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 24 #include <asm/fixmap.h> 25 #include <asm/tlbflush.h> 26 #include <asm/sections.h> 27 #include <asm/soc.h> 28 #include <asm/io.h> 29 #include <asm/ptdump.h> 30 #include <asm/numa.h> 31 32 #include "../kernel/head.h" 33 34 struct kernel_mapping kernel_map __ro_after_init; 35 EXPORT_SYMBOL(kernel_map); 36 #ifdef CONFIG_XIP_KERNEL 37 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map)) 38 #endif 39 40 #ifdef CONFIG_64BIT 41 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39; 42 #else 43 u64 satp_mode __ro_after_init = SATP_MODE_32; 44 #endif 45 EXPORT_SYMBOL(satp_mode); 46 47 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 48 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL); 49 EXPORT_SYMBOL(pgtable_l4_enabled); 50 EXPORT_SYMBOL(pgtable_l5_enabled); 51 52 phys_addr_t phys_ram_base __ro_after_init; 53 EXPORT_SYMBOL(phys_ram_base); 54 55 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 56 __page_aligned_bss; 57 EXPORT_SYMBOL(empty_zero_page); 58 59 extern char _start[]; 60 #define DTB_EARLY_BASE_VA PGDIR_SIZE 61 void *_dtb_early_va __initdata; 62 uintptr_t _dtb_early_pa __initdata; 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 80 #define LOG2_SZ_1K ilog2(SZ_1K) 81 #define LOG2_SZ_1M ilog2(SZ_1M) 82 #define LOG2_SZ_1G ilog2(SZ_1G) 83 #define LOG2_SZ_1T ilog2(SZ_1T) 84 85 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 86 { 87 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 88 (((t) - (b)) >> LOG2_SZ_1K)); 89 } 90 91 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 92 { 93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 94 (((t) - (b)) >> LOG2_SZ_1M)); 95 } 96 97 static inline void print_mlg(char *name, unsigned long b, unsigned long t) 98 { 99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t, 100 (((t) - (b)) >> LOG2_SZ_1G)); 101 } 102 103 #ifdef CONFIG_64BIT 104 static inline void print_mlt(char *name, unsigned long b, unsigned long t) 105 { 106 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t, 107 (((t) - (b)) >> LOG2_SZ_1T)); 108 } 109 #else 110 #define print_mlt(n, b, t) do {} while (0) 111 #endif 112 113 static inline void print_ml(char *name, unsigned long b, unsigned long t) 114 { 115 unsigned long diff = t - b; 116 117 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10) 118 print_mlt(name, b, t); 119 else if ((diff >> LOG2_SZ_1G) >= 10) 120 print_mlg(name, b, t); 121 else if ((diff >> LOG2_SZ_1M) >= 10) 122 print_mlm(name, b, t); 123 else 124 print_mlk(name, b, t); 125 } 126 127 static void __init print_vm_layout(void) 128 { 129 pr_notice("Virtual kernel memory layout:\n"); 130 print_ml("fixmap", (unsigned long)FIXADDR_START, 131 (unsigned long)FIXADDR_TOP); 132 print_ml("pci io", (unsigned long)PCI_IO_START, 133 (unsigned long)PCI_IO_END); 134 print_ml("vmemmap", (unsigned long)VMEMMAP_START, 135 (unsigned long)VMEMMAP_END); 136 print_ml("vmalloc", (unsigned long)VMALLOC_START, 137 (unsigned long)VMALLOC_END); 138 print_ml("lowmem", (unsigned long)PAGE_OFFSET, 139 (unsigned long)high_memory); 140 if (IS_ENABLED(CONFIG_64BIT)) { 141 #ifdef CONFIG_KASAN 142 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 143 #endif 144 145 print_ml("kernel", (unsigned long)KERNEL_LINK_ADDR, 146 (unsigned long)ADDRESS_SPACE_END); 147 } 148 } 149 #else 150 static void print_vm_layout(void) { } 151 #endif /* CONFIG_DEBUG_VM */ 152 153 void __init mem_init(void) 154 { 155 #ifdef CONFIG_FLATMEM 156 BUG_ON(!mem_map); 157 #endif /* CONFIG_FLATMEM */ 158 159 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE); 160 memblock_free_all(); 161 162 print_vm_layout(); 163 } 164 165 /* Limit the memory size via mem. */ 166 static phys_addr_t memory_limit; 167 168 static int __init early_mem(char *p) 169 { 170 u64 size; 171 172 if (!p) 173 return 1; 174 175 size = memparse(p, &p) & PAGE_MASK; 176 memory_limit = min_t(u64, size, memory_limit); 177 178 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 179 180 return 0; 181 } 182 early_param("mem", early_mem); 183 184 static void __init setup_bootmem(void) 185 { 186 phys_addr_t vmlinux_end = __pa_symbol(&_end); 187 phys_addr_t max_mapped_addr; 188 phys_addr_t phys_ram_end, vmlinux_start; 189 190 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 191 vmlinux_start = __pa_symbol(&_sdata); 192 else 193 vmlinux_start = __pa_symbol(&_start); 194 195 memblock_enforce_memory_limit(memory_limit); 196 197 /* 198 * Make sure we align the reservation on PMD_SIZE since we will 199 * map the kernel in the linear mapping as read-only: we do not want 200 * any allocation to happen between _end and the next pmd aligned page. 201 */ 202 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 203 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 204 /* 205 * Reserve from the start of the kernel to the end of the kernel 206 */ 207 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 208 209 phys_ram_end = memblock_end_of_DRAM(); 210 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 211 phys_ram_base = memblock_start_of_DRAM(); 212 /* 213 * memblock allocator is not aware of the fact that last 4K bytes of 214 * the addressable memory can not be mapped because of IS_ERR_VALUE 215 * macro. Make sure that last 4k bytes are not usable by memblock 216 * if end of dram is equal to maximum addressable memory. For 64-bit 217 * kernel, this problem can't happen here as the end of the virtual 218 * address space is occupied by the kernel mapping then this check must 219 * be done as soon as the kernel mapping base address is determined. 220 */ 221 if (!IS_ENABLED(CONFIG_64BIT)) { 222 max_mapped_addr = __pa(~(ulong)0); 223 if (max_mapped_addr == (phys_ram_end - 1)) 224 memblock_set_current_limit(max_mapped_addr - 4096); 225 } 226 227 min_low_pfn = PFN_UP(phys_ram_base); 228 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 229 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 230 231 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 232 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 233 234 reserve_initrd_mem(); 235 /* 236 * If DTB is built in, no need to reserve its memblock. 237 * Otherwise, do reserve it but avoid using 238 * early_init_fdt_reserve_self() since __pa() does 239 * not work for DTB pointers that are fixmap addresses 240 */ 241 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) { 242 /* 243 * In case the DTB is not located in a memory region we won't 244 * be able to locate it later on via the linear mapping and 245 * get a segfault when accessing it via __va(dtb_early_pa). 246 * To avoid this situation copy DTB to a memory region. 247 * Note that memblock_phys_alloc will also reserve DTB region. 248 */ 249 if (!memblock_is_memory(dtb_early_pa)) { 250 size_t fdt_size = fdt_totalsize(dtb_early_va); 251 phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE); 252 void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size); 253 254 memcpy(new_dtb_early_va, dtb_early_va, fdt_size); 255 early_memunmap(new_dtb_early_va, fdt_size); 256 _dtb_early_pa = new_dtb_early_pa; 257 } else 258 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 259 } 260 261 early_init_fdt_scan_reserved_mem(); 262 dma_contiguous_reserve(dma32_phys_limit); 263 if (IS_ENABLED(CONFIG_64BIT)) 264 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 265 memblock_allow_resize(); 266 } 267 268 #ifdef CONFIG_MMU 269 struct pt_alloc_ops pt_ops __initdata; 270 271 unsigned long riscv_pfn_base __ro_after_init; 272 EXPORT_SYMBOL(riscv_pfn_base); 273 274 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 275 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 276 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 277 278 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 279 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 280 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 281 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 282 283 #ifdef CONFIG_XIP_KERNEL 284 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 285 #define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base)) 286 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 287 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 288 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 289 #endif /* CONFIG_XIP_KERNEL */ 290 291 static const pgprot_t protection_map[16] = { 292 [VM_NONE] = PAGE_NONE, 293 [VM_READ] = PAGE_READ, 294 [VM_WRITE] = PAGE_COPY, 295 [VM_WRITE | VM_READ] = PAGE_COPY, 296 [VM_EXEC] = PAGE_EXEC, 297 [VM_EXEC | VM_READ] = PAGE_READ_EXEC, 298 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 299 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_READ_EXEC, 300 [VM_SHARED] = PAGE_NONE, 301 [VM_SHARED | VM_READ] = PAGE_READ, 302 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 303 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 304 [VM_SHARED | VM_EXEC] = PAGE_EXEC, 305 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC, 306 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC, 307 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC 308 }; 309 DECLARE_VM_GET_PAGE_PROT 310 311 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 312 { 313 unsigned long addr = __fix_to_virt(idx); 314 pte_t *ptep; 315 316 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 317 318 ptep = &fixmap_pte[pte_index(addr)]; 319 320 if (pgprot_val(prot)) 321 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 322 else 323 pte_clear(&init_mm, addr, ptep); 324 local_flush_tlb_page(addr); 325 } 326 327 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 328 { 329 return (pte_t *)((uintptr_t)pa); 330 } 331 332 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 333 { 334 clear_fixmap(FIX_PTE); 335 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 336 } 337 338 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 339 { 340 return (pte_t *) __va(pa); 341 } 342 343 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 344 { 345 /* 346 * We only create PMD or PGD early mappings so we 347 * should never reach here with MMU disabled. 348 */ 349 BUG(); 350 } 351 352 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 353 { 354 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 355 } 356 357 static phys_addr_t __init alloc_pte_late(uintptr_t va) 358 { 359 unsigned long vaddr; 360 361 vaddr = __get_free_page(GFP_KERNEL); 362 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr))); 363 364 return __pa(vaddr); 365 } 366 367 static void __init create_pte_mapping(pte_t *ptep, 368 uintptr_t va, phys_addr_t pa, 369 phys_addr_t sz, pgprot_t prot) 370 { 371 uintptr_t pte_idx = pte_index(va); 372 373 BUG_ON(sz != PAGE_SIZE); 374 375 if (pte_none(ptep[pte_idx])) 376 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 377 } 378 379 #ifndef __PAGETABLE_PMD_FOLDED 380 381 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 382 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 383 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 384 385 #ifdef CONFIG_XIP_KERNEL 386 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 387 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 388 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 389 #endif /* CONFIG_XIP_KERNEL */ 390 391 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss; 392 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss; 393 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); 394 395 #ifdef CONFIG_XIP_KERNEL 396 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d)) 397 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d)) 398 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d)) 399 #endif /* CONFIG_XIP_KERNEL */ 400 401 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 402 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 403 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 404 405 #ifdef CONFIG_XIP_KERNEL 406 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 407 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 408 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 409 #endif /* CONFIG_XIP_KERNEL */ 410 411 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 412 { 413 /* Before MMU is enabled */ 414 return (pmd_t *)((uintptr_t)pa); 415 } 416 417 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 418 { 419 clear_fixmap(FIX_PMD); 420 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 421 } 422 423 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 424 { 425 return (pmd_t *) __va(pa); 426 } 427 428 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 429 { 430 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 431 432 return (uintptr_t)early_pmd; 433 } 434 435 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 436 { 437 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 438 } 439 440 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 441 { 442 unsigned long vaddr; 443 444 vaddr = __get_free_page(GFP_KERNEL); 445 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr))); 446 447 return __pa(vaddr); 448 } 449 450 static void __init create_pmd_mapping(pmd_t *pmdp, 451 uintptr_t va, phys_addr_t pa, 452 phys_addr_t sz, pgprot_t prot) 453 { 454 pte_t *ptep; 455 phys_addr_t pte_phys; 456 uintptr_t pmd_idx = pmd_index(va); 457 458 if (sz == PMD_SIZE) { 459 if (pmd_none(pmdp[pmd_idx])) 460 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 461 return; 462 } 463 464 if (pmd_none(pmdp[pmd_idx])) { 465 pte_phys = pt_ops.alloc_pte(va); 466 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 467 ptep = pt_ops.get_pte_virt(pte_phys); 468 memset(ptep, 0, PAGE_SIZE); 469 } else { 470 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 471 ptep = pt_ops.get_pte_virt(pte_phys); 472 } 473 474 create_pte_mapping(ptep, va, pa, sz, prot); 475 } 476 477 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 478 { 479 return (pud_t *)((uintptr_t)pa); 480 } 481 482 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 483 { 484 clear_fixmap(FIX_PUD); 485 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 486 } 487 488 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 489 { 490 return (pud_t *)__va(pa); 491 } 492 493 static phys_addr_t __init alloc_pud_early(uintptr_t va) 494 { 495 /* Only one PUD is available for early mapping */ 496 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 497 498 return (uintptr_t)early_pud; 499 } 500 501 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 502 { 503 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 504 } 505 506 static phys_addr_t alloc_pud_late(uintptr_t va) 507 { 508 unsigned long vaddr; 509 510 vaddr = __get_free_page(GFP_KERNEL); 511 BUG_ON(!vaddr); 512 return __pa(vaddr); 513 } 514 515 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa) 516 { 517 return (p4d_t *)((uintptr_t)pa); 518 } 519 520 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa) 521 { 522 clear_fixmap(FIX_P4D); 523 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa); 524 } 525 526 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa) 527 { 528 return (p4d_t *)__va(pa); 529 } 530 531 static phys_addr_t __init alloc_p4d_early(uintptr_t va) 532 { 533 /* Only one P4D is available for early mapping */ 534 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 535 536 return (uintptr_t)early_p4d; 537 } 538 539 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va) 540 { 541 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 542 } 543 544 static phys_addr_t alloc_p4d_late(uintptr_t va) 545 { 546 unsigned long vaddr; 547 548 vaddr = __get_free_page(GFP_KERNEL); 549 BUG_ON(!vaddr); 550 return __pa(vaddr); 551 } 552 553 static void __init create_pud_mapping(pud_t *pudp, 554 uintptr_t va, phys_addr_t pa, 555 phys_addr_t sz, pgprot_t prot) 556 { 557 pmd_t *nextp; 558 phys_addr_t next_phys; 559 uintptr_t pud_index = pud_index(va); 560 561 if (sz == PUD_SIZE) { 562 if (pud_val(pudp[pud_index]) == 0) 563 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 564 return; 565 } 566 567 if (pud_val(pudp[pud_index]) == 0) { 568 next_phys = pt_ops.alloc_pmd(va); 569 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 570 nextp = pt_ops.get_pmd_virt(next_phys); 571 memset(nextp, 0, PAGE_SIZE); 572 } else { 573 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 574 nextp = pt_ops.get_pmd_virt(next_phys); 575 } 576 577 create_pmd_mapping(nextp, va, pa, sz, prot); 578 } 579 580 static void __init create_p4d_mapping(p4d_t *p4dp, 581 uintptr_t va, phys_addr_t pa, 582 phys_addr_t sz, pgprot_t prot) 583 { 584 pud_t *nextp; 585 phys_addr_t next_phys; 586 uintptr_t p4d_index = p4d_index(va); 587 588 if (sz == P4D_SIZE) { 589 if (p4d_val(p4dp[p4d_index]) == 0) 590 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot); 591 return; 592 } 593 594 if (p4d_val(p4dp[p4d_index]) == 0) { 595 next_phys = pt_ops.alloc_pud(va); 596 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE); 597 nextp = pt_ops.get_pud_virt(next_phys); 598 memset(nextp, 0, PAGE_SIZE); 599 } else { 600 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index])); 601 nextp = pt_ops.get_pud_virt(next_phys); 602 } 603 604 create_pud_mapping(nextp, va, pa, sz, prot); 605 } 606 607 #define pgd_next_t p4d_t 608 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \ 609 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \ 610 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))) 611 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \ 612 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \ 613 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa))) 614 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 615 (pgtable_l5_enabled ? \ 616 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \ 617 (pgtable_l4_enabled ? \ 618 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \ 619 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))) 620 #define fixmap_pgd_next (pgtable_l5_enabled ? \ 621 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \ 622 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)) 623 #define trampoline_pgd_next (pgtable_l5_enabled ? \ 624 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \ 625 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)) 626 #define early_dtb_pgd_next (pgtable_l5_enabled ? \ 627 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \ 628 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd)) 629 #else 630 #define pgd_next_t pte_t 631 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 632 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 633 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 634 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 635 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 636 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd) 637 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 638 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 639 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0) 640 #endif /* __PAGETABLE_PMD_FOLDED */ 641 642 void __init create_pgd_mapping(pgd_t *pgdp, 643 uintptr_t va, phys_addr_t pa, 644 phys_addr_t sz, pgprot_t prot) 645 { 646 pgd_next_t *nextp; 647 phys_addr_t next_phys; 648 uintptr_t pgd_idx = pgd_index(va); 649 650 if (sz == PGDIR_SIZE) { 651 if (pgd_val(pgdp[pgd_idx]) == 0) 652 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 653 return; 654 } 655 656 if (pgd_val(pgdp[pgd_idx]) == 0) { 657 next_phys = alloc_pgd_next(va); 658 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 659 nextp = get_pgd_next_virt(next_phys); 660 memset(nextp, 0, PAGE_SIZE); 661 } else { 662 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 663 nextp = get_pgd_next_virt(next_phys); 664 } 665 666 create_pgd_next_mapping(nextp, va, pa, sz, prot); 667 } 668 669 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size) 670 { 671 /* Upgrade to PMD_SIZE mappings whenever possible */ 672 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1))) 673 return PAGE_SIZE; 674 675 return PMD_SIZE; 676 } 677 678 #ifdef CONFIG_XIP_KERNEL 679 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base)) 680 extern char _xiprom[], _exiprom[], __data_loc; 681 682 /* called from head.S with MMU off */ 683 asmlinkage void __init __copy_data(void) 684 { 685 void *from = (void *)(&__data_loc); 686 void *to = (void *)CONFIG_PHYS_RAM_BASE; 687 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 688 689 memcpy(to, from, sz); 690 } 691 #endif 692 693 #ifdef CONFIG_STRICT_KERNEL_RWX 694 static __init pgprot_t pgprot_from_va(uintptr_t va) 695 { 696 if (is_va_kernel_text(va)) 697 return PAGE_KERNEL_READ_EXEC; 698 699 /* 700 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 701 * we must protect its linear mapping alias from being executed and 702 * written. 703 * And rodata section is marked readonly in mark_rodata_ro. 704 */ 705 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 706 return PAGE_KERNEL_READ; 707 708 return PAGE_KERNEL; 709 } 710 711 void mark_rodata_ro(void) 712 { 713 set_kernel_memory(__start_rodata, _data, set_memory_ro); 714 if (IS_ENABLED(CONFIG_64BIT)) 715 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 716 set_memory_ro); 717 718 debug_checkwx(); 719 } 720 #else 721 static __init pgprot_t pgprot_from_va(uintptr_t va) 722 { 723 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 724 return PAGE_KERNEL; 725 726 return PAGE_KERNEL_EXEC; 727 } 728 #endif /* CONFIG_STRICT_KERNEL_RWX */ 729 730 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 731 static void __init disable_pgtable_l5(void) 732 { 733 pgtable_l5_enabled = false; 734 kernel_map.page_offset = PAGE_OFFSET_L4; 735 satp_mode = SATP_MODE_48; 736 } 737 738 static void __init disable_pgtable_l4(void) 739 { 740 pgtable_l4_enabled = false; 741 kernel_map.page_offset = PAGE_OFFSET_L3; 742 satp_mode = SATP_MODE_39; 743 } 744 745 /* 746 * There is a simple way to determine if 4-level is supported by the 747 * underlying hardware: establish 1:1 mapping in 4-level page table mode 748 * then read SATP to see if the configuration was taken into account 749 * meaning sv48 is supported. 750 */ 751 static __init void set_satp_mode(void) 752 { 753 u64 identity_satp, hw_satp; 754 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 755 bool check_l4 = false; 756 757 create_p4d_mapping(early_p4d, 758 set_satp_mode_pmd, (uintptr_t)early_pud, 759 P4D_SIZE, PAGE_TABLE); 760 create_pud_mapping(early_pud, 761 set_satp_mode_pmd, (uintptr_t)early_pmd, 762 PUD_SIZE, PAGE_TABLE); 763 /* Handle the case where set_satp_mode straddles 2 PMDs */ 764 create_pmd_mapping(early_pmd, 765 set_satp_mode_pmd, set_satp_mode_pmd, 766 PMD_SIZE, PAGE_KERNEL_EXEC); 767 create_pmd_mapping(early_pmd, 768 set_satp_mode_pmd + PMD_SIZE, 769 set_satp_mode_pmd + PMD_SIZE, 770 PMD_SIZE, PAGE_KERNEL_EXEC); 771 retry: 772 create_pgd_mapping(early_pg_dir, 773 set_satp_mode_pmd, 774 check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d, 775 PGDIR_SIZE, PAGE_TABLE); 776 777 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 778 779 local_flush_tlb_all(); 780 csr_write(CSR_SATP, identity_satp); 781 hw_satp = csr_swap(CSR_SATP, 0ULL); 782 local_flush_tlb_all(); 783 784 if (hw_satp != identity_satp) { 785 if (!check_l4) { 786 disable_pgtable_l5(); 787 check_l4 = true; 788 memset(early_pg_dir, 0, PAGE_SIZE); 789 goto retry; 790 } 791 disable_pgtable_l4(); 792 } 793 794 memset(early_pg_dir, 0, PAGE_SIZE); 795 memset(early_p4d, 0, PAGE_SIZE); 796 memset(early_pud, 0, PAGE_SIZE); 797 memset(early_pmd, 0, PAGE_SIZE); 798 } 799 #endif 800 801 /* 802 * setup_vm() is called from head.S with MMU-off. 803 * 804 * Following requirements should be honoured for setup_vm() to work 805 * correctly: 806 * 1) It should use PC-relative addressing for accessing kernel symbols. 807 * To achieve this we always use GCC cmodel=medany. 808 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 809 * so disable compiler instrumentation when FTRACE is enabled. 810 * 811 * Currently, the above requirements are honoured by using custom CFLAGS 812 * for init.o in mm/Makefile. 813 */ 814 815 #ifndef __riscv_cmodel_medany 816 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 817 #endif 818 819 #ifdef CONFIG_XIP_KERNEL 820 static void __init create_kernel_page_table(pgd_t *pgdir, 821 __always_unused bool early) 822 { 823 uintptr_t va, end_va; 824 825 /* Map the flash resident part */ 826 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 827 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 828 create_pgd_mapping(pgdir, va, 829 kernel_map.xiprom + (va - kernel_map.virt_addr), 830 PMD_SIZE, PAGE_KERNEL_EXEC); 831 832 /* Map the data in RAM */ 833 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 834 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 835 create_pgd_mapping(pgdir, va, 836 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 837 PMD_SIZE, PAGE_KERNEL); 838 } 839 #else 840 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 841 { 842 uintptr_t va, end_va; 843 844 end_va = kernel_map.virt_addr + kernel_map.size; 845 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 846 create_pgd_mapping(pgdir, va, 847 kernel_map.phys_addr + (va - kernel_map.virt_addr), 848 PMD_SIZE, 849 early ? 850 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 851 } 852 #endif 853 854 /* 855 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 856 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 857 * entry. 858 */ 859 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa) 860 { 861 #ifndef CONFIG_BUILTIN_DTB 862 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 863 864 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 865 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa, 866 PGDIR_SIZE, 867 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL); 868 869 if (pgtable_l5_enabled) 870 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA, 871 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE); 872 873 if (pgtable_l4_enabled) 874 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA, 875 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE); 876 877 if (IS_ENABLED(CONFIG_64BIT)) { 878 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA, 879 pa, PMD_SIZE, PAGE_KERNEL); 880 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE, 881 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 882 } 883 884 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1)); 885 #else 886 /* 887 * For 64-bit kernel, __va can't be used since it would return a linear 888 * mapping address whereas dtb_early_va will be used before 889 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 890 * kernel is mapped in the linear mapping, that makes no difference. 891 */ 892 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 893 #endif 894 895 dtb_early_pa = dtb_pa; 896 } 897 898 /* 899 * MMU is not enabled, the page tables are allocated directly using 900 * early_pmd/pud/p4d and the address returned is the physical one. 901 */ 902 static void __init pt_ops_set_early(void) 903 { 904 pt_ops.alloc_pte = alloc_pte_early; 905 pt_ops.get_pte_virt = get_pte_virt_early; 906 #ifndef __PAGETABLE_PMD_FOLDED 907 pt_ops.alloc_pmd = alloc_pmd_early; 908 pt_ops.get_pmd_virt = get_pmd_virt_early; 909 pt_ops.alloc_pud = alloc_pud_early; 910 pt_ops.get_pud_virt = get_pud_virt_early; 911 pt_ops.alloc_p4d = alloc_p4d_early; 912 pt_ops.get_p4d_virt = get_p4d_virt_early; 913 #endif 914 } 915 916 /* 917 * MMU is enabled but page table setup is not complete yet. 918 * fixmap page table alloc functions must be used as a means to temporarily 919 * map the allocated physical pages since the linear mapping does not exist yet. 920 * 921 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 922 * but it will be used as described above. 923 */ 924 static void __init pt_ops_set_fixmap(void) 925 { 926 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap); 927 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap); 928 #ifndef __PAGETABLE_PMD_FOLDED 929 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap); 930 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap); 931 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap); 932 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap); 933 pt_ops.alloc_p4d = kernel_mapping_pa_to_va((uintptr_t)alloc_p4d_fixmap); 934 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va((uintptr_t)get_p4d_virt_fixmap); 935 #endif 936 } 937 938 /* 939 * MMU is enabled and page table setup is complete, so from now, we can use 940 * generic page allocation functions to setup page table. 941 */ 942 static void __init pt_ops_set_late(void) 943 { 944 pt_ops.alloc_pte = alloc_pte_late; 945 pt_ops.get_pte_virt = get_pte_virt_late; 946 #ifndef __PAGETABLE_PMD_FOLDED 947 pt_ops.alloc_pmd = alloc_pmd_late; 948 pt_ops.get_pmd_virt = get_pmd_virt_late; 949 pt_ops.alloc_pud = alloc_pud_late; 950 pt_ops.get_pud_virt = get_pud_virt_late; 951 pt_ops.alloc_p4d = alloc_p4d_late; 952 pt_ops.get_p4d_virt = get_p4d_virt_late; 953 #endif 954 } 955 956 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 957 { 958 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 959 960 kernel_map.virt_addr = KERNEL_LINK_ADDR; 961 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 962 963 #ifdef CONFIG_XIP_KERNEL 964 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 965 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 966 967 phys_ram_base = CONFIG_PHYS_RAM_BASE; 968 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 969 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 970 971 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 972 #else 973 kernel_map.phys_addr = (uintptr_t)(&_start); 974 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 975 #endif 976 977 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 978 set_satp_mode(); 979 #endif 980 981 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr; 982 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 983 984 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr); 985 986 /* 987 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 988 * kernel, whereas for 64-bit kernel, the end of the virtual address 989 * space is occupied by the modules/BPF/kernel mappings which reduces 990 * the available size of the linear mapping. 991 */ 992 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 993 994 /* Sanity check alignment and size */ 995 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 996 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 997 998 #ifdef CONFIG_64BIT 999 /* 1000 * The last 4K bytes of the addressable memory can not be mapped because 1001 * of IS_ERR_VALUE macro. 1002 */ 1003 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 1004 #endif 1005 1006 apply_early_boot_alternatives(); 1007 pt_ops_set_early(); 1008 1009 /* Setup early PGD for fixmap */ 1010 create_pgd_mapping(early_pg_dir, FIXADDR_START, 1011 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1012 1013 #ifndef __PAGETABLE_PMD_FOLDED 1014 /* Setup fixmap P4D and PUD */ 1015 if (pgtable_l5_enabled) 1016 create_p4d_mapping(fixmap_p4d, FIXADDR_START, 1017 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE); 1018 /* Setup fixmap PUD and PMD */ 1019 if (pgtable_l4_enabled) 1020 create_pud_mapping(fixmap_pud, FIXADDR_START, 1021 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 1022 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 1023 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 1024 /* Setup trampoline PGD and PMD */ 1025 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1026 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 1027 if (pgtable_l5_enabled) 1028 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr, 1029 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE); 1030 if (pgtable_l4_enabled) 1031 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 1032 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 1033 #ifdef CONFIG_XIP_KERNEL 1034 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1035 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 1036 #else 1037 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 1038 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 1039 #endif 1040 #else 1041 /* Setup trampoline PGD */ 1042 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 1043 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 1044 #endif 1045 1046 /* 1047 * Setup early PGD covering entire kernel which will allow 1048 * us to reach paging_init(). We map all memory banks later 1049 * in setup_vm_final() below. 1050 */ 1051 create_kernel_page_table(early_pg_dir, true); 1052 1053 /* Setup early mapping for FDT early scan */ 1054 create_fdt_early_page_table(early_pg_dir, dtb_pa); 1055 1056 /* 1057 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 1058 * range can not span multiple pmds. 1059 */ 1060 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 1061 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 1062 1063 #ifndef __PAGETABLE_PMD_FOLDED 1064 /* 1065 * Early ioremap fixmap is already created as it lies within first 2MB 1066 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 1067 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 1068 * the user if not. 1069 */ 1070 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 1071 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 1072 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 1073 WARN_ON(1); 1074 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 1075 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 1076 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 1077 fix_to_virt(FIX_BTMAP_BEGIN)); 1078 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 1079 fix_to_virt(FIX_BTMAP_END)); 1080 1081 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 1082 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 1083 } 1084 #endif 1085 1086 pt_ops_set_fixmap(); 1087 } 1088 1089 static void __init setup_vm_final(void) 1090 { 1091 uintptr_t va, map_size; 1092 phys_addr_t pa, start, end; 1093 u64 i; 1094 1095 /* Setup swapper PGD for fixmap */ 1096 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 1097 __pa_symbol(fixmap_pgd_next), 1098 PGDIR_SIZE, PAGE_TABLE); 1099 1100 /* Map all memory banks in the linear mapping */ 1101 for_each_mem_range(i, &start, &end) { 1102 if (start >= end) 1103 break; 1104 if (start <= __pa(PAGE_OFFSET) && 1105 __pa(PAGE_OFFSET) < end) 1106 start = __pa(PAGE_OFFSET); 1107 if (end >= __pa(PAGE_OFFSET) + memory_limit) 1108 end = __pa(PAGE_OFFSET) + memory_limit; 1109 1110 map_size = best_map_size(start, end - start); 1111 for (pa = start; pa < end; pa += map_size) { 1112 va = (uintptr_t)__va(pa); 1113 1114 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 1115 pgprot_from_va(va)); 1116 } 1117 } 1118 1119 /* Map the kernel */ 1120 if (IS_ENABLED(CONFIG_64BIT)) 1121 create_kernel_page_table(swapper_pg_dir, false); 1122 1123 #ifdef CONFIG_KASAN 1124 kasan_swapper_init(); 1125 #endif 1126 1127 /* Clear fixmap PTE and PMD mappings */ 1128 clear_fixmap(FIX_PTE); 1129 clear_fixmap(FIX_PMD); 1130 clear_fixmap(FIX_PUD); 1131 clear_fixmap(FIX_P4D); 1132 1133 /* Move to swapper page table */ 1134 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 1135 local_flush_tlb_all(); 1136 1137 pt_ops_set_late(); 1138 } 1139 #else 1140 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 1141 { 1142 dtb_early_va = (void *)dtb_pa; 1143 dtb_early_pa = dtb_pa; 1144 } 1145 1146 static inline void setup_vm_final(void) 1147 { 1148 } 1149 #endif /* CONFIG_MMU */ 1150 1151 /* 1152 * reserve_crashkernel() - reserves memory for crash kernel 1153 * 1154 * This function reserves memory area given in "crashkernel=" kernel command 1155 * line parameter. The memory reserved is used by dump capture kernel when 1156 * primary kernel is crashing. 1157 */ 1158 static void __init reserve_crashkernel(void) 1159 { 1160 unsigned long long crash_base = 0; 1161 unsigned long long crash_size = 0; 1162 unsigned long search_start = memblock_start_of_DRAM(); 1163 unsigned long search_end = memblock_end_of_DRAM(); 1164 1165 int ret = 0; 1166 1167 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 1168 return; 1169 /* 1170 * Don't reserve a region for a crash kernel on a crash kernel 1171 * since it doesn't make much sense and we have limited memory 1172 * resources. 1173 */ 1174 if (is_kdump_kernel()) { 1175 pr_info("crashkernel: ignoring reservation request\n"); 1176 return; 1177 } 1178 1179 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 1180 &crash_size, &crash_base); 1181 if (ret || !crash_size) 1182 return; 1183 1184 crash_size = PAGE_ALIGN(crash_size); 1185 1186 if (crash_base) { 1187 search_start = crash_base; 1188 search_end = crash_base + crash_size; 1189 } 1190 1191 /* 1192 * Current riscv boot protocol requires 2MB alignment for 1193 * RV64 and 4MB alignment for RV32 (hugepage size) 1194 * 1195 * Try to alloc from 32bit addressible physical memory so that 1196 * swiotlb can work on the crash kernel. 1197 */ 1198 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1199 search_start, 1200 min(search_end, (unsigned long) SZ_4G)); 1201 if (crash_base == 0) { 1202 /* Try again without restricting region to 32bit addressible memory */ 1203 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1204 search_start, search_end); 1205 if (crash_base == 0) { 1206 pr_warn("crashkernel: couldn't allocate %lldKB\n", 1207 crash_size >> 10); 1208 return; 1209 } 1210 } 1211 1212 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 1213 crash_base, crash_base + crash_size, crash_size >> 20); 1214 1215 crashk_res.start = crash_base; 1216 crashk_res.end = crash_base + crash_size - 1; 1217 } 1218 1219 void __init paging_init(void) 1220 { 1221 setup_bootmem(); 1222 setup_vm_final(); 1223 } 1224 1225 void __init misc_mem_init(void) 1226 { 1227 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1228 arch_numa_init(); 1229 sparse_init(); 1230 zone_sizes_init(); 1231 reserve_crashkernel(); 1232 memblock_dump_all(); 1233 } 1234 1235 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1236 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1237 struct vmem_altmap *altmap) 1238 { 1239 return vmemmap_populate_basepages(start, end, node, NULL); 1240 } 1241 #endif 1242