arch/riscv/mm/init.c

   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  */
   6
   7 #include <linux/init.h>
   8 #include <linux/mm.h>
   9 #include <linux/memblock.h>
  10 #include <linux/initrd.h>
  11 #include <linux/swap.h>
  12 #include <linux/sizes.h>
  13 #include <linux/of_fdt.h>
  14 #include <linux/libfdt.h>
  15 #include <linux/set_memory.h>
  16
  17 #include <asm/fixmap.h>
  18 #include <asm/tlbflush.h>
  19 #include <asm/sections.h>
  20 #include <asm/soc.h>
  21 #include <asm/io.h>
  22 #include <asm/ptdump.h>
  23
  24 #include "../kernel/head.h"
  25
  26 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
  27                                                         __page_aligned_bss;
  28 EXPORT_SYMBOL(empty_zero_page);
  29
  30 extern char _start[];
  31 void *dtb_early_va;
  32
  33 static void __init zone_sizes_init(void)
  34 {
  35         unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
  36
  37 #ifdef CONFIG_ZONE_DMA32
  38         max_zone_pfns[ZONE_DMA32] = PFN_DOWN(min(4UL * SZ_1G,
  39                         (unsigned long) PFN_PHYS(max_low_pfn)));
  40 #endif
  41         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
  42
  43         free_area_init(max_zone_pfns);
  44 }
  45
  46 static void setup_zero_page(void)
  47 {
  48         memset((void *)empty_zero_page, 0, PAGE_SIZE);
  49 }
  50
  51 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
  52 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
  53 {
  54         pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
  55                   (((t) - (b)) >> 10));
  56 }
  57
  58 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
  59 {
  60         pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
  61                   (((t) - (b)) >> 20));
  62 }
  63
  64 static void print_vm_layout(void)
  65 {
  66         pr_notice("Virtual kernel memory layout:\n");
  67         print_mlk("fixmap", (unsigned long)FIXADDR_START,
  68                   (unsigned long)FIXADDR_TOP);
  69         print_mlm("pci io", (unsigned long)PCI_IO_START,
  70                   (unsigned long)PCI_IO_END);
  71         print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
  72                   (unsigned long)VMEMMAP_END);
  73         print_mlm("vmalloc", (unsigned long)VMALLOC_START,
  74                   (unsigned long)VMALLOC_END);
  75         print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
  76                   (unsigned long)high_memory);
  77 }
  78 #else
  79 static void print_vm_layout(void) { }
  80 #endif /* CONFIG_DEBUG_VM */
  81
  82 void __init mem_init(void)
  83 {
  84 #ifdef CONFIG_FLATMEM
  85         BUG_ON(!mem_map);
  86 #endif /* CONFIG_FLATMEM */
  87
  88         high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
  89         memblock_free_all();
  90
  91         mem_init_print_info(NULL);
  92         print_vm_layout();
  93 }
  94
  95 #ifdef CONFIG_BLK_DEV_INITRD
  96 static void __init setup_initrd(void)
  97 {
  98         phys_addr_t start;
  99         unsigned long size;
 100
 101         /* Ignore the virtul address computed during device tree parsing */
 102         initrd_start = initrd_end = 0;
 103
 104         if (!phys_initrd_size)
 105                 return;
 106         /*
 107          * Round the memory region to page boundaries as per free_initrd_mem()
 108          * This allows us to detect whether the pages overlapping the initrd
 109          * are in use, but more importantly, reserves the entire set of pages
 110          * as we don't want these pages allocated for other purposes.
 111          */
 112         start = round_down(phys_initrd_start, PAGE_SIZE);
 113         size = phys_initrd_size + (phys_initrd_start - start);
 114         size = round_up(size, PAGE_SIZE);
 115
 116         if (!memblock_is_region_memory(start, size)) {
 117                 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
 118                        (u64)start, size);
 119                 goto disable;
 120         }
 121
 122         if (memblock_is_region_reserved(start, size)) {
 123                 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
 124                        (u64)start, size);
 125                 goto disable;
 126         }
 127
 128         memblock_reserve(start, size);
 129         /* Now convert initrd to virtual addresses */
 130         initrd_start = (unsigned long)__va(phys_initrd_start);
 131         initrd_end = initrd_start + phys_initrd_size;
 132         initrd_below_start_ok = 1;
 133
 134         pr_info("Initial ramdisk at: 0x%p (%lu bytes)\n",
 135                 (void *)(initrd_start), size);
 136         return;
 137 disable:
 138         pr_cont(" - disabling initrd\n");
 139         initrd_start = 0;
 140         initrd_end = 0;
 141 }
 142 #endif /* CONFIG_BLK_DEV_INITRD */
 143
 144 static phys_addr_t dtb_early_pa __initdata;
 145
 146 void __init setup_bootmem(void)
 147 {
 148         phys_addr_t mem_size = 0;
 149         phys_addr_t total_mem = 0;
 150         phys_addr_t mem_start, start, end = 0;
 151         phys_addr_t vmlinux_end = __pa_symbol(&_end);
 152         phys_addr_t vmlinux_start = __pa_symbol(&_start);
 153         u64 i;
 154
 155         /* Find the memory region containing the kernel */
 156         for_each_mem_range(i, &start, &end) {
 157                 phys_addr_t size = end - start;
 158                 if (!total_mem)
 159                         mem_start = start;
 160                 if (start <= vmlinux_start && vmlinux_end <= end)
 161                         BUG_ON(size == 0);
 162                 total_mem = total_mem + size;
 163         }
 164
 165         /*
 166          * Remove memblock from the end of usable area to the
 167          * end of region
 168          */
 169         mem_size = min(total_mem, (phys_addr_t)-PAGE_OFFSET);
 170         if (mem_start + mem_size < end)
 171                 memblock_remove(mem_start + mem_size,
 172                                 end - mem_start - mem_size);
 173
 174         /* Reserve from the start of the kernel to the end of the kernel */
 175         memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
 176
 177         max_pfn = PFN_DOWN(memblock_end_of_DRAM());
 178         max_low_pfn = max_pfn;
 179         set_max_mapnr(max_low_pfn);
 180
 181 #ifdef CONFIG_BLK_DEV_INITRD
 182         setup_initrd();
 183 #endif /* CONFIG_BLK_DEV_INITRD */
 184
 185         /*
 186          * Avoid using early_init_fdt_reserve_self() since __pa() does
 187          * not work for DTB pointers that are fixmap addresses
 188          */
 189         memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
 190
 191         early_init_fdt_scan_reserved_mem();
 192         memblock_allow_resize();
 193         memblock_dump_all();
 194 }
 195
 196 #ifdef CONFIG_MMU
 197 unsigned long va_pa_offset;
 198 EXPORT_SYMBOL(va_pa_offset);
 199 unsigned long pfn_base;
 200 EXPORT_SYMBOL(pfn_base);
 201
 202 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
 203 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
 204 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
 205 static bool mmu_enabled;
 206
 207 #define MAX_EARLY_MAPPING_SIZE  SZ_128M
 208
 209 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
 210
 211 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
 212 {
 213         unsigned long addr = __fix_to_virt(idx);
 214         pte_t *ptep;
 215
 216         BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 217
 218         ptep = &fixmap_pte[pte_index(addr)];
 219
 220         if (pgprot_val(prot))
 221                 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
 222         else
 223                 pte_clear(&init_mm, addr, ptep);
 224         local_flush_tlb_page(addr);
 225 }
 226
 227 static pte_t *__init get_pte_virt(phys_addr_t pa)
 228 {
 229         if (mmu_enabled) {
 230                 clear_fixmap(FIX_PTE);
 231                 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
 232         } else {
 233                 return (pte_t *)((uintptr_t)pa);
 234         }
 235 }
 236
 237 static phys_addr_t __init alloc_pte(uintptr_t va)
 238 {
 239         /*
 240          * We only create PMD or PGD early mappings so we
 241          * should never reach here with MMU disabled.
 242          */
 243         BUG_ON(!mmu_enabled);
 244
 245         return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
 246 }
 247
 248 static void __init create_pte_mapping(pte_t *ptep,
 249                                       uintptr_t va, phys_addr_t pa,
 250                                       phys_addr_t sz, pgprot_t prot)
 251 {
 252         uintptr_t pte_idx = pte_index(va);
 253
 254         BUG_ON(sz != PAGE_SIZE);
 255
 256         if (pte_none(ptep[pte_idx]))
 257                 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
 258 }
 259
 260 #ifndef __PAGETABLE_PMD_FOLDED
 261
 262 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
 263 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
 264
 265 #if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
 266 #define NUM_EARLY_PMDS          1UL
 267 #else
 268 #define NUM_EARLY_PMDS          (1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
 269 #endif
 270 pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
 271
 272 static pmd_t *__init get_pmd_virt(phys_addr_t pa)
 273 {
 274         if (mmu_enabled) {
 275                 clear_fixmap(FIX_PMD);
 276                 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
 277         } else {
 278                 return (pmd_t *)((uintptr_t)pa);
 279         }
 280 }
 281
 282 static phys_addr_t __init alloc_pmd(uintptr_t va)
 283 {
 284         uintptr_t pmd_num;
 285
 286         if (mmu_enabled)
 287                 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
 288
 289         pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
 290         BUG_ON(pmd_num >= NUM_EARLY_PMDS);
 291         return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
 292 }
 293
 294 static void __init create_pmd_mapping(pmd_t *pmdp,
 295                                       uintptr_t va, phys_addr_t pa,
 296                                       phys_addr_t sz, pgprot_t prot)
 297 {
 298         pte_t *ptep;
 299         phys_addr_t pte_phys;
 300         uintptr_t pmd_idx = pmd_index(va);
 301
 302         if (sz == PMD_SIZE) {
 303                 if (pmd_none(pmdp[pmd_idx]))
 304                         pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
 305                 return;
 306         }
 307
 308         if (pmd_none(pmdp[pmd_idx])) {
 309                 pte_phys = alloc_pte(va);
 310                 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
 311                 ptep = get_pte_virt(pte_phys);
 312                 memset(ptep, 0, PAGE_SIZE);
 313         } else {
 314                 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
 315                 ptep = get_pte_virt(pte_phys);
 316         }
 317
 318         create_pte_mapping(ptep, va, pa, sz, prot);
 319 }
 320
 321 #define pgd_next_t              pmd_t
 322 #define alloc_pgd_next(__va)    alloc_pmd(__va)
 323 #define get_pgd_next_virt(__pa) get_pmd_virt(__pa)
 324 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)      \
 325         create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
 326 #define fixmap_pgd_next         fixmap_pmd
 327 #else
 328 #define pgd_next_t              pte_t
 329 #define alloc_pgd_next(__va)    alloc_pte(__va)
 330 #define get_pgd_next_virt(__pa) get_pte_virt(__pa)
 331 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)      \
 332         create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
 333 #define fixmap_pgd_next         fixmap_pte
 334 #endif
 335
 336 static void __init create_pgd_mapping(pgd_t *pgdp,
 337                                       uintptr_t va, phys_addr_t pa,
 338                                       phys_addr_t sz, pgprot_t prot)
 339 {
 340         pgd_next_t *nextp;
 341         phys_addr_t next_phys;
 342         uintptr_t pgd_idx = pgd_index(va);
 343
 344         if (sz == PGDIR_SIZE) {
 345                 if (pgd_val(pgdp[pgd_idx]) == 0)
 346                         pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
 347                 return;
 348         }
 349
 350         if (pgd_val(pgdp[pgd_idx]) == 0) {
 351                 next_phys = alloc_pgd_next(va);
 352                 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
 353                 nextp = get_pgd_next_virt(next_phys);
 354                 memset(nextp, 0, PAGE_SIZE);
 355         } else {
 356                 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
 357                 nextp = get_pgd_next_virt(next_phys);
 358         }
 359
 360         create_pgd_next_mapping(nextp, va, pa, sz, prot);
 361 }
 362
 363 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
 364 {
 365         /* Upgrade to PMD_SIZE mappings whenever possible */
 366         if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
 367                 return PAGE_SIZE;
 368
 369         return PMD_SIZE;
 370 }
 371
 372 /*
 373  * setup_vm() is called from head.S with MMU-off.
 374  *
 375  * Following requirements should be honoured for setup_vm() to work
 376  * correctly:
 377  * 1) It should use PC-relative addressing for accessing kernel symbols.
 378  *    To achieve this we always use GCC cmodel=medany.
 379  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
 380  *    so disable compiler instrumentation when FTRACE is enabled.
 381  *
 382  * Currently, the above requirements are honoured by using custom CFLAGS
 383  * for init.o in mm/Makefile.
 384  */
 385
 386 #ifndef __riscv_cmodel_medany
 387 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
 388 #endif
 389
 390 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
 391 {
 392         uintptr_t va, end_va;
 393         uintptr_t load_pa = (uintptr_t)(&_start);
 394         uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
 395         uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
 396
 397         va_pa_offset = PAGE_OFFSET - load_pa;
 398         pfn_base = PFN_DOWN(load_pa);
 399
 400         /*
 401          * Enforce boot alignment requirements of RV32 and
 402          * RV64 by only allowing PMD or PGD mappings.
 403          */
 404         BUG_ON(map_size == PAGE_SIZE);
 405
 406         /* Sanity check alignment and size */
 407         BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
 408         BUG_ON((load_pa % map_size) != 0);
 409         BUG_ON(load_sz > MAX_EARLY_MAPPING_SIZE);
 410
 411         /* Setup early PGD for fixmap */
 412         create_pgd_mapping(early_pg_dir, FIXADDR_START,
 413                            (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
 414
 415 #ifndef __PAGETABLE_PMD_FOLDED
 416         /* Setup fixmap PMD */
 417         create_pmd_mapping(fixmap_pmd, FIXADDR_START,
 418                            (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
 419         /* Setup trampoline PGD and PMD */
 420         create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
 421                            (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
 422         create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
 423                            load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
 424 #else
 425         /* Setup trampoline PGD */
 426         create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
 427                            load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
 428 #endif
 429
 430         /*
 431          * Setup early PGD covering entire kernel which will allows
 432          * us to reach paging_init(). We map all memory banks later
 433          * in setup_vm_final() below.
 434          */
 435         end_va = PAGE_OFFSET + load_sz;
 436         for (va = PAGE_OFFSET; va < end_va; va += map_size)
 437                 create_pgd_mapping(early_pg_dir, va,
 438                                    load_pa + (va - PAGE_OFFSET),
 439                                    map_size, PAGE_KERNEL_EXEC);
 440
 441         /* Create fixed mapping for early FDT parsing */
 442         end_va = __fix_to_virt(FIX_FDT) + FIX_FDT_SIZE;
 443         for (va = __fix_to_virt(FIX_FDT); va < end_va; va += PAGE_SIZE)
 444                 create_pte_mapping(fixmap_pte, va,
 445                                    dtb_pa + (va - __fix_to_virt(FIX_FDT)),
 446                                    PAGE_SIZE, PAGE_KERNEL);
 447
 448         /* Save pointer to DTB for early FDT parsing */
 449         dtb_early_va = (void *)fix_to_virt(FIX_FDT) + (dtb_pa & ~PAGE_MASK);
 450         /* Save physical address for memblock reservation */
 451         dtb_early_pa = dtb_pa;
 452 }
 453
 454 static void __init setup_vm_final(void)
 455 {
 456         uintptr_t va, map_size;
 457         phys_addr_t pa, start, end;
 458         u64 i;
 459
 460         /* Set mmu_enabled flag */
 461         mmu_enabled = true;
 462
 463         /* Setup swapper PGD for fixmap */
 464         create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
 465                            __pa_symbol(fixmap_pgd_next),
 466                            PGDIR_SIZE, PAGE_TABLE);
 467
 468         /* Map all memory banks */
 469         for_each_mem_range(i, &start, &end) {
 470                 if (start >= end)
 471                         break;
 472                 if (start <= __pa(PAGE_OFFSET) &&
 473                     __pa(PAGE_OFFSET) < end)
 474                         start = __pa(PAGE_OFFSET);
 475
 476                 map_size = best_map_size(start, end - start);
 477                 for (pa = start; pa < end; pa += map_size) {
 478                         va = (uintptr_t)__va(pa);
 479                         create_pgd_mapping(swapper_pg_dir, va, pa,
 480                                            map_size, PAGE_KERNEL_EXEC);
 481                 }
 482         }
 483
 484         /* Clear fixmap PTE and PMD mappings */
 485         clear_fixmap(FIX_PTE);
 486         clear_fixmap(FIX_PMD);
 487
 488         /* Move to swapper page table */
 489         csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
 490         local_flush_tlb_all();
 491 }
 492 #else
 493 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
 494 {
 495 #ifdef CONFIG_BUILTIN_DTB
 496         dtb_early_va = soc_lookup_builtin_dtb();
 497         if (!dtb_early_va) {
 498                 /* Fallback to first available DTS */
 499                 dtb_early_va = (void *) __dtb_start;
 500         }
 501 #else
 502         dtb_early_va = (void *)dtb_pa;
 503 #endif
 504         dtb_early_pa = dtb_pa;
 505 }
 506
 507 static inline void setup_vm_final(void)
 508 {
 509 }
 510 #endif /* CONFIG_MMU */
 511
 512 #ifdef CONFIG_STRICT_KERNEL_RWX
 513 void mark_rodata_ro(void)
 514 {
 515         unsigned long text_start = (unsigned long)_text;
 516         unsigned long text_end = (unsigned long)_etext;
 517         unsigned long rodata_start = (unsigned long)__start_rodata;
 518         unsigned long data_start = (unsigned long)_data;
 519         unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
 520
 521         set_memory_ro(text_start, (text_end - text_start) >> PAGE_SHIFT);
 522         set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
 523         set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
 524         set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
 525
 526         debug_checkwx();
 527 }
 528 #endif
 529
 530 static void __init resource_init(void)
 531 {
 532         struct memblock_region *region;
 533
 534         for_each_mem_region(region) {
 535                 struct resource *res;
 536
 537                 res = memblock_alloc(sizeof(struct resource), SMP_CACHE_BYTES);
 538                 if (!res)
 539                         panic("%s: Failed to allocate %zu bytes\n", __func__,
 540                               sizeof(struct resource));
 541
 542                 if (memblock_is_nomap(region)) {
 543                         res->name = "reserved";
 544                         res->flags = IORESOURCE_MEM;
 545                 } else {
 546                         res->name = "System RAM";
 547                         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 548                 }
 549                 res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
 550                 res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
 551
 552                 request_resource(&iomem_resource, res);
 553         }
 554 }
 555
 556 void __init paging_init(void)
 557 {
 558         setup_vm_final();
 559         sparse_init();
 560         setup_zero_page();
 561         zone_sizes_init();
 562         resource_init();
 563 }
 564
 565 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 566 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
 567                                struct vmem_altmap *altmap)
 568 {
 569         return vmemmap_populate_basepages(start, end, node, NULL);
 570 }
 571 #endif