1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2012 Regents of the University of California 4 * Copyright (C) 2019 Western Digital Corporation or its affiliates. 5 * Copyright (C) 2020 FORTH-ICS/CARV 6 * Nick Kossifidis <mick@ics.forth.gr> 7 */ 8 9 #include <linux/init.h> 10 #include <linux/mm.h> 11 #include <linux/memblock.h> 12 #include <linux/initrd.h> 13 #include <linux/swap.h> 14 #include <linux/swiotlb.h> 15 #include <linux/sizes.h> 16 #include <linux/of_fdt.h> 17 #include <linux/of_reserved_mem.h> 18 #include <linux/libfdt.h> 19 #include <linux/set_memory.h> 20 #include <linux/dma-map-ops.h> 21 #include <linux/crash_dump.h> 22 #include <linux/hugetlb.h> 23 #ifdef CONFIG_RELOCATABLE 24 #include <linux/elf.h> 25 #endif 26 #include <linux/kfence.h> 27 #include <linux/execmem.h> 28 29 #include <asm/fixmap.h> 30 #include <asm/io.h> 31 #include <asm/kasan.h> 32 #include <asm/numa.h> 33 #include <asm/pgtable.h> 34 #include <asm/sections.h> 35 #include <asm/soc.h> 36 #include <asm/sparsemem.h> 37 #include <asm/tlbflush.h> 38 39 #include "../kernel/head.h" 40 41 u64 new_vmalloc[NR_CPUS / sizeof(u64) + 1]; 42 43 struct kernel_mapping kernel_map __ro_after_init; 44 EXPORT_SYMBOL(kernel_map); 45 #ifdef CONFIG_XIP_KERNEL 46 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 47 #endif 48 49 #ifdef CONFIG_64BIT 50 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 51 #else 52 u64 satp_mode __ro_after_init = SATP_MODE_32; 53 #endif 54 EXPORT_SYMBOL(satp_mode); 55 56 #ifdef CONFIG_64BIT 57 bool pgtable_l4_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 58 bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 59 EXPORT_SYMBOL(pgtable_l4_enabled); 60 EXPORT_SYMBOL(pgtable_l5_enabled); 61 #endif 62 63 phys_addr_t phys_ram_base __ro_after_init; 64 EXPORT_SYMBOL(phys_ram_base); 65 66 #ifdef CONFIG_SPARSEMEM_VMEMMAP 67 #define VMEMMAP_ADDR_ALIGN (1ULL << SECTION_SIZE_BITS) 68 69 unsigned long vmemmap_start_pfn __ro_after_init; 70 EXPORT_SYMBOL(vmemmap_start_pfn); 71 #endif 72 73 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 74 __page_aligned_bss; 75 EXPORT_SYMBOL(empty_zero_page); 76 77 extern char _start[]; 78 void *_dtb_early_va __initdata; 79 uintptr_t _dtb_early_pa __initdata; 80 81 phys_addr_t dma32_phys_limit __initdata; 82 83 static void __init zone_sizes_init(void) 84 { 85 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 86 87 #ifdef CONFIG_ZONE_DMA32 88 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 89 #endif 90 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 91 92 free_area_init(max_zone_pfns); 93 } 94 95 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 96 97 #define LOG2_SZ_1K ilog2(SZ_1K) 98 #define LOG2_SZ_1M ilog2(SZ_1M) 99 #define LOG2_SZ_1G ilog2(SZ_1G) 100 #define LOG2_SZ_1T ilog2(SZ_1T) 101 102 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 103 { 104 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 105 (((t) - (b)) >> LOG2_SZ_1K)); 106 } 107 108 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 109 { 110 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 111 (((t) - (b)) >> LOG2_SZ_1M)); 112 } 113 114 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 115 { 116 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 117 (((t) - (b)) >> LOG2_SZ_1G)); 118 } 119 120 #ifdef CONFIG_64BIT 121 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 122 { 123 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 124 (((t) - (b)) >> LOG2_SZ_1T)); 125 } 126 #else 127 #define print_mlt(n, b, t) do {} while (0) 128 #endif 129 130 static inline void print_ml(char *name, unsigned long b, unsigned long t) 131 { 132 unsigned long diff = t - b; 133 134 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 135 print_mlt(name, b, t); 136 else if ((diff >> LOG2_SZ_1G) >= 10) 137 print_mlg(name, b, t); 138 else if ((diff >> LOG2_SZ_1M) >= 10) 139 print_mlm(name, b, t); 140 else 141 print_mlk(name, b, t); 142 } 143 144 static void __init print_vm_layout(void) 145 { 146 pr_notice("Virtual kernel memory layout:\n"); 147 print_ml("fixmap", (unsigned long)FIXADDR_START, 148 (unsigned long)FIXADDR_TOP); 149 print_ml("pci io", (unsigned long)PCI_IO_START, 150 (unsigned long)PCI_IO_END); 151 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 152 (unsigned long)VMEMMAP_END); 153 print_ml("vmalloc", (unsigned long)VMALLOC_START, 154 (unsigned long)VMALLOC_END); 155 #ifdef CONFIG_64BIT 156 print_ml("modules", (unsigned long)MODULES_VADDR, 157 (unsigned long)MODULES_END); 158 #endif 159 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 160 (unsigned long)high_memory); 161 if (IS_ENABLED(CONFIG_64BIT)) { 162 #ifdef CONFIG_KASAN 163 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 164 #endif 165 166 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 167 (unsigned long)ADDRESS_SPACE_END); 168 } 169 } 170 #else 171 static void print_vm_layout(void) { } 172 #endif /* CONFIG_DEBUG_VM */ 173 174 void __init arch_mm_preinit(void) 175 { 176 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 177 #ifdef CONFIG_FLATMEM 178 BUG_ON(!mem_map); 179 #endif /* CONFIG_FLATMEM */ 180 181 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 182 dma_cache_alignment != 1) { 183 /* 184 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 185 * buffer per 1GB of RAM for kmalloc() bouncing on 186 * non-coherent platforms. 187 */ 188 unsigned long size = 189 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 190 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 191 swiotlb = true; 192 } 193 194 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 195 196 print_vm_layout(); 197 } 198 199 /* Limit the memory size via mem. */ 200 static phys_addr_t memory_limit; 201 #ifdef CONFIG_XIP_KERNEL 202 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 203 #endif /* CONFIG_XIP_KERNEL */ 204 205 static int __init early_mem(char *p) 206 { 207 u64 size; 208 209 if (!p) 210 return 1; 211 212 size = memparse(p, &p) & PAGE_MASK; 213 memory_limit = min_t(u64, size, memory_limit); 214 215 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 216 217 return 0; 218 } 219 early_param("mem", early_mem); 220 221 static void __init setup_bootmem(void) 222 { 223 phys_addr_t vmlinux_end = __pa_symbol(&_end); 224 phys_addr_t max_mapped_addr; 225 phys_addr_t phys_ram_end, vmlinux_start; 226 227 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 228 vmlinux_start = __pa_symbol(&_sdata); 229 else 230 vmlinux_start = __pa_symbol(&_start); 231 232 memblock_enforce_memory_limit(memory_limit); 233 234 /* 235 * Make sure we align the reservation on PMD_SIZE since we will 236 * map the kernel in the linear mapping as read-only: we do not want 237 * any allocation to happen between _end and the next pmd aligned page. 238 */ 239 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 240 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 241 /* 242 * Reserve from the start of the kernel to the end of the kernel 243 */ 244 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 245 246 /* 247 * Make sure we align the start of the memory on a PMD boundary so that 248 * at worst, we map the linear mapping with PMD mappings. 249 */ 250 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) { 251 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 252 #ifdef CONFIG_SPARSEMEM_VMEMMAP 253 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 254 #endif 255 } 256 257 /* 258 * In 64-bit, any use of __va/__pa before this point is wrong as we 259 * did not know the start of DRAM before. 260 */ 261 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 262 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 263 264 /* 265 * The size of the linear page mapping may restrict the amount of 266 * usable RAM. 267 */ 268 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) { 269 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE; 270 if (memblock_end_of_DRAM() > max_mapped_addr) { 271 memblock_cap_memory_range(phys_ram_base, 272 max_mapped_addr - phys_ram_base); 273 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed", 274 &max_mapped_addr); 275 } 276 } 277 278 /* 279 * Reserve physical address space that would be mapped to virtual 280 * addresses greater than (void *)(-PAGE_SIZE) because: 281 * - This memory would overlap with ERR_PTR 282 * - This memory belongs to high memory, which is not supported 283 * 284 * This is not applicable to 64-bit kernel, because virtual addresses 285 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 286 * occupied by kernel mapping. Also it is unrealistic for high memory 287 * to exist on 64-bit platforms. 288 */ 289 if (!IS_ENABLED(CONFIG_64BIT)) { 290 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 291 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 292 } 293 294 phys_ram_end = memblock_end_of_DRAM(); 295 min_low_pfn = PFN_UP(phys_ram_base); 296 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 297 298 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 299 300 reserve_initrd_mem(); 301 302 /* 303 * No allocation should be done before reserving the memory as defined 304 * in the device tree, otherwise the allocation could end up in a 305 * reserved region. 306 */ 307 early_init_fdt_scan_reserved_mem(); 308 309 /* 310 * If DTB is built in, no need to reserve its memblock. 311 * Otherwise, do reserve it but avoid using 312 * early_init_fdt_reserve_self() since __pa() does 313 * not work for DTB pointers that are fixmap addresses 314 */ 315 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 316 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 317 318 dma_contiguous_reserve(dma32_phys_limit); 319 if (IS_ENABLED(CONFIG_64BIT)) 320 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 321 } 322 323 #ifdef CONFIG_MMU 324 struct pt_alloc_ops pt_ops __meminitdata; 325 326 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 327 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 328 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 329 330 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 331 332 #ifdef CONFIG_XIP_KERNEL 333 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 334 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 335 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 336 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 337 #endif /* CONFIG_XIP_KERNEL */ 338 339 static const pgprot_t protection_map[16] = { 340 [VM_NONE] = PAGE_NONE, 341 [VM_READ] = PAGE_READ, 342 [VM_WRITE] = PAGE_COPY, 343 [VM_WRITE | VM_READ] = PAGE_COPY, 344 [VM_EXEC] = PAGE_EXEC, 345 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 346 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 347 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 348 [VM_SHARED] = PAGE_NONE, 349 [VM_SHARED | VM_READ] = PAGE_READ, 350 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 351 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 352 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 353 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 354 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 355 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 356 }; 357 DECLARE_VM_GET_PAGE_PROT 358 359 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 360 { 361 unsigned long addr = __fix_to_virt(idx); 362 pte_t *ptep; 363 364 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 365 366 ptep = &fixmap_pte[pte_index(addr)]; 367 368 if (pgprot_val(prot)) 369 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 370 else 371 pte_clear(&init_mm, addr, ptep); 372 local_flush_tlb_page(addr); 373 } 374 375 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 376 { 377 return (pte_t *)((uintptr_t)pa); 378 } 379 380 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 381 { 382 clear_fixmap(FIX_PTE); 383 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 384 } 385 386 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa) 387 { 388 return (pte_t *) __va(pa); 389 } 390 391 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 392 { 393 /* 394 * We only create PMD or PGD early mappings so we 395 * should never reach here with MMU disabled. 396 */ 397 BUG(); 398 } 399 400 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 401 { 402 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 403 } 404 405 static phys_addr_t __meminit alloc_pte_late(uintptr_t va) 406 { 407 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 408 409 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 410 return __pa((pte_t *)ptdesc_address(ptdesc)); 411 } 412 413 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 414 pgprot_t prot) 415 { 416 uintptr_t pte_idx = pte_index(va); 417 418 BUG_ON(sz != PAGE_SIZE); 419 420 if (pte_none(ptep[pte_idx])) 421 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 422 } 423 424 #ifndef __PAGETABLE_PMD_FOLDED 425 426 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 427 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 428 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 429 430 #ifdef CONFIG_XIP_KERNEL 431 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 432 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 433 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 434 #endif /* CONFIG_XIP_KERNEL */ 435 436 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 437 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 438 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 439 440 #ifdef CONFIG_XIP_KERNEL 441 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 442 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 443 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 444 #endif /* CONFIG_XIP_KERNEL */ 445 446 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 447 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 448 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 449 450 #ifdef CONFIG_XIP_KERNEL 451 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 452 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 453 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 454 #endif /* CONFIG_XIP_KERNEL */ 455 456 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 457 { 458 /* Before MMU is enabled */ 459 return (pmd_t *)((uintptr_t)pa); 460 } 461 462 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 463 { 464 clear_fixmap(FIX_PMD); 465 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 466 } 467 468 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa) 469 { 470 return (pmd_t *) __va(pa); 471 } 472 473 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 474 { 475 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 476 477 return (uintptr_t)early_pmd; 478 } 479 480 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 481 { 482 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 483 } 484 485 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va) 486 { 487 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 488 489 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 490 return __pa((pmd_t *)ptdesc_address(ptdesc)); 491 } 492 493 static void __meminit create_pmd_mapping(pmd_t *pmdp, 494 uintptr_t va, phys_addr_t pa, 495 phys_addr_t sz, pgprot_t prot) 496 { 497 pte_t *ptep; 498 phys_addr_t pte_phys; 499 uintptr_t pmd_idx = pmd_index(va); 500 501 if (sz == PMD_SIZE) { 502 if (pmd_none(pmdp[pmd_idx])) 503 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 504 return; 505 } 506 507 if (pmd_none(pmdp[pmd_idx])) { 508 pte_phys = pt_ops.alloc_pte(va); 509 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 510 ptep = pt_ops.get_pte_virt(pte_phys); 511 memset(ptep, 0, PAGE_SIZE); 512 } else { 513 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 514 ptep = pt_ops.get_pte_virt(pte_phys); 515 } 516 517 create_pte_mapping(ptep, va, pa, sz, prot); 518 } 519 520 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 521 { 522 return (pud_t *)((uintptr_t)pa); 523 } 524 525 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 526 { 527 clear_fixmap(FIX_PUD); 528 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 529 } 530 531 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 532 { 533 return (pud_t *)__va(pa); 534 } 535 536 static phys_addr_t __init alloc_pud_early(uintptr_t va) 537 { 538 /* Only one PUD is available for early mapping */ 539 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 540 541 return (uintptr_t)early_pud; 542 } 543 544 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 545 { 546 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 547 } 548 549 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 550 { 551 unsigned long vaddr; 552 553 vaddr = __get_free_page(GFP_KERNEL); 554 BUG_ON(!vaddr); 555 return __pa(vaddr); 556 } 557 558 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 559 { 560 return (p4d_t *)((uintptr_t)pa); 561 } 562 563 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 564 { 565 clear_fixmap(FIX_P4D); 566 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 567 } 568 569 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 570 { 571 return (p4d_t *)__va(pa); 572 } 573 574 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 575 { 576 /* Only one P4D is available for early mapping */ 577 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 578 579 return (uintptr_t)early_p4d; 580 } 581 582 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 583 { 584 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 585 } 586 587 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 588 { 589 unsigned long vaddr; 590 591 vaddr = __get_free_page(GFP_KERNEL); 592 BUG_ON(!vaddr); 593 return __pa(vaddr); 594 } 595 596 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 597 pgprot_t prot) 598 { 599 pmd_t *nextp; 600 phys_addr_t next_phys; 601 uintptr_t pud_index = pud_index(va); 602 603 if (sz == PUD_SIZE) { 604 if (pud_val(pudp[pud_index]) == 0) 605 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 606 return; 607 } 608 609 if (pud_val(pudp[pud_index]) == 0) { 610 next_phys = pt_ops.alloc_pmd(va); 611 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 612 nextp = pt_ops.get_pmd_virt(next_phys); 613 memset(nextp, 0, PAGE_SIZE); 614 } else { 615 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 616 nextp = pt_ops.get_pmd_virt(next_phys); 617 } 618 619 create_pmd_mapping(nextp, va, pa, sz, prot); 620 } 621 622 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 623 pgprot_t prot) 624 { 625 pud_t *nextp; 626 phys_addr_t next_phys; 627 uintptr_t p4d_index = p4d_index(va); 628 629 if (sz == P4D_SIZE) { 630 if (p4d_val(p4dp[p4d_index]) == 0) 631 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 632 return; 633 } 634 635 if (p4d_val(p4dp[p4d_index]) == 0) { 636 next_phys = pt_ops.alloc_pud(va); 637 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 638 nextp = pt_ops.get_pud_virt(next_phys); 639 memset(nextp, 0, PAGE_SIZE); 640 } else { 641 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 642 nextp = pt_ops.get_pud_virt(next_phys); 643 } 644 645 create_pud_mapping(nextp, va, pa, sz, prot); 646 } 647 648 #define pgd_next_t p4d_t 649 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 650 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 651 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 652 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 653 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 654 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 655 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 656 (pgtable_l5_enabled ? \ 657 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 658 (pgtable_l4_enabled ? \ 659 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 660 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 661 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 662 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 663 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 664 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 665 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 666 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 667 #else 668 #define pgd_next_t pte_t 669 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 670 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 671 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 672 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 673 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 674 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 675 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 676 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 677 #endif /* __PAGETABLE_PMD_FOLDED */ 678 679 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 680 pgprot_t prot) 681 { 682 pgd_next_t *nextp; 683 phys_addr_t next_phys; 684 uintptr_t pgd_idx = pgd_index(va); 685 686 if (sz == PGDIR_SIZE) { 687 if (pgd_val(pgdp[pgd_idx]) == 0) 688 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 689 return; 690 } 691 692 if (pgd_val(pgdp[pgd_idx]) == 0) { 693 next_phys = alloc_pgd_next(va); 694 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 695 nextp = get_pgd_next_virt(next_phys); 696 memset(nextp, 0, PAGE_SIZE); 697 } else { 698 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 699 nextp = get_pgd_next_virt(next_phys); 700 } 701 702 create_pgd_next_mapping(nextp, va, pa, sz, prot); 703 } 704 705 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 706 { 707 if (debug_pagealloc_enabled()) 708 return PAGE_SIZE; 709 710 if (pgtable_l5_enabled && 711 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 712 return P4D_SIZE; 713 714 if (pgtable_l4_enabled && 715 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 716 return PUD_SIZE; 717 718 if (IS_ENABLED(CONFIG_64BIT) && 719 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 720 return PMD_SIZE; 721 722 return PAGE_SIZE; 723 } 724 725 #ifdef CONFIG_XIP_KERNEL 726 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 727 extern char _xiprom[], _exiprom[], __data_loc; 728 729 /* called from head.S with MMU off */ 730 asmlinkage void __init __copy_data(void) 731 { 732 void *from = (void *)(&__data_loc); 733 void *to = (void *)CONFIG_PHYS_RAM_BASE; 734 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 735 736 memcpy(to, from, sz); 737 } 738 #endif 739 740 #ifdef CONFIG_STRICT_KERNEL_RWX 741 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 742 { 743 if (is_va_kernel_text(va)) 744 return PAGE_KERNEL_READ_EXEC; 745 746 /* 747 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 748 * we must protect its linear mapping alias from being executed and 749 * written. 750 * And rodata section is marked readonly in mark_rodata_ro. 751 */ 752 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 753 return PAGE_KERNEL_READ; 754 755 return PAGE_KERNEL; 756 } 757 758 void mark_rodata_ro(void) 759 { 760 set_kernel_memory(__start_rodata, _data, set_memory_ro); 761 if (IS_ENABLED(CONFIG_64BIT)) 762 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 763 set_memory_ro); 764 } 765 #else 766 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 767 { 768 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 769 return PAGE_KERNEL; 770 771 return PAGE_KERNEL_EXEC; 772 } 773 #endif /* CONFIG_STRICT_KERNEL_RWX */ 774 775 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 776 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 777 778 static void __init disable_pgtable_l5(void) 779 { 780 pgtable_l5_enabled = false; 781 kernel_map.page_offset = PAGE_OFFSET_L4; 782 satp_mode = SATP_MODE_48; 783 } 784 785 static void __init disable_pgtable_l4(void) 786 { 787 pgtable_l4_enabled = false; 788 kernel_map.page_offset = PAGE_OFFSET_L3; 789 satp_mode = SATP_MODE_39; 790 } 791 792 static int __init print_no4lvl(char *p) 793 { 794 pr_info("Disabled 4-level and 5-level paging"); 795 return 0; 796 } 797 early_param("no4lvl", print_no4lvl); 798 799 static int __init print_no5lvl(char *p) 800 { 801 pr_info("Disabled 5-level paging"); 802 return 0; 803 } 804 early_param("no5lvl", print_no5lvl); 805 806 static void __init set_mmap_rnd_bits_max(void) 807 { 808 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 809 } 810 811 /* 812 * There is a simple way to determine if 4-level is supported by the 813 * underlying hardware: establish 1:1 mapping in 4-level page table mode 814 * then read SATP to see if the configuration was taken into account 815 * meaning sv48 is supported. 816 */ 817 static __init void set_satp_mode(uintptr_t dtb_pa) 818 { 819 u64 identity_satp, hw_satp; 820 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 821 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 822 823 if (satp_mode_cmdline == SATP_MODE_57) { 824 disable_pgtable_l5(); 825 } else if (satp_mode_cmdline == SATP_MODE_48) { 826 disable_pgtable_l5(); 827 disable_pgtable_l4(); 828 return; 829 } 830 831 create_p4d_mapping(early_p4d, 832 set_satp_mode_pmd, (uintptr_t)early_pud, 833 P4D_SIZE, PAGE_TABLE); 834 create_pud_mapping(early_pud, 835 set_satp_mode_pmd, (uintptr_t)early_pmd, 836 PUD_SIZE, PAGE_TABLE); 837 /* Handle the case where set_satp_mode straddles 2 PMDs */ 838 create_pmd_mapping(early_pmd, 839 set_satp_mode_pmd, set_satp_mode_pmd, 840 PMD_SIZE, PAGE_KERNEL_EXEC); 841 create_pmd_mapping(early_pmd, 842 set_satp_mode_pmd + PMD_SIZE, 843 set_satp_mode_pmd + PMD_SIZE, 844 PMD_SIZE, PAGE_KERNEL_EXEC); 845 retry: 846 create_pgd_mapping(early_pg_dir, 847 set_satp_mode_pmd, 848 pgtable_l5_enabled ? 849 (uintptr_t)early_p4d : (uintptr_t)early_pud, 850 PGDIR_SIZE, PAGE_TABLE); 851 852 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 853 854 local_flush_tlb_all(); 855 csr_write(CSR_SATP, identity_satp); 856 hw_satp = csr_swap(CSR_SATP, 0ULL); 857 local_flush_tlb_all(); 858 859 if (hw_satp != identity_satp) { 860 if (pgtable_l5_enabled) { 861 disable_pgtable_l5(); 862 memset(early_pg_dir, 0, PAGE_SIZE); 863 goto retry; 864 } 865 disable_pgtable_l4(); 866 } 867 868 memset(early_pg_dir, 0, PAGE_SIZE); 869 memset(early_p4d, 0, PAGE_SIZE); 870 memset(early_pud, 0, PAGE_SIZE); 871 memset(early_pmd, 0, PAGE_SIZE); 872 } 873 #endif 874 875 /* 876 * setup_vm() is called from head.S with MMU-off. 877 * 878 * Following requirements should be honoured for setup_vm() to work 879 * correctly: 880 * 1) It should use PC-relative addressing for accessing kernel symbols. 881 * To achieve this we always use GCC cmodel=medany. 882 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 883 * so disable compiler instrumentation when FTRACE is enabled. 884 * 885 * Currently, the above requirements are honoured by using custom CFLAGS 886 * for init.o in mm/Makefile. 887 */ 888 889 #ifndef __riscv_cmodel_medany 890 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 891 #endif 892 893 #ifdef CONFIG_RELOCATABLE 894 extern unsigned long __rela_dyn_start, __rela_dyn_end; 895 896 static void __init relocate_kernel(void) 897 { 898 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 899 /* 900 * This holds the offset between the linked virtual address and the 901 * relocated virtual address. 902 */ 903 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 904 /* 905 * This holds the offset between kernel linked virtual address and 906 * physical address. 907 */ 908 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 909 910 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 911 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 912 Elf64_Addr relocated_addr = rela->r_addend; 913 914 if (rela->r_info != R_RISCV_RELATIVE) 915 continue; 916 917 /* 918 * Make sure to not relocate vdso symbols like rt_sigreturn 919 * which are linked from the address 0 in vmlinux since 920 * vdso symbol addresses are actually used as an offset from 921 * mm->context.vdso in VDSO_OFFSET macro. 922 */ 923 if (relocated_addr >= KERNEL_LINK_ADDR) 924 relocated_addr += reloc_offset; 925 926 *(Elf64_Addr *)addr = relocated_addr; 927 } 928 } 929 #endif /* CONFIG_RELOCATABLE */ 930 931 #ifdef CONFIG_XIP_KERNEL 932 static void __init create_kernel_page_table(pgd_t *pgdir, 933 __always_unused bool early) 934 { 935 uintptr_t va, start_va, end_va; 936 937 /* Map the flash resident part */ 938 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 939 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 940 create_pgd_mapping(pgdir, va, 941 kernel_map.xiprom + (va - kernel_map.virt_addr), 942 PMD_SIZE, PAGE_KERNEL_EXEC); 943 944 /* Map the data in RAM */ 945 start_va = kernel_map.virt_addr + (uintptr_t)&_sdata - (uintptr_t)&_start; 946 end_va = kernel_map.virt_addr + kernel_map.size; 947 for (va = start_va; va < end_va; va += PMD_SIZE) 948 create_pgd_mapping(pgdir, va, 949 kernel_map.phys_addr + (va - start_va), 950 PMD_SIZE, PAGE_KERNEL); 951 } 952 #else 953 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 954 { 955 uintptr_t va, end_va; 956 957 end_va = kernel_map.virt_addr + kernel_map.size; 958 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 959 create_pgd_mapping(pgdir, va, 960 kernel_map.phys_addr + (va - kernel_map.virt_addr), 961 PMD_SIZE, 962 early ? 963 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 964 } 965 #endif 966 967 /* 968 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 969 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 970 * entry. 971 */ 972 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 973 uintptr_t dtb_pa) 974 { 975 #ifndef CONFIG_BUILTIN_DTB 976 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 977 978 /* Make sure the fdt fixmap address is always aligned on PMD size */ 979 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 980 981 /* In 32-bit only, the fdt lies in its own PGD */ 982 if (!IS_ENABLED(CONFIG_64BIT)) { 983 create_pgd_mapping(early_pg_dir, fix_fdt_va, 984 pa, MAX_FDT_SIZE, PAGE_KERNEL); 985 } else { 986 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 987 pa, PMD_SIZE, PAGE_KERNEL); 988 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 989 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 990 } 991 992 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 993 #else 994 /* 995 * For 64-bit kernel, __va can't be used since it would return a linear 996 * mapping address whereas dtb_early_va will be used before 997 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 998 * kernel is mapped in the linear mapping, that makes no difference. 999 */ 1000 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 1001 #endif 1002 1003 dtb_early_pa = dtb_pa; 1004 } 1005 1006 /* 1007 * MMU is not enabled, the page tables are allocated directly using 1008 * early_pmd/pud/p4d and the address returned is the physical one. 1009 */ 1010 static void __init pt_ops_set_early(void) 1011 { 1012 pt_ops.alloc_pte = alloc_pte_early; 1013 pt_ops.get_pte_virt = get_pte_virt_early; 1014 #ifndef __PAGETABLE_PMD_FOLDED 1015 pt_ops.alloc_pmd = alloc_pmd_early; 1016 pt_ops.get_pmd_virt = get_pmd_virt_early; 1017 pt_ops.alloc_pud = alloc_pud_early; 1018 pt_ops.get_pud_virt = get_pud_virt_early; 1019 pt_ops.alloc_p4d = alloc_p4d_early; 1020 pt_ops.get_p4d_virt = get_p4d_virt_early; 1021 #endif 1022 } 1023 1024 /* 1025 * MMU is enabled but page table setup is not complete yet. 1026 * fixmap page table alloc functions must be used as a means to temporarily 1027 * map the allocated physical pages since the linear mapping does not exist yet. 1028 * 1029 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1030 * but it will be used as described above. 1031 */ 1032 static void __init pt_ops_set_fixmap(void) 1033 { 1034 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1035 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1036 #ifndef __PAGETABLE_PMD_FOLDED 1037 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1038 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1039 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1040 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1041 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1042 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1043 #endif 1044 } 1045 1046 /* 1047 * MMU is enabled and page table setup is complete, so from now, we can use 1048 * generic page allocation functions to setup page table. 1049 */ 1050 static void __init pt_ops_set_late(void) 1051 { 1052 pt_ops.alloc_pte = alloc_pte_late; 1053 pt_ops.get_pte_virt = get_pte_virt_late; 1054 #ifndef __PAGETABLE_PMD_FOLDED 1055 pt_ops.alloc_pmd = alloc_pmd_late; 1056 pt_ops.get_pmd_virt = get_pmd_virt_late; 1057 pt_ops.alloc_pud = alloc_pud_late; 1058 pt_ops.get_pud_virt = get_pud_virt_late; 1059 pt_ops.alloc_p4d = alloc_p4d_late; 1060 pt_ops.get_p4d_virt = get_p4d_virt_late; 1061 #endif 1062 } 1063 1064 #ifdef CONFIG_RANDOMIZE_BASE 1065 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1066 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1067 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa); 1068 1069 static int __init print_nokaslr(char *p) 1070 { 1071 pr_info("Disabled KASLR"); 1072 return 0; 1073 } 1074 early_param("nokaslr", print_nokaslr); 1075 1076 unsigned long kaslr_offset(void) 1077 { 1078 return kernel_map.virt_offset; 1079 } 1080 #endif 1081 1082 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1083 { 1084 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1085 1086 #ifdef CONFIG_RANDOMIZE_BASE 1087 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1088 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa); 1089 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1090 u32 nr_pos; 1091 1092 if (kaslr_seed == 0) 1093 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1094 /* 1095 * Compute the number of positions available: we are limited 1096 * by the early page table that only has one PUD and we must 1097 * be aligned on PMD_SIZE. 1098 */ 1099 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1100 1101 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1102 } 1103 #endif 1104 1105 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1106 1107 #ifdef CONFIG_XIP_KERNEL 1108 #ifdef CONFIG_64BIT 1109 kernel_map.page_offset = PAGE_OFFSET_L3; 1110 #else 1111 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1112 #endif 1113 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1114 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1115 1116 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1117 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1118 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 1119 #endif 1120 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1121 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1122 1123 kernel_map.va_kernel_xip_text_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1124 kernel_map.va_kernel_xip_data_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr 1125 + (uintptr_t)&_sdata - (uintptr_t)&_start; 1126 #else 1127 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1128 kernel_map.phys_addr = (uintptr_t)(&_start); 1129 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1130 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1131 #endif 1132 1133 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1134 set_satp_mode(dtb_pa); 1135 set_mmap_rnd_bits_max(); 1136 #endif 1137 1138 /* 1139 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1140 * where we have the system memory layout: this allows us to align 1141 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1142 * for the linear mapping. This is only possible because the kernel 1143 * mapping lies outside the linear mapping. 1144 * In 32-bit however, as the kernel resides in the linear mapping, 1145 * setup_vm_final can not change the mapping established here, 1146 * otherwise the same kernel addresses would get mapped to different 1147 * physical addresses (if the start of dram is different from the 1148 * kernel physical address start). 1149 */ 1150 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1151 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1152 1153 memory_limit = KERN_VIRT_SIZE; 1154 1155 /* Sanity check alignment and size */ 1156 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1157 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1158 1159 #ifdef CONFIG_64BIT 1160 /* 1161 * The last 4K bytes of the addressable memory can not be mapped because 1162 * of IS_ERR_VALUE macro. 1163 */ 1164 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1165 #endif 1166 1167 #ifdef CONFIG_RELOCATABLE 1168 /* 1169 * Early page table uses only one PUD, which makes it possible 1170 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1171 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1172 * since a part of the kernel would not get mapped. 1173 */ 1174 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1175 relocate_kernel(); 1176 #endif 1177 1178 apply_early_boot_alternatives(); 1179 pt_ops_set_early(); 1180 1181 /* Setup early PGD for fixmap */ 1182 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1183 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1184 1185 #ifndef __PAGETABLE_PMD_FOLDED 1186 /* Setup fixmap P4D and PUD */ 1187 if (pgtable_l5_enabled) 1188 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1189 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1190 /* Setup fixmap PUD and PMD */ 1191 if (pgtable_l4_enabled) 1192 create_pud_mapping(fixmap_pud, FIXADDR_START, 1193 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1194 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1195 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1196 /* Setup trampoline PGD and PMD */ 1197 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1198 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1199 if (pgtable_l5_enabled) 1200 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1201 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1202 if (pgtable_l4_enabled) 1203 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1204 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1205 #ifdef CONFIG_XIP_KERNEL 1206 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1207 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1208 #else 1209 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1210 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1211 #endif 1212 #else 1213 /* Setup trampoline PGD */ 1214 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1215 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1216 #endif 1217 1218 /* 1219 * Setup early PGD covering entire kernel which will allow 1220 * us to reach paging_init(). We map all memory banks later 1221 * in setup_vm_final() below. 1222 */ 1223 create_kernel_page_table(early_pg_dir, true); 1224 1225 /* Setup early mapping for FDT early scan */ 1226 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1227 1228 /* 1229 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1230 * range can not span multiple pmds. 1231 */ 1232 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1233 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1234 1235 #ifndef __PAGETABLE_PMD_FOLDED 1236 /* 1237 * Early ioremap fixmap is already created as it lies within first 2MB 1238 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1239 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1240 * the user if not. 1241 */ 1242 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1243 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1244 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1245 WARN_ON(1); 1246 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1247 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1248 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1249 fix_to_virt(FIX_BTMAP_BEGIN)); 1250 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1251 fix_to_virt(FIX_BTMAP_END)); 1252 1253 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1254 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1255 } 1256 #endif 1257 1258 pt_ops_set_fixmap(); 1259 } 1260 1261 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end, 1262 uintptr_t fixed_map_size, const pgprot_t *pgprot) 1263 { 1264 phys_addr_t pa; 1265 uintptr_t va, map_size; 1266 1267 for (pa = start; pa < end; pa += map_size) { 1268 va = (uintptr_t)__va(pa); 1269 map_size = fixed_map_size ? fixed_map_size : 1270 best_map_size(pa, va, end - pa); 1271 1272 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1273 pgprot ? *pgprot : pgprot_from_va(va)); 1274 } 1275 } 1276 1277 static void __init create_linear_mapping_page_table(void) 1278 { 1279 phys_addr_t start, end; 1280 phys_addr_t kfence_pool __maybe_unused; 1281 u64 i; 1282 1283 #ifdef CONFIG_STRICT_KERNEL_RWX 1284 phys_addr_t ktext_start = __pa_symbol(_start); 1285 phys_addr_t ktext_size = __init_data_begin - _start; 1286 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1287 phys_addr_t krodata_size = _data - __start_rodata; 1288 1289 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1290 memblock_mark_nomap(ktext_start, ktext_size); 1291 memblock_mark_nomap(krodata_start, krodata_size); 1292 #endif 1293 1294 #ifdef CONFIG_KFENCE 1295 /* 1296 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1297 * before we setup the linear mapping so that we avoid using hugepages 1298 * for this region. 1299 */ 1300 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1301 BUG_ON(!kfence_pool); 1302 1303 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1304 __kfence_pool = __va(kfence_pool); 1305 #endif 1306 1307 /* Map all memory banks in the linear mapping */ 1308 for_each_mem_range(i, &start, &end) { 1309 if (start >= end) 1310 break; 1311 if (start <= __pa(PAGE_OFFSET) && 1312 __pa(PAGE_OFFSET) < end) 1313 start = __pa(PAGE_OFFSET); 1314 1315 create_linear_mapping_range(start, end, 0, NULL); 1316 } 1317 1318 #ifdef CONFIG_STRICT_KERNEL_RWX 1319 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL); 1320 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL); 1321 1322 memblock_clear_nomap(ktext_start, ktext_size); 1323 memblock_clear_nomap(krodata_start, krodata_size); 1324 #endif 1325 1326 #ifdef CONFIG_KFENCE 1327 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL); 1328 1329 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1330 #endif 1331 } 1332 1333 static void __init setup_vm_final(void) 1334 { 1335 /* Setup swapper PGD for fixmap */ 1336 #if !defined(CONFIG_64BIT) 1337 /* 1338 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1339 * directly in swapper_pg_dir in addition to the pgd entry that points 1340 * to fixmap_pte. 1341 */ 1342 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1343 1344 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1345 #endif 1346 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1347 __pa_symbol(fixmap_pgd_next), 1348 PGDIR_SIZE, PAGE_TABLE); 1349 1350 /* Map the linear mapping */ 1351 create_linear_mapping_page_table(); 1352 1353 /* Map the kernel */ 1354 if (IS_ENABLED(CONFIG_64BIT)) 1355 create_kernel_page_table(swapper_pg_dir, false); 1356 1357 #ifdef CONFIG_KASAN 1358 kasan_swapper_init(); 1359 #endif 1360 1361 /* Clear fixmap PTE and PMD mappings */ 1362 clear_fixmap(FIX_PTE); 1363 clear_fixmap(FIX_PMD); 1364 clear_fixmap(FIX_PUD); 1365 clear_fixmap(FIX_P4D); 1366 1367 /* Move to swapper page table */ 1368 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1369 local_flush_tlb_all(); 1370 1371 pt_ops_set_late(); 1372 } 1373 #else 1374 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1375 { 1376 dtb_early_va = (void *)dtb_pa; 1377 dtb_early_pa = dtb_pa; 1378 } 1379 1380 static inline void setup_vm_final(void) 1381 { 1382 } 1383 #endif /* CONFIG_MMU */ 1384 1385 /* 1386 * reserve_crashkernel() - reserves memory for crash kernel 1387 * 1388 * This function reserves memory area given in "crashkernel=" kernel command 1389 * line parameter. The memory reserved is used by dump capture kernel when 1390 * primary kernel is crashing. 1391 */ 1392 static void __init arch_reserve_crashkernel(void) 1393 { 1394 unsigned long long low_size = 0; 1395 unsigned long long crash_base, crash_size; 1396 bool high = false; 1397 int ret; 1398 1399 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1400 return; 1401 1402 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1403 &crash_size, &crash_base, 1404 &low_size, &high); 1405 if (ret) 1406 return; 1407 1408 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 1409 } 1410 1411 void __init paging_init(void) 1412 { 1413 setup_bootmem(); 1414 setup_vm_final(); 1415 1416 /* Depend on that Linear Mapping is ready */ 1417 memblock_allow_resize(); 1418 } 1419 1420 void __init misc_mem_init(void) 1421 { 1422 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1423 arch_numa_init(); 1424 sparse_init(); 1425 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1426 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1427 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1428 #endif 1429 zone_sizes_init(); 1430 arch_reserve_crashkernel(); 1431 memblock_dump_all(); 1432 } 1433 1434 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1435 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1436 unsigned long addr, unsigned long next) 1437 { 1438 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1439 } 1440 1441 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1442 unsigned long addr, unsigned long next) 1443 { 1444 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1445 return 1; 1446 } 1447 1448 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1449 struct vmem_altmap *altmap) 1450 { 1451 /* 1452 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1453 * can't use hugepage mappings for 2-level page table because in case of 1454 * memory hotplug, we are not able to update all the page tables with 1455 * the new PMDs. 1456 */ 1457 return vmemmap_populate_hugepages(start, end, node, altmap); 1458 } 1459 #endif 1460 1461 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1462 /* 1463 * Pre-allocates page-table pages for a specific area in the kernel 1464 * page-table. Only the level which needs to be synchronized between 1465 * all page-tables is allocated because the synchronization can be 1466 * expensive. 1467 */ 1468 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1469 const char *area) 1470 { 1471 unsigned long addr; 1472 const char *lvl; 1473 1474 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1475 pgd_t *pgd = pgd_offset_k(addr); 1476 p4d_t *p4d; 1477 pud_t *pud; 1478 pmd_t *pmd; 1479 1480 lvl = "p4d"; 1481 p4d = p4d_alloc(&init_mm, pgd, addr); 1482 if (!p4d) 1483 goto failed; 1484 1485 if (pgtable_l5_enabled) 1486 continue; 1487 1488 lvl = "pud"; 1489 pud = pud_alloc(&init_mm, p4d, addr); 1490 if (!pud) 1491 goto failed; 1492 1493 if (pgtable_l4_enabled) 1494 continue; 1495 1496 lvl = "pmd"; 1497 pmd = pmd_alloc(&init_mm, pud, addr); 1498 if (!pmd) 1499 goto failed; 1500 } 1501 return; 1502 1503 failed: 1504 /* 1505 * The pages have to be there now or they will be missing in 1506 * process page-tables later. 1507 */ 1508 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1509 } 1510 1511 #define PAGE_END KASAN_SHADOW_START 1512 1513 void __init pgtable_cache_init(void) 1514 { 1515 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1516 if (IS_ENABLED(CONFIG_MODULES)) 1517 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1518 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1519 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1520 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1521 if (IS_ENABLED(CONFIG_KASAN)) 1522 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1523 } 1524 } 1525 #endif 1526 1527 #ifdef CONFIG_EXECMEM 1528 #ifdef CONFIG_MMU 1529 static struct execmem_info execmem_info __ro_after_init; 1530 1531 struct execmem_info __init *execmem_arch_setup(void) 1532 { 1533 execmem_info = (struct execmem_info){ 1534 .ranges = { 1535 [EXECMEM_DEFAULT] = { 1536 .start = MODULES_VADDR, 1537 .end = MODULES_END, 1538 .pgprot = PAGE_KERNEL, 1539 .alignment = 1, 1540 }, 1541 [EXECMEM_KPROBES] = { 1542 .start = VMALLOC_START, 1543 .end = VMALLOC_END, 1544 .pgprot = PAGE_KERNEL_READ_EXEC, 1545 .alignment = 1, 1546 }, 1547 [EXECMEM_BPF] = { 1548 .start = BPF_JIT_REGION_START, 1549 .end = BPF_JIT_REGION_END, 1550 .pgprot = PAGE_KERNEL, 1551 .alignment = PAGE_SIZE, 1552 }, 1553 }, 1554 }; 1555 1556 return &execmem_info; 1557 } 1558 #endif /* CONFIG_MMU */ 1559 #endif /* CONFIG_EXECMEM */ 1560 1561 #ifdef CONFIG_MEMORY_HOTPLUG 1562 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1563 { 1564 struct page *page = pmd_page(*pmd); 1565 struct ptdesc *ptdesc = page_ptdesc(page); 1566 pte_t *pte; 1567 int i; 1568 1569 for (i = 0; i < PTRS_PER_PTE; i++) { 1570 pte = pte_start + i; 1571 if (!pte_none(*pte)) 1572 return; 1573 } 1574 1575 pagetable_dtor(ptdesc); 1576 if (PageReserved(page)) 1577 free_reserved_page(page); 1578 else 1579 pagetable_free(ptdesc); 1580 pmd_clear(pmd); 1581 } 1582 1583 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1584 { 1585 struct page *page = pud_page(*pud); 1586 struct ptdesc *ptdesc = page_ptdesc(page); 1587 pmd_t *pmd; 1588 int i; 1589 1590 for (i = 0; i < PTRS_PER_PMD; i++) { 1591 pmd = pmd_start + i; 1592 if (!pmd_none(*pmd)) 1593 return; 1594 } 1595 1596 if (!is_vmemmap) 1597 pagetable_dtor(ptdesc); 1598 if (PageReserved(page)) 1599 free_reserved_page(page); 1600 else 1601 pagetable_free(ptdesc); 1602 pud_clear(pud); 1603 } 1604 1605 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1606 { 1607 struct page *page = p4d_page(*p4d); 1608 pud_t *pud; 1609 int i; 1610 1611 for (i = 0; i < PTRS_PER_PUD; i++) { 1612 pud = pud_start + i; 1613 if (!pud_none(*pud)) 1614 return; 1615 } 1616 1617 if (PageReserved(page)) 1618 free_reserved_page(page); 1619 else 1620 free_pages((unsigned long)page_address(page), 0); 1621 p4d_clear(p4d); 1622 } 1623 1624 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1625 struct vmem_altmap *altmap) 1626 { 1627 int order = get_order(size); 1628 1629 if (altmap) { 1630 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1631 return; 1632 } 1633 1634 if (PageReserved(page)) { 1635 unsigned int nr_pages = 1 << order; 1636 1637 while (nr_pages--) 1638 free_reserved_page(page++); 1639 return; 1640 } 1641 1642 free_pages((unsigned long)page_address(page), order); 1643 } 1644 1645 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1646 bool is_vmemmap, struct vmem_altmap *altmap) 1647 { 1648 unsigned long next; 1649 pte_t *ptep, pte; 1650 1651 for (; addr < end; addr = next) { 1652 next = (addr + PAGE_SIZE) & PAGE_MASK; 1653 if (next > end) 1654 next = end; 1655 1656 ptep = pte_base + pte_index(addr); 1657 pte = ptep_get(ptep); 1658 if (!pte_present(*ptep)) 1659 continue; 1660 1661 pte_clear(&init_mm, addr, ptep); 1662 if (is_vmemmap) 1663 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1664 } 1665 } 1666 1667 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1668 bool is_vmemmap, struct vmem_altmap *altmap) 1669 { 1670 unsigned long next; 1671 pte_t *pte_base; 1672 pmd_t *pmdp, pmd; 1673 1674 for (; addr < end; addr = next) { 1675 next = pmd_addr_end(addr, end); 1676 pmdp = pmd_base + pmd_index(addr); 1677 pmd = pmdp_get(pmdp); 1678 if (!pmd_present(pmd)) 1679 continue; 1680 1681 if (pmd_leaf(pmd)) { 1682 pmd_clear(pmdp); 1683 if (is_vmemmap) 1684 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1685 continue; 1686 } 1687 1688 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1689 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1690 free_pte_table(pte_base, pmdp); 1691 } 1692 } 1693 1694 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1695 bool is_vmemmap, struct vmem_altmap *altmap) 1696 { 1697 unsigned long next; 1698 pud_t *pudp, pud; 1699 pmd_t *pmd_base; 1700 1701 for (; addr < end; addr = next) { 1702 next = pud_addr_end(addr, end); 1703 pudp = pud_base + pud_index(addr); 1704 pud = pudp_get(pudp); 1705 if (!pud_present(pud)) 1706 continue; 1707 1708 if (pud_leaf(pud)) { 1709 if (pgtable_l4_enabled) { 1710 pud_clear(pudp); 1711 if (is_vmemmap) 1712 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1713 } 1714 continue; 1715 } 1716 1717 pmd_base = pmd_offset(pudp, 0); 1718 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1719 1720 if (pgtable_l4_enabled) 1721 free_pmd_table(pmd_base, pudp, is_vmemmap); 1722 } 1723 } 1724 1725 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1726 bool is_vmemmap, struct vmem_altmap *altmap) 1727 { 1728 unsigned long next; 1729 p4d_t *p4dp, p4d; 1730 pud_t *pud_base; 1731 1732 for (; addr < end; addr = next) { 1733 next = p4d_addr_end(addr, end); 1734 p4dp = p4d_base + p4d_index(addr); 1735 p4d = p4dp_get(p4dp); 1736 if (!p4d_present(p4d)) 1737 continue; 1738 1739 if (p4d_leaf(p4d)) { 1740 if (pgtable_l5_enabled) { 1741 p4d_clear(p4dp); 1742 if (is_vmemmap) 1743 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1744 } 1745 continue; 1746 } 1747 1748 pud_base = pud_offset(p4dp, 0); 1749 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1750 1751 if (pgtable_l5_enabled) 1752 free_pud_table(pud_base, p4dp); 1753 } 1754 } 1755 1756 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1757 struct vmem_altmap *altmap) 1758 { 1759 unsigned long addr, next; 1760 p4d_t *p4d_base; 1761 pgd_t *pgd; 1762 1763 for (addr = va; addr < end; addr = next) { 1764 next = pgd_addr_end(addr, end); 1765 pgd = pgd_offset_k(addr); 1766 1767 if (!pgd_present(*pgd)) 1768 continue; 1769 1770 if (pgd_leaf(*pgd)) 1771 continue; 1772 1773 p4d_base = p4d_offset(pgd, 0); 1774 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1775 } 1776 1777 flush_tlb_all(); 1778 } 1779 1780 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1781 { 1782 unsigned long va = (unsigned long)__va(start); 1783 unsigned long end = (unsigned long)__va(start + size); 1784 1785 remove_pgd_mapping(va, end, false, NULL); 1786 } 1787 1788 struct range arch_get_mappable_range(void) 1789 { 1790 struct range mhp_range; 1791 1792 mhp_range.start = __pa(PAGE_OFFSET); 1793 mhp_range.end = __pa(PAGE_END - 1); 1794 return mhp_range; 1795 } 1796 1797 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1798 { 1799 int ret = 0; 1800 1801 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1802 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1803 if (ret) { 1804 remove_linear_mapping(start, size); 1805 goto out; 1806 } 1807 1808 max_pfn = PFN_UP(start + size); 1809 max_low_pfn = max_pfn; 1810 1811 out: 1812 flush_tlb_all(); 1813 return ret; 1814 } 1815 1816 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 1817 { 1818 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap); 1819 remove_linear_mapping(start, size); 1820 flush_tlb_all(); 1821 } 1822 1823 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1824 { 1825 remove_pgd_mapping(start, end, true, altmap); 1826 } 1827 #endif /* CONFIG_MEMORY_HOTPLUG */ 1828