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 int page_size_mask;
  39 /*
  40  * First calculate space needed for kernel direct mapping page tables to cover
  41  * mr[0].start to mr[nr_range - 1].end, while accounting for possible 2M and 1GB
  42  * pages. Then find enough contiguous space for those page tables.
  43  */
  44 static void __init find_early_table_space(struct map_range *mr, int nr_range)
  45 {
  46         int i;
  47         unsigned long puds = 0, pmds = 0, ptes = 0, tables;
  48         unsigned long start = 0, good_end;
  49         phys_addr_t base;
  50
  51         for (i = 0; i < nr_range; i++) {
  52                 unsigned long range, extra;
  53
  54                 range = mr[i].end - mr[i].start;
  55                 puds += (range + PUD_SIZE - 1) >> PUD_SHIFT;
  56
  57                 if (mr[i].page_size_mask & (1 << PG_LEVEL_1G)) {
  58                         extra = range - ((range >> PUD_SHIFT) << PUD_SHIFT);
  59                         pmds += (extra + PMD_SIZE - 1) >> PMD_SHIFT;
  60                 } else {
  61                         pmds += (range + PMD_SIZE - 1) >> PMD_SHIFT;
  62                 }
  63
  64                 if (mr[i].page_size_mask & (1 << PG_LEVEL_2M)) {
  65                         extra = range - ((range >> PMD_SHIFT) << PMD_SHIFT);
  66 #ifdef CONFIG_X86_32
  67                         extra += PMD_SIZE;
  68 #endif
  69                         ptes += (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
  70                 } else {
  71                         ptes += (range + PAGE_SIZE - 1) >> PAGE_SHIFT;
  72                 }
  73         }
  74
  75         tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
  76         tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
  77         tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
  78
  79 #ifdef CONFIG_X86_32
  80         /* for fixmap */
  81         tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
  82 #endif
  83         good_end = max_pfn_mapped << PAGE_SHIFT;
  84
  85         base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE);
  86         if (!base)
  87                 panic("Cannot find space for the kernel page tables");
  88
  89         pgt_buf_start = base >> PAGE_SHIFT;
  90         pgt_buf_end = pgt_buf_start;
  91         pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
  92
  93         printk(KERN_DEBUG "kernel direct mapping tables up to %#lx @ [mem %#010lx-%#010lx]\n",
  94                 mr[nr_range - 1].end - 1, pgt_buf_start << PAGE_SHIFT,
  95                 (pgt_buf_top << PAGE_SHIFT) - 1);
  96 }
  97
  98 void probe_page_size_mask(void)
  99 {
 100 #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK)
 101         /*
 102          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
 103          * This will simplify cpa(), which otherwise needs to support splitting
 104          * large pages into small in interrupt context, etc.
 105          */
 106         if (direct_gbpages)
 107                 page_size_mask |= 1 << PG_LEVEL_1G;
 108         if (cpu_has_pse)
 109                 page_size_mask |= 1 << PG_LEVEL_2M;
 110 #endif
 111
 112         /* Enable PSE if available */
 113         if (cpu_has_pse)
 114                 set_in_cr4(X86_CR4_PSE);
 115
 116         /* Enable PGE if available */
 117         if (cpu_has_pge) {
 118                 set_in_cr4(X86_CR4_PGE);
 119                 __supported_pte_mask |= _PAGE_GLOBAL;
 120         }
 121 }
 122 void __init native_pagetable_reserve(u64 start, u64 end)
 123 {
 124         memblock_reserve(start, end - start);
 125 }
 126
 127 #ifdef CONFIG_X86_32
 128 #define NR_RANGE_MR 3
 129 #else /* CONFIG_X86_64 */
 130 #define NR_RANGE_MR 5
 131 #endif
 132
 133 static int __meminit save_mr(struct map_range *mr, int nr_range,
 134                              unsigned long start_pfn, unsigned long end_pfn,
 135                              unsigned long page_size_mask)
 136 {
 137         if (start_pfn < end_pfn) {
 138                 if (nr_range >= NR_RANGE_MR)
 139                         panic("run out of range for init_memory_mapping\n");
 140                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
 141                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
 142                 mr[nr_range].page_size_mask = page_size_mask;
 143                 nr_range++;
 144         }
 145
 146         return nr_range;
 147 }
 148
 149 static int __meminit split_mem_range(struct map_range *mr, int nr_range,
 150                                      unsigned long start,
 151                                      unsigned long end)
 152 {
 153         unsigned long start_pfn, end_pfn;
 154         unsigned long pos;
 155         int i;
 156
 157         /* head if not big page alignment ? */
 158         start_pfn = start >> PAGE_SHIFT;
 159         pos = start_pfn << PAGE_SHIFT;
 160 #ifdef CONFIG_X86_32
 161         /*
 162          * Don't use a large page for the first 2/4MB of memory
 163          * because there are often fixed size MTRRs in there
 164          * and overlapping MTRRs into large pages can cause
 165          * slowdowns.
 166          */
 167         if (pos == 0)
 168                 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
 169         else
 170                 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 171                                  << (PMD_SHIFT - PAGE_SHIFT);
 172 #else /* CONFIG_X86_64 */
 173         end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
 174                         << (PMD_SHIFT - PAGE_SHIFT);
 175 #endif
 176         if (end_pfn > (end >> PAGE_SHIFT))
 177                 end_pfn = end >> PAGE_SHIFT;
 178         if (start_pfn < end_pfn) {
 179                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 180                 pos = end_pfn << PAGE_SHIFT;
 181         }
 182
 183         /* big page (2M) range */
 184         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 185                          << (PMD_SHIFT - PAGE_SHIFT);
 186 #ifdef CONFIG_X86_32
 187         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 188 #else /* CONFIG_X86_64 */
 189         end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
 190                          << (PUD_SHIFT - PAGE_SHIFT);
 191         if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
 192                 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
 193 #endif
 194
 195         if (start_pfn < end_pfn) {
 196                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 197                                 page_size_mask & (1<<PG_LEVEL_2M));
 198                 pos = end_pfn << PAGE_SHIFT;
 199         }
 200
 201 #ifdef CONFIG_X86_64
 202         /* big page (1G) range */
 203         start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
 204                          << (PUD_SHIFT - PAGE_SHIFT);
 205         end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
 206         if (start_pfn < end_pfn) {
 207                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 208                                 page_size_mask &
 209                                  ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
 210                 pos = end_pfn << PAGE_SHIFT;
 211         }
 212
 213         /* tail is not big page (1G) alignment */
 214         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 215                          << (PMD_SHIFT - PAGE_SHIFT);
 216         end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 217         if (start_pfn < end_pfn) {
 218                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 219                                 page_size_mask & (1<<PG_LEVEL_2M));
 220                 pos = end_pfn << PAGE_SHIFT;
 221         }
 222 #endif
 223
 224         /* tail is not big page (2M) alignment */
 225         start_pfn = pos>>PAGE_SHIFT;
 226         end_pfn = end>>PAGE_SHIFT;
 227         nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 228
 229         /* try to merge same page size and continuous */
 230         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
 231                 unsigned long old_start;
 232                 if (mr[i].end != mr[i+1].start ||
 233                     mr[i].page_size_mask != mr[i+1].page_size_mask)
 234                         continue;
 235                 /* move it */
 236                 old_start = mr[i].start;
 237                 memmove(&mr[i], &mr[i+1],
 238                         (nr_range - 1 - i) * sizeof(struct map_range));
 239                 mr[i--].start = old_start;
 240                 nr_range--;
 241         }
 242
 243         for (i = 0; i < nr_range; i++)
 244                 printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n",
 245                                 mr[i].start, mr[i].end - 1,
 246                         (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
 247                          (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
 248
 249         return nr_range;
 250 }
 251
 252 /*
 253  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 254  * This runs before bootmem is initialized and gets pages directly from
 255  * the physical memory. To access them they are temporarily mapped.
 256  */
 257 unsigned long __init_refok init_memory_mapping(unsigned long start,
 258                                                unsigned long end)
 259 {
 260         struct map_range mr[NR_RANGE_MR];
 261         unsigned long ret = 0;
 262         int nr_range, i;
 263
 264         pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n",
 265                start, end - 1);
 266
 267         memset(mr, 0, sizeof(mr));
 268         nr_range = split_mem_range(mr, 0, start, end);
 269
 270         /*
 271          * Find space for the kernel direct mapping tables.
 272          *
 273          * Later we should allocate these tables in the local node of the
 274          * memory mapped. Unfortunately this is done currently before the
 275          * nodes are discovered.
 276          */
 277         if (!after_bootmem)
 278                 find_early_table_space(mr, nr_range);
 279
 280         for (i = 0; i < nr_range; i++)
 281                 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
 282                                                    mr[i].page_size_mask);
 283
 284 #ifdef CONFIG_X86_32
 285         early_ioremap_page_table_range_init();
 286
 287         load_cr3(swapper_pg_dir);
 288 #endif
 289
 290         __flush_tlb_all();
 291
 292         /*
 293          * Reserve the kernel pagetable pages we used (pgt_buf_start -
 294          * pgt_buf_end) and free the other ones (pgt_buf_end - pgt_buf_top)
 295          * so that they can be reused for other purposes.
 296          *
 297          * On native it just means calling memblock_reserve, on Xen it also
 298          * means marking RW the pagetable pages that we allocated before
 299          * but that haven't been used.
 300          *
 301          * In fact on xen we mark RO the whole range pgt_buf_start -
 302          * pgt_buf_top, because we have to make sure that when
 303          * init_memory_mapping reaches the pagetable pages area, it maps
 304          * RO all the pagetable pages, including the ones that are beyond
 305          * pgt_buf_end at that time.
 306          */
 307         if (!after_bootmem && pgt_buf_end > pgt_buf_start)
 308                 x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
 309                                 PFN_PHYS(pgt_buf_end));
 310
 311         if (!after_bootmem)
 312                 early_memtest(start, end);
 313
 314         return ret >> PAGE_SHIFT;
 315 }
 316
 317
 318 /*
 319  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 320  * is valid. The argument is a physical page number.
 321  *
 322  *
 323  * On x86, access has to be given to the first megabyte of ram because that area
 324  * contains bios code and data regions used by X and dosemu and similar apps.
 325  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 326  * mmio resources as well as potential bios/acpi data regions.
 327  */
 328 int devmem_is_allowed(unsigned long pagenr)
 329 {
 330         if (pagenr < 256)
 331                 return 1;
 332         if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
 333                 return 0;
 334         if (!page_is_ram(pagenr))
 335                 return 1;
 336         return 0;
 337 }
 338
 339 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 340 {
 341         unsigned long addr;
 342         unsigned long begin_aligned, end_aligned;
 343
 344         /* Make sure boundaries are page aligned */
 345         begin_aligned = PAGE_ALIGN(begin);
 346         end_aligned   = end & PAGE_MASK;
 347
 348         if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
 349                 begin = begin_aligned;
 350                 end   = end_aligned;
 351         }
 352
 353         if (begin >= end)
 354                 return;
 355
 356         addr = begin;
 357
 358         /*
 359          * If debugging page accesses then do not free this memory but
 360          * mark them not present - any buggy init-section access will
 361          * create a kernel page fault:
 362          */
 363 #ifdef CONFIG_DEBUG_PAGEALLOC
 364         printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
 365                 begin, end - 1);
 366         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 367 #else
 368         /*
 369          * We just marked the kernel text read only above, now that
 370          * we are going to free part of that, we need to make that
 371          * writeable and non-executable first.
 372          */
 373         set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
 374         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
 375
 376         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 377
 378         for (; addr < end; addr += PAGE_SIZE) {
 379                 ClearPageReserved(virt_to_page(addr));
 380                 init_page_count(virt_to_page(addr));
 381                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 382                 free_page(addr);
 383                 totalram_pages++;
 384         }
 385 #endif
 386 }
 387
 388 void free_initmem(void)
 389 {
 390         free_init_pages("unused kernel memory",
 391                         (unsigned long)(&__init_begin),
 392                         (unsigned long)(&__init_end));
 393 }
 394
 395 #ifdef CONFIG_BLK_DEV_INITRD
 396 void __init free_initrd_mem(unsigned long start, unsigned long end)
 397 {
 398         /*
 399          * end could be not aligned, and We can not align that,
 400          * decompresser could be confused by aligned initrd_end
 401          * We already reserve the end partial page before in
 402          *   - i386_start_kernel()
 403          *   - x86_64_start_kernel()
 404          *   - relocate_initrd()
 405          * So here We can do PAGE_ALIGN() safely to get partial page to be freed
 406          */
 407         free_init_pages("initrd memory", start, PAGE_ALIGN(end));
 408 }
 409 #endif
 410
 411 void __init zone_sizes_init(void)
 412 {
 413         unsigned long max_zone_pfns[MAX_NR_ZONES];
 414
 415         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 416
 417 #ifdef CONFIG_ZONE_DMA
 418         max_zone_pfns[ZONE_DMA]         = MAX_DMA_PFN;
 419 #endif
 420 #ifdef CONFIG_ZONE_DMA32
 421         max_zone_pfns[ZONE_DMA32]       = MAX_DMA32_PFN;
 422 #endif
 423         max_zone_pfns[ZONE_NORMAL]      = max_low_pfn;
 424 #ifdef CONFIG_HIGHMEM
 425         max_zone_pfns[ZONE_HIGHMEM]     = max_pfn;
 426 #endif
 427
 428         free_area_init_nodes(max_zone_pfns);
 429 }
 430