Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 8 | * Numa awareness, Christoph Lameter, SGI, June 2005 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | */ |
| 10 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 11 | #include <linux/vmalloc.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | #include <linux/mm.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/highmem.h> |
Alexey Dobriyan | d43c36d | 2009-10-07 17:09:06 +0400 | [diff] [blame] | 15 | #include <linux/sched.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/slab.h> |
| 17 | #include <linux/spinlock.h> |
| 18 | #include <linux/interrupt.h> |
Alexey Dobriyan | 5f6a6a9 | 2008-10-06 03:50:47 +0400 | [diff] [blame] | 19 | #include <linux/proc_fs.h> |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 20 | #include <linux/seq_file.h> |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 21 | #include <linux/debugobjects.h> |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 22 | #include <linux/kallsyms.h> |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 23 | #include <linux/list.h> |
| 24 | #include <linux/rbtree.h> |
| 25 | #include <linux/radix-tree.h> |
| 26 | #include <linux/rcupdate.h> |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 27 | #include <linux/pfn.h> |
Catalin Marinas | 89219d3 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 28 | #include <linux/kmemleak.h> |
Arun Sharma | 6006349 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 29 | #include <linux/atomic.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <asm/uaccess.h> |
| 31 | #include <asm/tlbflush.h> |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 32 | #include <asm/shmparam.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 34 | /*** Page table manipulation functions ***/ |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 35 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) |
| 37 | { |
| 38 | pte_t *pte; |
| 39 | |
| 40 | pte = pte_offset_kernel(pmd, addr); |
| 41 | do { |
| 42 | pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte); |
| 43 | WARN_ON(!pte_none(ptent) && !pte_present(ptent)); |
| 44 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 45 | } |
| 46 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 47 | static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | { |
| 49 | pmd_t *pmd; |
| 50 | unsigned long next; |
| 51 | |
| 52 | pmd = pmd_offset(pud, addr); |
| 53 | do { |
| 54 | next = pmd_addr_end(addr, end); |
| 55 | if (pmd_none_or_clear_bad(pmd)) |
| 56 | continue; |
| 57 | vunmap_pte_range(pmd, addr, next); |
| 58 | } while (pmd++, addr = next, addr != end); |
| 59 | } |
| 60 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 61 | static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | { |
| 63 | pud_t *pud; |
| 64 | unsigned long next; |
| 65 | |
| 66 | pud = pud_offset(pgd, addr); |
| 67 | do { |
| 68 | next = pud_addr_end(addr, end); |
| 69 | if (pud_none_or_clear_bad(pud)) |
| 70 | continue; |
| 71 | vunmap_pmd_range(pud, addr, next); |
| 72 | } while (pud++, addr = next, addr != end); |
| 73 | } |
| 74 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 75 | static void vunmap_page_range(unsigned long addr, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | { |
| 77 | pgd_t *pgd; |
| 78 | unsigned long next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | |
| 80 | BUG_ON(addr >= end); |
| 81 | pgd = pgd_offset_k(addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | do { |
| 83 | next = pgd_addr_end(addr, end); |
| 84 | if (pgd_none_or_clear_bad(pgd)) |
| 85 | continue; |
| 86 | vunmap_pud_range(pgd, addr, next); |
| 87 | } while (pgd++, addr = next, addr != end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | } |
| 89 | |
| 90 | static int vmap_pte_range(pmd_t *pmd, unsigned long addr, |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 91 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | { |
| 93 | pte_t *pte; |
| 94 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 95 | /* |
| 96 | * nr is a running index into the array which helps higher level |
| 97 | * callers keep track of where we're up to. |
| 98 | */ |
| 99 | |
Hugh Dickins | 872fec1 | 2005-10-29 18:16:21 -0700 | [diff] [blame] | 100 | pte = pte_alloc_kernel(pmd, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | if (!pte) |
| 102 | return -ENOMEM; |
| 103 | do { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 104 | struct page *page = pages[*nr]; |
| 105 | |
| 106 | if (WARN_ON(!pte_none(*pte))) |
| 107 | return -EBUSY; |
| 108 | if (WARN_ON(!page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | return -ENOMEM; |
| 110 | set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 111 | (*nr)++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 113 | return 0; |
| 114 | } |
| 115 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 116 | static int vmap_pmd_range(pud_t *pud, unsigned long addr, |
| 117 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | { |
| 119 | pmd_t *pmd; |
| 120 | unsigned long next; |
| 121 | |
| 122 | pmd = pmd_alloc(&init_mm, pud, addr); |
| 123 | if (!pmd) |
| 124 | return -ENOMEM; |
| 125 | do { |
| 126 | next = pmd_addr_end(addr, end); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 127 | if (vmap_pte_range(pmd, addr, next, prot, pages, nr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | return -ENOMEM; |
| 129 | } while (pmd++, addr = next, addr != end); |
| 130 | return 0; |
| 131 | } |
| 132 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 133 | static int vmap_pud_range(pgd_t *pgd, unsigned long addr, |
| 134 | unsigned long end, pgprot_t prot, struct page **pages, int *nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | { |
| 136 | pud_t *pud; |
| 137 | unsigned long next; |
| 138 | |
| 139 | pud = pud_alloc(&init_mm, pgd, addr); |
| 140 | if (!pud) |
| 141 | return -ENOMEM; |
| 142 | do { |
| 143 | next = pud_addr_end(addr, end); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 144 | if (vmap_pmd_range(pud, addr, next, prot, pages, nr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | return -ENOMEM; |
| 146 | } while (pud++, addr = next, addr != end); |
| 147 | return 0; |
| 148 | } |
| 149 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 150 | /* |
| 151 | * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and |
| 152 | * will have pfns corresponding to the "pages" array. |
| 153 | * |
| 154 | * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N] |
| 155 | */ |
Tejun Heo | 8fc4898 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 156 | static int vmap_page_range_noflush(unsigned long start, unsigned long end, |
| 157 | pgprot_t prot, struct page **pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | { |
| 159 | pgd_t *pgd; |
| 160 | unsigned long next; |
Adam Lackorzynski | 2e4e27c | 2009-01-04 12:00:46 -0800 | [diff] [blame] | 161 | unsigned long addr = start; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 162 | int err = 0; |
| 163 | int nr = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | |
| 165 | BUG_ON(addr >= end); |
| 166 | pgd = pgd_offset_k(addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | do { |
| 168 | next = pgd_addr_end(addr, end); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 169 | err = vmap_pud_range(pgd, addr, next, prot, pages, &nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | if (err) |
Figo.zhang | bf88c8c | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 171 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | } while (pgd++, addr = next, addr != end); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 173 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 174 | return nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | } |
| 176 | |
Tejun Heo | 8fc4898 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 177 | static int vmap_page_range(unsigned long start, unsigned long end, |
| 178 | pgprot_t prot, struct page **pages) |
| 179 | { |
| 180 | int ret; |
| 181 | |
| 182 | ret = vmap_page_range_noflush(start, end, prot, pages); |
| 183 | flush_cache_vmap(start, end); |
| 184 | return ret; |
| 185 | } |
| 186 | |
KAMEZAWA Hiroyuki | 81ac3ad | 2009-09-22 16:45:49 -0700 | [diff] [blame] | 187 | int is_vmalloc_or_module_addr(const void *x) |
Linus Torvalds | 73bdf0a | 2008-10-15 08:35:12 -0700 | [diff] [blame] | 188 | { |
| 189 | /* |
Russell King | ab4f2ee | 2008-11-06 17:11:07 +0000 | [diff] [blame] | 190 | * ARM, x86-64 and sparc64 put modules in a special place, |
Linus Torvalds | 73bdf0a | 2008-10-15 08:35:12 -0700 | [diff] [blame] | 191 | * and fall back on vmalloc() if that fails. Others |
| 192 | * just put it in the vmalloc space. |
| 193 | */ |
| 194 | #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) |
| 195 | unsigned long addr = (unsigned long)x; |
| 196 | if (addr >= MODULES_VADDR && addr < MODULES_END) |
| 197 | return 1; |
| 198 | #endif |
| 199 | return is_vmalloc_addr(x); |
| 200 | } |
| 201 | |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 202 | /* |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 203 | * Walk a vmap address to the struct page it maps. |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 204 | */ |
Christoph Lameter | b3bdda0 | 2008-02-04 22:28:32 -0800 | [diff] [blame] | 205 | struct page *vmalloc_to_page(const void *vmalloc_addr) |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 206 | { |
| 207 | unsigned long addr = (unsigned long) vmalloc_addr; |
| 208 | struct page *page = NULL; |
| 209 | pgd_t *pgd = pgd_offset_k(addr); |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 210 | |
Ingo Molnar | 7aa413d | 2008-06-19 13:28:11 +0200 | [diff] [blame] | 211 | /* |
| 212 | * XXX we might need to change this if we add VIRTUAL_BUG_ON for |
| 213 | * architectures that do not vmalloc module space |
| 214 | */ |
Linus Torvalds | 73bdf0a | 2008-10-15 08:35:12 -0700 | [diff] [blame] | 215 | VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr)); |
Jiri Slaby | 59ea746 | 2008-06-12 13:56:40 +0200 | [diff] [blame] | 216 | |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 217 | if (!pgd_none(*pgd)) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 218 | pud_t *pud = pud_offset(pgd, addr); |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 219 | if (!pud_none(*pud)) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 220 | pmd_t *pmd = pmd_offset(pud, addr); |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 221 | if (!pmd_none(*pmd)) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 222 | pte_t *ptep, pte; |
| 223 | |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 224 | ptep = pte_offset_map(pmd, addr); |
| 225 | pte = *ptep; |
| 226 | if (pte_present(pte)) |
| 227 | page = pte_page(pte); |
| 228 | pte_unmap(ptep); |
| 229 | } |
| 230 | } |
| 231 | } |
| 232 | return page; |
| 233 | } |
| 234 | EXPORT_SYMBOL(vmalloc_to_page); |
| 235 | |
| 236 | /* |
| 237 | * Map a vmalloc()-space virtual address to the physical page frame number. |
| 238 | */ |
Christoph Lameter | b3bdda0 | 2008-02-04 22:28:32 -0800 | [diff] [blame] | 239 | unsigned long vmalloc_to_pfn(const void *vmalloc_addr) |
Christoph Lameter | 48667e7 | 2008-02-04 22:28:31 -0800 | [diff] [blame] | 240 | { |
| 241 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); |
| 242 | } |
| 243 | EXPORT_SYMBOL(vmalloc_to_pfn); |
| 244 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 245 | |
| 246 | /*** Global kva allocator ***/ |
| 247 | |
| 248 | #define VM_LAZY_FREE 0x01 |
| 249 | #define VM_LAZY_FREEING 0x02 |
| 250 | #define VM_VM_AREA 0x04 |
| 251 | |
| 252 | struct vmap_area { |
| 253 | unsigned long va_start; |
| 254 | unsigned long va_end; |
| 255 | unsigned long flags; |
| 256 | struct rb_node rb_node; /* address sorted rbtree */ |
| 257 | struct list_head list; /* address sorted list */ |
| 258 | struct list_head purge_list; /* "lazy purge" list */ |
| 259 | void *private; |
| 260 | struct rcu_head rcu_head; |
| 261 | }; |
| 262 | |
| 263 | static DEFINE_SPINLOCK(vmap_area_lock); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 264 | static LIST_HEAD(vmap_area_list); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 265 | static struct rb_root vmap_area_root = RB_ROOT; |
| 266 | |
| 267 | /* The vmap cache globals are protected by vmap_area_lock */ |
| 268 | static struct rb_node *free_vmap_cache; |
| 269 | static unsigned long cached_hole_size; |
| 270 | static unsigned long cached_vstart; |
| 271 | static unsigned long cached_align; |
| 272 | |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 273 | static unsigned long vmap_area_pcpu_hole; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 274 | |
| 275 | static struct vmap_area *__find_vmap_area(unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 277 | struct rb_node *n = vmap_area_root.rb_node; |
| 278 | |
| 279 | while (n) { |
| 280 | struct vmap_area *va; |
| 281 | |
| 282 | va = rb_entry(n, struct vmap_area, rb_node); |
| 283 | if (addr < va->va_start) |
| 284 | n = n->rb_left; |
| 285 | else if (addr > va->va_start) |
| 286 | n = n->rb_right; |
| 287 | else |
| 288 | return va; |
| 289 | } |
| 290 | |
| 291 | return NULL; |
| 292 | } |
| 293 | |
| 294 | static void __insert_vmap_area(struct vmap_area *va) |
| 295 | { |
| 296 | struct rb_node **p = &vmap_area_root.rb_node; |
| 297 | struct rb_node *parent = NULL; |
| 298 | struct rb_node *tmp; |
| 299 | |
| 300 | while (*p) { |
Namhyung Kim | 170168d | 2010-10-26 14:22:02 -0700 | [diff] [blame] | 301 | struct vmap_area *tmp_va; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 302 | |
| 303 | parent = *p; |
Namhyung Kim | 170168d | 2010-10-26 14:22:02 -0700 | [diff] [blame] | 304 | tmp_va = rb_entry(parent, struct vmap_area, rb_node); |
| 305 | if (va->va_start < tmp_va->va_end) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 306 | p = &(*p)->rb_left; |
Namhyung Kim | 170168d | 2010-10-26 14:22:02 -0700 | [diff] [blame] | 307 | else if (va->va_end > tmp_va->va_start) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 308 | p = &(*p)->rb_right; |
| 309 | else |
| 310 | BUG(); |
| 311 | } |
| 312 | |
| 313 | rb_link_node(&va->rb_node, parent, p); |
| 314 | rb_insert_color(&va->rb_node, &vmap_area_root); |
| 315 | |
| 316 | /* address-sort this list so it is usable like the vmlist */ |
| 317 | tmp = rb_prev(&va->rb_node); |
| 318 | if (tmp) { |
| 319 | struct vmap_area *prev; |
| 320 | prev = rb_entry(tmp, struct vmap_area, rb_node); |
| 321 | list_add_rcu(&va->list, &prev->list); |
| 322 | } else |
| 323 | list_add_rcu(&va->list, &vmap_area_list); |
| 324 | } |
| 325 | |
| 326 | static void purge_vmap_area_lazy(void); |
| 327 | |
| 328 | /* |
| 329 | * Allocate a region of KVA of the specified size and alignment, within the |
| 330 | * vstart and vend. |
| 331 | */ |
| 332 | static struct vmap_area *alloc_vmap_area(unsigned long size, |
| 333 | unsigned long align, |
| 334 | unsigned long vstart, unsigned long vend, |
| 335 | int node, gfp_t gfp_mask) |
| 336 | { |
| 337 | struct vmap_area *va; |
| 338 | struct rb_node *n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | unsigned long addr; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 340 | int purged = 0; |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 341 | struct vmap_area *first; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 342 | |
Nick Piggin | 7766970 | 2009-02-27 14:03:03 -0800 | [diff] [blame] | 343 | BUG_ON(!size); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 344 | BUG_ON(size & ~PAGE_MASK); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 345 | BUG_ON(!is_power_of_2(align)); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 346 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 347 | va = kmalloc_node(sizeof(struct vmap_area), |
| 348 | gfp_mask & GFP_RECLAIM_MASK, node); |
| 349 | if (unlikely(!va)) |
| 350 | return ERR_PTR(-ENOMEM); |
| 351 | |
| 352 | retry: |
| 353 | spin_lock(&vmap_area_lock); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 354 | /* |
| 355 | * Invalidate cache if we have more permissive parameters. |
| 356 | * cached_hole_size notes the largest hole noticed _below_ |
| 357 | * the vmap_area cached in free_vmap_cache: if size fits |
| 358 | * into that hole, we want to scan from vstart to reuse |
| 359 | * the hole instead of allocating above free_vmap_cache. |
| 360 | * Note that __free_vmap_area may update free_vmap_cache |
| 361 | * without updating cached_hole_size or cached_align. |
| 362 | */ |
| 363 | if (!free_vmap_cache || |
| 364 | size < cached_hole_size || |
| 365 | vstart < cached_vstart || |
| 366 | align < cached_align) { |
| 367 | nocache: |
| 368 | cached_hole_size = 0; |
| 369 | free_vmap_cache = NULL; |
| 370 | } |
| 371 | /* record if we encounter less permissive parameters */ |
| 372 | cached_vstart = vstart; |
| 373 | cached_align = align; |
Nick Piggin | 7766970 | 2009-02-27 14:03:03 -0800 | [diff] [blame] | 374 | |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 375 | /* find starting point for our search */ |
| 376 | if (free_vmap_cache) { |
| 377 | first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); |
Johannes Weiner | 248ac0e | 2011-05-24 17:11:43 -0700 | [diff] [blame] | 378 | addr = ALIGN(first->va_end, align); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 379 | if (addr < vstart) |
| 380 | goto nocache; |
| 381 | if (addr + size - 1 < addr) |
| 382 | goto overflow; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 383 | |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 384 | } else { |
| 385 | addr = ALIGN(vstart, align); |
| 386 | if (addr + size - 1 < addr) |
| 387 | goto overflow; |
| 388 | |
| 389 | n = vmap_area_root.rb_node; |
| 390 | first = NULL; |
| 391 | |
| 392 | while (n) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 393 | struct vmap_area *tmp; |
| 394 | tmp = rb_entry(n, struct vmap_area, rb_node); |
| 395 | if (tmp->va_end >= addr) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 396 | first = tmp; |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 397 | if (tmp->va_start <= addr) |
| 398 | break; |
| 399 | n = n->rb_left; |
| 400 | } else |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 401 | n = n->rb_right; |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 402 | } |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 403 | |
| 404 | if (!first) |
| 405 | goto found; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 406 | } |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 407 | |
| 408 | /* from the starting point, walk areas until a suitable hole is found */ |
Johannes Weiner | 248ac0e | 2011-05-24 17:11:43 -0700 | [diff] [blame] | 409 | while (addr + size > first->va_start && addr + size <= vend) { |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 410 | if (addr + cached_hole_size < first->va_start) |
| 411 | cached_hole_size = first->va_start - addr; |
Johannes Weiner | 248ac0e | 2011-05-24 17:11:43 -0700 | [diff] [blame] | 412 | addr = ALIGN(first->va_end, align); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 413 | if (addr + size - 1 < addr) |
| 414 | goto overflow; |
| 415 | |
| 416 | n = rb_next(&first->rb_node); |
| 417 | if (n) |
| 418 | first = rb_entry(n, struct vmap_area, rb_node); |
| 419 | else |
| 420 | goto found; |
| 421 | } |
| 422 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 423 | found: |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 424 | if (addr + size > vend) |
| 425 | goto overflow; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 426 | |
| 427 | va->va_start = addr; |
| 428 | va->va_end = addr + size; |
| 429 | va->flags = 0; |
| 430 | __insert_vmap_area(va); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 431 | free_vmap_cache = &va->rb_node; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 432 | spin_unlock(&vmap_area_lock); |
| 433 | |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 434 | BUG_ON(va->va_start & (align-1)); |
| 435 | BUG_ON(va->va_start < vstart); |
| 436 | BUG_ON(va->va_end > vend); |
| 437 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 438 | return va; |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 439 | |
| 440 | overflow: |
| 441 | spin_unlock(&vmap_area_lock); |
| 442 | if (!purged) { |
| 443 | purge_vmap_area_lazy(); |
| 444 | purged = 1; |
| 445 | goto retry; |
| 446 | } |
| 447 | if (printk_ratelimit()) |
| 448 | printk(KERN_WARNING |
| 449 | "vmap allocation for size %lu failed: " |
| 450 | "use vmalloc=<size> to increase size.\n", size); |
| 451 | kfree(va); |
| 452 | return ERR_PTR(-EBUSY); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 453 | } |
| 454 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 455 | static void __free_vmap_area(struct vmap_area *va) |
| 456 | { |
| 457 | BUG_ON(RB_EMPTY_NODE(&va->rb_node)); |
Nick Piggin | 8969960 | 2011-03-22 16:30:36 -0700 | [diff] [blame] | 458 | |
| 459 | if (free_vmap_cache) { |
| 460 | if (va->va_end < cached_vstart) { |
| 461 | free_vmap_cache = NULL; |
| 462 | } else { |
| 463 | struct vmap_area *cache; |
| 464 | cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); |
| 465 | if (va->va_start <= cache->va_start) { |
| 466 | free_vmap_cache = rb_prev(&va->rb_node); |
| 467 | /* |
| 468 | * We don't try to update cached_hole_size or |
| 469 | * cached_align, but it won't go very wrong. |
| 470 | */ |
| 471 | } |
| 472 | } |
| 473 | } |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 474 | rb_erase(&va->rb_node, &vmap_area_root); |
| 475 | RB_CLEAR_NODE(&va->rb_node); |
| 476 | list_del_rcu(&va->list); |
| 477 | |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 478 | /* |
| 479 | * Track the highest possible candidate for pcpu area |
| 480 | * allocation. Areas outside of vmalloc area can be returned |
| 481 | * here too, consider only end addresses which fall inside |
| 482 | * vmalloc area proper. |
| 483 | */ |
| 484 | if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) |
| 485 | vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); |
| 486 | |
Lai Jiangshan | 14769de | 2011-03-18 12:12:19 +0800 | [diff] [blame] | 487 | kfree_rcu(va, rcu_head); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 488 | } |
| 489 | |
| 490 | /* |
| 491 | * Free a region of KVA allocated by alloc_vmap_area |
| 492 | */ |
| 493 | static void free_vmap_area(struct vmap_area *va) |
| 494 | { |
| 495 | spin_lock(&vmap_area_lock); |
| 496 | __free_vmap_area(va); |
| 497 | spin_unlock(&vmap_area_lock); |
| 498 | } |
| 499 | |
| 500 | /* |
| 501 | * Clear the pagetable entries of a given vmap_area |
| 502 | */ |
| 503 | static void unmap_vmap_area(struct vmap_area *va) |
| 504 | { |
| 505 | vunmap_page_range(va->va_start, va->va_end); |
| 506 | } |
| 507 | |
Nick Piggin | cd52858 | 2009-01-06 14:39:20 -0800 | [diff] [blame] | 508 | static void vmap_debug_free_range(unsigned long start, unsigned long end) |
| 509 | { |
| 510 | /* |
| 511 | * Unmap page tables and force a TLB flush immediately if |
| 512 | * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free |
| 513 | * bugs similarly to those in linear kernel virtual address |
| 514 | * space after a page has been freed. |
| 515 | * |
| 516 | * All the lazy freeing logic is still retained, in order to |
| 517 | * minimise intrusiveness of this debugging feature. |
| 518 | * |
| 519 | * This is going to be *slow* (linear kernel virtual address |
| 520 | * debugging doesn't do a broadcast TLB flush so it is a lot |
| 521 | * faster). |
| 522 | */ |
| 523 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 524 | vunmap_page_range(start, end); |
| 525 | flush_tlb_kernel_range(start, end); |
| 526 | #endif |
| 527 | } |
| 528 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 529 | /* |
| 530 | * lazy_max_pages is the maximum amount of virtual address space we gather up |
| 531 | * before attempting to purge with a TLB flush. |
| 532 | * |
| 533 | * There is a tradeoff here: a larger number will cover more kernel page tables |
| 534 | * and take slightly longer to purge, but it will linearly reduce the number of |
| 535 | * global TLB flushes that must be performed. It would seem natural to scale |
| 536 | * this number up linearly with the number of CPUs (because vmapping activity |
| 537 | * could also scale linearly with the number of CPUs), however it is likely |
| 538 | * that in practice, workloads might be constrained in other ways that mean |
| 539 | * vmap activity will not scale linearly with CPUs. Also, I want to be |
| 540 | * conservative and not introduce a big latency on huge systems, so go with |
| 541 | * a less aggressive log scale. It will still be an improvement over the old |
| 542 | * code, and it will be simple to change the scale factor if we find that it |
| 543 | * becomes a problem on bigger systems. |
| 544 | */ |
| 545 | static unsigned long lazy_max_pages(void) |
| 546 | { |
| 547 | unsigned int log; |
| 548 | |
| 549 | log = fls(num_online_cpus()); |
| 550 | |
| 551 | return log * (32UL * 1024 * 1024 / PAGE_SIZE); |
| 552 | } |
| 553 | |
| 554 | static atomic_t vmap_lazy_nr = ATOMIC_INIT(0); |
| 555 | |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 556 | /* for per-CPU blocks */ |
| 557 | static void purge_fragmented_blocks_allcpus(void); |
| 558 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 559 | /* |
Cliff Wickman | 3ee48b6 | 2010-09-16 11:44:02 -0500 | [diff] [blame] | 560 | * called before a call to iounmap() if the caller wants vm_area_struct's |
| 561 | * immediately freed. |
| 562 | */ |
| 563 | void set_iounmap_nonlazy(void) |
| 564 | { |
| 565 | atomic_set(&vmap_lazy_nr, lazy_max_pages()+1); |
| 566 | } |
| 567 | |
| 568 | /* |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 569 | * Purges all lazily-freed vmap areas. |
| 570 | * |
| 571 | * If sync is 0 then don't purge if there is already a purge in progress. |
| 572 | * If force_flush is 1, then flush kernel TLBs between *start and *end even |
| 573 | * if we found no lazy vmap areas to unmap (callers can use this to optimise |
| 574 | * their own TLB flushing). |
| 575 | * Returns with *start = min(*start, lowest purged address) |
| 576 | * *end = max(*end, highest purged address) |
| 577 | */ |
| 578 | static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end, |
| 579 | int sync, int force_flush) |
| 580 | { |
Andrew Morton | 46666d8 | 2009-01-15 13:51:15 -0800 | [diff] [blame] | 581 | static DEFINE_SPINLOCK(purge_lock); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 582 | LIST_HEAD(valist); |
| 583 | struct vmap_area *va; |
Vegard Nossum | cbb7667 | 2009-02-27 14:03:04 -0800 | [diff] [blame] | 584 | struct vmap_area *n_va; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 585 | int nr = 0; |
| 586 | |
| 587 | /* |
| 588 | * If sync is 0 but force_flush is 1, we'll go sync anyway but callers |
| 589 | * should not expect such behaviour. This just simplifies locking for |
| 590 | * the case that isn't actually used at the moment anyway. |
| 591 | */ |
| 592 | if (!sync && !force_flush) { |
Andrew Morton | 46666d8 | 2009-01-15 13:51:15 -0800 | [diff] [blame] | 593 | if (!spin_trylock(&purge_lock)) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 594 | return; |
| 595 | } else |
Andrew Morton | 46666d8 | 2009-01-15 13:51:15 -0800 | [diff] [blame] | 596 | spin_lock(&purge_lock); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 597 | |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 598 | if (sync) |
| 599 | purge_fragmented_blocks_allcpus(); |
| 600 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 601 | rcu_read_lock(); |
| 602 | list_for_each_entry_rcu(va, &vmap_area_list, list) { |
| 603 | if (va->flags & VM_LAZY_FREE) { |
| 604 | if (va->va_start < *start) |
| 605 | *start = va->va_start; |
| 606 | if (va->va_end > *end) |
| 607 | *end = va->va_end; |
| 608 | nr += (va->va_end - va->va_start) >> PAGE_SHIFT; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 609 | list_add_tail(&va->purge_list, &valist); |
| 610 | va->flags |= VM_LAZY_FREEING; |
| 611 | va->flags &= ~VM_LAZY_FREE; |
| 612 | } |
| 613 | } |
| 614 | rcu_read_unlock(); |
| 615 | |
Yongseok Koh | 88f5004 | 2010-01-19 17:33:49 +0900 | [diff] [blame] | 616 | if (nr) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 617 | atomic_sub(nr, &vmap_lazy_nr); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 618 | |
| 619 | if (nr || force_flush) |
| 620 | flush_tlb_kernel_range(*start, *end); |
| 621 | |
| 622 | if (nr) { |
| 623 | spin_lock(&vmap_area_lock); |
Vegard Nossum | cbb7667 | 2009-02-27 14:03:04 -0800 | [diff] [blame] | 624 | list_for_each_entry_safe(va, n_va, &valist, purge_list) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 625 | __free_vmap_area(va); |
| 626 | spin_unlock(&vmap_area_lock); |
| 627 | } |
Andrew Morton | 46666d8 | 2009-01-15 13:51:15 -0800 | [diff] [blame] | 628 | spin_unlock(&purge_lock); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 629 | } |
| 630 | |
| 631 | /* |
Nick Piggin | 496850e | 2008-11-19 15:36:33 -0800 | [diff] [blame] | 632 | * Kick off a purge of the outstanding lazy areas. Don't bother if somebody |
| 633 | * is already purging. |
| 634 | */ |
| 635 | static void try_purge_vmap_area_lazy(void) |
| 636 | { |
| 637 | unsigned long start = ULONG_MAX, end = 0; |
| 638 | |
| 639 | __purge_vmap_area_lazy(&start, &end, 0, 0); |
| 640 | } |
| 641 | |
| 642 | /* |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 643 | * Kick off a purge of the outstanding lazy areas. |
| 644 | */ |
| 645 | static void purge_vmap_area_lazy(void) |
| 646 | { |
| 647 | unsigned long start = ULONG_MAX, end = 0; |
| 648 | |
Nick Piggin | 496850e | 2008-11-19 15:36:33 -0800 | [diff] [blame] | 649 | __purge_vmap_area_lazy(&start, &end, 1, 0); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 650 | } |
| 651 | |
| 652 | /* |
Jeremy Fitzhardinge | 64141da | 2010-12-02 14:31:18 -0800 | [diff] [blame] | 653 | * Free a vmap area, caller ensuring that the area has been unmapped |
| 654 | * and flush_cache_vunmap had been called for the correct range |
| 655 | * previously. |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 656 | */ |
Jeremy Fitzhardinge | 64141da | 2010-12-02 14:31:18 -0800 | [diff] [blame] | 657 | static void free_vmap_area_noflush(struct vmap_area *va) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 658 | { |
| 659 | va->flags |= VM_LAZY_FREE; |
| 660 | atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr); |
| 661 | if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages())) |
Nick Piggin | 496850e | 2008-11-19 15:36:33 -0800 | [diff] [blame] | 662 | try_purge_vmap_area_lazy(); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 663 | } |
| 664 | |
Nick Piggin | b29acbd | 2008-12-01 13:13:47 -0800 | [diff] [blame] | 665 | /* |
Jeremy Fitzhardinge | 64141da | 2010-12-02 14:31:18 -0800 | [diff] [blame] | 666 | * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been |
| 667 | * called for the correct range previously. |
| 668 | */ |
| 669 | static void free_unmap_vmap_area_noflush(struct vmap_area *va) |
| 670 | { |
| 671 | unmap_vmap_area(va); |
| 672 | free_vmap_area_noflush(va); |
| 673 | } |
| 674 | |
| 675 | /* |
Nick Piggin | b29acbd | 2008-12-01 13:13:47 -0800 | [diff] [blame] | 676 | * Free and unmap a vmap area |
| 677 | */ |
| 678 | static void free_unmap_vmap_area(struct vmap_area *va) |
| 679 | { |
| 680 | flush_cache_vunmap(va->va_start, va->va_end); |
| 681 | free_unmap_vmap_area_noflush(va); |
| 682 | } |
| 683 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 684 | static struct vmap_area *find_vmap_area(unsigned long addr) |
| 685 | { |
| 686 | struct vmap_area *va; |
| 687 | |
| 688 | spin_lock(&vmap_area_lock); |
| 689 | va = __find_vmap_area(addr); |
| 690 | spin_unlock(&vmap_area_lock); |
| 691 | |
| 692 | return va; |
| 693 | } |
| 694 | |
| 695 | static void free_unmap_vmap_area_addr(unsigned long addr) |
| 696 | { |
| 697 | struct vmap_area *va; |
| 698 | |
| 699 | va = find_vmap_area(addr); |
| 700 | BUG_ON(!va); |
| 701 | free_unmap_vmap_area(va); |
| 702 | } |
| 703 | |
| 704 | |
| 705 | /*** Per cpu kva allocator ***/ |
| 706 | |
| 707 | /* |
| 708 | * vmap space is limited especially on 32 bit architectures. Ensure there is |
| 709 | * room for at least 16 percpu vmap blocks per CPU. |
| 710 | */ |
| 711 | /* |
| 712 | * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able |
| 713 | * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess |
| 714 | * instead (we just need a rough idea) |
| 715 | */ |
| 716 | #if BITS_PER_LONG == 32 |
| 717 | #define VMALLOC_SPACE (128UL*1024*1024) |
| 718 | #else |
| 719 | #define VMALLOC_SPACE (128UL*1024*1024*1024) |
| 720 | #endif |
| 721 | |
| 722 | #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE) |
| 723 | #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */ |
| 724 | #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */ |
| 725 | #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2) |
| 726 | #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */ |
| 727 | #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */ |
Clemens Ladisch | f982f91 | 2011-06-21 22:09:50 +0200 | [diff] [blame] | 728 | #define VMAP_BBMAP_BITS \ |
| 729 | VMAP_MIN(VMAP_BBMAP_BITS_MAX, \ |
| 730 | VMAP_MAX(VMAP_BBMAP_BITS_MIN, \ |
| 731 | VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16)) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 732 | |
| 733 | #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) |
| 734 | |
Jeremy Fitzhardinge | 9b46333 | 2008-10-28 19:22:34 +1100 | [diff] [blame] | 735 | static bool vmap_initialized __read_mostly = false; |
| 736 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 737 | struct vmap_block_queue { |
| 738 | spinlock_t lock; |
| 739 | struct list_head free; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 740 | }; |
| 741 | |
| 742 | struct vmap_block { |
| 743 | spinlock_t lock; |
| 744 | struct vmap_area *va; |
| 745 | struct vmap_block_queue *vbq; |
| 746 | unsigned long free, dirty; |
| 747 | DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS); |
| 748 | DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS); |
Nick Piggin | de56042 | 2010-02-01 22:24:18 +1100 | [diff] [blame] | 749 | struct list_head free_list; |
| 750 | struct rcu_head rcu_head; |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 751 | struct list_head purge; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 752 | }; |
| 753 | |
| 754 | /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */ |
| 755 | static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue); |
| 756 | |
| 757 | /* |
| 758 | * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block |
| 759 | * in the free path. Could get rid of this if we change the API to return a |
| 760 | * "cookie" from alloc, to be passed to free. But no big deal yet. |
| 761 | */ |
| 762 | static DEFINE_SPINLOCK(vmap_block_tree_lock); |
| 763 | static RADIX_TREE(vmap_block_tree, GFP_ATOMIC); |
| 764 | |
| 765 | /* |
| 766 | * We should probably have a fallback mechanism to allocate virtual memory |
| 767 | * out of partially filled vmap blocks. However vmap block sizing should be |
| 768 | * fairly reasonable according to the vmalloc size, so it shouldn't be a |
| 769 | * big problem. |
| 770 | */ |
| 771 | |
| 772 | static unsigned long addr_to_vb_idx(unsigned long addr) |
| 773 | { |
| 774 | addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1); |
| 775 | addr /= VMAP_BLOCK_SIZE; |
| 776 | return addr; |
| 777 | } |
| 778 | |
| 779 | static struct vmap_block *new_vmap_block(gfp_t gfp_mask) |
| 780 | { |
| 781 | struct vmap_block_queue *vbq; |
| 782 | struct vmap_block *vb; |
| 783 | struct vmap_area *va; |
| 784 | unsigned long vb_idx; |
| 785 | int node, err; |
| 786 | |
| 787 | node = numa_node_id(); |
| 788 | |
| 789 | vb = kmalloc_node(sizeof(struct vmap_block), |
| 790 | gfp_mask & GFP_RECLAIM_MASK, node); |
| 791 | if (unlikely(!vb)) |
| 792 | return ERR_PTR(-ENOMEM); |
| 793 | |
| 794 | va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE, |
| 795 | VMALLOC_START, VMALLOC_END, |
| 796 | node, gfp_mask); |
Tobias Klauser | ddf9c6d | 2011-01-13 15:46:15 -0800 | [diff] [blame] | 797 | if (IS_ERR(va)) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 798 | kfree(vb); |
Julia Lawall | e7d8634 | 2010-08-09 17:18:28 -0700 | [diff] [blame] | 799 | return ERR_CAST(va); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 800 | } |
| 801 | |
| 802 | err = radix_tree_preload(gfp_mask); |
| 803 | if (unlikely(err)) { |
| 804 | kfree(vb); |
| 805 | free_vmap_area(va); |
| 806 | return ERR_PTR(err); |
| 807 | } |
| 808 | |
| 809 | spin_lock_init(&vb->lock); |
| 810 | vb->va = va; |
| 811 | vb->free = VMAP_BBMAP_BITS; |
| 812 | vb->dirty = 0; |
| 813 | bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS); |
| 814 | bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS); |
| 815 | INIT_LIST_HEAD(&vb->free_list); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 816 | |
| 817 | vb_idx = addr_to_vb_idx(va->va_start); |
| 818 | spin_lock(&vmap_block_tree_lock); |
| 819 | err = radix_tree_insert(&vmap_block_tree, vb_idx, vb); |
| 820 | spin_unlock(&vmap_block_tree_lock); |
| 821 | BUG_ON(err); |
| 822 | radix_tree_preload_end(); |
| 823 | |
| 824 | vbq = &get_cpu_var(vmap_block_queue); |
| 825 | vb->vbq = vbq; |
| 826 | spin_lock(&vbq->lock); |
Nick Piggin | de56042 | 2010-02-01 22:24:18 +1100 | [diff] [blame] | 827 | list_add_rcu(&vb->free_list, &vbq->free); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 828 | spin_unlock(&vbq->lock); |
Tejun Heo | 3f04ba8 | 2009-10-29 22:34:12 +0900 | [diff] [blame] | 829 | put_cpu_var(vmap_block_queue); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 830 | |
| 831 | return vb; |
| 832 | } |
| 833 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 834 | static void free_vmap_block(struct vmap_block *vb) |
| 835 | { |
| 836 | struct vmap_block *tmp; |
| 837 | unsigned long vb_idx; |
| 838 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 839 | vb_idx = addr_to_vb_idx(vb->va->va_start); |
| 840 | spin_lock(&vmap_block_tree_lock); |
| 841 | tmp = radix_tree_delete(&vmap_block_tree, vb_idx); |
| 842 | spin_unlock(&vmap_block_tree_lock); |
| 843 | BUG_ON(tmp != vb); |
| 844 | |
Jeremy Fitzhardinge | 64141da | 2010-12-02 14:31:18 -0800 | [diff] [blame] | 845 | free_vmap_area_noflush(vb->va); |
Lai Jiangshan | 22a3c7d | 2011-03-18 12:13:08 +0800 | [diff] [blame] | 846 | kfree_rcu(vb, rcu_head); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 847 | } |
| 848 | |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 849 | static void purge_fragmented_blocks(int cpu) |
| 850 | { |
| 851 | LIST_HEAD(purge); |
| 852 | struct vmap_block *vb; |
| 853 | struct vmap_block *n_vb; |
| 854 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); |
| 855 | |
| 856 | rcu_read_lock(); |
| 857 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
| 858 | |
| 859 | if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS)) |
| 860 | continue; |
| 861 | |
| 862 | spin_lock(&vb->lock); |
| 863 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) { |
| 864 | vb->free = 0; /* prevent further allocs after releasing lock */ |
| 865 | vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */ |
| 866 | bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS); |
| 867 | bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS); |
| 868 | spin_lock(&vbq->lock); |
| 869 | list_del_rcu(&vb->free_list); |
| 870 | spin_unlock(&vbq->lock); |
| 871 | spin_unlock(&vb->lock); |
| 872 | list_add_tail(&vb->purge, &purge); |
| 873 | } else |
| 874 | spin_unlock(&vb->lock); |
| 875 | } |
| 876 | rcu_read_unlock(); |
| 877 | |
| 878 | list_for_each_entry_safe(vb, n_vb, &purge, purge) { |
| 879 | list_del(&vb->purge); |
| 880 | free_vmap_block(vb); |
| 881 | } |
| 882 | } |
| 883 | |
| 884 | static void purge_fragmented_blocks_thiscpu(void) |
| 885 | { |
| 886 | purge_fragmented_blocks(smp_processor_id()); |
| 887 | } |
| 888 | |
| 889 | static void purge_fragmented_blocks_allcpus(void) |
| 890 | { |
| 891 | int cpu; |
| 892 | |
| 893 | for_each_possible_cpu(cpu) |
| 894 | purge_fragmented_blocks(cpu); |
| 895 | } |
| 896 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 897 | static 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; |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 903 | int purge = 0; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 904 | |
| 905 | BUG_ON(size & ~PAGE_MASK); |
| 906 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
| 907 | order = get_order(size); |
| 908 | |
| 909 | again: |
| 910 | rcu_read_lock(); |
| 911 | vbq = &get_cpu_var(vmap_block_queue); |
| 912 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
| 913 | int i; |
| 914 | |
| 915 | spin_lock(&vb->lock); |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 916 | if (vb->free < 1UL << order) |
| 917 | goto next; |
| 918 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 919 | i = bitmap_find_free_region(vb->alloc_map, |
| 920 | VMAP_BBMAP_BITS, order); |
| 921 | |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 922 | if (i < 0) { |
| 923 | if (vb->free + vb->dirty == VMAP_BBMAP_BITS) { |
| 924 | /* fragmented and no outstanding allocations */ |
| 925 | BUG_ON(vb->dirty != VMAP_BBMAP_BITS); |
| 926 | purge = 1; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 927 | } |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 928 | goto next; |
| 929 | } |
| 930 | addr = vb->va->va_start + (i << PAGE_SHIFT); |
| 931 | BUG_ON(addr_to_vb_idx(addr) != |
| 932 | addr_to_vb_idx(vb->va->va_start)); |
| 933 | vb->free -= 1UL << order; |
| 934 | if (vb->free == 0) { |
| 935 | spin_lock(&vbq->lock); |
| 936 | list_del_rcu(&vb->free_list); |
| 937 | spin_unlock(&vbq->lock); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 938 | } |
| 939 | spin_unlock(&vb->lock); |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 940 | break; |
| 941 | next: |
| 942 | spin_unlock(&vb->lock); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 943 | } |
Nick Piggin | 02b709d | 2010-02-01 22:25:57 +1100 | [diff] [blame] | 944 | |
| 945 | if (purge) |
| 946 | purge_fragmented_blocks_thiscpu(); |
| 947 | |
Tejun Heo | 3f04ba8 | 2009-10-29 22:34:12 +0900 | [diff] [blame] | 948 | put_cpu_var(vmap_block_queue); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 949 | rcu_read_unlock(); |
| 950 | |
| 951 | if (!addr) { |
| 952 | vb = new_vmap_block(gfp_mask); |
| 953 | if (IS_ERR(vb)) |
| 954 | return vb; |
| 955 | goto again; |
| 956 | } |
| 957 | |
| 958 | return (void *)addr; |
| 959 | } |
| 960 | |
| 961 | static void vb_free(const void *addr, unsigned long size) |
| 962 | { |
| 963 | unsigned long offset; |
| 964 | unsigned long vb_idx; |
| 965 | unsigned int order; |
| 966 | struct vmap_block *vb; |
| 967 | |
| 968 | BUG_ON(size & ~PAGE_MASK); |
| 969 | BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC); |
Nick Piggin | b29acbd | 2008-12-01 13:13:47 -0800 | [diff] [blame] | 970 | |
| 971 | flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size); |
| 972 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 973 | order = get_order(size); |
| 974 | |
| 975 | offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1); |
| 976 | |
| 977 | vb_idx = addr_to_vb_idx((unsigned long)addr); |
| 978 | rcu_read_lock(); |
| 979 | vb = radix_tree_lookup(&vmap_block_tree, vb_idx); |
| 980 | rcu_read_unlock(); |
| 981 | BUG_ON(!vb); |
| 982 | |
Jeremy Fitzhardinge | 64141da | 2010-12-02 14:31:18 -0800 | [diff] [blame] | 983 | vunmap_page_range((unsigned long)addr, (unsigned long)addr + size); |
| 984 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 985 | spin_lock(&vb->lock); |
Nick Piggin | de56042 | 2010-02-01 22:24:18 +1100 | [diff] [blame] | 986 | BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order)); |
MinChan Kim | d086817 | 2009-03-31 15:19:26 -0700 | [diff] [blame] | 987 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 988 | vb->dirty += 1UL << order; |
| 989 | if (vb->dirty == VMAP_BBMAP_BITS) { |
Nick Piggin | de56042 | 2010-02-01 22:24:18 +1100 | [diff] [blame] | 990 | BUG_ON(vb->free); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 991 | spin_unlock(&vb->lock); |
| 992 | free_vmap_block(vb); |
| 993 | } else |
| 994 | spin_unlock(&vb->lock); |
| 995 | } |
| 996 | |
| 997 | /** |
| 998 | * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer |
| 999 | * |
| 1000 | * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily |
| 1001 | * to amortize TLB flushing overheads. What this means is that any page you |
| 1002 | * have now, may, in a former life, have been mapped into kernel virtual |
| 1003 | * address by the vmap layer and so there might be some CPUs with TLB entries |
| 1004 | * still referencing that page (additional to the regular 1:1 kernel mapping). |
| 1005 | * |
| 1006 | * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can |
| 1007 | * be sure that none of the pages we have control over will have any aliases |
| 1008 | * from the vmap layer. |
| 1009 | */ |
| 1010 | void vm_unmap_aliases(void) |
| 1011 | { |
| 1012 | unsigned long start = ULONG_MAX, end = 0; |
| 1013 | int cpu; |
| 1014 | int flush = 0; |
| 1015 | |
Jeremy Fitzhardinge | 9b46333 | 2008-10-28 19:22:34 +1100 | [diff] [blame] | 1016 | if (unlikely(!vmap_initialized)) |
| 1017 | return; |
| 1018 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1019 | for_each_possible_cpu(cpu) { |
| 1020 | struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu); |
| 1021 | struct vmap_block *vb; |
| 1022 | |
| 1023 | rcu_read_lock(); |
| 1024 | list_for_each_entry_rcu(vb, &vbq->free, free_list) { |
| 1025 | int i; |
| 1026 | |
| 1027 | spin_lock(&vb->lock); |
| 1028 | i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS); |
| 1029 | while (i < VMAP_BBMAP_BITS) { |
| 1030 | unsigned long s, e; |
| 1031 | int j; |
| 1032 | j = find_next_zero_bit(vb->dirty_map, |
| 1033 | VMAP_BBMAP_BITS, i); |
| 1034 | |
| 1035 | s = vb->va->va_start + (i << PAGE_SHIFT); |
| 1036 | e = vb->va->va_start + (j << PAGE_SHIFT); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1037 | flush = 1; |
| 1038 | |
| 1039 | if (s < start) |
| 1040 | start = s; |
| 1041 | if (e > end) |
| 1042 | end = e; |
| 1043 | |
| 1044 | i = j; |
| 1045 | i = find_next_bit(vb->dirty_map, |
| 1046 | VMAP_BBMAP_BITS, i); |
| 1047 | } |
| 1048 | spin_unlock(&vb->lock); |
| 1049 | } |
| 1050 | rcu_read_unlock(); |
| 1051 | } |
| 1052 | |
| 1053 | __purge_vmap_area_lazy(&start, &end, 1, flush); |
| 1054 | } |
| 1055 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
| 1056 | |
| 1057 | /** |
| 1058 | * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram |
| 1059 | * @mem: the pointer returned by vm_map_ram |
| 1060 | * @count: the count passed to that vm_map_ram call (cannot unmap partial) |
| 1061 | */ |
| 1062 | void vm_unmap_ram(const void *mem, unsigned int count) |
| 1063 | { |
| 1064 | unsigned long size = count << PAGE_SHIFT; |
| 1065 | unsigned long addr = (unsigned long)mem; |
| 1066 | |
| 1067 | BUG_ON(!addr); |
| 1068 | BUG_ON(addr < VMALLOC_START); |
| 1069 | BUG_ON(addr > VMALLOC_END); |
| 1070 | BUG_ON(addr & (PAGE_SIZE-1)); |
| 1071 | |
| 1072 | debug_check_no_locks_freed(mem, size); |
Nick Piggin | cd52858 | 2009-01-06 14:39:20 -0800 | [diff] [blame] | 1073 | vmap_debug_free_range(addr, addr+size); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1074 | |
| 1075 | if (likely(count <= VMAP_MAX_ALLOC)) |
| 1076 | vb_free(mem, size); |
| 1077 | else |
| 1078 | free_unmap_vmap_area_addr(addr); |
| 1079 | } |
| 1080 | EXPORT_SYMBOL(vm_unmap_ram); |
| 1081 | |
| 1082 | /** |
| 1083 | * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space) |
| 1084 | * @pages: an array of pointers to the pages to be mapped |
| 1085 | * @count: number of pages |
| 1086 | * @node: prefer to allocate data structures on this node |
| 1087 | * @prot: memory protection to use. PAGE_KERNEL for regular RAM |
Randy Dunlap | e99c97a | 2008-10-29 14:01:09 -0700 | [diff] [blame] | 1088 | * |
| 1089 | * Returns: a pointer to the address that has been mapped, or %NULL on failure |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1090 | */ |
| 1091 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
| 1092 | { |
| 1093 | unsigned long size = count << PAGE_SHIFT; |
| 1094 | unsigned long addr; |
| 1095 | void *mem; |
| 1096 | |
| 1097 | if (likely(count <= VMAP_MAX_ALLOC)) { |
| 1098 | mem = vb_alloc(size, GFP_KERNEL); |
| 1099 | if (IS_ERR(mem)) |
| 1100 | return NULL; |
| 1101 | addr = (unsigned long)mem; |
| 1102 | } else { |
| 1103 | struct vmap_area *va; |
| 1104 | va = alloc_vmap_area(size, PAGE_SIZE, |
| 1105 | VMALLOC_START, VMALLOC_END, node, GFP_KERNEL); |
| 1106 | if (IS_ERR(va)) |
| 1107 | return NULL; |
| 1108 | |
| 1109 | addr = va->va_start; |
| 1110 | mem = (void *)addr; |
| 1111 | } |
| 1112 | if (vmap_page_range(addr, addr + size, prot, pages) < 0) { |
| 1113 | vm_unmap_ram(mem, count); |
| 1114 | return NULL; |
| 1115 | } |
| 1116 | return mem; |
| 1117 | } |
| 1118 | EXPORT_SYMBOL(vm_map_ram); |
| 1119 | |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1120 | /** |
Nicolas Pitre | be9b733 | 2011-08-25 00:24:21 -0400 | [diff] [blame^] | 1121 | * vm_area_add_early - add vmap area early during boot |
| 1122 | * @vm: vm_struct to add |
| 1123 | * |
| 1124 | * This function is used to add fixed kernel vm area to vmlist before |
| 1125 | * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags |
| 1126 | * should contain proper values and the other fields should be zero. |
| 1127 | * |
| 1128 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. |
| 1129 | */ |
| 1130 | void __init vm_area_add_early(struct vm_struct *vm) |
| 1131 | { |
| 1132 | struct vm_struct *tmp, **p; |
| 1133 | |
| 1134 | BUG_ON(vmap_initialized); |
| 1135 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { |
| 1136 | if (tmp->addr >= vm->addr) { |
| 1137 | BUG_ON(tmp->addr < vm->addr + vm->size); |
| 1138 | break; |
| 1139 | } else |
| 1140 | BUG_ON(tmp->addr + tmp->size > vm->addr); |
| 1141 | } |
| 1142 | vm->next = *p; |
| 1143 | *p = vm; |
| 1144 | } |
| 1145 | |
| 1146 | /** |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1147 | * vm_area_register_early - register vmap area early during boot |
| 1148 | * @vm: vm_struct to register |
Tejun Heo | c0c0a29 | 2009-02-24 11:57:21 +0900 | [diff] [blame] | 1149 | * @align: requested alignment |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1150 | * |
| 1151 | * This function is used to register kernel vm area before |
| 1152 | * vmalloc_init() is called. @vm->size and @vm->flags should contain |
| 1153 | * proper values on entry and other fields should be zero. On return, |
| 1154 | * vm->addr contains the allocated address. |
| 1155 | * |
| 1156 | * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING. |
| 1157 | */ |
Tejun Heo | c0c0a29 | 2009-02-24 11:57:21 +0900 | [diff] [blame] | 1158 | void __init vm_area_register_early(struct vm_struct *vm, size_t align) |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1159 | { |
| 1160 | static size_t vm_init_off __initdata; |
Tejun Heo | c0c0a29 | 2009-02-24 11:57:21 +0900 | [diff] [blame] | 1161 | unsigned long addr; |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1162 | |
Tejun Heo | c0c0a29 | 2009-02-24 11:57:21 +0900 | [diff] [blame] | 1163 | addr = ALIGN(VMALLOC_START + vm_init_off, align); |
| 1164 | vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START; |
| 1165 | |
| 1166 | vm->addr = (void *)addr; |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1167 | |
Nicolas Pitre | be9b733 | 2011-08-25 00:24:21 -0400 | [diff] [blame^] | 1168 | vm_area_add_early(vm); |
Tejun Heo | f0aa661 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1169 | } |
| 1170 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1171 | void __init vmalloc_init(void) |
| 1172 | { |
Ivan Kokshaysky | 822c18f | 2009-01-15 13:50:48 -0800 | [diff] [blame] | 1173 | struct vmap_area *va; |
| 1174 | struct vm_struct *tmp; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1175 | int i; |
| 1176 | |
| 1177 | for_each_possible_cpu(i) { |
| 1178 | struct vmap_block_queue *vbq; |
| 1179 | |
| 1180 | vbq = &per_cpu(vmap_block_queue, i); |
| 1181 | spin_lock_init(&vbq->lock); |
| 1182 | INIT_LIST_HEAD(&vbq->free); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1183 | } |
Jeremy Fitzhardinge | 9b46333 | 2008-10-28 19:22:34 +1100 | [diff] [blame] | 1184 | |
Ivan Kokshaysky | 822c18f | 2009-01-15 13:50:48 -0800 | [diff] [blame] | 1185 | /* Import existing vmlist entries. */ |
| 1186 | for (tmp = vmlist; tmp; tmp = tmp->next) { |
Pekka Enberg | 43ebdac | 2009-05-25 15:01:35 +0300 | [diff] [blame] | 1187 | va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); |
Ivan Kokshaysky | 822c18f | 2009-01-15 13:50:48 -0800 | [diff] [blame] | 1188 | va->flags = tmp->flags | VM_VM_AREA; |
| 1189 | va->va_start = (unsigned long)tmp->addr; |
| 1190 | va->va_end = va->va_start + tmp->size; |
| 1191 | __insert_vmap_area(va); |
| 1192 | } |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 1193 | |
| 1194 | vmap_area_pcpu_hole = VMALLOC_END; |
| 1195 | |
Jeremy Fitzhardinge | 9b46333 | 2008-10-28 19:22:34 +1100 | [diff] [blame] | 1196 | vmap_initialized = true; |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1197 | } |
| 1198 | |
Tejun Heo | 8fc4898 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1199 | /** |
| 1200 | * map_kernel_range_noflush - map kernel VM area with the specified pages |
| 1201 | * @addr: start of the VM area to map |
| 1202 | * @size: size of the VM area to map |
| 1203 | * @prot: page protection flags to use |
| 1204 | * @pages: pages to map |
| 1205 | * |
| 1206 | * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size |
| 1207 | * specify should have been allocated using get_vm_area() and its |
| 1208 | * friends. |
| 1209 | * |
| 1210 | * NOTE: |
| 1211 | * This function does NOT do any cache flushing. The caller is |
| 1212 | * responsible for calling flush_cache_vmap() on to-be-mapped areas |
| 1213 | * before calling this function. |
| 1214 | * |
| 1215 | * RETURNS: |
| 1216 | * The number of pages mapped on success, -errno on failure. |
| 1217 | */ |
| 1218 | int map_kernel_range_noflush(unsigned long addr, unsigned long size, |
| 1219 | pgprot_t prot, struct page **pages) |
| 1220 | { |
| 1221 | return vmap_page_range_noflush(addr, addr + size, prot, pages); |
| 1222 | } |
| 1223 | |
| 1224 | /** |
| 1225 | * unmap_kernel_range_noflush - unmap kernel VM area |
| 1226 | * @addr: start of the VM area to unmap |
| 1227 | * @size: size of the VM area to unmap |
| 1228 | * |
| 1229 | * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size |
| 1230 | * specify should have been allocated using get_vm_area() and its |
| 1231 | * friends. |
| 1232 | * |
| 1233 | * NOTE: |
| 1234 | * This function does NOT do any cache flushing. The caller is |
| 1235 | * responsible for calling flush_cache_vunmap() on to-be-mapped areas |
| 1236 | * before calling this function and flush_tlb_kernel_range() after. |
| 1237 | */ |
| 1238 | void unmap_kernel_range_noflush(unsigned long addr, unsigned long size) |
| 1239 | { |
| 1240 | vunmap_page_range(addr, addr + size); |
| 1241 | } |
Huang Ying | 81e88fd | 2011-01-12 14:44:55 +0800 | [diff] [blame] | 1242 | EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush); |
Tejun Heo | 8fc4898 | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 1243 | |
| 1244 | /** |
| 1245 | * unmap_kernel_range - unmap kernel VM area and flush cache and TLB |
| 1246 | * @addr: start of the VM area to unmap |
| 1247 | * @size: size of the VM area to unmap |
| 1248 | * |
| 1249 | * Similar to unmap_kernel_range_noflush() but flushes vcache before |
| 1250 | * the unmapping and tlb after. |
| 1251 | */ |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1252 | void unmap_kernel_range(unsigned long addr, unsigned long size) |
| 1253 | { |
| 1254 | unsigned long end = addr + size; |
Tejun Heo | f6fcba7 | 2009-02-20 15:38:48 -0800 | [diff] [blame] | 1255 | |
| 1256 | flush_cache_vunmap(addr, end); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1257 | vunmap_page_range(addr, end); |
| 1258 | flush_tlb_kernel_range(addr, end); |
| 1259 | } |
| 1260 | |
| 1261 | int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) |
| 1262 | { |
| 1263 | unsigned long addr = (unsigned long)area->addr; |
| 1264 | unsigned long end = addr + area->size - PAGE_SIZE; |
| 1265 | int err; |
| 1266 | |
| 1267 | err = vmap_page_range(addr, end, prot, *pages); |
| 1268 | if (err > 0) { |
| 1269 | *pages += err; |
| 1270 | err = 0; |
| 1271 | } |
| 1272 | |
| 1273 | return err; |
| 1274 | } |
| 1275 | EXPORT_SYMBOL_GPL(map_vm_area); |
| 1276 | |
| 1277 | /*** Old vmalloc interfaces ***/ |
| 1278 | DEFINE_RWLOCK(vmlist_lock); |
| 1279 | struct vm_struct *vmlist; |
| 1280 | |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1281 | static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
Tejun Heo | cf88c79 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 1282 | unsigned long flags, void *caller) |
| 1283 | { |
Tejun Heo | cf88c79 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 1284 | vm->flags = flags; |
| 1285 | vm->addr = (void *)va->va_start; |
| 1286 | vm->size = va->va_end - va->va_start; |
| 1287 | vm->caller = caller; |
| 1288 | va->private = vm; |
| 1289 | va->flags |= VM_VM_AREA; |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1290 | } |
Tejun Heo | cf88c79 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 1291 | |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1292 | static void insert_vmalloc_vmlist(struct vm_struct *vm) |
| 1293 | { |
| 1294 | struct vm_struct *tmp, **p; |
| 1295 | |
| 1296 | vm->flags &= ~VM_UNLIST; |
Tejun Heo | cf88c79 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 1297 | write_lock(&vmlist_lock); |
| 1298 | for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) { |
| 1299 | if (tmp->addr >= vm->addr) |
| 1300 | break; |
| 1301 | } |
| 1302 | vm->next = *p; |
| 1303 | *p = vm; |
| 1304 | write_unlock(&vmlist_lock); |
| 1305 | } |
| 1306 | |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1307 | static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va, |
| 1308 | unsigned long flags, void *caller) |
| 1309 | { |
| 1310 | setup_vmalloc_vm(vm, va, flags, caller); |
| 1311 | insert_vmalloc_vmlist(vm); |
| 1312 | } |
| 1313 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1314 | static struct vm_struct *__get_vm_area_node(unsigned long size, |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1315 | unsigned long align, unsigned long flags, unsigned long start, |
| 1316 | unsigned long end, int node, gfp_t gfp_mask, void *caller) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1317 | { |
| 1318 | static struct vmap_area *va; |
| 1319 | struct vm_struct *area; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | |
Giridhar Pemmasani | 52fd24c | 2006-10-28 10:38:34 -0700 | [diff] [blame] | 1321 | BUG_ON(in_interrupt()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1322 | if (flags & VM_IOREMAP) { |
| 1323 | int bit = fls(size); |
| 1324 | |
| 1325 | if (bit > IOREMAP_MAX_ORDER) |
| 1326 | bit = IOREMAP_MAX_ORDER; |
| 1327 | else if (bit < PAGE_SHIFT) |
| 1328 | bit = PAGE_SHIFT; |
| 1329 | |
| 1330 | align = 1ul << bit; |
| 1331 | } |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1332 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | size = PAGE_ALIGN(size); |
OGAWA Hirofumi | 31be830 | 2006-11-16 01:19:29 -0800 | [diff] [blame] | 1334 | if (unlikely(!size)) |
| 1335 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1336 | |
Tejun Heo | cf88c79 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 1337 | area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | if (unlikely(!area)) |
| 1339 | return NULL; |
| 1340 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | /* |
| 1342 | * We always allocate a guard page. |
| 1343 | */ |
| 1344 | size += PAGE_SIZE; |
| 1345 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1346 | va = alloc_vmap_area(size, align, start, end, node, gfp_mask); |
| 1347 | if (IS_ERR(va)) { |
| 1348 | kfree(area); |
| 1349 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1352 | /* |
| 1353 | * When this function is called from __vmalloc_node_range, |
| 1354 | * we do not add vm_struct to vmlist here to avoid |
| 1355 | * accessing uninitialized members of vm_struct such as |
| 1356 | * pages and nr_pages fields. They will be set later. |
| 1357 | * To distinguish it from others, we use a VM_UNLIST flag. |
| 1358 | */ |
| 1359 | if (flags & VM_UNLIST) |
| 1360 | setup_vmalloc_vm(area, va, flags, caller); |
| 1361 | else |
| 1362 | insert_vmalloc_vm(area, va, flags, caller); |
| 1363 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | return area; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | } |
| 1366 | |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1367 | struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, |
| 1368 | unsigned long start, unsigned long end) |
| 1369 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1370 | return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1371 | __builtin_return_address(0)); |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1372 | } |
Rusty Russell | 5992b6d | 2007-07-19 01:49:21 -0700 | [diff] [blame] | 1373 | EXPORT_SYMBOL_GPL(__get_vm_area); |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1374 | |
Benjamin Herrenschmidt | c296861 | 2009-02-18 14:48:12 -0800 | [diff] [blame] | 1375 | struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags, |
| 1376 | unsigned long start, unsigned long end, |
| 1377 | void *caller) |
| 1378 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1379 | return __get_vm_area_node(size, 1, flags, start, end, -1, GFP_KERNEL, |
Benjamin Herrenschmidt | c296861 | 2009-02-18 14:48:12 -0800 | [diff] [blame] | 1380 | caller); |
| 1381 | } |
| 1382 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | /** |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 1384 | * get_vm_area - reserve a contiguous kernel virtual area |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | * @size: size of the area |
| 1386 | * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC |
| 1387 | * |
| 1388 | * Search an area of @size in the kernel virtual mapping area, |
| 1389 | * and reserved it for out purposes. Returns the area descriptor |
| 1390 | * on success or %NULL on failure. |
| 1391 | */ |
| 1392 | struct vm_struct *get_vm_area(unsigned long size, unsigned long flags) |
| 1393 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1394 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1395 | -1, GFP_KERNEL, __builtin_return_address(0)); |
| 1396 | } |
| 1397 | |
| 1398 | struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags, |
| 1399 | void *caller) |
| 1400 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1401 | return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1402 | -1, GFP_KERNEL, caller); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | } |
| 1404 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1405 | static struct vm_struct *find_vm_area(const void *addr) |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1406 | { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1407 | struct vmap_area *va; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1408 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1409 | va = find_vmap_area((unsigned long)addr); |
| 1410 | if (va && va->flags & VM_VM_AREA) |
| 1411 | return va->private; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1412 | |
Andi Kleen | 7856dfe | 2005-05-20 14:27:57 -0700 | [diff] [blame] | 1413 | return NULL; |
Andi Kleen | 7856dfe | 2005-05-20 14:27:57 -0700 | [diff] [blame] | 1414 | } |
| 1415 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1416 | /** |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 1417 | * remove_vm_area - find and remove a continuous kernel virtual area |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | * @addr: base address |
| 1419 | * |
| 1420 | * Search for the kernel VM area starting at @addr, and remove it. |
| 1421 | * This function returns the found VM area, but using it is NOT safe |
Andi Kleen | 7856dfe | 2005-05-20 14:27:57 -0700 | [diff] [blame] | 1422 | * on SMP machines, except for its size or flags. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1423 | */ |
Christoph Lameter | b3bdda0 | 2008-02-04 22:28:32 -0800 | [diff] [blame] | 1424 | struct vm_struct *remove_vm_area(const void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1425 | { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1426 | struct vmap_area *va; |
| 1427 | |
| 1428 | va = find_vmap_area((unsigned long)addr); |
| 1429 | if (va && va->flags & VM_VM_AREA) { |
| 1430 | struct vm_struct *vm = va->private; |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1431 | |
| 1432 | if (!(vm->flags & VM_UNLIST)) { |
| 1433 | struct vm_struct *tmp, **p; |
| 1434 | /* |
| 1435 | * remove from list and disallow access to |
| 1436 | * this vm_struct before unmap. (address range |
| 1437 | * confliction is maintained by vmap.) |
| 1438 | */ |
| 1439 | write_lock(&vmlist_lock); |
| 1440 | for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next) |
| 1441 | ; |
| 1442 | *p = tmp->next; |
| 1443 | write_unlock(&vmlist_lock); |
| 1444 | } |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1445 | |
KAMEZAWA Hiroyuki | dd32c27 | 2009-09-21 17:02:32 -0700 | [diff] [blame] | 1446 | vmap_debug_free_range(va->va_start, va->va_end); |
| 1447 | free_unmap_vmap_area(va); |
| 1448 | vm->size -= PAGE_SIZE; |
| 1449 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1450 | return vm; |
| 1451 | } |
| 1452 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | } |
| 1454 | |
Christoph Lameter | b3bdda0 | 2008-02-04 22:28:32 -0800 | [diff] [blame] | 1455 | static void __vunmap(const void *addr, int deallocate_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1456 | { |
| 1457 | struct vm_struct *area; |
| 1458 | |
| 1459 | if (!addr) |
| 1460 | return; |
| 1461 | |
| 1462 | if ((PAGE_SIZE-1) & (unsigned long)addr) { |
Arjan van de Ven | 4c8573e | 2008-07-25 19:45:37 -0700 | [diff] [blame] | 1463 | WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1464 | return; |
| 1465 | } |
| 1466 | |
| 1467 | area = remove_vm_area(addr); |
| 1468 | if (unlikely(!area)) { |
Arjan van de Ven | 4c8573e | 2008-07-25 19:45:37 -0700 | [diff] [blame] | 1469 | WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | return; |
| 1472 | } |
| 1473 | |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 1474 | debug_check_no_locks_freed(addr, area->size); |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 1475 | debug_check_no_obj_freed(addr, area->size); |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 1476 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | if (deallocate_pages) { |
| 1478 | int i; |
| 1479 | |
| 1480 | for (i = 0; i < area->nr_pages; i++) { |
Christoph Lameter | bf53d6f | 2008-02-04 22:28:34 -0800 | [diff] [blame] | 1481 | struct page *page = area->pages[i]; |
| 1482 | |
| 1483 | BUG_ON(!page); |
| 1484 | __free_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | } |
| 1486 | |
Jan Kiszka | 8757d5f | 2006-07-14 00:23:56 -0700 | [diff] [blame] | 1487 | if (area->flags & VM_VPAGES) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | vfree(area->pages); |
| 1489 | else |
| 1490 | kfree(area->pages); |
| 1491 | } |
| 1492 | |
| 1493 | kfree(area); |
| 1494 | return; |
| 1495 | } |
| 1496 | |
| 1497 | /** |
| 1498 | * vfree - release memory allocated by vmalloc() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1499 | * @addr: memory base address |
| 1500 | * |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 1501 | * Free the virtually continuous memory area starting at @addr, as |
Pekka Enberg | 80e93ef | 2005-09-09 13:10:16 -0700 | [diff] [blame] | 1502 | * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is |
| 1503 | * NULL, no operation is performed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | * |
Pekka Enberg | 80e93ef | 2005-09-09 13:10:16 -0700 | [diff] [blame] | 1505 | * Must not be called in interrupt context. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | */ |
Christoph Lameter | b3bdda0 | 2008-02-04 22:28:32 -0800 | [diff] [blame] | 1507 | void vfree(const void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | { |
| 1509 | BUG_ON(in_interrupt()); |
Catalin Marinas | 89219d3 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 1510 | |
| 1511 | kmemleak_free(addr); |
| 1512 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1513 | __vunmap(addr, 1); |
| 1514 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | EXPORT_SYMBOL(vfree); |
| 1516 | |
| 1517 | /** |
| 1518 | * vunmap - release virtual mapping obtained by vmap() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | * @addr: memory base address |
| 1520 | * |
| 1521 | * Free the virtually contiguous memory area starting at @addr, |
| 1522 | * which was created from the page array passed to vmap(). |
| 1523 | * |
Pekka Enberg | 80e93ef | 2005-09-09 13:10:16 -0700 | [diff] [blame] | 1524 | * Must not be called in interrupt context. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | */ |
Christoph Lameter | b3bdda0 | 2008-02-04 22:28:32 -0800 | [diff] [blame] | 1526 | void vunmap(const void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | { |
| 1528 | BUG_ON(in_interrupt()); |
Peter Zijlstra | 34754b6 | 2009-02-25 16:04:03 +0100 | [diff] [blame] | 1529 | might_sleep(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1530 | __vunmap(addr, 0); |
| 1531 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1532 | EXPORT_SYMBOL(vunmap); |
| 1533 | |
| 1534 | /** |
| 1535 | * vmap - map an array of pages into virtually contiguous space |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | * @pages: array of page pointers |
| 1537 | * @count: number of pages to map |
| 1538 | * @flags: vm_area->flags |
| 1539 | * @prot: page protection for the mapping |
| 1540 | * |
| 1541 | * Maps @count pages from @pages into contiguous kernel virtual |
| 1542 | * space. |
| 1543 | */ |
| 1544 | void *vmap(struct page **pages, unsigned int count, |
| 1545 | unsigned long flags, pgprot_t prot) |
| 1546 | { |
| 1547 | struct vm_struct *area; |
| 1548 | |
Peter Zijlstra | 34754b6 | 2009-02-25 16:04:03 +0100 | [diff] [blame] | 1549 | might_sleep(); |
| 1550 | |
Jan Beulich | 4481374 | 2009-09-21 17:03:05 -0700 | [diff] [blame] | 1551 | if (count > totalram_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1552 | return NULL; |
| 1553 | |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1554 | area = get_vm_area_caller((count << PAGE_SHIFT), flags, |
| 1555 | __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1556 | if (!area) |
| 1557 | return NULL; |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1558 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1559 | if (map_vm_area(area, prot, &pages)) { |
| 1560 | vunmap(area->addr); |
| 1561 | return NULL; |
| 1562 | } |
| 1563 | |
| 1564 | return area->addr; |
| 1565 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | EXPORT_SYMBOL(vmap); |
| 1567 | |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1568 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
| 1569 | gfp_t gfp_mask, pgprot_t prot, |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1570 | int node, void *caller); |
Adrian Bunk | e31d9eb | 2008-02-04 22:29:09 -0800 | [diff] [blame] | 1571 | static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1572 | pgprot_t prot, int node, void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1573 | { |
Dave Hansen | 22943ab | 2011-05-24 17:12:18 -0700 | [diff] [blame] | 1574 | const int order = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1575 | struct page **pages; |
| 1576 | unsigned int nr_pages, array_size, i; |
Jan Beulich | 976d6df | 2009-12-14 17:58:39 -0800 | [diff] [blame] | 1577 | gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1578 | |
| 1579 | nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT; |
| 1580 | array_size = (nr_pages * sizeof(struct page *)); |
| 1581 | |
| 1582 | area->nr_pages = nr_pages; |
| 1583 | /* Please note that the recursion is strictly bounded. */ |
Jan Kiszka | 8757d5f | 2006-07-14 00:23:56 -0700 | [diff] [blame] | 1584 | if (array_size > PAGE_SIZE) { |
Jan Beulich | 976d6df | 2009-12-14 17:58:39 -0800 | [diff] [blame] | 1585 | pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1586 | PAGE_KERNEL, node, caller); |
Jan Kiszka | 8757d5f | 2006-07-14 00:23:56 -0700 | [diff] [blame] | 1587 | area->flags |= VM_VPAGES; |
Andrew Morton | 286e1ea | 2006-10-17 00:09:57 -0700 | [diff] [blame] | 1588 | } else { |
Jan Beulich | 976d6df | 2009-12-14 17:58:39 -0800 | [diff] [blame] | 1589 | pages = kmalloc_node(array_size, nested_gfp, node); |
Andrew Morton | 286e1ea | 2006-10-17 00:09:57 -0700 | [diff] [blame] | 1590 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1591 | area->pages = pages; |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1592 | area->caller = caller; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1593 | if (!area->pages) { |
| 1594 | remove_vm_area(area->addr); |
| 1595 | kfree(area); |
| 1596 | return NULL; |
| 1597 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | |
| 1599 | for (i = 0; i < area->nr_pages; i++) { |
Christoph Lameter | bf53d6f | 2008-02-04 22:28:34 -0800 | [diff] [blame] | 1600 | struct page *page; |
Dave Hansen | 22943ab | 2011-05-24 17:12:18 -0700 | [diff] [blame] | 1601 | gfp_t tmp_mask = gfp_mask | __GFP_NOWARN; |
Christoph Lameter | bf53d6f | 2008-02-04 22:28:34 -0800 | [diff] [blame] | 1602 | |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1603 | if (node < 0) |
Dave Hansen | 22943ab | 2011-05-24 17:12:18 -0700 | [diff] [blame] | 1604 | page = alloc_page(tmp_mask); |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1605 | else |
Dave Hansen | 22943ab | 2011-05-24 17:12:18 -0700 | [diff] [blame] | 1606 | page = alloc_pages_node(node, tmp_mask, order); |
Christoph Lameter | bf53d6f | 2008-02-04 22:28:34 -0800 | [diff] [blame] | 1607 | |
| 1608 | if (unlikely(!page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1609 | /* Successfully allocated i pages, free them in __vunmap() */ |
| 1610 | area->nr_pages = i; |
| 1611 | goto fail; |
| 1612 | } |
Christoph Lameter | bf53d6f | 2008-02-04 22:28:34 -0800 | [diff] [blame] | 1613 | area->pages[i] = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | } |
| 1615 | |
| 1616 | if (map_vm_area(area, prot, &pages)) |
| 1617 | goto fail; |
| 1618 | return area->addr; |
| 1619 | |
| 1620 | fail: |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 1621 | warn_alloc_failed(gfp_mask, order, |
| 1622 | "vmalloc: allocation failure, allocated %ld of %ld bytes\n", |
Dave Hansen | 22943ab | 2011-05-24 17:12:18 -0700 | [diff] [blame] | 1623 | (area->nr_pages*PAGE_SIZE), area->size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1624 | vfree(area->addr); |
| 1625 | return NULL; |
| 1626 | } |
| 1627 | |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1628 | /** |
| 1629 | * __vmalloc_node_range - allocate virtually contiguous memory |
| 1630 | * @size: allocation size |
| 1631 | * @align: desired alignment |
| 1632 | * @start: vm area range start |
| 1633 | * @end: vm area range end |
| 1634 | * @gfp_mask: flags for the page level allocator |
| 1635 | * @prot: protection mask for the allocated pages |
| 1636 | * @node: node to use for allocation or -1 |
| 1637 | * @caller: caller's return address |
| 1638 | * |
| 1639 | * Allocate enough pages to cover @size from the page level |
| 1640 | * allocator with @gfp_mask flags. Map them into contiguous |
| 1641 | * kernel virtual space, using a pagetable protection of @prot. |
| 1642 | */ |
| 1643 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
| 1644 | unsigned long start, unsigned long end, gfp_t gfp_mask, |
| 1645 | pgprot_t prot, int node, void *caller) |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1646 | { |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1647 | struct vm_struct *area; |
| 1648 | void *addr; |
| 1649 | unsigned long real_size = size; |
| 1650 | |
| 1651 | size = PAGE_ALIGN(size); |
| 1652 | if (!size || (size >> PAGE_SHIFT) > totalram_pages) |
Joe Perches | de7d2b5 | 2011-10-31 17:08:48 -0700 | [diff] [blame] | 1653 | goto fail; |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1654 | |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1655 | area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST, |
| 1656 | start, end, node, gfp_mask, caller); |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1657 | if (!area) |
Joe Perches | de7d2b5 | 2011-10-31 17:08:48 -0700 | [diff] [blame] | 1658 | goto fail; |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1659 | |
| 1660 | addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller); |
Catalin Marinas | 89219d3 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 1661 | |
| 1662 | /* |
Mitsuo Hayasaka | f5252e0 | 2011-10-31 17:08:13 -0700 | [diff] [blame] | 1663 | * In this function, newly allocated vm_struct is not added |
| 1664 | * to vmlist at __get_vm_area_node(). so, it is added here. |
| 1665 | */ |
| 1666 | insert_vmalloc_vmlist(area); |
| 1667 | |
| 1668 | /* |
Catalin Marinas | 89219d3 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 1669 | * A ref_count = 3 is needed because the vm_struct and vmap_area |
| 1670 | * structures allocated in the __get_vm_area_node() function contain |
| 1671 | * references to the virtual address of the vmalloc'ed block. |
| 1672 | */ |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1673 | kmemleak_alloc(addr, real_size, 3, gfp_mask); |
Catalin Marinas | 89219d3 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 1674 | |
| 1675 | return addr; |
Joe Perches | de7d2b5 | 2011-10-31 17:08:48 -0700 | [diff] [blame] | 1676 | |
| 1677 | fail: |
| 1678 | warn_alloc_failed(gfp_mask, 0, |
| 1679 | "vmalloc: allocation failure: %lu bytes\n", |
| 1680 | real_size); |
| 1681 | return NULL; |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1682 | } |
| 1683 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | /** |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1685 | * __vmalloc_node - allocate virtually contiguous memory |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1686 | * @size: allocation size |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1687 | * @align: desired alignment |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1688 | * @gfp_mask: flags for the page level allocator |
| 1689 | * @prot: protection mask for the allocated pages |
Randy Dunlap | d44e078 | 2005-11-07 01:01:10 -0800 | [diff] [blame] | 1690 | * @node: node to use for allocation or -1 |
Randy Dunlap | c85d194 | 2008-05-01 04:34:48 -0700 | [diff] [blame] | 1691 | * @caller: caller's return address |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | * |
| 1693 | * Allocate enough pages to cover @size from the page level |
| 1694 | * allocator with @gfp_mask flags. Map them into contiguous |
| 1695 | * kernel virtual space, using a pagetable protection of @prot. |
| 1696 | */ |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1697 | static void *__vmalloc_node(unsigned long size, unsigned long align, |
| 1698 | gfp_t gfp_mask, pgprot_t prot, |
| 1699 | int node, void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1700 | { |
David Rientjes | d0a2126 | 2011-01-13 15:46:02 -0800 | [diff] [blame] | 1701 | return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END, |
| 1702 | gfp_mask, prot, node, caller); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1703 | } |
| 1704 | |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1705 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
| 1706 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1707 | return __vmalloc_node(size, 1, gfp_mask, prot, -1, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1708 | __builtin_return_address(0)); |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1709 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1710 | EXPORT_SYMBOL(__vmalloc); |
| 1711 | |
Dave Young | e1ca778 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1712 | static inline void *__vmalloc_node_flags(unsigned long size, |
| 1713 | int node, gfp_t flags) |
| 1714 | { |
| 1715 | return __vmalloc_node(size, 1, flags, PAGE_KERNEL, |
| 1716 | node, __builtin_return_address(0)); |
| 1717 | } |
| 1718 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1719 | /** |
| 1720 | * vmalloc - allocate virtually contiguous memory |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1721 | * @size: allocation size |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | * Allocate enough pages to cover @size from the page level |
| 1723 | * allocator and map them into contiguous kernel virtual space. |
| 1724 | * |
Michael Opdenacker | c1c8897 | 2006-10-03 23:21:02 +0200 | [diff] [blame] | 1725 | * For tight control over page level allocator and protection flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1726 | * use __vmalloc() instead. |
| 1727 | */ |
| 1728 | void *vmalloc(unsigned long size) |
| 1729 | { |
Dave Young | e1ca778 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1730 | return __vmalloc_node_flags(size, -1, GFP_KERNEL | __GFP_HIGHMEM); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1731 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1732 | EXPORT_SYMBOL(vmalloc); |
| 1733 | |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1734 | /** |
Dave Young | e1ca778 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1735 | * vzalloc - allocate virtually contiguous memory with zero fill |
| 1736 | * @size: allocation size |
| 1737 | * Allocate enough pages to cover @size from the page level |
| 1738 | * allocator and map them into contiguous kernel virtual space. |
| 1739 | * The memory allocated is set to zero. |
| 1740 | * |
| 1741 | * For tight control over page level allocator and protection flags |
| 1742 | * use __vmalloc() instead. |
| 1743 | */ |
| 1744 | void *vzalloc(unsigned long size) |
| 1745 | { |
| 1746 | return __vmalloc_node_flags(size, -1, |
| 1747 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
| 1748 | } |
| 1749 | EXPORT_SYMBOL(vzalloc); |
| 1750 | |
| 1751 | /** |
Rolf Eike Beer | ead0408 | 2006-09-27 01:50:13 -0700 | [diff] [blame] | 1752 | * vmalloc_user - allocate zeroed virtually contiguous memory for userspace |
| 1753 | * @size: allocation size |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1754 | * |
Rolf Eike Beer | ead0408 | 2006-09-27 01:50:13 -0700 | [diff] [blame] | 1755 | * The resulting memory area is zeroed so it can be mapped to userspace |
| 1756 | * without leaking data. |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1757 | */ |
| 1758 | void *vmalloc_user(unsigned long size) |
| 1759 | { |
| 1760 | struct vm_struct *area; |
| 1761 | void *ret; |
| 1762 | |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1763 | ret = __vmalloc_node(size, SHMLBA, |
| 1764 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
Glauber Costa | 8487784 | 2009-01-06 14:39:19 -0800 | [diff] [blame] | 1765 | PAGE_KERNEL, -1, __builtin_return_address(0)); |
Eric Dumazet | 2b4ac44 | 2006-11-10 12:27:48 -0800 | [diff] [blame] | 1766 | if (ret) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1767 | area = find_vm_area(ret); |
Eric Dumazet | 2b4ac44 | 2006-11-10 12:27:48 -0800 | [diff] [blame] | 1768 | area->flags |= VM_USERMAP; |
Eric Dumazet | 2b4ac44 | 2006-11-10 12:27:48 -0800 | [diff] [blame] | 1769 | } |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1770 | return ret; |
| 1771 | } |
| 1772 | EXPORT_SYMBOL(vmalloc_user); |
| 1773 | |
| 1774 | /** |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1775 | * vmalloc_node - allocate memory on a specific node |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1776 | * @size: allocation size |
Randy Dunlap | d44e078 | 2005-11-07 01:01:10 -0800 | [diff] [blame] | 1777 | * @node: numa node |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1778 | * |
| 1779 | * Allocate enough pages to cover @size from the page level |
| 1780 | * allocator and map them into contiguous kernel virtual space. |
| 1781 | * |
Michael Opdenacker | c1c8897 | 2006-10-03 23:21:02 +0200 | [diff] [blame] | 1782 | * For tight control over page level allocator and protection flags |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1783 | * use __vmalloc() instead. |
| 1784 | */ |
| 1785 | void *vmalloc_node(unsigned long size, int node) |
| 1786 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1787 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 1788 | node, __builtin_return_address(0)); |
Christoph Lameter | 930fc45 | 2005-10-29 18:15:41 -0700 | [diff] [blame] | 1789 | } |
| 1790 | EXPORT_SYMBOL(vmalloc_node); |
| 1791 | |
Dave Young | e1ca778 | 2010-10-26 14:22:06 -0700 | [diff] [blame] | 1792 | /** |
| 1793 | * vzalloc_node - allocate memory on a specific node with zero fill |
| 1794 | * @size: allocation size |
| 1795 | * @node: numa node |
| 1796 | * |
| 1797 | * Allocate enough pages to cover @size from the page level |
| 1798 | * allocator and map them into contiguous kernel virtual space. |
| 1799 | * The memory allocated is set to zero. |
| 1800 | * |
| 1801 | * For tight control over page level allocator and protection flags |
| 1802 | * use __vmalloc_node() instead. |
| 1803 | */ |
| 1804 | void *vzalloc_node(unsigned long size, int node) |
| 1805 | { |
| 1806 | return __vmalloc_node_flags(size, node, |
| 1807 | GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); |
| 1808 | } |
| 1809 | EXPORT_SYMBOL(vzalloc_node); |
| 1810 | |
Pavel Pisa | 4dc3b16 | 2005-05-01 08:59:25 -0700 | [diff] [blame] | 1811 | #ifndef PAGE_KERNEL_EXEC |
| 1812 | # define PAGE_KERNEL_EXEC PAGE_KERNEL |
| 1813 | #endif |
| 1814 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1815 | /** |
| 1816 | * vmalloc_exec - allocate virtually contiguous, executable memory |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | * @size: allocation size |
| 1818 | * |
| 1819 | * Kernel-internal function to allocate enough pages to cover @size |
| 1820 | * the page level allocator and map them into contiguous and |
| 1821 | * executable kernel virtual space. |
| 1822 | * |
Michael Opdenacker | c1c8897 | 2006-10-03 23:21:02 +0200 | [diff] [blame] | 1823 | * For tight control over page level allocator and protection flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1824 | * use __vmalloc() instead. |
| 1825 | */ |
| 1826 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1827 | void *vmalloc_exec(unsigned long size) |
| 1828 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1829 | return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, |
Glauber Costa | 8487784 | 2009-01-06 14:39:19 -0800 | [diff] [blame] | 1830 | -1, __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1831 | } |
| 1832 | |
Andi Kleen | 0d08e0d | 2007-05-02 19:27:12 +0200 | [diff] [blame] | 1833 | #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32) |
Benjamin Herrenschmidt | 7ac674f | 2007-07-19 01:49:10 -0700 | [diff] [blame] | 1834 | #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL |
Andi Kleen | 0d08e0d | 2007-05-02 19:27:12 +0200 | [diff] [blame] | 1835 | #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA) |
Benjamin Herrenschmidt | 7ac674f | 2007-07-19 01:49:10 -0700 | [diff] [blame] | 1836 | #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL |
Andi Kleen | 0d08e0d | 2007-05-02 19:27:12 +0200 | [diff] [blame] | 1837 | #else |
| 1838 | #define GFP_VMALLOC32 GFP_KERNEL |
| 1839 | #endif |
| 1840 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1841 | /** |
| 1842 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1843 | * @size: allocation size |
| 1844 | * |
| 1845 | * Allocate enough 32bit PA addressable pages to cover @size from the |
| 1846 | * page level allocator and map them into contiguous kernel virtual space. |
| 1847 | */ |
| 1848 | void *vmalloc_32(unsigned long size) |
| 1849 | { |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1850 | return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL, |
Glauber Costa | 8487784 | 2009-01-06 14:39:19 -0800 | [diff] [blame] | 1851 | -1, __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1852 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1853 | EXPORT_SYMBOL(vmalloc_32); |
| 1854 | |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1855 | /** |
Rolf Eike Beer | ead0408 | 2006-09-27 01:50:13 -0700 | [diff] [blame] | 1856 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1857 | * @size: allocation size |
Rolf Eike Beer | ead0408 | 2006-09-27 01:50:13 -0700 | [diff] [blame] | 1858 | * |
| 1859 | * The resulting memory area is 32bit addressable and zeroed so it can be |
| 1860 | * mapped to userspace without leaking data. |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1861 | */ |
| 1862 | void *vmalloc_32_user(unsigned long size) |
| 1863 | { |
| 1864 | struct vm_struct *area; |
| 1865 | void *ret; |
| 1866 | |
David Miller | 2dca699 | 2009-09-21 12:22:34 -0700 | [diff] [blame] | 1867 | ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL, |
Glauber Costa | 8487784 | 2009-01-06 14:39:19 -0800 | [diff] [blame] | 1868 | -1, __builtin_return_address(0)); |
Eric Dumazet | 2b4ac44 | 2006-11-10 12:27:48 -0800 | [diff] [blame] | 1869 | if (ret) { |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 1870 | area = find_vm_area(ret); |
Eric Dumazet | 2b4ac44 | 2006-11-10 12:27:48 -0800 | [diff] [blame] | 1871 | area->flags |= VM_USERMAP; |
Eric Dumazet | 2b4ac44 | 2006-11-10 12:27:48 -0800 | [diff] [blame] | 1872 | } |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 1873 | return ret; |
| 1874 | } |
| 1875 | EXPORT_SYMBOL(vmalloc_32_user); |
| 1876 | |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 1877 | /* |
| 1878 | * small helper routine , copy contents to buf from addr. |
| 1879 | * If the page is not present, fill zero. |
| 1880 | */ |
| 1881 | |
| 1882 | static int aligned_vread(char *buf, char *addr, unsigned long count) |
| 1883 | { |
| 1884 | struct page *p; |
| 1885 | int copied = 0; |
| 1886 | |
| 1887 | while (count) { |
| 1888 | unsigned long offset, length; |
| 1889 | |
| 1890 | offset = (unsigned long)addr & ~PAGE_MASK; |
| 1891 | length = PAGE_SIZE - offset; |
| 1892 | if (length > count) |
| 1893 | length = count; |
| 1894 | p = vmalloc_to_page(addr); |
| 1895 | /* |
| 1896 | * To do safe access to this _mapped_ area, we need |
| 1897 | * lock. But adding lock here means that we need to add |
| 1898 | * overhead of vmalloc()/vfree() calles for this _debug_ |
| 1899 | * interface, rarely used. Instead of that, we'll use |
| 1900 | * kmap() and get small overhead in this access function. |
| 1901 | */ |
| 1902 | if (p) { |
| 1903 | /* |
| 1904 | * we can expect USER0 is not used (see vread/vwrite's |
| 1905 | * function description) |
| 1906 | */ |
| 1907 | void *map = kmap_atomic(p, KM_USER0); |
| 1908 | memcpy(buf, map + offset, length); |
| 1909 | kunmap_atomic(map, KM_USER0); |
| 1910 | } else |
| 1911 | memset(buf, 0, length); |
| 1912 | |
| 1913 | addr += length; |
| 1914 | buf += length; |
| 1915 | copied += length; |
| 1916 | count -= length; |
| 1917 | } |
| 1918 | return copied; |
| 1919 | } |
| 1920 | |
| 1921 | static int aligned_vwrite(char *buf, char *addr, unsigned long count) |
| 1922 | { |
| 1923 | struct page *p; |
| 1924 | int copied = 0; |
| 1925 | |
| 1926 | while (count) { |
| 1927 | unsigned long offset, length; |
| 1928 | |
| 1929 | offset = (unsigned long)addr & ~PAGE_MASK; |
| 1930 | length = PAGE_SIZE - offset; |
| 1931 | if (length > count) |
| 1932 | length = count; |
| 1933 | p = vmalloc_to_page(addr); |
| 1934 | /* |
| 1935 | * To do safe access to this _mapped_ area, we need |
| 1936 | * lock. But adding lock here means that we need to add |
| 1937 | * overhead of vmalloc()/vfree() calles for this _debug_ |
| 1938 | * interface, rarely used. Instead of that, we'll use |
| 1939 | * kmap() and get small overhead in this access function. |
| 1940 | */ |
| 1941 | if (p) { |
| 1942 | /* |
| 1943 | * we can expect USER0 is not used (see vread/vwrite's |
| 1944 | * function description) |
| 1945 | */ |
| 1946 | void *map = kmap_atomic(p, KM_USER0); |
| 1947 | memcpy(map + offset, buf, length); |
| 1948 | kunmap_atomic(map, KM_USER0); |
| 1949 | } |
| 1950 | addr += length; |
| 1951 | buf += length; |
| 1952 | copied += length; |
| 1953 | count -= length; |
| 1954 | } |
| 1955 | return copied; |
| 1956 | } |
| 1957 | |
| 1958 | /** |
| 1959 | * vread() - read vmalloc area in a safe way. |
| 1960 | * @buf: buffer for reading data |
| 1961 | * @addr: vm address. |
| 1962 | * @count: number of bytes to be read. |
| 1963 | * |
| 1964 | * Returns # of bytes which addr and buf should be increased. |
| 1965 | * (same number to @count). Returns 0 if [addr...addr+count) doesn't |
| 1966 | * includes any intersect with alive vmalloc area. |
| 1967 | * |
| 1968 | * This function checks that addr is a valid vmalloc'ed area, and |
| 1969 | * copy data from that area to a given buffer. If the given memory range |
| 1970 | * of [addr...addr+count) includes some valid address, data is copied to |
| 1971 | * proper area of @buf. If there are memory holes, they'll be zero-filled. |
| 1972 | * IOREMAP area is treated as memory hole and no copy is done. |
| 1973 | * |
| 1974 | * If [addr...addr+count) doesn't includes any intersects with alive |
| 1975 | * vm_struct area, returns 0. |
| 1976 | * @buf should be kernel's buffer. Because this function uses KM_USER0, |
| 1977 | * the caller should guarantee KM_USER0 is not used. |
| 1978 | * |
| 1979 | * Note: In usual ops, vread() is never necessary because the caller |
| 1980 | * should know vmalloc() area is valid and can use memcpy(). |
| 1981 | * This is for routines which have to access vmalloc area without |
| 1982 | * any informaion, as /dev/kmem. |
| 1983 | * |
| 1984 | */ |
| 1985 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1986 | long vread(char *buf, char *addr, unsigned long count) |
| 1987 | { |
| 1988 | struct vm_struct *tmp; |
| 1989 | char *vaddr, *buf_start = buf; |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 1990 | unsigned long buflen = count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1991 | unsigned long n; |
| 1992 | |
| 1993 | /* Don't allow overflow */ |
| 1994 | if ((unsigned long) addr + count < count) |
| 1995 | count = -(unsigned long) addr; |
| 1996 | |
| 1997 | read_lock(&vmlist_lock); |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 1998 | for (tmp = vmlist; count && tmp; tmp = tmp->next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1999 | vaddr = (char *) tmp->addr; |
| 2000 | if (addr >= vaddr + tmp->size - PAGE_SIZE) |
| 2001 | continue; |
| 2002 | while (addr < vaddr) { |
| 2003 | if (count == 0) |
| 2004 | goto finished; |
| 2005 | *buf = '\0'; |
| 2006 | buf++; |
| 2007 | addr++; |
| 2008 | count--; |
| 2009 | } |
| 2010 | n = vaddr + tmp->size - PAGE_SIZE - addr; |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2011 | if (n > count) |
| 2012 | n = count; |
| 2013 | if (!(tmp->flags & VM_IOREMAP)) |
| 2014 | aligned_vread(buf, addr, n); |
| 2015 | else /* IOREMAP area is treated as memory hole */ |
| 2016 | memset(buf, 0, n); |
| 2017 | buf += n; |
| 2018 | addr += n; |
| 2019 | count -= n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2020 | } |
| 2021 | finished: |
| 2022 | read_unlock(&vmlist_lock); |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2023 | |
| 2024 | if (buf == buf_start) |
| 2025 | return 0; |
| 2026 | /* zero-fill memory holes */ |
| 2027 | if (buf != buf_start + buflen) |
| 2028 | memset(buf, 0, buflen - (buf - buf_start)); |
| 2029 | |
| 2030 | return buflen; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2031 | } |
| 2032 | |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2033 | /** |
| 2034 | * vwrite() - write vmalloc area in a safe way. |
| 2035 | * @buf: buffer for source data |
| 2036 | * @addr: vm address. |
| 2037 | * @count: number of bytes to be read. |
| 2038 | * |
| 2039 | * Returns # of bytes which addr and buf should be incresed. |
| 2040 | * (same number to @count). |
| 2041 | * If [addr...addr+count) doesn't includes any intersect with valid |
| 2042 | * vmalloc area, returns 0. |
| 2043 | * |
| 2044 | * This function checks that addr is a valid vmalloc'ed area, and |
| 2045 | * copy data from a buffer to the given addr. If specified range of |
| 2046 | * [addr...addr+count) includes some valid address, data is copied from |
| 2047 | * proper area of @buf. If there are memory holes, no copy to hole. |
| 2048 | * IOREMAP area is treated as memory hole and no copy is done. |
| 2049 | * |
| 2050 | * If [addr...addr+count) doesn't includes any intersects with alive |
| 2051 | * vm_struct area, returns 0. |
| 2052 | * @buf should be kernel's buffer. Because this function uses KM_USER0, |
| 2053 | * the caller should guarantee KM_USER0 is not used. |
| 2054 | * |
| 2055 | * Note: In usual ops, vwrite() is never necessary because the caller |
| 2056 | * should know vmalloc() area is valid and can use memcpy(). |
| 2057 | * This is for routines which have to access vmalloc area without |
| 2058 | * any informaion, as /dev/kmem. |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2059 | */ |
| 2060 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | long vwrite(char *buf, char *addr, unsigned long count) |
| 2062 | { |
| 2063 | struct vm_struct *tmp; |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2064 | char *vaddr; |
| 2065 | unsigned long n, buflen; |
| 2066 | int copied = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2067 | |
| 2068 | /* Don't allow overflow */ |
| 2069 | if ((unsigned long) addr + count < count) |
| 2070 | count = -(unsigned long) addr; |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2071 | buflen = count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2072 | |
| 2073 | read_lock(&vmlist_lock); |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2074 | for (tmp = vmlist; count && tmp; tmp = tmp->next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2075 | vaddr = (char *) tmp->addr; |
| 2076 | if (addr >= vaddr + tmp->size - PAGE_SIZE) |
| 2077 | continue; |
| 2078 | while (addr < vaddr) { |
| 2079 | if (count == 0) |
| 2080 | goto finished; |
| 2081 | buf++; |
| 2082 | addr++; |
| 2083 | count--; |
| 2084 | } |
| 2085 | n = vaddr + tmp->size - PAGE_SIZE - addr; |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2086 | if (n > count) |
| 2087 | n = count; |
| 2088 | if (!(tmp->flags & VM_IOREMAP)) { |
| 2089 | aligned_vwrite(buf, addr, n); |
| 2090 | copied++; |
| 2091 | } |
| 2092 | buf += n; |
| 2093 | addr += n; |
| 2094 | count -= n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | } |
| 2096 | finished: |
| 2097 | read_unlock(&vmlist_lock); |
KAMEZAWA Hiroyuki | d0107eb | 2009-09-21 17:02:34 -0700 | [diff] [blame] | 2098 | if (!copied) |
| 2099 | return 0; |
| 2100 | return buflen; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2101 | } |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2102 | |
| 2103 | /** |
| 2104 | * remap_vmalloc_range - map vmalloc pages to userspace |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2105 | * @vma: vma to cover (map full range of vma) |
| 2106 | * @addr: vmalloc memory |
| 2107 | * @pgoff: number of pages into addr before first page to map |
Randy Dunlap | 7682486 | 2008-03-19 17:00:40 -0700 | [diff] [blame] | 2108 | * |
| 2109 | * Returns: 0 for success, -Exxx on failure |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2110 | * |
| 2111 | * This function checks that addr is a valid vmalloc'ed area, and |
| 2112 | * that it is big enough to cover the vma. Will return failure if |
| 2113 | * that criteria isn't met. |
| 2114 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 2115 | * Similar to remap_pfn_range() (see mm/memory.c) |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2116 | */ |
| 2117 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
| 2118 | unsigned long pgoff) |
| 2119 | { |
| 2120 | struct vm_struct *area; |
| 2121 | unsigned long uaddr = vma->vm_start; |
| 2122 | unsigned long usize = vma->vm_end - vma->vm_start; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2123 | |
| 2124 | if ((PAGE_SIZE-1) & (unsigned long)addr) |
| 2125 | return -EINVAL; |
| 2126 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 2127 | area = find_vm_area(addr); |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2128 | if (!area) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 2129 | return -EINVAL; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2130 | |
| 2131 | if (!(area->flags & VM_USERMAP)) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 2132 | return -EINVAL; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2133 | |
| 2134 | if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 2135 | return -EINVAL; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2136 | |
| 2137 | addr += pgoff << PAGE_SHIFT; |
| 2138 | do { |
| 2139 | struct page *page = vmalloc_to_page(addr); |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 2140 | int ret; |
| 2141 | |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2142 | ret = vm_insert_page(vma, uaddr, page); |
| 2143 | if (ret) |
| 2144 | return ret; |
| 2145 | |
| 2146 | uaddr += PAGE_SIZE; |
| 2147 | addr += PAGE_SIZE; |
| 2148 | usize -= PAGE_SIZE; |
| 2149 | } while (usize > 0); |
| 2150 | |
| 2151 | /* Prevent "things" like memory migration? VM_flags need a cleanup... */ |
| 2152 | vma->vm_flags |= VM_RESERVED; |
| 2153 | |
Nick Piggin | db64fe0 | 2008-10-18 20:27:03 -0700 | [diff] [blame] | 2154 | return 0; |
Nick Piggin | 8334231 | 2006-06-23 02:03:20 -0700 | [diff] [blame] | 2155 | } |
| 2156 | EXPORT_SYMBOL(remap_vmalloc_range); |
| 2157 | |
Christoph Hellwig | 1eeb66a | 2007-05-08 00:27:03 -0700 | [diff] [blame] | 2158 | /* |
| 2159 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to |
| 2160 | * have one. |
| 2161 | */ |
| 2162 | void __attribute__((weak)) vmalloc_sync_all(void) |
| 2163 | { |
| 2164 | } |
Jeremy Fitzhardinge | 5f4352f | 2007-07-17 18:37:04 -0700 | [diff] [blame] | 2165 | |
| 2166 | |
Martin Schwidefsky | 2f569af | 2008-02-08 04:22:04 -0800 | [diff] [blame] | 2167 | static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data) |
Jeremy Fitzhardinge | 5f4352f | 2007-07-17 18:37:04 -0700 | [diff] [blame] | 2168 | { |
| 2169 | /* apply_to_page_range() does all the hard work. */ |
| 2170 | return 0; |
| 2171 | } |
| 2172 | |
| 2173 | /** |
| 2174 | * alloc_vm_area - allocate a range of kernel address space |
| 2175 | * @size: size of the area |
Randy Dunlap | 7682486 | 2008-03-19 17:00:40 -0700 | [diff] [blame] | 2176 | * |
| 2177 | * Returns: NULL on failure, vm_struct on success |
Jeremy Fitzhardinge | 5f4352f | 2007-07-17 18:37:04 -0700 | [diff] [blame] | 2178 | * |
| 2179 | * This function reserves a range of kernel address space, and |
| 2180 | * allocates pagetables to map that range. No actual mappings |
| 2181 | * are created. If the kernel address space is not shared |
| 2182 | * between processes, it syncs the pagetable across all |
| 2183 | * processes. |
| 2184 | */ |
| 2185 | struct vm_struct *alloc_vm_area(size_t size) |
| 2186 | { |
| 2187 | struct vm_struct *area; |
| 2188 | |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 2189 | area = get_vm_area_caller(size, VM_IOREMAP, |
| 2190 | __builtin_return_address(0)); |
Jeremy Fitzhardinge | 5f4352f | 2007-07-17 18:37:04 -0700 | [diff] [blame] | 2191 | if (area == NULL) |
| 2192 | return NULL; |
| 2193 | |
| 2194 | /* |
| 2195 | * This ensures that page tables are constructed for this region |
| 2196 | * of kernel virtual address space and mapped into init_mm. |
| 2197 | */ |
| 2198 | if (apply_to_page_range(&init_mm, (unsigned long)area->addr, |
| 2199 | area->size, f, NULL)) { |
| 2200 | free_vm_area(area); |
| 2201 | return NULL; |
| 2202 | } |
| 2203 | |
David Vrabel | 461ae48 | 2011-09-14 16:22:02 -0700 | [diff] [blame] | 2204 | /* |
| 2205 | * If the allocated address space is passed to a hypercall |
| 2206 | * before being used then we cannot rely on a page fault to |
| 2207 | * trigger an update of the page tables. So sync all the page |
| 2208 | * tables here. |
| 2209 | */ |
| 2210 | vmalloc_sync_all(); |
| 2211 | |
Jeremy Fitzhardinge | 5f4352f | 2007-07-17 18:37:04 -0700 | [diff] [blame] | 2212 | return area; |
| 2213 | } |
| 2214 | EXPORT_SYMBOL_GPL(alloc_vm_area); |
| 2215 | |
| 2216 | void free_vm_area(struct vm_struct *area) |
| 2217 | { |
| 2218 | struct vm_struct *ret; |
| 2219 | ret = remove_vm_area(area->addr); |
| 2220 | BUG_ON(ret != area); |
| 2221 | kfree(area); |
| 2222 | } |
| 2223 | EXPORT_SYMBOL_GPL(free_vm_area); |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2224 | |
Tejun Heo | 4f8b02b | 2010-09-03 18:22:47 +0200 | [diff] [blame] | 2225 | #ifdef CONFIG_SMP |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2226 | static struct vmap_area *node_to_va(struct rb_node *n) |
| 2227 | { |
| 2228 | return n ? rb_entry(n, struct vmap_area, rb_node) : NULL; |
| 2229 | } |
| 2230 | |
| 2231 | /** |
| 2232 | * pvm_find_next_prev - find the next and prev vmap_area surrounding @end |
| 2233 | * @end: target address |
| 2234 | * @pnext: out arg for the next vmap_area |
| 2235 | * @pprev: out arg for the previous vmap_area |
| 2236 | * |
| 2237 | * Returns: %true if either or both of next and prev are found, |
| 2238 | * %false if no vmap_area exists |
| 2239 | * |
| 2240 | * Find vmap_areas end addresses of which enclose @end. ie. if not |
| 2241 | * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. |
| 2242 | */ |
| 2243 | static bool pvm_find_next_prev(unsigned long end, |
| 2244 | struct vmap_area **pnext, |
| 2245 | struct vmap_area **pprev) |
| 2246 | { |
| 2247 | struct rb_node *n = vmap_area_root.rb_node; |
| 2248 | struct vmap_area *va = NULL; |
| 2249 | |
| 2250 | while (n) { |
| 2251 | va = rb_entry(n, struct vmap_area, rb_node); |
| 2252 | if (end < va->va_end) |
| 2253 | n = n->rb_left; |
| 2254 | else if (end > va->va_end) |
| 2255 | n = n->rb_right; |
| 2256 | else |
| 2257 | break; |
| 2258 | } |
| 2259 | |
| 2260 | if (!va) |
| 2261 | return false; |
| 2262 | |
| 2263 | if (va->va_end > end) { |
| 2264 | *pnext = va; |
| 2265 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); |
| 2266 | } else { |
| 2267 | *pprev = va; |
| 2268 | *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); |
| 2269 | } |
| 2270 | return true; |
| 2271 | } |
| 2272 | |
| 2273 | /** |
| 2274 | * pvm_determine_end - find the highest aligned address between two vmap_areas |
| 2275 | * @pnext: in/out arg for the next vmap_area |
| 2276 | * @pprev: in/out arg for the previous vmap_area |
| 2277 | * @align: alignment |
| 2278 | * |
| 2279 | * Returns: determined end address |
| 2280 | * |
| 2281 | * Find the highest aligned address between *@pnext and *@pprev below |
| 2282 | * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned |
| 2283 | * down address is between the end addresses of the two vmap_areas. |
| 2284 | * |
| 2285 | * Please note that the address returned by this function may fall |
| 2286 | * inside *@pnext vmap_area. The caller is responsible for checking |
| 2287 | * that. |
| 2288 | */ |
| 2289 | static unsigned long pvm_determine_end(struct vmap_area **pnext, |
| 2290 | struct vmap_area **pprev, |
| 2291 | unsigned long align) |
| 2292 | { |
| 2293 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
| 2294 | unsigned long addr; |
| 2295 | |
| 2296 | if (*pnext) |
| 2297 | addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); |
| 2298 | else |
| 2299 | addr = vmalloc_end; |
| 2300 | |
| 2301 | while (*pprev && (*pprev)->va_end > addr) { |
| 2302 | *pnext = *pprev; |
| 2303 | *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); |
| 2304 | } |
| 2305 | |
| 2306 | return addr; |
| 2307 | } |
| 2308 | |
| 2309 | /** |
| 2310 | * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator |
| 2311 | * @offsets: array containing offset of each area |
| 2312 | * @sizes: array containing size of each area |
| 2313 | * @nr_vms: the number of areas to allocate |
| 2314 | * @align: alignment, all entries in @offsets and @sizes must be aligned to this |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2315 | * |
| 2316 | * Returns: kmalloc'd vm_struct pointer array pointing to allocated |
| 2317 | * vm_structs on success, %NULL on failure |
| 2318 | * |
| 2319 | * Percpu allocator wants to use congruent vm areas so that it can |
| 2320 | * maintain the offsets among percpu areas. This function allocates |
David Rientjes | ec3f64f | 2011-01-13 15:46:01 -0800 | [diff] [blame] | 2321 | * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to |
| 2322 | * be scattered pretty far, distance between two areas easily going up |
| 2323 | * to gigabytes. To avoid interacting with regular vmallocs, these |
| 2324 | * areas are allocated from top. |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2325 | * |
| 2326 | * Despite its complicated look, this allocator is rather simple. It |
| 2327 | * does everything top-down and scans areas from the end looking for |
| 2328 | * matching slot. While scanning, if any of the areas overlaps with |
| 2329 | * existing vmap_area, the base address is pulled down to fit the |
| 2330 | * area. Scanning is repeated till all the areas fit and then all |
| 2331 | * necessary data structres are inserted and the result is returned. |
| 2332 | */ |
| 2333 | struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, |
| 2334 | const size_t *sizes, int nr_vms, |
David Rientjes | ec3f64f | 2011-01-13 15:46:01 -0800 | [diff] [blame] | 2335 | size_t align) |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2336 | { |
| 2337 | const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); |
| 2338 | const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); |
| 2339 | struct vmap_area **vas, *prev, *next; |
| 2340 | struct vm_struct **vms; |
| 2341 | int area, area2, last_area, term_area; |
| 2342 | unsigned long base, start, end, last_end; |
| 2343 | bool purged = false; |
| 2344 | |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2345 | /* verify parameters and allocate data structures */ |
| 2346 | BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align)); |
| 2347 | for (last_area = 0, area = 0; area < nr_vms; area++) { |
| 2348 | start = offsets[area]; |
| 2349 | end = start + sizes[area]; |
| 2350 | |
| 2351 | /* is everything aligned properly? */ |
| 2352 | BUG_ON(!IS_ALIGNED(offsets[area], align)); |
| 2353 | BUG_ON(!IS_ALIGNED(sizes[area], align)); |
| 2354 | |
| 2355 | /* detect the area with the highest address */ |
| 2356 | if (start > offsets[last_area]) |
| 2357 | last_area = area; |
| 2358 | |
| 2359 | for (area2 = 0; area2 < nr_vms; area2++) { |
| 2360 | unsigned long start2 = offsets[area2]; |
| 2361 | unsigned long end2 = start2 + sizes[area2]; |
| 2362 | |
| 2363 | if (area2 == area) |
| 2364 | continue; |
| 2365 | |
| 2366 | BUG_ON(start2 >= start && start2 < end); |
| 2367 | BUG_ON(end2 <= end && end2 > start); |
| 2368 | } |
| 2369 | } |
| 2370 | last_end = offsets[last_area] + sizes[last_area]; |
| 2371 | |
| 2372 | if (vmalloc_end - vmalloc_start < last_end) { |
| 2373 | WARN_ON(true); |
| 2374 | return NULL; |
| 2375 | } |
| 2376 | |
David Rientjes | ec3f64f | 2011-01-13 15:46:01 -0800 | [diff] [blame] | 2377 | vms = kzalloc(sizeof(vms[0]) * nr_vms, GFP_KERNEL); |
| 2378 | vas = kzalloc(sizeof(vas[0]) * nr_vms, GFP_KERNEL); |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2379 | if (!vas || !vms) |
| 2380 | goto err_free; |
| 2381 | |
| 2382 | for (area = 0; area < nr_vms; area++) { |
David Rientjes | ec3f64f | 2011-01-13 15:46:01 -0800 | [diff] [blame] | 2383 | vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); |
| 2384 | vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); |
Tejun Heo | ca23e40 | 2009-08-14 15:00:52 +0900 | [diff] [blame] | 2385 | if (!vas[area] || !vms[area]) |
| 2386 | goto err_free; |
| 2387 | } |
| 2388 | retry: |
| 2389 | spin_lock(&vmap_area_lock); |
| 2390 | |
| 2391 | /* start scanning - we scan from the top, begin with the last area */ |
| 2392 | area = term_area = last_area; |
| 2393 | start = offsets[area]; |
| 2394 | end = start + sizes[area]; |
| 2395 | |
| 2396 | if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { |
| 2397 | base = vmalloc_end - last_end; |
| 2398 | goto found; |
| 2399 | } |
| 2400 | base = pvm_determine_end(&next, &prev, align) - end; |
| 2401 | |
| 2402 | while (true) { |
| 2403 | BUG_ON(next && next->va_end <= base + end); |
| 2404 | BUG_ON(prev && prev->va_end > base + end); |
| 2405 | |
| 2406 | /* |
| 2407 | * base might have underflowed, add last_end before |
| 2408 | * comparing. |
| 2409 | */ |
| 2410 | if (base + last_end < vmalloc_start + last_end) { |
| 2411 | spin_unlock(&vmap_area_lock); |
| 2412 | if (!purged) { |
| 2413 | purge_vmap_area_lazy(); |
| 2414 | purged = true; |
| 2415 | goto retry; |
| 2416 | } |
| 2417 | goto err_free; |
| 2418 | } |
| 2419 | |
| 2420 | /* |
| 2421 | * If next overlaps, move base downwards so that it's |
| 2422 | * right below next and then recheck. |
| 2423 | */ |
| 2424 | if (next && next->va_start < base + end) { |
| 2425 | base = pvm_determine_end(&next, &prev, align) - end; |
| 2426 | term_area = area; |
| 2427 | continue; |
| 2428 | } |
| 2429 | |
| 2430 | /* |
| 2431 | * If prev overlaps, shift down next and prev and move |
| 2432 | * base so that it's right below new next and then |
| 2433 | * recheck. |
| 2434 | */ |
| 2435 | if (prev && prev->va_end > base + start) { |
| 2436 | next = prev; |
| 2437 | prev = node_to_va(rb_prev(&next->rb_node)); |
| 2438 | base = pvm_determine_end(&next, &prev, align) - end; |
| 2439 | term_area = area; |
| 2440 | continue; |
| 2441 | } |
| 2442 | |
| 2443 | /* |
| 2444 | * This area fits, move on to the previous one. If |
| 2445 | * the previous one is the terminal one, we're done. |
| 2446 | */ |
| 2447 | area = (area + nr_vms - 1) % nr_vms; |
| 2448 | if (area == term_area) |
| 2449 | break; |
| 2450 | start = offsets[area]; |
| 2451 | end = start + sizes[area]; |
| 2452 | pvm_find_next_prev(base + end, &next, &prev); |
| 2453 | } |
| 2454 | found: |
| 2455 | /* we've found a fitting base, insert all va's */ |
| 2456 | for (area = 0; area < nr_vms; area++) { |
| 2457 | struct vmap_area *va = vas[area]; |
| 2458 | |
| 2459 | va->va_start = base + offsets[area]; |
| 2460 | va->va_end = va->va_start + sizes[area]; |
| 2461 | __insert_vmap_area(va); |
| 2462 | } |
| 2463 | |
| 2464 | vmap_area_pcpu_hole = base + offsets[last_area]; |
| 2465 | |
| 2466 | spin_unlock(&vmap_area_lock); |
| 2467 | |
| 2468 | /* insert all vm's */ |
| 2469 | for (area = 0; area < nr_vms; area++) |
| 2470 | insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC, |
| 2471 | pcpu_get_vm_areas); |
| 2472 | |
| 2473 | kfree(vas); |
| 2474 | return vms; |
| 2475 | |
| 2476 | err_free: |
| 2477 | for (area = 0; area < nr_vms; area++) { |
| 2478 | if (vas) |
| 2479 | kfree(vas[area]); |
| 2480 | if (vms) |
| 2481 | kfree(vms[area]); |
| 2482 | } |
| 2483 | kfree(vas); |
| 2484 | kfree(vms); |
| 2485 | return NULL; |
| 2486 | } |
| 2487 | |
| 2488 | /** |
| 2489 | * pcpu_free_vm_areas - free vmalloc areas for percpu allocator |
| 2490 | * @vms: vm_struct pointer array returned by pcpu_get_vm_areas() |
| 2491 | * @nr_vms: the number of allocated areas |
| 2492 | * |
| 2493 | * Free vm_structs and the array allocated by pcpu_get_vm_areas(). |
| 2494 | */ |
| 2495 | void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms) |
| 2496 | { |
| 2497 | int i; |
| 2498 | |
| 2499 | for (i = 0; i < nr_vms; i++) |
| 2500 | free_vm_area(vms[i]); |
| 2501 | kfree(vms); |
| 2502 | } |
Tejun Heo | 4f8b02b | 2010-09-03 18:22:47 +0200 | [diff] [blame] | 2503 | #endif /* CONFIG_SMP */ |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2504 | |
| 2505 | #ifdef CONFIG_PROC_FS |
| 2506 | static void *s_start(struct seq_file *m, loff_t *pos) |
Namhyung Kim | e199b5d | 2010-10-26 14:22:03 -0700 | [diff] [blame] | 2507 | __acquires(&vmlist_lock) |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2508 | { |
| 2509 | loff_t n = *pos; |
| 2510 | struct vm_struct *v; |
| 2511 | |
| 2512 | read_lock(&vmlist_lock); |
| 2513 | v = vmlist; |
| 2514 | while (n > 0 && v) { |
| 2515 | n--; |
| 2516 | v = v->next; |
| 2517 | } |
| 2518 | if (!n) |
| 2519 | return v; |
| 2520 | |
| 2521 | return NULL; |
| 2522 | |
| 2523 | } |
| 2524 | |
| 2525 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
| 2526 | { |
| 2527 | struct vm_struct *v = p; |
| 2528 | |
| 2529 | ++*pos; |
| 2530 | return v->next; |
| 2531 | } |
| 2532 | |
| 2533 | static void s_stop(struct seq_file *m, void *p) |
Namhyung Kim | e199b5d | 2010-10-26 14:22:03 -0700 | [diff] [blame] | 2534 | __releases(&vmlist_lock) |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2535 | { |
| 2536 | read_unlock(&vmlist_lock); |
| 2537 | } |
| 2538 | |
Eric Dumazet | a47a126 | 2008-07-23 21:27:38 -0700 | [diff] [blame] | 2539 | static void show_numa_info(struct seq_file *m, struct vm_struct *v) |
| 2540 | { |
| 2541 | if (NUMA_BUILD) { |
| 2542 | unsigned int nr, *counters = m->private; |
| 2543 | |
| 2544 | if (!counters) |
| 2545 | return; |
| 2546 | |
| 2547 | memset(counters, 0, nr_node_ids * sizeof(unsigned int)); |
| 2548 | |
| 2549 | for (nr = 0; nr < v->nr_pages; nr++) |
| 2550 | counters[page_to_nid(v->pages[nr])]++; |
| 2551 | |
| 2552 | for_each_node_state(nr, N_HIGH_MEMORY) |
| 2553 | if (counters[nr]) |
| 2554 | seq_printf(m, " N%u=%u", nr, counters[nr]); |
| 2555 | } |
| 2556 | } |
| 2557 | |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2558 | static int s_show(struct seq_file *m, void *p) |
| 2559 | { |
| 2560 | struct vm_struct *v = p; |
| 2561 | |
| 2562 | seq_printf(m, "0x%p-0x%p %7ld", |
| 2563 | v->addr, v->addr + v->size, v->size); |
| 2564 | |
Joe Perches | 62c70bc | 2011-01-13 15:45:52 -0800 | [diff] [blame] | 2565 | if (v->caller) |
| 2566 | seq_printf(m, " %pS", v->caller); |
Christoph Lameter | 2301696 | 2008-04-28 02:12:42 -0700 | [diff] [blame] | 2567 | |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2568 | if (v->nr_pages) |
| 2569 | seq_printf(m, " pages=%d", v->nr_pages); |
| 2570 | |
| 2571 | if (v->phys_addr) |
Kenji Kaneshige | ffa71f3 | 2010-06-18 12:22:40 +0900 | [diff] [blame] | 2572 | seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr); |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2573 | |
| 2574 | if (v->flags & VM_IOREMAP) |
| 2575 | seq_printf(m, " ioremap"); |
| 2576 | |
| 2577 | if (v->flags & VM_ALLOC) |
| 2578 | seq_printf(m, " vmalloc"); |
| 2579 | |
| 2580 | if (v->flags & VM_MAP) |
| 2581 | seq_printf(m, " vmap"); |
| 2582 | |
| 2583 | if (v->flags & VM_USERMAP) |
| 2584 | seq_printf(m, " user"); |
| 2585 | |
| 2586 | if (v->flags & VM_VPAGES) |
| 2587 | seq_printf(m, " vpages"); |
| 2588 | |
Eric Dumazet | a47a126 | 2008-07-23 21:27:38 -0700 | [diff] [blame] | 2589 | show_numa_info(m, v); |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2590 | seq_putc(m, '\n'); |
| 2591 | return 0; |
| 2592 | } |
| 2593 | |
Alexey Dobriyan | 5f6a6a9 | 2008-10-06 03:50:47 +0400 | [diff] [blame] | 2594 | static const struct seq_operations vmalloc_op = { |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2595 | .start = s_start, |
| 2596 | .next = s_next, |
| 2597 | .stop = s_stop, |
| 2598 | .show = s_show, |
| 2599 | }; |
Alexey Dobriyan | 5f6a6a9 | 2008-10-06 03:50:47 +0400 | [diff] [blame] | 2600 | |
| 2601 | static int vmalloc_open(struct inode *inode, struct file *file) |
| 2602 | { |
| 2603 | unsigned int *ptr = NULL; |
| 2604 | int ret; |
| 2605 | |
Kulikov Vasiliy | 51980ac | 2010-08-09 17:19:58 -0700 | [diff] [blame] | 2606 | if (NUMA_BUILD) { |
Alexey Dobriyan | 5f6a6a9 | 2008-10-06 03:50:47 +0400 | [diff] [blame] | 2607 | ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL); |
Kulikov Vasiliy | 51980ac | 2010-08-09 17:19:58 -0700 | [diff] [blame] | 2608 | if (ptr == NULL) |
| 2609 | return -ENOMEM; |
| 2610 | } |
Alexey Dobriyan | 5f6a6a9 | 2008-10-06 03:50:47 +0400 | [diff] [blame] | 2611 | ret = seq_open(file, &vmalloc_op); |
| 2612 | if (!ret) { |
| 2613 | struct seq_file *m = file->private_data; |
| 2614 | m->private = ptr; |
| 2615 | } else |
| 2616 | kfree(ptr); |
| 2617 | return ret; |
| 2618 | } |
| 2619 | |
| 2620 | static const struct file_operations proc_vmalloc_operations = { |
| 2621 | .open = vmalloc_open, |
| 2622 | .read = seq_read, |
| 2623 | .llseek = seq_lseek, |
| 2624 | .release = seq_release_private, |
| 2625 | }; |
| 2626 | |
| 2627 | static int __init proc_vmalloc_init(void) |
| 2628 | { |
| 2629 | proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations); |
| 2630 | return 0; |
| 2631 | } |
| 2632 | module_init(proc_vmalloc_init); |
Christoph Lameter | a10aa57 | 2008-04-28 02:12:40 -0700 | [diff] [blame] | 2633 | #endif |
| 2634 | |