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 | |
| 9 | #include <linux/config.h> |
| 10 | #include <linux/signal.h> |
| 11 | #include <linux/sched.h> |
| 12 | #include <linux/kernel.h> |
| 13 | #include <linux/errno.h> |
| 14 | #include <linux/string.h> |
| 15 | #include <linux/types.h> |
| 16 | #include <linux/ptrace.h> |
| 17 | #include <linux/mman.h> |
| 18 | #include <linux/mm.h> |
| 19 | #include <linux/swap.h> |
| 20 | #include <linux/smp.h> |
| 21 | #include <linux/init.h> |
| 22 | #include <linux/pagemap.h> |
| 23 | #include <linux/bootmem.h> |
| 24 | #include <linux/proc_fs.h> |
Andi Kleen | 5917089 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 25 | #include <linux/pci.h> |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 26 | #include <linux/dma-mapping.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | |
| 28 | #include <asm/processor.h> |
| 29 | #include <asm/system.h> |
| 30 | #include <asm/uaccess.h> |
| 31 | #include <asm/pgtable.h> |
| 32 | #include <asm/pgalloc.h> |
| 33 | #include <asm/dma.h> |
| 34 | #include <asm/fixmap.h> |
| 35 | #include <asm/e820.h> |
| 36 | #include <asm/apic.h> |
| 37 | #include <asm/tlb.h> |
| 38 | #include <asm/mmu_context.h> |
| 39 | #include <asm/proto.h> |
| 40 | #include <asm/smp.h> |
Andi Kleen | 2bc0414 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 41 | #include <asm/sections.h> |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 42 | #include <asm/dma-mapping.h> |
| 43 | #include <asm/swiotlb.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | |
| 45 | #ifndef Dprintk |
| 46 | #define Dprintk(x...) |
| 47 | #endif |
| 48 | |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 49 | struct dma_mapping_ops* dma_ops; |
| 50 | EXPORT_SYMBOL(dma_ops); |
| 51 | |
Andi Kleen | e18c687 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 52 | static unsigned long dma_reserve __initdata; |
| 53 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
| 55 | |
| 56 | /* |
| 57 | * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the |
| 58 | * physical space so we can cache the place of the first one and move |
| 59 | * around without checking the pgd every time. |
| 60 | */ |
| 61 | |
| 62 | void show_mem(void) |
| 63 | { |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 64 | long i, total = 0, reserved = 0; |
| 65 | long shared = 0, cached = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | pg_data_t *pgdat; |
| 67 | struct page *page; |
| 68 | |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 69 | printk(KERN_INFO "Mem-info:\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | show_free_areas(); |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 71 | 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] | 72 | |
| 73 | for_each_pgdat(pgdat) { |
| 74 | for (i = 0; i < pgdat->node_spanned_pages; ++i) { |
| 75 | page = pfn_to_page(pgdat->node_start_pfn + i); |
| 76 | total++; |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 77 | if (PageReserved(page)) |
| 78 | reserved++; |
| 79 | else if (PageSwapCache(page)) |
| 80 | cached++; |
| 81 | else if (page_count(page)) |
| 82 | shared += page_count(page) - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | } |
| 84 | } |
Andi Kleen | e92343c | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 85 | printk(KERN_INFO "%lu pages of RAM\n", total); |
| 86 | printk(KERN_INFO "%lu reserved pages\n",reserved); |
| 87 | printk(KERN_INFO "%lu pages shared\n",shared); |
| 88 | printk(KERN_INFO "%lu pages swap cached\n",cached); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | } |
| 90 | |
| 91 | /* References to section boundaries */ |
| 92 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | int after_bootmem; |
| 94 | |
| 95 | static void *spp_getpage(void) |
| 96 | { |
| 97 | void *ptr; |
| 98 | if (after_bootmem) |
| 99 | ptr = (void *) get_zeroed_page(GFP_ATOMIC); |
| 100 | else |
| 101 | ptr = alloc_bootmem_pages(PAGE_SIZE); |
| 102 | if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) |
| 103 | panic("set_pte_phys: cannot allocate page data %s\n", after_bootmem?"after bootmem":""); |
| 104 | |
| 105 | Dprintk("spp_getpage %p\n", ptr); |
| 106 | return ptr; |
| 107 | } |
| 108 | |
| 109 | static void set_pte_phys(unsigned long vaddr, |
| 110 | unsigned long phys, pgprot_t prot) |
| 111 | { |
| 112 | pgd_t *pgd; |
| 113 | pud_t *pud; |
| 114 | pmd_t *pmd; |
| 115 | pte_t *pte, new_pte; |
| 116 | |
| 117 | Dprintk("set_pte_phys %lx to %lx\n", vaddr, phys); |
| 118 | |
| 119 | pgd = pgd_offset_k(vaddr); |
| 120 | if (pgd_none(*pgd)) { |
| 121 | printk("PGD FIXMAP MISSING, it should be setup in head.S!\n"); |
| 122 | return; |
| 123 | } |
| 124 | pud = pud_offset(pgd, vaddr); |
| 125 | if (pud_none(*pud)) { |
| 126 | pmd = (pmd_t *) spp_getpage(); |
| 127 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER)); |
| 128 | if (pmd != pmd_offset(pud, 0)) { |
| 129 | printk("PAGETABLE BUG #01! %p <-> %p\n", pmd, pmd_offset(pud,0)); |
| 130 | return; |
| 131 | } |
| 132 | } |
| 133 | pmd = pmd_offset(pud, vaddr); |
| 134 | if (pmd_none(*pmd)) { |
| 135 | pte = (pte_t *) spp_getpage(); |
| 136 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER)); |
| 137 | if (pte != pte_offset_kernel(pmd, 0)) { |
| 138 | printk("PAGETABLE BUG #02!\n"); |
| 139 | return; |
| 140 | } |
| 141 | } |
| 142 | new_pte = pfn_pte(phys >> PAGE_SHIFT, prot); |
| 143 | |
| 144 | pte = pte_offset_kernel(pmd, vaddr); |
| 145 | if (!pte_none(*pte) && |
| 146 | pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) |
| 147 | pte_ERROR(*pte); |
| 148 | set_pte(pte, new_pte); |
| 149 | |
| 150 | /* |
| 151 | * It's enough to flush this one mapping. |
| 152 | * (PGE mappings get flushed as well) |
| 153 | */ |
| 154 | __flush_tlb_one(vaddr); |
| 155 | } |
| 156 | |
| 157 | /* NOTE: this is meant to be run only at boot */ |
| 158 | void __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t prot) |
| 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 | |
| 171 | extern pmd_t temp_boot_pmds[]; |
| 172 | |
| 173 | static struct temp_map { |
| 174 | pmd_t *pmd; |
| 175 | void *address; |
| 176 | int allocated; |
| 177 | } temp_mappings[] __initdata = { |
| 178 | { &temp_boot_pmds[0], (void *)(40UL * 1024 * 1024) }, |
| 179 | { &temp_boot_pmds[1], (void *)(42UL * 1024 * 1024) }, |
| 180 | {} |
| 181 | }; |
| 182 | |
| 183 | static __init void *alloc_low_page(int *index, unsigned long *phys) |
| 184 | { |
| 185 | struct temp_map *ti; |
| 186 | int i; |
| 187 | unsigned long pfn = table_end++, paddr; |
| 188 | void *adr; |
| 189 | |
| 190 | if (pfn >= end_pfn) |
| 191 | panic("alloc_low_page: ran out of memory"); |
| 192 | for (i = 0; temp_mappings[i].allocated; i++) { |
| 193 | if (!temp_mappings[i].pmd) |
| 194 | panic("alloc_low_page: ran out of temp mappings"); |
| 195 | } |
| 196 | ti = &temp_mappings[i]; |
| 197 | paddr = (pfn << PAGE_SHIFT) & PMD_MASK; |
| 198 | set_pmd(ti->pmd, __pmd(paddr | _KERNPG_TABLE | _PAGE_PSE)); |
| 199 | ti->allocated = 1; |
| 200 | __flush_tlb(); |
| 201 | adr = ti->address + ((pfn << PAGE_SHIFT) & ~PMD_MASK); |
| 202 | *index = i; |
| 203 | *phys = pfn * PAGE_SIZE; |
| 204 | return adr; |
| 205 | } |
| 206 | |
| 207 | static __init void unmap_low_page(int i) |
| 208 | { |
| 209 | struct temp_map *ti = &temp_mappings[i]; |
| 210 | set_pmd(ti->pmd, __pmd(0)); |
| 211 | ti->allocated = 0; |
| 212 | } |
| 213 | |
| 214 | static void __init phys_pud_init(pud_t *pud, unsigned long address, unsigned long end) |
| 215 | { |
| 216 | long i, j; |
| 217 | |
| 218 | i = pud_index(address); |
| 219 | pud = pud + i; |
| 220 | for (; i < PTRS_PER_PUD; pud++, i++) { |
| 221 | int map; |
| 222 | unsigned long paddr, pmd_phys; |
| 223 | pmd_t *pmd; |
| 224 | |
| 225 | paddr = address + i*PUD_SIZE; |
| 226 | if (paddr >= end) { |
| 227 | for (; i < PTRS_PER_PUD; i++, pud++) |
| 228 | set_pud(pud, __pud(0)); |
| 229 | break; |
| 230 | } |
| 231 | |
| 232 | if (!e820_mapped(paddr, paddr+PUD_SIZE, 0)) { |
| 233 | set_pud(pud, __pud(0)); |
| 234 | continue; |
| 235 | } |
| 236 | |
| 237 | pmd = alloc_low_page(&map, &pmd_phys); |
| 238 | set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE)); |
| 239 | for (j = 0; j < PTRS_PER_PMD; pmd++, j++, paddr += PMD_SIZE) { |
| 240 | unsigned long pe; |
| 241 | |
| 242 | if (paddr >= end) { |
| 243 | for (; j < PTRS_PER_PMD; j++, pmd++) |
| 244 | set_pmd(pmd, __pmd(0)); |
| 245 | break; |
| 246 | } |
| 247 | pe = _PAGE_NX|_PAGE_PSE | _KERNPG_TABLE | _PAGE_GLOBAL | paddr; |
| 248 | pe &= __supported_pte_mask; |
| 249 | set_pmd(pmd, __pmd(pe)); |
| 250 | } |
| 251 | unmap_low_page(map); |
| 252 | } |
| 253 | __flush_tlb(); |
| 254 | } |
| 255 | |
| 256 | static void __init find_early_table_space(unsigned long end) |
| 257 | { |
Andi Kleen | 6c5acd1 | 2006-01-11 22:46:57 +0100 | [diff] [blame] | 258 | unsigned long puds, pmds, tables, start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | |
| 260 | puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; |
| 261 | pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; |
| 262 | tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) + |
| 263 | round_up(pmds * sizeof(pmd_t), PAGE_SIZE); |
| 264 | |
Andi Kleen | ee408c7 | 2006-01-16 01:56:51 +0100 | [diff] [blame^] | 265 | /* RED-PEN putting page tables only on node 0 could |
| 266 | cause a hotspot and fill up ZONE_DMA. The page tables |
| 267 | need roughly 0.5KB per GB. */ |
| 268 | start = 0x8000; |
| 269 | table_start = find_e820_area(start, end, tables); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | if (table_start == -1UL) |
| 271 | panic("Cannot find space for the kernel page tables"); |
| 272 | |
| 273 | table_start >>= PAGE_SHIFT; |
| 274 | table_end = table_start; |
| 275 | } |
| 276 | |
| 277 | /* Setup the direct mapping of the physical memory at PAGE_OFFSET. |
| 278 | This runs before bootmem is initialized and gets pages directly from the |
| 279 | physical memory. To access them they are temporarily mapped. */ |
| 280 | void __init init_memory_mapping(unsigned long start, unsigned long end) |
| 281 | { |
| 282 | unsigned long next; |
| 283 | |
| 284 | Dprintk("init_memory_mapping\n"); |
| 285 | |
| 286 | /* |
| 287 | * Find space for the kernel direct mapping tables. |
| 288 | * Later we should allocate these tables in the local node of the memory |
| 289 | * mapped. Unfortunately this is done currently before the nodes are |
| 290 | * discovered. |
| 291 | */ |
| 292 | find_early_table_space(end); |
| 293 | |
| 294 | start = (unsigned long)__va(start); |
| 295 | end = (unsigned long)__va(end); |
| 296 | |
| 297 | for (; start < end; start = next) { |
| 298 | int map; |
| 299 | unsigned long pud_phys; |
| 300 | pud_t *pud = alloc_low_page(&map, &pud_phys); |
| 301 | next = start + PGDIR_SIZE; |
| 302 | if (next > end) |
| 303 | next = end; |
| 304 | phys_pud_init(pud, __pa(start), __pa(next)); |
| 305 | set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys)); |
| 306 | unmap_low_page(map); |
| 307 | } |
| 308 | |
| 309 | asm volatile("movq %%cr4,%0" : "=r" (mmu_cr4_features)); |
| 310 | __flush_tlb_all(); |
| 311 | early_printk("kernel direct mapping tables upto %lx @ %lx-%lx\n", end, |
| 312 | table_start<<PAGE_SHIFT, |
| 313 | table_end<<PAGE_SHIFT); |
| 314 | } |
| 315 | |
Siddha, Suresh B | f6c2e33 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 316 | void __cpuinit zap_low_mappings(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | { |
Siddha, Suresh B | f6c2e33 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 318 | if (cpu == 0) { |
| 319 | pgd_t *pgd = pgd_offset_k(0UL); |
| 320 | pgd_clear(pgd); |
| 321 | } else { |
| 322 | /* |
| 323 | * For AP's, zap the low identity mappings by changing the cr3 |
| 324 | * to init_level4_pgt and doing local flush tlb all |
| 325 | */ |
| 326 | asm volatile("movq %0,%%cr3" :: "r" (__pa_symbol(&init_level4_pgt))); |
| 327 | } |
| 328 | __flush_tlb_all(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | } |
| 330 | |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 331 | /* Compute zone sizes for the DMA and DMA32 zones in a node. */ |
| 332 | __init void |
| 333 | size_zones(unsigned long *z, unsigned long *h, |
| 334 | unsigned long start_pfn, unsigned long end_pfn) |
| 335 | { |
| 336 | int i; |
| 337 | unsigned long w; |
| 338 | |
| 339 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 340 | z[i] = 0; |
| 341 | |
| 342 | if (start_pfn < MAX_DMA_PFN) |
| 343 | z[ZONE_DMA] = MAX_DMA_PFN - start_pfn; |
| 344 | if (start_pfn < MAX_DMA32_PFN) { |
| 345 | unsigned long dma32_pfn = MAX_DMA32_PFN; |
| 346 | if (dma32_pfn > end_pfn) |
| 347 | dma32_pfn = end_pfn; |
| 348 | z[ZONE_DMA32] = dma32_pfn - start_pfn; |
| 349 | } |
| 350 | z[ZONE_NORMAL] = end_pfn - start_pfn; |
| 351 | |
| 352 | /* Remove lower zones from higher ones. */ |
| 353 | w = 0; |
| 354 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 355 | if (z[i]) |
| 356 | z[i] -= w; |
| 357 | w += z[i]; |
| 358 | } |
| 359 | |
| 360 | /* Compute holes */ |
Ravikiran G Thirumalai | 576fc09 | 2005-12-29 13:06:11 +0100 | [diff] [blame] | 361 | w = start_pfn; |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 362 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 363 | unsigned long s = w; |
| 364 | w += z[i]; |
| 365 | h[i] = e820_hole_size(s, w); |
| 366 | } |
Andi Kleen | e18c687 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 367 | |
| 368 | /* Add the space pace needed for mem_map to the holes too. */ |
| 369 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 370 | h[i] += (z[i] * sizeof(struct page)) / PAGE_SIZE; |
| 371 | |
| 372 | /* The 16MB DMA zone has the kernel and other misc mappings. |
| 373 | Account them too */ |
| 374 | if (h[ZONE_DMA]) { |
| 375 | h[ZONE_DMA] += dma_reserve; |
| 376 | if (h[ZONE_DMA] >= z[ZONE_DMA]) { |
| 377 | printk(KERN_WARNING |
| 378 | "Kernel too large and filling up ZONE_DMA?\n"); |
| 379 | h[ZONE_DMA] = z[ZONE_DMA]; |
| 380 | } |
| 381 | } |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 382 | } |
| 383 | |
Matt Tolentino | 2b97690 | 2005-06-23 00:08:06 -0700 | [diff] [blame] | 384 | #ifndef CONFIG_NUMA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | void __init paging_init(void) |
| 386 | { |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 387 | unsigned long zones[MAX_NR_ZONES], holes[MAX_NR_ZONES]; |
| 388 | size_zones(zones, holes, 0, end_pfn); |
| 389 | free_area_init_node(0, NODE_DATA(0), zones, |
| 390 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, holes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | } |
| 392 | #endif |
| 393 | |
| 394 | /* Unmap a kernel mapping if it exists. This is useful to avoid prefetches |
| 395 | from the CPU leading to inconsistent cache lines. address and size |
| 396 | must be aligned to 2MB boundaries. |
| 397 | Does nothing when the mapping doesn't exist. */ |
| 398 | void __init clear_kernel_mapping(unsigned long address, unsigned long size) |
| 399 | { |
| 400 | unsigned long end = address + size; |
| 401 | |
| 402 | BUG_ON(address & ~LARGE_PAGE_MASK); |
| 403 | BUG_ON(size & ~LARGE_PAGE_MASK); |
| 404 | |
| 405 | for (; address < end; address += LARGE_PAGE_SIZE) { |
| 406 | pgd_t *pgd = pgd_offset_k(address); |
| 407 | pud_t *pud; |
| 408 | pmd_t *pmd; |
| 409 | if (pgd_none(*pgd)) |
| 410 | continue; |
| 411 | pud = pud_offset(pgd, address); |
| 412 | if (pud_none(*pud)) |
| 413 | continue; |
| 414 | pmd = pmd_offset(pud, address); |
| 415 | if (!pmd || pmd_none(*pmd)) |
| 416 | continue; |
| 417 | if (0 == (pmd_val(*pmd) & _PAGE_PSE)) { |
| 418 | /* Could handle this, but it should not happen currently. */ |
| 419 | printk(KERN_ERR |
| 420 | "clear_kernel_mapping: mapping has been split. will leak memory\n"); |
| 421 | pmd_ERROR(*pmd); |
| 422 | } |
| 423 | set_pmd(pmd, __pmd(0)); |
| 424 | } |
| 425 | __flush_tlb_all(); |
| 426 | } |
| 427 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, kcore_modules, |
| 429 | kcore_vsyscall; |
| 430 | |
| 431 | void __init mem_init(void) |
| 432 | { |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 433 | long codesize, reservedpages, datasize, initsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | |
| 435 | #ifdef CONFIG_SWIOTLB |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 436 | pci_swiotlb_init(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | #endif |
Muli Ben-Yehuda | 17a941d | 2006-01-11 22:44:42 +0100 | [diff] [blame] | 438 | no_iommu_init(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | |
| 440 | /* How many end-of-memory variables you have, grandma! */ |
| 441 | max_low_pfn = end_pfn; |
| 442 | max_pfn = end_pfn; |
| 443 | num_physpages = end_pfn; |
| 444 | high_memory = (void *) __va(end_pfn * PAGE_SIZE); |
| 445 | |
| 446 | /* clear the zero-page */ |
| 447 | memset(empty_zero_page, 0, PAGE_SIZE); |
| 448 | |
| 449 | reservedpages = 0; |
| 450 | |
| 451 | /* this will put all low memory onto the freelists */ |
Matt Tolentino | 2b97690 | 2005-06-23 00:08:06 -0700 | [diff] [blame] | 452 | #ifdef CONFIG_NUMA |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 453 | totalram_pages = numa_free_all_bootmem(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | #else |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 455 | totalram_pages = free_all_bootmem(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | #endif |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 457 | reservedpages = end_pfn - totalram_pages - e820_hole_size(0, end_pfn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | |
| 459 | after_bootmem = 1; |
| 460 | |
| 461 | codesize = (unsigned long) &_etext - (unsigned long) &_text; |
| 462 | datasize = (unsigned long) &_edata - (unsigned long) &_etext; |
| 463 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; |
| 464 | |
| 465 | /* Register memory areas for /proc/kcore */ |
| 466 | kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); |
| 467 | kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, |
| 468 | VMALLOC_END-VMALLOC_START); |
| 469 | kclist_add(&kcore_kernel, &_stext, _end - _stext); |
| 470 | kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); |
| 471 | kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, |
| 472 | VSYSCALL_END - VSYSCALL_START); |
| 473 | |
Andi Kleen | 0a43e4b | 2005-09-12 18:49:24 +0200 | [diff] [blame] | 474 | 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] | 475 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
| 476 | end_pfn << (PAGE_SHIFT-10), |
| 477 | codesize >> 10, |
| 478 | reservedpages << (PAGE_SHIFT-10), |
| 479 | datasize >> 10, |
| 480 | initsize >> 10); |
| 481 | |
Siddha, Suresh B | f6c2e33 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 482 | #ifdef CONFIG_SMP |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | /* |
Siddha, Suresh B | f6c2e33 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 484 | * Sync boot_level4_pgt mappings with the init_level4_pgt |
| 485 | * except for the low identity mappings which are already zapped |
| 486 | * in init_level4_pgt. This sync-up is essential for AP's bringup |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | */ |
Siddha, Suresh B | f6c2e33 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 488 | memcpy(boot_level4_pgt+1, init_level4_pgt+1, (PTRS_PER_PGD-1)*sizeof(pgd_t)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | #endif |
| 490 | } |
| 491 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | void free_initmem(void) |
| 493 | { |
| 494 | unsigned long addr; |
| 495 | |
| 496 | addr = (unsigned long)(&__init_begin); |
| 497 | for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) { |
| 498 | ClearPageReserved(virt_to_page(addr)); |
| 499 | set_page_count(virt_to_page(addr), 1); |
| 500 | memset((void *)(addr & ~(PAGE_SIZE-1)), 0xcc, PAGE_SIZE); |
| 501 | free_page(addr); |
| 502 | totalram_pages++; |
| 503 | } |
| 504 | memset(__initdata_begin, 0xba, __initdata_end - __initdata_begin); |
Andi Kleen | 2bc0414 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 505 | printk ("Freeing unused kernel memory: %luk freed\n", (__init_end - __init_begin) >> 10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 506 | } |
| 507 | |
Arjan van de Ven | 67df197 | 2006-01-06 00:12:04 -0800 | [diff] [blame] | 508 | #ifdef CONFIG_DEBUG_RODATA |
| 509 | |
| 510 | extern char __start_rodata, __end_rodata; |
| 511 | void mark_rodata_ro(void) |
| 512 | { |
| 513 | unsigned long addr = (unsigned long)&__start_rodata; |
| 514 | |
| 515 | for (; addr < (unsigned long)&__end_rodata; addr += PAGE_SIZE) |
| 516 | change_page_attr_addr(addr, 1, PAGE_KERNEL_RO); |
| 517 | |
| 518 | printk ("Write protecting the kernel read-only data: %luk\n", |
| 519 | (&__end_rodata - &__start_rodata) >> 10); |
| 520 | |
| 521 | /* |
| 522 | * change_page_attr_addr() requires a global_flush_tlb() call after it. |
| 523 | * We do this after the printk so that if something went wrong in the |
| 524 | * change, the printk gets out at least to give a better debug hint |
| 525 | * of who is the culprit. |
| 526 | */ |
| 527 | global_flush_tlb(); |
| 528 | } |
| 529 | #endif |
| 530 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | #ifdef CONFIG_BLK_DEV_INITRD |
| 532 | void free_initrd_mem(unsigned long start, unsigned long end) |
| 533 | { |
| 534 | if (start < (unsigned long)&_end) |
| 535 | return; |
| 536 | printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10); |
| 537 | for (; start < end; start += PAGE_SIZE) { |
| 538 | ClearPageReserved(virt_to_page(start)); |
| 539 | set_page_count(virt_to_page(start), 1); |
| 540 | free_page(start); |
| 541 | totalram_pages++; |
| 542 | } |
| 543 | } |
| 544 | #endif |
| 545 | |
| 546 | void __init reserve_bootmem_generic(unsigned long phys, unsigned len) |
| 547 | { |
| 548 | /* Should check here against the e820 map to avoid double free */ |
Matt Tolentino | 2b97690 | 2005-06-23 00:08:06 -0700 | [diff] [blame] | 549 | #ifdef CONFIG_NUMA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | int nid = phys_to_nid(phys); |
| 551 | reserve_bootmem_node(NODE_DATA(nid), phys, len); |
| 552 | #else |
| 553 | reserve_bootmem(phys, len); |
| 554 | #endif |
Andi Kleen | e18c687 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 555 | if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) |
| 556 | dma_reserve += len / PAGE_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | } |
| 558 | |
| 559 | int kern_addr_valid(unsigned long addr) |
| 560 | { |
| 561 | unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; |
| 562 | pgd_t *pgd; |
| 563 | pud_t *pud; |
| 564 | pmd_t *pmd; |
| 565 | pte_t *pte; |
| 566 | |
| 567 | if (above != 0 && above != -1UL) |
| 568 | return 0; |
| 569 | |
| 570 | pgd = pgd_offset_k(addr); |
| 571 | if (pgd_none(*pgd)) |
| 572 | return 0; |
| 573 | |
| 574 | pud = pud_offset(pgd, addr); |
| 575 | if (pud_none(*pud)) |
| 576 | return 0; |
| 577 | |
| 578 | pmd = pmd_offset(pud, addr); |
| 579 | if (pmd_none(*pmd)) |
| 580 | return 0; |
| 581 | if (pmd_large(*pmd)) |
| 582 | return pfn_valid(pmd_pfn(*pmd)); |
| 583 | |
| 584 | pte = pte_offset_kernel(pmd, addr); |
| 585 | if (pte_none(*pte)) |
| 586 | return 0; |
| 587 | return pfn_valid(pte_pfn(*pte)); |
| 588 | } |
| 589 | |
| 590 | #ifdef CONFIG_SYSCTL |
| 591 | #include <linux/sysctl.h> |
| 592 | |
| 593 | extern int exception_trace, page_fault_trace; |
| 594 | |
| 595 | static ctl_table debug_table2[] = { |
| 596 | { 99, "exception-trace", &exception_trace, sizeof(int), 0644, NULL, |
| 597 | proc_dointvec }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | { 0, } |
| 599 | }; |
| 600 | |
| 601 | static ctl_table debug_root_table2[] = { |
| 602 | { .ctl_name = CTL_DEBUG, .procname = "debug", .mode = 0555, |
| 603 | .child = debug_table2 }, |
| 604 | { 0 }, |
| 605 | }; |
| 606 | |
| 607 | static __init int x8664_sysctl_init(void) |
| 608 | { |
| 609 | register_sysctl_table(debug_root_table2, 1); |
| 610 | return 0; |
| 611 | } |
| 612 | __initcall(x8664_sysctl_init); |
| 613 | #endif |
| 614 | |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 615 | /* A pseudo VMAs to allow ptrace access for the vsyscall page. This only |
| 616 | covers the 64bit vsyscall page now. 32bit has a real VMA now and does |
| 617 | not need special handling anymore. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | |
| 619 | static struct vm_area_struct gate_vma = { |
| 620 | .vm_start = VSYSCALL_START, |
| 621 | .vm_end = VSYSCALL_END, |
| 622 | .vm_page_prot = PAGE_READONLY |
| 623 | }; |
| 624 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
| 626 | { |
| 627 | #ifdef CONFIG_IA32_EMULATION |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 628 | if (test_tsk_thread_flag(tsk, TIF_IA32)) |
| 629 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | #endif |
| 631 | return &gate_vma; |
| 632 | } |
| 633 | |
| 634 | int in_gate_area(struct task_struct *task, unsigned long addr) |
| 635 | { |
| 636 | struct vm_area_struct *vma = get_gate_vma(task); |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 637 | if (!vma) |
| 638 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | return (addr >= vma->vm_start) && (addr < vma->vm_end); |
| 640 | } |
| 641 | |
| 642 | /* Use this when you have no reliable task/vma, typically from interrupt |
| 643 | * context. It is less reliable than using the task's vma and may give |
| 644 | * false positives. |
| 645 | */ |
| 646 | int in_gate_area_no_task(unsigned long addr) |
| 647 | { |
Andi Kleen | 1e01441 | 2005-04-16 15:24:55 -0700 | [diff] [blame] | 648 | return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | } |