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/numa.h> 32 #include <asm/pgtable.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 #ifdef CONFIG_64BIT 53 bool pgtable_l4_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 54 bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 55 EXPORT_SYMBOL(pgtable_l4_enabled); 56 EXPORT_SYMBOL(pgtable_l5_enabled); 57 #endif 58 59 phys_addr_t phys_ram_base __ro_after_init; 60 EXPORT_SYMBOL(phys_ram_base); 61 62 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 63 __page_aligned_bss; 64 EXPORT_SYMBOL(empty_zero_page); 65 66 extern char _start[]; 67 void *_dtb_early_va __initdata; 68 uintptr_t _dtb_early_pa __initdata; 69 70 phys_addr_t dma32_phys_limit __initdata; 71 72 static void __init zone_sizes_init(void) 73 { 74 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 75 76 #ifdef CONFIG_ZONE_DMA32 77 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 78 #endif 79 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 80 81 free_area_init(max_zone_pfns); 82 } 83 84 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 85 86 #define LOG2_SZ_1K ilog2(SZ_1K) 87 #define LOG2_SZ_1M ilog2(SZ_1M) 88 #define LOG2_SZ_1G ilog2(SZ_1G) 89 #define LOG2_SZ_1T ilog2(SZ_1T) 90 91 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 92 { 93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 94 (((t) - (b)) >> LOG2_SZ_1K)); 95 } 96 97 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 98 { 99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 100 (((t) - (b)) >> LOG2_SZ_1M)); 101 } 102 103 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 104 { 105 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 106 (((t) - (b)) >> LOG2_SZ_1G)); 107 } 108 109 #ifdef CONFIG_64BIT 110 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 111 { 112 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 113 (((t) - (b)) >> LOG2_SZ_1T)); 114 } 115 #else 116 #define print_mlt(n, b, t) do {} while (0) 117 #endif 118 119 static inline void print_ml(char *name, unsigned long b, unsigned long t) 120 { 121 unsigned long diff = t - b; 122 123 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 124 print_mlt(name, b, t); 125 else if ((diff >> LOG2_SZ_1G) >= 10) 126 print_mlg(name, b, t); 127 else if ((diff >> LOG2_SZ_1M) >= 10) 128 print_mlm(name, b, t); 129 else 130 print_mlk(name, b, t); 131 } 132 133 static void __init print_vm_layout(void) 134 { 135 pr_notice("Virtual kernel memory layout:\n"); 136 print_ml("fixmap", (unsigned long)FIXADDR_START, 137 (unsigned long)FIXADDR_TOP); 138 print_ml("pci io", (unsigned long)PCI_IO_START, 139 (unsigned long)PCI_IO_END); 140 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 141 (unsigned long)VMEMMAP_END); 142 print_ml("vmalloc", (unsigned long)VMALLOC_START, 143 (unsigned long)VMALLOC_END); 144 #ifdef CONFIG_64BIT 145 print_ml("modules", (unsigned long)MODULES_VADDR, 146 (unsigned long)MODULES_END); 147 #endif 148 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 149 (unsigned long)high_memory); 150 if (IS_ENABLED(CONFIG_64BIT)) { 151 #ifdef CONFIG_KASAN 152 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 153 #endif 154 155 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 156 (unsigned long)ADDRESS_SPACE_END); 157 } 158 } 159 #else 160 static void print_vm_layout(void) { } 161 #endif /* CONFIG_DEBUG_VM */ 162 163 void __init mem_init(void) 164 { 165 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 166 #ifdef CONFIG_FLATMEM 167 BUG_ON(!mem_map); 168 #endif /* CONFIG_FLATMEM */ 169 170 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 171 dma_cache_alignment != 1) { 172 /* 173 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 174 * buffer per 1GB of RAM for kmalloc() bouncing on 175 * non-coherent platforms. 176 */ 177 unsigned long size = 178 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 179 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 180 swiotlb = true; 181 } 182 183 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 184 memblock_free_all(); 185 186 print_vm_layout(); 187 } 188 189 /* Limit the memory size via mem. */ 190 static phys_addr_t memory_limit; 191 #ifdef CONFIG_XIP_KERNEL 192 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 193 #endif /* CONFIG_XIP_KERNEL */ 194 195 static int __init early_mem(char *p) 196 { 197 u64 size; 198 199 if (!p) 200 return 1; 201 202 size = memparse(p, &p) & PAGE_MASK; 203 memory_limit = min_t(u64, size, memory_limit); 204 205 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 206 207 return 0; 208 } 209 early_param("mem", early_mem); 210 211 static void __init setup_bootmem(void) 212 { 213 phys_addr_t vmlinux_end = __pa_symbol(&_end); 214 phys_addr_t max_mapped_addr; 215 phys_addr_t phys_ram_end, vmlinux_start; 216 217 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 218 vmlinux_start = __pa_symbol(&_sdata); 219 else 220 vmlinux_start = __pa_symbol(&_start); 221 222 memblock_enforce_memory_limit(memory_limit); 223 224 /* 225 * Make sure we align the reservation on PMD_SIZE since we will 226 * map the kernel in the linear mapping as read-only: we do not want 227 * any allocation to happen between _end and the next pmd aligned page. 228 */ 229 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 230 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 231 /* 232 * Reserve from the start of the kernel to the end of the kernel 233 */ 234 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 235 236 phys_ram_end = memblock_end_of_DRAM(); 237 238 /* 239 * Make sure we align the start of the memory on a PMD boundary so that 240 * at worst, we map the linear mapping with PMD mappings. 241 */ 242 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 243 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 244 245 /* 246 * In 64-bit, any use of __va/__pa before this point is wrong as we 247 * did not know the start of DRAM before. 248 */ 249 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 250 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 251 252 /* 253 * Reserve physical address space that would be mapped to virtual 254 * addresses greater than (void *)(-PAGE_SIZE) because: 255 * - This memory would overlap with ERR_PTR 256 * - This memory belongs to high memory, which is not supported 257 * 258 * This is not applicable to 64-bit kernel, because virtual addresses 259 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 260 * occupied by kernel mapping. Also it is unrealistic for high memory 261 * to exist on 64-bit platforms. 262 */ 263 if (!IS_ENABLED(CONFIG_64BIT)) { 264 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 265 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 266 } 267 268 min_low_pfn = PFN_UP(phys_ram_base); 269 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 270 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 271 272 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 273 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 274 275 reserve_initrd_mem(); 276 277 /* 278 * No allocation should be done before reserving the memory as defined 279 * in the device tree, otherwise the allocation could end up in a 280 * reserved region. 281 */ 282 early_init_fdt_scan_reserved_mem(); 283 284 /* 285 * If DTB is built in, no need to reserve its memblock. 286 * Otherwise, do reserve it but avoid using 287 * early_init_fdt_reserve_self() since __pa() does 288 * not work for DTB pointers that are fixmap addresses 289 */ 290 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 291 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 292 293 dma_contiguous_reserve(dma32_phys_limit); 294 if (IS_ENABLED(CONFIG_64BIT)) 295 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 296 } 297 298 #ifdef CONFIG_MMU 299 struct pt_alloc_ops pt_ops __initdata; 300 301 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 302 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 303 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 304 305 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 306 307 #ifdef CONFIG_XIP_KERNEL 308 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 309 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 310 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 311 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 312 #endif /* CONFIG_XIP_KERNEL */ 313 314 static const pgprot_t protection_map[16] = { 315 [VM_NONE] = PAGE_NONE, 316 [VM_READ] = PAGE_READ, 317 [VM_WRITE] = PAGE_COPY, 318 [VM_WRITE | VM_READ] = PAGE_COPY, 319 [VM_EXEC] = PAGE_EXEC, 320 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 321 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 322 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 323 [VM_SHARED] = PAGE_NONE, 324 [VM_SHARED | VM_READ] = PAGE_READ, 325 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 326 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 327 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 328 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 329 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 330 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 331 }; 332 DECLARE_VM_GET_PAGE_PROT 333 334 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 335 { 336 unsigned long addr = __fix_to_virt(idx); 337 pte_t *ptep; 338 339 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 340 341 ptep = &fixmap_pte[pte_index(addr)]; 342 343 if (pgprot_val(prot)) 344 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 345 else 346 pte_clear(&init_mm, addr, ptep); 347 local_flush_tlb_page(addr); 348 } 349 350 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 351 { 352 return (pte_t *)((uintptr_t)pa); 353 } 354 355 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 356 { 357 clear_fixmap(FIX_PTE); 358 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 359 } 360 361 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 362 { 363 return (pte_t *) __va(pa); 364 } 365 366 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 367 { 368 /* 369 * We only create PMD or PGD early mappings so we 370 * should never reach here with MMU disabled. 371 */ 372 BUG(); 373 } 374 375 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 376 { 377 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 378 } 379 380 static phys_addr_t __init alloc_pte_late(uintptr_t va) 381 { 382 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 383 384 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 385 return __pa((pte_t *)ptdesc_address(ptdesc)); 386 } 387 388 static void __init create_pte_mapping(pte_t *ptep, 389 uintptr_t va, phys_addr_t pa, 390 phys_addr_t sz, pgprot_t prot) 391 { 392 uintptr_t pte_idx = pte_index(va); 393 394 BUG_ON(sz != PAGE_SIZE); 395 396 if (pte_none(ptep[pte_idx])) 397 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 398 } 399 400 #ifndef __PAGETABLE_PMD_FOLDED 401 402 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 403 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 404 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 405 406 #ifdef CONFIG_XIP_KERNEL 407 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 408 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 409 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 410 #endif /* CONFIG_XIP_KERNEL */ 411 412 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 413 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 414 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 415 416 #ifdef CONFIG_XIP_KERNEL 417 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 418 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 419 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 420 #endif /* CONFIG_XIP_KERNEL */ 421 422 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 423 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 424 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 425 426 #ifdef CONFIG_XIP_KERNEL 427 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 428 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 429 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 430 #endif /* CONFIG_XIP_KERNEL */ 431 432 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 433 { 434 /* Before MMU is enabled */ 435 return (pmd_t *)((uintptr_t)pa); 436 } 437 438 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 439 { 440 clear_fixmap(FIX_PMD); 441 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 442 } 443 444 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 445 { 446 return (pmd_t *) __va(pa); 447 } 448 449 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 450 { 451 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 452 453 return (uintptr_t)early_pmd; 454 } 455 456 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 457 { 458 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 459 } 460 461 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 462 { 463 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 464 465 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 466 return __pa((pmd_t *)ptdesc_address(ptdesc)); 467 } 468 469 static void __init create_pmd_mapping(pmd_t *pmdp, 470 uintptr_t va, phys_addr_t pa, 471 phys_addr_t sz, pgprot_t prot) 472 { 473 pte_t *ptep; 474 phys_addr_t pte_phys; 475 uintptr_t pmd_idx = pmd_index(va); 476 477 if (sz == PMD_SIZE) { 478 if (pmd_none(pmdp[pmd_idx])) 479 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 480 return; 481 } 482 483 if (pmd_none(pmdp[pmd_idx])) { 484 pte_phys = pt_ops.alloc_pte(va); 485 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 486 ptep = pt_ops.get_pte_virt(pte_phys); 487 memset(ptep, 0, PAGE_SIZE); 488 } else { 489 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 490 ptep = pt_ops.get_pte_virt(pte_phys); 491 } 492 493 create_pte_mapping(ptep, va, pa, sz, prot); 494 } 495 496 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 497 { 498 return (pud_t *)((uintptr_t)pa); 499 } 500 501 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 502 { 503 clear_fixmap(FIX_PUD); 504 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 505 } 506 507 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 508 { 509 return (pud_t *)__va(pa); 510 } 511 512 static phys_addr_t __init alloc_pud_early(uintptr_t va) 513 { 514 /* Only one PUD is available for early mapping */ 515 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 516 517 return (uintptr_t)early_pud; 518 } 519 520 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 521 { 522 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 523 } 524 525 static phys_addr_t alloc_pud_late(uintptr_t va) 526 { 527 unsigned long vaddr; 528 529 vaddr = __get_free_page(GFP_KERNEL); 530 BUG_ON(!vaddr); 531 return __pa(vaddr); 532 } 533 534 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 535 { 536 return (p4d_t *)((uintptr_t)pa); 537 } 538 539 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 540 { 541 clear_fixmap(FIX_P4D); 542 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 543 } 544 545 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 546 { 547 return (p4d_t *)__va(pa); 548 } 549 550 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 551 { 552 /* Only one P4D is available for early mapping */ 553 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 554 555 return (uintptr_t)early_p4d; 556 } 557 558 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 559 { 560 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 561 } 562 563 static phys_addr_t alloc_p4d_late(uintptr_t va) 564 { 565 unsigned long vaddr; 566 567 vaddr = __get_free_page(GFP_KERNEL); 568 BUG_ON(!vaddr); 569 return __pa(vaddr); 570 } 571 572 static void __init create_pud_mapping(pud_t *pudp, 573 uintptr_t va, phys_addr_t pa, 574 phys_addr_t sz, pgprot_t prot) 575 { 576 pmd_t *nextp; 577 phys_addr_t next_phys; 578 uintptr_t pud_index = pud_index(va); 579 580 if (sz == PUD_SIZE) { 581 if (pud_val(pudp[pud_index]) == 0) 582 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 583 return; 584 } 585 586 if (pud_val(pudp[pud_index]) == 0) { 587 next_phys = pt_ops.alloc_pmd(va); 588 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 589 nextp = pt_ops.get_pmd_virt(next_phys); 590 memset(nextp, 0, PAGE_SIZE); 591 } else { 592 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 593 nextp = pt_ops.get_pmd_virt(next_phys); 594 } 595 596 create_pmd_mapping(nextp, va, pa, sz, prot); 597 } 598 599 static void __init create_p4d_mapping(p4d_t *p4dp, 600 uintptr_t va, phys_addr_t pa, 601 phys_addr_t sz, pgprot_t prot) 602 { 603 pud_t *nextp; 604 phys_addr_t next_phys; 605 uintptr_t p4d_index = p4d_index(va); 606 607 if (sz == P4D_SIZE) { 608 if (p4d_val(p4dp[p4d_index]) == 0) 609 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 610 return; 611 } 612 613 if (p4d_val(p4dp[p4d_index]) == 0) { 614 next_phys = pt_ops.alloc_pud(va); 615 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 616 nextp = pt_ops.get_pud_virt(next_phys); 617 memset(nextp, 0, PAGE_SIZE); 618 } else { 619 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 620 nextp = pt_ops.get_pud_virt(next_phys); 621 } 622 623 create_pud_mapping(nextp, va, pa, sz, prot); 624 } 625 626 #define pgd_next_t p4d_t 627 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 628 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 629 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 630 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 631 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 632 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 633 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 634 (pgtable_l5_enabled ? \ 635 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 636 (pgtable_l4_enabled ? \ 637 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 638 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 639 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 640 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 641 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 642 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 643 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 644 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 645 #else 646 #define pgd_next_t pte_t 647 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 648 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 649 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 650 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 651 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 652 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 653 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 654 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 655 #endif /* __PAGETABLE_PMD_FOLDED */ 656 657 void __init create_pgd_mapping(pgd_t *pgdp, 658 uintptr_t va, phys_addr_t pa, 659 phys_addr_t sz, pgprot_t prot) 660 { 661 pgd_next_t *nextp; 662 phys_addr_t next_phys; 663 uintptr_t pgd_idx = pgd_index(va); 664 665 if (sz == PGDIR_SIZE) { 666 if (pgd_val(pgdp[pgd_idx]) == 0) 667 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 668 return; 669 } 670 671 if (pgd_val(pgdp[pgd_idx]) == 0) { 672 next_phys = alloc_pgd_next(va); 673 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 674 nextp = get_pgd_next_virt(next_phys); 675 memset(nextp, 0, PAGE_SIZE); 676 } else { 677 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 678 nextp = get_pgd_next_virt(next_phys); 679 } 680 681 create_pgd_next_mapping(nextp, va, pa, sz, prot); 682 } 683 684 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va, 685 phys_addr_t size) 686 { 687 if (debug_pagealloc_enabled()) 688 return PAGE_SIZE; 689 690 if (pgtable_l5_enabled && 691 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 692 return P4D_SIZE; 693 694 if (pgtable_l4_enabled && 695 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 696 return PUD_SIZE; 697 698 if (IS_ENABLED(CONFIG_64BIT) && 699 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 700 return PMD_SIZE; 701 702 return PAGE_SIZE; 703 } 704 705 #ifdef CONFIG_XIP_KERNEL 706 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 707 extern char _xiprom[], _exiprom[], __data_loc; 708 709 /* called from head.S with MMU off */ 710 asmlinkage void __init __copy_data(void) 711 { 712 void *from = (void *)(&__data_loc); 713 void *to = (void *)CONFIG_PHYS_RAM_BASE; 714 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 715 716 memcpy(to, from, sz); 717 } 718 #endif 719 720 #ifdef CONFIG_STRICT_KERNEL_RWX 721 static __init pgprot_t pgprot_from_va(uintptr_t va) 722 { 723 if (is_va_kernel_text(va)) 724 return PAGE_KERNEL_READ_EXEC; 725 726 /* 727 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 728 * we must protect its linear mapping alias from being executed and 729 * written. 730 * And rodata section is marked readonly in mark_rodata_ro. 731 */ 732 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 733 return PAGE_KERNEL_READ; 734 735 return PAGE_KERNEL; 736 } 737 738 void mark_rodata_ro(void) 739 { 740 set_kernel_memory(__start_rodata, _data, set_memory_ro); 741 if (IS_ENABLED(CONFIG_64BIT)) 742 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 743 set_memory_ro); 744 } 745 #else 746 static __init pgprot_t pgprot_from_va(uintptr_t va) 747 { 748 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 749 return PAGE_KERNEL; 750 751 return PAGE_KERNEL_EXEC; 752 } 753 #endif /* CONFIG_STRICT_KERNEL_RWX */ 754 755 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 756 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 757 758 static void __init disable_pgtable_l5(void) 759 { 760 pgtable_l5_enabled = false; 761 kernel_map.page_offset = PAGE_OFFSET_L4; 762 satp_mode = SATP_MODE_48; 763 } 764 765 static void __init disable_pgtable_l4(void) 766 { 767 pgtable_l4_enabled = false; 768 kernel_map.page_offset = PAGE_OFFSET_L3; 769 satp_mode = SATP_MODE_39; 770 } 771 772 static int __init print_no4lvl(char *p) 773 { 774 pr_info("Disabled 4-level and 5-level paging"); 775 return 0; 776 } 777 early_param("no4lvl", print_no4lvl); 778 779 static int __init print_no5lvl(char *p) 780 { 781 pr_info("Disabled 5-level paging"); 782 return 0; 783 } 784 early_param("no5lvl", print_no5lvl); 785 786 static void __init set_mmap_rnd_bits_max(void) 787 { 788 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 789 } 790 791 /* 792 * There is a simple way to determine if 4-level is supported by the 793 * underlying hardware: establish 1:1 mapping in 4-level page table mode 794 * then read SATP to see if the configuration was taken into account 795 * meaning sv48 is supported. 796 */ 797 static __init void set_satp_mode(uintptr_t dtb_pa) 798 { 799 u64 identity_satp, hw_satp; 800 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 801 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 802 803 if (satp_mode_cmdline == SATP_MODE_57) { 804 disable_pgtable_l5(); 805 } else if (satp_mode_cmdline == SATP_MODE_48) { 806 disable_pgtable_l5(); 807 disable_pgtable_l4(); 808 return; 809 } 810 811 create_p4d_mapping(early_p4d, 812 set_satp_mode_pmd, (uintptr_t)early_pud, 813 P4D_SIZE, PAGE_TABLE); 814 create_pud_mapping(early_pud, 815 set_satp_mode_pmd, (uintptr_t)early_pmd, 816 PUD_SIZE, PAGE_TABLE); 817 /* Handle the case where set_satp_mode straddles 2 PMDs */ 818 create_pmd_mapping(early_pmd, 819 set_satp_mode_pmd, set_satp_mode_pmd, 820 PMD_SIZE, PAGE_KERNEL_EXEC); 821 create_pmd_mapping(early_pmd, 822 set_satp_mode_pmd + PMD_SIZE, 823 set_satp_mode_pmd + PMD_SIZE, 824 PMD_SIZE, PAGE_KERNEL_EXEC); 825 retry: 826 create_pgd_mapping(early_pg_dir, 827 set_satp_mode_pmd, 828 pgtable_l5_enabled ? 829 (uintptr_t)early_p4d : (uintptr_t)early_pud, 830 PGDIR_SIZE, PAGE_TABLE); 831 832 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 833 834 local_flush_tlb_all(); 835 csr_write(CSR_SATP, identity_satp); 836 hw_satp = csr_swap(CSR_SATP, 0ULL); 837 local_flush_tlb_all(); 838 839 if (hw_satp != identity_satp) { 840 if (pgtable_l5_enabled) { 841 disable_pgtable_l5(); 842 memset(early_pg_dir, 0, PAGE_SIZE); 843 goto retry; 844 } 845 disable_pgtable_l4(); 846 } 847 848 memset(early_pg_dir, 0, PAGE_SIZE); 849 memset(early_p4d, 0, PAGE_SIZE); 850 memset(early_pud, 0, PAGE_SIZE); 851 memset(early_pmd, 0, PAGE_SIZE); 852 } 853 #endif 854 855 /* 856 * setup_vm() is called from head.S with MMU-off. 857 * 858 * Following requirements should be honoured for setup_vm() to work 859 * correctly: 860 * 1) It should use PC-relative addressing for accessing kernel symbols. 861 * To achieve this we always use GCC cmodel=medany. 862 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 863 * so disable compiler instrumentation when FTRACE is enabled. 864 * 865 * Currently, the above requirements are honoured by using custom CFLAGS 866 * for init.o in mm/Makefile. 867 */ 868 869 #ifndef __riscv_cmodel_medany 870 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 871 #endif 872 873 #ifdef CONFIG_RELOCATABLE 874 extern unsigned long __rela_dyn_start, __rela_dyn_end; 875 876 static void __init relocate_kernel(void) 877 { 878 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 879 /* 880 * This holds the offset between the linked virtual address and the 881 * relocated virtual address. 882 */ 883 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 884 /* 885 * This holds the offset between kernel linked virtual address and 886 * physical address. 887 */ 888 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 889 890 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 891 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 892 Elf64_Addr relocated_addr = rela->r_addend; 893 894 if (rela->r_info != R_RISCV_RELATIVE) 895 continue; 896 897 /* 898 * Make sure to not relocate vdso symbols like rt_sigreturn 899 * which are linked from the address 0 in vmlinux since 900 * vdso symbol addresses are actually used as an offset from 901 * mm->context.vdso in VDSO_OFFSET macro. 902 */ 903 if (relocated_addr >= KERNEL_LINK_ADDR) 904 relocated_addr += reloc_offset; 905 906 *(Elf64_Addr *)addr = relocated_addr; 907 } 908 } 909 #endif /* CONFIG_RELOCATABLE */ 910 911 #ifdef CONFIG_XIP_KERNEL 912 static void __init create_kernel_page_table(pgd_t *pgdir, 913 __always_unused bool early) 914 { 915 uintptr_t va, end_va; 916 917 /* Map the flash resident part */ 918 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 919 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 920 create_pgd_mapping(pgdir, va, 921 kernel_map.xiprom + (va - kernel_map.virt_addr), 922 PMD_SIZE, PAGE_KERNEL_EXEC); 923 924 /* Map the data in RAM */ 925 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 926 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 927 create_pgd_mapping(pgdir, va, 928 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 929 PMD_SIZE, PAGE_KERNEL); 930 } 931 #else 932 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 933 { 934 uintptr_t va, end_va; 935 936 end_va = kernel_map.virt_addr + kernel_map.size; 937 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 938 create_pgd_mapping(pgdir, va, 939 kernel_map.phys_addr + (va - kernel_map.virt_addr), 940 PMD_SIZE, 941 early ? 942 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 943 } 944 #endif 945 946 /* 947 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 948 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 949 * entry. 950 */ 951 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 952 uintptr_t dtb_pa) 953 { 954 #ifndef CONFIG_BUILTIN_DTB 955 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 956 957 /* Make sure the fdt fixmap address is always aligned on PMD size */ 958 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 959 960 /* In 32-bit only, the fdt lies in its own PGD */ 961 if (!IS_ENABLED(CONFIG_64BIT)) { 962 create_pgd_mapping(early_pg_dir, fix_fdt_va, 963 pa, MAX_FDT_SIZE, PAGE_KERNEL); 964 } else { 965 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 966 pa, PMD_SIZE, PAGE_KERNEL); 967 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 968 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 969 } 970 971 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 972 #else 973 /* 974 * For 64-bit kernel, __va can't be used since it would return a linear 975 * mapping address whereas dtb_early_va will be used before 976 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 977 * kernel is mapped in the linear mapping, that makes no difference. 978 */ 979 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 980 #endif 981 982 dtb_early_pa = dtb_pa; 983 } 984 985 /* 986 * MMU is not enabled, the page tables are allocated directly using 987 * early_pmd/pud/p4d and the address returned is the physical one. 988 */ 989 static void __init pt_ops_set_early(void) 990 { 991 pt_ops.alloc_pte = alloc_pte_early; 992 pt_ops.get_pte_virt = get_pte_virt_early; 993 #ifndef __PAGETABLE_PMD_FOLDED 994 pt_ops.alloc_pmd = alloc_pmd_early; 995 pt_ops.get_pmd_virt = get_pmd_virt_early; 996 pt_ops.alloc_pud = alloc_pud_early; 997 pt_ops.get_pud_virt = get_pud_virt_early; 998 pt_ops.alloc_p4d = alloc_p4d_early; 999 pt_ops.get_p4d_virt = get_p4d_virt_early; 1000 #endif 1001 } 1002 1003 /* 1004 * MMU is enabled but page table setup is not complete yet. 1005 * fixmap page table alloc functions must be used as a means to temporarily 1006 * map the allocated physical pages since the linear mapping does not exist yet. 1007 * 1008 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1009 * but it will be used as described above. 1010 */ 1011 static void __init pt_ops_set_fixmap(void) 1012 { 1013 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1014 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1015 #ifndef __PAGETABLE_PMD_FOLDED 1016 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1017 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1018 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1019 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1020 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1021 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1022 #endif 1023 } 1024 1025 /* 1026 * MMU is enabled and page table setup is complete, so from now, we can use 1027 * generic page allocation functions to setup page table. 1028 */ 1029 static void __init pt_ops_set_late(void) 1030 { 1031 pt_ops.alloc_pte = alloc_pte_late; 1032 pt_ops.get_pte_virt = get_pte_virt_late; 1033 #ifndef __PAGETABLE_PMD_FOLDED 1034 pt_ops.alloc_pmd = alloc_pmd_late; 1035 pt_ops.get_pmd_virt = get_pmd_virt_late; 1036 pt_ops.alloc_pud = alloc_pud_late; 1037 pt_ops.get_pud_virt = get_pud_virt_late; 1038 pt_ops.alloc_p4d = alloc_p4d_late; 1039 pt_ops.get_p4d_virt = get_p4d_virt_late; 1040 #endif 1041 } 1042 1043 #ifdef CONFIG_RANDOMIZE_BASE 1044 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1045 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1046 1047 static int __init print_nokaslr(char *p) 1048 { 1049 pr_info("Disabled KASLR"); 1050 return 0; 1051 } 1052 early_param("nokaslr", print_nokaslr); 1053 1054 unsigned long kaslr_offset(void) 1055 { 1056 return kernel_map.virt_offset; 1057 } 1058 #endif 1059 1060 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1061 { 1062 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1063 1064 #ifdef CONFIG_RANDOMIZE_BASE 1065 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1066 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1067 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1068 u32 nr_pos; 1069 1070 /* 1071 * Compute the number of positions available: we are limited 1072 * by the early page table that only has one PUD and we must 1073 * be aligned on PMD_SIZE. 1074 */ 1075 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1076 1077 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1078 } 1079 #endif 1080 1081 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1082 1083 #ifdef CONFIG_XIP_KERNEL 1084 #ifdef CONFIG_64BIT 1085 kernel_map.page_offset = PAGE_OFFSET_L3; 1086 #else 1087 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1088 #endif 1089 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1090 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1091 1092 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1093 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1094 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1095 1096 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1097 #else 1098 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1099 kernel_map.phys_addr = (uintptr_t)(&_start); 1100 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1101 #endif 1102 1103 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1104 set_satp_mode(dtb_pa); 1105 set_mmap_rnd_bits_max(); 1106 #endif 1107 1108 /* 1109 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1110 * where we have the system memory layout: this allows us to align 1111 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1112 * for the linear mapping. This is only possible because the kernel 1113 * mapping lies outside the linear mapping. 1114 * In 32-bit however, as the kernel resides in the linear mapping, 1115 * setup_vm_final can not change the mapping established here, 1116 * otherwise the same kernel addresses would get mapped to different 1117 * physical addresses (if the start of dram is different from the 1118 * kernel physical address start). 1119 */ 1120 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1121 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1122 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1123 1124 /* 1125 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1126 * kernel, whereas for 64-bit kernel, the end of the virtual address 1127 * space is occupied by the modules/BPF/kernel mappings which reduces 1128 * the available size of the linear mapping. 1129 */ 1130 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1131 1132 /* Sanity check alignment and size */ 1133 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1134 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1135 1136 #ifdef CONFIG_64BIT 1137 /* 1138 * The last 4K bytes of the addressable memory can not be mapped because 1139 * of IS_ERR_VALUE macro. 1140 */ 1141 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1142 #endif 1143 1144 #ifdef CONFIG_RELOCATABLE 1145 /* 1146 * Early page table uses only one PUD, which makes it possible 1147 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1148 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1149 * since a part of the kernel would not get mapped. 1150 */ 1151 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1152 relocate_kernel(); 1153 #endif 1154 1155 apply_early_boot_alternatives(); 1156 pt_ops_set_early(); 1157 1158 /* Setup early PGD for fixmap */ 1159 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1160 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1161 1162 #ifndef __PAGETABLE_PMD_FOLDED 1163 /* Setup fixmap P4D and PUD */ 1164 if (pgtable_l5_enabled) 1165 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1166 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1167 /* Setup fixmap PUD and PMD */ 1168 if (pgtable_l4_enabled) 1169 create_pud_mapping(fixmap_pud, FIXADDR_START, 1170 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1171 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1172 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1173 /* Setup trampoline PGD and PMD */ 1174 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1175 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1176 if (pgtable_l5_enabled) 1177 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1178 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1179 if (pgtable_l4_enabled) 1180 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1181 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1182 #ifdef CONFIG_XIP_KERNEL 1183 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1184 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1185 #else 1186 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1187 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1188 #endif 1189 #else 1190 /* Setup trampoline PGD */ 1191 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1192 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1193 #endif 1194 1195 /* 1196 * Setup early PGD covering entire kernel which will allow 1197 * us to reach paging_init(). We map all memory banks later 1198 * in setup_vm_final() below. 1199 */ 1200 create_kernel_page_table(early_pg_dir, true); 1201 1202 /* Setup early mapping for FDT early scan */ 1203 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1204 1205 /* 1206 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1207 * range can not span multiple pmds. 1208 */ 1209 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1210 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1211 1212 #ifndef __PAGETABLE_PMD_FOLDED 1213 /* 1214 * Early ioremap fixmap is already created as it lies within first 2MB 1215 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1216 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1217 * the user if not. 1218 */ 1219 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1220 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1221 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1222 WARN_ON(1); 1223 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1224 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1225 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1226 fix_to_virt(FIX_BTMAP_BEGIN)); 1227 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1228 fix_to_virt(FIX_BTMAP_END)); 1229 1230 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1231 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1232 } 1233 #endif 1234 1235 pt_ops_set_fixmap(); 1236 } 1237 1238 static void __init create_linear_mapping_range(phys_addr_t start, 1239 phys_addr_t end, 1240 uintptr_t fixed_map_size) 1241 { 1242 phys_addr_t pa; 1243 uintptr_t va, map_size; 1244 1245 for (pa = start; pa < end; pa += map_size) { 1246 va = (uintptr_t)__va(pa); 1247 map_size = fixed_map_size ? fixed_map_size : 1248 best_map_size(pa, va, end - pa); 1249 1250 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1251 pgprot_from_va(va)); 1252 } 1253 } 1254 1255 static void __init create_linear_mapping_page_table(void) 1256 { 1257 phys_addr_t start, end; 1258 phys_addr_t kfence_pool __maybe_unused; 1259 u64 i; 1260 1261 #ifdef CONFIG_STRICT_KERNEL_RWX 1262 phys_addr_t ktext_start = __pa_symbol(_start); 1263 phys_addr_t ktext_size = __init_data_begin - _start; 1264 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1265 phys_addr_t krodata_size = _data - __start_rodata; 1266 1267 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1268 memblock_mark_nomap(ktext_start, ktext_size); 1269 memblock_mark_nomap(krodata_start, krodata_size); 1270 #endif 1271 1272 #ifdef CONFIG_KFENCE 1273 /* 1274 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1275 * before we setup the linear mapping so that we avoid using hugepages 1276 * for this region. 1277 */ 1278 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1279 BUG_ON(!kfence_pool); 1280 1281 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1282 __kfence_pool = __va(kfence_pool); 1283 #endif 1284 1285 /* Map all memory banks in the linear mapping */ 1286 for_each_mem_range(i, &start, &end) { 1287 if (start >= end) 1288 break; 1289 if (start <= __pa(PAGE_OFFSET) && 1290 __pa(PAGE_OFFSET) < end) 1291 start = __pa(PAGE_OFFSET); 1292 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1293 end = __pa(PAGE_OFFSET) + memory_limit; 1294 1295 create_linear_mapping_range(start, end, 0); 1296 } 1297 1298 #ifdef CONFIG_STRICT_KERNEL_RWX 1299 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0); 1300 create_linear_mapping_range(krodata_start, 1301 krodata_start + krodata_size, 0); 1302 1303 memblock_clear_nomap(ktext_start, ktext_size); 1304 memblock_clear_nomap(krodata_start, krodata_size); 1305 #endif 1306 1307 #ifdef CONFIG_KFENCE 1308 create_linear_mapping_range(kfence_pool, 1309 kfence_pool + KFENCE_POOL_SIZE, 1310 PAGE_SIZE); 1311 1312 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1313 #endif 1314 } 1315 1316 static void __init setup_vm_final(void) 1317 { 1318 /* Setup swapper PGD for fixmap */ 1319 #if !defined(CONFIG_64BIT) 1320 /* 1321 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1322 * directly in swapper_pg_dir in addition to the pgd entry that points 1323 * to fixmap_pte. 1324 */ 1325 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1326 1327 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1328 #endif 1329 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1330 __pa_symbol(fixmap_pgd_next), 1331 PGDIR_SIZE, PAGE_TABLE); 1332 1333 /* Map the linear mapping */ 1334 create_linear_mapping_page_table(); 1335 1336 /* Map the kernel */ 1337 if (IS_ENABLED(CONFIG_64BIT)) 1338 create_kernel_page_table(swapper_pg_dir, false); 1339 1340 #ifdef CONFIG_KASAN 1341 kasan_swapper_init(); 1342 #endif 1343 1344 /* Clear fixmap PTE and PMD mappings */ 1345 clear_fixmap(FIX_PTE); 1346 clear_fixmap(FIX_PMD); 1347 clear_fixmap(FIX_PUD); 1348 clear_fixmap(FIX_P4D); 1349 1350 /* Move to swapper page table */ 1351 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1352 local_flush_tlb_all(); 1353 1354 pt_ops_set_late(); 1355 } 1356 #else 1357 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1358 { 1359 dtb_early_va = (void *)dtb_pa; 1360 dtb_early_pa = dtb_pa; 1361 } 1362 1363 static inline void setup_vm_final(void) 1364 { 1365 } 1366 #endif /* CONFIG_MMU */ 1367 1368 /* 1369 * reserve_crashkernel() - reserves memory for crash kernel 1370 * 1371 * This function reserves memory area given in "crashkernel=" kernel command 1372 * line parameter. The memory reserved is used by dump capture kernel when 1373 * primary kernel is crashing. 1374 */ 1375 static void __init arch_reserve_crashkernel(void) 1376 { 1377 unsigned long long low_size = 0; 1378 unsigned long long crash_base, crash_size; 1379 char *cmdline = boot_command_line; 1380 bool high = false; 1381 int ret; 1382 1383 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1384 return; 1385 1386 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 1387 &crash_size, &crash_base, 1388 &low_size, &high); 1389 if (ret) 1390 return; 1391 1392 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 1393 low_size, high); 1394 } 1395 1396 void __init paging_init(void) 1397 { 1398 setup_bootmem(); 1399 setup_vm_final(); 1400 1401 /* Depend on that Linear Mapping is ready */ 1402 memblock_allow_resize(); 1403 } 1404 1405 void __init misc_mem_init(void) 1406 { 1407 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1408 arch_numa_init(); 1409 sparse_init(); 1410 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1411 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1412 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1413 #endif 1414 zone_sizes_init(); 1415 arch_reserve_crashkernel(); 1416 memblock_dump_all(); 1417 } 1418 1419 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1420 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1421 unsigned long addr, unsigned long next) 1422 { 1423 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1424 } 1425 1426 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1427 unsigned long addr, unsigned long next) 1428 { 1429 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1430 return 1; 1431 } 1432 1433 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1434 struct vmem_altmap *altmap) 1435 { 1436 /* 1437 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1438 * can't use hugepage mappings for 2-level page table because in case of 1439 * memory hotplug, we are not able to update all the page tables with 1440 * the new PMDs. 1441 */ 1442 return vmemmap_populate_hugepages(start, end, node, NULL); 1443 } 1444 #endif 1445 1446 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1447 /* 1448 * Pre-allocates page-table pages for a specific area in the kernel 1449 * page-table. Only the level which needs to be synchronized between 1450 * all page-tables is allocated because the synchronization can be 1451 * expensive. 1452 */ 1453 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1454 const char *area) 1455 { 1456 unsigned long addr; 1457 const char *lvl; 1458 1459 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1460 pgd_t *pgd = pgd_offset_k(addr); 1461 p4d_t *p4d; 1462 pud_t *pud; 1463 pmd_t *pmd; 1464 1465 lvl = "p4d"; 1466 p4d = p4d_alloc(&init_mm, pgd, addr); 1467 if (!p4d) 1468 goto failed; 1469 1470 if (pgtable_l5_enabled) 1471 continue; 1472 1473 lvl = "pud"; 1474 pud = pud_alloc(&init_mm, p4d, addr); 1475 if (!pud) 1476 goto failed; 1477 1478 if (pgtable_l4_enabled) 1479 continue; 1480 1481 lvl = "pmd"; 1482 pmd = pmd_alloc(&init_mm, pud, addr); 1483 if (!pmd) 1484 goto failed; 1485 } 1486 return; 1487 1488 failed: 1489 /* 1490 * The pages have to be there now or they will be missing in 1491 * process page-tables later. 1492 */ 1493 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1494 } 1495 1496 void __init pgtable_cache_init(void) 1497 { 1498 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1499 if (IS_ENABLED(CONFIG_MODULES)) 1500 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1501 } 1502 #endif 1503 1504 #ifdef CONFIG_EXECMEM 1505 #ifdef CONFIG_MMU 1506 static struct execmem_info execmem_info __ro_after_init; 1507 1508 struct execmem_info __init *execmem_arch_setup(void) 1509 { 1510 execmem_info = (struct execmem_info){ 1511 .ranges = { 1512 [EXECMEM_DEFAULT] = { 1513 .start = MODULES_VADDR, 1514 .end = MODULES_END, 1515 .pgprot = PAGE_KERNEL, 1516 .alignment = 1, 1517 }, 1518 [EXECMEM_KPROBES] = { 1519 .start = VMALLOC_START, 1520 .end = VMALLOC_END, 1521 .pgprot = PAGE_KERNEL_READ_EXEC, 1522 .alignment = 1, 1523 }, 1524 [EXECMEM_BPF] = { 1525 .start = BPF_JIT_REGION_START, 1526 .end = BPF_JIT_REGION_END, 1527 .pgprot = PAGE_KERNEL, 1528 .alignment = PAGE_SIZE, 1529 }, 1530 }, 1531 }; 1532 1533 return &execmem_info; 1534 } 1535 #endif /* CONFIG_MMU */ 1536 #endif /* CONFIG_EXECMEM */ 1537