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
  40 #include <asm/boot.h>
  41 #include <asm/fixmap.h>
  42 #include <asm/kasan.h>
  43 #include <asm/kernel-pgtable.h>
  44 #include <asm/memory.h>
  45 #include <asm/numa.h>
  46 #include <asm/sections.h>
  47 #include <asm/setup.h>
  48 #include <asm/sizes.h>
  49 #include <asm/tlb.h>
  50 #include <asm/alternative.h>
  51
  52 /*
  53  * We need to be able to catch inadvertent references to memstart_addr
  54  * that occur (potentially in generic code) before arm64_memblock_init()
  55  * executes, which assigns it its actual value. So use a default value
  56  * that cannot be mistaken for a real physical address.
  57  */
  58 s64 memstart_addr __ro_after_init = -1;
  59 phys_addr_t arm64_dma_phys_limit __ro_after_init;
  60
  61 #ifdef CONFIG_BLK_DEV_INITRD
  62 static int __init early_initrd(char *p)
  63 {
  64         unsigned long start, size;
  65         char *endp;
  66
  67         start = memparse(p, &endp);
  68         if (*endp == ',') {
  69                 size = memparse(endp + 1, NULL);
  70
  71                 initrd_start = start;
  72                 initrd_end = start + size;
  73         }
  74         return 0;
  75 }
  76 early_param("initrd", early_initrd);
  77 #endif
  78
  79 /*
  80  * Return the maximum physical address for ZONE_DMA (DMA_BIT_MASK(32)). It
  81  * currently assumes that for memory starting above 4G, 32-bit devices will
  82  * use a DMA offset.
  83  */
  84 static phys_addr_t __init max_zone_dma_phys(void)
  85 {
  86         phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
  87         return min(offset + (1ULL << 32), memblock_end_of_DRAM());
  88 }
  89
  90 #ifdef CONFIG_NUMA
  91
  92 static void __init zone_sizes_init(unsigned long min, unsigned long max)
  93 {
  94         unsigned long max_zone_pfns[MAX_NR_ZONES]  = {0};
  95
  96         if (IS_ENABLED(CONFIG_ZONE_DMA))
  97                 max_zone_pfns[ZONE_DMA] = PFN_DOWN(max_zone_dma_phys());
  98         max_zone_pfns[ZONE_NORMAL] = max;
  99
 100         free_area_init_nodes(max_zone_pfns);
 101 }
 102
 103 #else
 104
 105 static void __init zone_sizes_init(unsigned long min, unsigned long max)
 106 {
 107         struct memblock_region *reg;
 108         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
 109         unsigned long max_dma = min;
 110
 111         memset(zone_size, 0, sizeof(zone_size));
 112
 113         /* 4GB maximum for 32-bit only capable devices */
 114 #ifdef CONFIG_ZONE_DMA
 115         max_dma = PFN_DOWN(arm64_dma_phys_limit);
 116         zone_size[ZONE_DMA] = max_dma - min;
 117 #endif
 118         zone_size[ZONE_NORMAL] = max - max_dma;
 119
 120         memcpy(zhole_size, zone_size, sizeof(zhole_size));
 121
 122         for_each_memblock(memory, reg) {
 123                 unsigned long start = memblock_region_memory_base_pfn(reg);
 124                 unsigned long end = memblock_region_memory_end_pfn(reg);
 125
 126                 if (start >= max)
 127                         continue;
 128
 129 #ifdef CONFIG_ZONE_DMA
 130                 if (start < max_dma) {
 131                         unsigned long dma_end = min(end, max_dma);
 132                         zhole_size[ZONE_DMA] -= dma_end - start;
 133                 }
 134 #endif
 135                 if (end > max_dma) {
 136                         unsigned long normal_end = min(end, max);
 137                         unsigned long normal_start = max(start, max_dma);
 138                         zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
 139                 }
 140         }
 141
 142         free_area_init_node(0, zone_size, min, zhole_size);
 143 }
 144
 145 #endif /* CONFIG_NUMA */
 146
 147 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
 148 int pfn_valid(unsigned long pfn)
 149 {
 150         return memblock_is_map_memory(pfn << PAGE_SHIFT);
 151 }
 152 EXPORT_SYMBOL(pfn_valid);
 153 #endif
 154
 155 #ifndef CONFIG_SPARSEMEM
 156 static void __init arm64_memory_present(void)
 157 {
 158 }
 159 #else
 160 static void __init arm64_memory_present(void)
 161 {
 162         struct memblock_region *reg;
 163
 164         for_each_memblock(memory, reg) {
 165                 int nid = memblock_get_region_node(reg);
 166
 167                 memory_present(nid, memblock_region_memory_base_pfn(reg),
 168                                 memblock_region_memory_end_pfn(reg));
 169         }
 170 }
 171 #endif
 172
 173 static phys_addr_t memory_limit = (phys_addr_t)ULLONG_MAX;
 174
 175 /*
 176  * Limit the memory size that was specified via FDT.
 177  */
 178 static int __init early_mem(char *p)
 179 {
 180         if (!p)
 181                 return 1;
 182
 183         memory_limit = memparse(p, &p) & PAGE_MASK;
 184         pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
 185
 186         return 0;
 187 }
 188 early_param("mem", early_mem);
 189
 190 void __init arm64_memblock_init(void)
 191 {
 192         const s64 linear_region_size = -(s64)PAGE_OFFSET;
 193
 194         /*
 195          * Ensure that the linear region takes up exactly half of the kernel
 196          * virtual address space. This way, we can distinguish a linear address
 197          * from a kernel/module/vmalloc address by testing a single bit.
 198          */
 199         BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
 200
 201         /*
 202          * Select a suitable value for the base of physical memory.
 203          */
 204         memstart_addr = round_down(memblock_start_of_DRAM(),
 205                                    ARM64_MEMSTART_ALIGN);
 206
 207         /*
 208          * Remove the memory that we will not be able to cover with the
 209          * linear mapping. Take care not to clip the kernel which may be
 210          * high in memory.
 211          */
 212         memblock_remove(max_t(u64, memstart_addr + linear_region_size, __pa(_end)),
 213                         ULLONG_MAX);
 214         if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
 215                 /* ensure that memstart_addr remains sufficiently aligned */
 216                 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
 217                                          ARM64_MEMSTART_ALIGN);
 218                 memblock_remove(0, memstart_addr);
 219         }
 220
 221         /*
 222          * Apply the memory limit if it was set. Since the kernel may be loaded
 223          * high up in memory, add back the kernel region that must be accessible
 224          * via the linear mapping.
 225          */
 226         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 227                 memblock_mem_limit_remove_map(memory_limit);
 228                 memblock_add(__pa(_text), (u64)(_end - _text));
 229         }
 230
 231         if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
 232                 /*
 233                  * Add back the memory we just removed if it results in the
 234                  * initrd to become inaccessible via the linear mapping.
 235                  * Otherwise, this is a no-op
 236                  */
 237                 u64 base = initrd_start & PAGE_MASK;
 238                 u64 size = PAGE_ALIGN(initrd_end) - base;
 239
 240                 /*
 241                  * We can only add back the initrd memory if we don't end up
 242                  * with more memory than we can address via the linear mapping.
 243                  * It is up to the bootloader to position the kernel and the
 244                  * initrd reasonably close to each other (i.e., within 32 GB of
 245                  * each other) so that all granule/#levels combinations can
 246                  * always access both.
 247                  */
 248                 if (WARN(base < memblock_start_of_DRAM() ||
 249                          base + size > memblock_start_of_DRAM() +
 250                                        linear_region_size,
 251                         "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
 252                         initrd_start = 0;
 253                 } else {
 254                         memblock_remove(base, size); /* clear MEMBLOCK_ flags */
 255                         memblock_add(base, size);
 256                         memblock_reserve(base, size);
 257                 }
 258         }
 259
 260         if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
 261                 extern u16 memstart_offset_seed;
 262                 u64 range = linear_region_size -
 263                             (memblock_end_of_DRAM() - memblock_start_of_DRAM());
 264
 265                 /*
 266                  * If the size of the linear region exceeds, by a sufficient
 267                  * margin, the size of the region that the available physical
 268                  * memory spans, randomize the linear region as well.
 269                  */
 270                 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
 271                         range = range / ARM64_MEMSTART_ALIGN + 1;
 272                         memstart_addr -= ARM64_MEMSTART_ALIGN *
 273                                          ((range * memstart_offset_seed) >> 16);
 274                 }
 275         }
 276
 277         /*
 278          * Register the kernel text, kernel data, initrd, and initial
 279          * pagetables with memblock.
 280          */
 281         memblock_reserve(__pa(_text), _end - _text);
 282 #ifdef CONFIG_BLK_DEV_INITRD
 283         if (initrd_start) {
 284                 memblock_reserve(initrd_start, initrd_end - initrd_start);
 285
 286                 /* the generic initrd code expects virtual addresses */
 287                 initrd_start = __phys_to_virt(initrd_start);
 288                 initrd_end = __phys_to_virt(initrd_end);
 289         }
 290 #endif
 291
 292         early_init_fdt_scan_reserved_mem();
 293
 294         /* 4GB maximum for 32-bit only capable devices */
 295         if (IS_ENABLED(CONFIG_ZONE_DMA))
 296                 arm64_dma_phys_limit = max_zone_dma_phys();
 297         else
 298                 arm64_dma_phys_limit = PHYS_MASK + 1;
 299         dma_contiguous_reserve(arm64_dma_phys_limit);
 300
 301         memblock_allow_resize();
 302 }
 303
 304 void __init bootmem_init(void)
 305 {
 306         unsigned long min, max;
 307
 308         min = PFN_UP(memblock_start_of_DRAM());
 309         max = PFN_DOWN(memblock_end_of_DRAM());
 310
 311         early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
 312
 313         max_pfn = max_low_pfn = max;
 314
 315         arm64_numa_init();
 316         /*
 317          * Sparsemem tries to allocate bootmem in memory_present(), so must be
 318          * done after the fixed reservations.
 319          */
 320         arm64_memory_present();
 321
 322         sparse_init();
 323         zone_sizes_init(min, max);
 324
 325         high_memory = __va((max << PAGE_SHIFT) - 1) + 1;
 326         memblock_dump_all();
 327 }
 328
 329 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 330 static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
 331 {
 332         struct page *start_pg, *end_pg;
 333         unsigned long pg, pgend;
 334
 335         /*
 336          * Convert start_pfn/end_pfn to a struct page pointer.
 337          */
 338         start_pg = pfn_to_page(start_pfn - 1) + 1;
 339         end_pg = pfn_to_page(end_pfn - 1) + 1;
 340
 341         /*
 342          * Convert to physical addresses, and round start upwards and end
 343          * downwards.
 344          */
 345         pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
 346         pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
 347
 348         /*
 349          * If there are free pages between these, free the section of the
 350          * memmap array.
 351          */
 352         if (pg < pgend)
 353                 free_bootmem(pg, pgend - pg);
 354 }
 355
 356 /*
 357  * The mem_map array can get very big. Free the unused area of the memory map.
 358  */
 359 static void __init free_unused_memmap(void)
 360 {
 361         unsigned long start, prev_end = 0;
 362         struct memblock_region *reg;
 363
 364         for_each_memblock(memory, reg) {
 365                 start = __phys_to_pfn(reg->base);
 366
 367 #ifdef CONFIG_SPARSEMEM
 368                 /*
 369                  * Take care not to free memmap entries that don't exist due
 370                  * to SPARSEMEM sections which aren't present.
 371                  */
 372                 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
 373 #endif
 374                 /*
 375                  * If we had a previous bank, and there is a space between the
 376                  * current bank and the previous, free it.
 377                  */
 378                 if (prev_end && prev_end < start)
 379                         free_memmap(prev_end, start);
 380
 381                 /*
 382                  * Align up here since the VM subsystem insists that the
 383                  * memmap entries are valid from the bank end aligned to
 384                  * MAX_ORDER_NR_PAGES.
 385                  */
 386                 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
 387                                  MAX_ORDER_NR_PAGES);
 388         }
 389
 390 #ifdef CONFIG_SPARSEMEM
 391         if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
 392                 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
 393 #endif
 394 }
 395 #endif  /* !CONFIG_SPARSEMEM_VMEMMAP */
 396
 397 /*
 398  * mem_init() marks the free areas in the mem_map and tells us how much memory
 399  * is free.  This is done after various parts of the system have claimed their
 400  * memory after the kernel image.
 401  */
 402 void __init mem_init(void)
 403 {
 404         if (swiotlb_force == SWIOTLB_FORCE ||
 405             max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT))
 406                 swiotlb_init(1);
 407         else
 408                 swiotlb_force = SWIOTLB_NO_FORCE;
 409
 410         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 411
 412 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 413         free_unused_memmap();
 414 #endif
 415         /* this will put all unused low memory onto the freelists */
 416         free_all_bootmem();
 417
 418         mem_init_print_info(NULL);
 419
 420 #define MLK(b, t) b, t, ((t) - (b)) >> 10
 421 #define MLM(b, t) b, t, ((t) - (b)) >> 20
 422 #define MLG(b, t) b, t, ((t) - (b)) >> 30
 423 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
 424
 425         pr_notice("Virtual kernel memory layout:\n");
 426 #ifdef CONFIG_KASAN
 427         pr_notice("    kasan   : 0x%16lx - 0x%16lx   (%6ld GB)\n",
 428                 MLG(KASAN_SHADOW_START, KASAN_SHADOW_END));
 429 #endif
 430         pr_notice("    modules : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 431                 MLM(MODULES_VADDR, MODULES_END));
 432         pr_notice("    vmalloc : 0x%16lx - 0x%16lx   (%6ld GB)\n",
 433                 MLG(VMALLOC_START, VMALLOC_END));
 434         pr_notice("      .text : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 435                 MLK_ROUNDUP(_text, _etext));
 436         pr_notice("    .rodata : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 437                 MLK_ROUNDUP(__start_rodata, __init_begin));
 438         pr_notice("      .init : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 439                 MLK_ROUNDUP(__init_begin, __init_end));
 440         pr_notice("      .data : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 441                 MLK_ROUNDUP(_sdata, _edata));
 442         pr_notice("       .bss : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 443                 MLK_ROUNDUP(__bss_start, __bss_stop));
 444         pr_notice("    fixed   : 0x%16lx - 0x%16lx   (%6ld KB)\n",
 445                 MLK(FIXADDR_START, FIXADDR_TOP));
 446         pr_notice("    PCI I/O : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 447                 MLM(PCI_IO_START, PCI_IO_END));
 448 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 449         pr_notice("    vmemmap : 0x%16lx - 0x%16lx   (%6ld GB maximum)\n",
 450                 MLG(VMEMMAP_START, VMEMMAP_START + VMEMMAP_SIZE));
 451         pr_notice("              0x%16lx - 0x%16lx   (%6ld MB actual)\n",
 452                 MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
 453                     (unsigned long)virt_to_page(high_memory)));
 454 #endif
 455         pr_notice("    memory  : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 456                 MLM(__phys_to_virt(memblock_start_of_DRAM()),
 457                     (unsigned long)high_memory));
 458
 459 #undef MLK
 460 #undef MLM
 461 #undef MLK_ROUNDUP
 462
 463         /*
 464          * Check boundaries twice: Some fundamental inconsistencies can be
 465          * detected at build time already.
 466          */
 467 #ifdef CONFIG_COMPAT
 468         BUILD_BUG_ON(TASK_SIZE_32                       > TASK_SIZE_64);
 469 #endif
 470
 471         /*
 472          * Make sure we chose the upper bound of sizeof(struct page)
 473          * correctly.
 474          */
 475         BUILD_BUG_ON(sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT));
 476
 477         if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
 478                 extern int sysctl_overcommit_memory;
 479                 /*
 480                  * On a machine this small we won't get anywhere without
 481                  * overcommit, so turn it on by default.
 482                  */
 483                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 484         }
 485 }
 486
 487 void free_initmem(void)
 488 {
 489         free_reserved_area(__va(__pa(__init_begin)), __va(__pa(__init_end)),
 490                            0, "unused kernel");
 491         /*
 492          * Unmap the __init region but leave the VM area in place. This
 493          * prevents the region from being reused for kernel modules, which
 494          * is not supported by kallsyms.
 495          */
 496         unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
 497 }
 498
 499 #ifdef CONFIG_BLK_DEV_INITRD
 500
 501 static int keep_initrd __initdata;
 502
 503 void __init free_initrd_mem(unsigned long start, unsigned long end)
 504 {
 505         if (!keep_initrd)
 506                 free_reserved_area((void *)start, (void *)end, 0, "initrd");
 507 }
 508
 509 static int __init keepinitrd_setup(char *__unused)
 510 {
 511         keep_initrd = 1;
 512         return 1;
 513 }
 514
 515 __setup("keepinitrd", keepinitrd_setup);
 516 #endif
 517
 518 /*
 519  * Dump out memory limit information on panic.
 520  */
 521 static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
 522 {
 523         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 524                 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
 525         } else {
 526                 pr_emerg("Memory Limit: none\n");
 527         }
 528         return 0;
 529 }
 530
 531 static struct notifier_block mem_limit_notifier = {
 532         .notifier_call = dump_mem_limit,
 533 };
 534
 535 static int __init register_mem_limit_dumper(void)
 536 {
 537         atomic_notifier_chain_register(&panic_notifier_list,
 538                                        &mem_limit_notifier);
 539         return 0;
 540 }
 541 __initcall(register_mem_limit_dumper);