arch/x86/mm/init_32.c

   1 /*
   2  *
   3  *  Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6  */
   7
   8 #include <linux/signal.h>
   9 #include <linux/sched.h>
  10 #include <linux/kernel.h>
  11 #include <linux/errno.h>
  12 #include <linux/string.h>
  13 #include <linux/types.h>
  14 #include <linux/ptrace.h>
  15 #include <linux/mman.h>
  16 #include <linux/mm.h>
  17 #include <linux/hugetlb.h>
  18 #include <linux/swap.h>
  19 #include <linux/smp.h>
  20 #include <linux/init.h>
  21 #include <linux/highmem.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/pci.h>
  24 #include <linux/pfn.h>
  25 #include <linux/poison.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/memblock.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/memory_hotplug.h>
  30 #include <linux/initrd.h>
  31 #include <linux/cpumask.h>
  32 #include <linux/gfp.h>
  33
  34 #include <asm/asm.h>
  35 #include <asm/bios_ebda.h>
  36 #include <asm/processor.h>
  37 #include <linux/uaccess.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/dma.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/e820.h>
  42 #include <asm/apic.h>
  43 #include <asm/bugs.h>
  44 #include <asm/tlb.h>
  45 #include <asm/tlbflush.h>
  46 #include <asm/olpc_ofw.h>
  47 #include <asm/pgalloc.h>
  48 #include <asm/sections.h>
  49 #include <asm/paravirt.h>
  50 #include <asm/setup.h>
  51 #include <asm/cacheflush.h>
  52 #include <asm/page_types.h>
  53 #include <asm/init.h>
  54
  55 #include "mm_internal.h"
  56
  57 unsigned long highstart_pfn, highend_pfn;
  58
  59 static noinline int do_test_wp_bit(void);
  60
  61 bool __read_mostly __vmalloc_start_set = false;
  62
  63 /*
  64  * Creates a middle page table and puts a pointer to it in the
  65  * given global directory entry. This only returns the gd entry
  66  * in non-PAE compilation mode, since the middle layer is folded.
  67  */
  68 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  69 {
  70         pud_t *pud;
  71         pmd_t *pmd_table;
  72
  73 #ifdef CONFIG_X86_PAE
  74         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  75                 pmd_table = (pmd_t *)alloc_low_page();
  76                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  77                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  78                 pud = pud_offset(pgd, 0);
  79                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  80
  81                 return pmd_table;
  82         }
  83 #endif
  84         pud = pud_offset(pgd, 0);
  85         pmd_table = pmd_offset(pud, 0);
  86
  87         return pmd_table;
  88 }
  89
  90 /*
  91  * Create a page table and place a pointer to it in a middle page
  92  * directory entry:
  93  */
  94 static pte_t * __init one_page_table_init(pmd_t *pmd)
  95 {
  96         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
  97                 pte_t *page_table = (pte_t *)alloc_low_page();
  98
  99                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 100                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 101                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 102         }
 103
 104         return pte_offset_kernel(pmd, 0);
 105 }
 106
 107 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 108 {
 109         int pgd_idx = pgd_index(vaddr);
 110         int pmd_idx = pmd_index(vaddr);
 111
 112         return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
 113 }
 114
 115 pte_t * __init populate_extra_pte(unsigned long vaddr)
 116 {
 117         int pte_idx = pte_index(vaddr);
 118         pmd_t *pmd;
 119
 120         pmd = populate_extra_pmd(vaddr);
 121         return one_page_table_init(pmd) + pte_idx;
 122 }
 123
 124 static unsigned long __init
 125 page_table_range_init_count(unsigned long start, unsigned long end)
 126 {
 127         unsigned long count = 0;
 128 #ifdef CONFIG_HIGHMEM
 129         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 130         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 131         int pgd_idx, pmd_idx;
 132         unsigned long vaddr;
 133
 134         if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
 135                 return 0;
 136
 137         vaddr = start;
 138         pgd_idx = pgd_index(vaddr);
 139         pmd_idx = pmd_index(vaddr);
 140
 141         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
 142                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 143                                                         pmd_idx++) {
 144                         if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
 145                             (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
 146                                 count++;
 147                         vaddr += PMD_SIZE;
 148                 }
 149                 pmd_idx = 0;
 150         }
 151 #endif
 152         return count;
 153 }
 154
 155 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 156                                            unsigned long vaddr, pte_t *lastpte,
 157                                            void **adr)
 158 {
 159 #ifdef CONFIG_HIGHMEM
 160         /*
 161          * Something (early fixmap) may already have put a pte
 162          * page here, which causes the page table allocation
 163          * to become nonlinear. Attempt to fix it, and if it
 164          * is still nonlinear then we have to bug.
 165          */
 166         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 167         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 168
 169         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 170             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 171             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
 172                 pte_t *newpte;
 173                 int i;
 174
 175                 BUG_ON(after_bootmem);
 176                 newpte = *adr;
 177                 for (i = 0; i < PTRS_PER_PTE; i++)
 178                         set_pte(newpte + i, pte[i]);
 179                 *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
 180
 181                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 182                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 183                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 184                 __flush_tlb_all();
 185
 186                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 187                 pte = newpte;
 188         }
 189         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 190                && vaddr > fix_to_virt(FIX_KMAP_END)
 191                && lastpte && lastpte + PTRS_PER_PTE != pte);
 192 #endif
 193         return pte;
 194 }
 195
 196 /*
 197  * This function initializes a certain range of kernel virtual memory
 198  * with new bootmem page tables, everywhere page tables are missing in
 199  * the given range.
 200  *
 201  * NOTE: The pagetables are allocated contiguous on the physical space
 202  * so we can cache the place of the first one and move around without
 203  * checking the pgd every time.
 204  */
 205 static void __init
 206 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 207 {
 208         int pgd_idx, pmd_idx;
 209         unsigned long vaddr;
 210         pgd_t *pgd;
 211         pmd_t *pmd;
 212         pte_t *pte = NULL;
 213         unsigned long count = page_table_range_init_count(start, end);
 214         void *adr = NULL;
 215
 216         if (count)
 217                 adr = alloc_low_pages(count);
 218
 219         vaddr = start;
 220         pgd_idx = pgd_index(vaddr);
 221         pmd_idx = pmd_index(vaddr);
 222         pgd = pgd_base + pgd_idx;
 223
 224         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 225                 pmd = one_md_table_init(pgd);
 226                 pmd = pmd + pmd_index(vaddr);
 227                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 228                                                         pmd++, pmd_idx++) {
 229                         pte = page_table_kmap_check(one_page_table_init(pmd),
 230                                                     pmd, vaddr, pte, &adr);
 231
 232                         vaddr += PMD_SIZE;
 233                 }
 234                 pmd_idx = 0;
 235         }
 236 }
 237
 238 static inline int is_kernel_text(unsigned long addr)
 239 {
 240         if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
 241                 return 1;
 242         return 0;
 243 }
 244
 245 /*
 246  * This maps the physical memory to kernel virtual address space, a total
 247  * of max_low_pfn pages, by creating page tables starting from address
 248  * PAGE_OFFSET:
 249  */
 250 unsigned long __init
 251 kernel_physical_mapping_init(unsigned long start,
 252                              unsigned long end,
 253                              unsigned long page_size_mask)
 254 {
 255         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
 256         unsigned long last_map_addr = end;
 257         unsigned long start_pfn, end_pfn;
 258         pgd_t *pgd_base = swapper_pg_dir;
 259         int pgd_idx, pmd_idx, pte_ofs;
 260         unsigned long pfn;
 261         pgd_t *pgd;
 262         pmd_t *pmd;
 263         pte_t *pte;
 264         unsigned pages_2m, pages_4k;
 265         int mapping_iter;
 266
 267         start_pfn = start >> PAGE_SHIFT;
 268         end_pfn = end >> PAGE_SHIFT;
 269
 270         /*
 271          * First iteration will setup identity mapping using large/small pages
 272          * based on use_pse, with other attributes same as set by
 273          * the early code in head_32.S
 274          *
 275          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 276          * as desired for the kernel identity mapping.
 277          *
 278          * This two pass mechanism conforms to the TLB app note which says:
 279          *
 280          *     "Software should not write to a paging-structure entry in a way
 281          *      that would change, for any linear address, both the page size
 282          *      and either the page frame or attributes."
 283          */
 284         mapping_iter = 1;
 285
 286         if (!boot_cpu_has(X86_FEATURE_PSE))
 287                 use_pse = 0;
 288
 289 repeat:
 290         pages_2m = pages_4k = 0;
 291         pfn = start_pfn;
 292         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 293         pgd = pgd_base + pgd_idx;
 294         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 295                 pmd = one_md_table_init(pgd);
 296
 297                 if (pfn >= end_pfn)
 298                         continue;
 299 #ifdef CONFIG_X86_PAE
 300                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 301                 pmd += pmd_idx;
 302 #else
 303                 pmd_idx = 0;
 304 #endif
 305                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 306                      pmd++, pmd_idx++) {
 307                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 308
 309                         /*
 310                          * Map with big pages if possible, otherwise
 311                          * create normal page tables:
 312                          */
 313                         if (use_pse) {
 314                                 unsigned int addr2;
 315                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 316                                 /*
 317                                  * first pass will use the same initial
 318                                  * identity mapping attribute + _PAGE_PSE.
 319                                  */
 320                                 pgprot_t init_prot =
 321                                         __pgprot(PTE_IDENT_ATTR |
 322                                                  _PAGE_PSE);
 323
 324                                 pfn &= PMD_MASK >> PAGE_SHIFT;
 325                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 326                                         PAGE_OFFSET + PAGE_SIZE-1;
 327
 328                                 if (is_kernel_text(addr) ||
 329                                     is_kernel_text(addr2))
 330                                         prot = PAGE_KERNEL_LARGE_EXEC;
 331
 332                                 pages_2m++;
 333                                 if (mapping_iter == 1)
 334                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 335                                 else
 336                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 337
 338                                 pfn += PTRS_PER_PTE;
 339                                 continue;
 340                         }
 341                         pte = one_page_table_init(pmd);
 342
 343                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 344                         pte += pte_ofs;
 345                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 346                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 347                                 pgprot_t prot = PAGE_KERNEL;
 348                                 /*
 349                                  * first pass will use the same initial
 350                                  * identity mapping attribute.
 351                                  */
 352                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 353
 354                                 if (is_kernel_text(addr))
 355                                         prot = PAGE_KERNEL_EXEC;
 356
 357                                 pages_4k++;
 358                                 if (mapping_iter == 1) {
 359                                         set_pte(pte, pfn_pte(pfn, init_prot));
 360                                         last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
 361                                 } else
 362                                         set_pte(pte, pfn_pte(pfn, prot));
 363                         }
 364                 }
 365         }
 366         if (mapping_iter == 1) {
 367                 /*
 368                  * update direct mapping page count only in the first
 369                  * iteration.
 370                  */
 371                 update_page_count(PG_LEVEL_2M, pages_2m);
 372                 update_page_count(PG_LEVEL_4K, pages_4k);
 373
 374                 /*
 375                  * local global flush tlb, which will flush the previous
 376                  * mappings present in both small and large page TLB's.
 377                  */
 378                 __flush_tlb_all();
 379
 380                 /*
 381                  * Second iteration will set the actual desired PTE attributes.
 382                  */
 383                 mapping_iter = 2;
 384                 goto repeat;
 385         }
 386         return last_map_addr;
 387 }
 388
 389 pte_t *kmap_pte;
 390
 391 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 392 {
 393         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 394                         vaddr), vaddr), vaddr);
 395 }
 396
 397 static void __init kmap_init(void)
 398 {
 399         unsigned long kmap_vstart;
 400
 401         /*
 402          * Cache the first kmap pte:
 403          */
 404         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 405         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 406 }
 407
 408 #ifdef CONFIG_HIGHMEM
 409 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 410 {
 411         unsigned long vaddr;
 412         pgd_t *pgd;
 413         pud_t *pud;
 414         pmd_t *pmd;
 415         pte_t *pte;
 416
 417         vaddr = PKMAP_BASE;
 418         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 419
 420         pgd = swapper_pg_dir + pgd_index(vaddr);
 421         pud = pud_offset(pgd, vaddr);
 422         pmd = pmd_offset(pud, vaddr);
 423         pte = pte_offset_kernel(pmd, vaddr);
 424         pkmap_page_table = pte;
 425 }
 426
 427 void __init add_highpages_with_active_regions(int nid,
 428                          unsigned long start_pfn, unsigned long end_pfn)
 429 {
 430         phys_addr_t start, end;
 431         u64 i;
 432
 433         for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
 434                 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
 435                                             start_pfn, end_pfn);
 436                 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
 437                                               start_pfn, end_pfn);
 438                 for ( ; pfn < e_pfn; pfn++)
 439                         if (pfn_valid(pfn))
 440                                 free_highmem_page(pfn_to_page(pfn));
 441         }
 442 }
 443 #else
 444 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 445 {
 446 }
 447 #endif /* CONFIG_HIGHMEM */
 448
 449 void __init native_pagetable_init(void)
 450 {
 451         unsigned long pfn, va;
 452         pgd_t *pgd, *base = swapper_pg_dir;
 453         pud_t *pud;
 454         pmd_t *pmd;
 455         pte_t *pte;
 456
 457         /*
 458          * Remove any mappings which extend past the end of physical
 459          * memory from the boot time page table.
 460          * In virtual address space, we should have at least two pages
 461          * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
 462          * definition. And max_low_pfn is set to VMALLOC_END physical
 463          * address. If initial memory mapping is doing right job, we
 464          * should have pte used near max_low_pfn or one pmd is not present.
 465          */
 466         for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 467                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 468                 pgd = base + pgd_index(va);
 469                 if (!pgd_present(*pgd))
 470                         break;
 471
 472                 pud = pud_offset(pgd, va);
 473                 pmd = pmd_offset(pud, va);
 474                 if (!pmd_present(*pmd))
 475                         break;
 476
 477                 /* should not be large page here */
 478                 if (pmd_large(*pmd)) {
 479                         pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
 480                                 pfn, pmd, __pa(pmd));
 481                         BUG_ON(1);
 482                 }
 483
 484                 pte = pte_offset_kernel(pmd, va);
 485                 if (!pte_present(*pte))
 486                         break;
 487
 488                 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
 489                                 pfn, pmd, __pa(pmd), pte, __pa(pte));
 490                 pte_clear(NULL, va, pte);
 491         }
 492         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 493         paging_init();
 494 }
 495
 496 /*
 497  * Build a proper pagetable for the kernel mappings.  Up until this
 498  * point, we've been running on some set of pagetables constructed by
 499  * the boot process.
 500  *
 501  * If we're booting on native hardware, this will be a pagetable
 502  * constructed in arch/x86/kernel/head_32.S.  The root of the
 503  * pagetable will be swapper_pg_dir.
 504  *
 505  * If we're booting paravirtualized under a hypervisor, then there are
 506  * more options: we may already be running PAE, and the pagetable may
 507  * or may not be based in swapper_pg_dir.  In any case,
 508  * paravirt_pagetable_init() will set up swapper_pg_dir
 509  * appropriately for the rest of the initialization to work.
 510  *
 511  * In general, pagetable_init() assumes that the pagetable may already
 512  * be partially populated, and so it avoids stomping on any existing
 513  * mappings.
 514  */
 515 void __init early_ioremap_page_table_range_init(void)
 516 {
 517         pgd_t *pgd_base = swapper_pg_dir;
 518         unsigned long vaddr, end;
 519
 520         /*
 521          * Fixed mappings, only the page table structure has to be
 522          * created - mappings will be set by set_fixmap():
 523          */
 524         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 525         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 526         page_table_range_init(vaddr, end, pgd_base);
 527         early_ioremap_reset();
 528 }
 529
 530 static void __init pagetable_init(void)
 531 {
 532         pgd_t *pgd_base = swapper_pg_dir;
 533
 534         permanent_kmaps_init(pgd_base);
 535 }
 536
 537 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
 538 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 539
 540 /* user-defined highmem size */
 541 static unsigned int highmem_pages = -1;
 542
 543 /*
 544  * highmem=size forces highmem to be exactly 'size' bytes.
 545  * This works even on boxes that have no highmem otherwise.
 546  * This also works to reduce highmem size on bigger boxes.
 547  */
 548 static int __init parse_highmem(char *arg)
 549 {
 550         if (!arg)
 551                 return -EINVAL;
 552
 553         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 554         return 0;
 555 }
 556 early_param("highmem", parse_highmem);
 557
 558 #define MSG_HIGHMEM_TOO_BIG \
 559         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 560
 561 #define MSG_LOWMEM_TOO_SMALL \
 562         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 563 /*
 564  * All of RAM fits into lowmem - but if user wants highmem
 565  * artificially via the highmem=x boot parameter then create
 566  * it:
 567  */
 568 static void __init lowmem_pfn_init(void)
 569 {
 570         /* max_low_pfn is 0, we already have early_res support */
 571         max_low_pfn = max_pfn;
 572
 573         if (highmem_pages == -1)
 574                 highmem_pages = 0;
 575 #ifdef CONFIG_HIGHMEM
 576         if (highmem_pages >= max_pfn) {
 577                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 578                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 579                 highmem_pages = 0;
 580         }
 581         if (highmem_pages) {
 582                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 583                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 584                                 pages_to_mb(highmem_pages));
 585                         highmem_pages = 0;
 586                 }
 587                 max_low_pfn -= highmem_pages;
 588         }
 589 #else
 590         if (highmem_pages)
 591                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 592 #endif
 593 }
 594
 595 #define MSG_HIGHMEM_TOO_SMALL \
 596         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 597
 598 #define MSG_HIGHMEM_TRIMMED \
 599         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 600 /*
 601  * We have more RAM than fits into lowmem - we try to put it into
 602  * highmem, also taking the highmem=x boot parameter into account:
 603  */
 604 static void __init highmem_pfn_init(void)
 605 {
 606         max_low_pfn = MAXMEM_PFN;
 607
 608         if (highmem_pages == -1)
 609                 highmem_pages = max_pfn - MAXMEM_PFN;
 610
 611         if (highmem_pages + MAXMEM_PFN < max_pfn)
 612                 max_pfn = MAXMEM_PFN + highmem_pages;
 613
 614         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 615                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 616                         pages_to_mb(max_pfn - MAXMEM_PFN),
 617                         pages_to_mb(highmem_pages));
 618                 highmem_pages = 0;
 619         }
 620 #ifndef CONFIG_HIGHMEM
 621         /* Maximum memory usable is what is directly addressable */
 622         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 623         if (max_pfn > MAX_NONPAE_PFN)
 624                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 625         else
 626                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 627         max_pfn = MAXMEM_PFN;
 628 #else /* !CONFIG_HIGHMEM */
 629 #ifndef CONFIG_HIGHMEM64G
 630         if (max_pfn > MAX_NONPAE_PFN) {
 631                 max_pfn = MAX_NONPAE_PFN;
 632                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 633         }
 634 #endif /* !CONFIG_HIGHMEM64G */
 635 #endif /* !CONFIG_HIGHMEM */
 636 }
 637
 638 /*
 639  * Determine low and high memory ranges:
 640  */
 641 void __init find_low_pfn_range(void)
 642 {
 643         /* it could update max_pfn */
 644
 645         if (max_pfn <= MAXMEM_PFN)
 646                 lowmem_pfn_init();
 647         else
 648                 highmem_pfn_init();
 649 }
 650
 651 #ifndef CONFIG_NEED_MULTIPLE_NODES
 652 void __init initmem_init(void)
 653 {
 654 #ifdef CONFIG_HIGHMEM
 655         highstart_pfn = highend_pfn = max_pfn;
 656         if (max_pfn > max_low_pfn)
 657                 highstart_pfn = max_low_pfn;
 658         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 659                 pages_to_mb(highend_pfn - highstart_pfn));
 660         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 661 #else
 662         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 663 #endif
 664
 665         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
 666         sparse_memory_present_with_active_regions(0);
 667
 668 #ifdef CONFIG_FLATMEM
 669         max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
 670 #endif
 671         __vmalloc_start_set = true;
 672
 673         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 674                         pages_to_mb(max_low_pfn));
 675
 676         setup_bootmem_allocator();
 677 }
 678 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 679
 680 void __init setup_bootmem_allocator(void)
 681 {
 682         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 683                  max_pfn_mapped<<PAGE_SHIFT);
 684         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 685 }
 686
 687 /*
 688  * paging_init() sets up the page tables - note that the first 8MB are
 689  * already mapped by head.S.
 690  *
 691  * This routines also unmaps the page at virtual kernel address 0, so
 692  * that we can trap those pesky NULL-reference errors in the kernel.
 693  */
 694 void __init paging_init(void)
 695 {
 696         pagetable_init();
 697
 698         __flush_tlb_all();
 699
 700         kmap_init();
 701
 702         /*
 703          * NOTE: at this point the bootmem allocator is fully available.
 704          */
 705         olpc_dt_build_devicetree();
 706         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 707         sparse_init();
 708         zone_sizes_init();
 709 }
 710
 711 /*
 712  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 713  * and also on some strange 486's. All 586+'s are OK. This used to involve
 714  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 715  * switch to using exceptions got rid of all that.
 716  */
 717 static void __init test_wp_bit(void)
 718 {
 719         printk(KERN_INFO
 720   "Checking if this processor honours the WP bit even in supervisor mode...");
 721
 722         /* Any page-aligned address will do, the test is non-destructive */
 723         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_KERNEL_RO);
 724         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 725         clear_fixmap(FIX_WP_TEST);
 726
 727         if (!boot_cpu_data.wp_works_ok) {
 728                 printk(KERN_CONT "No.\n");
 729                 panic("Linux doesn't support CPUs with broken WP.");
 730         } else {
 731                 printk(KERN_CONT "Ok.\n");
 732         }
 733 }
 734
 735 void __init mem_init(void)
 736 {
 737         pci_iommu_alloc();
 738
 739 #ifdef CONFIG_FLATMEM
 740         BUG_ON(!mem_map);
 741 #endif
 742         /*
 743          * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
 744          * be done before free_all_bootmem(). Memblock use free low memory for
 745          * temporary data (see find_range_array()) and for this purpose can use
 746          * pages that was already passed to the buddy allocator, hence marked as
 747          * not accessible in the page tables when compiled with
 748          * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
 749          * important here.
 750          */
 751         set_highmem_pages_init();
 752
 753         /* this will put all low memory onto the freelists */
 754         free_all_bootmem();
 755
 756         after_bootmem = 1;
 757
 758         mem_init_print_info(NULL);
 759         printk(KERN_INFO "virtual kernel memory layout:\n"
 760                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 761 #ifdef CONFIG_HIGHMEM
 762                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 763 #endif
 764                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 765                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 766                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 767                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 768                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 769                 FIXADDR_START, FIXADDR_TOP,
 770                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 771
 772 #ifdef CONFIG_HIGHMEM
 773                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 774                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 775 #endif
 776
 777                 VMALLOC_START, VMALLOC_END,
 778                 (VMALLOC_END - VMALLOC_START) >> 20,
 779
 780                 (unsigned long)__va(0), (unsigned long)high_memory,
 781                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 782
 783                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 784                 ((unsigned long)&__init_end -
 785                  (unsigned long)&__init_begin) >> 10,
 786
 787                 (unsigned long)&_etext, (unsigned long)&_edata,
 788                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 789
 790                 (unsigned long)&_text, (unsigned long)&_etext,
 791                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 792
 793         /*
 794          * Check boundaries twice: Some fundamental inconsistencies can
 795          * be detected at build time already.
 796          */
 797 #define __FIXADDR_TOP (-PAGE_SIZE)
 798 #ifdef CONFIG_HIGHMEM
 799         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 800         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 801 #endif
 802 #define high_memory (-128UL << 20)
 803         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 804 #undef high_memory
 805 #undef __FIXADDR_TOP
 806
 807 #ifdef CONFIG_HIGHMEM
 808         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 809         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 810 #endif
 811         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 812         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 813
 814         if (boot_cpu_data.wp_works_ok < 0)
 815                 test_wp_bit();
 816 }
 817
 818 #ifdef CONFIG_MEMORY_HOTPLUG
 819 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
 820 {
 821         struct pglist_data *pgdata = NODE_DATA(nid);
 822         struct zone *zone = pgdata->node_zones +
 823                 zone_for_memory(nid, start, size, ZONE_HIGHMEM, for_device);
 824         unsigned long start_pfn = start >> PAGE_SHIFT;
 825         unsigned long nr_pages = size >> PAGE_SHIFT;
 826
 827         return __add_pages(nid, zone, start_pfn, nr_pages);
 828 }
 829
 830 #ifdef CONFIG_MEMORY_HOTREMOVE
 831 int arch_remove_memory(u64 start, u64 size)
 832 {
 833         unsigned long start_pfn = start >> PAGE_SHIFT;
 834         unsigned long nr_pages = size >> PAGE_SHIFT;
 835         struct zone *zone;
 836
 837         zone = page_zone(pfn_to_page(start_pfn));
 838         return __remove_pages(zone, start_pfn, nr_pages);
 839 }
 840 #endif
 841 #endif
 842
 843 /*
 844  * This function cannot be __init, since exceptions don't work in that
 845  * section.  Put this after the callers, so that it cannot be inlined.
 846  */
 847 static noinline int do_test_wp_bit(void)
 848 {
 849         char tmp_reg;
 850         int flag;
 851
 852         __asm__ __volatile__(
 853                 "       movb %0, %1     \n"
 854                 "1:     movb %1, %0     \n"
 855                 "       xorl %2, %2     \n"
 856                 "2:                     \n"
 857                 _ASM_EXTABLE(1b,2b)
 858                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 859                  "=q" (tmp_reg),
 860                  "=r" (flag)
 861                 :"2" (1)
 862                 :"memory");
 863
 864         return flag;
 865 }
 866
 867 int kernel_set_to_readonly __read_mostly;
 868
 869 void set_kernel_text_rw(void)
 870 {
 871         unsigned long start = PFN_ALIGN(_text);
 872         unsigned long size = PFN_ALIGN(_etext) - start;
 873
 874         if (!kernel_set_to_readonly)
 875                 return;
 876
 877         pr_debug("Set kernel text: %lx - %lx for read write\n",
 878                  start, start+size);
 879
 880         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 881 }
 882
 883 void set_kernel_text_ro(void)
 884 {
 885         unsigned long start = PFN_ALIGN(_text);
 886         unsigned long size = PFN_ALIGN(_etext) - start;
 887
 888         if (!kernel_set_to_readonly)
 889                 return;
 890
 891         pr_debug("Set kernel text: %lx - %lx for read only\n",
 892                  start, start+size);
 893
 894         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 895 }
 896
 897 static void mark_nxdata_nx(void)
 898 {
 899         /*
 900          * When this called, init has already been executed and released,
 901          * so everything past _etext should be NX.
 902          */
 903         unsigned long start = PFN_ALIGN(_etext);
 904         /*
 905          * This comes from is_kernel_text upper limit. Also HPAGE where used:
 906          */
 907         unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
 908
 909         if (__supported_pte_mask & _PAGE_NX)
 910                 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
 911         set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
 912 }
 913
 914 void mark_rodata_ro(void)
 915 {
 916         unsigned long start = PFN_ALIGN(_text);
 917         unsigned long size = PFN_ALIGN(_etext) - start;
 918
 919         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 920         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
 921                 size >> 10);
 922
 923         kernel_set_to_readonly = 1;
 924
 925 #ifdef CONFIG_CPA_DEBUG
 926         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
 927                 start, start+size);
 928         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
 929
 930         printk(KERN_INFO "Testing CPA: write protecting again\n");
 931         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
 932 #endif
 933
 934         start += size;
 935         size = (unsigned long)__end_rodata - start;
 936         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 937         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 938                 size >> 10);
 939
 940 #ifdef CONFIG_CPA_DEBUG
 941         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
 942         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 943
 944         printk(KERN_INFO "Testing CPA: write protecting again\n");
 945         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 946 #endif
 947         mark_nxdata_nx();
 948         if (__supported_pte_mask & _PAGE_NX)
 949                 debug_checkwx();
 950 }