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/sizes.h> 15 #include <linux/of_fdt.h> 16 #include <linux/of_reserved_mem.h> 17 #include <linux/libfdt.h> 18 #include <linux/set_memory.h> 19 #include <linux/dma-map-ops.h> 20 #include <linux/crash_dump.h> 21 22 #include <asm/fixmap.h> 23 #include <asm/tlbflush.h> 24 #include <asm/sections.h> 25 #include <asm/soc.h> 26 #include <asm/io.h> 27 #include <asm/ptdump.h> 28 #include <asm/numa.h> 29 30 #include "../kernel/head.h" 31 32 unsigned long kernel_virt_addr = KERNEL_LINK_ADDR; 33 EXPORT_SYMBOL(kernel_virt_addr); 34 #ifdef CONFIG_XIP_KERNEL 35 #define kernel_virt_addr (*((unsigned long *)XIP_FIXUP(&kernel_virt_addr))) 36 #endif 37 38 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] 39 __page_aligned_bss; 40 EXPORT_SYMBOL(empty_zero_page); 41 42 extern char _start[]; 43 #define DTB_EARLY_BASE_VA PGDIR_SIZE 44 void *_dtb_early_va __initdata; 45 uintptr_t _dtb_early_pa __initdata; 46 47 struct pt_alloc_ops { 48 pte_t *(*get_pte_virt)(phys_addr_t pa); 49 phys_addr_t (*alloc_pte)(uintptr_t va); 50 #ifndef __PAGETABLE_PMD_FOLDED 51 pmd_t *(*get_pmd_virt)(phys_addr_t pa); 52 phys_addr_t (*alloc_pmd)(uintptr_t va); 53 #endif 54 }; 55 56 static phys_addr_t dma32_phys_limit __ro_after_init; 57 58 static void __init zone_sizes_init(void) 59 { 60 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, }; 61 62 #ifdef CONFIG_ZONE_DMA32 63 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit); 64 #endif 65 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 66 67 free_area_init(max_zone_pfns); 68 } 69 70 static void __init setup_zero_page(void) 71 { 72 memset((void *)empty_zero_page, 0, PAGE_SIZE); 73 } 74 75 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM) 76 static inline void print_mlk(char *name, unsigned long b, unsigned long t) 77 { 78 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t, 79 (((t) - (b)) >> 10)); 80 } 81 82 static inline void print_mlm(char *name, unsigned long b, unsigned long t) 83 { 84 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t, 85 (((t) - (b)) >> 20)); 86 } 87 88 static void __init print_vm_layout(void) 89 { 90 pr_notice("Virtual kernel memory layout:\n"); 91 print_mlk("fixmap", (unsigned long)FIXADDR_START, 92 (unsigned long)FIXADDR_TOP); 93 print_mlm("pci io", (unsigned long)PCI_IO_START, 94 (unsigned long)PCI_IO_END); 95 print_mlm("vmemmap", (unsigned long)VMEMMAP_START, 96 (unsigned long)VMEMMAP_END); 97 print_mlm("vmalloc", (unsigned long)VMALLOC_START, 98 (unsigned long)VMALLOC_END); 99 print_mlm("lowmem", (unsigned long)PAGE_OFFSET, 100 (unsigned long)high_memory); 101 #ifdef CONFIG_64BIT 102 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR, 103 (unsigned long)ADDRESS_SPACE_END); 104 #endif 105 } 106 #else 107 static void print_vm_layout(void) { } 108 #endif /* CONFIG_DEBUG_VM */ 109 110 void __init mem_init(void) 111 { 112 #ifdef CONFIG_FLATMEM 113 BUG_ON(!mem_map); 114 #endif /* CONFIG_FLATMEM */ 115 116 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn))); 117 memblock_free_all(); 118 119 print_vm_layout(); 120 } 121 122 void __init setup_bootmem(void) 123 { 124 phys_addr_t vmlinux_end = __pa_symbol(&_end); 125 phys_addr_t vmlinux_start = __pa_symbol(&_start); 126 phys_addr_t dram_end = memblock_end_of_DRAM(); 127 phys_addr_t max_mapped_addr = __pa(~(ulong)0); 128 129 #ifdef CONFIG_XIP_KERNEL 130 vmlinux_start = __pa_symbol(&_sdata); 131 #endif 132 133 /* The maximal physical memory size is -PAGE_OFFSET. */ 134 memblock_enforce_memory_limit(-PAGE_OFFSET); 135 136 /* 137 * Reserve from the start of the kernel to the end of the kernel 138 */ 139 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX) 140 /* 141 * Make sure we align the reservation on PMD_SIZE since we will 142 * map the kernel in the linear mapping as read-only: we do not want 143 * any allocation to happen between _end and the next pmd aligned page. 144 */ 145 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK; 146 #endif 147 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start); 148 149 /* 150 * memblock allocator is not aware of the fact that last 4K bytes of 151 * the addressable memory can not be mapped because of IS_ERR_VALUE 152 * macro. Make sure that last 4k bytes are not usable by memblock 153 * if end of dram is equal to maximum addressable memory. 154 */ 155 if (max_mapped_addr == (dram_end - 1)) 156 memblock_set_current_limit(max_mapped_addr - 4096); 157 158 min_low_pfn = PFN_UP(memblock_start_of_DRAM()); 159 max_low_pfn = max_pfn = PFN_DOWN(dram_end); 160 161 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn)); 162 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET); 163 164 reserve_initrd_mem(); 165 /* 166 * If DTB is built in, no need to reserve its memblock. 167 * Otherwise, do reserve it but avoid using 168 * early_init_fdt_reserve_self() since __pa() does 169 * not work for DTB pointers that are fixmap addresses 170 */ 171 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) 172 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va)); 173 174 early_init_fdt_scan_reserved_mem(); 175 dma_contiguous_reserve(dma32_phys_limit); 176 memblock_allow_resize(); 177 } 178 179 #ifdef CONFIG_XIP_KERNEL 180 181 extern char _xiprom[], _exiprom[]; 182 extern char _sdata[], _edata[]; 183 184 #endif /* CONFIG_XIP_KERNEL */ 185 186 #ifdef CONFIG_MMU 187 static struct pt_alloc_ops _pt_ops __ro_after_init; 188 189 #ifdef CONFIG_XIP_KERNEL 190 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&_pt_ops)) 191 #else 192 #define pt_ops _pt_ops 193 #endif 194 195 /* Offset between linear mapping virtual address and kernel load address */ 196 unsigned long va_pa_offset __ro_after_init; 197 EXPORT_SYMBOL(va_pa_offset); 198 #ifdef CONFIG_XIP_KERNEL 199 #define va_pa_offset (*((unsigned long *)XIP_FIXUP(&va_pa_offset))) 200 #endif 201 /* Offset between kernel mapping virtual address and kernel load address */ 202 #ifdef CONFIG_64BIT 203 unsigned long va_kernel_pa_offset; 204 EXPORT_SYMBOL(va_kernel_pa_offset); 205 #endif 206 #ifdef CONFIG_XIP_KERNEL 207 #define va_kernel_pa_offset (*((unsigned long *)XIP_FIXUP(&va_kernel_pa_offset))) 208 #endif 209 unsigned long va_kernel_xip_pa_offset; 210 EXPORT_SYMBOL(va_kernel_xip_pa_offset); 211 #ifdef CONFIG_XIP_KERNEL 212 #define va_kernel_xip_pa_offset (*((unsigned long *)XIP_FIXUP(&va_kernel_xip_pa_offset))) 213 #endif 214 unsigned long pfn_base __ro_after_init; 215 EXPORT_SYMBOL(pfn_base); 216 217 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 218 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; 219 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss; 220 221 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); 222 223 #ifdef CONFIG_XIP_KERNEL 224 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir)) 225 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte)) 226 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir)) 227 #endif /* CONFIG_XIP_KERNEL */ 228 229 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) 230 { 231 unsigned long addr = __fix_to_virt(idx); 232 pte_t *ptep; 233 234 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); 235 236 ptep = &fixmap_pte[pte_index(addr)]; 237 238 if (pgprot_val(prot)) 239 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot)); 240 else 241 pte_clear(&init_mm, addr, ptep); 242 local_flush_tlb_page(addr); 243 } 244 245 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa) 246 { 247 return (pte_t *)((uintptr_t)pa); 248 } 249 250 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa) 251 { 252 clear_fixmap(FIX_PTE); 253 return (pte_t *)set_fixmap_offset(FIX_PTE, pa); 254 } 255 256 static inline pte_t *get_pte_virt_late(phys_addr_t pa) 257 { 258 return (pte_t *) __va(pa); 259 } 260 261 static inline phys_addr_t __init alloc_pte_early(uintptr_t va) 262 { 263 /* 264 * We only create PMD or PGD early mappings so we 265 * should never reach here with MMU disabled. 266 */ 267 BUG(); 268 } 269 270 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va) 271 { 272 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 273 } 274 275 static phys_addr_t alloc_pte_late(uintptr_t va) 276 { 277 unsigned long vaddr; 278 279 vaddr = __get_free_page(GFP_KERNEL); 280 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr))); 281 282 return __pa(vaddr); 283 } 284 285 static void __init create_pte_mapping(pte_t *ptep, 286 uintptr_t va, phys_addr_t pa, 287 phys_addr_t sz, pgprot_t prot) 288 { 289 uintptr_t pte_idx = pte_index(va); 290 291 BUG_ON(sz != PAGE_SIZE); 292 293 if (pte_none(ptep[pte_idx])) 294 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot); 295 } 296 297 #ifndef __PAGETABLE_PMD_FOLDED 298 299 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss; 300 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss; 301 pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 302 pmd_t early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE); 303 304 #ifdef CONFIG_XIP_KERNEL 305 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd)) 306 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd)) 307 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd)) 308 #endif /* CONFIG_XIP_KERNEL */ 309 310 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa) 311 { 312 /* Before MMU is enabled */ 313 return (pmd_t *)((uintptr_t)pa); 314 } 315 316 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa) 317 { 318 clear_fixmap(FIX_PMD); 319 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa); 320 } 321 322 static pmd_t *get_pmd_virt_late(phys_addr_t pa) 323 { 324 return (pmd_t *) __va(pa); 325 } 326 327 static phys_addr_t __init alloc_pmd_early(uintptr_t va) 328 { 329 BUG_ON((va - kernel_virt_addr) >> PGDIR_SHIFT); 330 331 return (uintptr_t)early_pmd; 332 } 333 334 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va) 335 { 336 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE); 337 } 338 339 static phys_addr_t alloc_pmd_late(uintptr_t va) 340 { 341 unsigned long vaddr; 342 343 vaddr = __get_free_page(GFP_KERNEL); 344 BUG_ON(!vaddr); 345 return __pa(vaddr); 346 } 347 348 static void __init create_pmd_mapping(pmd_t *pmdp, 349 uintptr_t va, phys_addr_t pa, 350 phys_addr_t sz, pgprot_t prot) 351 { 352 pte_t *ptep; 353 phys_addr_t pte_phys; 354 uintptr_t pmd_idx = pmd_index(va); 355 356 if (sz == PMD_SIZE) { 357 if (pmd_none(pmdp[pmd_idx])) 358 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot); 359 return; 360 } 361 362 if (pmd_none(pmdp[pmd_idx])) { 363 pte_phys = pt_ops.alloc_pte(va); 364 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE); 365 ptep = pt_ops.get_pte_virt(pte_phys); 366 memset(ptep, 0, PAGE_SIZE); 367 } else { 368 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx])); 369 ptep = pt_ops.get_pte_virt(pte_phys); 370 } 371 372 create_pte_mapping(ptep, va, pa, sz, prot); 373 } 374 375 #define pgd_next_t pmd_t 376 #define alloc_pgd_next(__va) pt_ops.alloc_pmd(__va) 377 #define get_pgd_next_virt(__pa) pt_ops.get_pmd_virt(__pa) 378 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 379 create_pmd_mapping(__nextp, __va, __pa, __sz, __prot) 380 #define fixmap_pgd_next fixmap_pmd 381 #else 382 #define pgd_next_t pte_t 383 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va) 384 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa) 385 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \ 386 create_pte_mapping(__nextp, __va, __pa, __sz, __prot) 387 #define fixmap_pgd_next fixmap_pte 388 #endif 389 390 void __init create_pgd_mapping(pgd_t *pgdp, 391 uintptr_t va, phys_addr_t pa, 392 phys_addr_t sz, pgprot_t prot) 393 { 394 pgd_next_t *nextp; 395 phys_addr_t next_phys; 396 uintptr_t pgd_idx = pgd_index(va); 397 398 if (sz == PGDIR_SIZE) { 399 if (pgd_val(pgdp[pgd_idx]) == 0) 400 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot); 401 return; 402 } 403 404 if (pgd_val(pgdp[pgd_idx]) == 0) { 405 next_phys = alloc_pgd_next(va); 406 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE); 407 nextp = get_pgd_next_virt(next_phys); 408 memset(nextp, 0, PAGE_SIZE); 409 } else { 410 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx])); 411 nextp = get_pgd_next_virt(next_phys); 412 } 413 414 create_pgd_next_mapping(nextp, va, pa, sz, prot); 415 } 416 417 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size) 418 { 419 /* Upgrade to PMD_SIZE mappings whenever possible */ 420 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1))) 421 return PAGE_SIZE; 422 423 return PMD_SIZE; 424 } 425 426 #ifdef CONFIG_XIP_KERNEL 427 /* called from head.S with MMU off */ 428 asmlinkage void __init __copy_data(void) 429 { 430 void *from = (void *)(&_sdata); 431 void *end = (void *)(&_end); 432 void *to = (void *)CONFIG_PHYS_RAM_BASE; 433 size_t sz = (size_t)(end - from + 1); 434 435 memcpy(to, from, sz); 436 } 437 #endif 438 439 /* 440 * setup_vm() is called from head.S with MMU-off. 441 * 442 * Following requirements should be honoured for setup_vm() to work 443 * correctly: 444 * 1) It should use PC-relative addressing for accessing kernel symbols. 445 * To achieve this we always use GCC cmodel=medany. 446 * 2) The compiler instrumentation for FTRACE will not work for setup_vm() 447 * so disable compiler instrumentation when FTRACE is enabled. 448 * 449 * Currently, the above requirements are honoured by using custom CFLAGS 450 * for init.o in mm/Makefile. 451 */ 452 453 #ifndef __riscv_cmodel_medany 454 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing." 455 #endif 456 457 uintptr_t load_pa, load_sz; 458 #ifdef CONFIG_XIP_KERNEL 459 #define load_pa (*((uintptr_t *)XIP_FIXUP(&load_pa))) 460 #define load_sz (*((uintptr_t *)XIP_FIXUP(&load_sz))) 461 #endif 462 463 #ifdef CONFIG_XIP_KERNEL 464 uintptr_t xiprom, xiprom_sz; 465 #define xiprom_sz (*((uintptr_t *)XIP_FIXUP(&xiprom_sz))) 466 #define xiprom (*((uintptr_t *)XIP_FIXUP(&xiprom))) 467 468 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size) 469 { 470 uintptr_t va, end_va; 471 472 /* Map the flash resident part */ 473 end_va = kernel_virt_addr + xiprom_sz; 474 for (va = kernel_virt_addr; va < end_va; va += map_size) 475 create_pgd_mapping(pgdir, va, 476 xiprom + (va - kernel_virt_addr), 477 map_size, PAGE_KERNEL_EXEC); 478 479 /* Map the data in RAM */ 480 end_va = kernel_virt_addr + XIP_OFFSET + load_sz; 481 for (va = kernel_virt_addr + XIP_OFFSET; va < end_va; va += map_size) 482 create_pgd_mapping(pgdir, va, 483 load_pa + (va - (kernel_virt_addr + XIP_OFFSET)), 484 map_size, PAGE_KERNEL); 485 } 486 #else 487 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size) 488 { 489 uintptr_t va, end_va; 490 491 end_va = kernel_virt_addr + load_sz; 492 for (va = kernel_virt_addr; va < end_va; va += map_size) 493 create_pgd_mapping(pgdir, va, 494 load_pa + (va - kernel_virt_addr), 495 map_size, PAGE_KERNEL_EXEC); 496 } 497 #endif 498 499 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 500 { 501 uintptr_t __maybe_unused pa; 502 uintptr_t map_size; 503 #ifndef __PAGETABLE_PMD_FOLDED 504 pmd_t fix_bmap_spmd, fix_bmap_epmd; 505 #endif 506 507 #ifdef CONFIG_XIP_KERNEL 508 xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR; 509 xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom); 510 511 load_pa = (uintptr_t)CONFIG_PHYS_RAM_BASE; 512 load_sz = (uintptr_t)(&_end) - (uintptr_t)(&_sdata); 513 514 va_kernel_xip_pa_offset = kernel_virt_addr - xiprom; 515 #else 516 load_pa = (uintptr_t)(&_start); 517 load_sz = (uintptr_t)(&_end) - load_pa; 518 #endif 519 520 va_pa_offset = PAGE_OFFSET - load_pa; 521 #ifdef CONFIG_64BIT 522 va_kernel_pa_offset = kernel_virt_addr - load_pa; 523 #endif 524 525 pfn_base = PFN_DOWN(load_pa); 526 527 /* 528 * Enforce boot alignment requirements of RV32 and 529 * RV64 by only allowing PMD or PGD mappings. 530 */ 531 map_size = PMD_SIZE; 532 533 /* Sanity check alignment and size */ 534 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0); 535 BUG_ON((load_pa % map_size) != 0); 536 537 pt_ops.alloc_pte = alloc_pte_early; 538 pt_ops.get_pte_virt = get_pte_virt_early; 539 #ifndef __PAGETABLE_PMD_FOLDED 540 pt_ops.alloc_pmd = alloc_pmd_early; 541 pt_ops.get_pmd_virt = get_pmd_virt_early; 542 #endif 543 /* Setup early PGD for fixmap */ 544 create_pgd_mapping(early_pg_dir, FIXADDR_START, 545 (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE); 546 547 #ifndef __PAGETABLE_PMD_FOLDED 548 /* Setup fixmap PMD */ 549 create_pmd_mapping(fixmap_pmd, FIXADDR_START, 550 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE); 551 /* Setup trampoline PGD and PMD */ 552 create_pgd_mapping(trampoline_pg_dir, kernel_virt_addr, 553 (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE); 554 #ifdef CONFIG_XIP_KERNEL 555 create_pmd_mapping(trampoline_pmd, kernel_virt_addr, 556 xiprom, PMD_SIZE, PAGE_KERNEL_EXEC); 557 #else 558 create_pmd_mapping(trampoline_pmd, kernel_virt_addr, 559 load_pa, PMD_SIZE, PAGE_KERNEL_EXEC); 560 #endif 561 #else 562 /* Setup trampoline PGD */ 563 create_pgd_mapping(trampoline_pg_dir, kernel_virt_addr, 564 load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC); 565 #endif 566 567 /* 568 * Setup early PGD covering entire kernel which will allow 569 * us to reach paging_init(). We map all memory banks later 570 * in setup_vm_final() below. 571 */ 572 create_kernel_page_table(early_pg_dir, map_size); 573 574 #ifndef __PAGETABLE_PMD_FOLDED 575 /* Setup early PMD for DTB */ 576 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 577 (uintptr_t)early_dtb_pmd, PGDIR_SIZE, PAGE_TABLE); 578 #ifndef CONFIG_BUILTIN_DTB 579 /* Create two consecutive PMD mappings for FDT early scan */ 580 pa = dtb_pa & ~(PMD_SIZE - 1); 581 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA, 582 pa, PMD_SIZE, PAGE_KERNEL); 583 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE, 584 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL); 585 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1)); 586 #else /* CONFIG_BUILTIN_DTB */ 587 #ifdef CONFIG_64BIT 588 /* 589 * __va can't be used since it would return a linear mapping address 590 * whereas dtb_early_va will be used before setup_vm_final installs 591 * the linear mapping. 592 */ 593 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 594 #else 595 dtb_early_va = __va(dtb_pa); 596 #endif /* CONFIG_64BIT */ 597 #endif /* CONFIG_BUILTIN_DTB */ 598 #else 599 #ifndef CONFIG_BUILTIN_DTB 600 /* Create two consecutive PGD mappings for FDT early scan */ 601 pa = dtb_pa & ~(PGDIR_SIZE - 1); 602 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA, 603 pa, PGDIR_SIZE, PAGE_KERNEL); 604 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA + PGDIR_SIZE, 605 pa + PGDIR_SIZE, PGDIR_SIZE, PAGE_KERNEL); 606 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PGDIR_SIZE - 1)); 607 #else /* CONFIG_BUILTIN_DTB */ 608 #ifdef CONFIG_64BIT 609 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa)); 610 #else 611 dtb_early_va = __va(dtb_pa); 612 #endif /* CONFIG_64BIT */ 613 #endif /* CONFIG_BUILTIN_DTB */ 614 #endif 615 dtb_early_pa = dtb_pa; 616 617 /* 618 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap 619 * range can not span multiple pmds. 620 */ 621 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) 622 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); 623 624 #ifndef __PAGETABLE_PMD_FOLDED 625 /* 626 * Early ioremap fixmap is already created as it lies within first 2MB 627 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END 628 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn 629 * the user if not. 630 */ 631 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))]; 632 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))]; 633 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) { 634 WARN_ON(1); 635 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n", 636 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd)); 637 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", 638 fix_to_virt(FIX_BTMAP_BEGIN)); 639 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", 640 fix_to_virt(FIX_BTMAP_END)); 641 642 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); 643 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); 644 } 645 #endif 646 } 647 648 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX) 649 void protect_kernel_linear_mapping_text_rodata(void) 650 { 651 unsigned long text_start = (unsigned long)lm_alias(_start); 652 unsigned long init_text_start = (unsigned long)lm_alias(__init_text_begin); 653 unsigned long rodata_start = (unsigned long)lm_alias(__start_rodata); 654 unsigned long data_start = (unsigned long)lm_alias(_data); 655 656 set_memory_ro(text_start, (init_text_start - text_start) >> PAGE_SHIFT); 657 set_memory_nx(text_start, (init_text_start - text_start) >> PAGE_SHIFT); 658 659 set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT); 660 set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT); 661 } 662 #endif 663 664 static void __init setup_vm_final(void) 665 { 666 uintptr_t va, map_size; 667 phys_addr_t pa, start, end; 668 u64 i; 669 670 /** 671 * MMU is enabled at this point. But page table setup is not complete yet. 672 * fixmap page table alloc functions should be used at this point 673 */ 674 pt_ops.alloc_pte = alloc_pte_fixmap; 675 pt_ops.get_pte_virt = get_pte_virt_fixmap; 676 #ifndef __PAGETABLE_PMD_FOLDED 677 pt_ops.alloc_pmd = alloc_pmd_fixmap; 678 pt_ops.get_pmd_virt = get_pmd_virt_fixmap; 679 #endif 680 /* Setup swapper PGD for fixmap */ 681 create_pgd_mapping(swapper_pg_dir, FIXADDR_START, 682 __pa_symbol(fixmap_pgd_next), 683 PGDIR_SIZE, PAGE_TABLE); 684 685 /* Map all memory banks in the linear mapping */ 686 for_each_mem_range(i, &start, &end) { 687 if (start >= end) 688 break; 689 if (start <= __pa(PAGE_OFFSET) && 690 __pa(PAGE_OFFSET) < end) 691 start = __pa(PAGE_OFFSET); 692 693 map_size = best_map_size(start, end - start); 694 for (pa = start; pa < end; pa += map_size) { 695 va = (uintptr_t)__va(pa); 696 create_pgd_mapping(swapper_pg_dir, va, pa, 697 map_size, 698 #ifdef CONFIG_64BIT 699 PAGE_KERNEL 700 #else 701 PAGE_KERNEL_EXEC 702 #endif 703 ); 704 705 } 706 } 707 708 #ifdef CONFIG_64BIT 709 /* Map the kernel */ 710 create_kernel_page_table(swapper_pg_dir, PMD_SIZE); 711 #endif 712 713 /* Clear fixmap PTE and PMD mappings */ 714 clear_fixmap(FIX_PTE); 715 clear_fixmap(FIX_PMD); 716 717 /* Move to swapper page table */ 718 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE); 719 local_flush_tlb_all(); 720 721 /* generic page allocation functions must be used to setup page table */ 722 pt_ops.alloc_pte = alloc_pte_late; 723 pt_ops.get_pte_virt = get_pte_virt_late; 724 #ifndef __PAGETABLE_PMD_FOLDED 725 pt_ops.alloc_pmd = alloc_pmd_late; 726 pt_ops.get_pmd_virt = get_pmd_virt_late; 727 #endif 728 } 729 #else 730 asmlinkage void __init setup_vm(uintptr_t dtb_pa) 731 { 732 dtb_early_va = (void *)dtb_pa; 733 dtb_early_pa = dtb_pa; 734 } 735 736 static inline void setup_vm_final(void) 737 { 738 } 739 #endif /* CONFIG_MMU */ 740 741 #ifdef CONFIG_STRICT_KERNEL_RWX 742 void __init protect_kernel_text_data(void) 743 { 744 unsigned long text_start = (unsigned long)_start; 745 unsigned long init_text_start = (unsigned long)__init_text_begin; 746 unsigned long init_data_start = (unsigned long)__init_data_begin; 747 unsigned long rodata_start = (unsigned long)__start_rodata; 748 unsigned long data_start = (unsigned long)_data; 749 unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn))); 750 751 set_memory_ro(text_start, (init_text_start - text_start) >> PAGE_SHIFT); 752 set_memory_ro(init_text_start, (init_data_start - init_text_start) >> PAGE_SHIFT); 753 set_memory_nx(init_data_start, (rodata_start - init_data_start) >> PAGE_SHIFT); 754 /* rodata section is marked readonly in mark_rodata_ro */ 755 set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT); 756 set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT); 757 } 758 759 void mark_rodata_ro(void) 760 { 761 unsigned long rodata_start = (unsigned long)__start_rodata; 762 unsigned long data_start = (unsigned long)_data; 763 764 set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT); 765 766 debug_checkwx(); 767 } 768 #endif 769 770 #ifdef CONFIG_KEXEC_CORE 771 /* 772 * reserve_crashkernel() - reserves memory for crash kernel 773 * 774 * This function reserves memory area given in "crashkernel=" kernel command 775 * line parameter. The memory reserved is used by dump capture kernel when 776 * primary kernel is crashing. 777 */ 778 static void __init reserve_crashkernel(void) 779 { 780 unsigned long long crash_base = 0; 781 unsigned long long crash_size = 0; 782 unsigned long search_start = memblock_start_of_DRAM(); 783 unsigned long search_end = memblock_end_of_DRAM(); 784 785 int ret = 0; 786 787 /* 788 * Don't reserve a region for a crash kernel on a crash kernel 789 * since it doesn't make much sense and we have limited memory 790 * resources. 791 */ 792 #ifdef CONFIG_CRASH_DUMP 793 if (is_kdump_kernel()) { 794 pr_info("crashkernel: ignoring reservation request\n"); 795 return; 796 } 797 #endif 798 799 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 800 &crash_size, &crash_base); 801 if (ret || !crash_size) 802 return; 803 804 crash_size = PAGE_ALIGN(crash_size); 805 806 if (crash_base == 0) { 807 /* 808 * Current riscv boot protocol requires 2MB alignment for 809 * RV64 and 4MB alignment for RV32 (hugepage size) 810 */ 811 crash_base = memblock_find_in_range(search_start, search_end, 812 crash_size, PMD_SIZE); 813 814 if (crash_base == 0) { 815 pr_warn("crashkernel: couldn't allocate %lldKB\n", 816 crash_size >> 10); 817 return; 818 } 819 } else { 820 /* User specifies base address explicitly. */ 821 if (!memblock_is_region_memory(crash_base, crash_size)) { 822 pr_warn("crashkernel: requested region is not memory\n"); 823 return; 824 } 825 826 if (memblock_is_region_reserved(crash_base, crash_size)) { 827 pr_warn("crashkernel: requested region is reserved\n"); 828 return; 829 } 830 831 832 if (!IS_ALIGNED(crash_base, PMD_SIZE)) { 833 pr_warn("crashkernel: requested region is misaligned\n"); 834 return; 835 } 836 } 837 memblock_reserve(crash_base, crash_size); 838 839 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n", 840 crash_base, crash_base + crash_size, crash_size >> 20); 841 842 crashk_res.start = crash_base; 843 crashk_res.end = crash_base + crash_size - 1; 844 } 845 #endif /* CONFIG_KEXEC_CORE */ 846 847 #ifdef CONFIG_CRASH_DUMP 848 /* 849 * We keep track of the ELF core header of the crashed 850 * kernel with a reserved-memory region with compatible 851 * string "linux,elfcorehdr". Here we register a callback 852 * to populate elfcorehdr_addr/size when this region is 853 * present. Note that this region will be marked as 854 * reserved once we call early_init_fdt_scan_reserved_mem() 855 * later on. 856 */ 857 static int elfcore_hdr_setup(struct reserved_mem *rmem) 858 { 859 elfcorehdr_addr = rmem->base; 860 elfcorehdr_size = rmem->size; 861 return 0; 862 } 863 864 RESERVEDMEM_OF_DECLARE(elfcorehdr, "linux,elfcorehdr", elfcore_hdr_setup); 865 #endif 866 867 void __init paging_init(void) 868 { 869 setup_vm_final(); 870 setup_zero_page(); 871 } 872 873 void __init misc_mem_init(void) 874 { 875 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT); 876 arch_numa_init(); 877 sparse_init(); 878 zone_sizes_init(); 879 #ifdef CONFIG_KEXEC_CORE 880 reserve_crashkernel(); 881 #endif 882 memblock_dump_all(); 883 } 884 885 #ifdef CONFIG_SPARSEMEM_VMEMMAP 886 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 887 struct vmem_altmap *altmap) 888 { 889 return vmemmap_populate_basepages(start, end, node, NULL); 890 } 891 #endif 892