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 (debug_pagealloc_enabled()) 687 return PAGE_SIZE; 688 689 if (pgtable_l5_enabled && 690 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 691 return P4D_SIZE; 692 693 if (pgtable_l4_enabled && 694 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 695 return PUD_SIZE; 696 697 if (IS_ENABLED(CONFIG_64BIT) && 698 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 699 return PMD_SIZE; 700 701 return PAGE_SIZE; 702 } 703 704 #ifdef CONFIG_XIP_KERNEL 705 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 706 extern char _xiprom[], _exiprom[], __data_loc; 707 708 /* called from head.S with MMU off */ 709 asmlinkage void __init __copy_data(void) 710 { 711 void *from = (void *)(&__data_loc); 712 void *to = (void *)CONFIG_PHYS_RAM_BASE; 713 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 714 715 memcpy(to, from, sz); 716 } 717 #endif 718 719 #ifdef CONFIG_STRICT_KERNEL_RWX 720 static __init pgprot_t pgprot_from_va(uintptr_t va) 721 { 722 if (is_va_kernel_text(va)) 723 return PAGE_KERNEL_READ_EXEC; 724 725 /* 726 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 727 * we must protect its linear mapping alias from being executed and 728 * written. 729 * And rodata section is marked readonly in mark_rodata_ro. 730 */ 731 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 732 return PAGE_KERNEL_READ; 733 734 return PAGE_KERNEL; 735 } 736 737 void mark_rodata_ro(void) 738 { 739 set_kernel_memory(__start_rodata, _data, set_memory_ro); 740 if (IS_ENABLED(CONFIG_64BIT)) 741 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 742 set_memory_ro); 743 } 744 #else 745 static __init pgprot_t pgprot_from_va(uintptr_t va) 746 { 747 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 748 return PAGE_KERNEL; 749 750 return PAGE_KERNEL_EXEC; 751 } 752 #endif /* CONFIG_STRICT_KERNEL_RWX */ 753 754 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 755 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 756 757 static void __init disable_pgtable_l5(void) 758 { 759 pgtable_l5_enabled = false; 760 kernel_map.page_offset = PAGE_OFFSET_L4; 761 satp_mode = SATP_MODE_48; 762 } 763 764 static void __init disable_pgtable_l4(void) 765 { 766 pgtable_l4_enabled = false; 767 kernel_map.page_offset = PAGE_OFFSET_L3; 768 satp_mode = SATP_MODE_39; 769 } 770 771 static int __init print_no4lvl(char *p) 772 { 773 pr_info("Disabled 4-level and 5-level paging"); 774 return 0; 775 } 776 early_param("no4lvl", print_no4lvl); 777 778 static int __init print_no5lvl(char *p) 779 { 780 pr_info("Disabled 5-level paging"); 781 return 0; 782 } 783 early_param("no5lvl", print_no5lvl); 784 785 static void __init set_mmap_rnd_bits_max(void) 786 { 787 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 788 } 789 790 /* 791 * There is a simple way to determine if 4-level is supported by the 792 * underlying hardware: establish 1:1 mapping in 4-level page table mode 793 * then read SATP to see if the configuration was taken into account 794 * meaning sv48 is supported. 795 */ 796 static __init void set_satp_mode(uintptr_t dtb_pa) 797 { 798 u64 identity_satp, hw_satp; 799 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 800 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 801 802 if (satp_mode_cmdline == SATP_MODE_57) { 803 disable_pgtable_l5(); 804 } else if (satp_mode_cmdline == SATP_MODE_48) { 805 disable_pgtable_l5(); 806 disable_pgtable_l4(); 807 return; 808 } 809 810 create_p4d_mapping(early_p4d, 811 set_satp_mode_pmd, (uintptr_t)early_pud, 812 P4D_SIZE, PAGE_TABLE); 813 create_pud_mapping(early_pud, 814 set_satp_mode_pmd, (uintptr_t)early_pmd, 815 PUD_SIZE, PAGE_TABLE); 816 /* Handle the case where set_satp_mode straddles 2 PMDs */ 817 create_pmd_mapping(early_pmd, 818 set_satp_mode_pmd, set_satp_mode_pmd, 819 PMD_SIZE, PAGE_KERNEL_EXEC); 820 create_pmd_mapping(early_pmd, 821 set_satp_mode_pmd + PMD_SIZE, 822 set_satp_mode_pmd + PMD_SIZE, 823 PMD_SIZE, PAGE_KERNEL_EXEC); 824 retry: 825 create_pgd_mapping(early_pg_dir, 826 set_satp_mode_pmd, 827 pgtable_l5_enabled ? 828 (uintptr_t)early_p4d : (uintptr_t)early_pud, 829 PGDIR_SIZE, PAGE_TABLE); 830 831 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 832 833 local_flush_tlb_all(); 834 csr_write(CSR_SATP, identity_satp); 835 hw_satp = csr_swap(CSR_SATP, 0ULL); 836 local_flush_tlb_all(); 837 838 if (hw_satp != identity_satp) { 839 if (pgtable_l5_enabled) { 840 disable_pgtable_l5(); 841 memset(early_pg_dir, 0, PAGE_SIZE); 842 goto retry; 843 } 844 disable_pgtable_l4(); 845 } 846 847 memset(early_pg_dir, 0, PAGE_SIZE); 848 memset(early_p4d, 0, PAGE_SIZE); 849 memset(early_pud, 0, PAGE_SIZE); 850 memset(early_pmd, 0, PAGE_SIZE); 851 } 852 #endif 853 854 /* 855 * setup_vm() is called from head.S with MMU-off. 856 * 857 * Following requirements should be honoured for setup_vm() to work 858 * correctly: 859 * 1) It should use PC-relative addressing for accessing kernel symbols. 860 * To achieve this we always use GCC cmodel=medany. 861 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 862 * so disable compiler instrumentation when FTRACE is enabled. 863 * 864 * Currently, the above requirements are honoured by using custom CFLAGS 865 * for init.o in mm/Makefile. 866 */ 867 868 #ifndef __riscv_cmodel_medany 869 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 870 #endif 871 872 #ifdef CONFIG_RELOCATABLE 873 extern unsigned long __rela_dyn_start, __rela_dyn_end; 874 875 static void __init relocate_kernel(void) 876 { 877 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start; 878 /* 879 * This holds the offset between the linked virtual address and the 880 * relocated virtual address. 881 */ 882 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 883 /* 884 * This holds the offset between kernel linked virtual address and 885 * physical address. 886 */ 887 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 888 889 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) { 890 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 891 Elf64_Addr relocated_addr = rela->r_addend; 892 893 if (rela->r_info != R_RISCV_RELATIVE) 894 continue; 895 896 /* 897 * Make sure to not relocate vdso symbols like rt_sigreturn 898 * which are linked from the address 0 in vmlinux since 899 * vdso symbol addresses are actually used as an offset from 900 * mm->context.vdso in VDSO_OFFSET macro. 901 */ 902 if (relocated_addr >= KERNEL_LINK_ADDR) 903 relocated_addr += reloc_offset; 904 905 *(Elf64_Addr *)addr = relocated_addr; 906 } 907 } 908 #endif /* CONFIG_RELOCATABLE */ 909 910 #ifdef CONFIG_XIP_KERNEL 911 static void __init create_kernel_page_table(pgd_t *pgdir, 912 __always_unused bool early) 913 { 914 uintptr_t va, end_va; 915 916 /* Map the flash resident part */ 917 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 918 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 919 create_pgd_mapping(pgdir, va, 920 kernel_map.xiprom + (va - kernel_map.virt_addr), 921 PMD_SIZE, PAGE_KERNEL_EXEC); 922 923 /* Map the data in RAM */ 924 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 925 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 926 create_pgd_mapping(pgdir, va, 927 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 928 PMD_SIZE, PAGE_KERNEL); 929 } 930 #else 931 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 932 { 933 uintptr_t va, end_va; 934 935 end_va = kernel_map.virt_addr + kernel_map.size; 936 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 937 create_pgd_mapping(pgdir, va, 938 kernel_map.phys_addr + (va - kernel_map.virt_addr), 939 PMD_SIZE, 940 early ? 941 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 942 } 943 #endif 944 945 /* 946 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 947 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 948 * entry. 949 */ 950 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 951 uintptr_t dtb_pa) 952 { 953 #ifndef CONFIG_BUILTIN_DTB 954 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 955 956 /* Make sure the fdt fixmap address is always aligned on PMD size */ 957 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 958 959 /* In 32-bit only, the fdt lies in its own PGD */ 960 if (!IS_ENABLED(CONFIG_64BIT)) { 961 create_pgd_mapping(early_pg_dir, fix_fdt_va, 962 pa, MAX_FDT_SIZE, PAGE_KERNEL); 963 } else { 964 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 965 pa, PMD_SIZE, PAGE_KERNEL); 966 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 967 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 968 } 969 970 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 971 #else 972 /* 973 * For 64-bit kernel, __va can't be used since it would return a linear 974 * mapping address whereas dtb_early_va will be used before 975 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 976 * kernel is mapped in the linear mapping, that makes no difference. 977 */ 978 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 979 #endif 980 981 dtb_early_pa = dtb_pa; 982 } 983 984 /* 985 * MMU is not enabled, the page tables are allocated directly using 986 * early_pmd/pud/p4d and the address returned is the physical one. 987 */ 988 static void __init pt_ops_set_early(void) 989 { 990 pt_ops.alloc_pte = alloc_pte_early; 991 pt_ops.get_pte_virt = get_pte_virt_early; 992 #ifndef __PAGETABLE_PMD_FOLDED 993 pt_ops.alloc_pmd = alloc_pmd_early; 994 pt_ops.get_pmd_virt = get_pmd_virt_early; 995 pt_ops.alloc_pud = alloc_pud_early; 996 pt_ops.get_pud_virt = get_pud_virt_early; 997 pt_ops.alloc_p4d = alloc_p4d_early; 998 pt_ops.get_p4d_virt = get_p4d_virt_early; 999 #endif 1000 } 1001 1002 /* 1003 * MMU is enabled but page table setup is not complete yet. 1004 * fixmap page table alloc functions must be used as a means to temporarily 1005 * map the allocated physical pages since the linear mapping does not exist yet. 1006 * 1007 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1008 * but it will be used as described above. 1009 */ 1010 static void __init pt_ops_set_fixmap(void) 1011 { 1012 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1013 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1014 #ifndef __PAGETABLE_PMD_FOLDED 1015 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1016 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1017 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1018 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1019 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1020 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1021 #endif 1022 } 1023 1024 /* 1025 * MMU is enabled and page table setup is complete, so from now, we can use 1026 * generic page allocation functions to setup page table. 1027 */ 1028 static void __init pt_ops_set_late(void) 1029 { 1030 pt_ops.alloc_pte = alloc_pte_late; 1031 pt_ops.get_pte_virt = get_pte_virt_late; 1032 #ifndef __PAGETABLE_PMD_FOLDED 1033 pt_ops.alloc_pmd = alloc_pmd_late; 1034 pt_ops.get_pmd_virt = get_pmd_virt_late; 1035 pt_ops.alloc_pud = alloc_pud_late; 1036 pt_ops.get_pud_virt = get_pud_virt_late; 1037 pt_ops.alloc_p4d = alloc_p4d_late; 1038 pt_ops.get_p4d_virt = get_p4d_virt_late; 1039 #endif 1040 } 1041 1042 #ifdef CONFIG_RANDOMIZE_BASE 1043 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1044 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1045 1046 static int __init print_nokaslr(char *p) 1047 { 1048 pr_info("Disabled KASLR"); 1049 return 0; 1050 } 1051 early_param("nokaslr", print_nokaslr); 1052 1053 unsigned long kaslr_offset(void) 1054 { 1055 return kernel_map.virt_offset; 1056 } 1057 #endif 1058 1059 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1060 { 1061 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1062 1063 #ifdef CONFIG_RANDOMIZE_BASE 1064 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1065 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1066 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1067 u32 nr_pos; 1068 1069 /* 1070 * Compute the number of positions available: we are limited 1071 * by the early page table that only has one PUD and we must 1072 * be aligned on PMD_SIZE. 1073 */ 1074 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1075 1076 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1077 } 1078 #endif 1079 1080 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1081 1082 #ifdef CONFIG_XIP_KERNEL 1083 #ifdef CONFIG_64BIT 1084 kernel_map.page_offset = PAGE_OFFSET_L3; 1085 #else 1086 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1087 #endif 1088 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1089 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1090 1091 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1092 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1093 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 1094 1095 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1096 #else 1097 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 1098 kernel_map.phys_addr = (uintptr_t)(&_start); 1099 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1100 #endif 1101 1102 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1103 set_satp_mode(dtb_pa); 1104 set_mmap_rnd_bits_max(); 1105 #endif 1106 1107 /* 1108 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1109 * where we have the system memory layout: this allows us to align 1110 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1111 * for the linear mapping. This is only possible because the kernel 1112 * mapping lies outside the linear mapping. 1113 * In 32-bit however, as the kernel resides in the linear mapping, 1114 * setup_vm_final can not change the mapping established here, 1115 * otherwise the same kernel addresses would get mapped to different 1116 * physical addresses (if the start of dram is different from the 1117 * kernel physical address start). 1118 */ 1119 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1120 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1121 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1122 1123 /* 1124 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 1125 * kernel, whereas for 64-bit kernel, the end of the virtual address 1126 * space is occupied by the modules/BPF/kernel mappings which reduces 1127 * the available size of the linear mapping. 1128 */ 1129 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 1130 1131 /* Sanity check alignment and size */ 1132 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1133 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1134 1135 #ifdef CONFIG_64BIT 1136 /* 1137 * The last 4K bytes of the addressable memory can not be mapped because 1138 * of IS_ERR_VALUE macro. 1139 */ 1140 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1141 #endif 1142 1143 #ifdef CONFIG_RELOCATABLE 1144 /* 1145 * Early page table uses only one PUD, which makes it possible 1146 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1147 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1148 * since a part of the kernel would not get mapped. 1149 */ 1150 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1151 relocate_kernel(); 1152 #endif 1153 1154 apply_early_boot_alternatives(); 1155 pt_ops_set_early(); 1156 1157 /* Setup early PGD for fixmap */ 1158 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1159 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1160 1161 #ifndef __PAGETABLE_PMD_FOLDED 1162 /* Setup fixmap P4D and PUD */ 1163 if (pgtable_l5_enabled) 1164 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1165 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1166 /* Setup fixmap PUD and PMD */ 1167 if (pgtable_l4_enabled) 1168 create_pud_mapping(fixmap_pud, FIXADDR_START, 1169 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1170 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1171 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1172 /* Setup trampoline PGD and PMD */ 1173 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1174 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1175 if (pgtable_l5_enabled) 1176 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1177 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1178 if (pgtable_l4_enabled) 1179 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1180 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1181 #ifdef CONFIG_XIP_KERNEL 1182 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1183 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1184 #else 1185 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1186 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1187 #endif 1188 #else 1189 /* Setup trampoline PGD */ 1190 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1191 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1192 #endif 1193 1194 /* 1195 * Setup early PGD covering entire kernel which will allow 1196 * us to reach paging_init(). We map all memory banks later 1197 * in setup_vm_final() below. 1198 */ 1199 create_kernel_page_table(early_pg_dir, true); 1200 1201 /* Setup early mapping for FDT early scan */ 1202 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1203 1204 /* 1205 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1206 * range can not span multiple pmds. 1207 */ 1208 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1209 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1210 1211 #ifndef __PAGETABLE_PMD_FOLDED 1212 /* 1213 * Early ioremap fixmap is already created as it lies within first 2MB 1214 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1215 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1216 * the user if not. 1217 */ 1218 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1219 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1220 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1221 WARN_ON(1); 1222 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1223 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1224 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1225 fix_to_virt(FIX_BTMAP_BEGIN)); 1226 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1227 fix_to_virt(FIX_BTMAP_END)); 1228 1229 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1230 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1231 } 1232 #endif 1233 1234 pt_ops_set_fixmap(); 1235 } 1236 1237 static void __init create_linear_mapping_range(phys_addr_t start, 1238 phys_addr_t end, 1239 uintptr_t fixed_map_size) 1240 { 1241 phys_addr_t pa; 1242 uintptr_t va, map_size; 1243 1244 for (pa = start; pa < end; pa += map_size) { 1245 va = (uintptr_t)__va(pa); 1246 map_size = fixed_map_size ? fixed_map_size : 1247 best_map_size(pa, va, end - pa); 1248 1249 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1250 pgprot_from_va(va)); 1251 } 1252 } 1253 1254 static void __init create_linear_mapping_page_table(void) 1255 { 1256 phys_addr_t start, end; 1257 phys_addr_t kfence_pool __maybe_unused; 1258 u64 i; 1259 1260 #ifdef CONFIG_STRICT_KERNEL_RWX 1261 phys_addr_t ktext_start = __pa_symbol(_start); 1262 phys_addr_t ktext_size = __init_data_begin - _start; 1263 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1264 phys_addr_t krodata_size = _data - __start_rodata; 1265 1266 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1267 memblock_mark_nomap(ktext_start, ktext_size); 1268 memblock_mark_nomap(krodata_start, krodata_size); 1269 #endif 1270 1271 #ifdef CONFIG_KFENCE 1272 /* 1273 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1274 * before we setup the linear mapping so that we avoid using hugepages 1275 * for this region. 1276 */ 1277 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1278 BUG_ON(!kfence_pool); 1279 1280 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1281 __kfence_pool = __va(kfence_pool); 1282 #endif 1283 1284 /* Map all memory banks in the linear mapping */ 1285 for_each_mem_range(i, &start, &end) { 1286 if (start >= end) 1287 break; 1288 if (start <= __pa(PAGE_OFFSET) && 1289 __pa(PAGE_OFFSET) < end) 1290 start = __pa(PAGE_OFFSET); 1291 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1292 end = __pa(PAGE_OFFSET) + memory_limit; 1293 1294 create_linear_mapping_range(start, end, 0); 1295 } 1296 1297 #ifdef CONFIG_STRICT_KERNEL_RWX 1298 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0); 1299 create_linear_mapping_range(krodata_start, 1300 krodata_start + krodata_size, 0); 1301 1302 memblock_clear_nomap(ktext_start, ktext_size); 1303 memblock_clear_nomap(krodata_start, krodata_size); 1304 #endif 1305 1306 #ifdef CONFIG_KFENCE 1307 create_linear_mapping_range(kfence_pool, 1308 kfence_pool + KFENCE_POOL_SIZE, 1309 PAGE_SIZE); 1310 1311 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1312 #endif 1313 } 1314 1315 static void __init setup_vm_final(void) 1316 { 1317 /* Setup swapper PGD for fixmap */ 1318 #if !defined(CONFIG_64BIT) 1319 /* 1320 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1321 * directly in swapper_pg_dir in addition to the pgd entry that points 1322 * to fixmap_pte. 1323 */ 1324 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1325 1326 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1327 #endif 1328 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1329 __pa_symbol(fixmap_pgd_next), 1330 PGDIR_SIZE, PAGE_TABLE); 1331 1332 /* Map the linear mapping */ 1333 create_linear_mapping_page_table(); 1334 1335 /* Map the kernel */ 1336 if (IS_ENABLED(CONFIG_64BIT)) 1337 create_kernel_page_table(swapper_pg_dir, false); 1338 1339 #ifdef CONFIG_KASAN 1340 kasan_swapper_init(); 1341 #endif 1342 1343 /* Clear fixmap PTE and PMD mappings */ 1344 clear_fixmap(FIX_PTE); 1345 clear_fixmap(FIX_PMD); 1346 clear_fixmap(FIX_PUD); 1347 clear_fixmap(FIX_P4D); 1348 1349 /* Move to swapper page table */ 1350 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1351 local_flush_tlb_all(); 1352 1353 pt_ops_set_late(); 1354 } 1355 #else 1356 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1357 { 1358 dtb_early_va = (void *)dtb_pa; 1359 dtb_early_pa = dtb_pa; 1360 } 1361 1362 static inline void setup_vm_final(void) 1363 { 1364 } 1365 #endif /* CONFIG_MMU */ 1366 1367 /* 1368 * reserve_crashkernel() - reserves memory for crash kernel 1369 * 1370 * This function reserves memory area given in "crashkernel=" kernel command 1371 * line parameter. The memory reserved is used by dump capture kernel when 1372 * primary kernel is crashing. 1373 */ 1374 static void __init arch_reserve_crashkernel(void) 1375 { 1376 unsigned long long low_size = 0; 1377 unsigned long long crash_base, crash_size; 1378 char *cmdline = boot_command_line; 1379 bool high = false; 1380 int ret; 1381 1382 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1383 return; 1384 1385 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 1386 &crash_size, &crash_base, 1387 &low_size, &high); 1388 if (ret) 1389 return; 1390 1391 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 1392 low_size, high); 1393 } 1394 1395 void __init paging_init(void) 1396 { 1397 setup_bootmem(); 1398 setup_vm_final(); 1399 1400 /* Depend on that Linear Mapping is ready */ 1401 memblock_allow_resize(); 1402 } 1403 1404 void __init misc_mem_init(void) 1405 { 1406 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1407 arch_numa_init(); 1408 sparse_init(); 1409 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1410 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1411 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1412 #endif 1413 zone_sizes_init(); 1414 arch_reserve_crashkernel(); 1415 memblock_dump_all(); 1416 } 1417 1418 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1419 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1420 unsigned long addr, unsigned long next) 1421 { 1422 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1423 } 1424 1425 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1426 unsigned long addr, unsigned long next) 1427 { 1428 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1429 return 1; 1430 } 1431 1432 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1433 struct vmem_altmap *altmap) 1434 { 1435 /* 1436 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1437 * can't use hugepage mappings for 2-level page table because in case of 1438 * memory hotplug, we are not able to update all the page tables with 1439 * the new PMDs. 1440 */ 1441 return vmemmap_populate_hugepages(start, end, node, NULL); 1442 } 1443 #endif 1444 1445 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1446 /* 1447 * Pre-allocates page-table pages for a specific area in the kernel 1448 * page-table. Only the level which needs to be synchronized between 1449 * all page-tables is allocated because the synchronization can be 1450 * expensive. 1451 */ 1452 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1453 const char *area) 1454 { 1455 unsigned long addr; 1456 const char *lvl; 1457 1458 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1459 pgd_t *pgd = pgd_offset_k(addr); 1460 p4d_t *p4d; 1461 pud_t *pud; 1462 pmd_t *pmd; 1463 1464 lvl = "p4d"; 1465 p4d = p4d_alloc(&init_mm, pgd, addr); 1466 if (!p4d) 1467 goto failed; 1468 1469 if (pgtable_l5_enabled) 1470 continue; 1471 1472 lvl = "pud"; 1473 pud = pud_alloc(&init_mm, p4d, addr); 1474 if (!pud) 1475 goto failed; 1476 1477 if (pgtable_l4_enabled) 1478 continue; 1479 1480 lvl = "pmd"; 1481 pmd = pmd_alloc(&init_mm, pud, addr); 1482 if (!pmd) 1483 goto failed; 1484 } 1485 return; 1486 1487 failed: 1488 /* 1489 * The pages have to be there now or they will be missing in 1490 * process page-tables later. 1491 */ 1492 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1493 } 1494 1495 void __init pgtable_cache_init(void) 1496 { 1497 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1498 if (IS_ENABLED(CONFIG_MODULES)) 1499 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1500 } 1501 #endif 1502 1503 #ifdef CONFIG_EXECMEM 1504 #ifdef CONFIG_MMU 1505 static struct execmem_info execmem_info __ro_after_init; 1506 1507 struct execmem_info __init *execmem_arch_setup(void) 1508 { 1509 execmem_info = (struct execmem_info){ 1510 .ranges = { 1511 [EXECMEM_DEFAULT] = { 1512 .start = MODULES_VADDR, 1513 .end = MODULES_END, 1514 .pgprot = PAGE_KERNEL, 1515 .alignment = 1, 1516 }, 1517 [EXECMEM_KPROBES] = { 1518 .start = VMALLOC_START, 1519 .end = VMALLOC_END, 1520 .pgprot = PAGE_KERNEL_READ_EXEC, 1521 .alignment = 1, 1522 }, 1523 [EXECMEM_BPF] = { 1524 .start = BPF_JIT_REGION_START, 1525 .end = BPF_JIT_REGION_END, 1526 .pgprot = PAGE_KERNEL, 1527 .alignment = PAGE_SIZE, 1528 }, 1529 }, 1530 }; 1531 1532 return &execmem_info; 1533 } 1534 #endif /* CONFIG_MMU */ 1535 #endif /* CONFIG_EXECMEM */ 1536