Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/arch/x86_64/mm/init.c |
| 3 | * |
| 4 | * Copyright (C) 1995 Linus Torvalds |
| 5 | * Copyright (C) 2000 Pavel Machek <pavel@suse.cz> |
| 6 | * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de> |
| 7 | */ |
| 8 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | #include <linux/signal.h> |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/errno.h> |
| 13 | #include <linux/string.h> |
| 14 | #include <linux/types.h> |
| 15 | #include <linux/ptrace.h> |
| 16 | #include <linux/mman.h> |
| 17 | #include <linux/mm.h> |
| 18 | #include <linux/swap.h> |
| 19 | #include <linux/smp.h> |
| 20 | #include <linux/init.h> |
| 21 | #include <linux/pagemap.h> |
| 22 | #include <linux/bootmem.h> |
| 23 | #include <linux/proc_fs.h> |
Andi Kleen | 5917089 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 24 | #include <linux/pci.h> |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 25 | #include <linux/pfn.h> |
Randy Dunlap | c9cf552 | 2006-06-27 02:53:52 -0700 | [diff] [blame] | 26 | #include <linux/poison.h> |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 27 | #include <linux/dma-mapping.h> |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 28 | #include <linux/module.h> |
| 29 | #include <linux/memory_hotplug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | |
| 31 | #include <asm/processor.h> |
| 32 | #include <asm/system.h> |
| 33 | #include <asm/uaccess.h> |
| 34 | #include <asm/pgtable.h> |
| 35 | #include <asm/pgalloc.h> |
| 36 | #include <asm/dma.h> |
| 37 | #include <asm/fixmap.h> |
| 38 | #include <asm/e820.h> |
| 39 | #include <asm/apic.h> |
| 40 | #include <asm/tlb.h> |
| 41 | #include <asm/mmu_context.h> |
| 42 | #include <asm/proto.h> |
| 43 | #include <asm/smp.h> |
Andi Kleen | 2bc0414 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 44 | #include <asm/sections.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | |
| 46 | #ifndef Dprintk |
| 47 | #define Dprintk(x...) |
| 48 | #endif |
| 49 | |
Stephen Hemminger | e658450 | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 50 | const struct dma_mapping_ops* dma_ops; |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 51 | EXPORT_SYMBOL(dma_ops); |
| 52 | |
Andi Kleen | e18c687 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 53 | static unsigned long dma_reserve __initdata; |
| 54 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
| 56 | |
| 57 | /* |
| 58 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the |
| 59 | * physical space so we can cache the place of the first one and move |
| 60 | * around without checking the pgd every time. |
| 61 | */ |
| 62 | |
| 63 | void show_mem(void) |
| 64 | { |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 65 | long i, total = 0, reserved = 0; |
| 66 | long shared = 0, cached = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | pg_data_t *pgdat; |
| 68 | struct page *page; |
| 69 | |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 70 | printk(KERN_INFO "Mem-info:\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | show_free_areas(); |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 72 | printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | |
KAMEZAWA Hiroyuki | ec936fc | 2006-03-27 01:15:59 -0800 | [diff] [blame] | 74 | for_each_online_pgdat(pgdat) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
| 76 | page = pfn_to_page(pgdat->node_start_pfn + i); |
| 77 | total++; |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 78 | if (PageReserved(page)) |
| 79 | reserved++; |
| 80 | else if (PageSwapCache(page)) |
| 81 | cached++; |
| 82 | else if (page_count(page)) |
| 83 | shared += page_count(page) - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | } |
| 85 | } |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 86 | printk(KERN_INFO "%lu pages of RAM\n", total); |
| 87 | printk(KERN_INFO "%lu reserved pages\n",reserved); |
| 88 | printk(KERN_INFO "%lu pages shared\n",shared); |
| 89 | printk(KERN_INFO "%lu pages swap cached\n",cached); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | } |
| 91 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | int after_bootmem; |
| 93 | |
Andi Kleen | 5f44a66 | 2006-03-25 16:30:25 +0100 | [diff] [blame] | 94 | static __init void *spp_getpage(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | { |
| 96 | void *ptr; |
| 97 | if (after_bootmem) |
| 98 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); |
| 99 | else |
| 100 | ptr = alloc_bootmem_pages(PAGE_SIZE); |
| 101 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) |
| 102 | panic("set_pte_phys: cannot allocate page data %s\n", after_bootmem?"after bootmem":""); |
| 103 | |
| 104 | Dprintk("spp_getpage %p\n", ptr); |
| 105 | return ptr; |
| 106 | } |
| 107 | |
Andi Kleen | 5f44a66 | 2006-03-25 16:30:25 +0100 | [diff] [blame] | 108 | static __init void set_pte_phys(unsigned long vaddr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | unsigned long phys, pgprot_t prot) |
| 110 | { |
| 111 | pgd_t *pgd; |
| 112 | pud_t *pud; |
| 113 | pmd_t *pmd; |
| 114 | pte_t *pte, new_pte; |
| 115 | |
| 116 | Dprintk("set_pte_phys %lx to %lx\n", vaddr, phys); |
| 117 | |
| 118 | pgd = pgd_offset_k(vaddr); |
| 119 | if (pgd_none(*pgd)) { |
| 120 | printk("PGD FIXMAP MISSING, it should be setup in head.S!\n"); |
| 121 | return; |
| 122 | } |
| 123 | pud = pud_offset(pgd, vaddr); |
| 124 | if (pud_none(*pud)) { |
| 125 | pmd = (pmd_t *) spp_getpage(); |
| 126 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); |
| 127 | if (pmd != pmd_offset(pud, 0)) { |
| 128 | printk("PAGETABLE BUG #01! %p <-> %p\n", pmd, pmd_offset(pud,0)); |
| 129 | return; |
| 130 | } |
| 131 | } |
| 132 | pmd = pmd_offset(pud, vaddr); |
| 133 | if (pmd_none(*pmd)) { |
| 134 | pte = (pte_t *) spp_getpage(); |
| 135 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER)); |
| 136 | if (pte != pte_offset_kernel(pmd, 0)) { |
| 137 | printk("PAGETABLE BUG #02!\n"); |
| 138 | return; |
| 139 | } |
| 140 | } |
| 141 | new_pte = pfn_pte(phys >> PAGE_SHIFT, prot); |
| 142 | |
| 143 | pte = pte_offset_kernel(pmd, vaddr); |
| 144 | if (!pte_none(*pte) && |
| 145 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) |
| 146 | pte_ERROR(*pte); |
| 147 | set_pte(pte, new_pte); |
| 148 | |
| 149 | /* |
| 150 | * It's enough to flush this one mapping. |
| 151 | * (PGE mappings get flushed as well) |
| 152 | */ |
| 153 | __flush_tlb_one(vaddr); |
| 154 | } |
| 155 | |
| 156 | /* NOTE: this is meant to be run only at boot */ |
Andi Kleen | 5f44a66 | 2006-03-25 16:30:25 +0100 | [diff] [blame] | 157 | void __init |
| 158 | __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t prot) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | { |
| 160 | unsigned long address = __fix_to_virt(idx); |
| 161 | |
| 162 | if (idx >= __end_of_fixed_addresses) { |
| 163 | printk("Invalid __set_fixmap\n"); |
| 164 | return; |
| 165 | } |
| 166 | set_pte_phys(address, phys, prot); |
| 167 | } |
| 168 | |
| 169 | unsigned long __initdata table_start, table_end; |
| 170 | |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 171 | static __meminit void *alloc_low_page(unsigned long *phys) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | { |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 173 | unsigned long pfn = table_end++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | void *adr; |
| 175 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 176 | if (after_bootmem) { |
| 177 | adr = (void *)get_zeroed_page(GFP_ATOMIC); |
| 178 | *phys = __pa(adr); |
| 179 | return adr; |
| 180 | } |
| 181 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | if (pfn >= end_pfn) |
| 183 | panic("alloc_low_page: ran out of memory"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 185 | adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); |
| 186 | memset(adr, 0, PAGE_SIZE); |
| 187 | *phys = pfn * PAGE_SIZE; |
| 188 | return adr; |
| 189 | } |
| 190 | |
| 191 | static __meminit void unmap_low_page(void *adr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | { |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 193 | |
| 194 | if (after_bootmem) |
| 195 | return; |
| 196 | |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 197 | early_iounmap(adr, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | } |
| 199 | |
Andi Kleen | f2d3efe | 2006-03-25 16:30:22 +0100 | [diff] [blame] | 200 | /* Must run before zap_low_mappings */ |
| 201 | __init void *early_ioremap(unsigned long addr, unsigned long size) |
| 202 | { |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 203 | unsigned long vaddr; |
| 204 | pmd_t *pmd, *last_pmd; |
| 205 | int i, pmds; |
Andi Kleen | f2d3efe | 2006-03-25 16:30:22 +0100 | [diff] [blame] | 206 | |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 207 | pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; |
| 208 | vaddr = __START_KERNEL_map; |
| 209 | pmd = level2_kernel_pgt; |
| 210 | last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1; |
| 211 | for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) { |
| 212 | for (i = 0; i < pmds; i++) { |
| 213 | if (pmd_present(pmd[i])) |
| 214 | goto next; |
| 215 | } |
| 216 | vaddr += addr & ~PMD_MASK; |
| 217 | addr &= PMD_MASK; |
| 218 | for (i = 0; i < pmds; i++, addr += PMD_SIZE) |
| 219 | set_pmd(pmd + i,__pmd(addr | _KERNPG_TABLE | _PAGE_PSE)); |
| 220 | __flush_tlb(); |
| 221 | return (void *)vaddr; |
| 222 | next: |
| 223 | ; |
Andi Kleen | f2d3efe | 2006-03-25 16:30:22 +0100 | [diff] [blame] | 224 | } |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 225 | printk("early_ioremap(0x%lx, %lu) failed\n", addr, size); |
| 226 | return NULL; |
Andi Kleen | f2d3efe | 2006-03-25 16:30:22 +0100 | [diff] [blame] | 227 | } |
| 228 | |
| 229 | /* To avoid virtual aliases later */ |
| 230 | __init void early_iounmap(void *addr, unsigned long size) |
| 231 | { |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 232 | unsigned long vaddr; |
| 233 | pmd_t *pmd; |
| 234 | int i, pmds; |
| 235 | |
| 236 | vaddr = (unsigned long)addr; |
| 237 | pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE; |
| 238 | pmd = level2_kernel_pgt + pmd_index(vaddr); |
| 239 | for (i = 0; i < pmds; i++) |
| 240 | pmd_clear(pmd + i); |
Andi Kleen | f2d3efe | 2006-03-25 16:30:22 +0100 | [diff] [blame] | 241 | __flush_tlb(); |
| 242 | } |
| 243 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 244 | static void __meminit |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 245 | phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end) |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 246 | { |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 247 | int i = pmd_index(address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 249 | for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 250 | unsigned long entry; |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 251 | pmd_t *pmd = pmd_page + pmd_index(address); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 252 | |
Jan Beulich | 5f51e13 | 2006-06-26 13:59:02 +0200 | [diff] [blame] | 253 | if (address >= end) { |
| 254 | if (!after_bootmem) |
| 255 | for (; i < PTRS_PER_PMD; i++, pmd++) |
| 256 | set_pmd(pmd, __pmd(0)); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 257 | break; |
| 258 | } |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 259 | |
| 260 | if (pmd_val(*pmd)) |
| 261 | continue; |
| 262 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 263 | entry = _PAGE_NX|_PAGE_PSE|_KERNPG_TABLE|_PAGE_GLOBAL|address; |
| 264 | entry &= __supported_pte_mask; |
| 265 | set_pmd(pmd, __pmd(entry)); |
| 266 | } |
| 267 | } |
| 268 | |
| 269 | static void __meminit |
| 270 | phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end) |
| 271 | { |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 272 | pmd_t *pmd = pmd_offset(pud,0); |
| 273 | spin_lock(&init_mm.page_table_lock); |
| 274 | phys_pmd_init(pmd, address, end); |
| 275 | spin_unlock(&init_mm.page_table_lock); |
| 276 | __flush_tlb_all(); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 277 | } |
| 278 | |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 279 | static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end) |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 280 | { |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 281 | int i = pud_index(addr); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 282 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 283 | |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 284 | for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE ) { |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 285 | unsigned long pmd_phys; |
| 286 | pud_t *pud = pud_page + pud_index(addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | pmd_t *pmd; |
| 288 | |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 289 | if (addr >= end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 292 | if (!after_bootmem && !e820_any_mapped(addr,addr+PUD_SIZE,0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | set_pud(pud, __pud(0)); |
| 294 | continue; |
| 295 | } |
| 296 | |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 297 | if (pud_val(*pud)) { |
| 298 | phys_pmd_update(pud, addr, end); |
| 299 | continue; |
| 300 | } |
| 301 | |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 302 | pmd = alloc_low_page(&pmd_phys); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 303 | spin_lock(&init_mm.page_table_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); |
Keith Mannthey | 6ad9165 | 2006-09-26 10:52:36 +0200 | [diff] [blame] | 305 | phys_pmd_init(pmd, addr, end); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 306 | spin_unlock(&init_mm.page_table_lock); |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 307 | unmap_low_page(pmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | } |
| 309 | __flush_tlb(); |
| 310 | } |
| 311 | |
| 312 | static void __init find_early_table_space(unsigned long end) |
| 313 | { |
Andi Kleen | 6c5acd1 | 2006-01-11 22:46:57 +0100 | [diff] [blame] | 314 | unsigned long puds, pmds, tables, start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | |
| 316 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; |
| 317 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; |
| 318 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) + |
| 319 | round_up(pmds * sizeof(pmd_t), PAGE_SIZE); |
| 320 | |
Andi Kleen | ee408c7 | 2006-01-16 01:56:51 +0100 | [diff] [blame] | 321 | /* RED-PEN putting page tables only on node 0 could |
| 322 | cause a hotspot and fill up ZONE_DMA. The page tables |
| 323 | need roughly 0.5KB per GB. */ |
| 324 | start = 0x8000; |
| 325 | table_start = find_e820_area(start, end, tables); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | if (table_start == -1UL) |
| 327 | panic("Cannot find space for the kernel page tables"); |
| 328 | |
| 329 | table_start >>= PAGE_SHIFT; |
| 330 | table_end = table_start; |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 331 | |
| 332 | early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n", |
Jan Beulich | 5f51e13 | 2006-06-26 13:59:02 +0200 | [diff] [blame] | 333 | end, table_start << PAGE_SHIFT, |
| 334 | (table_start << PAGE_SHIFT) + tables); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | } |
| 336 | |
| 337 | /* Setup the direct mapping of the physical memory at PAGE_OFFSET. |
| 338 | This runs before bootmem is initialized and gets pages directly from the |
| 339 | physical memory. To access them they are temporarily mapped. */ |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 340 | void __meminit init_memory_mapping(unsigned long start, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | { |
| 342 | unsigned long next; |
| 343 | |
| 344 | Dprintk("init_memory_mapping\n"); |
| 345 | |
| 346 | /* |
| 347 | * Find space for the kernel direct mapping tables. |
| 348 | * Later we should allocate these tables in the local node of the memory |
| 349 | * mapped. Unfortunately this is done currently before the nodes are |
| 350 | * discovered. |
| 351 | */ |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 352 | if (!after_bootmem) |
| 353 | find_early_table_space(end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | |
| 355 | start = (unsigned long)__va(start); |
| 356 | end = (unsigned long)__va(end); |
| 357 | |
| 358 | for (; start < end; start = next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | unsigned long pud_phys; |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 360 | pgd_t *pgd = pgd_offset_k(start); |
| 361 | pud_t *pud; |
| 362 | |
| 363 | if (after_bootmem) |
Andi Kleen | d2ae5b5 | 2006-06-26 13:57:56 +0200 | [diff] [blame] | 364 | pud = pud_offset(pgd, start & PGDIR_MASK); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 365 | else |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 366 | pud = alloc_low_page(&pud_phys); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 367 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | next = start + PGDIR_SIZE; |
| 369 | if (next > end) |
| 370 | next = end; |
| 371 | phys_pud_init(pud, __pa(start), __pa(next)); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 372 | if (!after_bootmem) |
| 373 | set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); |
Vivek Goyal | dafe41e | 2007-05-02 19:27:06 +0200 | [diff] [blame] | 374 | unmap_low_page(pud); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | } |
| 376 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 377 | if (!after_bootmem) |
| 378 | asm volatile("movq %%cr4,%0" : "=r" (mmu_cr4_features)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | __flush_tlb_all(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 380 | } |
| 381 | |
Matt Tolentino | 2b97690 | 2005-06-23 00:08:06 -0700 | [diff] [blame] | 382 | #ifndef CONFIG_NUMA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | void __init paging_init(void) |
| 384 | { |
Mel Gorman | 6391af1 | 2006-10-11 01:20:39 -0700 | [diff] [blame] | 385 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
| 386 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
| 387 | max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; |
| 388 | max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; |
| 389 | max_zone_pfns[ZONE_NORMAL] = end_pfn; |
| 390 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 391 | memory_present(0, 0, end_pfn); |
| 392 | sparse_init(); |
Mel Gorman | 5cb248a | 2006-09-27 01:49:52 -0700 | [diff] [blame] | 393 | free_area_init_nodes(max_zone_pfns); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | } |
| 395 | #endif |
| 396 | |
| 397 | /* Unmap a kernel mapping if it exists. This is useful to avoid prefetches |
| 398 | from the CPU leading to inconsistent cache lines. address and size |
| 399 | must be aligned to 2MB boundaries. |
| 400 | Does nothing when the mapping doesn't exist. */ |
| 401 | void __init clear_kernel_mapping(unsigned long address, unsigned long size) |
| 402 | { |
| 403 | unsigned long end = address + size; |
| 404 | |
| 405 | BUG_ON(address & ~LARGE_PAGE_MASK); |
| 406 | BUG_ON(size & ~LARGE_PAGE_MASK); |
| 407 | |
| 408 | for (; address < end; address += LARGE_PAGE_SIZE) { |
| 409 | pgd_t *pgd = pgd_offset_k(address); |
| 410 | pud_t *pud; |
| 411 | pmd_t *pmd; |
| 412 | if (pgd_none(*pgd)) |
| 413 | continue; |
| 414 | pud = pud_offset(pgd, address); |
| 415 | if (pud_none(*pud)) |
| 416 | continue; |
| 417 | pmd = pmd_offset(pud, address); |
| 418 | if (!pmd || pmd_none(*pmd)) |
| 419 | continue; |
| 420 | if (0 == (pmd_val(*pmd) & _PAGE_PSE)) { |
| 421 | /* Could handle this, but it should not happen currently. */ |
| 422 | printk(KERN_ERR |
| 423 | "clear_kernel_mapping: mapping has been split. will leak memory\n"); |
| 424 | pmd_ERROR(*pmd); |
| 425 | } |
| 426 | set_pmd(pmd, __pmd(0)); |
| 427 | } |
| 428 | __flush_tlb_all(); |
| 429 | } |
| 430 | |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 431 | /* |
| 432 | * Memory hotplug specific functions |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 433 | */ |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 434 | void online_page(struct page *page) |
| 435 | { |
| 436 | ClearPageReserved(page); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 437 | init_page_count(page); |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 438 | __free_page(page); |
| 439 | totalram_pages++; |
| 440 | num_physpages++; |
| 441 | } |
| 442 | |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 443 | #ifdef CONFIG_MEMORY_HOTPLUG |
| 444 | /* |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 445 | * Memory is added always to NORMAL zone. This means you will never get |
| 446 | * additional DMA/DMA32 memory. |
| 447 | */ |
| 448 | int arch_add_memory(int nid, u64 start, u64 size) |
| 449 | { |
| 450 | struct pglist_data *pgdat = NODE_DATA(nid); |
Christoph Lameter | 776ed98 | 2006-09-25 23:31:09 -0700 | [diff] [blame] | 451 | struct zone *zone = pgdat->node_zones + ZONE_NORMAL; |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 452 | unsigned long start_pfn = start >> PAGE_SHIFT; |
| 453 | unsigned long nr_pages = size >> PAGE_SHIFT; |
| 454 | int ret; |
| 455 | |
Keith Mannthey | 45e0b78 | 2006-09-30 23:27:09 -0700 | [diff] [blame] | 456 | init_memory_mapping(start, (start + size -1)); |
| 457 | |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 458 | ret = __add_pages(zone, start_pfn, nr_pages); |
| 459 | if (ret) |
| 460 | goto error; |
| 461 | |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 462 | return ret; |
| 463 | error: |
| 464 | printk("%s: Problem encountered in __add_pages!\n", __func__); |
| 465 | return ret; |
| 466 | } |
| 467 | EXPORT_SYMBOL_GPL(arch_add_memory); |
| 468 | |
| 469 | int remove_memory(u64 start, u64 size) |
| 470 | { |
| 471 | return -EINVAL; |
| 472 | } |
| 473 | EXPORT_SYMBOL_GPL(remove_memory); |
| 474 | |
Yasunori Goto | 8243229 | 2006-11-18 22:19:40 -0800 | [diff] [blame] | 475 | #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) |
Keith Mannthey | 4942e99 | 2006-09-30 23:27:06 -0700 | [diff] [blame] | 476 | int memory_add_physaddr_to_nid(u64 start) |
| 477 | { |
| 478 | return 0; |
| 479 | } |
Keith Mannthey | 8c2676a | 2006-09-30 23:27:07 -0700 | [diff] [blame] | 480 | EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); |
Keith Mannthey | 4942e99 | 2006-09-30 23:27:06 -0700 | [diff] [blame] | 481 | #endif |
| 482 | |
Keith Mannthey | 45e0b78 | 2006-09-30 23:27:09 -0700 | [diff] [blame] | 483 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
| 484 | |
| 485 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
Andi Kleen | 9d99aaa | 2006-04-07 19:49:15 +0200 | [diff] [blame] | 486 | /* |
| 487 | * Memory Hotadd without sparsemem. The mem_maps have been allocated in advance, |
| 488 | * just online the pages. |
| 489 | */ |
| 490 | int __add_pages(struct zone *z, unsigned long start_pfn, unsigned long nr_pages) |
| 491 | { |
| 492 | int err = -EIO; |
| 493 | unsigned long pfn; |
| 494 | unsigned long total = 0, mem = 0; |
| 495 | for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++) { |
Andi Kleen | 68a3a7f | 2006-04-07 19:49:18 +0200 | [diff] [blame] | 496 | if (pfn_valid(pfn)) { |
Andi Kleen | 9d99aaa | 2006-04-07 19:49:15 +0200 | [diff] [blame] | 497 | online_page(pfn_to_page(pfn)); |
| 498 | err = 0; |
| 499 | mem++; |
| 500 | } |
| 501 | total++; |
| 502 | } |
| 503 | if (!err) { |
| 504 | z->spanned_pages += total; |
| 505 | z->present_pages += mem; |
| 506 | z->zone_pgdat->node_spanned_pages += total; |
| 507 | z->zone_pgdat->node_present_pages += mem; |
| 508 | } |
| 509 | return err; |
| 510 | } |
Keith Mannthey | 45e0b78 | 2006-09-30 23:27:09 -0700 | [diff] [blame] | 511 | #endif |
Matt Tolentino | 44df75e | 2006-01-17 07:03:41 +0100 | [diff] [blame] | 512 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, kcore_modules, |
| 514 | kcore_vsyscall; |
| 515 | |
| 516 | void __init mem_init(void) |
| 517 | { |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 518 | long codesize, reservedpages, datasize, initsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | |
Jon Mason | 0dc243a | 2006-06-26 13:58:11 +0200 | [diff] [blame] | 520 | pci_iommu_alloc(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | /* clear the zero-page */ |
| 523 | memset(empty_zero_page, 0, PAGE_SIZE); |
| 524 | |
| 525 | reservedpages = 0; |
| 526 | |
| 527 | /* this will put all low memory onto the freelists */ |
Matt Tolentino | 2b97690 | 2005-06-23 00:08:06 -0700 | [diff] [blame] | 528 | #ifdef CONFIG_NUMA |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 529 | totalram_pages = numa_free_all_bootmem(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | #else |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 531 | totalram_pages = free_all_bootmem(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | #endif |
Mel Gorman | 5cb248a | 2006-09-27 01:49:52 -0700 | [diff] [blame] | 533 | reservedpages = end_pfn - totalram_pages - |
| 534 | absent_pages_in_range(0, end_pfn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | |
| 536 | after_bootmem = 1; |
| 537 | |
| 538 | codesize = (unsigned long) &_etext - (unsigned long) &_text; |
| 539 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; |
| 540 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; |
| 541 | |
| 542 | /* Register memory areas for /proc/kcore */ |
| 543 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
| 544 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, |
| 545 | VMALLOC_END-VMALLOC_START); |
| 546 | kclist_add(&kcore_kernel, &_stext, _end - _stext); |
| 547 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); |
| 548 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
| 549 | VSYSCALL_END - VSYSCALL_START); |
| 550 | |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 551 | printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, %ldk data, %ldk init)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
| 553 | end_pfn << (PAGE_SHIFT-10), |
| 554 | codesize >> 10, |
| 555 | reservedpages << (PAGE_SHIFT-10), |
| 556 | datasize >> 10, |
| 557 | initsize >> 10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | } |
| 559 | |
Gerd Hoffmann | d167a51 | 2006-06-26 13:56:16 +0200 | [diff] [blame] | 560 | void free_init_pages(char *what, unsigned long begin, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | { |
| 562 | unsigned long addr; |
| 563 | |
Gerd Hoffmann | d167a51 | 2006-06-26 13:56:16 +0200 | [diff] [blame] | 564 | if (begin >= end) |
| 565 | return; |
| 566 | |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 567 | printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); |
Gerd Hoffmann | d167a51 | 2006-06-26 13:56:16 +0200 | [diff] [blame] | 568 | for (addr = begin; addr < end; addr += PAGE_SIZE) { |
Vivek Goyal | 0dbf702 | 2007-05-02 19:27:07 +0200 | [diff] [blame] | 569 | struct page *page = pfn_to_page(addr >> PAGE_SHIFT); |
| 570 | ClearPageReserved(page); |
| 571 | init_page_count(page); |
| 572 | memset(page_address(page), POISON_FREE_INITMEM, PAGE_SIZE); |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 573 | if (addr >= __START_KERNEL_map) |
| 574 | change_page_attr_addr(addr, 1, __pgprot(0)); |
Vivek Goyal | 0dbf702 | 2007-05-02 19:27:07 +0200 | [diff] [blame] | 575 | __free_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | totalram_pages++; |
| 577 | } |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 578 | if (addr > __START_KERNEL_map) |
| 579 | global_flush_tlb(); |
Gerd Hoffmann | d167a51 | 2006-06-26 13:56:16 +0200 | [diff] [blame] | 580 | } |
| 581 | |
| 582 | void free_initmem(void) |
| 583 | { |
Gerd Hoffmann | d167a51 | 2006-06-26 13:56:16 +0200 | [diff] [blame] | 584 | free_init_pages("unused kernel memory", |
Vivek Goyal | 0dbf702 | 2007-05-02 19:27:07 +0200 | [diff] [blame] | 585 | __pa_symbol(&__init_begin), |
| 586 | __pa_symbol(&__init_end)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | } |
| 588 | |
Arjan van de Ven | 67df197 | 2006-01-06 00:12:04 -0800 | [diff] [blame] | 589 | #ifdef CONFIG_DEBUG_RODATA |
| 590 | |
Arjan van de Ven | 67df197 | 2006-01-06 00:12:04 -0800 | [diff] [blame] | 591 | void mark_rodata_ro(void) |
| 592 | { |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 593 | unsigned long start = PFN_ALIGN(__va(__pa_symbol(&_stext))), size; |
Arjan van de Ven | 67df197 | 2006-01-06 00:12:04 -0800 | [diff] [blame] | 594 | |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 595 | #ifdef CONFIG_HOTPLUG_CPU |
| 596 | /* It must still be possible to apply SMP alternatives. */ |
| 597 | if (num_possible_cpus() > 1) |
| 598 | start = PFN_ALIGN(__va(__pa_symbol(&_etext))); |
| 599 | #endif |
| 600 | size = (unsigned long)__va(__pa_symbol(&__end_rodata)) - start; |
| 601 | change_page_attr_addr(start, size >> PAGE_SHIFT, PAGE_KERNEL_RO); |
Arjan van de Ven | 67df197 | 2006-01-06 00:12:04 -0800 | [diff] [blame] | 602 | |
Jan Beulich | 6fb1475 | 2007-05-02 19:27:10 +0200 | [diff] [blame] | 603 | printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", |
| 604 | size >> 10); |
Arjan van de Ven | 67df197 | 2006-01-06 00:12:04 -0800 | [diff] [blame] | 605 | |
| 606 | /* |
| 607 | * change_page_attr_addr() requires a global_flush_tlb() call after it. |
| 608 | * We do this after the printk so that if something went wrong in the |
| 609 | * change, the printk gets out at least to give a better debug hint |
| 610 | * of who is the culprit. |
| 611 | */ |
| 612 | global_flush_tlb(); |
| 613 | } |
| 614 | #endif |
| 615 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | #ifdef CONFIG_BLK_DEV_INITRD |
| 617 | void free_initrd_mem(unsigned long start, unsigned long end) |
| 618 | { |
Vivek Goyal | 0dbf702 | 2007-05-02 19:27:07 +0200 | [diff] [blame] | 619 | free_init_pages("initrd memory", __pa(start), __pa(end)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | } |
| 621 | #endif |
| 622 | |
| 623 | void __init reserve_bootmem_generic(unsigned long phys, unsigned len) |
| 624 | { |
Matt Tolentino | 2b97690 | 2005-06-23 00:08:06 -0700 | [diff] [blame] | 625 | #ifdef CONFIG_NUMA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | int nid = phys_to_nid(phys); |
Andi Kleen | 5e58a02 | 2006-11-14 16:57:46 +0100 | [diff] [blame] | 627 | #endif |
| 628 | unsigned long pfn = phys >> PAGE_SHIFT; |
| 629 | if (pfn >= end_pfn) { |
| 630 | /* This can happen with kdump kernels when accessing firmware |
| 631 | tables. */ |
| 632 | if (pfn < end_pfn_map) |
| 633 | return; |
| 634 | printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n", |
| 635 | phys, len); |
| 636 | return; |
| 637 | } |
| 638 | |
| 639 | /* Should check here against the e820 map to avoid double free */ |
| 640 | #ifdef CONFIG_NUMA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | reserve_bootmem_node(NODE_DATA(nid), phys, len); |
| 642 | #else |
| 643 | reserve_bootmem(phys, len); |
| 644 | #endif |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 645 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { |
Andi Kleen | e18c687 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 646 | dma_reserve += len / PAGE_SIZE; |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 647 | set_dma_reserve(dma_reserve); |
| 648 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | } |
| 650 | |
| 651 | int kern_addr_valid(unsigned long addr) |
| 652 | { |
| 653 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
| 654 | pgd_t *pgd; |
| 655 | pud_t *pud; |
| 656 | pmd_t *pmd; |
| 657 | pte_t *pte; |
| 658 | |
| 659 | if (above != 0 && above != -1UL) |
| 660 | return 0; |
| 661 | |
| 662 | pgd = pgd_offset_k(addr); |
| 663 | if (pgd_none(*pgd)) |
| 664 | return 0; |
| 665 | |
| 666 | pud = pud_offset(pgd, addr); |
| 667 | if (pud_none(*pud)) |
| 668 | return 0; |
| 669 | |
| 670 | pmd = pmd_offset(pud, addr); |
| 671 | if (pmd_none(*pmd)) |
| 672 | return 0; |
| 673 | if (pmd_large(*pmd)) |
| 674 | return pfn_valid(pmd_pfn(*pmd)); |
| 675 | |
| 676 | pte = pte_offset_kernel(pmd, addr); |
| 677 | if (pte_none(*pte)) |
| 678 | return 0; |
| 679 | return pfn_valid(pte_pfn(*pte)); |
| 680 | } |
| 681 | |
| 682 | #ifdef CONFIG_SYSCTL |
| 683 | #include <linux/sysctl.h> |
| 684 | |
| 685 | extern int exception_trace, page_fault_trace; |
| 686 | |
| 687 | static ctl_table debug_table2[] = { |
Eric W. Biederman | c37ce03 | 2007-02-14 00:33:51 -0800 | [diff] [blame] | 688 | { |
| 689 | .ctl_name = 99, |
| 690 | .procname = "exception-trace", |
| 691 | .data = &exception_trace, |
| 692 | .maxlen = sizeof(int), |
| 693 | .mode = 0644, |
| 694 | .proc_handler = proc_dointvec |
| 695 | }, |
| 696 | {} |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | }; |
| 698 | |
| 699 | static ctl_table debug_root_table2[] = { |
Eric W. Biederman | c37ce03 | 2007-02-14 00:33:51 -0800 | [diff] [blame] | 700 | { |
| 701 | .ctl_name = CTL_DEBUG, |
| 702 | .procname = "debug", |
| 703 | .mode = 0555, |
| 704 | .child = debug_table2 |
| 705 | }, |
| 706 | {} |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 707 | }; |
| 708 | |
| 709 | static __init int x8664_sysctl_init(void) |
| 710 | { |
Eric W. Biederman | 0b4d414 | 2007-02-14 00:34:09 -0800 | [diff] [blame] | 711 | register_sysctl_table(debug_root_table2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | return 0; |
| 713 | } |
| 714 | __initcall(x8664_sysctl_init); |
| 715 | #endif |
| 716 | |
Ernie Petrides | 103efcd | 2006-12-07 02:14:09 +0100 | [diff] [blame] | 717 | /* A pseudo VMA to allow ptrace access for the vsyscall page. This only |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 718 | covers the 64bit vsyscall page now. 32bit has a real VMA now and does |
| 719 | not need special handling anymore. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 720 | |
| 721 | static struct vm_area_struct gate_vma = { |
| 722 | .vm_start = VSYSCALL_START, |
Ernie Petrides | 103efcd | 2006-12-07 02:14:09 +0100 | [diff] [blame] | 723 | .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES << PAGE_SHIFT), |
| 724 | .vm_page_prot = PAGE_READONLY_EXEC, |
| 725 | .vm_flags = VM_READ | VM_EXEC |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | }; |
| 727 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
| 729 | { |
| 730 | #ifdef CONFIG_IA32_EMULATION |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 731 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
| 732 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | #endif |
| 734 | return &gate_vma; |
| 735 | } |
| 736 | |
| 737 | int in_gate_area(struct task_struct *task, unsigned long addr) |
| 738 | { |
| 739 | struct vm_area_struct *vma = get_gate_vma(task); |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 740 | if (!vma) |
| 741 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
| 743 | } |
| 744 | |
| 745 | /* Use this when you have no reliable task/vma, typically from interrupt |
| 746 | * context. It is less reliable than using the task's vma and may give |
| 747 | * false positives. |
| 748 | */ |
| 749 | int in_gate_area_no_task(unsigned long addr) |
| 750 | { |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 751 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | } |