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 static void __init find_early_table_space(unsigned long end, int use_pse,
  33                                           int use_gbpages)
  34 {
  35         unsigned long puds, pmds, ptes, tables, start = 0, good_end = end;
  36         phys_addr_t base;
  37
  38         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
  39         tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
  40
  41         if (use_gbpages) {
  42                 unsigned long extra;
  43
  44                 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
  45                 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
  46         } else
  47                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
  48
  49         tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
  50
  51         if (use_pse) {
  52                 unsigned long extra;
  53
  54                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
  55 #ifdef CONFIG_X86_32
  56                 extra += PMD_SIZE;
  57 #endif
  58                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
  59         } else
  60                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
  61
  62         tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
  63
  64 #ifdef CONFIG_X86_32
  65         /* for fixmap */
  66         tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
  67 #endif
  68         good_end = max_pfn_mapped << PAGE_SHIFT;
  69
  70         base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
  71         if (!base)
  72                 panic("Cannot find space for the kernel page tables");
  73
  74         pgt_buf_start = base >> PAGE_SHIFT;
  75         pgt_buf_end = pgt_buf_start;
  76         pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
  77
  78         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
  79                 end, pgt_buf_start << PAGE_SHIFT, pgt_buf_top << PAGE_SHIFT);
  80 }
  81
  82 void __init native_pagetable_reserve(u64 start, u64 end)
  83 {
  84         memblock_reserve(start, end - start);
  85 }
  86
  87 struct map_range {
  88         unsigned long start;
  89         unsigned long end;
  90         unsigned page_size_mask;
  91 };
  92
  93 #ifdef CONFIG_X86_32
  94 #define NR_RANGE_MR 3
  95 #else /* CONFIG_X86_64 */
  96 #define NR_RANGE_MR 5
  97 #endif
  98
  99 static int __meminit save_mr(struct map_range *mr, int nr_range,
 100                              unsigned long start_pfn, unsigned long end_pfn,
 101                              unsigned long page_size_mask)
 102 {
 103         if (start_pfn < end_pfn) {
 104                 if (nr_range >= NR_RANGE_MR)
 105                         panic("run out of range for init_memory_mapping\n");
 106                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
 107                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
 108                 mr[nr_range].page_size_mask = page_size_mask;
 109                 nr_range++;
 110         }
 111
 112         return nr_range;
 113 }
 114
 115 /*
 116  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 117  * This runs before bootmem is initialized and gets pages directly from
 118  * the physical memory. To access them they are temporarily mapped.
 119  */
 120 unsigned long __init_refok init_memory_mapping(unsigned long start,
 121                                                unsigned long end)
 122 {
 123         unsigned long page_size_mask = 0;
 124         unsigned long start_pfn, end_pfn;
 125         unsigned long ret = 0;
 126         unsigned long pos;
 127
 128         struct map_range mr[NR_RANGE_MR];
 129         int nr_range, i;
 130         int use_pse, use_gbpages;
 131
 132         printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
 133
 134 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
 135         /*
 136          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
 137          * This will simplify cpa(), which otherwise needs to support splitting
 138          * large pages into small in interrupt context, etc.
 139          */
 140         use_pse = use_gbpages = 0;
 141 #else
 142         use_pse = cpu_has_pse;
 143         use_gbpages = direct_gbpages;
 144 #endif
 145
 146         /* Enable PSE if available */
 147         if (cpu_has_pse)
 148                 set_in_cr4(X86_CR4_PSE);
 149
 150         /* Enable PGE if available */
 151         if (cpu_has_pge) {
 152                 set_in_cr4(X86_CR4_PGE);
 153                 __supported_pte_mask |= _PAGE_GLOBAL;
 154         }
 155
 156         if (use_gbpages)
 157                 page_size_mask |= 1 << PG_LEVEL_1G;
 158         if (use_pse)
 159                 page_size_mask |= 1 << PG_LEVEL_2M;
 160
 161         memset(mr, 0, sizeof(mr));
 162         nr_range = 0;
 163
 164         /* head if not big page alignment ? */
 165         start_pfn = start >> PAGE_SHIFT;
 166         pos = start_pfn << PAGE_SHIFT;
 167 #ifdef CONFIG_X86_32
 168         /*
 169          * Don't use a large page for the first 2/4MB of memory
 170          * because there are often fixed size MTRRs in there
 171          * and overlapping MTRRs into large pages can cause
 172          * slowdowns.
 173          */
 174         if (pos == 0)
 175                 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
 176         else
 177                 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 178                                  << (PMD_SHIFT - PAGE_SHIFT);
 179 #else /* CONFIG_X86_64 */
 180         end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
 181                         << (PMD_SHIFT - PAGE_SHIFT);
 182 #endif
 183         if (end_pfn > (end >> PAGE_SHIFT))
 184                 end_pfn = end >> PAGE_SHIFT;
 185         if (start_pfn < end_pfn) {
 186                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 187                 pos = end_pfn << PAGE_SHIFT;
 188         }
 189
 190         /* big page (2M) range */
 191         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 192                          << (PMD_SHIFT - PAGE_SHIFT);
 193 #ifdef CONFIG_X86_32
 194         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 195 #else /* CONFIG_X86_64 */
 196         end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
 197                          << (PUD_SHIFT - PAGE_SHIFT);
 198         if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
 199                 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
 200 #endif
 201
 202         if (start_pfn < end_pfn) {
 203                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 204                                 page_size_mask & (1<<PG_LEVEL_2M));
 205                 pos = end_pfn << PAGE_SHIFT;
 206         }
 207
 208 #ifdef CONFIG_X86_64
 209         /* big page (1G) range */
 210         start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
 211                          << (PUD_SHIFT - PAGE_SHIFT);
 212         end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
 213         if (start_pfn < end_pfn) {
 214                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 215                                 page_size_mask &
 216                                  ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
 217                 pos = end_pfn << PAGE_SHIFT;
 218         }
 219
 220         /* tail is not big page (1G) alignment */
 221         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 222                          << (PMD_SHIFT - PAGE_SHIFT);
 223         end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 224         if (start_pfn < end_pfn) {
 225                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 226                                 page_size_mask & (1<<PG_LEVEL_2M));
 227                 pos = end_pfn << PAGE_SHIFT;
 228         }
 229 #endif
 230
 231         /* tail is not big page (2M) alignment */
 232         start_pfn = pos>>PAGE_SHIFT;
 233         end_pfn = end>>PAGE_SHIFT;
 234         nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 235
 236         /* try to merge same page size and continuous */
 237         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
 238                 unsigned long old_start;
 239                 if (mr[i].end != mr[i+1].start ||
 240                     mr[i].page_size_mask != mr[i+1].page_size_mask)
 241                         continue;
 242                 /* move it */
 243                 old_start = mr[i].start;
 244                 memmove(&mr[i], &mr[i+1],
 245                         (nr_range - 1 - i) * sizeof(struct map_range));
 246                 mr[i--].start = old_start;
 247                 nr_range--;
 248         }
 249
 250         for (i = 0; i < nr_range; i++)
 251                 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
 252                                 mr[i].start, mr[i].end,
 253                         (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
 254                          (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
 255
 256         /*
 257          * Find space for the kernel direct mapping tables.
 258          *
 259          * Later we should allocate these tables in the local node of the
 260          * memory mapped. Unfortunately this is done currently before the
 261          * nodes are discovered.
 262          */
 263         if (!after_bootmem)
 264                 find_early_table_space(end, use_pse, use_gbpages);
 265
 266         for (i = 0; i < nr_range; i++)
 267                 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
 268                                                    mr[i].page_size_mask);
 269
 270 #ifdef CONFIG_X86_32
 271         early_ioremap_page_table_range_init();
 272
 273         load_cr3(swapper_pg_dir);
 274 #endif
 275
 276         __flush_tlb_all();
 277
 278         /*
 279          * Reserve the kernel pagetable pages we used (pgt_buf_start -
 280          * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
 281          * so that they can be reused for other purposes.
 282          *
 283          * On native it just means calling memblock_reserve, on Xen it also
 284          * means marking RW the pagetable pages that we allocated before
 285          * but that haven't been used.
 286          *
 287          * In fact on xen we mark RO the whole range pgt_buf_start -
 288          * pgt_buf_top, because we have to make sure that when
 289          * init_memory_mapping reaches the pagetable pages area, it maps
 290          * RO all the pagetable pages, including the ones that are beyond
 291          * pgt_buf_end at that time.
 292          */
 293         if (!after_bootmem && pgt_buf_end > pgt_buf_start)
 294                 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
 295                                 PFN_PHYS(pgt_buf_end));
 296
 297         if (!after_bootmem)
 298                 early_memtest(start, end);
 299
 300         return ret >> PAGE_SHIFT;
 301 }
 302
 303
 304 /*
 305  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 306  * is valid. The argument is a physical page number.
 307  *
 308  *
 309  * On x86, access has to be given to the first megabyte of ram because that area
 310  * contains bios code and data regions used by X and dosemu and similar apps.
 311  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 312  * mmio resources as well as potential bios/acpi data regions.
 313  */
 314 int devmem_is_allowed(unsigned long pagenr)
 315 {
 316         if (pagenr <= 256)
 317                 return 1;
 318         if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
 319                 return 0;
 320         if (!page_is_ram(pagenr))
 321                 return 1;
 322         return 0;
 323 }
 324
 325 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 326 {
 327         unsigned long addr;
 328         unsigned long begin_aligned, end_aligned;
 329
 330         /* Make sure boundaries are page aligned */
 331         begin_aligned = PAGE_ALIGN(begin);
 332         end_aligned   = end & PAGE_MASK;
 333
 334         if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
 335                 begin = begin_aligned;
 336                 end   = end_aligned;
 337         }
 338
 339         if (begin >= end)
 340                 return;
 341
 342         addr = begin;
 343
 344         /*
 345          * If debugging page accesses then do not free this memory but
 346          * mark them not present - any buggy init-section access will
 347          * create a kernel page fault:
 348          */
 349 #ifdef CONFIG_DEBUG_PAGEALLOC
 350         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 351                 begin, end);
 352         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 353 #else
 354         /*
 355          * We just marked the kernel text read only above, now that
 356          * we are going to free part of that, we need to make that
 357          * writeable and non-executable first.
 358          */
 359         set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
 360         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
 361
 362         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 363
 364         for (; addr < end; addr += PAGE_SIZE) {
 365                 ClearPageReserved(virt_to_page(addr));
 366                 init_page_count(virt_to_page(addr));
 367                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 368                 free_page(addr);
 369                 totalram_pages++;
 370         }
 371 #endif
 372 }
 373
 374 void free_initmem(void)
 375 {
 376         free_init_pages("unused kernel memory",
 377                         (unsigned long)(&__init_begin),
 378                         (unsigned long)(&__init_end));
 379 }
 380
 381 #ifdef CONFIG_BLK_DEV_INITRD
 382 void free_initrd_mem(unsigned long start, unsigned long end)
 383 {
 384         /*
 385          * end could be not aligned, and We can not align that,
 386          * decompresser could be confused by aligned initrd_end
 387          * We already reserve the end partial page before in
 388          *   - i386_start_kernel()
 389          *   - x86_64_start_kernel()
 390          *   - relocate_initrd()
 391          * So here We can do PAGE_ALIGN() safely to get partial page to be freed
 392          */
 393         free_init_pages("initrd memory", start, PAGE_ALIGN(end));
 394 }
 395 #endif
 396
 397 void __init zone_sizes_init(void)
 398 {
 399         unsigned long max_zone_pfns[MAX_NR_ZONES];
 400
 401         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 402
 403 #ifdef CONFIG_ZONE_DMA
 404         max_zone_pfns[ZONE_DMA]         = MAX_DMA_PFN;
 405 #endif
 406 #ifdef CONFIG_ZONE_DMA32
 407         max_zone_pfns[ZONE_DMA32]       = MAX_DMA32_PFN;
 408 #endif
 409         max_zone_pfns[ZONE_NORMAL]      = max_low_pfn;
 410 #ifdef CONFIG_HIGHMEM
 411         max_zone_pfns[ZONE_HIGHMEM]     = max_pfn;
 412 #endif
 413
 414         free_area_init_nodes(max_zone_pfns);
 415 }
 416