1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2012 Regents of the University of California 4 * Copyright (C) 2019 Western Digital Corporation or its affiliates. 5 * Copyright (C) 2020 FORTH-ICS/CARV 6 * Nick Kossifidis <mick@ics.forth.gr> 7 */ 8 9 #include <linux/init.h> 10 #include <linux/mm.h> 11 #include <linux/memblock.h> 12 #include <linux/initrd.h> 13 #include <linux/swap.h> 14 #include <linux/swiotlb.h> 15 #include <linux/sizes.h> 16 #include <linux/of_fdt.h> 17 #include <linux/of_reserved_mem.h> 18 #include <linux/libfdt.h> 19 #include <linux/set_memory.h> 20 #include <linux/dma-map-ops.h> 21 #include <linux/crash_dump.h> 22 #include <linux/hugetlb.h> 23 #ifdef CONFIG_RELOCATABLE 24 #include <linux/elf.h> 25 #endif 26 #include <linux/kfence.h> 27 28 #include <asm/fixmap.h> 29 #include <asm/io.h> 30 #include <asm/numa.h> 31 #include <asm/pgtable.h> 32 #include <asm/ptdump.h> 33 #include <asm/sections.h> 34 #include <asm/soc.h> 35 #include <asm/tlbflush.h> 36 37 #include "../kernel/head.h" 38 39 struct kernel_mapping kernel_map __ro_after_init; 40 EXPORT_SYMBOL(kernel_map); 41 #ifdef CONFIG_XIP_KERNEL 42 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 43 #endif 44 45 #ifdef CONFIG_64BIT 46 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 47 #else 48 u64 satp_mode __ro_after_init = SATP_MODE_32; 49 #endif 50 EXPORT_SYMBOL(satp_mode); 51 52 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 53 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 54 EXPORT_SYMBOL(pgtable_l4_enabled); 55 EXPORT_SYMBOL(pgtable_l5_enabled); 56 57 phys_addr_t phys_ram_base __ro_after_init; 58 EXPORT_SYMBOL(phys_ram_base); 59 60 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 61 __page_aligned_bss; 62 EXPORT_SYMBOL(empty_zero_page); 63 64 extern char _start[]; 65 void *_dtb_early_va __initdata; 66 uintptr_t _dtb_early_pa __initdata; 67 68 static phys_addr_t dma32_phys_limit __initdata; 69 70 static void __init zone_sizes_init(void) 71 { 72 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 73 74 #ifdef CONFIG_ZONE_DMA32 75 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 76 #endif 77 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 78 79 free_area_init(max_zone_pfns); 80 } 81 82 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 83 84 #define LOG2_SZ_1K ilog2(SZ_1K) 85 #define LOG2_SZ_1M ilog2(SZ_1M) 86 #define LOG2_SZ_1G ilog2(SZ_1G) 87 #define LOG2_SZ_1T ilog2(SZ_1T) 88 89 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 90 { 91 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 92 (((t) - (b)) >> LOG2_SZ_1K)); 93 } 94 95 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 96 { 97 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 98 (((t) - (b)) >> LOG2_SZ_1M)); 99 } 100 101 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 102 { 103 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 104 (((t) - (b)) >> LOG2_SZ_1G)); 105 } 106 107 #ifdef CONFIG_64BIT 108 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 109 { 110 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 111 (((t) - (b)) >> LOG2_SZ_1T)); 112 } 113 #else 114 #define print_mlt(n, b, t) do {} while (0) 115 #endif 116 117 static inline void print_ml(char *name, unsigned long b, unsigned long t) 118 { 119 unsigned long diff = t - b; 120 121 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 122 print_mlt(name, b, t); 123 else if ((diff >> LOG2_SZ_1G) >= 10) 124 print_mlg(name, b, t); 125 else if ((diff >> LOG2_SZ_1M) >= 10) 126 print_mlm(name, b, t); 127 else 128 print_mlk(name, b, t); 129 } 130 131 static void __init print_vm_layout(void) 132 { 133 pr_notice("Virtual kernel memory layout:\n"); 134 print_ml("fixmap", (unsigned long)FIXADDR_START, 135 (unsigned long)FIXADDR_TOP); 136 print_ml("pci io", (unsigned long)PCI_IO_START, 137 (unsigned long)PCI_IO_END); 138 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 139 (unsigned long)VMEMMAP_END); 140 print_ml("vmalloc", (unsigned long)VMALLOC_START, 141 (unsigned long)VMALLOC_END); 142 #ifdef CONFIG_64BIT 143 print_ml("modules", (unsigned long)MODULES_VADDR, 144 (unsigned long)MODULES_END); 145 #endif 146 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 147 (unsigned long)high_memory); 148 if (IS_ENABLED(CONFIG_64BIT)) { 149 #ifdef CONFIG_KASAN 150 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 151 #endif 152 153 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 154 (unsigned long)ADDRESS_SPACE_END); 155 } 156 } 157 #else 158 static void print_vm_layout(void) { } 159 #endif /* CONFIG_DEBUG_VM */ 160 161 void __init mem_init(void) 162 { 163 #ifdef CONFIG_FLATMEM 164 BUG_ON(!mem_map); 165 #endif /* CONFIG_FLATMEM */ 166 167 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE); 168 memblock_free_all(); 169 170 print_vm_layout(); 171 } 172 173 /* Limit the memory size via mem. */ 174 static phys_addr_t memory_limit; 175 #ifdef CONFIG_XIP_KERNEL 176 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 177 #endif /* CONFIG_XIP_KERNEL */ 178 179 static int __init early_mem(char *p) 180 { 181 u64 size; 182 183 if (!p) 184 return 1; 185 186 size = memparse(p, &p) & PAGE_MASK; 187 memory_limit = min_t(u64, size, memory_limit); 188 189 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 190 191 return 0; 192 } 193 early_param("mem", early_mem); 194 195 static void __init setup_bootmem(void) 196 { 197 phys_addr_t vmlinux_end = __pa_symbol(&_end); 198 phys_addr_t max_mapped_addr; 199 phys_addr_t phys_ram_end, vmlinux_start; 200 201 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 202 vmlinux_start = __pa_symbol(&_sdata); 203 else 204 vmlinux_start = __pa_symbol(&_start); 205 206 memblock_enforce_memory_limit(memory_limit); 207 208 /* 209 * Make sure we align the reservation on PMD_SIZE since we will 210 * map the kernel in the linear mapping as read-only: we do not want 211 * any allocation to happen between _end and the next pmd aligned page. 212 */ 213 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 214 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 215 /* 216 * Reserve from the start of the kernel to the end of the kernel 217 */ 218 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 219 220 phys_ram_end = memblock_end_of_DRAM(); 221 222 /* 223 * Make sure we align the start of the memory on a PMD boundary so that 224 * at worst, we map the linear mapping with PMD mappings. 225 */ 226 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 227 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 228 229 /* 230 * In 64-bit, any use of __va/__pa before this point is wrong as we 231 * did not know the start of DRAM before. 232 */ 233 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 234 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 235 236 /* 237 * memblock allocator is not aware of the fact that last 4K bytes of 238 * the addressable memory can not be mapped because of IS_ERR_VALUE 239 * macro. Make sure that last 4k bytes are not usable by memblock 240 * if end of dram is equal to maximum addressable memory. For 64-bit 241 * kernel, this problem can't happen here as the end of the virtual 242 * address space is occupied by the kernel mapping then this check must 243 * be done as soon as the kernel mapping base address is determined. 244 */ 245 if (!IS_ENABLED(CONFIG_64BIT)) { 246 max_mapped_addr = __pa(~(ulong)0); 247 if (max_mapped_addr == (phys_ram_end - 1)) 248 memblock_set_current_limit(max_mapped_addr - 4096); 249 } 250 251 min_low_pfn = PFN_UP(phys_ram_base); 252 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 253 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 254 255 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 256 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 257 258 reserve_initrd_mem(); 259 260 /* 261 * No allocation should be done before reserving the memory as defined 262 * in the device tree, otherwise the allocation could end up in a 263 * reserved region. 264 */ 265 early_init_fdt_scan_reserved_mem(); 266 267 /* 268 * If DTB is built in, no need to reserve its memblock. 269 * Otherwise, do reserve it but avoid using 270 * early_init_fdt_reserve_self() since __pa() does 271 * not work for DTB pointers that are fixmap addresses 272 */ 273 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 274 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 275 276 dma_contiguous_reserve(dma32_phys_limit); 277 if (IS_ENABLED(CONFIG_64BIT)) 278 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 279 } 280 281 #ifdef CONFIG_MMU 282 struct pt_alloc_ops pt_ops __initdata; 283 284 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 285 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 286 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 287 288 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 289 290 #ifdef CONFIG_XIP_KERNEL 291 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 292 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 293 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 294 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 295 #endif /* CONFIG_XIP_KERNEL */ 296 297 static const pgprot_t protection_map[16] = { 298 [VM_NONE] = PAGE_NONE, 299 [VM_READ] = PAGE_READ, 300 [VM_WRITE] = PAGE_COPY, 301 [VM_WRITE | VM_READ] = PAGE_COPY, 302 [VM_EXEC] = PAGE_EXEC, 303 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 304 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 305 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 306 [VM_SHARED] = PAGE_NONE, 307 [VM_SHARED | VM_READ] = PAGE_READ, 308 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 309 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 310 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 311 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 312 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 313 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 314 }; 315 DECLARE_VM_GET_PAGE_PROT 316 317 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 318 { 319 unsigned long addr = __fix_to_virt(idx); 320 pte_t *ptep; 321 322 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 323 324 ptep = &fixmap_pte[pte_index(addr)]; 325 326 if (pgprot_val(prot)) 327 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 328 else 329 pte_clear(&init_mm, addr, ptep); 330 local_flush_tlb_page(addr); 331 } 332 333 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 334 { 335 return (pte_t *)((uintptr_t)pa); 336 } 337 338 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 339 { 340 clear_fixmap(FIX_PTE); 341 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 342 } 343 344 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 345 { 346 return (pte_t *) __va(pa); 347 } 348 349 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 350 { 351 /* 352 * We only create PMD or PGD early mappings so we 353 * should never reach here with MMU disabled. 354 */ 355 BUG(); 356 } 357 358 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 359 { 360 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 361 } 362 363 static phys_addr_t __init alloc_pte_late(uintptr_t va) 364 { 365 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 366 367 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 368 return __pa((pte_t *)ptdesc_address(ptdesc)); 369 } 370 371 static void __init create_pte_mapping(pte_t *ptep, 372 uintptr_t va, phys_addr_t pa, 373 phys_addr_t sz, pgprot_t prot) 374 { 375 uintptr_t pte_idx = pte_index(va); 376 377 BUG_ON(sz != PAGE_SIZE); 378 379 if (pte_none(ptep[pte_idx])) 380 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 381 } 382 383 #ifndef __PAGETABLE_PMD_FOLDED 384 385 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 386 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 387 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 388 389 #ifdef CONFIG_XIP_KERNEL 390 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 391 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 392 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 393 #endif /* CONFIG_XIP_KERNEL */ 394 395 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 396 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 397 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 398 399 #ifdef CONFIG_XIP_KERNEL 400 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 401 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 402 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 403 #endif /* CONFIG_XIP_KERNEL */ 404 405 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 406 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 407 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 408 409 #ifdef CONFIG_XIP_KERNEL 410 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 411 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 412 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 413 #endif /* CONFIG_XIP_KERNEL */ 414 415 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 416 { 417 /* Before MMU is enabled */ 418 return (pmd_t *)((uintptr_t)pa); 419 } 420 421 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 422 { 423 clear_fixmap(FIX_PMD); 424 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 425 } 426 427 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 428 { 429 return (pmd_t *) __va(pa); 430 } 431 432 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 433 { 434 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 435 436 return (uintptr_t)early_pmd; 437 } 438 439 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 440 { 441 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 442 } 443 444 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 445 { 446 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 447 448 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 449 return __pa((pmd_t *)ptdesc_address(ptdesc)); 450 } 451 452 static void __init create_pmd_mapping(pmd_t *pmdp, 453 uintptr_t va, phys_addr_t pa, 454 phys_addr_t sz, pgprot_t prot) 455 { 456 pte_t *ptep; 457 phys_addr_t pte_phys; 458 uintptr_t pmd_idx = pmd_index(va); 459 460 if (sz == PMD_SIZE) { 461 if (pmd_none(pmdp[pmd_idx])) 462 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 463 return; 464 } 465 466 if (pmd_none(pmdp[pmd_idx])) { 467 pte_phys = pt_ops.alloc_pte(va); 468 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 469 ptep = pt_ops.get_pte_virt(pte_phys); 470 memset(ptep, 0, PAGE_SIZE); 471 } else { 472 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 473 ptep = pt_ops.get_pte_virt(pte_phys); 474 } 475 476 create_pte_mapping(ptep, va, pa, sz, prot); 477 } 478 479 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 480 { 481 return (pud_t *)((uintptr_t)pa); 482 } 483 484 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 485 { 486 clear_fixmap(FIX_PUD); 487 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 488 } 489 490 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 491 { 492 return (pud_t *)__va(pa); 493 } 494 495 static phys_addr_t __init alloc_pud_early(uintptr_t va) 496 { 497 /* Only one PUD is available for early mapping */ 498 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 499 500 return (uintptr_t)early_pud; 501 } 502 503 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 504 { 505 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 506 } 507 508 static phys_addr_t alloc_pud_late(uintptr_t va) 509 { 510 unsigned long vaddr; 511 512 vaddr = __get_free_page(GFP_KERNEL); 513 BUG_ON(!vaddr); 514 return __pa(vaddr); 515 } 516 517 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 518 { 519 return (p4d_t *)((uintptr_t)pa); 520 } 521 522 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 523 { 524 clear_fixmap(FIX_P4D); 525 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 526 } 527 528 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 529 { 530 return (p4d_t *)__va(pa); 531 } 532 533 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 534 { 535 /* Only one P4D is available for early mapping */ 536 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 537 538 return (uintptr_t)early_p4d; 539 } 540 541 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 542 { 543 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 544 } 545 546 static phys_addr_t alloc_p4d_late(uintptr_t va) 547 { 548 unsigned long vaddr; 549 550 vaddr = __get_free_page(GFP_KERNEL); 551 BUG_ON(!vaddr); 552 return __pa(vaddr); 553 } 554 555 static void __init create_pud_mapping(pud_t *pudp, 556 uintptr_t va, phys_addr_t pa, 557 phys_addr_t sz, pgprot_t prot) 558 { 559 pmd_t *nextp; 560 phys_addr_t next_phys; 561 uintptr_t pud_index = pud_index(va); 562 563 if (sz == PUD_SIZE) { 564 if (pud_val(pudp[pud_index]) == 0) 565 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 566 return; 567 } 568 569 if (pud_val(pudp[pud_index]) == 0) { 570 next_phys = pt_ops.alloc_pmd(va); 571 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 572 nextp = pt_ops.get_pmd_virt(next_phys); 573 memset(nextp, 0, PAGE_SIZE); 574 } else { 575 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 576 nextp = pt_ops.get_pmd_virt(next_phys); 577 } 578 579 create_pmd_mapping(nextp, va, pa, sz, prot); 580 } 581 582 static void __init create_p4d_mapping(p4d_t *p4dp, 583 uintptr_t va, phys_addr_t pa, 584 phys_addr_t sz, pgprot_t prot) 585 { 586 pud_t *nextp; 587 phys_addr_t next_phys; 588 uintptr_t p4d_index = p4d_index(va); 589 590 if (sz == P4D_SIZE) { 591 if (p4d_val(p4dp[p4d_index]) == 0) 592 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 593 return; 594 } 595 596 if (p4d_val(p4dp[p4d_index]) == 0) { 597 next_phys = pt_ops.alloc_pud(va); 598 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 599 nextp = pt_ops.get_pud_virt(next_phys); 600 memset(nextp, 0, PAGE_SIZE); 601 } else { 602 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 603 nextp = pt_ops.get_pud_virt(next_phys); 604 } 605 606 create_pud_mapping(nextp, va, pa, sz, prot); 607 } 608 609 #define pgd_next_t p4d_t 610 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 611 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 612 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 613 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 614 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 615 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 616 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 617 (pgtable_l5_enabled ? \ 618 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 619 (pgtable_l4_enabled ? \ 620 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 621 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 622 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 623 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 624 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 625 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 626 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 627 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 628 #else 629 #define pgd_next_t pte_t 630 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 631 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 632 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 633 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 634 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 635 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 636 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 637 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 638 #endif /* __PAGETABLE_PMD_FOLDED */ 639 640 void __init create_pgd_mapping(pgd_t *pgdp, 641 uintptr_t va, phys_addr_t pa, 642 phys_addr_t sz, pgprot_t prot) 643 { 644 pgd_next_t *nextp; 645 phys_addr_t next_phys; 646 uintptr_t pgd_idx = pgd_index(va); 647 648 if (sz == PGDIR_SIZE) { 649 if (pgd_val(pgdp[pgd_idx]) == 0) 650 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 651 return; 652 } 653 654 if (pgd_val(pgdp[pgd_idx]) == 0) { 655 next_phys = alloc_pgd_next(va); 656 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 657 nextp = get_pgd_next_virt(next_phys); 658 memset(nextp, 0, PAGE_SIZE); 659 } else { 660 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 661 nextp = get_pgd_next_virt(next_phys); 662 } 663 664 create_pgd_next_mapping(nextp, va, pa, sz, prot); 665 } 666 667 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va, 668 phys_addr_t size) 669 { 670 if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE) 671 return PGDIR_SIZE; 672 673 if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 674 return P4D_SIZE; 675 676 if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 677 return PUD_SIZE; 678 679 if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 680 return PMD_SIZE; 681 682 return PAGE_SIZE; 683 } 684 685 #ifdef CONFIG_XIP_KERNEL 686 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 687 extern char _xiprom[], _exiprom[], __data_loc; 688 689 /* called from head.S with MMU off */ 690 asmlinkage void __init __copy_data(void) 691 { 692 void *from = (void *)(&__data_loc); 693 void *to = (void *)CONFIG_PHYS_RAM_BASE; 694 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 695 696 memcpy(to, from, sz); 697 } 698 #endif 699 700 #ifdef CONFIG_STRICT_KERNEL_RWX 701 static __init pgprot_t pgprot_from_va(uintptr_t va) 702 { 703 if (is_va_kernel_text(va)) 704 return PAGE_KERNEL_READ_EXEC; 705 706 /* 707 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 708 * we must protect its linear mapping alias from being executed and 709 * written. 710 * And rodata section is marked readonly in mark_rodata_ro. 711 */ 712 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 713 return PAGE_KERNEL_READ; 714 715 return PAGE_KERNEL; 716 } 717 718 void mark_rodata_ro(void) 719 { 720 set_kernel_memory(__start_rodata, _data, set_memory_ro); 721 if (IS_ENABLED(CONFIG_64BIT)) 722 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 723 set_memory_ro); 724 725 debug_checkwx(); 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 /* 769 * There is a simple way to determine if 4-level is supported by the 770 * underlying hardware: establish 1:1 mapping in 4-level page table mode 771 * then read SATP to see if the configuration was taken into account 772 * meaning sv48 is supported. 773 */ 774 static __init void set_satp_mode(uintptr_t dtb_pa) 775 { 776 u64 identity_satp, hw_satp; 777 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 778 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 779 780 if (satp_mode_cmdline == SATP_MODE_57) { 781 disable_pgtable_l5(); 782 } else if (satp_mode_cmdline == SATP_MODE_48) { 783 disable_pgtable_l5(); 784 disable_pgtable_l4(); 785 return; 786 } 787 788 create_p4d_mapping(early_p4d, 789 set_satp_mode_pmd, (uintptr_t)early_pud, 790 P4D_SIZE, PAGE_TABLE); 791 create_pud_mapping(early_pud, 792 set_satp_mode_pmd, (uintptr_t)early_pmd, 793 PUD_SIZE, PAGE_TABLE); 794 /* Handle the case where set_satp_mode straddles 2 PMDs */ 795 create_pmd_mapping(early_pmd, 796 set_satp_mode_pmd, set_satp_mode_pmd, 797 PMD_SIZE, PAGE_KERNEL_EXEC); 798 create_pmd_mapping(early_pmd, 799 set_satp_mode_pmd + PMD_SIZE, 800 set_satp_mode_pmd + PMD_SIZE, 801 PMD_SIZE, PAGE_KERNEL_EXEC); 802 retry: 803 create_pgd_mapping(early_pg_dir, 804 set_satp_mode_pmd, 805 pgtable_l5_enabled ? 806 (uintptr_t)early_p4d : (uintptr_t)early_pud, 807 PGDIR_SIZE, PAGE_TABLE); 808 809 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 810 811 local_flush_tlb_all(); 812 csr_write(CSR_SATP, identity_satp); 813 hw_satp = csr_swap(CSR_SATP, 0ULL); 814 local_flush_tlb_all(); 815 816 if (hw_satp != identity_satp) { 817 if (pgtable_l5_enabled) { 818 disable_pgtable_l5(); 819 memset(early_pg_dir, 0, PAGE_SIZE); 820 goto retry; 821 } 822 disable_pgtable_l4(); 823 } 824 825 memset(early_pg_dir, 0, PAGE_SIZE); 826 memset(early_p4d, 0, PAGE_SIZE); 827 memset(early_pud, 0, PAGE_SIZE); 828 memset(early_pmd, 0, PAGE_SIZE); 829 } 830 #endif 831 832 /* 833 * setup_vm() is called from head.S with MMU-off. 834 * 835 * Following requirements should be honoured for setup_vm() to work 836 * correctly: 837 * 1) It should use PC-relative addressing for accessing kernel symbols. 838 * To achieve this we always use GCC cmodel=medany. 839 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 840 * so disable compiler instrumentation when FTRACE is enabled. 841 * 842 * Currently, the above requirements are honoured by using custom CFLAGS 843 * for init.o in mm/Makefile. 844 */ 845 846 #ifndef __riscv_cmodel_medany 847 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 848 #endif 849 850 #ifdef CONFIG_RELOCATABLE 851 extern unsigned long __rela_dyn_start, __rela_dyn_end; 852 853 static void __init relocate_kernel(void) 854 { 855 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 856 /* 857 * This holds the offset between the linked virtual address and the 858 * relocated virtual address. 859 */ 860 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 861 /* 862 * This holds the offset between kernel linked virtual address and 863 * physical address. 864 */ 865 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 866 867 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 868 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 869 Elf64_Addr relocated_addr = rela->r_addend; 870 871 if (rela->r_info != R_RISCV_RELATIVE) 872 continue; 873 874 /* 875 * Make sure to not relocate vdso symbols like rt_sigreturn 876 * which are linked from the address 0 in vmlinux since 877 * vdso symbol addresses are actually used as an offset from 878 * mm->context.vdso in VDSO_OFFSET macro. 879 */ 880 if (relocated_addr >= KERNEL_LINK_ADDR) 881 relocated_addr += reloc_offset; 882 883 *(Elf64_Addr *)addr = relocated_addr; 884 } 885 } 886 #endif /* CONFIG_RELOCATABLE */ 887 888 #ifdef CONFIG_XIP_KERNEL 889 static void __init create_kernel_page_table(pgd_t *pgdir, 890 __always_unused bool early) 891 { 892 uintptr_t va, end_va; 893 894 /* Map the flash resident part */ 895 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 896 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 897 create_pgd_mapping(pgdir, va, 898 kernel_map.xiprom + (va - kernel_map.virt_addr), 899 PMD_SIZE, PAGE_KERNEL_EXEC); 900 901 /* Map the data in RAM */ 902 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 903 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 904 create_pgd_mapping(pgdir, va, 905 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 906 PMD_SIZE, PAGE_KERNEL); 907 } 908 #else 909 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 910 { 911 uintptr_t va, end_va; 912 913 end_va = kernel_map.virt_addr + kernel_map.size; 914 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 915 create_pgd_mapping(pgdir, va, 916 kernel_map.phys_addr + (va - kernel_map.virt_addr), 917 PMD_SIZE, 918 early ? 919 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 920 } 921 #endif 922 923 /* 924 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 925 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 926 * entry. 927 */ 928 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 929 uintptr_t dtb_pa) 930 { 931 #ifndef CONFIG_BUILTIN_DTB 932 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 933 934 /* Make sure the fdt fixmap address is always aligned on PMD size */ 935 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 936 937 /* In 32-bit only, the fdt lies in its own PGD */ 938 if (!IS_ENABLED(CONFIG_64BIT)) { 939 create_pgd_mapping(early_pg_dir, fix_fdt_va, 940 pa, MAX_FDT_SIZE, PAGE_KERNEL); 941 } else { 942 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 943 pa, PMD_SIZE, PAGE_KERNEL); 944 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 945 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 946 } 947 948 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 949 #else 950 /* 951 * For 64-bit kernel, __va can't be used since it would return a linear 952 * mapping address whereas dtb_early_va will be used before 953 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 954 * kernel is mapped in the linear mapping, that makes no difference. 955 */ 956 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 957 #endif 958 959 dtb_early_pa = dtb_pa; 960 } 961 962 /* 963 * MMU is not enabled, the page tables are allocated directly using 964 * early_pmd/pud/p4d and the address returned is the physical one. 965 */ 966 static void __init pt_ops_set_early(void) 967 { 968 pt_ops.alloc_pte = alloc_pte_early; 969 pt_ops.get_pte_virt = get_pte_virt_early; 970 #ifndef __PAGETABLE_PMD_FOLDED 971 pt_ops.alloc_pmd = alloc_pmd_early; 972 pt_ops.get_pmd_virt = get_pmd_virt_early; 973 pt_ops.alloc_pud = alloc_pud_early; 974 pt_ops.get_pud_virt = get_pud_virt_early; 975 pt_ops.alloc_p4d = alloc_p4d_early; 976 pt_ops.get_p4d_virt = get_p4d_virt_early; 977 #endif 978 } 979 980 /* 981 * MMU is enabled but page table setup is not complete yet. 982 * fixmap page table alloc functions must be used as a means to temporarily 983 * map the allocated physical pages since the linear mapping does not exist yet. 984 * 985 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 986 * but it will be used as described above. 987 */ 988 static void __init pt_ops_set_fixmap(void) 989 { 990 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 991 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 992 #ifndef __PAGETABLE_PMD_FOLDED 993 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 994 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 995 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 996 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 997 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 998 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 999 #endif 1000 } 1001 1002 /* 1003 * MMU is enabled and page table setup is complete, so from now, we can use 1004 * generic page allocation functions to setup page table. 1005 */ 1006 static void __init pt_ops_set_late(void) 1007 { 1008 pt_ops.alloc_pte = alloc_pte_late; 1009 pt_ops.get_pte_virt = get_pte_virt_late; 1010 #ifndef __PAGETABLE_PMD_FOLDED 1011 pt_ops.alloc_pmd = alloc_pmd_late; 1012 pt_ops.get_pmd_virt = get_pmd_virt_late; 1013 pt_ops.alloc_pud = alloc_pud_late; 1014 pt_ops.get_pud_virt = get_pud_virt_late; 1015 pt_ops.alloc_p4d = alloc_p4d_late; 1016 pt_ops.get_p4d_virt = get_p4d_virt_late; 1017 #endif 1018 } 1019 1020 #ifdef CONFIG_RANDOMIZE_BASE 1021 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1022 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1023 1024 static int __init print_nokaslr(char *p) 1025 { 1026 pr_info("Disabled KASLR"); 1027 return 0; 1028 } 1029 early_param("nokaslr", print_nokaslr); 1030 1031 unsigned long kaslr_offset(void) 1032 { 1033 return kernel_map.virt_offset; 1034 } 1035 #endif 1036 1037 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1038 { 1039 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1040 1041 #ifdef CONFIG_RANDOMIZE_BASE 1042 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1043 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1044 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1045 u32 nr_pos; 1046 1047 /* 1048 * Compute the number of positions available: we are limited 1049 * by the early page table that only has one PUD and we must 1050 * be aligned on PMD_SIZE. 1051 */ 1052 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1053 1054 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1055 } 1056 #endif 1057 1058 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1059 1060 #ifdef CONFIG_XIP_KERNEL 1061 #ifdef CONFIG_64BIT 1062 kernel_map.page_offset = PAGE_OFFSET_L3; 1063 #else 1064 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1065 #endif 1066 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1067 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1068 1069 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1070 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1071 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1072 1073 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1074 #else 1075 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1076 kernel_map.phys_addr = (uintptr_t)(&_start); 1077 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1078 #endif 1079 1080 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1081 set_satp_mode(dtb_pa); 1082 #endif 1083 1084 /* 1085 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1086 * where we have the system memory layout: this allows us to align 1087 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1088 * for the linear mapping. This is only possible because the kernel 1089 * mapping lies outside the linear mapping. 1090 * In 32-bit however, as the kernel resides in the linear mapping, 1091 * setup_vm_final can not change the mapping established here, 1092 * otherwise the same kernel addresses would get mapped to different 1093 * physical addresses (if the start of dram is different from the 1094 * kernel physical address start). 1095 */ 1096 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1097 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1098 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1099 1100 /* 1101 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1102 * kernel, whereas for 64-bit kernel, the end of the virtual address 1103 * space is occupied by the modules/BPF/kernel mappings which reduces 1104 * the available size of the linear mapping. 1105 */ 1106 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1107 1108 /* Sanity check alignment and size */ 1109 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1110 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1111 1112 #ifdef CONFIG_64BIT 1113 /* 1114 * The last 4K bytes of the addressable memory can not be mapped because 1115 * of IS_ERR_VALUE macro. 1116 */ 1117 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1118 #endif 1119 1120 #ifdef CONFIG_RELOCATABLE 1121 /* 1122 * Early page table uses only one PUD, which makes it possible 1123 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1124 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1125 * since a part of the kernel would not get mapped. 1126 */ 1127 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1128 relocate_kernel(); 1129 #endif 1130 1131 apply_early_boot_alternatives(); 1132 pt_ops_set_early(); 1133 1134 /* Setup early PGD for fixmap */ 1135 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1136 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1137 1138 #ifndef __PAGETABLE_PMD_FOLDED 1139 /* Setup fixmap P4D and PUD */ 1140 if (pgtable_l5_enabled) 1141 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1142 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1143 /* Setup fixmap PUD and PMD */ 1144 if (pgtable_l4_enabled) 1145 create_pud_mapping(fixmap_pud, FIXADDR_START, 1146 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1147 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1148 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1149 /* Setup trampoline PGD and PMD */ 1150 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1151 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1152 if (pgtable_l5_enabled) 1153 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1154 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1155 if (pgtable_l4_enabled) 1156 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1157 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1158 #ifdef CONFIG_XIP_KERNEL 1159 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1160 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1161 #else 1162 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1163 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1164 #endif 1165 #else 1166 /* Setup trampoline PGD */ 1167 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1168 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1169 #endif 1170 1171 /* 1172 * Setup early PGD covering entire kernel which will allow 1173 * us to reach paging_init(). We map all memory banks later 1174 * in setup_vm_final() below. 1175 */ 1176 create_kernel_page_table(early_pg_dir, true); 1177 1178 /* Setup early mapping for FDT early scan */ 1179 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1180 1181 /* 1182 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1183 * range can not span multiple pmds. 1184 */ 1185 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1186 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1187 1188 #ifndef __PAGETABLE_PMD_FOLDED 1189 /* 1190 * Early ioremap fixmap is already created as it lies within first 2MB 1191 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1192 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1193 * the user if not. 1194 */ 1195 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1196 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1197 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1198 WARN_ON(1); 1199 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1200 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1201 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1202 fix_to_virt(FIX_BTMAP_BEGIN)); 1203 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1204 fix_to_virt(FIX_BTMAP_END)); 1205 1206 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1207 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1208 } 1209 #endif 1210 1211 pt_ops_set_fixmap(); 1212 } 1213 1214 static void __init create_linear_mapping_range(phys_addr_t start, 1215 phys_addr_t end, 1216 uintptr_t fixed_map_size) 1217 { 1218 phys_addr_t pa; 1219 uintptr_t va, map_size; 1220 1221 for (pa = start; pa < end; pa += map_size) { 1222 va = (uintptr_t)__va(pa); 1223 map_size = fixed_map_size ? fixed_map_size : 1224 best_map_size(pa, va, end - pa); 1225 1226 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1227 pgprot_from_va(va)); 1228 } 1229 } 1230 1231 static void __init create_linear_mapping_page_table(void) 1232 { 1233 phys_addr_t start, end; 1234 phys_addr_t kfence_pool __maybe_unused; 1235 u64 i; 1236 1237 #ifdef CONFIG_STRICT_KERNEL_RWX 1238 phys_addr_t ktext_start = __pa_symbol(_start); 1239 phys_addr_t ktext_size = __init_data_begin - _start; 1240 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1241 phys_addr_t krodata_size = _data - __start_rodata; 1242 1243 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1244 memblock_mark_nomap(ktext_start, ktext_size); 1245 memblock_mark_nomap(krodata_start, krodata_size); 1246 #endif 1247 1248 #ifdef CONFIG_KFENCE 1249 /* 1250 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1251 * before we setup the linear mapping so that we avoid using hugepages 1252 * for this region. 1253 */ 1254 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1255 BUG_ON(!kfence_pool); 1256 1257 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1258 __kfence_pool = __va(kfence_pool); 1259 #endif 1260 1261 /* Map all memory banks in the linear mapping */ 1262 for_each_mem_range(i, &start, &end) { 1263 if (start >= end) 1264 break; 1265 if (start <= __pa(PAGE_OFFSET) && 1266 __pa(PAGE_OFFSET) < end) 1267 start = __pa(PAGE_OFFSET); 1268 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1269 end = __pa(PAGE_OFFSET) + memory_limit; 1270 1271 create_linear_mapping_range(start, end, 0); 1272 } 1273 1274 #ifdef CONFIG_STRICT_KERNEL_RWX 1275 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0); 1276 create_linear_mapping_range(krodata_start, 1277 krodata_start + krodata_size, 0); 1278 1279 memblock_clear_nomap(ktext_start, ktext_size); 1280 memblock_clear_nomap(krodata_start, krodata_size); 1281 #endif 1282 1283 #ifdef CONFIG_KFENCE 1284 create_linear_mapping_range(kfence_pool, 1285 kfence_pool + KFENCE_POOL_SIZE, 1286 PAGE_SIZE); 1287 1288 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1289 #endif 1290 } 1291 1292 static void __init setup_vm_final(void) 1293 { 1294 /* Setup swapper PGD for fixmap */ 1295 #if !defined(CONFIG_64BIT) 1296 /* 1297 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1298 * directly in swapper_pg_dir in addition to the pgd entry that points 1299 * to fixmap_pte. 1300 */ 1301 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1302 1303 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1304 #endif 1305 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1306 __pa_symbol(fixmap_pgd_next), 1307 PGDIR_SIZE, PAGE_TABLE); 1308 1309 /* Map the linear mapping */ 1310 create_linear_mapping_page_table(); 1311 1312 /* Map the kernel */ 1313 if (IS_ENABLED(CONFIG_64BIT)) 1314 create_kernel_page_table(swapper_pg_dir, false); 1315 1316 #ifdef CONFIG_KASAN 1317 kasan_swapper_init(); 1318 #endif 1319 1320 /* Clear fixmap PTE and PMD mappings */ 1321 clear_fixmap(FIX_PTE); 1322 clear_fixmap(FIX_PMD); 1323 clear_fixmap(FIX_PUD); 1324 clear_fixmap(FIX_P4D); 1325 1326 /* Move to swapper page table */ 1327 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1328 local_flush_tlb_all(); 1329 1330 pt_ops_set_late(); 1331 } 1332 #else 1333 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1334 { 1335 dtb_early_va = (void *)dtb_pa; 1336 dtb_early_pa = dtb_pa; 1337 } 1338 1339 static inline void setup_vm_final(void) 1340 { 1341 } 1342 #endif /* CONFIG_MMU */ 1343 1344 /* Reserve 128M low memory by default for swiotlb buffer */ 1345 #define DEFAULT_CRASH_KERNEL_LOW_SIZE (128UL << 20) 1346 1347 static int __init reserve_crashkernel_low(unsigned long long low_size) 1348 { 1349 unsigned long long low_base; 1350 1351 low_base = memblock_phys_alloc_range(low_size, PMD_SIZE, 0, dma32_phys_limit); 1352 if (!low_base) { 1353 pr_err("cannot allocate crashkernel low memory (size:0x%llx).\n", low_size); 1354 return -ENOMEM; 1355 } 1356 1357 pr_info("crashkernel low memory reserved: 0x%016llx - 0x%016llx (%lld MB)\n", 1358 low_base, low_base + low_size, low_size >> 20); 1359 1360 crashk_low_res.start = low_base; 1361 crashk_low_res.end = low_base + low_size - 1; 1362 1363 return 0; 1364 } 1365 1366 /* 1367 * reserve_crashkernel() - reserves memory for crash kernel 1368 * 1369 * This function reserves memory area given in "crashkernel=" kernel command 1370 * line parameter. The memory reserved is used by dump capture kernel when 1371 * primary kernel is crashing. 1372 */ 1373 static void __init reserve_crashkernel(void) 1374 { 1375 unsigned long long crash_base = 0; 1376 unsigned long long crash_size = 0; 1377 unsigned long long crash_low_size = 0; 1378 unsigned long search_start = memblock_start_of_DRAM(); 1379 unsigned long search_end = (unsigned long)dma32_phys_limit; 1380 char *cmdline = boot_command_line; 1381 bool fixed_base = false; 1382 bool high = false; 1383 1384 int ret = 0; 1385 1386 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 1387 return; 1388 /* 1389 * Don't reserve a region for a crash kernel on a crash kernel 1390 * since it doesn't make much sense and we have limited memory 1391 * resources. 1392 */ 1393 if (is_kdump_kernel()) { 1394 pr_info("crashkernel: ignoring reservation request\n"); 1395 return; 1396 } 1397 1398 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 1399 &crash_size, &crash_base); 1400 if (ret == -ENOENT) { 1401 /* Fallback to crashkernel=X,[high,low] */ 1402 ret = parse_crashkernel_high(cmdline, 0, &crash_size, &crash_base); 1403 if (ret || !crash_size) 1404 return; 1405 1406 /* 1407 * crashkernel=Y,low is valid only when crashkernel=X,high 1408 * is passed. 1409 */ 1410 ret = parse_crashkernel_low(cmdline, 0, &crash_low_size, &crash_base); 1411 if (ret == -ENOENT) 1412 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE; 1413 else if (ret) 1414 return; 1415 1416 search_start = (unsigned long)dma32_phys_limit; 1417 search_end = memblock_end_of_DRAM(); 1418 high = true; 1419 } else if (ret || !crash_size) { 1420 /* Invalid argument value specified */ 1421 return; 1422 } 1423 1424 crash_size = PAGE_ALIGN(crash_size); 1425 1426 if (crash_base) { 1427 fixed_base = true; 1428 search_start = crash_base; 1429 search_end = crash_base + crash_size; 1430 } 1431 1432 /* 1433 * Current riscv boot protocol requires 2MB alignment for 1434 * RV64 and 4MB alignment for RV32 (hugepage size) 1435 * 1436 * Try to alloc from 32bit addressible physical memory so that 1437 * swiotlb can work on the crash kernel. 1438 */ 1439 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1440 search_start, search_end); 1441 if (crash_base == 0) { 1442 /* 1443 * For crashkernel=size[KMG]@offset[KMG], print out failure 1444 * message if can't reserve the specified region. 1445 */ 1446 if (fixed_base) { 1447 pr_warn("crashkernel: allocating failed with given size@offset\n"); 1448 return; 1449 } 1450 1451 if (high) { 1452 /* 1453 * For crashkernel=size[KMG],high, if the first attempt was 1454 * for high memory, fall back to low memory. 1455 */ 1456 search_start = memblock_start_of_DRAM(); 1457 search_end = (unsigned long)dma32_phys_limit; 1458 } else { 1459 /* 1460 * For crashkernel=size[KMG], if the first attempt was for 1461 * low memory, fall back to high memory, the minimum required 1462 * low memory will be reserved later. 1463 */ 1464 search_start = (unsigned long)dma32_phys_limit; 1465 search_end = memblock_end_of_DRAM(); 1466 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE; 1467 } 1468 1469 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1470 search_start, search_end); 1471 if (crash_base == 0) { 1472 pr_warn("crashkernel: couldn't allocate %lldKB\n", 1473 crash_size >> 10); 1474 return; 1475 } 1476 } 1477 1478 if ((crash_base >= dma32_phys_limit) && crash_low_size && 1479 reserve_crashkernel_low(crash_low_size)) { 1480 memblock_phys_free(crash_base, crash_size); 1481 return; 1482 } 1483 1484 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 1485 crash_base, crash_base + crash_size, crash_size >> 20); 1486 1487 crashk_res.start = crash_base; 1488 crashk_res.end = crash_base + crash_size - 1; 1489 } 1490 1491 void __init paging_init(void) 1492 { 1493 setup_bootmem(); 1494 setup_vm_final(); 1495 1496 /* Depend on that Linear Mapping is ready */ 1497 memblock_allow_resize(); 1498 } 1499 1500 void __init misc_mem_init(void) 1501 { 1502 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1503 arch_numa_init(); 1504 sparse_init(); 1505 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1506 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1507 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1508 #endif 1509 zone_sizes_init(); 1510 reserve_crashkernel(); 1511 memblock_dump_all(); 1512 } 1513 1514 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1515 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1516 struct vmem_altmap *altmap) 1517 { 1518 return vmemmap_populate_basepages(start, end, node, NULL); 1519 } 1520 #endif 1521 1522 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1523 /* 1524 * Pre-allocates page-table pages for a specific area in the kernel 1525 * page-table. Only the level which needs to be synchronized between 1526 * all page-tables is allocated because the synchronization can be 1527 * expensive. 1528 */ 1529 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1530 const char *area) 1531 { 1532 unsigned long addr; 1533 const char *lvl; 1534 1535 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1536 pgd_t *pgd = pgd_offset_k(addr); 1537 p4d_t *p4d; 1538 pud_t *pud; 1539 pmd_t *pmd; 1540 1541 lvl = "p4d"; 1542 p4d = p4d_alloc(&init_mm, pgd, addr); 1543 if (!p4d) 1544 goto failed; 1545 1546 if (pgtable_l5_enabled) 1547 continue; 1548 1549 lvl = "pud"; 1550 pud = pud_alloc(&init_mm, p4d, addr); 1551 if (!pud) 1552 goto failed; 1553 1554 if (pgtable_l4_enabled) 1555 continue; 1556 1557 lvl = "pmd"; 1558 pmd = pmd_alloc(&init_mm, pud, addr); 1559 if (!pmd) 1560 goto failed; 1561 } 1562 return; 1563 1564 failed: 1565 /* 1566 * The pages have to be there now or they will be missing in 1567 * process page-tables later. 1568 */ 1569 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1570 } 1571 1572 void __init pgtable_cache_init(void) 1573 { 1574 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1575 if (IS_ENABLED(CONFIG_MODULES)) 1576 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1577 } 1578 #endif 1579