1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2012 Regents of the University of California 4 * Copyright (C) 2019 Western Digital Corporation or its affiliates. 5 * Copyright (C) 2020 FORTH-ICS/CARV 6 * Nick Kossifidis <mick@ics.forth.gr> 7 */ 8 9 #include <linux/init.h> 10 #include <linux/mm.h> 11 #include <linux/memblock.h> 12 #include <linux/initrd.h> 13 #include <linux/swap.h> 14 #include <linux/swiotlb.h> 15 #include <linux/sizes.h> 16 #include <linux/of_fdt.h> 17 #include <linux/of_reserved_mem.h> 18 #include <linux/libfdt.h> 19 #include <linux/set_memory.h> 20 #include <linux/dma-map-ops.h> 21 #include <linux/crash_dump.h> 22 #include <linux/hugetlb.h> 23 #include <linux/kfence.h> 24 #include <linux/execmem.h> 25 26 #include <asm/alternative.h> 27 #include <asm/fixmap.h> 28 #include <asm/io.h> 29 #include <asm/kasan.h> 30 #include <asm/module.h> 31 #include <asm/numa.h> 32 #include <asm/pgtable.h> 33 #include <asm/sections.h> 34 #include <asm/soc.h> 35 #include <asm/sparsemem.h> 36 #include <asm/tlbflush.h> 37 38 #include "../kernel/head.h" 39 40 u64 new_vmalloc[NR_CPUS / sizeof(u64) + 1]; 41 42 struct kernel_mapping kernel_map __ro_after_init; 43 EXPORT_SYMBOL(kernel_map); 44 #ifdef CONFIG_XIP_KERNEL 45 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 46 #endif 47 48 #ifdef CONFIG_64BIT 49 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 50 #else 51 u64 satp_mode __ro_after_init = SATP_MODE_32; 52 #endif 53 EXPORT_SYMBOL(satp_mode); 54 55 #ifdef CONFIG_64BIT 56 bool pgtable_l4_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 57 bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL); 58 EXPORT_SYMBOL(pgtable_l4_enabled); 59 EXPORT_SYMBOL(pgtable_l5_enabled); 60 #endif 61 62 phys_addr_t phys_ram_base __ro_after_init; 63 EXPORT_SYMBOL(phys_ram_base); 64 65 #ifdef CONFIG_SPARSEMEM_VMEMMAP 66 #define VMEMMAP_ADDR_ALIGN (1ULL << SECTION_SIZE_BITS) 67 68 unsigned long vmemmap_start_pfn __ro_after_init; 69 EXPORT_SYMBOL(vmemmap_start_pfn); 70 #endif 71 72 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 73 __page_aligned_bss; 74 EXPORT_SYMBOL(empty_zero_page); 75 76 extern char _start[]; 77 void *_dtb_early_va __initdata; 78 uintptr_t _dtb_early_pa __initdata; 79 80 phys_addr_t dma32_phys_limit __initdata; 81 82 void __init arch_zone_limits_init(unsigned long *max_zone_pfns) 83 { 84 #ifdef CONFIG_ZONE_DMA32 85 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 86 #endif 87 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 88 } 89 90 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 91 92 #define LOG2_SZ_1K ilog2(SZ_1K) 93 #define LOG2_SZ_1M ilog2(SZ_1M) 94 #define LOG2_SZ_1G ilog2(SZ_1G) 95 #define LOG2_SZ_1T ilog2(SZ_1T) 96 97 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 98 { 99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 100 (((t) - (b)) >> LOG2_SZ_1K)); 101 } 102 103 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 104 { 105 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 106 (((t) - (b)) >> LOG2_SZ_1M)); 107 } 108 109 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 110 { 111 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 112 (((t) - (b)) >> LOG2_SZ_1G)); 113 } 114 115 #ifdef CONFIG_64BIT 116 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 117 { 118 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 119 (((t) - (b)) >> LOG2_SZ_1T)); 120 } 121 #else 122 #define print_mlt(n, b, t) do {} while (0) 123 #endif 124 125 static inline void print_ml(char *name, unsigned long b, unsigned long t) 126 { 127 unsigned long diff = t - b; 128 129 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 130 print_mlt(name, b, t); 131 else if ((diff >> LOG2_SZ_1G) >= 10) 132 print_mlg(name, b, t); 133 else if ((diff >> LOG2_SZ_1M) >= 10) 134 print_mlm(name, b, t); 135 else 136 print_mlk(name, b, t); 137 } 138 139 static void __init print_vm_layout(void) 140 { 141 pr_notice("Virtual kernel memory layout:\n"); 142 print_ml("fixmap", (unsigned long)FIXADDR_START, 143 (unsigned long)FIXADDR_TOP); 144 print_ml("pci io", (unsigned long)PCI_IO_START, 145 (unsigned long)PCI_IO_END); 146 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 147 (unsigned long)VMEMMAP_END); 148 print_ml("vmalloc", (unsigned long)VMALLOC_START, 149 (unsigned long)VMALLOC_END); 150 #ifdef CONFIG_64BIT 151 print_ml("modules", (unsigned long)MODULES_VADDR, 152 (unsigned long)MODULES_END); 153 #endif 154 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 155 (unsigned long)high_memory); 156 if (IS_ENABLED(CONFIG_64BIT)) { 157 #ifdef CONFIG_KASAN 158 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 159 #endif 160 161 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 162 (unsigned long)ADDRESS_SPACE_END); 163 } 164 } 165 #else 166 static void print_vm_layout(void) { } 167 #endif /* CONFIG_DEBUG_VM */ 168 169 void __init arch_mm_preinit(void) 170 { 171 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 172 #ifdef CONFIG_FLATMEM 173 BUG_ON(!mem_map); 174 #endif /* CONFIG_FLATMEM */ 175 176 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 177 dma_cache_alignment != 1) { 178 /* 179 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 180 * buffer per 1GB of RAM for kmalloc() bouncing on 181 * non-coherent platforms. 182 */ 183 unsigned long size = 184 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 185 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 186 swiotlb = true; 187 } 188 189 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 190 191 print_vm_layout(); 192 } 193 194 /* Limit the memory size via mem. */ 195 static phys_addr_t memory_limit; 196 #ifdef CONFIG_XIP_KERNEL 197 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 198 #endif /* CONFIG_XIP_KERNEL */ 199 200 static int __init early_mem(char *p) 201 { 202 u64 size; 203 204 if (!p) 205 return 1; 206 207 size = memparse(p, &p) & PAGE_MASK; 208 memory_limit = min_t(u64, size, memory_limit); 209 210 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 211 212 return 0; 213 } 214 early_param("mem", early_mem); 215 216 static void __init setup_bootmem(void) 217 { 218 phys_addr_t vmlinux_end = __pa_symbol(&_end); 219 phys_addr_t max_mapped_addr; 220 phys_addr_t phys_ram_end, vmlinux_start; 221 222 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 223 vmlinux_start = __pa_symbol(&_sdata); 224 else 225 vmlinux_start = __pa_symbol(&_start); 226 227 memblock_enforce_memory_limit(memory_limit); 228 229 /* 230 * Make sure we align the reservation on PMD_SIZE since we will 231 * map the kernel in the linear mapping as read-only: we do not want 232 * any allocation to happen between _end and the next pmd aligned page. 233 */ 234 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 235 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 236 /* 237 * Reserve from the start of the kernel to the end of the kernel 238 */ 239 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 240 241 /* 242 * Make sure we align the start of the memory on a PMD boundary so that 243 * at worst, we map the linear mapping with PMD mappings. 244 */ 245 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) { 246 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 247 #ifdef CONFIG_SPARSEMEM_VMEMMAP 248 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 249 #endif 250 } 251 252 /* 253 * In 64-bit, any use of __va/__pa before this point is wrong as we 254 * did not know the start of DRAM before. 255 */ 256 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 257 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 258 259 /* 260 * The size of the linear page mapping may restrict the amount of 261 * usable RAM. 262 */ 263 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) { 264 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE; 265 if (memblock_end_of_DRAM() > max_mapped_addr) { 266 memblock_cap_memory_range(phys_ram_base, 267 max_mapped_addr - phys_ram_base); 268 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed", 269 &max_mapped_addr); 270 } 271 } 272 273 /* 274 * Reserve physical address space that would be mapped to virtual 275 * addresses greater than (void *)(-PAGE_SIZE) because: 276 * - This memory would overlap with ERR_PTR 277 * - This memory belongs to high memory, which is not supported 278 * 279 * This is not applicable to 64-bit kernel, because virtual addresses 280 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 281 * occupied by kernel mapping. Also it is unrealistic for high memory 282 * to exist on 64-bit platforms. 283 */ 284 if (!IS_ENABLED(CONFIG_64BIT)) { 285 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 286 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 287 } 288 289 phys_ram_end = memblock_end_of_DRAM(); 290 min_low_pfn = PFN_UP(phys_ram_base); 291 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 292 293 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 294 295 reserve_initrd_mem(); 296 297 /* 298 * No allocation should be done before reserving the memory as defined 299 * in the device tree, otherwise the allocation could end up in a 300 * reserved region. 301 */ 302 early_init_fdt_scan_reserved_mem(); 303 304 /* 305 * If DTB is built in, no need to reserve its memblock. 306 * Otherwise, do reserve it but avoid using 307 * early_init_fdt_reserve_self() since __pa() does 308 * not work for DTB pointers that are fixmap addresses 309 */ 310 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 311 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 312 313 dma_contiguous_reserve(dma32_phys_limit); 314 } 315 316 #ifdef CONFIG_RELOCATABLE 317 extern unsigned long __rela_dyn_start, __rela_dyn_end; 318 319 static void __init relocate_kernel(void) 320 { 321 Elf_Rela *rela = (Elf_Rela *)&__rela_dyn_start; 322 /* 323 * This holds the offset between the linked virtual address and the 324 * relocated virtual address. 325 */ 326 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 327 /* 328 * This holds the offset between kernel linked virtual address and 329 * physical address. 330 */ 331 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 332 333 for ( ; rela < (Elf_Rela *)&__rela_dyn_end; rela++) { 334 Elf_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 335 Elf_Addr relocated_addr = rela->r_addend; 336 337 if (rela->r_info != R_RISCV_RELATIVE) 338 continue; 339 340 /* 341 * Make sure to not relocate vdso symbols like rt_sigreturn 342 * which are linked from the address 0 in vmlinux since 343 * vdso symbol addresses are actually used as an offset from 344 * mm->context.vdso in VDSO_OFFSET macro. 345 */ 346 if (relocated_addr >= KERNEL_LINK_ADDR) 347 relocated_addr += reloc_offset; 348 349 *(Elf_Addr *)addr = relocated_addr; 350 } 351 } 352 #endif /* CONFIG_RELOCATABLE */ 353 354 #ifdef CONFIG_MMU 355 struct pt_alloc_ops pt_ops __meminitdata; 356 357 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 358 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 359 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 360 361 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 362 363 #ifdef CONFIG_XIP_KERNEL 364 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 365 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 366 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 367 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 368 #endif /* CONFIG_XIP_KERNEL */ 369 370 static const pgprot_t protection_map[16] = { 371 [VM_NONE] = PAGE_NONE, 372 [VM_READ] = PAGE_READ, 373 [VM_WRITE] = PAGE_SHADOWSTACK, 374 [VM_WRITE | VM_READ] = PAGE_COPY, 375 [VM_EXEC] = PAGE_EXEC, 376 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 377 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 378 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 379 [VM_SHARED] = PAGE_NONE, 380 [VM_SHARED | VM_READ] = PAGE_READ, 381 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 382 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 383 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 384 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 385 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 386 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 387 }; 388 DECLARE_VM_GET_PAGE_PROT 389 390 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 391 { 392 unsigned long addr = __fix_to_virt(idx); 393 pte_t *ptep; 394 395 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 396 397 ptep = &fixmap_pte[pte_index(addr)]; 398 399 if (pgprot_val(prot)) 400 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 401 else 402 pte_clear(&init_mm, addr, ptep); 403 local_flush_tlb_page(addr); 404 } 405 406 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 407 { 408 return (pte_t *)((uintptr_t)pa); 409 } 410 411 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 412 { 413 clear_fixmap(FIX_PTE); 414 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 415 } 416 417 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa) 418 { 419 return (pte_t *) __va(pa); 420 } 421 422 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 423 { 424 /* 425 * We only create PMD or PGD early mappings so we 426 * should never reach here with MMU disabled. 427 */ 428 BUG(); 429 } 430 431 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 432 { 433 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 434 } 435 436 static phys_addr_t __meminit alloc_pte_late(uintptr_t va) 437 { 438 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 439 440 /* 441 * We do not know which mm the PTE page is associated to at this point. 442 * Passing NULL to the ctor is the safe option, though it may result 443 * in unnecessary work (e.g. initialising the ptlock for init_mm). 444 */ 445 BUG_ON(!ptdesc || !pagetable_pte_ctor(NULL, 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, 0); 524 525 /* See comment in alloc_pte_late() regarding NULL passed the ctor */ 526 BUG_ON(!ptdesc || !pagetable_pmd_ctor(NULL, ptdesc)); 527 return __pa((pmd_t *)ptdesc_address(ptdesc)); 528 } 529 530 static void __meminit create_pmd_mapping(pmd_t *pmdp, 531 uintptr_t va, phys_addr_t pa, 532 phys_addr_t sz, pgprot_t prot) 533 { 534 pte_t *ptep; 535 phys_addr_t pte_phys; 536 uintptr_t pmd_idx = pmd_index(va); 537 538 if (sz == PMD_SIZE) { 539 if (pmd_none(pmdp[pmd_idx])) 540 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 541 return; 542 } 543 544 if (pmd_none(pmdp[pmd_idx])) { 545 pte_phys = pt_ops.alloc_pte(va); 546 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 547 ptep = pt_ops.get_pte_virt(pte_phys); 548 memset(ptep, 0, PAGE_SIZE); 549 } else { 550 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 551 ptep = pt_ops.get_pte_virt(pte_phys); 552 } 553 554 create_pte_mapping(ptep, va, pa, sz, prot); 555 } 556 557 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 558 { 559 return (pud_t *)((uintptr_t)pa); 560 } 561 562 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 563 { 564 clear_fixmap(FIX_PUD); 565 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 566 } 567 568 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 569 { 570 return (pud_t *)__va(pa); 571 } 572 573 static phys_addr_t __init alloc_pud_early(uintptr_t va) 574 { 575 /* Only one PUD is available for early mapping */ 576 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 577 578 return (uintptr_t)early_pud; 579 } 580 581 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 582 { 583 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 584 } 585 586 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 587 { 588 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 589 590 BUG_ON(!ptdesc); 591 pagetable_pud_ctor(ptdesc); 592 return __pa((pud_t *)ptdesc_address(ptdesc)); 593 } 594 595 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 596 { 597 return (p4d_t *)((uintptr_t)pa); 598 } 599 600 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 601 { 602 clear_fixmap(FIX_P4D); 603 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 604 } 605 606 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 607 { 608 return (p4d_t *)__va(pa); 609 } 610 611 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 612 { 613 /* Only one P4D is available for early mapping */ 614 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 615 616 return (uintptr_t)early_p4d; 617 } 618 619 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 620 { 621 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 622 } 623 624 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 625 { 626 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 627 628 BUG_ON(!ptdesc); 629 pagetable_p4d_ctor(ptdesc); 630 return __pa((p4d_t *)ptdesc_address(ptdesc)); 631 } 632 633 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 634 pgprot_t prot) 635 { 636 pmd_t *nextp; 637 phys_addr_t next_phys; 638 uintptr_t pud_index = pud_index(va); 639 640 if (sz == PUD_SIZE) { 641 if (pud_val(pudp[pud_index]) == 0) 642 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 643 return; 644 } 645 646 if (pud_val(pudp[pud_index]) == 0) { 647 next_phys = pt_ops.alloc_pmd(va); 648 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 649 nextp = pt_ops.get_pmd_virt(next_phys); 650 memset(nextp, 0, PAGE_SIZE); 651 } else { 652 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 653 nextp = pt_ops.get_pmd_virt(next_phys); 654 } 655 656 create_pmd_mapping(nextp, va, pa, sz, prot); 657 } 658 659 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 660 pgprot_t prot) 661 { 662 pud_t *nextp; 663 phys_addr_t next_phys; 664 uintptr_t p4d_index = p4d_index(va); 665 666 if (sz == P4D_SIZE) { 667 if (p4d_val(p4dp[p4d_index]) == 0) 668 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 669 return; 670 } 671 672 if (p4d_val(p4dp[p4d_index]) == 0) { 673 next_phys = pt_ops.alloc_pud(va); 674 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 675 nextp = pt_ops.get_pud_virt(next_phys); 676 memset(nextp, 0, PAGE_SIZE); 677 } else { 678 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 679 nextp = pt_ops.get_pud_virt(next_phys); 680 } 681 682 create_pud_mapping(nextp, va, pa, sz, prot); 683 } 684 685 #define pgd_next_t p4d_t 686 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 687 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 688 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 689 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 690 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 691 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 692 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 693 (pgtable_l5_enabled ? \ 694 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 695 (pgtable_l4_enabled ? \ 696 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 697 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 698 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 699 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 700 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 701 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 702 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 703 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 704 #else 705 #define pgd_next_t pte_t 706 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 707 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 708 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 709 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 710 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 711 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 712 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 713 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 714 #endif /* __PAGETABLE_PMD_FOLDED */ 715 716 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 717 pgprot_t prot) 718 { 719 pgd_next_t *nextp; 720 phys_addr_t next_phys; 721 uintptr_t pgd_idx = pgd_index(va); 722 723 if (sz == PGDIR_SIZE) { 724 if (pgd_val(pgdp[pgd_idx]) == 0) 725 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 726 return; 727 } 728 729 if (pgd_val(pgdp[pgd_idx]) == 0) { 730 next_phys = alloc_pgd_next(va); 731 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 732 nextp = get_pgd_next_virt(next_phys); 733 memset(nextp, 0, PAGE_SIZE); 734 } else { 735 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 736 nextp = get_pgd_next_virt(next_phys); 737 } 738 739 create_pgd_next_mapping(nextp, va, pa, sz, prot); 740 } 741 742 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 743 { 744 if (debug_pagealloc_enabled()) 745 return PAGE_SIZE; 746 747 if (pgtable_l5_enabled && 748 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 749 return P4D_SIZE; 750 751 if (pgtable_l4_enabled && 752 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 753 return PUD_SIZE; 754 755 if (IS_ENABLED(CONFIG_64BIT) && 756 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 757 return PMD_SIZE; 758 759 return PAGE_SIZE; 760 } 761 762 #ifdef CONFIG_XIP_KERNEL 763 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 764 extern char _xiprom[], _exiprom[], __data_loc; 765 766 /* called from head.S with MMU off */ 767 asmlinkage void __init __copy_data(void) 768 { 769 void *from = (void *)(&__data_loc); 770 void *to = (void *)CONFIG_PHYS_RAM_BASE; 771 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 772 773 memcpy(to, from, sz); 774 } 775 #endif 776 777 #ifdef CONFIG_STRICT_KERNEL_RWX 778 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 779 { 780 if (is_va_kernel_text(va)) 781 return PAGE_KERNEL_READ_EXEC; 782 783 /* 784 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 785 * we must protect its linear mapping alias from being executed and 786 * written. 787 * And rodata section is marked readonly in mark_rodata_ro. 788 */ 789 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 790 return PAGE_KERNEL_READ; 791 792 return PAGE_KERNEL; 793 } 794 795 void mark_rodata_ro(void) 796 { 797 set_kernel_memory(__start_rodata, _data, set_memory_ro); 798 if (IS_ENABLED(CONFIG_64BIT)) 799 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 800 set_memory_ro); 801 } 802 #else 803 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 804 { 805 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 806 return PAGE_KERNEL; 807 808 return PAGE_KERNEL_EXEC; 809 } 810 #endif /* CONFIG_STRICT_KERNEL_RWX */ 811 812 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 813 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 814 u64 __pi_set_satp_mode_from_fdt(uintptr_t dtb_pa); 815 816 static void __init disable_pgtable_l5(void) 817 { 818 pgtable_l5_enabled = false; 819 kernel_map.page_offset = PAGE_OFFSET_L4; 820 satp_mode = SATP_MODE_48; 821 } 822 823 static void __init disable_pgtable_l4(void) 824 { 825 pgtable_l4_enabled = false; 826 kernel_map.page_offset = PAGE_OFFSET_L3; 827 satp_mode = SATP_MODE_39; 828 } 829 830 static int __init print_no4lvl(char *p) 831 { 832 pr_info("Disabled 4-level and 5-level paging"); 833 return 0; 834 } 835 early_param("no4lvl", print_no4lvl); 836 837 static int __init print_no5lvl(char *p) 838 { 839 pr_info("Disabled 5-level paging"); 840 return 0; 841 } 842 early_param("no5lvl", print_no5lvl); 843 844 static void __init set_mmap_rnd_bits_max(void) 845 { 846 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 847 } 848 849 /* 850 * There is a simple way to determine if 4-level is supported by the 851 * underlying hardware: establish 1:1 mapping in 4-level page table mode 852 * then read SATP to see if the configuration was taken into account 853 * meaning sv48 is supported. 854 * The maximum SATP mode is limited by both the command line and the "mmu-type" 855 * property in the device tree, since some platforms may hang if an unsupported 856 * SATP mode is attempted. 857 */ 858 static __init void set_satp_mode(uintptr_t dtb_pa) 859 { 860 u64 identity_satp, hw_satp; 861 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 862 u64 satp_mode_limit = min_not_zero(__pi_set_satp_mode_from_cmdline(dtb_pa), 863 __pi_set_satp_mode_from_fdt(dtb_pa)); 864 865 kernel_map.page_offset = PAGE_OFFSET_L5; 866 867 if (satp_mode_limit == SATP_MODE_48) { 868 disable_pgtable_l5(); 869 } else if (satp_mode_limit == SATP_MODE_39) { 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, NULL, &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 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1432 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1433 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1434 #endif 1435 arch_reserve_crashkernel(); 1436 memblock_dump_all(); 1437 } 1438 1439 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1440 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1441 unsigned long addr, unsigned long next) 1442 { 1443 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1444 } 1445 1446 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1447 unsigned long addr, unsigned long next) 1448 { 1449 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1450 return 1; 1451 } 1452 1453 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1454 struct vmem_altmap *altmap) 1455 { 1456 /* 1457 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1458 * can't use hugepage mappings for 2-level page table because in case of 1459 * memory hotplug, we are not able to update all the page tables with 1460 * the new PMDs. 1461 */ 1462 return vmemmap_populate_hugepages(start, end, node, altmap); 1463 } 1464 #endif 1465 1466 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1467 /* 1468 * Pre-allocates page-table pages for a specific area in the kernel 1469 * page-table. Only the level which needs to be synchronized between 1470 * all page-tables is allocated because the synchronization can be 1471 * expensive. 1472 */ 1473 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1474 const char *area) 1475 { 1476 unsigned long addr; 1477 const char *lvl; 1478 1479 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1480 pgd_t *pgd = pgd_offset_k(addr); 1481 p4d_t *p4d; 1482 pud_t *pud; 1483 pmd_t *pmd; 1484 1485 lvl = "p4d"; 1486 p4d = p4d_alloc(&init_mm, pgd, addr); 1487 if (!p4d) 1488 goto failed; 1489 1490 if (pgtable_l5_enabled) 1491 continue; 1492 1493 lvl = "pud"; 1494 pud = pud_alloc(&init_mm, p4d, addr); 1495 if (!pud) 1496 goto failed; 1497 1498 if (pgtable_l4_enabled) 1499 continue; 1500 1501 lvl = "pmd"; 1502 pmd = pmd_alloc(&init_mm, pud, addr); 1503 if (!pmd) 1504 goto failed; 1505 } 1506 return; 1507 1508 failed: 1509 /* 1510 * The pages have to be there now or they will be missing in 1511 * process page-tables later. 1512 */ 1513 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1514 } 1515 1516 #define PAGE_END KASAN_SHADOW_START 1517 1518 void __init pgtable_cache_init(void) 1519 { 1520 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1521 if (IS_ENABLED(CONFIG_MODULES)) 1522 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1523 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1524 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1525 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1526 if (IS_ENABLED(CONFIG_KASAN)) 1527 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1528 } 1529 } 1530 #endif 1531 1532 #ifdef CONFIG_EXECMEM 1533 #ifdef CONFIG_MMU 1534 static struct execmem_info execmem_info __ro_after_init; 1535 1536 struct execmem_info __init *execmem_arch_setup(void) 1537 { 1538 execmem_info = (struct execmem_info){ 1539 .ranges = { 1540 [EXECMEM_DEFAULT] = { 1541 .start = MODULES_VADDR, 1542 .end = MODULES_END, 1543 .pgprot = PAGE_KERNEL, 1544 .alignment = 1, 1545 }, 1546 [EXECMEM_KPROBES] = { 1547 .start = VMALLOC_START, 1548 .end = VMALLOC_END, 1549 .pgprot = PAGE_KERNEL_READ_EXEC, 1550 .alignment = 1, 1551 }, 1552 [EXECMEM_BPF] = { 1553 .start = BPF_JIT_REGION_START, 1554 .end = BPF_JIT_REGION_END, 1555 .pgprot = PAGE_KERNEL, 1556 .alignment = PAGE_SIZE, 1557 }, 1558 }, 1559 }; 1560 1561 return &execmem_info; 1562 } 1563 #endif /* CONFIG_MMU */ 1564 #endif /* CONFIG_EXECMEM */ 1565 1566 #ifdef CONFIG_MEMORY_HOTPLUG 1567 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1568 { 1569 struct page *page = pmd_page(*pmd); 1570 struct ptdesc *ptdesc = page_ptdesc(page); 1571 pte_t *pte; 1572 int i; 1573 1574 for (i = 0; i < PTRS_PER_PTE; i++) { 1575 pte = pte_start + i; 1576 if (!pte_none(*pte)) 1577 return; 1578 } 1579 1580 pagetable_dtor(ptdesc); 1581 if (PageReserved(page)) 1582 free_reserved_page(page); 1583 else 1584 pagetable_free(ptdesc); 1585 pmd_clear(pmd); 1586 } 1587 1588 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1589 { 1590 struct page *page = pud_page(*pud); 1591 struct ptdesc *ptdesc = page_ptdesc(page); 1592 pmd_t *pmd; 1593 int i; 1594 1595 for (i = 0; i < PTRS_PER_PMD; i++) { 1596 pmd = pmd_start + i; 1597 if (!pmd_none(*pmd)) 1598 return; 1599 } 1600 1601 if (!is_vmemmap) 1602 pagetable_dtor(ptdesc); 1603 if (PageReserved(page)) 1604 free_reserved_page(page); 1605 else 1606 pagetable_free(ptdesc); 1607 pud_clear(pud); 1608 } 1609 1610 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1611 { 1612 struct page *page = p4d_page(*p4d); 1613 pud_t *pud; 1614 int i; 1615 1616 for (i = 0; i < PTRS_PER_PUD; i++) { 1617 pud = pud_start + i; 1618 if (!pud_none(*pud)) 1619 return; 1620 } 1621 1622 if (PageReserved(page)) 1623 free_reserved_page(page); 1624 else 1625 __free_pages(page, 0); 1626 p4d_clear(p4d); 1627 } 1628 1629 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1630 struct vmem_altmap *altmap) 1631 { 1632 int order = get_order(size); 1633 1634 if (altmap) { 1635 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1636 return; 1637 } 1638 1639 if (PageReserved(page)) { 1640 unsigned int nr_pages = 1 << order; 1641 1642 while (nr_pages--) 1643 free_reserved_page(page++); 1644 return; 1645 } 1646 1647 __free_pages(page, order); 1648 } 1649 1650 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1651 bool is_vmemmap, struct vmem_altmap *altmap) 1652 { 1653 unsigned long next; 1654 pte_t *ptep, pte; 1655 1656 for (; addr < end; addr = next) { 1657 next = (addr + PAGE_SIZE) & PAGE_MASK; 1658 if (next > end) 1659 next = end; 1660 1661 ptep = pte_base + pte_index(addr); 1662 pte = ptep_get(ptep); 1663 if (!pte_present(*ptep)) 1664 continue; 1665 1666 pte_clear(&init_mm, addr, ptep); 1667 if (is_vmemmap) 1668 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1669 } 1670 } 1671 1672 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1673 bool is_vmemmap, struct vmem_altmap *altmap) 1674 { 1675 unsigned long next; 1676 pte_t *pte_base; 1677 pmd_t *pmdp, pmd; 1678 1679 for (; addr < end; addr = next) { 1680 next = pmd_addr_end(addr, end); 1681 pmdp = pmd_base + pmd_index(addr); 1682 pmd = pmdp_get(pmdp); 1683 if (!pmd_present(pmd)) 1684 continue; 1685 1686 if (pmd_leaf(pmd)) { 1687 pmd_clear(pmdp); 1688 if (is_vmemmap) 1689 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1690 continue; 1691 } 1692 1693 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1694 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1695 free_pte_table(pte_base, pmdp); 1696 } 1697 } 1698 1699 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1700 bool is_vmemmap, struct vmem_altmap *altmap) 1701 { 1702 unsigned long next; 1703 pud_t *pudp, pud; 1704 pmd_t *pmd_base; 1705 1706 for (; addr < end; addr = next) { 1707 next = pud_addr_end(addr, end); 1708 pudp = pud_base + pud_index(addr); 1709 pud = pudp_get(pudp); 1710 if (!pud_present(pud)) 1711 continue; 1712 1713 if (pud_leaf(pud)) { 1714 if (pgtable_l4_enabled) { 1715 pud_clear(pudp); 1716 if (is_vmemmap) 1717 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1718 } 1719 continue; 1720 } 1721 1722 pmd_base = pmd_offset(pudp, 0); 1723 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1724 1725 if (pgtable_l4_enabled) 1726 free_pmd_table(pmd_base, pudp, is_vmemmap); 1727 } 1728 } 1729 1730 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1731 bool is_vmemmap, struct vmem_altmap *altmap) 1732 { 1733 unsigned long next; 1734 p4d_t *p4dp, p4d; 1735 pud_t *pud_base; 1736 1737 for (; addr < end; addr = next) { 1738 next = p4d_addr_end(addr, end); 1739 p4dp = p4d_base + p4d_index(addr); 1740 p4d = p4dp_get(p4dp); 1741 if (!p4d_present(p4d)) 1742 continue; 1743 1744 if (p4d_leaf(p4d)) { 1745 if (pgtable_l5_enabled) { 1746 p4d_clear(p4dp); 1747 if (is_vmemmap) 1748 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1749 } 1750 continue; 1751 } 1752 1753 pud_base = pud_offset(p4dp, 0); 1754 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1755 1756 if (pgtable_l5_enabled) 1757 free_pud_table(pud_base, p4dp); 1758 } 1759 } 1760 1761 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1762 struct vmem_altmap *altmap) 1763 { 1764 unsigned long addr, next; 1765 p4d_t *p4d_base; 1766 pgd_t *pgd; 1767 1768 for (addr = va; addr < end; addr = next) { 1769 next = pgd_addr_end(addr, end); 1770 pgd = pgd_offset_k(addr); 1771 1772 if (!pgd_present(*pgd)) 1773 continue; 1774 1775 if (pgd_leaf(*pgd)) 1776 continue; 1777 1778 p4d_base = p4d_offset(pgd, 0); 1779 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1780 } 1781 1782 flush_tlb_all(); 1783 } 1784 1785 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1786 { 1787 unsigned long va = (unsigned long)__va(start); 1788 unsigned long end = (unsigned long)__va(start + size); 1789 1790 remove_pgd_mapping(va, end, false, NULL); 1791 } 1792 1793 struct range arch_get_mappable_range(void) 1794 { 1795 struct range mhp_range; 1796 1797 mhp_range.start = __pa(PAGE_OFFSET); 1798 mhp_range.end = __pa(PAGE_END - 1); 1799 return mhp_range; 1800 } 1801 1802 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1803 { 1804 int ret = 0; 1805 1806 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1807 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1808 if (ret) { 1809 remove_linear_mapping(start, size); 1810 goto out; 1811 } 1812 1813 max_pfn = PFN_UP(start + size); 1814 max_low_pfn = max_pfn; 1815 1816 out: 1817 flush_tlb_all(); 1818 return ret; 1819 } 1820 1821 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 1822 { 1823 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap); 1824 remove_linear_mapping(start, size); 1825 flush_tlb_all(); 1826 } 1827 1828 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1829 { 1830 remove_pgd_mapping(start, end, true, altmap); 1831 } 1832 #endif /* CONFIG_MEMORY_HOTPLUG */ 1833