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/libfdt.h>
  25 #include <linux/mman.h>
  26 #include <linux/nodemask.h>
  27 #include <linux/memblock.h>
  28 #include <linux/fs.h>
  29 #include <linux/io.h>
  30 #include <linux/slab.h>
  31 #include <linux/stop_machine.h>
  32
  33 #include <asm/barrier.h>
  34 #include <asm/cputype.h>
  35 #include <asm/fixmap.h>
  36 #include <asm/kasan.h>
  37 #include <asm/kernel-pgtable.h>
  38 #include <asm/sections.h>
  39 #include <asm/setup.h>
  40 #include <asm/sizes.h>
  41 #include <asm/tlb.h>
  42 #include <asm/memblock.h>
  43 #include <asm/mmu_context.h>
  44
  45 #include "mm.h"
  46
  47 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
  48
  49 u64 kimage_voffset __read_mostly;
  50 EXPORT_SYMBOL(kimage_voffset);
  51
  52 /*
  53  * Empty_zero_page is a special page that is used for zero-initialized data
  54  * and COW.
  55  */
  56 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
  57 EXPORT_SYMBOL(empty_zero_page);
  58
  59 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
  60 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
  61 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
  62
  63 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  64                               unsigned long size, pgprot_t vma_prot)
  65 {
  66         if (!pfn_valid(pfn))
  67                 return pgprot_noncached(vma_prot);
  68         else if (file->f_flags & O_SYNC)
  69                 return pgprot_writecombine(vma_prot);
  70         return vma_prot;
  71 }
  72 EXPORT_SYMBOL(phys_mem_access_prot);
  73
  74 static phys_addr_t __init early_pgtable_alloc(void)
  75 {
  76         phys_addr_t phys;
  77         void *ptr;
  78
  79         phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
  80
  81         /*
  82          * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
  83          * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
  84          * any level of table.
  85          */
  86         ptr = pte_set_fixmap(phys);
  87
  88         memset(ptr, 0, PAGE_SIZE);
  89
  90         /*
  91          * Implicit barriers also ensure the zeroed page is visible to the page
  92          * table walker
  93          */
  94         pte_clear_fixmap();
  95
  96         return phys;
  97 }
  98
  99 /*
 100  * remap a PMD into pages
 101  */
 102 static void split_pmd(pmd_t *pmd, pte_t *pte)
 103 {
 104         unsigned long pfn = pmd_pfn(*pmd);
 105         int i = 0;
 106
 107         do {
 108                 /*
 109                  * Need to have the least restrictive permissions available
 110                  * permissions will be fixed up later
 111                  */
 112                 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 113                 pfn++;
 114         } while (pte++, i++, i < PTRS_PER_PTE);
 115 }
 116
 117 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
 118                                   unsigned long end, unsigned long pfn,
 119                                   pgprot_t prot,
 120                                   phys_addr_t (*pgtable_alloc)(void))
 121 {
 122         pte_t *pte;
 123
 124         if (pmd_none(*pmd) || pmd_sect(*pmd)) {
 125                 phys_addr_t pte_phys;
 126                 BUG_ON(!pgtable_alloc);
 127                 pte_phys = pgtable_alloc();
 128                 pte = pte_set_fixmap(pte_phys);
 129                 if (pmd_sect(*pmd))
 130                         split_pmd(pmd, pte);
 131                 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
 132                 flush_tlb_all();
 133                 pte_clear_fixmap();
 134         }
 135         BUG_ON(pmd_bad(*pmd));
 136
 137         pte = pte_set_fixmap_offset(pmd, addr);
 138         do {
 139                 set_pte(pte, pfn_pte(pfn, prot));
 140                 pfn++;
 141         } while (pte++, addr += PAGE_SIZE, addr != end);
 142
 143         pte_clear_fixmap();
 144 }
 145
 146 static void split_pud(pud_t *old_pud, pmd_t *pmd)
 147 {
 148         unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
 149         pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
 150         int i = 0;
 151
 152         do {
 153                 set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
 154                 addr += PMD_SIZE;
 155         } while (pmd++, i++, i < PTRS_PER_PMD);
 156 }
 157
 158 static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
 159                                   phys_addr_t phys, pgprot_t prot,
 160                                   phys_addr_t (*pgtable_alloc)(void),
 161                                   bool allow_block_mappings)
 162 {
 163         pmd_t *pmd;
 164         unsigned long next;
 165
 166         /*
 167          * Check for initial section mappings in the pgd/pud and remove them.
 168          */
 169         if (pud_none(*pud) || pud_sect(*pud)) {
 170                 phys_addr_t pmd_phys;
 171                 BUG_ON(!pgtable_alloc);
 172                 pmd_phys = pgtable_alloc();
 173                 pmd = pmd_set_fixmap(pmd_phys);
 174                 if (pud_sect(*pud)) {
 175                         /*
 176                          * need to have the 1G of mappings continue to be
 177                          * present
 178                          */
 179                         split_pud(pud, pmd);
 180                 }
 181                 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
 182                 flush_tlb_all();
 183                 pmd_clear_fixmap();
 184         }
 185         BUG_ON(pud_bad(*pud));
 186
 187         pmd = pmd_set_fixmap_offset(pud, addr);
 188         do {
 189                 next = pmd_addr_end(addr, end);
 190                 /* try section mapping first */
 191                 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
 192                       allow_block_mappings) {
 193                         pmd_t old_pmd =*pmd;
 194                         pmd_set_huge(pmd, phys, prot);
 195                         /*
 196                          * Check for previous table entries created during
 197                          * boot (__create_page_tables) and flush them.
 198                          */
 199                         if (!pmd_none(old_pmd)) {
 200                                 flush_tlb_all();
 201                                 if (pmd_table(old_pmd)) {
 202                                         phys_addr_t table = pmd_page_paddr(old_pmd);
 203                                         if (!WARN_ON_ONCE(slab_is_available()))
 204                                                 memblock_free(table, PAGE_SIZE);
 205                                 }
 206                         }
 207                 } else {
 208                         alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
 209                                        prot, pgtable_alloc);
 210                 }
 211                 phys += next - addr;
 212         } while (pmd++, addr = next, addr != end);
 213
 214         pmd_clear_fixmap();
 215 }
 216
 217 static inline bool use_1G_block(unsigned long addr, unsigned long next,
 218                         unsigned long phys)
 219 {
 220         if (PAGE_SHIFT != 12)
 221                 return false;
 222
 223         if (((addr | next | phys) & ~PUD_MASK) != 0)
 224                 return false;
 225
 226         return true;
 227 }
 228
 229 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
 230                                   phys_addr_t phys, pgprot_t prot,
 231                                   phys_addr_t (*pgtable_alloc)(void),
 232                                   bool allow_block_mappings)
 233 {
 234         pud_t *pud;
 235         unsigned long next;
 236
 237         if (pgd_none(*pgd)) {
 238                 phys_addr_t pud_phys;
 239                 BUG_ON(!pgtable_alloc);
 240                 pud_phys = pgtable_alloc();
 241                 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
 242         }
 243         BUG_ON(pgd_bad(*pgd));
 244
 245         pud = pud_set_fixmap_offset(pgd, addr);
 246         do {
 247                 next = pud_addr_end(addr, end);
 248
 249                 /*
 250                  * For 4K granule only, attempt to put down a 1GB block
 251                  */
 252                 if (use_1G_block(addr, next, phys) && allow_block_mappings) {
 253                         pud_t old_pud = *pud;
 254                         pud_set_huge(pud, phys, prot);
 255
 256                         /*
 257                          * If we have an old value for a pud, it will
 258                          * be pointing to a pmd table that we no longer
 259                          * need (from swapper_pg_dir).
 260                          *
 261                          * Look up the old pmd table and free it.
 262                          */
 263                         if (!pud_none(old_pud)) {
 264                                 flush_tlb_all();
 265                                 if (pud_table(old_pud)) {
 266                                         phys_addr_t table = pud_page_paddr(old_pud);
 267                                         if (!WARN_ON_ONCE(slab_is_available()))
 268                                                 memblock_free(table, PAGE_SIZE);
 269                                 }
 270                         }
 271                 } else {
 272                         alloc_init_pmd(pud, addr, next, phys, prot,
 273                                        pgtable_alloc, allow_block_mappings);
 274                 }
 275                 phys += next - addr;
 276         } while (pud++, addr = next, addr != end);
 277
 278         pud_clear_fixmap();
 279 }
 280
 281 /*
 282  * Create the page directory entries and any necessary page tables for the
 283  * mapping specified by 'md'.
 284  */
 285 static void init_pgd(pgd_t *pgd, phys_addr_t phys, unsigned long virt,
 286                                     phys_addr_t size, pgprot_t prot,
 287                                     phys_addr_t (*pgtable_alloc)(void),
 288                                     bool allow_block_mappings)
 289 {
 290         unsigned long addr, length, end, next;
 291
 292         /*
 293          * If the virtual and physical address don't have the same offset
 294          * within a page, we cannot map the region as the caller expects.
 295          */
 296         if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
 297                 return;
 298
 299         phys &= PAGE_MASK;
 300         addr = virt & PAGE_MASK;
 301         length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
 302
 303         end = addr + length;
 304         do {
 305                 next = pgd_addr_end(addr, end);
 306                 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
 307                                allow_block_mappings);
 308                 phys += next - addr;
 309         } while (pgd++, addr = next, addr != end);
 310 }
 311
 312 static phys_addr_t late_pgtable_alloc(void)
 313 {
 314         void *ptr = (void *)__get_free_page(PGALLOC_GFP);
 315         BUG_ON(!ptr);
 316
 317         /* Ensure the zeroed page is visible to the page table walker */
 318         dsb(ishst);
 319         return __pa(ptr);
 320 }
 321
 322 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
 323                                  unsigned long virt, phys_addr_t size,
 324                                  pgprot_t prot,
 325                                  phys_addr_t (*alloc)(void),
 326                                  bool allow_block_mappings)
 327 {
 328         init_pgd(pgd_offset_raw(pgdir, virt), phys, virt, size, prot, alloc,
 329                  allow_block_mappings);
 330 }
 331
 332 /*
 333  * This function can only be used to modify existing table entries,
 334  * without allocating new levels of table. Note that this permits the
 335  * creation of new section or page entries.
 336  */
 337 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
 338                                   phys_addr_t size, pgprot_t prot)
 339 {
 340         if (virt < VMALLOC_START) {
 341                 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
 342                         &phys, virt);
 343                 return;
 344         }
 345         __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, true);
 346 }
 347
 348 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
 349                                unsigned long virt, phys_addr_t size,
 350                                pgprot_t prot, bool allow_block_mappings)
 351 {
 352         __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
 353                              late_pgtable_alloc, allow_block_mappings);
 354 }
 355
 356 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
 357                                   phys_addr_t size, pgprot_t prot)
 358 {
 359         if (virt < VMALLOC_START) {
 360                 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
 361                         &phys, virt);
 362                 return;
 363         }
 364
 365         __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
 366                              late_pgtable_alloc, !debug_pagealloc_enabled());
 367 }
 368
 369 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
 370 {
 371         unsigned long kernel_start = __pa(_text);
 372         unsigned long kernel_end = __pa(__init_begin);
 373
 374         /*
 375          * Take care not to create a writable alias for the
 376          * read-only text and rodata sections of the kernel image.
 377          */
 378
 379         /* No overlap with the kernel text/rodata */
 380         if (end < kernel_start || start >= kernel_end) {
 381                 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
 382                                      end - start, PAGE_KERNEL,
 383                                      early_pgtable_alloc,
 384                                      !debug_pagealloc_enabled());
 385                 return;
 386         }
 387
 388         /*
 389          * This block overlaps the kernel text/rodata mappings.
 390          * Map the portion(s) which don't overlap.
 391          */
 392         if (start < kernel_start)
 393                 __create_pgd_mapping(pgd, start,
 394                                      __phys_to_virt(start),
 395                                      kernel_start - start, PAGE_KERNEL,
 396                                      early_pgtable_alloc,
 397                                      !debug_pagealloc_enabled());
 398         if (kernel_end < end)
 399                 __create_pgd_mapping(pgd, kernel_end,
 400                                      __phys_to_virt(kernel_end),
 401                                      end - kernel_end, PAGE_KERNEL,
 402                                      early_pgtable_alloc,
 403                                      !debug_pagealloc_enabled());
 404
 405         /*
 406          * Map the linear alias of the [_text, __init_begin) interval as
 407          * read-only/non-executable. This makes the contents of the
 408          * region accessible to subsystems such as hibernate, but
 409          * protects it from inadvertent modification or execution.
 410          */
 411         __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
 412                              kernel_end - kernel_start, PAGE_KERNEL_RO,
 413                              early_pgtable_alloc, !debug_pagealloc_enabled());
 414 }
 415
 416 static void __init map_mem(pgd_t *pgd)
 417 {
 418         struct memblock_region *reg;
 419
 420         /* map all the memory banks */
 421         for_each_memblock(memory, reg) {
 422                 phys_addr_t start = reg->base;
 423                 phys_addr_t end = start + reg->size;
 424
 425                 if (start >= end)
 426                         break;
 427                 if (memblock_is_nomap(reg))
 428                         continue;
 429
 430                 __map_memblock(pgd, start, end);
 431         }
 432 }
 433
 434 void mark_rodata_ro(void)
 435 {
 436         unsigned long section_size;
 437
 438         section_size = (unsigned long)_etext - (unsigned long)_text;
 439         create_mapping_late(__pa(_text), (unsigned long)_text,
 440                             section_size, PAGE_KERNEL_ROX);
 441         /*
 442          * mark .rodata as read only. Use __init_begin rather than __end_rodata
 443          * to cover NOTES and EXCEPTION_TABLE.
 444          */
 445         section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
 446         create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
 447                             section_size, PAGE_KERNEL_RO);
 448 }
 449
 450 void fixup_init(void)
 451 {
 452         /*
 453          * Unmap the __init region but leave the VM area in place. This
 454          * prevents the region from being reused for kernel modules, which
 455          * is not supported by kallsyms.
 456          */
 457         unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
 458 }
 459
 460 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
 461                                       pgprot_t prot, struct vm_struct *vma)
 462 {
 463         phys_addr_t pa_start = __pa(va_start);
 464         unsigned long size = va_end - va_start;
 465
 466         BUG_ON(!PAGE_ALIGNED(pa_start));
 467         BUG_ON(!PAGE_ALIGNED(size));
 468
 469         __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
 470                              early_pgtable_alloc, !debug_pagealloc_enabled());
 471
 472         vma->addr       = va_start;
 473         vma->phys_addr  = pa_start;
 474         vma->size       = size;
 475         vma->flags      = VM_MAP;
 476         vma->caller     = __builtin_return_address(0);
 477
 478         vm_area_add_early(vma);
 479 }
 480
 481 /*
 482  * Create fine-grained mappings for the kernel.
 483  */
 484 static void __init map_kernel(pgd_t *pgd)
 485 {
 486         static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
 487
 488         map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
 489         map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
 490         map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
 491                            &vmlinux_init);
 492         map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
 493
 494         if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
 495                 /*
 496                  * The fixmap falls in a separate pgd to the kernel, and doesn't
 497                  * live in the carveout for the swapper_pg_dir. We can simply
 498                  * re-use the existing dir for the fixmap.
 499                  */
 500                 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
 501                         *pgd_offset_k(FIXADDR_START));
 502         } else if (CONFIG_PGTABLE_LEVELS > 3) {
 503                 /*
 504                  * The fixmap shares its top level pgd entry with the kernel
 505                  * mapping. This can really only occur when we are running
 506                  * with 16k/4 levels, so we can simply reuse the pud level
 507                  * entry instead.
 508                  */
 509                 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
 510                 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
 511                         __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
 512                 pud_clear_fixmap();
 513         } else {
 514                 BUG();
 515         }
 516
 517         kasan_copy_shadow(pgd);
 518 }
 519
 520 /*
 521  * paging_init() sets up the page tables, initialises the zone memory
 522  * maps and sets up the zero page.
 523  */
 524 void __init paging_init(void)
 525 {
 526         phys_addr_t pgd_phys = early_pgtable_alloc();
 527         pgd_t *pgd = pgd_set_fixmap(pgd_phys);
 528
 529         map_kernel(pgd);
 530         map_mem(pgd);
 531
 532         /*
 533          * We want to reuse the original swapper_pg_dir so we don't have to
 534          * communicate the new address to non-coherent secondaries in
 535          * secondary_entry, and so cpu_switch_mm can generate the address with
 536          * adrp+add rather than a load from some global variable.
 537          *
 538          * To do this we need to go via a temporary pgd.
 539          */
 540         cpu_replace_ttbr1(__va(pgd_phys));
 541         memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
 542         cpu_replace_ttbr1(swapper_pg_dir);
 543
 544         pgd_clear_fixmap();
 545         memblock_free(pgd_phys, PAGE_SIZE);
 546
 547         /*
 548          * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
 549          * allocated with it.
 550          */
 551         memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
 552                       SWAPPER_DIR_SIZE - PAGE_SIZE);
 553 }
 554
 555 /*
 556  * Check whether a kernel address is valid (derived from arch/x86/).
 557  */
 558 int kern_addr_valid(unsigned long addr)
 559 {
 560         pgd_t *pgd;
 561         pud_t *pud;
 562         pmd_t *pmd;
 563         pte_t *pte;
 564
 565         if ((((long)addr) >> VA_BITS) != -1UL)
 566                 return 0;
 567
 568         pgd = pgd_offset_k(addr);
 569         if (pgd_none(*pgd))
 570                 return 0;
 571
 572         pud = pud_offset(pgd, addr);
 573         if (pud_none(*pud))
 574                 return 0;
 575
 576         if (pud_sect(*pud))
 577                 return pfn_valid(pud_pfn(*pud));
 578
 579         pmd = pmd_offset(pud, addr);
 580         if (pmd_none(*pmd))
 581                 return 0;
 582
 583         if (pmd_sect(*pmd))
 584                 return pfn_valid(pmd_pfn(*pmd));
 585
 586         pte = pte_offset_kernel(pmd, addr);
 587         if (pte_none(*pte))
 588                 return 0;
 589
 590         return pfn_valid(pte_pfn(*pte));
 591 }
 592 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 593 #if !ARM64_SWAPPER_USES_SECTION_MAPS
 594 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 595 {
 596         return vmemmap_populate_basepages(start, end, node);
 597 }
 598 #else   /* !ARM64_SWAPPER_USES_SECTION_MAPS */
 599 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 600 {
 601         unsigned long addr = start;
 602         unsigned long next;
 603         pgd_t *pgd;
 604         pud_t *pud;
 605         pmd_t *pmd;
 606
 607         do {
 608                 next = pmd_addr_end(addr, end);
 609
 610                 pgd = vmemmap_pgd_populate(addr, node);
 611                 if (!pgd)
 612                         return -ENOMEM;
 613
 614                 pud = vmemmap_pud_populate(pgd, addr, node);
 615                 if (!pud)
 616                         return -ENOMEM;
 617
 618                 pmd = pmd_offset(pud, addr);
 619                 if (pmd_none(*pmd)) {
 620                         void *p = NULL;
 621
 622                         p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 623                         if (!p)
 624                                 return -ENOMEM;
 625
 626                         set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
 627                 } else
 628                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 629         } while (addr = next, addr != end);
 630
 631         return 0;
 632 }
 633 #endif  /* CONFIG_ARM64_64K_PAGES */
 634 void vmemmap_free(unsigned long start, unsigned long end)
 635 {
 636 }
 637 #endif  /* CONFIG_SPARSEMEM_VMEMMAP */
 638
 639 static inline pud_t * fixmap_pud(unsigned long addr)
 640 {
 641         pgd_t *pgd = pgd_offset_k(addr);
 642
 643         BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
 644
 645         return pud_offset_kimg(pgd, addr);
 646 }
 647
 648 static inline pmd_t * fixmap_pmd(unsigned long addr)
 649 {
 650         pud_t *pud = fixmap_pud(addr);
 651
 652         BUG_ON(pud_none(*pud) || pud_bad(*pud));
 653
 654         return pmd_offset_kimg(pud, addr);
 655 }
 656
 657 static inline pte_t * fixmap_pte(unsigned long addr)
 658 {
 659         return &bm_pte[pte_index(addr)];
 660 }
 661
 662 void __init early_fixmap_init(void)
 663 {
 664         pgd_t *pgd;
 665         pud_t *pud;
 666         pmd_t *pmd;
 667         unsigned long addr = FIXADDR_START;
 668
 669         pgd = pgd_offset_k(addr);
 670         if (CONFIG_PGTABLE_LEVELS > 3 &&
 671             !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
 672                 /*
 673                  * We only end up here if the kernel mapping and the fixmap
 674                  * share the top level pgd entry, which should only happen on
 675                  * 16k/4 levels configurations.
 676                  */
 677                 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
 678                 pud = pud_offset_kimg(pgd, addr);
 679         } else {
 680                 pgd_populate(&init_mm, pgd, bm_pud);
 681                 pud = fixmap_pud(addr);
 682         }
 683         pud_populate(&init_mm, pud, bm_pmd);
 684         pmd = fixmap_pmd(addr);
 685         pmd_populate_kernel(&init_mm, pmd, bm_pte);
 686
 687         /*
 688          * The boot-ioremap range spans multiple pmds, for which
 689          * we are not prepared:
 690          */
 691         BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
 692                      != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
 693
 694         if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
 695              || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
 696                 WARN_ON(1);
 697                 pr_warn("pmd %p != %p, %p\n",
 698                         pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
 699                         fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
 700                 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
 701                         fix_to_virt(FIX_BTMAP_BEGIN));
 702                 pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
 703                         fix_to_virt(FIX_BTMAP_END));
 704
 705                 pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
 706                 pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
 707         }
 708 }
 709
 710 void __set_fixmap(enum fixed_addresses idx,
 711                                phys_addr_t phys, pgprot_t flags)
 712 {
 713         unsigned long addr = __fix_to_virt(idx);
 714         pte_t *pte;
 715
 716         BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 717
 718         pte = fixmap_pte(addr);
 719
 720         if (pgprot_val(flags)) {
 721                 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
 722         } else {
 723                 pte_clear(&init_mm, addr, pte);
 724                 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
 725         }
 726 }
 727
 728 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
 729 {
 730         const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
 731         int offset;
 732         void *dt_virt;
 733
 734         /*
 735          * Check whether the physical FDT address is set and meets the minimum
 736          * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
 737          * at least 8 bytes so that we can always access the size field of the
 738          * FDT header after mapping the first chunk, double check here if that
 739          * is indeed the case.
 740          */
 741         BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
 742         if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
 743                 return NULL;
 744
 745         /*
 746          * Make sure that the FDT region can be mapped without the need to
 747          * allocate additional translation table pages, so that it is safe
 748          * to call create_mapping_noalloc() this early.
 749          *
 750          * On 64k pages, the FDT will be mapped using PTEs, so we need to
 751          * be in the same PMD as the rest of the fixmap.
 752          * On 4k pages, we'll use section mappings for the FDT so we only
 753          * have to be in the same PUD.
 754          */
 755         BUILD_BUG_ON(dt_virt_base % SZ_2M);
 756
 757         BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
 758                      __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
 759
 760         offset = dt_phys % SWAPPER_BLOCK_SIZE;
 761         dt_virt = (void *)dt_virt_base + offset;
 762
 763         /* map the first chunk so we can read the size from the header */
 764         create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
 765                         dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
 766
 767         if (fdt_check_header(dt_virt) != 0)
 768                 return NULL;
 769
 770         *size = fdt_totalsize(dt_virt);
 771         if (*size > MAX_FDT_SIZE)
 772                 return NULL;
 773
 774         if (offset + *size > SWAPPER_BLOCK_SIZE)
 775                 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
 776                                round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
 777
 778         return dt_virt;
 779 }
 780
 781 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
 782 {
 783         void *dt_virt;
 784         int size;
 785
 786         dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
 787         if (!dt_virt)
 788                 return NULL;
 789
 790         memblock_reserve(dt_phys, size);
 791         return dt_virt;
 792 }
 793
 794 int __init arch_ioremap_pud_supported(void)
 795 {
 796         /* only 4k granule supports level 1 block mappings */
 797         return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
 798 }
 799
 800 int __init arch_ioremap_pmd_supported(void)
 801 {
 802         return 1;
 803 }
 804
 805 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
 806 {
 807         BUG_ON(phys & ~PUD_MASK);
 808         set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
 809         return 1;
 810 }
 811
 812 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
 813 {
 814         BUG_ON(phys & ~PMD_MASK);
 815         set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
 816         return 1;
 817 }
 818
 819 int pud_clear_huge(pud_t *pud)
 820 {
 821         if (!pud_sect(*pud))
 822                 return 0;
 823         pud_clear(pud);
 824         return 1;
 825 }
 826
 827 int pmd_clear_huge(pmd_t *pmd)
 828 {
 829         if (!pmd_sect(*pmd))
 830                 return 0;
 831         pmd_clear(pmd);
 832         return 1;
 833 }