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 phys_addr_t memstart_addr __read_mostly = ~0ULL;
  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          * Select a suitable value for the base of physical memory.
 175          */
 176         memstart_addr = round_down(memblock_start_of_DRAM(),
 177                                    ARM64_MEMSTART_ALIGN);
 178
 179         /*
 180          * Remove the memory that we will not be able to cover with the
 181          * linear mapping. Take care not to clip the kernel which may be
 182          * high in memory.
 183          */
 184         memblock_remove(max(memstart_addr + linear_region_size, __pa(_end)),
 185                         ULLONG_MAX);
 186         if (memblock_end_of_DRAM() > linear_region_size)
 187                 memblock_remove(0, memblock_end_of_DRAM() - linear_region_size);
 188
 189         /*
 190          * Apply the memory limit if it was set. Since the kernel may be loaded
 191          * high up in memory, add back the kernel region that must be accessible
 192          * via the linear mapping.
 193          */
 194         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 195                 memblock_enforce_memory_limit(memory_limit);
 196                 memblock_add(__pa(_text), (u64)(_end - _text));
 197         }
 198
 199         /*
 200          * Register the kernel text, kernel data, initrd, and initial
 201          * pagetables with memblock.
 202          */
 203         memblock_reserve(__pa(_text), _end - _text);
 204 #ifdef CONFIG_BLK_DEV_INITRD
 205         if (initrd_start) {
 206                 memblock_reserve(initrd_start, initrd_end - initrd_start);
 207
 208                 /* the generic initrd code expects virtual addresses */
 209                 initrd_start = __phys_to_virt(initrd_start);
 210                 initrd_end = __phys_to_virt(initrd_end);
 211         }
 212 #endif
 213
 214         early_init_fdt_scan_reserved_mem();
 215
 216         /* 4GB maximum for 32-bit only capable devices */
 217         if (IS_ENABLED(CONFIG_ZONE_DMA))
 218                 arm64_dma_phys_limit = max_zone_dma_phys();
 219         else
 220                 arm64_dma_phys_limit = PHYS_MASK + 1;
 221         dma_contiguous_reserve(arm64_dma_phys_limit);
 222
 223         memblock_allow_resize();
 224         memblock_dump_all();
 225 }
 226
 227 void __init bootmem_init(void)
 228 {
 229         unsigned long min, max;
 230
 231         min = PFN_UP(memblock_start_of_DRAM());
 232         max = PFN_DOWN(memblock_end_of_DRAM());
 233
 234         early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
 235
 236         /*
 237          * Sparsemem tries to allocate bootmem in memory_present(), so must be
 238          * done after the fixed reservations.
 239          */
 240         arm64_memory_present();
 241
 242         sparse_init();
 243         zone_sizes_init(min, max);
 244
 245         high_memory = __va((max << PAGE_SHIFT) - 1) + 1;
 246         max_pfn = max_low_pfn = max;
 247 }
 248
 249 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 250 static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
 251 {
 252         struct page *start_pg, *end_pg;
 253         unsigned long pg, pgend;
 254
 255         /*
 256          * Convert start_pfn/end_pfn to a struct page pointer.
 257          */
 258         start_pg = pfn_to_page(start_pfn - 1) + 1;
 259         end_pg = pfn_to_page(end_pfn - 1) + 1;
 260
 261         /*
 262          * Convert to physical addresses, and round start upwards and end
 263          * downwards.
 264          */
 265         pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
 266         pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
 267
 268         /*
 269          * If there are free pages between these, free the section of the
 270          * memmap array.
 271          */
 272         if (pg < pgend)
 273                 free_bootmem(pg, pgend - pg);
 274 }
 275
 276 /*
 277  * The mem_map array can get very big. Free the unused area of the memory map.
 278  */
 279 static void __init free_unused_memmap(void)
 280 {
 281         unsigned long start, prev_end = 0;
 282         struct memblock_region *reg;
 283
 284         for_each_memblock(memory, reg) {
 285                 start = __phys_to_pfn(reg->base);
 286
 287 #ifdef CONFIG_SPARSEMEM
 288                 /*
 289                  * Take care not to free memmap entries that don't exist due
 290                  * to SPARSEMEM sections which aren't present.
 291                  */
 292                 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
 293 #endif
 294                 /*
 295                  * If we had a previous bank, and there is a space between the
 296                  * current bank and the previous, free it.
 297                  */
 298                 if (prev_end && prev_end < start)
 299                         free_memmap(prev_end, start);
 300
 301                 /*
 302                  * Align up here since the VM subsystem insists that the
 303                  * memmap entries are valid from the bank end aligned to
 304                  * MAX_ORDER_NR_PAGES.
 305                  */
 306                 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
 307                                  MAX_ORDER_NR_PAGES);
 308         }
 309
 310 #ifdef CONFIG_SPARSEMEM
 311         if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
 312                 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
 313 #endif
 314 }
 315 #endif  /* !CONFIG_SPARSEMEM_VMEMMAP */
 316
 317 /*
 318  * mem_init() marks the free areas in the mem_map and tells us how much memory
 319  * is free.  This is done after various parts of the system have claimed their
 320  * memory after the kernel image.
 321  */
 322 void __init mem_init(void)
 323 {
 324         swiotlb_init(1);
 325
 326         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 327
 328 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 329         free_unused_memmap();
 330 #endif
 331         /* this will put all unused low memory onto the freelists */
 332         free_all_bootmem();
 333
 334         mem_init_print_info(NULL);
 335
 336 #define MLK(b, t) b, t, ((t) - (b)) >> 10
 337 #define MLM(b, t) b, t, ((t) - (b)) >> 20
 338 #define MLG(b, t) b, t, ((t) - (b)) >> 30
 339 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
 340
 341         pr_notice("Virtual kernel memory layout:\n"
 342 #ifdef CONFIG_KASAN
 343                   "    kasan   : 0x%16lx - 0x%16lx   (%6ld GB)\n"
 344 #endif
 345                   "    modules : 0x%16lx - 0x%16lx   (%6ld MB)\n"
 346                   "    vmalloc : 0x%16lx - 0x%16lx   (%6ld GB)\n"
 347                   "      .init : 0x%p" " - 0x%p" "   (%6ld KB)\n"
 348                   "      .text : 0x%p" " - 0x%p" "   (%6ld KB)\n"
 349                   "      .data : 0x%p" " - 0x%p" "   (%6ld KB)\n"
 350 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 351                   "    vmemmap : 0x%16lx - 0x%16lx   (%6ld GB maximum)\n"
 352                   "              0x%16lx - 0x%16lx   (%6ld MB actual)\n"
 353 #endif
 354                   "    fixed   : 0x%16lx - 0x%16lx   (%6ld KB)\n"
 355                   "    PCI I/O : 0x%16lx - 0x%16lx   (%6ld MB)\n"
 356                   "    memory  : 0x%16lx - 0x%16lx   (%6ld MB)\n",
 357 #ifdef CONFIG_KASAN
 358                   MLG(KASAN_SHADOW_START, KASAN_SHADOW_END),
 359 #endif
 360                   MLM(MODULES_VADDR, MODULES_END),
 361                   MLG(VMALLOC_START, VMALLOC_END),
 362                   MLK_ROUNDUP(__init_begin, __init_end),
 363                   MLK_ROUNDUP(_text, _etext),
 364                   MLK_ROUNDUP(_sdata, _edata),
 365 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 366                   MLG((unsigned long)vmemmap,
 367                       (unsigned long)vmemmap + VMEMMAP_SIZE),
 368                   MLM((unsigned long)virt_to_page(PAGE_OFFSET),
 369                       (unsigned long)virt_to_page(high_memory)),
 370 #endif
 371                   MLK(FIXADDR_START, FIXADDR_TOP),
 372                   MLM(PCI_IO_START, PCI_IO_END),
 373                   MLM(PAGE_OFFSET, (unsigned long)high_memory));
 374
 375 #undef MLK
 376 #undef MLM
 377 #undef MLK_ROUNDUP
 378
 379         /*
 380          * Check boundaries twice: Some fundamental inconsistencies can be
 381          * detected at build time already.
 382          */
 383 #ifdef CONFIG_COMPAT
 384         BUILD_BUG_ON(TASK_SIZE_32                       > TASK_SIZE_64);
 385 #endif
 386         BUILD_BUG_ON(TASK_SIZE_64                       > MODULES_VADDR);
 387         BUG_ON(TASK_SIZE_64                             > MODULES_VADDR);
 388
 389         if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
 390                 extern int sysctl_overcommit_memory;
 391                 /*
 392                  * On a machine this small we won't get anywhere without
 393                  * overcommit, so turn it on by default.
 394                  */
 395                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 396         }
 397 }
 398
 399 void free_initmem(void)
 400 {
 401         free_initmem_default(0);
 402         fixup_init();
 403 }
 404
 405 #ifdef CONFIG_BLK_DEV_INITRD
 406
 407 static int keep_initrd __initdata;
 408
 409 void __init free_initrd_mem(unsigned long start, unsigned long end)
 410 {
 411         if (!keep_initrd)
 412                 free_reserved_area((void *)start, (void *)end, 0, "initrd");
 413 }
 414
 415 static int __init keepinitrd_setup(char *__unused)
 416 {
 417         keep_initrd = 1;
 418         return 1;
 419 }
 420
 421 __setup("keepinitrd", keepinitrd_setup);
 422 #endif
 423
 424 /*
 425  * Dump out memory limit information on panic.
 426  */
 427 static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
 428 {
 429         if (memory_limit != (phys_addr_t)ULLONG_MAX) {
 430                 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
 431         } else {
 432                 pr_emerg("Memory Limit: none\n");
 433         }
 434         return 0;
 435 }
 436
 437 static struct notifier_block mem_limit_notifier = {
 438         .notifier_call = dump_mem_limit,
 439 };
 440
 441 static int __init register_mem_limit_dumper(void)
 442 {
 443         atomic_notifier_chain_register(&panic_notifier_list,
 444                                        &mem_limit_notifier);
 445         return 0;
 446 }
 447 __initcall(register_mem_limit_dumper);