arch/sparc64/mm/fault.c

   1 /* $Id: fault.c,v 1.59 2002/02/09 19:49:31 davem Exp $
   2  * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
   3  *
   4  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
   5  * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
   6  */
   7
   8 #include <asm/head.h>
   9
  10 #include <linux/string.h>
  11 #include <linux/types.h>
  12 #include <linux/sched.h>
  13 #include <linux/ptrace.h>
  14 #include <linux/mman.h>
  15 #include <linux/signal.h>
  16 #include <linux/mm.h>
  17 #include <linux/module.h>
  18 #include <linux/smp_lock.h>
  19 #include <linux/init.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/kprobes.h>
  22 #include <linux/kallsyms.h>
  23
  24 #include <asm/page.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/openprom.h>
  27 #include <asm/oplib.h>
  28 #include <asm/uaccess.h>
  29 #include <asm/asi.h>
  30 #include <asm/lsu.h>
  31 #include <asm/sections.h>
  32 #include <asm/kdebug.h>
  33 #include <asm/mmu_context.h>
  34
  35 #ifdef CONFIG_KPROBES
  36 ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
  37
  38 /* Hook to register for page fault notifications */
  39 int register_page_fault_notifier(struct notifier_block *nb)
  40 {
  41         return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
  42 }
  43
  44 int unregister_page_fault_notifier(struct notifier_block *nb)
  45 {
  46         return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
  47 }
  48
  49 static inline int notify_page_fault(enum die_val val, const char *str,
  50                         struct pt_regs *regs, long err, int trap, int sig)
  51 {
  52         struct die_args args = {
  53                 .regs = regs,
  54                 .str = str,
  55                 .err = err,
  56                 .trapnr = trap,
  57                 .signr = sig
  58         };
  59         return atomic_notifier_call_chain(&notify_page_fault_chain, val, &args);
  60 }
  61 #else
  62 static inline int notify_page_fault(enum die_val val, const char *str,
  63                         struct pt_regs *regs, long err, int trap, int sig)
  64 {
  65         return NOTIFY_DONE;
  66 }
  67 #endif
  68
  69 /*
  70  * To debug kernel to catch accesses to certain virtual/physical addresses.
  71  * Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints.
  72  * flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses.
  73  * Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be
  74  * watched. This is only useful on a single cpu machine for now. After the watchpoint
  75  * is detected, the process causing it will be killed, thus preventing an infinite loop.
  76  */
  77 void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode)
  78 {
  79         unsigned long lsubits;
  80
  81         __asm__ __volatile__("ldxa [%%g0] %1, %0"
  82                              : "=r" (lsubits)
  83                              : "i" (ASI_LSU_CONTROL));
  84         lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM |
  85                      LSU_CONTROL_PR | LSU_CONTROL_VR |
  86                      LSU_CONTROL_PW | LSU_CONTROL_VW);
  87
  88         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
  89                              "membar    #Sync"
  90                              : /* no outputs */
  91                              : "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT),
  92                                "i" (ASI_DMMU));
  93
  94         lsubits |= ((unsigned long)mask << (mode ? 25 : 33));
  95         if (flags & VM_READ)
  96                 lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR);
  97         if (flags & VM_WRITE)
  98                 lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW);
  99         __asm__ __volatile__("stxa %0, [%%g0] %1\n\t"
 100                              "membar #Sync"
 101                              : /* no outputs */
 102                              : "r" (lsubits), "i" (ASI_LSU_CONTROL)
 103                              : "memory");
 104 }
 105
 106 static void __kprobes unhandled_fault(unsigned long address,
 107                                       struct task_struct *tsk,
 108                                       struct pt_regs *regs)
 109 {
 110         if ((unsigned long) address < PAGE_SIZE) {
 111                 printk(KERN_ALERT "Unable to handle kernel NULL "
 112                        "pointer dereference\n");
 113         } else {
 114                 printk(KERN_ALERT "Unable to handle kernel paging request "
 115                        "at virtual address %016lx\n", (unsigned long)address);
 116         }
 117         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
 118                (tsk->mm ?
 119                 CTX_HWBITS(tsk->mm->context) :
 120                 CTX_HWBITS(tsk->active_mm->context)));
 121         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
 122                (tsk->mm ? (unsigned long) tsk->mm->pgd :
 123                           (unsigned long) tsk->active_mm->pgd));
 124         if (notify_die(DIE_GPF, "general protection fault", regs,
 125                        0, 0, SIGSEGV) == NOTIFY_STOP)
 126                 return;
 127         die_if_kernel("Oops", regs);
 128 }
 129
 130 static void bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
 131 {
 132         unsigned long *ksp;
 133
 134         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
 135                regs->tpc);
 136         printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
 137         print_symbol("RPC: <%s>\n", regs->u_regs[15]);
 138         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
 139         __asm__("mov %%sp, %0" : "=r" (ksp));
 140         show_stack(current, ksp);
 141         unhandled_fault(regs->tpc, current, regs);
 142 }
 143
 144 /*
 145  * We now make sure that mmap_sem is held in all paths that call
 146  * this. Additionally, to prevent kswapd from ripping ptes from
 147  * under us, raise interrupts around the time that we look at the
 148  * pte, kswapd will have to wait to get his smp ipi response from
 149  * us. vmtruncate likewise. This saves us having to get pte lock.
 150  */
 151 static unsigned int get_user_insn(unsigned long tpc)
 152 {
 153         pgd_t *pgdp = pgd_offset(current->mm, tpc);
 154         pud_t *pudp;
 155         pmd_t *pmdp;
 156         pte_t *ptep, pte;
 157         unsigned long pa;
 158         u32 insn = 0;
 159         unsigned long pstate;
 160
 161         if (pgd_none(*pgdp))
 162                 goto outret;
 163         pudp = pud_offset(pgdp, tpc);
 164         if (pud_none(*pudp))
 165                 goto outret;
 166         pmdp = pmd_offset(pudp, tpc);
 167         if (pmd_none(*pmdp))
 168                 goto outret;
 169
 170         /* This disables preemption for us as well. */
 171         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 172         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 173                                 : : "r" (pstate), "i" (PSTATE_IE));
 174         ptep = pte_offset_map(pmdp, tpc);
 175         pte = *ptep;
 176         if (!pte_present(pte))
 177                 goto out;
 178
 179         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 180         pa += (tpc & ~PAGE_MASK);
 181
 182         /* Use phys bypass so we don't pollute dtlb/dcache. */
 183         __asm__ __volatile__("lduwa [%1] %2, %0"
 184                              : "=r" (insn)
 185                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 186
 187 out:
 188         pte_unmap(ptep);
 189         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 190 outret:
 191         return insn;
 192 }
 193
 194 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 195
 196 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 197                              unsigned int insn, int fault_code)
 198 {
 199         siginfo_t info;
 200
 201         info.si_code = code;
 202         info.si_signo = sig;
 203         info.si_errno = 0;
 204         if (fault_code & FAULT_CODE_ITLB)
 205                 info.si_addr = (void __user *) regs->tpc;
 206         else
 207                 info.si_addr = (void __user *)
 208                         compute_effective_address(regs, insn, 0);
 209         info.si_trapno = 0;
 210         force_sig_info(sig, &info, current);
 211 }
 212
 213 extern int handle_ldf_stq(u32, struct pt_regs *);
 214 extern int handle_ld_nf(u32, struct pt_regs *);
 215
 216 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 217 {
 218         if (!insn) {
 219                 if (!regs->tpc || (regs->tpc & 0x3))
 220                         return 0;
 221                 if (regs->tstate & TSTATE_PRIV) {
 222                         insn = *(unsigned int *) regs->tpc;
 223                 } else {
 224                         insn = get_user_insn(regs->tpc);
 225                 }
 226         }
 227         return insn;
 228 }
 229
 230 static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
 231                             unsigned int insn, unsigned long address)
 232 {
 233         unsigned char asi = ASI_P;
 234
 235         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 236                 goto cannot_handle;
 237
 238         /* If user insn could be read (thus insn is zero), that
 239          * is fine.  We will just gun down the process with a signal
 240          * in that case.
 241          */
 242
 243         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 244             (insn & 0xc0800000) == 0xc0800000) {
 245                 if (insn & 0x2000)
 246                         asi = (regs->tstate >> 24);
 247                 else
 248                         asi = (insn >> 5);
 249                 if ((asi & 0xf2) == 0x82) {
 250                         if (insn & 0x1000000) {
 251                                 handle_ldf_stq(insn, regs);
 252                         } else {
 253                                 /* This was a non-faulting load. Just clear the
 254                                  * destination register(s) and continue with the next
 255                                  * instruction. -jj
 256                                  */
 257                                 handle_ld_nf(insn, regs);
 258                         }
 259                         return;
 260                 }
 261         }
 262
 263         /* Is this in ex_table? */
 264         if (regs->tstate & TSTATE_PRIV) {
 265                 const struct exception_table_entry *entry;
 266
 267                 if (asi == ASI_P && (insn & 0xc0800000) == 0xc0800000) {
 268                         if (insn & 0x2000)
 269                                 asi = (regs->tstate >> 24);
 270                         else
 271                                 asi = (insn >> 5);
 272                 }
 273
 274                 /* Look in asi.h: All _S asis have LS bit set */
 275                 if ((asi & 0x1) &&
 276                     (entry = search_exception_tables(regs->tpc))) {
 277                         regs->tpc = entry->fixup;
 278                         regs->tnpc = regs->tpc + 4;
 279                         return;
 280                 }
 281         } else {
 282                 /* The si_code was set to make clear whether
 283                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 284                  */
 285                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 286                 return;
 287         }
 288
 289 cannot_handle:
 290         unhandled_fault (address, current, regs);
 291 }
 292
 293 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 294 {
 295         struct mm_struct *mm = current->mm;
 296         struct vm_area_struct *vma;
 297         unsigned int insn = 0;
 298         int si_code, fault_code;
 299         unsigned long address, mm_rss;
 300
 301         fault_code = get_thread_fault_code();
 302
 303         if (notify_page_fault(DIE_PAGE_FAULT, "page_fault", regs,
 304                        fault_code, 0, SIGSEGV) == NOTIFY_STOP)
 305                 return;
 306
 307         si_code = SEGV_MAPERR;
 308         address = current_thread_info()->fault_address;
 309
 310         if ((fault_code & FAULT_CODE_ITLB) &&
 311             (fault_code & FAULT_CODE_DTLB))
 312                 BUG();
 313
 314         if (regs->tstate & TSTATE_PRIV) {
 315                 unsigned long tpc = regs->tpc;
 316
 317                 /* Sanity check the PC. */
 318                 if ((tpc >= KERNBASE && tpc < (unsigned long) _etext) ||
 319                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 320                         /* Valid, no problems... */
 321                 } else {
 322                         bad_kernel_pc(regs, address);
 323                         return;
 324                 }
 325         }
 326
 327         /*
 328          * If we're in an interrupt or have no user
 329          * context, we must not take the fault..
 330          */
 331         if (in_atomic() || !mm)
 332                 goto intr_or_no_mm;
 333
 334         if (test_thread_flag(TIF_32BIT)) {
 335                 if (!(regs->tstate & TSTATE_PRIV))
 336                         regs->tpc &= 0xffffffff;
 337                 address &= 0xffffffff;
 338         }
 339
 340         if (!down_read_trylock(&mm->mmap_sem)) {
 341                 if ((regs->tstate & TSTATE_PRIV) &&
 342                     !search_exception_tables(regs->tpc)) {
 343                         insn = get_fault_insn(regs, insn);
 344                         goto handle_kernel_fault;
 345                 }
 346                 down_read(&mm->mmap_sem);
 347         }
 348
 349         vma = find_vma(mm, address);
 350         if (!vma)
 351                 goto bad_area;
 352
 353         /* Pure DTLB misses do not tell us whether the fault causing
 354          * load/store/atomic was a write or not, it only says that there
 355          * was no match.  So in such a case we (carefully) read the
 356          * instruction to try and figure this out.  It's an optimization
 357          * so it's ok if we can't do this.
 358          *
 359          * Special hack, window spill/fill knows the exact fault type.
 360          */
 361         if (((fault_code &
 362               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 363             (vma->vm_flags & VM_WRITE) != 0) {
 364                 insn = get_fault_insn(regs, 0);
 365                 if (!insn)
 366                         goto continue_fault;
 367                 /* All loads, stores and atomics have bits 30 and 31 both set
 368                  * in the instruction.  Bit 21 is set in all stores, but we
 369                  * have to avoid prefetches which also have bit 21 set.
 370                  */
 371                 if ((insn & 0xc0200000) == 0xc0200000 &&
 372                     (insn & 0x01780000) != 0x01680000) {
 373                         /* Don't bother updating thread struct value,
 374                          * because update_mmu_cache only cares which tlb
 375                          * the access came from.
 376                          */
 377                         fault_code |= FAULT_CODE_WRITE;
 378                 }
 379         }
 380 continue_fault:
 381
 382         if (vma->vm_start <= address)
 383                 goto good_area;
 384         if (!(vma->vm_flags & VM_GROWSDOWN))
 385                 goto bad_area;
 386         if (!(fault_code & FAULT_CODE_WRITE)) {
 387                 /* Non-faulting loads shouldn't expand stack. */
 388                 insn = get_fault_insn(regs, insn);
 389                 if ((insn & 0xc0800000) == 0xc0800000) {
 390                         unsigned char asi;
 391
 392                         if (insn & 0x2000)
 393                                 asi = (regs->tstate >> 24);
 394                         else
 395                                 asi = (insn >> 5);
 396                         if ((asi & 0xf2) == 0x82)
 397                                 goto bad_area;
 398                 }
 399         }
 400         if (expand_stack(vma, address))
 401                 goto bad_area;
 402         /*
 403          * Ok, we have a good vm_area for this memory access, so
 404          * we can handle it..
 405          */
 406 good_area:
 407         si_code = SEGV_ACCERR;
 408
 409         /* If we took a ITLB miss on a non-executable page, catch
 410          * that here.
 411          */
 412         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 413                 BUG_ON(address != regs->tpc);
 414                 BUG_ON(regs->tstate & TSTATE_PRIV);
 415                 goto bad_area;
 416         }
 417
 418         if (fault_code & FAULT_CODE_WRITE) {
 419                 if (!(vma->vm_flags & VM_WRITE))
 420                         goto bad_area;
 421
 422                 /* Spitfire has an icache which does not snoop
 423                  * processor stores.  Later processors do...
 424                  */
 425                 if (tlb_type == spitfire &&
 426                     (vma->vm_flags & VM_EXEC) != 0 &&
 427                     vma->vm_file != NULL)
 428                         set_thread_fault_code(fault_code |
 429                                               FAULT_CODE_BLKCOMMIT);
 430         } else {
 431                 /* Allow reads even for write-only mappings */
 432                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 433                         goto bad_area;
 434         }
 435
 436         switch (handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE))) {
 437         case VM_FAULT_MINOR:
 438                 current->min_flt++;
 439                 break;
 440         case VM_FAULT_MAJOR:
 441                 current->maj_flt++;
 442                 break;
 443         case VM_FAULT_SIGBUS:
 444                 goto do_sigbus;
 445         case VM_FAULT_OOM:
 446                 goto out_of_memory;
 447         default:
 448                 BUG();
 449         }
 450
 451         up_read(&mm->mmap_sem);
 452
 453         mm_rss = get_mm_rss(mm);
 454 #ifdef CONFIG_HUGETLB_PAGE
 455         mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
 456 #endif
 457         if (unlikely(mm_rss >
 458                      mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
 459                 tsb_grow(mm, MM_TSB_BASE, mm_rss);
 460 #ifdef CONFIG_HUGETLB_PAGE
 461         mm_rss = mm->context.huge_pte_count;
 462         if (unlikely(mm_rss >
 463                      mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
 464                 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
 465 #endif
 466         return;
 467
 468         /*
 469          * Something tried to access memory that isn't in our memory map..
 470          * Fix it, but check if it's kernel or user first..
 471          */
 472 bad_area:
 473         insn = get_fault_insn(regs, insn);
 474         up_read(&mm->mmap_sem);
 475
 476 handle_kernel_fault:
 477         do_kernel_fault(regs, si_code, fault_code, insn, address);
 478         return;
 479
 480 /*
 481  * We ran out of memory, or some other thing happened to us that made
 482  * us unable to handle the page fault gracefully.
 483  */
 484 out_of_memory:
 485         insn = get_fault_insn(regs, insn);
 486         up_read(&mm->mmap_sem);
 487         printk("VM: killing process %s\n", current->comm);
 488         if (!(regs->tstate & TSTATE_PRIV))
 489                 do_exit(SIGKILL);
 490         goto handle_kernel_fault;
 491
 492 intr_or_no_mm:
 493         insn = get_fault_insn(regs, 0);
 494         goto handle_kernel_fault;
 495
 496 do_sigbus:
 497         insn = get_fault_insn(regs, insn);
 498         up_read(&mm->mmap_sem);
 499
 500         /*
 501          * Send a sigbus, regardless of whether we were in kernel
 502          * or user mode.
 503          */
 504         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 505
 506         /* Kernel mode? Handle exceptions or die */
 507         if (regs->tstate & TSTATE_PRIV)
 508                 goto handle_kernel_fault;
 509 }