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/fixmap.h> 27 #include <asm/io.h> 28 #include <asm/kasan.h> 29 #include <asm/module.h> 30 #include <asm/numa.h> 31 #include <asm/pgtable.h> 32 #include <asm/sections.h> 33 #include <asm/soc.h> 34 #include <asm/sparsemem.h> 35 #include <asm/tlbflush.h> 36 37 #include "../kernel/head.h" 38 39 u64 new_vmalloc[NR_CPUS / sizeof(u64) + 1]; 40 41 struct kernel_mapping kernel_map __ro_after_init; 42 EXPORT_SYMBOL(kernel_map); 43 #ifdef CONFIG_XIP_KERNEL 44 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 45 #endif 46 47 #ifdef CONFIG_64BIT 48 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 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 = !IS_ENABLED(CONFIG_XIP_KERNEL); 56 bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 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 (1ULL << SECTION_SIZE_BITS) 66 67 unsigned long vmemmap_start_pfn __ro_after_init; 68 EXPORT_SYMBOL(vmemmap_start_pfn); 69 #endif 70 71 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 72 __page_aligned_bss; 73 EXPORT_SYMBOL(empty_zero_page); 74 75 extern char _start[]; 76 void *_dtb_early_va __initdata; 77 uintptr_t _dtb_early_pa __initdata; 78 79 phys_addr_t dma32_phys_limit __initdata; 80 81 static void __init zone_sizes_init(void) 82 { 83 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 84 85 #ifdef CONFIG_ZONE_DMA32 86 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 87 #endif 88 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 89 90 free_area_init(max_zone_pfns); 91 } 92 93 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 94 95 #define LOG2_SZ_1K ilog2(SZ_1K) 96 #define LOG2_SZ_1M ilog2(SZ_1M) 97 #define LOG2_SZ_1G ilog2(SZ_1G) 98 #define LOG2_SZ_1T ilog2(SZ_1T) 99 100 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 101 { 102 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 103 (((t) - (b)) >> LOG2_SZ_1K)); 104 } 105 106 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 107 { 108 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 109 (((t) - (b)) >> LOG2_SZ_1M)); 110 } 111 112 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 113 { 114 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 115 (((t) - (b)) >> LOG2_SZ_1G)); 116 } 117 118 #ifdef CONFIG_64BIT 119 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 120 { 121 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 122 (((t) - (b)) >> LOG2_SZ_1T)); 123 } 124 #else 125 #define print_mlt(n, b, t) do {} while (0) 126 #endif 127 128 static inline void print_ml(char *name, unsigned long b, unsigned long t) 129 { 130 unsigned long diff = t - b; 131 132 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 133 print_mlt(name, b, t); 134 else if ((diff >> LOG2_SZ_1G) >= 10) 135 print_mlg(name, b, t); 136 else if ((diff >> LOG2_SZ_1M) >= 10) 137 print_mlm(name, b, t); 138 else 139 print_mlk(name, b, t); 140 } 141 142 static void __init print_vm_layout(void) 143 { 144 pr_notice("Virtual kernel memory layout:\n"); 145 print_ml("fixmap", (unsigned long)FIXADDR_START, 146 (unsigned long)FIXADDR_TOP); 147 print_ml("pci io", (unsigned long)PCI_IO_START, 148 (unsigned long)PCI_IO_END); 149 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 150 (unsigned long)VMEMMAP_END); 151 print_ml("vmalloc", (unsigned long)VMALLOC_START, 152 (unsigned long)VMALLOC_END); 153 #ifdef CONFIG_64BIT 154 print_ml("modules", (unsigned long)MODULES_VADDR, 155 (unsigned long)MODULES_END); 156 #endif 157 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 158 (unsigned long)high_memory); 159 if (IS_ENABLED(CONFIG_64BIT)) { 160 #ifdef CONFIG_KASAN 161 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 162 #endif 163 164 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 165 (unsigned long)ADDRESS_SPACE_END); 166 } 167 } 168 #else 169 static void print_vm_layout(void) { } 170 #endif /* CONFIG_DEBUG_VM */ 171 172 void __init arch_mm_preinit(void) 173 { 174 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 175 #ifdef CONFIG_FLATMEM 176 BUG_ON(!mem_map); 177 #endif /* CONFIG_FLATMEM */ 178 179 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 180 dma_cache_alignment != 1) { 181 /* 182 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 183 * buffer per 1GB of RAM for kmalloc() bouncing on 184 * non-coherent platforms. 185 */ 186 unsigned long size = 187 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 188 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 189 swiotlb = true; 190 } 191 192 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 193 194 print_vm_layout(); 195 } 196 197 /* Limit the memory size via mem. */ 198 static phys_addr_t memory_limit; 199 #ifdef CONFIG_XIP_KERNEL 200 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 201 #endif /* CONFIG_XIP_KERNEL */ 202 203 static int __init early_mem(char *p) 204 { 205 u64 size; 206 207 if (!p) 208 return 1; 209 210 size = memparse(p, &p) & PAGE_MASK; 211 memory_limit = min_t(u64, size, memory_limit); 212 213 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 214 215 return 0; 216 } 217 early_param("mem", early_mem); 218 219 static void __init setup_bootmem(void) 220 { 221 phys_addr_t vmlinux_end = __pa_symbol(&_end); 222 phys_addr_t max_mapped_addr; 223 phys_addr_t phys_ram_end, vmlinux_start; 224 225 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 226 vmlinux_start = __pa_symbol(&_sdata); 227 else 228 vmlinux_start = __pa_symbol(&_start); 229 230 memblock_enforce_memory_limit(memory_limit); 231 232 /* 233 * Make sure we align the reservation on PMD_SIZE since we will 234 * map the kernel in the linear mapping as read-only: we do not want 235 * any allocation to happen between _end and the next pmd aligned page. 236 */ 237 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 238 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 239 /* 240 * Reserve from the start of the kernel to the end of the kernel 241 */ 242 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 243 244 /* 245 * Make sure we align the start of the memory on a PMD boundary so that 246 * at worst, we map the linear mapping with PMD mappings. 247 */ 248 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) { 249 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 250 #ifdef CONFIG_SPARSEMEM_VMEMMAP 251 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 252 #endif 253 } 254 255 /* 256 * In 64-bit, any use of __va/__pa before this point is wrong as we 257 * did not know the start of DRAM before. 258 */ 259 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 260 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 261 262 /* 263 * The size of the linear page mapping may restrict the amount of 264 * usable RAM. 265 */ 266 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) { 267 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE; 268 if (memblock_end_of_DRAM() > max_mapped_addr) { 269 memblock_cap_memory_range(phys_ram_base, 270 max_mapped_addr - phys_ram_base); 271 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed", 272 &max_mapped_addr); 273 } 274 } 275 276 /* 277 * Reserve physical address space that would be mapped to virtual 278 * addresses greater than (void *)(-PAGE_SIZE) because: 279 * - This memory would overlap with ERR_PTR 280 * - This memory belongs to high memory, which is not supported 281 * 282 * This is not applicable to 64-bit kernel, because virtual addresses 283 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 284 * occupied by kernel mapping. Also it is unrealistic for high memory 285 * to exist on 64-bit platforms. 286 */ 287 if (!IS_ENABLED(CONFIG_64BIT)) { 288 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 289 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 290 } 291 292 phys_ram_end = memblock_end_of_DRAM(); 293 min_low_pfn = PFN_UP(phys_ram_base); 294 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 295 296 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 297 298 reserve_initrd_mem(); 299 300 /* 301 * No allocation should be done before reserving the memory as defined 302 * in the device tree, otherwise the allocation could end up in a 303 * reserved region. 304 */ 305 early_init_fdt_scan_reserved_mem(); 306 307 /* 308 * If DTB is built in, no need to reserve its memblock. 309 * Otherwise, do reserve it but avoid using 310 * early_init_fdt_reserve_self() since __pa() does 311 * not work for DTB pointers that are fixmap addresses 312 */ 313 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 314 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 315 316 dma_contiguous_reserve(dma32_phys_limit); 317 if (IS_ENABLED(CONFIG_64BIT)) 318 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 319 } 320 321 #ifdef CONFIG_RELOCATABLE 322 extern unsigned long __rela_dyn_start, __rela_dyn_end; 323 324 static void __init relocate_kernel(void) 325 { 326 Elf_Rela *rela = (Elf_Rela *)&__rela_dyn_start; 327 /* 328 * This holds the offset between the linked virtual address and the 329 * relocated virtual address. 330 */ 331 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 332 /* 333 * This holds the offset between kernel linked virtual address and 334 * physical address. 335 */ 336 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 337 338 for ( ; rela < (Elf_Rela *)&__rela_dyn_end; rela++) { 339 Elf_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 340 Elf_Addr relocated_addr = rela->r_addend; 341 342 if (rela->r_info != R_RISCV_RELATIVE) 343 continue; 344 345 /* 346 * Make sure to not relocate vdso symbols like rt_sigreturn 347 * which are linked from the address 0 in vmlinux since 348 * vdso symbol addresses are actually used as an offset from 349 * mm->context.vdso in VDSO_OFFSET macro. 350 */ 351 if (relocated_addr >= KERNEL_LINK_ADDR) 352 relocated_addr += reloc_offset; 353 354 *(Elf_Addr *)addr = relocated_addr; 355 } 356 } 357 #endif /* CONFIG_RELOCATABLE */ 358 359 #ifdef CONFIG_MMU 360 struct pt_alloc_ops pt_ops __meminitdata; 361 362 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 363 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 364 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 365 366 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 367 368 #ifdef CONFIG_XIP_KERNEL 369 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 370 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 371 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 372 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 373 #endif /* CONFIG_XIP_KERNEL */ 374 375 static const pgprot_t protection_map[16] = { 376 [VM_NONE] = PAGE_NONE, 377 [VM_READ] = PAGE_READ, 378 [VM_WRITE] = PAGE_COPY, 379 [VM_WRITE | VM_READ] = PAGE_COPY, 380 [VM_EXEC] = PAGE_EXEC, 381 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 382 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 383 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 384 [VM_SHARED] = PAGE_NONE, 385 [VM_SHARED | VM_READ] = PAGE_READ, 386 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 387 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 388 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 389 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 390 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 391 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 392 }; 393 DECLARE_VM_GET_PAGE_PROT 394 395 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 396 { 397 unsigned long addr = __fix_to_virt(idx); 398 pte_t *ptep; 399 400 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 401 402 ptep = &fixmap_pte[pte_index(addr)]; 403 404 if (pgprot_val(prot)) 405 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 406 else 407 pte_clear(&init_mm, addr, ptep); 408 local_flush_tlb_page(addr); 409 } 410 411 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 412 { 413 return (pte_t *)((uintptr_t)pa); 414 } 415 416 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 417 { 418 clear_fixmap(FIX_PTE); 419 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 420 } 421 422 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa) 423 { 424 return (pte_t *) __va(pa); 425 } 426 427 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 428 { 429 /* 430 * We only create PMD or PGD early mappings so we 431 * should never reach here with MMU disabled. 432 */ 433 BUG(); 434 } 435 436 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 437 { 438 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 439 } 440 441 static phys_addr_t __meminit alloc_pte_late(uintptr_t va) 442 { 443 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 444 445 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc)); 446 return __pa((pte_t *)ptdesc_address(ptdesc)); 447 } 448 449 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 450 pgprot_t prot) 451 { 452 uintptr_t pte_idx = pte_index(va); 453 454 BUG_ON(sz != PAGE_SIZE); 455 456 if (pte_none(ptep[pte_idx])) 457 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 458 } 459 460 #ifndef __PAGETABLE_PMD_FOLDED 461 462 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 463 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 464 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 465 466 #ifdef CONFIG_XIP_KERNEL 467 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 468 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 469 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 470 #endif /* CONFIG_XIP_KERNEL */ 471 472 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 473 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 474 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 475 476 #ifdef CONFIG_XIP_KERNEL 477 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 478 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 479 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 480 #endif /* CONFIG_XIP_KERNEL */ 481 482 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 483 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 484 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 485 486 #ifdef CONFIG_XIP_KERNEL 487 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 488 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 489 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 490 #endif /* CONFIG_XIP_KERNEL */ 491 492 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 493 { 494 /* Before MMU is enabled */ 495 return (pmd_t *)((uintptr_t)pa); 496 } 497 498 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 499 { 500 clear_fixmap(FIX_PMD); 501 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 502 } 503 504 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa) 505 { 506 return (pmd_t *) __va(pa); 507 } 508 509 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 510 { 511 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 512 513 return (uintptr_t)early_pmd; 514 } 515 516 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 517 { 518 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 519 } 520 521 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va) 522 { 523 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 524 525 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc)); 526 return __pa((pmd_t *)ptdesc_address(ptdesc)); 527 } 528 529 static void __meminit create_pmd_mapping(pmd_t *pmdp, 530 uintptr_t va, phys_addr_t pa, 531 phys_addr_t sz, pgprot_t prot) 532 { 533 pte_t *ptep; 534 phys_addr_t pte_phys; 535 uintptr_t pmd_idx = pmd_index(va); 536 537 if (sz == PMD_SIZE) { 538 if (pmd_none(pmdp[pmd_idx])) 539 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 540 return; 541 } 542 543 if (pmd_none(pmdp[pmd_idx])) { 544 pte_phys = pt_ops.alloc_pte(va); 545 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 546 ptep = pt_ops.get_pte_virt(pte_phys); 547 memset(ptep, 0, PAGE_SIZE); 548 } else { 549 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 550 ptep = pt_ops.get_pte_virt(pte_phys); 551 } 552 553 create_pte_mapping(ptep, va, pa, sz, prot); 554 } 555 556 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 557 { 558 return (pud_t *)((uintptr_t)pa); 559 } 560 561 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 562 { 563 clear_fixmap(FIX_PUD); 564 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 565 } 566 567 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 568 { 569 return (pud_t *)__va(pa); 570 } 571 572 static phys_addr_t __init alloc_pud_early(uintptr_t va) 573 { 574 /* Only one PUD is available for early mapping */ 575 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 576 577 return (uintptr_t)early_pud; 578 } 579 580 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 581 { 582 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 583 } 584 585 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 586 { 587 unsigned long vaddr; 588 589 vaddr = __get_free_page(GFP_KERNEL); 590 BUG_ON(!vaddr); 591 return __pa(vaddr); 592 } 593 594 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 595 { 596 return (p4d_t *)((uintptr_t)pa); 597 } 598 599 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 600 { 601 clear_fixmap(FIX_P4D); 602 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 603 } 604 605 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 606 { 607 return (p4d_t *)__va(pa); 608 } 609 610 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 611 { 612 /* Only one P4D is available for early mapping */ 613 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 614 615 return (uintptr_t)early_p4d; 616 } 617 618 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 619 { 620 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 621 } 622 623 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 624 { 625 unsigned long vaddr; 626 627 vaddr = __get_free_page(GFP_KERNEL); 628 BUG_ON(!vaddr); 629 return __pa(vaddr); 630 } 631 632 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 633 pgprot_t prot) 634 { 635 pmd_t *nextp; 636 phys_addr_t next_phys; 637 uintptr_t pud_index = pud_index(va); 638 639 if (sz == PUD_SIZE) { 640 if (pud_val(pudp[pud_index]) == 0) 641 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 642 return; 643 } 644 645 if (pud_val(pudp[pud_index]) == 0) { 646 next_phys = pt_ops.alloc_pmd(va); 647 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 648 nextp = pt_ops.get_pmd_virt(next_phys); 649 memset(nextp, 0, PAGE_SIZE); 650 } else { 651 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 652 nextp = pt_ops.get_pmd_virt(next_phys); 653 } 654 655 create_pmd_mapping(nextp, va, pa, sz, prot); 656 } 657 658 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 659 pgprot_t prot) 660 { 661 pud_t *nextp; 662 phys_addr_t next_phys; 663 uintptr_t p4d_index = p4d_index(va); 664 665 if (sz == P4D_SIZE) { 666 if (p4d_val(p4dp[p4d_index]) == 0) 667 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 668 return; 669 } 670 671 if (p4d_val(p4dp[p4d_index]) == 0) { 672 next_phys = pt_ops.alloc_pud(va); 673 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 674 nextp = pt_ops.get_pud_virt(next_phys); 675 memset(nextp, 0, PAGE_SIZE); 676 } else { 677 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 678 nextp = pt_ops.get_pud_virt(next_phys); 679 } 680 681 create_pud_mapping(nextp, va, pa, sz, prot); 682 } 683 684 #define pgd_next_t p4d_t 685 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 686 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 687 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 688 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 689 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 690 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 691 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 692 (pgtable_l5_enabled ? \ 693 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 694 (pgtable_l4_enabled ? \ 695 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 696 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 697 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 698 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 699 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 700 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 701 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 702 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 703 #else 704 #define pgd_next_t pte_t 705 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 706 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 707 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 708 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 709 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 710 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 711 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 712 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 713 #endif /* __PAGETABLE_PMD_FOLDED */ 714 715 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 716 pgprot_t prot) 717 { 718 pgd_next_t *nextp; 719 phys_addr_t next_phys; 720 uintptr_t pgd_idx = pgd_index(va); 721 722 if (sz == PGDIR_SIZE) { 723 if (pgd_val(pgdp[pgd_idx]) == 0) 724 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 725 return; 726 } 727 728 if (pgd_val(pgdp[pgd_idx]) == 0) { 729 next_phys = alloc_pgd_next(va); 730 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 731 nextp = get_pgd_next_virt(next_phys); 732 memset(nextp, 0, PAGE_SIZE); 733 } else { 734 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 735 nextp = get_pgd_next_virt(next_phys); 736 } 737 738 create_pgd_next_mapping(nextp, va, pa, sz, prot); 739 } 740 741 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 742 { 743 if (debug_pagealloc_enabled()) 744 return PAGE_SIZE; 745 746 if (pgtable_l5_enabled && 747 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 748 return P4D_SIZE; 749 750 if (pgtable_l4_enabled && 751 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 752 return PUD_SIZE; 753 754 if (IS_ENABLED(CONFIG_64BIT) && 755 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 756 return PMD_SIZE; 757 758 return PAGE_SIZE; 759 } 760 761 #ifdef CONFIG_XIP_KERNEL 762 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 763 extern char _xiprom[], _exiprom[], __data_loc; 764 765 /* called from head.S with MMU off */ 766 asmlinkage void __init __copy_data(void) 767 { 768 void *from = (void *)(&__data_loc); 769 void *to = (void *)CONFIG_PHYS_RAM_BASE; 770 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 771 772 memcpy(to, from, sz); 773 } 774 #endif 775 776 #ifdef CONFIG_STRICT_KERNEL_RWX 777 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 778 { 779 if (is_va_kernel_text(va)) 780 return PAGE_KERNEL_READ_EXEC; 781 782 /* 783 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 784 * we must protect its linear mapping alias from being executed and 785 * written. 786 * And rodata section is marked readonly in mark_rodata_ro. 787 */ 788 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 789 return PAGE_KERNEL_READ; 790 791 return PAGE_KERNEL; 792 } 793 794 void mark_rodata_ro(void) 795 { 796 set_kernel_memory(__start_rodata, _data, set_memory_ro); 797 if (IS_ENABLED(CONFIG_64BIT)) 798 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 799 set_memory_ro); 800 } 801 #else 802 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 803 { 804 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 805 return PAGE_KERNEL; 806 807 return PAGE_KERNEL_EXEC; 808 } 809 #endif /* CONFIG_STRICT_KERNEL_RWX */ 810 811 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 812 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 813 814 static void __init disable_pgtable_l5(void) 815 { 816 pgtable_l5_enabled = false; 817 kernel_map.page_offset = PAGE_OFFSET_L4; 818 satp_mode = SATP_MODE_48; 819 } 820 821 static void __init disable_pgtable_l4(void) 822 { 823 pgtable_l4_enabled = false; 824 kernel_map.page_offset = PAGE_OFFSET_L3; 825 satp_mode = SATP_MODE_39; 826 } 827 828 static int __init print_no4lvl(char *p) 829 { 830 pr_info("Disabled 4-level and 5-level paging"); 831 return 0; 832 } 833 early_param("no4lvl", print_no4lvl); 834 835 static int __init print_no5lvl(char *p) 836 { 837 pr_info("Disabled 5-level paging"); 838 return 0; 839 } 840 early_param("no5lvl", print_no5lvl); 841 842 static void __init set_mmap_rnd_bits_max(void) 843 { 844 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 845 } 846 847 /* 848 * There is a simple way to determine if 4-level is supported by the 849 * underlying hardware: establish 1:1 mapping in 4-level page table mode 850 * then read SATP to see if the configuration was taken into account 851 * meaning sv48 is supported. 852 */ 853 static __init void set_satp_mode(uintptr_t dtb_pa) 854 { 855 u64 identity_satp, hw_satp; 856 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 857 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 858 859 kernel_map.page_offset = PAGE_OFFSET_L5; 860 861 if (satp_mode_cmdline == SATP_MODE_57) { 862 disable_pgtable_l5(); 863 } else if (satp_mode_cmdline == SATP_MODE_48) { 864 disable_pgtable_l5(); 865 disable_pgtable_l4(); 866 return; 867 } 868 869 create_p4d_mapping(early_p4d, 870 set_satp_mode_pmd, (uintptr_t)early_pud, 871 P4D_SIZE, PAGE_TABLE); 872 create_pud_mapping(early_pud, 873 set_satp_mode_pmd, (uintptr_t)early_pmd, 874 PUD_SIZE, PAGE_TABLE); 875 /* Handle the case where set_satp_mode straddles 2 PMDs */ 876 create_pmd_mapping(early_pmd, 877 set_satp_mode_pmd, set_satp_mode_pmd, 878 PMD_SIZE, PAGE_KERNEL_EXEC); 879 create_pmd_mapping(early_pmd, 880 set_satp_mode_pmd + PMD_SIZE, 881 set_satp_mode_pmd + PMD_SIZE, 882 PMD_SIZE, PAGE_KERNEL_EXEC); 883 retry: 884 create_pgd_mapping(early_pg_dir, 885 set_satp_mode_pmd, 886 pgtable_l5_enabled ? 887 (uintptr_t)early_p4d : (uintptr_t)early_pud, 888 PGDIR_SIZE, PAGE_TABLE); 889 890 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 891 892 local_flush_tlb_all(); 893 csr_write(CSR_SATP, identity_satp); 894 hw_satp = csr_swap(CSR_SATP, 0ULL); 895 local_flush_tlb_all(); 896 897 if (hw_satp != identity_satp) { 898 if (pgtable_l5_enabled) { 899 disable_pgtable_l5(); 900 memset(early_pg_dir, 0, PAGE_SIZE); 901 goto retry; 902 } 903 disable_pgtable_l4(); 904 } 905 906 memset(early_pg_dir, 0, PAGE_SIZE); 907 memset(early_p4d, 0, PAGE_SIZE); 908 memset(early_pud, 0, PAGE_SIZE); 909 memset(early_pmd, 0, PAGE_SIZE); 910 } 911 #endif 912 913 /* 914 * setup_vm() is called from head.S with MMU-off. 915 * 916 * Following requirements should be honoured for setup_vm() to work 917 * correctly: 918 * 1) It should use PC-relative addressing for accessing kernel symbols. 919 * To achieve this we always use GCC cmodel=medany. 920 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 921 * so disable compiler instrumentation when FTRACE is enabled. 922 * 923 * Currently, the above requirements are honoured by using custom CFLAGS 924 * for init.o in mm/Makefile. 925 */ 926 927 #ifndef __riscv_cmodel_medany 928 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 929 #endif 930 931 #ifdef CONFIG_XIP_KERNEL 932 static void __init create_kernel_page_table(pgd_t *pgdir, 933 __always_unused bool early) 934 { 935 uintptr_t va, start_va, end_va; 936 937 /* Map the flash resident part */ 938 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 939 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 940 create_pgd_mapping(pgdir, va, 941 kernel_map.xiprom + (va - kernel_map.virt_addr), 942 PMD_SIZE, PAGE_KERNEL_EXEC); 943 944 /* Map the data in RAM */ 945 start_va = kernel_map.virt_addr + (uintptr_t)&_sdata - (uintptr_t)&_start; 946 end_va = kernel_map.virt_addr + kernel_map.size; 947 for (va = start_va; va < end_va; va += PMD_SIZE) 948 create_pgd_mapping(pgdir, va, 949 kernel_map.phys_addr + (va - start_va), 950 PMD_SIZE, PAGE_KERNEL); 951 } 952 #else 953 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 954 { 955 uintptr_t va, end_va; 956 957 end_va = kernel_map.virt_addr + kernel_map.size; 958 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 959 create_pgd_mapping(pgdir, va, 960 kernel_map.phys_addr + (va - kernel_map.virt_addr), 961 PMD_SIZE, 962 early ? 963 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 964 } 965 #endif 966 967 /* 968 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 969 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 970 * entry. 971 */ 972 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 973 uintptr_t dtb_pa) 974 { 975 #ifndef CONFIG_BUILTIN_DTB 976 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 977 978 /* Make sure the fdt fixmap address is always aligned on PMD size */ 979 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 980 981 /* In 32-bit only, the fdt lies in its own PGD */ 982 if (!IS_ENABLED(CONFIG_64BIT)) { 983 create_pgd_mapping(early_pg_dir, fix_fdt_va, 984 pa, MAX_FDT_SIZE, PAGE_KERNEL); 985 } else { 986 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 987 pa, PMD_SIZE, PAGE_KERNEL); 988 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 989 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 990 } 991 992 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 993 #else 994 /* 995 * For 64-bit kernel, __va can't be used since it would return a linear 996 * mapping address whereas dtb_early_va will be used before 997 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 998 * kernel is mapped in the linear mapping, that makes no difference. 999 */ 1000 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 1001 #endif 1002 1003 dtb_early_pa = dtb_pa; 1004 } 1005 1006 /* 1007 * MMU is not enabled, the page tables are allocated directly using 1008 * early_pmd/pud/p4d and the address returned is the physical one. 1009 */ 1010 static void __init pt_ops_set_early(void) 1011 { 1012 pt_ops.alloc_pte = alloc_pte_early; 1013 pt_ops.get_pte_virt = get_pte_virt_early; 1014 #ifndef __PAGETABLE_PMD_FOLDED 1015 pt_ops.alloc_pmd = alloc_pmd_early; 1016 pt_ops.get_pmd_virt = get_pmd_virt_early; 1017 pt_ops.alloc_pud = alloc_pud_early; 1018 pt_ops.get_pud_virt = get_pud_virt_early; 1019 pt_ops.alloc_p4d = alloc_p4d_early; 1020 pt_ops.get_p4d_virt = get_p4d_virt_early; 1021 #endif 1022 } 1023 1024 /* 1025 * MMU is enabled but page table setup is not complete yet. 1026 * fixmap page table alloc functions must be used as a means to temporarily 1027 * map the allocated physical pages since the linear mapping does not exist yet. 1028 * 1029 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1030 * but it will be used as described above. 1031 */ 1032 static void __init pt_ops_set_fixmap(void) 1033 { 1034 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1035 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1036 #ifndef __PAGETABLE_PMD_FOLDED 1037 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1038 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1039 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1040 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1041 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1042 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1043 #endif 1044 } 1045 1046 /* 1047 * MMU is enabled and page table setup is complete, so from now, we can use 1048 * generic page allocation functions to setup page table. 1049 */ 1050 static void __init pt_ops_set_late(void) 1051 { 1052 pt_ops.alloc_pte = alloc_pte_late; 1053 pt_ops.get_pte_virt = get_pte_virt_late; 1054 #ifndef __PAGETABLE_PMD_FOLDED 1055 pt_ops.alloc_pmd = alloc_pmd_late; 1056 pt_ops.get_pmd_virt = get_pmd_virt_late; 1057 pt_ops.alloc_pud = alloc_pud_late; 1058 pt_ops.get_pud_virt = get_pud_virt_late; 1059 pt_ops.alloc_p4d = alloc_p4d_late; 1060 pt_ops.get_p4d_virt = get_p4d_virt_late; 1061 #endif 1062 } 1063 1064 #ifdef CONFIG_RANDOMIZE_BASE 1065 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1066 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1067 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa); 1068 1069 static int __init print_nokaslr(char *p) 1070 { 1071 pr_info("Disabled KASLR"); 1072 return 0; 1073 } 1074 early_param("nokaslr", print_nokaslr); 1075 1076 unsigned long kaslr_offset(void) 1077 { 1078 return kernel_map.virt_offset; 1079 } 1080 #endif 1081 1082 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1083 { 1084 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1085 1086 #ifdef CONFIG_RANDOMIZE_BASE 1087 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1088 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa); 1089 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1090 u32 nr_pos; 1091 1092 if (kaslr_seed == 0) 1093 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1094 /* 1095 * Compute the number of positions available: we are limited 1096 * by the early page table that only has one PUD and we must 1097 * be aligned on PMD_SIZE. 1098 */ 1099 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1100 1101 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1102 } 1103 #endif 1104 1105 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1106 1107 #ifdef CONFIG_XIP_KERNEL 1108 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1109 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1110 1111 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1112 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1113 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 1114 #endif 1115 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1116 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1117 1118 kernel_map.va_kernel_xip_text_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1119 kernel_map.va_kernel_xip_data_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr 1120 + (uintptr_t)&_sdata - (uintptr_t)&_start; 1121 #else 1122 kernel_map.phys_addr = (uintptr_t)(&_start); 1123 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1124 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1125 #endif 1126 1127 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1128 set_satp_mode(dtb_pa); 1129 set_mmap_rnd_bits_max(); 1130 #endif 1131 1132 /* 1133 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1134 * where we have the system memory layout: this allows us to align 1135 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1136 * for the linear mapping. This is only possible because the kernel 1137 * mapping lies outside the linear mapping. 1138 * In 32-bit however, as the kernel resides in the linear mapping, 1139 * setup_vm_final can not change the mapping established here, 1140 * otherwise the same kernel addresses would get mapped to different 1141 * physical addresses (if the start of dram is different from the 1142 * kernel physical address start). 1143 */ 1144 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1145 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1146 1147 memory_limit = KERN_VIRT_SIZE; 1148 1149 /* Sanity check alignment and size */ 1150 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1151 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1152 1153 #ifdef CONFIG_64BIT 1154 /* 1155 * The last 4K bytes of the addressable memory can not be mapped because 1156 * of IS_ERR_VALUE macro. 1157 */ 1158 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1159 #endif 1160 1161 #ifdef CONFIG_RELOCATABLE 1162 /* 1163 * Early page table uses only one PUD, which makes it possible 1164 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1165 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1166 * since a part of the kernel would not get mapped. 1167 */ 1168 if (IS_ENABLED(CONFIG_64BIT)) 1169 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1170 relocate_kernel(); 1171 #endif 1172 1173 apply_early_boot_alternatives(); 1174 pt_ops_set_early(); 1175 1176 /* Setup early PGD for fixmap */ 1177 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1178 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1179 1180 #ifndef __PAGETABLE_PMD_FOLDED 1181 /* Setup fixmap P4D and PUD */ 1182 if (pgtable_l5_enabled) 1183 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1184 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1185 /* Setup fixmap PUD and PMD */ 1186 if (pgtable_l4_enabled) 1187 create_pud_mapping(fixmap_pud, FIXADDR_START, 1188 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1189 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1190 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1191 /* Setup trampoline PGD and PMD */ 1192 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1193 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1194 if (pgtable_l5_enabled) 1195 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1196 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1197 if (pgtable_l4_enabled) 1198 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1199 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1200 #ifdef CONFIG_XIP_KERNEL 1201 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1202 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1203 #else 1204 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1205 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1206 #endif 1207 #else 1208 /* Setup trampoline PGD */ 1209 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1210 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1211 #endif 1212 1213 /* 1214 * Setup early PGD covering entire kernel which will allow 1215 * us to reach paging_init(). We map all memory banks later 1216 * in setup_vm_final() below. 1217 */ 1218 create_kernel_page_table(early_pg_dir, true); 1219 1220 /* Setup early mapping for FDT early scan */ 1221 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1222 1223 /* 1224 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1225 * range can not span multiple pmds. 1226 */ 1227 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1228 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1229 1230 #ifndef __PAGETABLE_PMD_FOLDED 1231 /* 1232 * Early ioremap fixmap is already created as it lies within first 2MB 1233 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1234 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1235 * the user if not. 1236 */ 1237 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1238 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1239 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1240 WARN_ON(1); 1241 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1242 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1243 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1244 fix_to_virt(FIX_BTMAP_BEGIN)); 1245 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1246 fix_to_virt(FIX_BTMAP_END)); 1247 1248 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1249 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1250 } 1251 #endif 1252 1253 pt_ops_set_fixmap(); 1254 } 1255 1256 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end, 1257 uintptr_t fixed_map_size, const pgprot_t *pgprot) 1258 { 1259 phys_addr_t pa; 1260 uintptr_t va, map_size; 1261 1262 for (pa = start; pa < end; pa += map_size) { 1263 va = (uintptr_t)__va(pa); 1264 map_size = fixed_map_size ? fixed_map_size : 1265 best_map_size(pa, va, end - pa); 1266 1267 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1268 pgprot ? *pgprot : pgprot_from_va(va)); 1269 } 1270 } 1271 1272 static void __init create_linear_mapping_page_table(void) 1273 { 1274 phys_addr_t start, end; 1275 phys_addr_t kfence_pool __maybe_unused; 1276 u64 i; 1277 1278 #ifdef CONFIG_STRICT_KERNEL_RWX 1279 phys_addr_t ktext_start = __pa_symbol(_start); 1280 phys_addr_t ktext_size = __init_data_begin - _start; 1281 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1282 phys_addr_t krodata_size = _data - __start_rodata; 1283 1284 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1285 memblock_mark_nomap(ktext_start, ktext_size); 1286 memblock_mark_nomap(krodata_start, krodata_size); 1287 #endif 1288 1289 #ifdef CONFIG_KFENCE 1290 /* 1291 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1292 * before we setup the linear mapping so that we avoid using hugepages 1293 * for this region. 1294 */ 1295 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1296 BUG_ON(!kfence_pool); 1297 1298 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1299 __kfence_pool = __va(kfence_pool); 1300 #endif 1301 1302 /* Map all memory banks in the linear mapping */ 1303 for_each_mem_range(i, &start, &end) { 1304 if (start >= end) 1305 break; 1306 if (start <= __pa(PAGE_OFFSET) && 1307 __pa(PAGE_OFFSET) < end) 1308 start = __pa(PAGE_OFFSET); 1309 1310 create_linear_mapping_range(start, end, 0, NULL); 1311 } 1312 1313 #ifdef CONFIG_STRICT_KERNEL_RWX 1314 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL); 1315 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL); 1316 1317 memblock_clear_nomap(ktext_start, ktext_size); 1318 memblock_clear_nomap(krodata_start, krodata_size); 1319 #endif 1320 1321 #ifdef CONFIG_KFENCE 1322 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL); 1323 1324 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1325 #endif 1326 } 1327 1328 static void __init setup_vm_final(void) 1329 { 1330 /* Setup swapper PGD for fixmap */ 1331 #if !defined(CONFIG_64BIT) 1332 /* 1333 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1334 * directly in swapper_pg_dir in addition to the pgd entry that points 1335 * to fixmap_pte. 1336 */ 1337 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1338 1339 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1340 #endif 1341 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1342 __pa_symbol(fixmap_pgd_next), 1343 PGDIR_SIZE, PAGE_TABLE); 1344 1345 /* Map the linear mapping */ 1346 create_linear_mapping_page_table(); 1347 1348 /* Map the kernel */ 1349 if (IS_ENABLED(CONFIG_64BIT)) 1350 create_kernel_page_table(swapper_pg_dir, false); 1351 1352 #ifdef CONFIG_KASAN 1353 kasan_swapper_init(); 1354 #endif 1355 1356 /* Clear fixmap PTE and PMD mappings */ 1357 clear_fixmap(FIX_PTE); 1358 clear_fixmap(FIX_PMD); 1359 clear_fixmap(FIX_PUD); 1360 clear_fixmap(FIX_P4D); 1361 1362 /* Move to swapper page table */ 1363 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1364 local_flush_tlb_all(); 1365 1366 pt_ops_set_late(); 1367 } 1368 #else 1369 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1370 { 1371 dtb_early_va = (void *)dtb_pa; 1372 dtb_early_pa = dtb_pa; 1373 1374 #ifdef CONFIG_RELOCATABLE 1375 kernel_map.virt_addr = (uintptr_t)_start; 1376 kernel_map.phys_addr = (uintptr_t)_start; 1377 relocate_kernel(); 1378 #endif 1379 } 1380 1381 static inline void setup_vm_final(void) 1382 { 1383 } 1384 #endif /* CONFIG_MMU */ 1385 1386 /* 1387 * reserve_crashkernel() - reserves memory for crash kernel 1388 * 1389 * This function reserves memory area given in "crashkernel=" kernel command 1390 * line parameter. The memory reserved is used by dump capture kernel when 1391 * primary kernel is crashing. 1392 */ 1393 static void __init arch_reserve_crashkernel(void) 1394 { 1395 unsigned long long low_size = 0; 1396 unsigned long long crash_base, crash_size; 1397 bool high = false; 1398 int ret; 1399 1400 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1401 return; 1402 1403 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1404 &crash_size, &crash_base, 1405 &low_size, &high); 1406 if (ret) 1407 return; 1408 1409 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 1410 } 1411 1412 void __init paging_init(void) 1413 { 1414 setup_bootmem(); 1415 setup_vm_final(); 1416 1417 /* Depend on that Linear Mapping is ready */ 1418 memblock_allow_resize(); 1419 } 1420 1421 void __init misc_mem_init(void) 1422 { 1423 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1424 arch_numa_init(); 1425 sparse_init(); 1426 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1427 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1428 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1429 #endif 1430 zone_sizes_init(); 1431 arch_reserve_crashkernel(); 1432 memblock_dump_all(); 1433 } 1434 1435 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1436 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1437 unsigned long addr, unsigned long next) 1438 { 1439 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1440 } 1441 1442 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1443 unsigned long addr, unsigned long next) 1444 { 1445 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1446 return 1; 1447 } 1448 1449 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1450 struct vmem_altmap *altmap) 1451 { 1452 /* 1453 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1454 * can't use hugepage mappings for 2-level page table because in case of 1455 * memory hotplug, we are not able to update all the page tables with 1456 * the new PMDs. 1457 */ 1458 return vmemmap_populate_hugepages(start, end, node, altmap); 1459 } 1460 #endif 1461 1462 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1463 /* 1464 * Pre-allocates page-table pages for a specific area in the kernel 1465 * page-table. Only the level which needs to be synchronized between 1466 * all page-tables is allocated because the synchronization can be 1467 * expensive. 1468 */ 1469 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1470 const char *area) 1471 { 1472 unsigned long addr; 1473 const char *lvl; 1474 1475 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1476 pgd_t *pgd = pgd_offset_k(addr); 1477 p4d_t *p4d; 1478 pud_t *pud; 1479 pmd_t *pmd; 1480 1481 lvl = "p4d"; 1482 p4d = p4d_alloc(&init_mm, pgd, addr); 1483 if (!p4d) 1484 goto failed; 1485 1486 if (pgtable_l5_enabled) 1487 continue; 1488 1489 lvl = "pud"; 1490 pud = pud_alloc(&init_mm, p4d, addr); 1491 if (!pud) 1492 goto failed; 1493 1494 if (pgtable_l4_enabled) 1495 continue; 1496 1497 lvl = "pmd"; 1498 pmd = pmd_alloc(&init_mm, pud, addr); 1499 if (!pmd) 1500 goto failed; 1501 } 1502 return; 1503 1504 failed: 1505 /* 1506 * The pages have to be there now or they will be missing in 1507 * process page-tables later. 1508 */ 1509 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1510 } 1511 1512 #define PAGE_END KASAN_SHADOW_START 1513 1514 void __init pgtable_cache_init(void) 1515 { 1516 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1517 if (IS_ENABLED(CONFIG_MODULES)) 1518 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1519 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1520 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1521 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1522 if (IS_ENABLED(CONFIG_KASAN)) 1523 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1524 } 1525 } 1526 #endif 1527 1528 #ifdef CONFIG_EXECMEM 1529 #ifdef CONFIG_MMU 1530 static struct execmem_info execmem_info __ro_after_init; 1531 1532 struct execmem_info __init *execmem_arch_setup(void) 1533 { 1534 execmem_info = (struct execmem_info){ 1535 .ranges = { 1536 [EXECMEM_DEFAULT] = { 1537 .start = MODULES_VADDR, 1538 .end = MODULES_END, 1539 .pgprot = PAGE_KERNEL, 1540 .alignment = 1, 1541 }, 1542 [EXECMEM_KPROBES] = { 1543 .start = VMALLOC_START, 1544 .end = VMALLOC_END, 1545 .pgprot = PAGE_KERNEL_READ_EXEC, 1546 .alignment = 1, 1547 }, 1548 [EXECMEM_BPF] = { 1549 .start = BPF_JIT_REGION_START, 1550 .end = BPF_JIT_REGION_END, 1551 .pgprot = PAGE_KERNEL, 1552 .alignment = PAGE_SIZE, 1553 }, 1554 }, 1555 }; 1556 1557 return &execmem_info; 1558 } 1559 #endif /* CONFIG_MMU */ 1560 #endif /* CONFIG_EXECMEM */ 1561 1562 #ifdef CONFIG_MEMORY_HOTPLUG 1563 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1564 { 1565 struct page *page = pmd_page(*pmd); 1566 struct ptdesc *ptdesc = page_ptdesc(page); 1567 pte_t *pte; 1568 int i; 1569 1570 for (i = 0; i < PTRS_PER_PTE; i++) { 1571 pte = pte_start + i; 1572 if (!pte_none(*pte)) 1573 return; 1574 } 1575 1576 pagetable_dtor(ptdesc); 1577 if (PageReserved(page)) 1578 free_reserved_page(page); 1579 else 1580 pagetable_free(ptdesc); 1581 pmd_clear(pmd); 1582 } 1583 1584 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1585 { 1586 struct page *page = pud_page(*pud); 1587 struct ptdesc *ptdesc = page_ptdesc(page); 1588 pmd_t *pmd; 1589 int i; 1590 1591 for (i = 0; i < PTRS_PER_PMD; i++) { 1592 pmd = pmd_start + i; 1593 if (!pmd_none(*pmd)) 1594 return; 1595 } 1596 1597 if (!is_vmemmap) 1598 pagetable_dtor(ptdesc); 1599 if (PageReserved(page)) 1600 free_reserved_page(page); 1601 else 1602 pagetable_free(ptdesc); 1603 pud_clear(pud); 1604 } 1605 1606 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1607 { 1608 struct page *page = p4d_page(*p4d); 1609 pud_t *pud; 1610 int i; 1611 1612 for (i = 0; i < PTRS_PER_PUD; i++) { 1613 pud = pud_start + i; 1614 if (!pud_none(*pud)) 1615 return; 1616 } 1617 1618 if (PageReserved(page)) 1619 free_reserved_page(page); 1620 else 1621 free_pages((unsigned long)page_address(page), 0); 1622 p4d_clear(p4d); 1623 } 1624 1625 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1626 struct vmem_altmap *altmap) 1627 { 1628 int order = get_order(size); 1629 1630 if (altmap) { 1631 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1632 return; 1633 } 1634 1635 if (PageReserved(page)) { 1636 unsigned int nr_pages = 1 << order; 1637 1638 while (nr_pages--) 1639 free_reserved_page(page++); 1640 return; 1641 } 1642 1643 free_pages((unsigned long)page_address(page), order); 1644 } 1645 1646 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1647 bool is_vmemmap, struct vmem_altmap *altmap) 1648 { 1649 unsigned long next; 1650 pte_t *ptep, pte; 1651 1652 for (; addr < end; addr = next) { 1653 next = (addr + PAGE_SIZE) & PAGE_MASK; 1654 if (next > end) 1655 next = end; 1656 1657 ptep = pte_base + pte_index(addr); 1658 pte = ptep_get(ptep); 1659 if (!pte_present(*ptep)) 1660 continue; 1661 1662 pte_clear(&init_mm, addr, ptep); 1663 if (is_vmemmap) 1664 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1665 } 1666 } 1667 1668 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1669 bool is_vmemmap, struct vmem_altmap *altmap) 1670 { 1671 unsigned long next; 1672 pte_t *pte_base; 1673 pmd_t *pmdp, pmd; 1674 1675 for (; addr < end; addr = next) { 1676 next = pmd_addr_end(addr, end); 1677 pmdp = pmd_base + pmd_index(addr); 1678 pmd = pmdp_get(pmdp); 1679 if (!pmd_present(pmd)) 1680 continue; 1681 1682 if (pmd_leaf(pmd)) { 1683 pmd_clear(pmdp); 1684 if (is_vmemmap) 1685 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1686 continue; 1687 } 1688 1689 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1690 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1691 free_pte_table(pte_base, pmdp); 1692 } 1693 } 1694 1695 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1696 bool is_vmemmap, struct vmem_altmap *altmap) 1697 { 1698 unsigned long next; 1699 pud_t *pudp, pud; 1700 pmd_t *pmd_base; 1701 1702 for (; addr < end; addr = next) { 1703 next = pud_addr_end(addr, end); 1704 pudp = pud_base + pud_index(addr); 1705 pud = pudp_get(pudp); 1706 if (!pud_present(pud)) 1707 continue; 1708 1709 if (pud_leaf(pud)) { 1710 if (pgtable_l4_enabled) { 1711 pud_clear(pudp); 1712 if (is_vmemmap) 1713 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1714 } 1715 continue; 1716 } 1717 1718 pmd_base = pmd_offset(pudp, 0); 1719 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1720 1721 if (pgtable_l4_enabled) 1722 free_pmd_table(pmd_base, pudp, is_vmemmap); 1723 } 1724 } 1725 1726 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1727 bool is_vmemmap, struct vmem_altmap *altmap) 1728 { 1729 unsigned long next; 1730 p4d_t *p4dp, p4d; 1731 pud_t *pud_base; 1732 1733 for (; addr < end; addr = next) { 1734 next = p4d_addr_end(addr, end); 1735 p4dp = p4d_base + p4d_index(addr); 1736 p4d = p4dp_get(p4dp); 1737 if (!p4d_present(p4d)) 1738 continue; 1739 1740 if (p4d_leaf(p4d)) { 1741 if (pgtable_l5_enabled) { 1742 p4d_clear(p4dp); 1743 if (is_vmemmap) 1744 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1745 } 1746 continue; 1747 } 1748 1749 pud_base = pud_offset(p4dp, 0); 1750 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1751 1752 if (pgtable_l5_enabled) 1753 free_pud_table(pud_base, p4dp); 1754 } 1755 } 1756 1757 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1758 struct vmem_altmap *altmap) 1759 { 1760 unsigned long addr, next; 1761 p4d_t *p4d_base; 1762 pgd_t *pgd; 1763 1764 for (addr = va; addr < end; addr = next) { 1765 next = pgd_addr_end(addr, end); 1766 pgd = pgd_offset_k(addr); 1767 1768 if (!pgd_present(*pgd)) 1769 continue; 1770 1771 if (pgd_leaf(*pgd)) 1772 continue; 1773 1774 p4d_base = p4d_offset(pgd, 0); 1775 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1776 } 1777 1778 flush_tlb_all(); 1779 } 1780 1781 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1782 { 1783 unsigned long va = (unsigned long)__va(start); 1784 unsigned long end = (unsigned long)__va(start + size); 1785 1786 remove_pgd_mapping(va, end, false, NULL); 1787 } 1788 1789 struct range arch_get_mappable_range(void) 1790 { 1791 struct range mhp_range; 1792 1793 mhp_range.start = __pa(PAGE_OFFSET); 1794 mhp_range.end = __pa(PAGE_END - 1); 1795 return mhp_range; 1796 } 1797 1798 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1799 { 1800 int ret = 0; 1801 1802 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1803 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1804 if (ret) { 1805 remove_linear_mapping(start, size); 1806 goto out; 1807 } 1808 1809 max_pfn = PFN_UP(start + size); 1810 max_low_pfn = max_pfn; 1811 1812 out: 1813 flush_tlb_all(); 1814 return ret; 1815 } 1816 1817 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 1818 { 1819 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap); 1820 remove_linear_mapping(start, size); 1821 flush_tlb_all(); 1822 } 1823 1824 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1825 { 1826 remove_pgd_mapping(start, end, true, altmap); 1827 } 1828 #endif /* CONFIG_MEMORY_HOTPLUG */ 1829