SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/powerpc/mm/hash_utils_64.c

   1 /*
   2  * PowerPC64 port by Mike Corrigan and Dave Engebretsen
   3  *   {mikejc|engebret}@us.ibm.com
   4  *
   5  *    Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com>
   6  *
   7  * SMP scalability work:
   8  *    Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
   9  *
  10  *    Module name: htab.c
  11  *
  12  *    Description:
  13  *      PowerPC Hashed Page Table functions
  14  *
  15  * This program is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU General Public License
  17  * as published by the Free Software Foundation; either version
  18  * 2 of the License, or (at your option) any later version.
  19  */
  20
  21 #undef DEBUG
  22 #undef DEBUG_LOW
  23
  24 #include <linux/config.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/errno.h>
  27 #include <linux/sched.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/stat.h>
  30 #include <linux/sysctl.h>
  31 #include <linux/ctype.h>
  32 #include <linux/cache.h>
  33 #include <linux/init.h>
  34 #include <linux/signal.h>
  35
  36 #include <asm/processor.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/mmu.h>
  39 #include <asm/mmu_context.h>
  40 #include <asm/page.h>
  41 #include <asm/types.h>
  42 #include <asm/system.h>
  43 #include <asm/uaccess.h>
  44 #include <asm/machdep.h>
  45 #include <asm/lmb.h>
  46 #include <asm/abs_addr.h>
  47 #include <asm/tlbflush.h>
  48 #include <asm/io.h>
  49 #include <asm/eeh.h>
  50 #include <asm/tlb.h>
  51 #include <asm/cacheflush.h>
  52 #include <asm/cputable.h>
  53 #include <asm/abs_addr.h>
  54 #include <asm/sections.h>
  55
  56 #ifdef DEBUG
  57 #define DBG(fmt...) udbg_printf(fmt)
  58 #else
  59 #define DBG(fmt...)
  60 #endif
  61
  62 #ifdef DEBUG_LOW
  63 #define DBG_LOW(fmt...) udbg_printf(fmt)
  64 #else
  65 #define DBG_LOW(fmt...)
  66 #endif
  67
  68 #define KB (1024)
  69 #define MB (1024*KB)
  70
  71 /*
  72  * Note:  pte   --> Linux PTE
  73  *        HPTE  --> PowerPC Hashed Page Table Entry
  74  *
  75  * Execution context:
  76  *   htab_initialize is called with the MMU off (of course), but
  77  *   the kernel has been copied down to zero so it can directly
  78  *   reference global data.  At this point it is very difficult
  79  *   to print debug info.
  80  *
  81  */
  82
  83 #ifdef CONFIG_U3_DART
  84 extern unsigned long dart_tablebase;
  85 #endif /* CONFIG_U3_DART */
  86
  87 static unsigned long _SDR1;
  88 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
  89
  90 hpte_t *htab_address;
  91 unsigned long htab_size_bytes;
  92 unsigned long htab_hash_mask;
  93 int mmu_linear_psize = MMU_PAGE_4K;
  94 int mmu_virtual_psize = MMU_PAGE_4K;
  95 #ifdef CONFIG_HUGETLB_PAGE
  96 int mmu_huge_psize = MMU_PAGE_16M;
  97 unsigned int HPAGE_SHIFT;
  98 #endif
  99
 100 /* There are definitions of page sizes arrays to be used when none
 101  * is provided by the firmware.
 102  */
 103
 104 /* Pre-POWER4 CPUs (4k pages only)
 105  */
 106 struct mmu_psize_def mmu_psize_defaults_old[] = {
 107         [MMU_PAGE_4K] = {
 108                 .shift  = 12,
 109                 .sllp   = 0,
 110                 .penc   = 0,
 111                 .avpnm  = 0,
 112                 .tlbiel = 0,
 113         },
 114 };
 115
 116 /* POWER4, GPUL, POWER5
 117  *
 118  * Support for 16Mb large pages
 119  */
 120 struct mmu_psize_def mmu_psize_defaults_gp[] = {
 121         [MMU_PAGE_4K] = {
 122                 .shift  = 12,
 123                 .sllp   = 0,
 124                 .penc   = 0,
 125                 .avpnm  = 0,
 126                 .tlbiel = 1,
 127         },
 128         [MMU_PAGE_16M] = {
 129                 .shift  = 24,
 130                 .sllp   = SLB_VSID_L,
 131                 .penc   = 0,
 132                 .avpnm  = 0x1UL,
 133                 .tlbiel = 0,
 134         },
 135 };
 136
 137
 138 int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
 139                       unsigned long pstart, unsigned long mode, int psize)
 140 {
 141         unsigned long vaddr, paddr;
 142         unsigned int step, shift;
 143         unsigned long tmp_mode;
 144         int ret = 0;
 145
 146         shift = mmu_psize_defs[psize].shift;
 147         step = 1 << shift;
 148
 149         for (vaddr = vstart, paddr = pstart; vaddr < vend;
 150              vaddr += step, paddr += step) {
 151                 unsigned long vpn, hash, hpteg;
 152                 unsigned long vsid = get_kernel_vsid(vaddr);
 153                 unsigned long va = (vsid << 28) | (vaddr & 0x0fffffff);
 154
 155                 vpn = va >> shift;
 156                 tmp_mode = mode;
 157
 158                 /* Make non-kernel text non-executable */
 159                 if (!in_kernel_text(vaddr))
 160                         tmp_mode = mode | HPTE_R_N;
 161
 162                 hash = hpt_hash(va, shift);
 163                 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
 164
 165                 /* The crap below can be cleaned once ppd_md.probe() can
 166                  * set up the hash callbacks, thus we can just used the
 167                  * normal insert callback here.
 168                  */
 169 #ifdef CONFIG_PPC_ISERIES
 170                 if (machine_is(iseries))
 171                         ret = iSeries_hpte_insert(hpteg, va,
 172                                                   paddr,
 173                                                   tmp_mode,
 174                                                   HPTE_V_BOLTED,
 175                                                   psize);
 176                 else
 177 #endif
 178 #ifdef CONFIG_PPC_PSERIES
 179                 if (machine_is(pseries) && firmware_has_feature(FW_FEATURE_LPAR))
 180                         ret = pSeries_lpar_hpte_insert(hpteg, va,
 181                                                        paddr,
 182                                                        tmp_mode,
 183                                                        HPTE_V_BOLTED,
 184                                                        psize);
 185                 else
 186 #endif
 187 #ifdef CONFIG_PPC_MULTIPLATFORM
 188                         ret = native_hpte_insert(hpteg, va,
 189                                                  paddr,
 190                                                  tmp_mode, HPTE_V_BOLTED,
 191                                                  psize);
 192 #endif
 193                 if (ret < 0)
 194                         break;
 195         }
 196         return ret < 0 ? ret : 0;
 197 }
 198
 199 static int __init htab_dt_scan_page_sizes(unsigned long node,
 200                                           const char *uname, int depth,
 201                                           void *data)
 202 {
 203         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 204         u32 *prop;
 205         unsigned long size = 0;
 206
 207         /* We are scanning "cpu" nodes only */
 208         if (type == NULL || strcmp(type, "cpu") != 0)
 209                 return 0;
 210
 211         prop = (u32 *)of_get_flat_dt_prop(node,
 212                                           "ibm,segment-page-sizes", &size);
 213         if (prop != NULL) {
 214                 DBG("Page sizes from device-tree:\n");
 215                 size /= 4;
 216                 cur_cpu_spec->cpu_features &= ~(CPU_FTR_16M_PAGE);
 217                 while(size > 0) {
 218                         unsigned int shift = prop[0];
 219                         unsigned int slbenc = prop[1];
 220                         unsigned int lpnum = prop[2];
 221                         unsigned int lpenc = 0;
 222                         struct mmu_psize_def *def;
 223                         int idx = -1;
 224
 225                         size -= 3; prop += 3;
 226                         while(size > 0 && lpnum) {
 227                                 if (prop[0] == shift)
 228                                         lpenc = prop[1];
 229                                 prop += 2; size -= 2;
 230                                 lpnum--;
 231                         }
 232                         switch(shift) {
 233                         case 0xc:
 234                                 idx = MMU_PAGE_4K;
 235                                 break;
 236                         case 0x10:
 237                                 idx = MMU_PAGE_64K;
 238                                 break;
 239                         case 0x14:
 240                                 idx = MMU_PAGE_1M;
 241                                 break;
 242                         case 0x18:
 243                                 idx = MMU_PAGE_16M;
 244                                 cur_cpu_spec->cpu_features |= CPU_FTR_16M_PAGE;
 245                                 break;
 246                         case 0x22:
 247                                 idx = MMU_PAGE_16G;
 248                                 break;
 249                         }
 250                         if (idx < 0)
 251                                 continue;
 252                         def = &mmu_psize_defs[idx];
 253                         def->shift = shift;
 254                         if (shift <= 23)
 255                                 def->avpnm = 0;
 256                         else
 257                                 def->avpnm = (1 << (shift - 23)) - 1;
 258                         def->sllp = slbenc;
 259                         def->penc = lpenc;
 260                         /* We don't know for sure what's up with tlbiel, so
 261                          * for now we only set it for 4K and 64K pages
 262                          */
 263                         if (idx == MMU_PAGE_4K || idx == MMU_PAGE_64K)
 264                                 def->tlbiel = 1;
 265                         else
 266                                 def->tlbiel = 0;
 267
 268                         DBG(" %d: shift=%02x, sllp=%04x, avpnm=%08x, "
 269                             "tlbiel=%d, penc=%d\n",
 270                             idx, shift, def->sllp, def->avpnm, def->tlbiel,
 271                             def->penc);
 272                 }
 273                 return 1;
 274         }
 275         return 0;
 276 }
 277
 278
 279 static void __init htab_init_page_sizes(void)
 280 {
 281         int rc;
 282
 283         /* Default to 4K pages only */
 284         memcpy(mmu_psize_defs, mmu_psize_defaults_old,
 285                sizeof(mmu_psize_defaults_old));
 286
 287         /*
 288          * Try to find the available page sizes in the device-tree
 289          */
 290         rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
 291         if (rc != 0)  /* Found */
 292                 goto found;
 293
 294         /*
 295          * Not in the device-tree, let's fallback on known size
 296          * list for 16M capable GP & GR
 297          */
 298         if (cpu_has_feature(CPU_FTR_16M_PAGE) && !machine_is(iseries))
 299                 memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
 300                        sizeof(mmu_psize_defaults_gp));
 301  found:
 302         /*
 303          * Pick a size for the linear mapping. Currently, we only support
 304          * 16M, 1M and 4K which is the default
 305          */
 306         if (mmu_psize_defs[MMU_PAGE_16M].shift)
 307                 mmu_linear_psize = MMU_PAGE_16M;
 308         else if (mmu_psize_defs[MMU_PAGE_1M].shift)
 309                 mmu_linear_psize = MMU_PAGE_1M;
 310
 311         /*
 312          * Pick a size for the ordinary pages. Default is 4K, we support
 313          * 64K if cache inhibited large pages are supported by the
 314          * processor
 315          */
 316 #ifdef CONFIG_PPC_64K_PAGES
 317         if (mmu_psize_defs[MMU_PAGE_64K].shift &&
 318             cpu_has_feature(CPU_FTR_CI_LARGE_PAGE))
 319                 mmu_virtual_psize = MMU_PAGE_64K;
 320 #endif
 321
 322         printk(KERN_INFO "Page orders: linear mapping = %d, others = %d\n",
 323                mmu_psize_defs[mmu_linear_psize].shift,
 324                mmu_psize_defs[mmu_virtual_psize].shift);
 325
 326 #ifdef CONFIG_HUGETLB_PAGE
 327         /* Init large page size. Currently, we pick 16M or 1M depending
 328          * on what is available
 329          */
 330         if (mmu_psize_defs[MMU_PAGE_16M].shift)
 331                 mmu_huge_psize = MMU_PAGE_16M;
 332         /* With 4k/4level pagetables, we can't (for now) cope with a
 333          * huge page size < PMD_SIZE */
 334         else if (mmu_psize_defs[MMU_PAGE_1M].shift)
 335                 mmu_huge_psize = MMU_PAGE_1M;
 336
 337         /* Calculate HPAGE_SHIFT and sanity check it */
 338         if (mmu_psize_defs[mmu_huge_psize].shift > MIN_HUGEPTE_SHIFT &&
 339             mmu_psize_defs[mmu_huge_psize].shift < SID_SHIFT)
 340                 HPAGE_SHIFT = mmu_psize_defs[mmu_huge_psize].shift;
 341         else
 342                 HPAGE_SHIFT = 0; /* No huge pages dude ! */
 343 #endif /* CONFIG_HUGETLB_PAGE */
 344 }
 345
 346 static int __init htab_dt_scan_pftsize(unsigned long node,
 347                                        const char *uname, int depth,
 348                                        void *data)
 349 {
 350         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 351         u32 *prop;
 352
 353         /* We are scanning "cpu" nodes only */
 354         if (type == NULL || strcmp(type, "cpu") != 0)
 355                 return 0;
 356
 357         prop = (u32 *)of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
 358         if (prop != NULL) {
 359                 /* pft_size[0] is the NUMA CEC cookie */
 360                 ppc64_pft_size = prop[1];
 361                 return 1;
 362         }
 363         return 0;
 364 }
 365
 366 static unsigned long __init htab_get_table_size(void)
 367 {
 368         unsigned long mem_size, rnd_mem_size, pteg_count;
 369
 370         /* If hash size isn't already provided by the platform, we try to
 371          * retrieve it from the device-tree. If it's not there neither, we
 372          * calculate it now based on the total RAM size
 373          */
 374         if (ppc64_pft_size == 0)
 375                 of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
 376         if (ppc64_pft_size)
 377                 return 1UL << ppc64_pft_size;
 378
 379         /* round mem_size up to next power of 2 */
 380         mem_size = lmb_phys_mem_size();
 381         rnd_mem_size = 1UL << __ilog2(mem_size);
 382         if (rnd_mem_size < mem_size)
 383                 rnd_mem_size <<= 1;
 384
 385         /* # pages / 2 */
 386         pteg_count = max(rnd_mem_size >> (12 + 1), 1UL << 11);
 387
 388         return pteg_count << 7;
 389 }
 390
 391 #ifdef CONFIG_MEMORY_HOTPLUG
 392 void create_section_mapping(unsigned long start, unsigned long end)
 393 {
 394                 BUG_ON(htab_bolt_mapping(start, end, __pa(start),
 395                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX,
 396                         mmu_linear_psize));
 397 }
 398 #endif /* CONFIG_MEMORY_HOTPLUG */
 399
 400 void __init htab_initialize(void)
 401 {
 402         unsigned long table;
 403         unsigned long pteg_count;
 404         unsigned long mode_rw;
 405         unsigned long base = 0, size = 0;
 406         int i;
 407
 408         extern unsigned long tce_alloc_start, tce_alloc_end;
 409
 410         DBG(" -> htab_initialize()\n");
 411
 412         /* Initialize page sizes */
 413         htab_init_page_sizes();
 414
 415         /*
 416          * Calculate the required size of the htab.  We want the number of
 417          * PTEGs to equal one half the number of real pages.
 418          */
 419         htab_size_bytes = htab_get_table_size();
 420         pteg_count = htab_size_bytes >> 7;
 421
 422         htab_hash_mask = pteg_count - 1;
 423
 424         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 425                 /* Using a hypervisor which owns the htab */
 426                 htab_address = NULL;
 427                 _SDR1 = 0;
 428         } else {
 429                 /* Find storage for the HPT.  Must be contiguous in
 430                  * the absolute address space.
 431                  */
 432                 table = lmb_alloc(htab_size_bytes, htab_size_bytes);
 433
 434                 DBG("Hash table allocated at %lx, size: %lx\n", table,
 435                     htab_size_bytes);
 436
 437                 htab_address = abs_to_virt(table);
 438
 439                 /* htab absolute addr + encoded htabsize */
 440                 _SDR1 = table + __ilog2(pteg_count) - 11;
 441
 442                 /* Initialize the HPT with no entries */
 443                 memset((void *)table, 0, htab_size_bytes);
 444
 445                 /* Set SDR1 */
 446                 mtspr(SPRN_SDR1, _SDR1);
 447         }
 448
 449         mode_rw = _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX;
 450
 451         /* On U3 based machines, we need to reserve the DART area and
 452          * _NOT_ map it to avoid cache paradoxes as it's remapped non
 453          * cacheable later on
 454          */
 455
 456         /* create bolted the linear mapping in the hash table */
 457         for (i=0; i < lmb.memory.cnt; i++) {
 458                 base = (unsigned long)__va(lmb.memory.region[i].base);
 459                 size = lmb.memory.region[i].size;
 460
 461                 DBG("creating mapping for region: %lx : %lx\n", base, size);
 462
 463 #ifdef CONFIG_U3_DART
 464                 /* Do not map the DART space. Fortunately, it will be aligned
 465                  * in such a way that it will not cross two lmb regions and
 466                  * will fit within a single 16Mb page.
 467                  * The DART space is assumed to be a full 16Mb region even if
 468                  * we only use 2Mb of that space. We will use more of it later
 469                  * for AGP GART. We have to use a full 16Mb large page.
 470                  */
 471                 DBG("DART base: %lx\n", dart_tablebase);
 472
 473                 if (dart_tablebase != 0 && dart_tablebase >= base
 474                     && dart_tablebase < (base + size)) {
 475                         unsigned long dart_table_end = dart_tablebase + 16 * MB;
 476                         if (base != dart_tablebase)
 477                                 BUG_ON(htab_bolt_mapping(base, dart_tablebase,
 478                                                         __pa(base), mode_rw,
 479                                                         mmu_linear_psize));
 480                         if ((base + size) > dart_table_end)
 481                                 BUG_ON(htab_bolt_mapping(dart_tablebase+16*MB,
 482                                                         base + size,
 483                                                         __pa(dart_table_end),
 484                                                          mode_rw,
 485                                                          mmu_linear_psize));
 486                         continue;
 487                 }
 488 #endif /* CONFIG_U3_DART */
 489                 BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
 490                                         mode_rw, mmu_linear_psize));
 491        }
 492
 493         /*
 494          * If we have a memory_limit and we've allocated TCEs then we need to
 495          * explicitly map the TCE area at the top of RAM. We also cope with the
 496          * case that the TCEs start below memory_limit.
 497          * tce_alloc_start/end are 16MB aligned so the mapping should work
 498          * for either 4K or 16MB pages.
 499          */
 500         if (tce_alloc_start) {
 501                 tce_alloc_start = (unsigned long)__va(tce_alloc_start);
 502                 tce_alloc_end = (unsigned long)__va(tce_alloc_end);
 503
 504                 if (base + size >= tce_alloc_start)
 505                         tce_alloc_start = base + size + 1;
 506
 507                 BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
 508                                          __pa(tce_alloc_start), mode_rw,
 509                                          mmu_linear_psize));
 510         }
 511
 512         DBG(" <- htab_initialize()\n");
 513 }
 514 #undef KB
 515 #undef MB
 516
 517 void htab_initialize_secondary(void)
 518 {
 519         if (!firmware_has_feature(FW_FEATURE_LPAR))
 520                 mtspr(SPRN_SDR1, _SDR1);
 521 }
 522
 523 /*
 524  * Called by asm hashtable.S for doing lazy icache flush
 525  */
 526 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
 527 {
 528         struct page *page;
 529
 530         if (!pfn_valid(pte_pfn(pte)))
 531                 return pp;
 532
 533         page = pte_page(pte);
 534
 535         /* page is dirty */
 536         if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
 537                 if (trap == 0x400) {
 538                         __flush_dcache_icache(page_address(page));
 539                         set_bit(PG_arch_1, &page->flags);
 540                 } else
 541                         pp |= HPTE_R_N;
 542         }
 543         return pp;
 544 }
 545
 546 /* Result code is:
 547  *  0 - handled
 548  *  1 - normal page fault
 549  * -1 - critical hash insertion error
 550  */
 551 int hash_page(unsigned long ea, unsigned long access, unsigned long trap)
 552 {
 553         void *pgdir;
 554         unsigned long vsid;
 555         struct mm_struct *mm;
 556         pte_t *ptep;
 557         cpumask_t tmp;
 558         int rc, user_region = 0, local = 0;
 559
 560         DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
 561                 ea, access, trap);
 562
 563         if ((ea & ~REGION_MASK) >= PGTABLE_RANGE) {
 564                 DBG_LOW(" out of pgtable range !\n");
 565                 return 1;
 566         }
 567
 568         /* Get region & vsid */
 569         switch (REGION_ID(ea)) {
 570         case USER_REGION_ID:
 571                 user_region = 1;
 572                 mm = current->mm;
 573                 if (! mm) {
 574                         DBG_LOW(" user region with no mm !\n");
 575                         return 1;
 576                 }
 577                 vsid = get_vsid(mm->context.id, ea);
 578                 break;
 579         case VMALLOC_REGION_ID:
 580                 mm = &init_mm;
 581                 vsid = get_kernel_vsid(ea);
 582                 break;
 583         default:
 584                 /* Not a valid range
 585                  * Send the problem up to do_page_fault
 586                  */
 587                 return 1;
 588         }
 589         DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
 590
 591         /* Get pgdir */
 592         pgdir = mm->pgd;
 593         if (pgdir == NULL)
 594                 return 1;
 595
 596         /* Check CPU locality */
 597         tmp = cpumask_of_cpu(smp_processor_id());
 598         if (user_region && cpus_equal(mm->cpu_vm_mask, tmp))
 599                 local = 1;
 600
 601         /* Handle hugepage regions */
 602         if (unlikely(in_hugepage_area(mm->context, ea))) {
 603                 DBG_LOW(" -> huge page !\n");
 604                 return hash_huge_page(mm, access, ea, vsid, local, trap);
 605         }
 606
 607         /* Get PTE and page size from page tables */
 608         ptep = find_linux_pte(pgdir, ea);
 609         if (ptep == NULL || !pte_present(*ptep)) {
 610                 DBG_LOW(" no PTE !\n");
 611                 return 1;
 612         }
 613
 614 #ifndef CONFIG_PPC_64K_PAGES
 615         DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
 616 #else
 617         DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
 618                 pte_val(*(ptep + PTRS_PER_PTE)));
 619 #endif
 620         /* Pre-check access permissions (will be re-checked atomically
 621          * in __hash_page_XX but this pre-check is a fast path
 622          */
 623         if (access & ~pte_val(*ptep)) {
 624                 DBG_LOW(" no access !\n");
 625                 return 1;
 626         }
 627
 628         /* Do actual hashing */
 629 #ifndef CONFIG_PPC_64K_PAGES
 630         rc = __hash_page_4K(ea, access, vsid, ptep, trap, local);
 631 #else
 632         if (mmu_virtual_psize == MMU_PAGE_64K)
 633                 rc = __hash_page_64K(ea, access, vsid, ptep, trap, local);
 634         else
 635                 rc = __hash_page_4K(ea, access, vsid, ptep, trap, local);
 636 #endif /* CONFIG_PPC_64K_PAGES */
 637
 638 #ifndef CONFIG_PPC_64K_PAGES
 639         DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
 640 #else
 641         DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
 642                 pte_val(*(ptep + PTRS_PER_PTE)));
 643 #endif
 644         DBG_LOW(" -> rc=%d\n", rc);
 645         return rc;
 646 }
 647 EXPORT_SYMBOL_GPL(hash_page);
 648
 649 void hash_preload(struct mm_struct *mm, unsigned long ea,
 650                   unsigned long access, unsigned long trap)
 651 {
 652         unsigned long vsid;
 653         void *pgdir;
 654         pte_t *ptep;
 655         cpumask_t mask;
 656         unsigned long flags;
 657         int local = 0;
 658
 659         /* We don't want huge pages prefaulted for now
 660          */
 661         if (unlikely(in_hugepage_area(mm->context, ea)))
 662                 return;
 663
 664         DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
 665                 " trap=%lx\n", mm, mm->pgd, ea, access, trap);
 666
 667         /* Get PTE, VSID, access mask */
 668         pgdir = mm->pgd;
 669         if (pgdir == NULL)
 670                 return;
 671         ptep = find_linux_pte(pgdir, ea);
 672         if (!ptep)
 673                 return;
 674         vsid = get_vsid(mm->context.id, ea);
 675
 676         /* Hash it in */
 677         local_irq_save(flags);
 678         mask = cpumask_of_cpu(smp_processor_id());
 679         if (cpus_equal(mm->cpu_vm_mask, mask))
 680                 local = 1;
 681 #ifndef CONFIG_PPC_64K_PAGES
 682         __hash_page_4K(ea, access, vsid, ptep, trap, local);
 683 #else
 684         if (mmu_virtual_psize == MMU_PAGE_64K)
 685                 __hash_page_64K(ea, access, vsid, ptep, trap, local);
 686         else
 687                 __hash_page_4K(ea, access, vsid, ptep, trap, local);
 688 #endif /* CONFIG_PPC_64K_PAGES */
 689         local_irq_restore(flags);
 690 }
 691
 692 void flush_hash_page(unsigned long va, real_pte_t pte, int psize, int local)
 693 {
 694         unsigned long hash, index, shift, hidx, slot;
 695
 696         DBG_LOW("flush_hash_page(va=%016x)\n", va);
 697         pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
 698                 hash = hpt_hash(va, shift);
 699                 hidx = __rpte_to_hidx(pte, index);
 700                 if (hidx & _PTEIDX_SECONDARY)
 701                         hash = ~hash;
 702                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 703                 slot += hidx & _PTEIDX_GROUP_IX;
 704                 DBG_LOW(" sub %d: hash=%x, hidx=%x\n", index, slot, hidx);
 705                 ppc_md.hpte_invalidate(slot, va, psize, local);
 706         } pte_iterate_hashed_end();
 707 }
 708
 709 void flush_hash_range(unsigned long number, int local)
 710 {
 711         if (ppc_md.flush_hash_range)
 712                 ppc_md.flush_hash_range(number, local);
 713         else {
 714                 int i;
 715                 struct ppc64_tlb_batch *batch =
 716                         &__get_cpu_var(ppc64_tlb_batch);
 717
 718                 for (i = 0; i < number; i++)
 719                         flush_hash_page(batch->vaddr[i], batch->pte[i],
 720                                         batch->psize, local);
 721         }
 722 }
 723
 724 static inline void make_bl(unsigned int *insn_addr, void *func)
 725 {
 726         unsigned long funcp = *((unsigned long *)func);
 727         int offset = funcp - (unsigned long)insn_addr;
 728
 729         *insn_addr = (unsigned int)(0x48000001 | (offset & 0x03fffffc));
 730         flush_icache_range((unsigned long)insn_addr, 4+
 731                            (unsigned long)insn_addr);
 732 }
 733
 734 /*
 735  * low_hash_fault is called when we the low level hash code failed
 736  * to instert a PTE due to an hypervisor error
 737  */
 738 void low_hash_fault(struct pt_regs *regs, unsigned long address)
 739 {
 740         if (user_mode(regs)) {
 741                 siginfo_t info;
 742
 743                 info.si_signo = SIGBUS;
 744                 info.si_errno = 0;
 745                 info.si_code = BUS_ADRERR;
 746                 info.si_addr = (void __user *)address;
 747                 force_sig_info(SIGBUS, &info, current);
 748                 return;
 749         }
 750         bad_page_fault(regs, address, SIGBUS);
 751 }
 752
 753 void __init htab_finish_init(void)
 754 {
 755         extern unsigned int *htab_call_hpte_insert1;
 756         extern unsigned int *htab_call_hpte_insert2;
 757         extern unsigned int *htab_call_hpte_remove;
 758         extern unsigned int *htab_call_hpte_updatepp;
 759
 760 #ifdef CONFIG_PPC_64K_PAGES
 761         extern unsigned int *ht64_call_hpte_insert1;
 762         extern unsigned int *ht64_call_hpte_insert2;
 763         extern unsigned int *ht64_call_hpte_remove;
 764         extern unsigned int *ht64_call_hpte_updatepp;
 765
 766         make_bl(ht64_call_hpte_insert1, ppc_md.hpte_insert);
 767         make_bl(ht64_call_hpte_insert2, ppc_md.hpte_insert);
 768         make_bl(ht64_call_hpte_remove, ppc_md.hpte_remove);
 769         make_bl(ht64_call_hpte_updatepp, ppc_md.hpte_updatepp);
 770 #endif /* CONFIG_PPC_64K_PAGES */
 771
 772         make_bl(htab_call_hpte_insert1, ppc_md.hpte_insert);
 773         make_bl(htab_call_hpte_insert2, ppc_md.hpte_insert);
 774         make_bl(htab_call_hpte_remove, ppc_md.hpte_remove);
 775         make_bl(htab_call_hpte_updatepp, ppc_md.hpte_updatepp);
 776 }