| 1 | /* |
| 2 | * linux/mm/vmalloc.c |
| 3 | * |
| 4 | * Copyright (C) 1993 Linus Torvalds |
| 5 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 |
| 6 | * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000 |
| 7 | * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002 |
| 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
| 9 | */ |
| 10 | |
| 11 | #include <linux/vmalloc.h> |
| 12 | #include <linux/mm.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/highmem.h> |
| 15 | #include <linux/sched.h> |
| 16 | #include <linux/slab.h> |
| 17 | #include <linux/spinlock.h> |
| 18 | #include <linux/interrupt.h> |
| 19 | #include <linux/proc_fs.h> |
| 20 | #include <linux/seq_file.h> |
| 21 | #include <linux/debugobjects.h> |
| 22 | #include <linux/kallsyms.h> |
| 23 | #include <linux/list.h> |
| 24 | #include <linux/rbtree.h> |
| 25 | #include <linux/radix-tree.h> |
| 26 | #include <linux/rcupdate.h> |
| 27 | #include <linux/pfn.h> |
| 28 | #include <linux/kmemleak.h> |
| 29 | #include <linux/atomic.h> |
| 30 | #include <linux/llist.h> |
| 31 | #include <asm/uaccess.h> |
| 32 | #include <asm/tlbflush.h> |
| 33 | #include <asm/shmparam.h> |
| 34 | |
| 35 | struct vfree_deferred { |
| 36 | struct llist_head list; |
| 37 | struct work_struct wq; |
| 38 | }; |
| 39 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred); |
| 40 | |
| 41 | static void __vunmap(const void *, int); |
| 42 | |
| 43 | static void free_work(struct work_struct *w) |
| 44 | { |
| 45 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq); |
| 46 | struct llist_node *llnode = llist_del_all(&p->list); |
| 47 | while (llnode) { |
| 48 | void *p = llnode; |
| 49 | llnode = llist_next(llnode); |
| 50 | __vunmap(p, 1); |
| 51 | } |
| 52 | } |
| 53 | |
| 54 | /*** Page table manipulation functions ***/ |
| 55 | |
| 56 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
| 57 | { |
| 58 | pte_t *pte; |
| 59 | |
| 60 | pte = pte_offset_kernel(pmd, addr); |
| 61 | do { |
| 62 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); |
| 63 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); |
| 64 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 65 | } |
| 66 | |
| 67 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
| 68 | { |
| 69 | pmd_t *pmd; |
| 70 | unsigned long next; |
| 71 | |
| 72 | pmd = pmd_offset(pud, addr); |
| 73 | do { |
| 74 | next = pmd_addr_end(addr, end); |
| 75 | if (pmd_none_or_clear_bad(pmd)) |
| 76 | continue; |
| 77 | vunmap_pte_range(pmd, addr, next); |
| 78 | } while (pmd++, addr = next, addr != end); |
| 79 | } |
| 80 | |
| 81 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
| 82 | { |
| 83 | pud_t *pud; |
| 84 | unsigned long next; |
| 85 | |
| 86 | pud = pud_offset(pgd, addr); |
| 87 | do { |
| 88 | next = pud_addr_end(addr, end); |
| 89 | if (pud_none_or_clear_bad(pud)) |
| 90 | continue; |
| 91 | vunmap_pmd_range(pud, addr, next); |
| 92 | } while (pud++, addr = next, addr != end); |
| 93 | } |
| 94 | |
| 95 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
| 96 | { |
| 97 | pgd_t *pgd; |
| 98 | unsigned long next; |
| 99 | |
| 100 | BUG_ON(addr >= end); |
| 101 | pgd = pgd_offset_k(addr); |
| 102 | do { |
| 103 | next = pgd_addr_end(addr, end); |
| 104 | if (pgd_none_or_clear_bad(pgd)) |
| 105 | continue; |
| 106 | vunmap_pud_range(pgd, addr, next); |
| 107 | } while (pgd++, addr = next, addr != end); |
| 108 | } |
| 109 | |
| 110 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, |
| 111 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
| 112 | { |
| 113 | pte_t *pte; |
| 114 | |
| 115 | /* |
| 116 | * nr is a running index into the array which helps higher level |
| 117 | * callers keep track of where we're up to. |
| 118 | */ |
| 119 | |
| 120 | pte = pte_alloc_kernel(pmd, addr); |
| 121 | if (!pte) |
| 122 | return -ENOMEM; |
| 123 | do { |
| 124 | struct page *page = pages[*nr]; |
| 125 | |
| 126 | if (WARN_ON(!pte_none(*pte))) |
| 127 | return -EBUSY; |
| 128 | if (WARN_ON(!page)) |
| 129 | return -ENOMEM; |
| 130 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); |
| 131 | (*nr)++; |
| 132 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 133 | return 0; |
| 134 | } |
| 135 | |
| 136 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
| 137 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
| 138 | { |
| 139 | pmd_t *pmd; |
| 140 | unsigned long next; |
| 141 | |
| 142 | pmd = pmd_alloc(&init_mm, pud, addr); |
| 143 | if (!pmd) |
| 144 | return -ENOMEM; |
| 145 | do { |
| 146 | next = pmd_addr_end(addr, end); |
| 147 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
| 148 | return -ENOMEM; |
| 149 | } while (pmd++, addr = next, addr != end); |
| 150 | return 0; |
| 151 | } |
| 152 | |
| 153 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
| 154 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
| 155 | { |
| 156 | pud_t *pud; |
| 157 | unsigned long next; |
| 158 | |
| 159 | pud = pud_alloc(&init_mm, pgd, addr); |
| 160 | if (!pud) |
| 161 | return -ENOMEM; |
| 162 | do { |
| 163 | next = pud_addr_end(addr, end); |
| 164 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
| 165 | return -ENOMEM; |
| 166 | } while (pud++, addr = next, addr != end); |
| 167 | return 0; |
| 168 | } |
| 169 | |
| 170 | /* |
| 171 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and |
| 172 | * will have pfns corresponding to the "pages" array. |
| 173 | * |
| 174 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] |
| 175 | */ |
| 176 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
| 177 | pgprot_t prot, struct page **pages) |
| 178 | { |
| 179 | pgd_t *pgd; |
| 180 | unsigned long next; |
| 181 | unsigned long addr = start; |
| 182 | int err = 0; |
| 183 | int nr = 0; |
| 184 | |
| 185 | BUG_ON(addr >= end); |
| 186 | pgd = pgd_offset_k(addr); |
| 187 | do { |
| 188 | next = pgd_addr_end(addr, end); |
| 189 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
| 190 | if (err) |
| 191 | return err; |
| 192 | } while (pgd++, addr = next, addr != end); |
| 193 | |
| 194 | return nr; |
| 195 | } |
| 196 | |
| 197 | static int vmap_page_range(unsigned long start, unsigned long end, |
| 198 | pgprot_t prot, struct page **pages) |
| 199 | { |
| 200 | int ret; |
| 201 | |
| 202 | ret = vmap_page_range_noflush(start, end, prot, pages); |
| 203 | flush_cache_vmap(start, end); |
| 204 | return ret; |
| 205 | } |
| 206 | |
| 207 | int is_vmalloc_or_module_addr(const void *x) |
| 208 | { |
| 209 | /* |
| 210 | * ARM, x86-64 and sparc64 put modules in a special place, |
| 211 | * and fall back on vmalloc() if that fails. Others |
| 212 | * just put it in the vmalloc space. |
| 213 | */ |
| 214 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) |
| 215 | unsigned long addr = (unsigned long)x; |
| 216 | if (addr >= MODULES_VADDR && addr < MODULES_END) |
| 217 | return 1; |
| 218 | #endif |
| 219 | return is_vmalloc_addr(x); |
| 220 | } |
| 221 | |
| 222 | /* |
| 223 | * Walk a vmap address to the struct page it maps. |
| 224 | */ |
| 225 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
| 226 | { |
| 227 | unsigned long addr = (unsigned long) vmalloc_addr; |
| 228 | struct page *page = NULL; |
| 229 | pgd_t *pgd = pgd_offset_k(addr); |
| 230 | |
| 231 | /* |
| 232 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for |
| 233 | * architectures that do not vmalloc module space |
| 234 | */ |
| 235 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
| 236 | |
| 237 | if (!pgd_none(*pgd)) { |
| 238 | pud_t *pud = pud_offset(pgd, addr); |
| 239 | if (!pud_none(*pud)) { |
| 240 | pmd_t *pmd = pmd_offset(pud, addr); |
| 241 | if (!pmd_none(*pmd)) { |
| 242 | pte_t *ptep, pte; |
| 243 | |
| 244 | ptep = pte_offset_map(pmd, addr); |
| 245 | pte = *ptep; |
| 246 | if (pte_present(pte)) |
| 247 | page = pte_page(pte); |
| 248 | pte_unmap(ptep); |
| 249 | } |
| 250 | } |
| 251 | } |
| 252 | return page; |
| 253 | } |
| 254 | EXPORT_SYMBOL(vmalloc_to_page); |
| 255 | |
| 256 | /* |
| 257 | * Map a vmalloc()-space virtual address to the physical page frame number. |
| 258 | */ |
| 259 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
| 260 | { |
| 261 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
| 262 | } |
| 263 | EXPORT_SYMBOL(vmalloc_to_pfn); |
| 264 | |
| 265 | |
| 266 | /*** Global kva allocator ***/ |
| 267 | |
| 268 | #define VM_LAZY_FREE 0x01 |
| 269 | #define VM_LAZY_FREEING 0x02 |
| 270 | #define VM_VM_AREA 0x04 |
| 271 | |
| 272 | static DEFINE_SPINLOCK(vmap_area_lock); |
| 273 | /* Export for kexec only */ |
| 274 | LIST_HEAD(vmap_area_list); |
| 275 | static struct rb_root vmap_area_root = RB_ROOT; |
| 276 | |
| 277 | /* The vmap cache globals are protected by vmap_area_lock */ |
| 278 | static struct rb_node *free_vmap_cache; |
| 279 | static unsigned long cached_hole_size; |
| 280 | static unsigned long cached_vstart; |
| 281 | static unsigned long cached_align; |
| 282 | |
| 283 | static unsigned long vmap_area_pcpu_hole; |
| 284 | |
| 285 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
| 286 | { |
| 287 | struct rb_node *n = vmap_area_root.rb_node; |
| 288 | |
| 289 | while (n) { |
| 290 | struct vmap_area *va; |
| 291 | |
| 292 | va = rb_entry(n, struct vmap_area, rb_node); |
| 293 | if (addr < va->va_start) |
| 294 | n = n->rb_left; |
| 295 | else if (addr > va->va_start) |
| 296 | n = n->rb_right; |
| 297 | else |
| 298 | return va; |
| 299 | } |
| 300 | |
| 301 | return NULL; |
| 302 | } |
| 303 | |
| 304 | static void __insert_vmap_area(struct vmap_area *va) |
| 305 | { |
| 306 | struct rb_node **p = &vmap_area_root.rb_node; |
| 307 | struct rb_node *parent = NULL; |
| 308 | struct rb_node *tmp; |
| 309 | |
| 310 | while (*p) { |
| 311 | struct vmap_area *tmp_va; |
| 312 | |
| 313 | parent = *p; |
| 314 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
| 315 | if (va->va_start < tmp_va->va_end) |
| 316 | p = &(*p)->rb_left; |
| 317 | else if (va->va_end > tmp_va->va_start) |
| 318 | p = &(*p)->rb_right; |
| 319 | else |
| 320 | BUG(); |
| 321 | } |
| 322 | |
| 323 | rb_link_node(&va->rb_node, parent, p); |
| 324 | rb_insert_color(&va->rb_node, &vmap_area_root); |
| 325 | |
| 326 | /* address-sort this list */ |
| 327 | tmp = rb_prev(&va->rb_node); |
| 328 | if (tmp) { |
| 329 | struct vmap_area *prev; |
| 330 | prev = rb_entry(tmp, struct vmap_area, rb_node); |
| 331 | list_add_rcu(&va->list, &prev->list); |
| 332 | } else |
| 333 | list_add_rcu(&va->list, &vmap_area_list); |
| 334 | } |
| 335 | |
| 336 | static void purge_vmap_area_lazy(void); |
| 337 | |
| 338 | /* |
| 339 | * Allocate a region of KVA of the specified size and alignment, within the |
| 340 | * vstart and vend. |
| 341 | */ |
| 342 | static struct vmap_area *alloc_vmap_area(unsigned long size, |
| 343 | unsigned long align, |
| 344 | unsigned long vstart, unsigned long vend, |
| 345 | int node, gfp_t gfp_mask) |
| 346 | { |
| 347 | struct vmap_area *va; |
| 348 | struct rb_node *n; |
| 349 | unsigned long addr; |
| 350 | int purged = 0; |
| 351 | struct vmap_area *first; |
| 352 | |
| 353 | BUG_ON(!size); |
| 354 | BUG_ON(size & ~PAGE_MASK); |
| 355 | BUG_ON(!is_power_of_2(align)); |
| 356 | |
| 357 | va = kmalloc_node(sizeof(struct vmap_area), |
| 358 | gfp_mask & GFP_RECLAIM_MASK, node); |
| 359 | if (unlikely(!va)) |
| 360 | return ERR_PTR(-ENOMEM); |
| 361 | |
| 362 | retry: |
| 363 | spin_lock(&vmap_area_lock); |
| 364 | /* |
| 365 | * Invalidate cache if we have more permissive parameters. |
| 366 | * cached_hole_size notes the largest hole noticed _below_ |
| 367 | * the vmap_area cached in free_vmap_cache: if size fits |
| 368 | * into that hole, we want to scan from vstart to reuse |
| 369 | * the hole instead of allocating above free_vmap_cache. |
| 370 | * Note that __free_vmap_area may update free_vmap_cache |
| 371 | * without updating cached_hole_size or cached_align. |
| 372 | */ |
| 373 | if (!free_vmap_cache || |
| 374 | size < cached_hole_size || |
| 375 | vstart < cached_vstart || |
| 376 | align < cached_align) { |
| 377 | nocache: |
| 378 | cached_hole_size = 0; |
| 379 | free_vmap_cache = NULL; |
| 380 | } |
| 381 | /* record if we encounter less permissive parameters */ |
| 382 | cached_vstart = vstart; |
| 383 | cached_align = align; |
| 384 | |
| 385 | /* find starting point for our search */ |
| 386 | if (free_vmap_cache) { |
| 387 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); |
| 388 | addr = ALIGN(first->va_end, align); |
| 389 | if (addr < vstart) |
| 390 | goto nocache; |
| 391 | if (addr + size < addr) |
| 392 | goto overflow; |
| 393 | |
| 394 | } else { |
| 395 | addr = ALIGN(vstart, align); |
| 396 | if (addr + size < addr) |
| 397 | goto overflow; |
| 398 | |
| 399 | n = vmap_area_root.rb_node; |
| 400 | first = NULL; |
| 401 | |
| 402 | while (n) { |
| 403 | struct vmap_area *tmp; |
| 404 | tmp = rb_entry(n, struct vmap_area, rb_node); |
| 405 | if (tmp->va_end >= addr) { |
| 406 | first = tmp; |
| 407 | if (tmp->va_start <= addr) |
| 408 | break; |
| 409 | n = n->rb_left; |
| 410 | } else |
| 411 | n = n->rb_right; |
| 412 | } |
| 413 | |
| 414 | if (!first) |
| 415 | goto found; |
| 416 | } |
| 417 | |
| 418 | /* from the starting point, walk areas until a suitable hole is found */ |
| 419 | while (addr + size > first->va_start && addr + size <= vend) { |
| 420 | if (addr + cached_hole_size < first->va_start) |
| 421 | cached_hole_size = first->va_start - addr; |
| 422 | addr = ALIGN(first->va_end, align); |
| 423 | if (addr + size < addr) |
| 424 | goto overflow; |
| 425 | |
| 426 | if (list_is_last(&first->list, &vmap_area_list)) |
| 427 | goto found; |
| 428 | |
| 429 | first = list_entry(first->list.next, |
| 430 | struct vmap_area, list); |
| 431 | } |
| 432 | |
| 433 | found: |
| 434 | if (addr + size > vend) |
| 435 | goto overflow; |
| 436 | |
| 437 | va->va_start = addr; |
| 438 | va->va_end = addr + size; |
| 439 | va->flags = 0; |
| 440 | __insert_vmap_area(va); |
| 441 | free_vmap_cache = &va->rb_node; |
| 442 | spin_unlock(&vmap_area_lock); |
| 443 | |
| 444 | BUG_ON(va->va_start & (align-1)); |
| 445 | BUG_ON(va->va_start < vstart); |
| 446 | BUG_ON(va->va_end > vend); |
| 447 | |
| 448 | return va; |
| 449 | |
| 450 | overflow: |
| 451 | spin_unlock(&vmap_area_lock); |
| 452 | if (!purged) { |
| 453 | purge_vmap_area_lazy(); |
| 454 | purged = 1; |
| 455 | goto retry; |
| 456 | } |
| 457 | if (printk_ratelimit()) |
| 458 | printk(KERN_WARNING |
| 459 | "vmap allocation for size %lu failed: " |
| 460 | "use vmalloc=<size> to increase size.\n", size); |
| 461 | kfree(va); |
| 462 | return ERR_PTR(-EBUSY); |
| 463 | } |
| 464 | |
| 465 | static void __free_vmap_area(struct vmap_area *va) |
| 466 | { |
| 467 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); |
| 468 | |
| 469 | if (free_vmap_cache) { |
| 470 | if (va->va_end < cached_vstart) { |
| 471 | free_vmap_cache = NULL; |
| 472 | } else { |
| 473 | struct vmap_area *cache; |
| 474 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); |
| 475 | if (va->va_start <= cache->va_start) { |
| 476 | free_vmap_cache = rb_prev(&va->rb_node); |
| 477 | /* |
| 478 | * We don't try to update cached_hole_size or |
| 479 | * cached_align, but it won't go very wrong. |
| 480 | */ |
| 481 | } |
| 482 | } |
| 483 | } |
| 484 | rb_erase(&va->rb_node, &vmap_area_root); |
| 485 | RB_CLEAR_NODE(&va->rb_node); |
| 486 | list_del_rcu(&va->list); |
| 487 | |
| 488 | /* |
| 489 | * Track the highest possible candidate for pcpu area |
| 490 | * allocation. Areas outside of vmalloc area can be returned |
| 491 | * here too, consider only end addresses which fall inside |
| 492 | * vmalloc area proper. |
| 493 | */ |
| 494 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) |
| 495 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); |
| 496 | |
| 497 | kfree_rcu(va, rcu_head); |
| 498 | } |
| 499 | |
| 500 | /* |
| 501 | * Free a region of KVA allocated by alloc_vmap_area |
| 502 | */ |
| 503 | static void free_vmap_area(struct vmap_area *va) |
| 504 | { |
| 505 | spin_lock(&vmap_area_lock); |
| 506 | __free_vmap_area(va); |
| 507 | spin_unlock(&vmap_area_lock); |
| 508 | } |
| 509 | |
| 510 | /* |
| 511 | * Clear the pagetable entries of a given vmap_area |
| 512 | */ |
| 513 | static void unmap_vmap_area(struct vmap_area *va) |
| 514 | { |
| 515 | vunmap_page_range(va->va_start, va->va_end); |
| 516 | } |
| 517 | |
| 518 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
| 519 | { |
| 520 | /* |
| 521 | * Unmap page tables and force a TLB flush immediately if |
| 522 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free |
| 523 | * bugs similarly to those in linear kernel virtual address |
| 524 | * space after a page has been freed. |
| 525 | * |
| 526 | * All the lazy freeing logic is still retained, in order to |
| 527 | * minimise intrusiveness of this debugging feature. |
| 528 | * |
| 529 | * This is going to be *slow* (linear kernel virtual address |
| 530 | * debugging doesn't do a broadcast TLB flush so it is a lot |
| 531 | * faster). |
| 532 | */ |
| 533 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 534 | vunmap_page_range(start, end); |
| 535 | flush_tlb_kernel_range(start, end); |
| 536 | #endif |
| 537 | } |
| 538 | |
| 539 | /* |
| 540 | * lazy_max_pages is the maximum amount of virtual address space we gather up |
| 541 | * before attempting to purge with a TLB flush. |
| 542 | * |
| 543 | * There is a tradeoff here: a larger number will cover more kernel page tables |
| 544 | * and take slightly longer to purge, but it will linearly reduce the number of |
| 545 | * global TLB flushes that must be performed. It would seem natural to scale |
| 546 | * this number up linearly with the number of CPUs (because vmapping activity |
| 547 | * could also scale linearly with the number of CPUs), however it is likely |
| 548 | * that in practice, workloads might be constrained in other ways that mean |
| 549 | * vmap activity will not scale linearly with CPUs. Also, I want to be |
| 550 | * conservative and not introduce a big latency on huge systems, so go with |
| 551 | * a less aggressive log scale. It will still be an improvement over the old |
| 552 | * code, and it will be simple to change the scale factor if we find that it |
| 553 | * becomes a problem on bigger systems. |
| 554 | */ |
| 555 | static unsigned long lazy_max_pages(void) |
| 556 | { |
| 557 | unsigned int log; |
| 558 | |
| 559 | log = fls(num_online_cpus()); |
| 560 | |
| 561 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); |
| 562 | } |
| 563 | |
| 564 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); |
| 565 | |
| 566 | /* for per-CPU blocks */ |
| 567 | static void purge_fragmented_blocks_allcpus(void); |
| 568 | |
| 569 | /* |
| 570 | * called before a call to iounmap() if the caller wants vm_area_struct's |
| 571 | * immediately freed. |
| 572 | */ |
| 573 | void set_iounmap_nonlazy(void) |
| 574 | { |
| 575 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); |
| 576 | } |
| 577 | |
| 578 | /* |
| 579 | * Purges all lazily-freed vmap areas. |
| 580 | * |
| 581 | * If sync is 0 then don't purge if there is already a purge in progress. |
| 582 | * If force_flush is 1, then flush kernel TLBs between *start and *end even |
| 583 | * if we found no lazy vmap areas to unmap (callers can use this to optimise |
| 584 | * their own TLB flushing). |
| 585 | * Returns with *start = min(*start, lowest purged address) |
| 586 | * *end = max(*end, highest purged address) |
| 587 | */ |
| 588 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, |
| 589 | int sync, int force_flush) |
| 590 | { |
| 591 | static DEFINE_SPINLOCK(purge_lock); |
| 592 | LIST_HEAD(valist); |
| 593 | struct vmap_area *va; |
| 594 | struct vmap_area *n_va; |
| 595 | int nr = 0; |
| 596 | |
| 597 | /* |
| 598 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers |
| 599 | * should not expect such behaviour. This just simplifies locking for |
| 600 | * the case that isn't actually used at the moment anyway. |
| 601 | */ |
| 602 | if (!sync && !force_flush) { |
| 603 | if (!spin_trylock(&purge_lock)) |
| 604 | return; |
| 605 | } else |
| 606 | spin_lock(&purge_lock); |
| 607 | |
| 608 | if (sync) |
| 609 | purge_fragmented_blocks_allcpus(); |
| 610 | |
| 611 | rcu_read_lock(); |
| 612 | list_for_each_entry_rcu(va, &vmap_area_list, list) { |
| 613 | if (va->flags & VM_LAZY_FREE) { |
| 614 | if (va->va_start < *start) |
| 615 | *start = va->va_start; |
| 616 | if (va->va_end > *end) |
| 617 | *end = va->va_end; |
| 618 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; |
| 619 | list_add_tail(&va->purge_list, &valist); |
| 620 | va->flags |= VM_LAZY_FREEING; |
| 621 | va->flags &= ~VM_LAZY_FREE; |
| 622 | } |
| 623 | } |
| 624 | rcu_read_unlock(); |
| 625 | |
| 626 | if (nr) |
| 627 | atomic_sub(nr, &vmap_lazy_nr); |
| 628 | |
| 629 | if (nr || force_flush) |
| 630 | flush_tlb_kernel_range(*start, *end); |
| 631 | |
| 632 | if (nr) { |
| 633 | spin_lock(&vmap_area_lock); |
| 634 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
| 635 | __free_vmap_area(va); |
| 636 | spin_unlock(&vmap_area_lock); |
| 637 | } |
| 638 | spin_unlock(&purge_lock); |
| 639 | } |
| 640 | |
| 641 | /* |
| 642 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody |
| 643 | * is already purging. |
| 644 | */ |
| 645 | static void try_purge_vmap_area_lazy(void) |
| 646 | { |
| 647 | unsigned long start = ULONG_MAX, end = 0; |
| 648 | |
| 649 | __purge_vmap_area_lazy(&start, &end, 0, 0); |
| 650 | } |
| 651 | |
| 652 | /* |
| 653 | * Kick off a purge of the outstanding lazy areas. |
| 654 | */ |
| 655 | static void purge_vmap_area_lazy(void) |
| 656 | { |
| 657 | unsigned long start = ULONG_MAX, end = 0; |
| 658 | |
| 659 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
| 660 | } |
| 661 | |
| 662 | /* |
| 663 | * Free a vmap area, caller ensuring that the area has been unmapped |
| 664 | * and flush_cache_vunmap had been called for the correct range |
| 665 | * previously. |
| 666 | */ |
| 667 | static void free_vmap_area_noflush(struct vmap_area *va) |
| 668 | { |
| 669 | va->flags |= VM_LAZY_FREE; |
| 670 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); |
| 671 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) |
| 672 | try_purge_vmap_area_lazy(); |
| 673 | } |
| 674 | |
| 675 | /* |
| 676 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been |
| 677 | * called for the correct range previously. |
| 678 | */ |
| 679 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) |
| 680 | { |
| 681 | unmap_vmap_area(va); |
| 682 | free_vmap_area_noflush(va); |
| 683 | } |
| 684 | |
| 685 | /* |
| 686 | * Free and unmap a vmap area |
| 687 | */ |
| 688 | static void free_unmap_vmap_area(struct vmap_area *va) |
| 689 | { |
| 690 | flush_cache_vunmap(va->va_start, va->va_end); |
| 691 | free_unmap_vmap_area_noflush(va); |
| 692 | } |
| 693 | |
| 694 | static struct vmap_area *find_vmap_area(unsigned long addr) |
| 695 | { |
| 696 | struct vmap_area *va; |
| 697 | |
| 698 | spin_lock(&vmap_area_lock); |
| 699 | va = __find_vmap_area(addr); |
| 700 | spin_unlock(&vmap_area_lock); |
| 701 | |
| 702 | return va; |
| 703 | } |
| 704 | |
| 705 | static void free_unmap_vmap_area_addr(unsigned long addr) |
| 706 | { |
| 707 | struct vmap_area *va; |
| 708 | |
| 709 | va = find_vmap_area(addr); |
| 710 | BUG_ON(!va); |
| 711 | free_unmap_vmap_area(va); |
| 712 | } |
| 713 | |
| 714 | |
| 715 | /*** Per cpu kva allocator ***/ |
| 716 | |
| 717 | /* |
| 718 | * vmap space is limited especially on 32 bit architectures. Ensure there is |
| 719 | * room for at least 16 percpu vmap blocks per CPU. |
| 720 | */ |
| 721 | /* |
| 722 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able |
| 723 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess |
| 724 | * instead (we just need a rough idea) |
| 725 | */ |
| 726 | #if BITS_PER_LONG == 32 |
| 727 | #define VMALLOC_SPACE (128UL*1024*1024) |
| 728 | #else |
| 729 | #define VMALLOC_SPACE (128UL*1024*1024*1024) |
| 730 | #endif |
| 731 | |
| 732 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) |
| 733 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ |
| 734 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ |
| 735 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) |
| 736 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ |
| 737 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ |
| 738 | #define VMAP_BBMAP_BITS \ |
| 739 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ |
| 740 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ |
| 741 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) |
| 742 | |
| 743 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) |
| 744 | |
| 745 | static bool vmap_initialized __read_mostly = false; |
| 746 | |
| 747 | struct vmap_block_queue { |
| 748 | spinlock_t lock; |
| 749 | struct list_head free; |
| 750 | }; |
| 751 | |
| 752 | struct vmap_block { |
| 753 | spinlock_t lock; |
| 754 | struct vmap_area *va; |
| 755 | struct vmap_block_queue *vbq; |
| 756 | unsigned long free, dirty; |
| 757 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); |
| 758 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); |
| 759 | struct list_head free_list; |
| 760 | struct rcu_head rcu_head; |
| 761 | struct list_head purge; |
| 762 | }; |
| 763 | |
| 764 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ |
| 765 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); |
| 766 | |
| 767 | /* |
| 768 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block |
| 769 | * in the free path. Could get rid of this if we change the API to return a |
| 770 | * "cookie" from alloc, to be passed to free. But no big deal yet. |
| 771 | */ |
| 772 | static DEFINE_SPINLOCK(vmap_block_tree_lock); |
| 773 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); |
| 774 | |
| 775 | /* |
| 776 | * We should probably have a fallback mechanism to allocate virtual memory |
| 777 | * out of partially filled vmap blocks. However vmap block sizing should be |
| 778 | * fairly reasonable according to the vmalloc size, so it shouldn't be a |
| 779 | * big problem. |
| 780 | */ |
| 781 | |
| 782 | static unsigned long addr_to_vb_idx(unsigned long addr) |
| 783 | { |
| 784 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); |
| 785 | addr /= VMAP_BLOCK_SIZE; |
| 786 | return addr; |
| 787 | } |
| 788 | |
| 789 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) |
| 790 | { |
| 791 | struct vmap_block_queue *vbq; |
| 792 | struct vmap_block *vb; |
| 793 | struct vmap_area *va; |
| 794 | unsigned long vb_idx; |
| 795 | int node, err; |
| 796 | |
| 797 | node = numa_node_id(); |
| 798 | |
| 799 | vb = kmalloc_node(sizeof(struct vmap_block), |
| 800 | gfp_mask & GFP_RECLAIM_MASK, node); |
| 801 | if (unlikely(!vb)) |
| 802 | return ERR_PTR(-ENOMEM); |
| 803 | |
| 804 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, |
| 805 | VMALLOC_START, VMALLOC_END, |
| 806 | node, gfp_mask); |
| 807 | if (IS_ERR(va)) { |
| 808 | kfree(vb); |
| 809 | return ERR_CAST(va); |
| 810 | } |
| 811 | |
| 812 | err = radix_tree_preload(gfp_mask); |
| 813 | if (unlikely(err)) { |
| 814 | kfree(vb); |
| 815 | free_vmap_area(va); |
| 816 | return ERR_PTR(err); |
| 817 | } |
| 818 | |
| 819 | spin_lock_init(&vb->lock); |
| 820 | vb->va = va; |
| 821 | vb->free = VMAP_BBMAP_BITS; |
| 822 | vb->dirty = 0; |
| 823 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); |
| 824 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); |
| 825 | INIT_LIST_HEAD(&vb->free_list); |
| 826 | |
| 827 | vb_idx = addr_to_vb_idx(va->va_start); |
| 828 | spin_lock(&vmap_block_tree_lock); |
| 829 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); |
| 830 | spin_unlock(&vmap_block_tree_lock); |
| 831 | BUG_ON(err); |
| 832 | radix_tree_preload_end(); |
| 833 | |
| 834 | vbq = &get_cpu_var(vmap_block_queue); |
| 835 | vb->vbq = vbq; |
| 836 | spin_lock(&vbq->lock); |
| 837 | list_add_rcu(&vb->free_list, &vbq->free); |
| 838 | spin_unlock(&vbq->lock); |
| 839 | put_cpu_var(vmap_block_queue); |
| 840 | |
| 841 | return vb; |
| 842 | } |
| 843 | |
| 844 | static void free_vmap_block(struct vmap_block *vb) |
| 845 | { |
| 846 | struct vmap_block *tmp; |
| 847 | unsigned long vb_idx; |
| 848 | |
| 849 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
| 850 | spin_lock(&vmap_block_tree_lock); |
| 851 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); |
| 852 | spin_unlock(&vmap_block_tree_lock); |
| 853 | BUG_ON(tmp != vb); |
| 854 | |
| 855 | free_vmap_area_noflush(vb->va); |
| 856 | kfree_rcu(vb, rcu_head); |
| 857 | } |
| 858 | |
| 859 | static void purge_fragmented_blocks(int cpu) |
| 860 | { |
| 861 | LIST_HEAD(purge); |
| 862 | struct vmap_block *vb; |
| 863 | struct vmap_block *n_vb; |
| 864 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); |
| 865 | |
| 866 | rcu_read_lock(); |
| 867 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
| 868 | |
| 869 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) |
| 870 | continue; |
| 871 | |
| 872 | spin_lock(&vb->lock); |
| 873 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { |
| 874 | vb->free = 0; /* prevent further allocs after releasing lock */ |
| 875 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ |
| 876 | bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS); |
| 877 | bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS); |
| 878 | spin_lock(&vbq->lock); |
| 879 | list_del_rcu(&vb->free_list); |
| 880 | spin_unlock(&vbq->lock); |
| 881 | spin_unlock(&vb->lock); |
| 882 | list_add_tail(&vb->purge, &purge); |
| 883 | } else |
| 884 | spin_unlock(&vb->lock); |
| 885 | } |
| 886 | rcu_read_unlock(); |
| 887 | |
| 888 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { |
| 889 | list_del(&vb->purge); |
| 890 | free_vmap_block(vb); |
| 891 | } |
| 892 | } |
| 893 | |
| 894 | static void purge_fragmented_blocks_thiscpu(void) |
| 895 | { |
| 896 | purge_fragmented_blocks(smp_processor_id()); |
| 897 | } |
| 898 | |
| 899 | static void purge_fragmented_blocks_allcpus(void) |
| 900 | { |
| 901 | int cpu; |
| 902 | |
| 903 | for_each_possible_cpu(cpu) |
| 904 | purge_fragmented_blocks(cpu); |
| 905 | } |
| 906 | |
| 907 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask) |
| 908 | { |
| 909 | struct vmap_block_queue *vbq; |
| 910 | struct vmap_block *vb; |
| 911 | unsigned long addr = 0; |
| 912 | unsigned int order; |
| 913 | int purge = 0; |
| 914 | |
| 915 | BUG_ON(size & ~PAGE_MASK); |
| 916 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
| 917 | if (WARN_ON(size == 0)) { |
| 918 | /* |
| 919 | * Allocating 0 bytes isn't what caller wants since |
| 920 | * get_order(0) returns funny result. Just warn and terminate |
| 921 | * early. |
| 922 | */ |
| 923 | return NULL; |
| 924 | } |
| 925 | order = get_order(size); |
| 926 | |
| 927 | again: |
| 928 | rcu_read_lock(); |
| 929 | vbq = &get_cpu_var(vmap_block_queue); |
| 930 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
| 931 | int i; |
| 932 | |
| 933 | spin_lock(&vb->lock); |
| 934 | if (vb->free < 1UL << order) |
| 935 | goto next; |
| 936 | |
| 937 | i = bitmap_find_free_region(vb->alloc_map, |
| 938 | VMAP_BBMAP_BITS, order); |
| 939 | |
| 940 | if (i < 0) { |
| 941 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS) { |
| 942 | /* fragmented and no outstanding allocations */ |
| 943 | BUG_ON(vb->dirty != VMAP_BBMAP_BITS); |
| 944 | purge = 1; |
| 945 | } |
| 946 | goto next; |
| 947 | } |
| 948 | addr = vb->va->va_start + (i << PAGE_SHIFT); |
| 949 | BUG_ON(addr_to_vb_idx(addr) != |
| 950 | addr_to_vb_idx(vb->va->va_start)); |
| 951 | vb->free -= 1UL << order; |
| 952 | if (vb->free == 0) { |
| 953 | spin_lock(&vbq->lock); |
| 954 | list_del_rcu(&vb->free_list); |
| 955 | spin_unlock(&vbq->lock); |
| 956 | } |
| 957 | spin_unlock(&vb->lock); |
| 958 | break; |
| 959 | next: |
| 960 | spin_unlock(&vb->lock); |
| 961 | } |
| 962 | |
| 963 | if (purge) |
| 964 | purge_fragmented_blocks_thiscpu(); |
| 965 | |
| 966 | put_cpu_var(vmap_block_queue); |
| 967 | rcu_read_unlock(); |
| 968 | |
| 969 | if (!addr) { |
| 970 | vb = new_vmap_block(gfp_mask); |
| 971 | if (IS_ERR(vb)) |
| 972 | return vb; |
| 973 | goto again; |
| 974 | } |
| 975 | |
| 976 | return (void *)addr; |
| 977 | } |
| 978 | |
| 979 | static void vb_free(const void *addr, unsigned long size) |
| 980 | { |
| 981 | unsigned long offset; |
| 982 | unsigned long vb_idx; |
| 983 | unsigned int order; |
| 984 | struct vmap_block *vb; |
| 985 | |
| 986 | BUG_ON(size & ~PAGE_MASK); |
| 987 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
| 988 | |
| 989 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); |
| 990 | |
| 991 | order = get_order(size); |
| 992 | |
| 993 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); |
| 994 | |
| 995 | vb_idx = addr_to_vb_idx((unsigned long)addr); |
| 996 | rcu_read_lock(); |
| 997 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); |
| 998 | rcu_read_unlock(); |
| 999 | BUG_ON(!vb); |
| 1000 | |
| 1001 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
| 1002 | |
| 1003 | spin_lock(&vb->lock); |
| 1004 | BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order)); |
| 1005 | |
| 1006 | vb->dirty += 1UL << order; |
| 1007 | if (vb->dirty == VMAP_BBMAP_BITS) { |
| 1008 | BUG_ON(vb->free); |
| 1009 | spin_unlock(&vb->lock); |
| 1010 | free_vmap_block(vb); |
| 1011 | } else |
| 1012 | spin_unlock(&vb->lock); |
| 1013 | } |
| 1014 | |
| 1015 | /** |
| 1016 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer |
| 1017 | * |
| 1018 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily |
| 1019 | * to amortize TLB flushing overheads. What this means is that any page you |
| 1020 | * have now, may, in a former life, have been mapped into kernel virtual |
| 1021 | * address by the vmap layer and so there might be some CPUs with TLB entries |
| 1022 | * still referencing that page (additional to the regular 1:1 kernel mapping). |
| 1023 | * |
| 1024 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can |
| 1025 | * be sure that none of the pages we have control over will have any aliases |
| 1026 | * from the vmap layer. |
| 1027 | */ |
| 1028 | void vm_unmap_aliases(void) |
| 1029 | { |
| 1030 | unsigned long start = ULONG_MAX, end = 0; |
| 1031 | int cpu; |
| 1032 | int flush = 0; |
| 1033 | |
| 1034 | if (unlikely(!vmap_initialized)) |
| 1035 | return; |
| 1036 | |
| 1037 | for_each_possible_cpu(cpu) { |
| 1038 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); |
| 1039 | struct vmap_block *vb; |
| 1040 | |
| 1041 | rcu_read_lock(); |
| 1042 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
| 1043 | int i; |
| 1044 | |
| 1045 | spin_lock(&vb->lock); |
| 1046 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); |
| 1047 | while (i < VMAP_BBMAP_BITS) { |
| 1048 | unsigned long s, e; |
| 1049 | int j; |
| 1050 | j = find_next_zero_bit(vb->dirty_map, |
| 1051 | VMAP_BBMAP_BITS, i); |
| 1052 | |
| 1053 | s = vb->va->va_start + (i << PAGE_SHIFT); |
| 1054 | e = vb->va->va_start + (j << PAGE_SHIFT); |
| 1055 | flush = 1; |
| 1056 | |
| 1057 | if (s < start) |
| 1058 | start = s; |
| 1059 | if (e > end) |
| 1060 | end = e; |
| 1061 | |
| 1062 | i = j; |
| 1063 | i = find_next_bit(vb->dirty_map, |
| 1064 | VMAP_BBMAP_BITS, i); |
| 1065 | } |
| 1066 | spin_unlock(&vb->lock); |
| 1067 | } |
| 1068 | rcu_read_unlock(); |
| 1069 | } |
| 1070 | |
| 1071 | __purge_vmap_area_lazy(&start, &end, 1, flush); |
| 1072 | } |
| 1073 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
| 1074 | |
| 1075 | /** |
| 1076 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram |
| 1077 | * @mem: the pointer returned by vm_map_ram |
| 1078 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) |
| 1079 | */ |
| 1080 | void vm_unmap_ram(const void *mem, unsigned int count) |
| 1081 | { |
| 1082 | unsigned long size = count << PAGE_SHIFT; |
| 1083 | unsigned long addr = (unsigned long)mem; |
| 1084 | |
| 1085 | BUG_ON(!addr); |
| 1086 | BUG_ON(addr < VMALLOC_START); |
| 1087 | BUG_ON(addr > VMALLOC_END); |
| 1088 | BUG_ON(addr & (PAGE_SIZE-1)); |
| 1089 | |
| 1090 | debug_check_no_locks_freed(mem, size); |
| 1091 | vmap_debug_free_range(addr, addr+size); |
| 1092 | |
| 1093 | if (likely(count <= VMAP_MAX_ALLOC)) |
| 1094 | vb_free(mem, size); |
| 1095 | else |
| 1096 | free_unmap_vmap_area_addr(addr); |
| 1097 | } |
| 1098 | EXPORT_SYMBOL(vm_unmap_ram); |
| 1099 | |
| 1100 | /** |
| 1101 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) |
| 1102 | * @pages: an array of pointers to the pages to be mapped |
| 1103 | * @count: number of pages |
| 1104 | * @node: prefer to allocate data structures on this node |
| 1105 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM |
| 1106 | * |
| 1107 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
| 1108 | */ |
| 1109 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
| 1110 | { |
| 1111 | unsigned long size = count << PAGE_SHIFT; |
| 1112 | unsigned long addr; |
| 1113 | void *mem; |
| 1114 | |
| 1115 | if (likely(count <= VMAP_MAX_ALLOC)) { |
| 1116 | mem = vb_alloc(size, GFP_KERNEL); |
| 1117 | if (IS_ERR(mem)) |
| 1118 | return NULL; |
| 1119 | addr = (unsigned long)mem; |
| 1120 | } else { |
| 1121 | struct vmap_area *va; |
| 1122 | va = alloc_vmap_area(size, PAGE_SIZE, |
| 1123 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); |
| 1124 | if (IS_ERR(va)) |
| 1125 | return NULL; |
| 1126 | |
| 1127 | addr = va->va_start; |
| 1128 | mem = (void *)addr; |
| 1129 | } |
| 1130 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { |
| 1131 | vm_unmap_ram(mem, count); |
| 1132 | return NULL; |
| 1133 | } |
| 1134 | return mem; |
| 1135 | } |
| 1136 | EXPORT_SYMBOL(vm_map_ram); |
| 1137 | |
| 1138 | static struct vm_struct *vmlist __initdata; |
| 1139 | /** |
| 1140 | * vm_area_add_early - add vmap area early during boot |
| 1141 | * @vm: vm_struct to add |
| 1142 | * |
| 1143 | * This function is used to add fixed kernel vm area to vmlist before |
| 1144 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags |
| 1145 | * should contain proper values and the other fields should be zero. |
| 1146 | * |
| 1147 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. |
| 1148 | */ |
| 1149 | void __init vm_area_add_early(struct vm_struct *vm) |
| 1150 | { |
| 1151 | struct vm_struct *tmp, **p; |
| 1152 | |
| 1153 | BUG_ON(vmap_initialized); |
| 1154 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { |
| 1155 | if (tmp->addr >= vm->addr) { |
| 1156 | BUG_ON(tmp->addr < vm->addr + vm->size); |
| 1157 | break; |
| 1158 | } else |
| 1159 | BUG_ON(tmp->addr + tmp->size > vm->addr); |
| 1160 | } |
| 1161 | vm->next = *p; |
| 1162 | *p = vm; |
| 1163 | } |
| 1164 | |
| 1165 | /** |
| 1166 | * vm_area_register_early - register vmap area early during boot |
| 1167 | * @vm: vm_struct to register |
| 1168 | * @align: requested alignment |
| 1169 | * |
| 1170 | * This function is used to register kernel vm area before |
| 1171 | * vmalloc_init() is called. @vm->size and @vm->flags should contain |
| 1172 | * proper values on entry and other fields should be zero. On return, |
| 1173 | * vm->addr contains the allocated address. |
| 1174 | * |
| 1175 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. |
| 1176 | */ |
| 1177 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
| 1178 | { |
| 1179 | static size_t vm_init_off __initdata; |
| 1180 | unsigned long addr; |
| 1181 | |
| 1182 | addr = ALIGN(VMALLOC_START + vm_init_off, align); |
| 1183 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; |
| 1184 | |
| 1185 | vm->addr = (void *)addr; |
| 1186 | |
| 1187 | vm_area_add_early(vm); |
| 1188 | } |
| 1189 | |
| 1190 | void __init vmalloc_init(void) |
| 1191 | { |
| 1192 | struct vmap_area *va; |
| 1193 | struct vm_struct *tmp; |
| 1194 | int i; |
| 1195 | |
| 1196 | for_each_possible_cpu(i) { |
| 1197 | struct vmap_block_queue *vbq; |
| 1198 | struct vfree_deferred *p; |
| 1199 | |
| 1200 | vbq = &per_cpu(vmap_block_queue, i); |
| 1201 | spin_lock_init(&vbq->lock); |
| 1202 | INIT_LIST_HEAD(&vbq->free); |
| 1203 | p = &per_cpu(vfree_deferred, i); |
| 1204 | init_llist_head(&p->list); |
| 1205 | INIT_WORK(&p->wq, free_work); |
| 1206 | } |
| 1207 | |
| 1208 | /* Import existing vmlist entries. */ |
| 1209 | for (tmp = vmlist; tmp; tmp = tmp->next) { |
| 1210 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
| 1211 | va->flags = VM_VM_AREA; |
| 1212 | va->va_start = (unsigned long)tmp->addr; |
| 1213 | va->va_end = va->va_start + tmp->size; |
| 1214 | va->vm = tmp; |
| 1215 | __insert_vmap_area(va); |
| 1216 | } |
| 1217 | |
| 1218 | vmap_area_pcpu_hole = VMALLOC_END; |
| 1219 | |
| 1220 | vmap_initialized = true; |
| 1221 | } |
| 1222 | |
| 1223 | /** |
| 1224 | * map_kernel_range_noflush - map kernel VM area with the specified pages |
| 1225 | * @addr: start of the VM area to map |
| 1226 | * @size: size of the VM area to map |
| 1227 | * @prot: page protection flags to use |
| 1228 | * @pages: pages to map |
| 1229 | * |
| 1230 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size |
| 1231 | * specify should have been allocated using get_vm_area() and its |
| 1232 | * friends. |
| 1233 | * |
| 1234 | * NOTE: |
| 1235 | * This function does NOT do any cache flushing. The caller is |
| 1236 | * responsible for calling flush_cache_vmap() on to-be-mapped areas |
| 1237 | * before calling this function. |
| 1238 | * |
| 1239 | * RETURNS: |
| 1240 | * The number of pages mapped on success, -errno on failure. |
| 1241 | */ |
| 1242 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, |
| 1243 | pgprot_t prot, struct page **pages) |
| 1244 | { |
| 1245 | return vmap_page_range_noflush(addr, addr + size, prot, pages); |
| 1246 | } |
| 1247 | |
| 1248 | /** |
| 1249 | * unmap_kernel_range_noflush - unmap kernel VM area |
| 1250 | * @addr: start of the VM area to unmap |
| 1251 | * @size: size of the VM area to unmap |
| 1252 | * |
| 1253 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size |
| 1254 | * specify should have been allocated using get_vm_area() and its |
| 1255 | * friends. |
| 1256 | * |
| 1257 | * NOTE: |
| 1258 | * This function does NOT do any cache flushing. The caller is |
| 1259 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas |
| 1260 | * before calling this function and flush_tlb_kernel_range() after. |
| 1261 | */ |
| 1262 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) |
| 1263 | { |
| 1264 | vunmap_page_range(addr, addr + size); |
| 1265 | } |
| 1266 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
| 1267 | |
| 1268 | /** |
| 1269 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB |
| 1270 | * @addr: start of the VM area to unmap |
| 1271 | * @size: size of the VM area to unmap |
| 1272 | * |
| 1273 | * Similar to unmap_kernel_range_noflush() but flushes vcache before |
| 1274 | * the unmapping and tlb after. |
| 1275 | */ |
| 1276 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
| 1277 | { |
| 1278 | unsigned long end = addr + size; |
| 1279 | |
| 1280 | flush_cache_vunmap(addr, end); |
| 1281 | vunmap_page_range(addr, end); |
| 1282 | flush_tlb_kernel_range(addr, end); |
| 1283 | } |
| 1284 | |
| 1285 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages) |
| 1286 | { |
| 1287 | unsigned long addr = (unsigned long)area->addr; |
| 1288 | unsigned long end = addr + area->size - PAGE_SIZE; |
| 1289 | int err; |
| 1290 | |
| 1291 | err = vmap_page_range(addr, end, prot, pages); |
| 1292 | |
| 1293 | return err > 0 ? 0 : err; |
| 1294 | } |
| 1295 | EXPORT_SYMBOL_GPL(map_vm_area); |
| 1296 | |
| 1297 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
| 1298 | unsigned long flags, const void *caller) |
| 1299 | { |
| 1300 | spin_lock(&vmap_area_lock); |
| 1301 | vm->flags = flags; |
| 1302 | vm->addr = (void *)va->va_start; |
| 1303 | vm->size = va->va_end - va->va_start; |
| 1304 | vm->caller = caller; |
| 1305 | va->vm = vm; |
| 1306 | va->flags |= VM_VM_AREA; |
| 1307 | spin_unlock(&vmap_area_lock); |
| 1308 | } |
| 1309 | |
| 1310 | static void clear_vm_unlist(struct vm_struct *vm) |
| 1311 | { |
| 1312 | /* |
| 1313 | * Before removing VM_UNLIST, |
| 1314 | * we should make sure that vm has proper values. |
| 1315 | * Pair with smp_rmb() in show_numa_info(). |
| 1316 | */ |
| 1317 | smp_wmb(); |
| 1318 | vm->flags &= ~VM_UNLIST; |
| 1319 | } |
| 1320 | |
| 1321 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
| 1322 | unsigned long flags, const void *caller) |
| 1323 | { |
| 1324 | setup_vmalloc_vm(vm, va, flags, caller); |
| 1325 | clear_vm_unlist(vm); |
| 1326 | } |
| 1327 | |
| 1328 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
| 1329 | unsigned long align, unsigned long flags, unsigned long start, |
| 1330 | unsigned long end, int node, gfp_t gfp_mask, const void *caller) |
| 1331 | { |
| 1332 | struct vmap_area *va; |
| 1333 | struct vm_struct *area; |
| 1334 | |
| 1335 | BUG_ON(in_interrupt()); |
| 1336 | if (flags & VM_IOREMAP) { |
| 1337 | int bit = fls(size); |
| 1338 | |
| 1339 | if (bit > IOREMAP_MAX_ORDER) |
| 1340 | bit = IOREMAP_MAX_ORDER; |
| 1341 | else if (bit < PAGE_SHIFT) |
| 1342 | bit = PAGE_SHIFT; |
| 1343 | |
| 1344 | align = 1ul << bit; |
| 1345 | } |
| 1346 | |
| 1347 | size = PAGE_ALIGN(size); |
| 1348 | if (unlikely(!size)) |
| 1349 | return NULL; |
| 1350 | |
| 1351 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
| 1352 | if (unlikely(!area)) |
| 1353 | return NULL; |
| 1354 | |
| 1355 | /* |
| 1356 | * We always allocate a guard page. |
| 1357 | */ |
| 1358 | size += PAGE_SIZE; |
| 1359 | |
| 1360 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
| 1361 | if (IS_ERR(va)) { |
| 1362 | kfree(area); |
| 1363 | return NULL; |
| 1364 | } |
| 1365 | |
| 1366 | /* |
| 1367 | * When this function is called from __vmalloc_node_range, |
| 1368 | * we add VM_UNLIST flag to avoid accessing uninitialized |
| 1369 | * members of vm_struct such as pages and nr_pages fields. |
| 1370 | * They will be set later. |
| 1371 | */ |
| 1372 | if (flags & VM_UNLIST) |
| 1373 | setup_vmalloc_vm(area, va, flags, caller); |
| 1374 | else |
| 1375 | insert_vmalloc_vm(area, va, flags, caller); |
| 1376 | |
| 1377 | return area; |
| 1378 | } |
| 1379 | |
| 1380 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
| 1381 | unsigned long start, unsigned long end) |
| 1382 | { |
| 1383 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
| 1384 | GFP_KERNEL, __builtin_return_address(0)); |
| 1385 | } |
| 1386 | EXPORT_SYMBOL_GPL(__get_vm_area); |
| 1387 | |
| 1388 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
| 1389 | unsigned long start, unsigned long end, |
| 1390 | const void *caller) |
| 1391 | { |
| 1392 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE, |
| 1393 | GFP_KERNEL, caller); |
| 1394 | } |
| 1395 | |
| 1396 | /** |
| 1397 | * get_vm_area - reserve a contiguous kernel virtual area |
| 1398 | * @size: size of the area |
| 1399 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC |
| 1400 | * |
| 1401 | * Search an area of @size in the kernel virtual mapping area, |
| 1402 | * and reserved it for out purposes. Returns the area descriptor |
| 1403 | * on success or %NULL on failure. |
| 1404 | */ |
| 1405 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) |
| 1406 | { |
| 1407 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
| 1408 | NUMA_NO_NODE, GFP_KERNEL, |
| 1409 | __builtin_return_address(0)); |
| 1410 | } |
| 1411 | |
| 1412 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, |
| 1413 | const void *caller) |
| 1414 | { |
| 1415 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
| 1416 | NUMA_NO_NODE, GFP_KERNEL, caller); |
| 1417 | } |
| 1418 | |
| 1419 | /** |
| 1420 | * find_vm_area - find a continuous kernel virtual area |
| 1421 | * @addr: base address |
| 1422 | * |
| 1423 | * Search for the kernel VM area starting at @addr, and return it. |
| 1424 | * It is up to the caller to do all required locking to keep the returned |
| 1425 | * pointer valid. |
| 1426 | */ |
| 1427 | struct vm_struct *find_vm_area(const void *addr) |
| 1428 | { |
| 1429 | struct vmap_area *va; |
| 1430 | |
| 1431 | va = find_vmap_area((unsigned long)addr); |
| 1432 | if (va && va->flags & VM_VM_AREA) |
| 1433 | return va->vm; |
| 1434 | |
| 1435 | return NULL; |
| 1436 | } |
| 1437 | |
| 1438 | /** |
| 1439 | * remove_vm_area - find and remove a continuous kernel virtual area |
| 1440 | * @addr: base address |
| 1441 | * |
| 1442 | * Search for the kernel VM area starting at @addr, and remove it. |
| 1443 | * This function returns the found VM area, but using it is NOT safe |
| 1444 | * on SMP machines, except for its size or flags. |
| 1445 | */ |
| 1446 | struct vm_struct *remove_vm_area(const void *addr) |
| 1447 | { |
| 1448 | struct vmap_area *va; |
| 1449 | |
| 1450 | va = find_vmap_area((unsigned long)addr); |
| 1451 | if (va && va->flags & VM_VM_AREA) { |
| 1452 | struct vm_struct *vm = va->vm; |
| 1453 | |
| 1454 | spin_lock(&vmap_area_lock); |
| 1455 | va->vm = NULL; |
| 1456 | va->flags &= ~VM_VM_AREA; |
| 1457 | spin_unlock(&vmap_area_lock); |
| 1458 | |
| 1459 | vmap_debug_free_range(va->va_start, va->va_end); |
| 1460 | free_unmap_vmap_area(va); |
| 1461 | vm->size -= PAGE_SIZE; |
| 1462 | |
| 1463 | return vm; |
| 1464 | } |
| 1465 | return NULL; |
| 1466 | } |
| 1467 | |
| 1468 | static void __vunmap(const void *addr, int deallocate_pages) |
| 1469 | { |
| 1470 | struct vm_struct *area; |
| 1471 | |
| 1472 | if (!addr) |
| 1473 | return; |
| 1474 | |
| 1475 | if ((PAGE_SIZE-1) & (unsigned long)addr) { |
| 1476 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
| 1477 | return; |
| 1478 | } |
| 1479 | |
| 1480 | area = remove_vm_area(addr); |
| 1481 | if (unlikely(!area)) { |
| 1482 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
| 1483 | addr); |
| 1484 | return; |
| 1485 | } |
| 1486 | |
| 1487 | debug_check_no_locks_freed(addr, area->size); |
| 1488 | debug_check_no_obj_freed(addr, area->size); |
| 1489 | |
| 1490 | if (deallocate_pages) { |
| 1491 | int i; |
| 1492 | |
| 1493 | for (i = 0; i < area->nr_pages; i++) { |
| 1494 | struct page *page = area->pages[i]; |
| 1495 | |
| 1496 | BUG_ON(!page); |
| 1497 | __free_page(page); |
| 1498 | } |
| 1499 | |
| 1500 | if (area->flags & VM_VPAGES) |
| 1501 | vfree(area->pages); |
| 1502 | else |
| 1503 | kfree(area->pages); |
| 1504 | } |
| 1505 | |
| 1506 | kfree(area); |
| 1507 | return; |
| 1508 | } |
| 1509 | |
| 1510 | /** |
| 1511 | * vfree - release memory allocated by vmalloc() |
| 1512 | * @addr: memory base address |
| 1513 | * |
| 1514 | * Free the virtually continuous memory area starting at @addr, as |
| 1515 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
| 1516 | * NULL, no operation is performed. |
| 1517 | * |
| 1518 | * Must not be called in NMI context (strictly speaking, only if we don't |
| 1519 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling |
| 1520 | * conventions for vfree() arch-depenedent would be a really bad idea) |
| 1521 | * |
| 1522 | * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node) |
| 1523 | * |
| 1524 | */ |
| 1525 | void vfree(const void *addr) |
| 1526 | { |
| 1527 | BUG_ON(in_nmi()); |
| 1528 | |
| 1529 | kmemleak_free(addr); |
| 1530 | |
| 1531 | if (!addr) |
| 1532 | return; |
| 1533 | if (unlikely(in_interrupt())) { |
| 1534 | struct vfree_deferred *p = &__get_cpu_var(vfree_deferred); |
| 1535 | llist_add((struct llist_node *)addr, &p->list); |
| 1536 | schedule_work(&p->wq); |
| 1537 | } else |
| 1538 | __vunmap(addr, 1); |
| 1539 | } |
| 1540 | EXPORT_SYMBOL(vfree); |
| 1541 | |
| 1542 | /** |
| 1543 | * vunmap - release virtual mapping obtained by vmap() |
| 1544 | * @addr: memory base address |
| 1545 | * |
| 1546 | * Free the virtually contiguous memory area starting at @addr, |
| 1547 | * which was created from the page array passed to vmap(). |
| 1548 | * |
| 1549 | * Must not be called in interrupt context. |
| 1550 | */ |
| 1551 | void vunmap(const void *addr) |
| 1552 | { |
| 1553 | BUG_ON(in_interrupt()); |
| 1554 | might_sleep(); |
| 1555 | if (addr) |
| 1556 | __vunmap(addr, 0); |
| 1557 | } |
| 1558 | EXPORT_SYMBOL(vunmap); |
| 1559 | |
| 1560 | /** |
| 1561 | * vmap - map an array of pages into virtually contiguous space |
| 1562 | * @pages: array of page pointers |
| 1563 | * @count: number of pages to map |
| 1564 | * @flags: vm_area->flags |
| 1565 | * @prot: page protection for the mapping |
| 1566 | * |
| 1567 | * Maps @count pages from @pages into contiguous kernel virtual |
| 1568 | * space. |
| 1569 | */ |
| 1570 | void *vmap(struct page **pages, unsigned int count, |
| 1571 | unsigned long flags, pgprot_t prot) |
| 1572 | { |
| 1573 | struct vm_struct *area; |
| 1574 | |
| 1575 | might_sleep(); |
| 1576 | |
| 1577 | if (count > totalram_pages) |
| 1578 | return NULL; |
| 1579 | |
| 1580 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
| 1581 | __builtin_return_address(0)); |
| 1582 | if (!area) |
| 1583 | return NULL; |
| 1584 | |
| 1585 | if (map_vm_area(area, prot, pages)) { |
| 1586 | vunmap(area->addr); |
| 1587 | return NULL; |
| 1588 | } |
| 1589 | |
| 1590 | return area->addr; |
| 1591 | } |
| 1592 | EXPORT_SYMBOL(vmap); |
| 1593 | |
| 1594 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
| 1595 | gfp_t gfp_mask, pgprot_t prot, |
| 1596 | int node, const void *caller); |
| 1597 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
| 1598 | pgprot_t prot, int node, const void *caller) |
| 1599 | { |
| 1600 | const int order = 0; |
| 1601 | struct page **pages; |
| 1602 | unsigned int nr_pages, array_size, i; |
| 1603 | gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
| 1604 | |
| 1605 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; |
| 1606 | array_size = (nr_pages * sizeof(struct page *)); |
| 1607 | |
| 1608 | area->nr_pages = nr_pages; |
| 1609 | /* Please note that the recursion is strictly bounded. */ |
| 1610 | if (array_size > PAGE_SIZE) { |
| 1611 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
| 1612 | PAGE_KERNEL, node, caller); |
| 1613 | area->flags |= VM_VPAGES; |
| 1614 | } else { |
| 1615 | pages = kmalloc_node(array_size, nested_gfp, node); |
| 1616 | } |
| 1617 | area->pages = pages; |
| 1618 | area->caller = caller; |
| 1619 | if (!area->pages) { |
| 1620 | remove_vm_area(area->addr); |
| 1621 | kfree(area); |
| 1622 | return NULL; |
| 1623 | } |
| 1624 | |
| 1625 | for (i = 0; i < area->nr_pages; i++) { |
| 1626 | struct page *page; |
| 1627 | gfp_t tmp_mask = gfp_mask | __GFP_NOWARN; |
| 1628 | |
| 1629 | if (node < 0) |
| 1630 | page = alloc_page(tmp_mask); |
| 1631 | else |
| 1632 | page = alloc_pages_node(node, tmp_mask, order); |
| 1633 | |
| 1634 | if (unlikely(!page)) { |
| 1635 | /* Successfully allocated i pages, free them in __vunmap() */ |
| 1636 | area->nr_pages = i; |
| 1637 | goto fail; |
| 1638 | } |
| 1639 | area->pages[i] = page; |
| 1640 | } |
| 1641 | |
| 1642 | if (map_vm_area(area, prot, pages)) |
| 1643 | goto fail; |
| 1644 | return area->addr; |
| 1645 | |
| 1646 | fail: |
| 1647 | warn_alloc_failed(gfp_mask, order, |
| 1648 | "vmalloc: allocation failure, allocated %ld of %ld bytes\n", |
| 1649 | (area->nr_pages*PAGE_SIZE), area->size); |
| 1650 | vfree(area->addr); |
| 1651 | return NULL; |
| 1652 | } |
| 1653 | |
| 1654 | /** |
| 1655 | * __vmalloc_node_range - allocate virtually contiguous memory |
| 1656 | * @size: allocation size |
| 1657 | * @align: desired alignment |
| 1658 | * @start: vm area range start |
| 1659 | * @end: vm area range end |
| 1660 | * @gfp_mask: flags for the page level allocator |
| 1661 | * @prot: protection mask for the allocated pages |
| 1662 | * @node: node to use for allocation or NUMA_NO_NODE |
| 1663 | * @caller: caller's return address |
| 1664 | * |
| 1665 | * Allocate enough pages to cover @size from the page level |
| 1666 | * allocator with @gfp_mask flags. Map them into contiguous |
| 1667 | * kernel virtual space, using a pagetable protection of @prot. |
| 1668 | */ |
| 1669 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
| 1670 | unsigned long start, unsigned long end, gfp_t gfp_mask, |
| 1671 | pgprot_t prot, int node, const void *caller) |
| 1672 | { |
| 1673 | struct vm_struct *area; |
| 1674 | void *addr; |
| 1675 | unsigned long real_size = size; |
| 1676 | |
| 1677 | size = PAGE_ALIGN(size); |
| 1678 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
| 1679 | goto fail; |
| 1680 | |
| 1681 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST, |
| 1682 | start, end, node, gfp_mask, caller); |
| 1683 | if (!area) |
| 1684 | goto fail; |
| 1685 | |
| 1686 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
| 1687 | if (!addr) |
| 1688 | return NULL; |
| 1689 | |
| 1690 | /* |
| 1691 | * In this function, newly allocated vm_struct has VM_UNLIST flag. |
| 1692 | * It means that vm_struct is not fully initialized. |
| 1693 | * Now, it is fully initialized, so remove this flag here. |
| 1694 | */ |
| 1695 | clear_vm_unlist(area); |
| 1696 | |
| 1697 | /* |
| 1698 | * A ref_count = 3 is needed because the vm_struct and vmap_area |
| 1699 | * structures allocated in the __get_vm_area_node() function contain |
| 1700 | * references to the virtual address of the vmalloc'ed block. |
| 1701 | */ |
| 1702 | kmemleak_alloc(addr, real_size, 3, gfp_mask); |
| 1703 | |
| 1704 | return addr; |
| 1705 | |
| 1706 | fail: |
| 1707 | warn_alloc_failed(gfp_mask, 0, |
| 1708 | "vmalloc: allocation failure: %lu bytes\n", |
| 1709 | real_size); |
| 1710 | return NULL; |
| 1711 | } |
| 1712 | |
| 1713 | /** |
| 1714 | * __vmalloc_node - allocate virtually contiguous memory |
| 1715 | * @size: allocation size |
| 1716 | * @align: desired alignment |
| 1717 | * @gfp_mask: flags for the page level allocator |
| 1718 | * @prot: protection mask for the allocated pages |
| 1719 | * @node: node to use for allocation or NUMA_NO_NODE |
| 1720 | * @caller: caller's return address |
| 1721 | * |
| 1722 | * Allocate enough pages to cover @size from the page level |
| 1723 | * allocator with @gfp_mask flags. Map them into contiguous |
| 1724 | * kernel virtual space, using a pagetable protection of @prot. |
| 1725 | */ |
| 1726 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
| 1727 | gfp_t gfp_mask, pgprot_t prot, |
| 1728 | int node, const void *caller) |
| 1729 | { |
| 1730 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, |
| 1731 | gfp_mask, prot, node, caller); |
| 1732 | } |
| 1733 | |
| 1734 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
| 1735 | { |
| 1736 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE, |
| 1737 | __builtin_return_address(0)); |
| 1738 | } |
| 1739 | EXPORT_SYMBOL(__vmalloc); |
| 1740 | |
| 1741 | static inline void *__vmalloc_node_flags(unsigned long size, |
| 1742 | int node, gfp_t flags) |
| 1743 | { |
| 1744 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, |
| 1745 | node, __builtin_return_address(0)); |
| 1746 | } |
| 1747 | |
| 1748 | /** |
| 1749 | * vmalloc - allocate virtually contiguous memory |
| 1750 | * @size: allocation size |
| 1751 | * Allocate enough pages to cover @size from the page level |
| 1752 | * allocator and map them into contiguous kernel virtual space. |
| 1753 | * |
| 1754 | * For tight control over page level allocator and protection flags |
| 1755 | * use __vmalloc() instead. |
| 1756 | */ |
| 1757 | void *vmalloc(unsigned long size) |
| 1758 | { |
| 1759 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
| 1760 | GFP_KERNEL | __GFP_HIGHMEM); |
| 1761 | } |
| 1762 | EXPORT_SYMBOL(vmalloc); |
| 1763 | |
| 1764 | /** |
| 1765 | * vzalloc - allocate virtually contiguous memory with zero fill |
| 1766 | * @size: allocation size |
| 1767 | * Allocate enough pages to cover @size from the page level |
| 1768 | * allocator and map them into contiguous kernel virtual space. |
| 1769 | * The memory allocated is set to zero. |
| 1770 | * |
| 1771 | * For tight control over page level allocator and protection flags |
| 1772 | * use __vmalloc() instead. |
| 1773 | */ |
| 1774 | void *vzalloc(unsigned long size) |
| 1775 | { |
| 1776 | return __vmalloc_node_flags(size, NUMA_NO_NODE, |
| 1777 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
| 1778 | } |
| 1779 | EXPORT_SYMBOL(vzalloc); |
| 1780 | |
| 1781 | /** |
| 1782 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
| 1783 | * @size: allocation size |
| 1784 | * |
| 1785 | * The resulting memory area is zeroed so it can be mapped to userspace |
| 1786 | * without leaking data. |
| 1787 | */ |
| 1788 | void *vmalloc_user(unsigned long size) |
| 1789 | { |
| 1790 | struct vm_struct *area; |
| 1791 | void *ret; |
| 1792 | |
| 1793 | ret = __vmalloc_node(size, SHMLBA, |
| 1794 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
| 1795 | PAGE_KERNEL, NUMA_NO_NODE, |
| 1796 | __builtin_return_address(0)); |
| 1797 | if (ret) { |
| 1798 | area = find_vm_area(ret); |
| 1799 | area->flags |= VM_USERMAP; |
| 1800 | } |
| 1801 | return ret; |
| 1802 | } |
| 1803 | EXPORT_SYMBOL(vmalloc_user); |
| 1804 | |
| 1805 | /** |
| 1806 | * vmalloc_node - allocate memory on a specific node |
| 1807 | * @size: allocation size |
| 1808 | * @node: numa node |
| 1809 | * |
| 1810 | * Allocate enough pages to cover @size from the page level |
| 1811 | * allocator and map them into contiguous kernel virtual space. |
| 1812 | * |
| 1813 | * For tight control over page level allocator and protection flags |
| 1814 | * use __vmalloc() instead. |
| 1815 | */ |
| 1816 | void *vmalloc_node(unsigned long size, int node) |
| 1817 | { |
| 1818 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
| 1819 | node, __builtin_return_address(0)); |
| 1820 | } |
| 1821 | EXPORT_SYMBOL(vmalloc_node); |
| 1822 | |
| 1823 | /** |
| 1824 | * vzalloc_node - allocate memory on a specific node with zero fill |
| 1825 | * @size: allocation size |
| 1826 | * @node: numa node |
| 1827 | * |
| 1828 | * Allocate enough pages to cover @size from the page level |
| 1829 | * allocator and map them into contiguous kernel virtual space. |
| 1830 | * The memory allocated is set to zero. |
| 1831 | * |
| 1832 | * For tight control over page level allocator and protection flags |
| 1833 | * use __vmalloc_node() instead. |
| 1834 | */ |
| 1835 | void *vzalloc_node(unsigned long size, int node) |
| 1836 | { |
| 1837 | return __vmalloc_node_flags(size, node, |
| 1838 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
| 1839 | } |
| 1840 | EXPORT_SYMBOL(vzalloc_node); |
| 1841 | |
| 1842 | #ifndef PAGE_KERNEL_EXEC |
| 1843 | # define PAGE_KERNEL_EXEC PAGE_KERNEL |
| 1844 | #endif |
| 1845 | |
| 1846 | /** |
| 1847 | * vmalloc_exec - allocate virtually contiguous, executable memory |
| 1848 | * @size: allocation size |
| 1849 | * |
| 1850 | * Kernel-internal function to allocate enough pages to cover @size |
| 1851 | * the page level allocator and map them into contiguous and |
| 1852 | * executable kernel virtual space. |
| 1853 | * |
| 1854 | * For tight control over page level allocator and protection flags |
| 1855 | * use __vmalloc() instead. |
| 1856 | */ |
| 1857 | |
| 1858 | void *vmalloc_exec(unsigned long size) |
| 1859 | { |
| 1860 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
| 1861 | NUMA_NO_NODE, __builtin_return_address(0)); |
| 1862 | } |
| 1863 | |
| 1864 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
| 1865 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
| 1866 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
| 1867 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
| 1868 | #else |
| 1869 | #define GFP_VMALLOC32 GFP_KERNEL |
| 1870 | #endif |
| 1871 | |
| 1872 | /** |
| 1873 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
| 1874 | * @size: allocation size |
| 1875 | * |
| 1876 | * Allocate enough 32bit PA addressable pages to cover @size from the |
| 1877 | * page level allocator and map them into contiguous kernel virtual space. |
| 1878 | */ |
| 1879 | void *vmalloc_32(unsigned long size) |
| 1880 | { |
| 1881 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
| 1882 | NUMA_NO_NODE, __builtin_return_address(0)); |
| 1883 | } |
| 1884 | EXPORT_SYMBOL(vmalloc_32); |
| 1885 | |
| 1886 | /** |
| 1887 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
| 1888 | * @size: allocation size |
| 1889 | * |
| 1890 | * The resulting memory area is 32bit addressable and zeroed so it can be |
| 1891 | * mapped to userspace without leaking data. |
| 1892 | */ |
| 1893 | void *vmalloc_32_user(unsigned long size) |
| 1894 | { |
| 1895 | struct vm_struct *area; |
| 1896 | void *ret; |
| 1897 | |
| 1898 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
| 1899 | NUMA_NO_NODE, __builtin_return_address(0)); |
| 1900 | if (ret) { |
| 1901 | area = find_vm_area(ret); |
| 1902 | area->flags |= VM_USERMAP; |
| 1903 | } |
| 1904 | return ret; |
| 1905 | } |
| 1906 | EXPORT_SYMBOL(vmalloc_32_user); |
| 1907 | |
| 1908 | /* |
| 1909 | * small helper routine , copy contents to buf from addr. |
| 1910 | * If the page is not present, fill zero. |
| 1911 | */ |
| 1912 | |
| 1913 | static int aligned_vread(char *buf, char *addr, unsigned long count) |
| 1914 | { |
| 1915 | struct page *p; |
| 1916 | int copied = 0; |
| 1917 | |
| 1918 | while (count) { |
| 1919 | unsigned long offset, length; |
| 1920 | |
| 1921 | offset = (unsigned long)addr & ~PAGE_MASK; |
| 1922 | length = PAGE_SIZE - offset; |
| 1923 | if (length > count) |
| 1924 | length = count; |
| 1925 | p = vmalloc_to_page(addr); |
| 1926 | /* |
| 1927 | * To do safe access to this _mapped_ area, we need |
| 1928 | * lock. But adding lock here means that we need to add |
| 1929 | * overhead of vmalloc()/vfree() calles for this _debug_ |
| 1930 | * interface, rarely used. Instead of that, we'll use |
| 1931 | * kmap() and get small overhead in this access function. |
| 1932 | */ |
| 1933 | if (p) { |
| 1934 | /* |
| 1935 | * we can expect USER0 is not used (see vread/vwrite's |
| 1936 | * function description) |
| 1937 | */ |
| 1938 | void *map = kmap_atomic(p); |
| 1939 | memcpy(buf, map + offset, length); |
| 1940 | kunmap_atomic(map); |
| 1941 | } else |
| 1942 | memset(buf, 0, length); |
| 1943 | |
| 1944 | addr += length; |
| 1945 | buf += length; |
| 1946 | copied += length; |
| 1947 | count -= length; |
| 1948 | } |
| 1949 | return copied; |
| 1950 | } |
| 1951 | |
| 1952 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) |
| 1953 | { |
| 1954 | struct page *p; |
| 1955 | int copied = 0; |
| 1956 | |
| 1957 | while (count) { |
| 1958 | unsigned long offset, length; |
| 1959 | |
| 1960 | offset = (unsigned long)addr & ~PAGE_MASK; |
| 1961 | length = PAGE_SIZE - offset; |
| 1962 | if (length > count) |
| 1963 | length = count; |
| 1964 | p = vmalloc_to_page(addr); |
| 1965 | /* |
| 1966 | * To do safe access to this _mapped_ area, we need |
| 1967 | * lock. But adding lock here means that we need to add |
| 1968 | * overhead of vmalloc()/vfree() calles for this _debug_ |
| 1969 | * interface, rarely used. Instead of that, we'll use |
| 1970 | * kmap() and get small overhead in this access function. |
| 1971 | */ |
| 1972 | if (p) { |
| 1973 | /* |
| 1974 | * we can expect USER0 is not used (see vread/vwrite's |
| 1975 | * function description) |
| 1976 | */ |
| 1977 | void *map = kmap_atomic(p); |
| 1978 | memcpy(map + offset, buf, length); |
| 1979 | kunmap_atomic(map); |
| 1980 | } |
| 1981 | addr += length; |
| 1982 | buf += length; |
| 1983 | copied += length; |
| 1984 | count -= length; |
| 1985 | } |
| 1986 | return copied; |
| 1987 | } |
| 1988 | |
| 1989 | /** |
| 1990 | * vread() - read vmalloc area in a safe way. |
| 1991 | * @buf: buffer for reading data |
| 1992 | * @addr: vm address. |
| 1993 | * @count: number of bytes to be read. |
| 1994 | * |
| 1995 | * Returns # of bytes which addr and buf should be increased. |
| 1996 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't |
| 1997 | * includes any intersect with alive vmalloc area. |
| 1998 | * |
| 1999 | * This function checks that addr is a valid vmalloc'ed area, and |
| 2000 | * copy data from that area to a given buffer. If the given memory range |
| 2001 | * of [addr...addr+count) includes some valid address, data is copied to |
| 2002 | * proper area of @buf. If there are memory holes, they'll be zero-filled. |
| 2003 | * IOREMAP area is treated as memory hole and no copy is done. |
| 2004 | * |
| 2005 | * If [addr...addr+count) doesn't includes any intersects with alive |
| 2006 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
| 2007 | * |
| 2008 | * Note: In usual ops, vread() is never necessary because the caller |
| 2009 | * should know vmalloc() area is valid and can use memcpy(). |
| 2010 | * This is for routines which have to access vmalloc area without |
| 2011 | * any informaion, as /dev/kmem. |
| 2012 | * |
| 2013 | */ |
| 2014 | |
| 2015 | long vread(char *buf, char *addr, unsigned long count) |
| 2016 | { |
| 2017 | struct vmap_area *va; |
| 2018 | struct vm_struct *vm; |
| 2019 | char *vaddr, *buf_start = buf; |
| 2020 | unsigned long buflen = count; |
| 2021 | unsigned long n; |
| 2022 | |
| 2023 | /* Don't allow overflow */ |
| 2024 | if ((unsigned long) addr + count < count) |
| 2025 | count = -(unsigned long) addr; |
| 2026 | |
| 2027 | spin_lock(&vmap_area_lock); |
| 2028 | list_for_each_entry(va, &vmap_area_list, list) { |
| 2029 | if (!count) |
| 2030 | break; |
| 2031 | |
| 2032 | if (!(va->flags & VM_VM_AREA)) |
| 2033 | continue; |
| 2034 | |
| 2035 | vm = va->vm; |
| 2036 | vaddr = (char *) vm->addr; |
| 2037 | if (addr >= vaddr + vm->size - PAGE_SIZE) |
| 2038 | continue; |
| 2039 | while (addr < vaddr) { |
| 2040 | if (count == 0) |
| 2041 | goto finished; |
| 2042 | *buf = '\0'; |
| 2043 | buf++; |
| 2044 | addr++; |
| 2045 | count--; |
| 2046 | } |
| 2047 | n = vaddr + vm->size - PAGE_SIZE - addr; |
| 2048 | if (n > count) |
| 2049 | n = count; |
| 2050 | if (!(vm->flags & VM_IOREMAP)) |
| 2051 | aligned_vread(buf, addr, n); |
| 2052 | else /* IOREMAP area is treated as memory hole */ |
| 2053 | memset(buf, 0, n); |
| 2054 | buf += n; |
| 2055 | addr += n; |
| 2056 | count -= n; |
| 2057 | } |
| 2058 | finished: |
| 2059 | spin_unlock(&vmap_area_lock); |
| 2060 | |
| 2061 | if (buf == buf_start) |
| 2062 | return 0; |
| 2063 | /* zero-fill memory holes */ |
| 2064 | if (buf != buf_start + buflen) |
| 2065 | memset(buf, 0, buflen - (buf - buf_start)); |
| 2066 | |
| 2067 | return buflen; |
| 2068 | } |
| 2069 | |
| 2070 | /** |
| 2071 | * vwrite() - write vmalloc area in a safe way. |
| 2072 | * @buf: buffer for source data |
| 2073 | * @addr: vm address. |
| 2074 | * @count: number of bytes to be read. |
| 2075 | * |
| 2076 | * Returns # of bytes which addr and buf should be incresed. |
| 2077 | * (same number to @count). |
| 2078 | * If [addr...addr+count) doesn't includes any intersect with valid |
| 2079 | * vmalloc area, returns 0. |
| 2080 | * |
| 2081 | * This function checks that addr is a valid vmalloc'ed area, and |
| 2082 | * copy data from a buffer to the given addr. If specified range of |
| 2083 | * [addr...addr+count) includes some valid address, data is copied from |
| 2084 | * proper area of @buf. If there are memory holes, no copy to hole. |
| 2085 | * IOREMAP area is treated as memory hole and no copy is done. |
| 2086 | * |
| 2087 | * If [addr...addr+count) doesn't includes any intersects with alive |
| 2088 | * vm_struct area, returns 0. @buf should be kernel's buffer. |
| 2089 | * |
| 2090 | * Note: In usual ops, vwrite() is never necessary because the caller |
| 2091 | * should know vmalloc() area is valid and can use memcpy(). |
| 2092 | * This is for routines which have to access vmalloc area without |
| 2093 | * any informaion, as /dev/kmem. |
| 2094 | */ |
| 2095 | |
| 2096 | long vwrite(char *buf, char *addr, unsigned long count) |
| 2097 | { |
| 2098 | struct vmap_area *va; |
| 2099 | struct vm_struct *vm; |
| 2100 | char *vaddr; |
| 2101 | unsigned long n, buflen; |
| 2102 | int copied = 0; |
| 2103 | |
| 2104 | /* Don't allow overflow */ |
| 2105 | if ((unsigned long) addr + count < count) |
| 2106 | count = -(unsigned long) addr; |
| 2107 | buflen = count; |
| 2108 | |
| 2109 | spin_lock(&vmap_area_lock); |
| 2110 | list_for_each_entry(va, &vmap_area_list, list) { |
| 2111 | if (!count) |
| 2112 | break; |
| 2113 | |
| 2114 | if (!(va->flags & VM_VM_AREA)) |
| 2115 | continue; |
| 2116 | |
| 2117 | vm = va->vm; |
| 2118 | vaddr = (char *) vm->addr; |
| 2119 | if (addr >= vaddr + vm->size - PAGE_SIZE) |
| 2120 | continue; |
| 2121 | while (addr < vaddr) { |
| 2122 | if (count == 0) |
| 2123 | goto finished; |
| 2124 | buf++; |
| 2125 | addr++; |
| 2126 | count--; |
| 2127 | } |
| 2128 | n = vaddr + vm->size - PAGE_SIZE - addr; |
| 2129 | if (n > count) |
| 2130 | n = count; |
| 2131 | if (!(vm->flags & VM_IOREMAP)) { |
| 2132 | aligned_vwrite(buf, addr, n); |
| 2133 | copied++; |
| 2134 | } |
| 2135 | buf += n; |
| 2136 | addr += n; |
| 2137 | count -= n; |
| 2138 | } |
| 2139 | finished: |
| 2140 | spin_unlock(&vmap_area_lock); |
| 2141 | if (!copied) |
| 2142 | return 0; |
| 2143 | return buflen; |
| 2144 | } |
| 2145 | |
| 2146 | /** |
| 2147 | * remap_vmalloc_range - map vmalloc pages to userspace |
| 2148 | * @vma: vma to cover (map full range of vma) |
| 2149 | * @addr: vmalloc memory |
| 2150 | * @pgoff: number of pages into addr before first page to map |
| 2151 | * |
| 2152 | * Returns: 0 for success, -Exxx on failure |
| 2153 | * |
| 2154 | * This function checks that addr is a valid vmalloc'ed area, and |
| 2155 | * that it is big enough to cover the vma. Will return failure if |
| 2156 | * that criteria isn't met. |
| 2157 | * |
| 2158 | * Similar to remap_pfn_range() (see mm/memory.c) |
| 2159 | */ |
| 2160 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
| 2161 | unsigned long pgoff) |
| 2162 | { |
| 2163 | struct vm_struct *area; |
| 2164 | unsigned long uaddr = vma->vm_start; |
| 2165 | unsigned long usize = vma->vm_end - vma->vm_start; |
| 2166 | |
| 2167 | if ((PAGE_SIZE-1) & (unsigned long)addr) |
| 2168 | return -EINVAL; |
| 2169 | |
| 2170 | area = find_vm_area(addr); |
| 2171 | if (!area) |
| 2172 | return -EINVAL; |
| 2173 | |
| 2174 | if (!(area->flags & VM_USERMAP)) |
| 2175 | return -EINVAL; |
| 2176 | |
| 2177 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) |
| 2178 | return -EINVAL; |
| 2179 | |
| 2180 | addr += pgoff << PAGE_SHIFT; |
| 2181 | do { |
| 2182 | struct page *page = vmalloc_to_page(addr); |
| 2183 | int ret; |
| 2184 | |
| 2185 | ret = vm_insert_page(vma, uaddr, page); |
| 2186 | if (ret) |
| 2187 | return ret; |
| 2188 | |
| 2189 | uaddr += PAGE_SIZE; |
| 2190 | addr += PAGE_SIZE; |
| 2191 | usize -= PAGE_SIZE; |
| 2192 | } while (usize > 0); |
| 2193 | |
| 2194 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
| 2195 | |
| 2196 | return 0; |
| 2197 | } |
| 2198 | EXPORT_SYMBOL(remap_vmalloc_range); |
| 2199 | |
| 2200 | /* |
| 2201 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to |
| 2202 | * have one. |
| 2203 | */ |
| 2204 | void __attribute__((weak)) vmalloc_sync_all(void) |
| 2205 | { |
| 2206 | } |
| 2207 | |
| 2208 | |
| 2209 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
| 2210 | { |
| 2211 | pte_t ***p = data; |
| 2212 | |
| 2213 | if (p) { |
| 2214 | *(*p) = pte; |
| 2215 | (*p)++; |
| 2216 | } |
| 2217 | return 0; |
| 2218 | } |
| 2219 | |
| 2220 | /** |
| 2221 | * alloc_vm_area - allocate a range of kernel address space |
| 2222 | * @size: size of the area |
| 2223 | * @ptes: returns the PTEs for the address space |
| 2224 | * |
| 2225 | * Returns: NULL on failure, vm_struct on success |
| 2226 | * |
| 2227 | * This function reserves a range of kernel address space, and |
| 2228 | * allocates pagetables to map that range. No actual mappings |
| 2229 | * are created. |
| 2230 | * |
| 2231 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm) |
| 2232 | * allocated for the VM area are returned. |
| 2233 | */ |
| 2234 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
| 2235 | { |
| 2236 | struct vm_struct *area; |
| 2237 | |
| 2238 | area = get_vm_area_caller(size, VM_IOREMAP, |
| 2239 | __builtin_return_address(0)); |
| 2240 | if (area == NULL) |
| 2241 | return NULL; |
| 2242 | |
| 2243 | /* |
| 2244 | * This ensures that page tables are constructed for this region |
| 2245 | * of kernel virtual address space and mapped into init_mm. |
| 2246 | */ |
| 2247 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, |
| 2248 | size, f, ptes ? &ptes : NULL)) { |
| 2249 | free_vm_area(area); |
| 2250 | return NULL; |
| 2251 | } |
| 2252 | |
| 2253 | return area; |
| 2254 | } |
| 2255 | EXPORT_SYMBOL_GPL(alloc_vm_area); |
| 2256 | |
| 2257 | void free_vm_area(struct vm_struct *area) |
| 2258 | { |
| 2259 | struct vm_struct *ret; |
| 2260 | ret = remove_vm_area(area->addr); |
| 2261 | BUG_ON(ret != area); |
| 2262 | kfree(area); |
| 2263 | } |
| 2264 | EXPORT_SYMBOL_GPL(free_vm_area); |
| 2265 | |
| 2266 | #ifdef CONFIG_SMP |
| 2267 | static struct vmap_area *node_to_va(struct rb_node *n) |
| 2268 | { |
| 2269 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; |
| 2270 | } |
| 2271 | |
| 2272 | /** |
| 2273 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end |
| 2274 | * @end: target address |
| 2275 | * @pnext: out arg for the next vmap_area |
| 2276 | * @pprev: out arg for the previous vmap_area |
| 2277 | * |
| 2278 | * Returns: %true if either or both of next and prev are found, |
| 2279 | * %false if no vmap_area exists |
| 2280 | * |
| 2281 | * Find vmap_areas end addresses of which enclose @end. ie. if not |
| 2282 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. |
| 2283 | */ |
| 2284 | static bool pvm_find_next_prev(unsigned long end, |
| 2285 | struct vmap_area **pnext, |
| 2286 | struct vmap_area **pprev) |
| 2287 | { |
| 2288 | struct rb_node *n = vmap_area_root.rb_node; |
| 2289 | struct vmap_area *va = NULL; |
| 2290 | |
| 2291 | while (n) { |
| 2292 | va = rb_entry(n, struct vmap_area, rb_node); |
| 2293 | if (end < va->va_end) |
| 2294 | n = n->rb_left; |
| 2295 | else if (end > va->va_end) |
| 2296 | n = n->rb_right; |
| 2297 | else |
| 2298 | break; |
| 2299 | } |
| 2300 | |
| 2301 | if (!va) |
| 2302 | return false; |
| 2303 | |
| 2304 | if (va->va_end > end) { |
| 2305 | *pnext = va; |
| 2306 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); |
| 2307 | } else { |
| 2308 | *pprev = va; |
| 2309 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); |
| 2310 | } |
| 2311 | return true; |
| 2312 | } |
| 2313 | |
| 2314 | /** |
| 2315 | * pvm_determine_end - find the highest aligned address between two vmap_areas |
| 2316 | * @pnext: in/out arg for the next vmap_area |
| 2317 | * @pprev: in/out arg for the previous vmap_area |
| 2318 | * @align: alignment |
| 2319 | * |
| 2320 | * Returns: determined end address |
| 2321 | * |
| 2322 | * Find the highest aligned address between *@pnext and *@pprev below |
| 2323 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned |
| 2324 | * down address is between the end addresses of the two vmap_areas. |
| 2325 | * |
| 2326 | * Please note that the address returned by this function may fall |
| 2327 | * inside *@pnext vmap_area. The caller is responsible for checking |
| 2328 | * that. |
| 2329 | */ |
| 2330 | static unsigned long pvm_determine_end(struct vmap_area **pnext, |
| 2331 | struct vmap_area **pprev, |
| 2332 | unsigned long align) |
| 2333 | { |
| 2334 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
| 2335 | unsigned long addr; |
| 2336 | |
| 2337 | if (*pnext) |
| 2338 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); |
| 2339 | else |
| 2340 | addr = vmalloc_end; |
| 2341 | |
| 2342 | while (*pprev && (*pprev)->va_end > addr) { |
| 2343 | *pnext = *pprev; |
| 2344 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); |
| 2345 | } |
| 2346 | |
| 2347 | return addr; |
| 2348 | } |
| 2349 | |
| 2350 | /** |
| 2351 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator |
| 2352 | * @offsets: array containing offset of each area |
| 2353 | * @sizes: array containing size of each area |
| 2354 | * @nr_vms: the number of areas to allocate |
| 2355 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this |
| 2356 | * |
| 2357 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated |
| 2358 | * vm_structs on success, %NULL on failure |
| 2359 | * |
| 2360 | * Percpu allocator wants to use congruent vm areas so that it can |
| 2361 | * maintain the offsets among percpu areas. This function allocates |
| 2362 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
| 2363 | * be scattered pretty far, distance between two areas easily going up |
| 2364 | * to gigabytes. To avoid interacting with regular vmallocs, these |
| 2365 | * areas are allocated from top. |
| 2366 | * |
| 2367 | * Despite its complicated look, this allocator is rather simple. It |
| 2368 | * does everything top-down and scans areas from the end looking for |
| 2369 | * matching slot. While scanning, if any of the areas overlaps with |
| 2370 | * existing vmap_area, the base address is pulled down to fit the |
| 2371 | * area. Scanning is repeated till all the areas fit and then all |
| 2372 | * necessary data structres are inserted and the result is returned. |
| 2373 | */ |
| 2374 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, |
| 2375 | const size_t *sizes, int nr_vms, |
| 2376 | size_t align) |
| 2377 | { |
| 2378 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); |
| 2379 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
| 2380 | struct vmap_area **vas, *prev, *next; |
| 2381 | struct vm_struct **vms; |
| 2382 | int area, area2, last_area, term_area; |
| 2383 | unsigned long base, start, end, last_end; |
| 2384 | bool purged = false; |
| 2385 | |
| 2386 | /* verify parameters and allocate data structures */ |
| 2387 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); |
| 2388 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
| 2389 | start = offsets[area]; |
| 2390 | end = start + sizes[area]; |
| 2391 | |
| 2392 | /* is everything aligned properly? */ |
| 2393 | BUG_ON(!IS_ALIGNED(offsets[area], align)); |
| 2394 | BUG_ON(!IS_ALIGNED(sizes[area], align)); |
| 2395 | |
| 2396 | /* detect the area with the highest address */ |
| 2397 | if (start > offsets[last_area]) |
| 2398 | last_area = area; |
| 2399 | |
| 2400 | for (area2 = 0; area2 < nr_vms; area2++) { |
| 2401 | unsigned long start2 = offsets[area2]; |
| 2402 | unsigned long end2 = start2 + sizes[area2]; |
| 2403 | |
| 2404 | if (area2 == area) |
| 2405 | continue; |
| 2406 | |
| 2407 | BUG_ON(start2 >= start && start2 < end); |
| 2408 | BUG_ON(end2 <= end && end2 > start); |
| 2409 | } |
| 2410 | } |
| 2411 | last_end = offsets[last_area] + sizes[last_area]; |
| 2412 | |
| 2413 | if (vmalloc_end - vmalloc_start < last_end) { |
| 2414 | WARN_ON(true); |
| 2415 | return NULL; |
| 2416 | } |
| 2417 | |
| 2418 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL); |
| 2419 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL); |
| 2420 | if (!vas || !vms) |
| 2421 | goto err_free2; |
| 2422 | |
| 2423 | for (area = 0; area < nr_vms; area++) { |
| 2424 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
| 2425 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
| 2426 | if (!vas[area] || !vms[area]) |
| 2427 | goto err_free; |
| 2428 | } |
| 2429 | retry: |
| 2430 | spin_lock(&vmap_area_lock); |
| 2431 | |
| 2432 | /* start scanning - we scan from the top, begin with the last area */ |
| 2433 | area = term_area = last_area; |
| 2434 | start = offsets[area]; |
| 2435 | end = start + sizes[area]; |
| 2436 | |
| 2437 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { |
| 2438 | base = vmalloc_end - last_end; |
| 2439 | goto found; |
| 2440 | } |
| 2441 | base = pvm_determine_end(&next, &prev, align) - end; |
| 2442 | |
| 2443 | while (true) { |
| 2444 | BUG_ON(next && next->va_end <= base + end); |
| 2445 | BUG_ON(prev && prev->va_end > base + end); |
| 2446 | |
| 2447 | /* |
| 2448 | * base might have underflowed, add last_end before |
| 2449 | * comparing. |
| 2450 | */ |
| 2451 | if (base + last_end < vmalloc_start + last_end) { |
| 2452 | spin_unlock(&vmap_area_lock); |
| 2453 | if (!purged) { |
| 2454 | purge_vmap_area_lazy(); |
| 2455 | purged = true; |
| 2456 | goto retry; |
| 2457 | } |
| 2458 | goto err_free; |
| 2459 | } |
| 2460 | |
| 2461 | /* |
| 2462 | * If next overlaps, move base downwards so that it's |
| 2463 | * right below next and then recheck. |
| 2464 | */ |
| 2465 | if (next && next->va_start < base + end) { |
| 2466 | base = pvm_determine_end(&next, &prev, align) - end; |
| 2467 | term_area = area; |
| 2468 | continue; |
| 2469 | } |
| 2470 | |
| 2471 | /* |
| 2472 | * If prev overlaps, shift down next and prev and move |
| 2473 | * base so that it's right below new next and then |
| 2474 | * recheck. |
| 2475 | */ |
| 2476 | if (prev && prev->va_end > base + start) { |
| 2477 | next = prev; |
| 2478 | prev = node_to_va(rb_prev(&next->rb_node)); |
| 2479 | base = pvm_determine_end(&next, &prev, align) - end; |
| 2480 | term_area = area; |
| 2481 | continue; |
| 2482 | } |
| 2483 | |
| 2484 | /* |
| 2485 | * This area fits, move on to the previous one. If |
| 2486 | * the previous one is the terminal one, we're done. |
| 2487 | */ |
| 2488 | area = (area + nr_vms - 1) % nr_vms; |
| 2489 | if (area == term_area) |
| 2490 | break; |
| 2491 | start = offsets[area]; |
| 2492 | end = start + sizes[area]; |
| 2493 | pvm_find_next_prev(base + end, &next, &prev); |
| 2494 | } |
| 2495 | found: |
| 2496 | /* we've found a fitting base, insert all va's */ |
| 2497 | for (area = 0; area < nr_vms; area++) { |
| 2498 | struct vmap_area *va = vas[area]; |
| 2499 | |
| 2500 | va->va_start = base + offsets[area]; |
| 2501 | va->va_end = va->va_start + sizes[area]; |
| 2502 | __insert_vmap_area(va); |
| 2503 | } |
| 2504 | |
| 2505 | vmap_area_pcpu_hole = base + offsets[last_area]; |
| 2506 | |
| 2507 | spin_unlock(&vmap_area_lock); |
| 2508 | |
| 2509 | /* insert all vm's */ |
| 2510 | for (area = 0; area < nr_vms; area++) |
| 2511 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
| 2512 | pcpu_get_vm_areas); |
| 2513 | |
| 2514 | kfree(vas); |
| 2515 | return vms; |
| 2516 | |
| 2517 | err_free: |
| 2518 | for (area = 0; area < nr_vms; area++) { |
| 2519 | kfree(vas[area]); |
| 2520 | kfree(vms[area]); |
| 2521 | } |
| 2522 | err_free2: |
| 2523 | kfree(vas); |
| 2524 | kfree(vms); |
| 2525 | return NULL; |
| 2526 | } |
| 2527 | |
| 2528 | /** |
| 2529 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator |
| 2530 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() |
| 2531 | * @nr_vms: the number of allocated areas |
| 2532 | * |
| 2533 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). |
| 2534 | */ |
| 2535 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) |
| 2536 | { |
| 2537 | int i; |
| 2538 | |
| 2539 | for (i = 0; i < nr_vms; i++) |
| 2540 | free_vm_area(vms[i]); |
| 2541 | kfree(vms); |
| 2542 | } |
| 2543 | #endif /* CONFIG_SMP */ |
| 2544 | |
| 2545 | #ifdef CONFIG_PROC_FS |
| 2546 | static void *s_start(struct seq_file *m, loff_t *pos) |
| 2547 | __acquires(&vmap_area_lock) |
| 2548 | { |
| 2549 | loff_t n = *pos; |
| 2550 | struct vmap_area *va; |
| 2551 | |
| 2552 | spin_lock(&vmap_area_lock); |
| 2553 | va = list_entry((&vmap_area_list)->next, typeof(*va), list); |
| 2554 | while (n > 0 && &va->list != &vmap_area_list) { |
| 2555 | n--; |
| 2556 | va = list_entry(va->list.next, typeof(*va), list); |
| 2557 | } |
| 2558 | if (!n && &va->list != &vmap_area_list) |
| 2559 | return va; |
| 2560 | |
| 2561 | return NULL; |
| 2562 | |
| 2563 | } |
| 2564 | |
| 2565 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
| 2566 | { |
| 2567 | struct vmap_area *va = p, *next; |
| 2568 | |
| 2569 | ++*pos; |
| 2570 | next = list_entry(va->list.next, typeof(*va), list); |
| 2571 | if (&next->list != &vmap_area_list) |
| 2572 | return next; |
| 2573 | |
| 2574 | return NULL; |
| 2575 | } |
| 2576 | |
| 2577 | static void s_stop(struct seq_file *m, void *p) |
| 2578 | __releases(&vmap_area_lock) |
| 2579 | { |
| 2580 | spin_unlock(&vmap_area_lock); |
| 2581 | } |
| 2582 | |
| 2583 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
| 2584 | { |
| 2585 | if (IS_ENABLED(CONFIG_NUMA)) { |
| 2586 | unsigned int nr, *counters = m->private; |
| 2587 | |
| 2588 | if (!counters) |
| 2589 | return; |
| 2590 | |
| 2591 | /* Pair with smp_wmb() in clear_vm_unlist() */ |
| 2592 | smp_rmb(); |
| 2593 | if (v->flags & VM_UNLIST) |
| 2594 | return; |
| 2595 | |
| 2596 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
| 2597 | |
| 2598 | for (nr = 0; nr < v->nr_pages; nr++) |
| 2599 | counters[page_to_nid(v->pages[nr])]++; |
| 2600 | |
| 2601 | for_each_node_state(nr, N_HIGH_MEMORY) |
| 2602 | if (counters[nr]) |
| 2603 | seq_printf(m, " N%u=%u", nr, counters[nr]); |
| 2604 | } |
| 2605 | } |
| 2606 | |
| 2607 | static int s_show(struct seq_file *m, void *p) |
| 2608 | { |
| 2609 | struct vmap_area *va = p; |
| 2610 | struct vm_struct *v; |
| 2611 | |
| 2612 | if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING)) |
| 2613 | return 0; |
| 2614 | |
| 2615 | if (!(va->flags & VM_VM_AREA)) { |
| 2616 | seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n", |
| 2617 | (void *)va->va_start, (void *)va->va_end, |
| 2618 | va->va_end - va->va_start); |
| 2619 | return 0; |
| 2620 | } |
| 2621 | |
| 2622 | v = va->vm; |
| 2623 | |
| 2624 | seq_printf(m, "0x%pK-0x%pK %7ld", |
| 2625 | v->addr, v->addr + v->size, v->size); |
| 2626 | |
| 2627 | if (v->caller) |
| 2628 | seq_printf(m, " %pS", v->caller); |
| 2629 | |
| 2630 | if (v->nr_pages) |
| 2631 | seq_printf(m, " pages=%d", v->nr_pages); |
| 2632 | |
| 2633 | if (v->phys_addr) |
| 2634 | seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr); |
| 2635 | |
| 2636 | if (v->flags & VM_IOREMAP) |
| 2637 | seq_printf(m, " ioremap"); |
| 2638 | |
| 2639 | if (v->flags & VM_ALLOC) |
| 2640 | seq_printf(m, " vmalloc"); |
| 2641 | |
| 2642 | if (v->flags & VM_MAP) |
| 2643 | seq_printf(m, " vmap"); |
| 2644 | |
| 2645 | if (v->flags & VM_USERMAP) |
| 2646 | seq_printf(m, " user"); |
| 2647 | |
| 2648 | if (v->flags & VM_VPAGES) |
| 2649 | seq_printf(m, " vpages"); |
| 2650 | |
| 2651 | show_numa_info(m, v); |
| 2652 | seq_putc(m, '\n'); |
| 2653 | return 0; |
| 2654 | } |
| 2655 | |
| 2656 | static const struct seq_operations vmalloc_op = { |
| 2657 | .start = s_start, |
| 2658 | .next = s_next, |
| 2659 | .stop = s_stop, |
| 2660 | .show = s_show, |
| 2661 | }; |
| 2662 | |
| 2663 | static int vmalloc_open(struct inode *inode, struct file *file) |
| 2664 | { |
| 2665 | unsigned int *ptr = NULL; |
| 2666 | int ret; |
| 2667 | |
| 2668 | if (IS_ENABLED(CONFIG_NUMA)) { |
| 2669 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); |
| 2670 | if (ptr == NULL) |
| 2671 | return -ENOMEM; |
| 2672 | } |
| 2673 | ret = seq_open(file, &vmalloc_op); |
| 2674 | if (!ret) { |
| 2675 | struct seq_file *m = file->private_data; |
| 2676 | m->private = ptr; |
| 2677 | } else |
| 2678 | kfree(ptr); |
| 2679 | return ret; |
| 2680 | } |
| 2681 | |
| 2682 | static const struct file_operations proc_vmalloc_operations = { |
| 2683 | .open = vmalloc_open, |
| 2684 | .read = seq_read, |
| 2685 | .llseek = seq_lseek, |
| 2686 | .release = seq_release_private, |
| 2687 | }; |
| 2688 | |
| 2689 | static int __init proc_vmalloc_init(void) |
| 2690 | { |
| 2691 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); |
| 2692 | return 0; |
| 2693 | } |
| 2694 | module_init(proc_vmalloc_init); |
| 2695 | |
| 2696 | void get_vmalloc_info(struct vmalloc_info *vmi) |
| 2697 | { |
| 2698 | struct vmap_area *va; |
| 2699 | unsigned long free_area_size; |
| 2700 | unsigned long prev_end; |
| 2701 | |
| 2702 | vmi->used = 0; |
| 2703 | vmi->largest_chunk = 0; |
| 2704 | |
| 2705 | prev_end = VMALLOC_START; |
| 2706 | |
| 2707 | spin_lock(&vmap_area_lock); |
| 2708 | |
| 2709 | if (list_empty(&vmap_area_list)) { |
| 2710 | vmi->largest_chunk = VMALLOC_TOTAL; |
| 2711 | goto out; |
| 2712 | } |
| 2713 | |
| 2714 | list_for_each_entry(va, &vmap_area_list, list) { |
| 2715 | unsigned long addr = va->va_start; |
| 2716 | |
| 2717 | /* |
| 2718 | * Some archs keep another range for modules in vmalloc space |
| 2719 | */ |
| 2720 | if (addr < VMALLOC_START) |
| 2721 | continue; |
| 2722 | if (addr >= VMALLOC_END) |
| 2723 | break; |
| 2724 | |
| 2725 | if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING)) |
| 2726 | continue; |
| 2727 | |
| 2728 | vmi->used += (va->va_end - va->va_start); |
| 2729 | |
| 2730 | free_area_size = addr - prev_end; |
| 2731 | if (vmi->largest_chunk < free_area_size) |
| 2732 | vmi->largest_chunk = free_area_size; |
| 2733 | |
| 2734 | prev_end = va->va_end; |
| 2735 | } |
| 2736 | |
| 2737 | if (VMALLOC_END - prev_end > vmi->largest_chunk) |
| 2738 | vmi->largest_chunk = VMALLOC_END - prev_end; |
| 2739 | |
| 2740 | out: |
| 2741 | spin_unlock(&vmap_area_lock); |
| 2742 | } |
| 2743 | #endif |
| 2744 | |