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