| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/arch/arm/mm/init.c |
| 4 | * |
| 5 | * Copyright (C) 1995-2005 Russell King |
| 6 | */ |
| 7 | #include <linux/kernel.h> |
| 8 | #include <linux/errno.h> |
| 9 | #include <linux/swap.h> |
| 10 | #include <linux/init.h> |
| 11 | #include <linux/mman.h> |
| 12 | #include <linux/sched/signal.h> |
| 13 | #include <linux/sched/task.h> |
| 14 | #include <linux/export.h> |
| 15 | #include <linux/nodemask.h> |
| 16 | #include <linux/initrd.h> |
| 17 | #include <linux/of_fdt.h> |
| 18 | #include <linux/highmem.h> |
| 19 | #include <linux/gfp.h> |
| 20 | #include <linux/memblock.h> |
| 21 | #include <linux/dma-map-ops.h> |
| 22 | #include <linux/sizes.h> |
| 23 | #include <linux/stop_machine.h> |
| 24 | #include <linux/swiotlb.h> |
| 25 | #include <linux/execmem.h> |
| 26 | |
| 27 | #include <asm/cp15.h> |
| 28 | #include <asm/mach-types.h> |
| 29 | #include <asm/memblock.h> |
| 30 | #include <asm/page.h> |
| 31 | #include <asm/prom.h> |
| 32 | #include <asm/sections.h> |
| 33 | #include <asm/setup.h> |
| 34 | #include <asm/set_memory.h> |
| 35 | #include <asm/system_info.h> |
| 36 | #include <asm/tlb.h> |
| 37 | #include <asm/fixmap.h> |
| 38 | #include <asm/ptdump.h> |
| 39 | |
| 40 | #include <asm/mach/arch.h> |
| 41 | #include <asm/mach/map.h> |
| 42 | |
| 43 | #include "mm.h" |
| 44 | |
| 45 | #ifdef CONFIG_CPU_CP15_MMU |
| 46 | unsigned long __init __clear_cr(unsigned long mask) |
| 47 | { |
| 48 | cr_alignment = cr_alignment & ~mask; |
| 49 | return cr_alignment; |
| 50 | } |
| 51 | #endif |
| 52 | |
| 53 | #ifdef CONFIG_BLK_DEV_INITRD |
| 54 | static int __init parse_tag_initrd(const struct tag *tag) |
| 55 | { |
| 56 | pr_warn("ATAG_INITRD is deprecated; " |
| 57 | "please update your bootloader.\n" ); |
| 58 | phys_initrd_start = __virt_to_phys(tag->u.initrd.start); |
| 59 | phys_initrd_size = tag->u.initrd.size; |
| 60 | return 0; |
| 61 | } |
| 62 | |
| 63 | __tagtable(ATAG_INITRD, parse_tag_initrd); |
| 64 | |
| 65 | static int __init parse_tag_initrd2(const struct tag *tag) |
| 66 | { |
| 67 | phys_initrd_start = tag->u.initrd.start; |
| 68 | phys_initrd_size = tag->u.initrd.size; |
| 69 | return 0; |
| 70 | } |
| 71 | |
| 72 | __tagtable(ATAG_INITRD2, parse_tag_initrd2); |
| 73 | #endif |
| 74 | |
| 75 | static void __init find_limits(unsigned long *min, unsigned long *max_low, |
| 76 | unsigned long *max_high) |
| 77 | { |
| 78 | *max_low = PFN_DOWN(memblock_get_current_limit()); |
| 79 | *min = PFN_UP(memblock_start_of_DRAM()); |
| 80 | *max_high = PFN_DOWN(memblock_end_of_DRAM()); |
| 81 | } |
| 82 | |
| 83 | #ifdef CONFIG_ZONE_DMA |
| 84 | |
| 85 | phys_addr_t arm_dma_zone_size __read_mostly; |
| 86 | EXPORT_SYMBOL(arm_dma_zone_size); |
| 87 | |
| 88 | /* |
| 89 | * The DMA mask corresponding to the maximum bus address allocatable |
| 90 | * using GFP_DMA. The default here places no restriction on DMA |
| 91 | * allocations. This must be the smallest DMA mask in the system, |
| 92 | * so a successful GFP_DMA allocation will always satisfy this. |
| 93 | */ |
| 94 | phys_addr_t arm_dma_limit; |
| 95 | unsigned long arm_dma_pfn_limit; |
| 96 | #endif |
| 97 | |
| 98 | void __init setup_dma_zone(const struct machine_desc *mdesc) |
| 99 | { |
| 100 | #ifdef CONFIG_ZONE_DMA |
| 101 | if (mdesc->dma_zone_size) { |
| 102 | arm_dma_zone_size = mdesc->dma_zone_size; |
| 103 | arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1; |
| 104 | } else |
| 105 | arm_dma_limit = 0xffffffff; |
| 106 | arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT; |
| 107 | #endif |
| 108 | } |
| 109 | |
| 110 | static void __init zone_sizes_init(unsigned long min, unsigned long max_low, |
| 111 | unsigned long max_high) |
| 112 | { |
| 113 | unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0 }; |
| 114 | |
| 115 | #ifdef CONFIG_ZONE_DMA |
| 116 | max_zone_pfn[ZONE_DMA] = min(arm_dma_pfn_limit, max_low); |
| 117 | #endif |
| 118 | max_zone_pfn[ZONE_NORMAL] = max_low; |
| 119 | #ifdef CONFIG_HIGHMEM |
| 120 | max_zone_pfn[ZONE_HIGHMEM] = max_high; |
| 121 | #endif |
| 122 | free_area_init(max_zone_pfn); |
| 123 | } |
| 124 | |
| 125 | #ifdef CONFIG_HAVE_ARCH_PFN_VALID |
| 126 | int pfn_valid(unsigned long pfn) |
| 127 | { |
| 128 | phys_addr_t addr = __pfn_to_phys(pfn); |
| 129 | unsigned long pageblock_size = PAGE_SIZE * pageblock_nr_pages; |
| 130 | |
| 131 | if (__phys_to_pfn(addr) != pfn) |
| 132 | return 0; |
| 133 | |
| 134 | /* |
| 135 | * If address less than pageblock_size bytes away from a present |
| 136 | * memory chunk there still will be a memory map entry for it |
| 137 | * because we round freed memory map to the pageblock boundaries. |
| 138 | */ |
| 139 | if (memblock_overlaps_region(&memblock.memory, |
| 140 | ALIGN_DOWN(addr, pageblock_size), |
| 141 | pageblock_size)) |
| 142 | return 1; |
| 143 | |
| 144 | return 0; |
| 145 | } |
| 146 | EXPORT_SYMBOL(pfn_valid); |
| 147 | #endif |
| 148 | |
| 149 | static bool arm_memblock_steal_permitted = true; |
| 150 | |
| 151 | phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align) |
| 152 | { |
| 153 | phys_addr_t phys; |
| 154 | |
| 155 | BUG_ON(!arm_memblock_steal_permitted); |
| 156 | |
| 157 | phys = memblock_phys_alloc(size, align); |
| 158 | if (!phys) |
| 159 | panic(fmt: "Failed to steal %pa bytes at %pS\n" , |
| 160 | &size, (void *)_RET_IP_); |
| 161 | |
| 162 | memblock_phys_free(base: phys, size); |
| 163 | memblock_remove(base: phys, size); |
| 164 | |
| 165 | return phys; |
| 166 | } |
| 167 | |
| 168 | #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND |
| 169 | void check_cpu_icache_size(int cpuid) |
| 170 | { |
| 171 | u32 size, ctr; |
| 172 | |
| 173 | asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr)); |
| 174 | |
| 175 | size = 1 << ((ctr & 0xf) + 2); |
| 176 | if (cpuid != 0 && icache_size != size) |
| 177 | pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n" , |
| 178 | cpuid); |
| 179 | if (icache_size > size) |
| 180 | icache_size = size; |
| 181 | } |
| 182 | #endif |
| 183 | |
| 184 | void __init arm_memblock_init(const struct machine_desc *mdesc) |
| 185 | { |
| 186 | /* Register the kernel text, kernel data and initrd with memblock. */ |
| 187 | memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START); |
| 188 | |
| 189 | reserve_initrd_mem(); |
| 190 | |
| 191 | arm_mm_memblock_reserve(); |
| 192 | |
| 193 | /* reserve any platform specific memblock areas */ |
| 194 | if (mdesc->reserve) |
| 195 | mdesc->reserve(); |
| 196 | |
| 197 | early_init_fdt_scan_reserved_mem(); |
| 198 | |
| 199 | /* reserve memory for DMA contiguous allocations */ |
| 200 | dma_contiguous_reserve(addr_limit: arm_dma_limit); |
| 201 | |
| 202 | arm_memblock_steal_permitted = false; |
| 203 | memblock_dump_all(); |
| 204 | } |
| 205 | |
| 206 | void __init bootmem_init(void) |
| 207 | { |
| 208 | memblock_allow_resize(); |
| 209 | |
| 210 | find_limits(min: &min_low_pfn, max_low: &max_low_pfn, max_high: &max_pfn); |
| 211 | |
| 212 | early_memtest(start: (phys_addr_t)min_low_pfn << PAGE_SHIFT, |
| 213 | end: (phys_addr_t)max_low_pfn << PAGE_SHIFT); |
| 214 | |
| 215 | /* |
| 216 | * sparse_init() tries to allocate memory from memblock, so must be |
| 217 | * done after the fixed reservations |
| 218 | */ |
| 219 | sparse_init(); |
| 220 | |
| 221 | /* |
| 222 | * Now free the memory - free_area_init needs |
| 223 | * the sparse mem_map arrays initialized by sparse_init() |
| 224 | * for memmap_init_zone(), otherwise all PFNs are invalid. |
| 225 | */ |
| 226 | zone_sizes_init(min: min_low_pfn, max_low: max_low_pfn, max_high: max_pfn); |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * Poison init memory with an undefined instruction (ARM) or a branch to an |
| 231 | * undefined instruction (Thumb). |
| 232 | */ |
| 233 | static inline void poison_init_mem(void *s, size_t count) |
| 234 | { |
| 235 | u32 *p = (u32 *)s; |
| 236 | for (; count != 0; count -= 4) |
| 237 | *p++ = 0xe7fddef0; |
| 238 | } |
| 239 | |
| 240 | void __init arch_mm_preinit(void) |
| 241 | { |
| 242 | #ifdef CONFIG_ARM_LPAE |
| 243 | swiotlb_init(max_pfn > arm_dma_pfn_limit, SWIOTLB_VERBOSE); |
| 244 | #endif |
| 245 | |
| 246 | #ifdef CONFIG_SA1111 |
| 247 | /* now that our DMA memory is actually so designated, we can free it */ |
| 248 | memblock_phys_free(PHYS_OFFSET, __pa(swapper_pg_dir) - PHYS_OFFSET); |
| 249 | #endif |
| 250 | |
| 251 | /* |
| 252 | * Check boundaries twice: Some fundamental inconsistencies can |
| 253 | * be detected at build time already. |
| 254 | */ |
| 255 | #ifdef CONFIG_MMU |
| 256 | BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR); |
| 257 | BUG_ON(TASK_SIZE > MODULES_VADDR); |
| 258 | #endif |
| 259 | |
| 260 | #ifdef CONFIG_HIGHMEM |
| 261 | BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET); |
| 262 | BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET); |
| 263 | #endif |
| 264 | } |
| 265 | |
| 266 | #ifdef CONFIG_STRICT_KERNEL_RWX |
| 267 | struct section_perm { |
| 268 | const char *name; |
| 269 | unsigned long start; |
| 270 | unsigned long end; |
| 271 | pmdval_t mask; |
| 272 | pmdval_t prot; |
| 273 | pmdval_t clear; |
| 274 | }; |
| 275 | |
| 276 | /* First section-aligned location at or after __start_rodata. */ |
| 277 | extern char __start_rodata_section_aligned[]; |
| 278 | |
| 279 | static struct section_perm nx_perms[] = { |
| 280 | /* Make pages tables, etc before _stext RW (set NX). */ |
| 281 | { |
| 282 | .name = "pre-text NX" , |
| 283 | .start = PAGE_OFFSET, |
| 284 | .end = (unsigned long)_stext, |
| 285 | .mask = ~PMD_SECT_XN, |
| 286 | .prot = PMD_SECT_XN, |
| 287 | }, |
| 288 | /* Make init RW (set NX). */ |
| 289 | { |
| 290 | .name = "init NX" , |
| 291 | .start = (unsigned long)__init_begin, |
| 292 | .end = (unsigned long)_sdata, |
| 293 | .mask = ~PMD_SECT_XN, |
| 294 | .prot = PMD_SECT_XN, |
| 295 | }, |
| 296 | /* Make rodata NX (set RO in ro_perms below). */ |
| 297 | { |
| 298 | .name = "rodata NX" , |
| 299 | .start = (unsigned long)__start_rodata_section_aligned, |
| 300 | .end = (unsigned long)__init_begin, |
| 301 | .mask = ~PMD_SECT_XN, |
| 302 | .prot = PMD_SECT_XN, |
| 303 | }, |
| 304 | }; |
| 305 | |
| 306 | static struct section_perm ro_perms[] = { |
| 307 | /* Make kernel code and rodata RX (set RO). */ |
| 308 | { |
| 309 | .name = "text/rodata RO" , |
| 310 | .start = (unsigned long)_stext, |
| 311 | .end = (unsigned long)__init_begin, |
| 312 | #ifdef CONFIG_ARM_LPAE |
| 313 | .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2), |
| 314 | .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2, |
| 315 | #else |
| 316 | .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE), |
| 317 | .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE, |
| 318 | .clear = PMD_SECT_AP_WRITE, |
| 319 | #endif |
| 320 | }, |
| 321 | }; |
| 322 | |
| 323 | /* |
| 324 | * Updates section permissions only for the current mm (sections are |
| 325 | * copied into each mm). During startup, this is the init_mm. Is only |
| 326 | * safe to be called with preemption disabled, as under stop_machine(). |
| 327 | */ |
| 328 | static inline void section_update(unsigned long addr, pmdval_t mask, |
| 329 | pmdval_t prot, struct mm_struct *mm) |
| 330 | { |
| 331 | pmd_t *pmd; |
| 332 | |
| 333 | pmd = pmd_offset(pud_offset(p4d: p4d_offset(pgd_offset(mm, addr), address: addr), address: addr), address: addr); |
| 334 | |
| 335 | #ifdef CONFIG_ARM_LPAE |
| 336 | pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot); |
| 337 | #else |
| 338 | if (addr & SECTION_SIZE) |
| 339 | pmd[1] = __pmd(val: (pmd_val(pmd: pmd[1]) & mask) | prot); |
| 340 | else |
| 341 | pmd[0] = __pmd(val: (pmd_val(pmd: pmd[0]) & mask) | prot); |
| 342 | #endif |
| 343 | flush_pmd_entry(pmd); |
| 344 | local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE); |
| 345 | } |
| 346 | |
| 347 | /* Make sure extended page tables are in use. */ |
| 348 | static inline bool arch_has_strict_perms(void) |
| 349 | { |
| 350 | if (cpu_architecture() < CPU_ARCH_ARMv6) |
| 351 | return false; |
| 352 | |
| 353 | return !!(get_cr() & CR_XP); |
| 354 | } |
| 355 | |
| 356 | static void set_section_perms(struct section_perm *perms, int n, bool set, |
| 357 | struct mm_struct *mm) |
| 358 | { |
| 359 | size_t i; |
| 360 | unsigned long addr; |
| 361 | |
| 362 | if (!arch_has_strict_perms()) |
| 363 | return; |
| 364 | |
| 365 | for (i = 0; i < n; i++) { |
| 366 | if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) || |
| 367 | !IS_ALIGNED(perms[i].end, SECTION_SIZE)) { |
| 368 | pr_err("BUG: %s section %lx-%lx not aligned to %lx\n" , |
| 369 | perms[i].name, perms[i].start, perms[i].end, |
| 370 | SECTION_SIZE); |
| 371 | continue; |
| 372 | } |
| 373 | |
| 374 | for (addr = perms[i].start; |
| 375 | addr < perms[i].end; |
| 376 | addr += SECTION_SIZE) |
| 377 | section_update(addr, mask: perms[i].mask, |
| 378 | prot: set ? perms[i].prot : perms[i].clear, mm); |
| 379 | } |
| 380 | |
| 381 | } |
| 382 | |
| 383 | /* |
| 384 | * update_sections_early intended to be called only through stop_machine |
| 385 | * framework and executed by only one CPU while all other CPUs will spin and |
| 386 | * wait, so no locking is required in this function. |
| 387 | */ |
| 388 | static void update_sections_early(struct section_perm perms[], int n) |
| 389 | { |
| 390 | struct task_struct *t, *s; |
| 391 | |
| 392 | for_each_process(t) { |
| 393 | if (t->flags & PF_KTHREAD) |
| 394 | continue; |
| 395 | for_each_thread(t, s) |
| 396 | if (s->mm) |
| 397 | set_section_perms(perms, n, set: true, mm: s->mm); |
| 398 | } |
| 399 | set_section_perms(perms, n, set: true, current->active_mm); |
| 400 | set_section_perms(perms, n, set: true, mm: &init_mm); |
| 401 | } |
| 402 | |
| 403 | static int __fix_kernmem_perms(void *unused) |
| 404 | { |
| 405 | update_sections_early(nx_perms, ARRAY_SIZE(nx_perms)); |
| 406 | return 0; |
| 407 | } |
| 408 | |
| 409 | static void fix_kernmem_perms(void) |
| 410 | { |
| 411 | stop_machine(fn: __fix_kernmem_perms, NULL, NULL); |
| 412 | } |
| 413 | |
| 414 | static int __mark_rodata_ro(void *unused) |
| 415 | { |
| 416 | update_sections_early(ro_perms, ARRAY_SIZE(ro_perms)); |
| 417 | return 0; |
| 418 | } |
| 419 | |
| 420 | void mark_rodata_ro(void) |
| 421 | { |
| 422 | stop_machine(fn: __mark_rodata_ro, NULL, NULL); |
| 423 | arm_debug_checkwx(); |
| 424 | } |
| 425 | |
| 426 | #else |
| 427 | static inline void fix_kernmem_perms(void) { } |
| 428 | #endif /* CONFIG_STRICT_KERNEL_RWX */ |
| 429 | |
| 430 | void free_initmem(void) |
| 431 | { |
| 432 | fix_kernmem_perms(); |
| 433 | |
| 434 | poison_init_mem(s: __init_begin, count: __init_end - __init_begin); |
| 435 | if (!machine_is_integrator() && !machine_is_cintegrator()) |
| 436 | free_initmem_default(poison: -1); |
| 437 | } |
| 438 | |
| 439 | #ifdef CONFIG_BLK_DEV_INITRD |
| 440 | void free_initrd_mem(unsigned long start, unsigned long end) |
| 441 | { |
| 442 | if (start == initrd_start) |
| 443 | start = round_down(start, PAGE_SIZE); |
| 444 | if (end == initrd_end) |
| 445 | end = round_up(end, PAGE_SIZE); |
| 446 | |
| 447 | poison_init_mem(s: (void *)start, PAGE_ALIGN(end) - start); |
| 448 | free_reserved_area(start: (void *)start, end: (void *)end, poison: -1, s: "initrd" ); |
| 449 | } |
| 450 | #endif |
| 451 | |
| 452 | #ifdef CONFIG_EXECMEM |
| 453 | |
| 454 | #ifdef CONFIG_XIP_KERNEL |
| 455 | /* |
| 456 | * The XIP kernel text is mapped in the module area for modules and |
| 457 | * some other stuff to work without any indirect relocations. |
| 458 | * MODULES_VADDR is redefined here and not in asm/memory.h to avoid |
| 459 | * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off. |
| 460 | */ |
| 461 | #undef MODULES_VADDR |
| 462 | #define MODULES_VADDR (((unsigned long)_exiprom + ~PMD_MASK) & PMD_MASK) |
| 463 | #endif |
| 464 | |
| 465 | #ifdef CONFIG_MMU |
| 466 | static struct execmem_info execmem_info __ro_after_init; |
| 467 | |
| 468 | struct execmem_info __init *execmem_arch_setup(void) |
| 469 | { |
| 470 | unsigned long fallback_start = 0, fallback_end = 0; |
| 471 | |
| 472 | if (IS_ENABLED(CONFIG_ARM_MODULE_PLTS)) { |
| 473 | fallback_start = VMALLOC_START; |
| 474 | fallback_end = VMALLOC_END; |
| 475 | } |
| 476 | |
| 477 | execmem_info = (struct execmem_info){ |
| 478 | .ranges = { |
| 479 | [EXECMEM_DEFAULT] = { |
| 480 | .start = MODULES_VADDR, |
| 481 | .end = MODULES_END, |
| 482 | .pgprot = PAGE_KERNEL_EXEC, |
| 483 | .alignment = 1, |
| 484 | .fallback_start = fallback_start, |
| 485 | .fallback_end = fallback_end, |
| 486 | }, |
| 487 | }, |
| 488 | }; |
| 489 | |
| 490 | return &execmem_info; |
| 491 | } |
| 492 | #endif /* CONFIG_MMU */ |
| 493 | |
| 494 | #endif /* CONFIG_EXECMEM */ |
| 495 | |