arch/ppc64/mm/hash_utils.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
  23 #include <linux/config.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/errno.h>
  26 #include <linux/sched.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/stat.h>
  29 #include <linux/sysctl.h>
  30 #include <linux/ctype.h>
  31 #include <linux/cache.h>
  32 #include <linux/init.h>
  33 #include <linux/signal.h>
  34
  35 #include <asm/ppcdebug.h>
  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 /*
  63  * Note:  pte   --> Linux PTE
  64  *        HPTE  --> PowerPC Hashed Page Table Entry
  65  *
  66  * Execution context:
  67  *   htab_initialize is called with the MMU off (of course), but
  68  *   the kernel has been copied down to zero so it can directly
  69  *   reference global data.  At this point it is very difficult
  70  *   to print debug info.
  71  *
  72  */
  73
  74 #ifdef CONFIG_U3_DART
  75 extern unsigned long dart_tablebase;
  76 #endif /* CONFIG_U3_DART */
  77
  78 hpte_t *htab_address;
  79 unsigned long htab_hash_mask;
  80
  81 extern unsigned long _SDR1;
  82
  83 #define KB (1024)
  84 #define MB (1024*KB)
  85
  86 static inline void loop_forever(void)
  87 {
  88         volatile unsigned long x = 1;
  89         for(;x;x|=1)
  90                 ;
  91 }
  92
  93 #ifdef CONFIG_PPC_MULTIPLATFORM
  94 static inline void create_pte_mapping(unsigned long start, unsigned long end,
  95                                       unsigned long mode, int large)
  96 {
  97         unsigned long addr;
  98         unsigned int step;
  99         unsigned long tmp_mode;
 100         unsigned long vflags;
 101
 102         if (large) {
 103                 step = 16*MB;
 104                 vflags = HPTE_V_BOLTED | HPTE_V_LARGE;
 105         } else {
 106                 step = 4*KB;
 107                 vflags = HPTE_V_BOLTED;
 108         }
 109
 110         for (addr = start; addr < end; addr += step) {
 111                 unsigned long vpn, hash, hpteg;
 112                 unsigned long vsid = get_kernel_vsid(addr);
 113                 unsigned long va = (vsid << 28) | (addr & 0xfffffff);
 114                 int ret;
 115
 116                 if (large)
 117                         vpn = va >> HPAGE_SHIFT;
 118                 else
 119                         vpn = va >> PAGE_SHIFT;
 120
 121
 122                 tmp_mode = mode;
 123
 124                 /* Make non-kernel text non-executable */
 125                 if (!in_kernel_text(addr))
 126                         tmp_mode = mode | HW_NO_EXEC;
 127
 128                 hash = hpt_hash(vpn, large);
 129
 130                 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
 131
 132 #ifdef CONFIG_PPC_PSERIES
 133                 if (systemcfg->platform & PLATFORM_LPAR)
 134                         ret = pSeries_lpar_hpte_insert(hpteg, va,
 135                                 virt_to_abs(addr) >> PAGE_SHIFT,
 136                                 vflags, tmp_mode);
 137                 else
 138 #endif /* CONFIG_PPC_PSERIES */
 139                         ret = native_hpte_insert(hpteg, va,
 140                                 virt_to_abs(addr) >> PAGE_SHIFT,
 141                                 vflags, tmp_mode);
 142
 143                 if (ret == -1) {
 144                         ppc64_terminate_msg(0x20, "create_pte_mapping");
 145                         loop_forever();
 146                 }
 147         }
 148 }
 149
 150 void __init htab_initialize(void)
 151 {
 152         unsigned long table, htab_size_bytes;
 153         unsigned long pteg_count;
 154         unsigned long mode_rw;
 155         int i, use_largepages = 0;
 156         unsigned long base = 0, size = 0;
 157         extern unsigned long tce_alloc_start, tce_alloc_end;
 158
 159         DBG(" -> htab_initialize()\n");
 160
 161         /*
 162          * Calculate the required size of the htab.  We want the number of
 163          * PTEGs to equal one half the number of real pages.
 164          */
 165         htab_size_bytes = 1UL << ppc64_pft_size;
 166         pteg_count = htab_size_bytes >> 7;
 167
 168         /* For debug, make the HTAB 1/8 as big as it normally would be. */
 169         ifppcdebug(PPCDBG_HTABSIZE) {
 170                 pteg_count >>= 3;
 171                 htab_size_bytes = pteg_count << 7;
 172         }
 173
 174         htab_hash_mask = pteg_count - 1;
 175
 176         if (systemcfg->platform & PLATFORM_LPAR) {
 177                 /* Using a hypervisor which owns the htab */
 178                 htab_address = NULL;
 179                 _SDR1 = 0;
 180         } else {
 181                 /* Find storage for the HPT.  Must be contiguous in
 182                  * the absolute address space.
 183                  */
 184                 table = lmb_alloc(htab_size_bytes, htab_size_bytes);
 185
 186                 DBG("Hash table allocated at %lx, size: %lx\n", table,
 187                     htab_size_bytes);
 188
 189                 if ( !table ) {
 190                         ppc64_terminate_msg(0x20, "hpt space");
 191                         loop_forever();
 192                 }
 193                 htab_address = abs_to_virt(table);
 194
 195                 /* htab absolute addr + encoded htabsize */
 196                 _SDR1 = table + __ilog2(pteg_count) - 11;
 197
 198                 /* Initialize the HPT with no entries */
 199                 memset((void *)table, 0, htab_size_bytes);
 200         }
 201
 202         mode_rw = _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX;
 203
 204         /* On U3 based machines, we need to reserve the DART area and
 205          * _NOT_ map it to avoid cache paradoxes as it's remapped non
 206          * cacheable later on
 207          */
 208         if (cpu_has_feature(CPU_FTR_16M_PAGE))
 209                 use_largepages = 1;
 210
 211         /* create bolted the linear mapping in the hash table */
 212         for (i=0; i < lmb.memory.cnt; i++) {
 213                 base = lmb.memory.region[i].physbase + KERNELBASE;
 214                 size = lmb.memory.region[i].size;
 215
 216                 DBG("creating mapping for region: %lx : %lx\n", base, size);
 217
 218 #ifdef CONFIG_U3_DART
 219                 /* Do not map the DART space. Fortunately, it will be aligned
 220                  * in such a way that it will not cross two lmb regions and will
 221                  * fit within a single 16Mb page.
 222                  * The DART space is assumed to be a full 16Mb region even if we
 223                  * only use 2Mb of that space. We will use more of it later for
 224                  * AGP GART. We have to use a full 16Mb large page.
 225                  */
 226                 DBG("DART base: %lx\n", dart_tablebase);
 227
 228                 if (dart_tablebase != 0 && dart_tablebase >= base
 229                     && dart_tablebase < (base + size)) {
 230                         if (base != dart_tablebase)
 231                                 create_pte_mapping(base, dart_tablebase, mode_rw,
 232                                                    use_largepages);
 233                         if ((base + size) > (dart_tablebase + 16*MB))
 234                                 create_pte_mapping(dart_tablebase + 16*MB, base + size,
 235                                                    mode_rw, use_largepages);
 236                         continue;
 237                 }
 238 #endif /* CONFIG_U3_DART */
 239                 create_pte_mapping(base, base + size, mode_rw, use_largepages);
 240         }
 241
 242         /*
 243          * If we have a memory_limit and we've allocated TCEs then we need to
 244          * explicitly map the TCE area at the top of RAM. We also cope with the
 245          * case that the TCEs start below memory_limit.
 246          * tce_alloc_start/end are 16MB aligned so the mapping should work
 247          * for either 4K or 16MB pages.
 248          */
 249         if (tce_alloc_start) {
 250                 tce_alloc_start += KERNELBASE;
 251                 tce_alloc_end += KERNELBASE;
 252
 253                 if (base + size >= tce_alloc_start)
 254                         tce_alloc_start = base + size + 1;
 255
 256                 create_pte_mapping(tce_alloc_start, tce_alloc_end,
 257                         mode_rw, use_largepages);
 258         }
 259
 260         DBG(" <- htab_initialize()\n");
 261 }
 262 #undef KB
 263 #undef MB
 264 #endif /* CONFIG_PPC_MULTIPLATFORM */
 265
 266 /*
 267  * Called by asm hashtable.S for doing lazy icache flush
 268  */
 269 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
 270 {
 271         struct page *page;
 272
 273         if (!pfn_valid(pte_pfn(pte)))
 274                 return pp;
 275
 276         page = pte_page(pte);
 277
 278         /* page is dirty */
 279         if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
 280                 if (trap == 0x400) {
 281                         __flush_dcache_icache(page_address(page));
 282                         set_bit(PG_arch_1, &page->flags);
 283                 } else
 284                         pp |= HW_NO_EXEC;
 285         }
 286         return pp;
 287 }
 288
 289 /* Result code is:
 290  *  0 - handled
 291  *  1 - normal page fault
 292  * -1 - critical hash insertion error
 293  */
 294 int hash_page(unsigned long ea, unsigned long access, unsigned long trap)
 295 {
 296         void *pgdir;
 297         unsigned long vsid;
 298         struct mm_struct *mm;
 299         pte_t *ptep;
 300         int ret;
 301         int user_region = 0;
 302         int local = 0;
 303         cpumask_t tmp;
 304
 305         if ((ea & ~REGION_MASK) > EADDR_MASK)
 306                 return 1;
 307
 308         switch (REGION_ID(ea)) {
 309         case USER_REGION_ID:
 310                 user_region = 1;
 311                 mm = current->mm;
 312                 if (! mm)
 313                         return 1;
 314
 315                 vsid = get_vsid(mm->context.id, ea);
 316                 break;
 317         case VMALLOC_REGION_ID:
 318                 mm = &init_mm;
 319                 vsid = get_kernel_vsid(ea);
 320                 break;
 321 #if 0
 322         case KERNEL_REGION_ID:
 323                 /*
 324                  * Should never get here - entire 0xC0... region is bolted.
 325                  * Send the problem up to do_page_fault
 326                  */
 327 #endif
 328         default:
 329                 /* Not a valid range
 330                  * Send the problem up to do_page_fault
 331                  */
 332                 return 1;
 333                 break;
 334         }
 335
 336         pgdir = mm->pgd;
 337
 338         if (pgdir == NULL)
 339                 return 1;
 340
 341         tmp = cpumask_of_cpu(smp_processor_id());
 342         if (user_region && cpus_equal(mm->cpu_vm_mask, tmp))
 343                 local = 1;
 344
 345         /* Is this a huge page ? */
 346         if (unlikely(in_hugepage_area(mm->context, ea)))
 347                 ret = hash_huge_page(mm, access, ea, vsid, local);
 348         else {
 349                 ptep = find_linux_pte(pgdir, ea);
 350                 if (ptep == NULL)
 351                         return 1;
 352                 ret = __hash_page(ea, access, vsid, ptep, trap, local);
 353         }
 354
 355         return ret;
 356 }
 357
 358 void flush_hash_page(unsigned long context, unsigned long ea, pte_t pte,
 359                      int local)
 360 {
 361         unsigned long vsid, vpn, va, hash, secondary, slot;
 362         unsigned long huge = pte_huge(pte);
 363
 364         if (ea < KERNELBASE)
 365                 vsid = get_vsid(context, ea);
 366         else
 367                 vsid = get_kernel_vsid(ea);
 368
 369         va = (vsid << 28) | (ea & 0x0fffffff);
 370         if (huge)
 371                 vpn = va >> HPAGE_SHIFT;
 372         else
 373                 vpn = va >> PAGE_SHIFT;
 374         hash = hpt_hash(vpn, huge);
 375         secondary = (pte_val(pte) & _PAGE_SECONDARY) >> 15;
 376         if (secondary)
 377                 hash = ~hash;
 378         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 379         slot += (pte_val(pte) & _PAGE_GROUP_IX) >> 12;
 380
 381         ppc_md.hpte_invalidate(slot, va, huge, local);
 382 }
 383
 384 void flush_hash_range(unsigned long context, unsigned long number, int local)
 385 {
 386         if (ppc_md.flush_hash_range) {
 387                 ppc_md.flush_hash_range(context, number, local);
 388         } else {
 389                 int i;
 390                 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
 391
 392                 for (i = 0; i < number; i++)
 393                         flush_hash_page(context, batch->addr[i], batch->pte[i],
 394                                         local);
 395         }
 396 }
 397
 398 static inline void make_bl(unsigned int *insn_addr, void *func)
 399 {
 400         unsigned long funcp = *((unsigned long *)func);
 401         int offset = funcp - (unsigned long)insn_addr;
 402
 403         *insn_addr = (unsigned int)(0x48000001 | (offset & 0x03fffffc));
 404         flush_icache_range((unsigned long)insn_addr, 4+
 405                            (unsigned long)insn_addr);
 406 }
 407
 408 /*
 409  * low_hash_fault is called when we the low level hash code failed
 410  * to instert a PTE due to an hypervisor error
 411  */
 412 void low_hash_fault(struct pt_regs *regs, unsigned long address)
 413 {
 414         if (user_mode(regs)) {
 415                 siginfo_t info;
 416
 417                 info.si_signo = SIGBUS;
 418                 info.si_errno = 0;
 419                 info.si_code = BUS_ADRERR;
 420                 info.si_addr = (void __user *)address;
 421                 force_sig_info(SIGBUS, &info, current);
 422                 return;
 423         }
 424         bad_page_fault(regs, address, SIGBUS);
 425 }
 426
 427 void __init htab_finish_init(void)
 428 {
 429         extern unsigned int *htab_call_hpte_insert1;
 430         extern unsigned int *htab_call_hpte_insert2;
 431         extern unsigned int *htab_call_hpte_remove;
 432         extern unsigned int *htab_call_hpte_updatepp;
 433
 434         make_bl(htab_call_hpte_insert1, ppc_md.hpte_insert);
 435         make_bl(htab_call_hpte_insert2, ppc_md.hpte_insert);
 436         make_bl(htab_call_hpte_remove, ppc_md.hpte_remove);
 437         make_bl(htab_call_hpte_updatepp, ppc_md.hpte_updatepp);
 438 }