1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Based on arch/arm/mm/mmu.c 4 * 5 * Copyright (C) 1995-2005 Russell King 6 * Copyright (C) 2012 ARM Ltd. 7 */ 8 9 #include <linux/cache.h> 10 #include <linux/export.h> 11 #include <linux/kernel.h> 12 #include <linux/errno.h> 13 #include <linux/init.h> 14 #include <linux/ioport.h> 15 #include <linux/kexec.h> 16 #include <linux/libfdt.h> 17 #include <linux/mman.h> 18 #include <linux/nodemask.h> 19 #include <linux/memblock.h> 20 #include <linux/memory.h> 21 #include <linux/fs.h> 22 #include <linux/io.h> 23 #include <linux/mm.h> 24 #include <linux/vmalloc.h> 25 26 #include <asm/barrier.h> 27 #include <asm/cputype.h> 28 #include <asm/fixmap.h> 29 #include <asm/kasan.h> 30 #include <asm/kernel-pgtable.h> 31 #include <asm/sections.h> 32 #include <asm/setup.h> 33 #include <linux/sizes.h> 34 #include <asm/tlb.h> 35 #include <asm/mmu_context.h> 36 #include <asm/ptdump.h> 37 #include <asm/tlbflush.h> 38 #include <asm/pgalloc.h> 39 40 #define NO_BLOCK_MAPPINGS BIT(0) 41 #define NO_CONT_MAPPINGS BIT(1) 42 43 u64 idmap_t0sz = TCR_T0SZ(VA_BITS); 44 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD; 45 46 u64 __section(".mmuoff.data.write") vabits_actual; 47 EXPORT_SYMBOL(vabits_actual); 48 49 u64 kimage_voffset __ro_after_init; 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 static DEFINE_SPINLOCK(swapper_pgdir_lock); 64 65 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd) 66 { 67 pgd_t *fixmap_pgdp; 68 69 spin_lock(&swapper_pgdir_lock); 70 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp)); 71 WRITE_ONCE(*fixmap_pgdp, pgd); 72 /* 73 * We need dsb(ishst) here to ensure the page-table-walker sees 74 * our new entry before set_p?d() returns. The fixmap's 75 * flush_tlb_kernel_range() via clear_fixmap() does this for us. 76 */ 77 pgd_clear_fixmap(); 78 spin_unlock(&swapper_pgdir_lock); 79 } 80 81 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, 82 unsigned long size, pgprot_t vma_prot) 83 { 84 if (!pfn_valid(pfn)) 85 return pgprot_noncached(vma_prot); 86 else if (file->f_flags & O_SYNC) 87 return pgprot_writecombine(vma_prot); 88 return vma_prot; 89 } 90 EXPORT_SYMBOL(phys_mem_access_prot); 91 92 static phys_addr_t __init early_pgtable_alloc(int shift) 93 { 94 phys_addr_t phys; 95 void *ptr; 96 97 phys = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 98 if (!phys) 99 panic("Failed to allocate page table page\n"); 100 101 /* 102 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE 103 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise 104 * any level of table. 105 */ 106 ptr = pte_set_fixmap(phys); 107 108 memset(ptr, 0, PAGE_SIZE); 109 110 /* 111 * Implicit barriers also ensure the zeroed page is visible to the page 112 * table walker 113 */ 114 pte_clear_fixmap(); 115 116 return phys; 117 } 118 119 static bool pgattr_change_is_safe(u64 old, u64 new) 120 { 121 /* 122 * The following mapping attributes may be updated in live 123 * kernel mappings without the need for break-before-make. 124 */ 125 pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG; 126 127 /* creating or taking down mappings is always safe */ 128 if (old == 0 || new == 0) 129 return true; 130 131 /* live contiguous mappings may not be manipulated at all */ 132 if ((old | new) & PTE_CONT) 133 return false; 134 135 /* Transitioning from Non-Global to Global is unsafe */ 136 if (old & ~new & PTE_NG) 137 return false; 138 139 /* 140 * Changing the memory type between Normal and Normal-Tagged is safe 141 * since Tagged is considered a permission attribute from the 142 * mismatched attribute aliases perspective. 143 */ 144 if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) || 145 (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) && 146 ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) || 147 (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED))) 148 mask |= PTE_ATTRINDX_MASK; 149 150 return ((old ^ new) & ~mask) == 0; 151 } 152 153 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end, 154 phys_addr_t phys, pgprot_t prot) 155 { 156 pte_t *ptep; 157 158 ptep = pte_set_fixmap_offset(pmdp, addr); 159 do { 160 pte_t old_pte = READ_ONCE(*ptep); 161 162 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot)); 163 164 /* 165 * After the PTE entry has been populated once, we 166 * only allow updates to the permission attributes. 167 */ 168 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), 169 READ_ONCE(pte_val(*ptep)))); 170 171 phys += PAGE_SIZE; 172 } while (ptep++, addr += PAGE_SIZE, addr != end); 173 174 pte_clear_fixmap(); 175 } 176 177 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr, 178 unsigned long end, phys_addr_t phys, 179 pgprot_t prot, 180 phys_addr_t (*pgtable_alloc)(int), 181 int flags) 182 { 183 unsigned long next; 184 pmd_t pmd = READ_ONCE(*pmdp); 185 186 BUG_ON(pmd_sect(pmd)); 187 if (pmd_none(pmd)) { 188 phys_addr_t pte_phys; 189 BUG_ON(!pgtable_alloc); 190 pte_phys = pgtable_alloc(PAGE_SHIFT); 191 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE); 192 pmd = READ_ONCE(*pmdp); 193 } 194 BUG_ON(pmd_bad(pmd)); 195 196 do { 197 pgprot_t __prot = prot; 198 199 next = pte_cont_addr_end(addr, end); 200 201 /* use a contiguous mapping if the range is suitably aligned */ 202 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) && 203 (flags & NO_CONT_MAPPINGS) == 0) 204 __prot = __pgprot(pgprot_val(prot) | PTE_CONT); 205 206 init_pte(pmdp, addr, next, phys, __prot); 207 208 phys += next - addr; 209 } while (addr = next, addr != end); 210 } 211 212 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end, 213 phys_addr_t phys, pgprot_t prot, 214 phys_addr_t (*pgtable_alloc)(int), int flags) 215 { 216 unsigned long next; 217 pmd_t *pmdp; 218 219 pmdp = pmd_set_fixmap_offset(pudp, addr); 220 do { 221 pmd_t old_pmd = READ_ONCE(*pmdp); 222 223 next = pmd_addr_end(addr, end); 224 225 /* try section mapping first */ 226 if (((addr | next | phys) & ~SECTION_MASK) == 0 && 227 (flags & NO_BLOCK_MAPPINGS) == 0) { 228 pmd_set_huge(pmdp, phys, prot); 229 230 /* 231 * After the PMD entry has been populated once, we 232 * only allow updates to the permission attributes. 233 */ 234 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd), 235 READ_ONCE(pmd_val(*pmdp)))); 236 } else { 237 alloc_init_cont_pte(pmdp, addr, next, phys, prot, 238 pgtable_alloc, flags); 239 240 BUG_ON(pmd_val(old_pmd) != 0 && 241 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp))); 242 } 243 phys += next - addr; 244 } while (pmdp++, addr = next, addr != end); 245 246 pmd_clear_fixmap(); 247 } 248 249 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr, 250 unsigned long end, phys_addr_t phys, 251 pgprot_t prot, 252 phys_addr_t (*pgtable_alloc)(int), int flags) 253 { 254 unsigned long next; 255 pud_t pud = READ_ONCE(*pudp); 256 257 /* 258 * Check for initial section mappings in the pgd/pud. 259 */ 260 BUG_ON(pud_sect(pud)); 261 if (pud_none(pud)) { 262 phys_addr_t pmd_phys; 263 BUG_ON(!pgtable_alloc); 264 pmd_phys = pgtable_alloc(PMD_SHIFT); 265 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE); 266 pud = READ_ONCE(*pudp); 267 } 268 BUG_ON(pud_bad(pud)); 269 270 do { 271 pgprot_t __prot = prot; 272 273 next = pmd_cont_addr_end(addr, end); 274 275 /* use a contiguous mapping if the range is suitably aligned */ 276 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) && 277 (flags & NO_CONT_MAPPINGS) == 0) 278 __prot = __pgprot(pgprot_val(prot) | PTE_CONT); 279 280 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags); 281 282 phys += next - addr; 283 } while (addr = next, addr != end); 284 } 285 286 static inline bool use_1G_block(unsigned long addr, unsigned long next, 287 unsigned long phys) 288 { 289 if (PAGE_SHIFT != 12) 290 return false; 291 292 if (((addr | next | phys) & ~PUD_MASK) != 0) 293 return false; 294 295 return true; 296 } 297 298 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end, 299 phys_addr_t phys, pgprot_t prot, 300 phys_addr_t (*pgtable_alloc)(int), 301 int flags) 302 { 303 unsigned long next; 304 pud_t *pudp; 305 p4d_t *p4dp = p4d_offset(pgdp, addr); 306 p4d_t p4d = READ_ONCE(*p4dp); 307 308 if (p4d_none(p4d)) { 309 phys_addr_t pud_phys; 310 BUG_ON(!pgtable_alloc); 311 pud_phys = pgtable_alloc(PUD_SHIFT); 312 __p4d_populate(p4dp, pud_phys, PUD_TYPE_TABLE); 313 p4d = READ_ONCE(*p4dp); 314 } 315 BUG_ON(p4d_bad(p4d)); 316 317 pudp = pud_set_fixmap_offset(p4dp, addr); 318 do { 319 pud_t old_pud = READ_ONCE(*pudp); 320 321 next = pud_addr_end(addr, end); 322 323 /* 324 * For 4K granule only, attempt to put down a 1GB block 325 */ 326 if (use_1G_block(addr, next, phys) && 327 (flags & NO_BLOCK_MAPPINGS) == 0) { 328 pud_set_huge(pudp, phys, prot); 329 330 /* 331 * After the PUD entry has been populated once, we 332 * only allow updates to the permission attributes. 333 */ 334 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud), 335 READ_ONCE(pud_val(*pudp)))); 336 } else { 337 alloc_init_cont_pmd(pudp, addr, next, phys, prot, 338 pgtable_alloc, flags); 339 340 BUG_ON(pud_val(old_pud) != 0 && 341 pud_val(old_pud) != READ_ONCE(pud_val(*pudp))); 342 } 343 phys += next - addr; 344 } while (pudp++, addr = next, addr != end); 345 346 pud_clear_fixmap(); 347 } 348 349 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys, 350 unsigned long virt, phys_addr_t size, 351 pgprot_t prot, 352 phys_addr_t (*pgtable_alloc)(int), 353 int flags) 354 { 355 unsigned long addr, end, next; 356 pgd_t *pgdp = pgd_offset_pgd(pgdir, virt); 357 358 /* 359 * If the virtual and physical address don't have the same offset 360 * within a page, we cannot map the region as the caller expects. 361 */ 362 if (WARN_ON((phys ^ virt) & ~PAGE_MASK)) 363 return; 364 365 phys &= PAGE_MASK; 366 addr = virt & PAGE_MASK; 367 end = PAGE_ALIGN(virt + size); 368 369 do { 370 next = pgd_addr_end(addr, end); 371 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc, 372 flags); 373 phys += next - addr; 374 } while (pgdp++, addr = next, addr != end); 375 } 376 377 static phys_addr_t __pgd_pgtable_alloc(int shift) 378 { 379 void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL); 380 BUG_ON(!ptr); 381 382 /* Ensure the zeroed page is visible to the page table walker */ 383 dsb(ishst); 384 return __pa(ptr); 385 } 386 387 static phys_addr_t pgd_pgtable_alloc(int shift) 388 { 389 phys_addr_t pa = __pgd_pgtable_alloc(shift); 390 391 /* 392 * Call proper page table ctor in case later we need to 393 * call core mm functions like apply_to_page_range() on 394 * this pre-allocated page table. 395 * 396 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is 397 * folded, and if so pgtable_pmd_page_ctor() becomes nop. 398 */ 399 if (shift == PAGE_SHIFT) 400 BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa))); 401 else if (shift == PMD_SHIFT) 402 BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa))); 403 404 return pa; 405 } 406 407 /* 408 * This function can only be used to modify existing table entries, 409 * without allocating new levels of table. Note that this permits the 410 * creation of new section or page entries. 411 */ 412 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt, 413 phys_addr_t size, pgprot_t prot) 414 { 415 if ((virt >= PAGE_END) && (virt < VMALLOC_START)) { 416 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n", 417 &phys, virt); 418 return; 419 } 420 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, 421 NO_CONT_MAPPINGS); 422 } 423 424 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys, 425 unsigned long virt, phys_addr_t size, 426 pgprot_t prot, bool page_mappings_only) 427 { 428 int flags = 0; 429 430 BUG_ON(mm == &init_mm); 431 432 if (page_mappings_only) 433 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; 434 435 __create_pgd_mapping(mm->pgd, phys, virt, size, prot, 436 pgd_pgtable_alloc, flags); 437 } 438 439 static void update_mapping_prot(phys_addr_t phys, unsigned long virt, 440 phys_addr_t size, pgprot_t prot) 441 { 442 if ((virt >= PAGE_END) && (virt < VMALLOC_START)) { 443 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n", 444 &phys, virt); 445 return; 446 } 447 448 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, 449 NO_CONT_MAPPINGS); 450 451 /* flush the TLBs after updating live kernel mappings */ 452 flush_tlb_kernel_range(virt, virt + size); 453 } 454 455 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start, 456 phys_addr_t end, pgprot_t prot, int flags) 457 { 458 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start, 459 prot, early_pgtable_alloc, flags); 460 } 461 462 void __init mark_linear_text_alias_ro(void) 463 { 464 /* 465 * Remove the write permissions from the linear alias of .text/.rodata 466 */ 467 update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext), 468 (unsigned long)__init_begin - (unsigned long)_stext, 469 PAGE_KERNEL_RO); 470 } 471 472 static bool crash_mem_map __initdata; 473 474 static int __init enable_crash_mem_map(char *arg) 475 { 476 /* 477 * Proper parameter parsing is done by reserve_crashkernel(). We only 478 * need to know if the linear map has to avoid block mappings so that 479 * the crashkernel reservations can be unmapped later. 480 */ 481 crash_mem_map = true; 482 483 return 0; 484 } 485 early_param("crashkernel", enable_crash_mem_map); 486 487 static void __init map_mem(pgd_t *pgdp) 488 { 489 phys_addr_t kernel_start = __pa_symbol(_stext); 490 phys_addr_t kernel_end = __pa_symbol(__init_begin); 491 phys_addr_t start, end; 492 int flags = 0; 493 u64 i; 494 495 if (rodata_full || crash_mem_map || debug_pagealloc_enabled()) 496 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; 497 498 /* 499 * Take care not to create a writable alias for the 500 * read-only text and rodata sections of the kernel image. 501 * So temporarily mark them as NOMAP to skip mappings in 502 * the following for-loop 503 */ 504 memblock_mark_nomap(kernel_start, kernel_end - kernel_start); 505 506 /* map all the memory banks */ 507 for_each_mem_range(i, &start, &end) { 508 if (start >= end) 509 break; 510 /* 511 * The linear map must allow allocation tags reading/writing 512 * if MTE is present. Otherwise, it has the same attributes as 513 * PAGE_KERNEL. 514 */ 515 __map_memblock(pgdp, start, end, PAGE_KERNEL_TAGGED, flags); 516 } 517 518 /* 519 * Map the linear alias of the [_stext, __init_begin) interval 520 * as non-executable now, and remove the write permission in 521 * mark_linear_text_alias_ro() below (which will be called after 522 * alternative patching has completed). This makes the contents 523 * of the region accessible to subsystems such as hibernate, 524 * but protects it from inadvertent modification or execution. 525 * Note that contiguous mappings cannot be remapped in this way, 526 * so we should avoid them here. 527 */ 528 __map_memblock(pgdp, kernel_start, kernel_end, 529 PAGE_KERNEL, NO_CONT_MAPPINGS); 530 memblock_clear_nomap(kernel_start, kernel_end - kernel_start); 531 } 532 533 void mark_rodata_ro(void) 534 { 535 unsigned long section_size; 536 537 /* 538 * mark .rodata as read only. Use __init_begin rather than __end_rodata 539 * to cover NOTES and EXCEPTION_TABLE. 540 */ 541 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata; 542 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata, 543 section_size, PAGE_KERNEL_RO); 544 545 debug_checkwx(); 546 } 547 548 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end, 549 pgprot_t prot, struct vm_struct *vma, 550 int flags, unsigned long vm_flags) 551 { 552 phys_addr_t pa_start = __pa_symbol(va_start); 553 unsigned long size = va_end - va_start; 554 555 BUG_ON(!PAGE_ALIGNED(pa_start)); 556 BUG_ON(!PAGE_ALIGNED(size)); 557 558 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot, 559 early_pgtable_alloc, flags); 560 561 if (!(vm_flags & VM_NO_GUARD)) 562 size += PAGE_SIZE; 563 564 vma->addr = va_start; 565 vma->phys_addr = pa_start; 566 vma->size = size; 567 vma->flags = VM_MAP | vm_flags; 568 vma->caller = __builtin_return_address(0); 569 570 vm_area_add_early(vma); 571 } 572 573 static int __init parse_rodata(char *arg) 574 { 575 int ret = strtobool(arg, &rodata_enabled); 576 if (!ret) { 577 rodata_full = false; 578 return 0; 579 } 580 581 /* permit 'full' in addition to boolean options */ 582 if (strcmp(arg, "full")) 583 return -EINVAL; 584 585 rodata_enabled = true; 586 rodata_full = true; 587 return 0; 588 } 589 early_param("rodata", parse_rodata); 590 591 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0 592 static int __init map_entry_trampoline(void) 593 { 594 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC; 595 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start); 596 597 /* The trampoline is always mapped and can therefore be global */ 598 pgprot_val(prot) &= ~PTE_NG; 599 600 /* Map only the text into the trampoline page table */ 601 memset(tramp_pg_dir, 0, PGD_SIZE); 602 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE, 603 prot, __pgd_pgtable_alloc, 0); 604 605 /* Map both the text and data into the kernel page table */ 606 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot); 607 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) { 608 extern char __entry_tramp_data_start[]; 609 610 __set_fixmap(FIX_ENTRY_TRAMP_DATA, 611 __pa_symbol(__entry_tramp_data_start), 612 PAGE_KERNEL_RO); 613 } 614 615 return 0; 616 } 617 core_initcall(map_entry_trampoline); 618 #endif 619 620 /* 621 * Open coded check for BTI, only for use to determine configuration 622 * for early mappings for before the cpufeature code has run. 623 */ 624 static bool arm64_early_this_cpu_has_bti(void) 625 { 626 u64 pfr1; 627 628 if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL)) 629 return false; 630 631 pfr1 = read_sysreg_s(SYS_ID_AA64PFR1_EL1); 632 return cpuid_feature_extract_unsigned_field(pfr1, 633 ID_AA64PFR1_BT_SHIFT); 634 } 635 636 /* 637 * Create fine-grained mappings for the kernel. 638 */ 639 static void __init map_kernel(pgd_t *pgdp) 640 { 641 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext, 642 vmlinux_initdata, vmlinux_data; 643 644 /* 645 * External debuggers may need to write directly to the text 646 * mapping to install SW breakpoints. Allow this (only) when 647 * explicitly requested with rodata=off. 648 */ 649 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC; 650 651 /* 652 * If we have a CPU that supports BTI and a kernel built for 653 * BTI then mark the kernel executable text as guarded pages 654 * now so we don't have to rewrite the page tables later. 655 */ 656 if (arm64_early_this_cpu_has_bti()) 657 text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP); 658 659 /* 660 * Only rodata will be remapped with different permissions later on, 661 * all other segments are allowed to use contiguous mappings. 662 */ 663 map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0, 664 VM_NO_GUARD); 665 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL, 666 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD); 667 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot, 668 &vmlinux_inittext, 0, VM_NO_GUARD); 669 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL, 670 &vmlinux_initdata, 0, VM_NO_GUARD); 671 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0); 672 673 if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) { 674 /* 675 * The fixmap falls in a separate pgd to the kernel, and doesn't 676 * live in the carveout for the swapper_pg_dir. We can simply 677 * re-use the existing dir for the fixmap. 678 */ 679 set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START), 680 READ_ONCE(*pgd_offset_k(FIXADDR_START))); 681 } else if (CONFIG_PGTABLE_LEVELS > 3) { 682 pgd_t *bm_pgdp; 683 p4d_t *bm_p4dp; 684 pud_t *bm_pudp; 685 /* 686 * The fixmap shares its top level pgd entry with the kernel 687 * mapping. This can really only occur when we are running 688 * with 16k/4 levels, so we can simply reuse the pud level 689 * entry instead. 690 */ 691 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES)); 692 bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START); 693 bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START); 694 bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START); 695 pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd)); 696 pud_clear_fixmap(); 697 } else { 698 BUG(); 699 } 700 701 kasan_copy_shadow(pgdp); 702 } 703 704 void __init paging_init(void) 705 { 706 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir)); 707 708 map_kernel(pgdp); 709 map_mem(pgdp); 710 711 pgd_clear_fixmap(); 712 713 cpu_replace_ttbr1(lm_alias(swapper_pg_dir)); 714 init_mm.pgd = swapper_pg_dir; 715 716 memblock_free(__pa_symbol(init_pg_dir), 717 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir)); 718 719 memblock_allow_resize(); 720 } 721 722 /* 723 * Check whether a kernel address is valid (derived from arch/x86/). 724 */ 725 int kern_addr_valid(unsigned long addr) 726 { 727 pgd_t *pgdp; 728 p4d_t *p4dp; 729 pud_t *pudp, pud; 730 pmd_t *pmdp, pmd; 731 pte_t *ptep, pte; 732 733 addr = arch_kasan_reset_tag(addr); 734 if ((((long)addr) >> VA_BITS) != -1UL) 735 return 0; 736 737 pgdp = pgd_offset_k(addr); 738 if (pgd_none(READ_ONCE(*pgdp))) 739 return 0; 740 741 p4dp = p4d_offset(pgdp, addr); 742 if (p4d_none(READ_ONCE(*p4dp))) 743 return 0; 744 745 pudp = pud_offset(p4dp, addr); 746 pud = READ_ONCE(*pudp); 747 if (pud_none(pud)) 748 return 0; 749 750 if (pud_sect(pud)) 751 return pfn_valid(pud_pfn(pud)); 752 753 pmdp = pmd_offset(pudp, addr); 754 pmd = READ_ONCE(*pmdp); 755 if (pmd_none(pmd)) 756 return 0; 757 758 if (pmd_sect(pmd)) 759 return pfn_valid(pmd_pfn(pmd)); 760 761 ptep = pte_offset_kernel(pmdp, addr); 762 pte = READ_ONCE(*ptep); 763 if (pte_none(pte)) 764 return 0; 765 766 return pfn_valid(pte_pfn(pte)); 767 } 768 769 #ifdef CONFIG_MEMORY_HOTPLUG 770 static void free_hotplug_page_range(struct page *page, size_t size, 771 struct vmem_altmap *altmap) 772 { 773 if (altmap) { 774 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 775 } else { 776 WARN_ON(PageReserved(page)); 777 free_pages((unsigned long)page_address(page), get_order(size)); 778 } 779 } 780 781 static void free_hotplug_pgtable_page(struct page *page) 782 { 783 free_hotplug_page_range(page, PAGE_SIZE, NULL); 784 } 785 786 static bool pgtable_range_aligned(unsigned long start, unsigned long end, 787 unsigned long floor, unsigned long ceiling, 788 unsigned long mask) 789 { 790 start &= mask; 791 if (start < floor) 792 return false; 793 794 if (ceiling) { 795 ceiling &= mask; 796 if (!ceiling) 797 return false; 798 } 799 800 if (end - 1 > ceiling - 1) 801 return false; 802 return true; 803 } 804 805 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr, 806 unsigned long end, bool free_mapped, 807 struct vmem_altmap *altmap) 808 { 809 pte_t *ptep, pte; 810 811 do { 812 ptep = pte_offset_kernel(pmdp, addr); 813 pte = READ_ONCE(*ptep); 814 if (pte_none(pte)) 815 continue; 816 817 WARN_ON(!pte_present(pte)); 818 pte_clear(&init_mm, addr, ptep); 819 flush_tlb_kernel_range(addr, addr + PAGE_SIZE); 820 if (free_mapped) 821 free_hotplug_page_range(pte_page(pte), 822 PAGE_SIZE, altmap); 823 } while (addr += PAGE_SIZE, addr < end); 824 } 825 826 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr, 827 unsigned long end, bool free_mapped, 828 struct vmem_altmap *altmap) 829 { 830 unsigned long next; 831 pmd_t *pmdp, pmd; 832 833 do { 834 next = pmd_addr_end(addr, end); 835 pmdp = pmd_offset(pudp, addr); 836 pmd = READ_ONCE(*pmdp); 837 if (pmd_none(pmd)) 838 continue; 839 840 WARN_ON(!pmd_present(pmd)); 841 if (pmd_sect(pmd)) { 842 pmd_clear(pmdp); 843 844 /* 845 * One TLBI should be sufficient here as the PMD_SIZE 846 * range is mapped with a single block entry. 847 */ 848 flush_tlb_kernel_range(addr, addr + PAGE_SIZE); 849 if (free_mapped) 850 free_hotplug_page_range(pmd_page(pmd), 851 PMD_SIZE, altmap); 852 continue; 853 } 854 WARN_ON(!pmd_table(pmd)); 855 unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap); 856 } while (addr = next, addr < end); 857 } 858 859 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr, 860 unsigned long end, bool free_mapped, 861 struct vmem_altmap *altmap) 862 { 863 unsigned long next; 864 pud_t *pudp, pud; 865 866 do { 867 next = pud_addr_end(addr, end); 868 pudp = pud_offset(p4dp, addr); 869 pud = READ_ONCE(*pudp); 870 if (pud_none(pud)) 871 continue; 872 873 WARN_ON(!pud_present(pud)); 874 if (pud_sect(pud)) { 875 pud_clear(pudp); 876 877 /* 878 * One TLBI should be sufficient here as the PUD_SIZE 879 * range is mapped with a single block entry. 880 */ 881 flush_tlb_kernel_range(addr, addr + PAGE_SIZE); 882 if (free_mapped) 883 free_hotplug_page_range(pud_page(pud), 884 PUD_SIZE, altmap); 885 continue; 886 } 887 WARN_ON(!pud_table(pud)); 888 unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap); 889 } while (addr = next, addr < end); 890 } 891 892 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr, 893 unsigned long end, bool free_mapped, 894 struct vmem_altmap *altmap) 895 { 896 unsigned long next; 897 p4d_t *p4dp, p4d; 898 899 do { 900 next = p4d_addr_end(addr, end); 901 p4dp = p4d_offset(pgdp, addr); 902 p4d = READ_ONCE(*p4dp); 903 if (p4d_none(p4d)) 904 continue; 905 906 WARN_ON(!p4d_present(p4d)); 907 unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap); 908 } while (addr = next, addr < end); 909 } 910 911 static void unmap_hotplug_range(unsigned long addr, unsigned long end, 912 bool free_mapped, struct vmem_altmap *altmap) 913 { 914 unsigned long next; 915 pgd_t *pgdp, pgd; 916 917 /* 918 * altmap can only be used as vmemmap mapping backing memory. 919 * In case the backing memory itself is not being freed, then 920 * altmap is irrelevant. Warn about this inconsistency when 921 * encountered. 922 */ 923 WARN_ON(!free_mapped && altmap); 924 925 do { 926 next = pgd_addr_end(addr, end); 927 pgdp = pgd_offset_k(addr); 928 pgd = READ_ONCE(*pgdp); 929 if (pgd_none(pgd)) 930 continue; 931 932 WARN_ON(!pgd_present(pgd)); 933 unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap); 934 } while (addr = next, addr < end); 935 } 936 937 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr, 938 unsigned long end, unsigned long floor, 939 unsigned long ceiling) 940 { 941 pte_t *ptep, pte; 942 unsigned long i, start = addr; 943 944 do { 945 ptep = pte_offset_kernel(pmdp, addr); 946 pte = READ_ONCE(*ptep); 947 948 /* 949 * This is just a sanity check here which verifies that 950 * pte clearing has been done by earlier unmap loops. 951 */ 952 WARN_ON(!pte_none(pte)); 953 } while (addr += PAGE_SIZE, addr < end); 954 955 if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK)) 956 return; 957 958 /* 959 * Check whether we can free the pte page if the rest of the 960 * entries are empty. Overlap with other regions have been 961 * handled by the floor/ceiling check. 962 */ 963 ptep = pte_offset_kernel(pmdp, 0UL); 964 for (i = 0; i < PTRS_PER_PTE; i++) { 965 if (!pte_none(READ_ONCE(ptep[i]))) 966 return; 967 } 968 969 pmd_clear(pmdp); 970 __flush_tlb_kernel_pgtable(start); 971 free_hotplug_pgtable_page(virt_to_page(ptep)); 972 } 973 974 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr, 975 unsigned long end, unsigned long floor, 976 unsigned long ceiling) 977 { 978 pmd_t *pmdp, pmd; 979 unsigned long i, next, start = addr; 980 981 do { 982 next = pmd_addr_end(addr, end); 983 pmdp = pmd_offset(pudp, addr); 984 pmd = READ_ONCE(*pmdp); 985 if (pmd_none(pmd)) 986 continue; 987 988 WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd)); 989 free_empty_pte_table(pmdp, addr, next, floor, ceiling); 990 } while (addr = next, addr < end); 991 992 if (CONFIG_PGTABLE_LEVELS <= 2) 993 return; 994 995 if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK)) 996 return; 997 998 /* 999 * Check whether we can free the pmd page if the rest of the 1000 * entries are empty. Overlap with other regions have been 1001 * handled by the floor/ceiling check. 1002 */ 1003 pmdp = pmd_offset(pudp, 0UL); 1004 for (i = 0; i < PTRS_PER_PMD; i++) { 1005 if (!pmd_none(READ_ONCE(pmdp[i]))) 1006 return; 1007 } 1008 1009 pud_clear(pudp); 1010 __flush_tlb_kernel_pgtable(start); 1011 free_hotplug_pgtable_page(virt_to_page(pmdp)); 1012 } 1013 1014 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr, 1015 unsigned long end, unsigned long floor, 1016 unsigned long ceiling) 1017 { 1018 pud_t *pudp, pud; 1019 unsigned long i, next, start = addr; 1020 1021 do { 1022 next = pud_addr_end(addr, end); 1023 pudp = pud_offset(p4dp, addr); 1024 pud = READ_ONCE(*pudp); 1025 if (pud_none(pud)) 1026 continue; 1027 1028 WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud)); 1029 free_empty_pmd_table(pudp, addr, next, floor, ceiling); 1030 } while (addr = next, addr < end); 1031 1032 if (CONFIG_PGTABLE_LEVELS <= 3) 1033 return; 1034 1035 if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK)) 1036 return; 1037 1038 /* 1039 * Check whether we can free the pud page if the rest of the 1040 * entries are empty. Overlap with other regions have been 1041 * handled by the floor/ceiling check. 1042 */ 1043 pudp = pud_offset(p4dp, 0UL); 1044 for (i = 0; i < PTRS_PER_PUD; i++) { 1045 if (!pud_none(READ_ONCE(pudp[i]))) 1046 return; 1047 } 1048 1049 p4d_clear(p4dp); 1050 __flush_tlb_kernel_pgtable(start); 1051 free_hotplug_pgtable_page(virt_to_page(pudp)); 1052 } 1053 1054 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr, 1055 unsigned long end, unsigned long floor, 1056 unsigned long ceiling) 1057 { 1058 unsigned long next; 1059 p4d_t *p4dp, p4d; 1060 1061 do { 1062 next = p4d_addr_end(addr, end); 1063 p4dp = p4d_offset(pgdp, addr); 1064 p4d = READ_ONCE(*p4dp); 1065 if (p4d_none(p4d)) 1066 continue; 1067 1068 WARN_ON(!p4d_present(p4d)); 1069 free_empty_pud_table(p4dp, addr, next, floor, ceiling); 1070 } while (addr = next, addr < end); 1071 } 1072 1073 static void free_empty_tables(unsigned long addr, unsigned long end, 1074 unsigned long floor, unsigned long ceiling) 1075 { 1076 unsigned long next; 1077 pgd_t *pgdp, pgd; 1078 1079 do { 1080 next = pgd_addr_end(addr, end); 1081 pgdp = pgd_offset_k(addr); 1082 pgd = READ_ONCE(*pgdp); 1083 if (pgd_none(pgd)) 1084 continue; 1085 1086 WARN_ON(!pgd_present(pgd)); 1087 free_empty_p4d_table(pgdp, addr, next, floor, ceiling); 1088 } while (addr = next, addr < end); 1089 } 1090 #endif 1091 1092 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1093 #if !ARM64_SWAPPER_USES_SECTION_MAPS 1094 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1095 struct vmem_altmap *altmap) 1096 { 1097 return vmemmap_populate_basepages(start, end, node, altmap); 1098 } 1099 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */ 1100 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1101 struct vmem_altmap *altmap) 1102 { 1103 unsigned long addr = start; 1104 unsigned long next; 1105 pgd_t *pgdp; 1106 p4d_t *p4dp; 1107 pud_t *pudp; 1108 pmd_t *pmdp; 1109 1110 do { 1111 next = pmd_addr_end(addr, end); 1112 1113 pgdp = vmemmap_pgd_populate(addr, node); 1114 if (!pgdp) 1115 return -ENOMEM; 1116 1117 p4dp = vmemmap_p4d_populate(pgdp, addr, node); 1118 if (!p4dp) 1119 return -ENOMEM; 1120 1121 pudp = vmemmap_pud_populate(p4dp, addr, node); 1122 if (!pudp) 1123 return -ENOMEM; 1124 1125 pmdp = pmd_offset(pudp, addr); 1126 if (pmd_none(READ_ONCE(*pmdp))) { 1127 void *p = NULL; 1128 1129 p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap); 1130 if (!p) { 1131 if (vmemmap_populate_basepages(addr, next, node, altmap)) 1132 return -ENOMEM; 1133 continue; 1134 } 1135 1136 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL)); 1137 } else 1138 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1139 } while (addr = next, addr != end); 1140 1141 return 0; 1142 } 1143 #endif /* !ARM64_SWAPPER_USES_SECTION_MAPS */ 1144 void vmemmap_free(unsigned long start, unsigned long end, 1145 struct vmem_altmap *altmap) 1146 { 1147 #ifdef CONFIG_MEMORY_HOTPLUG 1148 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END)); 1149 1150 unmap_hotplug_range(start, end, true, altmap); 1151 free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END); 1152 #endif 1153 } 1154 #endif /* CONFIG_SPARSEMEM_VMEMMAP */ 1155 1156 static inline pud_t * fixmap_pud(unsigned long addr) 1157 { 1158 pgd_t *pgdp = pgd_offset_k(addr); 1159 p4d_t *p4dp = p4d_offset(pgdp, addr); 1160 p4d_t p4d = READ_ONCE(*p4dp); 1161 1162 BUG_ON(p4d_none(p4d) || p4d_bad(p4d)); 1163 1164 return pud_offset_kimg(p4dp, addr); 1165 } 1166 1167 static inline pmd_t * fixmap_pmd(unsigned long addr) 1168 { 1169 pud_t *pudp = fixmap_pud(addr); 1170 pud_t pud = READ_ONCE(*pudp); 1171 1172 BUG_ON(pud_none(pud) || pud_bad(pud)); 1173 1174 return pmd_offset_kimg(pudp, addr); 1175 } 1176 1177 static inline pte_t * fixmap_pte(unsigned long addr) 1178 { 1179 return &bm_pte[pte_index(addr)]; 1180 } 1181 1182 /* 1183 * The p*d_populate functions call virt_to_phys implicitly so they can't be used 1184 * directly on kernel symbols (bm_p*d). This function is called too early to use 1185 * lm_alias so __p*d_populate functions must be used to populate with the 1186 * physical address from __pa_symbol. 1187 */ 1188 void __init early_fixmap_init(void) 1189 { 1190 pgd_t *pgdp; 1191 p4d_t *p4dp, p4d; 1192 pud_t *pudp; 1193 pmd_t *pmdp; 1194 unsigned long addr = FIXADDR_START; 1195 1196 pgdp = pgd_offset_k(addr); 1197 p4dp = p4d_offset(pgdp, addr); 1198 p4d = READ_ONCE(*p4dp); 1199 if (CONFIG_PGTABLE_LEVELS > 3 && 1200 !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) { 1201 /* 1202 * We only end up here if the kernel mapping and the fixmap 1203 * share the top level pgd entry, which should only happen on 1204 * 16k/4 levels configurations. 1205 */ 1206 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES)); 1207 pudp = pud_offset_kimg(p4dp, addr); 1208 } else { 1209 if (p4d_none(p4d)) 1210 __p4d_populate(p4dp, __pa_symbol(bm_pud), PUD_TYPE_TABLE); 1211 pudp = fixmap_pud(addr); 1212 } 1213 if (pud_none(READ_ONCE(*pudp))) 1214 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE); 1215 pmdp = fixmap_pmd(addr); 1216 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE); 1217 1218 /* 1219 * The boot-ioremap range spans multiple pmds, for which 1220 * we are not prepared: 1221 */ 1222 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1223 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1224 1225 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN))) 1226 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) { 1227 WARN_ON(1); 1228 pr_warn("pmdp %p != %p, %p\n", 1229 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)), 1230 fixmap_pmd(fix_to_virt(FIX_BTMAP_END))); 1231 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1232 fix_to_virt(FIX_BTMAP_BEGIN)); 1233 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1234 fix_to_virt(FIX_BTMAP_END)); 1235 1236 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1237 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1238 } 1239 } 1240 1241 /* 1242 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we 1243 * ever need to use IPIs for TLB broadcasting, then we're in trouble here. 1244 */ 1245 void __set_fixmap(enum fixed_addresses idx, 1246 phys_addr_t phys, pgprot_t flags) 1247 { 1248 unsigned long addr = __fix_to_virt(idx); 1249 pte_t *ptep; 1250 1251 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 1252 1253 ptep = fixmap_pte(addr); 1254 1255 if (pgprot_val(flags)) { 1256 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags)); 1257 } else { 1258 pte_clear(&init_mm, addr, ptep); 1259 flush_tlb_kernel_range(addr, addr+PAGE_SIZE); 1260 } 1261 } 1262 1263 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot) 1264 { 1265 const u64 dt_virt_base = __fix_to_virt(FIX_FDT); 1266 int offset; 1267 void *dt_virt; 1268 1269 /* 1270 * Check whether the physical FDT address is set and meets the minimum 1271 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be 1272 * at least 8 bytes so that we can always access the magic and size 1273 * fields of the FDT header after mapping the first chunk, double check 1274 * here if that is indeed the case. 1275 */ 1276 BUILD_BUG_ON(MIN_FDT_ALIGN < 8); 1277 if (!dt_phys || dt_phys % MIN_FDT_ALIGN) 1278 return NULL; 1279 1280 /* 1281 * Make sure that the FDT region can be mapped without the need to 1282 * allocate additional translation table pages, so that it is safe 1283 * to call create_mapping_noalloc() this early. 1284 * 1285 * On 64k pages, the FDT will be mapped using PTEs, so we need to 1286 * be in the same PMD as the rest of the fixmap. 1287 * On 4k pages, we'll use section mappings for the FDT so we only 1288 * have to be in the same PUD. 1289 */ 1290 BUILD_BUG_ON(dt_virt_base % SZ_2M); 1291 1292 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT != 1293 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT); 1294 1295 offset = dt_phys % SWAPPER_BLOCK_SIZE; 1296 dt_virt = (void *)dt_virt_base + offset; 1297 1298 /* map the first chunk so we can read the size from the header */ 1299 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), 1300 dt_virt_base, SWAPPER_BLOCK_SIZE, prot); 1301 1302 if (fdt_magic(dt_virt) != FDT_MAGIC) 1303 return NULL; 1304 1305 *size = fdt_totalsize(dt_virt); 1306 if (*size > MAX_FDT_SIZE) 1307 return NULL; 1308 1309 if (offset + *size > SWAPPER_BLOCK_SIZE) 1310 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base, 1311 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot); 1312 1313 return dt_virt; 1314 } 1315 1316 int __init arch_ioremap_p4d_supported(void) 1317 { 1318 return 0; 1319 } 1320 1321 int __init arch_ioremap_pud_supported(void) 1322 { 1323 /* 1324 * Only 4k granule supports level 1 block mappings. 1325 * SW table walks can't handle removal of intermediate entries. 1326 */ 1327 return IS_ENABLED(CONFIG_ARM64_4K_PAGES) && 1328 !IS_ENABLED(CONFIG_PTDUMP_DEBUGFS); 1329 } 1330 1331 int __init arch_ioremap_pmd_supported(void) 1332 { 1333 /* See arch_ioremap_pud_supported() */ 1334 return !IS_ENABLED(CONFIG_PTDUMP_DEBUGFS); 1335 } 1336 1337 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot) 1338 { 1339 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot)); 1340 1341 /* Only allow permission changes for now */ 1342 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)), 1343 pud_val(new_pud))) 1344 return 0; 1345 1346 VM_BUG_ON(phys & ~PUD_MASK); 1347 set_pud(pudp, new_pud); 1348 return 1; 1349 } 1350 1351 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot) 1352 { 1353 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot)); 1354 1355 /* Only allow permission changes for now */ 1356 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)), 1357 pmd_val(new_pmd))) 1358 return 0; 1359 1360 VM_BUG_ON(phys & ~PMD_MASK); 1361 set_pmd(pmdp, new_pmd); 1362 return 1; 1363 } 1364 1365 int pud_clear_huge(pud_t *pudp) 1366 { 1367 if (!pud_sect(READ_ONCE(*pudp))) 1368 return 0; 1369 pud_clear(pudp); 1370 return 1; 1371 } 1372 1373 int pmd_clear_huge(pmd_t *pmdp) 1374 { 1375 if (!pmd_sect(READ_ONCE(*pmdp))) 1376 return 0; 1377 pmd_clear(pmdp); 1378 return 1; 1379 } 1380 1381 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr) 1382 { 1383 pte_t *table; 1384 pmd_t pmd; 1385 1386 pmd = READ_ONCE(*pmdp); 1387 1388 if (!pmd_table(pmd)) { 1389 VM_WARN_ON(1); 1390 return 1; 1391 } 1392 1393 table = pte_offset_kernel(pmdp, addr); 1394 pmd_clear(pmdp); 1395 __flush_tlb_kernel_pgtable(addr); 1396 pte_free_kernel(NULL, table); 1397 return 1; 1398 } 1399 1400 int pud_free_pmd_page(pud_t *pudp, unsigned long addr) 1401 { 1402 pmd_t *table; 1403 pmd_t *pmdp; 1404 pud_t pud; 1405 unsigned long next, end; 1406 1407 pud = READ_ONCE(*pudp); 1408 1409 if (!pud_table(pud)) { 1410 VM_WARN_ON(1); 1411 return 1; 1412 } 1413 1414 table = pmd_offset(pudp, addr); 1415 pmdp = table; 1416 next = addr; 1417 end = addr + PUD_SIZE; 1418 do { 1419 pmd_free_pte_page(pmdp, next); 1420 } while (pmdp++, next += PMD_SIZE, next != end); 1421 1422 pud_clear(pudp); 1423 __flush_tlb_kernel_pgtable(addr); 1424 pmd_free(NULL, table); 1425 return 1; 1426 } 1427 1428 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr) 1429 { 1430 return 0; /* Don't attempt a block mapping */ 1431 } 1432 1433 #ifdef CONFIG_MEMORY_HOTPLUG 1434 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size) 1435 { 1436 unsigned long end = start + size; 1437 1438 WARN_ON(pgdir != init_mm.pgd); 1439 WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END)); 1440 1441 unmap_hotplug_range(start, end, false, NULL); 1442 free_empty_tables(start, end, PAGE_OFFSET, PAGE_END); 1443 } 1444 1445 static bool inside_linear_region(u64 start, u64 size) 1446 { 1447 /* 1448 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)] 1449 * accommodating both its ends but excluding PAGE_END. Max physical 1450 * range which can be mapped inside this linear mapping range, must 1451 * also be derived from its end points. 1452 */ 1453 return start >= __pa(_PAGE_OFFSET(vabits_actual)) && 1454 (start + size - 1) <= __pa(PAGE_END - 1); 1455 } 1456 1457 int arch_add_memory(int nid, u64 start, u64 size, 1458 struct mhp_params *params) 1459 { 1460 int ret, flags = 0; 1461 1462 if (!inside_linear_region(start, size)) { 1463 pr_err("[%llx %llx] is outside linear mapping region\n", start, start + size); 1464 return -EINVAL; 1465 } 1466 1467 if (rodata_full || debug_pagealloc_enabled()) 1468 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; 1469 1470 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start), 1471 size, params->pgprot, __pgd_pgtable_alloc, 1472 flags); 1473 1474 memblock_clear_nomap(start, size); 1475 1476 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, 1477 params); 1478 if (ret) 1479 __remove_pgd_mapping(swapper_pg_dir, 1480 __phys_to_virt(start), size); 1481 return ret; 1482 } 1483 1484 void arch_remove_memory(int nid, u64 start, u64 size, 1485 struct vmem_altmap *altmap) 1486 { 1487 unsigned long start_pfn = start >> PAGE_SHIFT; 1488 unsigned long nr_pages = size >> PAGE_SHIFT; 1489 1490 __remove_pages(start_pfn, nr_pages, altmap); 1491 __remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size); 1492 } 1493 1494 /* 1495 * This memory hotplug notifier helps prevent boot memory from being 1496 * inadvertently removed as it blocks pfn range offlining process in 1497 * __offline_pages(). Hence this prevents both offlining as well as 1498 * removal process for boot memory which is initially always online. 1499 * In future if and when boot memory could be removed, this notifier 1500 * should be dropped and free_hotplug_page_range() should handle any 1501 * reserved pages allocated during boot. 1502 */ 1503 static int prevent_bootmem_remove_notifier(struct notifier_block *nb, 1504 unsigned long action, void *data) 1505 { 1506 struct mem_section *ms; 1507 struct memory_notify *arg = data; 1508 unsigned long end_pfn = arg->start_pfn + arg->nr_pages; 1509 unsigned long pfn = arg->start_pfn; 1510 1511 if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE)) 1512 return NOTIFY_OK; 1513 1514 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 1515 unsigned long start = PFN_PHYS(pfn); 1516 unsigned long end = start + (1UL << PA_SECTION_SHIFT); 1517 1518 ms = __pfn_to_section(pfn); 1519 if (!early_section(ms)) 1520 continue; 1521 1522 if (action == MEM_GOING_OFFLINE) { 1523 /* 1524 * Boot memory removal is not supported. Prevent 1525 * it via blocking any attempted offline request 1526 * for the boot memory and just report it. 1527 */ 1528 pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end); 1529 return NOTIFY_BAD; 1530 } else if (action == MEM_OFFLINE) { 1531 /* 1532 * This should have never happened. Boot memory 1533 * offlining should have been prevented by this 1534 * very notifier. Probably some memory removal 1535 * procedure might have changed which would then 1536 * require further debug. 1537 */ 1538 pr_err("Boot memory [%lx %lx] offlined\n", start, end); 1539 1540 /* 1541 * Core memory hotplug does not process a return 1542 * code from the notifier for MEM_OFFLINE events. 1543 * The error condition has been reported. Return 1544 * from here as if ignored. 1545 */ 1546 return NOTIFY_DONE; 1547 } 1548 } 1549 return NOTIFY_OK; 1550 } 1551 1552 static struct notifier_block prevent_bootmem_remove_nb = { 1553 .notifier_call = prevent_bootmem_remove_notifier, 1554 }; 1555 1556 /* 1557 * This ensures that boot memory sections on the platform are online 1558 * from early boot. Memory sections could not be prevented from being 1559 * offlined, unless for some reason they are not online to begin with. 1560 * This helps validate the basic assumption on which the above memory 1561 * event notifier works to prevent boot memory section offlining and 1562 * its possible removal. 1563 */ 1564 static void validate_bootmem_online(void) 1565 { 1566 phys_addr_t start, end, addr; 1567 struct mem_section *ms; 1568 u64 i; 1569 1570 /* 1571 * Scanning across all memblock might be expensive 1572 * on some big memory systems. Hence enable this 1573 * validation only with DEBUG_VM. 1574 */ 1575 if (!IS_ENABLED(CONFIG_DEBUG_VM)) 1576 return; 1577 1578 for_each_mem_range(i, &start, &end) { 1579 for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) { 1580 ms = __pfn_to_section(PHYS_PFN(addr)); 1581 1582 /* 1583 * All memory ranges in the system at this point 1584 * should have been marked as early sections. 1585 */ 1586 WARN_ON(!early_section(ms)); 1587 1588 /* 1589 * Memory notifier mechanism here to prevent boot 1590 * memory offlining depends on the fact that each 1591 * early section memory on the system is initially 1592 * online. Otherwise a given memory section which 1593 * is already offline will be overlooked and can 1594 * be removed completely. Call out such sections. 1595 */ 1596 if (!online_section(ms)) 1597 pr_err("Boot memory [%llx %llx] is offline, can be removed\n", 1598 addr, addr + (1UL << PA_SECTION_SHIFT)); 1599 } 1600 } 1601 } 1602 1603 static int __init prevent_bootmem_remove_init(void) 1604 { 1605 int ret = 0; 1606 1607 if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) 1608 return ret; 1609 1610 validate_bootmem_online(); 1611 ret = register_memory_notifier(&prevent_bootmem_remove_nb); 1612 if (ret) 1613 pr_err("%s: Notifier registration failed %d\n", __func__, ret); 1614 1615 return ret; 1616 } 1617 early_initcall(prevent_bootmem_remove_init); 1618 #endif 1619