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 = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_48 : SATP_MODE_39; 42 #else 43 u64 satp_mode = 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 EXPORT_SYMBOL(pgtable_l4_enabled); 49 50 phys_addr_t phys_ram_base __ro_after_init; 51 EXPORT_SYMBOL(phys_ram_base); 52 53 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 54 __page_aligned_bss; 55 EXPORT_SYMBOL(empty_zero_page); 56 57 extern char _start[]; 58 #define DTB_EARLY_BASE_VA PGDIR_SIZE 59 void *_dtb_early_va __initdata; 60 uintptr_t _dtb_early_pa __initdata; 61 62 static phys_addr_t dma32_phys_limit __initdata; 63 64 static void __init zone_sizes_init(void) 65 { 66 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 67 68 #ifdef CONFIG_ZONE_DMA32 69 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 70 #endif 71 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 72 73 free_area_init(max_zone_pfns); 74 } 75 76 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 77 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 78 { 79 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 80 (((t) - (b)) >> 10)); 81 } 82 83 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 84 { 85 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 86 (((t) - (b)) >> 20)); 87 } 88 89 static void __init print_vm_layout(void) 90 { 91 pr_notice("Virtual kernel memory layout:\n"); 92 print_mlk("fixmap", (unsigned long)FIXADDR_START, 93 (unsigned long)FIXADDR_TOP); 94 print_mlm("pci io", (unsigned long)PCI_IO_START, 95 (unsigned long)PCI_IO_END); 96 print_mlm("vmemmap", (unsigned long)VMEMMAP_START, 97 (unsigned long)VMEMMAP_END); 98 print_mlm("vmalloc", (unsigned long)VMALLOC_START, 99 (unsigned long)VMALLOC_END); 100 print_mlm("lowmem", (unsigned long)PAGE_OFFSET, 101 (unsigned long)high_memory); 102 if (IS_ENABLED(CONFIG_64BIT)) { 103 #ifdef CONFIG_KASAN 104 print_mlm("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END); 105 #endif 106 107 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR, 108 (unsigned long)ADDRESS_SPACE_END); 109 } 110 } 111 #else 112 static void print_vm_layout(void) { } 113 #endif /* CONFIG_DEBUG_VM */ 114 115 void __init mem_init(void) 116 { 117 #ifdef CONFIG_FLATMEM 118 BUG_ON(!mem_map); 119 #endif /* CONFIG_FLATMEM */ 120 121 #ifdef CONFIG_SWIOTLB 122 if (swiotlb_force == SWIOTLB_FORCE || 123 max_pfn > PFN_DOWN(dma32_phys_limit)) 124 swiotlb_init(1); 125 else 126 swiotlb_force = SWIOTLB_NO_FORCE; 127 #endif 128 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 129 memblock_free_all(); 130 131 print_vm_layout(); 132 } 133 134 /* Limit the memory size via mem. */ 135 static phys_addr_t memory_limit; 136 137 static int __init early_mem(char *p) 138 { 139 u64 size; 140 141 if (!p) 142 return 1; 143 144 size = memparse(p, &p) & PAGE_MASK; 145 memory_limit = min_t(u64, size, memory_limit); 146 147 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20); 148 149 return 0; 150 } 151 early_param("mem", early_mem); 152 153 static void __init setup_bootmem(void) 154 { 155 phys_addr_t vmlinux_end = __pa_symbol(&_end); 156 phys_addr_t max_mapped_addr; 157 phys_addr_t phys_ram_end, vmlinux_start; 158 159 if (IS_ENABLED(CONFIG_XIP_KERNEL)) 160 vmlinux_start = __pa_symbol(&_sdata); 161 else 162 vmlinux_start = __pa_symbol(&_start); 163 164 memblock_enforce_memory_limit(memory_limit); 165 166 /* 167 * Make sure we align the reservation on PMD_SIZE since we will 168 * map the kernel in the linear mapping as read-only: we do not want 169 * any allocation to happen between _end and the next pmd aligned page. 170 */ 171 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) 172 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 173 /* 174 * Reserve from the start of the kernel to the end of the kernel 175 */ 176 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 177 178 phys_ram_end = memblock_end_of_DRAM(); 179 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) 180 phys_ram_base = memblock_start_of_DRAM(); 181 /* 182 * memblock allocator is not aware of the fact that last 4K bytes of 183 * the addressable memory can not be mapped because of IS_ERR_VALUE 184 * macro. Make sure that last 4k bytes are not usable by memblock 185 * if end of dram is equal to maximum addressable memory. For 64-bit 186 * kernel, this problem can't happen here as the end of the virtual 187 * address space is occupied by the kernel mapping then this check must 188 * be done as soon as the kernel mapping base address is determined. 189 */ 190 if (!IS_ENABLED(CONFIG_64BIT)) { 191 max_mapped_addr = __pa(~(ulong)0); 192 if (max_mapped_addr == (phys_ram_end - 1)) 193 memblock_set_current_limit(max_mapped_addr - 4096); 194 } 195 196 min_low_pfn = PFN_UP(phys_ram_base); 197 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end); 198 199 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 200 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 201 202 reserve_initrd_mem(); 203 /* 204 * If DTB is built in, no need to reserve its memblock. 205 * Otherwise, do reserve it but avoid using 206 * early_init_fdt_reserve_self() since __pa() does 207 * not work for DTB pointers that are fixmap addresses 208 */ 209 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 210 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 211 212 early_init_fdt_scan_reserved_mem(); 213 dma_contiguous_reserve(dma32_phys_limit); 214 if (IS_ENABLED(CONFIG_64BIT)) 215 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 216 memblock_allow_resize(); 217 } 218 219 #ifdef CONFIG_MMU 220 struct pt_alloc_ops pt_ops __initdata; 221 222 unsigned long riscv_pfn_base __ro_after_init; 223 EXPORT_SYMBOL(riscv_pfn_base); 224 225 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 226 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 227 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 228 229 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 230 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 231 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 232 233 #ifdef CONFIG_XIP_KERNEL 234 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops)) 235 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 236 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 237 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 238 #endif /* CONFIG_XIP_KERNEL */ 239 240 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 241 { 242 unsigned long addr = __fix_to_virt(idx); 243 pte_t *ptep; 244 245 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 246 247 ptep = &fixmap_pte[pte_index(addr)]; 248 249 if (pgprot_val(prot)) 250 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 251 else 252 pte_clear(&init_mm, addr, ptep); 253 local_flush_tlb_page(addr); 254 } 255 256 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 257 { 258 return (pte_t *)((uintptr_t)pa); 259 } 260 261 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 262 { 263 clear_fixmap(FIX_PTE); 264 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 265 } 266 267 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa) 268 { 269 return (pte_t *) __va(pa); 270 } 271 272 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 273 { 274 /* 275 * We only create PMD or PGD early mappings so we 276 * should never reach here with MMU disabled. 277 */ 278 BUG(); 279 } 280 281 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 282 { 283 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 284 } 285 286 static phys_addr_t __init alloc_pte_late(uintptr_t va) 287 { 288 unsigned long vaddr; 289 290 vaddr = __get_free_page(GFP_KERNEL); 291 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr))); 292 293 return __pa(vaddr); 294 } 295 296 static void __init create_pte_mapping(pte_t *ptep, 297 uintptr_t va, phys_addr_t pa, 298 phys_addr_t sz, pgprot_t prot) 299 { 300 uintptr_t pte_idx = pte_index(va); 301 302 BUG_ON(sz != PAGE_SIZE); 303 304 if (pte_none(ptep[pte_idx])) 305 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 306 } 307 308 #ifndef __PAGETABLE_PMD_FOLDED 309 310 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 311 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 312 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 313 314 #ifdef CONFIG_XIP_KERNEL 315 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 316 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 317 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 318 #endif /* CONFIG_XIP_KERNEL */ 319 320 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss; 321 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss; 322 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE); 323 324 #ifdef CONFIG_XIP_KERNEL 325 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud)) 326 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud)) 327 #define early_pud ((pud_t *)XIP_FIXUP(early_pud)) 328 #endif /* CONFIG_XIP_KERNEL */ 329 330 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 331 { 332 /* Before MMU is enabled */ 333 return (pmd_t *)((uintptr_t)pa); 334 } 335 336 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 337 { 338 clear_fixmap(FIX_PMD); 339 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 340 } 341 342 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa) 343 { 344 return (pmd_t *) __va(pa); 345 } 346 347 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 348 { 349 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT); 350 351 return (uintptr_t)early_pmd; 352 } 353 354 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 355 { 356 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 357 } 358 359 static phys_addr_t __init alloc_pmd_late(uintptr_t va) 360 { 361 unsigned long vaddr; 362 363 vaddr = __get_free_page(GFP_KERNEL); 364 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr))); 365 366 return __pa(vaddr); 367 } 368 369 static void __init create_pmd_mapping(pmd_t *pmdp, 370 uintptr_t va, phys_addr_t pa, 371 phys_addr_t sz, pgprot_t prot) 372 { 373 pte_t *ptep; 374 phys_addr_t pte_phys; 375 uintptr_t pmd_idx = pmd_index(va); 376 377 if (sz == PMD_SIZE) { 378 if (pmd_none(pmdp[pmd_idx])) 379 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 380 return; 381 } 382 383 if (pmd_none(pmdp[pmd_idx])) { 384 pte_phys = pt_ops.alloc_pte(va); 385 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 386 ptep = pt_ops.get_pte_virt(pte_phys); 387 memset(ptep, 0, PAGE_SIZE); 388 } else { 389 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 390 ptep = pt_ops.get_pte_virt(pte_phys); 391 } 392 393 create_pte_mapping(ptep, va, pa, sz, prot); 394 } 395 396 static pud_t *__init get_pud_virt_early(phys_addr_t pa) 397 { 398 return (pud_t *)((uintptr_t)pa); 399 } 400 401 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa) 402 { 403 clear_fixmap(FIX_PUD); 404 return (pud_t *)set_fixmap_offset(FIX_PUD, pa); 405 } 406 407 static pud_t *__init get_pud_virt_late(phys_addr_t pa) 408 { 409 return (pud_t *)__va(pa); 410 } 411 412 static phys_addr_t __init alloc_pud_early(uintptr_t va) 413 { 414 /* Only one PUD is available for early mapping */ 415 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT); 416 417 return (uintptr_t)early_pud; 418 } 419 420 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va) 421 { 422 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 423 } 424 425 static phys_addr_t alloc_pud_late(uintptr_t va) 426 { 427 unsigned long vaddr; 428 429 vaddr = __get_free_page(GFP_KERNEL); 430 BUG_ON(!vaddr); 431 return __pa(vaddr); 432 } 433 434 static void __init create_pud_mapping(pud_t *pudp, 435 uintptr_t va, phys_addr_t pa, 436 phys_addr_t sz, pgprot_t prot) 437 { 438 pmd_t *nextp; 439 phys_addr_t next_phys; 440 uintptr_t pud_index = pud_index(va); 441 442 if (sz == PUD_SIZE) { 443 if (pud_val(pudp[pud_index]) == 0) 444 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot); 445 return; 446 } 447 448 if (pud_val(pudp[pud_index]) == 0) { 449 next_phys = pt_ops.alloc_pmd(va); 450 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE); 451 nextp = pt_ops.get_pmd_virt(next_phys); 452 memset(nextp, 0, PAGE_SIZE); 453 } else { 454 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index])); 455 nextp = pt_ops.get_pmd_virt(next_phys); 456 } 457 458 create_pmd_mapping(nextp, va, pa, sz, prot); 459 } 460 461 #define pgd_next_t pud_t 462 #define alloc_pgd_next(__va) (pgtable_l4_enabled ? \ 463 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)) 464 #define get_pgd_next_virt(__pa) (pgtable_l4_enabled ? \ 465 pt_ops.get_pud_virt(__pa) : (pgd_next_t *)pt_ops.get_pmd_virt(__pa)) 466 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 467 (pgtable_l4_enabled ? \ 468 create_pud_mapping(__nextp, __va, __pa, __sz, __prot) : \ 469 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)) 470 #define fixmap_pgd_next (pgtable_l4_enabled ? \ 471 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd) 472 #define trampoline_pgd_next (pgtable_l4_enabled ? \ 473 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd) 474 #define early_dtb_pgd_next (pgtable_l4_enabled ? \ 475 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd) 476 #else 477 #define pgd_next_t pte_t 478 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 479 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 480 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 481 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 482 #define fixmap_pgd_next ((uintptr_t)fixmap_pte) 483 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd) 484 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) 485 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) 486 #endif /* __PAGETABLE_PMD_FOLDED */ 487 488 void __init create_pgd_mapping(pgd_t *pgdp, 489 uintptr_t va, phys_addr_t pa, 490 phys_addr_t sz, pgprot_t prot) 491 { 492 pgd_next_t *nextp; 493 phys_addr_t next_phys; 494 uintptr_t pgd_idx = pgd_index(va); 495 496 if (sz == PGDIR_SIZE) { 497 if (pgd_val(pgdp[pgd_idx]) == 0) 498 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 499 return; 500 } 501 502 if (pgd_val(pgdp[pgd_idx]) == 0) { 503 next_phys = alloc_pgd_next(va); 504 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 505 nextp = get_pgd_next_virt(next_phys); 506 memset(nextp, 0, PAGE_SIZE); 507 } else { 508 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 509 nextp = get_pgd_next_virt(next_phys); 510 } 511 512 create_pgd_next_mapping(nextp, va, pa, sz, prot); 513 } 514 515 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size) 516 { 517 /* Upgrade to PMD_SIZE mappings whenever possible */ 518 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1))) 519 return PAGE_SIZE; 520 521 return PMD_SIZE; 522 } 523 524 #ifdef CONFIG_XIP_KERNEL 525 extern char _xiprom[], _exiprom[], __data_loc; 526 527 /* called from head.S with MMU off */ 528 asmlinkage void __init __copy_data(void) 529 { 530 void *from = (void *)(&__data_loc); 531 void *to = (void *)CONFIG_PHYS_RAM_BASE; 532 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata)); 533 534 memcpy(to, from, sz); 535 } 536 #endif 537 538 #ifdef CONFIG_STRICT_KERNEL_RWX 539 static __init pgprot_t pgprot_from_va(uintptr_t va) 540 { 541 if (is_va_kernel_text(va)) 542 return PAGE_KERNEL_READ_EXEC; 543 544 /* 545 * In 64-bit kernel, the kernel mapping is outside the linear mapping so 546 * we must protect its linear mapping alias from being executed and 547 * written. 548 * And rodata section is marked readonly in mark_rodata_ro. 549 */ 550 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va)) 551 return PAGE_KERNEL_READ; 552 553 return PAGE_KERNEL; 554 } 555 556 void mark_rodata_ro(void) 557 { 558 set_kernel_memory(__start_rodata, _data, set_memory_ro); 559 if (IS_ENABLED(CONFIG_64BIT)) 560 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data), 561 set_memory_ro); 562 563 debug_checkwx(); 564 } 565 #else 566 static __init pgprot_t pgprot_from_va(uintptr_t va) 567 { 568 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va)) 569 return PAGE_KERNEL; 570 571 return PAGE_KERNEL_EXEC; 572 } 573 #endif /* CONFIG_STRICT_KERNEL_RWX */ 574 575 #ifdef CONFIG_64BIT 576 static void __init disable_pgtable_l4(void) 577 { 578 pgtable_l4_enabled = false; 579 kernel_map.page_offset = PAGE_OFFSET_L3; 580 satp_mode = SATP_MODE_39; 581 } 582 583 /* 584 * There is a simple way to determine if 4-level is supported by the 585 * underlying hardware: establish 1:1 mapping in 4-level page table mode 586 * then read SATP to see if the configuration was taken into account 587 * meaning sv48 is supported. 588 */ 589 static __init void set_satp_mode(void) 590 { 591 u64 identity_satp, hw_satp; 592 uintptr_t set_satp_mode_pmd; 593 594 set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK; 595 create_pgd_mapping(early_pg_dir, 596 set_satp_mode_pmd, (uintptr_t)early_pud, 597 PGDIR_SIZE, PAGE_TABLE); 598 create_pud_mapping(early_pud, 599 set_satp_mode_pmd, (uintptr_t)early_pmd, 600 PUD_SIZE, PAGE_TABLE); 601 /* Handle the case where set_satp_mode straddles 2 PMDs */ 602 create_pmd_mapping(early_pmd, 603 set_satp_mode_pmd, set_satp_mode_pmd, 604 PMD_SIZE, PAGE_KERNEL_EXEC); 605 create_pmd_mapping(early_pmd, 606 set_satp_mode_pmd + PMD_SIZE, 607 set_satp_mode_pmd + PMD_SIZE, 608 PMD_SIZE, PAGE_KERNEL_EXEC); 609 610 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode; 611 612 local_flush_tlb_all(); 613 csr_write(CSR_SATP, identity_satp); 614 hw_satp = csr_swap(CSR_SATP, 0ULL); 615 local_flush_tlb_all(); 616 617 if (hw_satp != identity_satp) 618 disable_pgtable_l4(); 619 620 memset(early_pg_dir, 0, PAGE_SIZE); 621 memset(early_pud, 0, PAGE_SIZE); 622 memset(early_pmd, 0, PAGE_SIZE); 623 } 624 #endif 625 626 /* 627 * setup_vm() is called from head.S with MMU-off. 628 * 629 * Following requirements should be honoured for setup_vm() to work 630 * correctly: 631 * 1) It should use PC-relative addressing for accessing kernel symbols. 632 * To achieve this we always use GCC cmodel=medany. 633 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 634 * so disable compiler instrumentation when FTRACE is enabled. 635 * 636 * Currently, the above requirements are honoured by using custom CFLAGS 637 * for init.o in mm/Makefile. 638 */ 639 640 #ifndef __riscv_cmodel_medany 641 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 642 #endif 643 644 #ifdef CONFIG_XIP_KERNEL 645 static void __init create_kernel_page_table(pgd_t *pgdir, 646 __always_unused bool early) 647 { 648 uintptr_t va, end_va; 649 650 /* Map the flash resident part */ 651 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz; 652 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 653 create_pgd_mapping(pgdir, va, 654 kernel_map.xiprom + (va - kernel_map.virt_addr), 655 PMD_SIZE, PAGE_KERNEL_EXEC); 656 657 /* Map the data in RAM */ 658 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size; 659 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE) 660 create_pgd_mapping(pgdir, va, 661 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)), 662 PMD_SIZE, PAGE_KERNEL); 663 } 664 #else 665 static void __init create_kernel_page_table(pgd_t *pgdir, bool early) 666 { 667 uintptr_t va, end_va; 668 669 end_va = kernel_map.virt_addr + kernel_map.size; 670 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE) 671 create_pgd_mapping(pgdir, va, 672 kernel_map.phys_addr + (va - kernel_map.virt_addr), 673 PMD_SIZE, 674 early ? 675 PAGE_KERNEL_EXEC : pgprot_from_va(va)); 676 } 677 #endif 678 679 /* 680 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel, 681 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR 682 * entry. 683 */ 684 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa) 685 { 686 #ifndef CONFIG_BUILTIN_DTB 687 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1); 688 689 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 690 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa, 691 PGDIR_SIZE, 692 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL); 693 694 if (pgtable_l4_enabled) { 695 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA, 696 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE); 697 } 698 699 if (IS_ENABLED(CONFIG_64BIT)) { 700 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA, 701 pa, PMD_SIZE, PAGE_KERNEL); 702 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE, 703 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 704 } 705 706 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1)); 707 #else 708 /* 709 * For 64-bit kernel, __va can't be used since it would return a linear 710 * mapping address whereas dtb_early_va will be used before 711 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the 712 * kernel is mapped in the linear mapping, that makes no difference. 713 */ 714 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 715 #endif 716 717 dtb_early_pa = dtb_pa; 718 } 719 720 /* 721 * MMU is not enabled, the page tables are allocated directly using 722 * early_pmd/pud/p4d and the address returned is the physical one. 723 */ 724 void __init pt_ops_set_early(void) 725 { 726 pt_ops.alloc_pte = alloc_pte_early; 727 pt_ops.get_pte_virt = get_pte_virt_early; 728 #ifndef __PAGETABLE_PMD_FOLDED 729 pt_ops.alloc_pmd = alloc_pmd_early; 730 pt_ops.get_pmd_virt = get_pmd_virt_early; 731 pt_ops.alloc_pud = alloc_pud_early; 732 pt_ops.get_pud_virt = get_pud_virt_early; 733 #endif 734 } 735 736 /* 737 * MMU is enabled but page table setup is not complete yet. 738 * fixmap page table alloc functions must be used as a means to temporarily 739 * map the allocated physical pages since the linear mapping does not exist yet. 740 * 741 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va, 742 * but it will be used as described above. 743 */ 744 void __init pt_ops_set_fixmap(void) 745 { 746 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap); 747 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap); 748 #ifndef __PAGETABLE_PMD_FOLDED 749 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap); 750 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap); 751 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap); 752 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap); 753 #endif 754 } 755 756 /* 757 * MMU is enabled and page table setup is complete, so from now, we can use 758 * generic page allocation functions to setup page table. 759 */ 760 void __init pt_ops_set_late(void) 761 { 762 pt_ops.alloc_pte = alloc_pte_late; 763 pt_ops.get_pte_virt = get_pte_virt_late; 764 #ifndef __PAGETABLE_PMD_FOLDED 765 pt_ops.alloc_pmd = alloc_pmd_late; 766 pt_ops.get_pmd_virt = get_pmd_virt_late; 767 pt_ops.alloc_pud = alloc_pud_late; 768 pt_ops.get_pud_virt = get_pud_virt_late; 769 #endif 770 } 771 772 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 773 { 774 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd; 775 776 kernel_map.virt_addr = KERNEL_LINK_ADDR; 777 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL); 778 779 #ifdef CONFIG_XIP_KERNEL 780 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 781 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 782 783 phys_ram_base = CONFIG_PHYS_RAM_BASE; 784 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE; 785 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 786 787 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom; 788 #else 789 kernel_map.phys_addr = (uintptr_t)(&_start); 790 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr; 791 #endif 792 793 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL) 794 set_satp_mode(); 795 #endif 796 797 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr; 798 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr; 799 800 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr); 801 802 /* 803 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit 804 * kernel, whereas for 64-bit kernel, the end of the virtual address 805 * space is occupied by the modules/BPF/kernel mappings which reduces 806 * the available size of the linear mapping. 807 */ 808 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0); 809 810 /* Sanity check alignment and size */ 811 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 812 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0); 813 814 #ifdef CONFIG_64BIT 815 /* 816 * The last 4K bytes of the addressable memory can not be mapped because 817 * of IS_ERR_VALUE macro. 818 */ 819 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K); 820 #endif 821 822 pt_ops_set_early(); 823 824 /* Setup early PGD for fixmap */ 825 create_pgd_mapping(early_pg_dir, FIXADDR_START, 826 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 827 828 #ifndef __PAGETABLE_PMD_FOLDED 829 /* Setup fixmap PUD and PMD */ 830 if (pgtable_l4_enabled) 831 create_pud_mapping(fixmap_pud, FIXADDR_START, 832 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE); 833 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 834 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 835 /* Setup trampoline PGD and PMD */ 836 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 837 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE); 838 if (pgtable_l4_enabled) 839 create_pud_mapping(trampoline_pud, kernel_map.virt_addr, 840 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE); 841 #ifdef CONFIG_XIP_KERNEL 842 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 843 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 844 #else 845 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr, 846 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC); 847 #endif 848 #else 849 /* Setup trampoline PGD */ 850 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr, 851 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC); 852 #endif 853 854 /* 855 * Setup early PGD covering entire kernel which will allow 856 * us to reach paging_init(). We map all memory banks later 857 * in setup_vm_final() below. 858 */ 859 create_kernel_page_table(early_pg_dir, true); 860 861 /* Setup early mapping for FDT early scan */ 862 create_fdt_early_page_table(early_pg_dir, dtb_pa); 863 864 /* 865 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 866 * range can not span multiple pmds. 867 */ 868 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 869 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 870 871 #ifndef __PAGETABLE_PMD_FOLDED 872 /* 873 * Early ioremap fixmap is already created as it lies within first 2MB 874 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 875 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 876 * the user if not. 877 */ 878 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 879 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 880 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 881 WARN_ON(1); 882 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 883 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 884 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 885 fix_to_virt(FIX_BTMAP_BEGIN)); 886 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 887 fix_to_virt(FIX_BTMAP_END)); 888 889 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 890 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 891 } 892 #endif 893 894 pt_ops_set_fixmap(); 895 } 896 897 static void __init setup_vm_final(void) 898 { 899 uintptr_t va, map_size; 900 phys_addr_t pa, start, end; 901 u64 i; 902 903 /* Setup swapper PGD for fixmap */ 904 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 905 __pa_symbol(fixmap_pgd_next), 906 PGDIR_SIZE, PAGE_TABLE); 907 908 /* Map all memory banks in the linear mapping */ 909 for_each_mem_range(i, &start, &end) { 910 if (start >= end) 911 break; 912 if (start <= __pa(PAGE_OFFSET) && 913 __pa(PAGE_OFFSET) < end) 914 start = __pa(PAGE_OFFSET); 915 if (end >= __pa(PAGE_OFFSET) + memory_limit) 916 end = __pa(PAGE_OFFSET) + memory_limit; 917 918 map_size = best_map_size(start, end - start); 919 for (pa = start; pa < end; pa += map_size) { 920 va = (uintptr_t)__va(pa); 921 922 create_pgd_mapping(swapper_pg_dir, va, pa, map_size, 923 pgprot_from_va(va)); 924 } 925 } 926 927 /* Map the kernel */ 928 if (IS_ENABLED(CONFIG_64BIT)) 929 create_kernel_page_table(swapper_pg_dir, false); 930 931 #ifdef CONFIG_KASAN 932 kasan_swapper_init(); 933 #endif 934 935 /* Clear fixmap PTE and PMD mappings */ 936 clear_fixmap(FIX_PTE); 937 clear_fixmap(FIX_PMD); 938 clear_fixmap(FIX_PUD); 939 940 /* Move to swapper page table */ 941 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode); 942 local_flush_tlb_all(); 943 944 pt_ops_set_late(); 945 } 946 #else 947 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 948 { 949 dtb_early_va = (void *)dtb_pa; 950 dtb_early_pa = dtb_pa; 951 } 952 953 static inline void setup_vm_final(void) 954 { 955 } 956 #endif /* CONFIG_MMU */ 957 958 #ifdef CONFIG_KEXEC_CORE 959 /* 960 * reserve_crashkernel() - reserves memory for crash kernel 961 * 962 * This function reserves memory area given in "crashkernel=" kernel command 963 * line parameter. The memory reserved is used by dump capture kernel when 964 * primary kernel is crashing. 965 */ 966 static void __init reserve_crashkernel(void) 967 { 968 unsigned long long crash_base = 0; 969 unsigned long long crash_size = 0; 970 unsigned long search_start = memblock_start_of_DRAM(); 971 unsigned long search_end = memblock_end_of_DRAM(); 972 973 int ret = 0; 974 975 /* 976 * Don't reserve a region for a crash kernel on a crash kernel 977 * since it doesn't make much sense and we have limited memory 978 * resources. 979 */ 980 if (is_kdump_kernel()) { 981 pr_info("crashkernel: ignoring reservation request\n"); 982 return; 983 } 984 985 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 986 &crash_size, &crash_base); 987 if (ret || !crash_size) 988 return; 989 990 crash_size = PAGE_ALIGN(crash_size); 991 992 if (crash_base) { 993 search_start = crash_base; 994 search_end = crash_base + crash_size; 995 } 996 997 /* 998 * Current riscv boot protocol requires 2MB alignment for 999 * RV64 and 4MB alignment for RV32 (hugepage size) 1000 * 1001 * Try to alloc from 32bit addressible physical memory so that 1002 * swiotlb can work on the crash kernel. 1003 */ 1004 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1005 search_start, 1006 min(search_end, (unsigned long) SZ_4G)); 1007 if (crash_base == 0) { 1008 /* Try again without restricting region to 32bit addressible memory */ 1009 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE, 1010 search_start, search_end); 1011 if (crash_base == 0) { 1012 pr_warn("crashkernel: couldn't allocate %lldKB\n", 1013 crash_size >> 10); 1014 return; 1015 } 1016 } 1017 1018 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 1019 crash_base, crash_base + crash_size, crash_size >> 20); 1020 1021 crashk_res.start = crash_base; 1022 crashk_res.end = crash_base + crash_size - 1; 1023 } 1024 #endif /* CONFIG_KEXEC_CORE */ 1025 1026 void __init paging_init(void) 1027 { 1028 setup_bootmem(); 1029 setup_vm_final(); 1030 } 1031 1032 void __init misc_mem_init(void) 1033 { 1034 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 1035 arch_numa_init(); 1036 sparse_init(); 1037 zone_sizes_init(); 1038 #ifdef CONFIG_KEXEC_CORE 1039 reserve_crashkernel(); 1040 #endif 1041 memblock_dump_all(); 1042 } 1043 1044 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1045 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 1046 struct vmem_altmap *altmap) 1047 { 1048 return vmemmap_populate_basepages(start, end, node, NULL); 1049 } 1050 #endif 1051