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 28 #include <asm/fixmap.h> 29 #include <asm/io.h> 30 #include <asm/numa.h> 31 #include <asm/pgtable.h> 32 #include <asm/ptdump.h> 33 #include <asm/sections.h> 34 #include <asm/soc.h> 35 #include <asm/tlbflush.h> 36 37 #include "../kernel/head.h" 38 39 struct kernel_mapping kernel_map __ro_after_init; 40 EXPORT_SYMBOL(kernel_map); 41 #ifdef CONFIG_XIP_KERNEL 42 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 43 #endif 44 45 #ifdef CONFIG_64BIT 46 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 47 #else 48 u64 satp_mode __ro_after_init = SATP_MODE_32; 49 #endif 50 EXPORT_SYMBOL(satp_mode); 51 52 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 53 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 54 EXPORT_SYMBOL(pgtable_l4_enabled); 55 EXPORT_SYMBOL(pgtable_l5_enabled); 56 57 phys_addr_t phys_ram_base __ro_after_init; 58 EXPORT_SYMBOL(phys_ram_base); 59 60 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 61 __page_aligned_bss; 62 EXPORT_SYMBOL(empty_zero_page); 63 64 extern char _start[]; 65 void *_dtb_early_va __initdata; 66 uintptr_t _dtb_early_pa __initdata; 67 68 static phys_addr_t dma32_phys_limit __initdata; 69 70 static void __init zone_sizes_init(void) 71 { 72 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 73 74 #ifdef CONFIG_ZONE_DMA32 75 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 76 #endif 77 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 78 79 free_area_init(max_zone_pfns); 80 } 81 82 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 83 84 #define LOG2_SZ_1K ilog2(SZ_1K) 85 #define LOG2_SZ_1M ilog2(SZ_1M) 86 #define LOG2_SZ_1G ilog2(SZ_1G) 87 #define LOG2_SZ_1T ilog2(SZ_1T) 88 89 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 90 { 91 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 92 (((t) - (b)) >> LOG2_SZ_1K)); 93 } 94 95 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 96 { 97 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 98 (((t) - (b)) >> LOG2_SZ_1M)); 99 } 100 101 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 102 { 103 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 104 (((t) - (b)) >> LOG2_SZ_1G)); 105 } 106 107 #ifdef CONFIG_64BIT 108 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 109 { 110 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 111 (((t) - (b)) >> LOG2_SZ_1T)); 112 } 113 #else 114 #define print_mlt(n, b, t) do {} while (0) 115 #endif 116 117 static inline void print_ml(char *name, unsigned long b, unsigned long t) 118 { 119 unsigned long diff = t - b; 120 121 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 122 print_mlt(name, b, t); 123 else if ((diff >> LOG2_SZ_1G) >= 10) 124 print_mlg(name, b, t); 125 else if ((diff >> LOG2_SZ_1M) >= 10) 126 print_mlm(name, b, t); 127 else 128 print_mlk(name, b, t); 129 } 130 131 static void __init print_vm_layout(void) 132 { 133 pr_notice("Virtual kernel memory layout:\n"); 134 print_ml("fixmap", (unsigned long)FIXADDR_START, 135 (unsigned long)FIXADDR_TOP); 136 print_ml("pci io", (unsigned long)PCI_IO_START, 137 (unsigned long)PCI_IO_END); 138 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 139 (unsigned long)VMEMMAP_END); 140 print_ml("vmalloc", (unsigned long)VMALLOC_START, 141 (unsigned long)VMALLOC_END); 142 #ifdef CONFIG_64BIT 143 print_ml("modules", (unsigned long)MODULES_VADDR, 144 (unsigned long)MODULES_END); 145 #endif 146 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 147 (unsigned long)high_memory); 148 if (IS_ENABLED(CONFIG_64BIT)) { 149 #ifdef CONFIG_KASAN 150 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 151 #endif 152 153 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 154 (unsigned long)ADDRESS_SPACE_END); 155 } 156 } 157 #else 158 static void print_vm_layout(void) { } 159 #endif /* CONFIG_DEBUG_VM */ 160 161 void __init mem_init(void) 162 { 163 #ifdef CONFIG_FLATMEM 164 BUG_ON(!mem_map); 165 #endif /* CONFIG_FLATMEM */ 166 167 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE); 168 memblock_free_all(); 169 170 print_vm_layout(); 171 } 172 173 /* Limit the memory size via mem. */ 174 static phys_addr_t memory_limit; 175 #ifdef CONFIG_XIP_KERNEL 176 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 177 #endif /* CONFIG_XIP_KERNEL */ 178 179 static int __init early_mem(char *p) 180 { 181 u64 size; 182 183 if (!p) 184 return 1; 185 186 size = memparse(p, &p) & PAGE_MASK; 187 memory_limit = min_t(u64, size, memory_limit); 188 189 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 190 191 return 0; 192 } 193 early_param("mem", early_mem); 194 195 static void __init setup_bootmem(void) 196 { 197 phys_addr_t vmlinux_end = __pa_symbol(&_end); 198 phys_addr_t max_mapped_addr; 199 phys_addr_t phys_ram_end, vmlinux_start; 200 201 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 202 vmlinux_start = __pa_symbol(&_sdata); 203 else 204 vmlinux_start = __pa_symbol(&_start); 205 206 memblock_enforce_memory_limit(memory_limit); 207 208 /* 209 * Make sure we align the reservation on PMD_SIZE since we will 210 * map the kernel in the linear mapping as read-only: we do not want 211 * any allocation to happen between _end and the next pmd aligned page. 212 */ 213 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 214 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 215 /* 216 * Reserve from the start of the kernel to the end of the kernel 217 */ 218 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 219 220 phys_ram_end = memblock_end_of_DRAM(); 221 222 /* 223 * Make sure we align the start of the memory on a PMD boundary so that 224 * at worst, we map the linear mapping with PMD mappings. 225 */ 226 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 227 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 228 229 /* 230 * In 64-bit, any use of __va/__pa before this point is wrong as we 231 * did not know the start of DRAM before. 232 */ 233 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 234 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 235 236 /* 237 * Reserve physical address space that would be mapped to virtual 238 * addresses greater than (void *)(-PAGE_SIZE) because: 239 * - This memory would overlap with ERR_PTR 240 * - This memory belongs to high memory, which is not supported 241 * 242 * This is not applicable to 64-bit kernel, because virtual addresses 243 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 244 * occupied by kernel mapping. Also it is unrealistic for high memory 245 * to exist on 64-bit platforms. 246 */ 247 if (!IS_ENABLED(CONFIG_64BIT)) { 248 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 249 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 250 } 251 252 min_low_pfn = PFN_UP(phys_ram_base); 253 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 254 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 255 256 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 257 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 258 259 reserve_initrd_mem(); 260 261 /* 262 * No allocation should be done before reserving the memory as defined 263 * in the device tree, otherwise the allocation could end up in a 264 * reserved region. 265 */ 266 early_init_fdt_scan_reserved_mem(); 267 268 /* 269 * If DTB is built in, no need to reserve its memblock. 270 * Otherwise, do reserve it but avoid using 271 * early_init_fdt_reserve_self() since __pa() does 272 * not work for DTB pointers that are fixmap addresses 273 */ 274 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 275 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 276 277 dma_contiguous_reserve(dma32_phys_limit); 278 if (IS_ENABLED(CONFIG_64BIT)) 279 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 280 } 281 282 #ifdef CONFIG_MMU 283 struct pt_alloc_ops pt_ops __initdata; 284 285 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 286 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 287 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 288 289 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 290 291 #ifdef CONFIG_XIP_KERNEL 292 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 293 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 294 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 295 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 296 #endif /* CONFIG_XIP_KERNEL */ 297 298 static const pgprot_t protection_map[16] = { 299 [VM_NONE] = PAGE_NONE, 300 [VM_READ] = PAGE_READ, 301 [VM_WRITE] = PAGE_COPY, 302 [VM_WRITE | VM_READ] = PAGE_COPY, 303 [VM_EXEC] = PAGE_EXEC, 304 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 305 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 306 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 307 [VM_SHARED] = PAGE_NONE, 308 [VM_SHARED | VM_READ] = PAGE_READ, 309 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 310 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 311 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 312 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 313 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 314 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 315 }; 316 DECLARE_VM_GET_PAGE_PROT 317 318 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 319 { 320 unsigned long addr = __fix_to_virt(idx); 321 pte_t *ptep; 322 323 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 324 325 ptep = &fixmap_pte[pte_index(addr)]; 326 327 if (pgprot_val(prot)) 328 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 329 else 330 pte_clear(&init_mm, addr, ptep); 331 local_flush_tlb_page(addr); 332 } 333 334 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 335 { 336 return (pte_t *)((uintptr_t)pa); 337 } 338 339 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 340 { 341 clear_fixmap(FIX_PTE); 342 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 343 } 344 345 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 346 { 347 return (pte_t *) __va(pa); 348 } 349 350 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 351 { 352 /* 353 * We only create PMD or PGD early mappings so we 354 * should never reach here with MMU disabled. 355 */ 356 BUG(); 357 } 358 359 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 360 { 361 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 362 } 363 364 static phys_addr_t __init alloc_pte_late(uintptr_t va) 365 { 366 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 367 368 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 369 return __pa((pte_t *)ptdesc_address(ptdesc)); 370 } 371 372 static void __init create_pte_mapping(pte_t *ptep, 373 uintptr_t va, phys_addr_t pa, 374 phys_addr_t sz, pgprot_t prot) 375 { 376 uintptr_t pte_idx = pte_index(va); 377 378 BUG_ON(sz != PAGE_SIZE); 379 380 if (pte_none(ptep[pte_idx])) 381 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 382 } 383 384 #ifndef __PAGETABLE_PMD_FOLDED 385 386 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 387 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 388 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 389 390 #ifdef CONFIG_XIP_KERNEL 391 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 392 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 393 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 394 #endif /* CONFIG_XIP_KERNEL */ 395 396 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 397 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 398 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 399 400 #ifdef CONFIG_XIP_KERNEL 401 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 402 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 403 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 404 #endif /* CONFIG_XIP_KERNEL */ 405 406 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 407 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 408 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 409 410 #ifdef CONFIG_XIP_KERNEL 411 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 412 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 413 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 414 #endif /* CONFIG_XIP_KERNEL */ 415 416 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 417 { 418 /* Before MMU is enabled */ 419 return (pmd_t *)((uintptr_t)pa); 420 } 421 422 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 423 { 424 clear_fixmap(FIX_PMD); 425 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 426 } 427 428 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 429 { 430 return (pmd_t *) __va(pa); 431 } 432 433 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 434 { 435 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 436 437 return (uintptr_t)early_pmd; 438 } 439 440 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 441 { 442 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 443 } 444 445 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 446 { 447 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 448 449 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 450 return __pa((pmd_t *)ptdesc_address(ptdesc)); 451 } 452 453 static void __init create_pmd_mapping(pmd_t *pmdp, 454 uintptr_t va, phys_addr_t pa, 455 phys_addr_t sz, pgprot_t prot) 456 { 457 pte_t *ptep; 458 phys_addr_t pte_phys; 459 uintptr_t pmd_idx = pmd_index(va); 460 461 if (sz == PMD_SIZE) { 462 if (pmd_none(pmdp[pmd_idx])) 463 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 464 return; 465 } 466 467 if (pmd_none(pmdp[pmd_idx])) { 468 pte_phys = pt_ops.alloc_pte(va); 469 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 470 ptep = pt_ops.get_pte_virt(pte_phys); 471 memset(ptep, 0, PAGE_SIZE); 472 } else { 473 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 474 ptep = pt_ops.get_pte_virt(pte_phys); 475 } 476 477 create_pte_mapping(ptep, va, pa, sz, prot); 478 } 479 480 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 481 { 482 return (pud_t *)((uintptr_t)pa); 483 } 484 485 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 486 { 487 clear_fixmap(FIX_PUD); 488 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 489 } 490 491 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 492 { 493 return (pud_t *)__va(pa); 494 } 495 496 static phys_addr_t __init alloc_pud_early(uintptr_t va) 497 { 498 /* Only one PUD is available for early mapping */ 499 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 500 501 return (uintptr_t)early_pud; 502 } 503 504 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 505 { 506 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 507 } 508 509 static phys_addr_t alloc_pud_late(uintptr_t va) 510 { 511 unsigned long vaddr; 512 513 vaddr = __get_free_page(GFP_KERNEL); 514 BUG_ON(!vaddr); 515 return __pa(vaddr); 516 } 517 518 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 519 { 520 return (p4d_t *)((uintptr_t)pa); 521 } 522 523 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 524 { 525 clear_fixmap(FIX_P4D); 526 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 527 } 528 529 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 530 { 531 return (p4d_t *)__va(pa); 532 } 533 534 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 535 { 536 /* Only one P4D is available for early mapping */ 537 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 538 539 return (uintptr_t)early_p4d; 540 } 541 542 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 543 { 544 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 545 } 546 547 static phys_addr_t alloc_p4d_late(uintptr_t va) 548 { 549 unsigned long vaddr; 550 551 vaddr = __get_free_page(GFP_KERNEL); 552 BUG_ON(!vaddr); 553 return __pa(vaddr); 554 } 555 556 static void __init create_pud_mapping(pud_t *pudp, 557 uintptr_t va, phys_addr_t pa, 558 phys_addr_t sz, pgprot_t prot) 559 { 560 pmd_t *nextp; 561 phys_addr_t next_phys; 562 uintptr_t pud_index = pud_index(va); 563 564 if (sz == PUD_SIZE) { 565 if (pud_val(pudp[pud_index]) == 0) 566 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 567 return; 568 } 569 570 if (pud_val(pudp[pud_index]) == 0) { 571 next_phys = pt_ops.alloc_pmd(va); 572 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 573 nextp = pt_ops.get_pmd_virt(next_phys); 574 memset(nextp, 0, PAGE_SIZE); 575 } else { 576 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 577 nextp = pt_ops.get_pmd_virt(next_phys); 578 } 579 580 create_pmd_mapping(nextp, va, pa, sz, prot); 581 } 582 583 static void __init create_p4d_mapping(p4d_t *p4dp, 584 uintptr_t va, phys_addr_t pa, 585 phys_addr_t sz, pgprot_t prot) 586 { 587 pud_t *nextp; 588 phys_addr_t next_phys; 589 uintptr_t p4d_index = p4d_index(va); 590 591 if (sz == P4D_SIZE) { 592 if (p4d_val(p4dp[p4d_index]) == 0) 593 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 594 return; 595 } 596 597 if (p4d_val(p4dp[p4d_index]) == 0) { 598 next_phys = pt_ops.alloc_pud(va); 599 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 600 nextp = pt_ops.get_pud_virt(next_phys); 601 memset(nextp, 0, PAGE_SIZE); 602 } else { 603 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 604 nextp = pt_ops.get_pud_virt(next_phys); 605 } 606 607 create_pud_mapping(nextp, va, pa, sz, prot); 608 } 609 610 #define pgd_next_t p4d_t 611 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 612 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 613 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 614 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 615 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 616 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 617 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 618 (pgtable_l5_enabled ? \ 619 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 620 (pgtable_l4_enabled ? \ 621 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 622 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 623 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 624 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 625 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 626 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 627 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 628 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 629 #else 630 #define pgd_next_t pte_t 631 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 632 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 633 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 634 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 635 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 636 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 637 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 638 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 639 #endif /* __PAGETABLE_PMD_FOLDED */ 640 641 void __init create_pgd_mapping(pgd_t *pgdp, 642 uintptr_t va, phys_addr_t pa, 643 phys_addr_t sz, pgprot_t prot) 644 { 645 pgd_next_t *nextp; 646 phys_addr_t next_phys; 647 uintptr_t pgd_idx = pgd_index(va); 648 649 if (sz == PGDIR_SIZE) { 650 if (pgd_val(pgdp[pgd_idx]) == 0) 651 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 652 return; 653 } 654 655 if (pgd_val(pgdp[pgd_idx]) == 0) { 656 next_phys = alloc_pgd_next(va); 657 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 658 nextp = get_pgd_next_virt(next_phys); 659 memset(nextp, 0, PAGE_SIZE); 660 } else { 661 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 662 nextp = get_pgd_next_virt(next_phys); 663 } 664 665 create_pgd_next_mapping(nextp, va, pa, sz, prot); 666 } 667 668 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va, 669 phys_addr_t size) 670 { 671 if (debug_pagealloc_enabled()) 672 return PAGE_SIZE; 673 674 if (pgtable_l5_enabled && 675 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 676 return P4D_SIZE; 677 678 if (pgtable_l4_enabled && 679 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 680 return PUD_SIZE; 681 682 if (IS_ENABLED(CONFIG_64BIT) && 683 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 684 return PMD_SIZE; 685 686 return PAGE_SIZE; 687 } 688 689 #ifdef CONFIG_XIP_KERNEL 690 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 691 extern char _xiprom[], _exiprom[], __data_loc; 692 693 /* called from head.S with MMU off */ 694 asmlinkage void __init __copy_data(void) 695 { 696 void *from = (void *)(&__data_loc); 697 void *to = (void *)CONFIG_PHYS_RAM_BASE; 698 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 699 700 memcpy(to, from, sz); 701 } 702 #endif 703 704 #ifdef CONFIG_STRICT_KERNEL_RWX 705 static __init pgprot_t pgprot_from_va(uintptr_t va) 706 { 707 if (is_va_kernel_text(va)) 708 return PAGE_KERNEL_READ_EXEC; 709 710 /* 711 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 712 * we must protect its linear mapping alias from being executed and 713 * written. 714 * And rodata section is marked readonly in mark_rodata_ro. 715 */ 716 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 717 return PAGE_KERNEL_READ; 718 719 return PAGE_KERNEL; 720 } 721 722 void mark_rodata_ro(void) 723 { 724 set_kernel_memory(__start_rodata, _data, set_memory_ro); 725 if (IS_ENABLED(CONFIG_64BIT)) 726 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 727 set_memory_ro); 728 729 debug_checkwx(); 730 } 731 #else 732 static __init pgprot_t pgprot_from_va(uintptr_t va) 733 { 734 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 735 return PAGE_KERNEL; 736 737 return PAGE_KERNEL_EXEC; 738 } 739 #endif /* CONFIG_STRICT_KERNEL_RWX */ 740 741 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 742 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 743 744 static void __init disable_pgtable_l5(void) 745 { 746 pgtable_l5_enabled = false; 747 kernel_map.page_offset = PAGE_OFFSET_L4; 748 satp_mode = SATP_MODE_48; 749 } 750 751 static void __init disable_pgtable_l4(void) 752 { 753 pgtable_l4_enabled = false; 754 kernel_map.page_offset = PAGE_OFFSET_L3; 755 satp_mode = SATP_MODE_39; 756 } 757 758 static int __init print_no4lvl(char *p) 759 { 760 pr_info("Disabled 4-level and 5-level paging"); 761 return 0; 762 } 763 early_param("no4lvl", print_no4lvl); 764 765 static int __init print_no5lvl(char *p) 766 { 767 pr_info("Disabled 5-level paging"); 768 return 0; 769 } 770 early_param("no5lvl", print_no5lvl); 771 772 /* 773 * There is a simple way to determine if 4-level is supported by the 774 * underlying hardware: establish 1:1 mapping in 4-level page table mode 775 * then read SATP to see if the configuration was taken into account 776 * meaning sv48 is supported. 777 */ 778 static __init void set_satp_mode(uintptr_t dtb_pa) 779 { 780 u64 identity_satp, hw_satp; 781 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 782 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 783 784 if (satp_mode_cmdline == SATP_MODE_57) { 785 disable_pgtable_l5(); 786 } else if (satp_mode_cmdline == SATP_MODE_48) { 787 disable_pgtable_l5(); 788 disable_pgtable_l4(); 789 return; 790 } 791 792 create_p4d_mapping(early_p4d, 793 set_satp_mode_pmd, (uintptr_t)early_pud, 794 P4D_SIZE, PAGE_TABLE); 795 create_pud_mapping(early_pud, 796 set_satp_mode_pmd, (uintptr_t)early_pmd, 797 PUD_SIZE, PAGE_TABLE); 798 /* Handle the case where set_satp_mode straddles 2 PMDs */ 799 create_pmd_mapping(early_pmd, 800 set_satp_mode_pmd, set_satp_mode_pmd, 801 PMD_SIZE, PAGE_KERNEL_EXEC); 802 create_pmd_mapping(early_pmd, 803 set_satp_mode_pmd + PMD_SIZE, 804 set_satp_mode_pmd + PMD_SIZE, 805 PMD_SIZE, PAGE_KERNEL_EXEC); 806 retry: 807 create_pgd_mapping(early_pg_dir, 808 set_satp_mode_pmd, 809 pgtable_l5_enabled ? 810 (uintptr_t)early_p4d : (uintptr_t)early_pud, 811 PGDIR_SIZE, PAGE_TABLE); 812 813 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 814 815 local_flush_tlb_all(); 816 csr_write(CSR_SATP, identity_satp); 817 hw_satp = csr_swap(CSR_SATP, 0ULL); 818 local_flush_tlb_all(); 819 820 if (hw_satp != identity_satp) { 821 if (pgtable_l5_enabled) { 822 disable_pgtable_l5(); 823 memset(early_pg_dir, 0, PAGE_SIZE); 824 goto retry; 825 } 826 disable_pgtable_l4(); 827 } 828 829 memset(early_pg_dir, 0, PAGE_SIZE); 830 memset(early_p4d, 0, PAGE_SIZE); 831 memset(early_pud, 0, PAGE_SIZE); 832 memset(early_pmd, 0, PAGE_SIZE); 833 } 834 #endif 835 836 /* 837 * setup_vm() is called from head.S with MMU-off. 838 * 839 * Following requirements should be honoured for setup_vm() to work 840 * correctly: 841 * 1) It should use PC-relative addressing for accessing kernel symbols. 842 * To achieve this we always use GCC cmodel=medany. 843 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 844 * so disable compiler instrumentation when FTRACE is enabled. 845 * 846 * Currently, the above requirements are honoured by using custom CFLAGS 847 * for init.o in mm/Makefile. 848 */ 849 850 #ifndef __riscv_cmodel_medany 851 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 852 #endif 853 854 #ifdef CONFIG_RELOCATABLE 855 extern unsigned long __rela_dyn_start, __rela_dyn_end; 856 857 static void __init relocate_kernel(void) 858 { 859 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 860 /* 861 * This holds the offset between the linked virtual address and the 862 * relocated virtual address. 863 */ 864 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 865 /* 866 * This holds the offset between kernel linked virtual address and 867 * physical address. 868 */ 869 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 870 871 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 872 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 873 Elf64_Addr relocated_addr = rela->r_addend; 874 875 if (rela->r_info != R_RISCV_RELATIVE) 876 continue; 877 878 /* 879 * Make sure to not relocate vdso symbols like rt_sigreturn 880 * which are linked from the address 0 in vmlinux since 881 * vdso symbol addresses are actually used as an offset from 882 * mm->context.vdso in VDSO_OFFSET macro. 883 */ 884 if (relocated_addr >= KERNEL_LINK_ADDR) 885 relocated_addr += reloc_offset; 886 887 *(Elf64_Addr *)addr = relocated_addr; 888 } 889 } 890 #endif /* CONFIG_RELOCATABLE */ 891 892 #ifdef CONFIG_XIP_KERNEL 893 static void __init create_kernel_page_table(pgd_t *pgdir, 894 __always_unused bool early) 895 { 896 uintptr_t va, end_va; 897 898 /* Map the flash resident part */ 899 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 900 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 901 create_pgd_mapping(pgdir, va, 902 kernel_map.xiprom + (va - kernel_map.virt_addr), 903 PMD_SIZE, PAGE_KERNEL_EXEC); 904 905 /* Map the data in RAM */ 906 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 907 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 908 create_pgd_mapping(pgdir, va, 909 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 910 PMD_SIZE, PAGE_KERNEL); 911 } 912 #else 913 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 914 { 915 uintptr_t va, end_va; 916 917 end_va = kernel_map.virt_addr + kernel_map.size; 918 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 919 create_pgd_mapping(pgdir, va, 920 kernel_map.phys_addr + (va - kernel_map.virt_addr), 921 PMD_SIZE, 922 early ? 923 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 924 } 925 #endif 926 927 /* 928 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 929 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 930 * entry. 931 */ 932 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 933 uintptr_t dtb_pa) 934 { 935 #ifndef CONFIG_BUILTIN_DTB 936 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 937 938 /* Make sure the fdt fixmap address is always aligned on PMD size */ 939 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 940 941 /* In 32-bit only, the fdt lies in its own PGD */ 942 if (!IS_ENABLED(CONFIG_64BIT)) { 943 create_pgd_mapping(early_pg_dir, fix_fdt_va, 944 pa, MAX_FDT_SIZE, PAGE_KERNEL); 945 } else { 946 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 947 pa, PMD_SIZE, PAGE_KERNEL); 948 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 949 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 950 } 951 952 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 953 #else 954 /* 955 * For 64-bit kernel, __va can't be used since it would return a linear 956 * mapping address whereas dtb_early_va will be used before 957 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 958 * kernel is mapped in the linear mapping, that makes no difference. 959 */ 960 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 961 #endif 962 963 dtb_early_pa = dtb_pa; 964 } 965 966 /* 967 * MMU is not enabled, the page tables are allocated directly using 968 * early_pmd/pud/p4d and the address returned is the physical one. 969 */ 970 static void __init pt_ops_set_early(void) 971 { 972 pt_ops.alloc_pte = alloc_pte_early; 973 pt_ops.get_pte_virt = get_pte_virt_early; 974 #ifndef __PAGETABLE_PMD_FOLDED 975 pt_ops.alloc_pmd = alloc_pmd_early; 976 pt_ops.get_pmd_virt = get_pmd_virt_early; 977 pt_ops.alloc_pud = alloc_pud_early; 978 pt_ops.get_pud_virt = get_pud_virt_early; 979 pt_ops.alloc_p4d = alloc_p4d_early; 980 pt_ops.get_p4d_virt = get_p4d_virt_early; 981 #endif 982 } 983 984 /* 985 * MMU is enabled but page table setup is not complete yet. 986 * fixmap page table alloc functions must be used as a means to temporarily 987 * map the allocated physical pages since the linear mapping does not exist yet. 988 * 989 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 990 * but it will be used as described above. 991 */ 992 static void __init pt_ops_set_fixmap(void) 993 { 994 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 995 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 996 #ifndef __PAGETABLE_PMD_FOLDED 997 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 998 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 999 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1000 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1001 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1002 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1003 #endif 1004 } 1005 1006 /* 1007 * MMU is enabled and page table setup is complete, so from now, we can use 1008 * generic page allocation functions to setup page table. 1009 */ 1010 static void __init pt_ops_set_late(void) 1011 { 1012 pt_ops.alloc_pte = alloc_pte_late; 1013 pt_ops.get_pte_virt = get_pte_virt_late; 1014 #ifndef __PAGETABLE_PMD_FOLDED 1015 pt_ops.alloc_pmd = alloc_pmd_late; 1016 pt_ops.get_pmd_virt = get_pmd_virt_late; 1017 pt_ops.alloc_pud = alloc_pud_late; 1018 pt_ops.get_pud_virt = get_pud_virt_late; 1019 pt_ops.alloc_p4d = alloc_p4d_late; 1020 pt_ops.get_p4d_virt = get_p4d_virt_late; 1021 #endif 1022 } 1023 1024 #ifdef CONFIG_RANDOMIZE_BASE 1025 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1026 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1027 1028 static int __init print_nokaslr(char *p) 1029 { 1030 pr_info("Disabled KASLR"); 1031 return 0; 1032 } 1033 early_param("nokaslr", print_nokaslr); 1034 1035 unsigned long kaslr_offset(void) 1036 { 1037 return kernel_map.virt_offset; 1038 } 1039 #endif 1040 1041 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1042 { 1043 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1044 1045 #ifdef CONFIG_RANDOMIZE_BASE 1046 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1047 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1048 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1049 u32 nr_pos; 1050 1051 /* 1052 * Compute the number of positions available: we are limited 1053 * by the early page table that only has one PUD and we must 1054 * be aligned on PMD_SIZE. 1055 */ 1056 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1057 1058 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1059 } 1060 #endif 1061 1062 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1063 1064 #ifdef CONFIG_XIP_KERNEL 1065 #ifdef CONFIG_64BIT 1066 kernel_map.page_offset = PAGE_OFFSET_L3; 1067 #else 1068 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1069 #endif 1070 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1071 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1072 1073 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1074 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1075 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1076 1077 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1078 #else 1079 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1080 kernel_map.phys_addr = (uintptr_t)(&_start); 1081 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1082 #endif 1083 1084 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1085 set_satp_mode(dtb_pa); 1086 #endif 1087 1088 /* 1089 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1090 * where we have the system memory layout: this allows us to align 1091 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1092 * for the linear mapping. This is only possible because the kernel 1093 * mapping lies outside the linear mapping. 1094 * In 32-bit however, as the kernel resides in the linear mapping, 1095 * setup_vm_final can not change the mapping established here, 1096 * otherwise the same kernel addresses would get mapped to different 1097 * physical addresses (if the start of dram is different from the 1098 * kernel physical address start). 1099 */ 1100 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1101 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1102 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1103 1104 /* 1105 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1106 * kernel, whereas for 64-bit kernel, the end of the virtual address 1107 * space is occupied by the modules/BPF/kernel mappings which reduces 1108 * the available size of the linear mapping. 1109 */ 1110 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1111 1112 /* Sanity check alignment and size */ 1113 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1114 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1115 1116 #ifdef CONFIG_64BIT 1117 /* 1118 * The last 4K bytes of the addressable memory can not be mapped because 1119 * of IS_ERR_VALUE macro. 1120 */ 1121 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1122 #endif 1123 1124 #ifdef CONFIG_RELOCATABLE 1125 /* 1126 * Early page table uses only one PUD, which makes it possible 1127 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1128 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1129 * since a part of the kernel would not get mapped. 1130 */ 1131 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1132 relocate_kernel(); 1133 #endif 1134 1135 apply_early_boot_alternatives(); 1136 pt_ops_set_early(); 1137 1138 /* Setup early PGD for fixmap */ 1139 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1140 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1141 1142 #ifndef __PAGETABLE_PMD_FOLDED 1143 /* Setup fixmap P4D and PUD */ 1144 if (pgtable_l5_enabled) 1145 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1146 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1147 /* Setup fixmap PUD and PMD */ 1148 if (pgtable_l4_enabled) 1149 create_pud_mapping(fixmap_pud, FIXADDR_START, 1150 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1151 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1152 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1153 /* Setup trampoline PGD and PMD */ 1154 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1155 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1156 if (pgtable_l5_enabled) 1157 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1158 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1159 if (pgtable_l4_enabled) 1160 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1161 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1162 #ifdef CONFIG_XIP_KERNEL 1163 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1164 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1165 #else 1166 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1167 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1168 #endif 1169 #else 1170 /* Setup trampoline PGD */ 1171 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1172 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1173 #endif 1174 1175 /* 1176 * Setup early PGD covering entire kernel which will allow 1177 * us to reach paging_init(). We map all memory banks later 1178 * in setup_vm_final() below. 1179 */ 1180 create_kernel_page_table(early_pg_dir, true); 1181 1182 /* Setup early mapping for FDT early scan */ 1183 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1184 1185 /* 1186 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1187 * range can not span multiple pmds. 1188 */ 1189 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1190 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1191 1192 #ifndef __PAGETABLE_PMD_FOLDED 1193 /* 1194 * Early ioremap fixmap is already created as it lies within first 2MB 1195 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1196 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1197 * the user if not. 1198 */ 1199 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1200 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1201 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1202 WARN_ON(1); 1203 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1204 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1205 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1206 fix_to_virt(FIX_BTMAP_BEGIN)); 1207 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1208 fix_to_virt(FIX_BTMAP_END)); 1209 1210 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1211 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1212 } 1213 #endif 1214 1215 pt_ops_set_fixmap(); 1216 } 1217 1218 static void __init create_linear_mapping_range(phys_addr_t start, 1219 phys_addr_t end, 1220 uintptr_t fixed_map_size) 1221 { 1222 phys_addr_t pa; 1223 uintptr_t va, map_size; 1224 1225 for (pa = start; pa < end; pa += map_size) { 1226 va = (uintptr_t)__va(pa); 1227 map_size = fixed_map_size ? fixed_map_size : 1228 best_map_size(pa, va, end - pa); 1229 1230 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1231 pgprot_from_va(va)); 1232 } 1233 } 1234 1235 static void __init create_linear_mapping_page_table(void) 1236 { 1237 phys_addr_t start, end; 1238 phys_addr_t kfence_pool __maybe_unused; 1239 u64 i; 1240 1241 #ifdef CONFIG_STRICT_KERNEL_RWX 1242 phys_addr_t ktext_start = __pa_symbol(_start); 1243 phys_addr_t ktext_size = __init_data_begin - _start; 1244 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1245 phys_addr_t krodata_size = _data - __start_rodata; 1246 1247 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1248 memblock_mark_nomap(ktext_start, ktext_size); 1249 memblock_mark_nomap(krodata_start, krodata_size); 1250 #endif 1251 1252 #ifdef CONFIG_KFENCE 1253 /* 1254 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1255 * before we setup the linear mapping so that we avoid using hugepages 1256 * for this region. 1257 */ 1258 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1259 BUG_ON(!kfence_pool); 1260 1261 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1262 __kfence_pool = __va(kfence_pool); 1263 #endif 1264 1265 /* Map all memory banks in the linear mapping */ 1266 for_each_mem_range(i, &start, &end) { 1267 if (start >= end) 1268 break; 1269 if (start <= __pa(PAGE_OFFSET) && 1270 __pa(PAGE_OFFSET) < end) 1271 start = __pa(PAGE_OFFSET); 1272 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1273 end = __pa(PAGE_OFFSET) + memory_limit; 1274 1275 create_linear_mapping_range(start, end, 0); 1276 } 1277 1278 #ifdef CONFIG_STRICT_KERNEL_RWX 1279 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0); 1280 create_linear_mapping_range(krodata_start, 1281 krodata_start + krodata_size, 0); 1282 1283 memblock_clear_nomap(ktext_start, ktext_size); 1284 memblock_clear_nomap(krodata_start, krodata_size); 1285 #endif 1286 1287 #ifdef CONFIG_KFENCE 1288 create_linear_mapping_range(kfence_pool, 1289 kfence_pool + KFENCE_POOL_SIZE, 1290 PAGE_SIZE); 1291 1292 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1293 #endif 1294 } 1295 1296 static void __init setup_vm_final(void) 1297 { 1298 /* Setup swapper PGD for fixmap */ 1299 #if !defined(CONFIG_64BIT) 1300 /* 1301 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1302 * directly in swapper_pg_dir in addition to the pgd entry that points 1303 * to fixmap_pte. 1304 */ 1305 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1306 1307 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1308 #endif 1309 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1310 __pa_symbol(fixmap_pgd_next), 1311 PGDIR_SIZE, PAGE_TABLE); 1312 1313 /* Map the linear mapping */ 1314 create_linear_mapping_page_table(); 1315 1316 /* Map the kernel */ 1317 if (IS_ENABLED(CONFIG_64BIT)) 1318 create_kernel_page_table(swapper_pg_dir, false); 1319 1320 #ifdef CONFIG_KASAN 1321 kasan_swapper_init(); 1322 #endif 1323 1324 /* Clear fixmap PTE and PMD mappings */ 1325 clear_fixmap(FIX_PTE); 1326 clear_fixmap(FIX_PMD); 1327 clear_fixmap(FIX_PUD); 1328 clear_fixmap(FIX_P4D); 1329 1330 /* Move to swapper page table */ 1331 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1332 local_flush_tlb_all(); 1333 1334 pt_ops_set_late(); 1335 } 1336 #else 1337 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1338 { 1339 dtb_early_va = (void *)dtb_pa; 1340 dtb_early_pa = dtb_pa; 1341 } 1342 1343 static inline void setup_vm_final(void) 1344 { 1345 } 1346 #endif /* CONFIG_MMU */ 1347 1348 /* Reserve 128M low memory by default for swiotlb buffer */ 1349 #define DEFAULT_CRASH_KERNEL_LOW_SIZE (128UL << 20) 1350 1351 static int __init reserve_crashkernel_low(unsigned long long low_size) 1352 { 1353 unsigned long long low_base; 1354 1355 low_base = memblock_phys_alloc_range(low_size, PMD_SIZE, 0, dma32_phys_limit); 1356 if (!low_base) { 1357 pr_err("cannot allocate crashkernel low memory (size:0x%llx).\n", low_size); 1358 return -ENOMEM; 1359 } 1360 1361 pr_info("crashkernel low memory reserved: 0x%016llx - 0x%016llx (%lld MB)\n", 1362 low_base, low_base + low_size, low_size >> 20); 1363 1364 crashk_low_res.start = low_base; 1365 crashk_low_res.end = low_base + low_size - 1; 1366 1367 return 0; 1368 } 1369 1370 /* 1371 * reserve_crashkernel() - reserves memory for crash kernel 1372 * 1373 * This function reserves memory area given in "crashkernel=" kernel command 1374 * line parameter. The memory reserved is used by dump capture kernel when 1375 * primary kernel is crashing. 1376 */ 1377 static void __init reserve_crashkernel(void) 1378 { 1379 unsigned long long crash_base = 0; 1380 unsigned long long crash_size = 0; 1381 unsigned long long crash_low_size = 0; 1382 unsigned long search_start = memblock_start_of_DRAM(); 1383 unsigned long search_end = (unsigned long)dma32_phys_limit; 1384 char *cmdline = boot_command_line; 1385 bool fixed_base = false; 1386 bool high = false; 1387 1388 int ret = 0; 1389 1390 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 1391 return; 1392 /* 1393 * Don't reserve a region for a crash kernel on a crash kernel 1394 * since it doesn't make much sense and we have limited memory 1395 * resources. 1396 */ 1397 if (is_kdump_kernel()) { 1398 pr_info("crashkernel: ignoring reservation request\n"); 1399 return; 1400 } 1401 1402 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 1403 &crash_size, &crash_base); 1404 if (ret == -ENOENT) { 1405 /* Fallback to crashkernel=X,[high,low] */ 1406 ret = parse_crashkernel_high(cmdline, 0, &crash_size, &crash_base); 1407 if (ret || !crash_size) 1408 return; 1409 1410 /* 1411 * crashkernel=Y,low is valid only when crashkernel=X,high 1412 * is passed. 1413 */ 1414 ret = parse_crashkernel_low(cmdline, 0, &crash_low_size, &crash_base); 1415 if (ret == -ENOENT) 1416 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE; 1417 else if (ret) 1418 return; 1419 1420 search_start = (unsigned long)dma32_phys_limit; 1421 search_end = memblock_end_of_DRAM(); 1422 high = true; 1423 } else if (ret || !crash_size) { 1424 /* Invalid argument value specified */ 1425 return; 1426 } 1427 1428 crash_size = PAGE_ALIGN(crash_size); 1429 1430 if (crash_base) { 1431 fixed_base = true; 1432 search_start = crash_base; 1433 search_end = crash_base + crash_size; 1434 } 1435 1436 /* 1437 * Current riscv boot protocol requires 2MB alignment for 1438 * RV64 and 4MB alignment for RV32 (hugepage size) 1439 * 1440 * Try to alloc from 32bit addressible physical memory so that 1441 * swiotlb can work on the crash kernel. 1442 */ 1443 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1444 search_start, search_end); 1445 if (crash_base == 0) { 1446 /* 1447 * For crashkernel=size[KMG]@offset[KMG], print out failure 1448 * message if can't reserve the specified region. 1449 */ 1450 if (fixed_base) { 1451 pr_warn("crashkernel: allocating failed with given size@offset\n"); 1452 return; 1453 } 1454 1455 if (high) { 1456 /* 1457 * For crashkernel=size[KMG],high, if the first attempt was 1458 * for high memory, fall back to low memory. 1459 */ 1460 search_start = memblock_start_of_DRAM(); 1461 search_end = (unsigned long)dma32_phys_limit; 1462 } else { 1463 /* 1464 * For crashkernel=size[KMG], if the first attempt was for 1465 * low memory, fall back to high memory, the minimum required 1466 * low memory will be reserved later. 1467 */ 1468 search_start = (unsigned long)dma32_phys_limit; 1469 search_end = memblock_end_of_DRAM(); 1470 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE; 1471 } 1472 1473 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1474 search_start, search_end); 1475 if (crash_base == 0) { 1476 pr_warn("crashkernel: couldn't allocate %lldKB\n", 1477 crash_size >> 10); 1478 return; 1479 } 1480 } 1481 1482 if ((crash_base >= dma32_phys_limit) && crash_low_size && 1483 reserve_crashkernel_low(crash_low_size)) { 1484 memblock_phys_free(crash_base, crash_size); 1485 return; 1486 } 1487 1488 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 1489 crash_base, crash_base + crash_size, crash_size >> 20); 1490 1491 crashk_res.start = crash_base; 1492 crashk_res.end = crash_base + crash_size - 1; 1493 } 1494 1495 void __init paging_init(void) 1496 { 1497 setup_bootmem(); 1498 setup_vm_final(); 1499 1500 /* Depend on that Linear Mapping is ready */ 1501 memblock_allow_resize(); 1502 } 1503 1504 void __init misc_mem_init(void) 1505 { 1506 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1507 arch_numa_init(); 1508 sparse_init(); 1509 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1510 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1511 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1512 #endif 1513 zone_sizes_init(); 1514 reserve_crashkernel(); 1515 memblock_dump_all(); 1516 } 1517 1518 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1519 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1520 struct vmem_altmap *altmap) 1521 { 1522 return vmemmap_populate_basepages(start, end, node, NULL); 1523 } 1524 #endif 1525 1526 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1527 /* 1528 * Pre-allocates page-table pages for a specific area in the kernel 1529 * page-table. Only the level which needs to be synchronized between 1530 * all page-tables is allocated because the synchronization can be 1531 * expensive. 1532 */ 1533 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1534 const char *area) 1535 { 1536 unsigned long addr; 1537 const char *lvl; 1538 1539 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1540 pgd_t *pgd = pgd_offset_k(addr); 1541 p4d_t *p4d; 1542 pud_t *pud; 1543 pmd_t *pmd; 1544 1545 lvl = "p4d"; 1546 p4d = p4d_alloc(&init_mm, pgd, addr); 1547 if (!p4d) 1548 goto failed; 1549 1550 if (pgtable_l5_enabled) 1551 continue; 1552 1553 lvl = "pud"; 1554 pud = pud_alloc(&init_mm, p4d, addr); 1555 if (!pud) 1556 goto failed; 1557 1558 if (pgtable_l4_enabled) 1559 continue; 1560 1561 lvl = "pmd"; 1562 pmd = pmd_alloc(&init_mm, pud, addr); 1563 if (!pmd) 1564 goto failed; 1565 } 1566 return; 1567 1568 failed: 1569 /* 1570 * The pages have to be there now or they will be missing in 1571 * process page-tables later. 1572 */ 1573 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1574 } 1575 1576 void __init pgtable_cache_init(void) 1577 { 1578 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1579 if (IS_ENABLED(CONFIG_MODULES)) 1580 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1581 } 1582 #endif 1583