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 27 #include <asm/fixmap.h> 28 #include <asm/tlbflush.h> 29 #include <asm/sections.h> 30 #include <asm/soc.h> 31 #include <asm/io.h> 32 #include <asm/ptdump.h> 33 #include <asm/numa.h> 34 35 #include "../kernel/head.h" 36 37 struct kernel_mapping kernel_map __ro_after_init; 38 EXPORT_SYMBOL(kernel_map); 39 #ifdef CONFIG_XIP_KERNEL 40 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 41 #endif 42 43 #ifdef CONFIG_64BIT 44 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 45 #else 46 u64 satp_mode __ro_after_init = SATP_MODE_32; 47 #endif 48 EXPORT_SYMBOL(satp_mode); 49 50 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 51 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 52 EXPORT_SYMBOL(pgtable_l4_enabled); 53 EXPORT_SYMBOL(pgtable_l5_enabled); 54 55 phys_addr_t phys_ram_base __ro_after_init; 56 EXPORT_SYMBOL(phys_ram_base); 57 58 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 59 __page_aligned_bss; 60 EXPORT_SYMBOL(empty_zero_page); 61 62 extern char _start[]; 63 #define DTB_EARLY_BASE_VA (ADDRESS_SPACE_END - (PTRS_PER_PGD / 2 * PGDIR_SIZE) + 1) 64 void *_dtb_early_va __initdata; 65 uintptr_t _dtb_early_pa __initdata; 66 67 static phys_addr_t dma32_phys_limit __initdata; 68 69 static void __init zone_sizes_init(void) 70 { 71 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 72 73 #ifdef CONFIG_ZONE_DMA32 74 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 75 #endif 76 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 77 78 free_area_init(max_zone_pfns); 79 } 80 81 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 82 83 #define LOG2_SZ_1K ilog2(SZ_1K) 84 #define LOG2_SZ_1M ilog2(SZ_1M) 85 #define LOG2_SZ_1G ilog2(SZ_1G) 86 #define LOG2_SZ_1T ilog2(SZ_1T) 87 88 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 89 { 90 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 91 (((t) - (b)) >> LOG2_SZ_1K)); 92 } 93 94 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 95 { 96 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 97 (((t) - (b)) >> LOG2_SZ_1M)); 98 } 99 100 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 101 { 102 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 103 (((t) - (b)) >> LOG2_SZ_1G)); 104 } 105 106 #ifdef CONFIG_64BIT 107 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 108 { 109 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 110 (((t) - (b)) >> LOG2_SZ_1T)); 111 } 112 #else 113 #define print_mlt(n, b, t) do {} while (0) 114 #endif 115 116 static inline void print_ml(char *name, unsigned long b, unsigned long t) 117 { 118 unsigned long diff = t - b; 119 120 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 121 print_mlt(name, b, t); 122 else if ((diff >> LOG2_SZ_1G) >= 10) 123 print_mlg(name, b, t); 124 else if ((diff >> LOG2_SZ_1M) >= 10) 125 print_mlm(name, b, t); 126 else 127 print_mlk(name, b, t); 128 } 129 130 static void __init print_vm_layout(void) 131 { 132 pr_notice("Virtual kernel memory layout:\n"); 133 print_ml("fixmap", (unsigned long)FIXADDR_START, 134 (unsigned long)FIXADDR_TOP); 135 print_ml("pci io", (unsigned long)PCI_IO_START, 136 (unsigned long)PCI_IO_END); 137 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 138 (unsigned long)VMEMMAP_END); 139 print_ml("vmalloc", (unsigned long)VMALLOC_START, 140 (unsigned long)VMALLOC_END); 141 #ifdef CONFIG_64BIT 142 print_ml("modules", (unsigned long)MODULES_VADDR, 143 (unsigned long)MODULES_END); 144 #endif 145 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 146 (unsigned long)high_memory); 147 if (IS_ENABLED(CONFIG_64BIT)) { 148 #ifdef CONFIG_KASAN 149 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 150 #endif 151 152 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 153 (unsigned long)ADDRESS_SPACE_END); 154 } 155 } 156 #else 157 static void print_vm_layout(void) { } 158 #endif /* CONFIG_DEBUG_VM */ 159 160 void __init mem_init(void) 161 { 162 #ifdef CONFIG_FLATMEM 163 BUG_ON(!mem_map); 164 #endif /* CONFIG_FLATMEM */ 165 166 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE); 167 memblock_free_all(); 168 169 print_vm_layout(); 170 } 171 172 /* Limit the memory size via mem. */ 173 static phys_addr_t memory_limit; 174 175 static int __init early_mem(char *p) 176 { 177 u64 size; 178 179 if (!p) 180 return 1; 181 182 size = memparse(p, &p) & PAGE_MASK; 183 memory_limit = min_t(u64, size, memory_limit); 184 185 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 186 187 return 0; 188 } 189 early_param("mem", early_mem); 190 191 static void __init setup_bootmem(void) 192 { 193 phys_addr_t vmlinux_end = __pa_symbol(&_end); 194 phys_addr_t max_mapped_addr; 195 phys_addr_t phys_ram_end, vmlinux_start; 196 197 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 198 vmlinux_start = __pa_symbol(&_sdata); 199 else 200 vmlinux_start = __pa_symbol(&_start); 201 202 memblock_enforce_memory_limit(memory_limit); 203 204 /* 205 * Make sure we align the reservation on PMD_SIZE since we will 206 * map the kernel in the linear mapping as read-only: we do not want 207 * any allocation to happen between _end and the next pmd aligned page. 208 */ 209 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 210 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 211 /* 212 * Reserve from the start of the kernel to the end of the kernel 213 */ 214 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 215 216 phys_ram_end = memblock_end_of_DRAM(); 217 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 218 phys_ram_base = memblock_start_of_DRAM(); 219 220 /* 221 * In 64-bit, any use of __va/__pa before this point is wrong as we 222 * did not know the start of DRAM before. 223 */ 224 if (IS_ENABLED(CONFIG_64BIT)) 225 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 226 227 /* 228 * memblock allocator is not aware of the fact that last 4K bytes of 229 * the addressable memory can not be mapped because of IS_ERR_VALUE 230 * macro. Make sure that last 4k bytes are not usable by memblock 231 * if end of dram is equal to maximum addressable memory. For 64-bit 232 * kernel, this problem can't happen here as the end of the virtual 233 * address space is occupied by the kernel mapping then this check must 234 * be done as soon as the kernel mapping base address is determined. 235 */ 236 if (!IS_ENABLED(CONFIG_64BIT)) { 237 max_mapped_addr = __pa(~(ulong)0); 238 if (max_mapped_addr == (phys_ram_end - 1)) 239 memblock_set_current_limit(max_mapped_addr - 4096); 240 } 241 242 min_low_pfn = PFN_UP(phys_ram_base); 243 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 244 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 245 246 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 247 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 248 249 reserve_initrd_mem(); 250 /* 251 * If DTB is built in, no need to reserve its memblock. 252 * Otherwise, do reserve it but avoid using 253 * early_init_fdt_reserve_self() since __pa() does 254 * not work for DTB pointers that are fixmap addresses 255 */ 256 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) { 257 /* 258 * In case the DTB is not located in a memory region we won't 259 * be able to locate it later on via the linear mapping and 260 * get a segfault when accessing it via __va(dtb_early_pa). 261 * To avoid this situation copy DTB to a memory region. 262 * Note that memblock_phys_alloc will also reserve DTB region. 263 */ 264 if (!memblock_is_memory(dtb_early_pa)) { 265 size_t fdt_size = fdt_totalsize(dtb_early_va); 266 phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE); 267 void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size); 268 269 memcpy(new_dtb_early_va, dtb_early_va, fdt_size); 270 early_memunmap(new_dtb_early_va, fdt_size); 271 _dtb_early_pa = new_dtb_early_pa; 272 } else 273 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 274 } 275 276 dma_contiguous_reserve(dma32_phys_limit); 277 if (IS_ENABLED(CONFIG_64BIT)) 278 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 279 memblock_allow_resize(); 280 } 281 282 #ifdef CONFIG_MMU 283 struct pt_alloc_ops pt_ops __initdata; 284 285 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 286 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 287 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 288 289 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 290 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 291 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 292 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 293 294 #ifdef CONFIG_XIP_KERNEL 295 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 296 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 297 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 298 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 299 #endif /* CONFIG_XIP_KERNEL */ 300 301 static const pgprot_t protection_map[16] = { 302 [VM_NONE] = PAGE_NONE, 303 [VM_READ] = PAGE_READ, 304 [VM_WRITE] = PAGE_COPY, 305 [VM_WRITE | VM_READ] = PAGE_COPY, 306 [VM_EXEC] = PAGE_EXEC, 307 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 308 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 309 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_READ_EXEC, 310 [VM_SHARED] = PAGE_NONE, 311 [VM_SHARED | VM_READ] = PAGE_READ, 312 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 313 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 314 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 315 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 316 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 317 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 318 }; 319 DECLARE_VM_GET_PAGE_PROT 320 321 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 322 { 323 unsigned long addr = __fix_to_virt(idx); 324 pte_t *ptep; 325 326 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 327 328 ptep = &fixmap_pte[pte_index(addr)]; 329 330 if (pgprot_val(prot)) 331 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 332 else 333 pte_clear(&init_mm, addr, ptep); 334 local_flush_tlb_page(addr); 335 } 336 337 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 338 { 339 return (pte_t *)((uintptr_t)pa); 340 } 341 342 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 343 { 344 clear_fixmap(FIX_PTE); 345 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 346 } 347 348 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 349 { 350 return (pte_t *) __va(pa); 351 } 352 353 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 354 { 355 /* 356 * We only create PMD or PGD early mappings so we 357 * should never reach here with MMU disabled. 358 */ 359 BUG(); 360 } 361 362 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 363 { 364 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 365 } 366 367 static phys_addr_t __init alloc_pte_late(uintptr_t va) 368 { 369 unsigned long vaddr; 370 371 vaddr = __get_free_page(GFP_KERNEL); 372 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr))); 373 374 return __pa(vaddr); 375 } 376 377 static void __init create_pte_mapping(pte_t *ptep, 378 uintptr_t va, phys_addr_t pa, 379 phys_addr_t sz, pgprot_t prot) 380 { 381 uintptr_t pte_idx = pte_index(va); 382 383 BUG_ON(sz != PAGE_SIZE); 384 385 if (pte_none(ptep[pte_idx])) 386 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 387 } 388 389 #ifndef __PAGETABLE_PMD_FOLDED 390 391 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 392 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 393 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 394 395 #ifdef CONFIG_XIP_KERNEL 396 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 397 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 398 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 399 #endif /* CONFIG_XIP_KERNEL */ 400 401 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 402 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 403 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 404 405 #ifdef CONFIG_XIP_KERNEL 406 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 407 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 408 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 409 #endif /* CONFIG_XIP_KERNEL */ 410 411 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 412 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 413 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 414 415 #ifdef CONFIG_XIP_KERNEL 416 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 417 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 418 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 419 #endif /* CONFIG_XIP_KERNEL */ 420 421 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 422 { 423 /* Before MMU is enabled */ 424 return (pmd_t *)((uintptr_t)pa); 425 } 426 427 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 428 { 429 clear_fixmap(FIX_PMD); 430 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 431 } 432 433 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 434 { 435 return (pmd_t *) __va(pa); 436 } 437 438 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 439 { 440 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 441 442 return (uintptr_t)early_pmd; 443 } 444 445 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 446 { 447 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 448 } 449 450 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 451 { 452 unsigned long vaddr; 453 454 vaddr = __get_free_page(GFP_KERNEL); 455 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr))); 456 457 return __pa(vaddr); 458 } 459 460 static void __init create_pmd_mapping(pmd_t *pmdp, 461 uintptr_t va, phys_addr_t pa, 462 phys_addr_t sz, pgprot_t prot) 463 { 464 pte_t *ptep; 465 phys_addr_t pte_phys; 466 uintptr_t pmd_idx = pmd_index(va); 467 468 if (sz == PMD_SIZE) { 469 if (pmd_none(pmdp[pmd_idx])) 470 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 471 return; 472 } 473 474 if (pmd_none(pmdp[pmd_idx])) { 475 pte_phys = pt_ops.alloc_pte(va); 476 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 477 ptep = pt_ops.get_pte_virt(pte_phys); 478 memset(ptep, 0, PAGE_SIZE); 479 } else { 480 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 481 ptep = pt_ops.get_pte_virt(pte_phys); 482 } 483 484 create_pte_mapping(ptep, va, pa, sz, prot); 485 } 486 487 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 488 { 489 return (pud_t *)((uintptr_t)pa); 490 } 491 492 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 493 { 494 clear_fixmap(FIX_PUD); 495 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 496 } 497 498 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 499 { 500 return (pud_t *)__va(pa); 501 } 502 503 static phys_addr_t __init alloc_pud_early(uintptr_t va) 504 { 505 /* Only one PUD is available for early mapping */ 506 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 507 508 return (uintptr_t)early_pud; 509 } 510 511 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 512 { 513 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 514 } 515 516 static phys_addr_t alloc_pud_late(uintptr_t va) 517 { 518 unsigned long vaddr; 519 520 vaddr = __get_free_page(GFP_KERNEL); 521 BUG_ON(!vaddr); 522 return __pa(vaddr); 523 } 524 525 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 526 { 527 return (p4d_t *)((uintptr_t)pa); 528 } 529 530 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 531 { 532 clear_fixmap(FIX_P4D); 533 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 534 } 535 536 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 537 { 538 return (p4d_t *)__va(pa); 539 } 540 541 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 542 { 543 /* Only one P4D is available for early mapping */ 544 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 545 546 return (uintptr_t)early_p4d; 547 } 548 549 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 550 { 551 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 552 } 553 554 static phys_addr_t alloc_p4d_late(uintptr_t va) 555 { 556 unsigned long vaddr; 557 558 vaddr = __get_free_page(GFP_KERNEL); 559 BUG_ON(!vaddr); 560 return __pa(vaddr); 561 } 562 563 static void __init create_pud_mapping(pud_t *pudp, 564 uintptr_t va, phys_addr_t pa, 565 phys_addr_t sz, pgprot_t prot) 566 { 567 pmd_t *nextp; 568 phys_addr_t next_phys; 569 uintptr_t pud_index = pud_index(va); 570 571 if (sz == PUD_SIZE) { 572 if (pud_val(pudp[pud_index]) == 0) 573 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 574 return; 575 } 576 577 if (pud_val(pudp[pud_index]) == 0) { 578 next_phys = pt_ops.alloc_pmd(va); 579 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 580 nextp = pt_ops.get_pmd_virt(next_phys); 581 memset(nextp, 0, PAGE_SIZE); 582 } else { 583 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 584 nextp = pt_ops.get_pmd_virt(next_phys); 585 } 586 587 create_pmd_mapping(nextp, va, pa, sz, prot); 588 } 589 590 static void __init create_p4d_mapping(p4d_t *p4dp, 591 uintptr_t va, phys_addr_t pa, 592 phys_addr_t sz, pgprot_t prot) 593 { 594 pud_t *nextp; 595 phys_addr_t next_phys; 596 uintptr_t p4d_index = p4d_index(va); 597 598 if (sz == P4D_SIZE) { 599 if (p4d_val(p4dp[p4d_index]) == 0) 600 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 601 return; 602 } 603 604 if (p4d_val(p4dp[p4d_index]) == 0) { 605 next_phys = pt_ops.alloc_pud(va); 606 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 607 nextp = pt_ops.get_pud_virt(next_phys); 608 memset(nextp, 0, PAGE_SIZE); 609 } else { 610 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 611 nextp = pt_ops.get_pud_virt(next_phys); 612 } 613 614 create_pud_mapping(nextp, va, pa, sz, prot); 615 } 616 617 #define pgd_next_t p4d_t 618 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 619 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 620 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 621 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 622 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 623 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 624 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 625 (pgtable_l5_enabled ? \ 626 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 627 (pgtable_l4_enabled ? \ 628 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 629 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 630 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 631 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 632 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 633 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 634 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 635 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 636 #define early_dtb_pgd_next (pgtable_l5_enabled ? \ 637 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \ 638 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd)) 639 #else 640 #define pgd_next_t pte_t 641 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 642 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 643 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 644 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 645 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 646 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd) 647 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 648 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 649 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 650 #endif /* __PAGETABLE_PMD_FOLDED */ 651 652 void __init create_pgd_mapping(pgd_t *pgdp, 653 uintptr_t va, phys_addr_t pa, 654 phys_addr_t sz, pgprot_t prot) 655 { 656 pgd_next_t *nextp; 657 phys_addr_t next_phys; 658 uintptr_t pgd_idx = pgd_index(va); 659 660 if (sz == PGDIR_SIZE) { 661 if (pgd_val(pgdp[pgd_idx]) == 0) 662 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 663 return; 664 } 665 666 if (pgd_val(pgdp[pgd_idx]) == 0) { 667 next_phys = alloc_pgd_next(va); 668 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 669 nextp = get_pgd_next_virt(next_phys); 670 memset(nextp, 0, PAGE_SIZE); 671 } else { 672 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 673 nextp = get_pgd_next_virt(next_phys); 674 } 675 676 create_pgd_next_mapping(nextp, va, pa, sz, prot); 677 } 678 679 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size) 680 { 681 if (!(base & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE) 682 return PGDIR_SIZE; 683 684 if (!(base & (P4D_SIZE - 1)) && size >= P4D_SIZE) 685 return P4D_SIZE; 686 687 if (!(base & (PUD_SIZE - 1)) && size >= PUD_SIZE) 688 return PUD_SIZE; 689 690 if (!(base & (PMD_SIZE - 1)) && size >= PMD_SIZE) 691 return PMD_SIZE; 692 693 return PAGE_SIZE; 694 } 695 696 #ifdef CONFIG_XIP_KERNEL 697 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 698 extern char _xiprom[], _exiprom[], __data_loc; 699 700 /* called from head.S with MMU off */ 701 asmlinkage void __init __copy_data(void) 702 { 703 void *from = (void *)(&__data_loc); 704 void *to = (void *)CONFIG_PHYS_RAM_BASE; 705 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 706 707 memcpy(to, from, sz); 708 } 709 #endif 710 711 #ifdef CONFIG_STRICT_KERNEL_RWX 712 static __init pgprot_t pgprot_from_va(uintptr_t va) 713 { 714 if (is_va_kernel_text(va)) 715 return PAGE_KERNEL_READ_EXEC; 716 717 /* 718 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 719 * we must protect its linear mapping alias from being executed and 720 * written. 721 * And rodata section is marked readonly in mark_rodata_ro. 722 */ 723 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 724 return PAGE_KERNEL_READ; 725 726 return PAGE_KERNEL; 727 } 728 729 void mark_rodata_ro(void) 730 { 731 set_kernel_memory(__start_rodata, _data, set_memory_ro); 732 if (IS_ENABLED(CONFIG_64BIT)) 733 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 734 set_memory_ro); 735 736 debug_checkwx(); 737 } 738 #else 739 static __init pgprot_t pgprot_from_va(uintptr_t va) 740 { 741 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 742 return PAGE_KERNEL; 743 744 return PAGE_KERNEL_EXEC; 745 } 746 #endif /* CONFIG_STRICT_KERNEL_RWX */ 747 748 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 749 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 750 751 static void __init disable_pgtable_l5(void) 752 { 753 pgtable_l5_enabled = false; 754 kernel_map.page_offset = PAGE_OFFSET_L4; 755 satp_mode = SATP_MODE_48; 756 } 757 758 static void __init disable_pgtable_l4(void) 759 { 760 pgtable_l4_enabled = false; 761 kernel_map.page_offset = PAGE_OFFSET_L3; 762 satp_mode = SATP_MODE_39; 763 } 764 765 static int __init print_no4lvl(char *p) 766 { 767 pr_info("Disabled 4-level and 5-level paging"); 768 return 0; 769 } 770 early_param("no4lvl", print_no4lvl); 771 772 static int __init print_no5lvl(char *p) 773 { 774 pr_info("Disabled 5-level paging"); 775 return 0; 776 } 777 early_param("no5lvl", print_no5lvl); 778 779 /* 780 * There is a simple way to determine if 4-level is supported by the 781 * underlying hardware: establish 1:1 mapping in 4-level page table mode 782 * then read SATP to see if the configuration was taken into account 783 * meaning sv48 is supported. 784 */ 785 static __init void set_satp_mode(uintptr_t dtb_pa) 786 { 787 u64 identity_satp, hw_satp; 788 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 789 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 790 791 if (satp_mode_cmdline == SATP_MODE_57) { 792 disable_pgtable_l5(); 793 } else if (satp_mode_cmdline == SATP_MODE_48) { 794 disable_pgtable_l5(); 795 disable_pgtable_l4(); 796 return; 797 } 798 799 create_p4d_mapping(early_p4d, 800 set_satp_mode_pmd, (uintptr_t)early_pud, 801 P4D_SIZE, PAGE_TABLE); 802 create_pud_mapping(early_pud, 803 set_satp_mode_pmd, (uintptr_t)early_pmd, 804 PUD_SIZE, PAGE_TABLE); 805 /* Handle the case where set_satp_mode straddles 2 PMDs */ 806 create_pmd_mapping(early_pmd, 807 set_satp_mode_pmd, set_satp_mode_pmd, 808 PMD_SIZE, PAGE_KERNEL_EXEC); 809 create_pmd_mapping(early_pmd, 810 set_satp_mode_pmd + PMD_SIZE, 811 set_satp_mode_pmd + PMD_SIZE, 812 PMD_SIZE, PAGE_KERNEL_EXEC); 813 retry: 814 create_pgd_mapping(early_pg_dir, 815 set_satp_mode_pmd, 816 pgtable_l5_enabled ? 817 (uintptr_t)early_p4d : (uintptr_t)early_pud, 818 PGDIR_SIZE, PAGE_TABLE); 819 820 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 821 822 local_flush_tlb_all(); 823 csr_write(CSR_SATP, identity_satp); 824 hw_satp = csr_swap(CSR_SATP, 0ULL); 825 local_flush_tlb_all(); 826 827 if (hw_satp != identity_satp) { 828 if (pgtable_l5_enabled) { 829 disable_pgtable_l5(); 830 memset(early_pg_dir, 0, PAGE_SIZE); 831 goto retry; 832 } 833 disable_pgtable_l4(); 834 } 835 836 memset(early_pg_dir, 0, PAGE_SIZE); 837 memset(early_p4d, 0, PAGE_SIZE); 838 memset(early_pud, 0, PAGE_SIZE); 839 memset(early_pmd, 0, PAGE_SIZE); 840 } 841 #endif 842 843 /* 844 * setup_vm() is called from head.S with MMU-off. 845 * 846 * Following requirements should be honoured for setup_vm() to work 847 * correctly: 848 * 1) It should use PC-relative addressing for accessing kernel symbols. 849 * To achieve this we always use GCC cmodel=medany. 850 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 851 * so disable compiler instrumentation when FTRACE is enabled. 852 * 853 * Currently, the above requirements are honoured by using custom CFLAGS 854 * for init.o in mm/Makefile. 855 */ 856 857 #ifndef __riscv_cmodel_medany 858 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 859 #endif 860 861 #ifdef CONFIG_RELOCATABLE 862 extern unsigned long __rela_dyn_start, __rela_dyn_end; 863 864 static void __init relocate_kernel(void) 865 { 866 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 867 /* 868 * This holds the offset between the linked virtual address and the 869 * relocated virtual address. 870 */ 871 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 872 /* 873 * This holds the offset between kernel linked virtual address and 874 * physical address. 875 */ 876 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 877 878 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 879 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 880 Elf64_Addr relocated_addr = rela->r_addend; 881 882 if (rela->r_info != R_RISCV_RELATIVE) 883 continue; 884 885 /* 886 * Make sure to not relocate vdso symbols like rt_sigreturn 887 * which are linked from the address 0 in vmlinux since 888 * vdso symbol addresses are actually used as an offset from 889 * mm->context.vdso in VDSO_OFFSET macro. 890 */ 891 if (relocated_addr >= KERNEL_LINK_ADDR) 892 relocated_addr += reloc_offset; 893 894 *(Elf64_Addr *)addr = relocated_addr; 895 } 896 } 897 #endif /* CONFIG_RELOCATABLE */ 898 899 #ifdef CONFIG_XIP_KERNEL 900 static void __init create_kernel_page_table(pgd_t *pgdir, 901 __always_unused bool early) 902 { 903 uintptr_t va, end_va; 904 905 /* Map the flash resident part */ 906 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 907 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 908 create_pgd_mapping(pgdir, va, 909 kernel_map.xiprom + (va - kernel_map.virt_addr), 910 PMD_SIZE, PAGE_KERNEL_EXEC); 911 912 /* Map the data in RAM */ 913 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 914 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 915 create_pgd_mapping(pgdir, va, 916 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 917 PMD_SIZE, PAGE_KERNEL); 918 } 919 #else 920 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 921 { 922 uintptr_t va, end_va; 923 924 end_va = kernel_map.virt_addr + kernel_map.size; 925 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 926 create_pgd_mapping(pgdir, va, 927 kernel_map.phys_addr + (va - kernel_map.virt_addr), 928 PMD_SIZE, 929 early ? 930 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 931 } 932 #endif 933 934 /* 935 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 936 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 937 * entry. 938 */ 939 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa) 940 { 941 #ifndef CONFIG_BUILTIN_DTB 942 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 943 944 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 945 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa, 946 PGDIR_SIZE, 947 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL); 948 949 if (pgtable_l5_enabled) 950 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA, 951 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE); 952 953 if (pgtable_l4_enabled) 954 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA, 955 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE); 956 957 if (IS_ENABLED(CONFIG_64BIT)) { 958 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA, 959 pa, PMD_SIZE, PAGE_KERNEL); 960 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE, 961 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 962 } 963 964 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1)); 965 #else 966 /* 967 * For 64-bit kernel, __va can't be used since it would return a linear 968 * mapping address whereas dtb_early_va will be used before 969 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 970 * kernel is mapped in the linear mapping, that makes no difference. 971 */ 972 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 973 #endif 974 975 dtb_early_pa = dtb_pa; 976 } 977 978 /* 979 * MMU is not enabled, the page tables are allocated directly using 980 * early_pmd/pud/p4d and the address returned is the physical one. 981 */ 982 static void __init pt_ops_set_early(void) 983 { 984 pt_ops.alloc_pte = alloc_pte_early; 985 pt_ops.get_pte_virt = get_pte_virt_early; 986 #ifndef __PAGETABLE_PMD_FOLDED 987 pt_ops.alloc_pmd = alloc_pmd_early; 988 pt_ops.get_pmd_virt = get_pmd_virt_early; 989 pt_ops.alloc_pud = alloc_pud_early; 990 pt_ops.get_pud_virt = get_pud_virt_early; 991 pt_ops.alloc_p4d = alloc_p4d_early; 992 pt_ops.get_p4d_virt = get_p4d_virt_early; 993 #endif 994 } 995 996 /* 997 * MMU is enabled but page table setup is not complete yet. 998 * fixmap page table alloc functions must be used as a means to temporarily 999 * map the allocated physical pages since the linear mapping does not exist yet. 1000 * 1001 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1002 * but it will be used as described above. 1003 */ 1004 static void __init pt_ops_set_fixmap(void) 1005 { 1006 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1007 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1008 #ifndef __PAGETABLE_PMD_FOLDED 1009 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1010 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1011 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1012 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1013 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1014 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1015 #endif 1016 } 1017 1018 /* 1019 * MMU is enabled and page table setup is complete, so from now, we can use 1020 * generic page allocation functions to setup page table. 1021 */ 1022 static void __init pt_ops_set_late(void) 1023 { 1024 pt_ops.alloc_pte = alloc_pte_late; 1025 pt_ops.get_pte_virt = get_pte_virt_late; 1026 #ifndef __PAGETABLE_PMD_FOLDED 1027 pt_ops.alloc_pmd = alloc_pmd_late; 1028 pt_ops.get_pmd_virt = get_pmd_virt_late; 1029 pt_ops.alloc_pud = alloc_pud_late; 1030 pt_ops.get_pud_virt = get_pud_virt_late; 1031 pt_ops.alloc_p4d = alloc_p4d_late; 1032 pt_ops.get_p4d_virt = get_p4d_virt_late; 1033 #endif 1034 } 1035 1036 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1037 { 1038 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1039 1040 kernel_map.virt_addr = KERNEL_LINK_ADDR; 1041 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1042 1043 #ifdef CONFIG_XIP_KERNEL 1044 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1045 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1046 1047 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1048 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1049 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1050 1051 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1052 #else 1053 kernel_map.phys_addr = (uintptr_t)(&_start); 1054 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1055 #endif 1056 1057 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1058 set_satp_mode(dtb_pa); 1059 #endif 1060 1061 /* 1062 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1063 * where we have the system memory layout: this allows us to align 1064 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1065 * for the linear mapping. This is only possible because the kernel 1066 * mapping lies outside the linear mapping. 1067 * In 32-bit however, as the kernel resides in the linear mapping, 1068 * setup_vm_final can not change the mapping established here, 1069 * otherwise the same kernel addresses would get mapped to different 1070 * physical addresses (if the start of dram is different from the 1071 * kernel physical address start). 1072 */ 1073 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1074 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1075 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1076 1077 /* 1078 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1079 * kernel, whereas for 64-bit kernel, the end of the virtual address 1080 * space is occupied by the modules/BPF/kernel mappings which reduces 1081 * the available size of the linear mapping. 1082 */ 1083 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1084 1085 /* Sanity check alignment and size */ 1086 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1087 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1088 1089 #ifdef CONFIG_64BIT 1090 /* 1091 * The last 4K bytes of the addressable memory can not be mapped because 1092 * of IS_ERR_VALUE macro. 1093 */ 1094 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1095 #endif 1096 1097 #ifdef CONFIG_RELOCATABLE 1098 /* 1099 * Early page table uses only one PUD, which makes it possible 1100 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1101 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1102 * since a part of the kernel would not get mapped. 1103 */ 1104 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1105 relocate_kernel(); 1106 #endif 1107 1108 apply_early_boot_alternatives(); 1109 pt_ops_set_early(); 1110 1111 /* Setup early PGD for fixmap */ 1112 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1113 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1114 1115 #ifndef __PAGETABLE_PMD_FOLDED 1116 /* Setup fixmap P4D and PUD */ 1117 if (pgtable_l5_enabled) 1118 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1119 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1120 /* Setup fixmap PUD and PMD */ 1121 if (pgtable_l4_enabled) 1122 create_pud_mapping(fixmap_pud, FIXADDR_START, 1123 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1124 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1125 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1126 /* Setup trampoline PGD and PMD */ 1127 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1128 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1129 if (pgtable_l5_enabled) 1130 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1131 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1132 if (pgtable_l4_enabled) 1133 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1134 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1135 #ifdef CONFIG_XIP_KERNEL 1136 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1137 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1138 #else 1139 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1140 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1141 #endif 1142 #else 1143 /* Setup trampoline PGD */ 1144 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1145 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1146 #endif 1147 1148 /* 1149 * Setup early PGD covering entire kernel which will allow 1150 * us to reach paging_init(). We map all memory banks later 1151 * in setup_vm_final() below. 1152 */ 1153 create_kernel_page_table(early_pg_dir, true); 1154 1155 /* Setup early mapping for FDT early scan */ 1156 create_fdt_early_page_table(early_pg_dir, dtb_pa); 1157 1158 /* 1159 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1160 * range can not span multiple pmds. 1161 */ 1162 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1163 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1164 1165 #ifndef __PAGETABLE_PMD_FOLDED 1166 /* 1167 * Early ioremap fixmap is already created as it lies within first 2MB 1168 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1169 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1170 * the user if not. 1171 */ 1172 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1173 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1174 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1175 WARN_ON(1); 1176 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1177 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1178 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1179 fix_to_virt(FIX_BTMAP_BEGIN)); 1180 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1181 fix_to_virt(FIX_BTMAP_END)); 1182 1183 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1184 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1185 } 1186 #endif 1187 1188 pt_ops_set_fixmap(); 1189 } 1190 1191 static void __init create_linear_mapping_range(phys_addr_t start, 1192 phys_addr_t end) 1193 { 1194 phys_addr_t pa; 1195 uintptr_t va, map_size; 1196 1197 for (pa = start; pa < end; pa += map_size) { 1198 va = (uintptr_t)__va(pa); 1199 map_size = best_map_size(pa, end - pa); 1200 1201 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1202 pgprot_from_va(va)); 1203 } 1204 } 1205 1206 static void __init create_linear_mapping_page_table(void) 1207 { 1208 phys_addr_t start, end; 1209 u64 i; 1210 1211 #ifdef CONFIG_STRICT_KERNEL_RWX 1212 phys_addr_t ktext_start = __pa_symbol(_start); 1213 phys_addr_t ktext_size = __init_data_begin - _start; 1214 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1215 phys_addr_t krodata_size = _data - __start_rodata; 1216 1217 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1218 memblock_mark_nomap(ktext_start, ktext_size); 1219 memblock_mark_nomap(krodata_start, krodata_size); 1220 #endif 1221 1222 /* Map all memory banks in the linear mapping */ 1223 for_each_mem_range(i, &start, &end) { 1224 if (start >= end) 1225 break; 1226 if (start <= __pa(PAGE_OFFSET) && 1227 __pa(PAGE_OFFSET) < end) 1228 start = __pa(PAGE_OFFSET); 1229 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1230 end = __pa(PAGE_OFFSET) + memory_limit; 1231 1232 create_linear_mapping_range(start, end); 1233 } 1234 1235 #ifdef CONFIG_STRICT_KERNEL_RWX 1236 create_linear_mapping_range(ktext_start, ktext_start + ktext_size); 1237 create_linear_mapping_range(krodata_start, 1238 krodata_start + krodata_size); 1239 1240 memblock_clear_nomap(ktext_start, ktext_size); 1241 memblock_clear_nomap(krodata_start, krodata_size); 1242 #endif 1243 } 1244 1245 static void __init setup_vm_final(void) 1246 { 1247 /* Setup swapper PGD for fixmap */ 1248 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1249 __pa_symbol(fixmap_pgd_next), 1250 PGDIR_SIZE, PAGE_TABLE); 1251 1252 /* Map the linear mapping */ 1253 create_linear_mapping_page_table(); 1254 1255 /* Map the kernel */ 1256 if (IS_ENABLED(CONFIG_64BIT)) 1257 create_kernel_page_table(swapper_pg_dir, false); 1258 1259 #ifdef CONFIG_KASAN 1260 kasan_swapper_init(); 1261 #endif 1262 1263 /* Clear fixmap PTE and PMD mappings */ 1264 clear_fixmap(FIX_PTE); 1265 clear_fixmap(FIX_PMD); 1266 clear_fixmap(FIX_PUD); 1267 clear_fixmap(FIX_P4D); 1268 1269 /* Move to swapper page table */ 1270 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1271 local_flush_tlb_all(); 1272 1273 pt_ops_set_late(); 1274 } 1275 #else 1276 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1277 { 1278 dtb_early_va = (void *)dtb_pa; 1279 dtb_early_pa = dtb_pa; 1280 } 1281 1282 static inline void setup_vm_final(void) 1283 { 1284 } 1285 #endif /* CONFIG_MMU */ 1286 1287 /* 1288 * reserve_crashkernel() - reserves memory for crash kernel 1289 * 1290 * This function reserves memory area given in "crashkernel=" kernel command 1291 * line parameter. The memory reserved is used by dump capture kernel when 1292 * primary kernel is crashing. 1293 */ 1294 static void __init reserve_crashkernel(void) 1295 { 1296 unsigned long long crash_base = 0; 1297 unsigned long long crash_size = 0; 1298 unsigned long search_start = memblock_start_of_DRAM(); 1299 unsigned long search_end = memblock_end_of_DRAM(); 1300 1301 int ret = 0; 1302 1303 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 1304 return; 1305 /* 1306 * Don't reserve a region for a crash kernel on a crash kernel 1307 * since it doesn't make much sense and we have limited memory 1308 * resources. 1309 */ 1310 if (is_kdump_kernel()) { 1311 pr_info("crashkernel: ignoring reservation request\n"); 1312 return; 1313 } 1314 1315 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1316 &crash_size, &crash_base); 1317 if (ret || !crash_size) 1318 return; 1319 1320 crash_size = PAGE_ALIGN(crash_size); 1321 1322 if (crash_base) { 1323 search_start = crash_base; 1324 search_end = crash_base + crash_size; 1325 } 1326 1327 /* 1328 * Current riscv boot protocol requires 2MB alignment for 1329 * RV64 and 4MB alignment for RV32 (hugepage size) 1330 * 1331 * Try to alloc from 32bit addressible physical memory so that 1332 * swiotlb can work on the crash kernel. 1333 */ 1334 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1335 search_start, 1336 min(search_end, (unsigned long) SZ_4G)); 1337 if (crash_base == 0) { 1338 /* Try again without restricting region to 32bit addressible memory */ 1339 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1340 search_start, search_end); 1341 if (crash_base == 0) { 1342 pr_warn("crashkernel: couldn't allocate %lldKB\n", 1343 crash_size >> 10); 1344 return; 1345 } 1346 } 1347 1348 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 1349 crash_base, crash_base + crash_size, crash_size >> 20); 1350 1351 crashk_res.start = crash_base; 1352 crashk_res.end = crash_base + crash_size - 1; 1353 } 1354 1355 void __init paging_init(void) 1356 { 1357 setup_bootmem(); 1358 setup_vm_final(); 1359 } 1360 1361 void __init misc_mem_init(void) 1362 { 1363 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1364 arch_numa_init(); 1365 sparse_init(); 1366 zone_sizes_init(); 1367 reserve_crashkernel(); 1368 memblock_dump_all(); 1369 } 1370 1371 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1372 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1373 struct vmem_altmap *altmap) 1374 { 1375 return vmemmap_populate_basepages(start, end, node, NULL); 1376 } 1377 #endif 1378