arch/arm64/mm/init.c

   1 /*
   2  * Based on arch/arm/mm/init.c
   3  *
   4  * Copyright (C) 1995-2005 Russell King
   5  * Copyright (C) 2012 ARM Ltd.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include <linux/kernel.h>
  21 #include <linux/export.h>
  22 #include <linux/errno.h>
  23 #include <linux/swap.h>
  24 #include <linux/init.h>
  25 #include <linux/bootmem.h>
  26 #include <linux/cache.h>
  27 #include <linux/mman.h>
  28 #include <linux/nodemask.h>
  29 #include <linux/initrd.h>
  30 #include <linux/gfp.h>
  31 #include <linux/memblock.h>
  32 #include <linux/sort.h>
  33 #include <linux/of_fdt.h>
  34 #include <linux/dma-mapping.h>
  35 #include <linux/dma-contiguous.h>
  36 #include <linux/efi.h>
  37 #include <linux/swiotlb.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/mm.h>
  40
  41 #include <asm/boot.h>
  42 #include <asm/fixmap.h>
  43 #include <asm/kasan.h>
  44 #include <asm/kernel-pgtable.h>
  45 #include <asm/memory.h>
  46 #include <asm/numa.h>
  47 #include <asm/sections.h>
  48 #include <asm/setup.h>
  49 #include <asm/sizes.h>
  50 #include <asm/tlb.h>
  51 #include <asm/alternative.h>
  52
  53 /*
  54  * We need to be able to catch inadvertent references to memstart_addr
  55  * that occur (potentially in generic code) before arm64_memblock_init()
  56  * executes, which assigns it its actual value. So use a default value
  57  * that cannot be mistaken for a real physical address.
  58  */
  59 s64 memstart_addr __ro_after_init = -1;
  60 phys_addr_t arm64_dma_phys_limit __ro_after_init;
  61
  62 #ifdef CONFIG_BLK_DEV_INITRD
  63 static int __init early_initrd(char *p)
  64 {
  65         unsigned long start, size;
  66         char *endp;
  67
  68         start = memparse(p, &endp);
  69         if (*endp == ',') {
  70                 size = memparse(endp + 1, NULL);
  71
  72                 initrd_start = start;
  73                 initrd_end = start + size;
  74         }
  75         return 0;
  76 }
  77 early_param("initrd", early_initrd);
  78 #endif
  79
  80 /*
  81  * Return the maximum physical address for ZONE_DMA (DMA_BIT_MASK(32)). It
  82  * currently assumes that for memory starting above 4G, 32-bit devices will
  83  * use a DMA offset.
  84  */
  85 static phys_addr_t __init max_zone_dma_phys(void)
  86 {
  87         phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
  88         return min(offset + (1ULL << 32), memblock_end_of_DRAM());
  89 }
  90
  91 #ifdef CONFIG_NUMA
  92
  93 static void __init zone_sizes_init(unsigned long min, unsigned long max)
  94 {
  95         unsigned long max_zone_pfns[MAX_NR_ZONES]  = {0};
  96
  97         if (IS_ENABLED(CONFIG_ZONE_DMA))
  98                 max_zone_pfns[ZONE_DMA] = PFN_DOWN(max_zone_dma_phys());
  99         max_zone_pfns[ZONE_NORMAL] = max;
 100
 101         free_area_init_nodes(max_zone_pfns);
 102 }
 103
 104 #else
 105
 106 static void __init zone_sizes_init(unsigned long min, unsigned long max)
 107 {
 108         struct memblock_region *reg;
 109         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
 110         unsigned long max_dma = min;
 111
 112         memset(zone_size, 0, sizeof(zone_size));
 113
 114         /* 4GB maximum for 32-bit only capable devices */
 115 #ifdef CONFIG_ZONE_DMA
 116         max_dma = PFN_DOWN(arm64_dma_phys_limit);
 117         zone_size[ZONE_DMA] = max_dma - min;
 118 #endif
 119         zone_size[ZONE_NORMAL] = max - max_dma;
 120
 121         memcpy(zhole_size, zone_size, sizeof(zhole_size));
 122
 123         for_each_memblock(memory, reg) {
 124                 unsigned long start = memblock_region_memory_base_pfn(reg);
 125                 unsigned long end = memblock_region_memory_end_pfn(reg);
 126
 127                 if (start >= max)
 128                         continue;
 129
 130 #ifdef CONFIG_ZONE_DMA
 131                 if (start < max_dma) {
 132                         unsigned long dma_end = min(end, max_dma);
 133                         zhole_size[ZONE_DMA] -= dma_end - start;
 134                 }
 135 #endif
 136                 if (end > max_dma) {
 137                         unsigned long normal_end = min(end, max);
 138                         unsigned long normal_start = max(start, max_dma);
 139                         zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
 140                 }
 141         }
 142
 143         free_area_init_node(0, zone_size, min, zhole_size);
 144 }
 145
 146 #endif /* CONFIG_NUMA */
 147
 148 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
 149 int pfn_valid(unsigned long pfn)
 150 {
 151         return memblock_is_map_memory(pfn << PAGE_SHIFT);
 152 }
 153 EXPORT_SYMBOL(pfn_valid);
 154 #endif
 155
 156 #ifndef CONFIG_SPARSEMEM
 157 static void __init arm64_memory_present(void)
 158 {
 159 }
 160 #else
 161 static void __init arm64_memory_present(void)
 162 {
 163         struct memblock_region *reg;
 164
 165         for_each_memblock(memory, reg) {
 166                 int nid = memblock_get_region_node(reg);
 167
 168                 memory_present(nid, memblock_region_memory_base_pfn(reg),
 169                                 memblock_region_memory_end_pfn(reg));
 170         }
 171 }
 172 #endif
 173
 174 static phys_addr_t memory_limit = (phys_addr_t)ULLONG_MAX;
 175
 176 /*
 177  * Limit the memory size that was specified via FDT.
 178  */
 179 static int __init early_mem(char *p)
 180 {
 181         if (!p)
 182                 return 1;
 183
 184         memory_limit = memparse(p, &p) & PAGE_MASK;
 185         pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
 186
 187         return 0;
 188 }
 189 early_param("mem", early_mem);
 190
 191 void __init arm64_memblock_init(void)
 192 {
 193         const s64 linear_region_size = -(s64)PAGE_OFFSET;
 194
 195         /*
 196          * Ensure that the linear region takes up exactly half of the kernel
 197          * virtual address space. This way, we can distinguish a linear address
 198          * from a kernel/module/vmalloc address by testing a single bit.
 199          */
 200         BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
 201
 202         /*
 203          * Select a suitable value for the base of physical memory.
 204          */
 205         memstart_addr = round_down(memblock_start_of_DRAM(),
 206                                    ARM64_MEMSTART_ALIGN);
 207
 208         /*
 209          * Remove the memory that we will not be able to cover with the
 210          * linear mapping. Take care not to clip the kernel which may be
 211          * high in memory.
 212          */
 213         memblock_remove(max_t(u64, memstart_addr + linear_region_size,
 214                         __pa_symbol(_end)), ULLONG_MAX);
 215         if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
 216                 /* ensure that memstart_addr remains sufficiently aligned */
 217                 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
 218                                          ARM64_MEMSTART_ALIGN);
 219                 memblock_remove(0, memstart_addr);
 220         }
 221
 222         /*
 223          * Apply the memory limit if it was set. Since the kernel may be loaded
 224          * high up in memory, add back the kernel region that must be accessible
 225          * via the linear mapping.
 226          */
 227         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 228                 memblock_mem_limit_remove_map(memory_limit);
 229                 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
 230         }
 231
 232         if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
 233                 /*
 234                  * Add back the memory we just removed if it results in the
 235                  * initrd to become inaccessible via the linear mapping.
 236                  * Otherwise, this is a no-op
 237                  */
 238                 u64 base = initrd_start & PAGE_MASK;
 239                 u64 size = PAGE_ALIGN(initrd_end) - base;
 240
 241                 /*
 242                  * We can only add back the initrd memory if we don't end up
 243                  * with more memory than we can address via the linear mapping.
 244                  * It is up to the bootloader to position the kernel and the
 245                  * initrd reasonably close to each other (i.e., within 32 GB of
 246                  * each other) so that all granule/#levels combinations can
 247                  * always access both.
 248                  */
 249                 if (WARN(base < memblock_start_of_DRAM() ||
 250                          base + size > memblock_start_of_DRAM() +
 251                                        linear_region_size,
 252                         "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
 253                         initrd_start = 0;
 254                 } else {
 255                         memblock_remove(base, size); /* clear MEMBLOCK_ flags */
 256                         memblock_add(base, size);
 257                         memblock_reserve(base, size);
 258                 }
 259         }
 260
 261         if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
 262                 extern u16 memstart_offset_seed;
 263                 u64 range = linear_region_size -
 264                             (memblock_end_of_DRAM() - memblock_start_of_DRAM());
 265
 266                 /*
 267                  * If the size of the linear region exceeds, by a sufficient
 268                  * margin, the size of the region that the available physical
 269                  * memory spans, randomize the linear region as well.
 270                  */
 271                 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
 272                         range = range / ARM64_MEMSTART_ALIGN + 1;
 273                         memstart_addr -= ARM64_MEMSTART_ALIGN *
 274                                          ((range * memstart_offset_seed) >> 16);
 275                 }
 276         }
 277
 278         /*
 279          * Register the kernel text, kernel data, initrd, and initial
 280          * pagetables with memblock.
 281          */
 282         memblock_reserve(__pa_symbol(_text), _end - _text);
 283 #ifdef CONFIG_BLK_DEV_INITRD
 284         if (initrd_start) {
 285                 memblock_reserve(initrd_start, initrd_end - initrd_start);
 286
 287                 /* the generic initrd code expects virtual addresses */
 288                 initrd_start = __phys_to_virt(initrd_start);
 289                 initrd_end = __phys_to_virt(initrd_end);
 290         }
 291 #endif
 292
 293         early_init_fdt_scan_reserved_mem();
 294
 295         /* 4GB maximum for 32-bit only capable devices */
 296         if (IS_ENABLED(CONFIG_ZONE_DMA))
 297                 arm64_dma_phys_limit = max_zone_dma_phys();
 298         else
 299                 arm64_dma_phys_limit = PHYS_MASK + 1;
 300         dma_contiguous_reserve(arm64_dma_phys_limit);
 301
 302         memblock_allow_resize();
 303 }
 304
 305 void __init bootmem_init(void)
 306 {
 307         unsigned long min, max;
 308
 309         min = PFN_UP(memblock_start_of_DRAM());
 310         max = PFN_DOWN(memblock_end_of_DRAM());
 311
 312         early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
 313
 314         max_pfn = max_low_pfn = max;
 315
 316         arm64_numa_init();
 317         /*
 318          * Sparsemem tries to allocate bootmem in memory_present(), so must be
 319          * done after the fixed reservations.
 320          */
 321         arm64_memory_present();
 322
 323         sparse_init();
 324         zone_sizes_init(min, max);
 325
 326         high_memory = __va((max << PAGE_SHIFT) - 1) + 1;
 327         memblock_dump_all();
 328 }
 329
 330 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 331 static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
 332 {
 333         struct page *start_pg, *end_pg;
 334         unsigned long pg, pgend;
 335
 336         /*
 337          * Convert start_pfn/end_pfn to a struct page pointer.
 338          */
 339         start_pg = pfn_to_page(start_pfn - 1) + 1;
 340         end_pg = pfn_to_page(end_pfn - 1) + 1;
 341
 342         /*
 343          * Convert to physical addresses, and round start upwards and end
 344          * downwards.
 345          */
 346         pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
 347         pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
 348
 349         /*
 350          * If there are free pages between these, free the section of the
 351          * memmap array.
 352          */
 353         if (pg < pgend)
 354                 free_bootmem(pg, pgend - pg);
 355 }
 356
 357 /*
 358  * The mem_map array can get very big. Free the unused area of the memory map.
 359  */
 360 static void __init free_unused_memmap(void)
 361 {
 362         unsigned long start, prev_end = 0;
 363         struct memblock_region *reg;
 364
 365         for_each_memblock(memory, reg) {
 366                 start = __phys_to_pfn(reg->base);
 367
 368 #ifdef CONFIG_SPARSEMEM
 369                 /*
 370                  * Take care not to free memmap entries that don't exist due
 371                  * to SPARSEMEM sections which aren't present.
 372                  */
 373                 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
 374 #endif
 375                 /*
 376                  * If we had a previous bank, and there is a space between the
 377                  * 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(__phys_to_pfn(reg->base + reg->size),
 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, ALIGN(prev_end, PAGES_PER_SECTION));
 394 #endif
 395 }
 396 #endif  /* !CONFIG_SPARSEMEM_VMEMMAP */
 397
 398 /*
 399  * mem_init() marks the free areas in the mem_map and tells us how much memory
 400  * is free.  This is done after various parts of the system have claimed their
 401  * memory after the kernel image.
 402  */
 403 void __init mem_init(void)
 404 {
 405         if (swiotlb_force == SWIOTLB_FORCE ||
 406             max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT))
 407                 swiotlb_init(1);
 408         else
 409                 swiotlb_force = SWIOTLB_NO_FORCE;
 410
 411         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 412
 413 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 414         free_unused_memmap();
 415 #endif
 416         /* this will put all unused low memory onto the freelists */
 417         free_all_bootmem();
 418
 419         mem_init_print_info(NULL);
 420
 421 #define MLK(b, t) b, t, ((t) - (b)) >> 10
 422 #define MLM(b, t) b, t, ((t) - (b)) >> 20
 423 #define MLG(b, t) b, t, ((t) - (b)) >> 30
 424 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
 425
 426         pr_notice("Virtual kernel memory layout:\n");
 427 #ifdef CONFIG_KASAN
 428         pr_notice("    kasan   : 0x%16lx - 0x%16lx   (%6ld GB)\n",
 429                 MLG(KASAN_SHADOW_START, KASAN_SHADOW_END));
 430 #endif
 431         pr_notice("    modules : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 432                 MLM(MODULES_VADDR, MODULES_END));
 433         pr_notice("    vmalloc : 0x%16lx - 0x%16lx   (%6ld GB)\n",
 434                 MLG(VMALLOC_START, VMALLOC_END));
 435         pr_notice("      .text : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 436                 MLK_ROUNDUP(_text, _etext));
 437         pr_notice("    .rodata : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 438                 MLK_ROUNDUP(__start_rodata, __init_begin));
 439         pr_notice("      .init : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 440                 MLK_ROUNDUP(__init_begin, __init_end));
 441         pr_notice("      .data : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 442                 MLK_ROUNDUP(_sdata, _edata));
 443         pr_notice("       .bss : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 444                 MLK_ROUNDUP(__bss_start, __bss_stop));
 445         pr_notice("    fixed   : 0x%16lx - 0x%16lx   (%6ld KB)\n",
 446                 MLK(FIXADDR_START, FIXADDR_TOP));
 447         pr_notice("    PCI I/O : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 448                 MLM(PCI_IO_START, PCI_IO_END));
 449 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 450         pr_notice("    vmemmap : 0x%16lx - 0x%16lx   (%6ld GB maximum)\n",
 451                 MLG(VMEMMAP_START, VMEMMAP_START + VMEMMAP_SIZE));
 452         pr_notice("              0x%16lx - 0x%16lx   (%6ld MB actual)\n",
 453                 MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
 454                     (unsigned long)virt_to_page(high_memory)));
 455 #endif
 456         pr_notice("    memory  : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 457                 MLM(__phys_to_virt(memblock_start_of_DRAM()),
 458                     (unsigned long)high_memory));
 459
 460 #undef MLK
 461 #undef MLM
 462 #undef MLK_ROUNDUP
 463
 464         /*
 465          * Check boundaries twice: Some fundamental inconsistencies can be
 466          * detected at build time already.
 467          */
 468 #ifdef CONFIG_COMPAT
 469         BUILD_BUG_ON(TASK_SIZE_32                       > TASK_SIZE_64);
 470 #endif
 471
 472         /*
 473          * Make sure we chose the upper bound of sizeof(struct page)
 474          * correctly.
 475          */
 476         BUILD_BUG_ON(sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT));
 477
 478         if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
 479                 extern int sysctl_overcommit_memory;
 480                 /*
 481                  * On a machine this small we won't get anywhere without
 482                  * overcommit, so turn it on by default.
 483                  */
 484                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 485         }
 486 }
 487
 488 void free_initmem(void)
 489 {
 490         free_reserved_area(lm_alias(__init_begin),
 491                            lm_alias(__init_end),
 492                            0, "unused kernel");
 493         /*
 494          * Unmap the __init region but leave the VM area in place. This
 495          * prevents the region from being reused for kernel modules, which
 496          * is not supported by kallsyms.
 497          */
 498         unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
 499 }
 500
 501 #ifdef CONFIG_BLK_DEV_INITRD
 502
 503 static int keep_initrd __initdata;
 504
 505 void __init free_initrd_mem(unsigned long start, unsigned long end)
 506 {
 507         if (!keep_initrd)
 508                 free_reserved_area((void *)start, (void *)end, 0, "initrd");
 509 }
 510
 511 static int __init keepinitrd_setup(char *__unused)
 512 {
 513         keep_initrd = 1;
 514         return 1;
 515 }
 516
 517 __setup("keepinitrd", keepinitrd_setup);
 518 #endif
 519
 520 /*
 521  * Dump out memory limit information on panic.
 522  */
 523 static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
 524 {
 525         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 526                 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
 527         } else {
 528                 pr_emerg("Memory Limit: none\n");
 529         }
 530         return 0;
 531 }
 532
 533 static struct notifier_block mem_limit_notifier = {
 534         .notifier_call = dump_mem_limit,
 535 };
 536
 537 static int __init register_mem_limit_dumper(void)
 538 {
 539         atomic_notifier_chain_register(&panic_notifier_list,
 540                                        &mem_limit_notifier);
 541         return 0;
 542 }
 543 __initcall(register_mem_limit_dumper);