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 static void __init zone_sizes_init(void) 83 { 84 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 85 86 #ifdef CONFIG_ZONE_DMA32 87 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 88 #endif 89 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 90 91 free_area_init(max_zone_pfns); 92 } 93 94 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 95 96 #define LOG2_SZ_1K ilog2(SZ_1K) 97 #define LOG2_SZ_1M ilog2(SZ_1M) 98 #define LOG2_SZ_1G ilog2(SZ_1G) 99 #define LOG2_SZ_1T ilog2(SZ_1T) 100 101 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 102 { 103 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 104 (((t) - (b)) >> LOG2_SZ_1K)); 105 } 106 107 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 108 { 109 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 110 (((t) - (b)) >> LOG2_SZ_1M)); 111 } 112 113 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 114 { 115 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 116 (((t) - (b)) >> LOG2_SZ_1G)); 117 } 118 119 #ifdef CONFIG_64BIT 120 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 121 { 122 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 123 (((t) - (b)) >> LOG2_SZ_1T)); 124 } 125 #else 126 #define print_mlt(n, b, t) do {} while (0) 127 #endif 128 129 static inline void print_ml(char *name, unsigned long b, unsigned long t) 130 { 131 unsigned long diff = t - b; 132 133 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 134 print_mlt(name, b, t); 135 else if ((diff >> LOG2_SZ_1G) >= 10) 136 print_mlg(name, b, t); 137 else if ((diff >> LOG2_SZ_1M) >= 10) 138 print_mlm(name, b, t); 139 else 140 print_mlk(name, b, t); 141 } 142 143 static void __init print_vm_layout(void) 144 { 145 pr_notice("Virtual kernel memory layout:\n"); 146 print_ml("fixmap", (unsigned long)FIXADDR_START, 147 (unsigned long)FIXADDR_TOP); 148 print_ml("pci io", (unsigned long)PCI_IO_START, 149 (unsigned long)PCI_IO_END); 150 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 151 (unsigned long)VMEMMAP_END); 152 print_ml("vmalloc", (unsigned long)VMALLOC_START, 153 (unsigned long)VMALLOC_END); 154 #ifdef CONFIG_64BIT 155 print_ml("modules", (unsigned long)MODULES_VADDR, 156 (unsigned long)MODULES_END); 157 #endif 158 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 159 (unsigned long)high_memory); 160 if (IS_ENABLED(CONFIG_64BIT)) { 161 #ifdef CONFIG_KASAN 162 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 163 #endif 164 165 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 166 (unsigned long)ADDRESS_SPACE_END); 167 } 168 } 169 #else 170 static void print_vm_layout(void) { } 171 #endif /* CONFIG_DEBUG_VM */ 172 173 void __init arch_mm_preinit(void) 174 { 175 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit); 176 #ifdef CONFIG_FLATMEM 177 BUG_ON(!mem_map); 178 #endif /* CONFIG_FLATMEM */ 179 180 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb && 181 dma_cache_alignment != 1) { 182 /* 183 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb 184 * buffer per 1GB of RAM for kmalloc() bouncing on 185 * non-coherent platforms. 186 */ 187 unsigned long size = 188 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 189 swiotlb_adjust_size(min(swiotlb_size_or_default(), size)); 190 swiotlb = true; 191 } 192 193 swiotlb_init(swiotlb, SWIOTLB_VERBOSE); 194 195 print_vm_layout(); 196 } 197 198 /* Limit the memory size via mem. */ 199 static phys_addr_t memory_limit; 200 #ifdef CONFIG_XIP_KERNEL 201 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit)) 202 #endif /* CONFIG_XIP_KERNEL */ 203 204 static int __init early_mem(char *p) 205 { 206 u64 size; 207 208 if (!p) 209 return 1; 210 211 size = memparse(p, &p) & PAGE_MASK; 212 memory_limit = min_t(u64, size, memory_limit); 213 214 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 215 216 return 0; 217 } 218 early_param("mem", early_mem); 219 220 static void __init setup_bootmem(void) 221 { 222 phys_addr_t vmlinux_end = __pa_symbol(&_end); 223 phys_addr_t max_mapped_addr; 224 phys_addr_t phys_ram_end, vmlinux_start; 225 226 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 227 vmlinux_start = __pa_symbol(&_sdata); 228 else 229 vmlinux_start = __pa_symbol(&_start); 230 231 memblock_enforce_memory_limit(memory_limit); 232 233 /* 234 * Make sure we align the reservation on PMD_SIZE since we will 235 * map the kernel in the linear mapping as read-only: we do not want 236 * any allocation to happen between _end and the next pmd aligned page. 237 */ 238 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 239 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 240 /* 241 * Reserve from the start of the kernel to the end of the kernel 242 */ 243 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 244 245 /* 246 * Make sure we align the start of the memory on a PMD boundary so that 247 * at worst, we map the linear mapping with PMD mappings. 248 */ 249 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) { 250 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 251 #ifdef CONFIG_SPARSEMEM_VMEMMAP 252 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 253 #endif 254 } 255 256 /* 257 * In 64-bit, any use of __va/__pa before this point is wrong as we 258 * did not know the start of DRAM before. 259 */ 260 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 261 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 262 263 /* 264 * The size of the linear page mapping may restrict the amount of 265 * usable RAM. 266 */ 267 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) { 268 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE; 269 if (memblock_end_of_DRAM() > max_mapped_addr) { 270 memblock_cap_memory_range(phys_ram_base, 271 max_mapped_addr - phys_ram_base); 272 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed", 273 &max_mapped_addr); 274 } 275 } 276 277 /* 278 * Reserve physical address space that would be mapped to virtual 279 * addresses greater than (void *)(-PAGE_SIZE) because: 280 * - This memory would overlap with ERR_PTR 281 * - This memory belongs to high memory, which is not supported 282 * 283 * This is not applicable to 64-bit kernel, because virtual addresses 284 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 285 * occupied by kernel mapping. Also it is unrealistic for high memory 286 * to exist on 64-bit platforms. 287 */ 288 if (!IS_ENABLED(CONFIG_64BIT)) { 289 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 290 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 291 } 292 293 phys_ram_end = memblock_end_of_DRAM(); 294 min_low_pfn = PFN_UP(phys_ram_base); 295 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 296 297 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 298 299 reserve_initrd_mem(); 300 301 /* 302 * No allocation should be done before reserving the memory as defined 303 * in the device tree, otherwise the allocation could end up in a 304 * reserved region. 305 */ 306 early_init_fdt_scan_reserved_mem(); 307 308 /* 309 * If DTB is built in, no need to reserve its memblock. 310 * Otherwise, do reserve it but avoid using 311 * early_init_fdt_reserve_self() since __pa() does 312 * not work for DTB pointers that are fixmap addresses 313 */ 314 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 315 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 316 317 dma_contiguous_reserve(dma32_phys_limit); 318 if (IS_ENABLED(CONFIG_64BIT)) 319 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 320 } 321 322 #ifdef CONFIG_RELOCATABLE 323 extern unsigned long __rela_dyn_start, __rela_dyn_end; 324 325 static void __init relocate_kernel(void) 326 { 327 Elf_Rela *rela = (Elf_Rela *)&__rela_dyn_start; 328 /* 329 * This holds the offset between the linked virtual address and the 330 * relocated virtual address. 331 */ 332 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 333 /* 334 * This holds the offset between kernel linked virtual address and 335 * physical address. 336 */ 337 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 338 339 for ( ; rela < (Elf_Rela *)&__rela_dyn_end; rela++) { 340 Elf_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 341 Elf_Addr relocated_addr = rela->r_addend; 342 343 if (rela->r_info != R_RISCV_RELATIVE) 344 continue; 345 346 /* 347 * Make sure to not relocate vdso symbols like rt_sigreturn 348 * which are linked from the address 0 in vmlinux since 349 * vdso symbol addresses are actually used as an offset from 350 * mm->context.vdso in VDSO_OFFSET macro. 351 */ 352 if (relocated_addr >= KERNEL_LINK_ADDR) 353 relocated_addr += reloc_offset; 354 355 *(Elf_Addr *)addr = relocated_addr; 356 } 357 } 358 #endif /* CONFIG_RELOCATABLE */ 359 360 #ifdef CONFIG_MMU 361 struct pt_alloc_ops pt_ops __meminitdata; 362 363 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 364 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 365 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 366 367 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 368 369 #ifdef CONFIG_XIP_KERNEL 370 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 371 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 372 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 373 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 374 #endif /* CONFIG_XIP_KERNEL */ 375 376 static const pgprot_t protection_map[16] = { 377 [VM_NONE] = PAGE_NONE, 378 [VM_READ] = PAGE_READ, 379 [VM_WRITE] = PAGE_COPY, 380 [VM_WRITE | VM_READ] = PAGE_COPY, 381 [VM_EXEC] = PAGE_EXEC, 382 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 383 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 384 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 385 [VM_SHARED] = PAGE_NONE, 386 [VM_SHARED | VM_READ] = PAGE_READ, 387 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 388 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 389 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 390 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 391 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 392 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 393 }; 394 DECLARE_VM_GET_PAGE_PROT 395 396 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 397 { 398 unsigned long addr = __fix_to_virt(idx); 399 pte_t *ptep; 400 401 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 402 403 ptep = &fixmap_pte[pte_index(addr)]; 404 405 if (pgprot_val(prot)) 406 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 407 else 408 pte_clear(&init_mm, addr, ptep); 409 local_flush_tlb_page(addr); 410 } 411 412 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 413 { 414 return (pte_t *)((uintptr_t)pa); 415 } 416 417 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 418 { 419 clear_fixmap(FIX_PTE); 420 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 421 } 422 423 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa) 424 { 425 return (pte_t *) __va(pa); 426 } 427 428 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 429 { 430 /* 431 * We only create PMD or PGD early mappings so we 432 * should never reach here with MMU disabled. 433 */ 434 BUG(); 435 } 436 437 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 438 { 439 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 440 } 441 442 static phys_addr_t __meminit alloc_pte_late(uintptr_t va) 443 { 444 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 445 446 /* 447 * We do not know which mm the PTE page is associated to at this point. 448 * Passing NULL to the ctor is the safe option, though it may result 449 * in unnecessary work (e.g. initialising the ptlock for init_mm). 450 */ 451 BUG_ON(!ptdesc || !pagetable_pte_ctor(NULL, ptdesc)); 452 return __pa((pte_t *)ptdesc_address(ptdesc)); 453 } 454 455 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 456 pgprot_t prot) 457 { 458 uintptr_t pte_idx = pte_index(va); 459 460 BUG_ON(sz != PAGE_SIZE); 461 462 if (pte_none(ptep[pte_idx])) 463 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 464 } 465 466 #ifndef __PAGETABLE_PMD_FOLDED 467 468 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 469 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 470 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 471 472 #ifdef CONFIG_XIP_KERNEL 473 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 474 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 475 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 476 #endif /* CONFIG_XIP_KERNEL */ 477 478 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 479 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 480 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 481 482 #ifdef CONFIG_XIP_KERNEL 483 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 484 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 485 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 486 #endif /* CONFIG_XIP_KERNEL */ 487 488 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 489 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 490 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 491 492 #ifdef CONFIG_XIP_KERNEL 493 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 494 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 495 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 496 #endif /* CONFIG_XIP_KERNEL */ 497 498 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 499 { 500 /* Before MMU is enabled */ 501 return (pmd_t *)((uintptr_t)pa); 502 } 503 504 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 505 { 506 clear_fixmap(FIX_PMD); 507 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 508 } 509 510 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa) 511 { 512 return (pmd_t *) __va(pa); 513 } 514 515 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 516 { 517 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 518 519 return (uintptr_t)early_pmd; 520 } 521 522 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 523 { 524 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 525 } 526 527 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va) 528 { 529 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0); 530 531 /* See comment in alloc_pte_late() regarding NULL passed the ctor */ 532 BUG_ON(!ptdesc || !pagetable_pmd_ctor(NULL, ptdesc)); 533 return __pa((pmd_t *)ptdesc_address(ptdesc)); 534 } 535 536 static void __meminit create_pmd_mapping(pmd_t *pmdp, 537 uintptr_t va, phys_addr_t pa, 538 phys_addr_t sz, pgprot_t prot) 539 { 540 pte_t *ptep; 541 phys_addr_t pte_phys; 542 uintptr_t pmd_idx = pmd_index(va); 543 544 if (sz == PMD_SIZE) { 545 if (pmd_none(pmdp[pmd_idx])) 546 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 547 return; 548 } 549 550 if (pmd_none(pmdp[pmd_idx])) { 551 pte_phys = pt_ops.alloc_pte(va); 552 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 553 ptep = pt_ops.get_pte_virt(pte_phys); 554 memset(ptep, 0, PAGE_SIZE); 555 } else { 556 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 557 ptep = pt_ops.get_pte_virt(pte_phys); 558 } 559 560 create_pte_mapping(ptep, va, pa, sz, prot); 561 } 562 563 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 564 { 565 return (pud_t *)((uintptr_t)pa); 566 } 567 568 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 569 { 570 clear_fixmap(FIX_PUD); 571 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 572 } 573 574 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 575 { 576 return (pud_t *)__va(pa); 577 } 578 579 static phys_addr_t __init alloc_pud_early(uintptr_t va) 580 { 581 /* Only one PUD is available for early mapping */ 582 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 583 584 return (uintptr_t)early_pud; 585 } 586 587 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 588 { 589 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 590 } 591 592 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 593 { 594 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 595 596 BUG_ON(!ptdesc); 597 pagetable_pud_ctor(ptdesc); 598 return __pa((pud_t *)ptdesc_address(ptdesc)); 599 } 600 601 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 602 { 603 return (p4d_t *)((uintptr_t)pa); 604 } 605 606 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 607 { 608 clear_fixmap(FIX_P4D); 609 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 610 } 611 612 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 613 { 614 return (p4d_t *)__va(pa); 615 } 616 617 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 618 { 619 /* Only one P4D is available for early mapping */ 620 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 621 622 return (uintptr_t)early_p4d; 623 } 624 625 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 626 { 627 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 628 } 629 630 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 631 { 632 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 633 634 BUG_ON(!ptdesc); 635 pagetable_p4d_ctor(ptdesc); 636 return __pa((p4d_t *)ptdesc_address(ptdesc)); 637 } 638 639 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 640 pgprot_t prot) 641 { 642 pmd_t *nextp; 643 phys_addr_t next_phys; 644 uintptr_t pud_index = pud_index(va); 645 646 if (sz == PUD_SIZE) { 647 if (pud_val(pudp[pud_index]) == 0) 648 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 649 return; 650 } 651 652 if (pud_val(pudp[pud_index]) == 0) { 653 next_phys = pt_ops.alloc_pmd(va); 654 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 655 nextp = pt_ops.get_pmd_virt(next_phys); 656 memset(nextp, 0, PAGE_SIZE); 657 } else { 658 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 659 nextp = pt_ops.get_pmd_virt(next_phys); 660 } 661 662 create_pmd_mapping(nextp, va, pa, sz, prot); 663 } 664 665 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 666 pgprot_t prot) 667 { 668 pud_t *nextp; 669 phys_addr_t next_phys; 670 uintptr_t p4d_index = p4d_index(va); 671 672 if (sz == P4D_SIZE) { 673 if (p4d_val(p4dp[p4d_index]) == 0) 674 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 675 return; 676 } 677 678 if (p4d_val(p4dp[p4d_index]) == 0) { 679 next_phys = pt_ops.alloc_pud(va); 680 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 681 nextp = pt_ops.get_pud_virt(next_phys); 682 memset(nextp, 0, PAGE_SIZE); 683 } else { 684 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 685 nextp = pt_ops.get_pud_virt(next_phys); 686 } 687 688 create_pud_mapping(nextp, va, pa, sz, prot); 689 } 690 691 #define pgd_next_t p4d_t 692 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 693 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 694 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 695 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 696 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 697 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 698 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 699 (pgtable_l5_enabled ? \ 700 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 701 (pgtable_l4_enabled ? \ 702 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 703 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 704 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 705 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 706 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 707 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 708 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 709 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 710 #else 711 #define pgd_next_t pte_t 712 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 713 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 714 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 715 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 716 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 717 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 718 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 719 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 720 #endif /* __PAGETABLE_PMD_FOLDED */ 721 722 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 723 pgprot_t prot) 724 { 725 pgd_next_t *nextp; 726 phys_addr_t next_phys; 727 uintptr_t pgd_idx = pgd_index(va); 728 729 if (sz == PGDIR_SIZE) { 730 if (pgd_val(pgdp[pgd_idx]) == 0) 731 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 732 return; 733 } 734 735 if (pgd_val(pgdp[pgd_idx]) == 0) { 736 next_phys = alloc_pgd_next(va); 737 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 738 nextp = get_pgd_next_virt(next_phys); 739 memset(nextp, 0, PAGE_SIZE); 740 } else { 741 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 742 nextp = get_pgd_next_virt(next_phys); 743 } 744 745 create_pgd_next_mapping(nextp, va, pa, sz, prot); 746 } 747 748 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 749 { 750 if (debug_pagealloc_enabled()) 751 return PAGE_SIZE; 752 753 if (pgtable_l5_enabled && 754 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 755 return P4D_SIZE; 756 757 if (pgtable_l4_enabled && 758 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 759 return PUD_SIZE; 760 761 if (IS_ENABLED(CONFIG_64BIT) && 762 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 763 return PMD_SIZE; 764 765 return PAGE_SIZE; 766 } 767 768 #ifdef CONFIG_XIP_KERNEL 769 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 770 extern char _xiprom[], _exiprom[], __data_loc; 771 772 /* called from head.S with MMU off */ 773 asmlinkage void __init __copy_data(void) 774 { 775 void *from = (void *)(&__data_loc); 776 void *to = (void *)CONFIG_PHYS_RAM_BASE; 777 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 778 779 memcpy(to, from, sz); 780 } 781 #endif 782 783 #ifdef CONFIG_STRICT_KERNEL_RWX 784 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 785 { 786 if (is_va_kernel_text(va)) 787 return PAGE_KERNEL_READ_EXEC; 788 789 /* 790 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 791 * we must protect its linear mapping alias from being executed and 792 * written. 793 * And rodata section is marked readonly in mark_rodata_ro. 794 */ 795 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 796 return PAGE_KERNEL_READ; 797 798 return PAGE_KERNEL; 799 } 800 801 void mark_rodata_ro(void) 802 { 803 set_kernel_memory(__start_rodata, _data, set_memory_ro); 804 if (IS_ENABLED(CONFIG_64BIT)) 805 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 806 set_memory_ro); 807 } 808 #else 809 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 810 { 811 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 812 return PAGE_KERNEL; 813 814 return PAGE_KERNEL_EXEC; 815 } 816 #endif /* CONFIG_STRICT_KERNEL_RWX */ 817 818 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 819 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 820 u64 __pi_set_satp_mode_from_fdt(uintptr_t dtb_pa); 821 822 static void __init disable_pgtable_l5(void) 823 { 824 pgtable_l5_enabled = false; 825 kernel_map.page_offset = PAGE_OFFSET_L4; 826 satp_mode = SATP_MODE_48; 827 } 828 829 static void __init disable_pgtable_l4(void) 830 { 831 pgtable_l4_enabled = false; 832 kernel_map.page_offset = PAGE_OFFSET_L3; 833 satp_mode = SATP_MODE_39; 834 } 835 836 static int __init print_no4lvl(char *p) 837 { 838 pr_info("Disabled 4-level and 5-level paging"); 839 return 0; 840 } 841 early_param("no4lvl", print_no4lvl); 842 843 static int __init print_no5lvl(char *p) 844 { 845 pr_info("Disabled 5-level paging"); 846 return 0; 847 } 848 early_param("no5lvl", print_no5lvl); 849 850 static void __init set_mmap_rnd_bits_max(void) 851 { 852 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 853 } 854 855 /* 856 * There is a simple way to determine if 4-level is supported by the 857 * underlying hardware: establish 1:1 mapping in 4-level page table mode 858 * then read SATP to see if the configuration was taken into account 859 * meaning sv48 is supported. 860 * The maximum SATP mode is limited by both the command line and the "mmu-type" 861 * property in the device tree, since some platforms may hang if an unsupported 862 * SATP mode is attempted. 863 */ 864 static __init void set_satp_mode(uintptr_t dtb_pa) 865 { 866 u64 identity_satp, hw_satp; 867 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 868 u64 satp_mode_limit = min_not_zero(__pi_set_satp_mode_from_cmdline(dtb_pa), 869 __pi_set_satp_mode_from_fdt(dtb_pa)); 870 871 kernel_map.page_offset = PAGE_OFFSET_L5; 872 873 if (satp_mode_limit == SATP_MODE_48) { 874 disable_pgtable_l5(); 875 } else if (satp_mode_limit == SATP_MODE_39) { 876 disable_pgtable_l5(); 877 disable_pgtable_l4(); 878 return; 879 } 880 881 create_p4d_mapping(early_p4d, 882 set_satp_mode_pmd, (uintptr_t)early_pud, 883 P4D_SIZE, PAGE_TABLE); 884 create_pud_mapping(early_pud, 885 set_satp_mode_pmd, (uintptr_t)early_pmd, 886 PUD_SIZE, PAGE_TABLE); 887 /* Handle the case where set_satp_mode straddles 2 PMDs */ 888 create_pmd_mapping(early_pmd, 889 set_satp_mode_pmd, set_satp_mode_pmd, 890 PMD_SIZE, PAGE_KERNEL_EXEC); 891 create_pmd_mapping(early_pmd, 892 set_satp_mode_pmd + PMD_SIZE, 893 set_satp_mode_pmd + PMD_SIZE, 894 PMD_SIZE, PAGE_KERNEL_EXEC); 895 retry: 896 create_pgd_mapping(early_pg_dir, 897 set_satp_mode_pmd, 898 pgtable_l5_enabled ? 899 (uintptr_t)early_p4d : (uintptr_t)early_pud, 900 PGDIR_SIZE, PAGE_TABLE); 901 902 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 903 904 local_flush_tlb_all(); 905 csr_write(CSR_SATP, identity_satp); 906 hw_satp = csr_swap(CSR_SATP, 0ULL); 907 local_flush_tlb_all(); 908 909 if (hw_satp != identity_satp) { 910 if (pgtable_l5_enabled) { 911 disable_pgtable_l5(); 912 memset(early_pg_dir, 0, PAGE_SIZE); 913 goto retry; 914 } 915 disable_pgtable_l4(); 916 } 917 918 memset(early_pg_dir, 0, PAGE_SIZE); 919 memset(early_p4d, 0, PAGE_SIZE); 920 memset(early_pud, 0, PAGE_SIZE); 921 memset(early_pmd, 0, PAGE_SIZE); 922 } 923 #endif 924 925 /* 926 * setup_vm() is called from head.S with MMU-off. 927 * 928 * Following requirements should be honoured for setup_vm() to work 929 * correctly: 930 * 1) It should use PC-relative addressing for accessing kernel symbols. 931 * To achieve this we always use GCC cmodel=medany. 932 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 933 * so disable compiler instrumentation when FTRACE is enabled. 934 * 935 * Currently, the above requirements are honoured by using custom CFLAGS 936 * for init.o in mm/Makefile. 937 */ 938 939 #ifndef __riscv_cmodel_medany 940 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 941 #endif 942 943 #ifdef CONFIG_XIP_KERNEL 944 static void __init create_kernel_page_table(pgd_t *pgdir, 945 __always_unused bool early) 946 { 947 uintptr_t va, start_va, end_va; 948 949 /* Map the flash resident part */ 950 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 951 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 952 create_pgd_mapping(pgdir, va, 953 kernel_map.xiprom + (va - kernel_map.virt_addr), 954 PMD_SIZE, PAGE_KERNEL_EXEC); 955 956 /* Map the data in RAM */ 957 start_va = kernel_map.virt_addr + (uintptr_t)&_sdata - (uintptr_t)&_start; 958 end_va = kernel_map.virt_addr + kernel_map.size; 959 for (va = start_va; va < end_va; va += PMD_SIZE) 960 create_pgd_mapping(pgdir, va, 961 kernel_map.phys_addr + (va - start_va), 962 PMD_SIZE, PAGE_KERNEL); 963 } 964 #else 965 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 966 { 967 uintptr_t va, end_va; 968 969 end_va = kernel_map.virt_addr + kernel_map.size; 970 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 971 create_pgd_mapping(pgdir, va, 972 kernel_map.phys_addr + (va - kernel_map.virt_addr), 973 PMD_SIZE, 974 early ? 975 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 976 } 977 #endif 978 979 /* 980 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 981 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 982 * entry. 983 */ 984 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 985 uintptr_t dtb_pa) 986 { 987 #ifndef CONFIG_BUILTIN_DTB 988 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 989 990 /* Make sure the fdt fixmap address is always aligned on PMD size */ 991 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 992 993 /* In 32-bit only, the fdt lies in its own PGD */ 994 if (!IS_ENABLED(CONFIG_64BIT)) { 995 create_pgd_mapping(early_pg_dir, fix_fdt_va, 996 pa, MAX_FDT_SIZE, PAGE_KERNEL); 997 } else { 998 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 999 pa, PMD_SIZE, PAGE_KERNEL); 1000 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 1001 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 1002 } 1003 1004 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 1005 #else 1006 /* 1007 * For 64-bit kernel, __va can't be used since it would return a linear 1008 * mapping address whereas dtb_early_va will be used before 1009 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 1010 * kernel is mapped in the linear mapping, that makes no difference. 1011 */ 1012 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 1013 #endif 1014 1015 dtb_early_pa = dtb_pa; 1016 } 1017 1018 /* 1019 * MMU is not enabled, the page tables are allocated directly using 1020 * early_pmd/pud/p4d and the address returned is the physical one. 1021 */ 1022 static void __init pt_ops_set_early(void) 1023 { 1024 pt_ops.alloc_pte = alloc_pte_early; 1025 pt_ops.get_pte_virt = get_pte_virt_early; 1026 #ifndef __PAGETABLE_PMD_FOLDED 1027 pt_ops.alloc_pmd = alloc_pmd_early; 1028 pt_ops.get_pmd_virt = get_pmd_virt_early; 1029 pt_ops.alloc_pud = alloc_pud_early; 1030 pt_ops.get_pud_virt = get_pud_virt_early; 1031 pt_ops.alloc_p4d = alloc_p4d_early; 1032 pt_ops.get_p4d_virt = get_p4d_virt_early; 1033 #endif 1034 } 1035 1036 /* 1037 * MMU is enabled but page table setup is not complete yet. 1038 * fixmap page table alloc functions must be used as a means to temporarily 1039 * map the allocated physical pages since the linear mapping does not exist yet. 1040 * 1041 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 1042 * but it will be used as described above. 1043 */ 1044 static void __init pt_ops_set_fixmap(void) 1045 { 1046 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 1047 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 1048 #ifndef __PAGETABLE_PMD_FOLDED 1049 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1050 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1051 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1052 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1053 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1054 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1055 #endif 1056 } 1057 1058 /* 1059 * MMU is enabled and page table setup is complete, so from now, we can use 1060 * generic page allocation functions to setup page table. 1061 */ 1062 static void __init pt_ops_set_late(void) 1063 { 1064 pt_ops.alloc_pte = alloc_pte_late; 1065 pt_ops.get_pte_virt = get_pte_virt_late; 1066 #ifndef __PAGETABLE_PMD_FOLDED 1067 pt_ops.alloc_pmd = alloc_pmd_late; 1068 pt_ops.get_pmd_virt = get_pmd_virt_late; 1069 pt_ops.alloc_pud = alloc_pud_late; 1070 pt_ops.get_pud_virt = get_pud_virt_late; 1071 pt_ops.alloc_p4d = alloc_p4d_late; 1072 pt_ops.get_p4d_virt = get_p4d_virt_late; 1073 #endif 1074 } 1075 1076 #ifdef CONFIG_RANDOMIZE_BASE 1077 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1078 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1079 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa); 1080 1081 static int __init print_nokaslr(char *p) 1082 { 1083 pr_info("Disabled KASLR"); 1084 return 0; 1085 } 1086 early_param("nokaslr", print_nokaslr); 1087 1088 unsigned long kaslr_offset(void) 1089 { 1090 return kernel_map.virt_offset; 1091 } 1092 #endif 1093 1094 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1095 { 1096 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1097 1098 #ifdef CONFIG_RANDOMIZE_BASE 1099 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1100 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa); 1101 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1102 u32 nr_pos; 1103 1104 if (kaslr_seed == 0) 1105 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1106 /* 1107 * Compute the number of positions available: we are limited 1108 * by the early page table that only has one PUD and we must 1109 * be aligned on PMD_SIZE. 1110 */ 1111 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1112 1113 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1114 } 1115 #endif 1116 1117 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1118 1119 #ifdef CONFIG_XIP_KERNEL 1120 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 1121 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 1122 1123 phys_ram_base = CONFIG_PHYS_RAM_BASE; 1124 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1125 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 1126 #endif 1127 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 1128 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1129 1130 kernel_map.va_kernel_xip_text_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 1131 kernel_map.va_kernel_xip_data_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr 1132 + (uintptr_t)&_sdata - (uintptr_t)&_start; 1133 #else 1134 kernel_map.phys_addr = (uintptr_t)(&_start); 1135 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1136 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1137 #endif 1138 1139 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 1140 set_satp_mode(dtb_pa); 1141 set_mmap_rnd_bits_max(); 1142 #endif 1143 1144 /* 1145 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1146 * where we have the system memory layout: this allows us to align 1147 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1148 * for the linear mapping. This is only possible because the kernel 1149 * mapping lies outside the linear mapping. 1150 * In 32-bit however, as the kernel resides in the linear mapping, 1151 * setup_vm_final can not change the mapping established here, 1152 * otherwise the same kernel addresses would get mapped to different 1153 * physical addresses (if the start of dram is different from the 1154 * kernel physical address start). 1155 */ 1156 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1157 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1158 1159 memory_limit = KERN_VIRT_SIZE; 1160 1161 /* Sanity check alignment and size */ 1162 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1163 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1164 1165 #ifdef CONFIG_64BIT 1166 /* 1167 * The last 4K bytes of the addressable memory can not be mapped because 1168 * of IS_ERR_VALUE macro. 1169 */ 1170 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1171 #endif 1172 1173 #ifdef CONFIG_RELOCATABLE 1174 /* 1175 * Early page table uses only one PUD, which makes it possible 1176 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1177 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1178 * since a part of the kernel would not get mapped. 1179 */ 1180 if (IS_ENABLED(CONFIG_64BIT)) 1181 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1182 relocate_kernel(); 1183 #endif 1184 1185 apply_early_boot_alternatives(); 1186 pt_ops_set_early(); 1187 1188 /* Setup early PGD for fixmap */ 1189 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1190 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1191 1192 #ifndef __PAGETABLE_PMD_FOLDED 1193 /* Setup fixmap P4D and PUD */ 1194 if (pgtable_l5_enabled) 1195 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1196 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1197 /* Setup fixmap PUD and PMD */ 1198 if (pgtable_l4_enabled) 1199 create_pud_mapping(fixmap_pud, FIXADDR_START, 1200 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1201 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1202 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1203 /* Setup trampoline PGD and PMD */ 1204 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1205 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1206 if (pgtable_l5_enabled) 1207 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1208 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1209 if (pgtable_l4_enabled) 1210 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1211 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1212 #ifdef CONFIG_XIP_KERNEL 1213 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1214 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1215 #else 1216 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1217 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1218 #endif 1219 #else 1220 /* Setup trampoline PGD */ 1221 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1222 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1223 #endif 1224 1225 /* 1226 * Setup early PGD covering entire kernel which will allow 1227 * us to reach paging_init(). We map all memory banks later 1228 * in setup_vm_final() below. 1229 */ 1230 create_kernel_page_table(early_pg_dir, true); 1231 1232 /* Setup early mapping for FDT early scan */ 1233 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1234 1235 /* 1236 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1237 * range can not span multiple pmds. 1238 */ 1239 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1240 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1241 1242 #ifndef __PAGETABLE_PMD_FOLDED 1243 /* 1244 * Early ioremap fixmap is already created as it lies within first 2MB 1245 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1246 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1247 * the user if not. 1248 */ 1249 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1250 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1251 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1252 WARN_ON(1); 1253 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1254 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1255 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1256 fix_to_virt(FIX_BTMAP_BEGIN)); 1257 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1258 fix_to_virt(FIX_BTMAP_END)); 1259 1260 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1261 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1262 } 1263 #endif 1264 1265 pt_ops_set_fixmap(); 1266 } 1267 1268 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end, 1269 uintptr_t fixed_map_size, const pgprot_t *pgprot) 1270 { 1271 phys_addr_t pa; 1272 uintptr_t va, map_size; 1273 1274 for (pa = start; pa < end; pa += map_size) { 1275 va = (uintptr_t)__va(pa); 1276 map_size = fixed_map_size ? fixed_map_size : 1277 best_map_size(pa, va, end - pa); 1278 1279 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1280 pgprot ? *pgprot : pgprot_from_va(va)); 1281 } 1282 } 1283 1284 static void __init create_linear_mapping_page_table(void) 1285 { 1286 phys_addr_t start, end; 1287 phys_addr_t kfence_pool __maybe_unused; 1288 u64 i; 1289 1290 #ifdef CONFIG_STRICT_KERNEL_RWX 1291 phys_addr_t ktext_start = __pa_symbol(_start); 1292 phys_addr_t ktext_size = __init_data_begin - _start; 1293 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1294 phys_addr_t krodata_size = _data - __start_rodata; 1295 1296 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1297 memblock_mark_nomap(ktext_start, ktext_size); 1298 memblock_mark_nomap(krodata_start, krodata_size); 1299 #endif 1300 1301 #ifdef CONFIG_KFENCE 1302 /* 1303 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1304 * before we setup the linear mapping so that we avoid using hugepages 1305 * for this region. 1306 */ 1307 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1308 BUG_ON(!kfence_pool); 1309 1310 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1311 __kfence_pool = __va(kfence_pool); 1312 #endif 1313 1314 /* Map all memory banks in the linear mapping */ 1315 for_each_mem_range(i, &start, &end) { 1316 if (start >= end) 1317 break; 1318 if (start <= __pa(PAGE_OFFSET) && 1319 __pa(PAGE_OFFSET) < end) 1320 start = __pa(PAGE_OFFSET); 1321 1322 create_linear_mapping_range(start, end, 0, NULL); 1323 } 1324 1325 #ifdef CONFIG_STRICT_KERNEL_RWX 1326 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL); 1327 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL); 1328 1329 memblock_clear_nomap(ktext_start, ktext_size); 1330 memblock_clear_nomap(krodata_start, krodata_size); 1331 #endif 1332 1333 #ifdef CONFIG_KFENCE 1334 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL); 1335 1336 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1337 #endif 1338 } 1339 1340 static void __init setup_vm_final(void) 1341 { 1342 /* Setup swapper PGD for fixmap */ 1343 #if !defined(CONFIG_64BIT) 1344 /* 1345 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1346 * directly in swapper_pg_dir in addition to the pgd entry that points 1347 * to fixmap_pte. 1348 */ 1349 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1350 1351 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1352 #endif 1353 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1354 __pa_symbol(fixmap_pgd_next), 1355 PGDIR_SIZE, PAGE_TABLE); 1356 1357 /* Map the linear mapping */ 1358 create_linear_mapping_page_table(); 1359 1360 /* Map the kernel */ 1361 if (IS_ENABLED(CONFIG_64BIT)) 1362 create_kernel_page_table(swapper_pg_dir, false); 1363 1364 #ifdef CONFIG_KASAN 1365 kasan_swapper_init(); 1366 #endif 1367 1368 /* Clear fixmap PTE and PMD mappings */ 1369 clear_fixmap(FIX_PTE); 1370 clear_fixmap(FIX_PMD); 1371 clear_fixmap(FIX_PUD); 1372 clear_fixmap(FIX_P4D); 1373 1374 /* Move to swapper page table */ 1375 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1376 local_flush_tlb_all(); 1377 1378 pt_ops_set_late(); 1379 } 1380 #else 1381 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1382 { 1383 dtb_early_va = (void *)dtb_pa; 1384 dtb_early_pa = dtb_pa; 1385 1386 #ifdef CONFIG_RELOCATABLE 1387 kernel_map.virt_addr = (uintptr_t)_start; 1388 kernel_map.phys_addr = (uintptr_t)_start; 1389 relocate_kernel(); 1390 #endif 1391 } 1392 1393 static inline void setup_vm_final(void) 1394 { 1395 } 1396 #endif /* CONFIG_MMU */ 1397 1398 /* 1399 * reserve_crashkernel() - reserves memory for crash kernel 1400 * 1401 * This function reserves memory area given in "crashkernel=" kernel command 1402 * line parameter. The memory reserved is used by dump capture kernel when 1403 * primary kernel is crashing. 1404 */ 1405 static void __init arch_reserve_crashkernel(void) 1406 { 1407 unsigned long long low_size = 0; 1408 unsigned long long crash_base, crash_size; 1409 bool high = false; 1410 int ret; 1411 1412 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1413 return; 1414 1415 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1416 &crash_size, &crash_base, 1417 &low_size, NULL, &high); 1418 if (ret) 1419 return; 1420 1421 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 1422 } 1423 1424 void __init paging_init(void) 1425 { 1426 setup_bootmem(); 1427 setup_vm_final(); 1428 1429 /* Depend on that Linear Mapping is ready */ 1430 memblock_allow_resize(); 1431 } 1432 1433 void __init misc_mem_init(void) 1434 { 1435 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1436 arch_numa_init(); 1437 sparse_init(); 1438 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1439 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1440 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1441 #endif 1442 zone_sizes_init(); 1443 arch_reserve_crashkernel(); 1444 memblock_dump_all(); 1445 } 1446 1447 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1448 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node, 1449 unsigned long addr, unsigned long next) 1450 { 1451 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL); 1452 } 1453 1454 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, 1455 unsigned long addr, unsigned long next) 1456 { 1457 vmemmap_verify((pte_t *)pmdp, node, addr, next); 1458 return 1; 1459 } 1460 1461 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1462 struct vmem_altmap *altmap) 1463 { 1464 /* 1465 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1466 * can't use hugepage mappings for 2-level page table because in case of 1467 * memory hotplug, we are not able to update all the page tables with 1468 * the new PMDs. 1469 */ 1470 return vmemmap_populate_hugepages(start, end, node, altmap); 1471 } 1472 #endif 1473 1474 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1475 /* 1476 * Pre-allocates page-table pages for a specific area in the kernel 1477 * page-table. Only the level which needs to be synchronized between 1478 * all page-tables is allocated because the synchronization can be 1479 * expensive. 1480 */ 1481 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1482 const char *area) 1483 { 1484 unsigned long addr; 1485 const char *lvl; 1486 1487 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1488 pgd_t *pgd = pgd_offset_k(addr); 1489 p4d_t *p4d; 1490 pud_t *pud; 1491 pmd_t *pmd; 1492 1493 lvl = "p4d"; 1494 p4d = p4d_alloc(&init_mm, pgd, addr); 1495 if (!p4d) 1496 goto failed; 1497 1498 if (pgtable_l5_enabled) 1499 continue; 1500 1501 lvl = "pud"; 1502 pud = pud_alloc(&init_mm, p4d, addr); 1503 if (!pud) 1504 goto failed; 1505 1506 if (pgtable_l4_enabled) 1507 continue; 1508 1509 lvl = "pmd"; 1510 pmd = pmd_alloc(&init_mm, pud, addr); 1511 if (!pmd) 1512 goto failed; 1513 } 1514 return; 1515 1516 failed: 1517 /* 1518 * The pages have to be there now or they will be missing in 1519 * process page-tables later. 1520 */ 1521 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1522 } 1523 1524 #define PAGE_END KASAN_SHADOW_START 1525 1526 void __init pgtable_cache_init(void) 1527 { 1528 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1529 if (IS_ENABLED(CONFIG_MODULES)) 1530 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1531 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1532 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1533 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1534 if (IS_ENABLED(CONFIG_KASAN)) 1535 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1536 } 1537 } 1538 #endif 1539 1540 #ifdef CONFIG_EXECMEM 1541 #ifdef CONFIG_MMU 1542 static struct execmem_info execmem_info __ro_after_init; 1543 1544 struct execmem_info __init *execmem_arch_setup(void) 1545 { 1546 execmem_info = (struct execmem_info){ 1547 .ranges = { 1548 [EXECMEM_DEFAULT] = { 1549 .start = MODULES_VADDR, 1550 .end = MODULES_END, 1551 .pgprot = PAGE_KERNEL, 1552 .alignment = 1, 1553 }, 1554 [EXECMEM_KPROBES] = { 1555 .start = VMALLOC_START, 1556 .end = VMALLOC_END, 1557 .pgprot = PAGE_KERNEL_READ_EXEC, 1558 .alignment = 1, 1559 }, 1560 [EXECMEM_BPF] = { 1561 .start = BPF_JIT_REGION_START, 1562 .end = BPF_JIT_REGION_END, 1563 .pgprot = PAGE_KERNEL, 1564 .alignment = PAGE_SIZE, 1565 }, 1566 }, 1567 }; 1568 1569 return &execmem_info; 1570 } 1571 #endif /* CONFIG_MMU */ 1572 #endif /* CONFIG_EXECMEM */ 1573 1574 #ifdef CONFIG_MEMORY_HOTPLUG 1575 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1576 { 1577 struct page *page = pmd_page(*pmd); 1578 struct ptdesc *ptdesc = page_ptdesc(page); 1579 pte_t *pte; 1580 int i; 1581 1582 for (i = 0; i < PTRS_PER_PTE; i++) { 1583 pte = pte_start + i; 1584 if (!pte_none(*pte)) 1585 return; 1586 } 1587 1588 pagetable_dtor(ptdesc); 1589 if (PageReserved(page)) 1590 free_reserved_page(page); 1591 else 1592 pagetable_free(ptdesc); 1593 pmd_clear(pmd); 1594 } 1595 1596 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1597 { 1598 struct page *page = pud_page(*pud); 1599 struct ptdesc *ptdesc = page_ptdesc(page); 1600 pmd_t *pmd; 1601 int i; 1602 1603 for (i = 0; i < PTRS_PER_PMD; i++) { 1604 pmd = pmd_start + i; 1605 if (!pmd_none(*pmd)) 1606 return; 1607 } 1608 1609 if (!is_vmemmap) 1610 pagetable_dtor(ptdesc); 1611 if (PageReserved(page)) 1612 free_reserved_page(page); 1613 else 1614 pagetable_free(ptdesc); 1615 pud_clear(pud); 1616 } 1617 1618 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1619 { 1620 struct page *page = p4d_page(*p4d); 1621 pud_t *pud; 1622 int i; 1623 1624 for (i = 0; i < PTRS_PER_PUD; i++) { 1625 pud = pud_start + i; 1626 if (!pud_none(*pud)) 1627 return; 1628 } 1629 1630 if (PageReserved(page)) 1631 free_reserved_page(page); 1632 else 1633 free_pages((unsigned long)page_address(page), 0); 1634 p4d_clear(p4d); 1635 } 1636 1637 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1638 struct vmem_altmap *altmap) 1639 { 1640 int order = get_order(size); 1641 1642 if (altmap) { 1643 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1644 return; 1645 } 1646 1647 if (PageReserved(page)) { 1648 unsigned int nr_pages = 1 << order; 1649 1650 while (nr_pages--) 1651 free_reserved_page(page++); 1652 return; 1653 } 1654 1655 free_pages((unsigned long)page_address(page), order); 1656 } 1657 1658 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1659 bool is_vmemmap, struct vmem_altmap *altmap) 1660 { 1661 unsigned long next; 1662 pte_t *ptep, pte; 1663 1664 for (; addr < end; addr = next) { 1665 next = (addr + PAGE_SIZE) & PAGE_MASK; 1666 if (next > end) 1667 next = end; 1668 1669 ptep = pte_base + pte_index(addr); 1670 pte = ptep_get(ptep); 1671 if (!pte_present(*ptep)) 1672 continue; 1673 1674 pte_clear(&init_mm, addr, ptep); 1675 if (is_vmemmap) 1676 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1677 } 1678 } 1679 1680 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1681 bool is_vmemmap, struct vmem_altmap *altmap) 1682 { 1683 unsigned long next; 1684 pte_t *pte_base; 1685 pmd_t *pmdp, pmd; 1686 1687 for (; addr < end; addr = next) { 1688 next = pmd_addr_end(addr, end); 1689 pmdp = pmd_base + pmd_index(addr); 1690 pmd = pmdp_get(pmdp); 1691 if (!pmd_present(pmd)) 1692 continue; 1693 1694 if (pmd_leaf(pmd)) { 1695 pmd_clear(pmdp); 1696 if (is_vmemmap) 1697 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1698 continue; 1699 } 1700 1701 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1702 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1703 free_pte_table(pte_base, pmdp); 1704 } 1705 } 1706 1707 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1708 bool is_vmemmap, struct vmem_altmap *altmap) 1709 { 1710 unsigned long next; 1711 pud_t *pudp, pud; 1712 pmd_t *pmd_base; 1713 1714 for (; addr < end; addr = next) { 1715 next = pud_addr_end(addr, end); 1716 pudp = pud_base + pud_index(addr); 1717 pud = pudp_get(pudp); 1718 if (!pud_present(pud)) 1719 continue; 1720 1721 if (pud_leaf(pud)) { 1722 if (pgtable_l4_enabled) { 1723 pud_clear(pudp); 1724 if (is_vmemmap) 1725 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1726 } 1727 continue; 1728 } 1729 1730 pmd_base = pmd_offset(pudp, 0); 1731 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1732 1733 if (pgtable_l4_enabled) 1734 free_pmd_table(pmd_base, pudp, is_vmemmap); 1735 } 1736 } 1737 1738 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1739 bool is_vmemmap, struct vmem_altmap *altmap) 1740 { 1741 unsigned long next; 1742 p4d_t *p4dp, p4d; 1743 pud_t *pud_base; 1744 1745 for (; addr < end; addr = next) { 1746 next = p4d_addr_end(addr, end); 1747 p4dp = p4d_base + p4d_index(addr); 1748 p4d = p4dp_get(p4dp); 1749 if (!p4d_present(p4d)) 1750 continue; 1751 1752 if (p4d_leaf(p4d)) { 1753 if (pgtable_l5_enabled) { 1754 p4d_clear(p4dp); 1755 if (is_vmemmap) 1756 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1757 } 1758 continue; 1759 } 1760 1761 pud_base = pud_offset(p4dp, 0); 1762 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1763 1764 if (pgtable_l5_enabled) 1765 free_pud_table(pud_base, p4dp); 1766 } 1767 } 1768 1769 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1770 struct vmem_altmap *altmap) 1771 { 1772 unsigned long addr, next; 1773 p4d_t *p4d_base; 1774 pgd_t *pgd; 1775 1776 for (addr = va; addr < end; addr = next) { 1777 next = pgd_addr_end(addr, end); 1778 pgd = pgd_offset_k(addr); 1779 1780 if (!pgd_present(*pgd)) 1781 continue; 1782 1783 if (pgd_leaf(*pgd)) 1784 continue; 1785 1786 p4d_base = p4d_offset(pgd, 0); 1787 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1788 } 1789 1790 flush_tlb_all(); 1791 } 1792 1793 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1794 { 1795 unsigned long va = (unsigned long)__va(start); 1796 unsigned long end = (unsigned long)__va(start + size); 1797 1798 remove_pgd_mapping(va, end, false, NULL); 1799 } 1800 1801 struct range arch_get_mappable_range(void) 1802 { 1803 struct range mhp_range; 1804 1805 mhp_range.start = __pa(PAGE_OFFSET); 1806 mhp_range.end = __pa(PAGE_END - 1); 1807 return mhp_range; 1808 } 1809 1810 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1811 { 1812 int ret = 0; 1813 1814 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1815 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1816 if (ret) { 1817 remove_linear_mapping(start, size); 1818 goto out; 1819 } 1820 1821 max_pfn = PFN_UP(start + size); 1822 max_low_pfn = max_pfn; 1823 1824 out: 1825 flush_tlb_all(); 1826 return ret; 1827 } 1828 1829 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 1830 { 1831 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap); 1832 remove_linear_mapping(start, size); 1833 flush_tlb_all(); 1834 } 1835 1836 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1837 { 1838 remove_pgd_mapping(start, end, true, altmap); 1839 } 1840 #endif /* CONFIG_MEMORY_HOTPLUG */ 1841