1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2012 Regents of the University of California 4 * Copyright (C) 2019 Western Digital Corporation or its affiliates. 5 * Copyright (C) 2020 FORTH-ICS/CARV 6 * Nick Kossifidis <mick@ics.forth.gr> 7 */ 8 9 #include <linux/init.h> 10 #include <linux/mm.h> 11 #include <linux/memblock.h> 12 #include <linux/initrd.h> 13 #include <linux/swap.h> 14 #include <linux/swiotlb.h> 15 #include <linux/sizes.h> 16 #include <linux/of_fdt.h> 17 #include <linux/of_reserved_mem.h> 18 #include <linux/libfdt.h> 19 #include <linux/set_memory.h> 20 #include <linux/dma-map-ops.h> 21 #include <linux/crash_dump.h> 22 #include <linux/hugetlb.h> 23 #ifdef CONFIG_RELOCATABLE 24 #include <linux/elf.h> 25 #endif 26 #include <linux/kfence.h> 27 #include <linux/execmem.h> 28 29 #include <asm/fixmap.h> 30 #include <asm/io.h> 31 #include <asm/numa.h> 32 #include <asm/pgtable.h> 33 #include <asm/sections.h> 34 #include <asm/soc.h> 35 #include <asm/tlbflush.h> 36 37 #include "../kernel/head.h" 38 39 struct kernel_mapping kernel_map __ro_after_init; 40 EXPORT_SYMBOL(kernel_map); 41 #ifdef CONFIG_XIP_KERNEL 42 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 43 #endif 44 45 #ifdef CONFIG_64BIT 46 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 47 #else 48 u64 satp_mode __ro_after_init = SATP_MODE_32; 49 #endif 50 EXPORT_SYMBOL(satp_mode); 51 52 #ifdef CONFIG_64BIT 53 bool pgtable_l4_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 54 bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 55 EXPORT_SYMBOL(pgtable_l4_enabled); 56 EXPORT_SYMBOL(pgtable_l5_enabled); 57 #endif 58 59 phys_addr_t phys_ram_base __ro_after_init; 60 EXPORT_SYMBOL(phys_ram_base); 61 62 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 63 __page_aligned_bss; 64 EXPORT_SYMBOL(empty_zero_page); 65 66 extern char _start[]; 67 void *_dtb_early_va __initdata; 68 uintptr_t _dtb_early_pa __initdata; 69 70 phys_addr_t dma32_phys_limit __initdata; 71 72 static void __init zone_sizes_init(void) 73 { 74 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 75 76 #ifdef CONFIG_ZONE_DMA32 77 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 78 #endif 79 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 80 81 free_area_init(max_zone_pfns); 82 } 83 84 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 85 86 #define LOG2_SZ_1K ilog2(SZ_1K) 87 #define LOG2_SZ_1M ilog2(SZ_1M) 88 #define LOG2_SZ_1G ilog2(SZ_1G) 89 #define LOG2_SZ_1T ilog2(SZ_1T) 90 91 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 92 { 93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 94 (((t) - (b)) >> LOG2_SZ_1K)); 95 } 96 97 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 98 { 99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 100 (((t) - (b)) >> LOG2_SZ_1M)); 101 } 102 103 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 104 { 105 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 106 (((t) - (b)) >> LOG2_SZ_1G)); 107 } 108 109 #ifdef CONFIG_64BIT 110 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 111 { 112 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 113 (((t) - (b)) >> LOG2_SZ_1T)); 114 } 115 #else 116 #define print_mlt(n, b, t) do {} while (0) 117 #endif 118 119 static inline void print_ml(char *name, unsigned long b, unsigned long t) 120 { 121 unsigned long diff = t - b; 122 123 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 124 print_mlt(name, b, t); 125 else if ((diff >> LOG2_SZ_1G) >= 10) 126 print_mlg(name, b, t); 127 else if ((diff >> LOG2_SZ_1M) >= 10) 128 print_mlm(name, b, t); 129 else 130 print_mlk(name, b, t); 131 } 132 133 static void __init print_vm_layout(void) 134 { 135 pr_notice("Virtual kernel memory layout:\n"); 136 print_ml("fixmap", (unsigned long)FIXADDR_START, 137 (unsigned long)FIXADDR_TOP); 138 print_ml("pci io", (unsigned long)PCI_IO_START, 139 (unsigned long)PCI_IO_END); 140 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 141 (unsigned long)VMEMMAP_END); 142 print_ml("vmalloc", (unsigned long)VMALLOC_START, 143 (unsigned long)VMALLOC_END); 144 #ifdef CONFIG_64BIT 145 print_ml("modules", (unsigned long)MODULES_VADDR, 146 (unsigned long)MODULES_END); 147 #endif 148 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 149 (unsigned long)high_memory); 150 if (IS_ENABLED(CONFIG_64BIT)) { 151 #ifdef CONFIG_KASAN 152 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 153 #endif 154 155 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 156 (unsigned long)ADDRESS_SPACE_END); 157 } 158 } 159 #else 160 static void print_vm_layout(void) { } 161 #endif /* CONFIG_DEBUG_VM */ 162 163 void __init mem_init(void) 164 { 165 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 166 #ifdef CONFIG_FLATMEM 167 BUG_ON(!mem_map); 168 #endif /* CONFIG_FLATMEM */ 169 170 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 171 dma_cache_alignment != 1) { 172 /* 173 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 174 * buffer per 1GB of RAM for kmalloc() bouncing on 175 * non-coherent platforms. 176 */ 177 unsigned long size = 178 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 179 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 180 swiotlb = true; 181 } 182 183 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 184 memblock_free_all(); 185 186 print_vm_layout(); 187 } 188 189 /* Limit the memory size via mem. */ 190 static phys_addr_t memory_limit; 191 #ifdef CONFIG_XIP_KERNEL 192 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 193 #endif /* CONFIG_XIP_KERNEL */ 194 195 static int __init early_mem(char *p) 196 { 197 u64 size; 198 199 if (!p) 200 return 1; 201 202 size = memparse(p, &p) & PAGE_MASK; 203 memory_limit = min_t(u64, size, memory_limit); 204 205 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 206 207 return 0; 208 } 209 early_param("mem", early_mem); 210 211 static void __init setup_bootmem(void) 212 { 213 phys_addr_t vmlinux_end = __pa_symbol(&_end); 214 phys_addr_t max_mapped_addr; 215 phys_addr_t phys_ram_end, vmlinux_start; 216 217 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 218 vmlinux_start = __pa_symbol(&_sdata); 219 else 220 vmlinux_start = __pa_symbol(&_start); 221 222 memblock_enforce_memory_limit(memory_limit); 223 224 /* 225 * Make sure we align the reservation on PMD_SIZE since we will 226 * map the kernel in the linear mapping as read-only: we do not want 227 * any allocation to happen between _end and the next pmd aligned page. 228 */ 229 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 230 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 231 /* 232 * Reserve from the start of the kernel to the end of the kernel 233 */ 234 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 235 236 phys_ram_end = memblock_end_of_DRAM(); 237 238 /* 239 * Make sure we align the start of the memory on a PMD boundary so that 240 * at worst, we map the linear mapping with PMD mappings. 241 */ 242 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 243 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 244 245 /* 246 * In 64-bit, any use of __va/__pa before this point is wrong as we 247 * did not know the start of DRAM before. 248 */ 249 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 250 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 251 252 /* 253 * memblock allocator is not aware of the fact that last 4K bytes of 254 * the addressable memory can not be mapped because of IS_ERR_VALUE 255 * macro. Make sure that last 4k bytes are not usable by memblock 256 * if end of dram is equal to maximum addressable memory. For 64-bit 257 * kernel, this problem can't happen here as the end of the virtual 258 * address space is occupied by the kernel mapping then this check must 259 * be done as soon as the kernel mapping base address is determined. 260 */ 261 if (!IS_ENABLED(CONFIG_64BIT)) { 262 max_mapped_addr = __pa(~(ulong)0); 263 if (max_mapped_addr == (phys_ram_end - 1)) 264 memblock_set_current_limit(max_mapped_addr - 4096); 265 } 266 267 min_low_pfn = PFN_UP(phys_ram_base); 268 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 269 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 270 271 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 272 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 273 274 reserve_initrd_mem(); 275 276 /* 277 * No allocation should be done before reserving the memory as defined 278 * in the device tree, otherwise the allocation could end up in a 279 * reserved region. 280 */ 281 early_init_fdt_scan_reserved_mem(); 282 283 /* 284 * If DTB is built in, no need to reserve its memblock. 285 * Otherwise, do reserve it but avoid using 286 * early_init_fdt_reserve_self() since __pa() does 287 * not work for DTB pointers that are fixmap addresses 288 */ 289 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 290 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 291 292 dma_contiguous_reserve(dma32_phys_limit); 293 if (IS_ENABLED(CONFIG_64BIT)) 294 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 295 } 296 297 #ifdef CONFIG_MMU 298 struct pt_alloc_ops pt_ops __initdata; 299 300 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 301 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 302 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 303 304 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 305 306 #ifdef CONFIG_XIP_KERNEL 307 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 308 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 309 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 310 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 311 #endif /* CONFIG_XIP_KERNEL */ 312 313 static const pgprot_t protection_map[16] = { 314 [VM_NONE] = PAGE_NONE, 315 [VM_READ] = PAGE_READ, 316 [VM_WRITE] = PAGE_COPY, 317 [VM_WRITE | VM_READ] = PAGE_COPY, 318 [VM_EXEC] = PAGE_EXEC, 319 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 320 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 321 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 322 [VM_SHARED] = PAGE_NONE, 323 [VM_SHARED | VM_READ] = PAGE_READ, 324 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 325 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 326 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 327 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 328 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 329 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 330 }; 331 DECLARE_VM_GET_PAGE_PROT 332 333 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 334 { 335 unsigned long addr = __fix_to_virt(idx); 336 pte_t *ptep; 337 338 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 339 340 ptep = &fixmap_pte[pte_index(addr)]; 341 342 if (pgprot_val(prot)) 343 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 344 else 345 pte_clear(&init_mm, addr, ptep); 346 local_flush_tlb_page(addr); 347 } 348 349 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 350 { 351 return (pte_t *)((uintptr_t)pa); 352 } 353 354 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 355 { 356 clear_fixmap(FIX_PTE); 357 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 358 } 359 360 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 361 { 362 return (pte_t *) __va(pa); 363 } 364 365 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 366 { 367 /* 368 * We only create PMD or PGD early mappings so we 369 * should never reach here with MMU disabled. 370 */ 371 BUG(); 372 } 373 374 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 375 { 376 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 377 } 378 379 static phys_addr_t __init alloc_pte_late(uintptr_t va) 380 { 381 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 382 383 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 384 return __pa((pte_t *)ptdesc_address(ptdesc)); 385 } 386 387 static void __init create_pte_mapping(pte_t *ptep, 388 uintptr_t va, phys_addr_t pa, 389 phys_addr_t sz, pgprot_t prot) 390 { 391 uintptr_t pte_idx = pte_index(va); 392 393 BUG_ON(sz != PAGE_SIZE); 394 395 if (pte_none(ptep[pte_idx])) 396 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 397 } 398 399 #ifndef __PAGETABLE_PMD_FOLDED 400 401 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 402 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 403 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 404 405 #ifdef CONFIG_XIP_KERNEL 406 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 407 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 408 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 409 #endif /* CONFIG_XIP_KERNEL */ 410 411 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 412 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 413 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 414 415 #ifdef CONFIG_XIP_KERNEL 416 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 417 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 418 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 419 #endif /* CONFIG_XIP_KERNEL */ 420 421 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 422 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 423 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 424 425 #ifdef CONFIG_XIP_KERNEL 426 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 427 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 428 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 429 #endif /* CONFIG_XIP_KERNEL */ 430 431 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 432 { 433 /* Before MMU is enabled */ 434 return (pmd_t *)((uintptr_t)pa); 435 } 436 437 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 438 { 439 clear_fixmap(FIX_PMD); 440 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 441 } 442 443 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 444 { 445 return (pmd_t *) __va(pa); 446 } 447 448 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 449 { 450 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 451 452 return (uintptr_t)early_pmd; 453 } 454 455 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 456 { 457 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 458 } 459 460 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 461 { 462 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 463 464 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 465 return __pa((pmd_t *)ptdesc_address(ptdesc)); 466 } 467 468 static void __init create_pmd_mapping(pmd_t *pmdp, 469 uintptr_t va, phys_addr_t pa, 470 phys_addr_t sz, pgprot_t prot) 471 { 472 pte_t *ptep; 473 phys_addr_t pte_phys; 474 uintptr_t pmd_idx = pmd_index(va); 475 476 if (sz == PMD_SIZE) { 477 if (pmd_none(pmdp[pmd_idx])) 478 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 479 return; 480 } 481 482 if (pmd_none(pmdp[pmd_idx])) { 483 pte_phys = pt_ops.alloc_pte(va); 484 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 485 ptep = pt_ops.get_pte_virt(pte_phys); 486 memset(ptep, 0, PAGE_SIZE); 487 } else { 488 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 489 ptep = pt_ops.get_pte_virt(pte_phys); 490 } 491 492 create_pte_mapping(ptep, va, pa, sz, prot); 493 } 494 495 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 496 { 497 return (pud_t *)((uintptr_t)pa); 498 } 499 500 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 501 { 502 clear_fixmap(FIX_PUD); 503 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 504 } 505 506 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 507 { 508 return (pud_t *)__va(pa); 509 } 510 511 static phys_addr_t __init alloc_pud_early(uintptr_t va) 512 { 513 /* Only one PUD is available for early mapping */ 514 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 515 516 return (uintptr_t)early_pud; 517 } 518 519 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 520 { 521 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 522 } 523 524 static phys_addr_t alloc_pud_late(uintptr_t va) 525 { 526 unsigned long vaddr; 527 528 vaddr = __get_free_page(GFP_KERNEL); 529 BUG_ON(!vaddr); 530 return __pa(vaddr); 531 } 532 533 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 534 { 535 return (p4d_t *)((uintptr_t)pa); 536 } 537 538 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 539 { 540 clear_fixmap(FIX_P4D); 541 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 542 } 543 544 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 545 { 546 return (p4d_t *)__va(pa); 547 } 548 549 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 550 { 551 /* Only one P4D is available for early mapping */ 552 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 553 554 return (uintptr_t)early_p4d; 555 } 556 557 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 558 { 559 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 560 } 561 562 static phys_addr_t alloc_p4d_late(uintptr_t va) 563 { 564 unsigned long vaddr; 565 566 vaddr = __get_free_page(GFP_KERNEL); 567 BUG_ON(!vaddr); 568 return __pa(vaddr); 569 } 570 571 static void __init create_pud_mapping(pud_t *pudp, 572 uintptr_t va, phys_addr_t pa, 573 phys_addr_t sz, pgprot_t prot) 574 { 575 pmd_t *nextp; 576 phys_addr_t next_phys; 577 uintptr_t pud_index = pud_index(va); 578 579 if (sz == PUD_SIZE) { 580 if (pud_val(pudp[pud_index]) == 0) 581 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 582 return; 583 } 584 585 if (pud_val(pudp[pud_index]) == 0) { 586 next_phys = pt_ops.alloc_pmd(va); 587 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 588 nextp = pt_ops.get_pmd_virt(next_phys); 589 memset(nextp, 0, PAGE_SIZE); 590 } else { 591 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 592 nextp = pt_ops.get_pmd_virt(next_phys); 593 } 594 595 create_pmd_mapping(nextp, va, pa, sz, prot); 596 } 597 598 static void __init create_p4d_mapping(p4d_t *p4dp, 599 uintptr_t va, phys_addr_t pa, 600 phys_addr_t sz, pgprot_t prot) 601 { 602 pud_t *nextp; 603 phys_addr_t next_phys; 604 uintptr_t p4d_index = p4d_index(va); 605 606 if (sz == P4D_SIZE) { 607 if (p4d_val(p4dp[p4d_index]) == 0) 608 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 609 return; 610 } 611 612 if (p4d_val(p4dp[p4d_index]) == 0) { 613 next_phys = pt_ops.alloc_pud(va); 614 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 615 nextp = pt_ops.get_pud_virt(next_phys); 616 memset(nextp, 0, PAGE_SIZE); 617 } else { 618 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 619 nextp = pt_ops.get_pud_virt(next_phys); 620 } 621 622 create_pud_mapping(nextp, va, pa, sz, prot); 623 } 624 625 #define pgd_next_t p4d_t 626 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 627 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 628 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 629 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 630 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 631 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 632 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 633 (pgtable_l5_enabled ? \ 634 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 635 (pgtable_l4_enabled ? \ 636 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 637 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 638 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 639 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 640 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 641 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 642 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 643 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 644 #else 645 #define pgd_next_t pte_t 646 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 647 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 648 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 649 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 650 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 651 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 652 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 653 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 654 #endif /* __PAGETABLE_PMD_FOLDED */ 655 656 void __init create_pgd_mapping(pgd_t *pgdp, 657 uintptr_t va, phys_addr_t pa, 658 phys_addr_t sz, pgprot_t prot) 659 { 660 pgd_next_t *nextp; 661 phys_addr_t next_phys; 662 uintptr_t pgd_idx = pgd_index(va); 663 664 if (sz == PGDIR_SIZE) { 665 if (pgd_val(pgdp[pgd_idx]) == 0) 666 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 667 return; 668 } 669 670 if (pgd_val(pgdp[pgd_idx]) == 0) { 671 next_phys = alloc_pgd_next(va); 672 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 673 nextp = get_pgd_next_virt(next_phys); 674 memset(nextp, 0, PAGE_SIZE); 675 } else { 676 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 677 nextp = get_pgd_next_virt(next_phys); 678 } 679 680 create_pgd_next_mapping(nextp, va, pa, sz, prot); 681 } 682 683 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va, 684 phys_addr_t size) 685 { 686 if (pgtable_l5_enabled && 687 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 688 return P4D_SIZE; 689 690 if (pgtable_l4_enabled && 691 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 692 return PUD_SIZE; 693 694 if (IS_ENABLED(CONFIG_64BIT) && 695 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 696 return PMD_SIZE; 697 698 return PAGE_SIZE; 699 } 700 701 #ifdef CONFIG_XIP_KERNEL 702 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 703 extern char _xiprom[], _exiprom[], __data_loc; 704 705 /* called from head.S with MMU off */ 706 asmlinkage void __init __copy_data(void) 707 { 708 void *from = (void *)(&__data_loc); 709 void *to = (void *)CONFIG_PHYS_RAM_BASE; 710 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 711 712 memcpy(to, from, sz); 713 } 714 #endif 715 716 #ifdef CONFIG_STRICT_KERNEL_RWX 717 static __init pgprot_t pgprot_from_va(uintptr_t va) 718 { 719 if (is_va_kernel_text(va)) 720 return PAGE_KERNEL_READ_EXEC; 721 722 /* 723 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 724 * we must protect its linear mapping alias from being executed and 725 * written. 726 * And rodata section is marked readonly in mark_rodata_ro. 727 */ 728 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 729 return PAGE_KERNEL_READ; 730 731 return PAGE_KERNEL; 732 } 733 734 void mark_rodata_ro(void) 735 { 736 set_kernel_memory(__start_rodata, _data, set_memory_ro); 737 if (IS_ENABLED(CONFIG_64BIT)) 738 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 739 set_memory_ro); 740 } 741 #else 742 static __init pgprot_t pgprot_from_va(uintptr_t va) 743 { 744 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 745 return PAGE_KERNEL; 746 747 return PAGE_KERNEL_EXEC; 748 } 749 #endif /* CONFIG_STRICT_KERNEL_RWX */ 750 751 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 752 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 753 754 static void __init disable_pgtable_l5(void) 755 { 756 pgtable_l5_enabled = false; 757 kernel_map.page_offset = PAGE_OFFSET_L4; 758 satp_mode = SATP_MODE_48; 759 } 760 761 static void __init disable_pgtable_l4(void) 762 { 763 pgtable_l4_enabled = false; 764 kernel_map.page_offset = PAGE_OFFSET_L3; 765 satp_mode = SATP_MODE_39; 766 } 767 768 static int __init print_no4lvl(char *p) 769 { 770 pr_info("Disabled 4-level and 5-level paging"); 771 return 0; 772 } 773 early_param("no4lvl", print_no4lvl); 774 775 static int __init print_no5lvl(char *p) 776 { 777 pr_info("Disabled 5-level paging"); 778 return 0; 779 } 780 early_param("no5lvl", print_no5lvl); 781 782 static void __init set_mmap_rnd_bits_max(void) 783 { 784 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 785 } 786 787 /* 788 * There is a simple way to determine if 4-level is supported by the 789 * underlying hardware: establish 1:1 mapping in 4-level page table mode 790 * then read SATP to see if the configuration was taken into account 791 * meaning sv48 is supported. 792 */ 793 static __init void set_satp_mode(uintptr_t dtb_pa) 794 { 795 u64 identity_satp, hw_satp; 796 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 797 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 798 799 if (satp_mode_cmdline == SATP_MODE_57) { 800 disable_pgtable_l5(); 801 } else if (satp_mode_cmdline == SATP_MODE_48) { 802 disable_pgtable_l5(); 803 disable_pgtable_l4(); 804 return; 805 } 806 807 create_p4d_mapping(early_p4d, 808 set_satp_mode_pmd, (uintptr_t)early_pud, 809 P4D_SIZE, PAGE_TABLE); 810 create_pud_mapping(early_pud, 811 set_satp_mode_pmd, (uintptr_t)early_pmd, 812 PUD_SIZE, PAGE_TABLE); 813 /* Handle the case where set_satp_mode straddles 2 PMDs */ 814 create_pmd_mapping(early_pmd, 815 set_satp_mode_pmd, set_satp_mode_pmd, 816 PMD_SIZE, PAGE_KERNEL_EXEC); 817 create_pmd_mapping(early_pmd, 818 set_satp_mode_pmd + PMD_SIZE, 819 set_satp_mode_pmd + PMD_SIZE, 820 PMD_SIZE, PAGE_KERNEL_EXEC); 821 retry: 822 create_pgd_mapping(early_pg_dir, 823 set_satp_mode_pmd, 824 pgtable_l5_enabled ? 825 (uintptr_t)early_p4d : (uintptr_t)early_pud, 826 PGDIR_SIZE, PAGE_TABLE); 827 828 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 829 830 local_flush_tlb_all(); 831 csr_write(CSR_SATP, identity_satp); 832 hw_satp = csr_swap(CSR_SATP, 0ULL); 833 local_flush_tlb_all(); 834 835 if (hw_satp != identity_satp) { 836 if (pgtable_l5_enabled) { 837 disable_pgtable_l5(); 838 memset(early_pg_dir, 0, PAGE_SIZE); 839 goto retry; 840 } 841 disable_pgtable_l4(); 842 } 843 844 memset(early_pg_dir, 0, PAGE_SIZE); 845 memset(early_p4d, 0, PAGE_SIZE); 846 memset(early_pud, 0, PAGE_SIZE); 847 memset(early_pmd, 0, PAGE_SIZE); 848 } 849 #endif 850 851 /* 852 * setup_vm() is called from head.S with MMU-off. 853 * 854 * Following requirements should be honoured for setup_vm() to work 855 * correctly: 856 * 1) It should use PC-relative addressing for accessing kernel symbols. 857 * To achieve this we always use GCC cmodel=medany. 858 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 859 * so disable compiler instrumentation when FTRACE is enabled. 860 * 861 * Currently, the above requirements are honoured by using custom CFLAGS 862 * for init.o in mm/Makefile. 863 */ 864 865 #ifndef __riscv_cmodel_medany 866 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 867 #endif 868 869 #ifdef CONFIG_RELOCATABLE 870 extern unsigned long __rela_dyn_start, __rela_dyn_end; 871 872 static void __init relocate_kernel(void) 873 { 874 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 875 /* 876 * This holds the offset between the linked virtual address and the 877 * relocated virtual address. 878 */ 879 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 880 /* 881 * This holds the offset between kernel linked virtual address and 882 * physical address. 883 */ 884 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 885 886 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 887 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 888 Elf64_Addr relocated_addr = rela->r_addend; 889 890 if (rela->r_info != R_RISCV_RELATIVE) 891 continue; 892 893 /* 894 * Make sure to not relocate vdso symbols like rt_sigreturn 895 * which are linked from the address 0 in vmlinux since 896 * vdso symbol addresses are actually used as an offset from 897 * mm->context.vdso in VDSO_OFFSET macro. 898 */ 899 if (relocated_addr >= KERNEL_LINK_ADDR) 900 relocated_addr += reloc_offset; 901 902 *(Elf64_Addr *)addr = relocated_addr; 903 } 904 } 905 #endif /* CONFIG_RELOCATABLE */ 906 907 #ifdef CONFIG_XIP_KERNEL 908 static void __init create_kernel_page_table(pgd_t *pgdir, 909 __always_unused bool early) 910 { 911 uintptr_t va, end_va; 912 913 /* Map the flash resident part */ 914 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 915 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 916 create_pgd_mapping(pgdir, va, 917 kernel_map.xiprom + (va - kernel_map.virt_addr), 918 PMD_SIZE, PAGE_KERNEL_EXEC); 919 920 /* Map the data in RAM */ 921 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 922 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 923 create_pgd_mapping(pgdir, va, 924 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 925 PMD_SIZE, PAGE_KERNEL); 926 } 927 #else 928 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 929 { 930 uintptr_t va, end_va; 931 932 end_va = kernel_map.virt_addr + kernel_map.size; 933 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 934 create_pgd_mapping(pgdir, va, 935 kernel_map.phys_addr + (va - kernel_map.virt_addr), 936 PMD_SIZE, 937 early ? 938 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 939 } 940 #endif 941 942 /* 943 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 944 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 945 * entry. 946 */ 947 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 948 uintptr_t dtb_pa) 949 { 950 #ifndef CONFIG_BUILTIN_DTB 951 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 952 953 /* Make sure the fdt fixmap address is always aligned on PMD size */ 954 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 955 956 /* In 32-bit only, the fdt lies in its own PGD */ 957 if (!IS_ENABLED(CONFIG_64BIT)) { 958 create_pgd_mapping(early_pg_dir, fix_fdt_va, 959 pa, MAX_FDT_SIZE, PAGE_KERNEL); 960 } else { 961 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 962 pa, PMD_SIZE, PAGE_KERNEL); 963 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 964 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 965 } 966 967 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 968 #else 969 /* 970 * For 64-bit kernel, __va can't be used since it would return a linear 971 * mapping address whereas dtb_early_va will be used before 972 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 973 * kernel is mapped in the linear mapping, that makes no difference. 974 */ 975 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 976 #endif 977 978 dtb_early_pa = dtb_pa; 979 } 980 981 /* 982 * MMU is not enabled, the page tables are allocated directly using 983 * early_pmd/pud/p4d and the address returned is the physical one. 984 */ 985 static void __init pt_ops_set_early(void) 986 { 987 pt_ops.alloc_pte = alloc_pte_early; 988 pt_ops.get_pte_virt = get_pte_virt_early; 989 #ifndef __PAGETABLE_PMD_FOLDED 990 pt_ops.alloc_pmd = alloc_pmd_early; 991 pt_ops.get_pmd_virt = get_pmd_virt_early; 992 pt_ops.alloc_pud = alloc_pud_early; 993 pt_ops.get_pud_virt = get_pud_virt_early; 994 pt_ops.alloc_p4d = alloc_p4d_early; 995 pt_ops.get_p4d_virt = get_p4d_virt_early; 996 #endif 997 } 998 999 /* 1000 * MMU is enabled but page table setup is not complete yet. 1001 * fixmap page table alloc functions must be used as a means to temporarily 1002 * map the allocated physical pages since the linear mapping does not exist yet. 1003 * 1004 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1005 * but it will be used as described above. 1006 */ 1007 static void __init pt_ops_set_fixmap(void) 1008 { 1009 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1010 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1011 #ifndef __PAGETABLE_PMD_FOLDED 1012 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1013 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1014 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1015 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1016 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1017 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1018 #endif 1019 } 1020 1021 /* 1022 * MMU is enabled and page table setup is complete, so from now, we can use 1023 * generic page allocation functions to setup page table. 1024 */ 1025 static void __init pt_ops_set_late(void) 1026 { 1027 pt_ops.alloc_pte = alloc_pte_late; 1028 pt_ops.get_pte_virt = get_pte_virt_late; 1029 #ifndef __PAGETABLE_PMD_FOLDED 1030 pt_ops.alloc_pmd = alloc_pmd_late; 1031 pt_ops.get_pmd_virt = get_pmd_virt_late; 1032 pt_ops.alloc_pud = alloc_pud_late; 1033 pt_ops.get_pud_virt = get_pud_virt_late; 1034 pt_ops.alloc_p4d = alloc_p4d_late; 1035 pt_ops.get_p4d_virt = get_p4d_virt_late; 1036 #endif 1037 } 1038 1039 #ifdef CONFIG_RANDOMIZE_BASE 1040 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1041 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1042 1043 static int __init print_nokaslr(char *p) 1044 { 1045 pr_info("Disabled KASLR"); 1046 return 0; 1047 } 1048 early_param("nokaslr", print_nokaslr); 1049 1050 unsigned long kaslr_offset(void) 1051 { 1052 return kernel_map.virt_offset; 1053 } 1054 #endif 1055 1056 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1057 { 1058 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1059 1060 #ifdef CONFIG_RANDOMIZE_BASE 1061 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1062 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1063 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1064 u32 nr_pos; 1065 1066 /* 1067 * Compute the number of positions available: we are limited 1068 * by the early page table that only has one PUD and we must 1069 * be aligned on PMD_SIZE. 1070 */ 1071 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1072 1073 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1074 } 1075 #endif 1076 1077 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1078 1079 #ifdef CONFIG_XIP_KERNEL 1080 #ifdef CONFIG_64BIT 1081 kernel_map.page_offset = PAGE_OFFSET_L3; 1082 #else 1083 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1084 #endif 1085 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1086 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1087 1088 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1089 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1090 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1091 1092 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1093 #else 1094 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1095 kernel_map.phys_addr = (uintptr_t)(&_start); 1096 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1097 #endif 1098 1099 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1100 set_satp_mode(dtb_pa); 1101 set_mmap_rnd_bits_max(); 1102 #endif 1103 1104 /* 1105 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1106 * where we have the system memory layout: this allows us to align 1107 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1108 * for the linear mapping. This is only possible because the kernel 1109 * mapping lies outside the linear mapping. 1110 * In 32-bit however, as the kernel resides in the linear mapping, 1111 * setup_vm_final can not change the mapping established here, 1112 * otherwise the same kernel addresses would get mapped to different 1113 * physical addresses (if the start of dram is different from the 1114 * kernel physical address start). 1115 */ 1116 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1117 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1118 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1119 1120 /* 1121 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1122 * kernel, whereas for 64-bit kernel, the end of the virtual address 1123 * space is occupied by the modules/BPF/kernel mappings which reduces 1124 * the available size of the linear mapping. 1125 */ 1126 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1127 1128 /* Sanity check alignment and size */ 1129 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1130 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1131 1132 #ifdef CONFIG_64BIT 1133 /* 1134 * The last 4K bytes of the addressable memory can not be mapped because 1135 * of IS_ERR_VALUE macro. 1136 */ 1137 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1138 #endif 1139 1140 #ifdef CONFIG_RELOCATABLE 1141 /* 1142 * Early page table uses only one PUD, which makes it possible 1143 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1144 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1145 * since a part of the kernel would not get mapped. 1146 */ 1147 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1148 relocate_kernel(); 1149 #endif 1150 1151 apply_early_boot_alternatives(); 1152 pt_ops_set_early(); 1153 1154 /* Setup early PGD for fixmap */ 1155 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1156 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1157 1158 #ifndef __PAGETABLE_PMD_FOLDED 1159 /* Setup fixmap P4D and PUD */ 1160 if (pgtable_l5_enabled) 1161 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1162 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1163 /* Setup fixmap PUD and PMD */ 1164 if (pgtable_l4_enabled) 1165 create_pud_mapping(fixmap_pud, FIXADDR_START, 1166 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1167 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1168 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1169 /* Setup trampoline PGD and PMD */ 1170 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1171 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1172 if (pgtable_l5_enabled) 1173 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1174 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1175 if (pgtable_l4_enabled) 1176 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1177 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1178 #ifdef CONFIG_XIP_KERNEL 1179 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1180 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1181 #else 1182 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1183 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1184 #endif 1185 #else 1186 /* Setup trampoline PGD */ 1187 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1188 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1189 #endif 1190 1191 /* 1192 * Setup early PGD covering entire kernel which will allow 1193 * us to reach paging_init(). We map all memory banks later 1194 * in setup_vm_final() below. 1195 */ 1196 create_kernel_page_table(early_pg_dir, true); 1197 1198 /* Setup early mapping for FDT early scan */ 1199 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1200 1201 /* 1202 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1203 * range can not span multiple pmds. 1204 */ 1205 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1206 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1207 1208 #ifndef __PAGETABLE_PMD_FOLDED 1209 /* 1210 * Early ioremap fixmap is already created as it lies within first 2MB 1211 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1212 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1213 * the user if not. 1214 */ 1215 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1216 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1217 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1218 WARN_ON(1); 1219 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1220 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1221 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1222 fix_to_virt(FIX_BTMAP_BEGIN)); 1223 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1224 fix_to_virt(FIX_BTMAP_END)); 1225 1226 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1227 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1228 } 1229 #endif 1230 1231 pt_ops_set_fixmap(); 1232 } 1233 1234 static void __init create_linear_mapping_range(phys_addr_t start, 1235 phys_addr_t end, 1236 uintptr_t fixed_map_size) 1237 { 1238 phys_addr_t pa; 1239 uintptr_t va, map_size; 1240 1241 for (pa = start; pa < end; pa += map_size) { 1242 va = (uintptr_t)__va(pa); 1243 map_size = fixed_map_size ? fixed_map_size : 1244 best_map_size(pa, va, end - pa); 1245 1246 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1247 pgprot_from_va(va)); 1248 } 1249 } 1250 1251 static void __init create_linear_mapping_page_table(void) 1252 { 1253 phys_addr_t start, end; 1254 phys_addr_t kfence_pool __maybe_unused; 1255 u64 i; 1256 1257 #ifdef CONFIG_STRICT_KERNEL_RWX 1258 phys_addr_t ktext_start = __pa_symbol(_start); 1259 phys_addr_t ktext_size = __init_data_begin - _start; 1260 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1261 phys_addr_t krodata_size = _data - __start_rodata; 1262 1263 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1264 memblock_mark_nomap(ktext_start, ktext_size); 1265 memblock_mark_nomap(krodata_start, krodata_size); 1266 #endif 1267 1268 #ifdef CONFIG_KFENCE 1269 /* 1270 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1271 * before we setup the linear mapping so that we avoid using hugepages 1272 * for this region. 1273 */ 1274 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1275 BUG_ON(!kfence_pool); 1276 1277 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1278 __kfence_pool = __va(kfence_pool); 1279 #endif 1280 1281 /* Map all memory banks in the linear mapping */ 1282 for_each_mem_range(i, &start, &end) { 1283 if (start >= end) 1284 break; 1285 if (start <= __pa(PAGE_OFFSET) && 1286 __pa(PAGE_OFFSET) < end) 1287 start = __pa(PAGE_OFFSET); 1288 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1289 end = __pa(PAGE_OFFSET) + memory_limit; 1290 1291 create_linear_mapping_range(start, end, 0); 1292 } 1293 1294 #ifdef CONFIG_STRICT_KERNEL_RWX 1295 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0); 1296 create_linear_mapping_range(krodata_start, 1297 krodata_start + krodata_size, 0); 1298 1299 memblock_clear_nomap(ktext_start, ktext_size); 1300 memblock_clear_nomap(krodata_start, krodata_size); 1301 #endif 1302 1303 #ifdef CONFIG_KFENCE 1304 create_linear_mapping_range(kfence_pool, 1305 kfence_pool + KFENCE_POOL_SIZE, 1306 PAGE_SIZE); 1307 1308 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1309 #endif 1310 } 1311 1312 static void __init setup_vm_final(void) 1313 { 1314 /* Setup swapper PGD for fixmap */ 1315 #if !defined(CONFIG_64BIT) 1316 /* 1317 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1318 * directly in swapper_pg_dir in addition to the pgd entry that points 1319 * to fixmap_pte. 1320 */ 1321 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1322 1323 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1324 #endif 1325 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1326 __pa_symbol(fixmap_pgd_next), 1327 PGDIR_SIZE, PAGE_TABLE); 1328 1329 /* Map the linear mapping */ 1330 create_linear_mapping_page_table(); 1331 1332 /* Map the kernel */ 1333 if (IS_ENABLED(CONFIG_64BIT)) 1334 create_kernel_page_table(swapper_pg_dir, false); 1335 1336 #ifdef CONFIG_KASAN 1337 kasan_swapper_init(); 1338 #endif 1339 1340 /* Clear fixmap PTE and PMD mappings */ 1341 clear_fixmap(FIX_PTE); 1342 clear_fixmap(FIX_PMD); 1343 clear_fixmap(FIX_PUD); 1344 clear_fixmap(FIX_P4D); 1345 1346 /* Move to swapper page table */ 1347 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1348 local_flush_tlb_all(); 1349 1350 pt_ops_set_late(); 1351 } 1352 #else 1353 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1354 { 1355 dtb_early_va = (void *)dtb_pa; 1356 dtb_early_pa = dtb_pa; 1357 } 1358 1359 static inline void setup_vm_final(void) 1360 { 1361 } 1362 #endif /* CONFIG_MMU */ 1363 1364 /* 1365 * reserve_crashkernel() - reserves memory for crash kernel 1366 * 1367 * This function reserves memory area given in "crashkernel=" kernel command 1368 * line parameter. The memory reserved is used by dump capture kernel when 1369 * primary kernel is crashing. 1370 */ 1371 static void __init arch_reserve_crashkernel(void) 1372 { 1373 unsigned long long low_size = 0; 1374 unsigned long long crash_base, crash_size; 1375 char *cmdline = boot_command_line; 1376 bool high = false; 1377 int ret; 1378 1379 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1380 return; 1381 1382 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 1383 &crash_size, &crash_base, 1384 &low_size, &high); 1385 if (ret) 1386 return; 1387 1388 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 1389 low_size, high); 1390 } 1391 1392 void __init paging_init(void) 1393 { 1394 setup_bootmem(); 1395 setup_vm_final(); 1396 1397 /* Depend on that Linear Mapping is ready */ 1398 memblock_allow_resize(); 1399 } 1400 1401 void __init misc_mem_init(void) 1402 { 1403 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1404 arch_numa_init(); 1405 sparse_init(); 1406 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1407 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1408 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1409 #endif 1410 zone_sizes_init(); 1411 arch_reserve_crashkernel(); 1412 memblock_dump_all(); 1413 } 1414 1415 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1416 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1417 unsigned long addr, unsigned long next) 1418 { 1419 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1420 } 1421 1422 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1423 unsigned long addr, unsigned long next) 1424 { 1425 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1426 return 1; 1427 } 1428 1429 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1430 struct vmem_altmap *altmap) 1431 { 1432 /* 1433 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1434 * can't use hugepage mappings for 2-level page table because in case of 1435 * memory hotplug, we are not able to update all the page tables with 1436 * the new PMDs. 1437 */ 1438 return vmemmap_populate_hugepages(start, end, node, NULL); 1439 } 1440 #endif 1441 1442 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1443 /* 1444 * Pre-allocates page-table pages for a specific area in the kernel 1445 * page-table. Only the level which needs to be synchronized between 1446 * all page-tables is allocated because the synchronization can be 1447 * expensive. 1448 */ 1449 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1450 const char *area) 1451 { 1452 unsigned long addr; 1453 const char *lvl; 1454 1455 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1456 pgd_t *pgd = pgd_offset_k(addr); 1457 p4d_t *p4d; 1458 pud_t *pud; 1459 pmd_t *pmd; 1460 1461 lvl = "p4d"; 1462 p4d = p4d_alloc(&init_mm, pgd, addr); 1463 if (!p4d) 1464 goto failed; 1465 1466 if (pgtable_l5_enabled) 1467 continue; 1468 1469 lvl = "pud"; 1470 pud = pud_alloc(&init_mm, p4d, addr); 1471 if (!pud) 1472 goto failed; 1473 1474 if (pgtable_l4_enabled) 1475 continue; 1476 1477 lvl = "pmd"; 1478 pmd = pmd_alloc(&init_mm, pud, addr); 1479 if (!pmd) 1480 goto failed; 1481 } 1482 return; 1483 1484 failed: 1485 /* 1486 * The pages have to be there now or they will be missing in 1487 * process page-tables later. 1488 */ 1489 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1490 } 1491 1492 void __init pgtable_cache_init(void) 1493 { 1494 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1495 if (IS_ENABLED(CONFIG_MODULES)) 1496 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1497 } 1498 #endif 1499 1500 #ifdef CONFIG_EXECMEM 1501 #ifdef CONFIG_MMU 1502 static struct execmem_info execmem_info __ro_after_init; 1503 1504 struct execmem_info __init *execmem_arch_setup(void) 1505 { 1506 execmem_info = (struct execmem_info){ 1507 .ranges = { 1508 [EXECMEM_DEFAULT] = { 1509 .start = MODULES_VADDR, 1510 .end = MODULES_END, 1511 .pgprot = PAGE_KERNEL, 1512 .alignment = 1, 1513 }, 1514 [EXECMEM_KPROBES] = { 1515 .start = VMALLOC_START, 1516 .end = VMALLOC_END, 1517 .pgprot = PAGE_KERNEL_READ_EXEC, 1518 .alignment = 1, 1519 }, 1520 [EXECMEM_BPF] = { 1521 .start = BPF_JIT_REGION_START, 1522 .end = BPF_JIT_REGION_END, 1523 .pgprot = PAGE_KERNEL, 1524 .alignment = PAGE_SIZE, 1525 }, 1526 }, 1527 }; 1528 1529 return &execmem_info; 1530 } 1531 #endif /* CONFIG_MMU */ 1532 #endif /* CONFIG_EXECMEM */ 1533