arch/x86_64/kernel/traps.c

   1 /*
   2  *  linux/arch/x86-64/traps.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   6  *
   7  *  Pentium III FXSR, SSE support
   8  *      Gareth Hughes <gareth@valinux.com>, May 2000
   9  */
  10
  11 /*
  12  * 'Traps.c' handles hardware traps and faults after we have saved some
  13  * state in 'entry.S'.
  14  */
  15 #include <linux/sched.h>
  16 #include <linux/kernel.h>
  17 #include <linux/string.h>
  18 #include <linux/errno.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/timer.h>
  21 #include <linux/mm.h>
  22 #include <linux/init.h>
  23 #include <linux/delay.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/kallsyms.h>
  27 #include <linux/module.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/nmi.h>
  30 #include <linux/kprobes.h>
  31 #include <linux/kexec.h>
  32 #include <linux/unwind.h>
  33 #include <linux/uaccess.h>
  34 #include <linux/bug.h>
  35
  36 #include <asm/system.h>
  37 #include <asm/io.h>
  38 #include <asm/atomic.h>
  39 #include <asm/debugreg.h>
  40 #include <asm/desc.h>
  41 #include <asm/i387.h>
  42 #include <asm/kdebug.h>
  43 #include <asm/processor.h>
  44 #include <asm/unwind.h>
  45 #include <asm/smp.h>
  46 #include <asm/pgalloc.h>
  47 #include <asm/pda.h>
  48 #include <asm/proto.h>
  49 #include <asm/nmi.h>
  50 #include <asm/stacktrace.h>
  51
  52 asmlinkage void divide_error(void);
  53 asmlinkage void debug(void);
  54 asmlinkage void nmi(void);
  55 asmlinkage void int3(void);
  56 asmlinkage void overflow(void);
  57 asmlinkage void bounds(void);
  58 asmlinkage void invalid_op(void);
  59 asmlinkage void device_not_available(void);
  60 asmlinkage void double_fault(void);
  61 asmlinkage void coprocessor_segment_overrun(void);
  62 asmlinkage void invalid_TSS(void);
  63 asmlinkage void segment_not_present(void);
  64 asmlinkage void stack_segment(void);
  65 asmlinkage void general_protection(void);
  66 asmlinkage void page_fault(void);
  67 asmlinkage void coprocessor_error(void);
  68 asmlinkage void simd_coprocessor_error(void);
  69 asmlinkage void reserved(void);
  70 asmlinkage void alignment_check(void);
  71 asmlinkage void machine_check(void);
  72 asmlinkage void spurious_interrupt_bug(void);
  73
  74 ATOMIC_NOTIFIER_HEAD(die_chain);
  75 EXPORT_SYMBOL(die_chain);
  76
  77 int register_die_notifier(struct notifier_block *nb)
  78 {
  79         vmalloc_sync_all();
  80         return atomic_notifier_chain_register(&die_chain, nb);
  81 }
  82 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
  83
  84 int unregister_die_notifier(struct notifier_block *nb)
  85 {
  86         return atomic_notifier_chain_unregister(&die_chain, nb);
  87 }
  88 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
  89
  90 static inline void conditional_sti(struct pt_regs *regs)
  91 {
  92         if (regs->eflags & X86_EFLAGS_IF)
  93                 local_irq_enable();
  94 }
  95
  96 static inline void preempt_conditional_sti(struct pt_regs *regs)
  97 {
  98         preempt_disable();
  99         if (regs->eflags & X86_EFLAGS_IF)
 100                 local_irq_enable();
 101 }
 102
 103 static inline void preempt_conditional_cli(struct pt_regs *regs)
 104 {
 105         if (regs->eflags & X86_EFLAGS_IF)
 106                 local_irq_disable();
 107         /* Make sure to not schedule here because we could be running
 108            on an exception stack. */
 109         preempt_enable_no_resched();
 110 }
 111
 112 int kstack_depth_to_print = 12;
 113
 114 #ifdef CONFIG_KALLSYMS
 115 void printk_address(unsigned long address)
 116 {
 117         unsigned long offset = 0, symsize;
 118         const char *symname;
 119         char *modname;
 120         char *delim = ":";
 121         char namebuf[128];
 122
 123         symname = kallsyms_lookup(address, &symsize, &offset,
 124                                         &modname, namebuf);
 125         if (!symname) {
 126                 printk(" [<%016lx>]\n", address);
 127                 return;
 128         }
 129         if (!modname)
 130                 modname = delim = "";
 131         printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
 132                 address, delim, modname, delim, symname, offset, symsize);
 133 }
 134 #else
 135 void printk_address(unsigned long address)
 136 {
 137         printk(" [<%016lx>]\n", address);
 138 }
 139 #endif
 140
 141 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 142                                         unsigned *usedp, char **idp)
 143 {
 144         static char ids[][8] = {
 145                 [DEBUG_STACK - 1] = "#DB",
 146                 [NMI_STACK - 1] = "NMI",
 147                 [DOUBLEFAULT_STACK - 1] = "#DF",
 148                 [STACKFAULT_STACK - 1] = "#SS",
 149                 [MCE_STACK - 1] = "#MC",
 150 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 151                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 152 #endif
 153         };
 154         unsigned k;
 155
 156         /*
 157          * Iterate over all exception stacks, and figure out whether
 158          * 'stack' is in one of them:
 159          */
 160         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 161                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
 162                 /*
 163                  * Is 'stack' above this exception frame's end?
 164                  * If yes then skip to the next frame.
 165                  */
 166                 if (stack >= end)
 167                         continue;
 168                 /*
 169                  * Is 'stack' above this exception frame's start address?
 170                  * If yes then we found the right frame.
 171                  */
 172                 if (stack >= end - EXCEPTION_STKSZ) {
 173                         /*
 174                          * Make sure we only iterate through an exception
 175                          * stack once. If it comes up for the second time
 176                          * then there's something wrong going on - just
 177                          * break out and return NULL:
 178                          */
 179                         if (*usedp & (1U << k))
 180                                 break;
 181                         *usedp |= 1U << k;
 182                         *idp = ids[k];
 183                         return (unsigned long *)end;
 184                 }
 185                 /*
 186                  * If this is a debug stack, and if it has a larger size than
 187                  * the usual exception stacks, then 'stack' might still
 188                  * be within the lower portion of the debug stack:
 189                  */
 190 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 191                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 192                         unsigned j = N_EXCEPTION_STACKS - 1;
 193
 194                         /*
 195                          * Black magic. A large debug stack is composed of
 196                          * multiple exception stack entries, which we
 197                          * iterate through now. Dont look:
 198                          */
 199                         do {
 200                                 ++j;
 201                                 end -= EXCEPTION_STKSZ;
 202                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 203                         } while (stack < end - EXCEPTION_STKSZ);
 204                         if (*usedp & (1U << j))
 205                                 break;
 206                         *usedp |= 1U << j;
 207                         *idp = ids[j];
 208                         return (unsigned long *)end;
 209                 }
 210 #endif
 211         }
 212         return NULL;
 213 }
 214
 215 #define MSG(txt) ops->warning(data, txt)
 216
 217 /*
 218  * x86-64 can have upto three kernel stacks:
 219  * process stack
 220  * interrupt stack
 221  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 222  */
 223
 224 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
 225 {
 226         void *t = (void *)tinfo;
 227         return p > t && p < t + THREAD_SIZE - 3;
 228 }
 229
 230 void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
 231                 unsigned long *stack,
 232                 struct stacktrace_ops *ops, void *data)
 233 {
 234         const unsigned cpu = get_cpu();
 235         unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
 236         unsigned used = 0;
 237         struct thread_info *tinfo;
 238
 239         if (!tsk)
 240                 tsk = current;
 241
 242         if (!stack) {
 243                 unsigned long dummy;
 244                 stack = &dummy;
 245                 if (tsk && tsk != current)
 246                         stack = (unsigned long *)tsk->thread.rsp;
 247         }
 248
 249         /*
 250          * Print function call entries within a stack. 'cond' is the
 251          * "end of stackframe" condition, that the 'stack++'
 252          * iteration will eventually trigger.
 253          */
 254 #define HANDLE_STACK(cond) \
 255         do while (cond) { \
 256                 unsigned long addr = *stack++; \
 257                 /* Use unlocked access here because except for NMIs     \
 258                    we should be already protected against module unloads */ \
 259                 if (__kernel_text_address(addr)) { \
 260                         /* \
 261                          * If the address is either in the text segment of the \
 262                          * kernel, or in the region which contains vmalloc'ed \
 263                          * memory, it *may* be the address of a calling \
 264                          * routine; if so, print it so that someone tracing \
 265                          * down the cause of the crash will be able to figure \
 266                          * out the call path that was taken. \
 267                          */ \
 268                         ops->address(data, addr);   \
 269                 } \
 270         } while (0)
 271
 272         /*
 273          * Print function call entries in all stacks, starting at the
 274          * current stack address. If the stacks consist of nested
 275          * exceptions
 276          */
 277         for (;;) {
 278                 char *id;
 279                 unsigned long *estack_end;
 280                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 281                                                 &used, &id);
 282
 283                 if (estack_end) {
 284                         if (ops->stack(data, id) < 0)
 285                                 break;
 286                         HANDLE_STACK (stack < estack_end);
 287                         ops->stack(data, "<EOE>");
 288                         /*
 289                          * We link to the next stack via the
 290                          * second-to-last pointer (index -2 to end) in the
 291                          * exception stack:
 292                          */
 293                         stack = (unsigned long *) estack_end[-2];
 294                         continue;
 295                 }
 296                 if (irqstack_end) {
 297                         unsigned long *irqstack;
 298                         irqstack = irqstack_end -
 299                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 300
 301                         if (stack >= irqstack && stack < irqstack_end) {
 302                                 if (ops->stack(data, "IRQ") < 0)
 303                                         break;
 304                                 HANDLE_STACK (stack < irqstack_end);
 305                                 /*
 306                                  * We link to the next stack (which would be
 307                                  * the process stack normally) the last
 308                                  * pointer (index -1 to end) in the IRQ stack:
 309                                  */
 310                                 stack = (unsigned long *) (irqstack_end[-1]);
 311                                 irqstack_end = NULL;
 312                                 ops->stack(data, "EOI");
 313                                 continue;
 314                         }
 315                 }
 316                 break;
 317         }
 318
 319         /*
 320          * This handles the process stack:
 321          */
 322         tinfo = task_thread_info(tsk);
 323         HANDLE_STACK (valid_stack_ptr(tinfo, stack));
 324 #undef HANDLE_STACK
 325         put_cpu();
 326 }
 327 EXPORT_SYMBOL(dump_trace);
 328
 329 static void
 330 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 331 {
 332         print_symbol(msg, symbol);
 333         printk("\n");
 334 }
 335
 336 static void print_trace_warning(void *data, char *msg)
 337 {
 338         printk("%s\n", msg);
 339 }
 340
 341 static int print_trace_stack(void *data, char *name)
 342 {
 343         printk(" <%s> ", name);
 344         return 0;
 345 }
 346
 347 static void print_trace_address(void *data, unsigned long addr)
 348 {
 349         printk_address(addr);
 350 }
 351
 352 static struct stacktrace_ops print_trace_ops = {
 353         .warning = print_trace_warning,
 354         .warning_symbol = print_trace_warning_symbol,
 355         .stack = print_trace_stack,
 356         .address = print_trace_address,
 357 };
 358
 359 void
 360 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 361 {
 362         printk("\nCall Trace:\n");
 363         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 364         printk("\n");
 365 }
 366
 367 static void
 368 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 369 {
 370         unsigned long *stack;
 371         int i;
 372         const int cpu = smp_processor_id();
 373         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 374         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 375
 376         // debugging aid: "show_stack(NULL, NULL);" prints the
 377         // back trace for this cpu.
 378
 379         if (rsp == NULL) {
 380                 if (tsk)
 381                         rsp = (unsigned long *)tsk->thread.rsp;
 382                 else
 383                         rsp = (unsigned long *)&rsp;
 384         }
 385
 386         stack = rsp;
 387         for(i=0; i < kstack_depth_to_print; i++) {
 388                 if (stack >= irqstack && stack <= irqstack_end) {
 389                         if (stack == irqstack_end) {
 390                                 stack = (unsigned long *) (irqstack_end[-1]);
 391                                 printk(" <EOI> ");
 392                         }
 393                 } else {
 394                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 395                         break;
 396                 }
 397                 if (i && ((i % 4) == 0))
 398                         printk("\n");
 399                 printk(" %016lx", *stack++);
 400                 touch_nmi_watchdog();
 401         }
 402         show_trace(tsk, regs, rsp);
 403 }
 404
 405 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 406 {
 407         _show_stack(tsk, NULL, rsp);
 408 }
 409
 410 /*
 411  * The architecture-independent dump_stack generator
 412  */
 413 void dump_stack(void)
 414 {
 415         unsigned long dummy;
 416         show_trace(NULL, NULL, &dummy);
 417 }
 418
 419 EXPORT_SYMBOL(dump_stack);
 420
 421 void show_registers(struct pt_regs *regs)
 422 {
 423         int i;
 424         int in_kernel = !user_mode(regs);
 425         unsigned long rsp;
 426         const int cpu = smp_processor_id();
 427         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 428
 429                 rsp = regs->rsp;
 430
 431         printk("CPU %d ", cpu);
 432         __show_regs(regs);
 433         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 434                 cur->comm, cur->pid, task_thread_info(cur), cur);
 435
 436         /*
 437          * When in-kernel, we also print out the stack and code at the
 438          * time of the fault..
 439          */
 440         if (in_kernel) {
 441
 442                 printk("Stack: ");
 443                 _show_stack(NULL, regs, (unsigned long*)rsp);
 444
 445                 printk("\nCode: ");
 446                 if (regs->rip < PAGE_OFFSET)
 447                         goto bad;
 448
 449                 for (i=0; i<20; i++) {
 450                         unsigned char c;
 451                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 452 bad:
 453                                 printk(" Bad RIP value.");
 454                                 break;
 455                         }
 456                         printk("%02x ", c);
 457                 }
 458         }
 459         printk("\n");
 460 }
 461
 462 int is_valid_bugaddr(unsigned long rip)
 463 {
 464         unsigned short ud2;
 465
 466         if (__copy_from_user(&ud2, (const void __user *) rip, sizeof(ud2)))
 467                 return 0;
 468
 469         return ud2 == 0x0b0f;
 470 }
 471
 472 #ifdef CONFIG_BUG
 473 void out_of_line_bug(void)
 474 {
 475         BUG();
 476 }
 477 EXPORT_SYMBOL(out_of_line_bug);
 478 #endif
 479
 480 static DEFINE_SPINLOCK(die_lock);
 481 static int die_owner = -1;
 482 static unsigned int die_nest_count;
 483
 484 unsigned __kprobes long oops_begin(void)
 485 {
 486         int cpu = smp_processor_id();
 487         unsigned long flags;
 488
 489         oops_enter();
 490
 491         /* racy, but better than risking deadlock. */
 492         local_irq_save(flags);
 493         if (!spin_trylock(&die_lock)) {
 494                 if (cpu == die_owner)
 495                         /* nested oops. should stop eventually */;
 496                 else
 497                         spin_lock(&die_lock);
 498         }
 499         die_nest_count++;
 500         die_owner = cpu;
 501         console_verbose();
 502         bust_spinlocks(1);
 503         return flags;
 504 }
 505
 506 void __kprobes oops_end(unsigned long flags)
 507 {
 508         die_owner = -1;
 509         bust_spinlocks(0);
 510         die_nest_count--;
 511         if (die_nest_count)
 512                 /* We still own the lock */
 513                 local_irq_restore(flags);
 514         else
 515                 /* Nest count reaches zero, release the lock. */
 516                 spin_unlock_irqrestore(&die_lock, flags);
 517         if (panic_on_oops)
 518                 panic("Fatal exception");
 519         oops_exit();
 520 }
 521
 522 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 523 {
 524         static int die_counter;
 525         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 526 #ifdef CONFIG_PREEMPT
 527         printk("PREEMPT ");
 528 #endif
 529 #ifdef CONFIG_SMP
 530         printk("SMP ");
 531 #endif
 532 #ifdef CONFIG_DEBUG_PAGEALLOC
 533         printk("DEBUG_PAGEALLOC");
 534 #endif
 535         printk("\n");
 536         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 537         show_registers(regs);
 538         /* Executive summary in case the oops scrolled away */
 539         printk(KERN_ALERT "RIP ");
 540         printk_address(regs->rip);
 541         printk(" RSP <%016lx>\n", regs->rsp);
 542         if (kexec_should_crash(current))
 543                 crash_kexec(regs);
 544 }
 545
 546 void die(const char * str, struct pt_regs * regs, long err)
 547 {
 548         unsigned long flags = oops_begin();
 549
 550         if (!user_mode(regs))
 551                 report_bug(regs->rip);
 552
 553         __die(str, regs, err);
 554         oops_end(flags);
 555         do_exit(SIGSEGV);
 556 }
 557
 558 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 559 {
 560         unsigned long flags = oops_begin();
 561
 562         /*
 563          * We are in trouble anyway, lets at least try
 564          * to get a message out.
 565          */
 566         printk(str, smp_processor_id());
 567         show_registers(regs);
 568         if (kexec_should_crash(current))
 569                 crash_kexec(regs);
 570         if (do_panic || panic_on_oops)
 571                 panic("Non maskable interrupt");
 572         oops_end(flags);
 573         nmi_exit();
 574         local_irq_enable();
 575         do_exit(SIGSEGV);
 576 }
 577
 578 static void __kprobes do_trap(int trapnr, int signr, char *str,
 579                               struct pt_regs * regs, long error_code,
 580                               siginfo_t *info)
 581 {
 582         struct task_struct *tsk = current;
 583
 584         tsk->thread.error_code = error_code;
 585         tsk->thread.trap_no = trapnr;
 586
 587         if (user_mode(regs)) {
 588                 if (exception_trace && unhandled_signal(tsk, signr))
 589                         printk(KERN_INFO
 590                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 591                                tsk->comm, tsk->pid, str,
 592                                regs->rip, regs->rsp, error_code);
 593
 594                 if (info)
 595                         force_sig_info(signr, info, tsk);
 596                 else
 597                         force_sig(signr, tsk);
 598                 return;
 599         }
 600
 601
 602         /* kernel trap */
 603         {
 604                 const struct exception_table_entry *fixup;
 605                 fixup = search_exception_tables(regs->rip);
 606                 if (fixup)
 607                         regs->rip = fixup->fixup;
 608                 else
 609                         die(str, regs, error_code);
 610                 return;
 611         }
 612 }
 613
 614 #define DO_ERROR(trapnr, signr, str, name) \
 615 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 616 { \
 617         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 618                                                         == NOTIFY_STOP) \
 619                 return; \
 620         conditional_sti(regs);                                          \
 621         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 622 }
 623
 624 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 625 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 626 { \
 627         siginfo_t info; \
 628         info.si_signo = signr; \
 629         info.si_errno = 0; \
 630         info.si_code = sicode; \
 631         info.si_addr = (void __user *)siaddr; \
 632         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 633                                                         == NOTIFY_STOP) \
 634                 return; \
 635         conditional_sti(regs);                                          \
 636         do_trap(trapnr, signr, str, regs, error_code, &info); \
 637 }
 638
 639 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 640 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 641 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 642 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 643 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 644 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 645 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 646 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 647 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 648 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 649
 650 /* Runs on IST stack */
 651 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 652 {
 653         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 654                         12, SIGBUS) == NOTIFY_STOP)
 655                 return;
 656         preempt_conditional_sti(regs);
 657         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 658         preempt_conditional_cli(regs);
 659 }
 660
 661 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 662 {
 663         static const char str[] = "double fault";
 664         struct task_struct *tsk = current;
 665
 666         /* Return not checked because double check cannot be ignored */
 667         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 668
 669         tsk->thread.error_code = error_code;
 670         tsk->thread.trap_no = 8;
 671
 672         /* This is always a kernel trap and never fixable (and thus must
 673            never return). */
 674         for (;;)
 675                 die(str, regs, error_code);
 676 }
 677
 678 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 679                                                 long error_code)
 680 {
 681         struct task_struct *tsk = current;
 682
 683         conditional_sti(regs);
 684
 685         tsk->thread.error_code = error_code;
 686         tsk->thread.trap_no = 13;
 687
 688         if (user_mode(regs)) {
 689                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 690                         printk(KERN_INFO
 691                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 692                                tsk->comm, tsk->pid,
 693                                regs->rip, regs->rsp, error_code);
 694
 695                 force_sig(SIGSEGV, tsk);
 696                 return;
 697         }
 698
 699         /* kernel gp */
 700         {
 701                 const struct exception_table_entry *fixup;
 702                 fixup = search_exception_tables(regs->rip);
 703                 if (fixup) {
 704                         regs->rip = fixup->fixup;
 705                         return;
 706                 }
 707                 if (notify_die(DIE_GPF, "general protection fault", regs,
 708                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 709                         return;
 710                 die("general protection fault", regs, error_code);
 711         }
 712 }
 713
 714 static __kprobes void
 715 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 716 {
 717         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 718                 reason);
 719         printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 720
 721         if (panic_on_unrecovered_nmi)
 722                 panic("NMI: Not continuing");
 723
 724         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 725
 726         /* Clear and disable the memory parity error line. */
 727         reason = (reason & 0xf) | 4;
 728         outb(reason, 0x61);
 729 }
 730
 731 static __kprobes void
 732 io_check_error(unsigned char reason, struct pt_regs * regs)
 733 {
 734         printk("NMI: IOCK error (debug interrupt?)\n");
 735         show_registers(regs);
 736
 737         /* Re-enable the IOCK line, wait for a few seconds */
 738         reason = (reason & 0xf) | 8;
 739         outb(reason, 0x61);
 740         mdelay(2000);
 741         reason &= ~8;
 742         outb(reason, 0x61);
 743 }
 744
 745 static __kprobes void
 746 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 747 {
 748         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 749                 reason);
 750         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 751
 752         if (panic_on_unrecovered_nmi)
 753                 panic("NMI: Not continuing");
 754
 755         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 756 }
 757
 758 /* Runs on IST stack. This code must keep interrupts off all the time.
 759    Nested NMIs are prevented by the CPU. */
 760 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 761 {
 762         unsigned char reason = 0;
 763         int cpu;
 764
 765         cpu = smp_processor_id();
 766
 767         /* Only the BSP gets external NMIs from the system.  */
 768         if (!cpu)
 769                 reason = get_nmi_reason();
 770
 771         if (!(reason & 0xc0)) {
 772                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 773                                                                 == NOTIFY_STOP)
 774                         return;
 775                 /*
 776                  * Ok, so this is none of the documented NMI sources,
 777                  * so it must be the NMI watchdog.
 778                  */
 779                 if (nmi_watchdog_tick(regs,reason))
 780                         return;
 781                 if (!do_nmi_callback(regs,cpu))
 782                         unknown_nmi_error(reason, regs);
 783
 784                 return;
 785         }
 786         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 787                 return;
 788
 789         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 790
 791         if (reason & 0x80)
 792                 mem_parity_error(reason, regs);
 793         if (reason & 0x40)
 794                 io_check_error(reason, regs);
 795 }
 796
 797 /* runs on IST stack. */
 798 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 799 {
 800         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 801                 return;
 802         }
 803         preempt_conditional_sti(regs);
 804         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 805         preempt_conditional_cli(regs);
 806 }
 807
 808 /* Help handler running on IST stack to switch back to user stack
 809    for scheduling or signal handling. The actual stack switch is done in
 810    entry.S */
 811 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 812 {
 813         struct pt_regs *regs = eregs;
 814         /* Did already sync */
 815         if (eregs == (struct pt_regs *)eregs->rsp)
 816                 ;
 817         /* Exception from user space */
 818         else if (user_mode(eregs))
 819                 regs = task_pt_regs(current);
 820         /* Exception from kernel and interrupts are enabled. Move to
 821            kernel process stack. */
 822         else if (eregs->eflags & X86_EFLAGS_IF)
 823                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 824         if (eregs != regs)
 825                 *regs = *eregs;
 826         return regs;
 827 }
 828
 829 /* runs on IST stack. */
 830 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 831                                    unsigned long error_code)
 832 {
 833         unsigned long condition;
 834         struct task_struct *tsk = current;
 835         siginfo_t info;
 836
 837         get_debugreg(condition, 6);
 838
 839         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 840                                                 SIGTRAP) == NOTIFY_STOP)
 841                 return;
 842
 843         preempt_conditional_sti(regs);
 844
 845         /* Mask out spurious debug traps due to lazy DR7 setting */
 846         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 847                 if (!tsk->thread.debugreg7) {
 848                         goto clear_dr7;
 849                 }
 850         }
 851
 852         tsk->thread.debugreg6 = condition;
 853
 854         /* Mask out spurious TF errors due to lazy TF clearing */
 855         if (condition & DR_STEP) {
 856                 /*
 857                  * The TF error should be masked out only if the current
 858                  * process is not traced and if the TRAP flag has been set
 859                  * previously by a tracing process (condition detected by
 860                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 861                  * can be modified by the process itself in user mode,
 862                  * allowing programs to debug themselves without the ptrace()
 863                  * interface.
 864                  */
 865                 if (!user_mode(regs))
 866                        goto clear_TF_reenable;
 867                 /*
 868                  * Was the TF flag set by a debugger? If so, clear it now,
 869                  * so that register information is correct.
 870                  */
 871                 if (tsk->ptrace & PT_DTRACE) {
 872                         regs->eflags &= ~TF_MASK;
 873                         tsk->ptrace &= ~PT_DTRACE;
 874                 }
 875         }
 876
 877         /* Ok, finally something we can handle */
 878         tsk->thread.trap_no = 1;
 879         tsk->thread.error_code = error_code;
 880         info.si_signo = SIGTRAP;
 881         info.si_errno = 0;
 882         info.si_code = TRAP_BRKPT;
 883         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 884         force_sig_info(SIGTRAP, &info, tsk);
 885
 886 clear_dr7:
 887         set_debugreg(0UL, 7);
 888         preempt_conditional_cli(regs);
 889         return;
 890
 891 clear_TF_reenable:
 892         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 893         regs->eflags &= ~TF_MASK;
 894         preempt_conditional_cli(regs);
 895 }
 896
 897 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 898 {
 899         const struct exception_table_entry *fixup;
 900         fixup = search_exception_tables(regs->rip);
 901         if (fixup) {
 902                 regs->rip = fixup->fixup;
 903                 return 1;
 904         }
 905         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 906         /* Illegal floating point operation in the kernel */
 907         current->thread.trap_no = trapnr;
 908         die(str, regs, 0);
 909         return 0;
 910 }
 911
 912 /*
 913  * Note that we play around with the 'TS' bit in an attempt to get
 914  * the correct behaviour even in the presence of the asynchronous
 915  * IRQ13 behaviour
 916  */
 917 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 918 {
 919         void __user *rip = (void __user *)(regs->rip);
 920         struct task_struct * task;
 921         siginfo_t info;
 922         unsigned short cwd, swd;
 923
 924         conditional_sti(regs);
 925         if (!user_mode(regs) &&
 926             kernel_math_error(regs, "kernel x87 math error", 16))
 927                 return;
 928
 929         /*
 930          * Save the info for the exception handler and clear the error.
 931          */
 932         task = current;
 933         save_init_fpu(task);
 934         task->thread.trap_no = 16;
 935         task->thread.error_code = 0;
 936         info.si_signo = SIGFPE;
 937         info.si_errno = 0;
 938         info.si_code = __SI_FAULT;
 939         info.si_addr = rip;
 940         /*
 941          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 942          * status.  0x3f is the exception bits in these regs, 0x200 is the
 943          * C1 reg you need in case of a stack fault, 0x040 is the stack
 944          * fault bit.  We should only be taking one exception at a time,
 945          * so if this combination doesn't produce any single exception,
 946          * then we have a bad program that isn't synchronizing its FPU usage
 947          * and it will suffer the consequences since we won't be able to
 948          * fully reproduce the context of the exception
 949          */
 950         cwd = get_fpu_cwd(task);
 951         swd = get_fpu_swd(task);
 952         switch (swd & ~cwd & 0x3f) {
 953                 case 0x000:
 954                 default:
 955                         break;
 956                 case 0x001: /* Invalid Op */
 957                         /*
 958                          * swd & 0x240 == 0x040: Stack Underflow
 959                          * swd & 0x240 == 0x240: Stack Overflow
 960                          * User must clear the SF bit (0x40) if set
 961                          */
 962                         info.si_code = FPE_FLTINV;
 963                         break;
 964                 case 0x002: /* Denormalize */
 965                 case 0x010: /* Underflow */
 966                         info.si_code = FPE_FLTUND;
 967                         break;
 968                 case 0x004: /* Zero Divide */
 969                         info.si_code = FPE_FLTDIV;
 970                         break;
 971                 case 0x008: /* Overflow */
 972                         info.si_code = FPE_FLTOVF;
 973                         break;
 974                 case 0x020: /* Precision */
 975                         info.si_code = FPE_FLTRES;
 976                         break;
 977         }
 978         force_sig_info(SIGFPE, &info, task);
 979 }
 980
 981 asmlinkage void bad_intr(void)
 982 {
 983         printk("bad interrupt");
 984 }
 985
 986 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 987 {
 988         void __user *rip = (void __user *)(regs->rip);
 989         struct task_struct * task;
 990         siginfo_t info;
 991         unsigned short mxcsr;
 992
 993         conditional_sti(regs);
 994         if (!user_mode(regs) &&
 995                 kernel_math_error(regs, "kernel simd math error", 19))
 996                 return;
 997
 998         /*
 999          * Save the info for the exception handler and clear the error.
1000          */
1001         task = current;
1002         save_init_fpu(task);
1003         task->thread.trap_no = 19;
1004         task->thread.error_code = 0;
1005         info.si_signo = SIGFPE;
1006         info.si_errno = 0;
1007         info.si_code = __SI_FAULT;
1008         info.si_addr = rip;
1009         /*
1010          * The SIMD FPU exceptions are handled a little differently, as there
1011          * is only a single status/control register.  Thus, to determine which
1012          * unmasked exception was caught we must mask the exception mask bits
1013          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1014          */
1015         mxcsr = get_fpu_mxcsr(task);
1016         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1017                 case 0x000:
1018                 default:
1019                         break;
1020                 case 0x001: /* Invalid Op */
1021                         info.si_code = FPE_FLTINV;
1022                         break;
1023                 case 0x002: /* Denormalize */
1024                 case 0x010: /* Underflow */
1025                         info.si_code = FPE_FLTUND;
1026                         break;
1027                 case 0x004: /* Zero Divide */
1028                         info.si_code = FPE_FLTDIV;
1029                         break;
1030                 case 0x008: /* Overflow */
1031                         info.si_code = FPE_FLTOVF;
1032                         break;
1033                 case 0x020: /* Precision */
1034                         info.si_code = FPE_FLTRES;
1035                         break;
1036         }
1037         force_sig_info(SIGFPE, &info, task);
1038 }
1039
1040 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1041 {
1042 }
1043
1044 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1045 {
1046 }
1047
1048 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1049 {
1050 }
1051
1052 /*
1053  *  'math_state_restore()' saves the current math information in the
1054  * old math state array, and gets the new ones from the current task
1055  *
1056  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1057  * Don't touch unless you *really* know how it works.
1058  */
1059 asmlinkage void math_state_restore(void)
1060 {
1061         struct task_struct *me = current;
1062         clts();                 /* Allow maths ops (or we recurse) */
1063
1064         if (!used_math())
1065                 init_fpu(me);
1066         restore_fpu_checking(&me->thread.i387.fxsave);
1067         task_thread_info(me)->status |= TS_USEDFPU;
1068         me->fpu_counter++;
1069 }
1070
1071 void __init trap_init(void)
1072 {
1073         set_intr_gate(0,&divide_error);
1074         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1075         set_intr_gate_ist(2,&nmi,NMI_STACK);
1076         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1077         set_system_gate(4,&overflow);   /* int4 can be called from all */
1078         set_intr_gate(5,&bounds);
1079         set_intr_gate(6,&invalid_op);
1080         set_intr_gate(7,&device_not_available);
1081         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1082         set_intr_gate(9,&coprocessor_segment_overrun);
1083         set_intr_gate(10,&invalid_TSS);
1084         set_intr_gate(11,&segment_not_present);
1085         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1086         set_intr_gate(13,&general_protection);
1087         set_intr_gate(14,&page_fault);
1088         set_intr_gate(15,&spurious_interrupt_bug);
1089         set_intr_gate(16,&coprocessor_error);
1090         set_intr_gate(17,&alignment_check);
1091 #ifdef CONFIG_X86_MCE
1092         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1093 #endif
1094         set_intr_gate(19,&simd_coprocessor_error);
1095
1096 #ifdef CONFIG_IA32_EMULATION
1097         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1098 #endif
1099
1100         /*
1101          * Should be a barrier for any external CPU state.
1102          */
1103         cpu_init();
1104 }
1105
1106
1107 static int __init oops_setup(char *s)
1108 {
1109         if (!s)
1110                 return -EINVAL;
1111         if (!strcmp(s, "panic"))
1112                 panic_on_oops = 1;
1113         return 0;
1114 }
1115 early_param("oops", oops_setup);
1116
1117 static int __init kstack_setup(char *s)
1118 {
1119         if (!s)
1120                 return -EINVAL;
1121         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1122         return 0;
1123 }
1124 early_param("kstack", kstack_setup);