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