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/mman.h>
  27 #include <linux/nodemask.h>
  28 #include <linux/initrd.h>
  29 #include <linux/gfp.h>
  30 #include <linux/memblock.h>
  31 #include <linux/sort.h>
  32 #include <linux/of_fdt.h>
  33 #include <linux/dma-mapping.h>
  34 #include <linux/dma-contiguous.h>
  35 #include <linux/efi.h>
  36 #include <linux/swiotlb.h>
  37
  38 #include <asm/boot.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/kasan.h>
  41 #include <asm/kernel-pgtable.h>
  42 #include <asm/memory.h>
  43 #include <asm/sections.h>
  44 #include <asm/setup.h>
  45 #include <asm/sizes.h>
  46 #include <asm/tlb.h>
  47 #include <asm/alternative.h>
  48
  49 #include "mm.h"
  50
  51 /*
  52  * We need to be able to catch inadvertent references to memstart_addr
  53  * that occur (potentially in generic code) before arm64_memblock_init()
  54  * executes, which assigns it its actual value. So use a default value
  55  * that cannot be mistaken for a real physical address.
  56  */
  57 s64 memstart_addr __read_mostly = -1;
  58 phys_addr_t arm64_dma_phys_limit __read_mostly;
  59
  60 #ifdef CONFIG_BLK_DEV_INITRD
  61 static int __init early_initrd(char *p)
  62 {
  63         unsigned long start, size;
  64         char *endp;
  65
  66         start = memparse(p, &endp);
  67         if (*endp == ',') {
  68                 size = memparse(endp + 1, NULL);
  69
  70                 initrd_start = start;
  71                 initrd_end = start + size;
  72         }
  73         return 0;
  74 }
  75 early_param("initrd", early_initrd);
  76 #endif
  77
  78 /*
  79  * Return the maximum physical address for ZONE_DMA (DMA_BIT_MASK(32)). It
  80  * currently assumes that for memory starting above 4G, 32-bit devices will
  81  * use a DMA offset.
  82  */
  83 static phys_addr_t __init max_zone_dma_phys(void)
  84 {
  85         phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
  86         return min(offset + (1ULL << 32), memblock_end_of_DRAM());
  87 }
  88
  89 static void __init zone_sizes_init(unsigned long min, unsigned long max)
  90 {
  91         struct memblock_region *reg;
  92         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
  93         unsigned long max_dma = min;
  94
  95         memset(zone_size, 0, sizeof(zone_size));
  96
  97         /* 4GB maximum for 32-bit only capable devices */
  98 #ifdef CONFIG_ZONE_DMA
  99         max_dma = PFN_DOWN(arm64_dma_phys_limit);
 100         zone_size[ZONE_DMA] = max_dma - min;
 101 #endif
 102         zone_size[ZONE_NORMAL] = max - max_dma;
 103
 104         memcpy(zhole_size, zone_size, sizeof(zhole_size));
 105
 106         for_each_memblock(memory, reg) {
 107                 unsigned long start = memblock_region_memory_base_pfn(reg);
 108                 unsigned long end = memblock_region_memory_end_pfn(reg);
 109
 110                 if (start >= max)
 111                         continue;
 112
 113 #ifdef CONFIG_ZONE_DMA
 114                 if (start < max_dma) {
 115                         unsigned long dma_end = min(end, max_dma);
 116                         zhole_size[ZONE_DMA] -= dma_end - start;
 117                 }
 118 #endif
 119                 if (end > max_dma) {
 120                         unsigned long normal_end = min(end, max);
 121                         unsigned long normal_start = max(start, max_dma);
 122                         zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
 123                 }
 124         }
 125
 126         free_area_init_node(0, zone_size, min, zhole_size);
 127 }
 128
 129 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
 130 int pfn_valid(unsigned long pfn)
 131 {
 132         return memblock_is_map_memory(pfn << PAGE_SHIFT);
 133 }
 134 EXPORT_SYMBOL(pfn_valid);
 135 #endif
 136
 137 #ifndef CONFIG_SPARSEMEM
 138 static void __init arm64_memory_present(void)
 139 {
 140 }
 141 #else
 142 static void __init arm64_memory_present(void)
 143 {
 144         struct memblock_region *reg;
 145
 146         for_each_memblock(memory, reg)
 147                 memory_present(0, memblock_region_memory_base_pfn(reg),
 148                                memblock_region_memory_end_pfn(reg));
 149 }
 150 #endif
 151
 152 static phys_addr_t memory_limit = (phys_addr_t)ULLONG_MAX;
 153
 154 /*
 155  * Limit the memory size that was specified via FDT.
 156  */
 157 static int __init early_mem(char *p)
 158 {
 159         if (!p)
 160                 return 1;
 161
 162         memory_limit = memparse(p, &p) & PAGE_MASK;
 163         pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
 164
 165         return 0;
 166 }
 167 early_param("mem", early_mem);
 168
 169 void __init arm64_memblock_init(void)
 170 {
 171         const s64 linear_region_size = -(s64)PAGE_OFFSET;
 172
 173         /*
 174          * Ensure that the linear region takes up exactly half of the kernel
 175          * virtual address space. This way, we can distinguish a linear address
 176          * from a kernel/module/vmalloc address by testing a single bit.
 177          */
 178         BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
 179
 180         /*
 181          * Select a suitable value for the base of physical memory.
 182          */
 183         memstart_addr = round_down(memblock_start_of_DRAM(),
 184                                    ARM64_MEMSTART_ALIGN);
 185
 186         /*
 187          * Remove the memory that we will not be able to cover with the
 188          * linear mapping. Take care not to clip the kernel which may be
 189          * high in memory.
 190          */
 191         memblock_remove(max_t(u64, memstart_addr + linear_region_size, __pa(_end)),
 192                         ULLONG_MAX);
 193         if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
 194                 /* ensure that memstart_addr remains sufficiently aligned */
 195                 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
 196                                          ARM64_MEMSTART_ALIGN);
 197                 memblock_remove(0, memstart_addr);
 198         }
 199
 200         /*
 201          * Apply the memory limit if it was set. Since the kernel may be loaded
 202          * high up in memory, add back the kernel region that must be accessible
 203          * via the linear mapping.
 204          */
 205         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 206                 memblock_enforce_memory_limit(memory_limit);
 207                 memblock_add(__pa(_text), (u64)(_end - _text));
 208         }
 209
 210         if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
 211                 /*
 212                  * Add back the memory we just removed if it results in the
 213                  * initrd to become inaccessible via the linear mapping.
 214                  * Otherwise, this is a no-op
 215                  */
 216                 u64 base = initrd_start & PAGE_MASK;
 217                 u64 size = PAGE_ALIGN(initrd_end) - base;
 218
 219                 /*
 220                  * We can only add back the initrd memory if we don't end up
 221                  * with more memory than we can address via the linear mapping.
 222                  * It is up to the bootloader to position the kernel and the
 223                  * initrd reasonably close to each other (i.e., within 32 GB of
 224                  * each other) so that all granule/#levels combinations can
 225                  * always access both.
 226                  */
 227                 if (WARN(base < memblock_start_of_DRAM() ||
 228                          base + size > memblock_start_of_DRAM() +
 229                                        linear_region_size,
 230                         "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
 231                         initrd_start = 0;
 232                 } else {
 233                         memblock_remove(base, size); /* clear MEMBLOCK_ flags */
 234                         memblock_add(base, size);
 235                         memblock_reserve(base, size);
 236                 }
 237         }
 238
 239         if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
 240                 extern u16 memstart_offset_seed;
 241                 u64 range = linear_region_size -
 242                             (memblock_end_of_DRAM() - memblock_start_of_DRAM());
 243
 244                 /*
 245                  * If the size of the linear region exceeds, by a sufficient
 246                  * margin, the size of the region that the available physical
 247                  * memory spans, randomize the linear region as well.
 248                  */
 249                 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
 250                         range = range / ARM64_MEMSTART_ALIGN + 1;
 251                         memstart_addr -= ARM64_MEMSTART_ALIGN *
 252                                          ((range * memstart_offset_seed) >> 16);
 253                 }
 254         }
 255
 256         /*
 257          * Register the kernel text, kernel data, initrd, and initial
 258          * pagetables with memblock.
 259          */
 260         memblock_reserve(__pa(_text), _end - _text);
 261 #ifdef CONFIG_BLK_DEV_INITRD
 262         if (initrd_start) {
 263                 memblock_reserve(initrd_start, initrd_end - initrd_start);
 264
 265                 /* the generic initrd code expects virtual addresses */
 266                 initrd_start = __phys_to_virt(initrd_start);
 267                 initrd_end = __phys_to_virt(initrd_end);
 268         }
 269 #endif
 270
 271         early_init_fdt_scan_reserved_mem();
 272
 273         /* 4GB maximum for 32-bit only capable devices */
 274         if (IS_ENABLED(CONFIG_ZONE_DMA))
 275                 arm64_dma_phys_limit = max_zone_dma_phys();
 276         else
 277                 arm64_dma_phys_limit = PHYS_MASK + 1;
 278         dma_contiguous_reserve(arm64_dma_phys_limit);
 279
 280         memblock_allow_resize();
 281         memblock_dump_all();
 282 }
 283
 284 void __init bootmem_init(void)
 285 {
 286         unsigned long min, max;
 287
 288         min = PFN_UP(memblock_start_of_DRAM());
 289         max = PFN_DOWN(memblock_end_of_DRAM());
 290
 291         early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
 292
 293         /*
 294          * Sparsemem tries to allocate bootmem in memory_present(), so must be
 295          * done after the fixed reservations.
 296          */
 297         arm64_memory_present();
 298
 299         sparse_init();
 300         zone_sizes_init(min, max);
 301
 302         high_memory = __va((max << PAGE_SHIFT) - 1) + 1;
 303         max_pfn = max_low_pfn = max;
 304 }
 305
 306 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 307 static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
 308 {
 309         struct page *start_pg, *end_pg;
 310         unsigned long pg, pgend;
 311
 312         /*
 313          * Convert start_pfn/end_pfn to a struct page pointer.
 314          */
 315         start_pg = pfn_to_page(start_pfn - 1) + 1;
 316         end_pg = pfn_to_page(end_pfn - 1) + 1;
 317
 318         /*
 319          * Convert to physical addresses, and round start upwards and end
 320          * downwards.
 321          */
 322         pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
 323         pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
 324
 325         /*
 326          * If there are free pages between these, free the section of the
 327          * memmap array.
 328          */
 329         if (pg < pgend)
 330                 free_bootmem(pg, pgend - pg);
 331 }
 332
 333 /*
 334  * The mem_map array can get very big. Free the unused area of the memory map.
 335  */
 336 static void __init free_unused_memmap(void)
 337 {
 338         unsigned long start, prev_end = 0;
 339         struct memblock_region *reg;
 340
 341         for_each_memblock(memory, reg) {
 342                 start = __phys_to_pfn(reg->base);
 343
 344 #ifdef CONFIG_SPARSEMEM
 345                 /*
 346                  * Take care not to free memmap entries that don't exist due
 347                  * to SPARSEMEM sections which aren't present.
 348                  */
 349                 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
 350 #endif
 351                 /*
 352                  * If we had a previous bank, and there is a space between the
 353                  * current bank and the previous, free it.
 354                  */
 355                 if (prev_end && prev_end < start)
 356                         free_memmap(prev_end, start);
 357
 358                 /*
 359                  * Align up here since the VM subsystem insists that the
 360                  * memmap entries are valid from the bank end aligned to
 361                  * MAX_ORDER_NR_PAGES.
 362                  */
 363                 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
 364                                  MAX_ORDER_NR_PAGES);
 365         }
 366
 367 #ifdef CONFIG_SPARSEMEM
 368         if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
 369                 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
 370 #endif
 371 }
 372 #endif  /* !CONFIG_SPARSEMEM_VMEMMAP */
 373
 374 /*
 375  * mem_init() marks the free areas in the mem_map and tells us how much memory
 376  * is free.  This is done after various parts of the system have claimed their
 377  * memory after the kernel image.
 378  */
 379 void __init mem_init(void)
 380 {
 381         swiotlb_init(1);
 382
 383         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 384
 385 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 386         free_unused_memmap();
 387 #endif
 388         /* this will put all unused low memory onto the freelists */
 389         free_all_bootmem();
 390
 391         mem_init_print_info(NULL);
 392
 393 #define MLK(b, t) b, t, ((t) - (b)) >> 10
 394 #define MLM(b, t) b, t, ((t) - (b)) >> 20
 395 #define MLG(b, t) b, t, ((t) - (b)) >> 30
 396 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
 397
 398         pr_notice("Virtual kernel memory layout:\n");
 399 #ifdef CONFIG_KASAN
 400         pr_cont("    kasan   : 0x%16lx - 0x%16lx   (%6ld GB)\n",
 401                 MLG(KASAN_SHADOW_START, KASAN_SHADOW_END));
 402 #endif
 403         pr_cont("    modules : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 404                 MLM(MODULES_VADDR, MODULES_END));
 405         pr_cont("    vmalloc : 0x%16lx - 0x%16lx   (%6ld GB)\n",
 406                 MLG(VMALLOC_START, VMALLOC_END));
 407         pr_cont("      .text : 0x%p" " - 0x%p" "   (%6ld KB)\n"
 408                 "    .rodata : 0x%p" " - 0x%p" "   (%6ld KB)\n"
 409                 "      .init : 0x%p" " - 0x%p" "   (%6ld KB)\n"
 410                 "      .data : 0x%p" " - 0x%p" "   (%6ld KB)\n",
 411                 MLK_ROUNDUP(_text, __start_rodata),
 412                 MLK_ROUNDUP(__start_rodata, _etext),
 413                 MLK_ROUNDUP(__init_begin, __init_end),
 414                 MLK_ROUNDUP(_sdata, _edata));
 415         pr_cont("    fixed   : 0x%16lx - 0x%16lx   (%6ld KB)\n",
 416                 MLK(FIXADDR_START, FIXADDR_TOP));
 417         pr_cont("    PCI I/O : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 418                 MLM(PCI_IO_START, PCI_IO_END));
 419 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 420         pr_cont("    vmemmap : 0x%16lx - 0x%16lx   (%6ld GB maximum)\n"
 421                 "              0x%16lx - 0x%16lx   (%6ld MB actual)\n",
 422                 MLG(VMEMMAP_START,
 423                     VMEMMAP_START + VMEMMAP_SIZE),
 424                 MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
 425                     (unsigned long)virt_to_page(high_memory)));
 426 #endif
 427         pr_cont("    memory  : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 428                 MLM(__phys_to_virt(memblock_start_of_DRAM()),
 429                     (unsigned long)high_memory));
 430
 431 #undef MLK
 432 #undef MLM
 433 #undef MLK_ROUNDUP
 434
 435         /*
 436          * Check boundaries twice: Some fundamental inconsistencies can be
 437          * detected at build time already.
 438          */
 439 #ifdef CONFIG_COMPAT
 440         BUILD_BUG_ON(TASK_SIZE_32                       > TASK_SIZE_64);
 441 #endif
 442
 443         /*
 444          * Make sure we chose the upper bound of sizeof(struct page)
 445          * correctly.
 446          */
 447         BUILD_BUG_ON(sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT));
 448
 449         if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
 450                 extern int sysctl_overcommit_memory;
 451                 /*
 452                  * On a machine this small we won't get anywhere without
 453                  * overcommit, so turn it on by default.
 454                  */
 455                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 456         }
 457 }
 458
 459 void free_initmem(void)
 460 {
 461         free_reserved_area(__va(__pa(__init_begin)), __va(__pa(__init_end)),
 462                            0, "unused kernel");
 463         fixup_init();
 464 }
 465
 466 #ifdef CONFIG_BLK_DEV_INITRD
 467
 468 static int keep_initrd __initdata;
 469
 470 void __init free_initrd_mem(unsigned long start, unsigned long end)
 471 {
 472         if (!keep_initrd)
 473                 free_reserved_area((void *)start, (void *)end, 0, "initrd");
 474 }
 475
 476 static int __init keepinitrd_setup(char *__unused)
 477 {
 478         keep_initrd = 1;
 479         return 1;
 480 }
 481
 482 __setup("keepinitrd", keepinitrd_setup);
 483 #endif
 484
 485 /*
 486  * Dump out memory limit information on panic.
 487  */
 488 static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
 489 {
 490         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 491                 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
 492         } else {
 493                 pr_emerg("Memory Limit: none\n");
 494         }
 495         return 0;
 496 }
 497
 498 static struct notifier_block mem_limit_notifier = {
 499         .notifier_call = dump_mem_limit,
 500 };
 501
 502 static int __init register_mem_limit_dumper(void)
 503 {
 504         atomic_notifier_chain_register(&panic_notifier_list,
 505                                        &mem_limit_notifier);
 506         return 0;
 507 }
 508 __initcall(register_mem_limit_dumper);