arch/powerpc/mm/pgtable-radix.c

   1 /*
   2  * Page table handling routines for radix page table.
   3  *
   4  * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11 #include <linux/sched.h>
  12 #include <linux/memblock.h>
  13 #include <linux/of_fdt.h>
  14
  15 #include <asm/pgtable.h>
  16 #include <asm/pgalloc.h>
  17 #include <asm/dma.h>
  18 #include <asm/machdep.h>
  19 #include <asm/mmu.h>
  20 #include <asm/firmware.h>
  21 #include <asm/powernv.h>
  22
  23 #include <trace/events/thp.h>
  24
  25 static int native_register_process_table(unsigned long base, unsigned long pg_sz,
  26                                          unsigned long table_size)
  27 {
  28         unsigned long patb1 = base | table_size | PATB_GR;
  29
  30         partition_tb->patb1 = cpu_to_be64(patb1);
  31         return 0;
  32 }
  33
  34 static __ref void *early_alloc_pgtable(unsigned long size)
  35 {
  36         void *pt;
  37
  38         pt = __va(memblock_alloc_base(size, size, MEMBLOCK_ALLOC_ANYWHERE));
  39         memset(pt, 0, size);
  40
  41         return pt;
  42 }
  43
  44 int radix__map_kernel_page(unsigned long ea, unsigned long pa,
  45                           pgprot_t flags,
  46                           unsigned int map_page_size)
  47 {
  48         pgd_t *pgdp;
  49         pud_t *pudp;
  50         pmd_t *pmdp;
  51         pte_t *ptep;
  52         /*
  53          * Make sure task size is correct as per the max adddr
  54          */
  55         BUILD_BUG_ON(TASK_SIZE_USER64 > RADIX_PGTABLE_RANGE);
  56         if (slab_is_available()) {
  57                 pgdp = pgd_offset_k(ea);
  58                 pudp = pud_alloc(&init_mm, pgdp, ea);
  59                 if (!pudp)
  60                         return -ENOMEM;
  61                 if (map_page_size == PUD_SIZE) {
  62                         ptep = (pte_t *)pudp;
  63                         goto set_the_pte;
  64                 }
  65                 pmdp = pmd_alloc(&init_mm, pudp, ea);
  66                 if (!pmdp)
  67                         return -ENOMEM;
  68                 if (map_page_size == PMD_SIZE) {
  69                         ptep = pmdp_ptep(pmdp);
  70                         goto set_the_pte;
  71                 }
  72                 ptep = pte_alloc_kernel(pmdp, ea);
  73                 if (!ptep)
  74                         return -ENOMEM;
  75         } else {
  76                 pgdp = pgd_offset_k(ea);
  77                 if (pgd_none(*pgdp)) {
  78                         pudp = early_alloc_pgtable(PUD_TABLE_SIZE);
  79                         BUG_ON(pudp == NULL);
  80                         pgd_populate(&init_mm, pgdp, pudp);
  81                 }
  82                 pudp = pud_offset(pgdp, ea);
  83                 if (map_page_size == PUD_SIZE) {
  84                         ptep = (pte_t *)pudp;
  85                         goto set_the_pte;
  86                 }
  87                 if (pud_none(*pudp)) {
  88                         pmdp = early_alloc_pgtable(PMD_TABLE_SIZE);
  89                         BUG_ON(pmdp == NULL);
  90                         pud_populate(&init_mm, pudp, pmdp);
  91                 }
  92                 pmdp = pmd_offset(pudp, ea);
  93                 if (map_page_size == PMD_SIZE) {
  94                         ptep = pmdp_ptep(pmdp);
  95                         goto set_the_pte;
  96                 }
  97                 if (!pmd_present(*pmdp)) {
  98                         ptep = early_alloc_pgtable(PAGE_SIZE);
  99                         BUG_ON(ptep == NULL);
 100                         pmd_populate_kernel(&init_mm, pmdp, ptep);
 101                 }
 102                 ptep = pte_offset_kernel(pmdp, ea);
 103         }
 104
 105 set_the_pte:
 106         set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, flags));
 107         smp_wmb();
 108         return 0;
 109 }
 110
 111 static inline void __meminit print_mapping(unsigned long start,
 112                                            unsigned long end,
 113                                            unsigned long size)
 114 {
 115         if (end <= start)
 116                 return;
 117
 118         pr_info("Mapped range 0x%lx - 0x%lx with 0x%lx\n", start, end, size);
 119 }
 120
 121 static int __meminit create_physical_mapping(unsigned long start,
 122                                              unsigned long end)
 123 {
 124         unsigned long addr, mapping_size = 0;
 125
 126         start = _ALIGN_UP(start, PAGE_SIZE);
 127         for (addr = start; addr < end; addr += mapping_size) {
 128                 unsigned long gap, previous_size;
 129                 int rc;
 130
 131                 gap = end - addr;
 132                 previous_size = mapping_size;
 133
 134                 if (IS_ALIGNED(addr, PUD_SIZE) && gap >= PUD_SIZE &&
 135                     mmu_psize_defs[MMU_PAGE_1G].shift)
 136                         mapping_size = PUD_SIZE;
 137                 else if (IS_ALIGNED(addr, PMD_SIZE) && gap >= PMD_SIZE &&
 138                          mmu_psize_defs[MMU_PAGE_2M].shift)
 139                         mapping_size = PMD_SIZE;
 140                 else
 141                         mapping_size = PAGE_SIZE;
 142
 143                 if (mapping_size != previous_size) {
 144                         print_mapping(start, addr, previous_size);
 145                         start = addr;
 146                 }
 147
 148                 rc = radix__map_kernel_page((unsigned long)__va(addr), addr,
 149                                             PAGE_KERNEL_X, mapping_size);
 150                 if (rc)
 151                         return rc;
 152         }
 153
 154         print_mapping(start, addr, mapping_size);
 155         return 0;
 156 }
 157
 158 static void __init radix_init_pgtable(void)
 159 {
 160         unsigned long rts_field;
 161         struct memblock_region *reg;
 162
 163         /* We don't support slb for radix */
 164         mmu_slb_size = 0;
 165         /*
 166          * Create the linear mapping, using standard page size for now
 167          */
 168         for_each_memblock(memory, reg)
 169                 WARN_ON(create_physical_mapping(reg->base,
 170                                                 reg->base + reg->size));
 171         /*
 172          * Allocate Partition table and process table for the
 173          * host.
 174          */
 175         BUILD_BUG_ON_MSG((PRTB_SIZE_SHIFT > 36), "Process table size too large.");
 176         process_tb = early_alloc_pgtable(1UL << PRTB_SIZE_SHIFT);
 177         /*
 178          * Fill in the process table.
 179          */
 180         rts_field = radix__get_tree_size();
 181         process_tb->prtb0 = cpu_to_be64(rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE);
 182         /*
 183          * Fill in the partition table. We are suppose to use effective address
 184          * of process table here. But our linear mapping also enable us to use
 185          * physical address here.
 186          */
 187         register_process_table(__pa(process_tb), 0, PRTB_SIZE_SHIFT - 12);
 188         pr_info("Process table %p and radix root for kernel: %p\n", process_tb, init_mm.pgd);
 189         asm volatile("ptesync" : : : "memory");
 190         asm volatile(PPC_TLBIE_5(%0,%1,2,1,1) : :
 191                      "r" (TLBIEL_INVAL_SET_LPID), "r" (0));
 192         asm volatile("eieio; tlbsync; ptesync" : : : "memory");
 193 }
 194
 195 static void __init radix_init_partition_table(void)
 196 {
 197         unsigned long rts_field, dw0;
 198
 199         mmu_partition_table_init();
 200         rts_field = radix__get_tree_size();
 201         dw0 = rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE | PATB_HR;
 202         mmu_partition_table_set_entry(0, dw0, 0);
 203
 204         pr_info("Initializing Radix MMU\n");
 205         pr_info("Partition table %p\n", partition_tb);
 206 }
 207
 208 void __init radix_init_native(void)
 209 {
 210         register_process_table = native_register_process_table;
 211 }
 212
 213 static int __init get_idx_from_shift(unsigned int shift)
 214 {
 215         int idx = -1;
 216
 217         switch (shift) {
 218         case 0xc:
 219                 idx = MMU_PAGE_4K;
 220                 break;
 221         case 0x10:
 222                 idx = MMU_PAGE_64K;
 223                 break;
 224         case 0x15:
 225                 idx = MMU_PAGE_2M;
 226                 break;
 227         case 0x1e:
 228                 idx = MMU_PAGE_1G;
 229                 break;
 230         }
 231         return idx;
 232 }
 233
 234 static int __init radix_dt_scan_page_sizes(unsigned long node,
 235                                            const char *uname, int depth,
 236                                            void *data)
 237 {
 238         int size = 0;
 239         int shift, idx;
 240         unsigned int ap;
 241         const __be32 *prop;
 242         const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 243
 244         /* We are scanning "cpu" nodes only */
 245         if (type == NULL || strcmp(type, "cpu") != 0)
 246                 return 0;
 247
 248         prop = of_get_flat_dt_prop(node, "ibm,processor-radix-AP-encodings", &size);
 249         if (!prop)
 250                 return 0;
 251
 252         pr_info("Page sizes from device-tree:\n");
 253         for (; size >= 4; size -= 4, ++prop) {
 254
 255                 struct mmu_psize_def *def;
 256
 257                 /* top 3 bit is AP encoding */
 258                 shift = be32_to_cpu(prop[0]) & ~(0xe << 28);
 259                 ap = be32_to_cpu(prop[0]) >> 29;
 260                 pr_info("Page size shift = %d AP=0x%x\n", shift, ap);
 261
 262                 idx = get_idx_from_shift(shift);
 263                 if (idx < 0)
 264                         continue;
 265
 266                 def = &mmu_psize_defs[idx];
 267                 def->shift = shift;
 268                 def->ap  = ap;
 269         }
 270
 271         /* needed ? */
 272         cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
 273         return 1;
 274 }
 275
 276 void __init radix__early_init_devtree(void)
 277 {
 278         int rc;
 279
 280         /*
 281          * Try to find the available page sizes in the device-tree
 282          */
 283         rc = of_scan_flat_dt(radix_dt_scan_page_sizes, NULL);
 284         if (rc != 0)  /* Found */
 285                 goto found;
 286         /*
 287          * let's assume we have page 4k and 64k support
 288          */
 289         mmu_psize_defs[MMU_PAGE_4K].shift = 12;
 290         mmu_psize_defs[MMU_PAGE_4K].ap = 0x0;
 291
 292         mmu_psize_defs[MMU_PAGE_64K].shift = 16;
 293         mmu_psize_defs[MMU_PAGE_64K].ap = 0x5;
 294 found:
 295 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 296         if (mmu_psize_defs[MMU_PAGE_2M].shift) {
 297                 /*
 298                  * map vmemmap using 2M if available
 299                  */
 300                 mmu_vmemmap_psize = MMU_PAGE_2M;
 301         }
 302 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
 303         return;
 304 }
 305
 306 static void update_hid_for_radix(void)
 307 {
 308         unsigned long hid0;
 309         unsigned long rb = 3UL << PPC_BITLSHIFT(53); /* IS = 3 */
 310
 311         asm volatile("ptesync": : :"memory");
 312         /* prs = 0, ric = 2, rs = 0, r = 1 is = 3 */
 313         asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
 314                      : : "r"(rb), "i"(1), "i"(0), "i"(2), "r"(0) : "memory");
 315         /* prs = 1, ric = 2, rs = 0, r = 1 is = 3 */
 316         asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
 317                      : : "r"(rb), "i"(1), "i"(1), "i"(2), "r"(0) : "memory");
 318         asm volatile("eieio; tlbsync; ptesync; isync; slbia": : :"memory");
 319         /*
 320          * now switch the HID
 321          */
 322         hid0  = mfspr(SPRN_HID0);
 323         hid0 |= HID0_POWER9_RADIX;
 324         mtspr(SPRN_HID0, hid0);
 325         asm volatile("isync": : :"memory");
 326
 327         /* Wait for it to happen */
 328         while (!(mfspr(SPRN_HID0) & HID0_POWER9_RADIX))
 329                 cpu_relax();
 330 }
 331
 332 static void radix_init_amor(void)
 333 {
 334         /*
 335         * In HV mode, we init AMOR (Authority Mask Override Register) so that
 336         * the hypervisor and guest can setup IAMR (Instruction Authority Mask
 337         * Register), enable key 0 and set it to 1.
 338         *
 339         * AMOR = 0b1100 .... 0000 (Mask for key 0 is 11)
 340         */
 341         mtspr(SPRN_AMOR, (3ul << 62));
 342 }
 343
 344 static void radix_init_iamr(void)
 345 {
 346         unsigned long iamr;
 347
 348         /*
 349          * The IAMR should set to 0 on DD1.
 350          */
 351         if (cpu_has_feature(CPU_FTR_POWER9_DD1))
 352                 iamr = 0;
 353         else
 354                 iamr = (1ul << 62);
 355
 356         /*
 357          * Radix always uses key0 of the IAMR to determine if an access is
 358          * allowed. We set bit 0 (IBM bit 1) of key0, to prevent instruction
 359          * fetch.
 360          */
 361         mtspr(SPRN_IAMR, iamr);
 362 }
 363
 364 void __init radix__early_init_mmu(void)
 365 {
 366         unsigned long lpcr;
 367
 368 #ifdef CONFIG_PPC_64K_PAGES
 369         /* PAGE_SIZE mappings */
 370         mmu_virtual_psize = MMU_PAGE_64K;
 371 #else
 372         mmu_virtual_psize = MMU_PAGE_4K;
 373 #endif
 374
 375 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 376         /* vmemmap mapping */
 377         mmu_vmemmap_psize = mmu_virtual_psize;
 378 #endif
 379         /*
 380          * initialize page table size
 381          */
 382         __pte_index_size = RADIX_PTE_INDEX_SIZE;
 383         __pmd_index_size = RADIX_PMD_INDEX_SIZE;
 384         __pud_index_size = RADIX_PUD_INDEX_SIZE;
 385         __pgd_index_size = RADIX_PGD_INDEX_SIZE;
 386         __pmd_cache_index = RADIX_PMD_INDEX_SIZE;
 387         __pte_table_size = RADIX_PTE_TABLE_SIZE;
 388         __pmd_table_size = RADIX_PMD_TABLE_SIZE;
 389         __pud_table_size = RADIX_PUD_TABLE_SIZE;
 390         __pgd_table_size = RADIX_PGD_TABLE_SIZE;
 391
 392         __pmd_val_bits = RADIX_PMD_VAL_BITS;
 393         __pud_val_bits = RADIX_PUD_VAL_BITS;
 394         __pgd_val_bits = RADIX_PGD_VAL_BITS;
 395
 396         __kernel_virt_start = RADIX_KERN_VIRT_START;
 397         __kernel_virt_size = RADIX_KERN_VIRT_SIZE;
 398         __vmalloc_start = RADIX_VMALLOC_START;
 399         __vmalloc_end = RADIX_VMALLOC_END;
 400         vmemmap = (struct page *)RADIX_VMEMMAP_BASE;
 401         ioremap_bot = IOREMAP_BASE;
 402
 403 #ifdef CONFIG_PCI
 404         pci_io_base = ISA_IO_BASE;
 405 #endif
 406
 407         /*
 408          * For now radix also use the same frag size
 409          */
 410         __pte_frag_nr = H_PTE_FRAG_NR;
 411         __pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
 412
 413         if (!firmware_has_feature(FW_FEATURE_LPAR)) {
 414                 radix_init_native();
 415                 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
 416                         update_hid_for_radix();
 417                 lpcr = mfspr(SPRN_LPCR);
 418                 mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
 419                 radix_init_partition_table();
 420                 radix_init_amor();
 421         } else {
 422                 radix_init_pseries();
 423         }
 424
 425         memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
 426
 427         radix_init_iamr();
 428         radix_init_pgtable();
 429 }
 430
 431 void radix__early_init_mmu_secondary(void)
 432 {
 433         unsigned long lpcr;
 434         /*
 435          * update partition table control register and UPRT
 436          */
 437         if (!firmware_has_feature(FW_FEATURE_LPAR)) {
 438
 439                 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
 440                         update_hid_for_radix();
 441
 442                 lpcr = mfspr(SPRN_LPCR);
 443                 mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
 444
 445                 mtspr(SPRN_PTCR,
 446                       __pa(partition_tb) | (PATB_SIZE_SHIFT - 12));
 447                 radix_init_amor();
 448         }
 449         radix_init_iamr();
 450 }
 451
 452 void radix__mmu_cleanup_all(void)
 453 {
 454         unsigned long lpcr;
 455
 456         if (!firmware_has_feature(FW_FEATURE_LPAR)) {
 457                 lpcr = mfspr(SPRN_LPCR);
 458                 mtspr(SPRN_LPCR, lpcr & ~LPCR_UPRT);
 459                 mtspr(SPRN_PTCR, 0);
 460                 powernv_set_nmmu_ptcr(0);
 461                 radix__flush_tlb_all();
 462         }
 463 }
 464
 465 void radix__setup_initial_memory_limit(phys_addr_t first_memblock_base,
 466                                 phys_addr_t first_memblock_size)
 467 {
 468         /* We don't currently support the first MEMBLOCK not mapping 0
 469          * physical on those processors
 470          */
 471         BUG_ON(first_memblock_base != 0);
 472         /*
 473          * We limit the allocation that depend on ppc64_rma_size
 474          * to first_memblock_size. We also clamp it to 1GB to
 475          * avoid some funky things such as RTAS bugs.
 476          *
 477          * On radix config we really don't have a limitation
 478          * on real mode access. But keeping it as above works
 479          * well enough.
 480          */
 481         ppc64_rma_size = min_t(u64, first_memblock_size, 0x40000000);
 482         /*
 483          * Finally limit subsequent allocations. We really don't want
 484          * to limit the memblock allocations to rma_size. FIXME!! should
 485          * we even limit at all ?
 486          */
 487         memblock_set_current_limit(first_memblock_base + first_memblock_size);
 488 }
 489
 490 #ifdef CONFIG_MEMORY_HOTPLUG
 491 static void free_pte_table(pte_t *pte_start, pmd_t *pmd)
 492 {
 493         pte_t *pte;
 494         int i;
 495
 496         for (i = 0; i < PTRS_PER_PTE; i++) {
 497                 pte = pte_start + i;
 498                 if (!pte_none(*pte))
 499                         return;
 500         }
 501
 502         pte_free_kernel(&init_mm, pte_start);
 503         pmd_clear(pmd);
 504 }
 505
 506 static void free_pmd_table(pmd_t *pmd_start, pud_t *pud)
 507 {
 508         pmd_t *pmd;
 509         int i;
 510
 511         for (i = 0; i < PTRS_PER_PMD; i++) {
 512                 pmd = pmd_start + i;
 513                 if (!pmd_none(*pmd))
 514                         return;
 515         }
 516
 517         pmd_free(&init_mm, pmd_start);
 518         pud_clear(pud);
 519 }
 520
 521 static void remove_pte_table(pte_t *pte_start, unsigned long addr,
 522                              unsigned long end)
 523 {
 524         unsigned long next;
 525         pte_t *pte;
 526
 527         pte = pte_start + pte_index(addr);
 528         for (; addr < end; addr = next, pte++) {
 529                 next = (addr + PAGE_SIZE) & PAGE_MASK;
 530                 if (next > end)
 531                         next = end;
 532
 533                 if (!pte_present(*pte))
 534                         continue;
 535
 536                 if (!PAGE_ALIGNED(addr) || !PAGE_ALIGNED(next)) {
 537                         /*
 538                          * The vmemmap_free() and remove_section_mapping()
 539                          * codepaths call us with aligned addresses.
 540                          */
 541                         WARN_ONCE(1, "%s: unaligned range\n", __func__);
 542                         continue;
 543                 }
 544
 545                 pte_clear(&init_mm, addr, pte);
 546         }
 547 }
 548
 549 static void remove_pmd_table(pmd_t *pmd_start, unsigned long addr,
 550                              unsigned long end)
 551 {
 552         unsigned long next;
 553         pte_t *pte_base;
 554         pmd_t *pmd;
 555
 556         pmd = pmd_start + pmd_index(addr);
 557         for (; addr < end; addr = next, pmd++) {
 558                 next = pmd_addr_end(addr, end);
 559
 560                 if (!pmd_present(*pmd))
 561                         continue;
 562
 563                 if (pmd_huge(*pmd)) {
 564                         if (!IS_ALIGNED(addr, PMD_SIZE) ||
 565                             !IS_ALIGNED(next, PMD_SIZE)) {
 566                                 WARN_ONCE(1, "%s: unaligned range\n", __func__);
 567                                 continue;
 568                         }
 569
 570                         pte_clear(&init_mm, addr, (pte_t *)pmd);
 571                         continue;
 572                 }
 573
 574                 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
 575                 remove_pte_table(pte_base, addr, next);
 576                 free_pte_table(pte_base, pmd);
 577         }
 578 }
 579
 580 static void remove_pud_table(pud_t *pud_start, unsigned long addr,
 581                              unsigned long end)
 582 {
 583         unsigned long next;
 584         pmd_t *pmd_base;
 585         pud_t *pud;
 586
 587         pud = pud_start + pud_index(addr);
 588         for (; addr < end; addr = next, pud++) {
 589                 next = pud_addr_end(addr, end);
 590
 591                 if (!pud_present(*pud))
 592                         continue;
 593
 594                 if (pud_huge(*pud)) {
 595                         if (!IS_ALIGNED(addr, PUD_SIZE) ||
 596                             !IS_ALIGNED(next, PUD_SIZE)) {
 597                                 WARN_ONCE(1, "%s: unaligned range\n", __func__);
 598                                 continue;
 599                         }
 600
 601                         pte_clear(&init_mm, addr, (pte_t *)pud);
 602                         continue;
 603                 }
 604
 605                 pmd_base = (pmd_t *)pud_page_vaddr(*pud);
 606                 remove_pmd_table(pmd_base, addr, next);
 607                 free_pmd_table(pmd_base, pud);
 608         }
 609 }
 610
 611 static void remove_pagetable(unsigned long start, unsigned long end)
 612 {
 613         unsigned long addr, next;
 614         pud_t *pud_base;
 615         pgd_t *pgd;
 616
 617         spin_lock(&init_mm.page_table_lock);
 618
 619         for (addr = start; addr < end; addr = next) {
 620                 next = pgd_addr_end(addr, end);
 621
 622                 pgd = pgd_offset_k(addr);
 623                 if (!pgd_present(*pgd))
 624                         continue;
 625
 626                 if (pgd_huge(*pgd)) {
 627                         if (!IS_ALIGNED(addr, PGDIR_SIZE) ||
 628                             !IS_ALIGNED(next, PGDIR_SIZE)) {
 629                                 WARN_ONCE(1, "%s: unaligned range\n", __func__);
 630                                 continue;
 631                         }
 632
 633                         pte_clear(&init_mm, addr, (pte_t *)pgd);
 634                         continue;
 635                 }
 636
 637                 pud_base = (pud_t *)pgd_page_vaddr(*pgd);
 638                 remove_pud_table(pud_base, addr, next);
 639         }
 640
 641         spin_unlock(&init_mm.page_table_lock);
 642         radix__flush_tlb_kernel_range(start, end);
 643 }
 644
 645 int __ref radix__create_section_mapping(unsigned long start, unsigned long end)
 646 {
 647         return create_physical_mapping(start, end);
 648 }
 649
 650 int radix__remove_section_mapping(unsigned long start, unsigned long end)
 651 {
 652         remove_pagetable(start, end);
 653         return 0;
 654 }
 655 #endif /* CONFIG_MEMORY_HOTPLUG */
 656
 657 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 658 int __meminit radix__vmemmap_create_mapping(unsigned long start,
 659                                       unsigned long page_size,
 660                                       unsigned long phys)
 661 {
 662         /* Create a PTE encoding */
 663         unsigned long flags = _PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_KERNEL_RW;
 664
 665         BUG_ON(radix__map_kernel_page(start, phys, __pgprot(flags), page_size));
 666         return 0;
 667 }
 668
 669 #ifdef CONFIG_MEMORY_HOTPLUG
 670 void radix__vmemmap_remove_mapping(unsigned long start, unsigned long page_size)
 671 {
 672         remove_pagetable(start, start + page_size);
 673 }
 674 #endif
 675 #endif
 676
 677 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 678
 679 unsigned long radix__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
 680                                   pmd_t *pmdp, unsigned long clr,
 681                                   unsigned long set)
 682 {
 683         unsigned long old;
 684
 685 #ifdef CONFIG_DEBUG_VM
 686         WARN_ON(!radix__pmd_trans_huge(*pmdp));
 687         assert_spin_locked(&mm->page_table_lock);
 688 #endif
 689
 690         old = radix__pte_update(mm, addr, (pte_t *)pmdp, clr, set, 1);
 691         trace_hugepage_update(addr, old, clr, set);
 692
 693         return old;
 694 }
 695
 696 pmd_t radix__pmdp_collapse_flush(struct vm_area_struct *vma, unsigned long address,
 697                         pmd_t *pmdp)
 698
 699 {
 700         pmd_t pmd;
 701
 702         VM_BUG_ON(address & ~HPAGE_PMD_MASK);
 703         VM_BUG_ON(radix__pmd_trans_huge(*pmdp));
 704         /*
 705          * khugepaged calls this for normal pmd
 706          */
 707         pmd = *pmdp;
 708         pmd_clear(pmdp);
 709
 710         /*FIXME!!  Verify whether we need this kick below */
 711         kick_all_cpus_sync();
 712
 713         radix__flush_tlb_collapsed_pmd(vma->vm_mm, address);
 714
 715         return pmd;
 716 }
 717
 718 /*
 719  * For us pgtable_t is pte_t *. Inorder to save the deposisted
 720  * page table, we consider the allocated page table as a list
 721  * head. On withdraw we need to make sure we zero out the used
 722  * list_head memory area.
 723  */
 724 void radix__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 725                                  pgtable_t pgtable)
 726 {
 727         struct list_head *lh = (struct list_head *) pgtable;
 728
 729         assert_spin_locked(pmd_lockptr(mm, pmdp));
 730
 731         /* FIFO */
 732         if (!pmd_huge_pte(mm, pmdp))
 733                 INIT_LIST_HEAD(lh);
 734         else
 735                 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
 736         pmd_huge_pte(mm, pmdp) = pgtable;
 737 }
 738
 739 pgtable_t radix__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
 740 {
 741         pte_t *ptep;
 742         pgtable_t pgtable;
 743         struct list_head *lh;
 744
 745         assert_spin_locked(pmd_lockptr(mm, pmdp));
 746
 747         /* FIFO */
 748         pgtable = pmd_huge_pte(mm, pmdp);
 749         lh = (struct list_head *) pgtable;
 750         if (list_empty(lh))
 751                 pmd_huge_pte(mm, pmdp) = NULL;
 752         else {
 753                 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
 754                 list_del(lh);
 755         }
 756         ptep = (pte_t *) pgtable;
 757         *ptep = __pte(0);
 758         ptep++;
 759         *ptep = __pte(0);
 760         return pgtable;
 761 }
 762
 763
 764 pmd_t radix__pmdp_huge_get_and_clear(struct mm_struct *mm,
 765                                unsigned long addr, pmd_t *pmdp)
 766 {
 767         pmd_t old_pmd;
 768         unsigned long old;
 769
 770         old = radix__pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
 771         old_pmd = __pmd(old);
 772         /*
 773          * Serialize against find_linux_pte_or_hugepte which does lock-less
 774          * lookup in page tables with local interrupts disabled. For huge pages
 775          * it casts pmd_t to pte_t. Since format of pte_t is different from
 776          * pmd_t we want to prevent transit from pmd pointing to page table
 777          * to pmd pointing to huge page (and back) while interrupts are disabled.
 778          * We clear pmd to possibly replace it with page table pointer in
 779          * different code paths. So make sure we wait for the parallel
 780          * find_linux_pte_or_hugepage to finish.
 781          */
 782         kick_all_cpus_sync();
 783         return old_pmd;
 784 }
 785
 786 int radix__has_transparent_hugepage(void)
 787 {
 788         /* For radix 2M at PMD level means thp */
 789         if (mmu_psize_defs[MMU_PAGE_2M].shift == PMD_SHIFT)
 790                 return 1;
 791         return 0;
 792 }
 793 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */