arch/x86/mm/init.c

   1 #include <linux/gfp.h>
   2 #include <linux/initrd.h>
   3 #include <linux/ioport.h>
   4 #include <linux/swap.h>
   5 #include <linux/memblock.h>
   6 #include <linux/bootmem.h>      /* for max_low_pfn */
   7
   8 #include <asm/cacheflush.h>
   9 #include <asm/e820.h>
  10 #include <asm/init.h>
  11 #include <asm/page.h>
  12 #include <asm/page_types.h>
  13 #include <asm/sections.h>
  14 #include <asm/setup.h>
  15 #include <asm/tlbflush.h>
  16 #include <asm/tlb.h>
  17 #include <asm/proto.h>
  18 #include <asm/dma.h>            /* for MAX_DMA_PFN */
  19
  20 unsigned long __initdata pgt_buf_start;
  21 unsigned long __meminitdata pgt_buf_end;
  22 unsigned long __meminitdata pgt_buf_top;
  23
  24 int after_bootmem;
  25
  26 int direct_gbpages
  27 #ifdef CONFIG_DIRECT_GBPAGES
  28                                 = 1
  29 #endif
  30 ;
  31
  32 struct map_range {
  33         unsigned long start;
  34         unsigned long end;
  35         unsigned page_size_mask;
  36 };
  37
  38 /*
  39  * First calculate space needed for kernel direct mapping page tables to cover
  40  * mr[0].start to mr[nr_range - 1].end, while accounting for possible 2M and 1GB
  41  * pages. Then find enough contiguous space for those page tables.
  42  */
  43 static void __init find_early_table_space(struct map_range *mr, int nr_range)
  44 {
  45         int i;
  46         unsigned long puds = 0, pmds = 0, ptes = 0, tables;
  47         unsigned long start = 0, good_end;
  48         phys_addr_t base;
  49
  50         for (i = 0; i < nr_range; i++) {
  51                 unsigned long range, extra;
  52
  53                 range = mr[i].end - mr[i].start;
  54                 puds += (range + PUD_SIZE - 1) >> PUD_SHIFT;
  55
  56                 if (mr[i].page_size_mask & (1 << PG_LEVEL_1G)) {
  57                         extra = range - ((range >> PUD_SHIFT) << PUD_SHIFT);
  58                         pmds += (extra + PMD_SIZE - 1) >> PMD_SHIFT;
  59                 } else {
  60                         pmds += (range + PMD_SIZE - 1) >> PMD_SHIFT;
  61                 }
  62
  63                 if (mr[i].page_size_mask & (1 << PG_LEVEL_2M)) {
  64                         extra = range - ((range >> PMD_SHIFT) << PMD_SHIFT);
  65 #ifdef CONFIG_X86_32
  66                         extra += PMD_SIZE;
  67 #endif
  68                         ptes += (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
  69                 } else {
  70                         ptes += (range + PAGE_SIZE - 1) >> PAGE_SHIFT;
  71                 }
  72         }
  73
  74         tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
  75         tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
  76         tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
  77
  78 #ifdef CONFIG_X86_32
  79         /* for fixmap */
  80         tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
  81 #endif
  82         good_end = max_pfn_mapped << PAGE_SHIFT;
  83
  84         base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
  85         if (!base)
  86                 panic("Cannot find space for the kernel page tables");
  87
  88         pgt_buf_start = base >> PAGE_SHIFT;
  89         pgt_buf_end = pgt_buf_start;
  90         pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
  91
  92         printk(KERN_DEBUG "kernel direct mapping tables up to %#lx @ [mem %#010lx-%#010lx]\n",
  93                 mr[nr_range - 1].end - 1, pgt_buf_start << PAGE_SHIFT,
  94                 (pgt_buf_top << PAGE_SHIFT) - 1);
  95 }
  96
  97 void __init native_pagetable_reserve(u64 start, u64 end)
  98 {
  99         memblock_reserve(start, end - start);
 100 }
 101
 102 #ifdef CONFIG_X86_32
 103 #define NR_RANGE_MR 3
 104 #else /* CONFIG_X86_64 */
 105 #define NR_RANGE_MR 5
 106 #endif
 107
 108 static int __meminit save_mr(struct map_range *mr, int nr_range,
 109                              unsigned long start_pfn, unsigned long end_pfn,
 110                              unsigned long page_size_mask)
 111 {
 112         if (start_pfn < end_pfn) {
 113                 if (nr_range >= NR_RANGE_MR)
 114                         panic("run out of range for init_memory_mapping\n");
 115                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
 116                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
 117                 mr[nr_range].page_size_mask = page_size_mask;
 118                 nr_range++;
 119         }
 120
 121         return nr_range;
 122 }
 123
 124 /*
 125  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 126  * This runs before bootmem is initialized and gets pages directly from
 127  * the physical memory. To access them they are temporarily mapped.
 128  */
 129 unsigned long __init_refok init_memory_mapping(unsigned long start,
 130                                                unsigned long end)
 131 {
 132         unsigned long page_size_mask = 0;
 133         unsigned long start_pfn, end_pfn;
 134         unsigned long ret = 0;
 135         unsigned long pos;
 136
 137         struct map_range mr[NR_RANGE_MR];
 138         int nr_range, i;
 139         int use_pse, use_gbpages;
 140
 141         printk(KERN_INFO "init_memory_mapping: [mem %#010lx-%#010lx]\n",
 142                start, end - 1);
 143
 144 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
 145         /*
 146          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
 147          * This will simplify cpa(), which otherwise needs to support splitting
 148          * large pages into small in interrupt context, etc.
 149          */
 150         use_pse = use_gbpages = 0;
 151 #else
 152         use_pse = cpu_has_pse;
 153         use_gbpages = direct_gbpages;
 154 #endif
 155
 156         /* Enable PSE if available */
 157         if (cpu_has_pse)
 158                 set_in_cr4(X86_CR4_PSE);
 159
 160         /* Enable PGE if available */
 161         if (cpu_has_pge) {
 162                 set_in_cr4(X86_CR4_PGE);
 163                 __supported_pte_mask |= _PAGE_GLOBAL;
 164         }
 165
 166         if (use_gbpages)
 167                 page_size_mask |= 1 << PG_LEVEL_1G;
 168         if (use_pse)
 169                 page_size_mask |= 1 << PG_LEVEL_2M;
 170
 171         memset(mr, 0, sizeof(mr));
 172         nr_range = 0;
 173
 174         /* head if not big page alignment ? */
 175         start_pfn = start >> PAGE_SHIFT;
 176         pos = start_pfn << PAGE_SHIFT;
 177 #ifdef CONFIG_X86_32
 178         /*
 179          * Don't use a large page for the first 2/4MB of memory
 180          * because there are often fixed size MTRRs in there
 181          * and overlapping MTRRs into large pages can cause
 182          * slowdowns.
 183          */
 184         if (pos == 0)
 185                 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
 186         else
 187                 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 188                                  << (PMD_SHIFT - PAGE_SHIFT);
 189 #else /* CONFIG_X86_64 */
 190         end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
 191                         << (PMD_SHIFT - PAGE_SHIFT);
 192 #endif
 193         if (end_pfn > (end >> PAGE_SHIFT))
 194                 end_pfn = end >> PAGE_SHIFT;
 195         if (start_pfn < end_pfn) {
 196                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 197                 pos = end_pfn << PAGE_SHIFT;
 198         }
 199
 200         /* big page (2M) range */
 201         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 202                          << (PMD_SHIFT - PAGE_SHIFT);
 203 #ifdef CONFIG_X86_32
 204         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 205 #else /* CONFIG_X86_64 */
 206         end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
 207                          << (PUD_SHIFT - PAGE_SHIFT);
 208         if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
 209                 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
 210 #endif
 211
 212         if (start_pfn < end_pfn) {
 213                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 214                                 page_size_mask & (1<<PG_LEVEL_2M));
 215                 pos = end_pfn << PAGE_SHIFT;
 216         }
 217
 218 #ifdef CONFIG_X86_64
 219         /* big page (1G) range */
 220         start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
 221                          << (PUD_SHIFT - PAGE_SHIFT);
 222         end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
 223         if (start_pfn < end_pfn) {
 224                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 225                                 page_size_mask &
 226                                  ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
 227                 pos = end_pfn << PAGE_SHIFT;
 228         }
 229
 230         /* tail is not big page (1G) alignment */
 231         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 232                          << (PMD_SHIFT - PAGE_SHIFT);
 233         end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 234         if (start_pfn < end_pfn) {
 235                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 236                                 page_size_mask & (1<<PG_LEVEL_2M));
 237                 pos = end_pfn << PAGE_SHIFT;
 238         }
 239 #endif
 240
 241         /* tail is not big page (2M) alignment */
 242         start_pfn = pos>>PAGE_SHIFT;
 243         end_pfn = end>>PAGE_SHIFT;
 244         nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 245
 246         /* try to merge same page size and continuous */
 247         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
 248                 unsigned long old_start;
 249                 if (mr[i].end != mr[i+1].start ||
 250                     mr[i].page_size_mask != mr[i+1].page_size_mask)
 251                         continue;
 252                 /* move it */
 253                 old_start = mr[i].start;
 254                 memmove(&mr[i], &mr[i+1],
 255                         (nr_range - 1 - i) * sizeof(struct map_range));
 256                 mr[i--].start = old_start;
 257                 nr_range--;
 258         }
 259
 260         for (i = 0; i < nr_range; i++)
 261                 printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n",
 262                                 mr[i].start, mr[i].end - 1,
 263                         (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
 264                          (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
 265
 266         /*
 267          * Find space for the kernel direct mapping tables.
 268          *
 269          * Later we should allocate these tables in the local node of the
 270          * memory mapped. Unfortunately this is done currently before the
 271          * nodes are discovered.
 272          */
 273         if (!after_bootmem)
 274                 find_early_table_space(mr, nr_range);
 275
 276         for (i = 0; i < nr_range; i++)
 277                 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
 278                                                    mr[i].page_size_mask);
 279
 280 #ifdef CONFIG_X86_32
 281         early_ioremap_page_table_range_init();
 282
 283         load_cr3(swapper_pg_dir);
 284 #endif
 285
 286         __flush_tlb_all();
 287
 288         /*
 289          * Reserve the kernel pagetable pages we used (pgt_buf_start -
 290          * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
 291          * so that they can be reused for other purposes.
 292          *
 293          * On native it just means calling memblock_reserve, on Xen it also
 294          * means marking RW the pagetable pages that we allocated before
 295          * but that haven't been used.
 296          *
 297          * In fact on xen we mark RO the whole range pgt_buf_start -
 298          * pgt_buf_top, because we have to make sure that when
 299          * init_memory_mapping reaches the pagetable pages area, it maps
 300          * RO all the pagetable pages, including the ones that are beyond
 301          * pgt_buf_end at that time.
 302          */
 303         if (!after_bootmem && pgt_buf_end > pgt_buf_start)
 304                 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
 305                                 PFN_PHYS(pgt_buf_end));
 306
 307         if (!after_bootmem)
 308                 early_memtest(start, end);
 309
 310         return ret >> PAGE_SHIFT;
 311 }
 312
 313
 314 /*
 315  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 316  * is valid. The argument is a physical page number.
 317  *
 318  *
 319  * On x86, access has to be given to the first megabyte of ram because that area
 320  * contains bios code and data regions used by X and dosemu and similar apps.
 321  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 322  * mmio resources as well as potential bios/acpi data regions.
 323  */
 324 int devmem_is_allowed(unsigned long pagenr)
 325 {
 326         if (pagenr < 256)
 327                 return 1;
 328         if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
 329                 return 0;
 330         if (!page_is_ram(pagenr))
 331                 return 1;
 332         return 0;
 333 }
 334
 335 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 336 {
 337         unsigned long addr;
 338         unsigned long begin_aligned, end_aligned;
 339
 340         /* Make sure boundaries are page aligned */
 341         begin_aligned = PAGE_ALIGN(begin);
 342         end_aligned   = end & PAGE_MASK;
 343
 344         if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
 345                 begin = begin_aligned;
 346                 end   = end_aligned;
 347         }
 348
 349         if (begin >= end)
 350                 return;
 351
 352         addr = begin;
 353
 354         /*
 355          * If debugging page accesses then do not free this memory but
 356          * mark them not present - any buggy init-section access will
 357          * create a kernel page fault:
 358          */
 359 #ifdef CONFIG_DEBUG_PAGEALLOC
 360         printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
 361                 begin, end - 1);
 362         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 363 #else
 364         /*
 365          * We just marked the kernel text read only above, now that
 366          * we are going to free part of that, we need to make that
 367          * writeable and non-executable first.
 368          */
 369         set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
 370         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
 371
 372         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 373
 374         for (; addr < end; addr += PAGE_SIZE) {
 375                 ClearPageReserved(virt_to_page(addr));
 376                 init_page_count(virt_to_page(addr));
 377                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 378                 free_page(addr);
 379                 totalram_pages++;
 380         }
 381 #endif
 382 }
 383
 384 void free_initmem(void)
 385 {
 386         free_init_pages("unused kernel memory",
 387                         (unsigned long)(&__init_begin),
 388                         (unsigned long)(&__init_end));
 389 }
 390
 391 #ifdef CONFIG_BLK_DEV_INITRD
 392 void __init free_initrd_mem(unsigned long start, unsigned long end)
 393 {
 394         /*
 395          * end could be not aligned, and We can not align that,
 396          * decompresser could be confused by aligned initrd_end
 397          * We already reserve the end partial page before in
 398          *   - i386_start_kernel()
 399          *   - x86_64_start_kernel()
 400          *   - relocate_initrd()
 401          * So here We can do PAGE_ALIGN() safely to get partial page to be freed
 402          */
 403         free_init_pages("initrd memory", start, PAGE_ALIGN(end));
 404 }
 405 #endif
 406
 407 void __init zone_sizes_init(void)
 408 {
 409         unsigned long max_zone_pfns[MAX_NR_ZONES];
 410
 411         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 412
 413 #ifdef CONFIG_ZONE_DMA
 414         max_zone_pfns[ZONE_DMA]         = MAX_DMA_PFN;
 415 #endif
 416 #ifdef CONFIG_ZONE_DMA32
 417         max_zone_pfns[ZONE_DMA32]       = MAX_DMA32_PFN;
 418 #endif
 419         max_zone_pfns[ZONE_NORMAL]      = max_low_pfn;
 420 #ifdef CONFIG_HIGHMEM
 421         max_zone_pfns[ZONE_HIGHMEM]     = max_pfn;
 422 #endif
 423
 424         free_area_init_nodes(max_zone_pfns);
 425 }
 426