arch/sh64/mm/fault.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * arch/sh64/mm/fault.c
   7  *
   8  * Copyright (C) 2000, 2001  Paolo Alberelli
   9  * Copyright (C) 2003  Richard Curnow (/proc/tlb, bug fixes)
  10  * Copyright (C) 2003  Paul Mundt
  11  *
  12  */
  13
  14 #include <linux/signal.h>
  15 #include <linux/rwsem.h>
  16 #include <linux/sched.h>
  17 #include <linux/kernel.h>
  18 #include <linux/errno.h>
  19 #include <linux/string.h>
  20 #include <linux/types.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/mman.h>
  23 #include <linux/mm.h>
  24 #include <linux/smp.h>
  25 #include <linux/interrupt.h>
  26
  27 #include <asm/system.h>
  28 #include <asm/io.h>
  29 #include <asm/tlb.h>
  30 #include <asm/uaccess.h>
  31 #include <asm/pgalloc.h>
  32 #include <asm/mmu_context.h>
  33 #include <asm/registers.h>              /* required by inline asm statements */
  34
  35 #if defined(CONFIG_SH64_PROC_TLB)
  36 #include <linux/init.h>
  37 #include <linux/proc_fs.h>
  38 /* Count numbers of tlb refills in each region */
  39 static unsigned long long calls_to_update_mmu_cache = 0ULL;
  40 static unsigned long long calls_to_flush_tlb_page   = 0ULL;
  41 static unsigned long long calls_to_flush_tlb_range  = 0ULL;
  42 static unsigned long long calls_to_flush_tlb_mm     = 0ULL;
  43 static unsigned long long calls_to_flush_tlb_all    = 0ULL;
  44 unsigned long long calls_to_do_slow_page_fault = 0ULL;
  45 unsigned long long calls_to_do_fast_page_fault = 0ULL;
  46
  47 /* Count size of ranges for flush_tlb_range */
  48 static unsigned long long flush_tlb_range_1         = 0ULL;
  49 static unsigned long long flush_tlb_range_2         = 0ULL;
  50 static unsigned long long flush_tlb_range_3_4       = 0ULL;
  51 static unsigned long long flush_tlb_range_5_7       = 0ULL;
  52 static unsigned long long flush_tlb_range_8_11      = 0ULL;
  53 static unsigned long long flush_tlb_range_12_15     = 0ULL;
  54 static unsigned long long flush_tlb_range_16_up     = 0ULL;
  55
  56 static unsigned long long page_not_present          = 0ULL;
  57
  58 #endif
  59
  60 extern void die(const char *,struct pt_regs *,long);
  61
  62 #define PFLAG(val,flag)   (( (val) & (flag) ) ? #flag : "" )
  63 #define PPROT(flag) PFLAG(pgprot_val(prot),flag)
  64
  65 static inline void print_prots(pgprot_t prot)
  66 {
  67         printk("prot is 0x%08lx\n",pgprot_val(prot));
  68
  69         printk("%s %s %s %s %s\n",PPROT(_PAGE_SHARED),PPROT(_PAGE_READ),
  70                PPROT(_PAGE_EXECUTE),PPROT(_PAGE_WRITE),PPROT(_PAGE_USER));
  71 }
  72
  73 static inline void print_vma(struct vm_area_struct *vma)
  74 {
  75         printk("vma start 0x%08lx\n", vma->vm_start);
  76         printk("vma end   0x%08lx\n", vma->vm_end);
  77
  78         print_prots(vma->vm_page_prot);
  79         printk("vm_flags 0x%08lx\n", vma->vm_flags);
  80 }
  81
  82 static inline void print_task(struct task_struct *tsk)
  83 {
  84         printk("Task pid %d\n", tsk->pid);
  85 }
  86
  87 static pte_t *lookup_pte(struct mm_struct *mm, unsigned long address)
  88 {
  89         pgd_t *dir;
  90         pmd_t *pmd;
  91         pte_t *pte;
  92         pte_t entry;
  93
  94         dir = pgd_offset(mm, address);
  95         if (pgd_none(*dir)) {
  96                 return NULL;
  97         }
  98
  99         pmd = pmd_offset(dir, address);
 100         if (pmd_none(*pmd)) {
 101                 return NULL;
 102         }
 103
 104         pte = pte_offset_kernel(pmd, address);
 105         entry = *pte;
 106
 107         if (pte_none(entry)) {
 108                 return NULL;
 109         }
 110         if (!pte_present(entry)) {
 111                 return NULL;
 112         }
 113
 114         return pte;
 115 }
 116
 117 /*
 118  * This routine handles page faults.  It determines the address,
 119  * and the problem, and then passes it off to one of the appropriate
 120  * routines.
 121  */
 122 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long writeaccess,
 123                               unsigned long textaccess, unsigned long address)
 124 {
 125         struct task_struct *tsk;
 126         struct mm_struct *mm;
 127         struct vm_area_struct * vma;
 128         const struct exception_table_entry *fixup;
 129         pte_t *pte;
 130
 131 #if defined(CONFIG_SH64_PROC_TLB)
 132         ++calls_to_do_slow_page_fault;
 133 #endif
 134
 135         /* SIM
 136          * Note this is now called with interrupts still disabled
 137          * This is to cope with being called for a missing IO port
 138          * address with interupts disabled. This should be fixed as
 139          * soon as we have a better 'fast path' miss handler.
 140          *
 141          * Plus take care how you try and debug this stuff.
 142          * For example, writing debug data to a port which you
 143          * have just faulted on is not going to work.
 144          */
 145
 146         tsk = current;
 147         mm = tsk->mm;
 148
 149         /* Not an IO address, so reenable interrupts */
 150         local_irq_enable();
 151
 152         /*
 153          * If we're in an interrupt or have no user
 154          * context, we must not take the fault..
 155          */
 156         if (in_atomic() || !mm)
 157                 goto no_context;
 158
 159         /* TLB misses upon some cache flushes get done under cli() */
 160         down_read(&mm->mmap_sem);
 161
 162         vma = find_vma(mm, address);
 163
 164         if (!vma) {
 165 #ifdef DEBUG_FAULT
 166                 print_task(tsk);
 167                 printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
 168                        __FUNCTION__,__LINE__,
 169                        address,regs->pc,textaccess,writeaccess);
 170                 show_regs(regs);
 171 #endif
 172                 goto bad_area;
 173         }
 174         if (vma->vm_start <= address) {
 175                 goto good_area;
 176         }
 177
 178         if (!(vma->vm_flags & VM_GROWSDOWN)) {
 179 #ifdef DEBUG_FAULT
 180                 print_task(tsk);
 181                 printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
 182                        __FUNCTION__,__LINE__,
 183                        address,regs->pc,textaccess,writeaccess);
 184                 show_regs(regs);
 185
 186                 print_vma(vma);
 187 #endif
 188                 goto bad_area;
 189         }
 190         if (expand_stack(vma, address)) {
 191 #ifdef DEBUG_FAULT
 192                 print_task(tsk);
 193                 printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
 194                        __FUNCTION__,__LINE__,
 195                        address,regs->pc,textaccess,writeaccess);
 196                 show_regs(regs);
 197 #endif
 198                 goto bad_area;
 199         }
 200 /*
 201  * Ok, we have a good vm_area for this memory access, so
 202  * we can handle it..
 203  */
 204 good_area:
 205         if (textaccess) {
 206                 if (!(vma->vm_flags & VM_EXEC))
 207                         goto bad_area;
 208         } else {
 209                 if (writeaccess) {
 210                         if (!(vma->vm_flags & VM_WRITE))
 211                                 goto bad_area;
 212                 } else {
 213                         if (!(vma->vm_flags & VM_READ))
 214                                 goto bad_area;
 215                 }
 216         }
 217
 218         /*
 219          * If for any reason at all we couldn't handle the fault,
 220          * make sure we exit gracefully rather than endlessly redo
 221          * the fault.
 222          */
 223 survive:
 224         switch (handle_mm_fault(mm, vma, address, writeaccess)) {
 225         case VM_FAULT_MINOR:
 226                 tsk->min_flt++;
 227                 break;
 228         case VM_FAULT_MAJOR:
 229                 tsk->maj_flt++;
 230                 break;
 231         case VM_FAULT_SIGBUS:
 232                 goto do_sigbus;
 233         default:
 234                 goto out_of_memory;
 235         }
 236         /* If we get here, the page fault has been handled.  Do the TLB refill
 237            now from the newly-setup PTE, to avoid having to fault again right
 238            away on the same instruction. */
 239         pte = lookup_pte (mm, address);
 240         if (!pte) {
 241                 /* From empirical evidence, we can get here, due to
 242                    !pte_present(pte).  (e.g. if a swap-in occurs, and the page
 243                    is swapped back out again before the process that wanted it
 244                    gets rescheduled?) */
 245                 goto no_pte;
 246         }
 247
 248         __do_tlb_refill(address, textaccess, pte);
 249
 250 no_pte:
 251
 252         up_read(&mm->mmap_sem);
 253         return;
 254
 255 /*
 256  * Something tried to access memory that isn't in our memory map..
 257  * Fix it, but check if it's kernel or user first..
 258  */
 259 bad_area:
 260 #ifdef DEBUG_FAULT
 261         printk("fault:bad area\n");
 262 #endif
 263         up_read(&mm->mmap_sem);
 264
 265         if (user_mode(regs)) {
 266                 static int count=0;
 267                 siginfo_t info;
 268                 if (count < 4) {
 269                         /* This is really to help debug faults when starting
 270                          * usermode, so only need a few */
 271                         count++;
 272                         printk("user mode bad_area address=%08lx pid=%d (%s) pc=%08lx\n",
 273                                 address, current->pid, current->comm,
 274                                 (unsigned long) regs->pc);
 275 #if 0
 276                         show_regs(regs);
 277 #endif
 278                 }
 279                 if (is_init(tsk)) {
 280                         panic("INIT had user mode bad_area\n");
 281                 }
 282                 tsk->thread.address = address;
 283                 tsk->thread.error_code = writeaccess;
 284                 info.si_signo = SIGSEGV;
 285                 info.si_errno = 0;
 286                 info.si_addr = (void *) address;
 287                 force_sig_info(SIGSEGV, &info, tsk);
 288                 return;
 289         }
 290
 291 no_context:
 292 #ifdef DEBUG_FAULT
 293         printk("fault:No context\n");
 294 #endif
 295         /* Are we prepared to handle this kernel fault?  */
 296         fixup = search_exception_tables(regs->pc);
 297         if (fixup) {
 298                 regs->pc = fixup->fixup;
 299                 return;
 300         }
 301
 302 /*
 303  * Oops. The kernel tried to access some bad page. We'll have to
 304  * terminate things with extreme prejudice.
 305  *
 306  */
 307         if (address < PAGE_SIZE)
 308                 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference");
 309         else
 310                 printk(KERN_ALERT "Unable to handle kernel paging request");
 311         printk(" at virtual address %08lx\n", address);
 312         printk(KERN_ALERT "pc = %08Lx%08Lx\n", regs->pc >> 32, regs->pc & 0xffffffff);
 313         die("Oops", regs, writeaccess);
 314         do_exit(SIGKILL);
 315
 316 /*
 317  * We ran out of memory, or some other thing happened to us that made
 318  * us unable to handle the page fault gracefully.
 319  */
 320 out_of_memory:
 321         if (is_init(current)) {
 322                 panic("INIT out of memory\n");
 323                 yield();
 324                 goto survive;
 325         }
 326         printk("fault:Out of memory\n");
 327         up_read(&mm->mmap_sem);
 328         if (is_init(current)) {
 329                 yield();
 330                 down_read(&mm->mmap_sem);
 331                 goto survive;
 332         }
 333         printk("VM: killing process %s\n", tsk->comm);
 334         if (user_mode(regs))
 335                 do_exit(SIGKILL);
 336         goto no_context;
 337
 338 do_sigbus:
 339         printk("fault:Do sigbus\n");
 340         up_read(&mm->mmap_sem);
 341
 342         /*
 343          * Send a sigbus, regardless of whether we were in kernel
 344          * or user mode.
 345          */
 346         tsk->thread.address = address;
 347         tsk->thread.error_code = writeaccess;
 348         tsk->thread.trap_no = 14;
 349         force_sig(SIGBUS, tsk);
 350
 351         /* Kernel mode? Handle exceptions or die */
 352         if (!user_mode(regs))
 353                 goto no_context;
 354 }
 355
 356
 357 void flush_tlb_all(void);
 358
 359 void update_mmu_cache(struct vm_area_struct * vma,
 360                         unsigned long address, pte_t pte)
 361 {
 362 #if defined(CONFIG_SH64_PROC_TLB)
 363         ++calls_to_update_mmu_cache;
 364 #endif
 365
 366         /*
 367          * This appears to get called once for every pte entry that gets
 368          * established => I don't think it's efficient to try refilling the
 369          * TLBs with the pages - some may not get accessed even.  Also, for
 370          * executable pages, it is impossible to determine reliably here which
 371          * TLB they should be mapped into (or both even).
 372          *
 373          * So, just do nothing here and handle faults on demand.  In the
 374          * TLBMISS handling case, the refill is now done anyway after the pte
 375          * has been fixed up, so that deals with most useful cases.
 376          */
 377 }
 378
 379 static void __flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 380 {
 381         unsigned long long match, pteh=0, lpage;
 382         unsigned long tlb;
 383         struct mm_struct *mm;
 384
 385         mm = vma->vm_mm;
 386
 387         if (mm->context == NO_CONTEXT)
 388                 return;
 389
 390         /*
 391          * Sign-extend based on neff.
 392          */
 393         lpage = (page & NEFF_SIGN) ? (page | NEFF_MASK) : page;
 394         match = ((mm->context & MMU_CONTEXT_ASID_MASK) << PTEH_ASID_SHIFT) | PTEH_VALID;
 395         match |= lpage;
 396
 397         /* Do ITLB : don't bother for pages in non-exectutable VMAs */
 398         if (vma->vm_flags & VM_EXEC) {
 399                 for_each_itlb_entry(tlb) {
 400                         asm volatile ("getcfg   %1, 0, %0"
 401                                       : "=r" (pteh)
 402                                       : "r" (tlb) );
 403
 404                         if (pteh == match) {
 405                                 __flush_tlb_slot(tlb);
 406                                 break;
 407                         }
 408
 409                 }
 410         }
 411
 412         /* Do DTLB : any page could potentially be in here. */
 413         for_each_dtlb_entry(tlb) {
 414                 asm volatile ("getcfg   %1, 0, %0"
 415                               : "=r" (pteh)
 416                               : "r" (tlb) );
 417
 418                 if (pteh == match) {
 419                         __flush_tlb_slot(tlb);
 420                         break;
 421                 }
 422
 423         }
 424 }
 425
 426 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 427 {
 428         unsigned long flags;
 429
 430 #if defined(CONFIG_SH64_PROC_TLB)
 431         ++calls_to_flush_tlb_page;
 432 #endif
 433
 434         if (vma->vm_mm) {
 435                 page &= PAGE_MASK;
 436                 local_irq_save(flags);
 437                 __flush_tlb_page(vma, page);
 438                 local_irq_restore(flags);
 439         }
 440 }
 441
 442 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 443                      unsigned long end)
 444 {
 445         unsigned long flags;
 446         unsigned long long match, pteh=0, pteh_epn, pteh_low;
 447         unsigned long tlb;
 448         struct mm_struct *mm;
 449
 450         mm = vma->vm_mm;
 451
 452 #if defined(CONFIG_SH64_PROC_TLB)
 453         ++calls_to_flush_tlb_range;
 454
 455         {
 456                 unsigned long size = (end - 1) - start;
 457                 size >>= 12; /* divide by PAGE_SIZE */
 458                 size++; /* end=start+4096 => 1 page */
 459                 switch (size) {
 460                   case  1        : flush_tlb_range_1++;     break;
 461                   case  2        : flush_tlb_range_2++;     break;
 462                   case  3 ...  4 : flush_tlb_range_3_4++;   break;
 463                   case  5 ...  7 : flush_tlb_range_5_7++;   break;
 464                   case  8 ... 11 : flush_tlb_range_8_11++;  break;
 465                   case 12 ... 15 : flush_tlb_range_12_15++; break;
 466                   default        : flush_tlb_range_16_up++; break;
 467                 }
 468         }
 469 #endif
 470
 471         if (mm->context == NO_CONTEXT)
 472                 return;
 473
 474         local_irq_save(flags);
 475
 476         start &= PAGE_MASK;
 477         end &= PAGE_MASK;
 478
 479         match = ((mm->context & MMU_CONTEXT_ASID_MASK) << PTEH_ASID_SHIFT) | PTEH_VALID;
 480
 481         /* Flush ITLB */
 482         for_each_itlb_entry(tlb) {
 483                 asm volatile ("getcfg   %1, 0, %0"
 484                               : "=r" (pteh)
 485                               : "r" (tlb) );
 486
 487                 pteh_epn = pteh & PAGE_MASK;
 488                 pteh_low = pteh & ~PAGE_MASK;
 489
 490                 if (pteh_low == match && pteh_epn >= start && pteh_epn <= end)
 491                         __flush_tlb_slot(tlb);
 492         }
 493
 494         /* Flush DTLB */
 495         for_each_dtlb_entry(tlb) {
 496                 asm volatile ("getcfg   %1, 0, %0"
 497                               : "=r" (pteh)
 498                               : "r" (tlb) );
 499
 500                 pteh_epn = pteh & PAGE_MASK;
 501                 pteh_low = pteh & ~PAGE_MASK;
 502
 503                 if (pteh_low == match && pteh_epn >= start && pteh_epn <= end)
 504                         __flush_tlb_slot(tlb);
 505         }
 506
 507         local_irq_restore(flags);
 508 }
 509
 510 void flush_tlb_mm(struct mm_struct *mm)
 511 {
 512         unsigned long flags;
 513
 514 #if defined(CONFIG_SH64_PROC_TLB)
 515         ++calls_to_flush_tlb_mm;
 516 #endif
 517
 518         if (mm->context == NO_CONTEXT)
 519                 return;
 520
 521         local_irq_save(flags);
 522
 523         mm->context=NO_CONTEXT;
 524         if(mm==current->mm)
 525                 activate_context(mm);
 526
 527         local_irq_restore(flags);
 528
 529 }
 530
 531 void flush_tlb_all(void)
 532 {
 533         /* Invalidate all, including shared pages, excluding fixed TLBs */
 534
 535         unsigned long flags, tlb;
 536
 537 #if defined(CONFIG_SH64_PROC_TLB)
 538         ++calls_to_flush_tlb_all;
 539 #endif
 540
 541         local_irq_save(flags);
 542
 543         /* Flush each ITLB entry */
 544         for_each_itlb_entry(tlb) {
 545                 __flush_tlb_slot(tlb);
 546         }
 547
 548         /* Flush each DTLB entry */
 549         for_each_dtlb_entry(tlb) {
 550                 __flush_tlb_slot(tlb);
 551         }
 552
 553         local_irq_restore(flags);
 554 }
 555
 556 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 557 {
 558         /* FIXME: Optimize this later.. */
 559         flush_tlb_all();
 560 }
 561
 562 #if defined(CONFIG_SH64_PROC_TLB)
 563 /* Procfs interface to read the performance information */
 564
 565 static int
 566 tlb_proc_info(char *buf, char **start, off_t fpos, int length, int *eof, void *data)
 567 {
 568   int len=0;
 569   len += sprintf(buf+len, "do_fast_page_fault   called %12lld times\n", calls_to_do_fast_page_fault);
 570   len += sprintf(buf+len, "do_slow_page_fault   called %12lld times\n", calls_to_do_slow_page_fault);
 571   len += sprintf(buf+len, "update_mmu_cache     called %12lld times\n", calls_to_update_mmu_cache);
 572   len += sprintf(buf+len, "flush_tlb_page       called %12lld times\n", calls_to_flush_tlb_page);
 573   len += sprintf(buf+len, "flush_tlb_range      called %12lld times\n", calls_to_flush_tlb_range);
 574   len += sprintf(buf+len, "flush_tlb_mm         called %12lld times\n", calls_to_flush_tlb_mm);
 575   len += sprintf(buf+len, "flush_tlb_all        called %12lld times\n", calls_to_flush_tlb_all);
 576   len += sprintf(buf+len, "flush_tlb_range_sizes\n"
 577                           " 1      : %12lld\n"
 578                           " 2      : %12lld\n"
 579                           " 3 -  4 : %12lld\n"
 580                           " 5 -  7 : %12lld\n"
 581                           " 8 - 11 : %12lld\n"
 582                           "12 - 15 : %12lld\n"
 583                           "16+     : %12lld\n",
 584                           flush_tlb_range_1, flush_tlb_range_2, flush_tlb_range_3_4,
 585                           flush_tlb_range_5_7, flush_tlb_range_8_11, flush_tlb_range_12_15,
 586                           flush_tlb_range_16_up);
 587   len += sprintf(buf+len, "page not present           %12lld times\n", page_not_present);
 588   *eof = 1;
 589   return len;
 590 }
 591
 592 static int __init register_proc_tlb(void)
 593 {
 594   create_proc_read_entry("tlb", 0, NULL, tlb_proc_info, NULL);
 595   return 0;
 596 }
 597
 598 __initcall(register_proc_tlb);
 599
 600 #endif