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
  23 #include <asm/page.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/openprom.h>
  26 #include <asm/oplib.h>
  27 #include <asm/uaccess.h>
  28 #include <asm/asi.h>
  29 #include <asm/lsu.h>
  30 #include <asm/sections.h>
  31 #include <asm/kdebug.h>
  32 #include <asm/mmu_context.h>
  33
  34 /*
  35  * To debug kernel to catch accesses to certain virtual/physical addresses.
  36  * Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints.
  37  * flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses.
  38  * Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be
  39  * watched. This is only useful on a single cpu machine for now. After the watchpoint
  40  * is detected, the process causing it will be killed, thus preventing an infinite loop.
  41  */
  42 void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode)
  43 {
  44         unsigned long lsubits;
  45
  46         __asm__ __volatile__("ldxa [%%g0] %1, %0"
  47                              : "=r" (lsubits)
  48                              : "i" (ASI_LSU_CONTROL));
  49         lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM |
  50                      LSU_CONTROL_PR | LSU_CONTROL_VR |
  51                      LSU_CONTROL_PW | LSU_CONTROL_VW);
  52
  53         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
  54                              "membar    #Sync"
  55                              : /* no outputs */
  56                              : "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT),
  57                                "i" (ASI_DMMU));
  58
  59         lsubits |= ((unsigned long)mask << (mode ? 25 : 33));
  60         if (flags & VM_READ)
  61                 lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR);
  62         if (flags & VM_WRITE)
  63                 lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW);
  64         __asm__ __volatile__("stxa %0, [%%g0] %1\n\t"
  65                              "membar #Sync"
  66                              : /* no outputs */
  67                              : "r" (lsubits), "i" (ASI_LSU_CONTROL)
  68                              : "memory");
  69 }
  70
  71 static void __kprobes unhandled_fault(unsigned long address,
  72                                       struct task_struct *tsk,
  73                                       struct pt_regs *regs)
  74 {
  75         if ((unsigned long) address < PAGE_SIZE) {
  76                 printk(KERN_ALERT "Unable to handle kernel NULL "
  77                        "pointer dereference\n");
  78         } else {
  79                 printk(KERN_ALERT "Unable to handle kernel paging request "
  80                        "at virtual address %016lx\n", (unsigned long)address);
  81         }
  82         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
  83                (tsk->mm ?
  84                 CTX_HWBITS(tsk->mm->context) :
  85                 CTX_HWBITS(tsk->active_mm->context)));
  86         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
  87                (tsk->mm ? (unsigned long) tsk->mm->pgd :
  88                           (unsigned long) tsk->active_mm->pgd));
  89         if (notify_die(DIE_GPF, "general protection fault", regs,
  90                        0, 0, SIGSEGV) == NOTIFY_STOP)
  91                 return;
  92         die_if_kernel("Oops", regs);
  93 }
  94
  95 static void bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
  96 {
  97         unsigned long *ksp;
  98
  99         printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
 100                regs->tpc);
 101         printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
 102         __asm__("mov %%sp, %0" : "=r" (ksp));
 103         show_stack(current, ksp);
 104         unhandled_fault(regs->tpc, current, regs);
 105 }
 106
 107 /*
 108  * We now make sure that mmap_sem is held in all paths that call
 109  * this. Additionally, to prevent kswapd from ripping ptes from
 110  * under us, raise interrupts around the time that we look at the
 111  * pte, kswapd will have to wait to get his smp ipi response from
 112  * us. vmtruncate likewise. This saves us having to get pte lock.
 113  */
 114 static unsigned int get_user_insn(unsigned long tpc)
 115 {
 116         pgd_t *pgdp = pgd_offset(current->mm, tpc);
 117         pud_t *pudp;
 118         pmd_t *pmdp;
 119         pte_t *ptep, pte;
 120         unsigned long pa;
 121         u32 insn = 0;
 122         unsigned long pstate;
 123
 124         if (pgd_none(*pgdp))
 125                 goto outret;
 126         pudp = pud_offset(pgdp, tpc);
 127         if (pud_none(*pudp))
 128                 goto outret;
 129         pmdp = pmd_offset(pudp, tpc);
 130         if (pmd_none(*pmdp))
 131                 goto outret;
 132
 133         /* This disables preemption for us as well. */
 134         __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 135         __asm__ __volatile__("wrpr %0, %1, %%pstate"
 136                                 : : "r" (pstate), "i" (PSTATE_IE));
 137         ptep = pte_offset_map(pmdp, tpc);
 138         pte = *ptep;
 139         if (!pte_present(pte))
 140                 goto out;
 141
 142         pa  = (pte_pfn(pte) << PAGE_SHIFT);
 143         pa += (tpc & ~PAGE_MASK);
 144
 145         /* Use phys bypass so we don't pollute dtlb/dcache. */
 146         __asm__ __volatile__("lduwa [%1] %2, %0"
 147                              : "=r" (insn)
 148                              : "r" (pa), "i" (ASI_PHYS_USE_EC));
 149
 150 out:
 151         pte_unmap(ptep);
 152         __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
 153 outret:
 154         return insn;
 155 }
 156
 157 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
 158
 159 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
 160                              unsigned int insn, int fault_code)
 161 {
 162         siginfo_t info;
 163
 164         info.si_code = code;
 165         info.si_signo = sig;
 166         info.si_errno = 0;
 167         if (fault_code & FAULT_CODE_ITLB)
 168                 info.si_addr = (void __user *) regs->tpc;
 169         else
 170                 info.si_addr = (void __user *)
 171                         compute_effective_address(regs, insn, 0);
 172         info.si_trapno = 0;
 173         force_sig_info(sig, &info, current);
 174 }
 175
 176 extern int handle_ldf_stq(u32, struct pt_regs *);
 177 extern int handle_ld_nf(u32, struct pt_regs *);
 178
 179 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
 180 {
 181         if (!insn) {
 182                 if (!regs->tpc || (regs->tpc & 0x3))
 183                         return 0;
 184                 if (regs->tstate & TSTATE_PRIV) {
 185                         insn = *(unsigned int *) regs->tpc;
 186                 } else {
 187                         insn = get_user_insn(regs->tpc);
 188                 }
 189         }
 190         return insn;
 191 }
 192
 193 static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
 194                             unsigned int insn, unsigned long address)
 195 {
 196         unsigned char asi = ASI_P;
 197
 198         if ((!insn) && (regs->tstate & TSTATE_PRIV))
 199                 goto cannot_handle;
 200
 201         /* If user insn could be read (thus insn is zero), that
 202          * is fine.  We will just gun down the process with a signal
 203          * in that case.
 204          */
 205
 206         if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
 207             (insn & 0xc0800000) == 0xc0800000) {
 208                 if (insn & 0x2000)
 209                         asi = (regs->tstate >> 24);
 210                 else
 211                         asi = (insn >> 5);
 212                 if ((asi & 0xf2) == 0x82) {
 213                         if (insn & 0x1000000) {
 214                                 handle_ldf_stq(insn, regs);
 215                         } else {
 216                                 /* This was a non-faulting load. Just clear the
 217                                  * destination register(s) and continue with the next
 218                                  * instruction. -jj
 219                                  */
 220                                 handle_ld_nf(insn, regs);
 221                         }
 222                         return;
 223                 }
 224         }
 225
 226         /* Is this in ex_table? */
 227         if (regs->tstate & TSTATE_PRIV) {
 228                 const struct exception_table_entry *entry;
 229
 230                 if (asi == ASI_P && (insn & 0xc0800000) == 0xc0800000) {
 231                         if (insn & 0x2000)
 232                                 asi = (regs->tstate >> 24);
 233                         else
 234                                 asi = (insn >> 5);
 235                 }
 236
 237                 /* Look in asi.h: All _S asis have LS bit set */
 238                 if ((asi & 0x1) &&
 239                     (entry = search_exception_tables(regs->tpc))) {
 240                         regs->tpc = entry->fixup;
 241                         regs->tnpc = regs->tpc + 4;
 242                         return;
 243                 }
 244         } else {
 245                 /* The si_code was set to make clear whether
 246                  * this was a SEGV_MAPERR or SEGV_ACCERR fault.
 247                  */
 248                 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
 249                 return;
 250         }
 251
 252 cannot_handle:
 253         unhandled_fault (address, current, regs);
 254 }
 255
 256 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 257 {
 258         struct mm_struct *mm = current->mm;
 259         struct vm_area_struct *vma;
 260         unsigned int insn = 0;
 261         int si_code, fault_code;
 262         unsigned long address, mm_rss;
 263
 264         fault_code = get_thread_fault_code();
 265
 266         if (notify_die(DIE_PAGE_FAULT, "page_fault", regs,
 267                        fault_code, 0, SIGSEGV) == NOTIFY_STOP)
 268                 return;
 269
 270         si_code = SEGV_MAPERR;
 271         address = current_thread_info()->fault_address;
 272
 273         if ((fault_code & FAULT_CODE_ITLB) &&
 274             (fault_code & FAULT_CODE_DTLB))
 275                 BUG();
 276
 277         if (regs->tstate & TSTATE_PRIV) {
 278                 unsigned long tpc = regs->tpc;
 279
 280                 /* Sanity check the PC. */
 281                 if ((tpc >= KERNBASE && tpc < (unsigned long) _etext) ||
 282                     (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
 283                         /* Valid, no problems... */
 284                 } else {
 285                         bad_kernel_pc(regs, address);
 286                         return;
 287                 }
 288         }
 289
 290         /*
 291          * If we're in an interrupt or have no user
 292          * context, we must not take the fault..
 293          */
 294         if (in_atomic() || !mm)
 295                 goto intr_or_no_mm;
 296
 297         if (test_thread_flag(TIF_32BIT)) {
 298                 if (!(regs->tstate & TSTATE_PRIV))
 299                         regs->tpc &= 0xffffffff;
 300                 address &= 0xffffffff;
 301         }
 302
 303         if (!down_read_trylock(&mm->mmap_sem)) {
 304                 if ((regs->tstate & TSTATE_PRIV) &&
 305                     !search_exception_tables(regs->tpc)) {
 306                         insn = get_fault_insn(regs, insn);
 307                         goto handle_kernel_fault;
 308                 }
 309                 down_read(&mm->mmap_sem);
 310         }
 311
 312         vma = find_vma(mm, address);
 313         if (!vma)
 314                 goto bad_area;
 315
 316         /* Pure DTLB misses do not tell us whether the fault causing
 317          * load/store/atomic was a write or not, it only says that there
 318          * was no match.  So in such a case we (carefully) read the
 319          * instruction to try and figure this out.  It's an optimization
 320          * so it's ok if we can't do this.
 321          *
 322          * Special hack, window spill/fill knows the exact fault type.
 323          */
 324         if (((fault_code &
 325               (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
 326             (vma->vm_flags & VM_WRITE) != 0) {
 327                 insn = get_fault_insn(regs, 0);
 328                 if (!insn)
 329                         goto continue_fault;
 330                 if ((insn & 0xc0200000) == 0xc0200000 &&
 331                     (insn & 0x1780000) != 0x1680000) {
 332                         /* Don't bother updating thread struct value,
 333                          * because update_mmu_cache only cares which tlb
 334                          * the access came from.
 335                          */
 336                         fault_code |= FAULT_CODE_WRITE;
 337                 }
 338         }
 339 continue_fault:
 340
 341         if (vma->vm_start <= address)
 342                 goto good_area;
 343         if (!(vma->vm_flags & VM_GROWSDOWN))
 344                 goto bad_area;
 345         if (!(fault_code & FAULT_CODE_WRITE)) {
 346                 /* Non-faulting loads shouldn't expand stack. */
 347                 insn = get_fault_insn(regs, insn);
 348                 if ((insn & 0xc0800000) == 0xc0800000) {
 349                         unsigned char asi;
 350
 351                         if (insn & 0x2000)
 352                                 asi = (regs->tstate >> 24);
 353                         else
 354                                 asi = (insn >> 5);
 355                         if ((asi & 0xf2) == 0x82)
 356                                 goto bad_area;
 357                 }
 358         }
 359         if (expand_stack(vma, address))
 360                 goto bad_area;
 361         /*
 362          * Ok, we have a good vm_area for this memory access, so
 363          * we can handle it..
 364          */
 365 good_area:
 366         si_code = SEGV_ACCERR;
 367
 368         /* If we took a ITLB miss on a non-executable page, catch
 369          * that here.
 370          */
 371         if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
 372                 BUG_ON(address != regs->tpc);
 373                 BUG_ON(regs->tstate & TSTATE_PRIV);
 374                 goto bad_area;
 375         }
 376
 377         if (fault_code & FAULT_CODE_WRITE) {
 378                 if (!(vma->vm_flags & VM_WRITE))
 379                         goto bad_area;
 380
 381                 /* Spitfire has an icache which does not snoop
 382                  * processor stores.  Later processors do...
 383                  */
 384                 if (tlb_type == spitfire &&
 385                     (vma->vm_flags & VM_EXEC) != 0 &&
 386                     vma->vm_file != NULL)
 387                         set_thread_fault_code(fault_code |
 388                                               FAULT_CODE_BLKCOMMIT);
 389         } else {
 390                 /* Allow reads even for write-only mappings */
 391                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 392                         goto bad_area;
 393         }
 394
 395         switch (handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE))) {
 396         case VM_FAULT_MINOR:
 397                 current->min_flt++;
 398                 break;
 399         case VM_FAULT_MAJOR:
 400                 current->maj_flt++;
 401                 break;
 402         case VM_FAULT_SIGBUS:
 403                 goto do_sigbus;
 404         case VM_FAULT_OOM:
 405                 goto out_of_memory;
 406         default:
 407                 BUG();
 408         }
 409
 410         up_read(&mm->mmap_sem);
 411
 412         mm_rss = get_mm_rss(mm);
 413         if (unlikely(mm_rss >= mm->context.tsb_rss_limit))
 414                 tsb_grow(mm, mm_rss);
 415
 416         return;
 417
 418         /*
 419          * Something tried to access memory that isn't in our memory map..
 420          * Fix it, but check if it's kernel or user first..
 421          */
 422 bad_area:
 423         insn = get_fault_insn(regs, insn);
 424         up_read(&mm->mmap_sem);
 425
 426 handle_kernel_fault:
 427         do_kernel_fault(regs, si_code, fault_code, insn, address);
 428         return;
 429
 430 /*
 431  * We ran out of memory, or some other thing happened to us that made
 432  * us unable to handle the page fault gracefully.
 433  */
 434 out_of_memory:
 435         insn = get_fault_insn(regs, insn);
 436         up_read(&mm->mmap_sem);
 437         printk("VM: killing process %s\n", current->comm);
 438         if (!(regs->tstate & TSTATE_PRIV))
 439                 do_exit(SIGKILL);
 440         goto handle_kernel_fault;
 441
 442 intr_or_no_mm:
 443         insn = get_fault_insn(regs, 0);
 444         goto handle_kernel_fault;
 445
 446 do_sigbus:
 447         insn = get_fault_insn(regs, insn);
 448         up_read(&mm->mmap_sem);
 449
 450         /*
 451          * Send a sigbus, regardless of whether we were in kernel
 452          * or user mode.
 453          */
 454         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
 455
 456         /* Kernel mode? Handle exceptions or die */
 457         if (regs->tstate & TSTATE_PRIV)
 458                 goto handle_kernel_fault;
 459 }