linux/arch/arm64/mm/mmu.c
<<
>>
Prefs
   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 */
  43struct page *empty_zero_page;
  44EXPORT_SYMBOL(empty_zero_page);
  45
  46pgprot_t pgprot_default;
  47EXPORT_SYMBOL(pgprot_default);
  48
  49static pmdval_t prot_sect_kernel;
  50
  51struct cachepolicy {
  52        const char      policy[16];
  53        u64             mair;
  54        u64             tcr;
  55};
  56
  57static struct cachepolicy cache_policies[] __initdata = {
  58        {
  59                .policy         = "uncached",
  60                .mair           = 0x44,                 /* inner, outer non-cacheable */
  61                .tcr            = TCR_IRGN_NC | TCR_ORGN_NC,
  62        }, {
  63                .policy         = "writethrough",
  64                .mair           = 0xaa,                 /* inner, outer write-through, read-allocate */
  65                .tcr            = TCR_IRGN_WT | TCR_ORGN_WT,
  66        }, {
  67                .policy         = "writeback",
  68                .mair           = 0xee,                 /* inner, outer write-back, read-allocate */
  69                .tcr            = TCR_IRGN_WBnWA | TCR_ORGN_WBnWA,
  70        }
  71};
  72
  73/*
  74 * These are useful for identifying cache coherency problems by allowing the
  75 * cache or the cache and writebuffer to be turned off. It changes the Normal
  76 * memory caching attributes in the MAIR_EL1 register.
  77 */
  78static int __init early_cachepolicy(char *p)
  79{
  80        int i;
  81        u64 tmp;
  82
  83        for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
  84                int len = strlen(cache_policies[i].policy);
  85
  86                if (memcmp(p, cache_policies[i].policy, len) == 0)
  87                        break;
  88        }
  89        if (i == ARRAY_SIZE(cache_policies)) {
  90                pr_err("ERROR: unknown or unsupported cache policy: %s\n", p);
  91                return 0;
  92        }
  93
  94        flush_cache_all();
  95
  96        /*
  97         * Modify MT_NORMAL attributes in MAIR_EL1.
  98         */
  99        asm volatile(
 100        "       mrs     %0, mair_el1\n"
 101        "       bfi     %0, %1, #%2, #8\n"
 102        "       msr     mair_el1, %0\n"
 103        "       isb\n"
 104        : "=&r" (tmp)
 105        : "r" (cache_policies[i].mair), "i" (MT_NORMAL * 8));
 106
 107        /*
 108         * Modify TCR PTW cacheability attributes.
 109         */
 110        asm volatile(
 111        "       mrs     %0, tcr_el1\n"
 112        "       bic     %0, %0, %2\n"
 113        "       orr     %0, %0, %1\n"
 114        "       msr     tcr_el1, %0\n"
 115        "       isb\n"
 116        : "=&r" (tmp)
 117        : "r" (cache_policies[i].tcr), "r" (TCR_IRGN_MASK | TCR_ORGN_MASK));
 118
 119        flush_cache_all();
 120
 121        return 0;
 122}
 123early_param("cachepolicy", early_cachepolicy);
 124
 125/*
 126 * Adjust the PMD section entries according to the CPU in use.
 127 */
 128static void __init init_mem_pgprot(void)
 129{
 130        pteval_t default_pgprot;
 131        int i;
 132
 133        default_pgprot = PTE_ATTRINDX(MT_NORMAL);
 134        prot_sect_kernel = PMD_TYPE_SECT | PMD_SECT_AF | PMD_ATTRINDX(MT_NORMAL);
 135
 136#ifdef CONFIG_SMP
 137        /*
 138         * Mark memory with the "shared" attribute for SMP systems
 139         */
 140        default_pgprot |= PTE_SHARED;
 141        prot_sect_kernel |= PMD_SECT_S;
 142#endif
 143
 144        for (i = 0; i < 16; i++) {
 145                unsigned long v = pgprot_val(protection_map[i]);
 146                protection_map[i] = __pgprot(v | default_pgprot);
 147        }
 148
 149        pgprot_default = __pgprot(PTE_TYPE_PAGE | PTE_AF | default_pgprot);
 150}
 151
 152pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 153                              unsigned long size, pgprot_t vma_prot)
 154{
 155        if (!pfn_valid(pfn))
 156                return pgprot_noncached(vma_prot);
 157        else if (file->f_flags & O_SYNC)
 158                return pgprot_writecombine(vma_prot);
 159        return vma_prot;
 160}
 161EXPORT_SYMBOL(phys_mem_access_prot);
 162
 163static void __init *early_alloc(unsigned long sz)
 164{
 165        void *ptr = __va(memblock_alloc(sz, sz));
 166        memset(ptr, 0, sz);
 167        return ptr;
 168}
 169
 170static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
 171                                  unsigned long end, unsigned long pfn)
 172{
 173        pte_t *pte;
 174
 175        if (pmd_none(*pmd)) {
 176                pte = early_alloc(PTRS_PER_PTE * sizeof(pte_t));
 177                __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
 178        }
 179        BUG_ON(pmd_bad(*pmd));
 180
 181        pte = pte_offset_kernel(pmd, addr);
 182        do {
 183                set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 184                pfn++;
 185        } while (pte++, addr += PAGE_SIZE, addr != end);
 186}
 187
 188static void __init alloc_init_pmd(pud_t *pud, unsigned long addr,
 189                                  unsigned long end, phys_addr_t phys)
 190{
 191        pmd_t *pmd;
 192        unsigned long next;
 193
 194        /*
 195         * Check for initial section mappings in the pgd/pud and remove them.
 196         */
 197        if (pud_none(*pud) || pud_bad(*pud)) {
 198                pmd = early_alloc(PTRS_PER_PMD * sizeof(pmd_t));
 199                pud_populate(&init_mm, pud, pmd);
 200        }
 201
 202        pmd = pmd_offset(pud, addr);
 203        do {
 204                next = pmd_addr_end(addr, end);
 205                /* try section mapping first */
 206                if (((addr | next | phys) & ~SECTION_MASK) == 0)
 207                        set_pmd(pmd, __pmd(phys | prot_sect_kernel));
 208                else
 209                        alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys));
 210                phys += next - addr;
 211        } while (pmd++, addr = next, addr != end);
 212}
 213
 214static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr,
 215                                  unsigned long end, unsigned long phys)
 216{
 217        pud_t *pud = pud_offset(pgd, addr);
 218        unsigned long next;
 219
 220        do {
 221                next = pud_addr_end(addr, end);
 222                alloc_init_pmd(pud, addr, next, phys);
 223                phys += next - addr;
 224        } while (pud++, addr = next, addr != end);
 225}
 226
 227/*
 228 * Create the page directory entries and any necessary page tables for the
 229 * mapping specified by 'md'.
 230 */
 231static void __init create_mapping(phys_addr_t phys, unsigned long virt,
 232                                  phys_addr_t size)
 233{
 234        unsigned long addr, length, end, next;
 235        pgd_t *pgd;
 236
 237        if (virt < VMALLOC_START) {
 238                pr_warning("BUG: not creating mapping for 0x%016llx at 0x%016lx - outside kernel range\n",
 239                           phys, virt);
 240                return;
 241        }
 242
 243        addr = virt & PAGE_MASK;
 244        length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
 245
 246        pgd = pgd_offset_k(addr);
 247        end = addr + length;
 248        do {
 249                next = pgd_addr_end(addr, end);
 250                alloc_init_pud(pgd, addr, next, phys);
 251                phys += next - addr;
 252        } while (pgd++, addr = next, addr != end);
 253}
 254
 255#ifdef CONFIG_EARLY_PRINTK
 256/*
 257 * Create an early I/O mapping using the pgd/pmd entries already populated
 258 * in head.S as this function is called too early to allocated any memory. The
 259 * mapping size is 2MB with 4KB pages or 64KB or 64KB pages.
 260 */
 261void __iomem * __init early_io_map(phys_addr_t phys, unsigned long virt)
 262{
 263        unsigned long size, mask;
 264        bool page64k = IS_ENABLED(CONFIG_ARM64_64K_PAGES);
 265        pgd_t *pgd;
 266        pud_t *pud;
 267        pmd_t *pmd;
 268        pte_t *pte;
 269
 270        /*
 271         * No early pte entries with !ARM64_64K_PAGES configuration, so using
 272         * sections (pmd).
 273         */
 274        size = page64k ? PAGE_SIZE : SECTION_SIZE;
 275        mask = ~(size - 1);
 276
 277        pgd = pgd_offset_k(virt);
 278        pud = pud_offset(pgd, virt);
 279        if (pud_none(*pud))
 280                return NULL;
 281        pmd = pmd_offset(pud, virt);
 282
 283        if (page64k) {
 284                if (pmd_none(*pmd))
 285                        return NULL;
 286                pte = pte_offset_kernel(pmd, virt);
 287                set_pte(pte, __pte((phys & mask) | PROT_DEVICE_nGnRE));
 288        } else {
 289                set_pmd(pmd, __pmd((phys & mask) | PROT_SECT_DEVICE_nGnRE));
 290        }
 291
 292        return (void __iomem *)((virt & mask) + (phys & ~mask));
 293}
 294#endif
 295
 296static void __init map_mem(void)
 297{
 298        struct memblock_region *reg;
 299
 300        /*
 301         * Temporarily limit the memblock range. We need to do this as
 302         * create_mapping requires puds, pmds and ptes to be allocated from
 303         * memory addressable from the initial direct kernel mapping.
 304         *
 305         * The initial direct kernel mapping, located at swapper_pg_dir,
 306         * gives us PGDIR_SIZE memory starting from PHYS_OFFSET (aligned).
 307         */
 308        memblock_set_current_limit((PHYS_OFFSET & PGDIR_MASK) + PGDIR_SIZE);
 309
 310        /* map all the memory banks */
 311        for_each_memblock(memory, reg) {
 312                phys_addr_t start = reg->base;
 313                phys_addr_t end = start + reg->size;
 314
 315                if (start >= end)
 316                        break;
 317
 318                create_mapping(start, __phys_to_virt(start), end - start);
 319        }
 320
 321        /* Limit no longer required. */
 322        memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
 323}
 324
 325/*
 326 * paging_init() sets up the page tables, initialises the zone memory
 327 * maps and sets up the zero page.
 328 */
 329void __init paging_init(void)
 330{
 331        void *zero_page;
 332
 333        init_mem_pgprot();
 334        map_mem();
 335
 336        /*
 337         * Finally flush the caches and tlb to ensure that we're in a
 338         * consistent state.
 339         */
 340        flush_cache_all();
 341        flush_tlb_all();
 342
 343        /* allocate the zero page. */
 344        zero_page = early_alloc(PAGE_SIZE);
 345
 346        bootmem_init();
 347
 348        empty_zero_page = virt_to_page(zero_page);
 349
 350        /*
 351         * TTBR0 is only used for the identity mapping at this stage. Make it
 352         * point to zero page to avoid speculatively fetching new entries.
 353         */
 354        cpu_set_reserved_ttbr0();
 355        flush_tlb_all();
 356}
 357
 358/*
 359 * Enable the identity mapping to allow the MMU disabling.
 360 */
 361void setup_mm_for_reboot(void)
 362{
 363        cpu_switch_mm(idmap_pg_dir, &init_mm);
 364        flush_tlb_all();
 365}
 366
 367/*
 368 * Check whether a kernel address is valid (derived from arch/x86/).
 369 */
 370int kern_addr_valid(unsigned long addr)
 371{
 372        pgd_t *pgd;
 373        pud_t *pud;
 374        pmd_t *pmd;
 375        pte_t *pte;
 376
 377        if ((((long)addr) >> VA_BITS) != -1UL)
 378                return 0;
 379
 380        pgd = pgd_offset_k(addr);
 381        if (pgd_none(*pgd))
 382                return 0;
 383
 384        pud = pud_offset(pgd, addr);
 385        if (pud_none(*pud))
 386                return 0;
 387
 388        pmd = pmd_offset(pud, addr);
 389        if (pmd_none(*pmd))
 390                return 0;
 391
 392        pte = pte_offset_kernel(pmd, addr);
 393        if (pte_none(*pte))
 394                return 0;
 395
 396        return pfn_valid(pte_pfn(*pte));
 397}
 398#ifdef CONFIG_SPARSEMEM_VMEMMAP
 399#ifdef CONFIG_ARM64_64K_PAGES
 400int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 401{
 402        return vmemmap_populate_basepages(start, end, node);
 403}
 404#else   /* !CONFIG_ARM64_64K_PAGES */
 405int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 406{
 407        unsigned long addr = start;
 408        unsigned long next;
 409        pgd_t *pgd;
 410        pud_t *pud;
 411        pmd_t *pmd;
 412
 413        do {
 414                next = pmd_addr_end(addr, end);
 415
 416                pgd = vmemmap_pgd_populate(addr, node);
 417                if (!pgd)
 418                        return -ENOMEM;
 419
 420                pud = vmemmap_pud_populate(pgd, addr, node);
 421                if (!pud)
 422                        return -ENOMEM;
 423
 424                pmd = pmd_offset(pud, addr);
 425                if (pmd_none(*pmd)) {
 426                        void *p = NULL;
 427
 428                        p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 429                        if (!p)
 430                                return -ENOMEM;
 431
 432                        set_pmd(pmd, __pmd(__pa(p) | prot_sect_kernel));
 433                } else
 434                        vmemmap_verify((pte_t *)pmd, node, addr, next);
 435        } while (addr = next, addr != end);
 436
 437        return 0;
 438}
 439#endif  /* CONFIG_ARM64_64K_PAGES */
 440void vmemmap_free(unsigned long start, unsigned long end)
 441{
 442}
 443#endif  /* CONFIG_SPARSEMEM_VMEMMAP */
 444