | if (!area) |
2398 | return NULL; | |
23016969 | 2399 | |
cca98e9f | 2400 | if (map_kernel_range((unsigned long)area->addr, size, pgprot_nx(prot), |
ed1f324c | 2401 | pages) < 0) { |
1da177e4 LT |
2402 | vunmap(area->addr); |
2403 | return NULL; | |
2404 | } | |
2405 | ||
2406 | return area->addr; | |
2407 | } | |
1da177e4 LT |
2408 | EXPORT_SYMBOL(vmap); |
2409 | ||
e31d9eb5 | 2410 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 2411 | pgprot_t prot, int node) |
1da177e4 LT |
2412 | { |
2413 | struct page **pages; | |
2414 | unsigned int nr_pages, array_size, i; | |
930f036b | 2415 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
704b862f LA |
2416 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; |
2417 | const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ? | |
2418 | 0 : | |
2419 | __GFP_HIGHMEM; | |
1da177e4 | 2420 | |
762216ab | 2421 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
2422 | array_size = (nr_pages * sizeof(struct page *)); |
2423 | ||
1da177e4 | 2424 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 2425 | if (array_size > PAGE_SIZE) { |
704b862f | 2426 | pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask, |
f38fcb9c | 2427 | node, area->caller); |
286e1ea3 | 2428 | } else { |
976d6dfb | 2429 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2430 | } |
7ea36242 AK |
2431 | |
2432 | if (!pages) { | |
1da177e4 LT |
2433 | remove_vm_area(area->addr); |
2434 | kfree(area); | |
2435 | return NULL; | |
2436 | } | |
1da177e4 | 2437 | |
7ea36242 AK |
2438 | area->pages = pages; |
2439 | area->nr_pages = nr_pages; | |
2440 | ||
1da177e4 | 2441 | for (i = 0; i < area->nr_pages; i++) { |
bf53d6f8 CL |
2442 | struct page *page; |
2443 | ||
4b90951c | 2444 | if (node == NUMA_NO_NODE) |
704b862f | 2445 | page = alloc_page(alloc_mask|highmem_mask); |
930fc45a | 2446 | else |
704b862f | 2447 | page = alloc_pages_node(node, alloc_mask|highmem_mask, 0); |
bf53d6f8 CL |
2448 | |
2449 | if (unlikely(!page)) { | |
82afbc32 | 2450 | /* Successfully allocated i pages, free them in __vfree() */ |
1da177e4 | 2451 | area->nr_pages = i; |
97105f0a | 2452 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 LT |
2453 | goto fail; |
2454 | } | |
bf53d6f8 | 2455 | area->pages[i] = page; |
dcf61ff0 | 2456 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 2457 | cond_resched(); |
1da177e4 | 2458 | } |
97105f0a | 2459 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2460 | |
ed1f324c CH |
2461 | if (map_kernel_range((unsigned long)area->addr, get_vm_area_size(area), |
2462 | prot, pages) < 0) | |
1da177e4 | 2463 | goto fail; |
ed1f324c | 2464 | |
1da177e4 LT |
2465 | return area->addr; |
2466 | ||
2467 | fail: | |
a8e99259 | 2468 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2469 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 2470 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 2471 | __vfree(area->addr); |
1da177e4 LT |
2472 | return NULL; |
2473 | } | |
2474 | ||
2475 | /** | |
92eac168 MR |
2476 | * __vmalloc_node_range - allocate virtually contiguous memory |
2477 | * @size: allocation size | |
2478 | * @align: desired alignment | |
2479 | * @start: vm area range start | |
2480 | * @end: vm area range end | |
2481 | * @gfp_mask: flags for the page level allocator | |
2482 | * @prot: protection mask for the allocated pages | |
2483 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2484 | * @node: node to use for allocation or NUMA_NO_NODE | |
2485 | * @caller: caller's return address | |
2486 | * | |
2487 | * Allocate enough pages to cover @size from the page level | |
2488 | * allocator with @gfp_mask flags. Map them into contiguous | |
2489 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2490 | * |
2491 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2492 | */ |
d0a21265 DR |
2493 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2494 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2495 | pgprot_t prot, unsigned long vm_flags, int node, |
2496 | const void *caller) | |
1da177e4 LT |
2497 | { |
2498 | struct vm_struct *area; | |
89219d37 CM |
2499 | void *addr; |
2500 | unsigned long real_size = size; | |
1da177e4 LT |
2501 | |
2502 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 2503 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 2504 | goto fail; |
1da177e4 | 2505 | |
d98c9e83 | 2506 | area = __get_vm_area_node(real_size, align, VM_ALLOC | VM_UNINITIALIZED | |
cb9e3c29 | 2507 | vm_flags, start, end, node, gfp_mask, caller); |
1da177e4 | 2508 | if (!area) |
de7d2b56 | 2509 | goto fail; |
1da177e4 | 2510 | |
3722e13c | 2511 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 2512 | if (!addr) |
b82225f3 | 2513 | return NULL; |
89219d37 | 2514 | |
f5252e00 | 2515 | /* |
20fc02b4 ZY |
2516 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2517 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2518 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2519 | */ |
20fc02b4 | 2520 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2521 | |
94f4a161 | 2522 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2523 | |
2524 | return addr; | |
de7d2b56 JP |
2525 | |
2526 | fail: | |
a8e99259 | 2527 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2528 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 2529 | return NULL; |
1da177e4 LT |
2530 | } |
2531 | ||
d0a21265 | 2532 | /** |
92eac168 MR |
2533 | * __vmalloc_node - allocate virtually contiguous memory |
2534 | * @size: allocation size | |
2535 | * @align: desired alignment | |
2536 | * @gfp_mask: flags for the page level allocator | |
92eac168 MR |
2537 | * @node: node to use for allocation or NUMA_NO_NODE |
2538 | * @caller: caller's return address | |
a7c3e901 | 2539 | * |
f38fcb9c CH |
2540 | * Allocate enough pages to cover @size from the page level allocator with |
2541 | * @gfp_mask flags. Map them into contiguous kernel virtual space. | |
a7c3e901 | 2542 | * |
92eac168 MR |
2543 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2544 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2545 | * |
92eac168 MR |
2546 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2547 | * with mm people. | |
a862f68a MR |
2548 | * |
2549 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2550 | */ |
2b905948 | 2551 | void *__vmalloc_node(unsigned long size, unsigned long align, |
f38fcb9c | 2552 | gfp_t gfp_mask, int node, const void *caller) |
d0a21265 DR |
2553 | { |
2554 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
f38fcb9c | 2555 | gfp_mask, PAGE_KERNEL, 0, node, caller); |
d0a21265 | 2556 | } |
c3f896dc CH |
2557 | /* |
2558 | * This is only for performance analysis of vmalloc and stress purpose. | |
2559 | * It is required by vmalloc test module, therefore do not use it other | |
2560 | * than that. | |
2561 | */ | |
2562 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2563 | EXPORT_SYMBOL_GPL(__vmalloc_node); | |
2564 | #endif | |
d0a21265 | 2565 | |
88dca4ca | 2566 | void *__vmalloc(unsigned long size, gfp_t gfp_mask) |
930fc45a | 2567 | { |
f38fcb9c | 2568 | return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE, |
23016969 | 2569 | __builtin_return_address(0)); |
930fc45a | 2570 | } |
1da177e4 LT |
2571 | EXPORT_SYMBOL(__vmalloc); |
2572 | ||
2573 | /** | |
92eac168 MR |
2574 | * vmalloc - allocate virtually contiguous memory |
2575 | * @size: allocation size | |
2576 | * | |
2577 | * Allocate enough pages to cover @size from the page level | |
2578 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2579 | * |
92eac168 MR |
2580 | * For tight control over page level allocator and protection flags |
2581 | * use __vmalloc() instead. | |
a862f68a MR |
2582 | * |
2583 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2584 | */ |
2585 | void *vmalloc(unsigned long size) | |
2586 | { | |
4d39d728 CH |
2587 | return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE, |
2588 | __builtin_return_address(0)); | |
1da177e4 | 2589 | } |
1da177e4 LT |
2590 | EXPORT_SYMBOL(vmalloc); |
2591 | ||
e1ca7788 | 2592 | /** |
92eac168 MR |
2593 | * vzalloc - allocate virtually contiguous memory with zero fill |
2594 | * @size: allocation size | |
2595 | * | |
2596 | * Allocate enough pages to cover @size from the page level | |
2597 | * allocator and map them into contiguous kernel virtual space. | |
2598 | * The memory allocated is set to zero. | |
2599 | * | |
2600 | * For tight control over page level allocator and protection flags | |
2601 | * use __vmalloc() instead. | |
a862f68a MR |
2602 | * |
2603 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2604 | */ |
2605 | void *vzalloc(unsigned long size) | |
2606 | { | |
4d39d728 CH |
2607 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE, |
2608 | __builtin_return_address(0)); | |
e1ca7788 DY |
2609 | } |
2610 | EXPORT_SYMBOL(vzalloc); | |
2611 | ||
83342314 | 2612 | /** |
ead04089 REB |
2613 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
2614 | * @size: allocation size | |
83342314 | 2615 | * |
ead04089 REB |
2616 | * The resulting memory area is zeroed so it can be mapped to userspace |
2617 | * without leaking data. | |
a862f68a MR |
2618 | * |
2619 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2620 | */ |
2621 | void *vmalloc_user(unsigned long size) | |
2622 | { | |
bc84c535 RP |
2623 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2624 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
2625 | VM_USERMAP, NUMA_NO_NODE, | |
2626 | __builtin_return_address(0)); | |
83342314 NP |
2627 | } |
2628 | EXPORT_SYMBOL(vmalloc_user); | |
2629 | ||
930fc45a | 2630 | /** |
92eac168 MR |
2631 | * vmalloc_node - allocate memory on a specific node |
2632 | * @size: allocation size | |
2633 | * @node: numa node | |
930fc45a | 2634 | * |
92eac168 MR |
2635 | * Allocate enough pages to cover @size from the page level |
2636 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 2637 | * |
92eac168 MR |
2638 | * For tight control over page level allocator and protection flags |
2639 | * use __vmalloc() instead. | |
a862f68a MR |
2640 | * |
2641 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2642 | */ |
2643 | void *vmalloc_node(unsigned long size, int node) | |
2644 | { | |
f38fcb9c CH |
2645 | return __vmalloc_node(size, 1, GFP_KERNEL, node, |
2646 | __builtin_return_address(0)); | |
930fc45a CL |
2647 | } |
2648 | EXPORT_SYMBOL(vmalloc_node); | |
2649 | ||
e1ca7788 DY |
2650 | /** |
2651 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2652 | * @size: allocation size | |
2653 | * @node: numa node | |
2654 | * | |
2655 | * Allocate enough pages to cover @size from the page level | |
2656 | * allocator and map them into contiguous kernel virtual space. | |
2657 | * The memory allocated is set to zero. | |
2658 | * | |
a862f68a | 2659 | * Return: pointer to the allocated memory or %NULL on error |
e1ca7788 DY |
2660 | */ |
2661 | void *vzalloc_node(unsigned long size, int node) | |
2662 | { | |
4d39d728 CH |
2663 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node, |
2664 | __builtin_return_address(0)); | |
e1ca7788 DY |
2665 | } |
2666 | EXPORT_SYMBOL(vzalloc_node); | |
2667 | ||
0d08e0d3 | 2668 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2669 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2670 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2671 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2672 | #else |
698d0831 MH |
2673 | /* |
2674 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2675 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2676 | */ | |
2677 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2678 | #endif |
2679 | ||
1da177e4 | 2680 | /** |
92eac168 MR |
2681 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2682 | * @size: allocation size | |
1da177e4 | 2683 | * |
92eac168 MR |
2684 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2685 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2686 | * |
2687 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2688 | */ |
2689 | void *vmalloc_32(unsigned long size) | |
2690 | { | |
f38fcb9c CH |
2691 | return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE, |
2692 | __builtin_return_address(0)); | |
1da177e4 | 2693 | } |
1da177e4 LT |
2694 | EXPORT_SYMBOL(vmalloc_32); |
2695 | ||
83342314 | 2696 | /** |
ead04089 | 2697 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2698 | * @size: allocation size |
ead04089 REB |
2699 | * |
2700 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2701 | * mapped to userspace without leaking data. | |
a862f68a MR |
2702 | * |
2703 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2704 | */ |
2705 | void *vmalloc_32_user(unsigned long size) | |
2706 | { | |
bc84c535 RP |
2707 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2708 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2709 | VM_USERMAP, NUMA_NO_NODE, | |
2710 | __builtin_return_address(0)); | |
83342314 NP |
2711 | } |
2712 | EXPORT_SYMBOL(vmalloc_32_user); | |
2713 | ||
d0107eb0 KH |
2714 | /* |
2715 | * small helper routine , copy contents to buf from addr. | |
2716 | * If the page is not present, fill zero. | |
2717 | */ | |
2718 | ||
2719 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2720 | { | |
2721 | struct page *p; | |
2722 | int copied = 0; | |
2723 | ||
2724 | while (count) { | |
2725 | unsigned long offset, length; | |
2726 | ||
891c49ab | 2727 | offset = offset_in_page(addr); |
d0107eb0 KH |
2728 | length = PAGE_SIZE - offset; |
2729 | if (length > count) | |
2730 | length = count; | |
2731 | p = vmalloc_to_page(addr); | |
2732 | /* | |
2733 | * To do safe access to this _mapped_ area, we need | |
2734 | * lock. But adding lock here means that we need to add | |
2735 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2736 | * interface, rarely used. Instead of that, we'll use | |
2737 | * kmap() and get small overhead in this access function. | |
2738 | */ | |
2739 | if (p) { | |
2740 | /* | |
2741 | * we can expect USER0 is not used (see vread/vwrite's | |
2742 | * function description) | |
2743 | */ | |
9b04c5fe | 2744 | void *map = kmap_atomic(p); |
d0107eb0 | 2745 | memcpy(buf, map + offset, length); |
9b04c5fe | 2746 | kunmap_atomic(map); |
d0107eb0 KH |
2747 | } else |
2748 | memset(buf, 0, length); | |
2749 | ||
2750 | addr += length; | |
2751 | buf += length; | |
2752 | copied += length; | |
2753 | count -= length; | |
2754 | } | |
2755 | return copied; | |
2756 | } | |
2757 | ||
2758 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2759 | { | |
2760 | struct page *p; | |
2761 | int copied = 0; | |
2762 | ||
2763 | while (count) { | |
2764 | unsigned long offset, length; | |
2765 | ||
891c49ab | 2766 | offset = offset_in_page(addr); |
d0107eb0 KH |
2767 | length = PAGE_SIZE - offset; |
2768 | if (length > count) | |
2769 | length = count; | |
2770 | p = vmalloc_to_page(addr); | |
2771 | /* | |
2772 | * To do safe access to this _mapped_ area, we need | |
2773 | * lock. But adding lock here means that we need to add | |
2774 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2775 | * interface, rarely used. Instead of that, we'll use | |
2776 | * kmap() and get small overhead in this access function. | |
2777 | */ | |
2778 | if (p) { | |
2779 | /* | |
2780 | * we can expect USER0 is not used (see vread/vwrite's | |
2781 | * function description) | |
2782 | */ | |
9b04c5fe | 2783 | void *map = kmap_atomic(p); |
d0107eb0 | 2784 | memcpy(map + offset, buf, length); |
9b04c5fe | 2785 | kunmap_atomic(map); |
d0107eb0 KH |
2786 | } |
2787 | addr += length; | |
2788 | buf += length; | |
2789 | copied += length; | |
2790 | count -= length; | |
2791 | } | |
2792 | return copied; | |
2793 | } | |
2794 | ||
2795 | /** | |
92eac168 MR |
2796 | * vread() - read vmalloc area in a safe way. |
2797 | * @buf: buffer for reading data | |
2798 | * @addr: vm address. | |
2799 | * @count: number of bytes to be read. | |
2800 | * | |
92eac168 MR |
2801 | * This function checks that addr is a valid vmalloc'ed area, and |
2802 | * copy data from that area to a given buffer. If the given memory range | |
2803 | * of [addr...addr+count) includes some valid address, data is copied to | |
2804 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2805 | * IOREMAP area is treated as memory hole and no copy is done. | |
2806 | * | |
2807 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2808 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2809 | * | |
2810 | * Note: In usual ops, vread() is never necessary because the caller | |
2811 | * should know vmalloc() area is valid and can use memcpy(). | |
2812 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2813 | * any information, as /dev/kmem. |
a862f68a MR |
2814 | * |
2815 | * Return: number of bytes for which addr and buf should be increased | |
2816 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
2817 | * include any intersection with valid vmalloc area | |
d0107eb0 | 2818 | */ |
1da177e4 LT |
2819 | long vread(char *buf, char *addr, unsigned long count) |
2820 | { | |
e81ce85f JK |
2821 | struct vmap_area *va; |
2822 | struct vm_struct *vm; | |
1da177e4 | 2823 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2824 | unsigned long buflen = count; |
1da177e4 LT |
2825 | unsigned long n; |
2826 | ||
2827 | /* Don't allow overflow */ | |
2828 | if ((unsigned long) addr + count < count) | |
2829 | count = -(unsigned long) addr; | |
2830 | ||
e81ce85f JK |
2831 | spin_lock(&vmap_area_lock); |
2832 | list_for_each_entry(va, &vmap_area_list, list) { | |
2833 | if (!count) | |
2834 | break; | |
2835 | ||
688fcbfc | 2836 | if (!va->vm) |
e81ce85f JK |
2837 | continue; |
2838 | ||
2839 | vm = va->vm; | |
2840 | vaddr = (char *) vm->addr; | |
762216ab | 2841 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2842 | continue; |
2843 | while (addr < vaddr) { | |
2844 | if (count == 0) | |
2845 | goto finished; | |
2846 | *buf = '\0'; | |
2847 | buf++; | |
2848 | addr++; | |
2849 | count--; | |
2850 | } | |
762216ab | 2851 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2852 | if (n > count) |
2853 | n = count; | |
e81ce85f | 2854 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2855 | aligned_vread(buf, addr, n); |
2856 | else /* IOREMAP area is treated as memory hole */ | |
2857 | memset(buf, 0, n); | |
2858 | buf += n; | |
2859 | addr += n; | |
2860 | count -= n; | |
1da177e4 LT |
2861 | } |
2862 | finished: | |
e81ce85f | 2863 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2864 | |
2865 | if (buf == buf_start) | |
2866 | return 0; | |
2867 | /* zero-fill memory holes */ | |
2868 | if (buf != buf_start + buflen) | |
2869 | memset(buf, 0, buflen - (buf - buf_start)); | |
2870 | ||
2871 | return buflen; | |
1da177e4 LT |
2872 | } |
2873 | ||
d0107eb0 | 2874 | /** |
92eac168 MR |
2875 | * vwrite() - write vmalloc area in a safe way. |
2876 | * @buf: buffer for source data | |
2877 | * @addr: vm address. | |
2878 | * @count: number of bytes to be read. | |
2879 | * | |
92eac168 MR |
2880 | * This function checks that addr is a valid vmalloc'ed area, and |
2881 | * copy data from a buffer to the given addr. If specified range of | |
2882 | * [addr...addr+count) includes some valid address, data is copied from | |
2883 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2884 | * IOREMAP area is treated as memory hole and no copy is done. | |
2885 | * | |
2886 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2887 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2888 | * | |
2889 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2890 | * should know vmalloc() area is valid and can use memcpy(). | |
2891 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2892 | * any information, as /dev/kmem. |
a862f68a MR |
2893 | * |
2894 | * Return: number of bytes for which addr and buf should be | |
2895 | * increased (same number as @count) or %0 if [addr...addr+count) | |
2896 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 2897 | */ |
1da177e4 LT |
2898 | long vwrite(char *buf, char *addr, unsigned long count) |
2899 | { | |
e81ce85f JK |
2900 | struct vmap_area *va; |
2901 | struct vm_struct *vm; | |
d0107eb0 KH |
2902 | char *vaddr; |
2903 | unsigned long n, buflen; | |
2904 | int copied = 0; | |
1da177e4 LT |
2905 | |
2906 | /* Don't allow overflow */ | |
2907 | if ((unsigned long) addr + count < count) | |
2908 | count = -(unsigned long) addr; | |
d0107eb0 | 2909 | buflen = count; |
1da177e4 | 2910 | |
e81ce85f JK |
2911 | spin_lock(&vmap_area_lock); |
2912 | list_for_each_entry(va, &vmap_area_list, list) { | |
2913 | if (!count) | |
2914 | break; | |
2915 | ||
688fcbfc | 2916 | if (!va->vm) |
e81ce85f JK |
2917 | continue; |
2918 | ||
2919 | vm = va->vm; | |
2920 | vaddr = (char *) vm->addr; | |
762216ab | 2921 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2922 | continue; |
2923 | while (addr < vaddr) { | |
2924 | if (count == 0) | |
2925 | goto finished; | |
2926 | buf++; | |
2927 | addr++; | |
2928 | count--; | |
2929 | } | |
762216ab | 2930 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2931 | if (n > count) |
2932 | n = count; | |
e81ce85f | 2933 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
2934 | aligned_vwrite(buf, addr, n); |
2935 | copied++; | |
2936 | } | |
2937 | buf += n; | |
2938 | addr += n; | |
2939 | count -= n; | |
1da177e4 LT |
2940 | } |
2941 | finished: | |
e81ce85f | 2942 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2943 | if (!copied) |
2944 | return 0; | |
2945 | return buflen; | |
1da177e4 | 2946 | } |
83342314 NP |
2947 | |
2948 | /** | |
92eac168 MR |
2949 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
2950 | * @vma: vma to cover | |
2951 | * @uaddr: target user address to start at | |
2952 | * @kaddr: virtual address of vmalloc kernel memory | |
bdebd6a2 | 2953 | * @pgoff: offset from @kaddr to start at |
92eac168 | 2954 | * @size: size of map area |
7682486b | 2955 | * |
92eac168 | 2956 | * Returns: 0 for success, -Exxx on failure |
83342314 | 2957 | * |
92eac168 MR |
2958 | * This function checks that @kaddr is a valid vmalloc'ed area, |
2959 | * and that it is big enough to cover the range starting at | |
2960 | * @uaddr in @vma. Will return failure if that criteria isn't | |
2961 | * met. | |
83342314 | 2962 | * |
92eac168 | 2963 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 2964 | */ |
e69e9d4a | 2965 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
bdebd6a2 JH |
2966 | void *kaddr, unsigned long pgoff, |
2967 | unsigned long size) | |
83342314 NP |
2968 | { |
2969 | struct vm_struct *area; | |
bdebd6a2 JH |
2970 | unsigned long off; |
2971 | unsigned long end_index; | |
2972 | ||
2973 | if (check_shl_overflow(pgoff, PAGE_SHIFT, &off)) | |
2974 | return -EINVAL; | |
83342314 | 2975 | |
e69e9d4a HD |
2976 | size = PAGE_ALIGN(size); |
2977 | ||
2978 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
2979 | return -EINVAL; |
2980 | ||
e69e9d4a | 2981 | area = find_vm_area(kaddr); |
83342314 | 2982 | if (!area) |
db64fe02 | 2983 | return -EINVAL; |
83342314 | 2984 | |
fe9041c2 | 2985 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 2986 | return -EINVAL; |
83342314 | 2987 | |
bdebd6a2 JH |
2988 | if (check_add_overflow(size, off, &end_index) || |
2989 | end_index > get_vm_area_size(area)) | |
db64fe02 | 2990 | return -EINVAL; |
bdebd6a2 | 2991 | kaddr += off; |
83342314 | 2992 | |
83342314 | 2993 | do { |
e69e9d4a | 2994 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
2995 | int ret; |
2996 | ||
83342314 NP |
2997 | ret = vm_insert_page(vma, uaddr, page); |
2998 | if (ret) | |
2999 | return ret; | |
3000 | ||
3001 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3002 | kaddr += PAGE_SIZE; |
3003 | size -= PAGE_SIZE; | |
3004 | } while (size > 0); | |
83342314 | 3005 | |
314e51b9 | 3006 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3007 | |
db64fe02 | 3008 | return 0; |
83342314 | 3009 | } |
e69e9d4a HD |
3010 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
3011 | ||
3012 | /** | |
92eac168 MR |
3013 | * remap_vmalloc_range - map vmalloc pages to userspace |
3014 | * @vma: vma to cover (map full range of vma) | |
3015 | * @addr: vmalloc memory | |
3016 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3017 | * |
92eac168 | 3018 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3019 | * |
92eac168 MR |
3020 | * This function checks that addr is a valid vmalloc'ed area, and |
3021 | * that it is big enough to cover the vma. Will return failure if | |
3022 | * that criteria isn't met. | |
e69e9d4a | 3023 | * |
92eac168 | 3024 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3025 | */ |
3026 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3027 | unsigned long pgoff) | |
3028 | { | |
3029 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
bdebd6a2 | 3030 | addr, pgoff, |
e69e9d4a HD |
3031 | vma->vm_end - vma->vm_start); |
3032 | } | |
83342314 NP |
3033 | EXPORT_SYMBOL(remap_vmalloc_range); |
3034 | ||
8b1e0f81 | 3035 | static int f(pte_t *pte, unsigned long addr, void *data) |
5f4352fb | 3036 | { |
cd12909c DV |
3037 | pte_t ***p = data; |
3038 | ||
3039 | if (p) { | |
3040 | *(*p) = pte; | |
3041 | (*p)++; | |
3042 | } | |
5f4352fb JF |
3043 | return 0; |
3044 | } | |
3045 | ||
3046 | /** | |
92eac168 MR |
3047 | * alloc_vm_area - allocate a range of kernel address space |
3048 | * @size: size of the area | |
3049 | * @ptes: returns the PTEs for the address space | |
7682486b | 3050 | * |
92eac168 | 3051 | * Returns: NULL on failure, vm_struct on success |
5f4352fb | 3052 | * |
92eac168 MR |
3053 | * This function reserves a range of kernel address space, and |
3054 | * allocates pagetables to map that range. No actual mappings | |
3055 | * are created. | |
cd12909c | 3056 | * |
92eac168 MR |
3057 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
3058 | * allocated for the VM area are returned. | |
5f4352fb | 3059 | */ |
cd12909c | 3060 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
3061 | { |
3062 | struct vm_struct *area; | |
3063 | ||
23016969 CL |
3064 | area = get_vm_area_caller(size, VM_IOREMAP, |
3065 | __builtin_return_address(0)); | |
5f4352fb JF |
3066 | if (area == NULL) |
3067 | return NULL; | |
3068 | ||
3069 | /* | |
3070 | * This ensures that page tables are constructed for this region | |
3071 | * of kernel virtual address space and mapped into init_mm. | |
3072 | */ | |
3073 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 3074 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
3075 | free_vm_area(area); |
3076 | return NULL; | |
3077 | } | |
3078 | ||
5f4352fb JF |
3079 | return area; |
3080 | } | |
3081 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
3082 | ||
3083 | void free_vm_area(struct vm_struct *area) | |
3084 | { | |
3085 | struct vm_struct *ret; | |
3086 | ret = remove_vm_area(area->addr); | |
3087 | BUG_ON(ret != area); | |
3088 | kfree(area); | |
3089 | } | |
3090 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3091 | |
4f8b02b4 | 3092 | #ifdef CONFIG_SMP |
ca23e405 TH |
3093 | static struct vmap_area *node_to_va(struct rb_node *n) |
3094 | { | |
4583e773 | 3095 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3096 | } |
3097 | ||
3098 | /** | |
68ad4a33 URS |
3099 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3100 | * @addr: target address | |
ca23e405 | 3101 | * |
68ad4a33 URS |
3102 | * Returns: vmap_area if it is found. If there is no such area |
3103 | * the first highest(reverse order) vmap_area is returned | |
3104 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3105 | * if there are no any areas before @addr. | |
ca23e405 | 3106 | */ |
68ad4a33 URS |
3107 | static struct vmap_area * |
3108 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3109 | { |
68ad4a33 URS |
3110 | struct vmap_area *va, *tmp; |
3111 | struct rb_node *n; | |
3112 | ||
3113 | n = free_vmap_area_root.rb_node; | |
3114 | va = NULL; | |
ca23e405 TH |
3115 | |
3116 | while (n) { | |
68ad4a33 URS |
3117 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3118 | if (tmp->va_start <= addr) { | |
3119 | va = tmp; | |
3120 | if (tmp->va_end >= addr) | |
3121 | break; | |
3122 | ||
ca23e405 | 3123 | n = n->rb_right; |
68ad4a33 URS |
3124 | } else { |
3125 | n = n->rb_left; | |
3126 | } | |
ca23e405 TH |
3127 | } |
3128 | ||
68ad4a33 | 3129 | return va; |
ca23e405 TH |
3130 | } |
3131 | ||
3132 | /** | |
68ad4a33 URS |
3133 | * pvm_determine_end_from_reverse - find the highest aligned address |
3134 | * of free block below VMALLOC_END | |
3135 | * @va: | |
3136 | * in - the VA we start the search(reverse order); | |
3137 | * out - the VA with the highest aligned end address. | |
ca23e405 | 3138 | * |
68ad4a33 | 3139 | * Returns: determined end address within vmap_area |
ca23e405 | 3140 | */ |
68ad4a33 URS |
3141 | static unsigned long |
3142 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3143 | { |
68ad4a33 | 3144 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3145 | unsigned long addr; |
3146 | ||
68ad4a33 URS |
3147 | if (likely(*va)) { |
3148 | list_for_each_entry_from_reverse((*va), | |
3149 | &free_vmap_area_list, list) { | |
3150 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3151 | if ((*va)->va_start < addr) | |
3152 | return addr; | |
3153 | } | |
ca23e405 TH |
3154 | } |
3155 | ||
68ad4a33 | 3156 | return 0; |
ca23e405 TH |
3157 | } |
3158 | ||
3159 | /** | |
3160 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3161 | * @offsets: array containing offset of each area | |
3162 | * @sizes: array containing size of each area | |
3163 | * @nr_vms: the number of areas to allocate | |
3164 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3165 | * |
3166 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3167 | * vm_structs on success, %NULL on failure | |
3168 | * | |
3169 | * Percpu allocator wants to use congruent vm areas so that it can | |
3170 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3171 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3172 | * be scattered pretty far, distance between two areas easily going up | |
3173 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3174 | * areas are allocated from top. | |
ca23e405 | 3175 | * |
68ad4a33 URS |
3176 | * Despite its complicated look, this allocator is rather simple. It |
3177 | * does everything top-down and scans free blocks from the end looking | |
3178 | * for matching base. While scanning, if any of the areas do not fit the | |
3179 | * base address is pulled down to fit the area. Scanning is repeated till | |
3180 | * all the areas fit and then all necessary data structures are inserted | |
3181 | * and the result is returned. | |
ca23e405 TH |
3182 | */ |
3183 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3184 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3185 | size_t align) |
ca23e405 TH |
3186 | { |
3187 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3188 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3189 | struct vmap_area **vas, *va; |
ca23e405 TH |
3190 | struct vm_struct **vms; |
3191 | int area, area2, last_area, term_area; | |
253a496d | 3192 | unsigned long base, start, size, end, last_end, orig_start, orig_end; |
ca23e405 | 3193 | bool purged = false; |
68ad4a33 | 3194 | enum fit_type type; |
ca23e405 | 3195 | |
ca23e405 | 3196 | /* verify parameters and allocate data structures */ |
891c49ab | 3197 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3198 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3199 | start = offsets[area]; | |
3200 | end = start + sizes[area]; | |
3201 | ||
3202 | /* is everything aligned properly? */ | |
3203 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3204 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3205 | ||
3206 | /* detect the area with the highest address */ | |
3207 | if (start > offsets[last_area]) | |
3208 | last_area = area; | |
3209 | ||
c568da28 | 3210 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3211 | unsigned long start2 = offsets[area2]; |
3212 | unsigned long end2 = start2 + sizes[area2]; | |
3213 | ||
c568da28 | 3214 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3215 | } |
3216 | } | |
3217 | last_end = offsets[last_area] + sizes[last_area]; | |
3218 | ||
3219 | if (vmalloc_end - vmalloc_start < last_end) { | |
3220 | WARN_ON(true); | |
3221 | return NULL; | |
3222 | } | |
3223 | ||
4d67d860 TM |
3224 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3225 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3226 | if (!vas || !vms) |
f1db7afd | 3227 | goto err_free2; |
ca23e405 TH |
3228 | |
3229 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3230 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3231 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3232 | if (!vas[area] || !vms[area]) |
3233 | goto err_free; | |
3234 | } | |
3235 | retry: | |
e36176be | 3236 | spin_lock(&free_vmap_area_lock); |
ca23e405 TH |
3237 | |
3238 | /* start scanning - we scan from the top, begin with the last area */ | |
3239 | area = term_area = last_area; | |
3240 | start = offsets[area]; | |
3241 | end = start + sizes[area]; | |
3242 | ||
68ad4a33 URS |
3243 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3244 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3245 | |
3246 | while (true) { | |
ca23e405 TH |
3247 | /* |
3248 | * base might have underflowed, add last_end before | |
3249 | * comparing. | |
3250 | */ | |
68ad4a33 URS |
3251 | if (base + last_end < vmalloc_start + last_end) |
3252 | goto overflow; | |
ca23e405 TH |
3253 | |
3254 | /* | |
68ad4a33 | 3255 | * Fitting base has not been found. |
ca23e405 | 3256 | */ |
68ad4a33 URS |
3257 | if (va == NULL) |
3258 | goto overflow; | |
ca23e405 | 3259 | |
5336e52c | 3260 | /* |
d8cc323d | 3261 | * If required width exceeds current VA block, move |
5336e52c KS |
3262 | * base downwards and then recheck. |
3263 | */ | |
3264 | if (base + end > va->va_end) { | |
3265 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3266 | term_area = area; | |
3267 | continue; | |
3268 | } | |
3269 | ||
ca23e405 | 3270 | /* |
68ad4a33 | 3271 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3272 | */ |
5336e52c | 3273 | if (base + start < va->va_start) { |
68ad4a33 URS |
3274 | va = node_to_va(rb_prev(&va->rb_node)); |
3275 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3276 | term_area = area; |
3277 | continue; | |
3278 | } | |
3279 | ||
3280 | /* | |
3281 | * This area fits, move on to the previous one. If | |
3282 | * the previous one is the terminal one, we're done. | |
3283 | */ | |
3284 | area = (area + nr_vms - 1) % nr_vms; | |
3285 | if (area == term_area) | |
3286 | break; | |
68ad4a33 | 3287 | |
ca23e405 TH |
3288 | start = offsets[area]; |
3289 | end = start + sizes[area]; | |
68ad4a33 | 3290 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3291 | } |
68ad4a33 | 3292 | |
ca23e405 TH |
3293 | /* we've found a fitting base, insert all va's */ |
3294 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3295 | int ret; |
ca23e405 | 3296 | |
68ad4a33 URS |
3297 | start = base + offsets[area]; |
3298 | size = sizes[area]; | |
ca23e405 | 3299 | |
68ad4a33 URS |
3300 | va = pvm_find_va_enclose_addr(start); |
3301 | if (WARN_ON_ONCE(va == NULL)) | |
3302 | /* It is a BUG(), but trigger recovery instead. */ | |
3303 | goto recovery; | |
3304 | ||
3305 | type = classify_va_fit_type(va, start, size); | |
3306 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3307 | /* It is a BUG(), but trigger recovery instead. */ | |
3308 | goto recovery; | |
3309 | ||
3310 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3311 | if (unlikely(ret)) | |
3312 | goto recovery; | |
3313 | ||
3314 | /* Allocated area. */ | |
3315 | va = vas[area]; | |
3316 | va->va_start = start; | |
3317 | va->va_end = start + size; | |
68ad4a33 | 3318 | } |
ca23e405 | 3319 | |
e36176be | 3320 | spin_unlock(&free_vmap_area_lock); |
ca23e405 | 3321 | |
253a496d DA |
3322 | /* populate the kasan shadow space */ |
3323 | for (area = 0; area < nr_vms; area++) { | |
3324 | if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area])) | |
3325 | goto err_free_shadow; | |
3326 | ||
3327 | kasan_unpoison_vmalloc((void *)vas[area]->va_start, | |
3328 | sizes[area]); | |
3329 | } | |
3330 | ||
ca23e405 | 3331 | /* insert all vm's */ |
e36176be URS |
3332 | spin_lock(&vmap_area_lock); |
3333 | for (area = 0; area < nr_vms; area++) { | |
3334 | insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list); | |
3335 | ||
3336 | setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC, | |
3645cb4a | 3337 | pcpu_get_vm_areas); |
e36176be URS |
3338 | } |
3339 | spin_unlock(&vmap_area_lock); | |
ca23e405 TH |
3340 | |
3341 | kfree(vas); | |
3342 | return vms; | |
3343 | ||
68ad4a33 | 3344 | recovery: |
e36176be URS |
3345 | /* |
3346 | * Remove previously allocated areas. There is no | |
3347 | * need in removing these areas from the busy tree, | |
3348 | * because they are inserted only on the final step | |
3349 | * and when pcpu_get_vm_areas() is success. | |
3350 | */ | |
68ad4a33 | 3351 | while (area--) { |
253a496d DA |
3352 | orig_start = vas[area]->va_start; |
3353 | orig_end = vas[area]->va_end; | |
3354 | va = merge_or_add_vmap_area(vas[area], &free_vmap_area_root, | |
3355 | &free_vmap_area_list); | |
9c801f61 URS |
3356 | if (va) |
3357 | kasan_release_vmalloc(orig_start, orig_end, | |
3358 | va->va_start, va->va_end); | |
68ad4a33 URS |
3359 | vas[area] = NULL; |
3360 | } | |
3361 | ||
3362 | overflow: | |
e36176be | 3363 | spin_unlock(&free_vmap_area_lock); |
68ad4a33 URS |
3364 | if (!purged) { |
3365 | purge_vmap_area_lazy(); | |
3366 | purged = true; | |
3367 | ||
3368 | /* Before "retry", check if we recover. */ | |
3369 | for (area = 0; area < nr_vms; area++) { | |
3370 | if (vas[area]) | |
3371 | continue; | |
3372 | ||
3373 | vas[area] = kmem_cache_zalloc( | |
3374 | vmap_area_cachep, GFP_KERNEL); | |
3375 | if (!vas[area]) | |
3376 | goto err_free; | |
3377 | } | |
3378 | ||
3379 | goto retry; | |
3380 | } | |
3381 | ||
ca23e405 TH |
3382 | err_free: |
3383 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3384 | if (vas[area]) |
3385 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3386 | ||
f1db7afd | 3387 | kfree(vms[area]); |
ca23e405 | 3388 | } |
f1db7afd | 3389 | err_free2: |
ca23e405 TH |
3390 | kfree(vas); |
3391 | kfree(vms); | |
3392 | return NULL; | |
253a496d DA |
3393 | |
3394 | err_free_shadow: | |
3395 | spin_lock(&free_vmap_area_lock); | |
3396 | /* | |
3397 | * We release all the vmalloc shadows, even the ones for regions that | |
3398 | * hadn't been successfully added. This relies on kasan_release_vmalloc | |
3399 | * being able to tolerate this case. | |
3400 | */ | |
3401 | for (area = 0; area < nr_vms; area++) { | |
3402 | orig_start = vas[area]->va_start; | |
3403 | orig_end = vas[area]->va_end; | |
3404 | va = merge_or_add_vmap_area(vas[area], &free_vmap_area_root, | |
3405 | &free_vmap_area_list); | |
9c801f61 URS |
3406 | if (va) |
3407 | kasan_release_vmalloc(orig_start, orig_end, | |
3408 | va->va_start, va->va_end); | |
253a496d DA |
3409 | vas[area] = NULL; |
3410 | kfree(vms[area]); | |
3411 | } | |
3412 | spin_unlock(&free_vmap_area_lock); | |
3413 | kfree(vas); | |
3414 | kfree(vms); | |
3415 | return NULL; | |
ca23e405 TH |
3416 | } |
3417 | ||
3418 | /** | |
3419 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3420 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3421 | * @nr_vms: the number of allocated areas | |
3422 | * | |
3423 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3424 | */ | |
3425 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3426 | { | |
3427 | int i; | |
3428 | ||
3429 | for (i = 0; i < nr_vms; i++) | |
3430 | free_vm_area(vms[i]); | |
3431 | kfree(vms); | |
3432 | } | |
4f8b02b4 | 3433 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
3434 | |
3435 | #ifdef CONFIG_PROC_FS | |
3436 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e36176be | 3437 | __acquires(&vmap_purge_lock) |
d4033afd | 3438 | __acquires(&vmap_area_lock) |
a10aa579 | 3439 | { |
e36176be | 3440 | mutex_lock(&vmap_purge_lock); |
d4033afd | 3441 | spin_lock(&vmap_area_lock); |
e36176be | 3442 | |
3f500069 | 3443 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3444 | } |
3445 | ||
3446 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3447 | { | |
3f500069 | 3448 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3449 | } |
3450 | ||
3451 | static void s_stop(struct seq_file *m, void *p) | |
e36176be | 3452 | __releases(&vmap_purge_lock) |
d4033afd | 3453 | __releases(&vmap_area_lock) |
a10aa579 | 3454 | { |
e36176be | 3455 | mutex_unlock(&vmap_purge_lock); |
d4033afd | 3456 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
3457 | } |
3458 | ||
a47a126a ED |
3459 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3460 | { | |
e5adfffc | 3461 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3462 | unsigned int nr, *counters = m->private; |
3463 | ||
3464 | if (!counters) | |
3465 | return; | |
3466 | ||
af12346c WL |
3467 | if (v->flags & VM_UNINITIALIZED) |
3468 | return; | |
7e5b528b DV |
3469 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3470 | smp_rmb(); | |
af12346c | 3471 | |
a47a126a ED |
3472 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3473 | ||
3474 | for (nr = 0; nr < v->nr_pages; nr++) | |
3475 | counters[page_to_nid(v->pages[nr])]++; | |
3476 | ||
3477 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3478 | if (counters[nr]) | |
3479 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3480 | } | |
3481 | } | |
3482 | ||
dd3b8353 URS |
3483 | static void show_purge_info(struct seq_file *m) |
3484 | { | |
3485 | struct llist_node *head; | |
3486 | struct vmap_area *va; | |
3487 | ||
3488 | head = READ_ONCE(vmap_purge_list.first); | |
3489 | if (head == NULL) | |
3490 | return; | |
3491 | ||
3492 | llist_for_each_entry(va, head, purge_list) { | |
3493 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", | |
3494 | (void *)va->va_start, (void *)va->va_end, | |
3495 | va->va_end - va->va_start); | |
3496 | } | |
3497 | } | |
3498 | ||
a10aa579 CL |
3499 | static int s_show(struct seq_file *m, void *p) |
3500 | { | |
3f500069 | 3501 | struct vmap_area *va; |
d4033afd JK |
3502 | struct vm_struct *v; |
3503 | ||
3f500069 | 3504 | va = list_entry(p, struct vmap_area, list); |
3505 | ||
c2ce8c14 | 3506 | /* |
688fcbfc PL |
3507 | * s_show can encounter race with remove_vm_area, !vm on behalf |
3508 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 3509 | */ |
688fcbfc | 3510 | if (!va->vm) { |
dd3b8353 | 3511 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 3512 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 3513 | va->va_end - va->va_start); |
78c72746 | 3514 | |
d4033afd | 3515 | return 0; |
78c72746 | 3516 | } |
d4033afd JK |
3517 | |
3518 | v = va->vm; | |
a10aa579 | 3519 | |
45ec1690 | 3520 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3521 | v->addr, v->addr + v->size, v->size); |
3522 | ||
62c70bce JP |
3523 | if (v->caller) |
3524 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3525 | |
a10aa579 CL |
3526 | if (v->nr_pages) |
3527 | seq_printf(m, " pages=%d", v->nr_pages); | |
3528 | ||
3529 | if (v->phys_addr) | |
199eaa05 | 3530 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3531 | |
3532 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3533 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3534 | |
3535 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3536 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3537 | |
3538 | if (v->flags & VM_MAP) | |
f4527c90 | 3539 | seq_puts(m, " vmap"); |
a10aa579 CL |
3540 | |
3541 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3542 | seq_puts(m, " user"); |
a10aa579 | 3543 | |
fe9041c2 CH |
3544 | if (v->flags & VM_DMA_COHERENT) |
3545 | seq_puts(m, " dma-coherent"); | |
3546 | ||
244d63ee | 3547 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3548 | seq_puts(m, " vpages"); |
a10aa579 | 3549 | |
a47a126a | 3550 | show_numa_info(m, v); |
a10aa579 | 3551 | seq_putc(m, '\n'); |
dd3b8353 URS |
3552 | |
3553 | /* | |
3554 | * As a final step, dump "unpurged" areas. Note, | |
3555 | * that entire "/proc/vmallocinfo" output will not | |
3556 | * be address sorted, because the purge list is not | |
3557 | * sorted. | |
3558 | */ | |
3559 | if (list_is_last(&va->list, &vmap_area_list)) | |
3560 | show_purge_info(m); | |
3561 | ||
a10aa579 CL |
3562 | return 0; |
3563 | } | |
3564 | ||
5f6a6a9c | 3565 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3566 | .start = s_start, |
3567 | .next = s_next, | |
3568 | .stop = s_stop, | |
3569 | .show = s_show, | |
3570 | }; | |
5f6a6a9c | 3571 | |
5f6a6a9c AD |
3572 | static int __init proc_vmalloc_init(void) |
3573 | { | |
fddda2b7 | 3574 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3575 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3576 | &vmalloc_op, |
3577 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3578 | else |
0825a6f9 | 3579 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3580 | return 0; |
3581 | } | |
3582 | module_init(proc_vmalloc_init); | |
db3808c1 | 3583 | |
a10aa579 | 3584 | #endif |