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 #include <linux/vmemmap-optimization.h> 26 27 #include <asm/alternative.h> 28 #include <asm/fixmap.h> 29 #include <asm/io.h> 30 #include <asm/kasan.h> 31 #include <asm/module.h> 32 #include <asm/numa.h> 33 #include <asm/pgtable.h> 34 #include <asm/sections.h> 35 #include <asm/soc.h> 36 #include <asm/sparsemem.h> 37 #include <asm/tlbflush.h> 38 39 #include "../kernel/head.h" 40 41 #if defined(CONFIG_64BIT) && defined(CONFIG_MMU) 42 DECLARE_BITMAP(new_valid_map_cpus, NR_CPUS); 43 #endif 44 45 struct kernel_mapping kernel_map __ro_after_init; 46 EXPORT_SYMBOL(kernel_map); 47 48 #ifdef CONFIG_64BIT 49 u64 satp_mode __ro_after_init = SATP_MODE_57; 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 = true; 57 bool pgtable_l5_enabled __ro_after_init = true; 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 max(1ULL << SECTION_SIZE_BITS, \ 67 PFN_PHYS(MAX_FOLIO_VMEMMAP_ALIGN / \ 68 sizeof(struct page))) 69 70 unsigned long vmemmap_start_pfn __ro_after_init; 71 EXPORT_SYMBOL(vmemmap_start_pfn); 72 #endif 73 74 extern char _start[]; 75 void *_dtb_early_va __initdata; 76 uintptr_t _dtb_early_pa __initdata; 77 78 phys_addr_t dma32_phys_limit __initdata; 79 80 void __init arch_zone_limits_init(unsigned long *max_zone_pfns) 81 { 82 #ifdef CONFIG_ZONE_DMA32 83 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 84 #endif 85 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 86 } 87 88 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 89 90 #define LOG2_SZ_1K ilog2(SZ_1K) 91 #define LOG2_SZ_1M ilog2(SZ_1M) 92 #define LOG2_SZ_1G ilog2(SZ_1G) 93 #define LOG2_SZ_1T ilog2(SZ_1T) 94 95 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 96 { 97 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 98 (((t) - (b)) >> LOG2_SZ_1K)); 99 } 100 101 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 102 { 103 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 104 (((t) - (b)) >> LOG2_SZ_1M)); 105 } 106 107 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 108 { 109 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 110 (((t) - (b)) >> LOG2_SZ_1G)); 111 } 112 113 #ifdef CONFIG_64BIT 114 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 115 { 116 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 117 (((t) - (b)) >> LOG2_SZ_1T)); 118 } 119 #else 120 #define print_mlt(n, b, t) do {} while (0) 121 #endif 122 123 static inline void print_ml(char *name, unsigned long b, unsigned long t) 124 { 125 unsigned long diff = t - b; 126 127 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 128 print_mlt(name, b, t); 129 else if ((diff >> LOG2_SZ_1G) >= 10) 130 print_mlg(name, b, t); 131 else if ((diff >> LOG2_SZ_1M) >= 10) 132 print_mlm(name, b, t); 133 else 134 print_mlk(name, b, t); 135 } 136 137 static void __init print_vm_layout(void) 138 { 139 pr_notice("Virtual kernel memory layout:\n"); 140 print_ml("fixmap", (unsigned long)FIXADDR_START, 141 (unsigned long)FIXADDR_TOP); 142 print_ml("pci io", (unsigned long)PCI_IO_START, 143 (unsigned long)PCI_IO_END); 144 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 145 (unsigned long)VMEMMAP_END); 146 print_ml("vmalloc", (unsigned long)VMALLOC_START, 147 (unsigned long)VMALLOC_END); 148 #ifdef CONFIG_64BIT 149 print_ml("modules", (unsigned long)MODULES_VADDR, 150 (unsigned long)MODULES_END); 151 #endif 152 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 153 (unsigned long)high_memory); 154 if (IS_ENABLED(CONFIG_64BIT)) { 155 #ifdef CONFIG_KASAN 156 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 157 #endif 158 159 print_ml("kernel", (unsigned long)kernel_map.virt_addr, 160 (unsigned long)ADDRESS_SPACE_END); 161 } 162 } 163 #else 164 static void print_vm_layout(void) { } 165 #endif /* CONFIG_DEBUG_VM */ 166 167 void __init arch_mm_preinit(void) 168 { 169 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit) && 170 memblock_start_of_DRAM() < dma32_phys_limit; 171 unsigned int swiotlb_flags = SWIOTLB_VERBOSE; 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 * No 32-bit DMA bouncing needed (either all DRAM is within 180 * the 32-bit limit, or it all starts above it), but 181 * kmalloc() buffers whose sizes are not cache-line-aligned 182 * still require bouncing for non-coherent DMA. Use 183 * SWIOTLB_ANY so that the buffer can be allocated from high 184 * memory when DRAM starts above dma32_phys_limit. Allocate 185 * ~1 MB per 1 GB of RAM. 186 */ 187 unsigned long size = 188 DIV_ROUND_UP(memblock_phys_mem_size(), 1024); 189 swiotlb_adjust_size(min(swiotlb_default_pool_size(), size)); 190 swiotlb = true; 191 swiotlb_flags |= SWIOTLB_ANY; 192 } 193 194 swiotlb_init(swiotlb, swiotlb_flags); 195 196 print_vm_layout(); 197 } 198 199 /* Limit the memory size via mem. */ 200 static phys_addr_t memory_limit; 201 202 static int __init early_mem(char *p) 203 { 204 u64 size; 205 206 if (!p) 207 return 1; 208 209 size = memparse(p, &p) & PAGE_MASK; 210 memory_limit = min_t(u64, size, memory_limit); 211 212 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 213 214 return 0; 215 } 216 early_param("mem", early_mem); 217 218 static void __init setup_bootmem(void) 219 { 220 phys_addr_t vmlinux_end = __pa_symbol(&_end); 221 phys_addr_t max_mapped_addr; 222 phys_addr_t phys_ram_end, vmlinux_start; 223 224 vmlinux_start = __pa_symbol(&_start); 225 226 memblock_enforce_memory_limit(memory_limit); 227 228 /* 229 * Make sure we align the reservation on PMD_SIZE since we will 230 * map the kernel in the linear mapping as read-only: we do not want 231 * any allocation to happen between _end and the next pmd aligned page. 232 */ 233 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 234 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 235 /* 236 * Reserve from the start of the kernel to the end of the kernel 237 */ 238 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 239 240 /* 241 * Make sure we align the start of the memory on a PMD boundary so that 242 * at worst, we map the linear mapping with PMD mappings. 243 */ 244 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK; 245 #ifdef CONFIG_SPARSEMEM_VMEMMAP 246 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT; 247 #endif 248 249 /* 250 * In 64-bit, any use of __va/__pa before this point is wrong as we 251 * did not know the start of DRAM before. 252 */ 253 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) 254 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base; 255 256 /* 257 * The size of the linear page mapping may restrict the amount of 258 * usable RAM. 259 */ 260 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) { 261 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE; 262 if (memblock_end_of_DRAM() > max_mapped_addr) { 263 memblock_cap_memory_range(phys_ram_base, 264 max_mapped_addr - phys_ram_base); 265 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed", 266 &max_mapped_addr); 267 } 268 } 269 270 /* 271 * Reserve physical address space that would be mapped to virtual 272 * addresses greater than (void *)(-PAGE_SIZE) because: 273 * - This memory would overlap with ERR_PTR 274 * - This memory belongs to high memory, which is not supported 275 * 276 * This is not applicable to 64-bit kernel, because virtual addresses 277 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are 278 * occupied by kernel mapping. Also it is unrealistic for high memory 279 * to exist on 64-bit platforms. 280 */ 281 if (!IS_ENABLED(CONFIG_64BIT)) { 282 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE); 283 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr); 284 } 285 286 phys_ram_end = memblock_end_of_DRAM(); 287 min_low_pfn = PFN_UP(phys_ram_base); 288 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 289 290 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 291 292 reserve_initrd_mem(); 293 294 /* 295 * No allocation should be done before reserving the memory as defined 296 * in the device tree, otherwise the allocation could end up in a 297 * reserved region. 298 */ 299 early_init_fdt_scan_reserved_mem(); 300 301 /* 302 * If DTB is built in, no need to reserve its memblock. 303 * Otherwise, do reserve it but avoid using 304 * early_init_fdt_reserve_self() since __pa() does 305 * not work for DTB pointers that are fixmap addresses 306 */ 307 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 308 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 309 } 310 311 #ifdef CONFIG_RELOCATABLE 312 extern unsigned long __rela_dyn_start, __rela_dyn_end; 313 314 static void __init relocate_kernel(void) 315 { 316 Elf_Rela *rela = (Elf_Rela *)&__rela_dyn_start; 317 /* 318 * This holds the offset between the linked virtual address and the 319 * relocated virtual address. 320 */ 321 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR; 322 /* 323 * This holds the offset between kernel linked virtual address and 324 * physical address. 325 */ 326 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr; 327 328 for ( ; rela < (Elf_Rela *)&__rela_dyn_end; rela++) { 329 Elf_Addr addr = (rela->r_offset - va_kernel_link_pa_offset); 330 Elf_Addr relocated_addr = rela->r_addend; 331 332 if (rela->r_info != R_RISCV_RELATIVE) 333 continue; 334 335 /* 336 * Make sure to not relocate vdso symbols like rt_sigreturn 337 * which are linked from the address 0 in vmlinux since 338 * vdso symbol addresses are actually used as an offset from 339 * mm->context.vdso in VDSO_OFFSET macro. 340 */ 341 if (relocated_addr >= KERNEL_LINK_ADDR) 342 relocated_addr += reloc_offset; 343 344 *(Elf_Addr *)addr = relocated_addr; 345 } 346 } 347 #endif /* CONFIG_RELOCATABLE */ 348 349 #ifdef CONFIG_MMU 350 struct pt_alloc_ops pt_ops __meminitdata; 351 352 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 353 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 354 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 355 356 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 357 358 static const pgprot_t protection_map[16] = { 359 [VM_NONE] = PAGE_NONE, 360 [VM_READ] = PAGE_READ, 361 [VM_WRITE] = PAGE_SHADOWSTACK, 362 [VM_WRITE | VM_READ] = PAGE_COPY, 363 [VM_EXEC] = PAGE_EXEC, 364 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 365 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 366 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 367 [VM_SHARED] = PAGE_NONE, 368 [VM_SHARED | VM_READ] = PAGE_READ, 369 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 370 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 371 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 372 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 373 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 374 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 375 }; 376 DECLARE_VM_GET_PAGE_PROT 377 378 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 379 { 380 unsigned long addr = __fix_to_virt(idx); 381 pte_t *ptep; 382 383 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 384 385 ptep = &fixmap_pte[pte_index(addr)]; 386 387 if (pgprot_val(prot)) 388 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 389 else 390 pte_clear(&init_mm, addr, ptep); 391 local_flush_tlb_page(addr); 392 } 393 394 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 395 { 396 return (pte_t *)((uintptr_t)pa); 397 } 398 399 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 400 { 401 clear_fixmap(FIX_PTE); 402 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 403 } 404 405 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa) 406 { 407 return (pte_t *) __va(pa); 408 } 409 410 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 411 { 412 /* 413 * We only create PMD or PGD early mappings so we 414 * should never reach here with MMU disabled. 415 */ 416 BUG(); 417 } 418 419 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 420 { 421 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 422 } 423 424 static phys_addr_t __meminit alloc_pte_late(uintptr_t va) 425 { 426 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 427 428 /* 429 * We do not know which mm the PTE page is associated to at this point. 430 * Passing NULL to the ctor is the safe option, though it may result 431 * in unnecessary work (e.g. initialising the ptlock for init_mm). 432 */ 433 BUG_ON(!ptdesc || !pagetable_pte_ctor(NULL, ptdesc)); 434 return __pa((pte_t *)ptdesc_address(ptdesc)); 435 } 436 437 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 438 pgprot_t prot) 439 { 440 uintptr_t pte_idx = pte_index(va); 441 442 BUG_ON(sz != PAGE_SIZE); 443 444 if (pte_none(ptep[pte_idx])) 445 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 446 } 447 448 #ifndef __PAGETABLE_PMD_FOLDED 449 450 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 451 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 452 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 453 454 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 455 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 456 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 457 458 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 459 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 460 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 461 462 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 463 { 464 /* Before MMU is enabled */ 465 return (pmd_t *)((uintptr_t)pa); 466 } 467 468 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 469 { 470 clear_fixmap(FIX_PMD); 471 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 472 } 473 474 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa) 475 { 476 return (pmd_t *) __va(pa); 477 } 478 479 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 480 { 481 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 482 483 return (uintptr_t)early_pmd; 484 } 485 486 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 487 { 488 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 489 } 490 491 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va) 492 { 493 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 494 495 /* See comment in alloc_pte_late() regarding NULL passed the ctor */ 496 BUG_ON(!ptdesc || !pagetable_pmd_ctor(NULL, ptdesc)); 497 return __pa((pmd_t *)ptdesc_address(ptdesc)); 498 } 499 500 static void __meminit create_pmd_mapping(pmd_t *pmdp, 501 uintptr_t va, phys_addr_t pa, 502 phys_addr_t sz, pgprot_t prot) 503 { 504 pte_t *ptep; 505 phys_addr_t pte_phys; 506 uintptr_t pmd_idx = pmd_index(va); 507 508 if (sz == PMD_SIZE) { 509 if (pmd_none(pmdp[pmd_idx])) 510 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 511 return; 512 } 513 514 if (pmd_none(pmdp[pmd_idx])) { 515 pte_phys = pt_ops.alloc_pte(va); 516 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 517 ptep = pt_ops.get_pte_virt(pte_phys); 518 memset(ptep, 0, PAGE_SIZE); 519 } else { 520 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 521 ptep = pt_ops.get_pte_virt(pte_phys); 522 } 523 524 create_pte_mapping(ptep, va, pa, sz, prot); 525 } 526 527 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 528 { 529 return (pud_t *)((uintptr_t)pa); 530 } 531 532 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 533 { 534 clear_fixmap(FIX_PUD); 535 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 536 } 537 538 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa) 539 { 540 return (pud_t *)__va(pa); 541 } 542 543 static phys_addr_t __init alloc_pud_early(uintptr_t va) 544 { 545 /* Only one PUD is available for early mapping */ 546 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 547 548 return (uintptr_t)early_pud; 549 } 550 551 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 552 { 553 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 554 } 555 556 static phys_addr_t __meminit alloc_pud_late(uintptr_t va) 557 { 558 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 559 560 BUG_ON(!ptdesc); 561 pagetable_pud_ctor(ptdesc); 562 return __pa((pud_t *)ptdesc_address(ptdesc)); 563 } 564 565 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 566 { 567 return (p4d_t *)((uintptr_t)pa); 568 } 569 570 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 571 { 572 clear_fixmap(FIX_P4D); 573 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 574 } 575 576 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa) 577 { 578 return (p4d_t *)__va(pa); 579 } 580 581 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 582 { 583 /* Only one P4D is available for early mapping */ 584 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 585 586 return (uintptr_t)early_p4d; 587 } 588 589 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 590 { 591 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 592 } 593 594 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va) 595 { 596 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0); 597 598 BUG_ON(!ptdesc); 599 pagetable_p4d_ctor(ptdesc); 600 return __pa((p4d_t *)ptdesc_address(ptdesc)); 601 } 602 603 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 604 pgprot_t prot) 605 { 606 pmd_t *nextp; 607 phys_addr_t next_phys; 608 uintptr_t pud_index = pud_index(va); 609 610 if (sz == PUD_SIZE) { 611 if (pud_val(pudp[pud_index]) == 0) 612 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 613 return; 614 } 615 616 if (pud_val(pudp[pud_index]) == 0) { 617 next_phys = pt_ops.alloc_pmd(va); 618 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 619 nextp = pt_ops.get_pmd_virt(next_phys); 620 memset(nextp, 0, PAGE_SIZE); 621 } else { 622 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 623 nextp = pt_ops.get_pmd_virt(next_phys); 624 } 625 626 create_pmd_mapping(nextp, va, pa, sz, prot); 627 } 628 629 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 630 pgprot_t prot) 631 { 632 pud_t *nextp; 633 phys_addr_t next_phys; 634 uintptr_t p4d_index = p4d_index(va); 635 636 if (sz == P4D_SIZE) { 637 if (p4d_val(p4dp[p4d_index]) == 0) 638 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 639 return; 640 } 641 642 if (p4d_val(p4dp[p4d_index]) == 0) { 643 next_phys = pt_ops.alloc_pud(va); 644 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 645 nextp = pt_ops.get_pud_virt(next_phys); 646 memset(nextp, 0, PAGE_SIZE); 647 } else { 648 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 649 nextp = pt_ops.get_pud_virt(next_phys); 650 } 651 652 create_pud_mapping(nextp, va, pa, sz, prot); 653 } 654 655 #define pgd_next_t p4d_t 656 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 657 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 658 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 659 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 660 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 661 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 662 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 663 (pgtable_l5_enabled ? \ 664 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 665 (pgtable_l4_enabled ? \ 666 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 667 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 668 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 669 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 670 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 671 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 672 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 673 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 674 #else 675 #define pgd_next_t pte_t 676 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 677 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 678 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 679 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 680 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 681 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 682 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 683 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 684 #endif /* __PAGETABLE_PMD_FOLDED */ 685 686 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz, 687 pgprot_t prot) 688 { 689 pgd_next_t *nextp; 690 phys_addr_t next_phys; 691 uintptr_t pgd_idx = pgd_index(va); 692 693 if (sz == PGDIR_SIZE) { 694 if (pgd_val(pgdp[pgd_idx]) == 0) 695 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 696 return; 697 } 698 699 if (pgd_val(pgdp[pgd_idx]) == 0) { 700 next_phys = alloc_pgd_next(va); 701 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 702 nextp = get_pgd_next_virt(next_phys); 703 memset(nextp, 0, PAGE_SIZE); 704 } else { 705 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 706 nextp = get_pgd_next_virt(next_phys); 707 } 708 709 create_pgd_next_mapping(nextp, va, pa, sz, prot); 710 } 711 712 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size) 713 { 714 if (debug_pagealloc_enabled()) 715 return PAGE_SIZE; 716 717 if (pgtable_l5_enabled && 718 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE) 719 return P4D_SIZE; 720 721 if (pgtable_l4_enabled && 722 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE) 723 return PUD_SIZE; 724 725 if (IS_ENABLED(CONFIG_64BIT) && 726 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE) 727 return PMD_SIZE; 728 729 return PAGE_SIZE; 730 } 731 732 #ifdef CONFIG_STRICT_KERNEL_RWX 733 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 734 { 735 if (is_va_kernel_text(va)) 736 return PAGE_KERNEL_READ_EXEC; 737 738 /* 739 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 740 * we must protect its linear mapping alias from being executed and 741 * written. 742 * And rodata section is marked readonly in mark_rodata_ro. 743 */ 744 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 745 return PAGE_KERNEL_READ; 746 747 return PAGE_KERNEL; 748 } 749 750 void mark_rodata_ro(void) 751 { 752 set_kernel_memory(__start_rodata, _data, set_memory_ro); 753 if (IS_ENABLED(CONFIG_64BIT)) 754 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 755 set_memory_ro); 756 } 757 #else 758 static __meminit pgprot_t pgprot_from_va(uintptr_t va) 759 { 760 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 761 return PAGE_KERNEL; 762 763 return PAGE_KERNEL_EXEC; 764 } 765 #endif /* CONFIG_STRICT_KERNEL_RWX */ 766 767 #if defined(CONFIG_64BIT) 768 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa); 769 u64 __pi_set_satp_mode_from_fdt(uintptr_t dtb_pa); 770 771 static void __init disable_pgtable_l5(void) 772 { 773 pgtable_l5_enabled = false; 774 kernel_map.page_offset = PAGE_OFFSET_L4; 775 satp_mode = SATP_MODE_48; 776 } 777 778 static void __init disable_pgtable_l4(void) 779 { 780 pgtable_l4_enabled = false; 781 kernel_map.page_offset = PAGE_OFFSET_L3; 782 satp_mode = SATP_MODE_39; 783 } 784 785 static int __init print_no4lvl(char *p) 786 { 787 pr_info("Disabled 4-level and 5-level paging"); 788 return 0; 789 } 790 early_param("no4lvl", print_no4lvl); 791 792 static int __init print_no5lvl(char *p) 793 { 794 pr_info("Disabled 5-level paging"); 795 return 0; 796 } 797 early_param("no5lvl", print_no5lvl); 798 799 static void __init set_mmap_rnd_bits_max(void) 800 { 801 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3; 802 } 803 804 static bool __init is_vaddr_valid(unsigned long va) 805 { 806 unsigned long up = 0; 807 808 switch (satp_mode) { 809 case SATP_MODE_39: 810 up = 1UL << 38; 811 break; 812 case SATP_MODE_48: 813 up = 1UL << 47; 814 break; 815 case SATP_MODE_57: 816 up = 1UL << 56; 817 break; 818 default: 819 return false; 820 } 821 822 return (va < up) || (va >= (ULONG_MAX - up + 1)); 823 } 824 825 /* 826 * There is a simple way to determine if 4-level is supported by the 827 * underlying hardware: establish 1:1 mapping in 4-level page table mode 828 * then read SATP to see if the configuration was taken into account 829 * meaning sv48 is supported. 830 * The maximum SATP mode is limited by both the command line and the "mmu-type" 831 * property in the device tree, since some platforms may hang if an unsupported 832 * SATP mode is attempted. 833 */ 834 static __init void set_satp_mode(uintptr_t dtb_pa) 835 { 836 u64 identity_satp, hw_satp; 837 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 838 u64 satp_mode_limit = min_not_zero(__pi_set_satp_mode_from_cmdline(dtb_pa), 839 __pi_set_satp_mode_from_fdt(dtb_pa)); 840 841 kernel_map.page_offset = PAGE_OFFSET_L5; 842 843 if (satp_mode_limit == SATP_MODE_48) { 844 disable_pgtable_l5(); 845 } else if (satp_mode_limit == SATP_MODE_39) { 846 disable_pgtable_l5(); 847 disable_pgtable_l4(); 848 return; 849 } 850 851 create_p4d_mapping(early_p4d, 852 set_satp_mode_pmd, (uintptr_t)early_pud, 853 P4D_SIZE, PAGE_TABLE); 854 create_pud_mapping(early_pud, 855 set_satp_mode_pmd, (uintptr_t)early_pmd, 856 PUD_SIZE, PAGE_TABLE); 857 /* Handle the case where set_satp_mode straddles 2 PMDs */ 858 create_pmd_mapping(early_pmd, 859 set_satp_mode_pmd, set_satp_mode_pmd, 860 PMD_SIZE, PAGE_KERNEL_EXEC); 861 create_pmd_mapping(early_pmd, 862 set_satp_mode_pmd + PMD_SIZE, 863 set_satp_mode_pmd + PMD_SIZE, 864 PMD_SIZE, PAGE_KERNEL_EXEC); 865 retry: 866 if (!is_vaddr_valid(set_satp_mode_pmd)) 867 goto out; 868 869 create_pgd_mapping(early_pg_dir, 870 set_satp_mode_pmd, 871 pgtable_l5_enabled ? 872 (uintptr_t)early_p4d : (uintptr_t)early_pud, 873 PGDIR_SIZE, PAGE_TABLE); 874 875 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 876 877 local_flush_tlb_all(); 878 csr_write(CSR_SATP, identity_satp); 879 hw_satp = csr_swap(CSR_SATP, 0ULL); 880 local_flush_tlb_all(); 881 882 if (hw_satp != identity_satp) { 883 if (pgtable_l5_enabled) { 884 disable_pgtable_l5(); 885 memset(early_pg_dir, 0, PAGE_SIZE); 886 goto retry; 887 } 888 disable_pgtable_l4(); 889 } 890 891 out: 892 memset(early_pg_dir, 0, PAGE_SIZE); 893 memset(early_p4d, 0, PAGE_SIZE); 894 memset(early_pud, 0, PAGE_SIZE); 895 memset(early_pmd, 0, PAGE_SIZE); 896 } 897 #endif 898 899 /* 900 * setup_vm() is called from head.S with MMU-off. 901 * 902 * Following requirements should be honoured for setup_vm() to work 903 * correctly: 904 * 1) It should use PC-relative addressing for accessing kernel symbols. 905 * To achieve this we always use GCC cmodel=medany. 906 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 907 * so disable compiler instrumentation when FTRACE is enabled. 908 * 909 * Currently, the above requirements are honoured by using custom CFLAGS 910 * for init.o in mm/Makefile. 911 */ 912 913 #ifndef __riscv_cmodel_medany 914 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 915 #endif 916 917 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 918 { 919 uintptr_t va, end_va; 920 921 end_va = kernel_map.virt_addr + kernel_map.size; 922 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 923 create_pgd_mapping(pgdir, va, 924 kernel_map.phys_addr + (va - kernel_map.virt_addr), 925 PMD_SIZE, 926 early ? 927 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 928 } 929 930 /* 931 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 932 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 933 * entry. 934 */ 935 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va, 936 uintptr_t dtb_pa) 937 { 938 #ifndef CONFIG_BUILTIN_DTB 939 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 940 941 /* Make sure the fdt fixmap address is always aligned on PMD size */ 942 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE)); 943 944 /* In 32-bit only, the fdt lies in its own PGD */ 945 if (!IS_ENABLED(CONFIG_64BIT)) { 946 create_pgd_mapping(early_pg_dir, fix_fdt_va, 947 pa, MAX_FDT_SIZE, PAGE_KERNEL); 948 } else { 949 create_pmd_mapping(fixmap_pmd, fix_fdt_va, 950 pa, PMD_SIZE, PAGE_KERNEL); 951 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE, 952 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 953 } 954 955 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1)); 956 #else 957 /* 958 * For 64-bit kernel, __va can't be used since it would return a linear 959 * mapping address whereas dtb_early_va will be used before 960 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 961 * kernel is mapped in the linear mapping, that makes no difference. 962 */ 963 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa); 964 #endif 965 966 dtb_early_pa = dtb_pa; 967 } 968 969 /* 970 * MMU is not enabled, the page tables are allocated directly using 971 * early_pmd/pud/p4d and the address returned is the physical one. 972 */ 973 static void __init pt_ops_set_early(void) 974 { 975 pt_ops.alloc_pte = alloc_pte_early; 976 pt_ops.get_pte_virt = get_pte_virt_early; 977 #ifndef __PAGETABLE_PMD_FOLDED 978 pt_ops.alloc_pmd = alloc_pmd_early; 979 pt_ops.get_pmd_virt = get_pmd_virt_early; 980 pt_ops.alloc_pud = alloc_pud_early; 981 pt_ops.get_pud_virt = get_pud_virt_early; 982 pt_ops.alloc_p4d = alloc_p4d_early; 983 pt_ops.get_p4d_virt = get_p4d_virt_early; 984 #endif 985 } 986 987 /* 988 * MMU is enabled but page table setup is not complete yet. 989 * fixmap page table alloc functions must be used as a means to temporarily 990 * map the allocated physical pages since the linear mapping does not exist yet. 991 * 992 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 993 * but it will be used as described above. 994 */ 995 static void __init pt_ops_set_fixmap(void) 996 { 997 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap); 998 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap); 999 #ifndef __PAGETABLE_PMD_FOLDED 1000 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap); 1001 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap); 1002 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap); 1003 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap); 1004 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap); 1005 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap); 1006 #endif 1007 } 1008 1009 /* 1010 * MMU is enabled and page table setup is complete, so from now, we can use 1011 * generic page allocation functions to setup page table. 1012 */ 1013 static void __init pt_ops_set_late(void) 1014 { 1015 pt_ops.alloc_pte = alloc_pte_late; 1016 pt_ops.get_pte_virt = get_pte_virt_late; 1017 #ifndef __PAGETABLE_PMD_FOLDED 1018 pt_ops.alloc_pmd = alloc_pmd_late; 1019 pt_ops.get_pmd_virt = get_pmd_virt_late; 1020 pt_ops.alloc_pud = alloc_pud_late; 1021 pt_ops.get_pud_virt = get_pud_virt_late; 1022 pt_ops.alloc_p4d = alloc_p4d_late; 1023 pt_ops.get_p4d_virt = get_p4d_virt_late; 1024 #endif 1025 } 1026 1027 #ifdef CONFIG_RANDOMIZE_BASE 1028 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa); 1029 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa); 1030 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa); 1031 1032 static int __init print_nokaslr(char *p) 1033 { 1034 pr_info("Disabled KASLR"); 1035 return 0; 1036 } 1037 early_param("nokaslr", print_nokaslr); 1038 #endif 1039 1040 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1041 { 1042 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 1043 1044 #ifdef CONFIG_RANDOMIZE_BASE 1045 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) { 1046 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa); 1047 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start); 1048 u32 nr_pos; 1049 1050 if (kaslr_seed == 0) 1051 kaslr_seed = __pi_get_kaslr_seed(dtb_pa); 1052 /* 1053 * Compute the number of positions available: we are limited 1054 * by the early page table that only has one PUD and we must 1055 * be aligned on PMD_SIZE. 1056 */ 1057 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE; 1058 1059 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE; 1060 } 1061 #endif 1062 1063 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset; 1064 1065 kernel_map.phys_addr = (uintptr_t)(&_start); 1066 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 1067 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 1068 1069 #if defined(CONFIG_64BIT) 1070 set_satp_mode(dtb_pa); 1071 set_mmap_rnd_bits_max(); 1072 #endif 1073 1074 /* 1075 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem, 1076 * where we have the system memory layout: this allows us to align 1077 * the physical and virtual mappings and then make use of PUD/P4D/PGD 1078 * for the linear mapping. This is only possible because the kernel 1079 * mapping lies outside the linear mapping. 1080 * In 32-bit however, as the kernel resides in the linear mapping, 1081 * setup_vm_final can not change the mapping established here, 1082 * otherwise the same kernel addresses would get mapped to different 1083 * physical addresses (if the start of dram is different from the 1084 * kernel physical address start). 1085 */ 1086 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ? 1087 0UL : PAGE_OFFSET - kernel_map.phys_addr; 1088 1089 memory_limit = KERN_VIRT_SIZE; 1090 1091 /* Sanity check alignment and size */ 1092 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 1093 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 1094 1095 #ifdef CONFIG_64BIT 1096 /* 1097 * The last 4K bytes of the addressable memory can not be mapped because 1098 * of IS_ERR_VALUE macro. 1099 */ 1100 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1101 #endif 1102 1103 #ifdef CONFIG_RELOCATABLE 1104 /* 1105 * Early page table uses only one PUD, which makes it possible 1106 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset 1107 * makes the kernel cross over a PUD_SIZE boundary, raise a bug 1108 * since a part of the kernel would not get mapped. 1109 */ 1110 if (IS_ENABLED(CONFIG_64BIT)) 1111 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size); 1112 relocate_kernel(); 1113 #endif 1114 1115 apply_early_boot_alternatives(); 1116 pt_ops_set_early(); 1117 1118 /* Setup early PGD for fixmap */ 1119 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1120 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1121 1122 #ifndef __PAGETABLE_PMD_FOLDED 1123 /* Setup fixmap P4D and PUD */ 1124 if (pgtable_l5_enabled) 1125 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1126 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1127 /* Setup fixmap PUD and PMD */ 1128 if (pgtable_l4_enabled) 1129 create_pud_mapping(fixmap_pud, FIXADDR_START, 1130 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1131 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1132 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1133 /* Setup trampoline PGD and PMD */ 1134 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1135 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1136 if (pgtable_l5_enabled) 1137 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1138 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1139 if (pgtable_l4_enabled) 1140 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1141 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1142 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1143 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1144 #else 1145 /* Setup trampoline PGD */ 1146 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1147 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1148 #endif 1149 1150 /* 1151 * Setup early PGD covering entire kernel which will allow 1152 * us to reach paging_init(). We map all memory banks later 1153 * in setup_vm_final() below. 1154 */ 1155 create_kernel_page_table(early_pg_dir, true); 1156 1157 /* Setup early mapping for FDT early scan */ 1158 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa); 1159 1160 /* 1161 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1162 * range can not span multiple pmds. 1163 */ 1164 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1165 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1166 1167 #ifndef __PAGETABLE_PMD_FOLDED 1168 /* 1169 * Early ioremap fixmap is already created as it lies within first 2MB 1170 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1171 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1172 * the user if not. 1173 */ 1174 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1175 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1176 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1177 WARN_ON(1); 1178 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1179 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1180 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1181 fix_to_virt(FIX_BTMAP_BEGIN)); 1182 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1183 fix_to_virt(FIX_BTMAP_END)); 1184 1185 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1186 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1187 } 1188 #endif 1189 1190 pt_ops_set_fixmap(); 1191 } 1192 1193 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end, 1194 uintptr_t fixed_map_size, const pgprot_t *pgprot) 1195 { 1196 phys_addr_t pa; 1197 uintptr_t va, map_size; 1198 1199 for (pa = start; pa < end; pa += map_size) { 1200 va = (uintptr_t)__va(pa); 1201 map_size = fixed_map_size ? fixed_map_size : 1202 best_map_size(pa, va, end - pa); 1203 1204 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1205 pgprot ? *pgprot : pgprot_from_va(va)); 1206 } 1207 } 1208 1209 static void __init create_linear_mapping_page_table(void) 1210 { 1211 phys_addr_t start, end; 1212 phys_addr_t kfence_pool __maybe_unused; 1213 u64 i; 1214 1215 #ifdef CONFIG_STRICT_KERNEL_RWX 1216 phys_addr_t ktext_start = __pa_symbol(_start); 1217 phys_addr_t ktext_size = __init_data_begin - _start; 1218 phys_addr_t krodata_start = __pa_symbol(__start_rodata); 1219 phys_addr_t krodata_size = _data - __start_rodata; 1220 1221 /* Isolate kernel text and rodata so they don't get mapped with a PUD */ 1222 memblock_mark_nomap(ktext_start, ktext_size); 1223 memblock_mark_nomap(krodata_start, krodata_size); 1224 #endif 1225 1226 #ifdef CONFIG_KFENCE 1227 /* 1228 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it 1229 * before we setup the linear mapping so that we avoid using hugepages 1230 * for this region. 1231 */ 1232 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE); 1233 BUG_ON(!kfence_pool); 1234 1235 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE); 1236 __kfence_pool = __va(kfence_pool); 1237 #endif 1238 1239 /* Map all memory banks in the linear mapping */ 1240 for_each_mem_range(i, &start, &end) { 1241 if (start <= __pa(PAGE_OFFSET) && 1242 __pa(PAGE_OFFSET) < end) 1243 start = __pa(PAGE_OFFSET); 1244 1245 create_linear_mapping_range(start, end, 0, NULL); 1246 } 1247 1248 #ifdef CONFIG_STRICT_KERNEL_RWX 1249 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL); 1250 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL); 1251 1252 memblock_clear_nomap(ktext_start, ktext_size); 1253 memblock_clear_nomap(krodata_start, krodata_size); 1254 #endif 1255 1256 #ifdef CONFIG_KFENCE 1257 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL); 1258 1259 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE); 1260 #endif 1261 } 1262 1263 static void __init setup_vm_final(void) 1264 { 1265 /* Setup swapper PGD for fixmap */ 1266 #if !defined(CONFIG_64BIT) 1267 /* 1268 * In 32-bit, the device tree lies in a pgd entry, so it must be copied 1269 * directly in swapper_pg_dir in addition to the pgd entry that points 1270 * to fixmap_pte. 1271 */ 1272 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT)); 1273 1274 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]); 1275 #endif 1276 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1277 __pa_symbol(fixmap_pgd_next), 1278 PGDIR_SIZE, PAGE_TABLE); 1279 1280 /* Map the linear mapping */ 1281 create_linear_mapping_page_table(); 1282 1283 /* Map the kernel */ 1284 if (IS_ENABLED(CONFIG_64BIT)) 1285 create_kernel_page_table(swapper_pg_dir, false); 1286 1287 #ifdef CONFIG_KASAN 1288 kasan_swapper_init(); 1289 #endif 1290 1291 /* Clear fixmap PTE and PMD mappings */ 1292 clear_fixmap(FIX_PTE); 1293 clear_fixmap(FIX_PMD); 1294 clear_fixmap(FIX_PUD); 1295 clear_fixmap(FIX_P4D); 1296 1297 /* Move to swapper page table */ 1298 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1299 local_flush_tlb_all(); 1300 1301 pt_ops_set_late(); 1302 } 1303 #else 1304 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1305 { 1306 dtb_early_va = (void *)dtb_pa; 1307 dtb_early_pa = dtb_pa; 1308 1309 #ifdef CONFIG_RELOCATABLE 1310 kernel_map.virt_addr = (uintptr_t)_start; 1311 kernel_map.phys_addr = (uintptr_t)_start; 1312 relocate_kernel(); 1313 #endif 1314 } 1315 1316 static inline void setup_vm_final(void) 1317 { 1318 } 1319 #endif /* CONFIG_MMU */ 1320 1321 /* 1322 * reserve_crashkernel() - reserves memory for crash kernel 1323 * 1324 * This function reserves memory area given in "crashkernel=" kernel command 1325 * line parameter. The memory reserved is used by dump capture kernel when 1326 * primary kernel is crashing. 1327 */ 1328 static void __init arch_reserve_crashkernel(void) 1329 { 1330 unsigned long long low_size = 0, cma_size = 0; 1331 unsigned long long crash_base, crash_size; 1332 bool high = false; 1333 int ret; 1334 1335 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 1336 return; 1337 1338 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1339 &crash_size, &crash_base, 1340 &low_size, &cma_size, &high); 1341 if (ret) 1342 return; 1343 1344 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 1345 reserve_crashkernel_cma(cma_size); 1346 } 1347 1348 void __init paging_init(void) 1349 { 1350 setup_bootmem(); 1351 setup_vm_final(); 1352 1353 /* Depend on that Linear Mapping is ready */ 1354 memblock_allow_resize(); 1355 } 1356 1357 void __init misc_mem_init(void) 1358 { 1359 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1360 arch_numa_init(); 1361 dma_contiguous_reserve(dma32_phys_limit); 1362 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1363 /* The entire VMEMMAP region has been populated. Flush TLB for this region */ 1364 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END); 1365 #endif 1366 arch_reserve_crashkernel(); 1367 memblock_dump_all(); 1368 } 1369 1370 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1371 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1372 struct vmem_altmap *altmap) 1373 { 1374 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END)); 1375 1376 /* 1377 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we 1378 * can't use hugepage mappings for 2-level page table because in case of 1379 * memory hotplug, we are not able to update all the page tables with 1380 * the new PMDs. 1381 */ 1382 return vmemmap_populate_hugepages(start, end, node, altmap); 1383 } 1384 #endif 1385 1386 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT) 1387 /* 1388 * Pre-allocates page-table pages for a specific area in the kernel 1389 * page-table. Only the level which needs to be synchronized between 1390 * all page-tables is allocated because the synchronization can be 1391 * expensive. 1392 */ 1393 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end, 1394 const char *area) 1395 { 1396 unsigned long addr; 1397 const char *lvl; 1398 1399 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) { 1400 pgd_t *pgd = pgd_offset_k(addr); 1401 p4d_t *p4d; 1402 pud_t *pud; 1403 pmd_t *pmd; 1404 1405 lvl = "p4d"; 1406 p4d = p4d_alloc(&init_mm, pgd, addr); 1407 if (!p4d) 1408 goto failed; 1409 1410 if (pgtable_l5_enabled) 1411 continue; 1412 1413 lvl = "pud"; 1414 pud = pud_alloc(&init_mm, p4d, addr); 1415 if (!pud) 1416 goto failed; 1417 1418 if (pgtable_l4_enabled) 1419 continue; 1420 1421 lvl = "pmd"; 1422 pmd = pmd_alloc(&init_mm, pud, addr); 1423 if (!pmd) 1424 goto failed; 1425 } 1426 return; 1427 1428 failed: 1429 /* 1430 * The pages have to be there now or they will be missing in 1431 * process page-tables later. 1432 */ 1433 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area); 1434 } 1435 1436 #define PAGE_END KASAN_SHADOW_START 1437 1438 void __init pgtable_cache_init(void) 1439 { 1440 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc"); 1441 if (IS_ENABLED(CONFIG_MODULES)) 1442 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules"); 1443 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) { 1444 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap"); 1445 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map"); 1446 if (IS_ENABLED(CONFIG_KASAN)) 1447 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan"); 1448 } 1449 } 1450 #endif 1451 1452 #ifdef CONFIG_EXECMEM 1453 #ifdef CONFIG_MMU 1454 static struct execmem_info execmem_info __ro_after_init; 1455 1456 struct execmem_info __init *execmem_arch_setup(void) 1457 { 1458 execmem_info = (struct execmem_info){ 1459 .ranges = { 1460 [EXECMEM_DEFAULT] = { 1461 .start = MODULES_VADDR, 1462 .end = MODULES_END, 1463 .pgprot = PAGE_KERNEL, 1464 .alignment = 1, 1465 }, 1466 [EXECMEM_KPROBES] = { 1467 .start = VMALLOC_START, 1468 .end = VMALLOC_END, 1469 .pgprot = IS_ENABLED(CONFIG_STRICT_MODULE_RWX) ? 1470 PAGE_KERNEL_READ_EXEC : 1471 PAGE_KERNEL_EXEC, 1472 .alignment = 1, 1473 }, 1474 [EXECMEM_BPF] = { 1475 .start = BPF_JIT_REGION_START, 1476 .end = BPF_JIT_REGION_END, 1477 .pgprot = PAGE_KERNEL, 1478 .alignment = PAGE_SIZE, 1479 }, 1480 }, 1481 }; 1482 1483 return &execmem_info; 1484 } 1485 #endif /* CONFIG_MMU */ 1486 #endif /* CONFIG_EXECMEM */ 1487 1488 #ifdef CONFIG_MEMORY_HOTPLUG 1489 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd) 1490 { 1491 struct page *page = pmd_page(*pmd); 1492 struct ptdesc *ptdesc = page_ptdesc(page); 1493 pte_t *pte; 1494 int i; 1495 1496 for (i = 0; i < PTRS_PER_PTE; i++) { 1497 pte = pte_start + i; 1498 if (!pte_none(*pte)) 1499 return; 1500 } 1501 1502 pagetable_dtor(ptdesc); 1503 if (PageReserved(page)) 1504 free_reserved_page(page); 1505 else 1506 pagetable_free(ptdesc); 1507 pmd_clear(pmd); 1508 } 1509 1510 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap) 1511 { 1512 struct page *page = pud_page(*pud); 1513 struct ptdesc *ptdesc = page_ptdesc(page); 1514 pmd_t *pmd; 1515 int i; 1516 1517 for (i = 0; i < PTRS_PER_PMD; i++) { 1518 pmd = pmd_start + i; 1519 if (!pmd_none(*pmd)) 1520 return; 1521 } 1522 1523 if (!is_vmemmap) 1524 pagetable_dtor(ptdesc); 1525 if (PageReserved(page)) 1526 free_reserved_page(page); 1527 else 1528 pagetable_free(ptdesc); 1529 pud_clear(pud); 1530 } 1531 1532 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d) 1533 { 1534 struct page *page = p4d_page(*p4d); 1535 pud_t *pud; 1536 int i; 1537 1538 for (i = 0; i < PTRS_PER_PUD; i++) { 1539 pud = pud_start + i; 1540 if (!pud_none(*pud)) 1541 return; 1542 } 1543 1544 if (PageReserved(page)) 1545 free_reserved_page(page); 1546 else 1547 __free_pages(page, 0); 1548 p4d_clear(p4d); 1549 } 1550 1551 static void __meminit free_vmemmap_storage(struct page *page, size_t size, 1552 struct vmem_altmap *altmap) 1553 { 1554 int order = get_order(size); 1555 1556 if (altmap) { 1557 vmem_altmap_free(altmap, size >> PAGE_SHIFT); 1558 return; 1559 } 1560 1561 if (PageReserved(page)) { 1562 unsigned int nr_pages = 1 << order; 1563 1564 while (nr_pages--) 1565 free_reserved_page(page++); 1566 return; 1567 } 1568 1569 __free_pages(page, order); 1570 } 1571 1572 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end, 1573 bool is_vmemmap, struct vmem_altmap *altmap) 1574 { 1575 unsigned long next; 1576 pte_t *ptep, pte; 1577 1578 for (; addr < end; addr = next) { 1579 next = (addr + PAGE_SIZE) & PAGE_MASK; 1580 if (next > end) 1581 next = end; 1582 1583 ptep = pte_base + pte_index(addr); 1584 pte = ptep_get(ptep); 1585 if (!pte_present(*ptep)) 1586 continue; 1587 1588 pte_clear(&init_mm, addr, ptep); 1589 if (is_vmemmap) 1590 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap); 1591 } 1592 } 1593 1594 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end, 1595 bool is_vmemmap, struct vmem_altmap *altmap) 1596 { 1597 unsigned long next; 1598 pte_t *pte_base; 1599 pmd_t *pmdp, pmd; 1600 1601 for (; addr < end; addr = next) { 1602 next = pmd_addr_end(addr, end); 1603 pmdp = pmd_base + pmd_index(addr); 1604 pmd = pmdp_get(pmdp); 1605 if (!pmd_present(pmd)) 1606 continue; 1607 1608 if (pmd_leaf(pmd)) { 1609 pmd_clear(pmdp); 1610 if (is_vmemmap) 1611 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap); 1612 continue; 1613 } 1614 1615 pte_base = (pte_t *)pmd_page_vaddr(*pmdp); 1616 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap); 1617 free_pte_table(pte_base, pmdp); 1618 } 1619 } 1620 1621 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end, 1622 bool is_vmemmap, struct vmem_altmap *altmap) 1623 { 1624 unsigned long next; 1625 pud_t *pudp, pud; 1626 pmd_t *pmd_base; 1627 1628 for (; addr < end; addr = next) { 1629 next = pud_addr_end(addr, end); 1630 pudp = pgtable_l4_enabled ? pud_base + pud_index(addr) : pud_base; 1631 pud = pudp_get(pudp); 1632 if (!pud_present(pud)) 1633 continue; 1634 1635 if (pud_leaf(pud)) { 1636 if (pgtable_l4_enabled) { 1637 pud_clear(pudp); 1638 if (is_vmemmap) 1639 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap); 1640 } 1641 continue; 1642 } 1643 1644 pmd_base = pmd_offset(pudp, 0); 1645 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap); 1646 1647 if (pgtable_l4_enabled) 1648 free_pmd_table(pmd_base, pudp, is_vmemmap); 1649 } 1650 } 1651 1652 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end, 1653 bool is_vmemmap, struct vmem_altmap *altmap) 1654 { 1655 unsigned long next; 1656 p4d_t *p4dp, p4d; 1657 pud_t *pud_base; 1658 1659 for (; addr < end; addr = next) { 1660 next = p4d_addr_end(addr, end); 1661 p4dp = pgtable_l5_enabled ? p4d_base + p4d_index(addr) : p4d_base; 1662 p4d = p4dp_get(p4dp); 1663 if (!p4d_present(p4d)) 1664 continue; 1665 1666 if (p4d_leaf(p4d)) { 1667 if (pgtable_l5_enabled) { 1668 p4d_clear(p4dp); 1669 if (is_vmemmap) 1670 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap); 1671 } 1672 continue; 1673 } 1674 1675 pud_base = pud_offset(p4dp, 0); 1676 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap); 1677 1678 if (pgtable_l5_enabled) 1679 free_pud_table(pud_base, p4dp); 1680 } 1681 } 1682 1683 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap, 1684 struct vmem_altmap *altmap) 1685 { 1686 unsigned long addr, next; 1687 p4d_t *p4d_base; 1688 pgd_t *pgd; 1689 1690 for (addr = va; addr < end; addr = next) { 1691 next = pgd_addr_end(addr, end); 1692 pgd = pgd_offset_k(addr); 1693 1694 if (!pgd_present(*pgd)) 1695 continue; 1696 1697 if (pgd_leaf(*pgd)) 1698 continue; 1699 1700 p4d_base = p4d_offset(pgd, 0); 1701 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap); 1702 } 1703 1704 flush_tlb_all(); 1705 } 1706 1707 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size) 1708 { 1709 unsigned long va = (unsigned long)__va(start); 1710 unsigned long end = (unsigned long)__va(start + size); 1711 1712 remove_pgd_mapping(va, end, false, NULL); 1713 } 1714 1715 struct range arch_get_mappable_range(void) 1716 { 1717 struct range mhp_range; 1718 1719 mhp_range.start = __pa(PAGE_OFFSET); 1720 mhp_range.end = __pa(PAGE_END - 1); 1721 return mhp_range; 1722 } 1723 1724 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params) 1725 { 1726 int ret = 0; 1727 1728 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot); 1729 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params); 1730 if (ret) { 1731 remove_linear_mapping(start, size); 1732 goto out; 1733 } 1734 1735 max_pfn = PFN_UP(start + size); 1736 max_low_pfn = max_pfn; 1737 1738 out: 1739 flush_tlb_all(); 1740 return ret; 1741 } 1742 1743 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap, 1744 struct dev_pagemap *pgmap) 1745 { 1746 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap, pgmap); 1747 remove_linear_mapping(start, size); 1748 flush_tlb_all(); 1749 } 1750 1751 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) 1752 { 1753 remove_pgd_mapping(start, end, true, altmap); 1754 } 1755 #endif /* CONFIG_MEMORY_HOTPLUG */ 1756