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 = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 54 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !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 #ifdef CONFIG_FLATMEM 166 BUG_ON(!mem_map); 167 #endif /* CONFIG_FLATMEM */ 168 169 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE); 170 memblock_free_all(); 171 172 print_vm_layout(); 173 } 174 175 /* Limit the memory size via mem. */ 176 static phys_addr_t memory_limit; 177 #ifdef CONFIG_XIP_KERNEL 178 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 179 #endif /* CONFIG_XIP_KERNEL */ 180 181 static int __init early_mem(char *p) 182 { 183 u64 size; 184 185 if (!p) 186 return 1; 187 188 size = memparse(p, &p) & PAGE_MASK; 189 memory_limit = min_t(u64, size, memory_limit); 190 191 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 192 193 return 0; 194 } 195 early_param("mem", early_mem); 196 197 static void __init setup_bootmem(void) 198 { 199 phys_addr_t vmlinux_end = __pa_symbol(&_end); 200 phys_addr_t max_mapped_addr; 201 phys_addr_t phys_ram_end, vmlinux_start; 202 203 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 204 vmlinux_start = __pa_symbol(&_sdata); 205 else 206 vmlinux_start = __pa_symbol(&_start); 207 208 memblock_enforce_memory_limit(memory_limit); 209 210 /* 211 * Make sure we align the reservation on PMD_SIZE since we will 212 * map the kernel in the linear mapping as read-only: we do not want 213 * any allocation to happen between _end and the next pmd aligned page. 214 */ 215 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 216 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 217 /* 218 * Reserve from the start of the kernel to the end of the kernel 219 */ 220 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 221 222 phys_ram_end = memblock_end_of_DRAM(); 223 224 /* 225 * Make sure we align the start of the memory on a PMD boundary so that 226 * at worst, we map the linear mapping with PMD mappings. 227 */ 228 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 229 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 230 231 /* 232 * In 64-bit, any use of __va/__pa before this point is wrong as we 233 * did not know the start of DRAM before. 234 */ 235 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 236 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 237 238 /* 239 * memblock allocator is not aware of the fact that last 4K bytes of 240 * the addressable memory can not be mapped because of IS_ERR_VALUE 241 * macro. Make sure that last 4k bytes are not usable by memblock 242 * if end of dram is equal to maximum addressable memory. For 64-bit 243 * kernel, this problem can't happen here as the end of the virtual 244 * address space is occupied by the kernel mapping then this check must 245 * be done as soon as the kernel mapping base address is determined. 246 */ 247 if (!IS_ENABLED(CONFIG_64BIT)) { 248 max_mapped_addr = __pa(~(ulong)0); 249 if (max_mapped_addr == (phys_ram_end - 1)) 250 memblock_set_current_limit(max_mapped_addr - 4096); 251 } 252 253 min_low_pfn = PFN_UP(phys_ram_base); 254 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 255 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 256 257 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 258 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 259 260 reserve_initrd_mem(); 261 262 /* 263 * No allocation should be done before reserving the memory as defined 264 * in the device tree, otherwise the allocation could end up in a 265 * reserved region. 266 */ 267 early_init_fdt_scan_reserved_mem(); 268 269 /* 270 * If DTB is built in, no need to reserve its memblock. 271 * Otherwise, do reserve it but avoid using 272 * early_init_fdt_reserve_self() since __pa() does 273 * not work for DTB pointers that are fixmap addresses 274 */ 275 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 276 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 277 278 dma_contiguous_reserve(dma32_phys_limit); 279 if (IS_ENABLED(CONFIG_64BIT)) 280 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 281 } 282 283 #ifdef CONFIG_MMU 284 struct pt_alloc_ops pt_ops __initdata; 285 286 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 287 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 288 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 289 290 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 291 292 #ifdef CONFIG_XIP_KERNEL 293 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 294 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 295 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 296 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 297 #endif /* CONFIG_XIP_KERNEL */ 298 299 static const pgprot_t protection_map[16] = { 300 [VM_NONE] = PAGE_NONE, 301 [VM_READ] = PAGE_READ, 302 [VM_WRITE] = PAGE_COPY, 303 [VM_WRITE | VM_READ] = PAGE_COPY, 304 [VM_EXEC] = PAGE_EXEC, 305 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 306 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 307 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 308 [VM_SHARED] = PAGE_NONE, 309 [VM_SHARED | VM_READ] = PAGE_READ, 310 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 311 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 312 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 313 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 314 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 315 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 316 }; 317 DECLARE_VM_GET_PAGE_PROT 318 319 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 320 { 321 unsigned long addr = __fix_to_virt(idx); 322 pte_t *ptep; 323 324 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 325 326 ptep = &fixmap_pte[pte_index(addr)]; 327 328 if (pgprot_val(prot)) 329 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 330 else 331 pte_clear(&init_mm, addr, ptep); 332 local_flush_tlb_page(addr); 333 } 334 335 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 336 { 337 return (pte_t *)((uintptr_t)pa); 338 } 339 340 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 341 { 342 clear_fixmap(FIX_PTE); 343 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 344 } 345 346 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 347 { 348 return (pte_t *) __va(pa); 349 } 350 351 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 352 { 353 /* 354 * We only create PMD or PGD early mappings so we 355 * should never reach here with MMU disabled. 356 */ 357 BUG(); 358 } 359 360 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 361 { 362 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 363 } 364 365 static phys_addr_t __init alloc_pte_late(uintptr_t va) 366 { 367 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 368 369 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 370 return __pa((pte_t *)ptdesc_address(ptdesc)); 371 } 372 373 static void __init create_pte_mapping(pte_t *ptep, 374 uintptr_t va, phys_addr_t pa, 375 phys_addr_t sz, pgprot_t prot) 376 { 377 uintptr_t pte_idx = pte_index(va); 378 379 BUG_ON(sz != PAGE_SIZE); 380 381 if (pte_none(ptep[pte_idx])) 382 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 383 } 384 385 #ifndef __PAGETABLE_PMD_FOLDED 386 387 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 388 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 389 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 390 391 #ifdef CONFIG_XIP_KERNEL 392 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 393 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 394 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 395 #endif /* CONFIG_XIP_KERNEL */ 396 397 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 398 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 399 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 400 401 #ifdef CONFIG_XIP_KERNEL 402 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 403 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 404 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 405 #endif /* CONFIG_XIP_KERNEL */ 406 407 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 408 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 409 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 410 411 #ifdef CONFIG_XIP_KERNEL 412 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 413 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 414 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 415 #endif /* CONFIG_XIP_KERNEL */ 416 417 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 418 { 419 /* Before MMU is enabled */ 420 return (pmd_t *)((uintptr_t)pa); 421 } 422 423 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 424 { 425 clear_fixmap(FIX_PMD); 426 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 427 } 428 429 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 430 { 431 return (pmd_t *) __va(pa); 432 } 433 434 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 435 { 436 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 437 438 return (uintptr_t)early_pmd; 439 } 440 441 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 442 { 443 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 444 } 445 446 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 447 { 448 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 449 450 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 451 return __pa((pmd_t *)ptdesc_address(ptdesc)); 452 } 453 454 static void __init create_pmd_mapping(pmd_t *pmdp, 455 uintptr_t va, phys_addr_t pa, 456 phys_addr_t sz, pgprot_t prot) 457 { 458 pte_t *ptep; 459 phys_addr_t pte_phys; 460 uintptr_t pmd_idx = pmd_index(va); 461 462 if (sz == PMD_SIZE) { 463 if (pmd_none(pmdp[pmd_idx])) 464 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 465 return; 466 } 467 468 if (pmd_none(pmdp[pmd_idx])) { 469 pte_phys = pt_ops.alloc_pte(va); 470 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 471 ptep = pt_ops.get_pte_virt(pte_phys); 472 memset(ptep, 0, PAGE_SIZE); 473 } else { 474 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 475 ptep = pt_ops.get_pte_virt(pte_phys); 476 } 477 478 create_pte_mapping(ptep, va, pa, sz, prot); 479 } 480 481 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 482 { 483 return (pud_t *)((uintptr_t)pa); 484 } 485 486 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 487 { 488 clear_fixmap(FIX_PUD); 489 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 490 } 491 492 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 493 { 494 return (pud_t *)__va(pa); 495 } 496 497 static phys_addr_t __init alloc_pud_early(uintptr_t va) 498 { 499 /* Only one PUD is available for early mapping */ 500 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 501 502 return (uintptr_t)early_pud; 503 } 504 505 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 506 { 507 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 508 } 509 510 static phys_addr_t alloc_pud_late(uintptr_t va) 511 { 512 unsigned long vaddr; 513 514 vaddr = __get_free_page(GFP_KERNEL); 515 BUG_ON(!vaddr); 516 return __pa(vaddr); 517 } 518 519 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 520 { 521 return (p4d_t *)((uintptr_t)pa); 522 } 523 524 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 525 { 526 clear_fixmap(FIX_P4D); 527 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 528 } 529 530 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 531 { 532 return (p4d_t *)__va(pa); 533 } 534 535 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 536 { 537 /* Only one P4D is available for early mapping */ 538 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 539 540 return (uintptr_t)early_p4d; 541 } 542 543 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 544 { 545 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 546 } 547 548 static phys_addr_t alloc_p4d_late(uintptr_t va) 549 { 550 unsigned long vaddr; 551 552 vaddr = __get_free_page(GFP_KERNEL); 553 BUG_ON(!vaddr); 554 return __pa(vaddr); 555 } 556 557 static void __init create_pud_mapping(pud_t *pudp, 558 uintptr_t va, phys_addr_t pa, 559 phys_addr_t sz, pgprot_t prot) 560 { 561 pmd_t *nextp; 562 phys_addr_t next_phys; 563 uintptr_t pud_index = pud_index(va); 564 565 if (sz == PUD_SIZE) { 566 if (pud_val(pudp[pud_index]) == 0) 567 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 568 return; 569 } 570 571 if (pud_val(pudp[pud_index]) == 0) { 572 next_phys = pt_ops.alloc_pmd(va); 573 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 574 nextp = pt_ops.get_pmd_virt(next_phys); 575 memset(nextp, 0, PAGE_SIZE); 576 } else { 577 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 578 nextp = pt_ops.get_pmd_virt(next_phys); 579 } 580 581 create_pmd_mapping(nextp, va, pa, sz, prot); 582 } 583 584 static void __init create_p4d_mapping(p4d_t *p4dp, 585 uintptr_t va, phys_addr_t pa, 586 phys_addr_t sz, pgprot_t prot) 587 { 588 pud_t *nextp; 589 phys_addr_t next_phys; 590 uintptr_t p4d_index = p4d_index(va); 591 592 if (sz == P4D_SIZE) { 593 if (p4d_val(p4dp[p4d_index]) == 0) 594 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 595 return; 596 } 597 598 if (p4d_val(p4dp[p4d_index]) == 0) { 599 next_phys = pt_ops.alloc_pud(va); 600 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 601 nextp = pt_ops.get_pud_virt(next_phys); 602 memset(nextp, 0, PAGE_SIZE); 603 } else { 604 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 605 nextp = pt_ops.get_pud_virt(next_phys); 606 } 607 608 create_pud_mapping(nextp, va, pa, sz, prot); 609 } 610 611 #define pgd_next_t p4d_t 612 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 613 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 614 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 615 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 616 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 617 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 618 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 619 (pgtable_l5_enabled ? \ 620 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 621 (pgtable_l4_enabled ? \ 622 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 623 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 624 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 625 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 626 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 627 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 628 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 629 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 630 #else 631 #define pgd_next_t pte_t 632 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 633 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 634 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 635 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 636 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 637 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 638 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 639 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 640 #endif /* __PAGETABLE_PMD_FOLDED */ 641 642 void __init create_pgd_mapping(pgd_t *pgdp, 643 uintptr_t va, phys_addr_t pa, 644 phys_addr_t sz, pgprot_t prot) 645 { 646 pgd_next_t *nextp; 647 phys_addr_t next_phys; 648 uintptr_t pgd_idx = pgd_index(va); 649 650 if (sz == PGDIR_SIZE) { 651 if (pgd_val(pgdp[pgd_idx]) == 0) 652 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 653 return; 654 } 655 656 if (pgd_val(pgdp[pgd_idx]) == 0) { 657 next_phys = alloc_pgd_next(va); 658 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 659 nextp = get_pgd_next_virt(next_phys); 660 memset(nextp, 0, PAGE_SIZE); 661 } else { 662 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 663 nextp = get_pgd_next_virt(next_phys); 664 } 665 666 create_pgd_next_mapping(nextp, va, pa, sz, prot); 667 } 668 669 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va, 670 phys_addr_t size) 671 { 672 if (pgtable_l5_enabled && 673 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 674 return P4D_SIZE; 675 676 if (pgtable_l4_enabled && 677 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 678 return PUD_SIZE; 679 680 if (IS_ENABLED(CONFIG_64BIT) && 681 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 682 return PMD_SIZE; 683 684 return PAGE_SIZE; 685 } 686 687 #ifdef CONFIG_XIP_KERNEL 688 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 689 extern char _xiprom[], _exiprom[], __data_loc; 690 691 /* called from head.S with MMU off */ 692 asmlinkage void __init __copy_data(void) 693 { 694 void *from = (void *)(&__data_loc); 695 void *to = (void *)CONFIG_PHYS_RAM_BASE; 696 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 697 698 memcpy(to, from, sz); 699 } 700 #endif 701 702 #ifdef CONFIG_STRICT_KERNEL_RWX 703 static __init pgprot_t pgprot_from_va(uintptr_t va) 704 { 705 if (is_va_kernel_text(va)) 706 return PAGE_KERNEL_READ_EXEC; 707 708 /* 709 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 710 * we must protect its linear mapping alias from being executed and 711 * written. 712 * And rodata section is marked readonly in mark_rodata_ro. 713 */ 714 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 715 return PAGE_KERNEL_READ; 716 717 return PAGE_KERNEL; 718 } 719 720 void mark_rodata_ro(void) 721 { 722 set_kernel_memory(__start_rodata, _data, set_memory_ro); 723 if (IS_ENABLED(CONFIG_64BIT)) 724 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 725 set_memory_ro); 726 } 727 #else 728 static __init pgprot_t pgprot_from_va(uintptr_t va) 729 { 730 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 731 return PAGE_KERNEL; 732 733 return PAGE_KERNEL_EXEC; 734 } 735 #endif /* CONFIG_STRICT_KERNEL_RWX */ 736 737 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 738 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 739 740 static void __init disable_pgtable_l5(void) 741 { 742 pgtable_l5_enabled = false; 743 kernel_map.page_offset = PAGE_OFFSET_L4; 744 satp_mode = SATP_MODE_48; 745 } 746 747 static void __init disable_pgtable_l4(void) 748 { 749 pgtable_l4_enabled = false; 750 kernel_map.page_offset = PAGE_OFFSET_L3; 751 satp_mode = SATP_MODE_39; 752 } 753 754 static int __init print_no4lvl(char *p) 755 { 756 pr_info("Disabled 4-level and 5-level paging"); 757 return 0; 758 } 759 early_param("no4lvl", print_no4lvl); 760 761 static int __init print_no5lvl(char *p) 762 { 763 pr_info("Disabled 5-level paging"); 764 return 0; 765 } 766 early_param("no5lvl", print_no5lvl); 767 768 static void __init set_mmap_rnd_bits_max(void) 769 { 770 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 771 } 772 773 /* 774 * There is a simple way to determine if 4-level is supported by the 775 * underlying hardware: establish 1:1 mapping in 4-level page table mode 776 * then read SATP to see if the configuration was taken into account 777 * meaning sv48 is supported. 778 */ 779 static __init void set_satp_mode(uintptr_t dtb_pa) 780 { 781 u64 identity_satp, hw_satp; 782 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 783 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 784 785 if (satp_mode_cmdline == SATP_MODE_57) { 786 disable_pgtable_l5(); 787 } else if (satp_mode_cmdline == SATP_MODE_48) { 788 disable_pgtable_l5(); 789 disable_pgtable_l4(); 790 return; 791 } 792 793 create_p4d_mapping(early_p4d, 794 set_satp_mode_pmd, (uintptr_t)early_pud, 795 P4D_SIZE, PAGE_TABLE); 796 create_pud_mapping(early_pud, 797 set_satp_mode_pmd, (uintptr_t)early_pmd, 798 PUD_SIZE, PAGE_TABLE); 799 /* Handle the case where set_satp_mode straddles 2 PMDs */ 800 create_pmd_mapping(early_pmd, 801 set_satp_mode_pmd, set_satp_mode_pmd, 802 PMD_SIZE, PAGE_KERNEL_EXEC); 803 create_pmd_mapping(early_pmd, 804 set_satp_mode_pmd + PMD_SIZE, 805 set_satp_mode_pmd + PMD_SIZE, 806 PMD_SIZE, PAGE_KERNEL_EXEC); 807 retry: 808 create_pgd_mapping(early_pg_dir, 809 set_satp_mode_pmd, 810 pgtable_l5_enabled ? 811 (uintptr_t)early_p4d : (uintptr_t)early_pud, 812 PGDIR_SIZE, PAGE_TABLE); 813 814 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 815 816 local_flush_tlb_all(); 817 csr_write(CSR_SATP, identity_satp); 818 hw_satp = csr_swap(CSR_SATP, 0ULL); 819 local_flush_tlb_all(); 820 821 if (hw_satp != identity_satp) { 822 if (pgtable_l5_enabled) { 823 disable_pgtable_l5(); 824 memset(early_pg_dir, 0, PAGE_SIZE); 825 goto retry; 826 } 827 disable_pgtable_l4(); 828 } 829 830 memset(early_pg_dir, 0, PAGE_SIZE); 831 memset(early_p4d, 0, PAGE_SIZE); 832 memset(early_pud, 0, PAGE_SIZE); 833 memset(early_pmd, 0, PAGE_SIZE); 834 } 835 #endif 836 837 /* 838 * setup_vm() is called from head.S with MMU-off. 839 * 840 * Following requirements should be honoured for setup_vm() to work 841 * correctly: 842 * 1) It should use PC-relative addressing for accessing kernel symbols. 843 * To achieve this we always use GCC cmodel=medany. 844 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 845 * so disable compiler instrumentation when FTRACE is enabled. 846 * 847 * Currently, the above requirements are honoured by using custom CFLAGS 848 * for init.o in mm/Makefile. 849 */ 850 851 #ifndef __riscv_cmodel_medany 852 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 853 #endif 854 855 #ifdef CONFIG_RELOCATABLE 856 extern unsigned long __rela_dyn_start, __rela_dyn_end; 857 858 static void __init relocate_kernel(void) 859 { 860 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 861 /* 862 * This holds the offset between the linked virtual address and the 863 * relocated virtual address. 864 */ 865 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 866 /* 867 * This holds the offset between kernel linked virtual address and 868 * physical address. 869 */ 870 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 871 872 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 873 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 874 Elf64_Addr relocated_addr = rela->r_addend; 875 876 if (rela->r_info != R_RISCV_RELATIVE) 877 continue; 878 879 /* 880 * Make sure to not relocate vdso symbols like rt_sigreturn 881 * which are linked from the address 0 in vmlinux since 882 * vdso symbol addresses are actually used as an offset from 883 * mm->context.vdso in VDSO_OFFSET macro. 884 */ 885 if (relocated_addr >= KERNEL_LINK_ADDR) 886 relocated_addr += reloc_offset; 887 888 *(Elf64_Addr *)addr = relocated_addr; 889 } 890 } 891 #endif /* CONFIG_RELOCATABLE */ 892 893 #ifdef CONFIG_XIP_KERNEL 894 static void __init create_kernel_page_table(pgd_t *pgdir, 895 __always_unused bool early) 896 { 897 uintptr_t va, end_va; 898 899 /* Map the flash resident part */ 900 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 901 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 902 create_pgd_mapping(pgdir, va, 903 kernel_map.xiprom + (va - kernel_map.virt_addr), 904 PMD_SIZE, PAGE_KERNEL_EXEC); 905 906 /* Map the data in RAM */ 907 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 908 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 909 create_pgd_mapping(pgdir, va, 910 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 911 PMD_SIZE, PAGE_KERNEL); 912 } 913 #else 914 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 915 { 916 uintptr_t va, end_va; 917 918 end_va = kernel_map.virt_addr + kernel_map.size; 919 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 920 create_pgd_mapping(pgdir, va, 921 kernel_map.phys_addr + (va - kernel_map.virt_addr), 922 PMD_SIZE, 923 early ? 924 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 925 } 926 #endif 927 928 /* 929 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 930 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 931 * entry. 932 */ 933 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 934 uintptr_t dtb_pa) 935 { 936 #ifndef CONFIG_BUILTIN_DTB 937 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 938 939 /* Make sure the fdt fixmap address is always aligned on PMD size */ 940 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 941 942 /* In 32-bit only, the fdt lies in its own PGD */ 943 if (!IS_ENABLED(CONFIG_64BIT)) { 944 create_pgd_mapping(early_pg_dir, fix_fdt_va, 945 pa, MAX_FDT_SIZE, PAGE_KERNEL); 946 } else { 947 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 948 pa, PMD_SIZE, PAGE_KERNEL); 949 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 950 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 951 } 952 953 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 954 #else 955 /* 956 * For 64-bit kernel, __va can't be used since it would return a linear 957 * mapping address whereas dtb_early_va will be used before 958 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 959 * kernel is mapped in the linear mapping, that makes no difference. 960 */ 961 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 962 #endif 963 964 dtb_early_pa = dtb_pa; 965 } 966 967 /* 968 * MMU is not enabled, the page tables are allocated directly using 969 * early_pmd/pud/p4d and the address returned is the physical one. 970 */ 971 static void __init pt_ops_set_early(void) 972 { 973 pt_ops.alloc_pte = alloc_pte_early; 974 pt_ops.get_pte_virt = get_pte_virt_early; 975 #ifndef __PAGETABLE_PMD_FOLDED 976 pt_ops.alloc_pmd = alloc_pmd_early; 977 pt_ops.get_pmd_virt = get_pmd_virt_early; 978 pt_ops.alloc_pud = alloc_pud_early; 979 pt_ops.get_pud_virt = get_pud_virt_early; 980 pt_ops.alloc_p4d = alloc_p4d_early; 981 pt_ops.get_p4d_virt = get_p4d_virt_early; 982 #endif 983 } 984 985 /* 986 * MMU is enabled but page table setup is not complete yet. 987 * fixmap page table alloc functions must be used as a means to temporarily 988 * map the allocated physical pages since the linear mapping does not exist yet. 989 * 990 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 991 * but it will be used as described above. 992 */ 993 static void __init pt_ops_set_fixmap(void) 994 { 995 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 996 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 997 #ifndef __PAGETABLE_PMD_FOLDED 998 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 999 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1000 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1001 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1002 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1003 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1004 #endif 1005 } 1006 1007 /* 1008 * MMU is enabled and page table setup is complete, so from now, we can use 1009 * generic page allocation functions to setup page table. 1010 */ 1011 static void __init pt_ops_set_late(void) 1012 { 1013 pt_ops.alloc_pte = alloc_pte_late; 1014 pt_ops.get_pte_virt = get_pte_virt_late; 1015 #ifndef __PAGETABLE_PMD_FOLDED 1016 pt_ops.alloc_pmd = alloc_pmd_late; 1017 pt_ops.get_pmd_virt = get_pmd_virt_late; 1018 pt_ops.alloc_pud = alloc_pud_late; 1019 pt_ops.get_pud_virt = get_pud_virt_late; 1020 pt_ops.alloc_p4d = alloc_p4d_late; 1021 pt_ops.get_p4d_virt = get_p4d_virt_late; 1022 #endif 1023 } 1024 1025 #ifdef CONFIG_RANDOMIZE_BASE 1026 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1027 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1028 1029 static int __init print_nokaslr(char *p) 1030 { 1031 pr_info("Disabled KASLR"); 1032 return 0; 1033 } 1034 early_param("nokaslr", print_nokaslr); 1035 1036 unsigned long kaslr_offset(void) 1037 { 1038 return kernel_map.virt_offset; 1039 } 1040 #endif 1041 1042 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1043 { 1044 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1045 1046 #ifdef CONFIG_RANDOMIZE_BASE 1047 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1048 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1049 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1050 u32 nr_pos; 1051 1052 /* 1053 * Compute the number of positions available: we are limited 1054 * by the early page table that only has one PUD and we must 1055 * be aligned on PMD_SIZE. 1056 */ 1057 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1058 1059 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1060 } 1061 #endif 1062 1063 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1064 1065 #ifdef CONFIG_XIP_KERNEL 1066 #ifdef CONFIG_64BIT 1067 kernel_map.page_offset = PAGE_OFFSET_L3; 1068 #else 1069 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1070 #endif 1071 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1072 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1073 1074 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1075 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1076 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1077 1078 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1079 #else 1080 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1081 kernel_map.phys_addr = (uintptr_t)(&_start); 1082 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1083 #endif 1084 1085 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1086 set_satp_mode(dtb_pa); 1087 set_mmap_rnd_bits_max(); 1088 #endif 1089 1090 /* 1091 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1092 * where we have the system memory layout: this allows us to align 1093 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1094 * for the linear mapping. This is only possible because the kernel 1095 * mapping lies outside the linear mapping. 1096 * In 32-bit however, as the kernel resides in the linear mapping, 1097 * setup_vm_final can not change the mapping established here, 1098 * otherwise the same kernel addresses would get mapped to different 1099 * physical addresses (if the start of dram is different from the 1100 * kernel physical address start). 1101 */ 1102 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1103 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1104 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1105 1106 /* 1107 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1108 * kernel, whereas for 64-bit kernel, the end of the virtual address 1109 * space is occupied by the modules/BPF/kernel mappings which reduces 1110 * the available size of the linear mapping. 1111 */ 1112 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1113 1114 /* Sanity check alignment and size */ 1115 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1116 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1117 1118 #ifdef CONFIG_64BIT 1119 /* 1120 * The last 4K bytes of the addressable memory can not be mapped because 1121 * of IS_ERR_VALUE macro. 1122 */ 1123 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1124 #endif 1125 1126 #ifdef CONFIG_RELOCATABLE 1127 /* 1128 * Early page table uses only one PUD, which makes it possible 1129 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1130 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1131 * since a part of the kernel would not get mapped. 1132 */ 1133 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1134 relocate_kernel(); 1135 #endif 1136 1137 apply_early_boot_alternatives(); 1138 pt_ops_set_early(); 1139 1140 /* Setup early PGD for fixmap */ 1141 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1142 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1143 1144 #ifndef __PAGETABLE_PMD_FOLDED 1145 /* Setup fixmap P4D and PUD */ 1146 if (pgtable_l5_enabled) 1147 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1148 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1149 /* Setup fixmap PUD and PMD */ 1150 if (pgtable_l4_enabled) 1151 create_pud_mapping(fixmap_pud, FIXADDR_START, 1152 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1153 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1154 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1155 /* Setup trampoline PGD and PMD */ 1156 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1157 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1158 if (pgtable_l5_enabled) 1159 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1160 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1161 if (pgtable_l4_enabled) 1162 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1163 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1164 #ifdef CONFIG_XIP_KERNEL 1165 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1166 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1167 #else 1168 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1169 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1170 #endif 1171 #else 1172 /* Setup trampoline PGD */ 1173 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1174 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1175 #endif 1176 1177 /* 1178 * Setup early PGD covering entire kernel which will allow 1179 * us to reach paging_init(). We map all memory banks later 1180 * in setup_vm_final() below. 1181 */ 1182 create_kernel_page_table(early_pg_dir, true); 1183 1184 /* Setup early mapping for FDT early scan */ 1185 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1186 1187 /* 1188 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1189 * range can not span multiple pmds. 1190 */ 1191 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1192 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1193 1194 #ifndef __PAGETABLE_PMD_FOLDED 1195 /* 1196 * Early ioremap fixmap is already created as it lies within first 2MB 1197 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1198 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1199 * the user if not. 1200 */ 1201 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1202 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1203 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1204 WARN_ON(1); 1205 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1206 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1207 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1208 fix_to_virt(FIX_BTMAP_BEGIN)); 1209 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1210 fix_to_virt(FIX_BTMAP_END)); 1211 1212 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1213 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1214 } 1215 #endif 1216 1217 pt_ops_set_fixmap(); 1218 } 1219 1220 static void __init create_linear_mapping_range(phys_addr_t start, 1221 phys_addr_t end, 1222 uintptr_t fixed_map_size) 1223 { 1224 phys_addr_t pa; 1225 uintptr_t va, map_size; 1226 1227 for (pa = start; pa < end; pa += map_size) { 1228 va = (uintptr_t)__va(pa); 1229 map_size = fixed_map_size ? fixed_map_size : 1230 best_map_size(pa, va, end - pa); 1231 1232 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1233 pgprot_from_va(va)); 1234 } 1235 } 1236 1237 static void __init create_linear_mapping_page_table(void) 1238 { 1239 phys_addr_t start, end; 1240 phys_addr_t kfence_pool __maybe_unused; 1241 u64 i; 1242 1243 #ifdef CONFIG_STRICT_KERNEL_RWX 1244 phys_addr_t ktext_start = __pa_symbol(_start); 1245 phys_addr_t ktext_size = __init_data_begin - _start; 1246 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1247 phys_addr_t krodata_size = _data - __start_rodata; 1248 1249 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1250 memblock_mark_nomap(ktext_start, ktext_size); 1251 memblock_mark_nomap(krodata_start, krodata_size); 1252 #endif 1253 1254 #ifdef CONFIG_KFENCE 1255 /* 1256 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1257 * before we setup the linear mapping so that we avoid using hugepages 1258 * for this region. 1259 */ 1260 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1261 BUG_ON(!kfence_pool); 1262 1263 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1264 __kfence_pool = __va(kfence_pool); 1265 #endif 1266 1267 /* Map all memory banks in the linear mapping */ 1268 for_each_mem_range(i, &start, &end) { 1269 if (start >= end) 1270 break; 1271 if (start <= __pa(PAGE_OFFSET) && 1272 __pa(PAGE_OFFSET) < end) 1273 start = __pa(PAGE_OFFSET); 1274 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1275 end = __pa(PAGE_OFFSET) + memory_limit; 1276 1277 create_linear_mapping_range(start, end, 0); 1278 } 1279 1280 #ifdef CONFIG_STRICT_KERNEL_RWX 1281 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0); 1282 create_linear_mapping_range(krodata_start, 1283 krodata_start + krodata_size, 0); 1284 1285 memblock_clear_nomap(ktext_start, ktext_size); 1286 memblock_clear_nomap(krodata_start, krodata_size); 1287 #endif 1288 1289 #ifdef CONFIG_KFENCE 1290 create_linear_mapping_range(kfence_pool, 1291 kfence_pool + KFENCE_POOL_SIZE, 1292 PAGE_SIZE); 1293 1294 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1295 #endif 1296 } 1297 1298 static void __init setup_vm_final(void) 1299 { 1300 /* Setup swapper PGD for fixmap */ 1301 #if !defined(CONFIG_64BIT) 1302 /* 1303 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1304 * directly in swapper_pg_dir in addition to the pgd entry that points 1305 * to fixmap_pte. 1306 */ 1307 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1308 1309 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1310 #endif 1311 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1312 __pa_symbol(fixmap_pgd_next), 1313 PGDIR_SIZE, PAGE_TABLE); 1314 1315 /* Map the linear mapping */ 1316 create_linear_mapping_page_table(); 1317 1318 /* Map the kernel */ 1319 if (IS_ENABLED(CONFIG_64BIT)) 1320 create_kernel_page_table(swapper_pg_dir, false); 1321 1322 #ifdef CONFIG_KASAN 1323 kasan_swapper_init(); 1324 #endif 1325 1326 /* Clear fixmap PTE and PMD mappings */ 1327 clear_fixmap(FIX_PTE); 1328 clear_fixmap(FIX_PMD); 1329 clear_fixmap(FIX_PUD); 1330 clear_fixmap(FIX_P4D); 1331 1332 /* Move to swapper page table */ 1333 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1334 local_flush_tlb_all(); 1335 1336 pt_ops_set_late(); 1337 } 1338 #else 1339 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1340 { 1341 dtb_early_va = (void *)dtb_pa; 1342 dtb_early_pa = dtb_pa; 1343 } 1344 1345 static inline void setup_vm_final(void) 1346 { 1347 } 1348 #endif /* CONFIG_MMU */ 1349 1350 /* 1351 * reserve_crashkernel() - reserves memory for crash kernel 1352 * 1353 * This function reserves memory area given in "crashkernel=" kernel command 1354 * line parameter. The memory reserved is used by dump capture kernel when 1355 * primary kernel is crashing. 1356 */ 1357 static void __init arch_reserve_crashkernel(void) 1358 { 1359 unsigned long long low_size = 0; 1360 unsigned long long crash_base, crash_size; 1361 char *cmdline = boot_command_line; 1362 bool high = false; 1363 int ret; 1364 1365 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1366 return; 1367 1368 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 1369 &crash_size, &crash_base, 1370 &low_size, &high); 1371 if (ret) 1372 return; 1373 1374 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 1375 low_size, high); 1376 } 1377 1378 void __init paging_init(void) 1379 { 1380 setup_bootmem(); 1381 setup_vm_final(); 1382 1383 /* Depend on that Linear Mapping is ready */ 1384 memblock_allow_resize(); 1385 } 1386 1387 void __init misc_mem_init(void) 1388 { 1389 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1390 arch_numa_init(); 1391 sparse_init(); 1392 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1393 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1394 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1395 #endif 1396 zone_sizes_init(); 1397 arch_reserve_crashkernel(); 1398 memblock_dump_all(); 1399 } 1400 1401 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1402 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1403 unsigned long addr, unsigned long next) 1404 { 1405 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1406 } 1407 1408 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1409 unsigned long addr, unsigned long next) 1410 { 1411 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1412 return 1; 1413 } 1414 1415 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1416 struct vmem_altmap *altmap) 1417 { 1418 /* 1419 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1420 * can't use hugepage mappings for 2-level page table because in case of 1421 * memory hotplug, we are not able to update all the page tables with 1422 * the new PMDs. 1423 */ 1424 return vmemmap_populate_hugepages(start, end, node, NULL); 1425 } 1426 #endif 1427 1428 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1429 /* 1430 * Pre-allocates page-table pages for a specific area in the kernel 1431 * page-table. Only the level which needs to be synchronized between 1432 * all page-tables is allocated because the synchronization can be 1433 * expensive. 1434 */ 1435 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1436 const char *area) 1437 { 1438 unsigned long addr; 1439 const char *lvl; 1440 1441 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1442 pgd_t *pgd = pgd_offset_k(addr); 1443 p4d_t *p4d; 1444 pud_t *pud; 1445 pmd_t *pmd; 1446 1447 lvl = "p4d"; 1448 p4d = p4d_alloc(&init_mm, pgd, addr); 1449 if (!p4d) 1450 goto failed; 1451 1452 if (pgtable_l5_enabled) 1453 continue; 1454 1455 lvl = "pud"; 1456 pud = pud_alloc(&init_mm, p4d, addr); 1457 if (!pud) 1458 goto failed; 1459 1460 if (pgtable_l4_enabled) 1461 continue; 1462 1463 lvl = "pmd"; 1464 pmd = pmd_alloc(&init_mm, pud, addr); 1465 if (!pmd) 1466 goto failed; 1467 } 1468 return; 1469 1470 failed: 1471 /* 1472 * The pages have to be there now or they will be missing in 1473 * process page-tables later. 1474 */ 1475 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1476 } 1477 1478 void __init pgtable_cache_init(void) 1479 { 1480 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1481 if (IS_ENABLED(CONFIG_MODULES)) 1482 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1483 } 1484 #endif 1485 1486 #ifdef CONFIG_EXECMEM 1487 #ifdef CONFIG_MMU 1488 static struct execmem_info execmem_info __ro_after_init; 1489 1490 struct execmem_info __init *execmem_arch_setup(void) 1491 { 1492 execmem_info = (struct execmem_info){ 1493 .ranges = { 1494 [EXECMEM_DEFAULT] = { 1495 .start = MODULES_VADDR, 1496 .end = MODULES_END, 1497 .pgprot = PAGE_KERNEL, 1498 .alignment = 1, 1499 }, 1500 [EXECMEM_KPROBES] = { 1501 .start = VMALLOC_START, 1502 .end = VMALLOC_END, 1503 .pgprot = PAGE_KERNEL_READ_EXEC, 1504 .alignment = 1, 1505 }, 1506 [EXECMEM_BPF] = { 1507 .start = BPF_JIT_REGION_START, 1508 .end = BPF_JIT_REGION_END, 1509 .pgprot = PAGE_KERNEL, 1510 .alignment = PAGE_SIZE, 1511 }, 1512 }, 1513 }; 1514 1515 return &execmem_info; 1516 } 1517 #endif /* CONFIG_MMU */ 1518 #endif /* CONFIG_EXECMEM */ 1519