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