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 23 #include <asm/fixmap.h> 24 #include <asm/tlbflush.h> 25 #include <asm/sections.h> 26 #include <asm/soc.h> 27 #include <asm/io.h> 28 #include <asm/ptdump.h> 29 #include <asm/numa.h> 30 31 #include "../kernel/head.h" 32 33 struct kernel_mapping kernel_map __ro_after_init; 34 EXPORT_SYMBOL(kernel_map); 35 #ifdef CONFIG_XIP_KERNEL 36 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 37 #endif 38 39 phys_addr_t phys_ram_base __ro_after_init; 40 EXPORT_SYMBOL(phys_ram_base); 41 42 #ifdef CONFIG_XIP_KERNEL 43 extern char _xiprom[], _exiprom[]; 44 #endif 45 46 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 47 __page_aligned_bss; 48 EXPORT_SYMBOL(empty_zero_page); 49 50 extern char _start[]; 51 #define DTB_EARLY_BASE_VA PGDIR_SIZE 52 void *_dtb_early_va __initdata; 53 uintptr_t _dtb_early_pa __initdata; 54 55 struct pt_alloc_ops { 56 pte_t *(*get_pte_virt)(phys_addr_t pa); 57 phys_addr_t (*alloc_pte)(uintptr_t va); 58 #ifndef __PAGETABLE_PMD_FOLDED 59 pmd_t *(*get_pmd_virt)(phys_addr_t pa); 60 phys_addr_t (*alloc_pmd)(uintptr_t va); 61 #endif 62 }; 63 64 static phys_addr_t dma32_phys_limit __initdata; 65 66 static void __init zone_sizes_init(void) 67 { 68 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 69 70 #ifdef CONFIG_ZONE_DMA32 71 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 72 #endif 73 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 74 75 free_area_init(max_zone_pfns); 76 } 77 78 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 79 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 80 { 81 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 82 (((t) - (b)) >> 10)); 83 } 84 85 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 86 { 87 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 88 (((t) - (b)) >> 20)); 89 } 90 91 static void __init print_vm_layout(void) 92 { 93 pr_notice("Virtual kernel memory layout:\n"); 94 print_mlk("fixmap", (unsigned long)FIXADDR_START, 95 (unsigned long)FIXADDR_TOP); 96 print_mlm("pci io", (unsigned long)PCI_IO_START, 97 (unsigned long)PCI_IO_END); 98 print_mlm("vmemmap", (unsigned long)VMEMMAP_START, 99 (unsigned long)VMEMMAP_END); 100 print_mlm("vmalloc", (unsigned long)VMALLOC_START, 101 (unsigned long)VMALLOC_END); 102 print_mlm("lowmem", (unsigned long)PAGE_OFFSET, 103 (unsigned long)high_memory); 104 #ifdef CONFIG_64BIT 105 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR, 106 (unsigned long)ADDRESS_SPACE_END); 107 #endif 108 } 109 #else 110 static void print_vm_layout(void) { } 111 #endif /* CONFIG_DEBUG_VM */ 112 113 void __init mem_init(void) 114 { 115 #ifdef CONFIG_FLATMEM 116 BUG_ON(!mem_map); 117 #endif /* CONFIG_FLATMEM */ 118 119 #ifdef CONFIG_SWIOTLB 120 if (swiotlb_force == SWIOTLB_FORCE || 121 max_pfn > PFN_DOWN(dma32_phys_limit)) 122 swiotlb_init(1); 123 else 124 swiotlb_force = SWIOTLB_NO_FORCE; 125 #endif 126 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 127 memblock_free_all(); 128 129 print_vm_layout(); 130 } 131 132 /* 133 * The default maximal physical memory size is -PAGE_OFFSET for 32-bit kernel, 134 * whereas for 64-bit kernel, the end of the virtual address space is occupied 135 * by the modules/BPF/kernel mappings which reduces the available size of the 136 * linear mapping. 137 * Limit the memory size via mem. 138 */ 139 #ifdef CONFIG_64BIT 140 static phys_addr_t memory_limit = -PAGE_OFFSET - SZ_4G; 141 #else 142 static phys_addr_t memory_limit = -PAGE_OFFSET; 143 #endif 144 145 static int __init early_mem(char *p) 146 { 147 u64 size; 148 149 if (!p) 150 return 1; 151 152 size = memparse(p, &p) & PAGE_MASK; 153 memory_limit = min_t(u64, size, memory_limit); 154 155 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 156 157 return 0; 158 } 159 early_param("mem", early_mem); 160 161 static void __init setup_bootmem(void) 162 { 163 phys_addr_t vmlinux_end = __pa_symbol(&_end); 164 phys_addr_t vmlinux_start = __pa_symbol(&_start); 165 phys_addr_t __maybe_unused max_mapped_addr; 166 phys_addr_t phys_ram_end; 167 168 #ifdef CONFIG_XIP_KERNEL 169 vmlinux_start = __pa_symbol(&_sdata); 170 #endif 171 172 memblock_enforce_memory_limit(memory_limit); 173 174 /* 175 * Reserve from the start of the kernel to the end of the kernel 176 */ 177 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX) 178 /* 179 * Make sure we align the reservation on PMD_SIZE since we will 180 * map the kernel in the linear mapping as read-only: we do not want 181 * any allocation to happen between _end and the next pmd aligned page. 182 */ 183 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 184 #endif 185 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 186 187 188 phys_ram_end = memblock_end_of_DRAM(); 189 #ifndef CONFIG_64BIT 190 #ifndef CONFIG_XIP_KERNEL 191 phys_ram_base = memblock_start_of_DRAM(); 192 #endif 193 /* 194 * memblock allocator is not aware of the fact that last 4K bytes of 195 * the addressable memory can not be mapped because of IS_ERR_VALUE 196 * macro. Make sure that last 4k bytes are not usable by memblock 197 * if end of dram is equal to maximum addressable memory. For 64-bit 198 * kernel, this problem can't happen here as the end of the virtual 199 * address space is occupied by the kernel mapping then this check must 200 * be done as soon as the kernel mapping base address is determined. 201 */ 202 max_mapped_addr = __pa(~(ulong)0); 203 if (max_mapped_addr == (phys_ram_end - 1)) 204 memblock_set_current_limit(max_mapped_addr - 4096); 205 #endif 206 207 min_low_pfn = PFN_UP(phys_ram_base); 208 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 209 210 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 211 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 212 213 reserve_initrd_mem(); 214 /* 215 * If DTB is built in, no need to reserve its memblock. 216 * Otherwise, do reserve it but avoid using 217 * early_init_fdt_reserve_self() since __pa() does 218 * not work for DTB pointers that are fixmap addresses 219 */ 220 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 221 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 222 223 early_init_fdt_scan_reserved_mem(); 224 dma_contiguous_reserve(dma32_phys_limit); 225 memblock_allow_resize(); 226 } 227 228 #ifdef CONFIG_MMU 229 static struct pt_alloc_ops _pt_ops __initdata; 230 231 #ifdef CONFIG_XIP_KERNEL 232 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&_pt_ops)) 233 #else 234 #define pt_ops _pt_ops 235 #endif 236 237 unsigned long pfn_base __ro_after_init; 238 EXPORT_SYMBOL(pfn_base); 239 240 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 241 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 242 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 243 244 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 245 246 #ifdef CONFIG_XIP_KERNEL 247 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 248 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 249 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 250 #endif /* CONFIG_XIP_KERNEL */ 251 252 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 253 { 254 unsigned long addr = __fix_to_virt(idx); 255 pte_t *ptep; 256 257 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 258 259 ptep = &fixmap_pte[pte_index(addr)]; 260 261 if (pgprot_val(prot)) 262 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 263 else 264 pte_clear(&init_mm, addr, ptep); 265 local_flush_tlb_page(addr); 266 } 267 268 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 269 { 270 return (pte_t *)((uintptr_t)pa); 271 } 272 273 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 274 { 275 clear_fixmap(FIX_PTE); 276 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 277 } 278 279 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 280 { 281 return (pte_t *) __va(pa); 282 } 283 284 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 285 { 286 /* 287 * We only create PMD or PGD early mappings so we 288 * should never reach here with MMU disabled. 289 */ 290 BUG(); 291 } 292 293 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 294 { 295 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 296 } 297 298 static phys_addr_t __init alloc_pte_late(uintptr_t va) 299 { 300 unsigned long vaddr; 301 302 vaddr = __get_free_page(GFP_KERNEL); 303 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr))); 304 305 return __pa(vaddr); 306 } 307 308 static void __init create_pte_mapping(pte_t *ptep, 309 uintptr_t va, phys_addr_t pa, 310 phys_addr_t sz, pgprot_t prot) 311 { 312 uintptr_t pte_idx = pte_index(va); 313 314 BUG_ON(sz != PAGE_SIZE); 315 316 if (pte_none(ptep[pte_idx])) 317 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 318 } 319 320 #ifndef __PAGETABLE_PMD_FOLDED 321 322 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 323 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 324 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 325 static pmd_t early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 326 327 #ifdef CONFIG_XIP_KERNEL 328 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 329 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 330 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 331 #endif /* CONFIG_XIP_KERNEL */ 332 333 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 334 { 335 /* Before MMU is enabled */ 336 return (pmd_t *)((uintptr_t)pa); 337 } 338 339 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 340 { 341 clear_fixmap(FIX_PMD); 342 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 343 } 344 345 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 346 { 347 return (pmd_t *) __va(pa); 348 } 349 350 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 351 { 352 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 353 354 return (uintptr_t)early_pmd; 355 } 356 357 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 358 { 359 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 360 } 361 362 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 363 { 364 unsigned long vaddr; 365 366 vaddr = __get_free_page(GFP_KERNEL); 367 BUG_ON(!vaddr); 368 return __pa(vaddr); 369 } 370 371 static void __init create_pmd_mapping(pmd_t *pmdp, 372 uintptr_t va, phys_addr_t pa, 373 phys_addr_t sz, pgprot_t prot) 374 { 375 pte_t *ptep; 376 phys_addr_t pte_phys; 377 uintptr_t pmd_idx = pmd_index(va); 378 379 if (sz == PMD_SIZE) { 380 if (pmd_none(pmdp[pmd_idx])) 381 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 382 return; 383 } 384 385 if (pmd_none(pmdp[pmd_idx])) { 386 pte_phys = pt_ops.alloc_pte(va); 387 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 388 ptep = pt_ops.get_pte_virt(pte_phys); 389 memset(ptep, 0, PAGE_SIZE); 390 } else { 391 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 392 ptep = pt_ops.get_pte_virt(pte_phys); 393 } 394 395 create_pte_mapping(ptep, va, pa, sz, prot); 396 } 397 398 #define pgd_next_t pmd_t 399 #define alloc_pgd_next(__va) pt_ops.alloc_pmd(__va) 400 #define get_pgd_next_virt(__pa) pt_ops.get_pmd_virt(__pa) 401 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 402 create_pmd_mapping(__nextp, __va, __pa, __sz, __prot) 403 #define fixmap_pgd_next fixmap_pmd 404 #else 405 #define pgd_next_t pte_t 406 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 407 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 408 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 409 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 410 #define fixmap_pgd_next fixmap_pte 411 #endif 412 413 void __init create_pgd_mapping(pgd_t *pgdp, 414 uintptr_t va, phys_addr_t pa, 415 phys_addr_t sz, pgprot_t prot) 416 { 417 pgd_next_t *nextp; 418 phys_addr_t next_phys; 419 uintptr_t pgd_idx = pgd_index(va); 420 421 if (sz == PGDIR_SIZE) { 422 if (pgd_val(pgdp[pgd_idx]) == 0) 423 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 424 return; 425 } 426 427 if (pgd_val(pgdp[pgd_idx]) == 0) { 428 next_phys = alloc_pgd_next(va); 429 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 430 nextp = get_pgd_next_virt(next_phys); 431 memset(nextp, 0, PAGE_SIZE); 432 } else { 433 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 434 nextp = get_pgd_next_virt(next_phys); 435 } 436 437 create_pgd_next_mapping(nextp, va, pa, sz, prot); 438 } 439 440 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size) 441 { 442 /* Upgrade to PMD_SIZE mappings whenever possible */ 443 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1))) 444 return PAGE_SIZE; 445 446 return PMD_SIZE; 447 } 448 449 #ifdef CONFIG_XIP_KERNEL 450 /* called from head.S with MMU off */ 451 asmlinkage void __init __copy_data(void) 452 { 453 void *from = (void *)(&_sdata); 454 void *end = (void *)(&_end); 455 void *to = (void *)CONFIG_PHYS_RAM_BASE; 456 size_t sz = (size_t)(end - from + 1); 457 458 memcpy(to, from, sz); 459 } 460 #endif 461 462 #ifdef CONFIG_STRICT_KERNEL_RWX 463 static __init pgprot_t pgprot_from_va(uintptr_t va) 464 { 465 if (is_va_kernel_text(va)) 466 return PAGE_KERNEL_READ_EXEC; 467 468 /* 469 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 470 * we must protect its linear mapping alias from being executed and 471 * written. 472 * And rodata section is marked readonly in mark_rodata_ro. 473 */ 474 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 475 return PAGE_KERNEL_READ; 476 477 return PAGE_KERNEL; 478 } 479 480 void mark_rodata_ro(void) 481 { 482 set_kernel_memory(__start_rodata, _data, set_memory_ro); 483 if (IS_ENABLED(CONFIG_64BIT)) 484 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 485 set_memory_ro); 486 487 debug_checkwx(); 488 } 489 #else 490 static __init pgprot_t pgprot_from_va(uintptr_t va) 491 { 492 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 493 return PAGE_KERNEL; 494 495 return PAGE_KERNEL_EXEC; 496 } 497 #endif /* CONFIG_STRICT_KERNEL_RWX */ 498 499 /* 500 * setup_vm() is called from head.S with MMU-off. 501 * 502 * Following requirements should be honoured for setup_vm() to work 503 * correctly: 504 * 1) It should use PC-relative addressing for accessing kernel symbols. 505 * To achieve this we always use GCC cmodel=medany. 506 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 507 * so disable compiler instrumentation when FTRACE is enabled. 508 * 509 * Currently, the above requirements are honoured by using custom CFLAGS 510 * for init.o in mm/Makefile. 511 */ 512 513 #ifndef __riscv_cmodel_medany 514 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 515 #endif 516 517 #ifdef CONFIG_XIP_KERNEL 518 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size, 519 __always_unused bool early) 520 { 521 uintptr_t va, end_va; 522 523 /* Map the flash resident part */ 524 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 525 for (va = kernel_map.virt_addr; va < end_va; va += map_size) 526 create_pgd_mapping(pgdir, va, 527 kernel_map.xiprom + (va - kernel_map.virt_addr), 528 map_size, PAGE_KERNEL_EXEC); 529 530 /* Map the data in RAM */ 531 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 532 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += map_size) 533 create_pgd_mapping(pgdir, va, 534 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 535 map_size, PAGE_KERNEL); 536 } 537 #else 538 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size, 539 bool early) 540 { 541 uintptr_t va, end_va; 542 543 end_va = kernel_map.virt_addr + kernel_map.size; 544 for (va = kernel_map.virt_addr; va < end_va; va += map_size) 545 create_pgd_mapping(pgdir, va, 546 kernel_map.phys_addr + (va - kernel_map.virt_addr), 547 map_size, 548 early ? 549 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 550 } 551 #endif 552 553 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 554 { 555 uintptr_t __maybe_unused pa; 556 uintptr_t map_size; 557 #ifndef __PAGETABLE_PMD_FOLDED 558 pmd_t fix_bmap_spmd, fix_bmap_epmd; 559 #endif 560 561 kernel_map.virt_addr = KERNEL_LINK_ADDR; 562 563 #ifdef CONFIG_XIP_KERNEL 564 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 565 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 566 567 phys_ram_base = CONFIG_PHYS_RAM_BASE; 568 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 569 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 570 571 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 572 #else 573 kernel_map.phys_addr = (uintptr_t)(&_start); 574 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 575 #endif 576 577 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr; 578 #ifdef CONFIG_64BIT 579 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 580 #endif 581 582 pfn_base = PFN_DOWN(kernel_map.phys_addr); 583 584 /* 585 * Enforce boot alignment requirements of RV32 and 586 * RV64 by only allowing PMD or PGD mappings. 587 */ 588 map_size = PMD_SIZE; 589 590 /* Sanity check alignment and size */ 591 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 592 BUG_ON((kernel_map.phys_addr % map_size) != 0); 593 594 #ifdef CONFIG_64BIT 595 /* 596 * The last 4K bytes of the addressable memory can not be mapped because 597 * of IS_ERR_VALUE macro. 598 */ 599 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 600 #endif 601 602 pt_ops.alloc_pte = alloc_pte_early; 603 pt_ops.get_pte_virt = get_pte_virt_early; 604 #ifndef __PAGETABLE_PMD_FOLDED 605 pt_ops.alloc_pmd = alloc_pmd_early; 606 pt_ops.get_pmd_virt = get_pmd_virt_early; 607 #endif 608 /* Setup early PGD for fixmap */ 609 create_pgd_mapping(early_pg_dir, FIXADDR_START, 610 (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 611 612 #ifndef __PAGETABLE_PMD_FOLDED 613 /* Setup fixmap PMD */ 614 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 615 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 616 /* Setup trampoline PGD and PMD */ 617 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 618 (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE); 619 #ifdef CONFIG_XIP_KERNEL 620 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 621 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 622 #else 623 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 624 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 625 #endif 626 #else 627 /* Setup trampoline PGD */ 628 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 629 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 630 #endif 631 632 /* 633 * Setup early PGD covering entire kernel which will allow 634 * us to reach paging_init(). We map all memory banks later 635 * in setup_vm_final() below. 636 */ 637 create_kernel_page_table(early_pg_dir, map_size, true); 638 639 #ifndef __PAGETABLE_PMD_FOLDED 640 /* Setup early PMD for DTB */ 641 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 642 (uintptr_t)early_dtb_pmd, PGDIR_SIZE, PAGE_TABLE); 643 #ifndef CONFIG_BUILTIN_DTB 644 /* Create two consecutive PMD mappings for FDT early scan */ 645 pa = dtb_pa & ~(PMD_SIZE - 1); 646 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA, 647 pa, PMD_SIZE, PAGE_KERNEL); 648 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE, 649 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 650 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1)); 651 #else /* CONFIG_BUILTIN_DTB */ 652 #ifdef CONFIG_64BIT 653 /* 654 * __va can't be used since it would return a linear mapping address 655 * whereas dtb_early_va will be used before setup_vm_final installs 656 * the linear mapping. 657 */ 658 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 659 #else 660 dtb_early_va = __va(dtb_pa); 661 #endif /* CONFIG_64BIT */ 662 #endif /* CONFIG_BUILTIN_DTB */ 663 #else 664 #ifndef CONFIG_BUILTIN_DTB 665 /* Create two consecutive PGD mappings for FDT early scan */ 666 pa = dtb_pa & ~(PGDIR_SIZE - 1); 667 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 668 pa, PGDIR_SIZE, PAGE_KERNEL); 669 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA + PGDIR_SIZE, 670 pa + PGDIR_SIZE, PGDIR_SIZE, PAGE_KERNEL); 671 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PGDIR_SIZE - 1)); 672 #else /* CONFIG_BUILTIN_DTB */ 673 #ifdef CONFIG_64BIT 674 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 675 #else 676 dtb_early_va = __va(dtb_pa); 677 #endif /* CONFIG_64BIT */ 678 #endif /* CONFIG_BUILTIN_DTB */ 679 #endif 680 dtb_early_pa = dtb_pa; 681 682 /* 683 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 684 * range can not span multiple pmds. 685 */ 686 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 687 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 688 689 #ifndef __PAGETABLE_PMD_FOLDED 690 /* 691 * Early ioremap fixmap is already created as it lies within first 2MB 692 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 693 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 694 * the user if not. 695 */ 696 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 697 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 698 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 699 WARN_ON(1); 700 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 701 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 702 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 703 fix_to_virt(FIX_BTMAP_BEGIN)); 704 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 705 fix_to_virt(FIX_BTMAP_END)); 706 707 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 708 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 709 } 710 #endif 711 } 712 713 static void __init setup_vm_final(void) 714 { 715 uintptr_t va, map_size; 716 phys_addr_t pa, start, end; 717 u64 i; 718 719 /** 720 * MMU is enabled at this point. But page table setup is not complete yet. 721 * fixmap page table alloc functions should be used at this point 722 */ 723 pt_ops.alloc_pte = alloc_pte_fixmap; 724 pt_ops.get_pte_virt = get_pte_virt_fixmap; 725 #ifndef __PAGETABLE_PMD_FOLDED 726 pt_ops.alloc_pmd = alloc_pmd_fixmap; 727 pt_ops.get_pmd_virt = get_pmd_virt_fixmap; 728 #endif 729 /* Setup swapper PGD for fixmap */ 730 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 731 __pa_symbol(fixmap_pgd_next), 732 PGDIR_SIZE, PAGE_TABLE); 733 734 /* Map all memory banks in the linear mapping */ 735 for_each_mem_range(i, &start, &end) { 736 if (start >= end) 737 break; 738 if (start <= __pa(PAGE_OFFSET) && 739 __pa(PAGE_OFFSET) < end) 740 start = __pa(PAGE_OFFSET); 741 if (end >= __pa(PAGE_OFFSET) + memory_limit) 742 end = __pa(PAGE_OFFSET) + memory_limit; 743 744 map_size = best_map_size(start, end - start); 745 for (pa = start; pa < end; pa += map_size) { 746 va = (uintptr_t)__va(pa); 747 748 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 749 pgprot_from_va(va)); 750 } 751 } 752 753 #ifdef CONFIG_64BIT 754 /* Map the kernel */ 755 create_kernel_page_table(swapper_pg_dir, PMD_SIZE, false); 756 #endif 757 758 /* Clear fixmap PTE and PMD mappings */ 759 clear_fixmap(FIX_PTE); 760 clear_fixmap(FIX_PMD); 761 762 /* Move to swapper page table */ 763 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE); 764 local_flush_tlb_all(); 765 766 /* generic page allocation functions must be used to setup page table */ 767 pt_ops.alloc_pte = alloc_pte_late; 768 pt_ops.get_pte_virt = get_pte_virt_late; 769 #ifndef __PAGETABLE_PMD_FOLDED 770 pt_ops.alloc_pmd = alloc_pmd_late; 771 pt_ops.get_pmd_virt = get_pmd_virt_late; 772 #endif 773 } 774 #else 775 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 776 { 777 dtb_early_va = (void *)dtb_pa; 778 dtb_early_pa = dtb_pa; 779 } 780 781 static inline void setup_vm_final(void) 782 { 783 } 784 #endif /* CONFIG_MMU */ 785 786 #ifdef CONFIG_KEXEC_CORE 787 /* 788 * reserve_crashkernel() - reserves memory for crash kernel 789 * 790 * This function reserves memory area given in "crashkernel=" kernel command 791 * line parameter. The memory reserved is used by dump capture kernel when 792 * primary kernel is crashing. 793 */ 794 static void __init reserve_crashkernel(void) 795 { 796 unsigned long long crash_base = 0; 797 unsigned long long crash_size = 0; 798 unsigned long search_start = memblock_start_of_DRAM(); 799 unsigned long search_end = memblock_end_of_DRAM(); 800 801 int ret = 0; 802 803 /* 804 * Don't reserve a region for a crash kernel on a crash kernel 805 * since it doesn't make much sense and we have limited memory 806 * resources. 807 */ 808 #ifdef CONFIG_CRASH_DUMP 809 if (is_kdump_kernel()) { 810 pr_info("crashkernel: ignoring reservation request\n"); 811 return; 812 } 813 #endif 814 815 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 816 &crash_size, &crash_base); 817 if (ret || !crash_size) 818 return; 819 820 crash_size = PAGE_ALIGN(crash_size); 821 822 if (crash_base == 0) { 823 /* 824 * Current riscv boot protocol requires 2MB alignment for 825 * RV64 and 4MB alignment for RV32 (hugepage size) 826 */ 827 crash_base = memblock_find_in_range(search_start, search_end, 828 crash_size, PMD_SIZE); 829 830 if (crash_base == 0) { 831 pr_warn("crashkernel: couldn't allocate %lldKB\n", 832 crash_size >> 10); 833 return; 834 } 835 } else { 836 /* User specifies base address explicitly. */ 837 if (!memblock_is_region_memory(crash_base, crash_size)) { 838 pr_warn("crashkernel: requested region is not memory\n"); 839 return; 840 } 841 842 if (memblock_is_region_reserved(crash_base, crash_size)) { 843 pr_warn("crashkernel: requested region is reserved\n"); 844 return; 845 } 846 847 848 if (!IS_ALIGNED(crash_base, PMD_SIZE)) { 849 pr_warn("crashkernel: requested region is misaligned\n"); 850 return; 851 } 852 } 853 memblock_reserve(crash_base, crash_size); 854 855 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 856 crash_base, crash_base + crash_size, crash_size >> 20); 857 858 crashk_res.start = crash_base; 859 crashk_res.end = crash_base + crash_size - 1; 860 } 861 #endif /* CONFIG_KEXEC_CORE */ 862 863 #ifdef CONFIG_CRASH_DUMP 864 /* 865 * We keep track of the ELF core header of the crashed 866 * kernel with a reserved-memory region with compatible 867 * string "linux,elfcorehdr". Here we register a callback 868 * to populate elfcorehdr_addr/size when this region is 869 * present. Note that this region will be marked as 870 * reserved once we call early_init_fdt_scan_reserved_mem() 871 * later on. 872 */ 873 static int __init elfcore_hdr_setup(struct reserved_mem *rmem) 874 { 875 elfcorehdr_addr = rmem->base; 876 elfcorehdr_size = rmem->size; 877 return 0; 878 } 879 880 RESERVEDMEM_OF_DECLARE(elfcorehdr, "linux,elfcorehdr", elfcore_hdr_setup); 881 #endif 882 883 void __init paging_init(void) 884 { 885 setup_bootmem(); 886 setup_vm_final(); 887 } 888 889 void __init misc_mem_init(void) 890 { 891 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 892 arch_numa_init(); 893 sparse_init(); 894 zone_sizes_init(); 895 #ifdef CONFIG_KEXEC_CORE 896 reserve_crashkernel(); 897 #endif 898 memblock_dump_all(); 899 } 900 901 #ifdef CONFIG_SPARSEMEM_VMEMMAP 902 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 903 struct vmem_altmap *altmap) 904 { 905 return vmemmap_populate_basepages(start, end, node, NULL); 906 } 907 #endif 908