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 |
1da177e4 LT |
10 | */ |
11 | ||
db64fe02 | 12 | #include <linux/vmalloc.h> |
1da177e4 LT |
13 | #include <linux/mm.h> |
14 | #include <linux/module.h> | |
15 | #include <linux/highmem.h> | |
c3edc401 | 16 | #include <linux/sched/signal.h> |
1da177e4 LT |
17 | #include <linux/slab.h> |
18 | #include <linux/spinlock.h> | |
19 | #include <linux/interrupt.h> | |
5f6a6a9c | 20 | #include <linux/proc_fs.h> |
a10aa579 | 21 | #include <linux/seq_file.h> |
868b104d | 22 | #include <linux/set_memory.h> |
3ac7fe5a | 23 | #include <linux/debugobjects.h> |
23016969 | 24 | #include <linux/kallsyms.h> |
db64fe02 | 25 | #include <linux/list.h> |
4da56b99 | 26 | #include <linux/notifier.h> |
db64fe02 NP |
27 | #include <linux/rbtree.h> |
28 | #include <linux/radix-tree.h> | |
29 | #include <linux/rcupdate.h> | |
f0aa6617 | 30 | #include <linux/pfn.h> |
89219d37 | 31 | #include <linux/kmemleak.h> |
60063497 | 32 | #include <linux/atomic.h> |
3b32123d | 33 | #include <linux/compiler.h> |
32fcfd40 | 34 | #include <linux/llist.h> |
0f616be1 | 35 | #include <linux/bitops.h> |
68ad4a33 | 36 | #include <linux/rbtree_augmented.h> |
bdebd6a2 | 37 | #include <linux/overflow.h> |
3b32123d | 38 | |
7c0f6ba6 | 39 | #include <linux/uaccess.h> |
1da177e4 | 40 | #include <asm/tlbflush.h> |
2dca6999 | 41 | #include <asm/shmparam.h> |
1da177e4 | 42 | |
dd56b046 MG |
43 | #include "internal.h" |
44 | ||
186525bd IM |
45 | bool is_vmalloc_addr(const void *x) |
46 | { | |
47 | unsigned long addr = (unsigned long)x; | |
48 | ||
49 | return addr >= VMALLOC_START && addr < VMALLOC_END; | |
50 | } | |
51 | EXPORT_SYMBOL(is_vmalloc_addr); | |
52 | ||
32fcfd40 AV |
53 | struct vfree_deferred { |
54 | struct llist_head list; | |
55 | struct work_struct wq; | |
56 | }; | |
57 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); | |
58 | ||
59 | static void __vunmap(const void *, int); | |
60 | ||
61 | static void free_work(struct work_struct *w) | |
62 | { | |
63 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); | |
894e58c1 BP |
64 | struct llist_node *t, *llnode; |
65 | ||
66 | llist_for_each_safe(llnode, t, llist_del_all(&p->list)) | |
67 | __vunmap((void *)llnode, 1); | |
32fcfd40 AV |
68 | } |
69 | ||
db64fe02 | 70 | /*** Page table manipulation functions ***/ |
b221385b | 71 | |
1da177e4 LT |
72 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
73 | { | |
74 | pte_t *pte; | |
75 | ||
76 | pte = pte_offset_kernel(pmd, addr); | |
77 | do { | |
78 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); | |
79 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); | |
80 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
81 | } | |
82 | ||
db64fe02 | 83 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
1da177e4 LT |
84 | { |
85 | pmd_t *pmd; | |
86 | unsigned long next; | |
87 | ||
88 | pmd = pmd_offset(pud, addr); | |
89 | do { | |
90 | next = pmd_addr_end(addr, end); | |
b9820d8f TK |
91 | if (pmd_clear_huge(pmd)) |
92 | continue; | |
1da177e4 LT |
93 | if (pmd_none_or_clear_bad(pmd)) |
94 | continue; | |
95 | vunmap_pte_range(pmd, addr, next); | |
96 | } while (pmd++, addr = next, addr != end); | |
97 | } | |
98 | ||
c2febafc | 99 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end) |
1da177e4 LT |
100 | { |
101 | pud_t *pud; | |
102 | unsigned long next; | |
103 | ||
c2febafc | 104 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
105 | do { |
106 | next = pud_addr_end(addr, end); | |
b9820d8f TK |
107 | if (pud_clear_huge(pud)) |
108 | continue; | |
1da177e4 LT |
109 | if (pud_none_or_clear_bad(pud)) |
110 | continue; | |
111 | vunmap_pmd_range(pud, addr, next); | |
112 | } while (pud++, addr = next, addr != end); | |
113 | } | |
114 | ||
c2febafc KS |
115 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
116 | { | |
117 | p4d_t *p4d; | |
118 | unsigned long next; | |
119 | ||
120 | p4d = p4d_offset(pgd, addr); | |
121 | do { | |
122 | next = p4d_addr_end(addr, end); | |
123 | if (p4d_clear_huge(p4d)) | |
124 | continue; | |
125 | if (p4d_none_or_clear_bad(p4d)) | |
126 | continue; | |
127 | vunmap_pud_range(p4d, addr, next); | |
128 | } while (p4d++, addr = next, addr != end); | |
129 | } | |
130 | ||
b521c43f CH |
131 | /** |
132 | * unmap_kernel_range_noflush - unmap kernel VM area | |
133 | * @addr: start of the VM area to unmap | |
134 | * @size: size of the VM area to unmap | |
135 | * | |
136 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size specify | |
137 | * should have been allocated using get_vm_area() and its friends. | |
138 | * | |
139 | * NOTE: | |
140 | * This function does NOT do any cache flushing. The caller is responsible | |
141 | * for calling flush_cache_vunmap() on to-be-mapped areas before calling this | |
142 | * function and flush_tlb_kernel_range() after. | |
143 | */ | |
144 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) | |
1da177e4 | 145 | { |
b521c43f | 146 | unsigned long end = addr + size; |
1da177e4 | 147 | unsigned long next; |
b521c43f | 148 | pgd_t *pgd; |
1da177e4 LT |
149 | |
150 | BUG_ON(addr >= end); | |
151 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
152 | do { |
153 | next = pgd_addr_end(addr, end); | |
154 | if (pgd_none_or_clear_bad(pgd)) | |
155 | continue; | |
c2febafc | 156 | vunmap_p4d_range(pgd, addr, next); |
1da177e4 | 157 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
158 | } |
159 | ||
160 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, | |
db64fe02 | 161 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
162 | { |
163 | pte_t *pte; | |
164 | ||
db64fe02 NP |
165 | /* |
166 | * nr is a running index into the array which helps higher level | |
167 | * callers keep track of where we're up to. | |
168 | */ | |
169 | ||
872fec16 | 170 | pte = pte_alloc_kernel(pmd, addr); |
1da177e4 LT |
171 | if (!pte) |
172 | return -ENOMEM; | |
173 | do { | |
db64fe02 NP |
174 | struct page *page = pages[*nr]; |
175 | ||
176 | if (WARN_ON(!pte_none(*pte))) | |
177 | return -EBUSY; | |
178 | if (WARN_ON(!page)) | |
1da177e4 LT |
179 | return -ENOMEM; |
180 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); | |
db64fe02 | 181 | (*nr)++; |
1da177e4 LT |
182 | } while (pte++, addr += PAGE_SIZE, addr != end); |
183 | return 0; | |
184 | } | |
185 | ||
db64fe02 NP |
186 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
187 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
1da177e4 LT |
188 | { |
189 | pmd_t *pmd; | |
190 | unsigned long next; | |
191 | ||
192 | pmd = pmd_alloc(&init_mm, pud, addr); | |
193 | if (!pmd) | |
194 | return -ENOMEM; | |
195 | do { | |
196 | next = pmd_addr_end(addr, end); | |
db64fe02 | 197 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
1da177e4 LT |
198 | return -ENOMEM; |
199 | } while (pmd++, addr = next, addr != end); | |
200 | return 0; | |
201 | } | |
202 | ||
c2febafc | 203 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr, |
db64fe02 | 204 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
1da177e4 LT |
205 | { |
206 | pud_t *pud; | |
207 | unsigned long next; | |
208 | ||
c2febafc | 209 | pud = pud_alloc(&init_mm, p4d, addr); |
1da177e4 LT |
210 | if (!pud) |
211 | return -ENOMEM; | |
212 | do { | |
213 | next = pud_addr_end(addr, end); | |
db64fe02 | 214 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
1da177e4 LT |
215 | return -ENOMEM; |
216 | } while (pud++, addr = next, addr != end); | |
217 | return 0; | |
218 | } | |
219 | ||
c2febafc KS |
220 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, |
221 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) | |
222 | { | |
223 | p4d_t *p4d; | |
224 | unsigned long next; | |
225 | ||
226 | p4d = p4d_alloc(&init_mm, pgd, addr); | |
227 | if (!p4d) | |
228 | return -ENOMEM; | |
229 | do { | |
230 | next = p4d_addr_end(addr, end); | |
231 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr)) | |
232 | return -ENOMEM; | |
233 | } while (p4d++, addr = next, addr != end); | |
234 | return 0; | |
235 | } | |
236 | ||
b521c43f CH |
237 | /** |
238 | * map_kernel_range_noflush - map kernel VM area with the specified pages | |
239 | * @addr: start of the VM area to map | |
240 | * @size: size of the VM area to map | |
241 | * @prot: page protection flags to use | |
242 | * @pages: pages to map | |
db64fe02 | 243 | * |
b521c43f CH |
244 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size specify should |
245 | * have been allocated using get_vm_area() and its friends. | |
246 | * | |
247 | * NOTE: | |
248 | * This function does NOT do any cache flushing. The caller is responsible for | |
249 | * calling flush_cache_vmap() on to-be-mapped areas before calling this | |
250 | * function. | |
251 | * | |
252 | * RETURNS: | |
253 | * The number of pages mapped on success, -errno on failure. | |
db64fe02 | 254 | */ |
b521c43f CH |
255 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, |
256 | pgprot_t prot, struct page **pages) | |
1da177e4 | 257 | { |
b521c43f | 258 | unsigned long end = addr + size; |
1da177e4 | 259 | unsigned long next; |
b521c43f | 260 | pgd_t *pgd; |
db64fe02 NP |
261 | int err = 0; |
262 | int nr = 0; | |
1da177e4 LT |
263 | |
264 | BUG_ON(addr >= end); | |
265 | pgd = pgd_offset_k(addr); | |
1da177e4 LT |
266 | do { |
267 | next = pgd_addr_end(addr, end); | |
c2febafc | 268 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr); |
1da177e4 | 269 | if (err) |
bf88c8c8 | 270 | return err; |
1da177e4 | 271 | } while (pgd++, addr = next, addr != end); |
db64fe02 | 272 | |
db64fe02 | 273 | return nr; |
1da177e4 LT |
274 | } |
275 | ||
8fc48985 TH |
276 | static int vmap_page_range(unsigned long start, unsigned long end, |
277 | pgprot_t prot, struct page **pages) | |
278 | { | |
279 | int ret; | |
280 | ||
b521c43f | 281 | ret = map_kernel_range_noflush(start, end - start, prot, pages); |
8fc48985 TH |
282 | flush_cache_vmap(start, end); |
283 | return ret; | |
284 | } | |
285 | ||
81ac3ad9 | 286 | int is_vmalloc_or_module_addr(const void *x) |
73bdf0a6 LT |
287 | { |
288 | /* | |
ab4f2ee1 | 289 | * ARM, x86-64 and sparc64 put modules in a special place, |
73bdf0a6 LT |
290 | * and fall back on vmalloc() if that fails. Others |
291 | * just put it in the vmalloc space. | |
292 | */ | |
293 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) | |
294 | unsigned long addr = (unsigned long)x; | |
295 | if (addr >= MODULES_VADDR && addr < MODULES_END) | |
296 | return 1; | |
297 | #endif | |
298 | return is_vmalloc_addr(x); | |
299 | } | |
300 | ||
48667e7a | 301 | /* |
add688fb | 302 | * Walk a vmap address to the struct page it maps. |
48667e7a | 303 | */ |
add688fb | 304 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
48667e7a CL |
305 | { |
306 | unsigned long addr = (unsigned long) vmalloc_addr; | |
add688fb | 307 | struct page *page = NULL; |
48667e7a | 308 | pgd_t *pgd = pgd_offset_k(addr); |
c2febafc KS |
309 | p4d_t *p4d; |
310 | pud_t *pud; | |
311 | pmd_t *pmd; | |
312 | pte_t *ptep, pte; | |
48667e7a | 313 | |
7aa413de IM |
314 | /* |
315 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for | |
316 | * architectures that do not vmalloc module space | |
317 | */ | |
73bdf0a6 | 318 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
59ea7463 | 319 | |
c2febafc KS |
320 | if (pgd_none(*pgd)) |
321 | return NULL; | |
322 | p4d = p4d_offset(pgd, addr); | |
323 | if (p4d_none(*p4d)) | |
324 | return NULL; | |
325 | pud = pud_offset(p4d, addr); | |
029c54b0 AB |
326 | |
327 | /* | |
328 | * Don't dereference bad PUD or PMD (below) entries. This will also | |
329 | * identify huge mappings, which we may encounter on architectures | |
330 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be | |
331 | * identified as vmalloc addresses by is_vmalloc_addr(), but are | |
332 | * not [unambiguously] associated with a struct page, so there is | |
333 | * no correct value to return for them. | |
334 | */ | |
335 | WARN_ON_ONCE(pud_bad(*pud)); | |
336 | if (pud_none(*pud) || pud_bad(*pud)) | |
c2febafc KS |
337 | return NULL; |
338 | pmd = pmd_offset(pud, addr); | |
029c54b0 AB |
339 | WARN_ON_ONCE(pmd_bad(*pmd)); |
340 | if (pmd_none(*pmd) || pmd_bad(*pmd)) | |
c2febafc KS |
341 | return NULL; |
342 | ||
343 | ptep = pte_offset_map(pmd, addr); | |
344 | pte = *ptep; | |
345 | if (pte_present(pte)) | |
346 | page = pte_page(pte); | |
347 | pte_unmap(ptep); | |
add688fb | 348 | return page; |
48667e7a | 349 | } |
add688fb | 350 | EXPORT_SYMBOL(vmalloc_to_page); |
48667e7a CL |
351 | |
352 | /* | |
add688fb | 353 | * Map a vmalloc()-space virtual address to the physical page frame number. |
48667e7a | 354 | */ |
add688fb | 355 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
48667e7a | 356 | { |
add688fb | 357 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
48667e7a | 358 | } |
add688fb | 359 | EXPORT_SYMBOL(vmalloc_to_pfn); |
48667e7a | 360 | |
db64fe02 NP |
361 | |
362 | /*** Global kva allocator ***/ | |
363 | ||
bb850f4d | 364 | #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 |
a6cf4e0f | 365 | #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 |
bb850f4d | 366 | |
db64fe02 | 367 | |
db64fe02 | 368 | static DEFINE_SPINLOCK(vmap_area_lock); |
e36176be | 369 | static DEFINE_SPINLOCK(free_vmap_area_lock); |
f1c4069e JK |
370 | /* Export for kexec only */ |
371 | LIST_HEAD(vmap_area_list); | |
80c4bd7a | 372 | static LLIST_HEAD(vmap_purge_list); |
89699605 | 373 | static struct rb_root vmap_area_root = RB_ROOT; |
68ad4a33 | 374 | static bool vmap_initialized __read_mostly; |
89699605 | 375 | |
68ad4a33 URS |
376 | /* |
377 | * This kmem_cache is used for vmap_area objects. Instead of | |
378 | * allocating from slab we reuse an object from this cache to | |
379 | * make things faster. Especially in "no edge" splitting of | |
380 | * free block. | |
381 | */ | |
382 | static struct kmem_cache *vmap_area_cachep; | |
383 | ||
384 | /* | |
385 | * This linked list is used in pair with free_vmap_area_root. | |
386 | * It gives O(1) access to prev/next to perform fast coalescing. | |
387 | */ | |
388 | static LIST_HEAD(free_vmap_area_list); | |
389 | ||
390 | /* | |
391 | * This augment red-black tree represents the free vmap space. | |
392 | * All vmap_area objects in this tree are sorted by va->va_start | |
393 | * address. It is used for allocation and merging when a vmap | |
394 | * object is released. | |
395 | * | |
396 | * Each vmap_area node contains a maximum available free block | |
397 | * of its sub-tree, right or left. Therefore it is possible to | |
398 | * find a lowest match of free area. | |
399 | */ | |
400 | static struct rb_root free_vmap_area_root = RB_ROOT; | |
401 | ||
82dd23e8 URS |
402 | /* |
403 | * Preload a CPU with one object for "no edge" split case. The | |
404 | * aim is to get rid of allocations from the atomic context, thus | |
405 | * to use more permissive allocation masks. | |
406 | */ | |
407 | static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node); | |
408 | ||
68ad4a33 URS |
409 | static __always_inline unsigned long |
410 | va_size(struct vmap_area *va) | |
411 | { | |
412 | return (va->va_end - va->va_start); | |
413 | } | |
414 | ||
415 | static __always_inline unsigned long | |
416 | get_subtree_max_size(struct rb_node *node) | |
417 | { | |
418 | struct vmap_area *va; | |
419 | ||
420 | va = rb_entry_safe(node, struct vmap_area, rb_node); | |
421 | return va ? va->subtree_max_size : 0; | |
422 | } | |
89699605 | 423 | |
68ad4a33 URS |
424 | /* |
425 | * Gets called when remove the node and rotate. | |
426 | */ | |
427 | static __always_inline unsigned long | |
428 | compute_subtree_max_size(struct vmap_area *va) | |
429 | { | |
430 | return max3(va_size(va), | |
431 | get_subtree_max_size(va->rb_node.rb_left), | |
432 | get_subtree_max_size(va->rb_node.rb_right)); | |
433 | } | |
434 | ||
315cc066 ML |
435 | RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb, |
436 | struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size) | |
68ad4a33 URS |
437 | |
438 | static void purge_vmap_area_lazy(void); | |
439 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); | |
440 | static unsigned long lazy_max_pages(void); | |
db64fe02 | 441 | |
97105f0a RG |
442 | static atomic_long_t nr_vmalloc_pages; |
443 | ||
444 | unsigned long vmalloc_nr_pages(void) | |
445 | { | |
446 | return atomic_long_read(&nr_vmalloc_pages); | |
447 | } | |
448 | ||
db64fe02 | 449 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
1da177e4 | 450 | { |
db64fe02 NP |
451 | struct rb_node *n = vmap_area_root.rb_node; |
452 | ||
453 | while (n) { | |
454 | struct vmap_area *va; | |
455 | ||
456 | va = rb_entry(n, struct vmap_area, rb_node); | |
457 | if (addr < va->va_start) | |
458 | n = n->rb_left; | |
cef2ac3f | 459 | else if (addr >= va->va_end) |
db64fe02 NP |
460 | n = n->rb_right; |
461 | else | |
462 | return va; | |
463 | } | |
464 | ||
465 | return NULL; | |
466 | } | |
467 | ||
68ad4a33 URS |
468 | /* |
469 | * This function returns back addresses of parent node | |
470 | * and its left or right link for further processing. | |
471 | */ | |
472 | static __always_inline struct rb_node ** | |
473 | find_va_links(struct vmap_area *va, | |
474 | struct rb_root *root, struct rb_node *from, | |
475 | struct rb_node **parent) | |
476 | { | |
477 | struct vmap_area *tmp_va; | |
478 | struct rb_node **link; | |
479 | ||
480 | if (root) { | |
481 | link = &root->rb_node; | |
482 | if (unlikely(!*link)) { | |
483 | *parent = NULL; | |
484 | return link; | |
485 | } | |
486 | } else { | |
487 | link = &from; | |
488 | } | |
db64fe02 | 489 | |
68ad4a33 URS |
490 | /* |
491 | * Go to the bottom of the tree. When we hit the last point | |
492 | * we end up with parent rb_node and correct direction, i name | |
493 | * it link, where the new va->rb_node will be attached to. | |
494 | */ | |
495 | do { | |
496 | tmp_va = rb_entry(*link, struct vmap_area, rb_node); | |
db64fe02 | 497 | |
68ad4a33 URS |
498 | /* |
499 | * During the traversal we also do some sanity check. | |
500 | * Trigger the BUG() if there are sides(left/right) | |
501 | * or full overlaps. | |
502 | */ | |
503 | if (va->va_start < tmp_va->va_end && | |
504 | va->va_end <= tmp_va->va_start) | |
505 | link = &(*link)->rb_left; | |
506 | else if (va->va_end > tmp_va->va_start && | |
507 | va->va_start >= tmp_va->va_end) | |
508 | link = &(*link)->rb_right; | |
db64fe02 NP |
509 | else |
510 | BUG(); | |
68ad4a33 URS |
511 | } while (*link); |
512 | ||
513 | *parent = &tmp_va->rb_node; | |
514 | return link; | |
515 | } | |
516 | ||
517 | static __always_inline struct list_head * | |
518 | get_va_next_sibling(struct rb_node *parent, struct rb_node **link) | |
519 | { | |
520 | struct list_head *list; | |
521 | ||
522 | if (unlikely(!parent)) | |
523 | /* | |
524 | * The red-black tree where we try to find VA neighbors | |
525 | * before merging or inserting is empty, i.e. it means | |
526 | * there is no free vmap space. Normally it does not | |
527 | * happen but we handle this case anyway. | |
528 | */ | |
529 | return NULL; | |
530 | ||
531 | list = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
532 | return (&parent->rb_right == link ? list->next : list); | |
533 | } | |
534 | ||
535 | static __always_inline void | |
536 | link_va(struct vmap_area *va, struct rb_root *root, | |
537 | struct rb_node *parent, struct rb_node **link, struct list_head *head) | |
538 | { | |
539 | /* | |
540 | * VA is still not in the list, but we can | |
541 | * identify its future previous list_head node. | |
542 | */ | |
543 | if (likely(parent)) { | |
544 | head = &rb_entry(parent, struct vmap_area, rb_node)->list; | |
545 | if (&parent->rb_right != link) | |
546 | head = head->prev; | |
db64fe02 NP |
547 | } |
548 | ||
68ad4a33 URS |
549 | /* Insert to the rb-tree */ |
550 | rb_link_node(&va->rb_node, parent, link); | |
551 | if (root == &free_vmap_area_root) { | |
552 | /* | |
553 | * Some explanation here. Just perform simple insertion | |
554 | * to the tree. We do not set va->subtree_max_size to | |
555 | * its current size before calling rb_insert_augmented(). | |
556 | * It is because of we populate the tree from the bottom | |
557 | * to parent levels when the node _is_ in the tree. | |
558 | * | |
559 | * Therefore we set subtree_max_size to zero after insertion, | |
560 | * to let __augment_tree_propagate_from() puts everything to | |
561 | * the correct order later on. | |
562 | */ | |
563 | rb_insert_augmented(&va->rb_node, | |
564 | root, &free_vmap_area_rb_augment_cb); | |
565 | va->subtree_max_size = 0; | |
566 | } else { | |
567 | rb_insert_color(&va->rb_node, root); | |
568 | } | |
db64fe02 | 569 | |
68ad4a33 URS |
570 | /* Address-sort this list */ |
571 | list_add(&va->list, head); | |
db64fe02 NP |
572 | } |
573 | ||
68ad4a33 URS |
574 | static __always_inline void |
575 | unlink_va(struct vmap_area *va, struct rb_root *root) | |
576 | { | |
460e42d1 URS |
577 | if (WARN_ON(RB_EMPTY_NODE(&va->rb_node))) |
578 | return; | |
db64fe02 | 579 | |
460e42d1 URS |
580 | if (root == &free_vmap_area_root) |
581 | rb_erase_augmented(&va->rb_node, | |
582 | root, &free_vmap_area_rb_augment_cb); | |
583 | else | |
584 | rb_erase(&va->rb_node, root); | |
585 | ||
586 | list_del(&va->list); | |
587 | RB_CLEAR_NODE(&va->rb_node); | |
68ad4a33 URS |
588 | } |
589 | ||
bb850f4d URS |
590 | #if DEBUG_AUGMENT_PROPAGATE_CHECK |
591 | static void | |
592 | augment_tree_propagate_check(struct rb_node *n) | |
593 | { | |
594 | struct vmap_area *va; | |
595 | struct rb_node *node; | |
596 | unsigned long size; | |
597 | bool found = false; | |
598 | ||
599 | if (n == NULL) | |
600 | return; | |
601 | ||
602 | va = rb_entry(n, struct vmap_area, rb_node); | |
603 | size = va->subtree_max_size; | |
604 | node = n; | |
605 | ||
606 | while (node) { | |
607 | va = rb_entry(node, struct vmap_area, rb_node); | |
608 | ||
609 | if (get_subtree_max_size(node->rb_left) == size) { | |
610 | node = node->rb_left; | |
611 | } else { | |
612 | if (va_size(va) == size) { | |
613 | found = true; | |
614 | break; | |
615 | } | |
616 | ||
617 | node = node->rb_right; | |
618 | } | |
619 | } | |
620 | ||
621 | if (!found) { | |
622 | va = rb_entry(n, struct vmap_area, rb_node); | |
623 | pr_emerg("tree is corrupted: %lu, %lu\n", | |
624 | va_size(va), va->subtree_max_size); | |
625 | } | |
626 | ||
627 | augment_tree_propagate_check(n->rb_left); | |
628 | augment_tree_propagate_check(n->rb_right); | |
629 | } | |
630 | #endif | |
631 | ||
68ad4a33 URS |
632 | /* |
633 | * This function populates subtree_max_size from bottom to upper | |
634 | * levels starting from VA point. The propagation must be done | |
635 | * when VA size is modified by changing its va_start/va_end. Or | |
636 | * in case of newly inserting of VA to the tree. | |
637 | * | |
638 | * It means that __augment_tree_propagate_from() must be called: | |
639 | * - After VA has been inserted to the tree(free path); | |
640 | * - After VA has been shrunk(allocation path); | |
641 | * - After VA has been increased(merging path). | |
642 | * | |
643 | * Please note that, it does not mean that upper parent nodes | |
644 | * and their subtree_max_size are recalculated all the time up | |
645 | * to the root node. | |
646 | * | |
647 | * 4--8 | |
648 | * /\ | |
649 | * / \ | |
650 | * / \ | |
651 | * 2--2 8--8 | |
652 | * | |
653 | * For example if we modify the node 4, shrinking it to 2, then | |
654 | * no any modification is required. If we shrink the node 2 to 1 | |
655 | * its subtree_max_size is updated only, and set to 1. If we shrink | |
656 | * the node 8 to 6, then its subtree_max_size is set to 6 and parent | |
657 | * node becomes 4--6. | |
658 | */ | |
659 | static __always_inline void | |
660 | augment_tree_propagate_from(struct vmap_area *va) | |
661 | { | |
662 | struct rb_node *node = &va->rb_node; | |
663 | unsigned long new_va_sub_max_size; | |
664 | ||
665 | while (node) { | |
666 | va = rb_entry(node, struct vmap_area, rb_node); | |
667 | new_va_sub_max_size = compute_subtree_max_size(va); | |
668 | ||
669 | /* | |
670 | * If the newly calculated maximum available size of the | |
671 | * subtree is equal to the current one, then it means that | |
672 | * the tree is propagated correctly. So we have to stop at | |
673 | * this point to save cycles. | |
674 | */ | |
675 | if (va->subtree_max_size == new_va_sub_max_size) | |
676 | break; | |
677 | ||
678 | va->subtree_max_size = new_va_sub_max_size; | |
679 | node = rb_parent(&va->rb_node); | |
680 | } | |
bb850f4d URS |
681 | |
682 | #if DEBUG_AUGMENT_PROPAGATE_CHECK | |
683 | augment_tree_propagate_check(free_vmap_area_root.rb_node); | |
684 | #endif | |
68ad4a33 URS |
685 | } |
686 | ||
687 | static void | |
688 | insert_vmap_area(struct vmap_area *va, | |
689 | struct rb_root *root, struct list_head *head) | |
690 | { | |
691 | struct rb_node **link; | |
692 | struct rb_node *parent; | |
693 | ||
694 | link = find_va_links(va, root, NULL, &parent); | |
695 | link_va(va, root, parent, link, head); | |
696 | } | |
697 | ||
698 | static void | |
699 | insert_vmap_area_augment(struct vmap_area *va, | |
700 | struct rb_node *from, struct rb_root *root, | |
701 | struct list_head *head) | |
702 | { | |
703 | struct rb_node **link; | |
704 | struct rb_node *parent; | |
705 | ||
706 | if (from) | |
707 | link = find_va_links(va, NULL, from, &parent); | |
708 | else | |
709 | link = find_va_links(va, root, NULL, &parent); | |
710 | ||
711 | link_va(va, root, parent, link, head); | |
712 | augment_tree_propagate_from(va); | |
713 | } | |
714 | ||
715 | /* | |
716 | * Merge de-allocated chunk of VA memory with previous | |
717 | * and next free blocks. If coalesce is not done a new | |
718 | * free area is inserted. If VA has been merged, it is | |
719 | * freed. | |
720 | */ | |
3c5c3cfb | 721 | static __always_inline struct vmap_area * |
68ad4a33 URS |
722 | merge_or_add_vmap_area(struct vmap_area *va, |
723 | struct rb_root *root, struct list_head *head) | |
724 | { | |
725 | struct vmap_area *sibling; | |
726 | struct list_head *next; | |
727 | struct rb_node **link; | |
728 | struct rb_node *parent; | |
729 | bool merged = false; | |
730 | ||
731 | /* | |
732 | * Find a place in the tree where VA potentially will be | |
733 | * inserted, unless it is merged with its sibling/siblings. | |
734 | */ | |
735 | link = find_va_links(va, root, NULL, &parent); | |
736 | ||
737 | /* | |
738 | * Get next node of VA to check if merging can be done. | |
739 | */ | |
740 | next = get_va_next_sibling(parent, link); | |
741 | if (unlikely(next == NULL)) | |
742 | goto insert; | |
743 | ||
744 | /* | |
745 | * start end | |
746 | * | | | |
747 | * |<------VA------>|<-----Next----->| | |
748 | * | | | |
749 | * start end | |
750 | */ | |
751 | if (next != head) { | |
752 | sibling = list_entry(next, struct vmap_area, list); | |
753 | if (sibling->va_start == va->va_end) { | |
754 | sibling->va_start = va->va_start; | |
755 | ||
756 | /* Check and update the tree if needed. */ | |
757 | augment_tree_propagate_from(sibling); | |
758 | ||
68ad4a33 URS |
759 | /* Free vmap_area object. */ |
760 | kmem_cache_free(vmap_area_cachep, va); | |
761 | ||
762 | /* Point to the new merged area. */ | |
763 | va = sibling; | |
764 | merged = true; | |
765 | } | |
766 | } | |
767 | ||
768 | /* | |
769 | * start end | |
770 | * | | | |
771 | * |<-----Prev----->|<------VA------>| | |
772 | * | | | |
773 | * start end | |
774 | */ | |
775 | if (next->prev != head) { | |
776 | sibling = list_entry(next->prev, struct vmap_area, list); | |
777 | if (sibling->va_end == va->va_start) { | |
778 | sibling->va_end = va->va_end; | |
779 | ||
780 | /* Check and update the tree if needed. */ | |
781 | augment_tree_propagate_from(sibling); | |
782 | ||
54f63d9d URS |
783 | if (merged) |
784 | unlink_va(va, root); | |
68ad4a33 URS |
785 | |
786 | /* Free vmap_area object. */ | |
787 | kmem_cache_free(vmap_area_cachep, va); | |
3c5c3cfb DA |
788 | |
789 | /* Point to the new merged area. */ | |
790 | va = sibling; | |
791 | merged = true; | |
68ad4a33 URS |
792 | } |
793 | } | |
794 | ||
795 | insert: | |
796 | if (!merged) { | |
797 | link_va(va, root, parent, link, head); | |
798 | augment_tree_propagate_from(va); | |
799 | } | |
3c5c3cfb DA |
800 | |
801 | return va; | |
68ad4a33 URS |
802 | } |
803 | ||
804 | static __always_inline bool | |
805 | is_within_this_va(struct vmap_area *va, unsigned long size, | |
806 | unsigned long align, unsigned long vstart) | |
807 | { | |
808 | unsigned long nva_start_addr; | |
809 | ||
810 | if (va->va_start > vstart) | |
811 | nva_start_addr = ALIGN(va->va_start, align); | |
812 | else | |
813 | nva_start_addr = ALIGN(vstart, align); | |
814 | ||
815 | /* Can be overflowed due to big size or alignment. */ | |
816 | if (nva_start_addr + size < nva_start_addr || | |
817 | nva_start_addr < vstart) | |
818 | return false; | |
819 | ||
820 | return (nva_start_addr + size <= va->va_end); | |
821 | } | |
822 | ||
823 | /* | |
824 | * Find the first free block(lowest start address) in the tree, | |
825 | * that will accomplish the request corresponding to passing | |
826 | * parameters. | |
827 | */ | |
828 | static __always_inline struct vmap_area * | |
829 | find_vmap_lowest_match(unsigned long size, | |
830 | unsigned long align, unsigned long vstart) | |
831 | { | |
832 | struct vmap_area *va; | |
833 | struct rb_node *node; | |
834 | unsigned long length; | |
835 | ||
836 | /* Start from the root. */ | |
837 | node = free_vmap_area_root.rb_node; | |
838 | ||
839 | /* Adjust the search size for alignment overhead. */ | |
840 | length = size + align - 1; | |
841 | ||
842 | while (node) { | |
843 | va = rb_entry(node, struct vmap_area, rb_node); | |
844 | ||
845 | if (get_subtree_max_size(node->rb_left) >= length && | |
846 | vstart < va->va_start) { | |
847 | node = node->rb_left; | |
848 | } else { | |
849 | if (is_within_this_va(va, size, align, vstart)) | |
850 | return va; | |
851 | ||
852 | /* | |
853 | * Does not make sense to go deeper towards the right | |
854 | * sub-tree if it does not have a free block that is | |
855 | * equal or bigger to the requested search length. | |
856 | */ | |
857 | if (get_subtree_max_size(node->rb_right) >= length) { | |
858 | node = node->rb_right; | |
859 | continue; | |
860 | } | |
861 | ||
862 | /* | |
3806b041 | 863 | * OK. We roll back and find the first right sub-tree, |
68ad4a33 URS |
864 | * that will satisfy the search criteria. It can happen |
865 | * only once due to "vstart" restriction. | |
866 | */ | |
867 | while ((node = rb_parent(node))) { | |
868 | va = rb_entry(node, struct vmap_area, rb_node); | |
869 | if (is_within_this_va(va, size, align, vstart)) | |
870 | return va; | |
871 | ||
872 | if (get_subtree_max_size(node->rb_right) >= length && | |
873 | vstart <= va->va_start) { | |
874 | node = node->rb_right; | |
875 | break; | |
876 | } | |
877 | } | |
878 | } | |
879 | } | |
880 | ||
881 | return NULL; | |
882 | } | |
883 | ||
a6cf4e0f URS |
884 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
885 | #include <linux/random.h> | |
886 | ||
887 | static struct vmap_area * | |
888 | find_vmap_lowest_linear_match(unsigned long size, | |
889 | unsigned long align, unsigned long vstart) | |
890 | { | |
891 | struct vmap_area *va; | |
892 | ||
893 | list_for_each_entry(va, &free_vmap_area_list, list) { | |
894 | if (!is_within_this_va(va, size, align, vstart)) | |
895 | continue; | |
896 | ||
897 | return va; | |
898 | } | |
899 | ||
900 | return NULL; | |
901 | } | |
902 | ||
903 | static void | |
904 | find_vmap_lowest_match_check(unsigned long size) | |
905 | { | |
906 | struct vmap_area *va_1, *va_2; | |
907 | unsigned long vstart; | |
908 | unsigned int rnd; | |
909 | ||
910 | get_random_bytes(&rnd, sizeof(rnd)); | |
911 | vstart = VMALLOC_START + rnd; | |
912 | ||
913 | va_1 = find_vmap_lowest_match(size, 1, vstart); | |
914 | va_2 = find_vmap_lowest_linear_match(size, 1, vstart); | |
915 | ||
916 | if (va_1 != va_2) | |
917 | pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", | |
918 | va_1, va_2, vstart); | |
919 | } | |
920 | #endif | |
921 | ||
68ad4a33 URS |
922 | enum fit_type { |
923 | NOTHING_FIT = 0, | |
924 | FL_FIT_TYPE = 1, /* full fit */ | |
925 | LE_FIT_TYPE = 2, /* left edge fit */ | |
926 | RE_FIT_TYPE = 3, /* right edge fit */ | |
927 | NE_FIT_TYPE = 4 /* no edge fit */ | |
928 | }; | |
929 | ||
930 | static __always_inline enum fit_type | |
931 | classify_va_fit_type(struct vmap_area *va, | |
932 | unsigned long nva_start_addr, unsigned long size) | |
933 | { | |
934 | enum fit_type type; | |
935 | ||
936 | /* Check if it is within VA. */ | |
937 | if (nva_start_addr < va->va_start || | |
938 | nva_start_addr + size > va->va_end) | |
939 | return NOTHING_FIT; | |
940 | ||
941 | /* Now classify. */ | |
942 | if (va->va_start == nva_start_addr) { | |
943 | if (va->va_end == nva_start_addr + size) | |
944 | type = FL_FIT_TYPE; | |
945 | else | |
946 | type = LE_FIT_TYPE; | |
947 | } else if (va->va_end == nva_start_addr + size) { | |
948 | type = RE_FIT_TYPE; | |
949 | } else { | |
950 | type = NE_FIT_TYPE; | |
951 | } | |
952 | ||
953 | return type; | |
954 | } | |
955 | ||
956 | static __always_inline int | |
957 | adjust_va_to_fit_type(struct vmap_area *va, | |
958 | unsigned long nva_start_addr, unsigned long size, | |
959 | enum fit_type type) | |
960 | { | |
2c929233 | 961 | struct vmap_area *lva = NULL; |
68ad4a33 URS |
962 | |
963 | if (type == FL_FIT_TYPE) { | |
964 | /* | |
965 | * No need to split VA, it fully fits. | |
966 | * | |
967 | * | | | |
968 | * V NVA V | |
969 | * |---------------| | |
970 | */ | |
971 | unlink_va(va, &free_vmap_area_root); | |
972 | kmem_cache_free(vmap_area_cachep, va); | |
973 | } else if (type == LE_FIT_TYPE) { | |
974 | /* | |
975 | * Split left edge of fit VA. | |
976 | * | |
977 | * | | | |
978 | * V NVA V R | |
979 | * |-------|-------| | |
980 | */ | |
981 | va->va_start += size; | |
982 | } else if (type == RE_FIT_TYPE) { | |
983 | /* | |
984 | * Split right edge of fit VA. | |
985 | * | |
986 | * | | | |
987 | * L V NVA V | |
988 | * |-------|-------| | |
989 | */ | |
990 | va->va_end = nva_start_addr; | |
991 | } else if (type == NE_FIT_TYPE) { | |
992 | /* | |
993 | * Split no edge of fit VA. | |
994 | * | |
995 | * | | | |
996 | * L V NVA V R | |
997 | * |---|-------|---| | |
998 | */ | |
82dd23e8 URS |
999 | lva = __this_cpu_xchg(ne_fit_preload_node, NULL); |
1000 | if (unlikely(!lva)) { | |
1001 | /* | |
1002 | * For percpu allocator we do not do any pre-allocation | |
1003 | * and leave it as it is. The reason is it most likely | |
1004 | * never ends up with NE_FIT_TYPE splitting. In case of | |
1005 | * percpu allocations offsets and sizes are aligned to | |
1006 | * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE | |
1007 | * are its main fitting cases. | |
1008 | * | |
1009 | * There are a few exceptions though, as an example it is | |
1010 | * a first allocation (early boot up) when we have "one" | |
1011 | * big free space that has to be split. | |
060650a2 URS |
1012 | * |
1013 | * Also we can hit this path in case of regular "vmap" | |
1014 | * allocations, if "this" current CPU was not preloaded. | |
1015 | * See the comment in alloc_vmap_area() why. If so, then | |
1016 | * GFP_NOWAIT is used instead to get an extra object for | |
1017 | * split purpose. That is rare and most time does not | |
1018 | * occur. | |
1019 | * | |
1020 | * What happens if an allocation gets failed. Basically, | |
1021 | * an "overflow" path is triggered to purge lazily freed | |
1022 | * areas to free some memory, then, the "retry" path is | |
1023 | * triggered to repeat one more time. See more details | |
1024 | * in alloc_vmap_area() function. | |
82dd23e8 URS |
1025 | */ |
1026 | lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); | |
1027 | if (!lva) | |
1028 | return -1; | |
1029 | } | |
68ad4a33 URS |
1030 | |
1031 | /* | |
1032 | * Build the remainder. | |
1033 | */ | |
1034 | lva->va_start = va->va_start; | |
1035 | lva->va_end = nva_start_addr; | |
1036 | ||
1037 | /* | |
1038 | * Shrink this VA to remaining size. | |
1039 | */ | |
1040 | va->va_start = nva_start_addr + size; | |
1041 | } else { | |
1042 | return -1; | |
1043 | } | |
1044 | ||
1045 | if (type != FL_FIT_TYPE) { | |
1046 | augment_tree_propagate_from(va); | |
1047 | ||
2c929233 | 1048 | if (lva) /* type == NE_FIT_TYPE */ |
68ad4a33 URS |
1049 | insert_vmap_area_augment(lva, &va->rb_node, |
1050 | &free_vmap_area_root, &free_vmap_area_list); | |
1051 | } | |
1052 | ||
1053 | return 0; | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Returns a start address of the newly allocated area, if success. | |
1058 | * Otherwise a vend is returned that indicates failure. | |
1059 | */ | |
1060 | static __always_inline unsigned long | |
1061 | __alloc_vmap_area(unsigned long size, unsigned long align, | |
cacca6ba | 1062 | unsigned long vstart, unsigned long vend) |
68ad4a33 URS |
1063 | { |
1064 | unsigned long nva_start_addr; | |
1065 | struct vmap_area *va; | |
1066 | enum fit_type type; | |
1067 | int ret; | |
1068 | ||
1069 | va = find_vmap_lowest_match(size, align, vstart); | |
1070 | if (unlikely(!va)) | |
1071 | return vend; | |
1072 | ||
1073 | if (va->va_start > vstart) | |
1074 | nva_start_addr = ALIGN(va->va_start, align); | |
1075 | else | |
1076 | nva_start_addr = ALIGN(vstart, align); | |
1077 | ||
1078 | /* Check the "vend" restriction. */ | |
1079 | if (nva_start_addr + size > vend) | |
1080 | return vend; | |
1081 | ||
1082 | /* Classify what we have found. */ | |
1083 | type = classify_va_fit_type(va, nva_start_addr, size); | |
1084 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
1085 | return vend; | |
1086 | ||
1087 | /* Update the free vmap_area. */ | |
1088 | ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); | |
1089 | if (ret) | |
1090 | return vend; | |
1091 | ||
a6cf4e0f URS |
1092 | #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK |
1093 | find_vmap_lowest_match_check(size); | |
1094 | #endif | |
1095 | ||
68ad4a33 URS |
1096 | return nva_start_addr; |
1097 | } | |
4da56b99 | 1098 | |
d98c9e83 AR |
1099 | /* |
1100 | * Free a region of KVA allocated by alloc_vmap_area | |
1101 | */ | |
1102 | static void free_vmap_area(struct vmap_area *va) | |
1103 | { | |
1104 | /* | |
1105 | * Remove from the busy tree/list. | |
1106 | */ | |
1107 | spin_lock(&vmap_area_lock); | |
1108 | unlink_va(va, &vmap_area_root); | |
1109 | spin_unlock(&vmap_area_lock); | |
1110 | ||
1111 | /* | |
1112 | * Insert/Merge it back to the free tree/list. | |
1113 | */ | |
1114 | spin_lock(&free_vmap_area_lock); | |
1115 | merge_or_add_vmap_area(va, &free_vmap_area_root, &free_vmap_area_list); | |
1116 | spin_unlock(&free_vmap_area_lock); | |
1117 | } | |
1118 | ||
db64fe02 NP |
1119 | /* |
1120 | * Allocate a region of KVA of the specified size and alignment, within the | |
1121 | * vstart and vend. | |
1122 | */ | |
1123 | static struct vmap_area *alloc_vmap_area(unsigned long size, | |
1124 | unsigned long align, | |
1125 | unsigned long vstart, unsigned long vend, | |
1126 | int node, gfp_t gfp_mask) | |
1127 | { | |
82dd23e8 | 1128 | struct vmap_area *va, *pva; |
1da177e4 | 1129 | unsigned long addr; |
db64fe02 | 1130 | int purged = 0; |
d98c9e83 | 1131 | int ret; |
db64fe02 | 1132 | |
7766970c | 1133 | BUG_ON(!size); |
891c49ab | 1134 | BUG_ON(offset_in_page(size)); |
89699605 | 1135 | BUG_ON(!is_power_of_2(align)); |
db64fe02 | 1136 | |
68ad4a33 URS |
1137 | if (unlikely(!vmap_initialized)) |
1138 | return ERR_PTR(-EBUSY); | |
1139 | ||
5803ed29 | 1140 | might_sleep(); |
f07116d7 | 1141 | gfp_mask = gfp_mask & GFP_RECLAIM_MASK; |
4da56b99 | 1142 | |
f07116d7 | 1143 | va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
db64fe02 NP |
1144 | if (unlikely(!va)) |
1145 | return ERR_PTR(-ENOMEM); | |
1146 | ||
7f88f88f CM |
1147 | /* |
1148 | * Only scan the relevant parts containing pointers to other objects | |
1149 | * to avoid false negatives. | |
1150 | */ | |
f07116d7 | 1151 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask); |
7f88f88f | 1152 | |
db64fe02 | 1153 | retry: |
82dd23e8 | 1154 | /* |
81f1ba58 URS |
1155 | * Preload this CPU with one extra vmap_area object. It is used |
1156 | * when fit type of free area is NE_FIT_TYPE. Please note, it | |
1157 | * does not guarantee that an allocation occurs on a CPU that | |
1158 | * is preloaded, instead we minimize the case when it is not. | |
1159 | * It can happen because of cpu migration, because there is a | |
1160 | * race until the below spinlock is taken. | |
82dd23e8 URS |
1161 | * |
1162 | * The preload is done in non-atomic context, thus it allows us | |
1163 | * to use more permissive allocation masks to be more stable under | |
81f1ba58 URS |
1164 | * low memory condition and high memory pressure. In rare case, |
1165 | * if not preloaded, GFP_NOWAIT is used. | |
82dd23e8 | 1166 | * |
81f1ba58 | 1167 | * Set "pva" to NULL here, because of "retry" path. |
82dd23e8 | 1168 | */ |
81f1ba58 | 1169 | pva = NULL; |
82dd23e8 | 1170 | |
81f1ba58 URS |
1171 | if (!this_cpu_read(ne_fit_preload_node)) |
1172 | /* | |
1173 | * Even if it fails we do not really care about that. | |
1174 | * Just proceed as it is. If needed "overflow" path | |
1175 | * will refill the cache we allocate from. | |
1176 | */ | |
f07116d7 | 1177 | pva = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node); |
82dd23e8 | 1178 | |
e36176be | 1179 | spin_lock(&free_vmap_area_lock); |
81f1ba58 URS |
1180 | |
1181 | if (pva && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, pva)) | |
1182 | kmem_cache_free(vmap_area_cachep, pva); | |
89699605 | 1183 | |
afd07389 | 1184 | /* |
68ad4a33 URS |
1185 | * If an allocation fails, the "vend" address is |
1186 | * returned. Therefore trigger the overflow path. | |
afd07389 | 1187 | */ |
cacca6ba | 1188 | addr = __alloc_vmap_area(size, align, vstart, vend); |
e36176be URS |
1189 | spin_unlock(&free_vmap_area_lock); |
1190 | ||
68ad4a33 | 1191 | if (unlikely(addr == vend)) |
89699605 | 1192 | goto overflow; |
db64fe02 NP |
1193 | |
1194 | va->va_start = addr; | |
1195 | va->va_end = addr + size; | |
688fcbfc | 1196 | va->vm = NULL; |
68ad4a33 | 1197 | |
d98c9e83 | 1198 | |
e36176be URS |
1199 | spin_lock(&vmap_area_lock); |
1200 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); | |
db64fe02 NP |
1201 | spin_unlock(&vmap_area_lock); |
1202 | ||
61e16557 | 1203 | BUG_ON(!IS_ALIGNED(va->va_start, align)); |
89699605 NP |
1204 | BUG_ON(va->va_start < vstart); |
1205 | BUG_ON(va->va_end > vend); | |
1206 | ||
d98c9e83 AR |
1207 | ret = kasan_populate_vmalloc(addr, size); |
1208 | if (ret) { | |
1209 | free_vmap_area(va); | |
1210 | return ERR_PTR(ret); | |
1211 | } | |
1212 | ||
db64fe02 | 1213 | return va; |
89699605 NP |
1214 | |
1215 | overflow: | |
89699605 NP |
1216 | if (!purged) { |
1217 | purge_vmap_area_lazy(); | |
1218 | purged = 1; | |
1219 | goto retry; | |
1220 | } | |
4da56b99 CW |
1221 | |
1222 | if (gfpflags_allow_blocking(gfp_mask)) { | |
1223 | unsigned long freed = 0; | |
1224 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed); | |
1225 | if (freed > 0) { | |
1226 | purged = 0; | |
1227 | goto retry; | |
1228 | } | |
1229 | } | |
1230 | ||
03497d76 | 1231 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) |
756a025f JP |
1232 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n", |
1233 | size); | |
68ad4a33 URS |
1234 | |
1235 | kmem_cache_free(vmap_area_cachep, va); | |
89699605 | 1236 | return ERR_PTR(-EBUSY); |
db64fe02 NP |
1237 | } |
1238 | ||
4da56b99 CW |
1239 | int register_vmap_purge_notifier(struct notifier_block *nb) |
1240 | { | |
1241 | return blocking_notifier_chain_register(&vmap_notify_list, nb); | |
1242 | } | |
1243 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier); | |
1244 | ||
1245 | int unregister_vmap_purge_notifier(struct notifier_block *nb) | |
1246 | { | |
1247 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb); | |
1248 | } | |
1249 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier); | |
1250 | ||
db64fe02 NP |
1251 | /* |
1252 | * Clear the pagetable entries of a given vmap_area | |
1253 | */ | |
1254 | static void unmap_vmap_area(struct vmap_area *va) | |
1255 | { | |
b521c43f | 1256 | unmap_kernel_range_noflush(va->va_start, va->va_end - va->va_start); |
db64fe02 NP |
1257 | } |
1258 | ||
1259 | /* | |
1260 | * lazy_max_pages is the maximum amount of virtual address space we gather up | |
1261 | * before attempting to purge with a TLB flush. | |
1262 | * | |
1263 | * There is a tradeoff here: a larger number will cover more kernel page tables | |
1264 | * and take slightly longer to purge, but it will linearly reduce the number of | |
1265 | * global TLB flushes that must be performed. It would seem natural to scale | |
1266 | * this number up linearly with the number of CPUs (because vmapping activity | |
1267 | * could also scale linearly with the number of CPUs), however it is likely | |
1268 | * that in practice, workloads might be constrained in other ways that mean | |
1269 | * vmap activity will not scale linearly with CPUs. Also, I want to be | |
1270 | * conservative and not introduce a big latency on huge systems, so go with | |
1271 | * a less aggressive log scale. It will still be an improvement over the old | |
1272 | * code, and it will be simple to change the scale factor if we find that it | |
1273 | * becomes a problem on bigger systems. | |
1274 | */ | |
1275 | static unsigned long lazy_max_pages(void) | |
1276 | { | |
1277 | unsigned int log; | |
1278 | ||
1279 | log = fls(num_online_cpus()); | |
1280 | ||
1281 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); | |
1282 | } | |
1283 | ||
4d36e6f8 | 1284 | static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0); |
db64fe02 | 1285 | |
0574ecd1 CH |
1286 | /* |
1287 | * Serialize vmap purging. There is no actual criticial section protected | |
1288 | * by this look, but we want to avoid concurrent calls for performance | |
1289 | * reasons and to make the pcpu_get_vm_areas more deterministic. | |
1290 | */ | |
f9e09977 | 1291 | static DEFINE_MUTEX(vmap_purge_lock); |
0574ecd1 | 1292 | |
02b709df NP |
1293 | /* for per-CPU blocks */ |
1294 | static void purge_fragmented_blocks_allcpus(void); | |
1295 | ||
3ee48b6a CW |
1296 | /* |
1297 | * called before a call to iounmap() if the caller wants vm_area_struct's | |
1298 | * immediately freed. | |
1299 | */ | |
1300 | void set_iounmap_nonlazy(void) | |
1301 | { | |
4d36e6f8 | 1302 | atomic_long_set(&vmap_lazy_nr, lazy_max_pages()+1); |
3ee48b6a CW |
1303 | } |
1304 | ||
db64fe02 NP |
1305 | /* |
1306 | * Purges all lazily-freed vmap areas. | |
db64fe02 | 1307 | */ |
0574ecd1 | 1308 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end) |
db64fe02 | 1309 | { |
4d36e6f8 | 1310 | unsigned long resched_threshold; |
80c4bd7a | 1311 | struct llist_node *valist; |
db64fe02 | 1312 | struct vmap_area *va; |
cbb76676 | 1313 | struct vmap_area *n_va; |
db64fe02 | 1314 | |
0574ecd1 | 1315 | lockdep_assert_held(&vmap_purge_lock); |
02b709df | 1316 | |
80c4bd7a | 1317 | valist = llist_del_all(&vmap_purge_list); |
68571be9 URS |
1318 | if (unlikely(valist == NULL)) |
1319 | return false; | |
1320 | ||
3f8fd02b JR |
1321 | /* |
1322 | * First make sure the mappings are removed from all page-tables | |
1323 | * before they are freed. | |
1324 | */ | |
763802b5 | 1325 | vmalloc_sync_unmappings(); |
3f8fd02b | 1326 | |
68571be9 URS |
1327 | /* |
1328 | * TODO: to calculate a flush range without looping. | |
1329 | * The list can be up to lazy_max_pages() elements. | |
1330 | */ | |
80c4bd7a | 1331 | llist_for_each_entry(va, valist, purge_list) { |
0574ecd1 CH |
1332 | if (va->va_start < start) |
1333 | start = va->va_start; | |
1334 | if (va->va_end > end) | |
1335 | end = va->va_end; | |
db64fe02 | 1336 | } |
db64fe02 | 1337 | |
0574ecd1 | 1338 | flush_tlb_kernel_range(start, end); |
4d36e6f8 | 1339 | resched_threshold = lazy_max_pages() << 1; |
db64fe02 | 1340 | |
e36176be | 1341 | spin_lock(&free_vmap_area_lock); |
763b218d | 1342 | llist_for_each_entry_safe(va, n_va, valist, purge_list) { |
4d36e6f8 | 1343 | unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT; |
3c5c3cfb DA |
1344 | unsigned long orig_start = va->va_start; |
1345 | unsigned long orig_end = va->va_end; | |
763b218d | 1346 | |
dd3b8353 URS |
1347 | /* |
1348 | * Finally insert or merge lazily-freed area. It is | |
1349 | * detached and there is no need to "unlink" it from | |
1350 | * anything. | |
1351 | */ | |
3c5c3cfb DA |
1352 | va = merge_or_add_vmap_area(va, &free_vmap_area_root, |
1353 | &free_vmap_area_list); | |
1354 | ||
1355 | if (is_vmalloc_or_module_addr((void *)orig_start)) | |
1356 | kasan_release_vmalloc(orig_start, orig_end, | |
1357 | va->va_start, va->va_end); | |
dd3b8353 | 1358 | |
4d36e6f8 | 1359 | atomic_long_sub(nr, &vmap_lazy_nr); |
68571be9 | 1360 | |
4d36e6f8 | 1361 | if (atomic_long_read(&vmap_lazy_nr) < resched_threshold) |
e36176be | 1362 | cond_resched_lock(&free_vmap_area_lock); |
763b218d | 1363 | } |
e36176be | 1364 | spin_unlock(&free_vmap_area_lock); |
0574ecd1 | 1365 | return true; |
db64fe02 NP |
1366 | } |
1367 | ||
496850e5 NP |
1368 | /* |
1369 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody | |
1370 | * is already purging. | |
1371 | */ | |
1372 | static void try_purge_vmap_area_lazy(void) | |
1373 | { | |
f9e09977 | 1374 | if (mutex_trylock(&vmap_purge_lock)) { |
0574ecd1 | 1375 | __purge_vmap_area_lazy(ULONG_MAX, 0); |
f9e09977 | 1376 | mutex_unlock(&vmap_purge_lock); |
0574ecd1 | 1377 | } |
496850e5 NP |
1378 | } |
1379 | ||
db64fe02 NP |
1380 | /* |
1381 | * Kick off a purge of the outstanding lazy areas. | |
1382 | */ | |
1383 | static void purge_vmap_area_lazy(void) | |
1384 | { | |
f9e09977 | 1385 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1386 | purge_fragmented_blocks_allcpus(); |
1387 | __purge_vmap_area_lazy(ULONG_MAX, 0); | |
f9e09977 | 1388 | mutex_unlock(&vmap_purge_lock); |
db64fe02 NP |
1389 | } |
1390 | ||
1391 | /* | |
64141da5 JF |
1392 | * Free a vmap area, caller ensuring that the area has been unmapped |
1393 | * and flush_cache_vunmap had been called for the correct range | |
1394 | * previously. | |
db64fe02 | 1395 | */ |
64141da5 | 1396 | static void free_vmap_area_noflush(struct vmap_area *va) |
db64fe02 | 1397 | { |
4d36e6f8 | 1398 | unsigned long nr_lazy; |
80c4bd7a | 1399 | |
dd3b8353 URS |
1400 | spin_lock(&vmap_area_lock); |
1401 | unlink_va(va, &vmap_area_root); | |
1402 | spin_unlock(&vmap_area_lock); | |
1403 | ||
4d36e6f8 URS |
1404 | nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >> |
1405 | PAGE_SHIFT, &vmap_lazy_nr); | |
80c4bd7a CW |
1406 | |
1407 | /* After this point, we may free va at any time */ | |
1408 | llist_add(&va->purge_list, &vmap_purge_list); | |
1409 | ||
1410 | if (unlikely(nr_lazy > lazy_max_pages())) | |
496850e5 | 1411 | try_purge_vmap_area_lazy(); |
db64fe02 NP |
1412 | } |
1413 | ||
b29acbdc NP |
1414 | /* |
1415 | * Free and unmap a vmap area | |
1416 | */ | |
1417 | static void free_unmap_vmap_area(struct vmap_area *va) | |
1418 | { | |
1419 | flush_cache_vunmap(va->va_start, va->va_end); | |
c8eef01e | 1420 | unmap_vmap_area(va); |
8e57f8ac | 1421 | if (debug_pagealloc_enabled_static()) |
82a2e924 CP |
1422 | flush_tlb_kernel_range(va->va_start, va->va_end); |
1423 | ||
c8eef01e | 1424 | free_vmap_area_noflush(va); |
b29acbdc NP |
1425 | } |
1426 | ||
db64fe02 NP |
1427 | static struct vmap_area *find_vmap_area(unsigned long addr) |
1428 | { | |
1429 | struct vmap_area *va; | |
1430 | ||
1431 | spin_lock(&vmap_area_lock); | |
1432 | va = __find_vmap_area(addr); | |
1433 | spin_unlock(&vmap_area_lock); | |
1434 | ||
1435 | return va; | |
1436 | } | |
1437 | ||
db64fe02 NP |
1438 | /*** Per cpu kva allocator ***/ |
1439 | ||
1440 | /* | |
1441 | * vmap space is limited especially on 32 bit architectures. Ensure there is | |
1442 | * room for at least 16 percpu vmap blocks per CPU. | |
1443 | */ | |
1444 | /* | |
1445 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able | |
1446 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess | |
1447 | * instead (we just need a rough idea) | |
1448 | */ | |
1449 | #if BITS_PER_LONG == 32 | |
1450 | #define VMALLOC_SPACE (128UL*1024*1024) | |
1451 | #else | |
1452 | #define VMALLOC_SPACE (128UL*1024*1024*1024) | |
1453 | #endif | |
1454 | ||
1455 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) | |
1456 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ | |
1457 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ | |
1458 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) | |
1459 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ | |
1460 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ | |
f982f915 CL |
1461 | #define VMAP_BBMAP_BITS \ |
1462 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ | |
1463 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ | |
1464 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) | |
db64fe02 NP |
1465 | |
1466 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) | |
1467 | ||
1468 | struct vmap_block_queue { | |
1469 | spinlock_t lock; | |
1470 | struct list_head free; | |
db64fe02 NP |
1471 | }; |
1472 | ||
1473 | struct vmap_block { | |
1474 | spinlock_t lock; | |
1475 | struct vmap_area *va; | |
db64fe02 | 1476 | unsigned long free, dirty; |
7d61bfe8 | 1477 | unsigned long dirty_min, dirty_max; /*< dirty range */ |
de560423 NP |
1478 | struct list_head free_list; |
1479 | struct rcu_head rcu_head; | |
02b709df | 1480 | struct list_head purge; |
db64fe02 NP |
1481 | }; |
1482 | ||
1483 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ | |
1484 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); | |
1485 | ||
1486 | /* | |
1487 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block | |
1488 | * in the free path. Could get rid of this if we change the API to return a | |
1489 | * "cookie" from alloc, to be passed to free. But no big deal yet. | |
1490 | */ | |
1491 | static DEFINE_SPINLOCK(vmap_block_tree_lock); | |
1492 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); | |
1493 | ||
1494 | /* | |
1495 | * We should probably have a fallback mechanism to allocate virtual memory | |
1496 | * out of partially filled vmap blocks. However vmap block sizing should be | |
1497 | * fairly reasonable according to the vmalloc size, so it shouldn't be a | |
1498 | * big problem. | |
1499 | */ | |
1500 | ||
1501 | static unsigned long addr_to_vb_idx(unsigned long addr) | |
1502 | { | |
1503 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); | |
1504 | addr /= VMAP_BLOCK_SIZE; | |
1505 | return addr; | |
1506 | } | |
1507 | ||
cf725ce2 RP |
1508 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off) |
1509 | { | |
1510 | unsigned long addr; | |
1511 | ||
1512 | addr = va_start + (pages_off << PAGE_SHIFT); | |
1513 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start)); | |
1514 | return (void *)addr; | |
1515 | } | |
1516 | ||
1517 | /** | |
1518 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this | |
1519 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS | |
1520 | * @order: how many 2^order pages should be occupied in newly allocated block | |
1521 | * @gfp_mask: flags for the page level allocator | |
1522 | * | |
a862f68a | 1523 | * Return: virtual address in a newly allocated block or ERR_PTR(-errno) |
cf725ce2 RP |
1524 | */ |
1525 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) | |
db64fe02 NP |
1526 | { |
1527 | struct vmap_block_queue *vbq; | |
1528 | struct vmap_block *vb; | |
1529 | struct vmap_area *va; | |
1530 | unsigned long vb_idx; | |
1531 | int node, err; | |
cf725ce2 | 1532 | void *vaddr; |
db64fe02 NP |
1533 | |
1534 | node = numa_node_id(); | |
1535 | ||
1536 | vb = kmalloc_node(sizeof(struct vmap_block), | |
1537 | gfp_mask & GFP_RECLAIM_MASK, node); | |
1538 | if (unlikely(!vb)) | |
1539 | return ERR_PTR(-ENOMEM); | |
1540 | ||
1541 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, | |
1542 | VMALLOC_START, VMALLOC_END, | |
1543 | node, gfp_mask); | |
ddf9c6d4 | 1544 | if (IS_ERR(va)) { |
db64fe02 | 1545 | kfree(vb); |
e7d86340 | 1546 | return ERR_CAST(va); |
db64fe02 NP |
1547 | } |
1548 | ||
1549 | err = radix_tree_preload(gfp_mask); | |
1550 | if (unlikely(err)) { | |
1551 | kfree(vb); | |
1552 | free_vmap_area(va); | |
1553 | return ERR_PTR(err); | |
1554 | } | |
1555 | ||
cf725ce2 | 1556 | vaddr = vmap_block_vaddr(va->va_start, 0); |
db64fe02 NP |
1557 | spin_lock_init(&vb->lock); |
1558 | vb->va = va; | |
cf725ce2 RP |
1559 | /* At least something should be left free */ |
1560 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order)); | |
1561 | vb->free = VMAP_BBMAP_BITS - (1UL << order); | |
db64fe02 | 1562 | vb->dirty = 0; |
7d61bfe8 RP |
1563 | vb->dirty_min = VMAP_BBMAP_BITS; |
1564 | vb->dirty_max = 0; | |
db64fe02 | 1565 | INIT_LIST_HEAD(&vb->free_list); |
db64fe02 NP |
1566 | |
1567 | vb_idx = addr_to_vb_idx(va->va_start); | |
1568 | spin_lock(&vmap_block_tree_lock); | |
1569 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); | |
1570 | spin_unlock(&vmap_block_tree_lock); | |
1571 | BUG_ON(err); | |
1572 | radix_tree_preload_end(); | |
1573 | ||
1574 | vbq = &get_cpu_var(vmap_block_queue); | |
db64fe02 | 1575 | spin_lock(&vbq->lock); |
68ac546f | 1576 | list_add_tail_rcu(&vb->free_list, &vbq->free); |
db64fe02 | 1577 | spin_unlock(&vbq->lock); |
3f04ba85 | 1578 | put_cpu_var(vmap_block_queue); |
db64fe02 | 1579 | |
cf725ce2 | 1580 | return vaddr; |
db64fe02 NP |
1581 | } |
1582 | ||
db64fe02 NP |
1583 | static void free_vmap_block(struct vmap_block *vb) |
1584 | { | |
1585 | struct vmap_block *tmp; | |
1586 | unsigned long vb_idx; | |
1587 | ||
db64fe02 NP |
1588 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
1589 | spin_lock(&vmap_block_tree_lock); | |
1590 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); | |
1591 | spin_unlock(&vmap_block_tree_lock); | |
1592 | BUG_ON(tmp != vb); | |
1593 | ||
64141da5 | 1594 | free_vmap_area_noflush(vb->va); |
22a3c7d1 | 1595 | kfree_rcu(vb, rcu_head); |
db64fe02 NP |
1596 | } |
1597 | ||
02b709df NP |
1598 | static void purge_fragmented_blocks(int cpu) |
1599 | { | |
1600 | LIST_HEAD(purge); | |
1601 | struct vmap_block *vb; | |
1602 | struct vmap_block *n_vb; | |
1603 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1604 | ||
1605 | rcu_read_lock(); | |
1606 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
1607 | ||
1608 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) | |
1609 | continue; | |
1610 | ||
1611 | spin_lock(&vb->lock); | |
1612 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { | |
1613 | vb->free = 0; /* prevent further allocs after releasing lock */ | |
1614 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ | |
7d61bfe8 RP |
1615 | vb->dirty_min = 0; |
1616 | vb->dirty_max = VMAP_BBMAP_BITS; | |
02b709df NP |
1617 | spin_lock(&vbq->lock); |
1618 | list_del_rcu(&vb->free_list); | |
1619 | spin_unlock(&vbq->lock); | |
1620 | spin_unlock(&vb->lock); | |
1621 | list_add_tail(&vb->purge, &purge); | |
1622 | } else | |
1623 | spin_unlock(&vb->lock); | |
1624 | } | |
1625 | rcu_read_unlock(); | |
1626 | ||
1627 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { | |
1628 | list_del(&vb->purge); | |
1629 | free_vmap_block(vb); | |
1630 | } | |
1631 | } | |
1632 | ||
02b709df NP |
1633 | static void purge_fragmented_blocks_allcpus(void) |
1634 | { | |
1635 | int cpu; | |
1636 | ||
1637 | for_each_possible_cpu(cpu) | |
1638 | purge_fragmented_blocks(cpu); | |
1639 | } | |
1640 | ||
db64fe02 NP |
1641 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
1642 | { | |
1643 | struct vmap_block_queue *vbq; | |
1644 | struct vmap_block *vb; | |
cf725ce2 | 1645 | void *vaddr = NULL; |
db64fe02 NP |
1646 | unsigned int order; |
1647 | ||
891c49ab | 1648 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1649 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
aa91c4d8 JK |
1650 | if (WARN_ON(size == 0)) { |
1651 | /* | |
1652 | * Allocating 0 bytes isn't what caller wants since | |
1653 | * get_order(0) returns funny result. Just warn and terminate | |
1654 | * early. | |
1655 | */ | |
1656 | return NULL; | |
1657 | } | |
db64fe02 NP |
1658 | order = get_order(size); |
1659 | ||
db64fe02 NP |
1660 | rcu_read_lock(); |
1661 | vbq = &get_cpu_var(vmap_block_queue); | |
1662 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
cf725ce2 | 1663 | unsigned long pages_off; |
db64fe02 NP |
1664 | |
1665 | spin_lock(&vb->lock); | |
cf725ce2 RP |
1666 | if (vb->free < (1UL << order)) { |
1667 | spin_unlock(&vb->lock); | |
1668 | continue; | |
1669 | } | |
02b709df | 1670 | |
cf725ce2 RP |
1671 | pages_off = VMAP_BBMAP_BITS - vb->free; |
1672 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off); | |
02b709df NP |
1673 | vb->free -= 1UL << order; |
1674 | if (vb->free == 0) { | |
1675 | spin_lock(&vbq->lock); | |
1676 | list_del_rcu(&vb->free_list); | |
1677 | spin_unlock(&vbq->lock); | |
1678 | } | |
cf725ce2 | 1679 | |
02b709df NP |
1680 | spin_unlock(&vb->lock); |
1681 | break; | |
db64fe02 | 1682 | } |
02b709df | 1683 | |
3f04ba85 | 1684 | put_cpu_var(vmap_block_queue); |
db64fe02 NP |
1685 | rcu_read_unlock(); |
1686 | ||
cf725ce2 RP |
1687 | /* Allocate new block if nothing was found */ |
1688 | if (!vaddr) | |
1689 | vaddr = new_vmap_block(order, gfp_mask); | |
db64fe02 | 1690 | |
cf725ce2 | 1691 | return vaddr; |
db64fe02 NP |
1692 | } |
1693 | ||
78a0e8c4 | 1694 | static void vb_free(unsigned long addr, unsigned long size) |
db64fe02 NP |
1695 | { |
1696 | unsigned long offset; | |
1697 | unsigned long vb_idx; | |
1698 | unsigned int order; | |
1699 | struct vmap_block *vb; | |
1700 | ||
891c49ab | 1701 | BUG_ON(offset_in_page(size)); |
db64fe02 | 1702 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
b29acbdc | 1703 | |
78a0e8c4 | 1704 | flush_cache_vunmap(addr, addr + size); |
b29acbdc | 1705 | |
db64fe02 NP |
1706 | order = get_order(size); |
1707 | ||
78a0e8c4 | 1708 | offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT; |
db64fe02 | 1709 | |
78a0e8c4 | 1710 | vb_idx = addr_to_vb_idx(addr); |
db64fe02 NP |
1711 | rcu_read_lock(); |
1712 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); | |
1713 | rcu_read_unlock(); | |
1714 | BUG_ON(!vb); | |
1715 | ||
b521c43f | 1716 | unmap_kernel_range_noflush(addr, size); |
64141da5 | 1717 | |
8e57f8ac | 1718 | if (debug_pagealloc_enabled_static()) |
78a0e8c4 | 1719 | flush_tlb_kernel_range(addr, addr + size); |
82a2e924 | 1720 | |
db64fe02 | 1721 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1722 | |
1723 | /* Expand dirty range */ | |
1724 | vb->dirty_min = min(vb->dirty_min, offset); | |
1725 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order)); | |
d086817d | 1726 | |
db64fe02 NP |
1727 | vb->dirty += 1UL << order; |
1728 | if (vb->dirty == VMAP_BBMAP_BITS) { | |
de560423 | 1729 | BUG_ON(vb->free); |
db64fe02 NP |
1730 | spin_unlock(&vb->lock); |
1731 | free_vmap_block(vb); | |
1732 | } else | |
1733 | spin_unlock(&vb->lock); | |
1734 | } | |
1735 | ||
868b104d | 1736 | static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush) |
db64fe02 | 1737 | { |
db64fe02 | 1738 | int cpu; |
db64fe02 | 1739 | |
9b463334 JF |
1740 | if (unlikely(!vmap_initialized)) |
1741 | return; | |
1742 | ||
5803ed29 CH |
1743 | might_sleep(); |
1744 | ||
db64fe02 NP |
1745 | for_each_possible_cpu(cpu) { |
1746 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); | |
1747 | struct vmap_block *vb; | |
1748 | ||
1749 | rcu_read_lock(); | |
1750 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { | |
db64fe02 | 1751 | spin_lock(&vb->lock); |
7d61bfe8 RP |
1752 | if (vb->dirty) { |
1753 | unsigned long va_start = vb->va->va_start; | |
db64fe02 | 1754 | unsigned long s, e; |
b136be5e | 1755 | |
7d61bfe8 RP |
1756 | s = va_start + (vb->dirty_min << PAGE_SHIFT); |
1757 | e = va_start + (vb->dirty_max << PAGE_SHIFT); | |
db64fe02 | 1758 | |
7d61bfe8 RP |
1759 | start = min(s, start); |
1760 | end = max(e, end); | |
db64fe02 | 1761 | |
7d61bfe8 | 1762 | flush = 1; |
db64fe02 NP |
1763 | } |
1764 | spin_unlock(&vb->lock); | |
1765 | } | |
1766 | rcu_read_unlock(); | |
1767 | } | |
1768 | ||
f9e09977 | 1769 | mutex_lock(&vmap_purge_lock); |
0574ecd1 CH |
1770 | purge_fragmented_blocks_allcpus(); |
1771 | if (!__purge_vmap_area_lazy(start, end) && flush) | |
1772 | flush_tlb_kernel_range(start, end); | |
f9e09977 | 1773 | mutex_unlock(&vmap_purge_lock); |
db64fe02 | 1774 | } |
868b104d RE |
1775 | |
1776 | /** | |
1777 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer | |
1778 | * | |
1779 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily | |
1780 | * to amortize TLB flushing overheads. What this means is that any page you | |
1781 | * have now, may, in a former life, have been mapped into kernel virtual | |
1782 | * address by the vmap layer and so there might be some CPUs with TLB entries | |
1783 | * still referencing that page (additional to the regular 1:1 kernel mapping). | |
1784 | * | |
1785 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can | |
1786 | * be sure that none of the pages we have control over will have any aliases | |
1787 | * from the vmap layer. | |
1788 | */ | |
1789 | void vm_unmap_aliases(void) | |
1790 | { | |
1791 | unsigned long start = ULONG_MAX, end = 0; | |
1792 | int flush = 0; | |
1793 | ||
1794 | _vm_unmap_aliases(start, end, flush); | |
1795 | } | |
db64fe02 NP |
1796 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
1797 | ||
1798 | /** | |
1799 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram | |
1800 | * @mem: the pointer returned by vm_map_ram | |
1801 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) | |
1802 | */ | |
1803 | void vm_unmap_ram(const void *mem, unsigned int count) | |
1804 | { | |
65ee03c4 | 1805 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 | 1806 | unsigned long addr = (unsigned long)mem; |
9c3acf60 | 1807 | struct vmap_area *va; |
db64fe02 | 1808 | |
5803ed29 | 1809 | might_sleep(); |
db64fe02 NP |
1810 | BUG_ON(!addr); |
1811 | BUG_ON(addr < VMALLOC_START); | |
1812 | BUG_ON(addr > VMALLOC_END); | |
a1c0b1a0 | 1813 | BUG_ON(!PAGE_ALIGNED(addr)); |
db64fe02 | 1814 | |
d98c9e83 AR |
1815 | kasan_poison_vmalloc(mem, size); |
1816 | ||
9c3acf60 | 1817 | if (likely(count <= VMAP_MAX_ALLOC)) { |
05e3ff95 | 1818 | debug_check_no_locks_freed(mem, size); |
78a0e8c4 | 1819 | vb_free(addr, size); |
9c3acf60 CH |
1820 | return; |
1821 | } | |
1822 | ||
1823 | va = find_vmap_area(addr); | |
1824 | BUG_ON(!va); | |
05e3ff95 CP |
1825 | debug_check_no_locks_freed((void *)va->va_start, |
1826 | (va->va_end - va->va_start)); | |
9c3acf60 | 1827 | free_unmap_vmap_area(va); |
db64fe02 NP |
1828 | } |
1829 | EXPORT_SYMBOL(vm_unmap_ram); | |
1830 | ||
1831 | /** | |
1832 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) | |
1833 | * @pages: an array of pointers to the pages to be mapped | |
1834 | * @count: number of pages | |
1835 | * @node: prefer to allocate data structures on this node | |
1836 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM | |
e99c97ad | 1837 | * |
36437638 GK |
1838 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be |
1839 | * faster than vmap so it's good. But if you mix long-life and short-life | |
1840 | * objects with vm_map_ram(), it could consume lots of address space through | |
1841 | * fragmentation (especially on a 32bit machine). You could see failures in | |
1842 | * the end. Please use this function for short-lived objects. | |
1843 | * | |
e99c97ad | 1844 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
db64fe02 NP |
1845 | */ |
1846 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) | |
1847 | { | |
65ee03c4 | 1848 | unsigned long size = (unsigned long)count << PAGE_SHIFT; |
db64fe02 NP |
1849 | unsigned long addr; |
1850 | void *mem; | |
1851 | ||
1852 | if (likely(count <= VMAP_MAX_ALLOC)) { | |
1853 | mem = vb_alloc(size, GFP_KERNEL); | |
1854 | if (IS_ERR(mem)) | |
1855 | return NULL; | |
1856 | addr = (unsigned long)mem; | |
1857 | } else { | |
1858 | struct vmap_area *va; | |
1859 | va = alloc_vmap_area(size, PAGE_SIZE, | |
1860 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); | |
1861 | if (IS_ERR(va)) | |
1862 | return NULL; | |
1863 | ||
1864 | addr = va->va_start; | |
1865 | mem = (void *)addr; | |
1866 | } | |
d98c9e83 AR |
1867 | |
1868 | kasan_unpoison_vmalloc(mem, size); | |
1869 | ||
db64fe02 NP |
1870 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { |
1871 | vm_unmap_ram(mem, count); | |
1872 | return NULL; | |
1873 | } | |
1874 | return mem; | |
1875 | } | |
1876 | EXPORT_SYMBOL(vm_map_ram); | |
1877 | ||
4341fa45 | 1878 | static struct vm_struct *vmlist __initdata; |
92eac168 | 1879 | |
be9b7335 NP |
1880 | /** |
1881 | * vm_area_add_early - add vmap area early during boot | |
1882 | * @vm: vm_struct to add | |
1883 | * | |
1884 | * This function is used to add fixed kernel vm area to vmlist before | |
1885 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags | |
1886 | * should contain proper values and the other fields should be zero. | |
1887 | * | |
1888 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1889 | */ | |
1890 | void __init vm_area_add_early(struct vm_struct *vm) | |
1891 | { | |
1892 | struct vm_struct *tmp, **p; | |
1893 | ||
1894 | BUG_ON(vmap_initialized); | |
1895 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { | |
1896 | if (tmp->addr >= vm->addr) { | |
1897 | BUG_ON(tmp->addr < vm->addr + vm->size); | |
1898 | break; | |
1899 | } else | |
1900 | BUG_ON(tmp->addr + tmp->size > vm->addr); | |
1901 | } | |
1902 | vm->next = *p; | |
1903 | *p = vm; | |
1904 | } | |
1905 | ||
f0aa6617 TH |
1906 | /** |
1907 | * vm_area_register_early - register vmap area early during boot | |
1908 | * @vm: vm_struct to register | |
c0c0a293 | 1909 | * @align: requested alignment |
f0aa6617 TH |
1910 | * |
1911 | * This function is used to register kernel vm area before | |
1912 | * vmalloc_init() is called. @vm->size and @vm->flags should contain | |
1913 | * proper values on entry and other fields should be zero. On return, | |
1914 | * vm->addr contains the allocated address. | |
1915 | * | |
1916 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. | |
1917 | */ | |
c0c0a293 | 1918 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
f0aa6617 TH |
1919 | { |
1920 | static size_t vm_init_off __initdata; | |
c0c0a293 TH |
1921 | unsigned long addr; |
1922 | ||
1923 | addr = ALIGN(VMALLOC_START + vm_init_off, align); | |
1924 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; | |
f0aa6617 | 1925 | |
c0c0a293 | 1926 | vm->addr = (void *)addr; |
f0aa6617 | 1927 | |
be9b7335 | 1928 | vm_area_add_early(vm); |
f0aa6617 TH |
1929 | } |
1930 | ||
68ad4a33 URS |
1931 | static void vmap_init_free_space(void) |
1932 | { | |
1933 | unsigned long vmap_start = 1; | |
1934 | const unsigned long vmap_end = ULONG_MAX; | |
1935 | struct vmap_area *busy, *free; | |
1936 | ||
1937 | /* | |
1938 | * B F B B B F | |
1939 | * -|-----|.....|-----|-----|-----|.....|- | |
1940 | * | The KVA space | | |
1941 | * |<--------------------------------->| | |
1942 | */ | |
1943 | list_for_each_entry(busy, &vmap_area_list, list) { | |
1944 | if (busy->va_start - vmap_start > 0) { | |
1945 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1946 | if (!WARN_ON_ONCE(!free)) { | |
1947 | free->va_start = vmap_start; | |
1948 | free->va_end = busy->va_start; | |
1949 | ||
1950 | insert_vmap_area_augment(free, NULL, | |
1951 | &free_vmap_area_root, | |
1952 | &free_vmap_area_list); | |
1953 | } | |
1954 | } | |
1955 | ||
1956 | vmap_start = busy->va_end; | |
1957 | } | |
1958 | ||
1959 | if (vmap_end - vmap_start > 0) { | |
1960 | free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); | |
1961 | if (!WARN_ON_ONCE(!free)) { | |
1962 | free->va_start = vmap_start; | |
1963 | free->va_end = vmap_end; | |
1964 | ||
1965 | insert_vmap_area_augment(free, NULL, | |
1966 | &free_vmap_area_root, | |
1967 | &free_vmap_area_list); | |
1968 | } | |
1969 | } | |
1970 | } | |
1971 | ||
db64fe02 NP |
1972 | void __init vmalloc_init(void) |
1973 | { | |
822c18f2 IK |
1974 | struct vmap_area *va; |
1975 | struct vm_struct *tmp; | |
db64fe02 NP |
1976 | int i; |
1977 | ||
68ad4a33 URS |
1978 | /* |
1979 | * Create the cache for vmap_area objects. | |
1980 | */ | |
1981 | vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); | |
1982 | ||
db64fe02 NP |
1983 | for_each_possible_cpu(i) { |
1984 | struct vmap_block_queue *vbq; | |
32fcfd40 | 1985 | struct vfree_deferred *p; |
db64fe02 NP |
1986 | |
1987 | vbq = &per_cpu(vmap_block_queue, i); | |
1988 | spin_lock_init(&vbq->lock); | |
1989 | INIT_LIST_HEAD(&vbq->free); | |
32fcfd40 AV |
1990 | p = &per_cpu(vfree_deferred, i); |
1991 | init_llist_head(&p->list); | |
1992 | INIT_WORK(&p->wq, free_work); | |
db64fe02 | 1993 | } |
9b463334 | 1994 | |
822c18f2 IK |
1995 | /* Import existing vmlist entries. */ |
1996 | for (tmp = vmlist; tmp; tmp = tmp->next) { | |
68ad4a33 URS |
1997 | va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); |
1998 | if (WARN_ON_ONCE(!va)) | |
1999 | continue; | |
2000 | ||
822c18f2 IK |
2001 | va->va_start = (unsigned long)tmp->addr; |
2002 | va->va_end = va->va_start + tmp->size; | |
dbda591d | 2003 | va->vm = tmp; |
68ad4a33 | 2004 | insert_vmap_area(va, &vmap_area_root, &vmap_area_list); |
822c18f2 | 2005 | } |
ca23e405 | 2006 | |
68ad4a33 URS |
2007 | /* |
2008 | * Now we can initialize a free vmap space. | |
2009 | */ | |
2010 | vmap_init_free_space(); | |
9b463334 | 2011 | vmap_initialized = true; |
db64fe02 NP |
2012 | } |
2013 | ||
8fc48985 TH |
2014 | /** |
2015 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB | |
2016 | * @addr: start of the VM area to unmap | |
2017 | * @size: size of the VM area to unmap | |
2018 | * | |
2019 | * Similar to unmap_kernel_range_noflush() but flushes vcache before | |
2020 | * the unmapping and tlb after. | |
2021 | */ | |
db64fe02 NP |
2022 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
2023 | { | |
2024 | unsigned long end = addr + size; | |
f6fcba70 TH |
2025 | |
2026 | flush_cache_vunmap(addr, end); | |
b521c43f | 2027 | unmap_kernel_range_noflush(addr, size); |
db64fe02 NP |
2028 | flush_tlb_kernel_range(addr, end); |
2029 | } | |
2030 | ||
f6f8ed47 | 2031 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
db64fe02 NP |
2032 | { |
2033 | unsigned long addr = (unsigned long)area->addr; | |
762216ab | 2034 | unsigned long end = addr + get_vm_area_size(area); |
db64fe02 NP |
2035 | int err; |
2036 | ||
f6f8ed47 | 2037 | err = vmap_page_range(addr, end, prot, pages); |
db64fe02 | 2038 | |
f6f8ed47 | 2039 | return err > 0 ? 0 : err; |
db64fe02 | 2040 | } |
db64fe02 | 2041 | |
e36176be URS |
2042 | static inline void setup_vmalloc_vm_locked(struct vm_struct *vm, |
2043 | struct vmap_area *va, unsigned long flags, const void *caller) | |
cf88c790 | 2044 | { |
cf88c790 TH |
2045 | vm->flags = flags; |
2046 | vm->addr = (void *)va->va_start; | |
2047 | vm->size = va->va_end - va->va_start; | |
2048 | vm->caller = caller; | |
db1aecaf | 2049 | va->vm = vm; |
e36176be URS |
2050 | } |
2051 | ||
2052 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, | |
2053 | unsigned long flags, const void *caller) | |
2054 | { | |
2055 | spin_lock(&vmap_area_lock); | |
2056 | setup_vmalloc_vm_locked(vm, va, flags, caller); | |
c69480ad | 2057 | spin_unlock(&vmap_area_lock); |
f5252e00 | 2058 | } |
cf88c790 | 2059 | |
20fc02b4 | 2060 | static void clear_vm_uninitialized_flag(struct vm_struct *vm) |
f5252e00 | 2061 | { |
d4033afd | 2062 | /* |
20fc02b4 | 2063 | * Before removing VM_UNINITIALIZED, |
d4033afd JK |
2064 | * we should make sure that vm has proper values. |
2065 | * Pair with smp_rmb() in show_numa_info(). | |
2066 | */ | |
2067 | smp_wmb(); | |
20fc02b4 | 2068 | vm->flags &= ~VM_UNINITIALIZED; |
cf88c790 TH |
2069 | } |
2070 | ||
db64fe02 | 2071 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
2dca6999 | 2072 | unsigned long align, unsigned long flags, unsigned long start, |
5e6cafc8 | 2073 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
db64fe02 | 2074 | { |
0006526d | 2075 | struct vmap_area *va; |
db64fe02 | 2076 | struct vm_struct *area; |
d98c9e83 | 2077 | unsigned long requested_size = size; |
1da177e4 | 2078 | |
52fd24ca | 2079 | BUG_ON(in_interrupt()); |
1da177e4 | 2080 | size = PAGE_ALIGN(size); |
31be8309 OH |
2081 | if (unlikely(!size)) |
2082 | return NULL; | |
1da177e4 | 2083 | |
252e5c6e | 2084 | if (flags & VM_IOREMAP) |
2085 | align = 1ul << clamp_t(int, get_count_order_long(size), | |
2086 | PAGE_SHIFT, IOREMAP_MAX_ORDER); | |
2087 | ||
cf88c790 | 2088 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
1da177e4 LT |
2089 | if (unlikely(!area)) |
2090 | return NULL; | |
2091 | ||
71394fe5 AR |
2092 | if (!(flags & VM_NO_GUARD)) |
2093 | size += PAGE_SIZE; | |
1da177e4 | 2094 | |
db64fe02 NP |
2095 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
2096 | if (IS_ERR(va)) { | |
2097 | kfree(area); | |
2098 | return NULL; | |
1da177e4 | 2099 | } |
1da177e4 | 2100 | |
d98c9e83 | 2101 | kasan_unpoison_vmalloc((void *)va->va_start, requested_size); |
f5252e00 | 2102 | |
d98c9e83 | 2103 | setup_vmalloc_vm(area, va, flags, caller); |
3c5c3cfb | 2104 | |
1da177e4 | 2105 | return area; |
1da177e4 LT |
2106 | } |
2107 | ||
c2968612 BH |
2108 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
2109 | unsigned long start, unsigned long end, | |
5e6cafc8 | 2110 | const void *caller) |
c2968612 | 2111 | { |
00ef2d2f DR |
2112 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
2113 | GFP_KERNEL, caller); | |
c2968612 BH |
2114 | } |
2115 | ||
1da177e4 | 2116 | /** |
92eac168 MR |
2117 | * get_vm_area - reserve a contiguous kernel virtual area |
2118 | * @size: size of the area | |
2119 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC | |
1da177e4 | 2120 | * |
92eac168 MR |
2121 | * Search an area of @size in the kernel virtual mapping area, |
2122 | * and reserved it for out purposes. Returns the area descriptor | |
2123 | * on success or %NULL on failure. | |
a862f68a MR |
2124 | * |
2125 | * Return: the area descriptor on success or %NULL on failure. | |
1da177e4 LT |
2126 | */ |
2127 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) | |
2128 | { | |
2dca6999 | 2129 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f DR |
2130 | NUMA_NO_NODE, GFP_KERNEL, |
2131 | __builtin_return_address(0)); | |
23016969 CL |
2132 | } |
2133 | ||
2134 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, | |
5e6cafc8 | 2135 | const void *caller) |
23016969 | 2136 | { |
2dca6999 | 2137 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
00ef2d2f | 2138 | NUMA_NO_NODE, GFP_KERNEL, caller); |
1da177e4 LT |
2139 | } |
2140 | ||
e9da6e99 | 2141 | /** |
92eac168 MR |
2142 | * find_vm_area - find a continuous kernel virtual area |
2143 | * @addr: base address | |
e9da6e99 | 2144 | * |
92eac168 MR |
2145 | * Search for the kernel VM area starting at @addr, and return it. |
2146 | * It is up to the caller to do all required locking to keep the returned | |
2147 | * pointer valid. | |
a862f68a MR |
2148 | * |
2149 | * Return: pointer to the found area or %NULL on faulure | |
e9da6e99 MS |
2150 | */ |
2151 | struct vm_struct *find_vm_area(const void *addr) | |
83342314 | 2152 | { |
db64fe02 | 2153 | struct vmap_area *va; |
83342314 | 2154 | |
db64fe02 | 2155 | va = find_vmap_area((unsigned long)addr); |
688fcbfc PL |
2156 | if (!va) |
2157 | return NULL; | |
1da177e4 | 2158 | |
688fcbfc | 2159 | return va->vm; |
1da177e4 LT |
2160 | } |
2161 | ||
7856dfeb | 2162 | /** |
92eac168 MR |
2163 | * remove_vm_area - find and remove a continuous kernel virtual area |
2164 | * @addr: base address | |
7856dfeb | 2165 | * |
92eac168 MR |
2166 | * Search for the kernel VM area starting at @addr, and remove it. |
2167 | * This function returns the found VM area, but using it is NOT safe | |
2168 | * on SMP machines, except for its size or flags. | |
a862f68a MR |
2169 | * |
2170 | * Return: pointer to the found area or %NULL on faulure | |
7856dfeb | 2171 | */ |
b3bdda02 | 2172 | struct vm_struct *remove_vm_area(const void *addr) |
7856dfeb | 2173 | { |
db64fe02 NP |
2174 | struct vmap_area *va; |
2175 | ||
5803ed29 CH |
2176 | might_sleep(); |
2177 | ||
dd3b8353 URS |
2178 | spin_lock(&vmap_area_lock); |
2179 | va = __find_vmap_area((unsigned long)addr); | |
688fcbfc | 2180 | if (va && va->vm) { |
db1aecaf | 2181 | struct vm_struct *vm = va->vm; |
f5252e00 | 2182 | |
c69480ad | 2183 | va->vm = NULL; |
c69480ad JK |
2184 | spin_unlock(&vmap_area_lock); |
2185 | ||
a5af5aa8 | 2186 | kasan_free_shadow(vm); |
dd32c279 | 2187 | free_unmap_vmap_area(va); |
dd32c279 | 2188 | |
db64fe02 NP |
2189 | return vm; |
2190 | } | |
dd3b8353 URS |
2191 | |
2192 | spin_unlock(&vmap_area_lock); | |
db64fe02 | 2193 | return NULL; |
7856dfeb AK |
2194 | } |
2195 | ||
868b104d RE |
2196 | static inline void set_area_direct_map(const struct vm_struct *area, |
2197 | int (*set_direct_map)(struct page *page)) | |
2198 | { | |
2199 | int i; | |
2200 | ||
2201 | for (i = 0; i < area->nr_pages; i++) | |
2202 | if (page_address(area->pages[i])) | |
2203 | set_direct_map(area->pages[i]); | |
2204 | } | |
2205 | ||
2206 | /* Handle removing and resetting vm mappings related to the vm_struct. */ | |
2207 | static void vm_remove_mappings(struct vm_struct *area, int deallocate_pages) | |
2208 | { | |
868b104d RE |
2209 | unsigned long start = ULONG_MAX, end = 0; |
2210 | int flush_reset = area->flags & VM_FLUSH_RESET_PERMS; | |
31e67340 | 2211 | int flush_dmap = 0; |
868b104d RE |
2212 | int i; |
2213 | ||
868b104d RE |
2214 | remove_vm_area(area->addr); |
2215 | ||
2216 | /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */ | |
2217 | if (!flush_reset) | |
2218 | return; | |
2219 | ||
2220 | /* | |
2221 | * If not deallocating pages, just do the flush of the VM area and | |
2222 | * return. | |
2223 | */ | |
2224 | if (!deallocate_pages) { | |
2225 | vm_unmap_aliases(); | |
2226 | return; | |
2227 | } | |
2228 | ||
2229 | /* | |
2230 | * If execution gets here, flush the vm mapping and reset the direct | |
2231 | * map. Find the start and end range of the direct mappings to make sure | |
2232 | * the vm_unmap_aliases() flush includes the direct map. | |
2233 | */ | |
2234 | for (i = 0; i < area->nr_pages; i++) { | |
8e41f872 RE |
2235 | unsigned long addr = (unsigned long)page_address(area->pages[i]); |
2236 | if (addr) { | |
868b104d | 2237 | start = min(addr, start); |
8e41f872 | 2238 | end = max(addr + PAGE_SIZE, end); |
31e67340 | 2239 | flush_dmap = 1; |
868b104d RE |
2240 | } |
2241 | } | |
2242 | ||
2243 | /* | |
2244 | * Set direct map to something invalid so that it won't be cached if | |
2245 | * there are any accesses after the TLB flush, then flush the TLB and | |
2246 | * reset the direct map permissions to the default. | |
2247 | */ | |
2248 | set_area_direct_map(area, set_direct_map_invalid_noflush); | |
31e67340 | 2249 | _vm_unmap_aliases(start, end, flush_dmap); |
868b104d RE |
2250 | set_area_direct_map(area, set_direct_map_default_noflush); |
2251 | } | |
2252 | ||
b3bdda02 | 2253 | static void __vunmap(const void *addr, int deallocate_pages) |
1da177e4 LT |
2254 | { |
2255 | struct vm_struct *area; | |
2256 | ||
2257 | if (!addr) | |
2258 | return; | |
2259 | ||
e69e9d4a | 2260 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n", |
ab15d9b4 | 2261 | addr)) |
1da177e4 | 2262 | return; |
1da177e4 | 2263 | |
6ade2032 | 2264 | area = find_vm_area(addr); |
1da177e4 | 2265 | if (unlikely(!area)) { |
4c8573e2 | 2266 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
1da177e4 | 2267 | addr); |
1da177e4 LT |
2268 | return; |
2269 | } | |
2270 | ||
05e3ff95 CP |
2271 | debug_check_no_locks_freed(area->addr, get_vm_area_size(area)); |
2272 | debug_check_no_obj_freed(area->addr, get_vm_area_size(area)); | |
9a11b49a | 2273 | |
d98c9e83 | 2274 | kasan_poison_vmalloc(area->addr, area->size); |
3c5c3cfb | 2275 | |
868b104d RE |
2276 | vm_remove_mappings(area, deallocate_pages); |
2277 | ||
1da177e4 LT |
2278 | if (deallocate_pages) { |
2279 | int i; | |
2280 | ||
2281 | for (i = 0; i < area->nr_pages; i++) { | |
bf53d6f8 CL |
2282 | struct page *page = area->pages[i]; |
2283 | ||
2284 | BUG_ON(!page); | |
4949148a | 2285 | __free_pages(page, 0); |
1da177e4 | 2286 | } |
97105f0a | 2287 | atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2288 | |
244d63ee | 2289 | kvfree(area->pages); |
1da177e4 LT |
2290 | } |
2291 | ||
2292 | kfree(area); | |
2293 | return; | |
2294 | } | |
bf22e37a AR |
2295 | |
2296 | static inline void __vfree_deferred(const void *addr) | |
2297 | { | |
2298 | /* | |
2299 | * Use raw_cpu_ptr() because this can be called from preemptible | |
2300 | * context. Preemption is absolutely fine here, because the llist_add() | |
2301 | * implementation is lockless, so it works even if we are adding to | |
2302 | * nother cpu's list. schedule_work() should be fine with this too. | |
2303 | */ | |
2304 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred); | |
2305 | ||
2306 | if (llist_add((struct llist_node *)addr, &p->list)) | |
2307 | schedule_work(&p->wq); | |
2308 | } | |
2309 | ||
2310 | /** | |
92eac168 MR |
2311 | * vfree_atomic - release memory allocated by vmalloc() |
2312 | * @addr: memory base address | |
bf22e37a | 2313 | * |
92eac168 MR |
2314 | * This one is just like vfree() but can be called in any atomic context |
2315 | * except NMIs. | |
bf22e37a AR |
2316 | */ |
2317 | void vfree_atomic(const void *addr) | |
2318 | { | |
2319 | BUG_ON(in_nmi()); | |
2320 | ||
2321 | kmemleak_free(addr); | |
2322 | ||
2323 | if (!addr) | |
2324 | return; | |
2325 | __vfree_deferred(addr); | |
2326 | } | |
2327 | ||
c67dc624 RP |
2328 | static void __vfree(const void *addr) |
2329 | { | |
2330 | if (unlikely(in_interrupt())) | |
2331 | __vfree_deferred(addr); | |
2332 | else | |
2333 | __vunmap(addr, 1); | |
2334 | } | |
2335 | ||
1da177e4 | 2336 | /** |
92eac168 MR |
2337 | * vfree - release memory allocated by vmalloc() |
2338 | * @addr: memory base address | |
1da177e4 | 2339 | * |
92eac168 MR |
2340 | * Free the virtually continuous memory area starting at @addr, as |
2341 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is | |
2342 | * NULL, no operation is performed. | |
1da177e4 | 2343 | * |
92eac168 MR |
2344 | * Must not be called in NMI context (strictly speaking, only if we don't |
2345 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling | |
2346 | * conventions for vfree() arch-depenedent would be a really bad idea) | |
c9fcee51 | 2347 | * |
92eac168 | 2348 | * May sleep if called *not* from interrupt context. |
3ca4ea3a | 2349 | * |
92eac168 | 2350 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node) |
1da177e4 | 2351 | */ |
b3bdda02 | 2352 | void vfree(const void *addr) |
1da177e4 | 2353 | { |
32fcfd40 | 2354 | BUG_ON(in_nmi()); |
89219d37 CM |
2355 | |
2356 | kmemleak_free(addr); | |
2357 | ||
a8dda165 AR |
2358 | might_sleep_if(!in_interrupt()); |
2359 | ||
32fcfd40 AV |
2360 | if (!addr) |
2361 | return; | |
c67dc624 RP |
2362 | |
2363 | __vfree(addr); | |
1da177e4 | 2364 | } |
1da177e4 LT |
2365 | EXPORT_SYMBOL(vfree); |
2366 | ||
2367 | /** | |
92eac168 MR |
2368 | * vunmap - release virtual mapping obtained by vmap() |
2369 | * @addr: memory base address | |
1da177e4 | 2370 | * |
92eac168 MR |
2371 | * Free the virtually contiguous memory area starting at @addr, |
2372 | * which was created from the page array passed to vmap(). | |
1da177e4 | 2373 | * |
92eac168 | 2374 | * Must not be called in interrupt context. |
1da177e4 | 2375 | */ |
b3bdda02 | 2376 | void vunmap(const void *addr) |
1da177e4 LT |
2377 | { |
2378 | BUG_ON(in_interrupt()); | |
34754b69 | 2379 | might_sleep(); |
32fcfd40 AV |
2380 | if (addr) |
2381 | __vunmap(addr, 0); | |
1da177e4 | 2382 | } |
1da177e4 LT |
2383 | EXPORT_SYMBOL(vunmap); |
2384 | ||
2385 | /** | |
92eac168 MR |
2386 | * vmap - map an array of pages into virtually contiguous space |
2387 | * @pages: array of page pointers | |
2388 | * @count: number of pages to map | |
2389 | * @flags: vm_area->flags | |
2390 | * @prot: page protection for the mapping | |
2391 | * | |
2392 | * Maps @count pages from @pages into contiguous kernel virtual | |
2393 | * space. | |
a862f68a MR |
2394 | * |
2395 | * Return: the address of the area or %NULL on failure | |
1da177e4 LT |
2396 | */ |
2397 | void *vmap(struct page **pages, unsigned int count, | |
92eac168 | 2398 | unsigned long flags, pgprot_t prot) |
1da177e4 LT |
2399 | { |
2400 | struct vm_struct *area; | |
65ee03c4 | 2401 | unsigned long size; /* In bytes */ |
1da177e4 | 2402 | |
34754b69 PZ |
2403 | might_sleep(); |
2404 | ||
ca79b0c2 | 2405 | if (count > totalram_pages()) |
1da177e4 LT |
2406 | return NULL; |
2407 | ||
65ee03c4 GJM |
2408 | size = (unsigned long)count << PAGE_SHIFT; |
2409 | area = get_vm_area_caller(size, flags, __builtin_return_address(0)); | |
1da177e4 LT |
2410 | if (!area) |
2411 | return NULL; | |
23016969 | 2412 | |
f6f8ed47 | 2413 | if (map_vm_area(area, prot, pages)) { |
1da177e4 LT |
2414 | vunmap(area->addr); |
2415 | return NULL; | |
2416 | } | |
2417 | ||
2418 | return area->addr; | |
2419 | } | |
1da177e4 LT |
2420 | EXPORT_SYMBOL(vmap); |
2421 | ||
8594a21c MH |
2422 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
2423 | gfp_t gfp_mask, pgprot_t prot, | |
2424 | int node, const void *caller); | |
e31d9eb5 | 2425 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
3722e13c | 2426 | pgprot_t prot, int node) |
1da177e4 LT |
2427 | { |
2428 | struct page **pages; | |
2429 | unsigned int nr_pages, array_size, i; | |
930f036b | 2430 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
704b862f LA |
2431 | const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; |
2432 | const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ? | |
2433 | 0 : | |
2434 | __GFP_HIGHMEM; | |
1da177e4 | 2435 | |
762216ab | 2436 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; |
1da177e4 LT |
2437 | array_size = (nr_pages * sizeof(struct page *)); |
2438 | ||
1da177e4 | 2439 | /* Please note that the recursion is strictly bounded. */ |
8757d5fa | 2440 | if (array_size > PAGE_SIZE) { |
704b862f | 2441 | pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask, |
3722e13c | 2442 | PAGE_KERNEL, node, area->caller); |
286e1ea3 | 2443 | } else { |
976d6dfb | 2444 | pages = kmalloc_node(array_size, nested_gfp, node); |
286e1ea3 | 2445 | } |
7ea36242 AK |
2446 | |
2447 | if (!pages) { | |
1da177e4 LT |
2448 | remove_vm_area(area->addr); |
2449 | kfree(area); | |
2450 | return NULL; | |
2451 | } | |
1da177e4 | 2452 | |
7ea36242 AK |
2453 | area->pages = pages; |
2454 | area->nr_pages = nr_pages; | |
2455 | ||
1da177e4 | 2456 | for (i = 0; i < area->nr_pages; i++) { |
bf53d6f8 CL |
2457 | struct page *page; |
2458 | ||
4b90951c | 2459 | if (node == NUMA_NO_NODE) |
704b862f | 2460 | page = alloc_page(alloc_mask|highmem_mask); |
930fc45a | 2461 | else |
704b862f | 2462 | page = alloc_pages_node(node, alloc_mask|highmem_mask, 0); |
bf53d6f8 CL |
2463 | |
2464 | if (unlikely(!page)) { | |
1da177e4 LT |
2465 | /* Successfully allocated i pages, free them in __vunmap() */ |
2466 | area->nr_pages = i; | |
97105f0a | 2467 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 LT |
2468 | goto fail; |
2469 | } | |
bf53d6f8 | 2470 | area->pages[i] = page; |
dcf61ff0 | 2471 | if (gfpflags_allow_blocking(gfp_mask)) |
660654f9 | 2472 | cond_resched(); |
1da177e4 | 2473 | } |
97105f0a | 2474 | atomic_long_add(area->nr_pages, &nr_vmalloc_pages); |
1da177e4 | 2475 | |
f6f8ed47 | 2476 | if (map_vm_area(area, prot, pages)) |
1da177e4 LT |
2477 | goto fail; |
2478 | return area->addr; | |
2479 | ||
2480 | fail: | |
a8e99259 | 2481 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2482 | "vmalloc: allocation failure, allocated %ld of %ld bytes", |
22943ab1 | 2483 | (area->nr_pages*PAGE_SIZE), area->size); |
c67dc624 | 2484 | __vfree(area->addr); |
1da177e4 LT |
2485 | return NULL; |
2486 | } | |
2487 | ||
2488 | /** | |
92eac168 MR |
2489 | * __vmalloc_node_range - allocate virtually contiguous memory |
2490 | * @size: allocation size | |
2491 | * @align: desired alignment | |
2492 | * @start: vm area range start | |
2493 | * @end: vm area range end | |
2494 | * @gfp_mask: flags for the page level allocator | |
2495 | * @prot: protection mask for the allocated pages | |
2496 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD) | |
2497 | * @node: node to use for allocation or NUMA_NO_NODE | |
2498 | * @caller: caller's return address | |
2499 | * | |
2500 | * Allocate enough pages to cover @size from the page level | |
2501 | * allocator with @gfp_mask flags. Map them into contiguous | |
2502 | * kernel virtual space, using a pagetable protection of @prot. | |
a862f68a MR |
2503 | * |
2504 | * Return: the address of the area or %NULL on failure | |
1da177e4 | 2505 | */ |
d0a21265 DR |
2506 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
2507 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
cb9e3c29 AR |
2508 | pgprot_t prot, unsigned long vm_flags, int node, |
2509 | const void *caller) | |
1da177e4 LT |
2510 | { |
2511 | struct vm_struct *area; | |
89219d37 CM |
2512 | void *addr; |
2513 | unsigned long real_size = size; | |
1da177e4 LT |
2514 | |
2515 | size = PAGE_ALIGN(size); | |
ca79b0c2 | 2516 | if (!size || (size >> PAGE_SHIFT) > totalram_pages()) |
de7d2b56 | 2517 | goto fail; |
1da177e4 | 2518 | |
d98c9e83 | 2519 | area = __get_vm_area_node(real_size, align, VM_ALLOC | VM_UNINITIALIZED | |
cb9e3c29 | 2520 | vm_flags, start, end, node, gfp_mask, caller); |
1da177e4 | 2521 | if (!area) |
de7d2b56 | 2522 | goto fail; |
1da177e4 | 2523 | |
3722e13c | 2524 | addr = __vmalloc_area_node(area, gfp_mask, prot, node); |
1368edf0 | 2525 | if (!addr) |
b82225f3 | 2526 | return NULL; |
89219d37 | 2527 | |
f5252e00 | 2528 | /* |
20fc02b4 ZY |
2529 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED |
2530 | * flag. It means that vm_struct is not fully initialized. | |
4341fa45 | 2531 | * Now, it is fully initialized, so remove this flag here. |
f5252e00 | 2532 | */ |
20fc02b4 | 2533 | clear_vm_uninitialized_flag(area); |
f5252e00 | 2534 | |
94f4a161 | 2535 | kmemleak_vmalloc(area, size, gfp_mask); |
89219d37 CM |
2536 | |
2537 | return addr; | |
de7d2b56 JP |
2538 | |
2539 | fail: | |
a8e99259 | 2540 | warn_alloc(gfp_mask, NULL, |
7877cdcc | 2541 | "vmalloc: allocation failure: %lu bytes", real_size); |
de7d2b56 | 2542 | return NULL; |
1da177e4 LT |
2543 | } |
2544 | ||
153178ed URS |
2545 | /* |
2546 | * This is only for performance analysis of vmalloc and stress purpose. | |
2547 | * It is required by vmalloc test module, therefore do not use it other | |
2548 | * than that. | |
2549 | */ | |
2550 | #ifdef CONFIG_TEST_VMALLOC_MODULE | |
2551 | EXPORT_SYMBOL_GPL(__vmalloc_node_range); | |
2552 | #endif | |
2553 | ||
d0a21265 | 2554 | /** |
92eac168 MR |
2555 | * __vmalloc_node - allocate virtually contiguous memory |
2556 | * @size: allocation size | |
2557 | * @align: desired alignment | |
2558 | * @gfp_mask: flags for the page level allocator | |
2559 | * @prot: protection mask for the allocated pages | |
2560 | * @node: node to use for allocation or NUMA_NO_NODE | |
2561 | * @caller: caller's return address | |
a7c3e901 | 2562 | * |
92eac168 MR |
2563 | * Allocate enough pages to cover @size from the page level |
2564 | * allocator with @gfp_mask flags. Map them into contiguous | |
2565 | * kernel virtual space, using a pagetable protection of @prot. | |
a7c3e901 | 2566 | * |
92eac168 MR |
2567 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL |
2568 | * and __GFP_NOFAIL are not supported | |
a7c3e901 | 2569 | * |
92eac168 MR |
2570 | * Any use of gfp flags outside of GFP_KERNEL should be consulted |
2571 | * with mm people. | |
a862f68a MR |
2572 | * |
2573 | * Return: pointer to the allocated memory or %NULL on error | |
d0a21265 | 2574 | */ |
8594a21c | 2575 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
d0a21265 | 2576 | gfp_t gfp_mask, pgprot_t prot, |
5e6cafc8 | 2577 | int node, const void *caller) |
d0a21265 DR |
2578 | { |
2579 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, | |
cb9e3c29 | 2580 | gfp_mask, prot, 0, node, caller); |
d0a21265 DR |
2581 | } |
2582 | ||
930fc45a CL |
2583 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
2584 | { | |
00ef2d2f | 2585 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
23016969 | 2586 | __builtin_return_address(0)); |
930fc45a | 2587 | } |
1da177e4 LT |
2588 | EXPORT_SYMBOL(__vmalloc); |
2589 | ||
8594a21c MH |
2590 | static inline void *__vmalloc_node_flags(unsigned long size, |
2591 | int node, gfp_t flags) | |
2592 | { | |
2593 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, | |
2594 | node, __builtin_return_address(0)); | |
2595 | } | |
2596 | ||
2597 | ||
2598 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags, | |
2599 | void *caller) | |
2600 | { | |
2601 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller); | |
2602 | } | |
2603 | ||
1da177e4 | 2604 | /** |
92eac168 MR |
2605 | * vmalloc - allocate virtually contiguous memory |
2606 | * @size: allocation size | |
2607 | * | |
2608 | * Allocate enough pages to cover @size from the page level | |
2609 | * allocator and map them into contiguous kernel virtual space. | |
1da177e4 | 2610 | * |
92eac168 MR |
2611 | * For tight control over page level allocator and protection flags |
2612 | * use __vmalloc() instead. | |
a862f68a MR |
2613 | * |
2614 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2615 | */ |
2616 | void *vmalloc(unsigned long size) | |
2617 | { | |
00ef2d2f | 2618 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2619 | GFP_KERNEL); |
1da177e4 | 2620 | } |
1da177e4 LT |
2621 | EXPORT_SYMBOL(vmalloc); |
2622 | ||
e1ca7788 | 2623 | /** |
92eac168 MR |
2624 | * vzalloc - allocate virtually contiguous memory with zero fill |
2625 | * @size: allocation size | |
2626 | * | |
2627 | * Allocate enough pages to cover @size from the page level | |
2628 | * allocator and map them into contiguous kernel virtual space. | |
2629 | * The memory allocated is set to zero. | |
2630 | * | |
2631 | * For tight control over page level allocator and protection flags | |
2632 | * use __vmalloc() instead. | |
a862f68a MR |
2633 | * |
2634 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2635 | */ |
2636 | void *vzalloc(unsigned long size) | |
2637 | { | |
00ef2d2f | 2638 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
19809c2d | 2639 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2640 | } |
2641 | EXPORT_SYMBOL(vzalloc); | |
2642 | ||
83342314 | 2643 | /** |
ead04089 REB |
2644 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
2645 | * @size: allocation size | |
83342314 | 2646 | * |
ead04089 REB |
2647 | * The resulting memory area is zeroed so it can be mapped to userspace |
2648 | * without leaking data. | |
a862f68a MR |
2649 | * |
2650 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2651 | */ |
2652 | void *vmalloc_user(unsigned long size) | |
2653 | { | |
bc84c535 RP |
2654 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2655 | GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL, | |
2656 | VM_USERMAP, NUMA_NO_NODE, | |
2657 | __builtin_return_address(0)); | |
83342314 NP |
2658 | } |
2659 | EXPORT_SYMBOL(vmalloc_user); | |
2660 | ||
930fc45a | 2661 | /** |
92eac168 MR |
2662 | * vmalloc_node - allocate memory on a specific node |
2663 | * @size: allocation size | |
2664 | * @node: numa node | |
930fc45a | 2665 | * |
92eac168 MR |
2666 | * Allocate enough pages to cover @size from the page level |
2667 | * allocator and map them into contiguous kernel virtual space. | |
930fc45a | 2668 | * |
92eac168 MR |
2669 | * For tight control over page level allocator and protection flags |
2670 | * use __vmalloc() instead. | |
a862f68a MR |
2671 | * |
2672 | * Return: pointer to the allocated memory or %NULL on error | |
930fc45a CL |
2673 | */ |
2674 | void *vmalloc_node(unsigned long size, int node) | |
2675 | { | |
19809c2d | 2676 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL, |
23016969 | 2677 | node, __builtin_return_address(0)); |
930fc45a CL |
2678 | } |
2679 | EXPORT_SYMBOL(vmalloc_node); | |
2680 | ||
e1ca7788 DY |
2681 | /** |
2682 | * vzalloc_node - allocate memory on a specific node with zero fill | |
2683 | * @size: allocation size | |
2684 | * @node: numa node | |
2685 | * | |
2686 | * Allocate enough pages to cover @size from the page level | |
2687 | * allocator and map them into contiguous kernel virtual space. | |
2688 | * The memory allocated is set to zero. | |
2689 | * | |
2690 | * For tight control over page level allocator and protection flags | |
2691 | * use __vmalloc_node() instead. | |
a862f68a MR |
2692 | * |
2693 | * Return: pointer to the allocated memory or %NULL on error | |
e1ca7788 DY |
2694 | */ |
2695 | void *vzalloc_node(unsigned long size, int node) | |
2696 | { | |
2697 | return __vmalloc_node_flags(size, node, | |
19809c2d | 2698 | GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
2699 | } |
2700 | EXPORT_SYMBOL(vzalloc_node); | |
2701 | ||
fc970227 AN |
2702 | /** |
2703 | * vmalloc_user_node_flags - allocate memory for userspace on a specific node | |
2704 | * @size: allocation size | |
2705 | * @node: numa node | |
2706 | * @flags: flags for the page level allocator | |
2707 | * | |
2708 | * The resulting memory area is zeroed so it can be mapped to userspace | |
2709 | * without leaking data. | |
2710 | * | |
2711 | * Return: pointer to the allocated memory or %NULL on error | |
2712 | */ | |
2713 | void *vmalloc_user_node_flags(unsigned long size, int node, gfp_t flags) | |
2714 | { | |
2715 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, | |
2716 | flags | __GFP_ZERO, PAGE_KERNEL, | |
2717 | VM_USERMAP, node, | |
2718 | __builtin_return_address(0)); | |
2719 | } | |
2720 | EXPORT_SYMBOL(vmalloc_user_node_flags); | |
2721 | ||
1da177e4 | 2722 | /** |
92eac168 MR |
2723 | * vmalloc_exec - allocate virtually contiguous, executable memory |
2724 | * @size: allocation size | |
1da177e4 | 2725 | * |
92eac168 MR |
2726 | * Kernel-internal function to allocate enough pages to cover @size |
2727 | * the page level allocator and map them into contiguous and | |
2728 | * executable kernel virtual space. | |
1da177e4 | 2729 | * |
92eac168 MR |
2730 | * For tight control over page level allocator and protection flags |
2731 | * use __vmalloc() instead. | |
a862f68a MR |
2732 | * |
2733 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 | 2734 | */ |
1da177e4 LT |
2735 | void *vmalloc_exec(unsigned long size) |
2736 | { | |
868b104d RE |
2737 | return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, |
2738 | GFP_KERNEL, PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS, | |
2739 | NUMA_NO_NODE, __builtin_return_address(0)); | |
1da177e4 LT |
2740 | } |
2741 | ||
0d08e0d3 | 2742 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
698d0831 | 2743 | #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL) |
0d08e0d3 | 2744 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
698d0831 | 2745 | #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL) |
0d08e0d3 | 2746 | #else |
698d0831 MH |
2747 | /* |
2748 | * 64b systems should always have either DMA or DMA32 zones. For others | |
2749 | * GFP_DMA32 should do the right thing and use the normal zone. | |
2750 | */ | |
2751 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL | |
0d08e0d3 AK |
2752 | #endif |
2753 | ||
1da177e4 | 2754 | /** |
92eac168 MR |
2755 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
2756 | * @size: allocation size | |
1da177e4 | 2757 | * |
92eac168 MR |
2758 | * Allocate enough 32bit PA addressable pages to cover @size from the |
2759 | * page level allocator and map them into contiguous kernel virtual space. | |
a862f68a MR |
2760 | * |
2761 | * Return: pointer to the allocated memory or %NULL on error | |
1da177e4 LT |
2762 | */ |
2763 | void *vmalloc_32(unsigned long size) | |
2764 | { | |
2dca6999 | 2765 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
00ef2d2f | 2766 | NUMA_NO_NODE, __builtin_return_address(0)); |
1da177e4 | 2767 | } |
1da177e4 LT |
2768 | EXPORT_SYMBOL(vmalloc_32); |
2769 | ||
83342314 | 2770 | /** |
ead04089 | 2771 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
92eac168 | 2772 | * @size: allocation size |
ead04089 REB |
2773 | * |
2774 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
2775 | * mapped to userspace without leaking data. | |
a862f68a MR |
2776 | * |
2777 | * Return: pointer to the allocated memory or %NULL on error | |
83342314 NP |
2778 | */ |
2779 | void *vmalloc_32_user(unsigned long size) | |
2780 | { | |
bc84c535 RP |
2781 | return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END, |
2782 | GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, | |
2783 | VM_USERMAP, NUMA_NO_NODE, | |
2784 | __builtin_return_address(0)); | |
83342314 NP |
2785 | } |
2786 | EXPORT_SYMBOL(vmalloc_32_user); | |
2787 | ||
d0107eb0 KH |
2788 | /* |
2789 | * small helper routine , copy contents to buf from addr. | |
2790 | * If the page is not present, fill zero. | |
2791 | */ | |
2792 | ||
2793 | static int aligned_vread(char *buf, char *addr, unsigned long count) | |
2794 | { | |
2795 | struct page *p; | |
2796 | int copied = 0; | |
2797 | ||
2798 | while (count) { | |
2799 | unsigned long offset, length; | |
2800 | ||
891c49ab | 2801 | offset = offset_in_page(addr); |
d0107eb0 KH |
2802 | length = PAGE_SIZE - offset; |
2803 | if (length > count) | |
2804 | length = count; | |
2805 | p = vmalloc_to_page(addr); | |
2806 | /* | |
2807 | * To do safe access to this _mapped_ area, we need | |
2808 | * lock. But adding lock here means that we need to add | |
2809 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2810 | * interface, rarely used. Instead of that, we'll use | |
2811 | * kmap() and get small overhead in this access function. | |
2812 | */ | |
2813 | if (p) { | |
2814 | /* | |
2815 | * we can expect USER0 is not used (see vread/vwrite's | |
2816 | * function description) | |
2817 | */ | |
9b04c5fe | 2818 | void *map = kmap_atomic(p); |
d0107eb0 | 2819 | memcpy(buf, map + offset, length); |
9b04c5fe | 2820 | kunmap_atomic(map); |
d0107eb0 KH |
2821 | } else |
2822 | memset(buf, 0, length); | |
2823 | ||
2824 | addr += length; | |
2825 | buf += length; | |
2826 | copied += length; | |
2827 | count -= length; | |
2828 | } | |
2829 | return copied; | |
2830 | } | |
2831 | ||
2832 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) | |
2833 | { | |
2834 | struct page *p; | |
2835 | int copied = 0; | |
2836 | ||
2837 | while (count) { | |
2838 | unsigned long offset, length; | |
2839 | ||
891c49ab | 2840 | offset = offset_in_page(addr); |
d0107eb0 KH |
2841 | length = PAGE_SIZE - offset; |
2842 | if (length > count) | |
2843 | length = count; | |
2844 | p = vmalloc_to_page(addr); | |
2845 | /* | |
2846 | * To do safe access to this _mapped_ area, we need | |
2847 | * lock. But adding lock here means that we need to add | |
2848 | * overhead of vmalloc()/vfree() calles for this _debug_ | |
2849 | * interface, rarely used. Instead of that, we'll use | |
2850 | * kmap() and get small overhead in this access function. | |
2851 | */ | |
2852 | if (p) { | |
2853 | /* | |
2854 | * we can expect USER0 is not used (see vread/vwrite's | |
2855 | * function description) | |
2856 | */ | |
9b04c5fe | 2857 | void *map = kmap_atomic(p); |
d0107eb0 | 2858 | memcpy(map + offset, buf, length); |
9b04c5fe | 2859 | kunmap_atomic(map); |
d0107eb0 KH |
2860 | } |
2861 | addr += length; | |
2862 | buf += length; | |
2863 | copied += length; | |
2864 | count -= length; | |
2865 | } | |
2866 | return copied; | |
2867 | } | |
2868 | ||
2869 | /** | |
92eac168 MR |
2870 | * vread() - read vmalloc area in a safe way. |
2871 | * @buf: buffer for reading data | |
2872 | * @addr: vm address. | |
2873 | * @count: number of bytes to be read. | |
2874 | * | |
92eac168 MR |
2875 | * This function checks that addr is a valid vmalloc'ed area, and |
2876 | * copy data from that area to a given buffer. If the given memory range | |
2877 | * of [addr...addr+count) includes some valid address, data is copied to | |
2878 | * proper area of @buf. If there are memory holes, they'll be zero-filled. | |
2879 | * IOREMAP area is treated as memory hole and no copy is done. | |
2880 | * | |
2881 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2882 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2883 | * | |
2884 | * Note: In usual ops, vread() is never necessary because the caller | |
2885 | * should know vmalloc() area is valid and can use memcpy(). | |
2886 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2887 | * any information, as /dev/kmem. |
a862f68a MR |
2888 | * |
2889 | * Return: number of bytes for which addr and buf should be increased | |
2890 | * (same number as @count) or %0 if [addr...addr+count) doesn't | |
2891 | * include any intersection with valid vmalloc area | |
d0107eb0 | 2892 | */ |
1da177e4 LT |
2893 | long vread(char *buf, char *addr, unsigned long count) |
2894 | { | |
e81ce85f JK |
2895 | struct vmap_area *va; |
2896 | struct vm_struct *vm; | |
1da177e4 | 2897 | char *vaddr, *buf_start = buf; |
d0107eb0 | 2898 | unsigned long buflen = count; |
1da177e4 LT |
2899 | unsigned long n; |
2900 | ||
2901 | /* Don't allow overflow */ | |
2902 | if ((unsigned long) addr + count < count) | |
2903 | count = -(unsigned long) addr; | |
2904 | ||
e81ce85f JK |
2905 | spin_lock(&vmap_area_lock); |
2906 | list_for_each_entry(va, &vmap_area_list, list) { | |
2907 | if (!count) | |
2908 | break; | |
2909 | ||
688fcbfc | 2910 | if (!va->vm) |
e81ce85f JK |
2911 | continue; |
2912 | ||
2913 | vm = va->vm; | |
2914 | vaddr = (char *) vm->addr; | |
762216ab | 2915 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2916 | continue; |
2917 | while (addr < vaddr) { | |
2918 | if (count == 0) | |
2919 | goto finished; | |
2920 | *buf = '\0'; | |
2921 | buf++; | |
2922 | addr++; | |
2923 | count--; | |
2924 | } | |
762216ab | 2925 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
2926 | if (n > count) |
2927 | n = count; | |
e81ce85f | 2928 | if (!(vm->flags & VM_IOREMAP)) |
d0107eb0 KH |
2929 | aligned_vread(buf, addr, n); |
2930 | else /* IOREMAP area is treated as memory hole */ | |
2931 | memset(buf, 0, n); | |
2932 | buf += n; | |
2933 | addr += n; | |
2934 | count -= n; | |
1da177e4 LT |
2935 | } |
2936 | finished: | |
e81ce85f | 2937 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
2938 | |
2939 | if (buf == buf_start) | |
2940 | return 0; | |
2941 | /* zero-fill memory holes */ | |
2942 | if (buf != buf_start + buflen) | |
2943 | memset(buf, 0, buflen - (buf - buf_start)); | |
2944 | ||
2945 | return buflen; | |
1da177e4 LT |
2946 | } |
2947 | ||
d0107eb0 | 2948 | /** |
92eac168 MR |
2949 | * vwrite() - write vmalloc area in a safe way. |
2950 | * @buf: buffer for source data | |
2951 | * @addr: vm address. | |
2952 | * @count: number of bytes to be read. | |
2953 | * | |
92eac168 MR |
2954 | * This function checks that addr is a valid vmalloc'ed area, and |
2955 | * copy data from a buffer to the given addr. If specified range of | |
2956 | * [addr...addr+count) includes some valid address, data is copied from | |
2957 | * proper area of @buf. If there are memory holes, no copy to hole. | |
2958 | * IOREMAP area is treated as memory hole and no copy is done. | |
2959 | * | |
2960 | * If [addr...addr+count) doesn't includes any intersects with alive | |
2961 | * vm_struct area, returns 0. @buf should be kernel's buffer. | |
2962 | * | |
2963 | * Note: In usual ops, vwrite() is never necessary because the caller | |
2964 | * should know vmalloc() area is valid and can use memcpy(). | |
2965 | * This is for routines which have to access vmalloc area without | |
d9009d67 | 2966 | * any information, as /dev/kmem. |
a862f68a MR |
2967 | * |
2968 | * Return: number of bytes for which addr and buf should be | |
2969 | * increased (same number as @count) or %0 if [addr...addr+count) | |
2970 | * doesn't include any intersection with valid vmalloc area | |
d0107eb0 | 2971 | */ |
1da177e4 LT |
2972 | long vwrite(char *buf, char *addr, unsigned long count) |
2973 | { | |
e81ce85f JK |
2974 | struct vmap_area *va; |
2975 | struct vm_struct *vm; | |
d0107eb0 KH |
2976 | char *vaddr; |
2977 | unsigned long n, buflen; | |
2978 | int copied = 0; | |
1da177e4 LT |
2979 | |
2980 | /* Don't allow overflow */ | |
2981 | if ((unsigned long) addr + count < count) | |
2982 | count = -(unsigned long) addr; | |
d0107eb0 | 2983 | buflen = count; |
1da177e4 | 2984 | |
e81ce85f JK |
2985 | spin_lock(&vmap_area_lock); |
2986 | list_for_each_entry(va, &vmap_area_list, list) { | |
2987 | if (!count) | |
2988 | break; | |
2989 | ||
688fcbfc | 2990 | if (!va->vm) |
e81ce85f JK |
2991 | continue; |
2992 | ||
2993 | vm = va->vm; | |
2994 | vaddr = (char *) vm->addr; | |
762216ab | 2995 | if (addr >= vaddr + get_vm_area_size(vm)) |
1da177e4 LT |
2996 | continue; |
2997 | while (addr < vaddr) { | |
2998 | if (count == 0) | |
2999 | goto finished; | |
3000 | buf++; | |
3001 | addr++; | |
3002 | count--; | |
3003 | } | |
762216ab | 3004 | n = vaddr + get_vm_area_size(vm) - addr; |
d0107eb0 KH |
3005 | if (n > count) |
3006 | n = count; | |
e81ce85f | 3007 | if (!(vm->flags & VM_IOREMAP)) { |
d0107eb0 KH |
3008 | aligned_vwrite(buf, addr, n); |
3009 | copied++; | |
3010 | } | |
3011 | buf += n; | |
3012 | addr += n; | |
3013 | count -= n; | |
1da177e4 LT |
3014 | } |
3015 | finished: | |
e81ce85f | 3016 | spin_unlock(&vmap_area_lock); |
d0107eb0 KH |
3017 | if (!copied) |
3018 | return 0; | |
3019 | return buflen; | |
1da177e4 | 3020 | } |
83342314 NP |
3021 | |
3022 | /** | |
92eac168 MR |
3023 | * remap_vmalloc_range_partial - map vmalloc pages to userspace |
3024 | * @vma: vma to cover | |
3025 | * @uaddr: target user address to start at | |
3026 | * @kaddr: virtual address of vmalloc kernel memory | |
bdebd6a2 | 3027 | * @pgoff: offset from @kaddr to start at |
92eac168 | 3028 | * @size: size of map area |
7682486b | 3029 | * |
92eac168 | 3030 | * Returns: 0 for success, -Exxx on failure |
83342314 | 3031 | * |
92eac168 MR |
3032 | * This function checks that @kaddr is a valid vmalloc'ed area, |
3033 | * and that it is big enough to cover the range starting at | |
3034 | * @uaddr in @vma. Will return failure if that criteria isn't | |
3035 | * met. | |
83342314 | 3036 | * |
92eac168 | 3037 | * Similar to remap_pfn_range() (see mm/memory.c) |
83342314 | 3038 | */ |
e69e9d4a | 3039 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr, |
bdebd6a2 JH |
3040 | void *kaddr, unsigned long pgoff, |
3041 | unsigned long size) | |
83342314 NP |
3042 | { |
3043 | struct vm_struct *area; | |
bdebd6a2 JH |
3044 | unsigned long off; |
3045 | unsigned long end_index; | |
3046 | ||
3047 | if (check_shl_overflow(pgoff, PAGE_SHIFT, &off)) | |
3048 | return -EINVAL; | |
83342314 | 3049 | |
e69e9d4a HD |
3050 | size = PAGE_ALIGN(size); |
3051 | ||
3052 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr)) | |
83342314 NP |
3053 | return -EINVAL; |
3054 | ||
e69e9d4a | 3055 | area = find_vm_area(kaddr); |
83342314 | 3056 | if (!area) |
db64fe02 | 3057 | return -EINVAL; |
83342314 | 3058 | |
fe9041c2 | 3059 | if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT))) |
db64fe02 | 3060 | return -EINVAL; |
83342314 | 3061 | |
bdebd6a2 JH |
3062 | if (check_add_overflow(size, off, &end_index) || |
3063 | end_index > get_vm_area_size(area)) | |
db64fe02 | 3064 | return -EINVAL; |
bdebd6a2 | 3065 | kaddr += off; |
83342314 | 3066 | |
83342314 | 3067 | do { |
e69e9d4a | 3068 | struct page *page = vmalloc_to_page(kaddr); |
db64fe02 NP |
3069 | int ret; |
3070 | ||
83342314 NP |
3071 | ret = vm_insert_page(vma, uaddr, page); |
3072 | if (ret) | |
3073 | return ret; | |
3074 | ||
3075 | uaddr += PAGE_SIZE; | |
e69e9d4a HD |
3076 | kaddr += PAGE_SIZE; |
3077 | size -= PAGE_SIZE; | |
3078 | } while (size > 0); | |
83342314 | 3079 | |
314e51b9 | 3080 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
83342314 | 3081 | |
db64fe02 | 3082 | return 0; |
83342314 | 3083 | } |
e69e9d4a HD |
3084 | EXPORT_SYMBOL(remap_vmalloc_range_partial); |
3085 | ||
3086 | /** | |
92eac168 MR |
3087 | * remap_vmalloc_range - map vmalloc pages to userspace |
3088 | * @vma: vma to cover (map full range of vma) | |
3089 | * @addr: vmalloc memory | |
3090 | * @pgoff: number of pages into addr before first page to map | |
e69e9d4a | 3091 | * |
92eac168 | 3092 | * Returns: 0 for success, -Exxx on failure |
e69e9d4a | 3093 | * |
92eac168 MR |
3094 | * This function checks that addr is a valid vmalloc'ed area, and |
3095 | * that it is big enough to cover the vma. Will return failure if | |
3096 | * that criteria isn't met. | |
e69e9d4a | 3097 | * |
92eac168 | 3098 | * Similar to remap_pfn_range() (see mm/memory.c) |
e69e9d4a HD |
3099 | */ |
3100 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, | |
3101 | unsigned long pgoff) | |
3102 | { | |
3103 | return remap_vmalloc_range_partial(vma, vma->vm_start, | |
bdebd6a2 | 3104 | addr, pgoff, |
e69e9d4a HD |
3105 | vma->vm_end - vma->vm_start); |
3106 | } | |
83342314 NP |
3107 | EXPORT_SYMBOL(remap_vmalloc_range); |
3108 | ||
1eeb66a1 | 3109 | /* |
763802b5 JR |
3110 | * Implement stubs for vmalloc_sync_[un]mappings () if the architecture chose |
3111 | * not to have one. | |
3f8fd02b JR |
3112 | * |
3113 | * The purpose of this function is to make sure the vmalloc area | |
3114 | * mappings are identical in all page-tables in the system. | |
1eeb66a1 | 3115 | */ |
763802b5 | 3116 | void __weak vmalloc_sync_mappings(void) |
1eeb66a1 CH |
3117 | { |
3118 | } | |
5f4352fb | 3119 | |
763802b5 JR |
3120 | void __weak vmalloc_sync_unmappings(void) |
3121 | { | |
3122 | } | |
5f4352fb | 3123 | |
8b1e0f81 | 3124 | static int f(pte_t *pte, unsigned long addr, void *data) |
5f4352fb | 3125 | { |
cd12909c DV |
3126 | pte_t ***p = data; |
3127 | ||
3128 | if (p) { | |
3129 | *(*p) = pte; | |
3130 | (*p)++; | |
3131 | } | |
5f4352fb JF |
3132 | return 0; |
3133 | } | |
3134 | ||
3135 | /** | |
92eac168 MR |
3136 | * alloc_vm_area - allocate a range of kernel address space |
3137 | * @size: size of the area | |
3138 | * @ptes: returns the PTEs for the address space | |
7682486b | 3139 | * |
92eac168 | 3140 | * Returns: NULL on failure, vm_struct on success |
5f4352fb | 3141 | * |
92eac168 MR |
3142 | * This function reserves a range of kernel address space, and |
3143 | * allocates pagetables to map that range. No actual mappings | |
3144 | * are created. | |
cd12909c | 3145 | * |
92eac168 MR |
3146 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
3147 | * allocated for the VM area are returned. | |
5f4352fb | 3148 | */ |
cd12909c | 3149 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
5f4352fb JF |
3150 | { |
3151 | struct vm_struct *area; | |
3152 | ||
23016969 CL |
3153 | area = get_vm_area_caller(size, VM_IOREMAP, |
3154 | __builtin_return_address(0)); | |
5f4352fb JF |
3155 | if (area == NULL) |
3156 | return NULL; | |
3157 | ||
3158 | /* | |
3159 | * This ensures that page tables are constructed for this region | |
3160 | * of kernel virtual address space and mapped into init_mm. | |
3161 | */ | |
3162 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, | |
cd12909c | 3163 | size, f, ptes ? &ptes : NULL)) { |
5f4352fb JF |
3164 | free_vm_area(area); |
3165 | return NULL; | |
3166 | } | |
3167 | ||
5f4352fb JF |
3168 | return area; |
3169 | } | |
3170 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
3171 | ||
3172 | void free_vm_area(struct vm_struct *area) | |
3173 | { | |
3174 | struct vm_struct *ret; | |
3175 | ret = remove_vm_area(area->addr); | |
3176 | BUG_ON(ret != area); | |
3177 | kfree(area); | |
3178 | } | |
3179 | EXPORT_SYMBOL_GPL(free_vm_area); | |
a10aa579 | 3180 | |
4f8b02b4 | 3181 | #ifdef CONFIG_SMP |
ca23e405 TH |
3182 | static struct vmap_area *node_to_va(struct rb_node *n) |
3183 | { | |
4583e773 | 3184 | return rb_entry_safe(n, struct vmap_area, rb_node); |
ca23e405 TH |
3185 | } |
3186 | ||
3187 | /** | |
68ad4a33 URS |
3188 | * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to |
3189 | * @addr: target address | |
ca23e405 | 3190 | * |
68ad4a33 URS |
3191 | * Returns: vmap_area if it is found. If there is no such area |
3192 | * the first highest(reverse order) vmap_area is returned | |
3193 | * i.e. va->va_start < addr && va->va_end < addr or NULL | |
3194 | * if there are no any areas before @addr. | |
ca23e405 | 3195 | */ |
68ad4a33 URS |
3196 | static struct vmap_area * |
3197 | pvm_find_va_enclose_addr(unsigned long addr) | |
ca23e405 | 3198 | { |
68ad4a33 URS |
3199 | struct vmap_area *va, *tmp; |
3200 | struct rb_node *n; | |
3201 | ||
3202 | n = free_vmap_area_root.rb_node; | |
3203 | va = NULL; | |
ca23e405 TH |
3204 | |
3205 | while (n) { | |
68ad4a33 URS |
3206 | tmp = rb_entry(n, struct vmap_area, rb_node); |
3207 | if (tmp->va_start <= addr) { | |
3208 | va = tmp; | |
3209 | if (tmp->va_end >= addr) | |
3210 | break; | |
3211 | ||
ca23e405 | 3212 | n = n->rb_right; |
68ad4a33 URS |
3213 | } else { |
3214 | n = n->rb_left; | |
3215 | } | |
ca23e405 TH |
3216 | } |
3217 | ||
68ad4a33 | 3218 | return va; |
ca23e405 TH |
3219 | } |
3220 | ||
3221 | /** | |
68ad4a33 URS |
3222 | * pvm_determine_end_from_reverse - find the highest aligned address |
3223 | * of free block below VMALLOC_END | |
3224 | * @va: | |
3225 | * in - the VA we start the search(reverse order); | |
3226 | * out - the VA with the highest aligned end address. | |
ca23e405 | 3227 | * |
68ad4a33 | 3228 | * Returns: determined end address within vmap_area |
ca23e405 | 3229 | */ |
68ad4a33 URS |
3230 | static unsigned long |
3231 | pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) | |
ca23e405 | 3232 | { |
68ad4a33 | 3233 | unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
ca23e405 TH |
3234 | unsigned long addr; |
3235 | ||
68ad4a33 URS |
3236 | if (likely(*va)) { |
3237 | list_for_each_entry_from_reverse((*va), | |
3238 | &free_vmap_area_list, list) { | |
3239 | addr = min((*va)->va_end & ~(align - 1), vmalloc_end); | |
3240 | if ((*va)->va_start < addr) | |
3241 | return addr; | |
3242 | } | |
ca23e405 TH |
3243 | } |
3244 | ||
68ad4a33 | 3245 | return 0; |
ca23e405 TH |
3246 | } |
3247 | ||
3248 | /** | |
3249 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator | |
3250 | * @offsets: array containing offset of each area | |
3251 | * @sizes: array containing size of each area | |
3252 | * @nr_vms: the number of areas to allocate | |
3253 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this | |
ca23e405 TH |
3254 | * |
3255 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated | |
3256 | * vm_structs on success, %NULL on failure | |
3257 | * | |
3258 | * Percpu allocator wants to use congruent vm areas so that it can | |
3259 | * maintain the offsets among percpu areas. This function allocates | |
ec3f64fc DR |
3260 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
3261 | * be scattered pretty far, distance between two areas easily going up | |
3262 | * to gigabytes. To avoid interacting with regular vmallocs, these | |
3263 | * areas are allocated from top. | |
ca23e405 | 3264 | * |
68ad4a33 URS |
3265 | * Despite its complicated look, this allocator is rather simple. It |
3266 | * does everything top-down and scans free blocks from the end looking | |
3267 | * for matching base. While scanning, if any of the areas do not fit the | |
3268 | * base address is pulled down to fit the area. Scanning is repeated till | |
3269 | * all the areas fit and then all necessary data structures are inserted | |
3270 | * and the result is returned. | |
ca23e405 TH |
3271 | */ |
3272 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, | |
3273 | const size_t *sizes, int nr_vms, | |
ec3f64fc | 3274 | size_t align) |
ca23e405 TH |
3275 | { |
3276 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); | |
3277 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); | |
68ad4a33 | 3278 | struct vmap_area **vas, *va; |
ca23e405 TH |
3279 | struct vm_struct **vms; |
3280 | int area, area2, last_area, term_area; | |
253a496d | 3281 | unsigned long base, start, size, end, last_end, orig_start, orig_end; |
ca23e405 | 3282 | bool purged = false; |
68ad4a33 | 3283 | enum fit_type type; |
ca23e405 | 3284 | |
ca23e405 | 3285 | /* verify parameters and allocate data structures */ |
891c49ab | 3286 | BUG_ON(offset_in_page(align) || !is_power_of_2(align)); |
ca23e405 TH |
3287 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
3288 | start = offsets[area]; | |
3289 | end = start + sizes[area]; | |
3290 | ||
3291 | /* is everything aligned properly? */ | |
3292 | BUG_ON(!IS_ALIGNED(offsets[area], align)); | |
3293 | BUG_ON(!IS_ALIGNED(sizes[area], align)); | |
3294 | ||
3295 | /* detect the area with the highest address */ | |
3296 | if (start > offsets[last_area]) | |
3297 | last_area = area; | |
3298 | ||
c568da28 | 3299 | for (area2 = area + 1; area2 < nr_vms; area2++) { |
ca23e405 TH |
3300 | unsigned long start2 = offsets[area2]; |
3301 | unsigned long end2 = start2 + sizes[area2]; | |
3302 | ||
c568da28 | 3303 | BUG_ON(start2 < end && start < end2); |
ca23e405 TH |
3304 | } |
3305 | } | |
3306 | last_end = offsets[last_area] + sizes[last_area]; | |
3307 | ||
3308 | if (vmalloc_end - vmalloc_start < last_end) { | |
3309 | WARN_ON(true); | |
3310 | return NULL; | |
3311 | } | |
3312 | ||
4d67d860 TM |
3313 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
3314 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); | |
ca23e405 | 3315 | if (!vas || !vms) |
f1db7afd | 3316 | goto err_free2; |
ca23e405 TH |
3317 | |
3318 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3319 | vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); |
ec3f64fc | 3320 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
ca23e405 TH |
3321 | if (!vas[area] || !vms[area]) |
3322 | goto err_free; | |
3323 | } | |
3324 | retry: | |
e36176be | 3325 | spin_lock(&free_vmap_area_lock); |
ca23e405 TH |
3326 | |
3327 | /* start scanning - we scan from the top, begin with the last area */ | |
3328 | area = term_area = last_area; | |
3329 | start = offsets[area]; | |
3330 | end = start + sizes[area]; | |
3331 | ||
68ad4a33 URS |
3332 | va = pvm_find_va_enclose_addr(vmalloc_end); |
3333 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3334 | |
3335 | while (true) { | |
ca23e405 TH |
3336 | /* |
3337 | * base might have underflowed, add last_end before | |
3338 | * comparing. | |
3339 | */ | |
68ad4a33 URS |
3340 | if (base + last_end < vmalloc_start + last_end) |
3341 | goto overflow; | |
ca23e405 TH |
3342 | |
3343 | /* | |
68ad4a33 | 3344 | * Fitting base has not been found. |
ca23e405 | 3345 | */ |
68ad4a33 URS |
3346 | if (va == NULL) |
3347 | goto overflow; | |
ca23e405 | 3348 | |
5336e52c | 3349 | /* |
d8cc323d | 3350 | * If required width exceeds current VA block, move |
5336e52c KS |
3351 | * base downwards and then recheck. |
3352 | */ | |
3353 | if (base + end > va->va_end) { | |
3354 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
3355 | term_area = area; | |
3356 | continue; | |
3357 | } | |
3358 | ||
ca23e405 | 3359 | /* |
68ad4a33 | 3360 | * If this VA does not fit, move base downwards and recheck. |
ca23e405 | 3361 | */ |
5336e52c | 3362 | if (base + start < va->va_start) { |
68ad4a33 URS |
3363 | va = node_to_va(rb_prev(&va->rb_node)); |
3364 | base = pvm_determine_end_from_reverse(&va, align) - end; | |
ca23e405 TH |
3365 | term_area = area; |
3366 | continue; | |
3367 | } | |
3368 | ||
3369 | /* | |
3370 | * This area fits, move on to the previous one. If | |
3371 | * the previous one is the terminal one, we're done. | |
3372 | */ | |
3373 | area = (area + nr_vms - 1) % nr_vms; | |
3374 | if (area == term_area) | |
3375 | break; | |
68ad4a33 | 3376 | |
ca23e405 TH |
3377 | start = offsets[area]; |
3378 | end = start + sizes[area]; | |
68ad4a33 | 3379 | va = pvm_find_va_enclose_addr(base + end); |
ca23e405 | 3380 | } |
68ad4a33 | 3381 | |
ca23e405 TH |
3382 | /* we've found a fitting base, insert all va's */ |
3383 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 | 3384 | int ret; |
ca23e405 | 3385 | |
68ad4a33 URS |
3386 | start = base + offsets[area]; |
3387 | size = sizes[area]; | |
ca23e405 | 3388 | |
68ad4a33 URS |
3389 | va = pvm_find_va_enclose_addr(start); |
3390 | if (WARN_ON_ONCE(va == NULL)) | |
3391 | /* It is a BUG(), but trigger recovery instead. */ | |
3392 | goto recovery; | |
3393 | ||
3394 | type = classify_va_fit_type(va, start, size); | |
3395 | if (WARN_ON_ONCE(type == NOTHING_FIT)) | |
3396 | /* It is a BUG(), but trigger recovery instead. */ | |
3397 | goto recovery; | |
3398 | ||
3399 | ret = adjust_va_to_fit_type(va, start, size, type); | |
3400 | if (unlikely(ret)) | |
3401 | goto recovery; | |
3402 | ||
3403 | /* Allocated area. */ | |
3404 | va = vas[area]; | |
3405 | va->va_start = start; | |
3406 | va->va_end = start + size; | |
68ad4a33 | 3407 | } |
ca23e405 | 3408 | |
e36176be | 3409 | spin_unlock(&free_vmap_area_lock); |
ca23e405 | 3410 | |
253a496d DA |
3411 | /* populate the kasan shadow space */ |
3412 | for (area = 0; area < nr_vms; area++) { | |
3413 | if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area])) | |
3414 | goto err_free_shadow; | |
3415 | ||
3416 | kasan_unpoison_vmalloc((void *)vas[area]->va_start, | |
3417 | sizes[area]); | |
3418 | } | |
3419 | ||
ca23e405 | 3420 | /* insert all vm's */ |
e36176be URS |
3421 | spin_lock(&vmap_area_lock); |
3422 | for (area = 0; area < nr_vms; area++) { | |
3423 | insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list); | |
3424 | ||
3425 | setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC, | |
3645cb4a | 3426 | pcpu_get_vm_areas); |
e36176be URS |
3427 | } |
3428 | spin_unlock(&vmap_area_lock); | |
ca23e405 TH |
3429 | |
3430 | kfree(vas); | |
3431 | return vms; | |
3432 | ||
68ad4a33 | 3433 | recovery: |
e36176be URS |
3434 | /* |
3435 | * Remove previously allocated areas. There is no | |
3436 | * need in removing these areas from the busy tree, | |
3437 | * because they are inserted only on the final step | |
3438 | * and when pcpu_get_vm_areas() is success. | |
3439 | */ | |
68ad4a33 | 3440 | while (area--) { |
253a496d DA |
3441 | orig_start = vas[area]->va_start; |
3442 | orig_end = vas[area]->va_end; | |
3443 | va = merge_or_add_vmap_area(vas[area], &free_vmap_area_root, | |
3444 | &free_vmap_area_list); | |
3445 | kasan_release_vmalloc(orig_start, orig_end, | |
3446 | va->va_start, va->va_end); | |
68ad4a33 URS |
3447 | vas[area] = NULL; |
3448 | } | |
3449 | ||
3450 | overflow: | |
e36176be | 3451 | spin_unlock(&free_vmap_area_lock); |
68ad4a33 URS |
3452 | if (!purged) { |
3453 | purge_vmap_area_lazy(); | |
3454 | purged = true; | |
3455 | ||
3456 | /* Before "retry", check if we recover. */ | |
3457 | for (area = 0; area < nr_vms; area++) { | |
3458 | if (vas[area]) | |
3459 | continue; | |
3460 | ||
3461 | vas[area] = kmem_cache_zalloc( | |
3462 | vmap_area_cachep, GFP_KERNEL); | |
3463 | if (!vas[area]) | |
3464 | goto err_free; | |
3465 | } | |
3466 | ||
3467 | goto retry; | |
3468 | } | |
3469 | ||
ca23e405 TH |
3470 | err_free: |
3471 | for (area = 0; area < nr_vms; area++) { | |
68ad4a33 URS |
3472 | if (vas[area]) |
3473 | kmem_cache_free(vmap_area_cachep, vas[area]); | |
3474 | ||
f1db7afd | 3475 | kfree(vms[area]); |
ca23e405 | 3476 | } |
f1db7afd | 3477 | err_free2: |
ca23e405 TH |
3478 | kfree(vas); |
3479 | kfree(vms); | |
3480 | return NULL; | |
253a496d DA |
3481 | |
3482 | err_free_shadow: | |
3483 | spin_lock(&free_vmap_area_lock); | |
3484 | /* | |
3485 | * We release all the vmalloc shadows, even the ones for regions that | |
3486 | * hadn't been successfully added. This relies on kasan_release_vmalloc | |
3487 | * being able to tolerate this case. | |
3488 | */ | |
3489 | for (area = 0; area < nr_vms; area++) { | |
3490 | orig_start = vas[area]->va_start; | |
3491 | orig_end = vas[area]->va_end; | |
3492 | va = merge_or_add_vmap_area(vas[area], &free_vmap_area_root, | |
3493 | &free_vmap_area_list); | |
3494 | kasan_release_vmalloc(orig_start, orig_end, | |
3495 | va->va_start, va->va_end); | |
3496 | vas[area] = NULL; | |
3497 | kfree(vms[area]); | |
3498 | } | |
3499 | spin_unlock(&free_vmap_area_lock); | |
3500 | kfree(vas); | |
3501 | kfree(vms); | |
3502 | return NULL; | |
ca23e405 TH |
3503 | } |
3504 | ||
3505 | /** | |
3506 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator | |
3507 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() | |
3508 | * @nr_vms: the number of allocated areas | |
3509 | * | |
3510 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). | |
3511 | */ | |
3512 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) | |
3513 | { | |
3514 | int i; | |
3515 | ||
3516 | for (i = 0; i < nr_vms; i++) | |
3517 | free_vm_area(vms[i]); | |
3518 | kfree(vms); | |
3519 | } | |
4f8b02b4 | 3520 | #endif /* CONFIG_SMP */ |
a10aa579 CL |
3521 | |
3522 | #ifdef CONFIG_PROC_FS | |
3523 | static void *s_start(struct seq_file *m, loff_t *pos) | |
e36176be | 3524 | __acquires(&vmap_purge_lock) |
d4033afd | 3525 | __acquires(&vmap_area_lock) |
a10aa579 | 3526 | { |
e36176be | 3527 | mutex_lock(&vmap_purge_lock); |
d4033afd | 3528 | spin_lock(&vmap_area_lock); |
e36176be | 3529 | |
3f500069 | 3530 | return seq_list_start(&vmap_area_list, *pos); |
a10aa579 CL |
3531 | } |
3532 | ||
3533 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3534 | { | |
3f500069 | 3535 | return seq_list_next(p, &vmap_area_list, pos); |
a10aa579 CL |
3536 | } |
3537 | ||
3538 | static void s_stop(struct seq_file *m, void *p) | |
e36176be | 3539 | __releases(&vmap_purge_lock) |
d4033afd | 3540 | __releases(&vmap_area_lock) |
a10aa579 | 3541 | { |
e36176be | 3542 | mutex_unlock(&vmap_purge_lock); |
d4033afd | 3543 | spin_unlock(&vmap_area_lock); |
a10aa579 CL |
3544 | } |
3545 | ||
a47a126a ED |
3546 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
3547 | { | |
e5adfffc | 3548 | if (IS_ENABLED(CONFIG_NUMA)) { |
a47a126a ED |
3549 | unsigned int nr, *counters = m->private; |
3550 | ||
3551 | if (!counters) | |
3552 | return; | |
3553 | ||
af12346c WL |
3554 | if (v->flags & VM_UNINITIALIZED) |
3555 | return; | |
7e5b528b DV |
3556 | /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */ |
3557 | smp_rmb(); | |
af12346c | 3558 | |
a47a126a ED |
3559 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
3560 | ||
3561 | for (nr = 0; nr < v->nr_pages; nr++) | |
3562 | counters[page_to_nid(v->pages[nr])]++; | |
3563 | ||
3564 | for_each_node_state(nr, N_HIGH_MEMORY) | |
3565 | if (counters[nr]) | |
3566 | seq_printf(m, " N%u=%u", nr, counters[nr]); | |
3567 | } | |
3568 | } | |
3569 | ||
dd3b8353 URS |
3570 | static void show_purge_info(struct seq_file *m) |
3571 | { | |
3572 | struct llist_node *head; | |
3573 | struct vmap_area *va; | |
3574 | ||
3575 | head = READ_ONCE(vmap_purge_list.first); | |
3576 | if (head == NULL) | |
3577 | return; | |
3578 | ||
3579 | llist_for_each_entry(va, head, purge_list) { | |
3580 | seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n", | |
3581 | (void *)va->va_start, (void *)va->va_end, | |
3582 | va->va_end - va->va_start); | |
3583 | } | |
3584 | } | |
3585 | ||
a10aa579 CL |
3586 | static int s_show(struct seq_file *m, void *p) |
3587 | { | |
3f500069 | 3588 | struct vmap_area *va; |
d4033afd JK |
3589 | struct vm_struct *v; |
3590 | ||
3f500069 | 3591 | va = list_entry(p, struct vmap_area, list); |
3592 | ||
c2ce8c14 | 3593 | /* |
688fcbfc PL |
3594 | * s_show can encounter race with remove_vm_area, !vm on behalf |
3595 | * of vmap area is being tear down or vm_map_ram allocation. | |
c2ce8c14 | 3596 | */ |
688fcbfc | 3597 | if (!va->vm) { |
dd3b8353 | 3598 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
78c72746 | 3599 | (void *)va->va_start, (void *)va->va_end, |
dd3b8353 | 3600 | va->va_end - va->va_start); |
78c72746 | 3601 | |
d4033afd | 3602 | return 0; |
78c72746 | 3603 | } |
d4033afd JK |
3604 | |
3605 | v = va->vm; | |
a10aa579 | 3606 | |
45ec1690 | 3607 | seq_printf(m, "0x%pK-0x%pK %7ld", |
a10aa579 CL |
3608 | v->addr, v->addr + v->size, v->size); |
3609 | ||
62c70bce JP |
3610 | if (v->caller) |
3611 | seq_printf(m, " %pS", v->caller); | |
23016969 | 3612 | |
a10aa579 CL |
3613 | if (v->nr_pages) |
3614 | seq_printf(m, " pages=%d", v->nr_pages); | |
3615 | ||
3616 | if (v->phys_addr) | |
199eaa05 | 3617 | seq_printf(m, " phys=%pa", &v->phys_addr); |
a10aa579 CL |
3618 | |
3619 | if (v->flags & VM_IOREMAP) | |
f4527c90 | 3620 | seq_puts(m, " ioremap"); |
a10aa579 CL |
3621 | |
3622 | if (v->flags & VM_ALLOC) | |
f4527c90 | 3623 | seq_puts(m, " vmalloc"); |
a10aa579 CL |
3624 | |
3625 | if (v->flags & VM_MAP) | |
f4527c90 | 3626 | seq_puts(m, " vmap"); |
a10aa579 CL |
3627 | |
3628 | if (v->flags & VM_USERMAP) | |
f4527c90 | 3629 | seq_puts(m, " user"); |
a10aa579 | 3630 | |
fe9041c2 CH |
3631 | if (v->flags & VM_DMA_COHERENT) |
3632 | seq_puts(m, " dma-coherent"); | |
3633 | ||
244d63ee | 3634 | if (is_vmalloc_addr(v->pages)) |
f4527c90 | 3635 | seq_puts(m, " vpages"); |
a10aa579 | 3636 | |
a47a126a | 3637 | show_numa_info(m, v); |
a10aa579 | 3638 | seq_putc(m, '\n'); |
dd3b8353 URS |
3639 | |
3640 | /* | |
3641 | * As a final step, dump "unpurged" areas. Note, | |
3642 | * that entire "/proc/vmallocinfo" output will not | |
3643 | * be address sorted, because the purge list is not | |
3644 | * sorted. | |
3645 | */ | |
3646 | if (list_is_last(&va->list, &vmap_area_list)) | |
3647 | show_purge_info(m); | |
3648 | ||
a10aa579 CL |
3649 | return 0; |
3650 | } | |
3651 | ||
5f6a6a9c | 3652 | static const struct seq_operations vmalloc_op = { |
a10aa579 CL |
3653 | .start = s_start, |
3654 | .next = s_next, | |
3655 | .stop = s_stop, | |
3656 | .show = s_show, | |
3657 | }; | |
5f6a6a9c | 3658 | |
5f6a6a9c AD |
3659 | static int __init proc_vmalloc_init(void) |
3660 | { | |
fddda2b7 | 3661 | if (IS_ENABLED(CONFIG_NUMA)) |
0825a6f9 | 3662 | proc_create_seq_private("vmallocinfo", 0400, NULL, |
44414d82 CH |
3663 | &vmalloc_op, |
3664 | nr_node_ids * sizeof(unsigned int), NULL); | |
fddda2b7 | 3665 | else |
0825a6f9 | 3666 | proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op); |
5f6a6a9c AD |
3667 | return 0; |
3668 | } | |
3669 | module_init(proc_vmalloc_init); | |
db3808c1 | 3670 | |
a10aa579 | 3671 | #endif |