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 #include <linux/kfence.h> 24 #include <linux/execmem.h> 25 26 #include <asm/alternative.h> 27 #include <asm/fixmap.h> 28 #include <asm/io.h> 29 #include <asm/kasan.h> 30 #include <asm/module.h> 31 #include <asm/numa.h> 32 #include <asm/pgtable.h> 33 #include <asm/sections.h> 34 #include <asm/soc.h> 35 #include <asm/sparsemem.h> 36 #include <asm/tlbflush.h> 37 38 #include "../kernel/head.h" 39 40 #if defined(CONFIG_64BIT) && defined(CONFIG_MMU) 41 DECLARE_BITMAP(new_valid_map_cpus, NR_CPUS); 42 #endif 43 44 struct kernel_mapping kernel_map __ro_after_init; 45 EXPORT_SYMBOL(kernel_map); 46 47 #ifdef CONFIG_64BIT 48 u64 satp_mode __ro_after_init = SATP_MODE_57; 49 #else 50 u64 satp_mode __ro_after_init = SATP_MODE_32; 51 #endif 52 EXPORT_SYMBOL(satp_mode); 53 54 #ifdef CONFIG_64BIT 55 bool pgtable_l4_enabled __ro_after_init = true; 56 bool pgtable_l5_enabled __ro_after_init = true; 57 EXPORT_SYMBOL(pgtable_l4_enabled); 58 EXPORT_SYMBOL(pgtable_l5_enabled); 59 #endif 60 61 phys_addr_t phys_ram_base __ro_after_init; 62 EXPORT_SYMBOL(phys_ram_base); 63 64 #ifdef CONFIG_SPARSEMEM_VMEMMAP 65 #define VMEMMAP_ADDR_ALIGN max(1ULL << SECTION_SIZE_BITS, \ 66 MAX_FOLIO_VMEMMAP_ALIGN) 67 68 unsigned long vmemmap_start_pfn __ro_after_init; 69 EXPORT_SYMBOL(vmemmap_start_pfn); 70 #endif 71 72 extern char _start[]; 73 void *_dtb_early_va __initdata; 74 uintptr_t _dtb_early_pa __initdata; 75 76 phys_addr_t dma32_phys_limit __initdata; 77 78 void __init arch_zone_limits_init(unsigned long *max_zone_pfns) 79 { 80 #ifdef CONFIG_ZONE_DMA32 81 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 82 #endif 83 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 84 } 85 86 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 87 88 #define LOG2_SZ_1K ilog2(SZ_1K) 89 #define LOG2_SZ_1M ilog2(SZ_1M) 90 #define LOG2_SZ_1G ilog2(SZ_1G) 91 #define LOG2_SZ_1T ilog2(SZ_1T) 92 93 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 94 { 95 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 96 (((t) - (b)) >> LOG2_SZ_1K)); 97 } 98 99 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 100 { 101 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 102 (((t) - (b)) >> LOG2_SZ_1M)); 103 } 104 105 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 106 { 107 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 108 (((t) - (b)) >> LOG2_SZ_1G)); 109 } 110 111 #ifdef CONFIG_64BIT 112 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 113 { 114 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 115 (((t) - (b)) >> LOG2_SZ_1T)); 116 } 117 #else 118 #define print_mlt(n, b, t) do {} while (0) 119 #endif 120 121 static inline void print_ml(char *name, unsigned long b, unsigned long t) 122 { 123 unsigned long diff = t - b; 124 125 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 126 print_mlt(name, b, t); 127 else if ((diff >> LOG2_SZ_1G) >= 10) 128 print_mlg(name, b, t); 129 else if ((diff >> LOG2_SZ_1M) >= 10) 130 print_mlm(name, b, t); 131 else 132 print_mlk(name, b, t); 133 } 134 135 static void __init print_vm_layout(void) 136 { 137 pr_notice("Virtual kernel memory layout:\n"); 138 print_ml("fixmap", (unsigned long)FIXADDR_START, 139 (unsigned long)FIXADDR_TOP); 140 print_ml("pci io", (unsigned long)PCI_IO_START, 141 (unsigned long)PCI_IO_END); 142 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 143 (unsigned long)VMEMMAP_END); 144 print_ml("vmalloc", (unsigned long)VMALLOC_START, 145 (unsigned long)VMALLOC_END); 146 #ifdef CONFIG_64BIT 147 print_ml("modules", (unsigned long)MODULES_VADDR, 148 (unsigned long)MODULES_END); 149 #endif 150 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 151 (unsigned long)high_memory); 152 if (IS_ENABLED(CONFIG_64BIT)) { 153 #ifdef CONFIG_KASAN 154 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 155 #endif 156 157 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 158 (unsigned long)ADDRESS_SPACE_END); 159 } 160 } 161 #else 162 static void print_vm_layout(void) { } 163 #endif /* CONFIG_DEBUG_VM */ 164 165 void __init arch_mm_preinit(void) 166 { 167 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 168 #ifdef CONFIG_FLATMEM 169 BUG_ON(!mem_map); 170 #endif /* CONFIG_FLATMEM */ 171 172 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 173 dma_cache_alignment != 1) { 174 /* 175 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 176 * buffer per 1GB of RAM for kmalloc() bouncing on 177 * non-coherent platforms. 178 */ 179 unsigned long size = 180 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 181 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 182 swiotlb = true; 183 } 184 185 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 186 187 print_vm_layout(); 188 } 189 190 /* Limit the memory size via mem. */ 191 static phys_addr_t memory_limit; 192 193 static int __init early_mem(char *p) 194 { 195 u64 size; 196 197 if (!p) 198 return 1; 199 200 size = memparse(p, &p) & PAGE_MASK; 201 memory_limit = min_t(u64, size, memory_limit); 202 203 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 204 205 return 0; 206 } 207 early_param("mem", early_mem); 208 209 static void __init setup_bootmem(void) 210 { 211 phys_addr_t vmlinux_end = __pa_symbol(&_end); 212 phys_addr_t max_mapped_addr; 213 phys_addr_t phys_ram_end, vmlinux_start; 214 215 vmlinux_start = __pa_symbol(&_start); 216 217 memblock_enforce_memory_limit(memory_limit); 218 219 /* 220 * Make sure we align the reservation on PMD_SIZE since we will 221 * map the kernel in the linear mapping as read-only: we do not want 222 * any allocation to happen between _end and the next pmd aligned page. 223 */ 224 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 225 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 226 /* 227 * Reserve from the start of the kernel to the end of the kernel 228 */ 229 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 230 231 /* 232 * Make sure we align the start of the memory on a PMD boundary so that 233 * at worst, we map the linear mapping with PMD mappings. 234 */ 235 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 236 #ifdef CONFIG_SPARSEMEM_VMEMMAP 237 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 238 #endif 239 240 /* 241 * In 64-bit, any use of __va/__pa before this point is wrong as we 242 * did not know the start of DRAM before. 243 */ 244 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 245 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 246 247 /* 248 * The size of the linear page mapping may restrict the amount of 249 * usable RAM. 250 */ 251 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) { 252 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE; 253 if (memblock_end_of_DRAM() > max_mapped_addr) { 254 memblock_cap_memory_range(phys_ram_base, 255 max_mapped_addr - phys_ram_base); 256 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed", 257 &max_mapped_addr); 258 } 259 } 260 261 /* 262 * Reserve physical address space that would be mapped to virtual 263 * addresses greater than (void *)(-PAGE_SIZE) because: 264 * - This memory would overlap with ERR_PTR 265 * - This memory belongs to high memory, which is not supported 266 * 267 * This is not applicable to 64-bit kernel, because virtual addresses 268 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 269 * occupied by kernel mapping. Also it is unrealistic for high memory 270 * to exist on 64-bit platforms. 271 */ 272 if (!IS_ENABLED(CONFIG_64BIT)) { 273 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 274 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 275 } 276 277 phys_ram_end = memblock_end_of_DRAM(); 278 min_low_pfn = PFN_UP(phys_ram_base); 279 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 280 281 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 282 283 reserve_initrd_mem(); 284 285 /* 286 * No allocation should be done before reserving the memory as defined 287 * in the device tree, otherwise the allocation could end up in a 288 * reserved region. 289 */ 290 early_init_fdt_scan_reserved_mem(); 291 292 /* 293 * If DTB is built in, no need to reserve its memblock. 294 * Otherwise, do reserve it but avoid using 295 * early_init_fdt_reserve_self() since __pa() does 296 * not work for DTB pointers that are fixmap addresses 297 */ 298 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 299 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 300 301 dma_contiguous_reserve(dma32_phys_limit); 302 } 303 304 #ifdef CONFIG_RELOCATABLE 305 extern unsigned long __rela_dyn_start, __rela_dyn_end; 306 307 static void __init relocate_kernel(void) 308 { 309 Elf_Rela *rela = (Elf_Rela *)&__rela_dyn_start; 310 /* 311 * This holds the offset between the linked virtual address and the 312 * relocated virtual address. 313 */ 314 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 315 /* 316 * This holds the offset between kernel linked virtual address and 317 * physical address. 318 */ 319 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 320 321 for ( ; rela < (Elf_Rela *)&__rela_dyn_end; rela++) { 322 Elf_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 323 Elf_Addr relocated_addr = rela->r_addend; 324 325 if (rela->r_info != R_RISCV_RELATIVE) 326 continue; 327 328 /* 329 * Make sure to not relocate vdso symbols like rt_sigreturn 330 * which are linked from the address 0 in vmlinux since 331 * vdso symbol addresses are actually used as an offset from 332 * mm->context.vdso in VDSO_OFFSET macro. 333 */ 334 if (relocated_addr >= KERNEL_LINK_ADDR) 335 relocated_addr += reloc_offset; 336 337 *(Elf_Addr *)addr = relocated_addr; 338 } 339 } 340 #endif /* CONFIG_RELOCATABLE */ 341 342 #ifdef CONFIG_MMU 343 struct pt_alloc_ops pt_ops __meminitdata; 344 345 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 346 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 347 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 348 349 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 350 351 static const pgprot_t protection_map[16] = { 352 [VM_NONE] = PAGE_NONE, 353 [VM_READ] = PAGE_READ, 354 [VM_WRITE] = PAGE_SHADOWSTACK, 355 [VM_WRITE | VM_READ] = PAGE_COPY, 356 [VM_EXEC] = PAGE_EXEC, 357 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 358 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 359 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 360 [VM_SHARED] = PAGE_NONE, 361 [VM_SHARED | VM_READ] = PAGE_READ, 362 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 363 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 364 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 365 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 366 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 367 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 368 }; 369 DECLARE_VM_GET_PAGE_PROT 370 371 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 372 { 373 unsigned long addr = __fix_to_virt(idx); 374 pte_t *ptep; 375 376 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 377 378 ptep = &fixmap_pte[pte_index(addr)]; 379 380 if (pgprot_val(prot)) 381 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 382 else 383 pte_clear(&init_mm, addr, ptep); 384 local_flush_tlb_page(addr); 385 } 386 387 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 388 { 389 return (pte_t *)((uintptr_t)pa); 390 } 391 392 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 393 { 394 clear_fixmap(FIX_PTE); 395 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 396 } 397 398 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa) 399 { 400 return (pte_t *) __va(pa); 401 } 402 403 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 404 { 405 /* 406 * We only create PMD or PGD early mappings so we 407 * should never reach here with MMU disabled. 408 */ 409 BUG(); 410 } 411 412 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 413 { 414 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 415 } 416 417 static phys_addr_t __meminit alloc_pte_late(uintptr_t va) 418 { 419 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 420 421 /* 422 * We do not know which mm the PTE page is associated to at this point. 423 * Passing NULL to the ctor is the safe option, though it may result 424 * in unnecessary work (e.g. initialising the ptlock for init_mm). 425 */ 426 BUG_ON(!ptdesc || !pagetable_pte_ctor(NULL, ptdesc)); 427 return __pa((pte_t *)ptdesc_address(ptdesc)); 428 } 429 430 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 431 pgprot_t prot) 432 { 433 uintptr_t pte_idx = pte_index(va); 434 435 BUG_ON(sz != PAGE_SIZE); 436 437 if (pte_none(ptep[pte_idx])) 438 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 439 } 440 441 #ifndef __PAGETABLE_PMD_FOLDED 442 443 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 444 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 445 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 446 447 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 448 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 449 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 450 451 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 452 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 453 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 454 455 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 456 { 457 /* Before MMU is enabled */ 458 return (pmd_t *)((uintptr_t)pa); 459 } 460 461 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 462 { 463 clear_fixmap(FIX_PMD); 464 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 465 } 466 467 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa) 468 { 469 return (pmd_t *) __va(pa); 470 } 471 472 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 473 { 474 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 475 476 return (uintptr_t)early_pmd; 477 } 478 479 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 480 { 481 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 482 } 483 484 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va) 485 { 486 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 487 488 /* See comment in alloc_pte_late() regarding NULL passed the ctor */ 489 BUG_ON(!ptdesc || !pagetable_pmd_ctor(NULL, ptdesc)); 490 return __pa((pmd_t *)ptdesc_address(ptdesc)); 491 } 492 493 static void __meminit create_pmd_mapping(pmd_t *pmdp, 494 uintptr_t va, phys_addr_t pa, 495 phys_addr_t sz, pgprot_t prot) 496 { 497 pte_t *ptep; 498 phys_addr_t pte_phys; 499 uintptr_t pmd_idx = pmd_index(va); 500 501 if (sz == PMD_SIZE) { 502 if (pmd_none(pmdp[pmd_idx])) 503 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 504 return; 505 } 506 507 if (pmd_none(pmdp[pmd_idx])) { 508 pte_phys = pt_ops.alloc_pte(va); 509 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 510 ptep = pt_ops.get_pte_virt(pte_phys); 511 memset(ptep, 0, PAGE_SIZE); 512 } else { 513 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 514 ptep = pt_ops.get_pte_virt(pte_phys); 515 } 516 517 create_pte_mapping(ptep, va, pa, sz, prot); 518 } 519 520 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 521 { 522 return (pud_t *)((uintptr_t)pa); 523 } 524 525 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 526 { 527 clear_fixmap(FIX_PUD); 528 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 529 } 530 531 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 532 { 533 return (pud_t *)__va(pa); 534 } 535 536 static phys_addr_t __init alloc_pud_early(uintptr_t va) 537 { 538 /* Only one PUD is available for early mapping */ 539 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 540 541 return (uintptr_t)early_pud; 542 } 543 544 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 545 { 546 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 547 } 548 549 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 550 { 551 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 552 553 BUG_ON(!ptdesc); 554 pagetable_pud_ctor(ptdesc); 555 return __pa((pud_t *)ptdesc_address(ptdesc)); 556 } 557 558 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 559 { 560 return (p4d_t *)((uintptr_t)pa); 561 } 562 563 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 564 { 565 clear_fixmap(FIX_P4D); 566 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 567 } 568 569 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 570 { 571 return (p4d_t *)__va(pa); 572 } 573 574 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 575 { 576 /* Only one P4D is available for early mapping */ 577 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 578 579 return (uintptr_t)early_p4d; 580 } 581 582 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 583 { 584 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 585 } 586 587 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 588 { 589 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 590 591 BUG_ON(!ptdesc); 592 pagetable_p4d_ctor(ptdesc); 593 return __pa((p4d_t *)ptdesc_address(ptdesc)); 594 } 595 596 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 597 pgprot_t prot) 598 { 599 pmd_t *nextp; 600 phys_addr_t next_phys; 601 uintptr_t pud_index = pud_index(va); 602 603 if (sz == PUD_SIZE) { 604 if (pud_val(pudp[pud_index]) == 0) 605 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 606 return; 607 } 608 609 if (pud_val(pudp[pud_index]) == 0) { 610 next_phys = pt_ops.alloc_pmd(va); 611 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 612 nextp = pt_ops.get_pmd_virt(next_phys); 613 memset(nextp, 0, PAGE_SIZE); 614 } else { 615 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 616 nextp = pt_ops.get_pmd_virt(next_phys); 617 } 618 619 create_pmd_mapping(nextp, va, pa, sz, prot); 620 } 621 622 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 623 pgprot_t prot) 624 { 625 pud_t *nextp; 626 phys_addr_t next_phys; 627 uintptr_t p4d_index = p4d_index(va); 628 629 if (sz == P4D_SIZE) { 630 if (p4d_val(p4dp[p4d_index]) == 0) 631 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 632 return; 633 } 634 635 if (p4d_val(p4dp[p4d_index]) == 0) { 636 next_phys = pt_ops.alloc_pud(va); 637 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 638 nextp = pt_ops.get_pud_virt(next_phys); 639 memset(nextp, 0, PAGE_SIZE); 640 } else { 641 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 642 nextp = pt_ops.get_pud_virt(next_phys); 643 } 644 645 create_pud_mapping(nextp, va, pa, sz, prot); 646 } 647 648 #define pgd_next_t p4d_t 649 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 650 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 651 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 652 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 653 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 654 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 655 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 656 (pgtable_l5_enabled ? \ 657 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 658 (pgtable_l4_enabled ? \ 659 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 660 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 661 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 662 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 663 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 664 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 665 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 666 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 667 #else 668 #define pgd_next_t pte_t 669 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 670 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 671 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 672 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 673 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 674 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 675 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 676 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 677 #endif /* __PAGETABLE_PMD_FOLDED */ 678 679 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 680 pgprot_t prot) 681 { 682 pgd_next_t *nextp; 683 phys_addr_t next_phys; 684 uintptr_t pgd_idx = pgd_index(va); 685 686 if (sz == PGDIR_SIZE) { 687 if (pgd_val(pgdp[pgd_idx]) == 0) 688 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 689 return; 690 } 691 692 if (pgd_val(pgdp[pgd_idx]) == 0) { 693 next_phys = alloc_pgd_next(va); 694 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 695 nextp = get_pgd_next_virt(next_phys); 696 memset(nextp, 0, PAGE_SIZE); 697 } else { 698 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 699 nextp = get_pgd_next_virt(next_phys); 700 } 701 702 create_pgd_next_mapping(nextp, va, pa, sz, prot); 703 } 704 705 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 706 { 707 if (debug_pagealloc_enabled()) 708 return PAGE_SIZE; 709 710 if (pgtable_l5_enabled && 711 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 712 return P4D_SIZE; 713 714 if (pgtable_l4_enabled && 715 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 716 return PUD_SIZE; 717 718 if (IS_ENABLED(CONFIG_64BIT) && 719 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 720 return PMD_SIZE; 721 722 return PAGE_SIZE; 723 } 724 725 #ifdef CONFIG_STRICT_KERNEL_RWX 726 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 727 { 728 if (is_va_kernel_text(va)) 729 return PAGE_KERNEL_READ_EXEC; 730 731 /* 732 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 733 * we must protect its linear mapping alias from being executed and 734 * written. 735 * And rodata section is marked readonly in mark_rodata_ro. 736 */ 737 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 738 return PAGE_KERNEL_READ; 739 740 return PAGE_KERNEL; 741 } 742 743 void mark_rodata_ro(void) 744 { 745 set_kernel_memory(__start_rodata, _data, set_memory_ro); 746 if (IS_ENABLED(CONFIG_64BIT)) 747 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 748 set_memory_ro); 749 } 750 #else 751 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 752 { 753 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 754 return PAGE_KERNEL; 755 756 return PAGE_KERNEL_EXEC; 757 } 758 #endif /* CONFIG_STRICT_KERNEL_RWX */ 759 760 #if defined(CONFIG_64BIT) 761 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 762 u64 __pi_set_satp_mode_from_fdt(uintptr_t dtb_pa); 763 764 static void __init disable_pgtable_l5(void) 765 { 766 pgtable_l5_enabled = false; 767 kernel_map.page_offset = PAGE_OFFSET_L4; 768 satp_mode = SATP_MODE_48; 769 } 770 771 static void __init disable_pgtable_l4(void) 772 { 773 pgtable_l4_enabled = false; 774 kernel_map.page_offset = PAGE_OFFSET_L3; 775 satp_mode = SATP_MODE_39; 776 } 777 778 static int __init print_no4lvl(char *p) 779 { 780 pr_info("Disabled 4-level and 5-level paging"); 781 return 0; 782 } 783 early_param("no4lvl", print_no4lvl); 784 785 static int __init print_no5lvl(char *p) 786 { 787 pr_info("Disabled 5-level paging"); 788 return 0; 789 } 790 early_param("no5lvl", print_no5lvl); 791 792 static void __init set_mmap_rnd_bits_max(void) 793 { 794 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 795 } 796 797 static bool __init is_vaddr_valid(unsigned long va) 798 { 799 unsigned long up = 0; 800 801 switch (satp_mode) { 802 case SATP_MODE_39: 803 up = 1UL << 38; 804 break; 805 case SATP_MODE_48: 806 up = 1UL << 47; 807 break; 808 case SATP_MODE_57: 809 up = 1UL << 56; 810 break; 811 default: 812 return false; 813 } 814 815 return (va < up) || (va >= (ULONG_MAX - up + 1)); 816 } 817 818 /* 819 * There is a simple way to determine if 4-level is supported by the 820 * underlying hardware: establish 1:1 mapping in 4-level page table mode 821 * then read SATP to see if the configuration was taken into account 822 * meaning sv48 is supported. 823 * The maximum SATP mode is limited by both the command line and the "mmu-type" 824 * property in the device tree, since some platforms may hang if an unsupported 825 * SATP mode is attempted. 826 */ 827 static __init void set_satp_mode(uintptr_t dtb_pa) 828 { 829 u64 identity_satp, hw_satp; 830 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 831 u64 satp_mode_limit = min_not_zero(__pi_set_satp_mode_from_cmdline(dtb_pa), 832 __pi_set_satp_mode_from_fdt(dtb_pa)); 833 834 kernel_map.page_offset = PAGE_OFFSET_L5; 835 836 if (satp_mode_limit == SATP_MODE_48) { 837 disable_pgtable_l5(); 838 } else if (satp_mode_limit == SATP_MODE_39) { 839 disable_pgtable_l5(); 840 disable_pgtable_l4(); 841 return; 842 } 843 844 create_p4d_mapping(early_p4d, 845 set_satp_mode_pmd, (uintptr_t)early_pud, 846 P4D_SIZE, PAGE_TABLE); 847 create_pud_mapping(early_pud, 848 set_satp_mode_pmd, (uintptr_t)early_pmd, 849 PUD_SIZE, PAGE_TABLE); 850 /* Handle the case where set_satp_mode straddles 2 PMDs */ 851 create_pmd_mapping(early_pmd, 852 set_satp_mode_pmd, set_satp_mode_pmd, 853 PMD_SIZE, PAGE_KERNEL_EXEC); 854 create_pmd_mapping(early_pmd, 855 set_satp_mode_pmd + PMD_SIZE, 856 set_satp_mode_pmd + PMD_SIZE, 857 PMD_SIZE, PAGE_KERNEL_EXEC); 858 retry: 859 if (!is_vaddr_valid(set_satp_mode_pmd)) 860 goto out; 861 862 create_pgd_mapping(early_pg_dir, 863 set_satp_mode_pmd, 864 pgtable_l5_enabled ? 865 (uintptr_t)early_p4d : (uintptr_t)early_pud, 866 PGDIR_SIZE, PAGE_TABLE); 867 868 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 869 870 local_flush_tlb_all(); 871 csr_write(CSR_SATP, identity_satp); 872 hw_satp = csr_swap(CSR_SATP, 0ULL); 873 local_flush_tlb_all(); 874 875 if (hw_satp != identity_satp) { 876 if (pgtable_l5_enabled) { 877 disable_pgtable_l5(); 878 memset(early_pg_dir, 0, PAGE_SIZE); 879 goto retry; 880 } 881 disable_pgtable_l4(); 882 } 883 884 out: 885 memset(early_pg_dir, 0, PAGE_SIZE); 886 memset(early_p4d, 0, PAGE_SIZE); 887 memset(early_pud, 0, PAGE_SIZE); 888 memset(early_pmd, 0, PAGE_SIZE); 889 } 890 #endif 891 892 /* 893 * setup_vm() is called from head.S with MMU-off. 894 * 895 * Following requirements should be honoured for setup_vm() to work 896 * correctly: 897 * 1) It should use PC-relative addressing for accessing kernel symbols. 898 * To achieve this we always use GCC cmodel=medany. 899 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 900 * so disable compiler instrumentation when FTRACE is enabled. 901 * 902 * Currently, the above requirements are honoured by using custom CFLAGS 903 * for init.o in mm/Makefile. 904 */ 905 906 #ifndef __riscv_cmodel_medany 907 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 908 #endif 909 910 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 911 { 912 uintptr_t va, end_va; 913 914 end_va = kernel_map.virt_addr + kernel_map.size; 915 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 916 create_pgd_mapping(pgdir, va, 917 kernel_map.phys_addr + (va - kernel_map.virt_addr), 918 PMD_SIZE, 919 early ? 920 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 921 } 922 923 /* 924 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 925 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 926 * entry. 927 */ 928 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 929 uintptr_t dtb_pa) 930 { 931 #ifndef CONFIG_BUILTIN_DTB 932 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 933 934 /* Make sure the fdt fixmap address is always aligned on PMD size */ 935 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 936 937 /* In 32-bit only, the fdt lies in its own PGD */ 938 if (!IS_ENABLED(CONFIG_64BIT)) { 939 create_pgd_mapping(early_pg_dir, fix_fdt_va, 940 pa, MAX_FDT_SIZE, PAGE_KERNEL); 941 } else { 942 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 943 pa, PMD_SIZE, PAGE_KERNEL); 944 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 945 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 946 } 947 948 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 949 #else 950 /* 951 * For 64-bit kernel, __va can't be used since it would return a linear 952 * mapping address whereas dtb_early_va will be used before 953 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 954 * kernel is mapped in the linear mapping, that makes no difference. 955 */ 956 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 957 #endif 958 959 dtb_early_pa = dtb_pa; 960 } 961 962 /* 963 * MMU is not enabled, the page tables are allocated directly using 964 * early_pmd/pud/p4d and the address returned is the physical one. 965 */ 966 static void __init pt_ops_set_early(void) 967 { 968 pt_ops.alloc_pte = alloc_pte_early; 969 pt_ops.get_pte_virt = get_pte_virt_early; 970 #ifndef __PAGETABLE_PMD_FOLDED 971 pt_ops.alloc_pmd = alloc_pmd_early; 972 pt_ops.get_pmd_virt = get_pmd_virt_early; 973 pt_ops.alloc_pud = alloc_pud_early; 974 pt_ops.get_pud_virt = get_pud_virt_early; 975 pt_ops.alloc_p4d = alloc_p4d_early; 976 pt_ops.get_p4d_virt = get_p4d_virt_early; 977 #endif 978 } 979 980 /* 981 * MMU is enabled but page table setup is not complete yet. 982 * fixmap page table alloc functions must be used as a means to temporarily 983 * map the allocated physical pages since the linear mapping does not exist yet. 984 * 985 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 986 * but it will be used as described above. 987 */ 988 static void __init pt_ops_set_fixmap(void) 989 { 990 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 991 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 992 #ifndef __PAGETABLE_PMD_FOLDED 993 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 994 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 995 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 996 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 997 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 998 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 999 #endif 1000 } 1001 1002 /* 1003 * MMU is enabled and page table setup is complete, so from now, we can use 1004 * generic page allocation functions to setup page table. 1005 */ 1006 static void __init pt_ops_set_late(void) 1007 { 1008 pt_ops.alloc_pte = alloc_pte_late; 1009 pt_ops.get_pte_virt = get_pte_virt_late; 1010 #ifndef __PAGETABLE_PMD_FOLDED 1011 pt_ops.alloc_pmd = alloc_pmd_late; 1012 pt_ops.get_pmd_virt = get_pmd_virt_late; 1013 pt_ops.alloc_pud = alloc_pud_late; 1014 pt_ops.get_pud_virt = get_pud_virt_late; 1015 pt_ops.alloc_p4d = alloc_p4d_late; 1016 pt_ops.get_p4d_virt = get_p4d_virt_late; 1017 #endif 1018 } 1019 1020 #ifdef CONFIG_RANDOMIZE_BASE 1021 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1022 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1023 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa); 1024 1025 static int __init print_nokaslr(char *p) 1026 { 1027 pr_info("Disabled KASLR"); 1028 return 0; 1029 } 1030 early_param("nokaslr", print_nokaslr); 1031 #endif 1032 1033 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1034 { 1035 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1036 1037 #ifdef CONFIG_RANDOMIZE_BASE 1038 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1039 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa); 1040 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1041 u32 nr_pos; 1042 1043 if (kaslr_seed == 0) 1044 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1045 /* 1046 * Compute the number of positions available: we are limited 1047 * by the early page table that only has one PUD and we must 1048 * be aligned on PMD_SIZE. 1049 */ 1050 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1051 1052 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1053 } 1054 #endif 1055 1056 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1057 1058 kernel_map.phys_addr = (uintptr_t)(&_start); 1059 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1060 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1061 1062 #if defined(CONFIG_64BIT) 1063 set_satp_mode(dtb_pa); 1064 set_mmap_rnd_bits_max(); 1065 #endif 1066 1067 /* 1068 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1069 * where we have the system memory layout: this allows us to align 1070 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1071 * for the linear mapping. This is only possible because the kernel 1072 * mapping lies outside the linear mapping. 1073 * In 32-bit however, as the kernel resides in the linear mapping, 1074 * setup_vm_final can not change the mapping established here, 1075 * otherwise the same kernel addresses would get mapped to different 1076 * physical addresses (if the start of dram is different from the 1077 * kernel physical address start). 1078 */ 1079 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1080 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1081 1082 memory_limit = KERN_VIRT_SIZE; 1083 1084 /* Sanity check alignment and size */ 1085 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1086 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1087 1088 #ifdef CONFIG_64BIT 1089 /* 1090 * The last 4K bytes of the addressable memory can not be mapped because 1091 * of IS_ERR_VALUE macro. 1092 */ 1093 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1094 #endif 1095 1096 #ifdef CONFIG_RELOCATABLE 1097 /* 1098 * Early page table uses only one PUD, which makes it possible 1099 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1100 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1101 * since a part of the kernel would not get mapped. 1102 */ 1103 if (IS_ENABLED(CONFIG_64BIT)) 1104 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1105 relocate_kernel(); 1106 #endif 1107 1108 apply_early_boot_alternatives(); 1109 pt_ops_set_early(); 1110 1111 /* Setup early PGD for fixmap */ 1112 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1113 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1114 1115 #ifndef __PAGETABLE_PMD_FOLDED 1116 /* Setup fixmap P4D and PUD */ 1117 if (pgtable_l5_enabled) 1118 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1119 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1120 /* Setup fixmap PUD and PMD */ 1121 if (pgtable_l4_enabled) 1122 create_pud_mapping(fixmap_pud, FIXADDR_START, 1123 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1124 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1125 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1126 /* Setup trampoline PGD and PMD */ 1127 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1128 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1129 if (pgtable_l5_enabled) 1130 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1131 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1132 if (pgtable_l4_enabled) 1133 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1134 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1135 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1136 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1137 #else 1138 /* Setup trampoline PGD */ 1139 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1140 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1141 #endif 1142 1143 /* 1144 * Setup early PGD covering entire kernel which will allow 1145 * us to reach paging_init(). We map all memory banks later 1146 * in setup_vm_final() below. 1147 */ 1148 create_kernel_page_table(early_pg_dir, true); 1149 1150 /* Setup early mapping for FDT early scan */ 1151 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1152 1153 /* 1154 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1155 * range can not span multiple pmds. 1156 */ 1157 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1158 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1159 1160 #ifndef __PAGETABLE_PMD_FOLDED 1161 /* 1162 * Early ioremap fixmap is already created as it lies within first 2MB 1163 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1164 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1165 * the user if not. 1166 */ 1167 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1168 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1169 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1170 WARN_ON(1); 1171 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1172 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1173 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1174 fix_to_virt(FIX_BTMAP_BEGIN)); 1175 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1176 fix_to_virt(FIX_BTMAP_END)); 1177 1178 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1179 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1180 } 1181 #endif 1182 1183 pt_ops_set_fixmap(); 1184 } 1185 1186 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end, 1187 uintptr_t fixed_map_size, const pgprot_t *pgprot) 1188 { 1189 phys_addr_t pa; 1190 uintptr_t va, map_size; 1191 1192 for (pa = start; pa < end; pa += map_size) { 1193 va = (uintptr_t)__va(pa); 1194 map_size = fixed_map_size ? fixed_map_size : 1195 best_map_size(pa, va, end - pa); 1196 1197 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1198 pgprot ? *pgprot : pgprot_from_va(va)); 1199 } 1200 } 1201 1202 static void __init create_linear_mapping_page_table(void) 1203 { 1204 phys_addr_t start, end; 1205 phys_addr_t kfence_pool __maybe_unused; 1206 u64 i; 1207 1208 #ifdef CONFIG_STRICT_KERNEL_RWX 1209 phys_addr_t ktext_start = __pa_symbol(_start); 1210 phys_addr_t ktext_size = __init_data_begin - _start; 1211 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1212 phys_addr_t krodata_size = _data - __start_rodata; 1213 1214 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1215 memblock_mark_nomap(ktext_start, ktext_size); 1216 memblock_mark_nomap(krodata_start, krodata_size); 1217 #endif 1218 1219 #ifdef CONFIG_KFENCE 1220 /* 1221 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1222 * before we setup the linear mapping so that we avoid using hugepages 1223 * for this region. 1224 */ 1225 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1226 BUG_ON(!kfence_pool); 1227 1228 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1229 __kfence_pool = __va(kfence_pool); 1230 #endif 1231 1232 /* Map all memory banks in the linear mapping */ 1233 for_each_mem_range(i, &start, &end) { 1234 if (start >= end) 1235 break; 1236 if (start <= __pa(PAGE_OFFSET) && 1237 __pa(PAGE_OFFSET) < end) 1238 start = __pa(PAGE_OFFSET); 1239 1240 create_linear_mapping_range(start, end, 0, NULL); 1241 } 1242 1243 #ifdef CONFIG_STRICT_KERNEL_RWX 1244 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL); 1245 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL); 1246 1247 memblock_clear_nomap(ktext_start, ktext_size); 1248 memblock_clear_nomap(krodata_start, krodata_size); 1249 #endif 1250 1251 #ifdef CONFIG_KFENCE 1252 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL); 1253 1254 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1255 #endif 1256 } 1257 1258 static void __init setup_vm_final(void) 1259 { 1260 /* Setup swapper PGD for fixmap */ 1261 #if !defined(CONFIG_64BIT) 1262 /* 1263 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1264 * directly in swapper_pg_dir in addition to the pgd entry that points 1265 * to fixmap_pte. 1266 */ 1267 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1268 1269 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1270 #endif 1271 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1272 __pa_symbol(fixmap_pgd_next), 1273 PGDIR_SIZE, PAGE_TABLE); 1274 1275 /* Map the linear mapping */ 1276 create_linear_mapping_page_table(); 1277 1278 /* Map the kernel */ 1279 if (IS_ENABLED(CONFIG_64BIT)) 1280 create_kernel_page_table(swapper_pg_dir, false); 1281 1282 #ifdef CONFIG_KASAN 1283 kasan_swapper_init(); 1284 #endif 1285 1286 /* Clear fixmap PTE and PMD mappings */ 1287 clear_fixmap(FIX_PTE); 1288 clear_fixmap(FIX_PMD); 1289 clear_fixmap(FIX_PUD); 1290 clear_fixmap(FIX_P4D); 1291 1292 /* Move to swapper page table */ 1293 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1294 local_flush_tlb_all(); 1295 1296 pt_ops_set_late(); 1297 } 1298 #else 1299 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1300 { 1301 dtb_early_va = (void *)dtb_pa; 1302 dtb_early_pa = dtb_pa; 1303 1304 #ifdef CONFIG_RELOCATABLE 1305 kernel_map.virt_addr = (uintptr_t)_start; 1306 kernel_map.phys_addr = (uintptr_t)_start; 1307 relocate_kernel(); 1308 #endif 1309 } 1310 1311 static inline void setup_vm_final(void) 1312 { 1313 } 1314 #endif /* CONFIG_MMU */ 1315 1316 /* 1317 * reserve_crashkernel() - reserves memory for crash kernel 1318 * 1319 * This function reserves memory area given in "crashkernel=" kernel command 1320 * line parameter. The memory reserved is used by dump capture kernel when 1321 * primary kernel is crashing. 1322 */ 1323 static void __init arch_reserve_crashkernel(void) 1324 { 1325 unsigned long long low_size = 0; 1326 unsigned long long crash_base, crash_size; 1327 bool high = false; 1328 int ret; 1329 1330 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1331 return; 1332 1333 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1334 &crash_size, &crash_base, 1335 &low_size, NULL, &high); 1336 if (ret) 1337 return; 1338 1339 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 1340 } 1341 1342 void __init paging_init(void) 1343 { 1344 setup_bootmem(); 1345 setup_vm_final(); 1346 1347 /* Depend on that Linear Mapping is ready */ 1348 memblock_allow_resize(); 1349 } 1350 1351 void __init misc_mem_init(void) 1352 { 1353 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1354 arch_numa_init(); 1355 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1356 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1357 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1358 #endif 1359 arch_reserve_crashkernel(); 1360 memblock_dump_all(); 1361 } 1362 1363 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1364 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1365 struct vmem_altmap *altmap) 1366 { 1367 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END)); 1368 1369 /* 1370 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1371 * can't use hugepage mappings for 2-level page table because in case of 1372 * memory hotplug, we are not able to update all the page tables with 1373 * the new PMDs. 1374 */ 1375 return vmemmap_populate_hugepages(start, end, node, altmap); 1376 } 1377 #endif 1378 1379 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1380 /* 1381 * Pre-allocates page-table pages for a specific area in the kernel 1382 * page-table. Only the level which needs to be synchronized between 1383 * all page-tables is allocated because the synchronization can be 1384 * expensive. 1385 */ 1386 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1387 const char *area) 1388 { 1389 unsigned long addr; 1390 const char *lvl; 1391 1392 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1393 pgd_t *pgd = pgd_offset_k(addr); 1394 p4d_t *p4d; 1395 pud_t *pud; 1396 pmd_t *pmd; 1397 1398 lvl = "p4d"; 1399 p4d = p4d_alloc(&init_mm, pgd, addr); 1400 if (!p4d) 1401 goto failed; 1402 1403 if (pgtable_l5_enabled) 1404 continue; 1405 1406 lvl = "pud"; 1407 pud = pud_alloc(&init_mm, p4d, addr); 1408 if (!pud) 1409 goto failed; 1410 1411 if (pgtable_l4_enabled) 1412 continue; 1413 1414 lvl = "pmd"; 1415 pmd = pmd_alloc(&init_mm, pud, addr); 1416 if (!pmd) 1417 goto failed; 1418 } 1419 return; 1420 1421 failed: 1422 /* 1423 * The pages have to be there now or they will be missing in 1424 * process page-tables later. 1425 */ 1426 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1427 } 1428 1429 #define PAGE_END KASAN_SHADOW_START 1430 1431 void __init pgtable_cache_init(void) 1432 { 1433 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1434 if (IS_ENABLED(CONFIG_MODULES)) 1435 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1436 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1437 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1438 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1439 if (IS_ENABLED(CONFIG_KASAN)) 1440 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1441 } 1442 } 1443 #endif 1444 1445 #ifdef CONFIG_EXECMEM 1446 #ifdef CONFIG_MMU 1447 static struct execmem_info execmem_info __ro_after_init; 1448 1449 struct execmem_info __init *execmem_arch_setup(void) 1450 { 1451 execmem_info = (struct execmem_info){ 1452 .ranges = { 1453 [EXECMEM_DEFAULT] = { 1454 .start = MODULES_VADDR, 1455 .end = MODULES_END, 1456 .pgprot = PAGE_KERNEL, 1457 .alignment = 1, 1458 }, 1459 [EXECMEM_KPROBES] = { 1460 .start = VMALLOC_START, 1461 .end = VMALLOC_END, 1462 .pgprot = PAGE_KERNEL_READ_EXEC, 1463 .alignment = 1, 1464 }, 1465 [EXECMEM_BPF] = { 1466 .start = BPF_JIT_REGION_START, 1467 .end = BPF_JIT_REGION_END, 1468 .pgprot = PAGE_KERNEL, 1469 .alignment = PAGE_SIZE, 1470 }, 1471 }, 1472 }; 1473 1474 return &execmem_info; 1475 } 1476 #endif /* CONFIG_MMU */ 1477 #endif /* CONFIG_EXECMEM */ 1478 1479 #ifdef CONFIG_MEMORY_HOTPLUG 1480 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1481 { 1482 struct page *page = pmd_page(*pmd); 1483 struct ptdesc *ptdesc = page_ptdesc(page); 1484 pte_t *pte; 1485 int i; 1486 1487 for (i = 0; i < PTRS_PER_PTE; i++) { 1488 pte = pte_start + i; 1489 if (!pte_none(*pte)) 1490 return; 1491 } 1492 1493 pagetable_dtor(ptdesc); 1494 if (PageReserved(page)) 1495 free_reserved_page(page); 1496 else 1497 pagetable_free(ptdesc); 1498 pmd_clear(pmd); 1499 } 1500 1501 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1502 { 1503 struct page *page = pud_page(*pud); 1504 struct ptdesc *ptdesc = page_ptdesc(page); 1505 pmd_t *pmd; 1506 int i; 1507 1508 for (i = 0; i < PTRS_PER_PMD; i++) { 1509 pmd = pmd_start + i; 1510 if (!pmd_none(*pmd)) 1511 return; 1512 } 1513 1514 if (!is_vmemmap) 1515 pagetable_dtor(ptdesc); 1516 if (PageReserved(page)) 1517 free_reserved_page(page); 1518 else 1519 pagetable_free(ptdesc); 1520 pud_clear(pud); 1521 } 1522 1523 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1524 { 1525 struct page *page = p4d_page(*p4d); 1526 pud_t *pud; 1527 int i; 1528 1529 for (i = 0; i < PTRS_PER_PUD; i++) { 1530 pud = pud_start + i; 1531 if (!pud_none(*pud)) 1532 return; 1533 } 1534 1535 if (PageReserved(page)) 1536 free_reserved_page(page); 1537 else 1538 __free_pages(page, 0); 1539 p4d_clear(p4d); 1540 } 1541 1542 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1543 struct vmem_altmap *altmap) 1544 { 1545 int order = get_order(size); 1546 1547 if (altmap) { 1548 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1549 return; 1550 } 1551 1552 if (PageReserved(page)) { 1553 unsigned int nr_pages = 1 << order; 1554 1555 while (nr_pages--) 1556 free_reserved_page(page++); 1557 return; 1558 } 1559 1560 __free_pages(page, order); 1561 } 1562 1563 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1564 bool is_vmemmap, struct vmem_altmap *altmap) 1565 { 1566 unsigned long next; 1567 pte_t *ptep, pte; 1568 1569 for (; addr < end; addr = next) { 1570 next = (addr + PAGE_SIZE) & PAGE_MASK; 1571 if (next > end) 1572 next = end; 1573 1574 ptep = pte_base + pte_index(addr); 1575 pte = ptep_get(ptep); 1576 if (!pte_present(*ptep)) 1577 continue; 1578 1579 pte_clear(&init_mm, addr, ptep); 1580 if (is_vmemmap) 1581 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1582 } 1583 } 1584 1585 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1586 bool is_vmemmap, struct vmem_altmap *altmap) 1587 { 1588 unsigned long next; 1589 pte_t *pte_base; 1590 pmd_t *pmdp, pmd; 1591 1592 for (; addr < end; addr = next) { 1593 next = pmd_addr_end(addr, end); 1594 pmdp = pmd_base + pmd_index(addr); 1595 pmd = pmdp_get(pmdp); 1596 if (!pmd_present(pmd)) 1597 continue; 1598 1599 if (pmd_leaf(pmd)) { 1600 pmd_clear(pmdp); 1601 if (is_vmemmap) 1602 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1603 continue; 1604 } 1605 1606 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1607 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1608 free_pte_table(pte_base, pmdp); 1609 } 1610 } 1611 1612 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1613 bool is_vmemmap, struct vmem_altmap *altmap) 1614 { 1615 unsigned long next; 1616 pud_t *pudp, pud; 1617 pmd_t *pmd_base; 1618 1619 for (; addr < end; addr = next) { 1620 next = pud_addr_end(addr, end); 1621 pudp = pud_base + pud_index(addr); 1622 pud = pudp_get(pudp); 1623 if (!pud_present(pud)) 1624 continue; 1625 1626 if (pud_leaf(pud)) { 1627 if (pgtable_l4_enabled) { 1628 pud_clear(pudp); 1629 if (is_vmemmap) 1630 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1631 } 1632 continue; 1633 } 1634 1635 pmd_base = pmd_offset(pudp, 0); 1636 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1637 1638 if (pgtable_l4_enabled) 1639 free_pmd_table(pmd_base, pudp, is_vmemmap); 1640 } 1641 } 1642 1643 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1644 bool is_vmemmap, struct vmem_altmap *altmap) 1645 { 1646 unsigned long next; 1647 p4d_t *p4dp, p4d; 1648 pud_t *pud_base; 1649 1650 for (; addr < end; addr = next) { 1651 next = p4d_addr_end(addr, end); 1652 p4dp = p4d_base + p4d_index(addr); 1653 p4d = p4dp_get(p4dp); 1654 if (!p4d_present(p4d)) 1655 continue; 1656 1657 if (p4d_leaf(p4d)) { 1658 if (pgtable_l5_enabled) { 1659 p4d_clear(p4dp); 1660 if (is_vmemmap) 1661 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1662 } 1663 continue; 1664 } 1665 1666 pud_base = pud_offset(p4dp, 0); 1667 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1668 1669 if (pgtable_l5_enabled) 1670 free_pud_table(pud_base, p4dp); 1671 } 1672 } 1673 1674 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1675 struct vmem_altmap *altmap) 1676 { 1677 unsigned long addr, next; 1678 p4d_t *p4d_base; 1679 pgd_t *pgd; 1680 1681 for (addr = va; addr < end; addr = next) { 1682 next = pgd_addr_end(addr, end); 1683 pgd = pgd_offset_k(addr); 1684 1685 if (!pgd_present(*pgd)) 1686 continue; 1687 1688 if (pgd_leaf(*pgd)) 1689 continue; 1690 1691 p4d_base = p4d_offset(pgd, 0); 1692 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1693 } 1694 1695 flush_tlb_all(); 1696 } 1697 1698 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1699 { 1700 unsigned long va = (unsigned long)__va(start); 1701 unsigned long end = (unsigned long)__va(start + size); 1702 1703 remove_pgd_mapping(va, end, false, NULL); 1704 } 1705 1706 struct range arch_get_mappable_range(void) 1707 { 1708 struct range mhp_range; 1709 1710 mhp_range.start = __pa(PAGE_OFFSET); 1711 mhp_range.end = __pa(PAGE_END - 1); 1712 return mhp_range; 1713 } 1714 1715 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1716 { 1717 int ret = 0; 1718 1719 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1720 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1721 if (ret) { 1722 remove_linear_mapping(start, size); 1723 goto out; 1724 } 1725 1726 max_pfn = PFN_UP(start + size); 1727 max_low_pfn = max_pfn; 1728 1729 out: 1730 flush_tlb_all(); 1731 return ret; 1732 } 1733 1734 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap, 1735 struct dev_pagemap *pgmap) 1736 { 1737 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap, pgmap); 1738 remove_linear_mapping(start, size); 1739 flush_tlb_all(); 1740 } 1741 1742 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1743 { 1744 remove_pgd_mapping(start, end, true, altmap); 1745 } 1746 #endif /* CONFIG_MEMORY_HOTPLUG */ 1747