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/cache.h> 21 #include <linux/export.h> 22 #include <linux/kernel.h> 23 #include <linux/errno.h> 24 #include <linux/init.h> 25 #include <linux/ioport.h> 26 #include <linux/kexec.h> 27 #include <linux/libfdt.h> 28 #include <linux/mman.h> 29 #include <linux/nodemask.h> 30 #include <linux/memblock.h> 31 #include <linux/fs.h> 32 #include <linux/io.h> 33 #include <linux/mm.h> 34 #include <linux/vmalloc.h> 35 36 #include <asm/barrier.h> 37 #include <asm/cputype.h> 38 #include <asm/fixmap.h> 39 #include <asm/kasan.h> 40 #include <asm/kernel-pgtable.h> 41 #include <asm/sections.h> 42 #include <asm/setup.h> 43 #include <asm/sizes.h> 44 #include <asm/tlb.h> 45 #include <asm/memblock.h> 46 #include <asm/mmu_context.h> 47 #include <asm/ptdump.h> 48 49 #define NO_BLOCK_MAPPINGS BIT(0) 50 #define NO_CONT_MAPPINGS BIT(1) 51 52 u64 idmap_t0sz = TCR_T0SZ(VA_BITS); 53 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD; 54 55 u64 kimage_voffset __ro_after_init; 56 EXPORT_SYMBOL(kimage_voffset); 57 58 /* 59 * Empty_zero_page is a special page that is used for zero-initialized data 60 * and COW. 61 */ 62 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss; 63 EXPORT_SYMBOL(empty_zero_page); 64 65 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss; 66 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused; 67 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; 68 69 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, 70 unsigned long size, pgprot_t vma_prot) 71 { 72 if (!pfn_valid(pfn)) 73 return pgprot_noncached(vma_prot); 74 else if (file->f_flags & O_SYNC) 75 return pgprot_writecombine(vma_prot); 76 return vma_prot; 77 } 78 EXPORT_SYMBOL(phys_mem_access_prot); 79 80 static phys_addr_t __init early_pgtable_alloc(void) 81 { 82 phys_addr_t phys; 83 void *ptr; 84 85 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE); 86 87 /* 88 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE 89 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise 90 * any level of table. 91 */ 92 ptr = pte_set_fixmap(phys); 93 94 memset(ptr, 0, PAGE_SIZE); 95 96 /* 97 * Implicit barriers also ensure the zeroed page is visible to the page 98 * table walker 99 */ 100 pte_clear_fixmap(); 101 102 return phys; 103 } 104 105 static bool pgattr_change_is_safe(u64 old, u64 new) 106 { 107 /* 108 * The following mapping attributes may be updated in live 109 * kernel mappings without the need for break-before-make. 110 */ 111 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE; 112 113 /* creating or taking down mappings is always safe */ 114 if (old == 0 || new == 0) 115 return true; 116 117 /* live contiguous mappings may not be manipulated at all */ 118 if ((old | new) & PTE_CONT) 119 return false; 120 121 /* Transitioning from Global to Non-Global is safe */ 122 if (((old ^ new) == PTE_NG) && (new & PTE_NG)) 123 return true; 124 125 return ((old ^ new) & ~mask) == 0; 126 } 127 128 static void init_pte(pmd_t *pmd, unsigned long addr, unsigned long end, 129 phys_addr_t phys, pgprot_t prot) 130 { 131 pte_t *pte; 132 133 pte = pte_set_fixmap_offset(pmd, addr); 134 do { 135 pte_t old_pte = *pte; 136 137 set_pte(pte, pfn_pte(__phys_to_pfn(phys), prot)); 138 139 /* 140 * After the PTE entry has been populated once, we 141 * only allow updates to the permission attributes. 142 */ 143 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte))); 144 145 phys += PAGE_SIZE; 146 } while (pte++, addr += PAGE_SIZE, addr != end); 147 148 pte_clear_fixmap(); 149 } 150 151 static void alloc_init_cont_pte(pmd_t *pmd, unsigned long addr, 152 unsigned long end, phys_addr_t phys, 153 pgprot_t prot, 154 phys_addr_t (*pgtable_alloc)(void), 155 int flags) 156 { 157 unsigned long next; 158 159 BUG_ON(pmd_sect(*pmd)); 160 if (pmd_none(*pmd)) { 161 phys_addr_t pte_phys; 162 BUG_ON(!pgtable_alloc); 163 pte_phys = pgtable_alloc(); 164 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE); 165 } 166 BUG_ON(pmd_bad(*pmd)); 167 168 do { 169 pgprot_t __prot = prot; 170 171 next = pte_cont_addr_end(addr, end); 172 173 /* use a contiguous mapping if the range is suitably aligned */ 174 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) && 175 (flags & NO_CONT_MAPPINGS) == 0) 176 __prot = __pgprot(pgprot_val(prot) | PTE_CONT); 177 178 init_pte(pmd, addr, next, phys, __prot); 179 180 phys += next - addr; 181 } while (addr = next, addr != end); 182 } 183 184 static void init_pmd(pud_t *pud, unsigned long addr, unsigned long end, 185 phys_addr_t phys, pgprot_t prot, 186 phys_addr_t (*pgtable_alloc)(void), int flags) 187 { 188 unsigned long next; 189 pmd_t *pmd; 190 191 pmd = pmd_set_fixmap_offset(pud, addr); 192 do { 193 pmd_t old_pmd = *pmd; 194 195 next = pmd_addr_end(addr, end); 196 197 /* try section mapping first */ 198 if (((addr | next | phys) & ~SECTION_MASK) == 0 && 199 (flags & NO_BLOCK_MAPPINGS) == 0) { 200 pmd_set_huge(pmd, phys, prot); 201 202 /* 203 * After the PMD entry has been populated once, we 204 * only allow updates to the permission attributes. 205 */ 206 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd), 207 pmd_val(*pmd))); 208 } else { 209 alloc_init_cont_pte(pmd, addr, next, phys, prot, 210 pgtable_alloc, flags); 211 212 BUG_ON(pmd_val(old_pmd) != 0 && 213 pmd_val(old_pmd) != pmd_val(*pmd)); 214 } 215 phys += next - addr; 216 } while (pmd++, addr = next, addr != end); 217 218 pmd_clear_fixmap(); 219 } 220 221 static void alloc_init_cont_pmd(pud_t *pud, unsigned long addr, 222 unsigned long end, phys_addr_t phys, 223 pgprot_t prot, 224 phys_addr_t (*pgtable_alloc)(void), int flags) 225 { 226 unsigned long next; 227 228 /* 229 * Check for initial section mappings in the pgd/pud. 230 */ 231 BUG_ON(pud_sect(*pud)); 232 if (pud_none(*pud)) { 233 phys_addr_t pmd_phys; 234 BUG_ON(!pgtable_alloc); 235 pmd_phys = pgtable_alloc(); 236 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE); 237 } 238 BUG_ON(pud_bad(*pud)); 239 240 do { 241 pgprot_t __prot = prot; 242 243 next = pmd_cont_addr_end(addr, end); 244 245 /* use a contiguous mapping if the range is suitably aligned */ 246 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) && 247 (flags & NO_CONT_MAPPINGS) == 0) 248 __prot = __pgprot(pgprot_val(prot) | PTE_CONT); 249 250 init_pmd(pud, addr, next, phys, __prot, pgtable_alloc, flags); 251 252 phys += next - addr; 253 } while (addr = next, addr != end); 254 } 255 256 static inline bool use_1G_block(unsigned long addr, unsigned long next, 257 unsigned long phys) 258 { 259 if (PAGE_SHIFT != 12) 260 return false; 261 262 if (((addr | next | phys) & ~PUD_MASK) != 0) 263 return false; 264 265 return true; 266 } 267 268 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end, 269 phys_addr_t phys, pgprot_t prot, 270 phys_addr_t (*pgtable_alloc)(void), 271 int flags) 272 { 273 pud_t *pud; 274 unsigned long next; 275 276 if (pgd_none(*pgd)) { 277 phys_addr_t pud_phys; 278 BUG_ON(!pgtable_alloc); 279 pud_phys = pgtable_alloc(); 280 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE); 281 } 282 BUG_ON(pgd_bad(*pgd)); 283 284 pud = pud_set_fixmap_offset(pgd, addr); 285 do { 286 pud_t old_pud = *pud; 287 288 next = pud_addr_end(addr, end); 289 290 /* 291 * For 4K granule only, attempt to put down a 1GB block 292 */ 293 if (use_1G_block(addr, next, phys) && 294 (flags & NO_BLOCK_MAPPINGS) == 0) { 295 pud_set_huge(pud, phys, prot); 296 297 /* 298 * After the PUD entry has been populated once, we 299 * only allow updates to the permission attributes. 300 */ 301 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud), 302 pud_val(*pud))); 303 } else { 304 alloc_init_cont_pmd(pud, addr, next, phys, prot, 305 pgtable_alloc, flags); 306 307 BUG_ON(pud_val(old_pud) != 0 && 308 pud_val(old_pud) != pud_val(*pud)); 309 } 310 phys += next - addr; 311 } while (pud++, addr = next, addr != end); 312 313 pud_clear_fixmap(); 314 } 315 316 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys, 317 unsigned long virt, phys_addr_t size, 318 pgprot_t prot, 319 phys_addr_t (*pgtable_alloc)(void), 320 int flags) 321 { 322 unsigned long addr, length, end, next; 323 pgd_t *pgd = pgd_offset_raw(pgdir, virt); 324 325 /* 326 * If the virtual and physical address don't have the same offset 327 * within a page, we cannot map the region as the caller expects. 328 */ 329 if (WARN_ON((phys ^ virt) & ~PAGE_MASK)) 330 return; 331 332 phys &= PAGE_MASK; 333 addr = virt & PAGE_MASK; 334 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK)); 335 336 end = addr + length; 337 do { 338 next = pgd_addr_end(addr, end); 339 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc, 340 flags); 341 phys += next - addr; 342 } while (pgd++, addr = next, addr != end); 343 } 344 345 static phys_addr_t pgd_pgtable_alloc(void) 346 { 347 void *ptr = (void *)__get_free_page(PGALLOC_GFP); 348 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr))) 349 BUG(); 350 351 /* Ensure the zeroed page is visible to the page table walker */ 352 dsb(ishst); 353 return __pa(ptr); 354 } 355 356 /* 357 * This function can only be used to modify existing table entries, 358 * without allocating new levels of table. Note that this permits the 359 * creation of new section or page entries. 360 */ 361 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt, 362 phys_addr_t size, pgprot_t prot) 363 { 364 if (virt < VMALLOC_START) { 365 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n", 366 &phys, virt); 367 return; 368 } 369 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, 370 NO_CONT_MAPPINGS); 371 } 372 373 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys, 374 unsigned long virt, phys_addr_t size, 375 pgprot_t prot, bool page_mappings_only) 376 { 377 int flags = 0; 378 379 BUG_ON(mm == &init_mm); 380 381 if (page_mappings_only) 382 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; 383 384 __create_pgd_mapping(mm->pgd, phys, virt, size, prot, 385 pgd_pgtable_alloc, flags); 386 } 387 388 static void update_mapping_prot(phys_addr_t phys, unsigned long virt, 389 phys_addr_t size, pgprot_t prot) 390 { 391 if (virt < VMALLOC_START) { 392 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n", 393 &phys, virt); 394 return; 395 } 396 397 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, 398 NO_CONT_MAPPINGS); 399 400 /* flush the TLBs after updating live kernel mappings */ 401 flush_tlb_kernel_range(virt, virt + size); 402 } 403 404 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, 405 phys_addr_t end, pgprot_t prot, int flags) 406 { 407 __create_pgd_mapping(pgd, start, __phys_to_virt(start), end - start, 408 prot, early_pgtable_alloc, flags); 409 } 410 411 void __init mark_linear_text_alias_ro(void) 412 { 413 /* 414 * Remove the write permissions from the linear alias of .text/.rodata 415 */ 416 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text), 417 (unsigned long)__init_begin - (unsigned long)_text, 418 PAGE_KERNEL_RO); 419 } 420 421 static void __init map_mem(pgd_t *pgd) 422 { 423 phys_addr_t kernel_start = __pa_symbol(_text); 424 phys_addr_t kernel_end = __pa_symbol(__init_begin); 425 struct memblock_region *reg; 426 int flags = 0; 427 428 if (debug_pagealloc_enabled()) 429 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; 430 431 /* 432 * Take care not to create a writable alias for the 433 * read-only text and rodata sections of the kernel image. 434 * So temporarily mark them as NOMAP to skip mappings in 435 * the following for-loop 436 */ 437 memblock_mark_nomap(kernel_start, kernel_end - kernel_start); 438 #ifdef CONFIG_KEXEC_CORE 439 if (crashk_res.end) 440 memblock_mark_nomap(crashk_res.start, 441 resource_size(&crashk_res)); 442 #endif 443 444 /* map all the memory banks */ 445 for_each_memblock(memory, reg) { 446 phys_addr_t start = reg->base; 447 phys_addr_t end = start + reg->size; 448 449 if (start >= end) 450 break; 451 if (memblock_is_nomap(reg)) 452 continue; 453 454 __map_memblock(pgd, start, end, PAGE_KERNEL, flags); 455 } 456 457 /* 458 * Map the linear alias of the [_text, __init_begin) interval 459 * as non-executable now, and remove the write permission in 460 * mark_linear_text_alias_ro() below (which will be called after 461 * alternative patching has completed). This makes the contents 462 * of the region accessible to subsystems such as hibernate, 463 * but protects it from inadvertent modification or execution. 464 * Note that contiguous mappings cannot be remapped in this way, 465 * so we should avoid them here. 466 */ 467 __map_memblock(pgd, kernel_start, kernel_end, 468 PAGE_KERNEL, NO_CONT_MAPPINGS); 469 memblock_clear_nomap(kernel_start, kernel_end - kernel_start); 470 471 #ifdef CONFIG_KEXEC_CORE 472 /* 473 * Use page-level mappings here so that we can shrink the region 474 * in page granularity and put back unused memory to buddy system 475 * through /sys/kernel/kexec_crash_size interface. 476 */ 477 if (crashk_res.end) { 478 __map_memblock(pgd, crashk_res.start, crashk_res.end + 1, 479 PAGE_KERNEL, 480 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS); 481 memblock_clear_nomap(crashk_res.start, 482 resource_size(&crashk_res)); 483 } 484 #endif 485 } 486 487 void mark_rodata_ro(void) 488 { 489 unsigned long section_size; 490 491 /* 492 * mark .rodata as read only. Use __init_begin rather than __end_rodata 493 * to cover NOTES and EXCEPTION_TABLE. 494 */ 495 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata; 496 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata, 497 section_size, PAGE_KERNEL_RO); 498 499 debug_checkwx(); 500 } 501 502 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end, 503 pgprot_t prot, struct vm_struct *vma, 504 int flags, unsigned long vm_flags) 505 { 506 phys_addr_t pa_start = __pa_symbol(va_start); 507 unsigned long size = va_end - va_start; 508 509 BUG_ON(!PAGE_ALIGNED(pa_start)); 510 BUG_ON(!PAGE_ALIGNED(size)); 511 512 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot, 513 early_pgtable_alloc, flags); 514 515 if (!(vm_flags & VM_NO_GUARD)) 516 size += PAGE_SIZE; 517 518 vma->addr = va_start; 519 vma->phys_addr = pa_start; 520 vma->size = size; 521 vma->flags = VM_MAP | vm_flags; 522 vma->caller = __builtin_return_address(0); 523 524 vm_area_add_early(vma); 525 } 526 527 static int __init parse_rodata(char *arg) 528 { 529 return strtobool(arg, &rodata_enabled); 530 } 531 early_param("rodata", parse_rodata); 532 533 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0 534 static int __init map_entry_trampoline(void) 535 { 536 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC; 537 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start); 538 539 /* The trampoline is always mapped and can therefore be global */ 540 pgprot_val(prot) &= ~PTE_NG; 541 542 /* Map only the text into the trampoline page table */ 543 memset(tramp_pg_dir, 0, PGD_SIZE); 544 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE, 545 prot, pgd_pgtable_alloc, 0); 546 547 /* Map both the text and data into the kernel page table */ 548 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot); 549 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) { 550 extern char __entry_tramp_data_start[]; 551 552 __set_fixmap(FIX_ENTRY_TRAMP_DATA, 553 __pa_symbol(__entry_tramp_data_start), 554 PAGE_KERNEL_RO); 555 } 556 557 return 0; 558 } 559 core_initcall(map_entry_trampoline); 560 #endif 561 562 /* 563 * Create fine-grained mappings for the kernel. 564 */ 565 static void __init map_kernel(pgd_t *pgd) 566 { 567 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext, 568 vmlinux_initdata, vmlinux_data; 569 570 /* 571 * External debuggers may need to write directly to the text 572 * mapping to install SW breakpoints. Allow this (only) when 573 * explicitly requested with rodata=off. 574 */ 575 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC; 576 577 /* 578 * Only rodata will be remapped with different permissions later on, 579 * all other segments are allowed to use contiguous mappings. 580 */ 581 map_kernel_segment(pgd, _text, _etext, text_prot, &vmlinux_text, 0, 582 VM_NO_GUARD); 583 map_kernel_segment(pgd, __start_rodata, __inittext_begin, PAGE_KERNEL, 584 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD); 585 map_kernel_segment(pgd, __inittext_begin, __inittext_end, text_prot, 586 &vmlinux_inittext, 0, VM_NO_GUARD); 587 map_kernel_segment(pgd, __initdata_begin, __initdata_end, PAGE_KERNEL, 588 &vmlinux_initdata, 0, VM_NO_GUARD); 589 map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0); 590 591 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) { 592 /* 593 * The fixmap falls in a separate pgd to the kernel, and doesn't 594 * live in the carveout for the swapper_pg_dir. We can simply 595 * re-use the existing dir for the fixmap. 596 */ 597 set_pgd(pgd_offset_raw(pgd, FIXADDR_START), 598 *pgd_offset_k(FIXADDR_START)); 599 } else if (CONFIG_PGTABLE_LEVELS > 3) { 600 /* 601 * The fixmap shares its top level pgd entry with the kernel 602 * mapping. This can really only occur when we are running 603 * with 16k/4 levels, so we can simply reuse the pud level 604 * entry instead. 605 */ 606 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES)); 607 pud_populate(&init_mm, pud_set_fixmap_offset(pgd, FIXADDR_START), 608 lm_alias(bm_pmd)); 609 pud_clear_fixmap(); 610 } else { 611 BUG(); 612 } 613 614 kasan_copy_shadow(pgd); 615 } 616 617 /* 618 * paging_init() sets up the page tables, initialises the zone memory 619 * maps and sets up the zero page. 620 */ 621 void __init paging_init(void) 622 { 623 phys_addr_t pgd_phys = early_pgtable_alloc(); 624 pgd_t *pgd = pgd_set_fixmap(pgd_phys); 625 626 map_kernel(pgd); 627 map_mem(pgd); 628 629 /* 630 * We want to reuse the original swapper_pg_dir so we don't have to 631 * communicate the new address to non-coherent secondaries in 632 * secondary_entry, and so cpu_switch_mm can generate the address with 633 * adrp+add rather than a load from some global variable. 634 * 635 * To do this we need to go via a temporary pgd. 636 */ 637 cpu_replace_ttbr1(__va(pgd_phys)); 638 memcpy(swapper_pg_dir, pgd, PGD_SIZE); 639 cpu_replace_ttbr1(lm_alias(swapper_pg_dir)); 640 641 pgd_clear_fixmap(); 642 memblock_free(pgd_phys, PAGE_SIZE); 643 644 /* 645 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd 646 * allocated with it. 647 */ 648 memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE, 649 __pa_symbol(swapper_pg_end) - __pa_symbol(swapper_pg_dir) 650 - PAGE_SIZE); 651 } 652 653 /* 654 * Check whether a kernel address is valid (derived from arch/x86/). 655 */ 656 int kern_addr_valid(unsigned long addr) 657 { 658 pgd_t *pgd; 659 pud_t *pud; 660 pmd_t *pmd; 661 pte_t *pte; 662 663 if ((((long)addr) >> VA_BITS) != -1UL) 664 return 0; 665 666 pgd = pgd_offset_k(addr); 667 if (pgd_none(*pgd)) 668 return 0; 669 670 pud = pud_offset(pgd, addr); 671 if (pud_none(*pud)) 672 return 0; 673 674 if (pud_sect(*pud)) 675 return pfn_valid(pud_pfn(*pud)); 676 677 pmd = pmd_offset(pud, addr); 678 if (pmd_none(*pmd)) 679 return 0; 680 681 if (pmd_sect(*pmd)) 682 return pfn_valid(pmd_pfn(*pmd)); 683 684 pte = pte_offset_kernel(pmd, addr); 685 if (pte_none(*pte)) 686 return 0; 687 688 return pfn_valid(pte_pfn(*pte)); 689 } 690 #ifdef CONFIG_SPARSEMEM_VMEMMAP 691 #if !ARM64_SWAPPER_USES_SECTION_MAPS 692 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 693 struct vmem_altmap *altmap) 694 { 695 return vmemmap_populate_basepages(start, end, node); 696 } 697 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */ 698 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 699 struct vmem_altmap *altmap) 700 { 701 unsigned long addr = start; 702 unsigned long next; 703 pgd_t *pgd; 704 pud_t *pud; 705 pmd_t *pmd; 706 707 do { 708 next = pmd_addr_end(addr, end); 709 710 pgd = vmemmap_pgd_populate(addr, node); 711 if (!pgd) 712 return -ENOMEM; 713 714 pud = vmemmap_pud_populate(pgd, addr, node); 715 if (!pud) 716 return -ENOMEM; 717 718 pmd = pmd_offset(pud, addr); 719 if (pmd_none(*pmd)) { 720 void *p = NULL; 721 722 p = vmemmap_alloc_block_buf(PMD_SIZE, node); 723 if (!p) 724 return -ENOMEM; 725 726 pmd_set_huge(pmd, __pa(p), __pgprot(PROT_SECT_NORMAL)); 727 } else 728 vmemmap_verify((pte_t *)pmd, node, addr, next); 729 } while (addr = next, addr != end); 730 731 return 0; 732 } 733 #endif /* CONFIG_ARM64_64K_PAGES */ 734 void vmemmap_free(unsigned long start, unsigned long end, 735 struct vmem_altmap *altmap) 736 { 737 } 738 #endif /* CONFIG_SPARSEMEM_VMEMMAP */ 739 740 static inline pud_t * fixmap_pud(unsigned long addr) 741 { 742 pgd_t *pgd = pgd_offset_k(addr); 743 744 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd)); 745 746 return pud_offset_kimg(pgd, addr); 747 } 748 749 static inline pmd_t * fixmap_pmd(unsigned long addr) 750 { 751 pud_t *pud = fixmap_pud(addr); 752 753 BUG_ON(pud_none(*pud) || pud_bad(*pud)); 754 755 return pmd_offset_kimg(pud, addr); 756 } 757 758 static inline pte_t * fixmap_pte(unsigned long addr) 759 { 760 return &bm_pte[pte_index(addr)]; 761 } 762 763 /* 764 * The p*d_populate functions call virt_to_phys implicitly so they can't be used 765 * directly on kernel symbols (bm_p*d). This function is called too early to use 766 * lm_alias so __p*d_populate functions must be used to populate with the 767 * physical address from __pa_symbol. 768 */ 769 void __init early_fixmap_init(void) 770 { 771 pgd_t *pgd; 772 pud_t *pud; 773 pmd_t *pmd; 774 unsigned long addr = FIXADDR_START; 775 776 pgd = pgd_offset_k(addr); 777 if (CONFIG_PGTABLE_LEVELS > 3 && 778 !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) { 779 /* 780 * We only end up here if the kernel mapping and the fixmap 781 * share the top level pgd entry, which should only happen on 782 * 16k/4 levels configurations. 783 */ 784 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES)); 785 pud = pud_offset_kimg(pgd, addr); 786 } else { 787 if (pgd_none(*pgd)) 788 __pgd_populate(pgd, __pa_symbol(bm_pud), PUD_TYPE_TABLE); 789 pud = fixmap_pud(addr); 790 } 791 if (pud_none(*pud)) 792 __pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE); 793 pmd = fixmap_pmd(addr); 794 __pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE); 795 796 /* 797 * The boot-ioremap range spans multiple pmds, for which 798 * we are not prepared: 799 */ 800 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 801 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 802 803 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN))) 804 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) { 805 WARN_ON(1); 806 pr_warn("pmd %p != %p, %p\n", 807 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)), 808 fixmap_pmd(fix_to_virt(FIX_BTMAP_END))); 809 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 810 fix_to_virt(FIX_BTMAP_BEGIN)); 811 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 812 fix_to_virt(FIX_BTMAP_END)); 813 814 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 815 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 816 } 817 } 818 819 /* 820 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we 821 * ever need to use IPIs for TLB broadcasting, then we're in trouble here. 822 */ 823 void __set_fixmap(enum fixed_addresses idx, 824 phys_addr_t phys, pgprot_t flags) 825 { 826 unsigned long addr = __fix_to_virt(idx); 827 pte_t *pte; 828 829 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 830 831 pte = fixmap_pte(addr); 832 833 if (pgprot_val(flags)) { 834 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags)); 835 } else { 836 pte_clear(&init_mm, addr, pte); 837 flush_tlb_kernel_range(addr, addr+PAGE_SIZE); 838 } 839 } 840 841 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot) 842 { 843 const u64 dt_virt_base = __fix_to_virt(FIX_FDT); 844 int offset; 845 void *dt_virt; 846 847 /* 848 * Check whether the physical FDT address is set and meets the minimum 849 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be 850 * at least 8 bytes so that we can always access the magic and size 851 * fields of the FDT header after mapping the first chunk, double check 852 * here if that is indeed the case. 853 */ 854 BUILD_BUG_ON(MIN_FDT_ALIGN < 8); 855 if (!dt_phys || dt_phys % MIN_FDT_ALIGN) 856 return NULL; 857 858 /* 859 * Make sure that the FDT region can be mapped without the need to 860 * allocate additional translation table pages, so that it is safe 861 * to call create_mapping_noalloc() this early. 862 * 863 * On 64k pages, the FDT will be mapped using PTEs, so we need to 864 * be in the same PMD as the rest of the fixmap. 865 * On 4k pages, we'll use section mappings for the FDT so we only 866 * have to be in the same PUD. 867 */ 868 BUILD_BUG_ON(dt_virt_base % SZ_2M); 869 870 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT != 871 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT); 872 873 offset = dt_phys % SWAPPER_BLOCK_SIZE; 874 dt_virt = (void *)dt_virt_base + offset; 875 876 /* map the first chunk so we can read the size from the header */ 877 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), 878 dt_virt_base, SWAPPER_BLOCK_SIZE, prot); 879 880 if (fdt_magic(dt_virt) != FDT_MAGIC) 881 return NULL; 882 883 *size = fdt_totalsize(dt_virt); 884 if (*size > MAX_FDT_SIZE) 885 return NULL; 886 887 if (offset + *size > SWAPPER_BLOCK_SIZE) 888 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base, 889 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot); 890 891 return dt_virt; 892 } 893 894 void *__init fixmap_remap_fdt(phys_addr_t dt_phys) 895 { 896 void *dt_virt; 897 int size; 898 899 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO); 900 if (!dt_virt) 901 return NULL; 902 903 memblock_reserve(dt_phys, size); 904 return dt_virt; 905 } 906 907 int __init arch_ioremap_pud_supported(void) 908 { 909 /* only 4k granule supports level 1 block mappings */ 910 return IS_ENABLED(CONFIG_ARM64_4K_PAGES); 911 } 912 913 int __init arch_ioremap_pmd_supported(void) 914 { 915 return 1; 916 } 917 918 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot) 919 { 920 pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT | 921 pgprot_val(mk_sect_prot(prot))); 922 BUG_ON(phys & ~PUD_MASK); 923 set_pud(pud, pfn_pud(__phys_to_pfn(phys), sect_prot)); 924 return 1; 925 } 926 927 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot) 928 { 929 pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT | 930 pgprot_val(mk_sect_prot(prot))); 931 BUG_ON(phys & ~PMD_MASK); 932 set_pmd(pmd, pfn_pmd(__phys_to_pfn(phys), sect_prot)); 933 return 1; 934 } 935 936 int pud_clear_huge(pud_t *pud) 937 { 938 if (!pud_sect(*pud)) 939 return 0; 940 pud_clear(pud); 941 return 1; 942 } 943 944 int pmd_clear_huge(pmd_t *pmd) 945 { 946 if (!pmd_sect(*pmd)) 947 return 0; 948 pmd_clear(pmd); 949 return 1; 950 } 951