arch/arm64/mm/mmu.c

   1 /*
   2  * Based on arch/arm/mm/mmu.c
   3  *
   4  * Copyright (C) 1995-2005 Russell King
   5  * Copyright (C) 2012 ARM Ltd.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include <linux/export.h>
  21 #include <linux/kernel.h>
  22 #include <linux/errno.h>
  23 #include <linux/init.h>
  24 #include <linux/mman.h>
  25 #include <linux/nodemask.h>
  26 #include <linux/memblock.h>
  27 #include <linux/fs.h>
  28 #include <linux/io.h>
  29
  30 #include <asm/cputype.h>
  31 #include <asm/sections.h>
  32 #include <asm/setup.h>
  33 #include <asm/sizes.h>
  34 #include <asm/tlb.h>
  35 #include <asm/mmu_context.h>
  36
  37 #include "mm.h"
  38
  39 /*
  40  * Empty_zero_page is a special page that is used for zero-initialized data
  41  * and COW.
  42  */
  43 struct page *empty_zero_page;
  44 EXPORT_SYMBOL(empty_zero_page);
  45
  46 struct cachepolicy {
  47         const char      policy[16];
  48         u64             mair;
  49         u64             tcr;
  50 };
  51
  52 static struct cachepolicy cache_policies[] __initdata = {
  53         {
  54                 .policy         = "uncached",
  55                 .mair           = 0x44,                 /* inner, outer non-cacheable */
  56                 .tcr            = TCR_IRGN_NC | TCR_ORGN_NC,
  57         }, {
  58                 .policy         = "writethrough",
  59                 .mair           = 0xaa,                 /* inner, outer write-through, read-allocate */
  60                 .tcr            = TCR_IRGN_WT | TCR_ORGN_WT,
  61         }, {
  62                 .policy         = "writeback",
  63                 .mair           = 0xee,                 /* inner, outer write-back, read-allocate */
  64                 .tcr            = TCR_IRGN_WBnWA | TCR_ORGN_WBnWA,
  65         }
  66 };
  67
  68 /*
  69  * These are useful for identifying cache coherency problems by allowing the
  70  * cache or the cache and writebuffer to be turned off. It changes the Normal
  71  * memory caching attributes in the MAIR_EL1 register.
  72  */
  73 static int __init early_cachepolicy(char *p)
  74 {
  75         int i;
  76         u64 tmp;
  77
  78         for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
  79                 int len = strlen(cache_policies[i].policy);
  80
  81                 if (memcmp(p, cache_policies[i].policy, len) == 0)
  82                         break;
  83         }
  84         if (i == ARRAY_SIZE(cache_policies)) {
  85                 pr_err("ERROR: unknown or unsupported cache policy: %s\n", p);
  86                 return 0;
  87         }
  88
  89         flush_cache_all();
  90
  91         /*
  92          * Modify MT_NORMAL attributes in MAIR_EL1.
  93          */
  94         asm volatile(
  95         "       mrs     %0, mair_el1\n"
  96         "       bfi     %0, %1, #%2, #8\n"
  97         "       msr     mair_el1, %0\n"
  98         "       isb\n"
  99         : "=&r" (tmp)
 100         : "r" (cache_policies[i].mair), "i" (MT_NORMAL * 8));
 101
 102         /*
 103          * Modify TCR PTW cacheability attributes.
 104          */
 105         asm volatile(
 106         "       mrs     %0, tcr_el1\n"
 107         "       bic     %0, %0, %2\n"
 108         "       orr     %0, %0, %1\n"
 109         "       msr     tcr_el1, %0\n"
 110         "       isb\n"
 111         : "=&r" (tmp)
 112         : "r" (cache_policies[i].tcr), "r" (TCR_IRGN_MASK | TCR_ORGN_MASK));
 113
 114         flush_cache_all();
 115
 116         return 0;
 117 }
 118 early_param("cachepolicy", early_cachepolicy);
 119
 120 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 121                               unsigned long size, pgprot_t vma_prot)
 122 {
 123         if (!pfn_valid(pfn))
 124                 return pgprot_noncached(vma_prot);
 125         else if (file->f_flags & O_SYNC)
 126                 return pgprot_writecombine(vma_prot);
 127         return vma_prot;
 128 }
 129 EXPORT_SYMBOL(phys_mem_access_prot);
 130
 131 static void __init *early_alloc(unsigned long sz)
 132 {
 133         void *ptr = __va(memblock_alloc(sz, sz));
 134         memset(ptr, 0, sz);
 135         return ptr;
 136 }
 137
 138 static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
 139                                   unsigned long end, unsigned long pfn,
 140                                   pgprot_t prot)
 141 {
 142         pte_t *pte;
 143
 144         if (pmd_none(*pmd)) {
 145                 pte = early_alloc(PTRS_PER_PTE * sizeof(pte_t));
 146                 __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
 147         }
 148         BUG_ON(pmd_bad(*pmd));
 149
 150         pte = pte_offset_kernel(pmd, addr);
 151         do {
 152                 set_pte(pte, pfn_pte(pfn, prot));
 153                 pfn++;
 154         } while (pte++, addr += PAGE_SIZE, addr != end);
 155 }
 156
 157 static void __init alloc_init_pmd(pud_t *pud, unsigned long addr,
 158                                   unsigned long end, phys_addr_t phys,
 159                                   int map_io)
 160 {
 161         pmd_t *pmd;
 162         unsigned long next;
 163         pmdval_t prot_sect;
 164         pgprot_t prot_pte;
 165
 166         if (map_io) {
 167                 prot_sect = PROT_SECT_DEVICE_nGnRE;
 168                 prot_pte = __pgprot(PROT_DEVICE_nGnRE);
 169         } else {
 170                 prot_sect = PROT_SECT_NORMAL_EXEC;
 171                 prot_pte = PAGE_KERNEL_EXEC;
 172         }
 173
 174         /*
 175          * Check for initial section mappings in the pgd/pud and remove them.
 176          */
 177         if (pud_none(*pud) || pud_bad(*pud)) {
 178                 pmd = early_alloc(PTRS_PER_PMD * sizeof(pmd_t));
 179                 pud_populate(&init_mm, pud, pmd);
 180         }
 181
 182         pmd = pmd_offset(pud, addr);
 183         do {
 184                 next = pmd_addr_end(addr, end);
 185                 /* try section mapping first */
 186                 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
 187                         pmd_t old_pmd =*pmd;
 188                         set_pmd(pmd, __pmd(phys | prot_sect));
 189                         /*
 190                          * Check for previous table entries created during
 191                          * boot (__create_page_tables) and flush them.
 192                          */
 193                         if (!pmd_none(old_pmd))
 194                                 flush_tlb_all();
 195                 } else {
 196                         alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
 197                                        prot_pte);
 198                 }
 199                 phys += next - addr;
 200         } while (pmd++, addr = next, addr != end);
 201 }
 202
 203 static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr,
 204                                   unsigned long end, unsigned long phys,
 205                                   int map_io)
 206 {
 207         pud_t *pud = pud_offset(pgd, addr);
 208         unsigned long next;
 209
 210         do {
 211                 next = pud_addr_end(addr, end);
 212
 213                 /*
 214                  * For 4K granule only, attempt to put down a 1GB block
 215                  */
 216                 if (!map_io && (PAGE_SHIFT == 12) &&
 217                     ((addr | next | phys) & ~PUD_MASK) == 0) {
 218                         pud_t old_pud = *pud;
 219                         set_pud(pud, __pud(phys | PROT_SECT_NORMAL_EXEC));
 220
 221                         /*
 222                          * If we have an old value for a pud, it will
 223                          * be pointing to a pmd table that we no longer
 224                          * need (from swapper_pg_dir).
 225                          *
 226                          * Look up the old pmd table and free it.
 227                          */
 228                         if (!pud_none(old_pud)) {
 229                                 phys_addr_t table = __pa(pmd_offset(&old_pud, 0));
 230                                 memblock_free(table, PAGE_SIZE);
 231                                 flush_tlb_all();
 232                         }
 233                 } else {
 234                         alloc_init_pmd(pud, addr, next, phys, map_io);
 235                 }
 236                 phys += next - addr;
 237         } while (pud++, addr = next, addr != end);
 238 }
 239
 240 /*
 241  * Create the page directory entries and any necessary page tables for the
 242  * mapping specified by 'md'.
 243  */
 244 static void __init __create_mapping(pgd_t *pgd, phys_addr_t phys,
 245                                     unsigned long virt, phys_addr_t size,
 246                                     int map_io)
 247 {
 248         unsigned long addr, length, end, next;
 249
 250         addr = virt & PAGE_MASK;
 251         length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
 252
 253         end = addr + length;
 254         do {
 255                 next = pgd_addr_end(addr, end);
 256                 alloc_init_pud(pgd, addr, next, phys, map_io);
 257                 phys += next - addr;
 258         } while (pgd++, addr = next, addr != end);
 259 }
 260
 261 static void __init create_mapping(phys_addr_t phys, unsigned long virt,
 262                                   phys_addr_t size)
 263 {
 264         if (virt < VMALLOC_START) {
 265                 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
 266                         &phys, virt);
 267                 return;
 268         }
 269         __create_mapping(pgd_offset_k(virt & PAGE_MASK), phys, virt, size, 0);
 270 }
 271
 272 void __init create_id_mapping(phys_addr_t addr, phys_addr_t size, int map_io)
 273 {
 274         if ((addr >> PGDIR_SHIFT) >= ARRAY_SIZE(idmap_pg_dir)) {
 275                 pr_warn("BUG: not creating id mapping for %pa\n", &addr);
 276                 return;
 277         }
 278         __create_mapping(&idmap_pg_dir[pgd_index(addr)],
 279                          addr, addr, size, map_io);
 280 }
 281
 282 static void __init map_mem(void)
 283 {
 284         struct memblock_region *reg;
 285         phys_addr_t limit;
 286
 287         /*
 288          * Temporarily limit the memblock range. We need to do this as
 289          * create_mapping requires puds, pmds and ptes to be allocated from
 290          * memory addressable from the initial direct kernel mapping.
 291          *
 292          * The initial direct kernel mapping, located at swapper_pg_dir,
 293          * gives us PGDIR_SIZE memory starting from PHYS_OFFSET (which must be
 294          * aligned to 2MB as per Documentation/arm64/booting.txt).
 295          */
 296         limit = PHYS_OFFSET + PGDIR_SIZE;
 297         memblock_set_current_limit(limit);
 298
 299         /* map all the memory banks */
 300         for_each_memblock(memory, reg) {
 301                 phys_addr_t start = reg->base;
 302                 phys_addr_t end = start + reg->size;
 303
 304                 if (start >= end)
 305                         break;
 306
 307 #ifndef CONFIG_ARM64_64K_PAGES
 308                 /*
 309                  * For the first memory bank align the start address and
 310                  * current memblock limit to prevent create_mapping() from
 311                  * allocating pte page tables from unmapped memory.
 312                  * When 64K pages are enabled, the pte page table for the
 313                  * first PGDIR_SIZE is already present in swapper_pg_dir.
 314                  */
 315                 if (start < limit)
 316                         start = ALIGN(start, PMD_SIZE);
 317                 if (end < limit) {
 318                         limit = end & PMD_MASK;
 319                         memblock_set_current_limit(limit);
 320                 }
 321 #endif
 322
 323                 create_mapping(start, __phys_to_virt(start), end - start);
 324         }
 325
 326         /* Limit no longer required. */
 327         memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
 328 }
 329
 330 /*
 331  * paging_init() sets up the page tables, initialises the zone memory
 332  * maps and sets up the zero page.
 333  */
 334 void __init paging_init(void)
 335 {
 336         void *zero_page;
 337
 338         map_mem();
 339
 340         /*
 341          * Finally flush the caches and tlb to ensure that we're in a
 342          * consistent state.
 343          */
 344         flush_cache_all();
 345         flush_tlb_all();
 346
 347         /* allocate the zero page. */
 348         zero_page = early_alloc(PAGE_SIZE);
 349
 350         bootmem_init();
 351
 352         empty_zero_page = virt_to_page(zero_page);
 353
 354         /*
 355          * TTBR0 is only used for the identity mapping at this stage. Make it
 356          * point to zero page to avoid speculatively fetching new entries.
 357          */
 358         cpu_set_reserved_ttbr0();
 359         flush_tlb_all();
 360 }
 361
 362 /*
 363  * Enable the identity mapping to allow the MMU disabling.
 364  */
 365 void setup_mm_for_reboot(void)
 366 {
 367         cpu_switch_mm(idmap_pg_dir, &init_mm);
 368         flush_tlb_all();
 369 }
 370
 371 /*
 372  * Check whether a kernel address is valid (derived from arch/x86/).
 373  */
 374 int kern_addr_valid(unsigned long addr)
 375 {
 376         pgd_t *pgd;
 377         pud_t *pud;
 378         pmd_t *pmd;
 379         pte_t *pte;
 380
 381         if ((((long)addr) >> VA_BITS) != -1UL)
 382                 return 0;
 383
 384         pgd = pgd_offset_k(addr);
 385         if (pgd_none(*pgd))
 386                 return 0;
 387
 388         pud = pud_offset(pgd, addr);
 389         if (pud_none(*pud))
 390                 return 0;
 391
 392         if (pud_sect(*pud))
 393                 return pfn_valid(pud_pfn(*pud));
 394
 395         pmd = pmd_offset(pud, addr);
 396         if (pmd_none(*pmd))
 397                 return 0;
 398
 399         if (pmd_sect(*pmd))
 400                 return pfn_valid(pmd_pfn(*pmd));
 401
 402         pte = pte_offset_kernel(pmd, addr);
 403         if (pte_none(*pte))
 404                 return 0;
 405
 406         return pfn_valid(pte_pfn(*pte));
 407 }
 408 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 409 #ifdef CONFIG_ARM64_64K_PAGES
 410 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 411 {
 412         return vmemmap_populate_basepages(start, end, node);
 413 }
 414 #else   /* !CONFIG_ARM64_64K_PAGES */
 415 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 416 {
 417         unsigned long addr = start;
 418         unsigned long next;
 419         pgd_t *pgd;
 420         pud_t *pud;
 421         pmd_t *pmd;
 422
 423         do {
 424                 next = pmd_addr_end(addr, end);
 425
 426                 pgd = vmemmap_pgd_populate(addr, node);
 427                 if (!pgd)
 428                         return -ENOMEM;
 429
 430                 pud = vmemmap_pud_populate(pgd, addr, node);
 431                 if (!pud)
 432                         return -ENOMEM;
 433
 434                 pmd = pmd_offset(pud, addr);
 435                 if (pmd_none(*pmd)) {
 436                         void *p = NULL;
 437
 438                         p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 439                         if (!p)
 440                                 return -ENOMEM;
 441
 442                         set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
 443                 } else
 444                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 445         } while (addr = next, addr != end);
 446
 447         return 0;
 448 }
 449 #endif  /* CONFIG_ARM64_64K_PAGES */
 450 void vmemmap_free(unsigned long start, unsigned long end)
 451 {
 452 }
 453 #endif  /* CONFIG_SPARSEMEM_VMEMMAP */