arch/sparc/mm/fault_32.c

   1 /*
   2  * fault.c:  Page fault handlers for the Sparc.
   3  *
   4  * Copyright (C) 1995 David S. Miller (davem@caip.rutgers.edu)
   5  * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be)
   6  * Copyright (C) 1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
   7  */
   8
   9 #include <asm/head.h>
  10
  11 #include <linux/string.h>
  12 #include <linux/types.h>
  13 #include <linux/sched.h>
  14 #include <linux/ptrace.h>
  15 #include <linux/mman.h>
  16 #include <linux/threads.h>
  17 #include <linux/kernel.h>
  18 #include <linux/signal.h>
  19 #include <linux/mm.h>
  20 #include <linux/smp.h>
  21 #include <linux/perf_event.h>
  22 #include <linux/interrupt.h>
  23 #include <linux/module.h>
  24 #include <linux/kdebug.h>
  25
  26 #include <asm/system.h>
  27 #include <asm/page.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/memreg.h>
  30 #include <asm/openprom.h>
  31 #include <asm/oplib.h>
  32 #include <asm/smp.h>
  33 #include <asm/traps.h>
  34 #include <asm/uaccess.h>
  35
  36 extern int prom_node_root;
  37
  38 /* At boot time we determine these two values necessary for setting
  39  * up the segment maps and page table entries (pte's).
  40  */
  41
  42 int num_segmaps, num_contexts;
  43 int invalid_segment;
  44
  45 /* various Virtual Address Cache parameters we find at boot time... */
  46
  47 int vac_size, vac_linesize, vac_do_hw_vac_flushes;
  48 int vac_entries_per_context, vac_entries_per_segment;
  49 int vac_entries_per_page;
  50
  51 /* Return how much physical memory we have.  */
  52 unsigned long probe_memory(void)
  53 {
  54         unsigned long total = 0;
  55         int i;
  56
  57         for (i = 0; sp_banks[i].num_bytes; i++)
  58                 total += sp_banks[i].num_bytes;
  59
  60         return total;
  61 }
  62
  63 extern void sun4c_complete_all_stores(void);
  64
  65 /* Whee, a level 15 NMI interrupt memory error.  Let's have fun... */
  66 asmlinkage void sparc_lvl15_nmi(struct pt_regs *regs, unsigned long serr,
  67                                 unsigned long svaddr, unsigned long aerr,
  68                                 unsigned long avaddr)
  69 {
  70         sun4c_complete_all_stores();
  71         printk("FAULT: NMI received\n");
  72         printk("SREGS: Synchronous Error %08lx\n", serr);
  73         printk("       Synchronous Vaddr %08lx\n", svaddr);
  74         printk("      Asynchronous Error %08lx\n", aerr);
  75         printk("      Asynchronous Vaddr %08lx\n", avaddr);
  76         if (sun4c_memerr_reg)
  77                 printk("     Memory Parity Error %08lx\n", *sun4c_memerr_reg);
  78         printk("REGISTER DUMP:\n");
  79         show_regs(regs);
  80         prom_halt();
  81 }
  82
  83 static void unhandled_fault(unsigned long, struct task_struct *,
  84                 struct pt_regs *) __attribute__ ((noreturn));
  85
  86 static void unhandled_fault(unsigned long address, struct task_struct *tsk,
  87                      struct pt_regs *regs)
  88 {
  89         if((unsigned long) address < PAGE_SIZE) {
  90                 printk(KERN_ALERT
  91                     "Unable to handle kernel NULL pointer dereference\n");
  92         } else {
  93                 printk(KERN_ALERT "Unable to handle kernel paging request "
  94                        "at virtual address %08lx\n", address);
  95         }
  96         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %08lx\n",
  97                 (tsk->mm ? tsk->mm->context : tsk->active_mm->context));
  98         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %08lx\n",
  99                 (tsk->mm ? (unsigned long) tsk->mm->pgd :
 100                         (unsigned long) tsk->active_mm->pgd));
 101         die_if_kernel("Oops", regs);
 102 }
 103
 104 asmlinkage int lookup_fault(unsigned long pc, unsigned long ret_pc,
 105                             unsigned long address)
 106 {
 107         struct pt_regs regs;
 108         unsigned long g2;
 109         unsigned int insn;
 110         int i;
 111
 112         i = search_extables_range(ret_pc, &g2);
 113         switch (i) {
 114         case 3:
 115                 /* load & store will be handled by fixup */
 116                 return 3;
 117
 118         case 1:
 119                 /* store will be handled by fixup, load will bump out */
 120                 /* for _to_ macros */
 121                 insn = *((unsigned int *) pc);
 122                 if ((insn >> 21) & 1)
 123                         return 1;
 124                 break;
 125
 126         case 2:
 127                 /* load will be handled by fixup, store will bump out */
 128                 /* for _from_ macros */
 129                 insn = *((unsigned int *) pc);
 130                 if (!((insn >> 21) & 1) || ((insn>>19)&0x3f) == 15)
 131                         return 2;
 132                 break;
 133
 134         default:
 135                 break;
 136         };
 137
 138         memset(&regs, 0, sizeof (regs));
 139         regs.pc = pc;
 140         regs.npc = pc + 4;
 141         __asm__ __volatile__(
 142                 "rd %%psr, %0\n\t"
 143                 "nop\n\t"
 144                 "nop\n\t"
 145                 "nop\n" : "=r" (regs.psr));
 146         unhandled_fault(address, current, &regs);
 147
 148         /* Not reached */
 149         return 0;
 150 }
 151
 152 extern unsigned long safe_compute_effective_address(struct pt_regs *,
 153                                                     unsigned int);
 154
 155 static unsigned long compute_si_addr(struct pt_regs *regs, int text_fault)
 156 {
 157         unsigned int insn;
 158
 159         if (text_fault)
 160                 return regs->pc;
 161
 162         if (regs->psr & PSR_PS) {
 163                 insn = *(unsigned int *) regs->pc;
 164         } else {
 165                 __get_user(insn, (unsigned int *) regs->pc);
 166         }
 167
 168         return safe_compute_effective_address(regs, insn);
 169 }
 170
 171 asmlinkage void do_sparc_fault(struct pt_regs *regs, int text_fault, int write,
 172                                unsigned long address)
 173 {
 174         struct vm_area_struct *vma;
 175         struct task_struct *tsk = current;
 176         struct mm_struct *mm = tsk->mm;
 177         unsigned int fixup;
 178         unsigned long g2;
 179         siginfo_t info;
 180         int from_user = !(regs->psr & PSR_PS);
 181         int fault;
 182
 183         if(text_fault)
 184                 address = regs->pc;
 185
 186         /*
 187          * We fault-in kernel-space virtual memory on-demand. The
 188          * 'reference' page table is init_mm.pgd.
 189          *
 190          * NOTE! We MUST NOT take any locks for this case. We may
 191          * be in an interrupt or a critical region, and should
 192          * only copy the information from the master page table,
 193          * nothing more.
 194          */
 195         if (!ARCH_SUN4C && address >= TASK_SIZE)
 196                 goto vmalloc_fault;
 197
 198         info.si_code = SEGV_MAPERR;
 199
 200         /*
 201          * If we're in an interrupt or have no user
 202          * context, we must not take the fault..
 203          */
 204         if (in_atomic() || !mm)
 205                 goto no_context;
 206
 207         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
 208
 209         down_read(&mm->mmap_sem);
 210
 211         /*
 212          * The kernel referencing a bad kernel pointer can lock up
 213          * a sun4c machine completely, so we must attempt recovery.
 214          */
 215         if(!from_user && address >= PAGE_OFFSET)
 216                 goto bad_area;
 217
 218         vma = find_vma(mm, address);
 219         if(!vma)
 220                 goto bad_area;
 221         if(vma->vm_start <= address)
 222                 goto good_area;
 223         if(!(vma->vm_flags & VM_GROWSDOWN))
 224                 goto bad_area;
 225         if(expand_stack(vma, address))
 226                 goto bad_area;
 227         /*
 228          * Ok, we have a good vm_area for this memory access, so
 229          * we can handle it..
 230          */
 231 good_area:
 232         info.si_code = SEGV_ACCERR;
 233         if(write) {
 234                 if(!(vma->vm_flags & VM_WRITE))
 235                         goto bad_area;
 236         } else {
 237                 /* Allow reads even for write-only mappings */
 238                 if(!(vma->vm_flags & (VM_READ | VM_EXEC)))
 239                         goto bad_area;
 240         }
 241
 242         /*
 243          * If for any reason at all we couldn't handle the fault,
 244          * make sure we exit gracefully rather than endlessly redo
 245          * the fault.
 246          */
 247         fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0);
 248         if (unlikely(fault & VM_FAULT_ERROR)) {
 249                 if (fault & VM_FAULT_OOM)
 250                         goto out_of_memory;
 251                 else if (fault & VM_FAULT_SIGBUS)
 252                         goto do_sigbus;
 253                 BUG();
 254         }
 255         if (fault & VM_FAULT_MAJOR) {
 256                 current->maj_flt++;
 257                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
 258                               regs, address);
 259         } else {
 260                 current->min_flt++;
 261                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
 262                               regs, address);
 263         }
 264         up_read(&mm->mmap_sem);
 265         return;
 266
 267         /*
 268          * Something tried to access memory that isn't in our memory map..
 269          * Fix it, but check if it's kernel or user first..
 270          */
 271 bad_area:
 272         up_read(&mm->mmap_sem);
 273
 274 bad_area_nosemaphore:
 275         /* User mode accesses just cause a SIGSEGV */
 276         if(from_user) {
 277 #if 0
 278                 printk("Fault whee %s [%d]: segfaults at %08lx pc=%08lx\n",
 279                        tsk->comm, tsk->pid, address, regs->pc);
 280 #endif
 281                 info.si_signo = SIGSEGV;
 282                 info.si_errno = 0;
 283                 /* info.si_code set above to make clear whether
 284                    this was a SEGV_MAPERR or SEGV_ACCERR fault.  */
 285                 info.si_addr = (void __user *)compute_si_addr(regs, text_fault);
 286                 info.si_trapno = 0;
 287                 force_sig_info (SIGSEGV, &info, tsk);
 288                 return;
 289         }
 290
 291         /* Is this in ex_table? */
 292 no_context:
 293         g2 = regs->u_regs[UREG_G2];
 294         if (!from_user) {
 295                 fixup = search_extables_range(regs->pc, &g2);
 296                 if (fixup > 10) { /* Values below are reserved for other things */
 297                         extern const unsigned __memset_start[];
 298                         extern const unsigned __memset_end[];
 299                         extern const unsigned __csum_partial_copy_start[];
 300                         extern const unsigned __csum_partial_copy_end[];
 301
 302 #ifdef DEBUG_EXCEPTIONS
 303                         printk("Exception: PC<%08lx> faddr<%08lx>\n", regs->pc, address);
 304                         printk("EX_TABLE: insn<%08lx> fixup<%08x> g2<%08lx>\n",
 305                                 regs->pc, fixup, g2);
 306 #endif
 307                         if ((regs->pc >= (unsigned long)__memset_start &&
 308                              regs->pc < (unsigned long)__memset_end) ||
 309                             (regs->pc >= (unsigned long)__csum_partial_copy_start &&
 310                              regs->pc < (unsigned long)__csum_partial_copy_end)) {
 311                                 regs->u_regs[UREG_I4] = address;
 312                                 regs->u_regs[UREG_I5] = regs->pc;
 313                         }
 314                         regs->u_regs[UREG_G2] = g2;
 315                         regs->pc = fixup;
 316                         regs->npc = regs->pc + 4;
 317                         return;
 318                 }
 319         }
 320
 321         unhandled_fault (address, tsk, regs);
 322         do_exit(SIGKILL);
 323
 324 /*
 325  * We ran out of memory, or some other thing happened to us that made
 326  * us unable to handle the page fault gracefully.
 327  */
 328 out_of_memory:
 329         up_read(&mm->mmap_sem);
 330         if (from_user) {
 331                 pagefault_out_of_memory();
 332                 return;
 333         }
 334         goto no_context;
 335
 336 do_sigbus:
 337         up_read(&mm->mmap_sem);
 338         info.si_signo = SIGBUS;
 339         info.si_errno = 0;
 340         info.si_code = BUS_ADRERR;
 341         info.si_addr = (void __user *) compute_si_addr(regs, text_fault);
 342         info.si_trapno = 0;
 343         force_sig_info (SIGBUS, &info, tsk);
 344         if (!from_user)
 345                 goto no_context;
 346
 347 vmalloc_fault:
 348         {
 349                 /*
 350                  * Synchronize this task's top level page-table
 351                  * with the 'reference' page table.
 352                  */
 353                 int offset = pgd_index(address);
 354                 pgd_t *pgd, *pgd_k;
 355                 pmd_t *pmd, *pmd_k;
 356
 357                 pgd = tsk->active_mm->pgd + offset;
 358                 pgd_k = init_mm.pgd + offset;
 359
 360                 if (!pgd_present(*pgd)) {
 361                         if (!pgd_present(*pgd_k))
 362                                 goto bad_area_nosemaphore;
 363                         pgd_val(*pgd) = pgd_val(*pgd_k);
 364                         return;
 365                 }
 366
 367                 pmd = pmd_offset(pgd, address);
 368                 pmd_k = pmd_offset(pgd_k, address);
 369
 370                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
 371                         goto bad_area_nosemaphore;
 372                 *pmd = *pmd_k;
 373                 return;
 374         }
 375 }
 376
 377 asmlinkage void do_sun4c_fault(struct pt_regs *regs, int text_fault, int write,
 378                                unsigned long address)
 379 {
 380         extern void sun4c_update_mmu_cache(struct vm_area_struct *,
 381                                            unsigned long,pte_t);
 382         extern pte_t *sun4c_pte_offset_kernel(pmd_t *,unsigned long);
 383         struct task_struct *tsk = current;
 384         struct mm_struct *mm = tsk->mm;
 385         pgd_t *pgdp;
 386         pte_t *ptep;
 387
 388         if (text_fault) {
 389                 address = regs->pc;
 390         } else if (!write &&
 391                    !(regs->psr & PSR_PS)) {
 392                 unsigned int insn, __user *ip;
 393
 394                 ip = (unsigned int __user *)regs->pc;
 395                 if (!get_user(insn, ip)) {
 396                         if ((insn & 0xc1680000) == 0xc0680000)
 397                                 write = 1;
 398                 }
 399         }
 400
 401         if (!mm) {
 402                 /* We are oopsing. */
 403                 do_sparc_fault(regs, text_fault, write, address);
 404                 BUG();  /* P3 Oops already, you bitch */
 405         }
 406
 407         pgdp = pgd_offset(mm, address);
 408         ptep = sun4c_pte_offset_kernel((pmd_t *) pgdp, address);
 409
 410         if (pgd_val(*pgdp)) {
 411             if (write) {
 412                 if ((pte_val(*ptep) & (_SUN4C_PAGE_WRITE|_SUN4C_PAGE_PRESENT))
 413                                    == (_SUN4C_PAGE_WRITE|_SUN4C_PAGE_PRESENT)) {
 414                         unsigned long flags;
 415
 416                         *ptep = __pte(pte_val(*ptep) | _SUN4C_PAGE_ACCESSED |
 417                                       _SUN4C_PAGE_MODIFIED |
 418                                       _SUN4C_PAGE_VALID |
 419                                       _SUN4C_PAGE_DIRTY);
 420
 421                         local_irq_save(flags);
 422                         if (sun4c_get_segmap(address) != invalid_segment) {
 423                                 sun4c_put_pte(address, pte_val(*ptep));
 424                                 local_irq_restore(flags);
 425                                 return;
 426                         }
 427                         local_irq_restore(flags);
 428                 }
 429             } else {
 430                 if ((pte_val(*ptep) & (_SUN4C_PAGE_READ|_SUN4C_PAGE_PRESENT))
 431                                    == (_SUN4C_PAGE_READ|_SUN4C_PAGE_PRESENT)) {
 432                         unsigned long flags;
 433
 434                         *ptep = __pte(pte_val(*ptep) | _SUN4C_PAGE_ACCESSED |
 435                                       _SUN4C_PAGE_VALID);
 436
 437                         local_irq_save(flags);
 438                         if (sun4c_get_segmap(address) != invalid_segment) {
 439                                 sun4c_put_pte(address, pte_val(*ptep));
 440                                 local_irq_restore(flags);
 441                                 return;
 442                         }
 443                         local_irq_restore(flags);
 444                 }
 445             }
 446         }
 447
 448         /* This conditional is 'interesting'. */
 449         if (pgd_val(*pgdp) && !(write && !(pte_val(*ptep) & _SUN4C_PAGE_WRITE))
 450             && (pte_val(*ptep) & _SUN4C_PAGE_VALID))
 451                 /* Note: It is safe to not grab the MMAP semaphore here because
 452                  *       we know that update_mmu_cache() will not sleep for
 453                  *       any reason (at least not in the current implementation)
 454                  *       and therefore there is no danger of another thread getting
 455                  *       on the CPU and doing a shrink_mmap() on this vma.
 456                  */
 457                 sun4c_update_mmu_cache (find_vma(current->mm, address), address,
 458                                         *ptep);
 459         else
 460                 do_sparc_fault(regs, text_fault, write, address);
 461 }
 462
 463 /* This always deals with user addresses. */
 464 static void force_user_fault(unsigned long address, int write)
 465 {
 466         struct vm_area_struct *vma;
 467         struct task_struct *tsk = current;
 468         struct mm_struct *mm = tsk->mm;
 469         siginfo_t info;
 470
 471         info.si_code = SEGV_MAPERR;
 472
 473 #if 0
 474         printk("wf<pid=%d,wr=%d,addr=%08lx>\n",
 475                tsk->pid, write, address);
 476 #endif
 477         down_read(&mm->mmap_sem);
 478         vma = find_vma(mm, address);
 479         if(!vma)
 480                 goto bad_area;
 481         if(vma->vm_start <= address)
 482                 goto good_area;
 483         if(!(vma->vm_flags & VM_GROWSDOWN))
 484                 goto bad_area;
 485         if(expand_stack(vma, address))
 486                 goto bad_area;
 487 good_area:
 488         info.si_code = SEGV_ACCERR;
 489         if(write) {
 490                 if(!(vma->vm_flags & VM_WRITE))
 491                         goto bad_area;
 492         } else {
 493                 if(!(vma->vm_flags & (VM_READ | VM_EXEC)))
 494                         goto bad_area;
 495         }
 496         switch (handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0)) {
 497         case VM_FAULT_SIGBUS:
 498         case VM_FAULT_OOM:
 499                 goto do_sigbus;
 500         }
 501         up_read(&mm->mmap_sem);
 502         return;
 503 bad_area:
 504         up_read(&mm->mmap_sem);
 505 #if 0
 506         printk("Window whee %s [%d]: segfaults at %08lx\n",
 507                tsk->comm, tsk->pid, address);
 508 #endif
 509         info.si_signo = SIGSEGV;
 510         info.si_errno = 0;
 511         /* info.si_code set above to make clear whether
 512            this was a SEGV_MAPERR or SEGV_ACCERR fault.  */
 513         info.si_addr = (void __user *) address;
 514         info.si_trapno = 0;
 515         force_sig_info (SIGSEGV, &info, tsk);
 516         return;
 517
 518 do_sigbus:
 519         up_read(&mm->mmap_sem);
 520         info.si_signo = SIGBUS;
 521         info.si_errno = 0;
 522         info.si_code = BUS_ADRERR;
 523         info.si_addr = (void __user *) address;
 524         info.si_trapno = 0;
 525         force_sig_info (SIGBUS, &info, tsk);
 526 }
 527
 528 void window_overflow_fault(void)
 529 {
 530         unsigned long sp;
 531
 532         sp = current_thread_info()->rwbuf_stkptrs[0];
 533         if(((sp + 0x38) & PAGE_MASK) != (sp & PAGE_MASK))
 534                 force_user_fault(sp + 0x38, 1);
 535         force_user_fault(sp, 1);
 536 }
 537
 538 void window_underflow_fault(unsigned long sp)
 539 {
 540         if(((sp + 0x38) & PAGE_MASK) != (sp & PAGE_MASK))
 541                 force_user_fault(sp + 0x38, 0);
 542         force_user_fault(sp, 0);
 543 }
 544
 545 void window_ret_fault(struct pt_regs *regs)
 546 {
 547         unsigned long sp;
 548
 549         sp = regs->u_regs[UREG_FP];
 550         if(((sp + 0x38) & PAGE_MASK) != (sp & PAGE_MASK))
 551                 force_user_fault(sp + 0x38, 0);
 552         force_user_fault(sp, 0);
 553 }