arch/x86/mm/init_64.c

   1 /*
   2  *  linux/arch/x86_64/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Copyright (C) 2000  Pavel Machek <pavel@ucw.cz>
   6  *  Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
   7  */
   8
   9 #include <linux/signal.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/swap.h>
  19 #include <linux/smp.h>
  20 #include <linux/init.h>
  21 #include <linux/initrd.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/memblock.h>
  24 #include <linux/proc_fs.h>
  25 #include <linux/pci.h>
  26 #include <linux/pfn.h>
  27 #include <linux/poison.h>
  28 #include <linux/dma-mapping.h>
  29 #include <linux/memory.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/memremap.h>
  32 #include <linux/nmi.h>
  33 #include <linux/gfp.h>
  34 #include <linux/kcore.h>
  35
  36 #include <asm/processor.h>
  37 #include <asm/bios_ebda.h>
  38 #include <linux/uaccess.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/pgalloc.h>
  41 #include <asm/dma.h>
  42 #include <asm/fixmap.h>
  43 #include <asm/e820/api.h>
  44 #include <asm/apic.h>
  45 #include <asm/tlb.h>
  46 #include <asm/mmu_context.h>
  47 #include <asm/proto.h>
  48 #include <asm/smp.h>
  49 #include <asm/sections.h>
  50 #include <asm/kdebug.h>
  51 #include <asm/numa.h>
  52 #include <asm/set_memory.h>
  53 #include <asm/init.h>
  54 #include <asm/uv/uv.h>
  55 #include <asm/setup.h>
  56
  57 #include "mm_internal.h"
  58
  59 #include "ident_map.c"
  60
  61 #define DEFINE_POPULATE(fname, type1, type2, init)              \
  62 static inline void fname##_init(struct mm_struct *mm,           \
  63                 type1##_t *arg1, type2##_t *arg2, bool init)    \
  64 {                                                               \
  65         if (init)                                               \
  66                 fname##_safe(mm, arg1, arg2);                   \
  67         else                                                    \
  68                 fname(mm, arg1, arg2);                          \
  69 }
  70
  71 DEFINE_POPULATE(p4d_populate, p4d, pud, init)
  72 DEFINE_POPULATE(pgd_populate, pgd, p4d, init)
  73 DEFINE_POPULATE(pud_populate, pud, pmd, init)
  74 DEFINE_POPULATE(pmd_populate_kernel, pmd, pte, init)
  75
  76 #define DEFINE_ENTRY(type1, type2, init)                        \
  77 static inline void set_##type1##_init(type1##_t *arg1,          \
  78                         type2##_t arg2, bool init)              \
  79 {                                                               \
  80         if (init)                                               \
  81                 set_##type1##_safe(arg1, arg2);                 \
  82         else                                                    \
  83                 set_##type1(arg1, arg2);                        \
  84 }
  85
  86 DEFINE_ENTRY(p4d, p4d, init)
  87 DEFINE_ENTRY(pud, pud, init)
  88 DEFINE_ENTRY(pmd, pmd, init)
  89 DEFINE_ENTRY(pte, pte, init)
  90
  91
  92 /*
  93  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  94  * physical space so we can cache the place of the first one and move
  95  * around without checking the pgd every time.
  96  */
  97
  98 /* Bits supported by the hardware: */
  99 pteval_t __supported_pte_mask __read_mostly = ~0;
 100 /* Bits allowed in normal kernel mappings: */
 101 pteval_t __default_kernel_pte_mask __read_mostly = ~0;
 102 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 103 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
 104 EXPORT_SYMBOL(__default_kernel_pte_mask);
 105
 106 int force_personality32;
 107
 108 /*
 109  * noexec32=on|off
 110  * Control non executable heap for 32bit processes.
 111  * To control the stack too use noexec=off
 112  *
 113  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
 114  * off  PROT_READ implies PROT_EXEC
 115  */
 116 static int __init nonx32_setup(char *str)
 117 {
 118         if (!strcmp(str, "on"))
 119                 force_personality32 &= ~READ_IMPLIES_EXEC;
 120         else if (!strcmp(str, "off"))
 121                 force_personality32 |= READ_IMPLIES_EXEC;
 122         return 1;
 123 }
 124 __setup("noexec32=", nonx32_setup);
 125
 126 static void sync_global_pgds_l5(unsigned long start, unsigned long end)
 127 {
 128         unsigned long addr;
 129
 130         for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
 131                 const pgd_t *pgd_ref = pgd_offset_k(addr);
 132                 struct page *page;
 133
 134                 /* Check for overflow */
 135                 if (addr < start)
 136                         break;
 137
 138                 if (pgd_none(*pgd_ref))
 139                         continue;
 140
 141                 spin_lock(&pgd_lock);
 142                 list_for_each_entry(page, &pgd_list, lru) {
 143                         pgd_t *pgd;
 144                         spinlock_t *pgt_lock;
 145
 146                         pgd = (pgd_t *)page_address(page) + pgd_index(addr);
 147                         /* the pgt_lock only for Xen */
 148                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
 149                         spin_lock(pgt_lock);
 150
 151                         if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
 152                                 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
 153
 154                         if (pgd_none(*pgd))
 155                                 set_pgd(pgd, *pgd_ref);
 156
 157                         spin_unlock(pgt_lock);
 158                 }
 159                 spin_unlock(&pgd_lock);
 160         }
 161 }
 162
 163 static void sync_global_pgds_l4(unsigned long start, unsigned long end)
 164 {
 165         unsigned long addr;
 166
 167         for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
 168                 pgd_t *pgd_ref = pgd_offset_k(addr);
 169                 const p4d_t *p4d_ref;
 170                 struct page *page;
 171
 172                 /*
 173                  * With folded p4d, pgd_none() is always false, we need to
 174                  * handle synchonization on p4d level.
 175                  */
 176                 MAYBE_BUILD_BUG_ON(pgd_none(*pgd_ref));
 177                 p4d_ref = p4d_offset(pgd_ref, addr);
 178
 179                 if (p4d_none(*p4d_ref))
 180                         continue;
 181
 182                 spin_lock(&pgd_lock);
 183                 list_for_each_entry(page, &pgd_list, lru) {
 184                         pgd_t *pgd;
 185                         p4d_t *p4d;
 186                         spinlock_t *pgt_lock;
 187
 188                         pgd = (pgd_t *)page_address(page) + pgd_index(addr);
 189                         p4d = p4d_offset(pgd, addr);
 190                         /* the pgt_lock only for Xen */
 191                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
 192                         spin_lock(pgt_lock);
 193
 194                         if (!p4d_none(*p4d_ref) && !p4d_none(*p4d))
 195                                 BUG_ON(p4d_page_vaddr(*p4d)
 196                                        != p4d_page_vaddr(*p4d_ref));
 197
 198                         if (p4d_none(*p4d))
 199                                 set_p4d(p4d, *p4d_ref);
 200
 201                         spin_unlock(pgt_lock);
 202                 }
 203                 spin_unlock(&pgd_lock);
 204         }
 205 }
 206
 207 /*
 208  * When memory was added make sure all the processes MM have
 209  * suitable PGD entries in the local PGD level page.
 210  */
 211 void sync_global_pgds(unsigned long start, unsigned long end)
 212 {
 213         if (pgtable_l5_enabled())
 214                 sync_global_pgds_l5(start, end);
 215         else
 216                 sync_global_pgds_l4(start, end);
 217 }
 218
 219 /*
 220  * NOTE: This function is marked __ref because it calls __init function
 221  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 222  */
 223 static __ref void *spp_getpage(void)
 224 {
 225         void *ptr;
 226
 227         if (after_bootmem)
 228                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 229         else
 230                 ptr = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 231
 232         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 233                 panic("set_pte_phys: cannot allocate page data %s\n",
 234                         after_bootmem ? "after bootmem" : "");
 235         }
 236
 237         pr_debug("spp_getpage %p\n", ptr);
 238
 239         return ptr;
 240 }
 241
 242 static p4d_t *fill_p4d(pgd_t *pgd, unsigned long vaddr)
 243 {
 244         if (pgd_none(*pgd)) {
 245                 p4d_t *p4d = (p4d_t *)spp_getpage();
 246                 pgd_populate(&init_mm, pgd, p4d);
 247                 if (p4d != p4d_offset(pgd, 0))
 248                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
 249                                p4d, p4d_offset(pgd, 0));
 250         }
 251         return p4d_offset(pgd, vaddr);
 252 }
 253
 254 static pud_t *fill_pud(p4d_t *p4d, unsigned long vaddr)
 255 {
 256         if (p4d_none(*p4d)) {
 257                 pud_t *pud = (pud_t *)spp_getpage();
 258                 p4d_populate(&init_mm, p4d, pud);
 259                 if (pud != pud_offset(p4d, 0))
 260                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 261                                pud, pud_offset(p4d, 0));
 262         }
 263         return pud_offset(p4d, vaddr);
 264 }
 265
 266 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
 267 {
 268         if (pud_none(*pud)) {
 269                 pmd_t *pmd = (pmd_t *) spp_getpage();
 270                 pud_populate(&init_mm, pud, pmd);
 271                 if (pmd != pmd_offset(pud, 0))
 272                         printk(KERN_ERR "PAGETABLE BUG #02! %p <-> %p\n",
 273                                pmd, pmd_offset(pud, 0));
 274         }
 275         return pmd_offset(pud, vaddr);
 276 }
 277
 278 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
 279 {
 280         if (pmd_none(*pmd)) {
 281                 pte_t *pte = (pte_t *) spp_getpage();
 282                 pmd_populate_kernel(&init_mm, pmd, pte);
 283                 if (pte != pte_offset_kernel(pmd, 0))
 284                         printk(KERN_ERR "PAGETABLE BUG #03!\n");
 285         }
 286         return pte_offset_kernel(pmd, vaddr);
 287 }
 288
 289 static void __set_pte_vaddr(pud_t *pud, unsigned long vaddr, pte_t new_pte)
 290 {
 291         pmd_t *pmd = fill_pmd(pud, vaddr);
 292         pte_t *pte = fill_pte(pmd, vaddr);
 293
 294         set_pte(pte, new_pte);
 295
 296         /*
 297          * It's enough to flush this one mapping.
 298          * (PGE mappings get flushed as well)
 299          */
 300         __flush_tlb_one_kernel(vaddr);
 301 }
 302
 303 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte)
 304 {
 305         p4d_t *p4d = p4d_page + p4d_index(vaddr);
 306         pud_t *pud = fill_pud(p4d, vaddr);
 307
 308         __set_pte_vaddr(pud, vaddr, new_pte);
 309 }
 310
 311 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 312 {
 313         pud_t *pud = pud_page + pud_index(vaddr);
 314
 315         __set_pte_vaddr(pud, vaddr, new_pte);
 316 }
 317
 318 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 319 {
 320         pgd_t *pgd;
 321         p4d_t *p4d_page;
 322
 323         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
 324
 325         pgd = pgd_offset_k(vaddr);
 326         if (pgd_none(*pgd)) {
 327                 printk(KERN_ERR
 328                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 329                 return;
 330         }
 331
 332         p4d_page = p4d_offset(pgd, 0);
 333         set_pte_vaddr_p4d(p4d_page, vaddr, pteval);
 334 }
 335
 336 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 337 {
 338         pgd_t *pgd;
 339         p4d_t *p4d;
 340         pud_t *pud;
 341
 342         pgd = pgd_offset_k(vaddr);
 343         p4d = fill_p4d(pgd, vaddr);
 344         pud = fill_pud(p4d, vaddr);
 345         return fill_pmd(pud, vaddr);
 346 }
 347
 348 pte_t * __init populate_extra_pte(unsigned long vaddr)
 349 {
 350         pmd_t *pmd;
 351
 352         pmd = populate_extra_pmd(vaddr);
 353         return fill_pte(pmd, vaddr);
 354 }
 355
 356 /*
 357  * Create large page table mappings for a range of physical addresses.
 358  */
 359 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 360                                         enum page_cache_mode cache)
 361 {
 362         pgd_t *pgd;
 363         p4d_t *p4d;
 364         pud_t *pud;
 365         pmd_t *pmd;
 366         pgprot_t prot;
 367
 368         pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
 369                 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
 370         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 371         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 372                 pgd = pgd_offset_k((unsigned long)__va(phys));
 373                 if (pgd_none(*pgd)) {
 374                         p4d = (p4d_t *) spp_getpage();
 375                         set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE |
 376                                                 _PAGE_USER));
 377                 }
 378                 p4d = p4d_offset(pgd, (unsigned long)__va(phys));
 379                 if (p4d_none(*p4d)) {
 380                         pud = (pud_t *) spp_getpage();
 381                         set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE |
 382                                                 _PAGE_USER));
 383                 }
 384                 pud = pud_offset(p4d, (unsigned long)__va(phys));
 385                 if (pud_none(*pud)) {
 386                         pmd = (pmd_t *) spp_getpage();
 387                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 388                                                 _PAGE_USER));
 389                 }
 390                 pmd = pmd_offset(pud, phys);
 391                 BUG_ON(!pmd_none(*pmd));
 392                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 393         }
 394 }
 395
 396 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 397 {
 398         __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
 399 }
 400
 401 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 402 {
 403         __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
 404 }
 405
 406 /*
 407  * The head.S code sets up the kernel high mapping:
 408  *
 409  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 410  *
 411  * phys_base holds the negative offset to the kernel, which is added
 412  * to the compile time generated pmds. This results in invalid pmds up
 413  * to the point where we hit the physaddr 0 mapping.
 414  *
 415  * We limit the mappings to the region from _text to _brk_end.  _brk_end
 416  * is rounded up to the 2MB boundary. This catches the invalid pmds as
 417  * well, as they are located before _text:
 418  */
 419 void __init cleanup_highmap(void)
 420 {
 421         unsigned long vaddr = __START_KERNEL_map;
 422         unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
 423         unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
 424         pmd_t *pmd = level2_kernel_pgt;
 425
 426         /*
 427          * Native path, max_pfn_mapped is not set yet.
 428          * Xen has valid max_pfn_mapped set in
 429          *      arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
 430          */
 431         if (max_pfn_mapped)
 432                 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
 433
 434         for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
 435                 if (pmd_none(*pmd))
 436                         continue;
 437                 if (vaddr < (unsigned long) _text || vaddr > end)
 438                         set_pmd(pmd, __pmd(0));
 439         }
 440 }
 441
 442 /*
 443  * Create PTE level page table mapping for physical addresses.
 444  * It returns the last physical address mapped.
 445  */
 446 static unsigned long __meminit
 447 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
 448               pgprot_t prot, bool init)
 449 {
 450         unsigned long pages = 0, paddr_next;
 451         unsigned long paddr_last = paddr_end;
 452         pte_t *pte;
 453         int i;
 454
 455         pte = pte_page + pte_index(paddr);
 456         i = pte_index(paddr);
 457
 458         for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
 459                 paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
 460                 if (paddr >= paddr_end) {
 461                         if (!after_bootmem &&
 462                             !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
 463                                              E820_TYPE_RAM) &&
 464                             !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
 465                                              E820_TYPE_RESERVED_KERN))
 466                                 set_pte_init(pte, __pte(0), init);
 467                         continue;
 468                 }
 469
 470                 /*
 471                  * We will re-use the existing mapping.
 472                  * Xen for example has some special requirements, like mapping
 473                  * pagetable pages as RO. So assume someone who pre-setup
 474                  * these mappings are more intelligent.
 475                  */
 476                 if (!pte_none(*pte)) {
 477                         if (!after_bootmem)
 478                                 pages++;
 479                         continue;
 480                 }
 481
 482                 if (0)
 483                         pr_info("   pte=%p addr=%lx pte=%016lx\n", pte, paddr,
 484                                 pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 485                 pages++;
 486                 set_pte_init(pte, pfn_pte(paddr >> PAGE_SHIFT, prot), init);
 487                 paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
 488         }
 489
 490         update_page_count(PG_LEVEL_4K, pages);
 491
 492         return paddr_last;
 493 }
 494
 495 /*
 496  * Create PMD level page table mapping for physical addresses. The virtual
 497  * and physical address have to be aligned at this level.
 498  * It returns the last physical address mapped.
 499  */
 500 static unsigned long __meminit
 501 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
 502               unsigned long page_size_mask, pgprot_t prot, bool init)
 503 {
 504         unsigned long pages = 0, paddr_next;
 505         unsigned long paddr_last = paddr_end;
 506
 507         int i = pmd_index(paddr);
 508
 509         for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
 510                 pmd_t *pmd = pmd_page + pmd_index(paddr);
 511                 pte_t *pte;
 512                 pgprot_t new_prot = prot;
 513
 514                 paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
 515                 if (paddr >= paddr_end) {
 516                         if (!after_bootmem &&
 517                             !e820__mapped_any(paddr & PMD_MASK, paddr_next,
 518                                              E820_TYPE_RAM) &&
 519                             !e820__mapped_any(paddr & PMD_MASK, paddr_next,
 520                                              E820_TYPE_RESERVED_KERN))
 521                                 set_pmd_init(pmd, __pmd(0), init);
 522                         continue;
 523                 }
 524
 525                 if (!pmd_none(*pmd)) {
 526                         if (!pmd_large(*pmd)) {
 527                                 spin_lock(&init_mm.page_table_lock);
 528                                 pte = (pte_t *)pmd_page_vaddr(*pmd);
 529                                 paddr_last = phys_pte_init(pte, paddr,
 530                                                            paddr_end, prot,
 531                                                            init);
 532                                 spin_unlock(&init_mm.page_table_lock);
 533                                 continue;
 534                         }
 535                         /*
 536                          * If we are ok with PG_LEVEL_2M mapping, then we will
 537                          * use the existing mapping,
 538                          *
 539                          * Otherwise, we will split the large page mapping but
 540                          * use the same existing protection bits except for
 541                          * large page, so that we don't violate Intel's TLB
 542                          * Application note (317080) which says, while changing
 543                          * the page sizes, new and old translations should
 544                          * not differ with respect to page frame and
 545                          * attributes.
 546                          */
 547                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 548                                 if (!after_bootmem)
 549                                         pages++;
 550                                 paddr_last = paddr_next;
 551                                 continue;
 552                         }
 553                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 554                 }
 555
 556                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 557                         pages++;
 558                         spin_lock(&init_mm.page_table_lock);
 559                         set_pte_init((pte_t *)pmd,
 560                                      pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
 561                                              __pgprot(pgprot_val(prot) | _PAGE_PSE)),
 562                                      init);
 563                         spin_unlock(&init_mm.page_table_lock);
 564                         paddr_last = paddr_next;
 565                         continue;
 566                 }
 567
 568                 pte = alloc_low_page();
 569                 paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot, init);
 570
 571                 spin_lock(&init_mm.page_table_lock);
 572                 pmd_populate_kernel_init(&init_mm, pmd, pte, init);
 573                 spin_unlock(&init_mm.page_table_lock);
 574         }
 575         update_page_count(PG_LEVEL_2M, pages);
 576         return paddr_last;
 577 }
 578
 579 /*
 580  * Create PUD level page table mapping for physical addresses. The virtual
 581  * and physical address do not have to be aligned at this level. KASLR can
 582  * randomize virtual addresses up to this level.
 583  * It returns the last physical address mapped.
 584  */
 585 static unsigned long __meminit
 586 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
 587               unsigned long page_size_mask, bool init)
 588 {
 589         unsigned long pages = 0, paddr_next;
 590         unsigned long paddr_last = paddr_end;
 591         unsigned long vaddr = (unsigned long)__va(paddr);
 592         int i = pud_index(vaddr);
 593
 594         for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
 595                 pud_t *pud;
 596                 pmd_t *pmd;
 597                 pgprot_t prot = PAGE_KERNEL;
 598
 599                 vaddr = (unsigned long)__va(paddr);
 600                 pud = pud_page + pud_index(vaddr);
 601                 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
 602
 603                 if (paddr >= paddr_end) {
 604                         if (!after_bootmem &&
 605                             !e820__mapped_any(paddr & PUD_MASK, paddr_next,
 606                                              E820_TYPE_RAM) &&
 607                             !e820__mapped_any(paddr & PUD_MASK, paddr_next,
 608                                              E820_TYPE_RESERVED_KERN))
 609                                 set_pud_init(pud, __pud(0), init);
 610                         continue;
 611                 }
 612
 613                 if (!pud_none(*pud)) {
 614                         if (!pud_large(*pud)) {
 615                                 pmd = pmd_offset(pud, 0);
 616                                 paddr_last = phys_pmd_init(pmd, paddr,
 617                                                            paddr_end,
 618                                                            page_size_mask,
 619                                                            prot, init);
 620                                 continue;
 621                         }
 622                         /*
 623                          * If we are ok with PG_LEVEL_1G mapping, then we will
 624                          * use the existing mapping.
 625                          *
 626                          * Otherwise, we will split the gbpage mapping but use
 627                          * the same existing protection  bits except for large
 628                          * page, so that we don't violate Intel's TLB
 629                          * Application note (317080) which says, while changing
 630                          * the page sizes, new and old translations should
 631                          * not differ with respect to page frame and
 632                          * attributes.
 633                          */
 634                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 635                                 if (!after_bootmem)
 636                                         pages++;
 637                                 paddr_last = paddr_next;
 638                                 continue;
 639                         }
 640                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 641                 }
 642
 643                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 644                         pages++;
 645                         spin_lock(&init_mm.page_table_lock);
 646                         set_pte_init((pte_t *)pud,
 647                                      pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
 648                                              PAGE_KERNEL_LARGE),
 649                                      init);
 650                         spin_unlock(&init_mm.page_table_lock);
 651                         paddr_last = paddr_next;
 652                         continue;
 653                 }
 654
 655                 pmd = alloc_low_page();
 656                 paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
 657                                            page_size_mask, prot, init);
 658
 659                 spin_lock(&init_mm.page_table_lock);
 660                 pud_populate_init(&init_mm, pud, pmd, init);
 661                 spin_unlock(&init_mm.page_table_lock);
 662         }
 663
 664         update_page_count(PG_LEVEL_1G, pages);
 665
 666         return paddr_last;
 667 }
 668
 669 static unsigned long __meminit
 670 phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
 671               unsigned long page_size_mask, bool init)
 672 {
 673         unsigned long paddr_next, paddr_last = paddr_end;
 674         unsigned long vaddr = (unsigned long)__va(paddr);
 675         int i = p4d_index(vaddr);
 676
 677         if (!pgtable_l5_enabled())
 678                 return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
 679                                      page_size_mask, init);
 680
 681         for (; i < PTRS_PER_P4D; i++, paddr = paddr_next) {
 682                 p4d_t *p4d;
 683                 pud_t *pud;
 684
 685                 vaddr = (unsigned long)__va(paddr);
 686                 p4d = p4d_page + p4d_index(vaddr);
 687                 paddr_next = (paddr & P4D_MASK) + P4D_SIZE;
 688
 689                 if (paddr >= paddr_end) {
 690                         if (!after_bootmem &&
 691                             !e820__mapped_any(paddr & P4D_MASK, paddr_next,
 692                                              E820_TYPE_RAM) &&
 693                             !e820__mapped_any(paddr & P4D_MASK, paddr_next,
 694                                              E820_TYPE_RESERVED_KERN))
 695                                 set_p4d_init(p4d, __p4d(0), init);
 696                         continue;
 697                 }
 698
 699                 if (!p4d_none(*p4d)) {
 700                         pud = pud_offset(p4d, 0);
 701                         paddr_last = phys_pud_init(pud, paddr, paddr_end,
 702                                                    page_size_mask, init);
 703                         continue;
 704                 }
 705
 706                 pud = alloc_low_page();
 707                 paddr_last = phys_pud_init(pud, paddr, paddr_end,
 708                                            page_size_mask, init);
 709
 710                 spin_lock(&init_mm.page_table_lock);
 711                 p4d_populate_init(&init_mm, p4d, pud, init);
 712                 spin_unlock(&init_mm.page_table_lock);
 713         }
 714
 715         return paddr_last;
 716 }
 717
 718 static unsigned long __meminit
 719 __kernel_physical_mapping_init(unsigned long paddr_start,
 720                                unsigned long paddr_end,
 721                                unsigned long page_size_mask,
 722                                bool init)
 723 {
 724         bool pgd_changed = false;
 725         unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
 726
 727         paddr_last = paddr_end;
 728         vaddr = (unsigned long)__va(paddr_start);
 729         vaddr_end = (unsigned long)__va(paddr_end);
 730         vaddr_start = vaddr;
 731
 732         for (; vaddr < vaddr_end; vaddr = vaddr_next) {
 733                 pgd_t *pgd = pgd_offset_k(vaddr);
 734                 p4d_t *p4d;
 735
 736                 vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
 737
 738                 if (pgd_val(*pgd)) {
 739                         p4d = (p4d_t *)pgd_page_vaddr(*pgd);
 740                         paddr_last = phys_p4d_init(p4d, __pa(vaddr),
 741                                                    __pa(vaddr_end),
 742                                                    page_size_mask,
 743                                                    init);
 744                         continue;
 745                 }
 746
 747                 p4d = alloc_low_page();
 748                 paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
 749                                            page_size_mask, init);
 750
 751                 spin_lock(&init_mm.page_table_lock);
 752                 if (pgtable_l5_enabled())
 753                         pgd_populate_init(&init_mm, pgd, p4d, init);
 754                 else
 755                         p4d_populate_init(&init_mm, p4d_offset(pgd, vaddr),
 756                                           (pud_t *) p4d, init);
 757
 758                 spin_unlock(&init_mm.page_table_lock);
 759                 pgd_changed = true;
 760         }
 761
 762         if (pgd_changed)
 763                 sync_global_pgds(vaddr_start, vaddr_end - 1);
 764
 765         return paddr_last;
 766 }
 767
 768
 769 /*
 770  * Create page table mapping for the physical memory for specific physical
 771  * addresses. Note that it can only be used to populate non-present entries.
 772  * The virtual and physical addresses have to be aligned on PMD level
 773  * down. It returns the last physical address mapped.
 774  */
 775 unsigned long __meminit
 776 kernel_physical_mapping_init(unsigned long paddr_start,
 777                              unsigned long paddr_end,
 778                              unsigned long page_size_mask)
 779 {
 780         return __kernel_physical_mapping_init(paddr_start, paddr_end,
 781                                               page_size_mask, true);
 782 }
 783
 784 /*
 785  * This function is similar to kernel_physical_mapping_init() above with the
 786  * exception that it uses set_{pud,pmd}() instead of the set_{pud,pte}_safe()
 787  * when updating the mapping. The caller is responsible to flush the TLBs after
 788  * the function returns.
 789  */
 790 unsigned long __meminit
 791 kernel_physical_mapping_change(unsigned long paddr_start,
 792                                unsigned long paddr_end,
 793                                unsigned long page_size_mask)
 794 {
 795         return __kernel_physical_mapping_init(paddr_start, paddr_end,
 796                                               page_size_mask, false);
 797 }
 798
 799 #ifndef CONFIG_NUMA
 800 void __init initmem_init(void)
 801 {
 802         memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
 803 }
 804 #endif
 805
 806 void __init paging_init(void)
 807 {
 808         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 809         sparse_init();
 810
 811         /*
 812          * clear the default setting with node 0
 813          * note: don't use nodes_clear here, that is really clearing when
 814          *       numa support is not compiled in, and later node_set_state
 815          *       will not set it back.
 816          */
 817         node_clear_state(0, N_MEMORY);
 818         if (N_MEMORY != N_NORMAL_MEMORY)
 819                 node_clear_state(0, N_NORMAL_MEMORY);
 820
 821         zone_sizes_init();
 822 }
 823
 824 /*
 825  * Memory hotplug specific functions
 826  */
 827 #ifdef CONFIG_MEMORY_HOTPLUG
 828 /*
 829  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
 830  * updating.
 831  */
 832 static void update_end_of_memory_vars(u64 start, u64 size)
 833 {
 834         unsigned long end_pfn = PFN_UP(start + size);
 835
 836         if (end_pfn > max_pfn) {
 837                 max_pfn = end_pfn;
 838                 max_low_pfn = end_pfn;
 839                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 840         }
 841 }
 842
 843 int add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
 844                                 struct mhp_restrictions *restrictions)
 845 {
 846         int ret;
 847
 848         ret = __add_pages(nid, start_pfn, nr_pages, restrictions);
 849         WARN_ON_ONCE(ret);
 850
 851         /* update max_pfn, max_low_pfn and high_memory */
 852         update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
 853                                   nr_pages << PAGE_SHIFT);
 854
 855         return ret;
 856 }
 857
 858 int arch_add_memory(int nid, u64 start, u64 size,
 859                         struct mhp_restrictions *restrictions)
 860 {
 861         unsigned long start_pfn = start >> PAGE_SHIFT;
 862         unsigned long nr_pages = size >> PAGE_SHIFT;
 863
 864         init_memory_mapping(start, start + size);
 865
 866         return add_pages(nid, start_pfn, nr_pages, restrictions);
 867 }
 868
 869 #define PAGE_INUSE 0xFD
 870
 871 static void __meminit free_pagetable(struct page *page, int order)
 872 {
 873         unsigned long magic;
 874         unsigned int nr_pages = 1 << order;
 875
 876         /* bootmem page has reserved flag */
 877         if (PageReserved(page)) {
 878                 __ClearPageReserved(page);
 879
 880                 magic = (unsigned long)page->freelist;
 881                 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
 882                         while (nr_pages--)
 883                                 put_page_bootmem(page++);
 884                 } else
 885                         while (nr_pages--)
 886                                 free_reserved_page(page++);
 887         } else
 888                 free_pages((unsigned long)page_address(page), order);
 889 }
 890
 891 static void __meminit free_hugepage_table(struct page *page,
 892                 struct vmem_altmap *altmap)
 893 {
 894         if (altmap)
 895                 vmem_altmap_free(altmap, PMD_SIZE / PAGE_SIZE);
 896         else
 897                 free_pagetable(page, get_order(PMD_SIZE));
 898 }
 899
 900 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
 901 {
 902         pte_t *pte;
 903         int i;
 904
 905         for (i = 0; i < PTRS_PER_PTE; i++) {
 906                 pte = pte_start + i;
 907                 if (!pte_none(*pte))
 908                         return;
 909         }
 910
 911         /* free a pte talbe */
 912         free_pagetable(pmd_page(*pmd), 0);
 913         spin_lock(&init_mm.page_table_lock);
 914         pmd_clear(pmd);
 915         spin_unlock(&init_mm.page_table_lock);
 916 }
 917
 918 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
 919 {
 920         pmd_t *pmd;
 921         int i;
 922
 923         for (i = 0; i < PTRS_PER_PMD; i++) {
 924                 pmd = pmd_start + i;
 925                 if (!pmd_none(*pmd))
 926                         return;
 927         }
 928
 929         /* free a pmd talbe */
 930         free_pagetable(pud_page(*pud), 0);
 931         spin_lock(&init_mm.page_table_lock);
 932         pud_clear(pud);
 933         spin_unlock(&init_mm.page_table_lock);
 934 }
 935
 936 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
 937 {
 938         pud_t *pud;
 939         int i;
 940
 941         for (i = 0; i < PTRS_PER_PUD; i++) {
 942                 pud = pud_start + i;
 943                 if (!pud_none(*pud))
 944                         return;
 945         }
 946
 947         /* free a pud talbe */
 948         free_pagetable(p4d_page(*p4d), 0);
 949         spin_lock(&init_mm.page_table_lock);
 950         p4d_clear(p4d);
 951         spin_unlock(&init_mm.page_table_lock);
 952 }
 953
 954 static void __meminit
 955 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
 956                  bool direct)
 957 {
 958         unsigned long next, pages = 0;
 959         pte_t *pte;
 960         void *page_addr;
 961         phys_addr_t phys_addr;
 962
 963         pte = pte_start + pte_index(addr);
 964         for (; addr < end; addr = next, pte++) {
 965                 next = (addr + PAGE_SIZE) & PAGE_MASK;
 966                 if (next > end)
 967                         next = end;
 968
 969                 if (!pte_present(*pte))
 970                         continue;
 971
 972                 /*
 973                  * We mapped [0,1G) memory as identity mapping when
 974                  * initializing, in arch/x86/kernel/head_64.S. These
 975                  * pagetables cannot be removed.
 976                  */
 977                 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
 978                 if (phys_addr < (phys_addr_t)0x40000000)
 979                         return;
 980
 981                 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
 982                         /*
 983                          * Do not free direct mapping pages since they were
 984                          * freed when offlining, or simplely not in use.
 985                          */
 986                         if (!direct)
 987                                 free_pagetable(pte_page(*pte), 0);
 988
 989                         spin_lock(&init_mm.page_table_lock);
 990                         pte_clear(&init_mm, addr, pte);
 991                         spin_unlock(&init_mm.page_table_lock);
 992
 993                         /* For non-direct mapping, pages means nothing. */
 994                         pages++;
 995                 } else {
 996                         /*
 997                          * If we are here, we are freeing vmemmap pages since
 998                          * direct mapped memory ranges to be freed are aligned.
 999                          *
1000                          * If we are not removing the whole page, it means
1001                          * other page structs in this page are being used and
1002                          * we canot remove them. So fill the unused page_structs
1003                          * with 0xFD, and remove the page when it is wholly
1004                          * filled with 0xFD.
1005                          */
1006                         memset((void *)addr, PAGE_INUSE, next - addr);
1007
1008                         page_addr = page_address(pte_page(*pte));
1009                         if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
1010                                 free_pagetable(pte_page(*pte), 0);
1011
1012                                 spin_lock(&init_mm.page_table_lock);
1013                                 pte_clear(&init_mm, addr, pte);
1014                                 spin_unlock(&init_mm.page_table_lock);
1015                         }
1016                 }
1017         }
1018
1019         /* Call free_pte_table() in remove_pmd_table(). */
1020         flush_tlb_all();
1021         if (direct)
1022                 update_page_count(PG_LEVEL_4K, -pages);
1023 }
1024
1025 static void __meminit
1026 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
1027                  bool direct, struct vmem_altmap *altmap)
1028 {
1029         unsigned long next, pages = 0;
1030         pte_t *pte_base;
1031         pmd_t *pmd;
1032         void *page_addr;
1033
1034         pmd = pmd_start + pmd_index(addr);
1035         for (; addr < end; addr = next, pmd++) {
1036                 next = pmd_addr_end(addr, end);
1037
1038                 if (!pmd_present(*pmd))
1039                         continue;
1040
1041                 if (pmd_large(*pmd)) {
1042                         if (IS_ALIGNED(addr, PMD_SIZE) &&
1043                             IS_ALIGNED(next, PMD_SIZE)) {
1044                                 if (!direct)
1045                                         free_hugepage_table(pmd_page(*pmd),
1046                                                             altmap);
1047
1048                                 spin_lock(&init_mm.page_table_lock);
1049                                 pmd_clear(pmd);
1050                                 spin_unlock(&init_mm.page_table_lock);
1051                                 pages++;
1052                         } else {
1053                                 /* If here, we are freeing vmemmap pages. */
1054                                 memset((void *)addr, PAGE_INUSE, next - addr);
1055
1056                                 page_addr = page_address(pmd_page(*pmd));
1057                                 if (!memchr_inv(page_addr, PAGE_INUSE,
1058                                                 PMD_SIZE)) {
1059                                         free_hugepage_table(pmd_page(*pmd),
1060                                                             altmap);
1061
1062                                         spin_lock(&init_mm.page_table_lock);
1063                                         pmd_clear(pmd);
1064                                         spin_unlock(&init_mm.page_table_lock);
1065                                 }
1066                         }
1067
1068                         continue;
1069                 }
1070
1071                 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
1072                 remove_pte_table(pte_base, addr, next, direct);
1073                 free_pte_table(pte_base, pmd);
1074         }
1075
1076         /* Call free_pmd_table() in remove_pud_table(). */
1077         if (direct)
1078                 update_page_count(PG_LEVEL_2M, -pages);
1079 }
1080
1081 static void __meminit
1082 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
1083                  struct vmem_altmap *altmap, bool direct)
1084 {
1085         unsigned long next, pages = 0;
1086         pmd_t *pmd_base;
1087         pud_t *pud;
1088         void *page_addr;
1089
1090         pud = pud_start + pud_index(addr);
1091         for (; addr < end; addr = next, pud++) {
1092                 next = pud_addr_end(addr, end);
1093
1094                 if (!pud_present(*pud))
1095                         continue;
1096
1097                 if (pud_large(*pud)) {
1098                         if (IS_ALIGNED(addr, PUD_SIZE) &&
1099                             IS_ALIGNED(next, PUD_SIZE)) {
1100                                 if (!direct)
1101                                         free_pagetable(pud_page(*pud),
1102                                                        get_order(PUD_SIZE));
1103
1104                                 spin_lock(&init_mm.page_table_lock);
1105                                 pud_clear(pud);
1106                                 spin_unlock(&init_mm.page_table_lock);
1107                                 pages++;
1108                         } else {
1109                                 /* If here, we are freeing vmemmap pages. */
1110                                 memset((void *)addr, PAGE_INUSE, next - addr);
1111
1112                                 page_addr = page_address(pud_page(*pud));
1113                                 if (!memchr_inv(page_addr, PAGE_INUSE,
1114                                                 PUD_SIZE)) {
1115                                         free_pagetable(pud_page(*pud),
1116                                                        get_order(PUD_SIZE));
1117
1118                                         spin_lock(&init_mm.page_table_lock);
1119                                         pud_clear(pud);
1120                                         spin_unlock(&init_mm.page_table_lock);
1121                                 }
1122                         }
1123
1124                         continue;
1125                 }
1126
1127                 pmd_base = pmd_offset(pud, 0);
1128                 remove_pmd_table(pmd_base, addr, next, direct, altmap);
1129                 free_pmd_table(pmd_base, pud);
1130         }
1131
1132         if (direct)
1133                 update_page_count(PG_LEVEL_1G, -pages);
1134 }
1135
1136 static void __meminit
1137 remove_p4d_table(p4d_t *p4d_start, unsigned long addr, unsigned long end,
1138                  struct vmem_altmap *altmap, bool direct)
1139 {
1140         unsigned long next, pages = 0;
1141         pud_t *pud_base;
1142         p4d_t *p4d;
1143
1144         p4d = p4d_start + p4d_index(addr);
1145         for (; addr < end; addr = next, p4d++) {
1146                 next = p4d_addr_end(addr, end);
1147
1148                 if (!p4d_present(*p4d))
1149                         continue;
1150
1151                 BUILD_BUG_ON(p4d_large(*p4d));
1152
1153                 pud_base = pud_offset(p4d, 0);
1154                 remove_pud_table(pud_base, addr, next, altmap, direct);
1155                 /*
1156                  * For 4-level page tables we do not want to free PUDs, but in the
1157                  * 5-level case we should free them. This code will have to change
1158                  * to adapt for boot-time switching between 4 and 5 level page tables.
1159                  */
1160                 if (pgtable_l5_enabled())
1161                         free_pud_table(pud_base, p4d);
1162         }
1163
1164         if (direct)
1165                 update_page_count(PG_LEVEL_512G, -pages);
1166 }
1167
1168 /* start and end are both virtual address. */
1169 static void __meminit
1170 remove_pagetable(unsigned long start, unsigned long end, bool direct,
1171                 struct vmem_altmap *altmap)
1172 {
1173         unsigned long next;
1174         unsigned long addr;
1175         pgd_t *pgd;
1176         p4d_t *p4d;
1177
1178         for (addr = start; addr < end; addr = next) {
1179                 next = pgd_addr_end(addr, end);
1180
1181                 pgd = pgd_offset_k(addr);
1182                 if (!pgd_present(*pgd))
1183                         continue;
1184
1185                 p4d = p4d_offset(pgd, 0);
1186                 remove_p4d_table(p4d, addr, next, altmap, direct);
1187         }
1188
1189         flush_tlb_all();
1190 }
1191
1192 void __ref vmemmap_free(unsigned long start, unsigned long end,
1193                 struct vmem_altmap *altmap)
1194 {
1195         remove_pagetable(start, end, false, altmap);
1196 }
1197
1198 #ifdef CONFIG_MEMORY_HOTREMOVE
1199 static void __meminit
1200 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1201 {
1202         start = (unsigned long)__va(start);
1203         end = (unsigned long)__va(end);
1204
1205         remove_pagetable(start, end, true, NULL);
1206 }
1207
1208 void __ref arch_remove_memory(int nid, u64 start, u64 size,
1209                               struct vmem_altmap *altmap)
1210 {
1211         unsigned long start_pfn = start >> PAGE_SHIFT;
1212         unsigned long nr_pages = size >> PAGE_SHIFT;
1213         struct page *page = pfn_to_page(start_pfn);
1214         struct zone *zone;
1215
1216         /* With altmap the first mapped page is offset from @start */
1217         if (altmap)
1218                 page += vmem_altmap_offset(altmap);
1219         zone = page_zone(page);
1220         __remove_pages(zone, start_pfn, nr_pages, altmap);
1221         kernel_physical_mapping_remove(start, start + size);
1222 }
1223 #endif
1224 #endif /* CONFIG_MEMORY_HOTPLUG */
1225
1226 static struct kcore_list kcore_vsyscall;
1227
1228 static void __init register_page_bootmem_info(void)
1229 {
1230 #ifdef CONFIG_NUMA
1231         int i;
1232
1233         for_each_online_node(i)
1234                 register_page_bootmem_info_node(NODE_DATA(i));
1235 #endif
1236 }
1237
1238 void __init mem_init(void)
1239 {
1240         pci_iommu_alloc();
1241
1242         /* clear_bss() already clear the empty_zero_page */
1243
1244         /* this will put all memory onto the freelists */
1245         memblock_free_all();
1246         after_bootmem = 1;
1247         x86_init.hyper.init_after_bootmem();
1248
1249         /*
1250          * Must be done after boot memory is put on freelist, because here we
1251          * might set fields in deferred struct pages that have not yet been
1252          * initialized, and memblock_free_all() initializes all the reserved
1253          * deferred pages for us.
1254          */
1255         register_page_bootmem_info();
1256
1257         /* Register memory areas for /proc/kcore */
1258         if (get_gate_vma(&init_mm))
1259                 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR, PAGE_SIZE, KCORE_USER);
1260
1261         mem_init_print_info(NULL);
1262 }
1263
1264 int kernel_set_to_readonly;
1265
1266 void set_kernel_text_rw(void)
1267 {
1268         unsigned long start = PFN_ALIGN(_text);
1269         unsigned long end = PFN_ALIGN(__stop___ex_table);
1270
1271         if (!kernel_set_to_readonly)
1272                 return;
1273
1274         pr_debug("Set kernel text: %lx - %lx for read write\n",
1275                  start, end);
1276
1277         /*
1278          * Make the kernel identity mapping for text RW. Kernel text
1279          * mapping will always be RO. Refer to the comment in
1280          * static_protections() in pageattr.c
1281          */
1282         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1283 }
1284
1285 void set_kernel_text_ro(void)
1286 {
1287         unsigned long start = PFN_ALIGN(_text);
1288         unsigned long end = PFN_ALIGN(__stop___ex_table);
1289
1290         if (!kernel_set_to_readonly)
1291                 return;
1292
1293         pr_debug("Set kernel text: %lx - %lx for read only\n",
1294                  start, end);
1295
1296         /*
1297          * Set the kernel identity mapping for text RO.
1298          */
1299         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1300 }
1301
1302 void mark_rodata_ro(void)
1303 {
1304         unsigned long start = PFN_ALIGN(_text);
1305         unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1306         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1307         unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1308         unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1309         unsigned long all_end;
1310
1311         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1312                (end - start) >> 10);
1313         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1314
1315         kernel_set_to_readonly = 1;
1316
1317         /*
1318          * The rodata/data/bss/brk section (but not the kernel text!)
1319          * should also be not-executable.
1320          *
1321          * We align all_end to PMD_SIZE because the existing mapping
1322          * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1323          * split the PMD and the reminder between _brk_end and the end
1324          * of the PMD will remain mapped executable.
1325          *
1326          * Any PMD which was setup after the one which covers _brk_end
1327          * has been zapped already via cleanup_highmem().
1328          */
1329         all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1330         set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1331
1332 #ifdef CONFIG_CPA_DEBUG
1333         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1334         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1335
1336         printk(KERN_INFO "Testing CPA: again\n");
1337         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1338 #endif
1339
1340         free_kernel_image_pages((void *)text_end, (void *)rodata_start);
1341         free_kernel_image_pages((void *)rodata_end, (void *)_sdata);
1342
1343         debug_checkwx();
1344 }
1345
1346 int kern_addr_valid(unsigned long addr)
1347 {
1348         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1349         pgd_t *pgd;
1350         p4d_t *p4d;
1351         pud_t *pud;
1352         pmd_t *pmd;
1353         pte_t *pte;
1354
1355         if (above != 0 && above != -1UL)
1356                 return 0;
1357
1358         pgd = pgd_offset_k(addr);
1359         if (pgd_none(*pgd))
1360                 return 0;
1361
1362         p4d = p4d_offset(pgd, addr);
1363         if (p4d_none(*p4d))
1364                 return 0;
1365
1366         pud = pud_offset(p4d, addr);
1367         if (pud_none(*pud))
1368                 return 0;
1369
1370         if (pud_large(*pud))
1371                 return pfn_valid(pud_pfn(*pud));
1372
1373         pmd = pmd_offset(pud, addr);
1374         if (pmd_none(*pmd))
1375                 return 0;
1376
1377         if (pmd_large(*pmd))
1378                 return pfn_valid(pmd_pfn(*pmd));
1379
1380         pte = pte_offset_kernel(pmd, addr);
1381         if (pte_none(*pte))
1382                 return 0;
1383
1384         return pfn_valid(pte_pfn(*pte));
1385 }
1386
1387 /*
1388  * Block size is the minimum amount of memory which can be hotplugged or
1389  * hotremoved. It must be power of two and must be equal or larger than
1390  * MIN_MEMORY_BLOCK_SIZE.
1391  */
1392 #define MAX_BLOCK_SIZE (2UL << 30)
1393
1394 /* Amount of ram needed to start using large blocks */
1395 #define MEM_SIZE_FOR_LARGE_BLOCK (64UL << 30)
1396
1397 /* Adjustable memory block size */
1398 static unsigned long set_memory_block_size;
1399 int __init set_memory_block_size_order(unsigned int order)
1400 {
1401         unsigned long size = 1UL << order;
1402
1403         if (size > MEM_SIZE_FOR_LARGE_BLOCK || size < MIN_MEMORY_BLOCK_SIZE)
1404                 return -EINVAL;
1405
1406         set_memory_block_size = size;
1407         return 0;
1408 }
1409
1410 static unsigned long probe_memory_block_size(void)
1411 {
1412         unsigned long boot_mem_end = max_pfn << PAGE_SHIFT;
1413         unsigned long bz;
1414
1415         /* If memory block size has been set, then use it */
1416         bz = set_memory_block_size;
1417         if (bz)
1418                 goto done;
1419
1420         /* Use regular block if RAM is smaller than MEM_SIZE_FOR_LARGE_BLOCK */
1421         if (boot_mem_end < MEM_SIZE_FOR_LARGE_BLOCK) {
1422                 bz = MIN_MEMORY_BLOCK_SIZE;
1423                 goto done;
1424         }
1425
1426         /* Find the largest allowed block size that aligns to memory end */
1427         for (bz = MAX_BLOCK_SIZE; bz > MIN_MEMORY_BLOCK_SIZE; bz >>= 1) {
1428                 if (IS_ALIGNED(boot_mem_end, bz))
1429                         break;
1430         }
1431 done:
1432         pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1433
1434         return bz;
1435 }
1436
1437 static unsigned long memory_block_size_probed;
1438 unsigned long memory_block_size_bytes(void)
1439 {
1440         if (!memory_block_size_probed)
1441                 memory_block_size_probed = probe_memory_block_size();
1442
1443         return memory_block_size_probed;
1444 }
1445
1446 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1447 /*
1448  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1449  */
1450 static long __meminitdata addr_start, addr_end;
1451 static void __meminitdata *p_start, *p_end;
1452 static int __meminitdata node_start;
1453
1454 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1455                 unsigned long end, int node, struct vmem_altmap *altmap)
1456 {
1457         unsigned long addr;
1458         unsigned long next;
1459         pgd_t *pgd;
1460         p4d_t *p4d;
1461         pud_t *pud;
1462         pmd_t *pmd;
1463
1464         for (addr = start; addr < end; addr = next) {
1465                 next = pmd_addr_end(addr, end);
1466
1467                 pgd = vmemmap_pgd_populate(addr, node);
1468                 if (!pgd)
1469                         return -ENOMEM;
1470
1471                 p4d = vmemmap_p4d_populate(pgd, addr, node);
1472                 if (!p4d)
1473                         return -ENOMEM;
1474
1475                 pud = vmemmap_pud_populate(p4d, addr, node);
1476                 if (!pud)
1477                         return -ENOMEM;
1478
1479                 pmd = pmd_offset(pud, addr);
1480                 if (pmd_none(*pmd)) {
1481                         void *p;
1482
1483                         if (altmap)
1484                                 p = altmap_alloc_block_buf(PMD_SIZE, altmap);
1485                         else
1486                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1487                         if (p) {
1488                                 pte_t entry;
1489
1490                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1491                                                 PAGE_KERNEL_LARGE);
1492                                 set_pmd(pmd, __pmd(pte_val(entry)));
1493
1494                                 /* check to see if we have contiguous blocks */
1495                                 if (p_end != p || node_start != node) {
1496                                         if (p_start)
1497                                                 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1498                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
1499                                         addr_start = addr;
1500                                         node_start = node;
1501                                         p_start = p;
1502                                 }
1503
1504                                 addr_end = addr + PMD_SIZE;
1505                                 p_end = p + PMD_SIZE;
1506                                 continue;
1507                         } else if (altmap)
1508                                 return -ENOMEM; /* no fallback */
1509                 } else if (pmd_large(*pmd)) {
1510                         vmemmap_verify((pte_t *)pmd, node, addr, next);
1511                         continue;
1512                 }
1513                 if (vmemmap_populate_basepages(addr, next, node))
1514                         return -ENOMEM;
1515         }
1516         return 0;
1517 }
1518
1519 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1520                 struct vmem_altmap *altmap)
1521 {
1522         int err;
1523
1524         if (boot_cpu_has(X86_FEATURE_PSE))
1525                 err = vmemmap_populate_hugepages(start, end, node, altmap);
1526         else if (altmap) {
1527                 pr_err_once("%s: no cpu support for altmap allocations\n",
1528                                 __func__);
1529                 err = -ENOMEM;
1530         } else
1531                 err = vmemmap_populate_basepages(start, end, node);
1532         if (!err)
1533                 sync_global_pgds(start, end - 1);
1534         return err;
1535 }
1536
1537 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1538 void register_page_bootmem_memmap(unsigned long section_nr,
1539                                   struct page *start_page, unsigned long nr_pages)
1540 {
1541         unsigned long addr = (unsigned long)start_page;
1542         unsigned long end = (unsigned long)(start_page + nr_pages);
1543         unsigned long next;
1544         pgd_t *pgd;
1545         p4d_t *p4d;
1546         pud_t *pud;
1547         pmd_t *pmd;
1548         unsigned int nr_pmd_pages;
1549         struct page *page;
1550
1551         for (; addr < end; addr = next) {
1552                 pte_t *pte = NULL;
1553
1554                 pgd = pgd_offset_k(addr);
1555                 if (pgd_none(*pgd)) {
1556                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1557                         continue;
1558                 }
1559                 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1560
1561                 p4d = p4d_offset(pgd, addr);
1562                 if (p4d_none(*p4d)) {
1563                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1564                         continue;
1565                 }
1566                 get_page_bootmem(section_nr, p4d_page(*p4d), MIX_SECTION_INFO);
1567
1568                 pud = pud_offset(p4d, addr);
1569                 if (pud_none(*pud)) {
1570                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1571                         continue;
1572                 }
1573                 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1574
1575                 if (!boot_cpu_has(X86_FEATURE_PSE)) {
1576                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1577                         pmd = pmd_offset(pud, addr);
1578                         if (pmd_none(*pmd))
1579                                 continue;
1580                         get_page_bootmem(section_nr, pmd_page(*pmd),
1581                                          MIX_SECTION_INFO);
1582
1583                         pte = pte_offset_kernel(pmd, addr);
1584                         if (pte_none(*pte))
1585                                 continue;
1586                         get_page_bootmem(section_nr, pte_page(*pte),
1587                                          SECTION_INFO);
1588                 } else {
1589                         next = pmd_addr_end(addr, end);
1590
1591                         pmd = pmd_offset(pud, addr);
1592                         if (pmd_none(*pmd))
1593                                 continue;
1594
1595                         nr_pmd_pages = 1 << get_order(PMD_SIZE);
1596                         page = pmd_page(*pmd);
1597                         while (nr_pmd_pages--)
1598                                 get_page_bootmem(section_nr, page++,
1599                                                  SECTION_INFO);
1600                 }
1601         }
1602 }
1603 #endif
1604
1605 void __meminit vmemmap_populate_print_last(void)
1606 {
1607         if (p_start) {
1608                 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1609                         addr_start, addr_end-1, p_start, p_end-1, node_start);
1610                 p_start = NULL;
1611                 p_end = NULL;
1612                 node_start = 0;
1613         }
1614 }
1615 #endif