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/memblock.h>
  36 #include <asm/mmu_context.h>
  37
  38 #include "mm.h"
  39
  40 /*
  41  * Empty_zero_page is a special page that is used for zero-initialized data
  42  * and COW.
  43  */
  44 struct page *empty_zero_page;
  45 EXPORT_SYMBOL(empty_zero_page);
  46
  47 struct cachepolicy {
  48         const char      policy[16];
  49         u64             mair;
  50         u64             tcr;
  51 };
  52
  53 static struct cachepolicy cache_policies[] __initdata = {
  54         {
  55                 .policy         = "uncached",
  56                 .mair           = 0x44,                 /* inner, outer non-cacheable */
  57                 .tcr            = TCR_IRGN_NC | TCR_ORGN_NC,
  58         }, {
  59                 .policy         = "writethrough",
  60                 .mair           = 0xaa,                 /* inner, outer write-through, read-allocate */
  61                 .tcr            = TCR_IRGN_WT | TCR_ORGN_WT,
  62         }, {
  63                 .policy         = "writeback",
  64                 .mair           = 0xee,                 /* inner, outer write-back, read-allocate */
  65                 .tcr            = TCR_IRGN_WBnWA | TCR_ORGN_WBnWA,
  66         }
  67 };
  68
  69 /*
  70  * These are useful for identifying cache coherency problems by allowing the
  71  * cache or the cache and writebuffer to be turned off. It changes the Normal
  72  * memory caching attributes in the MAIR_EL1 register.
  73  */
  74 static int __init early_cachepolicy(char *p)
  75 {
  76         int i;
  77         u64 tmp;
  78
  79         for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
  80                 int len = strlen(cache_policies[i].policy);
  81
  82                 if (memcmp(p, cache_policies[i].policy, len) == 0)
  83                         break;
  84         }
  85         if (i == ARRAY_SIZE(cache_policies)) {
  86                 pr_err("ERROR: unknown or unsupported cache policy: %s\n", p);
  87                 return 0;
  88         }
  89
  90         flush_cache_all();
  91
  92         /*
  93          * Modify MT_NORMAL attributes in MAIR_EL1.
  94          */
  95         asm volatile(
  96         "       mrs     %0, mair_el1\n"
  97         "       bfi     %0, %1, %2, #8\n"
  98         "       msr     mair_el1, %0\n"
  99         "       isb\n"
 100         : "=&r" (tmp)
 101         : "r" (cache_policies[i].mair), "i" (MT_NORMAL * 8));
 102
 103         /*
 104          * Modify TCR PTW cacheability attributes.
 105          */
 106         asm volatile(
 107         "       mrs     %0, tcr_el1\n"
 108         "       bic     %0, %0, %2\n"
 109         "       orr     %0, %0, %1\n"
 110         "       msr     tcr_el1, %0\n"
 111         "       isb\n"
 112         : "=&r" (tmp)
 113         : "r" (cache_policies[i].tcr), "r" (TCR_IRGN_MASK | TCR_ORGN_MASK));
 114
 115         flush_cache_all();
 116
 117         return 0;
 118 }
 119 early_param("cachepolicy", early_cachepolicy);
 120
 121 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 122                               unsigned long size, pgprot_t vma_prot)
 123 {
 124         if (!pfn_valid(pfn))
 125                 return pgprot_noncached(vma_prot);
 126         else if (file->f_flags & O_SYNC)
 127                 return pgprot_writecombine(vma_prot);
 128         return vma_prot;
 129 }
 130 EXPORT_SYMBOL(phys_mem_access_prot);
 131
 132 static void __init *early_alloc(unsigned long sz)
 133 {
 134         void *ptr = __va(memblock_alloc(sz, sz));
 135         memset(ptr, 0, sz);
 136         return ptr;
 137 }
 138
 139 static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
 140                                   unsigned long end, unsigned long pfn,
 141                                   pgprot_t prot)
 142 {
 143         pte_t *pte;
 144
 145         if (pmd_none(*pmd)) {
 146                 pte = early_alloc(PTRS_PER_PTE * sizeof(pte_t));
 147                 __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
 148         }
 149         BUG_ON(pmd_bad(*pmd));
 150
 151         pte = pte_offset_kernel(pmd, addr);
 152         do {
 153                 set_pte(pte, pfn_pte(pfn, prot));
 154                 pfn++;
 155         } while (pte++, addr += PAGE_SIZE, addr != end);
 156 }
 157
 158 static void __init alloc_init_pmd(pud_t *pud, unsigned long addr,
 159                                   unsigned long end, phys_addr_t phys,
 160                                   int map_io)
 161 {
 162         pmd_t *pmd;
 163         unsigned long next;
 164         pmdval_t prot_sect;
 165         pgprot_t prot_pte;
 166
 167         if (map_io) {
 168                 prot_sect = PROT_SECT_DEVICE_nGnRE;
 169                 prot_pte = __pgprot(PROT_DEVICE_nGnRE);
 170         } else {
 171                 prot_sect = PROT_SECT_NORMAL_EXEC;
 172                 prot_pte = PAGE_KERNEL_EXEC;
 173         }
 174
 175         /*
 176          * Check for initial section mappings in the pgd/pud and remove them.
 177          */
 178         if (pud_none(*pud) || pud_bad(*pud)) {
 179                 pmd = early_alloc(PTRS_PER_PMD * sizeof(pmd_t));
 180                 pud_populate(&init_mm, pud, pmd);
 181         }
 182
 183         pmd = pmd_offset(pud, addr);
 184         do {
 185                 next = pmd_addr_end(addr, end);
 186                 /* try section mapping first */
 187                 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
 188                         pmd_t old_pmd =*pmd;
 189                         set_pmd(pmd, __pmd(phys | prot_sect));
 190                         /*
 191                          * Check for previous table entries created during
 192                          * boot (__create_page_tables) and flush them.
 193                          */
 194                         if (!pmd_none(old_pmd))
 195                                 flush_tlb_all();
 196                 } else {
 197                         alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
 198                                        prot_pte);
 199                 }
 200                 phys += next - addr;
 201         } while (pmd++, addr = next, addr != end);
 202 }
 203
 204 static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr,
 205                                   unsigned long end, unsigned long phys,
 206                                   int map_io)
 207 {
 208         pud_t *pud;
 209         unsigned long next;
 210
 211         if (pgd_none(*pgd)) {
 212                 pud = early_alloc(PTRS_PER_PUD * sizeof(pud_t));
 213                 pgd_populate(&init_mm, pgd, pud);
 214         }
 215         BUG_ON(pgd_bad(*pgd));
 216
 217         pud = pud_offset(pgd, addr);
 218         do {
 219                 next = pud_addr_end(addr, end);
 220
 221                 /*
 222                  * For 4K granule only, attempt to put down a 1GB block
 223                  */
 224                 if (!map_io && (PAGE_SHIFT == 12) &&
 225                     ((addr | next | phys) & ~PUD_MASK) == 0) {
 226                         pud_t old_pud = *pud;
 227                         set_pud(pud, __pud(phys | PROT_SECT_NORMAL_EXEC));
 228
 229                         /*
 230                          * If we have an old value for a pud, it will
 231                          * be pointing to a pmd table that we no longer
 232                          * need (from swapper_pg_dir).
 233                          *
 234                          * Look up the old pmd table and free it.
 235                          */
 236                         if (!pud_none(old_pud)) {
 237                                 phys_addr_t table = __pa(pmd_offset(&old_pud, 0));
 238                                 memblock_free(table, PAGE_SIZE);
 239                                 flush_tlb_all();
 240                         }
 241                 } else {
 242                         alloc_init_pmd(pud, addr, next, phys, map_io);
 243                 }
 244                 phys += next - addr;
 245         } while (pud++, addr = next, addr != end);
 246 }
 247
 248 /*
 249  * Create the page directory entries and any necessary page tables for the
 250  * mapping specified by 'md'.
 251  */
 252 static void __init __create_mapping(pgd_t *pgd, phys_addr_t phys,
 253                                     unsigned long virt, phys_addr_t size,
 254                                     int map_io)
 255 {
 256         unsigned long addr, length, end, next;
 257
 258         addr = virt & PAGE_MASK;
 259         length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
 260
 261         end = addr + length;
 262         do {
 263                 next = pgd_addr_end(addr, end);
 264                 alloc_init_pud(pgd, addr, next, phys, map_io);
 265                 phys += next - addr;
 266         } while (pgd++, addr = next, addr != end);
 267 }
 268
 269 static void __init create_mapping(phys_addr_t phys, unsigned long virt,
 270                                   phys_addr_t size)
 271 {
 272         if (virt < VMALLOC_START) {
 273                 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
 274                         &phys, virt);
 275                 return;
 276         }
 277         __create_mapping(pgd_offset_k(virt & PAGE_MASK), phys, virt, size, 0);
 278 }
 279
 280 void __init create_id_mapping(phys_addr_t addr, phys_addr_t size, int map_io)
 281 {
 282         if ((addr >> PGDIR_SHIFT) >= ARRAY_SIZE(idmap_pg_dir)) {
 283                 pr_warn("BUG: not creating id mapping for %pa\n", &addr);
 284                 return;
 285         }
 286         __create_mapping(&idmap_pg_dir[pgd_index(addr)],
 287                          addr, addr, size, map_io);
 288 }
 289
 290 static void __init map_mem(void)
 291 {
 292         struct memblock_region *reg;
 293         phys_addr_t limit;
 294
 295         /*
 296          * Temporarily limit the memblock range. We need to do this as
 297          * create_mapping requires puds, pmds and ptes to be allocated from
 298          * memory addressable from the initial direct kernel mapping.
 299          *
 300          * The initial direct kernel mapping, located at swapper_pg_dir, gives
 301          * us PUD_SIZE (4K pages) or PMD_SIZE (64K pages) memory starting from
 302          * PHYS_OFFSET (which must be aligned to 2MB as per
 303          * Documentation/arm64/booting.txt).
 304          */
 305         if (IS_ENABLED(CONFIG_ARM64_64K_PAGES))
 306                 limit = PHYS_OFFSET + PMD_SIZE;
 307         else
 308                 limit = PHYS_OFFSET + PUD_SIZE;
 309         memblock_set_current_limit(limit);
 310
 311         /* map all the memory banks */
 312         for_each_memblock(memory, reg) {
 313                 phys_addr_t start = reg->base;
 314                 phys_addr_t end = start + reg->size;
 315
 316                 if (start >= end)
 317                         break;
 318
 319 #ifndef CONFIG_ARM64_64K_PAGES
 320                 /*
 321                  * For the first memory bank align the start address and
 322                  * current memblock limit to prevent create_mapping() from
 323                  * allocating pte page tables from unmapped memory.
 324                  * When 64K pages are enabled, the pte page table for the
 325                  * first PGDIR_SIZE is already present in swapper_pg_dir.
 326                  */
 327                 if (start < limit)
 328                         start = ALIGN(start, PMD_SIZE);
 329                 if (end < limit) {
 330                         limit = end & PMD_MASK;
 331                         memblock_set_current_limit(limit);
 332                 }
 333 #endif
 334
 335                 create_mapping(start, __phys_to_virt(start), end - start);
 336         }
 337
 338         /* Limit no longer required. */
 339         memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
 340 }
 341
 342 /*
 343  * paging_init() sets up the page tables, initialises the zone memory
 344  * maps and sets up the zero page.
 345  */
 346 void __init paging_init(void)
 347 {
 348         void *zero_page;
 349
 350         map_mem();
 351
 352         /*
 353          * Finally flush the caches and tlb to ensure that we're in a
 354          * consistent state.
 355          */
 356         flush_cache_all();
 357         flush_tlb_all();
 358
 359         /* allocate the zero page. */
 360         zero_page = early_alloc(PAGE_SIZE);
 361
 362         bootmem_init();
 363
 364         empty_zero_page = virt_to_page(zero_page);
 365
 366         /*
 367          * TTBR0 is only used for the identity mapping at this stage. Make it
 368          * point to zero page to avoid speculatively fetching new entries.
 369          */
 370         cpu_set_reserved_ttbr0();
 371         flush_tlb_all();
 372 }
 373
 374 /*
 375  * Enable the identity mapping to allow the MMU disabling.
 376  */
 377 void setup_mm_for_reboot(void)
 378 {
 379         cpu_switch_mm(idmap_pg_dir, &init_mm);
 380         flush_tlb_all();
 381 }
 382
 383 /*
 384  * Check whether a kernel address is valid (derived from arch/x86/).
 385  */
 386 int kern_addr_valid(unsigned long addr)
 387 {
 388         pgd_t *pgd;
 389         pud_t *pud;
 390         pmd_t *pmd;
 391         pte_t *pte;
 392
 393         if ((((long)addr) >> VA_BITS) != -1UL)
 394                 return 0;
 395
 396         pgd = pgd_offset_k(addr);
 397         if (pgd_none(*pgd))
 398                 return 0;
 399
 400         pud = pud_offset(pgd, addr);
 401         if (pud_none(*pud))
 402                 return 0;
 403
 404         if (pud_sect(*pud))
 405                 return pfn_valid(pud_pfn(*pud));
 406
 407         pmd = pmd_offset(pud, addr);
 408         if (pmd_none(*pmd))
 409                 return 0;
 410
 411         if (pmd_sect(*pmd))
 412                 return pfn_valid(pmd_pfn(*pmd));
 413
 414         pte = pte_offset_kernel(pmd, addr);
 415         if (pte_none(*pte))
 416                 return 0;
 417
 418         return pfn_valid(pte_pfn(*pte));
 419 }
 420 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 421 #ifdef CONFIG_ARM64_64K_PAGES
 422 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 423 {
 424         return vmemmap_populate_basepages(start, end, node);
 425 }
 426 #else   /* !CONFIG_ARM64_64K_PAGES */
 427 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 428 {
 429         unsigned long addr = start;
 430         unsigned long next;
 431         pgd_t *pgd;
 432         pud_t *pud;
 433         pmd_t *pmd;
 434
 435         do {
 436                 next = pmd_addr_end(addr, end);
 437
 438                 pgd = vmemmap_pgd_populate(addr, node);
 439                 if (!pgd)
 440                         return -ENOMEM;
 441
 442                 pud = vmemmap_pud_populate(pgd, addr, node);
 443                 if (!pud)
 444                         return -ENOMEM;
 445
 446                 pmd = pmd_offset(pud, addr);
 447                 if (pmd_none(*pmd)) {
 448                         void *p = NULL;
 449
 450                         p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 451                         if (!p)
 452                                 return -ENOMEM;
 453
 454                         set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
 455                 } else
 456                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 457         } while (addr = next, addr != end);
 458
 459         return 0;
 460 }
 461 #endif  /* CONFIG_ARM64_64K_PAGES */
 462 void vmemmap_free(unsigned long start, unsigned long end)
 463 {
 464 }
 465 #endif  /* CONFIG_SPARSEMEM_VMEMMAP */