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