linux/mm/vmalloc.c
<<
>>
Prefs
   1/*
   2 *  linux/mm/vmalloc.c
   3 *
   4 *  Copyright (C) 1993  Linus Torvalds
   5 *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6 *  SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
   7 *  Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
   8 *  Numa awareness, Christoph Lameter, SGI, June 2005
   9 */
  10
  11#include <linux/vmalloc.h>
  12#include <linux/mm.h>
  13#include <linux/module.h>
  14#include <linux/highmem.h>
  15#include <linux/slab.h>
  16#include <linux/spinlock.h>
  17#include <linux/interrupt.h>
  18#include <linux/proc_fs.h>
  19#include <linux/seq_file.h>
  20#include <linux/debugobjects.h>
  21#include <linux/kallsyms.h>
  22#include <linux/list.h>
  23#include <linux/rbtree.h>
  24#include <linux/radix-tree.h>
  25#include <linux/rcupdate.h>
  26
  27#include <asm/atomic.h>
  28#include <asm/uaccess.h>
  29#include <asm/tlbflush.h>
  30
  31
  32/*** Page table manipulation functions ***/
  33
  34static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  35{
  36        pte_t *pte;
  37
  38        pte = pte_offset_kernel(pmd, addr);
  39        do {
  40                pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  41                WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  42        } while (pte++, addr += PAGE_SIZE, addr != end);
  43}
  44
  45static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
  46{
  47        pmd_t *pmd;
  48        unsigned long next;
  49
  50        pmd = pmd_offset(pud, addr);
  51        do {
  52                next = pmd_addr_end(addr, end);
  53                if (pmd_none_or_clear_bad(pmd))
  54                        continue;
  55                vunmap_pte_range(pmd, addr, next);
  56        } while (pmd++, addr = next, addr != end);
  57}
  58
  59static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
  60{
  61        pud_t *pud;
  62        unsigned long next;
  63
  64        pud = pud_offset(pgd, addr);
  65        do {
  66                next = pud_addr_end(addr, end);
  67                if (pud_none_or_clear_bad(pud))
  68                        continue;
  69                vunmap_pmd_range(pud, addr, next);
  70        } while (pud++, addr = next, addr != end);
  71}
  72
  73static void vunmap_page_range(unsigned long addr, unsigned long end)
  74{
  75        pgd_t *pgd;
  76        unsigned long next;
  77
  78        BUG_ON(addr >= end);
  79        pgd = pgd_offset_k(addr);
  80        do {
  81                next = pgd_addr_end(addr, end);
  82                if (pgd_none_or_clear_bad(pgd))
  83                        continue;
  84                vunmap_pud_range(pgd, addr, next);
  85        } while (pgd++, addr = next, addr != end);
  86}
  87
  88static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  89                unsigned long end, pgprot_t prot, struct page **pages, int *nr)
  90{
  91        pte_t *pte;
  92
  93        /*
  94         * nr is a running index into the array which helps higher level
  95         * callers keep track of where we're up to.
  96         */
  97
  98        pte = pte_alloc_kernel(pmd, addr);
  99        if (!pte)
 100                return -ENOMEM;
 101        do {
 102                struct page *page = pages[*nr];
 103
 104                if (WARN_ON(!pte_none(*pte)))
 105                        return -EBUSY;
 106                if (WARN_ON(!page))
 107                        return -ENOMEM;
 108                set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 109                (*nr)++;
 110        } while (pte++, addr += PAGE_SIZE, addr != end);
 111        return 0;
 112}
 113
 114static int vmap_pmd_range(pud_t *pud, unsigned long addr,
 115                unsigned long end, pgprot_t prot, struct page **pages, int *nr)
 116{
 117        pmd_t *pmd;
 118        unsigned long next;
 119
 120        pmd = pmd_alloc(&init_mm, pud, addr);
 121        if (!pmd)
 122                return -ENOMEM;
 123        do {
 124                next = pmd_addr_end(addr, end);
 125                if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
 126                        return -ENOMEM;
 127        } while (pmd++, addr = next, addr != end);
 128        return 0;
 129}
 130
 131static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 132                unsigned long end, pgprot_t prot, struct page **pages, int *nr)
 133{
 134        pud_t *pud;
 135        unsigned long next;
 136
 137        pud = pud_alloc(&init_mm, pgd, addr);
 138        if (!pud)
 139                return -ENOMEM;
 140        do {
 141                next = pud_addr_end(addr, end);
 142                if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
 143                        return -ENOMEM;
 144        } while (pud++, addr = next, addr != end);
 145        return 0;
 146}
 147
 148/*
 149 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
 150 * will have pfns corresponding to the "pages" array.
 151 *
 152 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
 153 */
 154static int vmap_page_range(unsigned long start, unsigned long end,
 155                                pgprot_t prot, struct page **pages)
 156{
 157        pgd_t *pgd;
 158        unsigned long next;
 159        unsigned long addr = start;
 160        int err = 0;
 161        int nr = 0;
 162
 163        BUG_ON(addr >= end);
 164        pgd = pgd_offset_k(addr);
 165        do {
 166                next = pgd_addr_end(addr, end);
 167                err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
 168                if (err)
 169                        break;
 170        } while (pgd++, addr = next, addr != end);
 171        flush_cache_vmap(start, end);
 172
 173        if (unlikely(err))
 174                return err;
 175        return nr;
 176}
 177
 178static inline int is_vmalloc_or_module_addr(const void *x)
 179{
 180        /*
 181         * ARM, x86-64 and sparc64 put modules in a special place,
 182         * and fall back on vmalloc() if that fails. Others
 183         * just put it in the vmalloc space.
 184         */
 185#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
 186        unsigned long addr = (unsigned long)x;
 187        if (addr >= MODULES_VADDR && addr < MODULES_END)
 188                return 1;
 189#endif
 190        return is_vmalloc_addr(x);
 191}
 192
 193/*
 194 * Walk a vmap address to the struct page it maps.
 195 */
 196struct page *vmalloc_to_page(const void *vmalloc_addr)
 197{
 198        unsigned long addr = (unsigned long) vmalloc_addr;
 199        struct page *page = NULL;
 200        pgd_t *pgd = pgd_offset_k(addr);
 201
 202        /*
 203         * XXX we might need to change this if we add VIRTUAL_BUG_ON for
 204         * architectures that do not vmalloc module space
 205         */
 206        VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
 207
 208        if (!pgd_none(*pgd)) {
 209                pud_t *pud = pud_offset(pgd, addr);
 210                if (!pud_none(*pud)) {
 211                        pmd_t *pmd = pmd_offset(pud, addr);
 212                        if (!pmd_none(*pmd)) {
 213                                pte_t *ptep, pte;
 214
 215                                ptep = pte_offset_map(pmd, addr);
 216                                pte = *ptep;
 217                                if (pte_present(pte))
 218                                        page = pte_page(pte);
 219                                pte_unmap(ptep);
 220                        }
 221                }
 222        }
 223        return page;
 224}
 225EXPORT_SYMBOL(vmalloc_to_page);
 226
 227/*
 228 * Map a vmalloc()-space virtual address to the physical page frame number.
 229 */
 230unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
 231{
 232        return page_to_pfn(vmalloc_to_page(vmalloc_addr));
 233}
 234EXPORT_SYMBOL(vmalloc_to_pfn);
 235
 236
 237/*** Global kva allocator ***/
 238
 239#define VM_LAZY_FREE    0x01
 240#define VM_LAZY_FREEING 0x02
 241#define VM_VM_AREA      0x04
 242
 243struct vmap_area {
 244        unsigned long va_start;
 245        unsigned long va_end;
 246        unsigned long flags;
 247        struct rb_node rb_node;         /* address sorted rbtree */
 248        struct list_head list;          /* address sorted list */
 249        struct list_head purge_list;    /* "lazy purge" list */
 250        void *private;
 251        struct rcu_head rcu_head;
 252};
 253
 254static DEFINE_SPINLOCK(vmap_area_lock);
 255static struct rb_root vmap_area_root = RB_ROOT;
 256static LIST_HEAD(vmap_area_list);
 257
 258static struct vmap_area *__find_vmap_area(unsigned long addr)
 259{
 260        struct rb_node *n = vmap_area_root.rb_node;
 261
 262        while (n) {
 263                struct vmap_area *va;
 264
 265                va = rb_entry(n, struct vmap_area, rb_node);
 266                if (addr < va->va_start)
 267                        n = n->rb_left;
 268                else if (addr > va->va_start)
 269                        n = n->rb_right;
 270                else
 271                        return va;
 272        }
 273
 274        return NULL;
 275}
 276
 277static void __insert_vmap_area(struct vmap_area *va)
 278{
 279        struct rb_node **p = &vmap_area_root.rb_node;
 280        struct rb_node *parent = NULL;
 281        struct rb_node *tmp;
 282
 283        while (*p) {
 284                struct vmap_area *tmp;
 285
 286                parent = *p;
 287                tmp = rb_entry(parent, struct vmap_area, rb_node);
 288                if (va->va_start < tmp->va_end)
 289                        p = &(*p)->rb_left;
 290                else if (va->va_end > tmp->va_start)
 291                        p = &(*p)->rb_right;
 292                else
 293                        BUG();
 294        }
 295
 296        rb_link_node(&va->rb_node, parent, p);
 297        rb_insert_color(&va->rb_node, &vmap_area_root);
 298
 299        /* address-sort this list so it is usable like the vmlist */
 300        tmp = rb_prev(&va->rb_node);
 301        if (tmp) {
 302                struct vmap_area *prev;
 303                prev = rb_entry(tmp, struct vmap_area, rb_node);
 304                list_add_rcu(&va->list, &prev->list);
 305        } else
 306                list_add_rcu(&va->list, &vmap_area_list);
 307}
 308
 309static void purge_vmap_area_lazy(void);
 310
 311/*
 312 * Allocate a region of KVA of the specified size and alignment, within the
 313 * vstart and vend.
 314 */
 315static struct vmap_area *alloc_vmap_area(unsigned long size,
 316                                unsigned long align,
 317                                unsigned long vstart, unsigned long vend,
 318                                int node, gfp_t gfp_mask)
 319{
 320        struct vmap_area *va;
 321        struct rb_node *n;
 322        unsigned long addr;
 323        int purged = 0;
 324
 325        BUG_ON(size & ~PAGE_MASK);
 326
 327        va = kmalloc_node(sizeof(struct vmap_area),
 328                        gfp_mask & GFP_RECLAIM_MASK, node);
 329        if (unlikely(!va))
 330                return ERR_PTR(-ENOMEM);
 331
 332retry:
 333        addr = ALIGN(vstart, align);
 334
 335        spin_lock(&vmap_area_lock);
 336        /* XXX: could have a last_hole cache */
 337        n = vmap_area_root.rb_node;
 338        if (n) {
 339                struct vmap_area *first = NULL;
 340
 341                do {
 342                        struct vmap_area *tmp;
 343                        tmp = rb_entry(n, struct vmap_area, rb_node);
 344                        if (tmp->va_end >= addr) {
 345                                if (!first && tmp->va_start < addr + size)
 346                                        first = tmp;
 347                                n = n->rb_left;
 348                        } else {
 349                                first = tmp;
 350                                n = n->rb_right;
 351                        }
 352                } while (n);
 353
 354                if (!first)
 355                        goto found;
 356
 357                if (first->va_end < addr) {
 358                        n = rb_next(&first->rb_node);
 359                        if (n)
 360                                first = rb_entry(n, struct vmap_area, rb_node);
 361                        else
 362                                goto found;
 363                }
 364
 365                while (addr + size > first->va_start && addr + size <= vend) {
 366                        addr = ALIGN(first->va_end + PAGE_SIZE, align);
 367
 368                        n = rb_next(&first->rb_node);
 369                        if (n)
 370                                first = rb_entry(n, struct vmap_area, rb_node);
 371                        else
 372                                goto found;
 373                }
 374        }
 375found:
 376        if (addr + size > vend) {
 377                spin_unlock(&vmap_area_lock);
 378                if (!purged) {
 379                        purge_vmap_area_lazy();
 380                        purged = 1;
 381                        goto retry;
 382                }
 383                if (printk_ratelimit())
 384                        printk(KERN_WARNING "vmap allocation failed: "
 385                                 "use vmalloc=<size> to increase size.\n");
 386                return ERR_PTR(-EBUSY);
 387        }
 388
 389        BUG_ON(addr & (align-1));
 390
 391        va->va_start = addr;
 392        va->va_end = addr + size;
 393        va->flags = 0;
 394        __insert_vmap_area(va);
 395        spin_unlock(&vmap_area_lock);
 396
 397        return va;
 398}
 399
 400static void rcu_free_va(struct rcu_head *head)
 401{
 402        struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
 403
 404        kfree(va);
 405}
 406
 407static void __free_vmap_area(struct vmap_area *va)
 408{
 409        BUG_ON(RB_EMPTY_NODE(&va->rb_node));
 410        rb_erase(&va->rb_node, &vmap_area_root);
 411        RB_CLEAR_NODE(&va->rb_node);
 412        list_del_rcu(&va->list);
 413
 414        call_rcu(&va->rcu_head, rcu_free_va);
 415}
 416
 417/*
 418 * Free a region of KVA allocated by alloc_vmap_area
 419 */
 420static void free_vmap_area(struct vmap_area *va)
 421{
 422        spin_lock(&vmap_area_lock);
 423        __free_vmap_area(va);
 424        spin_unlock(&vmap_area_lock);
 425}
 426
 427/*
 428 * Clear the pagetable entries of a given vmap_area
 429 */
 430static void unmap_vmap_area(struct vmap_area *va)
 431{
 432        vunmap_page_range(va->va_start, va->va_end);
 433}
 434
 435/*
 436 * lazy_max_pages is the maximum amount of virtual address space we gather up
 437 * before attempting to purge with a TLB flush.
 438 *
 439 * There is a tradeoff here: a larger number will cover more kernel page tables
 440 * and take slightly longer to purge, but it will linearly reduce the number of
 441 * global TLB flushes that must be performed. It would seem natural to scale
 442 * this number up linearly with the number of CPUs (because vmapping activity
 443 * could also scale linearly with the number of CPUs), however it is likely
 444 * that in practice, workloads might be constrained in other ways that mean
 445 * vmap activity will not scale linearly with CPUs. Also, I want to be
 446 * conservative and not introduce a big latency on huge systems, so go with
 447 * a less aggressive log scale. It will still be an improvement over the old
 448 * code, and it will be simple to change the scale factor if we find that it
 449 * becomes a problem on bigger systems.
 450 */
 451static unsigned long lazy_max_pages(void)
 452{
 453        unsigned int log;
 454
 455        log = fls(num_online_cpus());
 456
 457        return log * (32UL * 1024 * 1024 / PAGE_SIZE);
 458}
 459
 460static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
 461
 462/*
 463 * Purges all lazily-freed vmap areas.
 464 *
 465 * If sync is 0 then don't purge if there is already a purge in progress.
 466 * If force_flush is 1, then flush kernel TLBs between *start and *end even
 467 * if we found no lazy vmap areas to unmap (callers can use this to optimise
 468 * their own TLB flushing).
 469 * Returns with *start = min(*start, lowest purged address)
 470 *              *end = max(*end, highest purged address)
 471 */
 472static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
 473                                        int sync, int force_flush)
 474{
 475        static DEFINE_SPINLOCK(purge_lock);
 476        LIST_HEAD(valist);
 477        struct vmap_area *va;
 478        int nr = 0;
 479
 480        /*
 481         * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
 482         * should not expect such behaviour. This just simplifies locking for
 483         * the case that isn't actually used at the moment anyway.
 484         */
 485        if (!sync && !force_flush) {
 486                if (!spin_trylock(&purge_lock))
 487                        return;
 488        } else
 489                spin_lock(&purge_lock);
 490
 491        rcu_read_lock();
 492        list_for_each_entry_rcu(va, &vmap_area_list, list) {
 493                if (va->flags & VM_LAZY_FREE) {
 494                        if (va->va_start < *start)
 495                                *start = va->va_start;
 496                        if (va->va_end > *end)
 497                                *end = va->va_end;
 498                        nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
 499                        unmap_vmap_area(va);
 500                        list_add_tail(&va->purge_list, &valist);
 501                        va->flags |= VM_LAZY_FREEING;
 502                        va->flags &= ~VM_LAZY_FREE;
 503                }
 504        }
 505        rcu_read_unlock();
 506
 507        if (nr) {
 508                BUG_ON(nr > atomic_read(&vmap_lazy_nr));
 509                atomic_sub(nr, &vmap_lazy_nr);
 510        }
 511
 512        if (nr || force_flush)
 513                flush_tlb_kernel_range(*start, *end);
 514
 515        if (nr) {
 516                spin_lock(&vmap_area_lock);
 517                list_for_each_entry(va, &valist, purge_list)
 518                        __free_vmap_area(va);
 519                spin_unlock(&vmap_area_lock);
 520        }
 521        spin_unlock(&purge_lock);
 522}
 523
 524/*
 525 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
 526 * is already purging.
 527 */
 528static void try_purge_vmap_area_lazy(void)
 529{
 530        unsigned long start = ULONG_MAX, end = 0;
 531
 532        __purge_vmap_area_lazy(&start, &end, 0, 0);
 533}
 534
 535/*
 536 * Kick off a purge of the outstanding lazy areas.
 537 */
 538static void purge_vmap_area_lazy(void)
 539{
 540        unsigned long start = ULONG_MAX, end = 0;
 541
 542        __purge_vmap_area_lazy(&start, &end, 1, 0);
 543}
 544
 545/*
 546 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
 547 * called for the correct range previously.
 548 */
 549static void free_unmap_vmap_area_noflush(struct vmap_area *va)
 550{
 551        va->flags |= VM_LAZY_FREE;
 552        atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
 553        if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
 554                try_purge_vmap_area_lazy();
 555}
 556
 557/*
 558 * Free and unmap a vmap area
 559 */
 560static void free_unmap_vmap_area(struct vmap_area *va)
 561{
 562        flush_cache_vunmap(va->va_start, va->va_end);
 563        free_unmap_vmap_area_noflush(va);
 564}
 565
 566static struct vmap_area *find_vmap_area(unsigned long addr)
 567{
 568        struct vmap_area *va;
 569
 570        spin_lock(&vmap_area_lock);
 571        va = __find_vmap_area(addr);
 572        spin_unlock(&vmap_area_lock);
 573
 574        return va;
 575}
 576
 577static void free_unmap_vmap_area_addr(unsigned long addr)
 578{
 579        struct vmap_area *va;
 580
 581        va = find_vmap_area(addr);
 582        BUG_ON(!va);
 583        free_unmap_vmap_area(va);
 584}
 585
 586
 587/*** Per cpu kva allocator ***/
 588
 589/*
 590 * vmap space is limited especially on 32 bit architectures. Ensure there is
 591 * room for at least 16 percpu vmap blocks per CPU.
 592 */
 593/*
 594 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
 595 * to #define VMALLOC_SPACE             (VMALLOC_END-VMALLOC_START). Guess
 596 * instead (we just need a rough idea)
 597 */
 598#if BITS_PER_LONG == 32
 599#define VMALLOC_SPACE           (128UL*1024*1024)
 600#else
 601#define VMALLOC_SPACE           (128UL*1024*1024*1024)
 602#endif
 603
 604#define VMALLOC_PAGES           (VMALLOC_SPACE / PAGE_SIZE)
 605#define VMAP_MAX_ALLOC          BITS_PER_LONG   /* 256K with 4K pages */
 606#define VMAP_BBMAP_BITS_MAX     1024    /* 4MB with 4K pages */
 607#define VMAP_BBMAP_BITS_MIN     (VMAP_MAX_ALLOC*2)
 608#define VMAP_MIN(x, y)          ((x) < (y) ? (x) : (y)) /* can't use min() */
 609#define VMAP_MAX(x, y)          ((x) > (y) ? (x) : (y)) /* can't use max() */
 610#define VMAP_BBMAP_BITS         VMAP_MIN(VMAP_BBMAP_BITS_MAX,           \
 611                                        VMAP_MAX(VMAP_BBMAP_BITS_MIN,   \
 612                                                VMALLOC_PAGES / NR_CPUS / 16))
 613
 614#define VMAP_BLOCK_SIZE         (VMAP_BBMAP_BITS * PAGE_SIZE)
 615
 616static bool vmap_initialized __read_mostly = false;
 617
 618struct vmap_block_queue {
 619        spinlock_t lock;
 620        struct list_head free;
 621        struct list_head dirty;
 622        unsigned int nr_dirty;
 623};
 624
 625struct vmap_block {
 626        spinlock_t lock;
 627        struct vmap_area *va;
 628        struct vmap_block_queue *vbq;
 629        unsigned long free, dirty;
 630        DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
 631        DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
 632        union {
 633                struct {
 634                        struct list_head free_list;
 635                        struct list_head dirty_list;
 636                };
 637                struct rcu_head rcu_head;
 638        };
 639};
 640
 641/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
 642static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
 643
 644/*
 645 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
 646 * in the free path. Could get rid of this if we change the API to return a
 647 * "cookie" from alloc, to be passed to free. But no big deal yet.
 648 */
 649static DEFINE_SPINLOCK(vmap_block_tree_lock);
 650static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
 651
 652/*
 653 * We should probably have a fallback mechanism to allocate virtual memory
 654 * out of partially filled vmap blocks. However vmap block sizing should be
 655 * fairly reasonable according to the vmalloc size, so it shouldn't be a
 656 * big problem.
 657 */
 658
 659static unsigned long addr_to_vb_idx(unsigned long addr)
 660{
 661        addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
 662        addr /= VMAP_BLOCK_SIZE;
 663        return addr;
 664}
 665
 666static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
 667{
 668        struct vmap_block_queue *vbq;
 669        struct vmap_block *vb;
 670        struct vmap_area *va;
 671        unsigned long vb_idx;
 672        int node, err;
 673
 674        node = numa_node_id();
 675
 676        vb = kmalloc_node(sizeof(struct vmap_block),
 677                        gfp_mask & GFP_RECLAIM_MASK, node);
 678        if (unlikely(!vb))
 679                return ERR_PTR(-ENOMEM);
 680
 681        va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
 682                                        VMALLOC_START, VMALLOC_END,
 683                                        node, gfp_mask);
 684        if (unlikely(IS_ERR(va))) {
 685                kfree(vb);
 686                return ERR_PTR(PTR_ERR(va));
 687        }
 688
 689        err = radix_tree_preload(gfp_mask);
 690        if (unlikely(err)) {
 691                kfree(vb);
 692                free_vmap_area(va);
 693                return ERR_PTR(err);
 694        }
 695
 696        spin_lock_init(&vb->lock);
 697        vb->va = va;
 698        vb->free = VMAP_BBMAP_BITS;
 699        vb->dirty = 0;
 700        bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
 701        bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
 702        INIT_LIST_HEAD(&vb->free_list);
 703        INIT_LIST_HEAD(&vb->dirty_list);
 704
 705        vb_idx = addr_to_vb_idx(va->va_start);
 706        spin_lock(&vmap_block_tree_lock);
 707        err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
 708        spin_unlock(&vmap_block_tree_lock);
 709        BUG_ON(err);
 710        radix_tree_preload_end();
 711
 712        vbq = &get_cpu_var(vmap_block_queue);
 713        vb->vbq = vbq;
 714        spin_lock(&vbq->lock);
 715        list_add(&vb->free_list, &vbq->free);
 716        spin_unlock(&vbq->lock);
 717        put_cpu_var(vmap_cpu_blocks);
 718
 719        return vb;
 720}
 721
 722static void rcu_free_vb(struct rcu_head *head)
 723{
 724        struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
 725
 726        kfree(vb);
 727}
 728
 729static void free_vmap_block(struct vmap_block *vb)
 730{
 731        struct vmap_block *tmp;
 732        unsigned long vb_idx;
 733
 734        spin_lock(&vb->vbq->lock);
 735        if (!list_empty(&vb->free_list))
 736                list_del(&vb->free_list);
 737        if (!list_empty(&vb->dirty_list))
 738                list_del(&vb->dirty_list);
 739        spin_unlock(&vb->vbq->lock);
 740
 741        vb_idx = addr_to_vb_idx(vb->va->va_start);
 742        spin_lock(&vmap_block_tree_lock);
 743        tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
 744        spin_unlock(&vmap_block_tree_lock);
 745        BUG_ON(tmp != vb);
 746
 747        free_unmap_vmap_area_noflush(vb->va);
 748        call_rcu(&vb->rcu_head, rcu_free_vb);
 749}
 750
 751static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
 752{
 753        struct vmap_block_queue *vbq;
 754        struct vmap_block *vb;
 755        unsigned long addr = 0;
 756        unsigned int order;
 757
 758        BUG_ON(size & ~PAGE_MASK);
 759        BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
 760        order = get_order(size);
 761
 762again:
 763        rcu_read_lock();
 764        vbq = &get_cpu_var(vmap_block_queue);
 765        list_for_each_entry_rcu(vb, &vbq->free, free_list) {
 766                int i;
 767
 768                spin_lock(&vb->lock);
 769                i = bitmap_find_free_region(vb->alloc_map,
 770                                                VMAP_BBMAP_BITS, order);
 771
 772                if (i >= 0) {
 773                        addr = vb->va->va_start + (i << PAGE_SHIFT);
 774                        BUG_ON(addr_to_vb_idx(addr) !=
 775                                        addr_to_vb_idx(vb->va->va_start));
 776                        vb->free -= 1UL << order;
 777                        if (vb->free == 0) {
 778                                spin_lock(&vbq->lock);
 779                                list_del_init(&vb->free_list);
 780                                spin_unlock(&vbq->lock);
 781                        }
 782                        spin_unlock(&vb->lock);
 783                        break;
 784                }
 785                spin_unlock(&vb->lock);
 786        }
 787        put_cpu_var(vmap_cpu_blocks);
 788        rcu_read_unlock();
 789
 790        if (!addr) {
 791                vb = new_vmap_block(gfp_mask);
 792                if (IS_ERR(vb))
 793                        return vb;
 794                goto again;
 795        }
 796
 797        return (void *)addr;
 798}
 799
 800static void vb_free(const void *addr, unsigned long size)
 801{
 802        unsigned long offset;
 803        unsigned long vb_idx;
 804        unsigned int order;
 805        struct vmap_block *vb;
 806
 807        BUG_ON(size & ~PAGE_MASK);
 808        BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
 809
 810        flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
 811
 812        order = get_order(size);
 813
 814        offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
 815
 816        vb_idx = addr_to_vb_idx((unsigned long)addr);
 817        rcu_read_lock();
 818        vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
 819        rcu_read_unlock();
 820        BUG_ON(!vb);
 821
 822        spin_lock(&vb->lock);
 823        bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
 824        if (!vb->dirty) {
 825                spin_lock(&vb->vbq->lock);
 826                list_add(&vb->dirty_list, &vb->vbq->dirty);
 827                spin_unlock(&vb->vbq->lock);
 828        }
 829        vb->dirty += 1UL << order;
 830        if (vb->dirty == VMAP_BBMAP_BITS) {
 831                BUG_ON(vb->free || !list_empty(&vb->free_list));
 832                spin_unlock(&vb->lock);
 833                free_vmap_block(vb);
 834        } else
 835                spin_unlock(&vb->lock);
 836}
 837
 838/**
 839 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
 840 *
 841 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
 842 * to amortize TLB flushing overheads. What this means is that any page you
 843 * have now, may, in a former life, have been mapped into kernel virtual
 844 * address by the vmap layer and so there might be some CPUs with TLB entries
 845 * still referencing that page (additional to the regular 1:1 kernel mapping).
 846 *
 847 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
 848 * be sure that none of the pages we have control over will have any aliases
 849 * from the vmap layer.
 850 */
 851void vm_unmap_aliases(void)
 852{
 853        unsigned long start = ULONG_MAX, end = 0;
 854        int cpu;
 855        int flush = 0;
 856
 857        if (unlikely(!vmap_initialized))
 858                return;
 859
 860        for_each_possible_cpu(cpu) {
 861                struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
 862                struct vmap_block *vb;
 863
 864                rcu_read_lock();
 865                list_for_each_entry_rcu(vb, &vbq->free, free_list) {
 866                        int i;
 867
 868                        spin_lock(&vb->lock);
 869                        i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
 870                        while (i < VMAP_BBMAP_BITS) {
 871                                unsigned long s, e;
 872                                int j;
 873                                j = find_next_zero_bit(vb->dirty_map,
 874                                        VMAP_BBMAP_BITS, i);
 875
 876                                s = vb->va->va_start + (i << PAGE_SHIFT);
 877                                e = vb->va->va_start + (j << PAGE_SHIFT);
 878                                vunmap_page_range(s, e);
 879                                flush = 1;
 880
 881                                if (s < start)
 882                                        start = s;
 883                                if (e > end)
 884                                        end = e;
 885
 886                                i = j;
 887                                i = find_next_bit(vb->dirty_map,
 888                                                        VMAP_BBMAP_BITS, i);
 889                        }
 890                        spin_unlock(&vb->lock);
 891                }
 892                rcu_read_unlock();
 893        }
 894
 895        __purge_vmap_area_lazy(&start, &end, 1, flush);
 896}
 897EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 898
 899/**
 900 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
 901 * @mem: the pointer returned by vm_map_ram
 902 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
 903 */
 904void vm_unmap_ram(const void *mem, unsigned int count)
 905{
 906        unsigned long size = count << PAGE_SHIFT;
 907        unsigned long addr = (unsigned long)mem;
 908
 909        BUG_ON(!addr);
 910        BUG_ON(addr < VMALLOC_START);
 911        BUG_ON(addr > VMALLOC_END);
 912        BUG_ON(addr & (PAGE_SIZE-1));
 913
 914        debug_check_no_locks_freed(mem, size);
 915
 916        if (likely(count <= VMAP_MAX_ALLOC))
 917                vb_free(mem, size);
 918        else
 919                free_unmap_vmap_area_addr(addr);
 920}
 921EXPORT_SYMBOL(vm_unmap_ram);
 922
 923/**
 924 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
 925 * @pages: an array of pointers to the pages to be mapped
 926 * @count: number of pages
 927 * @node: prefer to allocate data structures on this node
 928 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
 929 *
 930 * Returns: a pointer to the address that has been mapped, or %NULL on failure
 931 */
 932void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 933{
 934        unsigned long size = count << PAGE_SHIFT;
 935        unsigned long addr;
 936        void *mem;
 937
 938        if (likely(count <= VMAP_MAX_ALLOC)) {
 939                mem = vb_alloc(size, GFP_KERNEL);
 940                if (IS_ERR(mem))
 941                        return NULL;
 942                addr = (unsigned long)mem;
 943        } else {
 944                struct vmap_area *va;
 945                va = alloc_vmap_area(size, PAGE_SIZE,
 946                                VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
 947                if (IS_ERR(va))
 948                        return NULL;
 949
 950                addr = va->va_start;
 951                mem = (void *)addr;
 952        }
 953        if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
 954                vm_unmap_ram(mem, count);
 955                return NULL;
 956        }
 957        return mem;
 958}
 959EXPORT_SYMBOL(vm_map_ram);
 960
 961void __init vmalloc_init(void)
 962{
 963        int i;
 964
 965        for_each_possible_cpu(i) {
 966                struct vmap_block_queue *vbq;
 967
 968                vbq = &per_cpu(vmap_block_queue, i);
 969                spin_lock_init(&vbq->lock);
 970                INIT_LIST_HEAD(&vbq->free);
 971                INIT_LIST_HEAD(&vbq->dirty);
 972                vbq->nr_dirty = 0;
 973        }
 974
 975        vmap_initialized = true;
 976}
 977
 978void unmap_kernel_range(unsigned long addr, unsigned long size)
 979{
 980        unsigned long end = addr + size;
 981        vunmap_page_range(addr, end);
 982        flush_tlb_kernel_range(addr, end);
 983}
 984
 985int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 986{
 987        unsigned long addr = (unsigned long)area->addr;
 988        unsigned long end = addr + area->size - PAGE_SIZE;
 989        int err;
 990
 991        err = vmap_page_range(addr, end, prot, *pages);
 992        if (err > 0) {
 993                *pages += err;
 994                err = 0;
 995        }
 996
 997        return err;
 998}
 999EXPORT_SYMBOL_GPL(map_vm_area);
1000
1001/*** Old vmalloc interfaces ***/
1002DEFINE_RWLOCK(vmlist_lock);
1003struct vm_struct *vmlist;
1004
1005static struct vm_struct *__get_vm_area_node(unsigned long size,
1006                unsigned long flags, unsigned long start, unsigned long end,
1007                int node, gfp_t gfp_mask, void *caller)
1008{
1009        static struct vmap_area *va;
1010        struct vm_struct *area;
1011        struct vm_struct *tmp, **p;
1012        unsigned long align = 1;
1013
1014        BUG_ON(in_interrupt());
1015        if (flags & VM_IOREMAP) {
1016                int bit = fls(size);
1017
1018                if (bit > IOREMAP_MAX_ORDER)
1019                        bit = IOREMAP_MAX_ORDER;
1020                else if (bit < PAGE_SHIFT)
1021                        bit = PAGE_SHIFT;
1022
1023                align = 1ul << bit;
1024        }
1025
1026        size = PAGE_ALIGN(size);
1027        if (unlikely(!size))
1028                return NULL;
1029
1030        area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1031        if (unlikely(!area))
1032                return NULL;
1033
1034        /*
1035         * We always allocate a guard page.
1036         */
1037        size += PAGE_SIZE;
1038
1039        va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1040        if (IS_ERR(va)) {
1041                kfree(area);
1042                return NULL;
1043        }
1044
1045        area->flags = flags;
1046        area->addr = (void *)va->va_start;
1047        area->size = size;
1048        area->pages = NULL;
1049        area->nr_pages = 0;
1050        area->phys_addr = 0;
1051        area->caller = caller;
1052        va->private = area;
1053        va->flags |= VM_VM_AREA;
1054
1055        write_lock(&vmlist_lock);
1056        for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1057                if (tmp->addr >= area->addr)
1058                        break;
1059        }
1060        area->next = *p;
1061        *p = area;
1062        write_unlock(&vmlist_lock);
1063
1064        return area;
1065}
1066
1067struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1068                                unsigned long start, unsigned long end)
1069{
1070        return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1071                                                __builtin_return_address(0));
1072}
1073EXPORT_SYMBOL_GPL(__get_vm_area);
1074
1075/**
1076 *      get_vm_area  -  reserve a contiguous kernel virtual area
1077 *      @size:          size of the area
1078 *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
1079 *
1080 *      Search an area of @size in the kernel virtual mapping area,
1081 *      and reserved it for out purposes.  Returns the area descriptor
1082 *      on success or %NULL on failure.
1083 */
1084struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1085{
1086        return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1087                                -1, GFP_KERNEL, __builtin_return_address(0));
1088}
1089
1090struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1091                                void *caller)
1092{
1093        return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1094                                                -1, GFP_KERNEL, caller);
1095}
1096
1097struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1098                                   int node, gfp_t gfp_mask)
1099{
1100        return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
1101                                  gfp_mask, __builtin_return_address(0));
1102}
1103
1104static struct vm_struct *find_vm_area(const void *addr)
1105{
1106        struct vmap_area *va;
1107
1108        va = find_vmap_area((unsigned long)addr);
1109        if (va && va->flags & VM_VM_AREA)
1110                return va->private;
1111
1112        return NULL;
1113}
1114
1115/**
1116 *      remove_vm_area  -  find and remove a continuous kernel virtual area
1117 *      @addr:          base address
1118 *
1119 *      Search for the kernel VM area starting at @addr, and remove it.
1120 *      This function returns the found VM area, but using it is NOT safe
1121 *      on SMP machines, except for its size or flags.
1122 */
1123struct vm_struct *remove_vm_area(const void *addr)
1124{
1125        struct vmap_area *va;
1126
1127        va = find_vmap_area((unsigned long)addr);
1128        if (va && va->flags & VM_VM_AREA) {
1129                struct vm_struct *vm = va->private;
1130                struct vm_struct *tmp, **p;
1131                free_unmap_vmap_area(va);
1132                vm->size -= PAGE_SIZE;
1133
1134                write_lock(&vmlist_lock);
1135                for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1136                        ;
1137                *p = tmp->next;
1138                write_unlock(&vmlist_lock);
1139
1140                return vm;
1141        }
1142        return NULL;
1143}
1144
1145static void __vunmap(const void *addr, int deallocate_pages)
1146{
1147        struct vm_struct *area;
1148
1149        if (!addr)
1150                return;
1151
1152        if ((PAGE_SIZE-1) & (unsigned long)addr) {
1153                WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1154                return;
1155        }
1156
1157        area = remove_vm_area(addr);
1158        if (unlikely(!area)) {
1159                WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1160                                addr);
1161                return;
1162        }
1163
1164        debug_check_no_locks_freed(addr, area->size);
1165        debug_check_no_obj_freed(addr, area->size);
1166
1167        if (deallocate_pages) {
1168                int i;
1169
1170                for (i = 0; i < area->nr_pages; i++) {
1171                        struct page *page = area->pages[i];
1172
1173                        BUG_ON(!page);
1174                        __free_page(page);
1175                }
1176
1177                if (area->flags & VM_VPAGES)
1178                        vfree(area->pages);
1179                else
1180                        kfree(area->pages);
1181        }
1182
1183        kfree(area);
1184        return;
1185}
1186
1187/**
1188 *      vfree  -  release memory allocated by vmalloc()
1189 *      @addr:          memory base address
1190 *
1191 *      Free the virtually continuous memory area starting at @addr, as
1192 *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1193 *      NULL, no operation is performed.
1194 *
1195 *      Must not be called in interrupt context.
1196 */
1197void vfree(const void *addr)
1198{
1199        BUG_ON(in_interrupt());
1200        __vunmap(addr, 1);
1201}
1202EXPORT_SYMBOL(vfree);
1203
1204/**
1205 *      vunmap  -  release virtual mapping obtained by vmap()
1206 *      @addr:          memory base address
1207 *
1208 *      Free the virtually contiguous memory area starting at @addr,
1209 *      which was created from the page array passed to vmap().
1210 *
1211 *      Must not be called in interrupt context.
1212 */
1213void vunmap(const void *addr)
1214{
1215        BUG_ON(in_interrupt());
1216        __vunmap(addr, 0);
1217}
1218EXPORT_SYMBOL(vunmap);
1219
1220/**
1221 *      vmap  -  map an array of pages into virtually contiguous space
1222 *      @pages:         array of page pointers
1223 *      @count:         number of pages to map
1224 *      @flags:         vm_area->flags
1225 *      @prot:          page protection for the mapping
1226 *
1227 *      Maps @count pages from @pages into contiguous kernel virtual
1228 *      space.
1229 */
1230void *vmap(struct page **pages, unsigned int count,
1231                unsigned long flags, pgprot_t prot)
1232{
1233        struct vm_struct *area;
1234
1235        if (count > num_physpages)
1236                return NULL;
1237
1238        area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1239                                        __builtin_return_address(0));
1240        if (!area)
1241                return NULL;
1242
1243        if (map_vm_area(area, prot, &pages)) {
1244                vunmap(area->addr);
1245                return NULL;
1246        }
1247
1248        return area->addr;
1249}
1250EXPORT_SYMBOL(vmap);
1251
1252static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1253                            int node, void *caller);
1254static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1255                                 pgprot_t prot, int node, void *caller)
1256{
1257        struct page **pages;
1258        unsigned int nr_pages, array_size, i;
1259
1260        nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1261        array_size = (nr_pages * sizeof(struct page *));
1262
1263        area->nr_pages = nr_pages;
1264        /* Please note that the recursion is strictly bounded. */
1265        if (array_size > PAGE_SIZE) {
1266                pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
1267                                PAGE_KERNEL, node, caller);
1268                area->flags |= VM_VPAGES;
1269        } else {
1270                pages = kmalloc_node(array_size,
1271                                (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
1272                                node);
1273        }
1274        area->pages = pages;
1275        area->caller = caller;
1276        if (!area->pages) {
1277                remove_vm_area(area->addr);
1278                kfree(area);
1279                return NULL;
1280        }
1281
1282        for (i = 0; i < area->nr_pages; i++) {
1283                struct page *page;
1284
1285                if (node < 0)
1286                        page = alloc_page(gfp_mask);
1287                else
1288                        page = alloc_pages_node(node, gfp_mask, 0);
1289
1290                if (unlikely(!page)) {
1291                        /* Successfully allocated i pages, free them in __vunmap() */
1292                        area->nr_pages = i;
1293                        goto fail;
1294                }
1295                area->pages[i] = page;
1296        }
1297
1298        if (map_vm_area(area, prot, &pages))
1299                goto fail;
1300        return area->addr;
1301
1302fail:
1303        vfree(area->addr);
1304        return NULL;
1305}
1306
1307void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1308{
1309        return __vmalloc_area_node(area, gfp_mask, prot, -1,
1310                                        __builtin_return_address(0));
1311}
1312
1313/**
1314 *      __vmalloc_node  -  allocate virtually contiguous memory
1315 *      @size:          allocation size
1316 *      @gfp_mask:      flags for the page level allocator
1317 *      @prot:          protection mask for the allocated pages
1318 *      @node:          node to use for allocation or -1
1319 *      @caller:        caller's return address
1320 *
1321 *      Allocate enough pages to cover @size from the page level
1322 *      allocator with @gfp_mask flags.  Map them into contiguous
1323 *      kernel virtual space, using a pagetable protection of @prot.
1324 */
1325static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1326                                                int node, void *caller)
1327{
1328        struct vm_struct *area;
1329
1330        size = PAGE_ALIGN(size);
1331        if (!size || (size >> PAGE_SHIFT) > num_physpages)
1332                return NULL;
1333
1334        area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1335                                                node, gfp_mask, caller);
1336
1337        if (!area)
1338                return NULL;
1339
1340        return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1341}
1342
1343void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1344{
1345        return __vmalloc_node(size, gfp_mask, prot, -1,
1346                                __builtin_return_address(0));
1347}
1348EXPORT_SYMBOL(__vmalloc);
1349
1350/**
1351 *      vmalloc  -  allocate virtually contiguous memory
1352 *      @size:          allocation size
1353 *      Allocate enough pages to cover @size from the page level
1354 *      allocator and map them into contiguous kernel virtual space.
1355 *
1356 *      For tight control over page level allocator and protection flags
1357 *      use __vmalloc() instead.
1358 */
1359void *vmalloc(unsigned long size)
1360{
1361        return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1362                                        -1, __builtin_return_address(0));
1363}
1364EXPORT_SYMBOL(vmalloc);
1365
1366/**
1367 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1368 * @size: allocation size
1369 *
1370 * The resulting memory area is zeroed so it can be mapped to userspace
1371 * without leaking data.
1372 */
1373void *vmalloc_user(unsigned long size)
1374{
1375        struct vm_struct *area;
1376        void *ret;
1377
1378        ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
1379        if (ret) {
1380                area = find_vm_area(ret);
1381                area->flags |= VM_USERMAP;
1382        }
1383        return ret;
1384}
1385EXPORT_SYMBOL(vmalloc_user);
1386
1387/**
1388 *      vmalloc_node  -  allocate memory on a specific node
1389 *      @size:          allocation size
1390 *      @node:          numa node
1391 *
1392 *      Allocate enough pages to cover @size from the page level
1393 *      allocator and map them into contiguous kernel virtual space.
1394 *
1395 *      For tight control over page level allocator and protection flags
1396 *      use __vmalloc() instead.
1397 */
1398void *vmalloc_node(unsigned long size, int node)
1399{
1400        return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1401                                        node, __builtin_return_address(0));
1402}
1403EXPORT_SYMBOL(vmalloc_node);
1404
1405#ifndef PAGE_KERNEL_EXEC
1406# define PAGE_KERNEL_EXEC PAGE_KERNEL
1407#endif
1408
1409/**
1410 *      vmalloc_exec  -  allocate virtually contiguous, executable memory
1411 *      @size:          allocation size
1412 *
1413 *      Kernel-internal function to allocate enough pages to cover @size
1414 *      the page level allocator and map them into contiguous and
1415 *      executable kernel virtual space.
1416 *
1417 *      For tight control over page level allocator and protection flags
1418 *      use __vmalloc() instead.
1419 */
1420
1421void *vmalloc_exec(unsigned long size)
1422{
1423        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
1424}
1425
1426#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1427#define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1428#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1429#define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1430#else
1431#define GFP_VMALLOC32 GFP_KERNEL
1432#endif
1433
1434/**
1435 *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
1436 *      @size:          allocation size
1437 *
1438 *      Allocate enough 32bit PA addressable pages to cover @size from the
1439 *      page level allocator and map them into contiguous kernel virtual space.
1440 */
1441void *vmalloc_32(unsigned long size)
1442{
1443        return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
1444}
1445EXPORT_SYMBOL(vmalloc_32);
1446
1447/**
1448 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1449 *      @size:          allocation size
1450 *
1451 * The resulting memory area is 32bit addressable and zeroed so it can be
1452 * mapped to userspace without leaking data.
1453 */
1454void *vmalloc_32_user(unsigned long size)
1455{
1456        struct vm_struct *area;
1457        void *ret;
1458
1459        ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
1460        if (ret) {
1461                area = find_vm_area(ret);
1462                area->flags |= VM_USERMAP;
1463        }
1464        return ret;
1465}
1466EXPORT_SYMBOL(vmalloc_32_user);
1467
1468long vread(char *buf, char *addr, unsigned long count)
1469{
1470        struct vm_struct *tmp;
1471        char *vaddr, *buf_start = buf;
1472        unsigned long n;
1473
1474        /* Don't allow overflow */
1475        if ((unsigned long) addr + count < count)
1476                count = -(unsigned long) addr;
1477
1478        read_lock(&vmlist_lock);
1479        for (tmp = vmlist; tmp; tmp = tmp->next) {
1480                vaddr = (char *) tmp->addr;
1481                if (addr >= vaddr + tmp->size - PAGE_SIZE)
1482                        continue;
1483                while (addr < vaddr) {
1484                        if (count == 0)
1485                                goto finished;
1486                        *buf = '\0';
1487                        buf++;
1488                        addr++;
1489                        count--;
1490                }
1491                n = vaddr + tmp->size - PAGE_SIZE - addr;
1492                do {
1493                        if (count == 0)
1494                                goto finished;
1495                        *buf = *addr;
1496                        buf++;
1497                        addr++;
1498                        count--;
1499                } while (--n > 0);
1500        }
1501finished:
1502        read_unlock(&vmlist_lock);
1503        return buf - buf_start;
1504}
1505
1506long vwrite(char *buf, char *addr, unsigned long count)
1507{
1508        struct vm_struct *tmp;
1509        char *vaddr, *buf_start = buf;
1510        unsigned long n;
1511
1512        /* Don't allow overflow */
1513        if ((unsigned long) addr + count < count)
1514                count = -(unsigned long) addr;
1515
1516        read_lock(&vmlist_lock);
1517        for (tmp = vmlist; tmp; tmp = tmp->next) {
1518                vaddr = (char *) tmp->addr;
1519                if (addr >= vaddr + tmp->size - PAGE_SIZE)
1520                        continue;
1521                while (addr < vaddr) {
1522                        if (count == 0)
1523                                goto finished;
1524                        buf++;
1525                        addr++;
1526                        count--;
1527                }
1528                n = vaddr + tmp->size - PAGE_SIZE - addr;
1529                do {
1530                        if (count == 0)
1531                                goto finished;
1532                        *addr = *buf;
1533                        buf++;
1534                        addr++;
1535                        count--;
1536                } while (--n > 0);
1537        }
1538finished:
1539        read_unlock(&vmlist_lock);
1540        return buf - buf_start;
1541}
1542
1543/**
1544 *      remap_vmalloc_range  -  map vmalloc pages to userspace
1545 *      @vma:           vma to cover (map full range of vma)
1546 *      @addr:          vmalloc memory
1547 *      @pgoff:         number of pages into addr before first page to map
1548 *
1549 *      Returns:        0 for success, -Exxx on failure
1550 *
1551 *      This function checks that addr is a valid vmalloc'ed area, and
1552 *      that it is big enough to cover the vma. Will return failure if
1553 *      that criteria isn't met.
1554 *
1555 *      Similar to remap_pfn_range() (see mm/memory.c)
1556 */
1557int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1558                                                unsigned long pgoff)
1559{
1560        struct vm_struct *area;
1561        unsigned long uaddr = vma->vm_start;
1562        unsigned long usize = vma->vm_end - vma->vm_start;
1563
1564        if ((PAGE_SIZE-1) & (unsigned long)addr)
1565                return -EINVAL;
1566
1567        area = find_vm_area(addr);
1568        if (!area)
1569                return -EINVAL;
1570
1571        if (!(area->flags & VM_USERMAP))
1572                return -EINVAL;
1573
1574        if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
1575                return -EINVAL;
1576
1577        addr += pgoff << PAGE_SHIFT;
1578        do {
1579                struct page *page = vmalloc_to_page(addr);
1580                int ret;
1581
1582                ret = vm_insert_page(vma, uaddr, page);
1583                if (ret)
1584                        return ret;
1585
1586                uaddr += PAGE_SIZE;
1587                addr += PAGE_SIZE;
1588                usize -= PAGE_SIZE;
1589        } while (usize > 0);
1590
1591        /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1592        vma->vm_flags |= VM_RESERVED;
1593
1594        return 0;
1595}
1596EXPORT_SYMBOL(remap_vmalloc_range);
1597
1598/*
1599 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1600 * have one.
1601 */
1602void  __attribute__((weak)) vmalloc_sync_all(void)
1603{
1604}
1605
1606
1607static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
1608{
1609        /* apply_to_page_range() does all the hard work. */
1610        return 0;
1611}
1612
1613/**
1614 *      alloc_vm_area - allocate a range of kernel address space
1615 *      @size:          size of the area
1616 *
1617 *      Returns:        NULL on failure, vm_struct on success
1618 *
1619 *      This function reserves a range of kernel address space, and
1620 *      allocates pagetables to map that range.  No actual mappings
1621 *      are created.  If the kernel address space is not shared
1622 *      between processes, it syncs the pagetable across all
1623 *      processes.
1624 */
1625struct vm_struct *alloc_vm_area(size_t size)
1626{
1627        struct vm_struct *area;
1628
1629        area = get_vm_area_caller(size, VM_IOREMAP,
1630                                __builtin_return_address(0));
1631        if (area == NULL)
1632                return NULL;
1633
1634        /*
1635         * This ensures that page tables are constructed for this region
1636         * of kernel virtual address space and mapped into init_mm.
1637         */
1638        if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1639                                area->size, f, NULL)) {
1640                free_vm_area(area);
1641                return NULL;
1642        }
1643
1644        /* Make sure the pagetables are constructed in process kernel
1645           mappings */
1646        vmalloc_sync_all();
1647
1648        return area;
1649}
1650EXPORT_SYMBOL_GPL(alloc_vm_area);
1651
1652void free_vm_area(struct vm_struct *area)
1653{
1654        struct vm_struct *ret;
1655        ret = remove_vm_area(area->addr);
1656        BUG_ON(ret != area);
1657        kfree(area);
1658}
1659EXPORT_SYMBOL_GPL(free_vm_area);
1660
1661
1662#ifdef CONFIG_PROC_FS
1663static void *s_start(struct seq_file *m, loff_t *pos)
1664{
1665        loff_t n = *pos;
1666        struct vm_struct *v;
1667
1668        read_lock(&vmlist_lock);
1669        v = vmlist;
1670        while (n > 0 && v) {
1671                n--;
1672                v = v->next;
1673        }
1674        if (!n)
1675                return v;
1676
1677        return NULL;
1678
1679}
1680
1681static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1682{
1683        struct vm_struct *v = p;
1684
1685        ++*pos;
1686        return v->next;
1687}
1688
1689static void s_stop(struct seq_file *m, void *p)
1690{
1691        read_unlock(&vmlist_lock);
1692}
1693
1694static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1695{
1696        if (NUMA_BUILD) {
1697                unsigned int nr, *counters = m->private;
1698
1699                if (!counters)
1700                        return;
1701
1702                memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1703
1704                for (nr = 0; nr < v->nr_pages; nr++)
1705                        counters[page_to_nid(v->pages[nr])]++;
1706
1707                for_each_node_state(nr, N_HIGH_MEMORY)
1708                        if (counters[nr])
1709                                seq_printf(m, " N%u=%u", nr, counters[nr]);
1710        }
1711}
1712
1713static int s_show(struct seq_file *m, void *p)
1714{
1715        struct vm_struct *v = p;
1716
1717        seq_printf(m, "0x%p-0x%p %7ld",
1718                v->addr, v->addr + v->size, v->size);
1719
1720        if (v->caller) {
1721                char buff[KSYM_SYMBOL_LEN];
1722
1723                seq_putc(m, ' ');
1724                sprint_symbol(buff, (unsigned long)v->caller);
1725                seq_puts(m, buff);
1726        }
1727
1728        if (v->nr_pages)
1729                seq_printf(m, " pages=%d", v->nr_pages);
1730
1731        if (v->phys_addr)
1732                seq_printf(m, " phys=%lx", v->phys_addr);
1733
1734        if (v->flags & VM_IOREMAP)
1735                seq_printf(m, " ioremap");
1736
1737        if (v->flags & VM_ALLOC)
1738                seq_printf(m, " vmalloc");
1739
1740        if (v->flags & VM_MAP)
1741                seq_printf(m, " vmap");
1742
1743        if (v->flags & VM_USERMAP)
1744                seq_printf(m, " user");
1745
1746        if (v->flags & VM_VPAGES)
1747                seq_printf(m, " vpages");
1748
1749        show_numa_info(m, v);
1750        seq_putc(m, '\n');
1751        return 0;
1752}
1753
1754static const struct seq_operations vmalloc_op = {
1755        .start = s_start,
1756        .next = s_next,
1757        .stop = s_stop,
1758        .show = s_show,
1759};
1760
1761static int vmalloc_open(struct inode *inode, struct file *file)
1762{
1763        unsigned int *ptr = NULL;
1764        int ret;
1765
1766        if (NUMA_BUILD)
1767                ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1768        ret = seq_open(file, &vmalloc_op);
1769        if (!ret) {
1770                struct seq_file *m = file->private_data;
1771                m->private = ptr;
1772        } else
1773                kfree(ptr);
1774        return ret;
1775}
1776
1777static const struct file_operations proc_vmalloc_operations = {
1778        .open           = vmalloc_open,
1779        .read           = seq_read,
1780        .llseek         = seq_lseek,
1781        .release        = seq_release_private,
1782};
1783
1784static int __init proc_vmalloc_init(void)
1785{
1786        proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1787        return 0;
1788}
1789module_init(proc_vmalloc_init);
1790#endif
1791
1792