1 /* 2 * linux/arch/arm/mm/init.c 3 * 4 * Copyright (C) 1995-2005 Russell King 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License version 2 as 8 * published by the Free Software Foundation. 9 */ 10 #include <linux/kernel.h> 11 #include <linux/errno.h> 12 #include <linux/swap.h> 13 #include <linux/init.h> 14 #include <linux/mman.h> 15 #include <linux/sched/signal.h> 16 #include <linux/sched/task.h> 17 #include <linux/export.h> 18 #include <linux/nodemask.h> 19 #include <linux/initrd.h> 20 #include <linux/of_fdt.h> 21 #include <linux/highmem.h> 22 #include <linux/gfp.h> 23 #include <linux/memblock.h> 24 #include <linux/dma-contiguous.h> 25 #include <linux/sizes.h> 26 #include <linux/stop_machine.h> 27 28 #include <asm/cp15.h> 29 #include <asm/mach-types.h> 30 #include <asm/memblock.h> 31 #include <asm/memory.h> 32 #include <asm/prom.h> 33 #include <asm/sections.h> 34 #include <asm/setup.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 97 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole, 98 unsigned long dma_size) 99 { 100 if (size[0] <= dma_size) 101 return; 102 103 size[ZONE_NORMAL] = size[0] - dma_size; 104 size[ZONE_DMA] = dma_size; 105 hole[ZONE_NORMAL] = hole[0]; 106 hole[ZONE_DMA] = 0; 107 } 108 #endif 109 110 void __init setup_dma_zone(const struct machine_desc *mdesc) 111 { 112 #ifdef CONFIG_ZONE_DMA 113 if (mdesc->dma_zone_size) { 114 arm_dma_zone_size = mdesc->dma_zone_size; 115 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1; 116 } else 117 arm_dma_limit = 0xffffffff; 118 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT; 119 #endif 120 } 121 122 static void __init zone_sizes_init(unsigned long min, unsigned long max_low, 123 unsigned long max_high) 124 { 125 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES]; 126 struct memblock_region *reg; 127 128 /* 129 * initialise the zones. 130 */ 131 memset(zone_size, 0, sizeof(zone_size)); 132 133 /* 134 * The memory size has already been determined. If we need 135 * to do anything fancy with the allocation of this memory 136 * to the zones, now is the time to do it. 137 */ 138 zone_size[0] = max_low - min; 139 #ifdef CONFIG_HIGHMEM 140 zone_size[ZONE_HIGHMEM] = max_high - max_low; 141 #endif 142 143 /* 144 * Calculate the size of the holes. 145 * holes = node_size - sum(bank_sizes) 146 */ 147 memcpy(zhole_size, zone_size, sizeof(zhole_size)); 148 for_each_memblock(memory, reg) { 149 unsigned long start = memblock_region_memory_base_pfn(reg); 150 unsigned long end = memblock_region_memory_end_pfn(reg); 151 152 if (start < max_low) { 153 unsigned long low_end = min(end, max_low); 154 zhole_size[0] -= low_end - start; 155 } 156 #ifdef CONFIG_HIGHMEM 157 if (end > max_low) { 158 unsigned long high_start = max(start, max_low); 159 zhole_size[ZONE_HIGHMEM] -= end - high_start; 160 } 161 #endif 162 } 163 164 #ifdef CONFIG_ZONE_DMA 165 /* 166 * Adjust the sizes according to any special requirements for 167 * this machine type. 168 */ 169 if (arm_dma_zone_size) 170 arm_adjust_dma_zone(zone_size, zhole_size, 171 arm_dma_zone_size >> PAGE_SHIFT); 172 #endif 173 174 free_area_init_node(0, zone_size, min, zhole_size); 175 } 176 177 #ifdef CONFIG_HAVE_ARCH_PFN_VALID 178 int pfn_valid(unsigned long pfn) 179 { 180 return memblock_is_map_memory(__pfn_to_phys(pfn)); 181 } 182 EXPORT_SYMBOL(pfn_valid); 183 #endif 184 185 #ifndef CONFIG_SPARSEMEM 186 static void __init arm_memory_present(void) 187 { 188 } 189 #else 190 static void __init arm_memory_present(void) 191 { 192 struct memblock_region *reg; 193 194 for_each_memblock(memory, reg) 195 memory_present(0, memblock_region_memory_base_pfn(reg), 196 memblock_region_memory_end_pfn(reg)); 197 } 198 #endif 199 200 static bool arm_memblock_steal_permitted = true; 201 202 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align) 203 { 204 phys_addr_t phys; 205 206 BUG_ON(!arm_memblock_steal_permitted); 207 208 phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE); 209 memblock_free(phys, size); 210 memblock_remove(phys, size); 211 212 return phys; 213 } 214 215 static void __init arm_initrd_init(void) 216 { 217 #ifdef CONFIG_BLK_DEV_INITRD 218 phys_addr_t start; 219 unsigned long size; 220 221 initrd_start = initrd_end = 0; 222 223 if (!phys_initrd_size) 224 return; 225 226 /* 227 * Round the memory region to page boundaries as per free_initrd_mem() 228 * This allows us to detect whether the pages overlapping the initrd 229 * are in use, but more importantly, reserves the entire set of pages 230 * as we don't want these pages allocated for other purposes. 231 */ 232 start = round_down(phys_initrd_start, PAGE_SIZE); 233 size = phys_initrd_size + (phys_initrd_start - start); 234 size = round_up(size, PAGE_SIZE); 235 236 if (!memblock_is_region_memory(start, size)) { 237 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n", 238 (u64)start, size); 239 return; 240 } 241 242 if (memblock_is_region_reserved(start, size)) { 243 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n", 244 (u64)start, size); 245 return; 246 } 247 248 memblock_reserve(start, size); 249 250 /* Now convert initrd to virtual addresses */ 251 initrd_start = __phys_to_virt(phys_initrd_start); 252 initrd_end = initrd_start + phys_initrd_size; 253 #endif 254 } 255 256 void __init arm_memblock_init(const struct machine_desc *mdesc) 257 { 258 /* Register the kernel text, kernel data and initrd with memblock. */ 259 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START); 260 261 arm_initrd_init(); 262 263 arm_mm_memblock_reserve(); 264 265 /* reserve any platform specific memblock areas */ 266 if (mdesc->reserve) 267 mdesc->reserve(); 268 269 early_init_fdt_reserve_self(); 270 early_init_fdt_scan_reserved_mem(); 271 272 /* reserve memory for DMA contiguous allocations */ 273 dma_contiguous_reserve(arm_dma_limit); 274 275 arm_memblock_steal_permitted = false; 276 memblock_dump_all(); 277 } 278 279 void __init bootmem_init(void) 280 { 281 memblock_allow_resize(); 282 283 find_limits(&min_low_pfn, &max_low_pfn, &max_pfn); 284 285 early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT, 286 (phys_addr_t)max_low_pfn << PAGE_SHIFT); 287 288 /* 289 * Sparsemem tries to allocate bootmem in memory_present(), 290 * so must be done after the fixed reservations 291 */ 292 arm_memory_present(); 293 294 /* 295 * sparse_init() needs the bootmem allocator up and running. 296 */ 297 sparse_init(); 298 299 /* 300 * Now free the memory - free_area_init_node needs 301 * the sparse mem_map arrays initialized by sparse_init() 302 * for memmap_init_zone(), otherwise all PFNs are invalid. 303 */ 304 zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn); 305 } 306 307 /* 308 * Poison init memory with an undefined instruction (ARM) or a branch to an 309 * undefined instruction (Thumb). 310 */ 311 static inline void poison_init_mem(void *s, size_t count) 312 { 313 u32 *p = (u32 *)s; 314 for (; count != 0; count -= 4) 315 *p++ = 0xe7fddef0; 316 } 317 318 static inline void 319 free_memmap(unsigned long start_pfn, unsigned long end_pfn) 320 { 321 struct page *start_pg, *end_pg; 322 phys_addr_t pg, pgend; 323 324 /* 325 * Convert start_pfn/end_pfn to a struct page pointer. 326 */ 327 start_pg = pfn_to_page(start_pfn - 1) + 1; 328 end_pg = pfn_to_page(end_pfn - 1) + 1; 329 330 /* 331 * Convert to physical addresses, and 332 * round start upwards and end downwards. 333 */ 334 pg = PAGE_ALIGN(__pa(start_pg)); 335 pgend = __pa(end_pg) & PAGE_MASK; 336 337 /* 338 * If there are free pages between these, 339 * free the section of the memmap array. 340 */ 341 if (pg < pgend) 342 memblock_free_early(pg, pgend - pg); 343 } 344 345 /* 346 * The mem_map array can get very big. Free the unused area of the memory map. 347 */ 348 static void __init free_unused_memmap(void) 349 { 350 unsigned long start, prev_end = 0; 351 struct memblock_region *reg; 352 353 /* 354 * This relies on each bank being in address order. 355 * The banks are sorted previously in bootmem_init(). 356 */ 357 for_each_memblock(memory, reg) { 358 start = memblock_region_memory_base_pfn(reg); 359 360 #ifdef CONFIG_SPARSEMEM 361 /* 362 * Take care not to free memmap entries that don't exist 363 * due to SPARSEMEM sections which aren't present. 364 */ 365 start = min(start, 366 ALIGN(prev_end, PAGES_PER_SECTION)); 367 #else 368 /* 369 * Align down here since the VM subsystem insists that the 370 * memmap entries are valid from the bank start aligned to 371 * MAX_ORDER_NR_PAGES. 372 */ 373 start = round_down(start, MAX_ORDER_NR_PAGES); 374 #endif 375 /* 376 * If we had a previous bank, and there is a space 377 * between the current bank and the previous, free it. 378 */ 379 if (prev_end && prev_end < start) 380 free_memmap(prev_end, start); 381 382 /* 383 * Align up here since the VM subsystem insists that the 384 * memmap entries are valid from the bank end aligned to 385 * MAX_ORDER_NR_PAGES. 386 */ 387 prev_end = ALIGN(memblock_region_memory_end_pfn(reg), 388 MAX_ORDER_NR_PAGES); 389 } 390 391 #ifdef CONFIG_SPARSEMEM 392 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) 393 free_memmap(prev_end, 394 ALIGN(prev_end, PAGES_PER_SECTION)); 395 #endif 396 } 397 398 #ifdef CONFIG_HIGHMEM 399 static inline void free_area_high(unsigned long pfn, unsigned long end) 400 { 401 for (; pfn < end; pfn++) 402 free_highmem_page(pfn_to_page(pfn)); 403 } 404 #endif 405 406 static void __init free_highpages(void) 407 { 408 #ifdef CONFIG_HIGHMEM 409 unsigned long max_low = max_low_pfn; 410 struct memblock_region *mem, *res; 411 412 /* set highmem page free */ 413 for_each_memblock(memory, mem) { 414 unsigned long start = memblock_region_memory_base_pfn(mem); 415 unsigned long end = memblock_region_memory_end_pfn(mem); 416 417 /* Ignore complete lowmem entries */ 418 if (end <= max_low) 419 continue; 420 421 if (memblock_is_nomap(mem)) 422 continue; 423 424 /* Truncate partial highmem entries */ 425 if (start < max_low) 426 start = max_low; 427 428 /* Find and exclude any reserved regions */ 429 for_each_memblock(reserved, res) { 430 unsigned long res_start, res_end; 431 432 res_start = memblock_region_reserved_base_pfn(res); 433 res_end = memblock_region_reserved_end_pfn(res); 434 435 if (res_end < start) 436 continue; 437 if (res_start < start) 438 res_start = start; 439 if (res_start > end) 440 res_start = end; 441 if (res_end > end) 442 res_end = end; 443 if (res_start != start) 444 free_area_high(start, res_start); 445 start = res_end; 446 if (start == end) 447 break; 448 } 449 450 /* And now free anything which remains */ 451 if (start < end) 452 free_area_high(start, end); 453 } 454 #endif 455 } 456 457 /* 458 * mem_init() marks the free areas in the mem_map and tells us how much 459 * memory is free. This is done after various parts of the system have 460 * claimed their memory after the kernel image. 461 */ 462 void __init mem_init(void) 463 { 464 #ifdef CONFIG_HAVE_TCM 465 /* These pointers are filled in on TCM detection */ 466 extern u32 dtcm_end; 467 extern u32 itcm_end; 468 #endif 469 470 set_max_mapnr(pfn_to_page(max_pfn) - mem_map); 471 472 /* this will put all unused low memory onto the freelists */ 473 free_unused_memmap(); 474 memblock_free_all(); 475 476 #ifdef CONFIG_SA1111 477 /* now that our DMA memory is actually so designated, we can free it */ 478 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL); 479 #endif 480 481 free_highpages(); 482 483 mem_init_print_info(NULL); 484 485 /* 486 * Check boundaries twice: Some fundamental inconsistencies can 487 * be detected at build time already. 488 */ 489 #ifdef CONFIG_MMU 490 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR); 491 BUG_ON(TASK_SIZE > MODULES_VADDR); 492 #endif 493 494 #ifdef CONFIG_HIGHMEM 495 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET); 496 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET); 497 #endif 498 } 499 500 #ifdef CONFIG_STRICT_KERNEL_RWX 501 struct section_perm { 502 const char *name; 503 unsigned long start; 504 unsigned long end; 505 pmdval_t mask; 506 pmdval_t prot; 507 pmdval_t clear; 508 }; 509 510 /* First section-aligned location at or after __start_rodata. */ 511 extern char __start_rodata_section_aligned[]; 512 513 static struct section_perm nx_perms[] = { 514 /* Make pages tables, etc before _stext RW (set NX). */ 515 { 516 .name = "pre-text NX", 517 .start = PAGE_OFFSET, 518 .end = (unsigned long)_stext, 519 .mask = ~PMD_SECT_XN, 520 .prot = PMD_SECT_XN, 521 }, 522 /* Make init RW (set NX). */ 523 { 524 .name = "init NX", 525 .start = (unsigned long)__init_begin, 526 .end = (unsigned long)_sdata, 527 .mask = ~PMD_SECT_XN, 528 .prot = PMD_SECT_XN, 529 }, 530 /* Make rodata NX (set RO in ro_perms below). */ 531 { 532 .name = "rodata NX", 533 .start = (unsigned long)__start_rodata_section_aligned, 534 .end = (unsigned long)__init_begin, 535 .mask = ~PMD_SECT_XN, 536 .prot = PMD_SECT_XN, 537 }, 538 }; 539 540 static struct section_perm ro_perms[] = { 541 /* Make kernel code and rodata RX (set RO). */ 542 { 543 .name = "text/rodata RO", 544 .start = (unsigned long)_stext, 545 .end = (unsigned long)__init_begin, 546 #ifdef CONFIG_ARM_LPAE 547 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2), 548 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2, 549 #else 550 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE), 551 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE, 552 .clear = PMD_SECT_AP_WRITE, 553 #endif 554 }, 555 }; 556 557 /* 558 * Updates section permissions only for the current mm (sections are 559 * copied into each mm). During startup, this is the init_mm. Is only 560 * safe to be called with preemption disabled, as under stop_machine(). 561 */ 562 static inline void section_update(unsigned long addr, pmdval_t mask, 563 pmdval_t prot, struct mm_struct *mm) 564 { 565 pmd_t *pmd; 566 567 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr); 568 569 #ifdef CONFIG_ARM_LPAE 570 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot); 571 #else 572 if (addr & SECTION_SIZE) 573 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot); 574 else 575 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot); 576 #endif 577 flush_pmd_entry(pmd); 578 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE); 579 } 580 581 /* Make sure extended page tables are in use. */ 582 static inline bool arch_has_strict_perms(void) 583 { 584 if (cpu_architecture() < CPU_ARCH_ARMv6) 585 return false; 586 587 return !!(get_cr() & CR_XP); 588 } 589 590 void set_section_perms(struct section_perm *perms, int n, bool set, 591 struct mm_struct *mm) 592 { 593 size_t i; 594 unsigned long addr; 595 596 if (!arch_has_strict_perms()) 597 return; 598 599 for (i = 0; i < n; i++) { 600 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) || 601 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) { 602 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n", 603 perms[i].name, perms[i].start, perms[i].end, 604 SECTION_SIZE); 605 continue; 606 } 607 608 for (addr = perms[i].start; 609 addr < perms[i].end; 610 addr += SECTION_SIZE) 611 section_update(addr, perms[i].mask, 612 set ? perms[i].prot : perms[i].clear, mm); 613 } 614 615 } 616 617 /** 618 * update_sections_early intended to be called only through stop_machine 619 * framework and executed by only one CPU while all other CPUs will spin and 620 * wait, so no locking is required in this function. 621 */ 622 static void update_sections_early(struct section_perm perms[], int n) 623 { 624 struct task_struct *t, *s; 625 626 for_each_process(t) { 627 if (t->flags & PF_KTHREAD) 628 continue; 629 for_each_thread(t, s) 630 set_section_perms(perms, n, true, s->mm); 631 } 632 set_section_perms(perms, n, true, current->active_mm); 633 set_section_perms(perms, n, true, &init_mm); 634 } 635 636 static int __fix_kernmem_perms(void *unused) 637 { 638 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms)); 639 return 0; 640 } 641 642 static void fix_kernmem_perms(void) 643 { 644 stop_machine(__fix_kernmem_perms, NULL, NULL); 645 } 646 647 static int __mark_rodata_ro(void *unused) 648 { 649 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms)); 650 return 0; 651 } 652 653 static int kernel_set_to_readonly __read_mostly; 654 655 void mark_rodata_ro(void) 656 { 657 kernel_set_to_readonly = 1; 658 stop_machine(__mark_rodata_ro, NULL, NULL); 659 debug_checkwx(); 660 } 661 662 void set_kernel_text_rw(void) 663 { 664 if (!kernel_set_to_readonly) 665 return; 666 667 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false, 668 current->active_mm); 669 } 670 671 void set_kernel_text_ro(void) 672 { 673 if (!kernel_set_to_readonly) 674 return; 675 676 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true, 677 current->active_mm); 678 } 679 680 #else 681 static inline void fix_kernmem_perms(void) { } 682 #endif /* CONFIG_STRICT_KERNEL_RWX */ 683 684 void free_initmem(void) 685 { 686 fix_kernmem_perms(); 687 688 poison_init_mem(__init_begin, __init_end - __init_begin); 689 if (!machine_is_integrator() && !machine_is_cintegrator()) 690 free_initmem_default(-1); 691 } 692 693 #ifdef CONFIG_BLK_DEV_INITRD 694 695 static int keep_initrd; 696 697 void free_initrd_mem(unsigned long start, unsigned long end) 698 { 699 if (!keep_initrd) { 700 if (start == initrd_start) 701 start = round_down(start, PAGE_SIZE); 702 if (end == initrd_end) 703 end = round_up(end, PAGE_SIZE); 704 705 poison_init_mem((void *)start, PAGE_ALIGN(end) - start); 706 free_reserved_area((void *)start, (void *)end, -1, "initrd"); 707 } 708 } 709 710 static int __init keepinitrd_setup(char *__unused) 711 { 712 keep_initrd = 1; 713 return 1; 714 } 715 716 __setup("keepinitrd", keepinitrd_setup); 717 #endif 718