| 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 <[email protected]>, 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/signal.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/notifier.h>
|
| 25 | #include <linux/rbtree.h>
|
| 26 | #include <linux/radix-tree.h>
|
| 27 | #include <linux/rcupdate.h>
|
| 28 | #include <linux/pfn.h>
|
| 29 | #include <linux/kmemleak.h>
|
| 30 | #include <linux/atomic.h>
|
| 31 | #include <linux/compiler.h>
|
| 32 | #include <linux/llist.h>
|
| 33 | #include <linux/bitops.h>
|
| 34 |
|
| 35 | #include <linux/uaccess.h>
|
| 36 | #include <asm/tlbflush.h>
|
| 37 | #include <asm/shmparam.h>
|
| 38 |
|
| 39 | #include "internal.h"
|
| 40 |
|
| 41 | struct vfree_deferred {
|
| 42 | struct llist_head list;
|
| 43 | struct work_struct wq;
|
| 44 | };
|
| 45 | static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
|
| 46 |
|
| 47 | static void __vunmap(const void *, int);
|
| 48 |
|
| 49 | static void free_work(struct work_struct *w)
|
| 50 | {
|
| 51 | struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
|
| 52 | struct llist_node *llnode = llist_del_all(&p->list);
|
| 53 | while (llnode) {
|
| 54 | void *p = llnode;
|
| 55 | llnode = llist_next(llnode);
|
| 56 | __vunmap(p, 1);
|
| 57 | }
|
| 58 | }
|
| 59 |
|
| 60 |
|
| 61 |
|
| 62 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
|
| 63 | {
|
| 64 | pte_t *pte;
|
| 65 |
|
| 66 | pte = pte_offset_kernel(pmd, addr);
|
| 67 | do {
|
| 68 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
|
| 69 | WARN_ON(!pte_none(ptent) && !pte_present(ptent));
|
| 70 | } while (pte++, addr += PAGE_SIZE, addr != end);
|
| 71 | }
|
| 72 |
|
| 73 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
|
| 74 | {
|
| 75 | pmd_t *pmd;
|
| 76 | unsigned long next;
|
| 77 |
|
| 78 | pmd = pmd_offset(pud, addr);
|
| 79 | do {
|
| 80 | next = pmd_addr_end(addr, end);
|
| 81 | if (pmd_clear_huge(pmd))
|
| 82 | continue;
|
| 83 | if (pmd_none_or_clear_bad(pmd))
|
| 84 | continue;
|
| 85 | vunmap_pte_range(pmd, addr, next);
|
| 86 | } while (pmd++, addr = next, addr != end);
|
| 87 | }
|
| 88 |
|
| 89 | static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end)
|
| 90 | {
|
| 91 | pud_t *pud;
|
| 92 | unsigned long next;
|
| 93 |
|
| 94 | pud = pud_offset(p4d, addr);
|
| 95 | do {
|
| 96 | next = pud_addr_end(addr, end);
|
| 97 | if (pud_clear_huge(pud))
|
| 98 | continue;
|
| 99 | if (pud_none_or_clear_bad(pud))
|
| 100 | continue;
|
| 101 | vunmap_pmd_range(pud, addr, next);
|
| 102 | } while (pud++, addr = next, addr != end);
|
| 103 | }
|
| 104 |
|
| 105 | static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end)
|
| 106 | {
|
| 107 | p4d_t *p4d;
|
| 108 | unsigned long next;
|
| 109 |
|
| 110 | p4d = p4d_offset(pgd, addr);
|
| 111 | do {
|
| 112 | next = p4d_addr_end(addr, end);
|
| 113 | if (p4d_clear_huge(p4d))
|
| 114 | continue;
|
| 115 | if (p4d_none_or_clear_bad(p4d))
|
| 116 | continue;
|
| 117 | vunmap_pud_range(p4d, addr, next);
|
| 118 | } while (p4d++, addr = next, addr != end);
|
| 119 | }
|
| 120 |
|
| 121 | static void vunmap_page_range(unsigned long addr, unsigned long end)
|
| 122 | {
|
| 123 | pgd_t *pgd;
|
| 124 | unsigned long next;
|
| 125 |
|
| 126 | BUG_ON(addr >= end);
|
| 127 | pgd = pgd_offset_k(addr);
|
| 128 | do {
|
| 129 | next = pgd_addr_end(addr, end);
|
| 130 | if (pgd_none_or_clear_bad(pgd))
|
| 131 | continue;
|
| 132 | vunmap_p4d_range(pgd, addr, next);
|
| 133 | } while (pgd++, addr = next, addr != end);
|
| 134 | }
|
| 135 |
|
| 136 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
|
| 137 | unsigned long end, pgprot_t prot, struct page **pages, int *nr)
|
| 138 | {
|
| 139 | pte_t *pte;
|
| 140 |
|
| 141 | |
| 142 | * nr is a running index into the array which helps higher level
|
| 143 | * callers keep track of where we're up to.
|
| 144 |
|
| 145 |
|
| 146 | pte = pte_alloc_kernel(pmd, addr);
|
| 147 | if (!pte)
|
| 148 | return -ENOMEM;
|
| 149 | do {
|
| 150 | struct page *page = pages[*nr];
|
| 151 |
|
| 152 | if (WARN_ON(!pte_none(*pte)))
|
| 153 | return -EBUSY;
|
| 154 | if (WARN_ON(!page))
|
| 155 | return -ENOMEM;
|
| 156 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
|
| 157 | (*nr)++;
|
| 158 | } while (pte++, addr += PAGE_SIZE, addr != end);
|
| 159 | return 0;
|
| 160 | }
|
| 161 |
|
| 162 | static int vmap_pmd_range(pud_t *pud, unsigned long addr,
|
| 163 | unsigned long end, pgprot_t prot, struct page **pages, int *nr)
|
| 164 | {
|
| 165 | pmd_t *pmd;
|
| 166 | unsigned long next;
|
| 167 |
|
| 168 | pmd = pmd_alloc(&init_mm, pud, addr);
|
| 169 | if (!pmd)
|
| 170 | return -ENOMEM;
|
| 171 | do {
|
| 172 | next = pmd_addr_end(addr, end);
|
| 173 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
|
| 174 | return -ENOMEM;
|
| 175 | } while (pmd++, addr = next, addr != end);
|
| 176 | return 0;
|
| 177 | }
|
| 178 |
|
| 179 | static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
|
| 180 | unsigned long end, pgprot_t prot, struct page **pages, int *nr)
|
| 181 | {
|
| 182 | pud_t *pud;
|
| 183 | unsigned long next;
|
| 184 |
|
| 185 | pud = pud_alloc(&init_mm, p4d, addr);
|
| 186 | if (!pud)
|
| 187 | return -ENOMEM;
|
| 188 | do {
|
| 189 | next = pud_addr_end(addr, end);
|
| 190 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
|
| 191 | return -ENOMEM;
|
| 192 | } while (pud++, addr = next, addr != end);
|
| 193 | return 0;
|
| 194 | }
|
| 195 |
|
| 196 | static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
|
| 197 | unsigned long end, pgprot_t prot, struct page **pages, int *nr)
|
| 198 | {
|
| 199 | p4d_t *p4d;
|
| 200 | unsigned long next;
|
| 201 |
|
| 202 | p4d = p4d_alloc(&init_mm, pgd, addr);
|
| 203 | if (!p4d)
|
| 204 | return -ENOMEM;
|
| 205 | do {
|
| 206 | next = p4d_addr_end(addr, end);
|
| 207 | if (vmap_pud_range(p4d, addr, next, prot, pages, nr))
|
| 208 | return -ENOMEM;
|
| 209 | } while (p4d++, addr = next, addr != end);
|
| 210 | return 0;
|
| 211 | }
|
| 212 |
|
| 213 | |
| 214 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
|
| 215 | * will have pfns corresponding to the "pages" array.
|
| 216 | *
|
| 217 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
|
| 218 |
|
| 219 | static int vmap_page_range_noflush(unsigned long start, unsigned long end,
|
| 220 | pgprot_t prot, struct page **pages)
|
| 221 | {
|
| 222 | pgd_t *pgd;
|
| 223 | unsigned long next;
|
| 224 | unsigned long addr = start;
|
| 225 | int err = 0;
|
| 226 | int nr = 0;
|
| 227 |
|
| 228 | BUG_ON(addr >= end);
|
| 229 | pgd = pgd_offset_k(addr);
|
| 230 | do {
|
| 231 | next = pgd_addr_end(addr, end);
|
| 232 | err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr);
|
| 233 | if (err)
|
| 234 | return err;
|
| 235 | } while (pgd++, addr = next, addr != end);
|
| 236 |
|
| 237 | return nr;
|
| 238 | }
|
| 239 |
|
| 240 | static int vmap_page_range(unsigned long start, unsigned long end,
|
| 241 | pgprot_t prot, struct page **pages)
|
| 242 | {
|
| 243 | int ret;
|
| 244 |
|
| 245 | ret = vmap_page_range_noflush(start, end, prot, pages);
|
| 246 | flush_cache_vmap(start, end);
|
| 247 | return ret;
|
| 248 | }
|
| 249 |
|
| 250 | int is_vmalloc_or_module_addr(const void *x)
|
| 251 | {
|
| 252 | |
| 253 | * ARM, x86-64 and sparc64 put modules in a special place,
|
| 254 | * and fall back on vmalloc() if that fails. Others
|
| 255 | * just put it in the vmalloc space.
|
| 256 |
|
| 257 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
|
| 258 | unsigned long addr = (unsigned long)x;
|
| 259 | if (addr >= MODULES_VADDR && addr < MODULES_END)
|
| 260 | return 1;
|
| 261 | #endif
|
| 262 | return is_vmalloc_addr(x);
|
| 263 | }
|
| 264 |
|
| 265 | |
| 266 | * Walk a vmap address to the struct page it maps.
|
| 267 |
|
| 268 | struct page *vmalloc_to_page(const void *vmalloc_addr)
|
| 269 | {
|
| 270 | unsigned long addr = (unsigned long) vmalloc_addr;
|
| 271 | struct page *page = NULL;
|
| 272 | pgd_t *pgd = pgd_offset_k(addr);
|
| 273 | p4d_t *p4d;
|
| 274 | pud_t *pud;
|
| 275 | pmd_t *pmd;
|
| 276 | pte_t *ptep, pte;
|
| 277 |
|
| 278 | |
| 279 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for
|
| 280 | * architectures that do not vmalloc module space
|
| 281 |
|
| 282 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
|
| 283 |
|
| 284 | if (pgd_none(*pgd))
|
| 285 | return NULL;
|
| 286 | p4d = p4d_offset(pgd, addr);
|
| 287 | if (p4d_none(*p4d))
|
| 288 | return NULL;
|
| 289 | pud = pud_offset(p4d, addr);
|
| 290 |
|
| 291 | |
| 292 | * Don't dereference bad PUD or PMD (below) entries. This will also
|
| 293 | * identify huge mappings, which we may encounter on architectures
|
| 294 | * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
|
| 295 | * identified as vmalloc addresses by is_vmalloc_addr(), but are
|
| 296 | * not [unambiguously] associated with a struct page, so there is
|
| 297 | * no correct value to return for them.
|
| 298 |
|
| 299 | WARN_ON_ONCE(pud_bad(*pud));
|
| 300 | if (pud_none(*pud) || pud_bad(*pud))
|
| 301 | return NULL;
|
| 302 | pmd = pmd_offset(pud, addr);
|
| 303 | WARN_ON_ONCE(pmd_bad(*pmd));
|
| 304 | if (pmd_none(*pmd) || pmd_bad(*pmd))
|
| 305 | return NULL;
|
| 306 |
|
| 307 | ptep = pte_offset_map(pmd, addr);
|
| 308 | pte = *ptep;
|
| 309 | if (pte_present(pte))
|
| 310 | page = pte_page(pte);
|
| 311 | pte_unmap(ptep);
|
| 312 | return page;
|
| 313 | }
|
| 314 | EXPORT_SYMBOL(vmalloc_to_page);
|
| 315 |
|
| 316 | |
| 317 | * Map a vmalloc()-space virtual address to the physical page frame number.
|
| 318 |
|
| 319 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
|
| 320 | {
|
| 321 | return page_to_pfn(vmalloc_to_page(vmalloc_addr));
|
| 322 | }
|
| 323 | EXPORT_SYMBOL(vmalloc_to_pfn);
|
| 324 |
|
| 325 |
|
| 326 |
|
| 327 |
|
| 328 | #define VM_VM_AREA 0x04
|
| 329 |
|
| 330 | static DEFINE_SPINLOCK(vmap_area_lock);
|
| 331 |
|
| 332 | LIST_HEAD(vmap_area_list);
|
| 333 | static LLIST_HEAD(vmap_purge_list);
|
| 334 | static struct rb_root vmap_area_root = RB_ROOT;
|
| 335 |
|
| 336 |
|
| 337 | static struct rb_node *free_vmap_cache;
|
| 338 | static unsigned long cached_hole_size;
|
| 339 | static unsigned long cached_vstart;
|
| 340 | static unsigned long cached_align;
|
| 341 |
|
| 342 | static unsigned long vmap_area_pcpu_hole;
|
| 343 |
|
| 344 | static struct vmap_area *__find_vmap_area(unsigned long addr)
|
| 345 | {
|
| 346 | struct rb_node *n = vmap_area_root.rb_node;
|
| 347 |
|
| 348 | while (n) {
|
| 349 | struct vmap_area *va;
|
| 350 |
|
| 351 | va = rb_entry(n, struct vmap_area, rb_node);
|
| 352 | if (addr < va->va_start)
|
| 353 | n = n->rb_left;
|
| 354 | else if (addr >= va->va_end)
|
| 355 | n = n->rb_right;
|
| 356 | else
|
| 357 | return va;
|
| 358 | }
|
| 359 |
|
| 360 | return NULL;
|
| 361 | }
|
| 362 |
|
| 363 | static void __insert_vmap_area(struct vmap_area *va)
|
| 364 | {
|
| 365 | struct rb_node **p = &vmap_area_root.rb_node;
|
| 366 | struct rb_node *parent = NULL;
|
| 367 | struct rb_node *tmp;
|
| 368 |
|
| 369 | while (*p) {
|
| 370 | struct vmap_area *tmp_va;
|
| 371 |
|
| 372 | parent = *p;
|
| 373 | tmp_va = rb_entry(parent, struct vmap_area, rb_node);
|
| 374 | if (va->va_start < tmp_va->va_end)
|
| 375 | p = &(*p)->rb_left;
|
| 376 | else if (va->va_end > tmp_va->va_start)
|
| 377 | p = &(*p)->rb_right;
|
| 378 | else
|
| 379 | BUG();
|
| 380 | }
|
| 381 |
|
| 382 | rb_link_node(&va->rb_node, parent, p);
|
| 383 | rb_insert_color(&va->rb_node, &vmap_area_root);
|
| 384 |
|
| 385 |
|
| 386 | tmp = rb_prev(&va->rb_node);
|
| 387 | if (tmp) {
|
| 388 | struct vmap_area *prev;
|
| 389 | prev = rb_entry(tmp, struct vmap_area, rb_node);
|
| 390 | list_add_rcu(&va->list, &prev->list);
|
| 391 | } else
|
| 392 | list_add_rcu(&va->list, &vmap_area_list);
|
| 393 | }
|
| 394 |
|
| 395 | static void purge_vmap_area_lazy(void);
|
| 396 |
|
| 397 | static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
|
| 398 |
|
| 399 | |
| 400 | * Allocate a region of KVA of the specified size and alignment, within the
|
| 401 | * vstart and vend.
|
| 402 |
|
| 403 | static struct vmap_area *alloc_vmap_area(unsigned long size,
|
| 404 | unsigned long align,
|
| 405 | unsigned long vstart, unsigned long vend,
|
| 406 | int node, gfp_t gfp_mask)
|
| 407 | {
|
| 408 | struct vmap_area *va;
|
| 409 | struct rb_node *n;
|
| 410 | unsigned long addr;
|
| 411 | int purged = 0;
|
| 412 | struct vmap_area *first;
|
| 413 |
|
| 414 | BUG_ON(!size);
|
| 415 | BUG_ON(offset_in_page(size));
|
| 416 | BUG_ON(!is_power_of_2(align));
|
| 417 |
|
| 418 | might_sleep();
|
| 419 |
|
| 420 | va = kmalloc_node(sizeof(struct vmap_area),
|
| 421 | gfp_mask & GFP_RECLAIM_MASK, node);
|
| 422 | if (unlikely(!va))
|
| 423 | return ERR_PTR(-ENOMEM);
|
| 424 |
|
| 425 | |
| 426 | * Only scan the relevant parts containing pointers to other objects
|
| 427 | * to avoid false negatives.
|
| 428 |
|
| 429 | kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
|
| 430 |
|
| 431 | retry:
|
| 432 | spin_lock(&vmap_area_lock);
|
| 433 | |
| 434 | * Invalidate cache if we have more permissive parameters.
|
| 435 | * cached_hole_size notes the largest hole noticed _below_
|
| 436 | * the vmap_area cached in free_vmap_cache: if size fits
|
| 437 | * into that hole, we want to scan from vstart to reuse
|
| 438 | * the hole instead of allocating above free_vmap_cache.
|
| 439 | * Note that __free_vmap_area may update free_vmap_cache
|
| 440 | * without updating cached_hole_size or cached_align.
|
| 441 |
|
| 442 | if (!free_vmap_cache ||
|
| 443 | size < cached_hole_size ||
|
| 444 | vstart < cached_vstart ||
|
| 445 | align < cached_align) {
|
| 446 | nocache:
|
| 447 | cached_hole_size = 0;
|
| 448 | free_vmap_cache = NULL;
|
| 449 | }
|
| 450 |
|
| 451 | cached_vstart = vstart;
|
| 452 | cached_align = align;
|
| 453 |
|
| 454 |
|
| 455 | if (free_vmap_cache) {
|
| 456 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
|
| 457 | addr = ALIGN(first->va_end, align);
|
| 458 | if (addr < vstart)
|
| 459 | goto nocache;
|
| 460 | if (addr + size < addr)
|
| 461 | goto overflow;
|
| 462 |
|
| 463 | } else {
|
| 464 | addr = ALIGN(vstart, align);
|
| 465 | if (addr + size < addr)
|
| 466 | goto overflow;
|
| 467 |
|
| 468 | n = vmap_area_root.rb_node;
|
| 469 | first = NULL;
|
| 470 |
|
| 471 | while (n) {
|
| 472 | struct vmap_area *tmp;
|
| 473 | tmp = rb_entry(n, struct vmap_area, rb_node);
|
| 474 | if (tmp->va_end >= addr) {
|
| 475 | first = tmp;
|
| 476 | if (tmp->va_start <= addr)
|
| 477 | break;
|
| 478 | n = n->rb_left;
|
| 479 | } else
|
| 480 | n = n->rb_right;
|
| 481 | }
|
| 482 |
|
| 483 | if (!first)
|
| 484 | goto found;
|
| 485 | }
|
| 486 |
|
| 487 |
|
| 488 | while (addr + size > first->va_start && addr + size <= vend) {
|
| 489 | if (addr + cached_hole_size < first->va_start)
|
| 490 | cached_hole_size = first->va_start - addr;
|
| 491 | addr = ALIGN(first->va_end, align);
|
| 492 | if (addr + size < addr)
|
| 493 | goto overflow;
|
| 494 |
|
| 495 | if (list_is_last(&first->list, &vmap_area_list))
|
| 496 | goto found;
|
| 497 |
|
| 498 | first = list_next_entry(first, list);
|
| 499 | }
|
| 500 |
|
| 501 | found:
|
| 502 | if (addr + size > vend)
|
| 503 | goto overflow;
|
| 504 |
|
| 505 | va->va_start = addr;
|
| 506 | va->va_end = addr + size;
|
| 507 | va->flags = 0;
|
| 508 | __insert_vmap_area(va);
|
| 509 | free_vmap_cache = &va->rb_node;
|
| 510 | spin_unlock(&vmap_area_lock);
|
| 511 |
|
| 512 | BUG_ON(!IS_ALIGNED(va->va_start, align));
|
| 513 | BUG_ON(va->va_start < vstart);
|
| 514 | BUG_ON(va->va_end > vend);
|
| 515 |
|
| 516 | return va;
|
| 517 |
|
| 518 | overflow:
|
| 519 | spin_unlock(&vmap_area_lock);
|
| 520 | if (!purged) {
|
| 521 | purge_vmap_area_lazy();
|
| 522 | purged = 1;
|
| 523 | goto retry;
|
| 524 | }
|
| 525 |
|
| 526 | if (gfpflags_allow_blocking(gfp_mask)) {
|
| 527 | unsigned long freed = 0;
|
| 528 | blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
|
| 529 | if (freed > 0) {
|
| 530 | purged = 0;
|
| 531 | goto retry;
|
| 532 | }
|
| 533 | }
|
| 534 |
|
| 535 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
|
| 536 | pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
|
| 537 | size);
|
| 538 | kfree(va);
|
| 539 | return ERR_PTR(-EBUSY);
|
| 540 | }
|
| 541 |
|
| 542 | int register_vmap_purge_notifier(struct notifier_block *nb)
|
| 543 | {
|
| 544 | return blocking_notifier_chain_register(&vmap_notify_list, nb);
|
| 545 | }
|
| 546 | EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
|
| 547 |
|
| 548 | int unregister_vmap_purge_notifier(struct notifier_block *nb)
|
| 549 | {
|
| 550 | return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
|
| 551 | }
|
| 552 | EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
|
| 553 |
|
| 554 | static void __free_vmap_area(struct vmap_area *va)
|
| 555 | {
|
| 556 | BUG_ON(RB_EMPTY_NODE(&va->rb_node));
|
| 557 |
|
| 558 | if (free_vmap_cache) {
|
| 559 | if (va->va_end < cached_vstart) {
|
| 560 | free_vmap_cache = NULL;
|
| 561 | } else {
|
| 562 | struct vmap_area *cache;
|
| 563 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
|
| 564 | if (va->va_start <= cache->va_start) {
|
| 565 | free_vmap_cache = rb_prev(&va->rb_node);
|
| 566 | |
| 567 | * We don't try to update cached_hole_size or
|
| 568 | * cached_align, but it won't go very wrong.
|
| 569 |
|
| 570 | }
|
| 571 | }
|
| 572 | }
|
| 573 | rb_erase(&va->rb_node, &vmap_area_root);
|
| 574 | RB_CLEAR_NODE(&va->rb_node);
|
| 575 | list_del_rcu(&va->list);
|
| 576 |
|
| 577 | |
| 578 | * Track the highest possible candidate for pcpu area
|
| 579 | * allocation. Areas outside of vmalloc area can be returned
|
| 580 | * here too, consider only end addresses which fall inside
|
| 581 | * vmalloc area proper.
|
| 582 |
|
| 583 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
|
| 584 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
|
| 585 |
|
| 586 | kfree_rcu(va, rcu_head);
|
| 587 | }
|
| 588 |
|
| 589 | |
| 590 | * Free a region of KVA allocated by alloc_vmap_area
|
| 591 |
|
| 592 | static void free_vmap_area(struct vmap_area *va)
|
| 593 | {
|
| 594 | spin_lock(&vmap_area_lock);
|
| 595 | __free_vmap_area(va);
|
| 596 | spin_unlock(&vmap_area_lock);
|
| 597 | }
|
| 598 |
|
| 599 | |
| 600 | * Clear the pagetable entries of a given vmap_area
|
| 601 |
|
| 602 | static void unmap_vmap_area(struct vmap_area *va)
|
| 603 | {
|
| 604 | vunmap_page_range(va->va_start, va->va_end);
|
| 605 | }
|
| 606 |
|
| 607 | static void vmap_debug_free_range(unsigned long start, unsigned long end)
|
| 608 | {
|
| 609 | |
| 610 | * Unmap page tables and force a TLB flush immediately if pagealloc
|
| 611 | * debugging is enabled. This catches use after free bugs similarly to
|
| 612 | * those in linear kernel virtual address space after a page has been
|
| 613 | * freed.
|
| 614 | *
|
| 615 | * All the lazy freeing logic is still retained, in order to minimise
|
| 616 | * intrusiveness of this debugging feature.
|
| 617 | *
|
| 618 | * This is going to be *slow* (linear kernel virtual address debugging
|
| 619 | * doesn't do a broadcast TLB flush so it is a lot faster).
|
| 620 |
|
| 621 | if (debug_pagealloc_enabled()) {
|
| 622 | vunmap_page_range(start, end);
|
| 623 | flush_tlb_kernel_range(start, end);
|
| 624 | }
|
| 625 | }
|
| 626 |
|
| 627 | |
| 628 | * lazy_max_pages is the maximum amount of virtual address space we gather up
|
| 629 | * before attempting to purge with a TLB flush.
|
| 630 | *
|
| 631 | * There is a tradeoff here: a larger number will cover more kernel page tables
|
| 632 | * and take slightly longer to purge, but it will linearly reduce the number of
|
| 633 | * global TLB flushes that must be performed. It would seem natural to scale
|
| 634 | * this number up linearly with the number of CPUs (because vmapping activity
|
| 635 | * could also scale linearly with the number of CPUs), however it is likely
|
| 636 | * that in practice, workloads might be constrained in other ways that mean
|
| 637 | * vmap activity will not scale linearly with CPUs. Also, I want to be
|
| 638 | * conservative and not introduce a big latency on huge systems, so go with
|
| 639 | * a less aggressive log scale. It will still be an improvement over the old
|
| 640 | * code, and it will be simple to change the scale factor if we find that it
|
| 641 | * becomes a problem on bigger systems.
|
| 642 |
|
| 643 | static unsigned long lazy_max_pages(void)
|
| 644 | {
|
| 645 | unsigned int log;
|
| 646 |
|
| 647 | log = fls(num_online_cpus());
|
| 648 |
|
| 649 | return log * (32UL * 1024 * 1024 / PAGE_SIZE);
|
| 650 | }
|
| 651 |
|
| 652 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
|
| 653 |
|
| 654 | |
| 655 | * Serialize vmap purging. There is no actual criticial section protected
|
| 656 | * by this look, but we want to avoid concurrent calls for performance
|
| 657 | * reasons and to make the pcpu_get_vm_areas more deterministic.
|
| 658 |
|
| 659 | static DEFINE_MUTEX(vmap_purge_lock);
|
| 660 |
|
| 661 |
|
| 662 | static void purge_fragmented_blocks_allcpus(void);
|
| 663 |
|
| 664 | |
| 665 | * called before a call to iounmap() if the caller wants vm_area_struct's
|
| 666 | * immediately freed.
|
| 667 |
|
| 668 | void set_iounmap_nonlazy(void)
|
| 669 | {
|
| 670 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
|
| 671 | }
|
| 672 |
|
| 673 | |
| 674 | * Purges all lazily-freed vmap areas.
|
| 675 |
|
| 676 | static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
|
| 677 | {
|
| 678 | struct llist_node *valist;
|
| 679 | struct vmap_area *va;
|
| 680 | struct vmap_area *n_va;
|
| 681 | bool do_free = false;
|
| 682 |
|
| 683 | lockdep_assert_held(&vmap_purge_lock);
|
| 684 |
|
| 685 | valist = llist_del_all(&vmap_purge_list);
|
| 686 | llist_for_each_entry(va, valist, purge_list) {
|
| 687 | if (va->va_start < start)
|
| 688 | start = va->va_start;
|
| 689 | if (va->va_end > end)
|
| 690 | end = va->va_end;
|
| 691 | do_free = true;
|
| 692 | }
|
| 693 |
|
| 694 | if (!do_free)
|
| 695 | return false;
|
| 696 |
|
| 697 | flush_tlb_kernel_range(start, end);
|
| 698 |
|
| 699 | spin_lock(&vmap_area_lock);
|
| 700 | llist_for_each_entry_safe(va, n_va, valist, purge_list) {
|
| 701 | int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
|
| 702 |
|
| 703 | __free_vmap_area(va);
|
| 704 | atomic_sub(nr, &vmap_lazy_nr);
|
| 705 | cond_resched_lock(&vmap_area_lock);
|
| 706 | }
|
| 707 | spin_unlock(&vmap_area_lock);
|
| 708 | return true;
|
| 709 | }
|
| 710 |
|
| 711 | |
| 712 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
|
| 713 | * is already purging.
|
| 714 |
|
| 715 | static void try_purge_vmap_area_lazy(void)
|
| 716 | {
|
| 717 | if (mutex_trylock(&vmap_purge_lock)) {
|
| 718 | __purge_vmap_area_lazy(ULONG_MAX, 0);
|
| 719 | mutex_unlock(&vmap_purge_lock);
|
| 720 | }
|
| 721 | }
|
| 722 |
|
| 723 | |
| 724 | * Kick off a purge of the outstanding lazy areas.
|
| 725 |
|
| 726 | static void purge_vmap_area_lazy(void)
|
| 727 | {
|
| 728 | mutex_lock(&vmap_purge_lock);
|
| 729 | purge_fragmented_blocks_allcpus();
|
| 730 | __purge_vmap_area_lazy(ULONG_MAX, 0);
|
| 731 | mutex_unlock(&vmap_purge_lock);
|
| 732 | }
|
| 733 |
|
| 734 | |
| 735 | * Free a vmap area, caller ensuring that the area has been unmapped
|
| 736 | * and flush_cache_vunmap had been called for the correct range
|
| 737 | * previously.
|
| 738 |
|
| 739 | static void free_vmap_area_noflush(struct vmap_area *va)
|
| 740 | {
|
| 741 | int nr_lazy;
|
| 742 |
|
| 743 | nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
|
| 744 | &vmap_lazy_nr);
|
| 745 |
|
| 746 |
|
| 747 | llist_add(&va->purge_list, &vmap_purge_list);
|
| 748 |
|
| 749 | if (unlikely(nr_lazy > lazy_max_pages()))
|
| 750 | try_purge_vmap_area_lazy();
|
| 751 | }
|
| 752 |
|
| 753 | |
| 754 | * Free and unmap a vmap area
|
| 755 |
|
| 756 | static void free_unmap_vmap_area(struct vmap_area *va)
|
| 757 | {
|
| 758 | flush_cache_vunmap(va->va_start, va->va_end);
|
| 759 | unmap_vmap_area(va);
|
| 760 | free_vmap_area_noflush(va);
|
| 761 | }
|
| 762 |
|
| 763 | static struct vmap_area *find_vmap_area(unsigned long addr)
|
| 764 | {
|
| 765 | struct vmap_area *va;
|
| 766 |
|
| 767 | spin_lock(&vmap_area_lock);
|
| 768 | va = __find_vmap_area(addr);
|
| 769 | spin_unlock(&vmap_area_lock);
|
| 770 |
|
| 771 | return va;
|
| 772 | }
|
| 773 |
|
| 774 |
|
| 775 |
|
| 776 | |
| 777 | * vmap space is limited especially on 32 bit architectures. Ensure there is
|
| 778 | * room for at least 16 percpu vmap blocks per CPU.
|
| 779 |
|
| 780 | |
| 781 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
|
| 782 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
|
| 783 | * instead (we just need a rough idea)
|
| 784 |
|
| 785 | #if BITS_PER_LONG == 32
|
| 786 | #define VMALLOC_SPACE (128UL*1024*1024)
|
| 787 | #else
|
| 788 | #define VMALLOC_SPACE (128UL*1024*1024*1024)
|
| 789 | #endif
|
| 790 |
|
| 791 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
|
| 792 | #define VMAP_MAX_ALLOC BITS_PER_LONG
|
| 793 | #define VMAP_BBMAP_BITS_MAX 1024
|
| 794 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
|
| 795 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y))
|
| 796 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y))
|
| 797 | #define VMAP_BBMAP_BITS \
|
| 798 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
|
| 799 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
|
| 800 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
|
| 801 |
|
| 802 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
|
| 803 |
|
| 804 | static bool vmap_initialized __read_mostly = false;
|
| 805 |
|
| 806 | struct vmap_block_queue {
|
| 807 | spinlock_t lock;
|
| 808 | struct list_head free;
|
| 809 | };
|
| 810 |
|
| 811 | struct vmap_block {
|
| 812 | spinlock_t lock;
|
| 813 | struct vmap_area *va;
|
| 814 | unsigned long free, dirty;
|
| 815 | unsigned long dirty_min, dirty_max;
|
| 816 | struct list_head free_list;
|
| 817 | struct rcu_head rcu_head;
|
| 818 | struct list_head purge;
|
| 819 | };
|
| 820 |
|
| 821 |
|
| 822 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
|
| 823 |
|
| 824 | |
| 825 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
|
| 826 | * in the free path. Could get rid of this if we change the API to return a
|
| 827 | * "cookie" from alloc, to be passed to free. But no big deal yet.
|
| 828 |
|
| 829 | static DEFINE_SPINLOCK(vmap_block_tree_lock);
|
| 830 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
|
| 831 |
|
| 832 | |
| 833 | * We should probably have a fallback mechanism to allocate virtual memory
|
| 834 | * out of partially filled vmap blocks. However vmap block sizing should be
|
| 835 | * fairly reasonable according to the vmalloc size, so it shouldn't be a
|
| 836 | * big problem.
|
| 837 |
|
| 838 |
|
| 839 | static unsigned long addr_to_vb_idx(unsigned long addr)
|
| 840 | {
|
| 841 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
|
| 842 | addr /= VMAP_BLOCK_SIZE;
|
| 843 | return addr;
|
| 844 | }
|
| 845 |
|
| 846 | static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
|
| 847 | {
|
| 848 | unsigned long addr;
|
| 849 |
|
| 850 | addr = va_start + (pages_off << PAGE_SHIFT);
|
| 851 | BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
|
| 852 | return (void *)addr;
|
| 853 | }
|
| 854 |
|
| 855 | |
| 856 | * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
|
| 857 | * block. Of course pages number can't exceed VMAP_BBMAP_BITS
|
| 858 | * @order: how many 2^order pages should be occupied in newly allocated block
|
| 859 | * @gfp_mask: flags for the page level allocator
|
| 860 | *
|
| 861 | * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
|
| 862 |
|
| 863 | static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
|
| 864 | {
|
| 865 | struct vmap_block_queue *vbq;
|
| 866 | struct vmap_block *vb;
|
| 867 | struct vmap_area *va;
|
| 868 | unsigned long vb_idx;
|
| 869 | int node, err;
|
| 870 | void *vaddr;
|
| 871 |
|
| 872 | node = numa_node_id();
|
| 873 |
|
| 874 | vb = kmalloc_node(sizeof(struct vmap_block),
|
| 875 | gfp_mask & GFP_RECLAIM_MASK, node);
|
| 876 | if (unlikely(!vb))
|
| 877 | return ERR_PTR(-ENOMEM);
|
| 878 |
|
| 879 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
|
| 880 | VMALLOC_START, VMALLOC_END,
|
| 881 | node, gfp_mask);
|
| 882 | if (IS_ERR(va)) {
|
| 883 | kfree(vb);
|
| 884 | return ERR_CAST(va);
|
| 885 | }
|
| 886 |
|
| 887 | err = radix_tree_preload(gfp_mask);
|
| 888 | if (unlikely(err)) {
|
| 889 | kfree(vb);
|
| 890 | free_vmap_area(va);
|
| 891 | return ERR_PTR(err);
|
| 892 | }
|
| 893 |
|
| 894 | vaddr = vmap_block_vaddr(va->va_start, 0);
|
| 895 | spin_lock_init(&vb->lock);
|
| 896 | vb->va = va;
|
| 897 |
|
| 898 | BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
|
| 899 | vb->free = VMAP_BBMAP_BITS - (1UL << order);
|
| 900 | vb->dirty = 0;
|
| 901 | vb->dirty_min = VMAP_BBMAP_BITS;
|
| 902 | vb->dirty_max = 0;
|
| 903 | INIT_LIST_HEAD(&vb->free_list);
|
| 904 |
|
| 905 | vb_idx = addr_to_vb_idx(va->va_start);
|
| 906 | spin_lock(&vmap_block_tree_lock);
|
| 907 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
|
| 908 | spin_unlock(&vmap_block_tree_lock);
|
| 909 | BUG_ON(err);
|
| 910 | radix_tree_preload_end();
|
| 911 |
|
| 912 | vbq = &get_cpu_var(vmap_block_queue);
|
| 913 | spin_lock(&vbq->lock);
|
| 914 | list_add_tail_rcu(&vb->free_list, &vbq->free);
|
| 915 | spin_unlock(&vbq->lock);
|
| 916 | put_cpu_var(vmap_block_queue);
|
| 917 |
|
| 918 | return vaddr;
|
| 919 | }
|
| 920 |
|
| 921 | static void free_vmap_block(struct vmap_block *vb)
|
| 922 | {
|
| 923 | struct vmap_block *tmp;
|
| 924 | unsigned long vb_idx;
|
| 925 |
|
| 926 | vb_idx = addr_to_vb_idx(vb->va->va_start);
|
| 927 | spin_lock(&vmap_block_tree_lock);
|
| 928 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
|
| 929 | spin_unlock(&vmap_block_tree_lock);
|
| 930 | BUG_ON(tmp != vb);
|
| 931 |
|
| 932 | free_vmap_area_noflush(vb->va);
|
| 933 | kfree_rcu(vb, rcu_head);
|
| 934 | }
|
| 935 |
|
| 936 | static void purge_fragmented_blocks(int cpu)
|
| 937 | {
|
| 938 | LIST_HEAD(purge);
|
| 939 | struct vmap_block *vb;
|
| 940 | struct vmap_block *n_vb;
|
| 941 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
|
| 942 |
|
| 943 | rcu_read_lock();
|
| 944 | list_for_each_entry_rcu(vb, &vbq->free, free_list) {
|
| 945 |
|
| 946 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
|
| 947 | continue;
|
| 948 |
|
| 949 | spin_lock(&vb->lock);
|
| 950 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
|
| 951 | vb->free = 0;
|
| 952 | vb->dirty = VMAP_BBMAP_BITS;
|
| 953 | vb->dirty_min = 0;
|
| 954 | vb->dirty_max = VMAP_BBMAP_BITS;
|
| 955 | spin_lock(&vbq->lock);
|
| 956 | list_del_rcu(&vb->free_list);
|
| 957 | spin_unlock(&vbq->lock);
|
| 958 | spin_unlock(&vb->lock);
|
| 959 | list_add_tail(&vb->purge, &purge);
|
| 960 | } else
|
| 961 | spin_unlock(&vb->lock);
|
| 962 | }
|
| 963 | rcu_read_unlock();
|
| 964 |
|
| 965 | list_for_each_entry_safe(vb, n_vb, &purge, purge) {
|
| 966 | list_del(&vb->purge);
|
| 967 | free_vmap_block(vb);
|
| 968 | }
|
| 969 | }
|
| 970 |
|
| 971 | static void purge_fragmented_blocks_allcpus(void)
|
| 972 | {
|
| 973 | int cpu;
|
| 974 |
|
| 975 | for_each_possible_cpu(cpu)
|
| 976 | purge_fragmented_blocks(cpu);
|
| 977 | }
|
| 978 |
|
| 979 | static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
|
| 980 | {
|
| 981 | struct vmap_block_queue *vbq;
|
| 982 | struct vmap_block *vb;
|
| 983 | void *vaddr = NULL;
|
| 984 | unsigned int order;
|
| 985 |
|
| 986 | BUG_ON(offset_in_page(size));
|
| 987 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
|
| 988 | if (WARN_ON(size == 0)) {
|
| 989 | |
| 990 | * Allocating 0 bytes isn't what caller wants since
|
| 991 | * get_order(0) returns funny result. Just warn and terminate
|
| 992 | * early.
|
| 993 |
|
| 994 | return NULL;
|
| 995 | }
|
| 996 | order = get_order(size);
|
| 997 |
|
| 998 | rcu_read_lock();
|
| 999 | vbq = &get_cpu_var(vmap_block_queue);
|
| 1000 | list_for_each_entry_rcu(vb, &vbq->free, free_list) {
|
| 1001 | unsigned long pages_off;
|
| 1002 |
|
| 1003 | spin_lock(&vb->lock);
|
| 1004 | if (vb->free < (1UL << order)) {
|
| 1005 | spin_unlock(&vb->lock);
|
| 1006 | continue;
|
| 1007 | }
|
| 1008 |
|
| 1009 | pages_off = VMAP_BBMAP_BITS - vb->free;
|
| 1010 | vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
|
| 1011 | vb->free -= 1UL << order;
|
| 1012 | if (vb->free == 0) {
|
| 1013 | spin_lock(&vbq->lock);
|
| 1014 | list_del_rcu(&vb->free_list);
|
| 1015 | spin_unlock(&vbq->lock);
|
| 1016 | }
|
| 1017 |
|
| 1018 | spin_unlock(&vb->lock);
|
| 1019 | break;
|
| 1020 | }
|
| 1021 |
|
| 1022 | put_cpu_var(vmap_block_queue);
|
| 1023 | rcu_read_unlock();
|
| 1024 |
|
| 1025 |
|
| 1026 | if (!vaddr)
|
| 1027 | vaddr = new_vmap_block(order, gfp_mask);
|
| 1028 |
|
| 1029 | return vaddr;
|
| 1030 | }
|
| 1031 |
|
| 1032 | static void vb_free(const void *addr, unsigned long size)
|
| 1033 | {
|
| 1034 | unsigned long offset;
|
| 1035 | unsigned long vb_idx;
|
| 1036 | unsigned int order;
|
| 1037 | struct vmap_block *vb;
|
| 1038 |
|
| 1039 | BUG_ON(offset_in_page(size));
|
| 1040 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
|
| 1041 |
|
| 1042 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
|
| 1043 |
|
| 1044 | order = get_order(size);
|
| 1045 |
|
| 1046 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
|
| 1047 | offset >>= PAGE_SHIFT;
|
| 1048 |
|
| 1049 | vb_idx = addr_to_vb_idx((unsigned long)addr);
|
| 1050 | rcu_read_lock();
|
| 1051 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
|
| 1052 | rcu_read_unlock();
|
| 1053 | BUG_ON(!vb);
|
| 1054 |
|
| 1055 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
|
| 1056 |
|
| 1057 | spin_lock(&vb->lock);
|
| 1058 |
|
| 1059 |
|
| 1060 | vb->dirty_min = min(vb->dirty_min, offset);
|
| 1061 | vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
|
| 1062 |
|
| 1063 | vb->dirty += 1UL << order;
|
| 1064 | if (vb->dirty == VMAP_BBMAP_BITS) {
|
| 1065 | BUG_ON(vb->free);
|
| 1066 | spin_unlock(&vb->lock);
|
| 1067 | free_vmap_block(vb);
|
| 1068 | } else
|
| 1069 | spin_unlock(&vb->lock);
|
| 1070 | }
|
| 1071 |
|
| 1072 | |
| 1073 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
|
| 1074 | *
|
| 1075 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
|
| 1076 | * to amortize TLB flushing overheads. What this means is that any page you
|
| 1077 | * have now, may, in a former life, have been mapped into kernel virtual
|
| 1078 | * address by the vmap layer and so there might be some CPUs with TLB entries
|
| 1079 | * still referencing that page (additional to the regular 1:1 kernel mapping).
|
| 1080 | *
|
| 1081 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
|
| 1082 | * be sure that none of the pages we have control over will have any aliases
|
| 1083 | * from the vmap layer.
|
| 1084 |
|
| 1085 | void vm_unmap_aliases(void)
|
| 1086 | {
|
| 1087 | unsigned long start = ULONG_MAX, end = 0;
|
| 1088 | int cpu;
|
| 1089 | int flush = 0;
|
| 1090 |
|
| 1091 | if (unlikely(!vmap_initialized))
|
| 1092 | return;
|
| 1093 |
|
| 1094 | might_sleep();
|
| 1095 |
|
| 1096 | for_each_possible_cpu(cpu) {
|
| 1097 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
|
| 1098 | struct vmap_block *vb;
|
| 1099 |
|
| 1100 | rcu_read_lock();
|
| 1101 | list_for_each_entry_rcu(vb, &vbq->free, free_list) {
|
| 1102 | spin_lock(&vb->lock);
|
| 1103 | if (vb->dirty) {
|
| 1104 | unsigned long va_start = vb->va->va_start;
|
| 1105 | unsigned long s, e;
|
| 1106 |
|
| 1107 | s = va_start + (vb->dirty_min << PAGE_SHIFT);
|
| 1108 | e = va_start + (vb->dirty_max << PAGE_SHIFT);
|
| 1109 |
|
| 1110 | start = min(s, start);
|
| 1111 | end = max(e, end);
|
| 1112 |
|
| 1113 | flush = 1;
|
| 1114 | }
|
| 1115 | spin_unlock(&vb->lock);
|
| 1116 | }
|
| 1117 | rcu_read_unlock();
|
| 1118 | }
|
| 1119 |
|
| 1120 | mutex_lock(&vmap_purge_lock);
|
| 1121 | purge_fragmented_blocks_allcpus();
|
| 1122 | if (!__purge_vmap_area_lazy(start, end) && flush)
|
| 1123 | flush_tlb_kernel_range(start, end);
|
| 1124 | mutex_unlock(&vmap_purge_lock);
|
| 1125 | }
|
| 1126 | EXPORT_SYMBOL_GPL(vm_unmap_aliases);
|
| 1127 |
|
| 1128 | |
| 1129 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
|
| 1130 | * @mem: the pointer returned by vm_map_ram
|
| 1131 | * @count: the count passed to that vm_map_ram call (cannot unmap partial)
|
| 1132 |
|
| 1133 | void vm_unmap_ram(const void *mem, unsigned int count)
|
| 1134 | {
|
| 1135 | unsigned long size = (unsigned long)count << PAGE_SHIFT;
|
| 1136 | unsigned long addr = (unsigned long)mem;
|
| 1137 | struct vmap_area *va;
|
| 1138 |
|
| 1139 | might_sleep();
|
| 1140 | BUG_ON(!addr);
|
| 1141 | BUG_ON(addr < VMALLOC_START);
|
| 1142 | BUG_ON(addr > VMALLOC_END);
|
| 1143 | BUG_ON(!PAGE_ALIGNED(addr));
|
| 1144 |
|
| 1145 | debug_check_no_locks_freed(mem, size);
|
| 1146 | vmap_debug_free_range(addr, addr+size);
|
| 1147 |
|
| 1148 | if (likely(count <= VMAP_MAX_ALLOC)) {
|
| 1149 | vb_free(mem, size);
|
| 1150 | return;
|
| 1151 | }
|
| 1152 |
|
| 1153 | va = find_vmap_area(addr);
|
| 1154 | BUG_ON(!va);
|
| 1155 | free_unmap_vmap_area(va);
|
| 1156 | }
|
| 1157 | EXPORT_SYMBOL(vm_unmap_ram);
|
| 1158 |
|
| 1159 | |
| 1160 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
|
| 1161 | * @pages: an array of pointers to the pages to be mapped
|
| 1162 | * @count: number of pages
|
| 1163 | * @node: prefer to allocate data structures on this node
|
| 1164 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM
|
| 1165 | *
|
| 1166 | * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
|
| 1167 | * faster than vmap so it's good. But if you mix long-life and short-life
|
| 1168 | * objects with vm_map_ram(), it could consume lots of address space through
|
| 1169 | * fragmentation (especially on a 32bit machine). You could see failures in
|
| 1170 | * the end. Please use this function for short-lived objects.
|
| 1171 | *
|
| 1172 | * Returns: a pointer to the address that has been mapped, or %NULL on failure
|
| 1173 |
|
| 1174 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
|
| 1175 | {
|
| 1176 | unsigned long size = (unsigned long)count << PAGE_SHIFT;
|
| 1177 | unsigned long addr;
|
| 1178 | void *mem;
|
| 1179 |
|
| 1180 | if (likely(count <= VMAP_MAX_ALLOC)) {
|
| 1181 | mem = vb_alloc(size, GFP_KERNEL);
|
| 1182 | if (IS_ERR(mem))
|
| 1183 | return NULL;
|
| 1184 | addr = (unsigned long)mem;
|
| 1185 | } else {
|
| 1186 | struct vmap_area *va;
|
| 1187 | va = alloc_vmap_area(size, PAGE_SIZE,
|
| 1188 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
|
| 1189 | if (IS_ERR(va))
|
| 1190 | return NULL;
|
| 1191 |
|
| 1192 | addr = va->va_start;
|
| 1193 | mem = (void *)addr;
|
| 1194 | }
|
| 1195 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
|
| 1196 | vm_unmap_ram(mem, count);
|
| 1197 | return NULL;
|
| 1198 | }
|
| 1199 | return mem;
|
| 1200 | }
|
| 1201 | EXPORT_SYMBOL(vm_map_ram);
|
| 1202 |
|
| 1203 | static struct vm_struct *vmlist __initdata;
|
| 1204 | |
| 1205 | * vm_area_add_early - add vmap area early during boot
|
| 1206 | * @vm: vm_struct to add
|
| 1207 | *
|
| 1208 | * This function is used to add fixed kernel vm area to vmlist before
|
| 1209 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
|
| 1210 | * should contain proper values and the other fields should be zero.
|
| 1211 | *
|
| 1212 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
|
| 1213 |
|
| 1214 | void __init vm_area_add_early(struct vm_struct *vm)
|
| 1215 | {
|
| 1216 | struct vm_struct *tmp, **p;
|
| 1217 |
|
| 1218 | BUG_ON(vmap_initialized);
|
| 1219 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
|
| 1220 | if (tmp->addr >= vm->addr) {
|
| 1221 | BUG_ON(tmp->addr < vm->addr + vm->size);
|
| 1222 | break;
|
| 1223 | } else
|
| 1224 | BUG_ON(tmp->addr + tmp->size > vm->addr);
|
| 1225 | }
|
| 1226 | vm->next = *p;
|
| 1227 | *p = vm;
|
| 1228 | }
|
| 1229 |
|
| 1230 | |
| 1231 | * vm_area_register_early - register vmap area early during boot
|
| 1232 | * @vm: vm_struct to register
|
| 1233 | * @align: requested alignment
|
| 1234 | *
|
| 1235 | * This function is used to register kernel vm area before
|
| 1236 | * vmalloc_init() is called. @vm->size and @vm->flags should contain
|
| 1237 | * proper values on entry and other fields should be zero. On return,
|
| 1238 | * vm->addr contains the allocated address.
|
| 1239 | *
|
| 1240 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
|
| 1241 |
|
| 1242 | void __init vm_area_register_early(struct vm_struct *vm, size_t align)
|
| 1243 | {
|
| 1244 | static size_t vm_init_off __initdata;
|
| 1245 | unsigned long addr;
|
| 1246 |
|
| 1247 | addr = ALIGN(VMALLOC_START + vm_init_off, align);
|
| 1248 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
|
| 1249 |
|
| 1250 | vm->addr = (void *)addr;
|
| 1251 |
|
| 1252 | vm_area_add_early(vm);
|
| 1253 | }
|
| 1254 |
|
| 1255 | void __init vmalloc_init(void)
|
| 1256 | {
|
| 1257 | struct vmap_area *va;
|
| 1258 | struct vm_struct *tmp;
|
| 1259 | int i;
|
| 1260 |
|
| 1261 | for_each_possible_cpu(i) {
|
| 1262 | struct vmap_block_queue *vbq;
|
| 1263 | struct vfree_deferred *p;
|
| 1264 |
|
| 1265 | vbq = &per_cpu(vmap_block_queue, i);
|
| 1266 | spin_lock_init(&vbq->lock);
|
| 1267 | INIT_LIST_HEAD(&vbq->free);
|
| 1268 | p = &per_cpu(vfree_deferred, i);
|
| 1269 | init_llist_head(&p->list);
|
| 1270 | INIT_WORK(&p->wq, free_work);
|
| 1271 | }
|
| 1272 |
|
| 1273 |
|
| 1274 | for (tmp = vmlist; tmp; tmp = tmp->next) {
|
| 1275 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
|
| 1276 | va->flags = VM_VM_AREA;
|
| 1277 | va->va_start = (unsigned long)tmp->addr;
|
| 1278 | va->va_end = va->va_start + tmp->size;
|
| 1279 | va->vm = tmp;
|
| 1280 | __insert_vmap_area(va);
|
| 1281 | }
|
| 1282 |
|
| 1283 | vmap_area_pcpu_hole = VMALLOC_END;
|
| 1284 |
|
| 1285 | vmap_initialized = true;
|
| 1286 | }
|
| 1287 |
|
| 1288 | |
| 1289 | * map_kernel_range_noflush - map kernel VM area with the specified pages
|
| 1290 | * @addr: start of the VM area to map
|
| 1291 | * @size: size of the VM area to map
|
| 1292 | * @prot: page protection flags to use
|
| 1293 | * @pages: pages to map
|
| 1294 | *
|
| 1295 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
|
| 1296 | * specify should have been allocated using get_vm_area() and its
|
| 1297 | * friends.
|
| 1298 | *
|
| 1299 | * NOTE:
|
| 1300 | * This function does NOT do any cache flushing. The caller is
|
| 1301 | * responsible for calling flush_cache_vmap() on to-be-mapped areas
|
| 1302 | * before calling this function.
|
| 1303 | *
|
| 1304 | * RETURNS:
|
| 1305 | * The number of pages mapped on success, -errno on failure.
|
| 1306 |
|
| 1307 | int map_kernel_range_noflush(unsigned long addr, unsigned long size,
|
| 1308 | pgprot_t prot, struct page **pages)
|
| 1309 | {
|
| 1310 | return vmap_page_range_noflush(addr, addr + size, prot, pages);
|
| 1311 | }
|
| 1312 |
|
| 1313 | |
| 1314 | * unmap_kernel_range_noflush - unmap kernel VM area
|
| 1315 | * @addr: start of the VM area to unmap
|
| 1316 | * @size: size of the VM area to unmap
|
| 1317 | *
|
| 1318 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
|
| 1319 | * specify should have been allocated using get_vm_area() and its
|
| 1320 | * friends.
|
| 1321 | *
|
| 1322 | * NOTE:
|
| 1323 | * This function does NOT do any cache flushing. The caller is
|
| 1324 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas
|
| 1325 | * before calling this function and flush_tlb_kernel_range() after.
|
| 1326 |
|
| 1327 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
|
| 1328 | {
|
| 1329 | vunmap_page_range(addr, addr + size);
|
| 1330 | }
|
| 1331 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
|
| 1332 |
|
| 1333 | |
| 1334 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
|
| 1335 | * @addr: start of the VM area to unmap
|
| 1336 | * @size: size of the VM area to unmap
|
| 1337 | *
|
| 1338 | * Similar to unmap_kernel_range_noflush() but flushes vcache before
|
| 1339 | * the unmapping and tlb after.
|
| 1340 |
|
| 1341 | void unmap_kernel_range(unsigned long addr, unsigned long size)
|
| 1342 | {
|
| 1343 | unsigned long end = addr + size;
|
| 1344 |
|
| 1345 | flush_cache_vunmap(addr, end);
|
| 1346 | vunmap_page_range(addr, end);
|
| 1347 | flush_tlb_kernel_range(addr, end);
|
| 1348 | }
|
| 1349 | EXPORT_SYMBOL_GPL(unmap_kernel_range);
|
| 1350 |
|
| 1351 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
|
| 1352 | {
|
| 1353 | unsigned long addr = (unsigned long)area->addr;
|
| 1354 | unsigned long end = addr + get_vm_area_size(area);
|
| 1355 | int err;
|
| 1356 |
|
| 1357 | err = vmap_page_range(addr, end, prot, pages);
|
| 1358 |
|
| 1359 | return err > 0 ? 0 : err;
|
| 1360 | }
|
| 1361 | EXPORT_SYMBOL_GPL(map_vm_area);
|
| 1362 |
|
| 1363 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
|
| 1364 | unsigned long flags, const void *caller)
|
| 1365 | {
|
| 1366 | spin_lock(&vmap_area_lock);
|
| 1367 | vm->flags = flags;
|
| 1368 | vm->addr = (void *)va->va_start;
|
| 1369 | vm->size = va->va_end - va->va_start;
|
| 1370 | vm->caller = caller;
|
| 1371 | va->vm = vm;
|
| 1372 | va->flags |= VM_VM_AREA;
|
| 1373 | spin_unlock(&vmap_area_lock);
|
| 1374 | }
|
| 1375 |
|
| 1376 | static void clear_vm_uninitialized_flag(struct vm_struct *vm)
|
| 1377 | {
|
| 1378 | |
| 1379 | * Before removing VM_UNINITIALIZED,
|
| 1380 | * we should make sure that vm has proper values.
|
| 1381 | * Pair with smp_rmb() in show_numa_info().
|
| 1382 |
|
| 1383 | smp_wmb();
|
| 1384 | vm->flags &= ~VM_UNINITIALIZED;
|
| 1385 | }
|
| 1386 |
|
| 1387 | static struct vm_struct *__get_vm_area_node(unsigned long size,
|
| 1388 | unsigned long align, unsigned long flags, unsigned long start,
|
| 1389 | unsigned long end, int node, gfp_t gfp_mask, const void *caller)
|
| 1390 | {
|
| 1391 | struct vmap_area *va;
|
| 1392 | struct vm_struct *area;
|
| 1393 |
|
| 1394 | BUG_ON(in_interrupt());
|
| 1395 | size = PAGE_ALIGN(size);
|
| 1396 | if (unlikely(!size))
|
| 1397 | return NULL;
|
| 1398 |
|
| 1399 | if (flags & VM_IOREMAP)
|
| 1400 | align = 1ul << clamp_t(int, get_count_order_long(size),
|
| 1401 | PAGE_SHIFT, IOREMAP_MAX_ORDER);
|
| 1402 |
|
| 1403 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
|
| 1404 | if (unlikely(!area))
|
| 1405 | return NULL;
|
| 1406 |
|
| 1407 | if (!(flags & VM_NO_GUARD))
|
| 1408 | size += PAGE_SIZE;
|
| 1409 |
|
| 1410 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
|
| 1411 | if (IS_ERR(va)) {
|
| 1412 | kfree(area);
|
| 1413 | return NULL;
|
| 1414 | }
|
| 1415 |
|
| 1416 | setup_vmalloc_vm(area, va, flags, caller);
|
| 1417 |
|
| 1418 | return area;
|
| 1419 | }
|
| 1420 |
|
| 1421 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
|
| 1422 | unsigned long start, unsigned long end)
|
| 1423 | {
|
| 1424 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
|
| 1425 | GFP_KERNEL, __builtin_return_address(0));
|
| 1426 | }
|
| 1427 | EXPORT_SYMBOL_GPL(__get_vm_area);
|
| 1428 |
|
| 1429 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
|
| 1430 | unsigned long start, unsigned long end,
|
| 1431 | const void *caller)
|
| 1432 | {
|
| 1433 | return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
|
| 1434 | GFP_KERNEL, caller);
|
| 1435 | }
|
| 1436 |
|
| 1437 | |
| 1438 | * get_vm_area - reserve a contiguous kernel virtual area
|
| 1439 | * @size: size of the area
|
| 1440 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
|
| 1441 | *
|
| 1442 | * Search an area of @size in the kernel virtual mapping area,
|
| 1443 | * and reserved it for out purposes. Returns the area descriptor
|
| 1444 | * on success or %NULL on failure.
|
| 1445 |
|
| 1446 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
|
| 1447 | {
|
| 1448 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
|
| 1449 | NUMA_NO_NODE, GFP_KERNEL,
|
| 1450 | __builtin_return_address(0));
|
| 1451 | }
|
| 1452 |
|
| 1453 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
|
| 1454 | const void *caller)
|
| 1455 | {
|
| 1456 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
|
| 1457 | NUMA_NO_NODE, GFP_KERNEL, caller);
|
| 1458 | }
|
| 1459 |
|
| 1460 | |
| 1461 | * find_vm_area - find a continuous kernel virtual area
|
| 1462 | * @addr: base address
|
| 1463 | *
|
| 1464 | * Search for the kernel VM area starting at @addr, and return it.
|
| 1465 | * It is up to the caller to do all required locking to keep the returned
|
| 1466 | * pointer valid.
|
| 1467 |
|
| 1468 | struct vm_struct *find_vm_area(const void *addr)
|
| 1469 | {
|
| 1470 | struct vmap_area *va;
|
| 1471 |
|
| 1472 | va = find_vmap_area((unsigned long)addr);
|
| 1473 | if (va && va->flags & VM_VM_AREA)
|
| 1474 | return va->vm;
|
| 1475 |
|
| 1476 | return NULL;
|
| 1477 | }
|
| 1478 |
|
| 1479 | |
| 1480 | * remove_vm_area - find and remove a continuous kernel virtual area
|
| 1481 | * @addr: base address
|
| 1482 | *
|
| 1483 | * Search for the kernel VM area starting at @addr, and remove it.
|
| 1484 | * This function returns the found VM area, but using it is NOT safe
|
| 1485 | * on SMP machines, except for its size or flags.
|
| 1486 |
|
| 1487 | struct vm_struct *remove_vm_area(const void *addr)
|
| 1488 | {
|
| 1489 | struct vmap_area *va;
|
| 1490 |
|
| 1491 | might_sleep();
|
| 1492 |
|
| 1493 | va = find_vmap_area((unsigned long)addr);
|
| 1494 | if (va && va->flags & VM_VM_AREA) {
|
| 1495 | struct vm_struct *vm = va->vm;
|
| 1496 |
|
| 1497 | spin_lock(&vmap_area_lock);
|
| 1498 | va->vm = NULL;
|
| 1499 | va->flags &= ~VM_VM_AREA;
|
| 1500 | spin_unlock(&vmap_area_lock);
|
| 1501 |
|
| 1502 | vmap_debug_free_range(va->va_start, va->va_end);
|
| 1503 | kasan_free_shadow(vm);
|
| 1504 | free_unmap_vmap_area(va);
|
| 1505 |
|
| 1506 | return vm;
|
| 1507 | }
|
| 1508 | return NULL;
|
| 1509 | }
|
| 1510 |
|
| 1511 | static void __vunmap(const void *addr, int deallocate_pages)
|
| 1512 | {
|
| 1513 | struct vm_struct *area;
|
| 1514 |
|
| 1515 | if (!addr)
|
| 1516 | return;
|
| 1517 |
|
| 1518 | if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
|
| 1519 | addr))
|
| 1520 | return;
|
| 1521 |
|
| 1522 | area = remove_vm_area(addr);
|
| 1523 | if (unlikely(!area)) {
|
| 1524 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
|
| 1525 | addr);
|
| 1526 | return;
|
| 1527 | }
|
| 1528 |
|
| 1529 | debug_check_no_locks_freed(addr, get_vm_area_size(area));
|
| 1530 | debug_check_no_obj_freed(addr, get_vm_area_size(area));
|
| 1531 |
|
| 1532 | if (deallocate_pages) {
|
| 1533 | int i;
|
| 1534 |
|
| 1535 | for (i = 0; i < area->nr_pages; i++) {
|
| 1536 | struct page *page = area->pages[i];
|
| 1537 |
|
| 1538 | BUG_ON(!page);
|
| 1539 | __free_pages(page, 0);
|
| 1540 | }
|
| 1541 |
|
| 1542 | kvfree(area->pages);
|
| 1543 | }
|
| 1544 |
|
| 1545 | kfree(area);
|
| 1546 | return;
|
| 1547 | }
|
| 1548 |
|
| 1549 | static inline void __vfree_deferred(const void *addr)
|
| 1550 | {
|
| 1551 | |
| 1552 | * Use raw_cpu_ptr() because this can be called from preemptible
|
| 1553 | * context. Preemption is absolutely fine here, because the llist_add()
|
| 1554 | * implementation is lockless, so it works even if we are adding to
|
| 1555 | * nother cpu's list. schedule_work() should be fine with this too.
|
| 1556 |
|
| 1557 | struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
|
| 1558 |
|
| 1559 | if (llist_add((struct llist_node *)addr, &p->list))
|
| 1560 | schedule_work(&p->wq);
|
| 1561 | }
|
| 1562 |
|
| 1563 | |
| 1564 | * vfree_atomic - release memory allocated by vmalloc()
|
| 1565 | * @addr: memory base address
|
| 1566 | *
|
| 1567 | * This one is just like vfree() but can be called in any atomic context
|
| 1568 | * except NMIs.
|
| 1569 |
|
| 1570 | void vfree_atomic(const void *addr)
|
| 1571 | {
|
| 1572 | BUG_ON(in_nmi());
|
| 1573 |
|
| 1574 | kmemleak_free(addr);
|
| 1575 |
|
| 1576 | if (!addr)
|
| 1577 | return;
|
| 1578 | __vfree_deferred(addr);
|
| 1579 | }
|
| 1580 |
|
| 1581 | |
| 1582 | * vfree - release memory allocated by vmalloc()
|
| 1583 | * @addr: memory base address
|
| 1584 | *
|
| 1585 | * Free the virtually continuous memory area starting at @addr, as
|
| 1586 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
|
| 1587 | * NULL, no operation is performed.
|
| 1588 | *
|
| 1589 | * Must not be called in NMI context (strictly speaking, only if we don't
|
| 1590 | * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
|
| 1591 | * conventions for vfree() arch-depenedent would be a really bad idea)
|
| 1592 | *
|
| 1593 | * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
|
| 1594 |
|
| 1595 | void vfree(const void *addr)
|
| 1596 | {
|
| 1597 | BUG_ON(in_nmi());
|
| 1598 |
|
| 1599 | kmemleak_free(addr);
|
| 1600 |
|
| 1601 | if (!addr)
|
| 1602 | return;
|
| 1603 | if (unlikely(in_interrupt()))
|
| 1604 | __vfree_deferred(addr);
|
| 1605 | else
|
| 1606 | __vunmap(addr, 1);
|
| 1607 | }
|
| 1608 | EXPORT_SYMBOL(vfree);
|
| 1609 |
|
| 1610 | |
| 1611 | * vunmap - release virtual mapping obtained by vmap()
|
| 1612 | * @addr: memory base address
|
| 1613 | *
|
| 1614 | * Free the virtually contiguous memory area starting at @addr,
|
| 1615 | * which was created from the page array passed to vmap().
|
| 1616 | *
|
| 1617 | * Must not be called in interrupt context.
|
| 1618 |
|
| 1619 | void vunmap(const void *addr)
|
| 1620 | {
|
| 1621 | BUG_ON(in_interrupt());
|
| 1622 | might_sleep();
|
| 1623 | if (addr)
|
| 1624 | __vunmap(addr, 0);
|
| 1625 | }
|
| 1626 | EXPORT_SYMBOL(vunmap);
|
| 1627 |
|
| 1628 | |
| 1629 | * vmap - map an array of pages into virtually contiguous space
|
| 1630 | * @pages: array of page pointers
|
| 1631 | * @count: number of pages to map
|
| 1632 | * @flags: vm_area->flags
|
| 1633 | * @prot: page protection for the mapping
|
| 1634 | *
|
| 1635 | * Maps @count pages from @pages into contiguous kernel virtual
|
| 1636 | * space.
|
| 1637 |
|
| 1638 | void *vmap(struct page **pages, unsigned int count,
|
| 1639 | unsigned long flags, pgprot_t prot)
|
| 1640 | {
|
| 1641 | struct vm_struct *area;
|
| 1642 | unsigned long size;
|
| 1643 |
|
| 1644 | might_sleep();
|
| 1645 |
|
| 1646 | if (count > totalram_pages)
|
| 1647 | return NULL;
|
| 1648 |
|
| 1649 | size = (unsigned long)count << PAGE_SHIFT;
|
| 1650 | area = get_vm_area_caller(size, flags, __builtin_return_address(0));
|
| 1651 | if (!area)
|
| 1652 | return NULL;
|
| 1653 |
|
| 1654 | if (map_vm_area(area, prot, pages)) {
|
| 1655 | vunmap(area->addr);
|
| 1656 | return NULL;
|
| 1657 | }
|
| 1658 |
|
| 1659 | return area->addr;
|
| 1660 | }
|
| 1661 | EXPORT_SYMBOL(vmap);
|
| 1662 |
|
| 1663 | static void *__vmalloc_node(unsigned long size, unsigned long align,
|
| 1664 | gfp_t gfp_mask, pgprot_t prot,
|
| 1665 | int node, const void *caller);
|
| 1666 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
|
| 1667 | pgprot_t prot, int node)
|
| 1668 | {
|
| 1669 | struct page **pages;
|
| 1670 | unsigned int nr_pages, array_size, i;
|
| 1671 | const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
|
| 1672 | const gfp_t alloc_mask = gfp_mask | __GFP_HIGHMEM | __GFP_NOWARN;
|
| 1673 |
|
| 1674 | nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
|
| 1675 | array_size = (nr_pages * sizeof(struct page *));
|
| 1676 |
|
| 1677 | area->nr_pages = nr_pages;
|
| 1678 |
|
| 1679 | if (array_size > PAGE_SIZE) {
|
| 1680 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
|
| 1681 | PAGE_KERNEL, node, area->caller);
|
| 1682 | } else {
|
| 1683 | pages = kmalloc_node(array_size, nested_gfp, node);
|
| 1684 | }
|
| 1685 | area->pages = pages;
|
| 1686 | if (!area->pages) {
|
| 1687 | remove_vm_area(area->addr);
|
| 1688 | kfree(area);
|
| 1689 | return NULL;
|
| 1690 | }
|
| 1691 |
|
| 1692 | for (i = 0; i < area->nr_pages; i++) {
|
| 1693 | struct page *page;
|
| 1694 |
|
| 1695 | if (fatal_signal_pending(current)) {
|
| 1696 | area->nr_pages = i;
|
| 1697 | goto fail_no_warn;
|
| 1698 | }
|
| 1699 |
|
| 1700 | if (node == NUMA_NO_NODE)
|
| 1701 | page = alloc_page(alloc_mask);
|
| 1702 | else
|
| 1703 | page = alloc_pages_node(node, alloc_mask, 0);
|
| 1704 |
|
| 1705 | if (unlikely(!page)) {
|
| 1706 |
|
| 1707 | area->nr_pages = i;
|
| 1708 | goto fail;
|
| 1709 | }
|
| 1710 | area->pages[i] = page;
|
| 1711 | if (gfpflags_allow_blocking(gfp_mask))
|
| 1712 | cond_resched();
|
| 1713 | }
|
| 1714 |
|
| 1715 | if (map_vm_area(area, prot, pages))
|
| 1716 | goto fail;
|
| 1717 | return area->addr;
|
| 1718 |
|
| 1719 | fail:
|
| 1720 | warn_alloc(gfp_mask, NULL,
|
| 1721 | "vmalloc: allocation failure, allocated %ld of %ld bytes",
|
| 1722 | (area->nr_pages*PAGE_SIZE), area->size);
|
| 1723 | fail_no_warn:
|
| 1724 | vfree(area->addr);
|
| 1725 | return NULL;
|
| 1726 | }
|
| 1727 |
|
| 1728 | |
| 1729 | * __vmalloc_node_range - allocate virtually contiguous memory
|
| 1730 | * @size: allocation size
|
| 1731 | * @align: desired alignment
|
| 1732 | * @start: vm area range start
|
| 1733 | * @end: vm area range end
|
| 1734 | * @gfp_mask: flags for the page level allocator
|
| 1735 | * @prot: protection mask for the allocated pages
|
| 1736 | * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
|
| 1737 | * @node: node to use for allocation or NUMA_NO_NODE
|
| 1738 | * @caller: caller's return address
|
| 1739 | *
|
| 1740 | * Allocate enough pages to cover @size from the page level
|
| 1741 | * allocator with @gfp_mask flags. Map them into contiguous
|
| 1742 | * kernel virtual space, using a pagetable protection of @prot.
|
| 1743 |
|
| 1744 | void *__vmalloc_node_range(unsigned long size, unsigned long align,
|
| 1745 | unsigned long start, unsigned long end, gfp_t gfp_mask,
|
| 1746 | pgprot_t prot, unsigned long vm_flags, int node,
|
| 1747 | const void *caller)
|
| 1748 | {
|
| 1749 | struct vm_struct *area;
|
| 1750 | void *addr;
|
| 1751 | unsigned long real_size = size;
|
| 1752 |
|
| 1753 | size = PAGE_ALIGN(size);
|
| 1754 | if (!size || (size >> PAGE_SHIFT) > totalram_pages)
|
| 1755 | goto fail;
|
| 1756 |
|
| 1757 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
|
| 1758 | vm_flags, start, end, node, gfp_mask, caller);
|
| 1759 | if (!area)
|
| 1760 | goto fail;
|
| 1761 |
|
| 1762 | addr = __vmalloc_area_node(area, gfp_mask, prot, node);
|
| 1763 | if (!addr)
|
| 1764 | return NULL;
|
| 1765 |
|
| 1766 | |
| 1767 | * In this function, newly allocated vm_struct has VM_UNINITIALIZED
|
| 1768 | * flag. It means that vm_struct is not fully initialized.
|
| 1769 | * Now, it is fully initialized, so remove this flag here.
|
| 1770 |
|
| 1771 | clear_vm_uninitialized_flag(area);
|
| 1772 |
|
| 1773 | |
| 1774 | * A ref_count = 2 is needed because vm_struct allocated in
|
| 1775 | * __get_vm_area_node() contains a reference to the virtual address of
|
| 1776 | * the vmalloc'ed block.
|
| 1777 |
|
| 1778 | kmemleak_alloc(addr, real_size, 2, gfp_mask);
|
| 1779 |
|
| 1780 | return addr;
|
| 1781 |
|
| 1782 | fail:
|
| 1783 | warn_alloc(gfp_mask, NULL,
|
| 1784 | "vmalloc: allocation failure: %lu bytes", real_size);
|
| 1785 | return NULL;
|
| 1786 | }
|
| 1787 |
|
| 1788 | |
| 1789 | * __vmalloc_node - allocate virtually contiguous memory
|
| 1790 | * @size: allocation size
|
| 1791 | * @align: desired alignment
|
| 1792 | * @gfp_mask: flags for the page level allocator
|
| 1793 | * @prot: protection mask for the allocated pages
|
| 1794 | * @node: node to use for allocation or NUMA_NO_NODE
|
| 1795 | * @caller: caller's return address
|
| 1796 | *
|
| 1797 | * Allocate enough pages to cover @size from the page level
|
| 1798 | * allocator with @gfp_mask flags. Map them into contiguous
|
| 1799 | * kernel virtual space, using a pagetable protection of @prot.
|
| 1800 | *
|
| 1801 | * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_REPEAT
|
| 1802 | * and __GFP_NOFAIL are not supported
|
| 1803 | *
|
| 1804 | * Any use of gfp flags outside of GFP_KERNEL should be consulted
|
| 1805 | * with mm people.
|
| 1806 | *
|
| 1807 |
|
| 1808 | static void *__vmalloc_node(unsigned long size, unsigned long align,
|
| 1809 | gfp_t gfp_mask, pgprot_t prot,
|
| 1810 | int node, const void *caller)
|
| 1811 | {
|
| 1812 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
|
| 1813 | gfp_mask, prot, 0, node, caller);
|
| 1814 | }
|
| 1815 |
|
| 1816 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
|
| 1817 | {
|
| 1818 | return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
|
| 1819 | __builtin_return_address(0));
|
| 1820 | }
|
| 1821 | EXPORT_SYMBOL(__vmalloc);
|
| 1822 |
|
| 1823 | static inline void *__vmalloc_node_flags(unsigned long size,
|
| 1824 | int node, gfp_t flags)
|
| 1825 | {
|
| 1826 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
|
| 1827 | node, __builtin_return_address(0));
|
| 1828 | }
|
| 1829 |
|
| 1830 |
|
| 1831 | void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
|
| 1832 | void *caller)
|
| 1833 | {
|
| 1834 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
|
| 1835 | }
|
| 1836 |
|
| 1837 | |
| 1838 | * vmalloc - allocate virtually contiguous memory
|
| 1839 | * @size: allocation size
|
| 1840 | * Allocate enough pages to cover @size from the page level
|
| 1841 | * allocator and map them into contiguous kernel virtual space.
|
| 1842 | *
|
| 1843 | * For tight control over page level allocator and protection flags
|
| 1844 | * use __vmalloc() instead.
|
| 1845 |
|
| 1846 | void *vmalloc(unsigned long size)
|
| 1847 | {
|
| 1848 | return __vmalloc_node_flags(size, NUMA_NO_NODE,
|
| 1849 | GFP_KERNEL);
|
| 1850 | }
|
| 1851 | EXPORT_SYMBOL(vmalloc);
|
| 1852 |
|
| 1853 | |
| 1854 | * vzalloc - allocate virtually contiguous memory with zero fill
|
| 1855 | * @size: allocation size
|
| 1856 | * Allocate enough pages to cover @size from the page level
|
| 1857 | * allocator and map them into contiguous kernel virtual space.
|
| 1858 | * The memory allocated is set to zero.
|
| 1859 | *
|
| 1860 | * For tight control over page level allocator and protection flags
|
| 1861 | * use __vmalloc() instead.
|
| 1862 |
|
| 1863 | void *vzalloc(unsigned long size)
|
| 1864 | {
|
| 1865 | return __vmalloc_node_flags(size, NUMA_NO_NODE,
|
| 1866 | GFP_KERNEL | __GFP_ZERO);
|
| 1867 | }
|
| 1868 | EXPORT_SYMBOL(vzalloc);
|
| 1869 |
|
| 1870 | |
| 1871 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
|
| 1872 | * @size: allocation size
|
| 1873 | *
|
| 1874 | * The resulting memory area is zeroed so it can be mapped to userspace
|
| 1875 | * without leaking data.
|
| 1876 |
|
| 1877 | void *vmalloc_user(unsigned long size)
|
| 1878 | {
|
| 1879 | struct vm_struct *area;
|
| 1880 | void *ret;
|
| 1881 |
|
| 1882 | ret = __vmalloc_node(size, SHMLBA,
|
| 1883 | GFP_KERNEL | __GFP_ZERO,
|
| 1884 | PAGE_KERNEL, NUMA_NO_NODE,
|
| 1885 | __builtin_return_address(0));
|
| 1886 | if (ret) {
|
| 1887 | area = find_vm_area(ret);
|
| 1888 | area->flags |= VM_USERMAP;
|
| 1889 | }
|
| 1890 | return ret;
|
| 1891 | }
|
| 1892 | EXPORT_SYMBOL(vmalloc_user);
|
| 1893 |
|
| 1894 | |
| 1895 | * vmalloc_node - allocate memory on a specific node
|
| 1896 | * @size: allocation size
|
| 1897 | * @node: numa node
|
| 1898 | *
|
| 1899 | * Allocate enough pages to cover @size from the page level
|
| 1900 | * allocator and map them into contiguous kernel virtual space.
|
| 1901 | *
|
| 1902 | * For tight control over page level allocator and protection flags
|
| 1903 | * use __vmalloc() instead.
|
| 1904 |
|
| 1905 | void *vmalloc_node(unsigned long size, int node)
|
| 1906 | {
|
| 1907 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
|
| 1908 | node, __builtin_return_address(0));
|
| 1909 | }
|
| 1910 | EXPORT_SYMBOL(vmalloc_node);
|
| 1911 |
|
| 1912 | |
| 1913 | * vzalloc_node - allocate memory on a specific node with zero fill
|
| 1914 | * @size: allocation size
|
| 1915 | * @node: numa node
|
| 1916 | *
|
| 1917 | * Allocate enough pages to cover @size from the page level
|
| 1918 | * allocator and map them into contiguous kernel virtual space.
|
| 1919 | * The memory allocated is set to zero.
|
| 1920 | *
|
| 1921 | * For tight control over page level allocator and protection flags
|
| 1922 | * use __vmalloc_node() instead.
|
| 1923 |
|
| 1924 | void *vzalloc_node(unsigned long size, int node)
|
| 1925 | {
|
| 1926 | return __vmalloc_node_flags(size, node,
|
| 1927 | GFP_KERNEL | __GFP_ZERO);
|
| 1928 | }
|
| 1929 | EXPORT_SYMBOL(vzalloc_node);
|
| 1930 |
|
| 1931 | #ifndef PAGE_KERNEL_EXEC
|
| 1932 | # define PAGE_KERNEL_EXEC PAGE_KERNEL
|
| 1933 | #endif
|
| 1934 |
|
| 1935 | |
| 1936 | * vmalloc_exec - allocate virtually contiguous, executable memory
|
| 1937 | * @size: allocation size
|
| 1938 | *
|
| 1939 | * Kernel-internal function to allocate enough pages to cover @size
|
| 1940 | * the page level allocator and map them into contiguous and
|
| 1941 | * executable kernel virtual space.
|
| 1942 | *
|
| 1943 | * For tight control over page level allocator and protection flags
|
| 1944 | * use __vmalloc() instead.
|
| 1945 |
|
| 1946 |
|
| 1947 | void *vmalloc_exec(unsigned long size)
|
| 1948 | {
|
| 1949 | return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
|
| 1950 | NUMA_NO_NODE, __builtin_return_address(0));
|
| 1951 | }
|
| 1952 |
|
| 1953 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
|
| 1954 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
|
| 1955 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
|
| 1956 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
|
| 1957 | #else
|
| 1958 | #define GFP_VMALLOC32 GFP_KERNEL
|
| 1959 | #endif
|
| 1960 |
|
| 1961 | |
| 1962 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
|
| 1963 | * @size: allocation size
|
| 1964 | *
|
| 1965 | * Allocate enough 32bit PA addressable pages to cover @size from the
|
| 1966 | * page level allocator and map them into contiguous kernel virtual space.
|
| 1967 |
|
| 1968 | void *vmalloc_32(unsigned long size)
|
| 1969 | {
|
| 1970 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
|
| 1971 | NUMA_NO_NODE, __builtin_return_address(0));
|
| 1972 | }
|
| 1973 | EXPORT_SYMBOL(vmalloc_32);
|
| 1974 |
|
| 1975 | |
| 1976 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
|
| 1977 | * @size: allocation size
|
| 1978 | *
|
| 1979 | * The resulting memory area is 32bit addressable and zeroed so it can be
|
| 1980 | * mapped to userspace without leaking data.
|
| 1981 |
|
| 1982 | void *vmalloc_32_user(unsigned long size)
|
| 1983 | {
|
| 1984 | struct vm_struct *area;
|
| 1985 | void *ret;
|
| 1986 |
|
| 1987 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
|
| 1988 | NUMA_NO_NODE, __builtin_return_address(0));
|
| 1989 | if (ret) {
|
| 1990 | area = find_vm_area(ret);
|
| 1991 | area->flags |= VM_USERMAP;
|
| 1992 | }
|
| 1993 | return ret;
|
| 1994 | }
|
| 1995 | EXPORT_SYMBOL(vmalloc_32_user);
|
| 1996 |
|
| 1997 | |
| 1998 | * small helper routine , copy contents to buf from addr.
|
| 1999 | * If the page is not present, fill zero.
|
| 2000 |
|
| 2001 |
|
| 2002 | static int aligned_vread(char *buf, char *addr, unsigned long count)
|
| 2003 | {
|
| 2004 | struct page *p;
|
| 2005 | int copied = 0;
|
| 2006 |
|
| 2007 | while (count) {
|
| 2008 | unsigned long offset, length;
|
| 2009 |
|
| 2010 | offset = offset_in_page(addr);
|
| 2011 | length = PAGE_SIZE - offset;
|
| 2012 | if (length > count)
|
| 2013 | length = count;
|
| 2014 | p = vmalloc_to_page(addr);
|
| 2015 | |
| 2016 | * To do safe access to this _mapped_ area, we need
|
| 2017 | * lock. But adding lock here means that we need to add
|
| 2018 | * overhead of vmalloc()/vfree() calles for this _debug_
|
| 2019 | * interface, rarely used. Instead of that, we'll use
|
| 2020 | * kmap() and get small overhead in this access function.
|
| 2021 |
|
| 2022 | if (p) {
|
| 2023 | |
| 2024 | * we can expect USER0 is not used (see vread/vwrite's
|
| 2025 | * function description)
|
| 2026 |
|
| 2027 | void *map = kmap_atomic(p);
|
| 2028 | memcpy(buf, map + offset, length);
|
| 2029 | kunmap_atomic(map);
|
| 2030 | } else
|
| 2031 | memset(buf, 0, length);
|
| 2032 |
|
| 2033 | addr += length;
|
| 2034 | buf += length;
|
| 2035 | copied += length;
|
| 2036 | count -= length;
|
| 2037 | }
|
| 2038 | return copied;
|
| 2039 | }
|
| 2040 |
|
| 2041 | static int aligned_vwrite(char *buf, char *addr, unsigned long count)
|
| 2042 | {
|
| 2043 | struct page *p;
|
| 2044 | int copied = 0;
|
| 2045 |
|
| 2046 | while (count) {
|
| 2047 | unsigned long offset, length;
|
| 2048 |
|
| 2049 | offset = offset_in_page(addr);
|
| 2050 | length = PAGE_SIZE - offset;
|
| 2051 | if (length > count)
|
| 2052 | length = count;
|
| 2053 | p = vmalloc_to_page(addr);
|
| 2054 | |
| 2055 | * To do safe access to this _mapped_ area, we need
|
| 2056 | * lock. But adding lock here means that we need to add
|
| 2057 | * overhead of vmalloc()/vfree() calles for this _debug_
|
| 2058 | * interface, rarely used. Instead of that, we'll use
|
| 2059 | * kmap() and get small overhead in this access function.
|
| 2060 |
|
| 2061 | if (p) {
|
| 2062 | |
| 2063 | * we can expect USER0 is not used (see vread/vwrite's
|
| 2064 | * function description)
|
| 2065 |
|
| 2066 | void *map = kmap_atomic(p);
|
| 2067 | memcpy(map + offset, buf, length);
|
| 2068 | kunmap_atomic(map);
|
| 2069 | }
|
| 2070 | addr += length;
|
| 2071 | buf += length;
|
| 2072 | copied += length;
|
| 2073 | count -= length;
|
| 2074 | }
|
| 2075 | return copied;
|
| 2076 | }
|
| 2077 |
|
| 2078 | |
| 2079 | * vread() - read vmalloc area in a safe way.
|
| 2080 | * @buf: buffer for reading data
|
| 2081 | * @addr: vm address.
|
| 2082 | * @count: number of bytes to be read.
|
| 2083 | *
|
| 2084 | * Returns # of bytes which addr and buf should be increased.
|
| 2085 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't
|
| 2086 | * includes any intersect with alive vmalloc area.
|
| 2087 | *
|
| 2088 | * This function checks that addr is a valid vmalloc'ed area, and
|
| 2089 | * copy data from that area to a given buffer. If the given memory range
|
| 2090 | * of [addr...addr+count) includes some valid address, data is copied to
|
| 2091 | * proper area of @buf. If there are memory holes, they'll be zero-filled.
|
| 2092 | * IOREMAP area is treated as memory hole and no copy is done.
|
| 2093 | *
|
| 2094 | * If [addr...addr+count) doesn't includes any intersects with alive
|
| 2095 | * vm_struct area, returns 0. @buf should be kernel's buffer.
|
| 2096 | *
|
| 2097 | * Note: In usual ops, vread() is never necessary because the caller
|
| 2098 | * should know vmalloc() area is valid and can use memcpy().
|
| 2099 | * This is for routines which have to access vmalloc area without
|
| 2100 | * any informaion, as /dev/kmem.
|
| 2101 | *
|
| 2102 |
|
| 2103 |
|
| 2104 | long vread(char *buf, char *addr, unsigned long count)
|
| 2105 | {
|
| 2106 | struct vmap_area *va;
|
| 2107 | struct vm_struct *vm;
|
| 2108 | char *vaddr, *buf_start = buf;
|
| 2109 | unsigned long buflen = count;
|
| 2110 | unsigned long n;
|
| 2111 |
|
| 2112 |
|
| 2113 | if ((unsigned long) addr + count < count)
|
| 2114 | count = -(unsigned long) addr;
|
| 2115 |
|
| 2116 | spin_lock(&vmap_area_lock);
|
| 2117 | list_for_each_entry(va, &vmap_area_list, list) {
|
| 2118 | if (!count)
|
| 2119 | break;
|
| 2120 |
|
| 2121 | if (!(va->flags & VM_VM_AREA))
|
| 2122 | continue;
|
| 2123 |
|
| 2124 | vm = va->vm;
|
| 2125 | vaddr = (char *) vm->addr;
|
| 2126 | if (addr >= vaddr + get_vm_area_size(vm))
|
| 2127 | continue;
|
| 2128 | while (addr < vaddr) {
|
| 2129 | if (count == 0)
|
| 2130 | goto finished;
|
| 2131 | *buf = '\0';
|
| 2132 | buf++;
|
| 2133 | addr++;
|
| 2134 | count--;
|
| 2135 | }
|
| 2136 | n = vaddr + get_vm_area_size(vm) - addr;
|
| 2137 | if (n > count)
|
| 2138 | n = count;
|
| 2139 | if (!(vm->flags & VM_IOREMAP))
|
| 2140 | aligned_vread(buf, addr, n);
|
| 2141 | else
|
| 2142 | memset(buf, 0, n);
|
| 2143 | buf += n;
|
| 2144 | addr += n;
|
| 2145 | count -= n;
|
| 2146 | }
|
| 2147 | finished:
|
| 2148 | spin_unlock(&vmap_area_lock);
|
| 2149 |
|
| 2150 | if (buf == buf_start)
|
| 2151 | return 0;
|
| 2152 |
|
| 2153 | if (buf != buf_start + buflen)
|
| 2154 | memset(buf, 0, buflen - (buf - buf_start));
|
| 2155 |
|
| 2156 | return buflen;
|
| 2157 | }
|
| 2158 |
|
| 2159 | |
| 2160 | * vwrite() - write vmalloc area in a safe way.
|
| 2161 | * @buf: buffer for source data
|
| 2162 | * @addr: vm address.
|
| 2163 | * @count: number of bytes to be read.
|
| 2164 | *
|
| 2165 | * Returns # of bytes which addr and buf should be incresed.
|
| 2166 | * (same number to @count).
|
| 2167 | * If [addr...addr+count) doesn't includes any intersect with valid
|
| 2168 | * vmalloc area, returns 0.
|
| 2169 | *
|
| 2170 | * This function checks that addr is a valid vmalloc'ed area, and
|
| 2171 | * copy data from a buffer to the given addr. If specified range of
|
| 2172 | * [addr...addr+count) includes some valid address, data is copied from
|
| 2173 | * proper area of @buf. If there are memory holes, no copy to hole.
|
| 2174 | * IOREMAP area is treated as memory hole and no copy is done.
|
| 2175 | *
|
| 2176 | * If [addr...addr+count) doesn't includes any intersects with alive
|
| 2177 | * vm_struct area, returns 0. @buf should be kernel's buffer.
|
| 2178 | *
|
| 2179 | * Note: In usual ops, vwrite() is never necessary because the caller
|
| 2180 | * should know vmalloc() area is valid and can use memcpy().
|
| 2181 | * This is for routines which have to access vmalloc area without
|
| 2182 | * any informaion, as /dev/kmem.
|
| 2183 |
|
| 2184 |
|
| 2185 | long vwrite(char *buf, char *addr, unsigned long count)
|
| 2186 | {
|
| 2187 | struct vmap_area *va;
|
| 2188 | struct vm_struct *vm;
|
| 2189 | char *vaddr;
|
| 2190 | unsigned long n, buflen;
|
| 2191 | int copied = 0;
|
| 2192 |
|
| 2193 |
|
| 2194 | if ((unsigned long) addr + count < count)
|
| 2195 | count = -(unsigned long) addr;
|
| 2196 | buflen = count;
|
| 2197 |
|
| 2198 | spin_lock(&vmap_area_lock);
|
| 2199 | list_for_each_entry(va, &vmap_area_list, list) {
|
| 2200 | if (!count)
|
| 2201 | break;
|
| 2202 |
|
| 2203 | if (!(va->flags & VM_VM_AREA))
|
| 2204 | continue;
|
| 2205 |
|
| 2206 | vm = va->vm;
|
| 2207 | vaddr = (char *) vm->addr;
|
| 2208 | if (addr >= vaddr + get_vm_area_size(vm))
|
| 2209 | continue;
|
| 2210 | while (addr < vaddr) {
|
| 2211 | if (count == 0)
|
| 2212 | goto finished;
|
| 2213 | buf++;
|
| 2214 | addr++;
|
| 2215 | count--;
|
| 2216 | }
|
| 2217 | n = vaddr + get_vm_area_size(vm) - addr;
|
| 2218 | if (n > count)
|
| 2219 | n = count;
|
| 2220 | if (!(vm->flags & VM_IOREMAP)) {
|
| 2221 | aligned_vwrite(buf, addr, n);
|
| 2222 | copied++;
|
| 2223 | }
|
| 2224 | buf += n;
|
| 2225 | addr += n;
|
| 2226 | count -= n;
|
| 2227 | }
|
| 2228 | finished:
|
| 2229 | spin_unlock(&vmap_area_lock);
|
| 2230 | if (!copied)
|
| 2231 | return 0;
|
| 2232 | return buflen;
|
| 2233 | }
|
| 2234 |
|
| 2235 | |
| 2236 | * remap_vmalloc_range_partial - map vmalloc pages to userspace
|
| 2237 | * @vma: vma to cover
|
| 2238 | * @uaddr: target user address to start at
|
| 2239 | * @kaddr: virtual address of vmalloc kernel memory
|
| 2240 | * @size: size of map area
|
| 2241 | *
|
| 2242 | * Returns: 0 for success, -Exxx on failure
|
| 2243 | *
|
| 2244 | * This function checks that @kaddr is a valid vmalloc'ed area,
|
| 2245 | * and that it is big enough to cover the range starting at
|
| 2246 | * @uaddr in @vma. Will return failure if that criteria isn't
|
| 2247 | * met.
|
| 2248 | *
|
| 2249 | * Similar to remap_pfn_range() (see mm/memory.c)
|
| 2250 |
|
| 2251 | int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
|
| 2252 | void *kaddr, unsigned long size)
|
| 2253 | {
|
| 2254 | struct vm_struct *area;
|
| 2255 |
|
| 2256 | size = PAGE_ALIGN(size);
|
| 2257 |
|
| 2258 | if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
|
| 2259 | return -EINVAL;
|
| 2260 |
|
| 2261 | area = find_vm_area(kaddr);
|
| 2262 | if (!area)
|
| 2263 | return -EINVAL;
|
| 2264 |
|
| 2265 | if (!(area->flags & VM_USERMAP))
|
| 2266 | return -EINVAL;
|
| 2267 |
|
| 2268 | if (kaddr + size > area->addr + area->size)
|
| 2269 | return -EINVAL;
|
| 2270 |
|
| 2271 | do {
|
| 2272 | struct page *page = vmalloc_to_page(kaddr);
|
| 2273 | int ret;
|
| 2274 |
|
| 2275 | ret = vm_insert_page(vma, uaddr, page);
|
| 2276 | if (ret)
|
| 2277 | return ret;
|
| 2278 |
|
| 2279 | uaddr += PAGE_SIZE;
|
| 2280 | kaddr += PAGE_SIZE;
|
| 2281 | size -= PAGE_SIZE;
|
| 2282 | } while (size > 0);
|
| 2283 |
|
| 2284 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
|
| 2285 |
|
| 2286 | return 0;
|
| 2287 | }
|
| 2288 | EXPORT_SYMBOL(remap_vmalloc_range_partial);
|
| 2289 |
|
| 2290 | |
| 2291 | * remap_vmalloc_range - map vmalloc pages to userspace
|
| 2292 | * @vma: vma to cover (map full range of vma)
|
| 2293 | * @addr: vmalloc memory
|
| 2294 | * @pgoff: number of pages into addr before first page to map
|
| 2295 | *
|
| 2296 | * Returns: 0 for success, -Exxx on failure
|
| 2297 | *
|
| 2298 | * This function checks that addr is a valid vmalloc'ed area, and
|
| 2299 | * that it is big enough to cover the vma. Will return failure if
|
| 2300 | * that criteria isn't met.
|
| 2301 | *
|
| 2302 | * Similar to remap_pfn_range() (see mm/memory.c)
|
| 2303 |
|
| 2304 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
|
| 2305 | unsigned long pgoff)
|
| 2306 | {
|
| 2307 | return remap_vmalloc_range_partial(vma, vma->vm_start,
|
| 2308 | addr + (pgoff << PAGE_SHIFT),
|
| 2309 | vma->vm_end - vma->vm_start);
|
| 2310 | }
|
| 2311 | EXPORT_SYMBOL(remap_vmalloc_range);
|
| 2312 |
|
| 2313 | |
| 2314 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to
|
| 2315 | * have one.
|
| 2316 |
|
| 2317 | void __weak vmalloc_sync_all(void)
|
| 2318 | {
|
| 2319 | }
|
| 2320 |
|
| 2321 |
|
| 2322 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
|
| 2323 | {
|
| 2324 | pte_t ***p = data;
|
| 2325 |
|
| 2326 | if (p) {
|
| 2327 | *(*p) = pte;
|
| 2328 | (*p)++;
|
| 2329 | }
|
| 2330 | return 0;
|
| 2331 | }
|
| 2332 |
|
| 2333 | |
| 2334 | * alloc_vm_area - allocate a range of kernel address space
|
| 2335 | * @size: size of the area
|
| 2336 | * @ptes: returns the PTEs for the address space
|
| 2337 | *
|
| 2338 | * Returns: NULL on failure, vm_struct on success
|
| 2339 | *
|
| 2340 | * This function reserves a range of kernel address space, and
|
| 2341 | * allocates pagetables to map that range. No actual mappings
|
| 2342 | * are created.
|
| 2343 | *
|
| 2344 | * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
|
| 2345 | * allocated for the VM area are returned.
|
| 2346 |
|
| 2347 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
|
| 2348 | {
|
| 2349 | struct vm_struct *area;
|
| 2350 |
|
| 2351 | area = get_vm_area_caller(size, VM_IOREMAP,
|
| 2352 | __builtin_return_address(0));
|
| 2353 | if (area == NULL)
|
| 2354 | return NULL;
|
| 2355 |
|
| 2356 | |
| 2357 | * This ensures that page tables are constructed for this region
|
| 2358 | * of kernel virtual address space and mapped into init_mm.
|
| 2359 |
|
| 2360 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
|
| 2361 | size, f, ptes ? &ptes : NULL)) {
|
| 2362 | free_vm_area(area);
|
| 2363 | return NULL;
|
| 2364 | }
|
| 2365 |
|
| 2366 | return area;
|
| 2367 | }
|
| 2368 | EXPORT_SYMBOL_GPL(alloc_vm_area);
|
| 2369 |
|
| 2370 | void free_vm_area(struct vm_struct *area)
|
| 2371 | {
|
| 2372 | struct vm_struct *ret;
|
| 2373 | ret = remove_vm_area(area->addr);
|
| 2374 | BUG_ON(ret != area);
|
| 2375 | kfree(area);
|
| 2376 | }
|
| 2377 | EXPORT_SYMBOL_GPL(free_vm_area);
|
| 2378 |
|
| 2379 | #ifdef CONFIG_SMP
|
| 2380 | static struct vmap_area *node_to_va(struct rb_node *n)
|
| 2381 | {
|
| 2382 | return rb_entry_safe(n, struct vmap_area, rb_node);
|
| 2383 | }
|
| 2384 |
|
| 2385 | |
| 2386 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
|
| 2387 | * @end: target address
|
| 2388 | * @pnext: out arg for the next vmap_area
|
| 2389 | * @pprev: out arg for the previous vmap_area
|
| 2390 | *
|
| 2391 | * Returns: %true if either or both of next and prev are found,
|
| 2392 | * %false if no vmap_area exists
|
| 2393 | *
|
| 2394 | * Find vmap_areas end addresses of which enclose @end. ie. if not
|
| 2395 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
|
| 2396 |
|
| 2397 | static bool pvm_find_next_prev(unsigned long end,
|
| 2398 | struct vmap_area **pnext,
|
| 2399 | struct vmap_area **pprev)
|
| 2400 | {
|
| 2401 | struct rb_node *n = vmap_area_root.rb_node;
|
| 2402 | struct vmap_area *va = NULL;
|
| 2403 |
|
| 2404 | while (n) {
|
| 2405 | va = rb_entry(n, struct vmap_area, rb_node);
|
| 2406 | if (end < va->va_end)
|
| 2407 | n = n->rb_left;
|
| 2408 | else if (end > va->va_end)
|
| 2409 | n = n->rb_right;
|
| 2410 | else
|
| 2411 | break;
|
| 2412 | }
|
| 2413 |
|
| 2414 | if (!va)
|
| 2415 | return false;
|
| 2416 |
|
| 2417 | if (va->va_end > end) {
|
| 2418 | *pnext = va;
|
| 2419 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
|
| 2420 | } else {
|
| 2421 | *pprev = va;
|
| 2422 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
|
| 2423 | }
|
| 2424 | return true;
|
| 2425 | }
|
| 2426 |
|
| 2427 | |
| 2428 | * pvm_determine_end - find the highest aligned address between two vmap_areas
|
| 2429 | * @pnext: in/out arg for the next vmap_area
|
| 2430 | * @pprev: in/out arg for the previous vmap_area
|
| 2431 | * @align: alignment
|
| 2432 | *
|
| 2433 | * Returns: determined end address
|
| 2434 | *
|
| 2435 | * Find the highest aligned address between *@pnext and *@pprev below
|
| 2436 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
|
| 2437 | * down address is between the end addresses of the two vmap_areas.
|
| 2438 | *
|
| 2439 | * Please note that the address returned by this function may fall
|
| 2440 | * inside *@pnext vmap_area. The caller is responsible for checking
|
| 2441 | * that.
|
| 2442 |
|
| 2443 | static unsigned long pvm_determine_end(struct vmap_area **pnext,
|
| 2444 | struct vmap_area **pprev,
|
| 2445 | unsigned long align)
|
| 2446 | {
|
| 2447 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
|
| 2448 | unsigned long addr;
|
| 2449 |
|
| 2450 | if (*pnext)
|
| 2451 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
|
| 2452 | else
|
| 2453 | addr = vmalloc_end;
|
| 2454 |
|
| 2455 | while (*pprev && (*pprev)->va_end > addr) {
|
| 2456 | *pnext = *pprev;
|
| 2457 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
|
| 2458 | }
|
| 2459 |
|
| 2460 | return addr;
|
| 2461 | }
|
| 2462 |
|
| 2463 | |
| 2464 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
|
| 2465 | * @offsets: array containing offset of each area
|
| 2466 | * @sizes: array containing size of each area
|
| 2467 | * @nr_vms: the number of areas to allocate
|
| 2468 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this
|
| 2469 | *
|
| 2470 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated
|
| 2471 | * vm_structs on success, %NULL on failure
|
| 2472 | *
|
| 2473 | * Percpu allocator wants to use congruent vm areas so that it can
|
| 2474 | * maintain the offsets among percpu areas. This function allocates
|
| 2475 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
|
| 2476 | * be scattered pretty far, distance between two areas easily going up
|
| 2477 | * to gigabytes. To avoid interacting with regular vmallocs, these
|
| 2478 | * areas are allocated from top.
|
| 2479 | *
|
| 2480 | * Despite its complicated look, this allocator is rather simple. It
|
| 2481 | * does everything top-down and scans areas from the end looking for
|
| 2482 | * matching slot. While scanning, if any of the areas overlaps with
|
| 2483 | * existing vmap_area, the base address is pulled down to fit the
|
| 2484 | * area. Scanning is repeated till all the areas fit and then all
|
| 2485 | * necessary data structres are inserted and the result is returned.
|
| 2486 |
|
| 2487 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
|
| 2488 | const size_t *sizes, int nr_vms,
|
| 2489 | size_t align)
|
| 2490 | {
|
| 2491 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
|
| 2492 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
|
| 2493 | struct vmap_area **vas, *prev, *next;
|
| 2494 | struct vm_struct **vms;
|
| 2495 | int area, area2, last_area, term_area;
|
| 2496 | unsigned long base, start, end, last_end;
|
| 2497 | bool purged = false;
|
| 2498 |
|
| 2499 |
|
| 2500 | BUG_ON(offset_in_page(align) || !is_power_of_2(align));
|
| 2501 | for (last_area = 0, area = 0; area < nr_vms; area++) {
|
| 2502 | start = offsets[area];
|
| 2503 | end = start + sizes[area];
|
| 2504 |
|
| 2505 |
|
| 2506 | BUG_ON(!IS_ALIGNED(offsets[area], align));
|
| 2507 | BUG_ON(!IS_ALIGNED(sizes[area], align));
|
| 2508 |
|
| 2509 |
|
| 2510 | if (start > offsets[last_area])
|
| 2511 | last_area = area;
|
| 2512 |
|
| 2513 | for (area2 = 0; area2 < nr_vms; area2++) {
|
| 2514 | unsigned long start2 = offsets[area2];
|
| 2515 | unsigned long end2 = start2 + sizes[area2];
|
| 2516 |
|
| 2517 | if (area2 == area)
|
| 2518 | continue;
|
| 2519 |
|
| 2520 | BUG_ON(start2 >= start && start2 < end);
|
| 2521 | BUG_ON(end2 <= end && end2 > start);
|
| 2522 | }
|
| 2523 | }
|
| 2524 | last_end = offsets[last_area] + sizes[last_area];
|
| 2525 |
|
| 2526 | if (vmalloc_end - vmalloc_start < last_end) {
|
| 2527 | WARN_ON(true);
|
| 2528 | return NULL;
|
| 2529 | }
|
| 2530 |
|
| 2531 | vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
|
| 2532 | vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
|
| 2533 | if (!vas || !vms)
|
| 2534 | goto err_free2;
|
| 2535 |
|
| 2536 | for (area = 0; area < nr_vms; area++) {
|
| 2537 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
|
| 2538 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
|
| 2539 | if (!vas[area] || !vms[area])
|
| 2540 | goto err_free;
|
| 2541 | }
|
| 2542 | retry:
|
| 2543 | spin_lock(&vmap_area_lock);
|
| 2544 |
|
| 2545 |
|
| 2546 | area = term_area = last_area;
|
| 2547 | start = offsets[area];
|
| 2548 | end = start + sizes[area];
|
| 2549 |
|
| 2550 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
|
| 2551 | base = vmalloc_end - last_end;
|
| 2552 | goto found;
|
| 2553 | }
|
| 2554 | base = pvm_determine_end(&next, &prev, align) - end;
|
| 2555 |
|
| 2556 | while (true) {
|
| 2557 | BUG_ON(next && next->va_end <= base + end);
|
| 2558 | BUG_ON(prev && prev->va_end > base + end);
|
| 2559 |
|
| 2560 | |
| 2561 | * base might have underflowed, add last_end before
|
| 2562 | * comparing.
|
| 2563 |
|
| 2564 | if (base + last_end < vmalloc_start + last_end) {
|
| 2565 | spin_unlock(&vmap_area_lock);
|
| 2566 | if (!purged) {
|
| 2567 | purge_vmap_area_lazy();
|
| 2568 | purged = true;
|
| 2569 | goto retry;
|
| 2570 | }
|
| 2571 | goto err_free;
|
| 2572 | }
|
| 2573 |
|
| 2574 | |
| 2575 | * If next overlaps, move base downwards so that it's
|
| 2576 | * right below next and then recheck.
|
| 2577 |
|
| 2578 | if (next && next->va_start < base + end) {
|
| 2579 | base = pvm_determine_end(&next, &prev, align) - end;
|
| 2580 | term_area = area;
|
| 2581 | continue;
|
| 2582 | }
|
| 2583 |
|
| 2584 | |
| 2585 | * If prev overlaps, shift down next and prev and move
|
| 2586 | * base so that it's right below new next and then
|
| 2587 | * recheck.
|
| 2588 |
|
| 2589 | if (prev && prev->va_end > base + start) {
|
| 2590 | next = prev;
|
| 2591 | prev = node_to_va(rb_prev(&next->rb_node));
|
| 2592 | base = pvm_determine_end(&next, &prev, align) - end;
|
| 2593 | term_area = area;
|
| 2594 | continue;
|
| 2595 | }
|
| 2596 |
|
| 2597 | |
| 2598 | * This area fits, move on to the previous one. If
|
| 2599 | * the previous one is the terminal one, we're done.
|
| 2600 |
|
| 2601 | area = (area + nr_vms - 1) % nr_vms;
|
| 2602 | if (area == term_area)
|
| 2603 | break;
|
| 2604 | start = offsets[area];
|
| 2605 | end = start + sizes[area];
|
| 2606 | pvm_find_next_prev(base + end, &next, &prev);
|
| 2607 | }
|
| 2608 | found:
|
| 2609 |
|
| 2610 | for (area = 0; area < nr_vms; area++) {
|
| 2611 | struct vmap_area *va = vas[area];
|
| 2612 |
|
| 2613 | va->va_start = base + offsets[area];
|
| 2614 | va->va_end = va->va_start + sizes[area];
|
| 2615 | __insert_vmap_area(va);
|
| 2616 | }
|
| 2617 |
|
| 2618 | vmap_area_pcpu_hole = base + offsets[last_area];
|
| 2619 |
|
| 2620 | spin_unlock(&vmap_area_lock);
|
| 2621 |
|
| 2622 |
|
| 2623 | for (area = 0; area < nr_vms; area++)
|
| 2624 | setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
|
| 2625 | pcpu_get_vm_areas);
|
| 2626 |
|
| 2627 | kfree(vas);
|
| 2628 | return vms;
|
| 2629 |
|
| 2630 | err_free:
|
| 2631 | for (area = 0; area < nr_vms; area++) {
|
| 2632 | kfree(vas[area]);
|
| 2633 | kfree(vms[area]);
|
| 2634 | }
|
| 2635 | err_free2:
|
| 2636 | kfree(vas);
|
| 2637 | kfree(vms);
|
| 2638 | return NULL;
|
| 2639 | }
|
| 2640 |
|
| 2641 | |
| 2642 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
|
| 2643 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
|
| 2644 | * @nr_vms: the number of allocated areas
|
| 2645 | *
|
| 2646 | * Free vm_structs and the array allocated by pcpu_get_vm_areas().
|
| 2647 |
|
| 2648 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
|
| 2649 | {
|
| 2650 | int i;
|
| 2651 |
|
| 2652 | for (i = 0; i < nr_vms; i++)
|
| 2653 | free_vm_area(vms[i]);
|
| 2654 | kfree(vms);
|
| 2655 | }
|
| 2656 | #endif
|
| 2657 |
|
| 2658 | #ifdef CONFIG_PROC_FS
|
| 2659 | static void *s_start(struct seq_file *m, loff_t *pos)
|
| 2660 | __acquires(&vmap_area_lock)
|
| 2661 | {
|
| 2662 | spin_lock(&vmap_area_lock);
|
| 2663 | return seq_list_start(&vmap_area_list, *pos);
|
| 2664 | }
|
| 2665 |
|
| 2666 | static void *s_next(struct seq_file *m, void *p, loff_t *pos)
|
| 2667 | {
|
| 2668 | return seq_list_next(p, &vmap_area_list, pos);
|
| 2669 | }
|
| 2670 |
|
| 2671 | static void s_stop(struct seq_file *m, void *p)
|
| 2672 | __releases(&vmap_area_lock)
|
| 2673 | {
|
| 2674 | spin_unlock(&vmap_area_lock);
|
| 2675 | }
|
| 2676 |
|
| 2677 | static void show_numa_info(struct seq_file *m, struct vm_struct *v)
|
| 2678 | {
|
| 2679 | if (IS_ENABLED(CONFIG_NUMA)) {
|
| 2680 | unsigned int nr, *counters = m->private;
|
| 2681 |
|
| 2682 | if (!counters)
|
| 2683 | return;
|
| 2684 |
|
| 2685 | if (v->flags & VM_UNINITIALIZED)
|
| 2686 | return;
|
| 2687 |
|
| 2688 | smp_rmb();
|
| 2689 |
|
| 2690 | memset(counters, 0, nr_node_ids * sizeof(unsigned int));
|
| 2691 |
|
| 2692 | for (nr = 0; nr < v->nr_pages; nr++)
|
| 2693 | counters[page_to_nid(v->pages[nr])]++;
|
| 2694 |
|
| 2695 | for_each_node_state(nr, N_HIGH_MEMORY)
|
| 2696 | if (counters[nr])
|
| 2697 | seq_printf(m, " N%u=%u", nr, counters[nr]);
|
| 2698 | }
|
| 2699 | }
|
| 2700 |
|
| 2701 | static int s_show(struct seq_file *m, void *p)
|
| 2702 | {
|
| 2703 | struct vmap_area *va;
|
| 2704 | struct vm_struct *v;
|
| 2705 |
|
| 2706 | va = list_entry(p, struct vmap_area, list);
|
| 2707 |
|
| 2708 | |
| 2709 | * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
|
| 2710 | * behalf of vmap area is being tear down or vm_map_ram allocation.
|
| 2711 |
|
| 2712 | if (!(va->flags & VM_VM_AREA))
|
| 2713 | return 0;
|
| 2714 |
|
| 2715 | v = va->vm;
|
| 2716 |
|
| 2717 | seq_printf(m, "0x%pK-0x%pK %7ld",
|
| 2718 | v->addr, v->addr + v->size, v->size);
|
| 2719 |
|
| 2720 | if (v->caller)
|
| 2721 | seq_printf(m, " %pS", v->caller);
|
| 2722 |
|
| 2723 | if (v->nr_pages)
|
| 2724 | seq_printf(m, " pages=%d", v->nr_pages);
|
| 2725 |
|
| 2726 | if (v->phys_addr)
|
| 2727 | seq_printf(m, " phys=%pa", &v->phys_addr);
|
| 2728 |
|
| 2729 | if (v->flags & VM_IOREMAP)
|
| 2730 | seq_puts(m, " ioremap");
|
| 2731 |
|
| 2732 | if (v->flags & VM_ALLOC)
|
| 2733 | seq_puts(m, " vmalloc");
|
| 2734 |
|
| 2735 | if (v->flags & VM_MAP)
|
| 2736 | seq_puts(m, " vmap");
|
| 2737 |
|
| 2738 | if (v->flags & VM_USERMAP)
|
| 2739 | seq_puts(m, " user");
|
| 2740 |
|
| 2741 | if (is_vmalloc_addr(v->pages))
|
| 2742 | seq_puts(m, " vpages");
|
| 2743 |
|
| 2744 | show_numa_info(m, v);
|
| 2745 | seq_putc(m, '\n');
|
| 2746 | return 0;
|
| 2747 | }
|
| 2748 |
|
| 2749 | static const struct seq_operations vmalloc_op = {
|
| 2750 | .start = s_start,
|
| 2751 | .next = s_next,
|
| 2752 | .stop = s_stop,
|
| 2753 | .show = s_show,
|
| 2754 | };
|
| 2755 |
|
| 2756 | static int vmalloc_open(struct inode *inode, struct file *file)
|
| 2757 | {
|
| 2758 | if (IS_ENABLED(CONFIG_NUMA))
|
| 2759 | return seq_open_private(file, &vmalloc_op,
|
| 2760 | nr_node_ids * sizeof(unsigned int));
|
| 2761 | else
|
| 2762 | return seq_open(file, &vmalloc_op);
|
| 2763 | }
|
| 2764 |
|
| 2765 | static const struct file_operations proc_vmalloc_operations = {
|
| 2766 | .open = vmalloc_open,
|
| 2767 | .read = seq_read,
|
| 2768 | .llseek = seq_lseek,
|
| 2769 | .release = seq_release_private,
|
| 2770 | };
|
| 2771 |
|
| 2772 | static int __init proc_vmalloc_init(void)
|
| 2773 | {
|
| 2774 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
|
| 2775 | return 0;
|
| 2776 | }
|
| 2777 | module_init(proc_vmalloc_init);
|
| 2778 |
|
| 2779 | #endif
|
| 2780 |
|