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 /* 446 * We do not know which mm the PTE page is associated to at this point. 447 * Passing NULL to the ctor is the safe option, though it may result 448 * in unnecessary work (e.g. initialising the ptlock for init_mm). 449 */ 450 BUG_ON(!ptdesc || !pagetable_pte_ctor(NULL, ptdesc)); 451 return __pa((pte_t *)ptdesc_address(ptdesc)); 452 } 453 454 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 455 pgprot_t prot) 456 { 457 uintptr_t pte_idx = pte_index(va); 458 459 BUG_ON(sz != PAGE_SIZE); 460 461 if (pte_none(ptep[pte_idx])) 462 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 463 } 464 465 #ifndef __PAGETABLE_PMD_FOLDED 466 467 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 468 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 469 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 470 471 #ifdef CONFIG_XIP_KERNEL 472 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 473 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 474 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 475 #endif /* CONFIG_XIP_KERNEL */ 476 477 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 478 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 479 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 480 481 #ifdef CONFIG_XIP_KERNEL 482 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 483 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 484 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 485 #endif /* CONFIG_XIP_KERNEL */ 486 487 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 488 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 489 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 490 491 #ifdef CONFIG_XIP_KERNEL 492 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 493 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 494 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 495 #endif /* CONFIG_XIP_KERNEL */ 496 497 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 498 { 499 /* Before MMU is enabled */ 500 return (pmd_t *)((uintptr_t)pa); 501 } 502 503 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 504 { 505 clear_fixmap(FIX_PMD); 506 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 507 } 508 509 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa) 510 { 511 return (pmd_t *) __va(pa); 512 } 513 514 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 515 { 516 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 517 518 return (uintptr_t)early_pmd; 519 } 520 521 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 522 { 523 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 524 } 525 526 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va) 527 { 528 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 529 530 /* See comment in alloc_pte_late() regarding NULL passed the ctor */ 531 BUG_ON(!ptdesc || !pagetable_pmd_ctor(NULL, ptdesc)); 532 return __pa((pmd_t *)ptdesc_address(ptdesc)); 533 } 534 535 static void __meminit create_pmd_mapping(pmd_t *pmdp, 536 uintptr_t va, phys_addr_t pa, 537 phys_addr_t sz, pgprot_t prot) 538 { 539 pte_t *ptep; 540 phys_addr_t pte_phys; 541 uintptr_t pmd_idx = pmd_index(va); 542 543 if (sz == PMD_SIZE) { 544 if (pmd_none(pmdp[pmd_idx])) 545 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 546 return; 547 } 548 549 if (pmd_none(pmdp[pmd_idx])) { 550 pte_phys = pt_ops.alloc_pte(va); 551 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 552 ptep = pt_ops.get_pte_virt(pte_phys); 553 memset(ptep, 0, PAGE_SIZE); 554 } else { 555 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 556 ptep = pt_ops.get_pte_virt(pte_phys); 557 } 558 559 create_pte_mapping(ptep, va, pa, sz, prot); 560 } 561 562 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 563 { 564 return (pud_t *)((uintptr_t)pa); 565 } 566 567 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 568 { 569 clear_fixmap(FIX_PUD); 570 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 571 } 572 573 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 574 { 575 return (pud_t *)__va(pa); 576 } 577 578 static phys_addr_t __init alloc_pud_early(uintptr_t va) 579 { 580 /* Only one PUD is available for early mapping */ 581 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 582 583 return (uintptr_t)early_pud; 584 } 585 586 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 587 { 588 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 589 } 590 591 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 592 { 593 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 594 595 BUG_ON(!ptdesc); 596 pagetable_pud_ctor(ptdesc); 597 return __pa((pud_t *)ptdesc_address(ptdesc)); 598 } 599 600 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 601 { 602 return (p4d_t *)((uintptr_t)pa); 603 } 604 605 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 606 { 607 clear_fixmap(FIX_P4D); 608 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 609 } 610 611 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 612 { 613 return (p4d_t *)__va(pa); 614 } 615 616 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 617 { 618 /* Only one P4D is available for early mapping */ 619 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 620 621 return (uintptr_t)early_p4d; 622 } 623 624 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 625 { 626 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 627 } 628 629 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 630 { 631 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 632 633 BUG_ON(!ptdesc); 634 pagetable_p4d_ctor(ptdesc); 635 return __pa((p4d_t *)ptdesc_address(ptdesc)); 636 } 637 638 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 639 pgprot_t prot) 640 { 641 pmd_t *nextp; 642 phys_addr_t next_phys; 643 uintptr_t pud_index = pud_index(va); 644 645 if (sz == PUD_SIZE) { 646 if (pud_val(pudp[pud_index]) == 0) 647 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 648 return; 649 } 650 651 if (pud_val(pudp[pud_index]) == 0) { 652 next_phys = pt_ops.alloc_pmd(va); 653 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 654 nextp = pt_ops.get_pmd_virt(next_phys); 655 memset(nextp, 0, PAGE_SIZE); 656 } else { 657 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 658 nextp = pt_ops.get_pmd_virt(next_phys); 659 } 660 661 create_pmd_mapping(nextp, va, pa, sz, prot); 662 } 663 664 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 665 pgprot_t prot) 666 { 667 pud_t *nextp; 668 phys_addr_t next_phys; 669 uintptr_t p4d_index = p4d_index(va); 670 671 if (sz == P4D_SIZE) { 672 if (p4d_val(p4dp[p4d_index]) == 0) 673 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 674 return; 675 } 676 677 if (p4d_val(p4dp[p4d_index]) == 0) { 678 next_phys = pt_ops.alloc_pud(va); 679 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 680 nextp = pt_ops.get_pud_virt(next_phys); 681 memset(nextp, 0, PAGE_SIZE); 682 } else { 683 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 684 nextp = pt_ops.get_pud_virt(next_phys); 685 } 686 687 create_pud_mapping(nextp, va, pa, sz, prot); 688 } 689 690 #define pgd_next_t p4d_t 691 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 692 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 693 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 694 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 695 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 696 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 697 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 698 (pgtable_l5_enabled ? \ 699 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 700 (pgtable_l4_enabled ? \ 701 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 702 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 703 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 704 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 705 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 706 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 707 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 708 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 709 #else 710 #define pgd_next_t pte_t 711 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 712 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 713 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 714 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 715 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 716 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 717 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 718 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 719 #endif /* __PAGETABLE_PMD_FOLDED */ 720 721 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 722 pgprot_t prot) 723 { 724 pgd_next_t *nextp; 725 phys_addr_t next_phys; 726 uintptr_t pgd_idx = pgd_index(va); 727 728 if (sz == PGDIR_SIZE) { 729 if (pgd_val(pgdp[pgd_idx]) == 0) 730 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 731 return; 732 } 733 734 if (pgd_val(pgdp[pgd_idx]) == 0) { 735 next_phys = alloc_pgd_next(va); 736 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 737 nextp = get_pgd_next_virt(next_phys); 738 memset(nextp, 0, PAGE_SIZE); 739 } else { 740 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 741 nextp = get_pgd_next_virt(next_phys); 742 } 743 744 create_pgd_next_mapping(nextp, va, pa, sz, prot); 745 } 746 747 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 748 { 749 if (debug_pagealloc_enabled()) 750 return PAGE_SIZE; 751 752 if (pgtable_l5_enabled && 753 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 754 return P4D_SIZE; 755 756 if (pgtable_l4_enabled && 757 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 758 return PUD_SIZE; 759 760 if (IS_ENABLED(CONFIG_64BIT) && 761 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 762 return PMD_SIZE; 763 764 return PAGE_SIZE; 765 } 766 767 #ifdef CONFIG_XIP_KERNEL 768 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 769 extern char _xiprom[], _exiprom[], __data_loc; 770 771 /* called from head.S with MMU off */ 772 asmlinkage void __init __copy_data(void) 773 { 774 void *from = (void *)(&__data_loc); 775 void *to = (void *)CONFIG_PHYS_RAM_BASE; 776 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 777 778 memcpy(to, from, sz); 779 } 780 #endif 781 782 #ifdef CONFIG_STRICT_KERNEL_RWX 783 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 784 { 785 if (is_va_kernel_text(va)) 786 return PAGE_KERNEL_READ_EXEC; 787 788 /* 789 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 790 * we must protect its linear mapping alias from being executed and 791 * written. 792 * And rodata section is marked readonly in mark_rodata_ro. 793 */ 794 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 795 return PAGE_KERNEL_READ; 796 797 return PAGE_KERNEL; 798 } 799 800 void mark_rodata_ro(void) 801 { 802 set_kernel_memory(__start_rodata, _data, set_memory_ro); 803 if (IS_ENABLED(CONFIG_64BIT)) 804 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 805 set_memory_ro); 806 } 807 #else 808 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 809 { 810 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 811 return PAGE_KERNEL; 812 813 return PAGE_KERNEL_EXEC; 814 } 815 #endif /* CONFIG_STRICT_KERNEL_RWX */ 816 817 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 818 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 819 820 static void __init disable_pgtable_l5(void) 821 { 822 pgtable_l5_enabled = false; 823 kernel_map.page_offset = PAGE_OFFSET_L4; 824 satp_mode = SATP_MODE_48; 825 } 826 827 static void __init disable_pgtable_l4(void) 828 { 829 pgtable_l4_enabled = false; 830 kernel_map.page_offset = PAGE_OFFSET_L3; 831 satp_mode = SATP_MODE_39; 832 } 833 834 static int __init print_no4lvl(char *p) 835 { 836 pr_info("Disabled 4-level and 5-level paging"); 837 return 0; 838 } 839 early_param("no4lvl", print_no4lvl); 840 841 static int __init print_no5lvl(char *p) 842 { 843 pr_info("Disabled 5-level paging"); 844 return 0; 845 } 846 early_param("no5lvl", print_no5lvl); 847 848 static void __init set_mmap_rnd_bits_max(void) 849 { 850 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 851 } 852 853 /* 854 * There is a simple way to determine if 4-level is supported by the 855 * underlying hardware: establish 1:1 mapping in 4-level page table mode 856 * then read SATP to see if the configuration was taken into account 857 * meaning sv48 is supported. 858 */ 859 static __init void set_satp_mode(uintptr_t dtb_pa) 860 { 861 u64 identity_satp, hw_satp; 862 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 863 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa); 864 865 kernel_map.page_offset = PAGE_OFFSET_L5; 866 867 if (satp_mode_cmdline == SATP_MODE_57) { 868 disable_pgtable_l5(); 869 } else if (satp_mode_cmdline == SATP_MODE_48) { 870 disable_pgtable_l5(); 871 disable_pgtable_l4(); 872 return; 873 } 874 875 create_p4d_mapping(early_p4d, 876 set_satp_mode_pmd, (uintptr_t)early_pud, 877 P4D_SIZE, PAGE_TABLE); 878 create_pud_mapping(early_pud, 879 set_satp_mode_pmd, (uintptr_t)early_pmd, 880 PUD_SIZE, PAGE_TABLE); 881 /* Handle the case where set_satp_mode straddles 2 PMDs */ 882 create_pmd_mapping(early_pmd, 883 set_satp_mode_pmd, set_satp_mode_pmd, 884 PMD_SIZE, PAGE_KERNEL_EXEC); 885 create_pmd_mapping(early_pmd, 886 set_satp_mode_pmd + PMD_SIZE, 887 set_satp_mode_pmd + PMD_SIZE, 888 PMD_SIZE, PAGE_KERNEL_EXEC); 889 retry: 890 create_pgd_mapping(early_pg_dir, 891 set_satp_mode_pmd, 892 pgtable_l5_enabled ? 893 (uintptr_t)early_p4d : (uintptr_t)early_pud, 894 PGDIR_SIZE, PAGE_TABLE); 895 896 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 897 898 local_flush_tlb_all(); 899 csr_write(CSR_SATP, identity_satp); 900 hw_satp = csr_swap(CSR_SATP, 0ULL); 901 local_flush_tlb_all(); 902 903 if (hw_satp != identity_satp) { 904 if (pgtable_l5_enabled) { 905 disable_pgtable_l5(); 906 memset(early_pg_dir, 0, PAGE_SIZE); 907 goto retry; 908 } 909 disable_pgtable_l4(); 910 } 911 912 memset(early_pg_dir, 0, PAGE_SIZE); 913 memset(early_p4d, 0, PAGE_SIZE); 914 memset(early_pud, 0, PAGE_SIZE); 915 memset(early_pmd, 0, PAGE_SIZE); 916 } 917 #endif 918 919 /* 920 * setup_vm() is called from head.S with MMU-off. 921 * 922 * Following requirements should be honoured for setup_vm() to work 923 * correctly: 924 * 1) It should use PC-relative addressing for accessing kernel symbols. 925 * To achieve this we always use GCC cmodel=medany. 926 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 927 * so disable compiler instrumentation when FTRACE is enabled. 928 * 929 * Currently, the above requirements are honoured by using custom CFLAGS 930 * for init.o in mm/Makefile. 931 */ 932 933 #ifndef __riscv_cmodel_medany 934 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 935 #endif 936 937 #ifdef CONFIG_XIP_KERNEL 938 static void __init create_kernel_page_table(pgd_t *pgdir, 939 __always_unused bool early) 940 { 941 uintptr_t va, start_va, end_va; 942 943 /* Map the flash resident part */ 944 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 945 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 946 create_pgd_mapping(pgdir, va, 947 kernel_map.xiprom + (va - kernel_map.virt_addr), 948 PMD_SIZE, PAGE_KERNEL_EXEC); 949 950 /* Map the data in RAM */ 951 start_va = kernel_map.virt_addr + (uintptr_t)&_sdata - (uintptr_t)&_start; 952 end_va = kernel_map.virt_addr + kernel_map.size; 953 for (va = start_va; va < end_va; va += PMD_SIZE) 954 create_pgd_mapping(pgdir, va, 955 kernel_map.phys_addr + (va - start_va), 956 PMD_SIZE, PAGE_KERNEL); 957 } 958 #else 959 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 960 { 961 uintptr_t va, end_va; 962 963 end_va = kernel_map.virt_addr + kernel_map.size; 964 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 965 create_pgd_mapping(pgdir, va, 966 kernel_map.phys_addr + (va - kernel_map.virt_addr), 967 PMD_SIZE, 968 early ? 969 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 970 } 971 #endif 972 973 /* 974 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 975 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 976 * entry. 977 */ 978 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 979 uintptr_t dtb_pa) 980 { 981 #ifndef CONFIG_BUILTIN_DTB 982 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 983 984 /* Make sure the fdt fixmap address is always aligned on PMD size */ 985 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 986 987 /* In 32-bit only, the fdt lies in its own PGD */ 988 if (!IS_ENABLED(CONFIG_64BIT)) { 989 create_pgd_mapping(early_pg_dir, fix_fdt_va, 990 pa, MAX_FDT_SIZE, PAGE_KERNEL); 991 } else { 992 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 993 pa, PMD_SIZE, PAGE_KERNEL); 994 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 995 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 996 } 997 998 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 999 #else 1000 /* 1001 * For 64-bit kernel, __va can't be used since it would return a linear 1002 * mapping address whereas dtb_early_va will be used before 1003 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 1004 * kernel is mapped in the linear mapping, that makes no difference. 1005 */ 1006 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 1007 #endif 1008 1009 dtb_early_pa = dtb_pa; 1010 } 1011 1012 /* 1013 * MMU is not enabled, the page tables are allocated directly using 1014 * early_pmd/pud/p4d and the address returned is the physical one. 1015 */ 1016 static void __init pt_ops_set_early(void) 1017 { 1018 pt_ops.alloc_pte = alloc_pte_early; 1019 pt_ops.get_pte_virt = get_pte_virt_early; 1020 #ifndef __PAGETABLE_PMD_FOLDED 1021 pt_ops.alloc_pmd = alloc_pmd_early; 1022 pt_ops.get_pmd_virt = get_pmd_virt_early; 1023 pt_ops.alloc_pud = alloc_pud_early; 1024 pt_ops.get_pud_virt = get_pud_virt_early; 1025 pt_ops.alloc_p4d = alloc_p4d_early; 1026 pt_ops.get_p4d_virt = get_p4d_virt_early; 1027 #endif 1028 } 1029 1030 /* 1031 * MMU is enabled but page table setup is not complete yet. 1032 * fixmap page table alloc functions must be used as a means to temporarily 1033 * map the allocated physical pages since the linear mapping does not exist yet. 1034 * 1035 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1036 * but it will be used as described above. 1037 */ 1038 static void __init pt_ops_set_fixmap(void) 1039 { 1040 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1041 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1042 #ifndef __PAGETABLE_PMD_FOLDED 1043 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1044 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1045 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1046 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1047 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1048 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1049 #endif 1050 } 1051 1052 /* 1053 * MMU is enabled and page table setup is complete, so from now, we can use 1054 * generic page allocation functions to setup page table. 1055 */ 1056 static void __init pt_ops_set_late(void) 1057 { 1058 pt_ops.alloc_pte = alloc_pte_late; 1059 pt_ops.get_pte_virt = get_pte_virt_late; 1060 #ifndef __PAGETABLE_PMD_FOLDED 1061 pt_ops.alloc_pmd = alloc_pmd_late; 1062 pt_ops.get_pmd_virt = get_pmd_virt_late; 1063 pt_ops.alloc_pud = alloc_pud_late; 1064 pt_ops.get_pud_virt = get_pud_virt_late; 1065 pt_ops.alloc_p4d = alloc_p4d_late; 1066 pt_ops.get_p4d_virt = get_p4d_virt_late; 1067 #endif 1068 } 1069 1070 #ifdef CONFIG_RANDOMIZE_BASE 1071 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1072 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1073 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa); 1074 1075 static int __init print_nokaslr(char *p) 1076 { 1077 pr_info("Disabled KASLR"); 1078 return 0; 1079 } 1080 early_param("nokaslr", print_nokaslr); 1081 1082 unsigned long kaslr_offset(void) 1083 { 1084 return kernel_map.virt_offset; 1085 } 1086 #endif 1087 1088 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1089 { 1090 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1091 1092 #ifdef CONFIG_RANDOMIZE_BASE 1093 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1094 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa); 1095 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1096 u32 nr_pos; 1097 1098 if (kaslr_seed == 0) 1099 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1100 /* 1101 * Compute the number of positions available: we are limited 1102 * by the early page table that only has one PUD and we must 1103 * be aligned on PMD_SIZE. 1104 */ 1105 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1106 1107 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1108 } 1109 #endif 1110 1111 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1112 1113 #ifdef CONFIG_XIP_KERNEL 1114 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1115 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1116 1117 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1118 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1119 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 1120 #endif 1121 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1122 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1123 1124 kernel_map.va_kernel_xip_text_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1125 kernel_map.va_kernel_xip_data_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr 1126 + (uintptr_t)&_sdata - (uintptr_t)&_start; 1127 #else 1128 kernel_map.phys_addr = (uintptr_t)(&_start); 1129 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1130 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1131 #endif 1132 1133 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1134 set_satp_mode(dtb_pa); 1135 set_mmap_rnd_bits_max(); 1136 #endif 1137 1138 /* 1139 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1140 * where we have the system memory layout: this allows us to align 1141 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1142 * for the linear mapping. This is only possible because the kernel 1143 * mapping lies outside the linear mapping. 1144 * In 32-bit however, as the kernel resides in the linear mapping, 1145 * setup_vm_final can not change the mapping established here, 1146 * otherwise the same kernel addresses would get mapped to different 1147 * physical addresses (if the start of dram is different from the 1148 * kernel physical address start). 1149 */ 1150 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1151 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1152 1153 memory_limit = KERN_VIRT_SIZE; 1154 1155 /* Sanity check alignment and size */ 1156 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1157 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1158 1159 #ifdef CONFIG_64BIT 1160 /* 1161 * The last 4K bytes of the addressable memory can not be mapped because 1162 * of IS_ERR_VALUE macro. 1163 */ 1164 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1165 #endif 1166 1167 #ifdef CONFIG_RELOCATABLE 1168 /* 1169 * Early page table uses only one PUD, which makes it possible 1170 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1171 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1172 * since a part of the kernel would not get mapped. 1173 */ 1174 if (IS_ENABLED(CONFIG_64BIT)) 1175 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1176 relocate_kernel(); 1177 #endif 1178 1179 apply_early_boot_alternatives(); 1180 pt_ops_set_early(); 1181 1182 /* Setup early PGD for fixmap */ 1183 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1184 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1185 1186 #ifndef __PAGETABLE_PMD_FOLDED 1187 /* Setup fixmap P4D and PUD */ 1188 if (pgtable_l5_enabled) 1189 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1190 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1191 /* Setup fixmap PUD and PMD */ 1192 if (pgtable_l4_enabled) 1193 create_pud_mapping(fixmap_pud, FIXADDR_START, 1194 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1195 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1196 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1197 /* Setup trampoline PGD and PMD */ 1198 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1199 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1200 if (pgtable_l5_enabled) 1201 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1202 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1203 if (pgtable_l4_enabled) 1204 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1205 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1206 #ifdef CONFIG_XIP_KERNEL 1207 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1208 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1209 #else 1210 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1211 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1212 #endif 1213 #else 1214 /* Setup trampoline PGD */ 1215 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1216 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1217 #endif 1218 1219 /* 1220 * Setup early PGD covering entire kernel which will allow 1221 * us to reach paging_init(). We map all memory banks later 1222 * in setup_vm_final() below. 1223 */ 1224 create_kernel_page_table(early_pg_dir, true); 1225 1226 /* Setup early mapping for FDT early scan */ 1227 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1228 1229 /* 1230 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1231 * range can not span multiple pmds. 1232 */ 1233 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1234 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1235 1236 #ifndef __PAGETABLE_PMD_FOLDED 1237 /* 1238 * Early ioremap fixmap is already created as it lies within first 2MB 1239 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1240 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1241 * the user if not. 1242 */ 1243 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1244 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1245 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1246 WARN_ON(1); 1247 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1248 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1249 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1250 fix_to_virt(FIX_BTMAP_BEGIN)); 1251 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1252 fix_to_virt(FIX_BTMAP_END)); 1253 1254 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1255 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1256 } 1257 #endif 1258 1259 pt_ops_set_fixmap(); 1260 } 1261 1262 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end, 1263 uintptr_t fixed_map_size, const pgprot_t *pgprot) 1264 { 1265 phys_addr_t pa; 1266 uintptr_t va, map_size; 1267 1268 for (pa = start; pa < end; pa += map_size) { 1269 va = (uintptr_t)__va(pa); 1270 map_size = fixed_map_size ? fixed_map_size : 1271 best_map_size(pa, va, end - pa); 1272 1273 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1274 pgprot ? *pgprot : pgprot_from_va(va)); 1275 } 1276 } 1277 1278 static void __init create_linear_mapping_page_table(void) 1279 { 1280 phys_addr_t start, end; 1281 phys_addr_t kfence_pool __maybe_unused; 1282 u64 i; 1283 1284 #ifdef CONFIG_STRICT_KERNEL_RWX 1285 phys_addr_t ktext_start = __pa_symbol(_start); 1286 phys_addr_t ktext_size = __init_data_begin - _start; 1287 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1288 phys_addr_t krodata_size = _data - __start_rodata; 1289 1290 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1291 memblock_mark_nomap(ktext_start, ktext_size); 1292 memblock_mark_nomap(krodata_start, krodata_size); 1293 #endif 1294 1295 #ifdef CONFIG_KFENCE 1296 /* 1297 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1298 * before we setup the linear mapping so that we avoid using hugepages 1299 * for this region. 1300 */ 1301 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1302 BUG_ON(!kfence_pool); 1303 1304 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1305 __kfence_pool = __va(kfence_pool); 1306 #endif 1307 1308 /* Map all memory banks in the linear mapping */ 1309 for_each_mem_range(i, &start, &end) { 1310 if (start >= end) 1311 break; 1312 if (start <= __pa(PAGE_OFFSET) && 1313 __pa(PAGE_OFFSET) < end) 1314 start = __pa(PAGE_OFFSET); 1315 1316 create_linear_mapping_range(start, end, 0, NULL); 1317 } 1318 1319 #ifdef CONFIG_STRICT_KERNEL_RWX 1320 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL); 1321 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL); 1322 1323 memblock_clear_nomap(ktext_start, ktext_size); 1324 memblock_clear_nomap(krodata_start, krodata_size); 1325 #endif 1326 1327 #ifdef CONFIG_KFENCE 1328 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL); 1329 1330 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1331 #endif 1332 } 1333 1334 static void __init setup_vm_final(void) 1335 { 1336 /* Setup swapper PGD for fixmap */ 1337 #if !defined(CONFIG_64BIT) 1338 /* 1339 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1340 * directly in swapper_pg_dir in addition to the pgd entry that points 1341 * to fixmap_pte. 1342 */ 1343 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1344 1345 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1346 #endif 1347 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1348 __pa_symbol(fixmap_pgd_next), 1349 PGDIR_SIZE, PAGE_TABLE); 1350 1351 /* Map the linear mapping */ 1352 create_linear_mapping_page_table(); 1353 1354 /* Map the kernel */ 1355 if (IS_ENABLED(CONFIG_64BIT)) 1356 create_kernel_page_table(swapper_pg_dir, false); 1357 1358 #ifdef CONFIG_KASAN 1359 kasan_swapper_init(); 1360 #endif 1361 1362 /* Clear fixmap PTE and PMD mappings */ 1363 clear_fixmap(FIX_PTE); 1364 clear_fixmap(FIX_PMD); 1365 clear_fixmap(FIX_PUD); 1366 clear_fixmap(FIX_P4D); 1367 1368 /* Move to swapper page table */ 1369 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1370 local_flush_tlb_all(); 1371 1372 pt_ops_set_late(); 1373 } 1374 #else 1375 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1376 { 1377 dtb_early_va = (void *)dtb_pa; 1378 dtb_early_pa = dtb_pa; 1379 1380 #ifdef CONFIG_RELOCATABLE 1381 kernel_map.virt_addr = (uintptr_t)_start; 1382 kernel_map.phys_addr = (uintptr_t)_start; 1383 relocate_kernel(); 1384 #endif 1385 } 1386 1387 static inline void setup_vm_final(void) 1388 { 1389 } 1390 #endif /* CONFIG_MMU */ 1391 1392 /* 1393 * reserve_crashkernel() - reserves memory for crash kernel 1394 * 1395 * This function reserves memory area given in "crashkernel=" kernel command 1396 * line parameter. The memory reserved is used by dump capture kernel when 1397 * primary kernel is crashing. 1398 */ 1399 static void __init arch_reserve_crashkernel(void) 1400 { 1401 unsigned long long low_size = 0; 1402 unsigned long long crash_base, crash_size; 1403 bool high = false; 1404 int ret; 1405 1406 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1407 return; 1408 1409 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1410 &crash_size, &crash_base, 1411 &low_size, &high); 1412 if (ret) 1413 return; 1414 1415 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 1416 } 1417 1418 void __init paging_init(void) 1419 { 1420 setup_bootmem(); 1421 setup_vm_final(); 1422 1423 /* Depend on that Linear Mapping is ready */ 1424 memblock_allow_resize(); 1425 } 1426 1427 void __init misc_mem_init(void) 1428 { 1429 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1430 arch_numa_init(); 1431 sparse_init(); 1432 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1433 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1434 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1435 #endif 1436 zone_sizes_init(); 1437 arch_reserve_crashkernel(); 1438 memblock_dump_all(); 1439 } 1440 1441 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1442 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1443 unsigned long addr, unsigned long next) 1444 { 1445 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1446 } 1447 1448 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1449 unsigned long addr, unsigned long next) 1450 { 1451 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1452 return 1; 1453 } 1454 1455 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1456 struct vmem_altmap *altmap) 1457 { 1458 /* 1459 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1460 * can't use hugepage mappings for 2-level page table because in case of 1461 * memory hotplug, we are not able to update all the page tables with 1462 * the new PMDs. 1463 */ 1464 return vmemmap_populate_hugepages(start, end, node, altmap); 1465 } 1466 #endif 1467 1468 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1469 /* 1470 * Pre-allocates page-table pages for a specific area in the kernel 1471 * page-table. Only the level which needs to be synchronized between 1472 * all page-tables is allocated because the synchronization can be 1473 * expensive. 1474 */ 1475 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1476 const char *area) 1477 { 1478 unsigned long addr; 1479 const char *lvl; 1480 1481 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1482 pgd_t *pgd = pgd_offset_k(addr); 1483 p4d_t *p4d; 1484 pud_t *pud; 1485 pmd_t *pmd; 1486 1487 lvl = "p4d"; 1488 p4d = p4d_alloc(&init_mm, pgd, addr); 1489 if (!p4d) 1490 goto failed; 1491 1492 if (pgtable_l5_enabled) 1493 continue; 1494 1495 lvl = "pud"; 1496 pud = pud_alloc(&init_mm, p4d, addr); 1497 if (!pud) 1498 goto failed; 1499 1500 if (pgtable_l4_enabled) 1501 continue; 1502 1503 lvl = "pmd"; 1504 pmd = pmd_alloc(&init_mm, pud, addr); 1505 if (!pmd) 1506 goto failed; 1507 } 1508 return; 1509 1510 failed: 1511 /* 1512 * The pages have to be there now or they will be missing in 1513 * process page-tables later. 1514 */ 1515 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1516 } 1517 1518 #define PAGE_END KASAN_SHADOW_START 1519 1520 void __init pgtable_cache_init(void) 1521 { 1522 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1523 if (IS_ENABLED(CONFIG_MODULES)) 1524 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1525 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1526 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1527 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1528 if (IS_ENABLED(CONFIG_KASAN)) 1529 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1530 } 1531 } 1532 #endif 1533 1534 #ifdef CONFIG_EXECMEM 1535 #ifdef CONFIG_MMU 1536 static struct execmem_info execmem_info __ro_after_init; 1537 1538 struct execmem_info __init *execmem_arch_setup(void) 1539 { 1540 execmem_info = (struct execmem_info){ 1541 .ranges = { 1542 [EXECMEM_DEFAULT] = { 1543 .start = MODULES_VADDR, 1544 .end = MODULES_END, 1545 .pgprot = PAGE_KERNEL, 1546 .alignment = 1, 1547 }, 1548 [EXECMEM_KPROBES] = { 1549 .start = VMALLOC_START, 1550 .end = VMALLOC_END, 1551 .pgprot = PAGE_KERNEL_READ_EXEC, 1552 .alignment = 1, 1553 }, 1554 [EXECMEM_BPF] = { 1555 .start = BPF_JIT_REGION_START, 1556 .end = BPF_JIT_REGION_END, 1557 .pgprot = PAGE_KERNEL, 1558 .alignment = PAGE_SIZE, 1559 }, 1560 }, 1561 }; 1562 1563 return &execmem_info; 1564 } 1565 #endif /* CONFIG_MMU */ 1566 #endif /* CONFIG_EXECMEM */ 1567 1568 #ifdef CONFIG_MEMORY_HOTPLUG 1569 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1570 { 1571 struct page *page = pmd_page(*pmd); 1572 struct ptdesc *ptdesc = page_ptdesc(page); 1573 pte_t *pte; 1574 int i; 1575 1576 for (i = 0; i < PTRS_PER_PTE; i++) { 1577 pte = pte_start + i; 1578 if (!pte_none(*pte)) 1579 return; 1580 } 1581 1582 pagetable_dtor(ptdesc); 1583 if (PageReserved(page)) 1584 free_reserved_page(page); 1585 else 1586 pagetable_free(ptdesc); 1587 pmd_clear(pmd); 1588 } 1589 1590 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1591 { 1592 struct page *page = pud_page(*pud); 1593 struct ptdesc *ptdesc = page_ptdesc(page); 1594 pmd_t *pmd; 1595 int i; 1596 1597 for (i = 0; i < PTRS_PER_PMD; i++) { 1598 pmd = pmd_start + i; 1599 if (!pmd_none(*pmd)) 1600 return; 1601 } 1602 1603 if (!is_vmemmap) 1604 pagetable_dtor(ptdesc); 1605 if (PageReserved(page)) 1606 free_reserved_page(page); 1607 else 1608 pagetable_free(ptdesc); 1609 pud_clear(pud); 1610 } 1611 1612 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1613 { 1614 struct page *page = p4d_page(*p4d); 1615 pud_t *pud; 1616 int i; 1617 1618 for (i = 0; i < PTRS_PER_PUD; i++) { 1619 pud = pud_start + i; 1620 if (!pud_none(*pud)) 1621 return; 1622 } 1623 1624 if (PageReserved(page)) 1625 free_reserved_page(page); 1626 else 1627 free_pages((unsigned long)page_address(page), 0); 1628 p4d_clear(p4d); 1629 } 1630 1631 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1632 struct vmem_altmap *altmap) 1633 { 1634 int order = get_order(size); 1635 1636 if (altmap) { 1637 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1638 return; 1639 } 1640 1641 if (PageReserved(page)) { 1642 unsigned int nr_pages = 1 << order; 1643 1644 while (nr_pages--) 1645 free_reserved_page(page++); 1646 return; 1647 } 1648 1649 free_pages((unsigned long)page_address(page), order); 1650 } 1651 1652 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1653 bool is_vmemmap, struct vmem_altmap *altmap) 1654 { 1655 unsigned long next; 1656 pte_t *ptep, pte; 1657 1658 for (; addr < end; addr = next) { 1659 next = (addr + PAGE_SIZE) & PAGE_MASK; 1660 if (next > end) 1661 next = end; 1662 1663 ptep = pte_base + pte_index(addr); 1664 pte = ptep_get(ptep); 1665 if (!pte_present(*ptep)) 1666 continue; 1667 1668 pte_clear(&init_mm, addr, ptep); 1669 if (is_vmemmap) 1670 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1671 } 1672 } 1673 1674 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1675 bool is_vmemmap, struct vmem_altmap *altmap) 1676 { 1677 unsigned long next; 1678 pte_t *pte_base; 1679 pmd_t *pmdp, pmd; 1680 1681 for (; addr < end; addr = next) { 1682 next = pmd_addr_end(addr, end); 1683 pmdp = pmd_base + pmd_index(addr); 1684 pmd = pmdp_get(pmdp); 1685 if (!pmd_present(pmd)) 1686 continue; 1687 1688 if (pmd_leaf(pmd)) { 1689 pmd_clear(pmdp); 1690 if (is_vmemmap) 1691 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1692 continue; 1693 } 1694 1695 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1696 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1697 free_pte_table(pte_base, pmdp); 1698 } 1699 } 1700 1701 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1702 bool is_vmemmap, struct vmem_altmap *altmap) 1703 { 1704 unsigned long next; 1705 pud_t *pudp, pud; 1706 pmd_t *pmd_base; 1707 1708 for (; addr < end; addr = next) { 1709 next = pud_addr_end(addr, end); 1710 pudp = pud_base + pud_index(addr); 1711 pud = pudp_get(pudp); 1712 if (!pud_present(pud)) 1713 continue; 1714 1715 if (pud_leaf(pud)) { 1716 if (pgtable_l4_enabled) { 1717 pud_clear(pudp); 1718 if (is_vmemmap) 1719 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1720 } 1721 continue; 1722 } 1723 1724 pmd_base = pmd_offset(pudp, 0); 1725 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1726 1727 if (pgtable_l4_enabled) 1728 free_pmd_table(pmd_base, pudp, is_vmemmap); 1729 } 1730 } 1731 1732 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1733 bool is_vmemmap, struct vmem_altmap *altmap) 1734 { 1735 unsigned long next; 1736 p4d_t *p4dp, p4d; 1737 pud_t *pud_base; 1738 1739 for (; addr < end; addr = next) { 1740 next = p4d_addr_end(addr, end); 1741 p4dp = p4d_base + p4d_index(addr); 1742 p4d = p4dp_get(p4dp); 1743 if (!p4d_present(p4d)) 1744 continue; 1745 1746 if (p4d_leaf(p4d)) { 1747 if (pgtable_l5_enabled) { 1748 p4d_clear(p4dp); 1749 if (is_vmemmap) 1750 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1751 } 1752 continue; 1753 } 1754 1755 pud_base = pud_offset(p4dp, 0); 1756 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1757 1758 if (pgtable_l5_enabled) 1759 free_pud_table(pud_base, p4dp); 1760 } 1761 } 1762 1763 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1764 struct vmem_altmap *altmap) 1765 { 1766 unsigned long addr, next; 1767 p4d_t *p4d_base; 1768 pgd_t *pgd; 1769 1770 for (addr = va; addr < end; addr = next) { 1771 next = pgd_addr_end(addr, end); 1772 pgd = pgd_offset_k(addr); 1773 1774 if (!pgd_present(*pgd)) 1775 continue; 1776 1777 if (pgd_leaf(*pgd)) 1778 continue; 1779 1780 p4d_base = p4d_offset(pgd, 0); 1781 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1782 } 1783 1784 flush_tlb_all(); 1785 } 1786 1787 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1788 { 1789 unsigned long va = (unsigned long)__va(start); 1790 unsigned long end = (unsigned long)__va(start + size); 1791 1792 remove_pgd_mapping(va, end, false, NULL); 1793 } 1794 1795 struct range arch_get_mappable_range(void) 1796 { 1797 struct range mhp_range; 1798 1799 mhp_range.start = __pa(PAGE_OFFSET); 1800 mhp_range.end = __pa(PAGE_END - 1); 1801 return mhp_range; 1802 } 1803 1804 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1805 { 1806 int ret = 0; 1807 1808 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1809 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1810 if (ret) { 1811 remove_linear_mapping(start, size); 1812 goto out; 1813 } 1814 1815 max_pfn = PFN_UP(start + size); 1816 max_low_pfn = max_pfn; 1817 1818 out: 1819 flush_tlb_all(); 1820 return ret; 1821 } 1822 1823 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 1824 { 1825 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap); 1826 remove_linear_mapping(start, size); 1827 flush_tlb_all(); 1828 } 1829 1830 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1831 { 1832 remove_pgd_mapping(start, end, true, altmap); 1833 } 1834 #endif /* CONFIG_MEMORY_HOTPLUG */ 1835