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 #include <linux/kdebug.h>
  36
  37 #if defined(CONFIG_EDAC)
  38 #include <linux/edac.h>
  39 #endif
  40
  41 #include <asm/system.h>
  42 #include <asm/io.h>
  43 #include <asm/atomic.h>
  44 #include <asm/debugreg.h>
  45 #include <asm/desc.h>
  46 #include <asm/i387.h>
  47 #include <asm/processor.h>
  48 #include <asm/unwind.h>
  49 #include <asm/smp.h>
  50 #include <asm/pgalloc.h>
  51 #include <asm/pda.h>
  52 #include <asm/proto.h>
  53 #include <asm/nmi.h>
  54 #include <asm/stacktrace.h>
  55
  56 asmlinkage void divide_error(void);
  57 asmlinkage void debug(void);
  58 asmlinkage void nmi(void);
  59 asmlinkage void int3(void);
  60 asmlinkage void overflow(void);
  61 asmlinkage void bounds(void);
  62 asmlinkage void invalid_op(void);
  63 asmlinkage void device_not_available(void);
  64 asmlinkage void double_fault(void);
  65 asmlinkage void coprocessor_segment_overrun(void);
  66 asmlinkage void invalid_TSS(void);
  67 asmlinkage void segment_not_present(void);
  68 asmlinkage void stack_segment(void);
  69 asmlinkage void general_protection(void);
  70 asmlinkage void page_fault(void);
  71 asmlinkage void coprocessor_error(void);
  72 asmlinkage void simd_coprocessor_error(void);
  73 asmlinkage void reserved(void);
  74 asmlinkage void alignment_check(void);
  75 asmlinkage void machine_check(void);
  76 asmlinkage void spurious_interrupt_bug(void);
  77
  78 static inline void conditional_sti(struct pt_regs *regs)
  79 {
  80         if (regs->eflags & X86_EFLAGS_IF)
  81                 local_irq_enable();
  82 }
  83
  84 static inline void preempt_conditional_sti(struct pt_regs *regs)
  85 {
  86         preempt_disable();
  87         if (regs->eflags & X86_EFLAGS_IF)
  88                 local_irq_enable();
  89 }
  90
  91 static inline void preempt_conditional_cli(struct pt_regs *regs)
  92 {
  93         if (regs->eflags & X86_EFLAGS_IF)
  94                 local_irq_disable();
  95         /* Make sure to not schedule here because we could be running
  96            on an exception stack. */
  97         preempt_enable_no_resched();
  98 }
  99
 100 int kstack_depth_to_print = 12;
 101
 102 #ifdef CONFIG_KALLSYMS
 103 void printk_address(unsigned long address)
 104 {
 105         unsigned long offset = 0, symsize;
 106         const char *symname;
 107         char *modname;
 108         char *delim = ":";
 109         char namebuf[128];
 110
 111         symname = kallsyms_lookup(address, &symsize, &offset,
 112                                         &modname, namebuf);
 113         if (!symname) {
 114                 printk(" [<%016lx>]\n", address);
 115                 return;
 116         }
 117         if (!modname)
 118                 modname = delim = "";
 119         printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
 120                 address, delim, modname, delim, symname, offset, symsize);
 121 }
 122 #else
 123 void printk_address(unsigned long address)
 124 {
 125         printk(" [<%016lx>]\n", address);
 126 }
 127 #endif
 128
 129 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 130                                         unsigned *usedp, char **idp)
 131 {
 132         static char ids[][8] = {
 133                 [DEBUG_STACK - 1] = "#DB",
 134                 [NMI_STACK - 1] = "NMI",
 135                 [DOUBLEFAULT_STACK - 1] = "#DF",
 136                 [STACKFAULT_STACK - 1] = "#SS",
 137                 [MCE_STACK - 1] = "#MC",
 138 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 139                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 140 #endif
 141         };
 142         unsigned k;
 143
 144         /*
 145          * Iterate over all exception stacks, and figure out whether
 146          * 'stack' is in one of them:
 147          */
 148         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 149                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
 150                 /*
 151                  * Is 'stack' above this exception frame's end?
 152                  * If yes then skip to the next frame.
 153                  */
 154                 if (stack >= end)
 155                         continue;
 156                 /*
 157                  * Is 'stack' above this exception frame's start address?
 158                  * If yes then we found the right frame.
 159                  */
 160                 if (stack >= end - EXCEPTION_STKSZ) {
 161                         /*
 162                          * Make sure we only iterate through an exception
 163                          * stack once. If it comes up for the second time
 164                          * then there's something wrong going on - just
 165                          * break out and return NULL:
 166                          */
 167                         if (*usedp & (1U << k))
 168                                 break;
 169                         *usedp |= 1U << k;
 170                         *idp = ids[k];
 171                         return (unsigned long *)end;
 172                 }
 173                 /*
 174                  * If this is a debug stack, and if it has a larger size than
 175                  * the usual exception stacks, then 'stack' might still
 176                  * be within the lower portion of the debug stack:
 177                  */
 178 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 179                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 180                         unsigned j = N_EXCEPTION_STACKS - 1;
 181
 182                         /*
 183                          * Black magic. A large debug stack is composed of
 184                          * multiple exception stack entries, which we
 185                          * iterate through now. Dont look:
 186                          */
 187                         do {
 188                                 ++j;
 189                                 end -= EXCEPTION_STKSZ;
 190                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 191                         } while (stack < end - EXCEPTION_STKSZ);
 192                         if (*usedp & (1U << j))
 193                                 break;
 194                         *usedp |= 1U << j;
 195                         *idp = ids[j];
 196                         return (unsigned long *)end;
 197                 }
 198 #endif
 199         }
 200         return NULL;
 201 }
 202
 203 #define MSG(txt) ops->warning(data, txt)
 204
 205 /*
 206  * x86-64 can have upto three kernel stacks:
 207  * process stack
 208  * interrupt stack
 209  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 210  */
 211
 212 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
 213 {
 214         void *t = (void *)tinfo;
 215         return p > t && p < t + THREAD_SIZE - 3;
 216 }
 217
 218 void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
 219                 unsigned long *stack,
 220                 struct stacktrace_ops *ops, void *data)
 221 {
 222         const unsigned cpu = get_cpu();
 223         unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
 224         unsigned used = 0;
 225         struct thread_info *tinfo;
 226
 227         if (!tsk)
 228                 tsk = current;
 229
 230         if (!stack) {
 231                 unsigned long dummy;
 232                 stack = &dummy;
 233                 if (tsk && tsk != current)
 234                         stack = (unsigned long *)tsk->thread.rsp;
 235         }
 236
 237         /*
 238          * Print function call entries within a stack. 'cond' is the
 239          * "end of stackframe" condition, that the 'stack++'
 240          * iteration will eventually trigger.
 241          */
 242 #define HANDLE_STACK(cond) \
 243         do while (cond) { \
 244                 unsigned long addr = *stack++; \
 245                 /* Use unlocked access here because except for NMIs     \
 246                    we should be already protected against module unloads */ \
 247                 if (__kernel_text_address(addr)) { \
 248                         /* \
 249                          * If the address is either in the text segment of the \
 250                          * kernel, or in the region which contains vmalloc'ed \
 251                          * memory, it *may* be the address of a calling \
 252                          * routine; if so, print it so that someone tracing \
 253                          * down the cause of the crash will be able to figure \
 254                          * out the call path that was taken. \
 255                          */ \
 256                         ops->address(data, addr);   \
 257                 } \
 258         } while (0)
 259
 260         /*
 261          * Print function call entries in all stacks, starting at the
 262          * current stack address. If the stacks consist of nested
 263          * exceptions
 264          */
 265         for (;;) {
 266                 char *id;
 267                 unsigned long *estack_end;
 268                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 269                                                 &used, &id);
 270
 271                 if (estack_end) {
 272                         if (ops->stack(data, id) < 0)
 273                                 break;
 274                         HANDLE_STACK (stack < estack_end);
 275                         ops->stack(data, "<EOE>");
 276                         /*
 277                          * We link to the next stack via the
 278                          * second-to-last pointer (index -2 to end) in the
 279                          * exception stack:
 280                          */
 281                         stack = (unsigned long *) estack_end[-2];
 282                         continue;
 283                 }
 284                 if (irqstack_end) {
 285                         unsigned long *irqstack;
 286                         irqstack = irqstack_end -
 287                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 288
 289                         if (stack >= irqstack && stack < irqstack_end) {
 290                                 if (ops->stack(data, "IRQ") < 0)
 291                                         break;
 292                                 HANDLE_STACK (stack < irqstack_end);
 293                                 /*
 294                                  * We link to the next stack (which would be
 295                                  * the process stack normally) the last
 296                                  * pointer (index -1 to end) in the IRQ stack:
 297                                  */
 298                                 stack = (unsigned long *) (irqstack_end[-1]);
 299                                 irqstack_end = NULL;
 300                                 ops->stack(data, "EOI");
 301                                 continue;
 302                         }
 303                 }
 304                 break;
 305         }
 306
 307         /*
 308          * This handles the process stack:
 309          */
 310         tinfo = task_thread_info(tsk);
 311         HANDLE_STACK (valid_stack_ptr(tinfo, stack));
 312 #undef HANDLE_STACK
 313         put_cpu();
 314 }
 315 EXPORT_SYMBOL(dump_trace);
 316
 317 static void
 318 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 319 {
 320         print_symbol(msg, symbol);
 321         printk("\n");
 322 }
 323
 324 static void print_trace_warning(void *data, char *msg)
 325 {
 326         printk("%s\n", msg);
 327 }
 328
 329 static int print_trace_stack(void *data, char *name)
 330 {
 331         printk(" <%s> ", name);
 332         return 0;
 333 }
 334
 335 static void print_trace_address(void *data, unsigned long addr)
 336 {
 337         touch_nmi_watchdog();
 338         printk_address(addr);
 339 }
 340
 341 static struct stacktrace_ops print_trace_ops = {
 342         .warning = print_trace_warning,
 343         .warning_symbol = print_trace_warning_symbol,
 344         .stack = print_trace_stack,
 345         .address = print_trace_address,
 346 };
 347
 348 void
 349 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 350 {
 351         printk("\nCall Trace:\n");
 352         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 353         printk("\n");
 354 }
 355
 356 static void
 357 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 358 {
 359         unsigned long *stack;
 360         int i;
 361         const int cpu = smp_processor_id();
 362         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 363         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 364
 365         // debugging aid: "show_stack(NULL, NULL);" prints the
 366         // back trace for this cpu.
 367
 368         if (rsp == NULL) {
 369                 if (tsk)
 370                         rsp = (unsigned long *)tsk->thread.rsp;
 371                 else
 372                         rsp = (unsigned long *)&rsp;
 373         }
 374
 375         stack = rsp;
 376         for(i=0; i < kstack_depth_to_print; i++) {
 377                 if (stack >= irqstack && stack <= irqstack_end) {
 378                         if (stack == irqstack_end) {
 379                                 stack = (unsigned long *) (irqstack_end[-1]);
 380                                 printk(" <EOI> ");
 381                         }
 382                 } else {
 383                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 384                         break;
 385                 }
 386                 if (i && ((i % 4) == 0))
 387                         printk("\n");
 388                 printk(" %016lx", *stack++);
 389                 touch_nmi_watchdog();
 390         }
 391         show_trace(tsk, regs, rsp);
 392 }
 393
 394 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 395 {
 396         _show_stack(tsk, NULL, rsp);
 397 }
 398
 399 /*
 400  * The architecture-independent dump_stack generator
 401  */
 402 void dump_stack(void)
 403 {
 404         unsigned long dummy;
 405         show_trace(NULL, NULL, &dummy);
 406 }
 407
 408 EXPORT_SYMBOL(dump_stack);
 409
 410 void show_registers(struct pt_regs *regs)
 411 {
 412         int i;
 413         int in_kernel = !user_mode(regs);
 414         unsigned long rsp;
 415         const int cpu = smp_processor_id();
 416         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 417
 418         rsp = regs->rsp;
 419         printk("CPU %d ", cpu);
 420         __show_regs(regs);
 421         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 422                 cur->comm, cur->pid, task_thread_info(cur), cur);
 423
 424         /*
 425          * When in-kernel, we also print out the stack and code at the
 426          * time of the fault..
 427          */
 428         if (in_kernel) {
 429                 printk("Stack: ");
 430                 _show_stack(NULL, regs, (unsigned long*)rsp);
 431
 432                 printk("\nCode: ");
 433                 if (regs->rip < PAGE_OFFSET)
 434                         goto bad;
 435
 436                 for (i=0; i<20; i++) {
 437                         unsigned char c;
 438                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 439 bad:
 440                                 printk(" Bad RIP value.");
 441                                 break;
 442                         }
 443                         printk("%02x ", c);
 444                 }
 445         }
 446         printk("\n");
 447 }
 448
 449 int is_valid_bugaddr(unsigned long rip)
 450 {
 451         unsigned short ud2;
 452
 453         if (__copy_from_user(&ud2, (const void __user *) rip, sizeof(ud2)))
 454                 return 0;
 455
 456         return ud2 == 0x0b0f;
 457 }
 458
 459 #ifdef CONFIG_BUG
 460 void out_of_line_bug(void)
 461 {
 462         BUG();
 463 }
 464 EXPORT_SYMBOL(out_of_line_bug);
 465 #endif
 466
 467 static DEFINE_SPINLOCK(die_lock);
 468 static int die_owner = -1;
 469 static unsigned int die_nest_count;
 470
 471 unsigned __kprobes long oops_begin(void)
 472 {
 473         int cpu;
 474         unsigned long flags;
 475
 476         oops_enter();
 477
 478         /* racy, but better than risking deadlock. */
 479         local_irq_save(flags);
 480         cpu = smp_processor_id();
 481         if (!spin_trylock(&die_lock)) {
 482                 if (cpu == die_owner)
 483                         /* nested oops. should stop eventually */;
 484                 else
 485                         spin_lock(&die_lock);
 486         }
 487         die_nest_count++;
 488         die_owner = cpu;
 489         console_verbose();
 490         bust_spinlocks(1);
 491         return flags;
 492 }
 493
 494 void __kprobes oops_end(unsigned long flags)
 495 {
 496         die_owner = -1;
 497         bust_spinlocks(0);
 498         die_nest_count--;
 499         if (die_nest_count)
 500                 /* We still own the lock */
 501                 local_irq_restore(flags);
 502         else
 503                 /* Nest count reaches zero, release the lock. */
 504                 spin_unlock_irqrestore(&die_lock, flags);
 505         if (panic_on_oops)
 506                 panic("Fatal exception");
 507         oops_exit();
 508 }
 509
 510 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 511 {
 512         static int die_counter;
 513         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 514 #ifdef CONFIG_PREEMPT
 515         printk("PREEMPT ");
 516 #endif
 517 #ifdef CONFIG_SMP
 518         printk("SMP ");
 519 #endif
 520 #ifdef CONFIG_DEBUG_PAGEALLOC
 521         printk("DEBUG_PAGEALLOC");
 522 #endif
 523         printk("\n");
 524         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 525         show_registers(regs);
 526         add_taint(TAINT_DIE);
 527         /* Executive summary in case the oops scrolled away */
 528         printk(KERN_ALERT "RIP ");
 529         printk_address(regs->rip);
 530         printk(" RSP <%016lx>\n", regs->rsp);
 531         if (kexec_should_crash(current))
 532                 crash_kexec(regs);
 533 }
 534
 535 void die(const char * str, struct pt_regs * regs, long err)
 536 {
 537         unsigned long flags = oops_begin();
 538
 539         if (!user_mode(regs))
 540                 report_bug(regs->rip, regs);
 541
 542         __die(str, regs, err);
 543         oops_end(flags);
 544         do_exit(SIGSEGV);
 545 }
 546
 547 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 548 {
 549         unsigned long flags = oops_begin();
 550
 551         /*
 552          * We are in trouble anyway, lets at least try
 553          * to get a message out.
 554          */
 555         printk(str, smp_processor_id());
 556         show_registers(regs);
 557         if (kexec_should_crash(current))
 558                 crash_kexec(regs);
 559         if (do_panic || panic_on_oops)
 560                 panic("Non maskable interrupt");
 561         oops_end(flags);
 562         nmi_exit();
 563         local_irq_enable();
 564         do_exit(SIGSEGV);
 565 }
 566
 567 static void __kprobes do_trap(int trapnr, int signr, char *str,
 568                               struct pt_regs * regs, long error_code,
 569                               siginfo_t *info)
 570 {
 571         struct task_struct *tsk = current;
 572
 573         if (user_mode(regs)) {
 574                 /*
 575                  * We want error_code and trap_no set for userspace
 576                  * faults and kernelspace faults which result in
 577                  * die(), but not kernelspace faults which are fixed
 578                  * up.  die() gives the process no chance to handle
 579                  * the signal and notice the kernel fault information,
 580                  * so that won't result in polluting the information
 581                  * about previously queued, but not yet delivered,
 582                  * faults.  See also do_general_protection below.
 583                  */
 584                 tsk->thread.error_code = error_code;
 585                 tsk->thread.trap_no = trapnr;
 586
 587                 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 588                     printk_ratelimit())
 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                         tsk->thread.error_code = error_code;
 610                         tsk->thread.trap_no = trapnr;
 611                         die(str, regs, error_code);
 612                 }
 613                 return;
 614         }
 615 }
 616
 617 #define DO_ERROR(trapnr, signr, str, name) \
 618 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 619 { \
 620         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 621                                                         == NOTIFY_STOP) \
 622                 return; \
 623         conditional_sti(regs);                                          \
 624         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 625 }
 626
 627 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 628 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 629 { \
 630         siginfo_t info; \
 631         info.si_signo = signr; \
 632         info.si_errno = 0; \
 633         info.si_code = sicode; \
 634         info.si_addr = (void __user *)siaddr; \
 635         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 636                                                         == NOTIFY_STOP) \
 637                 return; \
 638         conditional_sti(regs);                                          \
 639         do_trap(trapnr, signr, str, regs, error_code, &info); \
 640 }
 641
 642 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 643 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 644 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 645 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 646 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 647 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 648 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 649 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 650 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 651 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 652
 653 /* Runs on IST stack */
 654 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 655 {
 656         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 657                         12, SIGBUS) == NOTIFY_STOP)
 658                 return;
 659         preempt_conditional_sti(regs);
 660         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 661         preempt_conditional_cli(regs);
 662 }
 663
 664 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 665 {
 666         static const char str[] = "double fault";
 667         struct task_struct *tsk = current;
 668
 669         /* Return not checked because double check cannot be ignored */
 670         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 671
 672         tsk->thread.error_code = error_code;
 673         tsk->thread.trap_no = 8;
 674
 675         /* This is always a kernel trap and never fixable (and thus must
 676            never return). */
 677         for (;;)
 678                 die(str, regs, error_code);
 679 }
 680
 681 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 682                                                 long error_code)
 683 {
 684         struct task_struct *tsk = current;
 685
 686         conditional_sti(regs);
 687
 688         if (user_mode(regs)) {
 689                 tsk->thread.error_code = error_code;
 690                 tsk->thread.trap_no = 13;
 691
 692                 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 693                     printk_ratelimit())
 694                         printk(KERN_INFO
 695                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 696                                tsk->comm, tsk->pid,
 697                                regs->rip, regs->rsp, error_code);
 698
 699                 force_sig(SIGSEGV, tsk);
 700                 return;
 701         }
 702
 703         /* kernel gp */
 704         {
 705                 const struct exception_table_entry *fixup;
 706                 fixup = search_exception_tables(regs->rip);
 707                 if (fixup) {
 708                         regs->rip = fixup->fixup;
 709                         return;
 710                 }
 711
 712                 tsk->thread.error_code = error_code;
 713                 tsk->thread.trap_no = 13;
 714                 if (notify_die(DIE_GPF, "general protection fault", regs,
 715                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 716                         return;
 717                 die("general protection fault", regs, error_code);
 718         }
 719 }
 720
 721 static __kprobes void
 722 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 723 {
 724         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 725                 reason);
 726         printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 727
 728 #if defined(CONFIG_EDAC)
 729         if(edac_handler_set()) {
 730                 edac_atomic_assert_error();
 731                 return;
 732         }
 733 #endif
 734
 735         if (panic_on_unrecovered_nmi)
 736                 panic("NMI: Not continuing");
 737
 738         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 739
 740         /* Clear and disable the memory parity error line. */
 741         reason = (reason & 0xf) | 4;
 742         outb(reason, 0x61);
 743 }
 744
 745 static __kprobes void
 746 io_check_error(unsigned char reason, struct pt_regs * regs)
 747 {
 748         printk("NMI: IOCK error (debug interrupt?)\n");
 749         show_registers(regs);
 750
 751         /* Re-enable the IOCK line, wait for a few seconds */
 752         reason = (reason & 0xf) | 8;
 753         outb(reason, 0x61);
 754         mdelay(2000);
 755         reason &= ~8;
 756         outb(reason, 0x61);
 757 }
 758
 759 static __kprobes void
 760 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 761 {
 762         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 763                 reason);
 764         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 765
 766         if (panic_on_unrecovered_nmi)
 767                 panic("NMI: Not continuing");
 768
 769         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 770 }
 771
 772 /* Runs on IST stack. This code must keep interrupts off all the time.
 773    Nested NMIs are prevented by the CPU. */
 774 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 775 {
 776         unsigned char reason = 0;
 777         int cpu;
 778
 779         cpu = smp_processor_id();
 780
 781         /* Only the BSP gets external NMIs from the system.  */
 782         if (!cpu)
 783                 reason = get_nmi_reason();
 784
 785         if (!(reason & 0xc0)) {
 786                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 787                                                                 == NOTIFY_STOP)
 788                         return;
 789                 /*
 790                  * Ok, so this is none of the documented NMI sources,
 791                  * so it must be the NMI watchdog.
 792                  */
 793                 if (nmi_watchdog_tick(regs,reason))
 794                         return;
 795                 if (!do_nmi_callback(regs,cpu))
 796                         unknown_nmi_error(reason, regs);
 797
 798                 return;
 799         }
 800         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 801                 return;
 802
 803         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 804
 805         if (reason & 0x80)
 806                 mem_parity_error(reason, regs);
 807         if (reason & 0x40)
 808                 io_check_error(reason, regs);
 809 }
 810
 811 /* runs on IST stack. */
 812 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 813 {
 814         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 815                 return;
 816         }
 817         preempt_conditional_sti(regs);
 818         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 819         preempt_conditional_cli(regs);
 820 }
 821
 822 /* Help handler running on IST stack to switch back to user stack
 823    for scheduling or signal handling. The actual stack switch is done in
 824    entry.S */
 825 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 826 {
 827         struct pt_regs *regs = eregs;
 828         /* Did already sync */
 829         if (eregs == (struct pt_regs *)eregs->rsp)
 830                 ;
 831         /* Exception from user space */
 832         else if (user_mode(eregs))
 833                 regs = task_pt_regs(current);
 834         /* Exception from kernel and interrupts are enabled. Move to
 835            kernel process stack. */
 836         else if (eregs->eflags & X86_EFLAGS_IF)
 837                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 838         if (eregs != regs)
 839                 *regs = *eregs;
 840         return regs;
 841 }
 842
 843 /* runs on IST stack. */
 844 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 845                                    unsigned long error_code)
 846 {
 847         unsigned long condition;
 848         struct task_struct *tsk = current;
 849         siginfo_t info;
 850
 851         get_debugreg(condition, 6);
 852
 853         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 854                                                 SIGTRAP) == NOTIFY_STOP)
 855                 return;
 856
 857         preempt_conditional_sti(regs);
 858
 859         /* Mask out spurious debug traps due to lazy DR7 setting */
 860         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 861                 if (!tsk->thread.debugreg7) {
 862                         goto clear_dr7;
 863                 }
 864         }
 865
 866         tsk->thread.debugreg6 = condition;
 867
 868         /* Mask out spurious TF errors due to lazy TF clearing */
 869         if (condition & DR_STEP) {
 870                 /*
 871                  * The TF error should be masked out only if the current
 872                  * process is not traced and if the TRAP flag has been set
 873                  * previously by a tracing process (condition detected by
 874                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 875                  * can be modified by the process itself in user mode,
 876                  * allowing programs to debug themselves without the ptrace()
 877                  * interface.
 878                  */
 879                 if (!user_mode(regs))
 880                        goto clear_TF_reenable;
 881                 /*
 882                  * Was the TF flag set by a debugger? If so, clear it now,
 883                  * so that register information is correct.
 884                  */
 885                 if (tsk->ptrace & PT_DTRACE) {
 886                         regs->eflags &= ~TF_MASK;
 887                         tsk->ptrace &= ~PT_DTRACE;
 888                 }
 889         }
 890
 891         /* Ok, finally something we can handle */
 892         tsk->thread.trap_no = 1;
 893         tsk->thread.error_code = error_code;
 894         info.si_signo = SIGTRAP;
 895         info.si_errno = 0;
 896         info.si_code = TRAP_BRKPT;
 897         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 898         force_sig_info(SIGTRAP, &info, tsk);
 899
 900 clear_dr7:
 901         set_debugreg(0UL, 7);
 902         preempt_conditional_cli(regs);
 903         return;
 904
 905 clear_TF_reenable:
 906         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 907         regs->eflags &= ~TF_MASK;
 908         preempt_conditional_cli(regs);
 909 }
 910
 911 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 912 {
 913         const struct exception_table_entry *fixup;
 914         fixup = search_exception_tables(regs->rip);
 915         if (fixup) {
 916                 regs->rip = fixup->fixup;
 917                 return 1;
 918         }
 919         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 920         /* Illegal floating point operation in the kernel */
 921         current->thread.trap_no = trapnr;
 922         die(str, regs, 0);
 923         return 0;
 924 }
 925
 926 /*
 927  * Note that we play around with the 'TS' bit in an attempt to get
 928  * the correct behaviour even in the presence of the asynchronous
 929  * IRQ13 behaviour
 930  */
 931 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 932 {
 933         void __user *rip = (void __user *)(regs->rip);
 934         struct task_struct * task;
 935         siginfo_t info;
 936         unsigned short cwd, swd;
 937
 938         conditional_sti(regs);
 939         if (!user_mode(regs) &&
 940             kernel_math_error(regs, "kernel x87 math error", 16))
 941                 return;
 942
 943         /*
 944          * Save the info for the exception handler and clear the error.
 945          */
 946         task = current;
 947         save_init_fpu(task);
 948         task->thread.trap_no = 16;
 949         task->thread.error_code = 0;
 950         info.si_signo = SIGFPE;
 951         info.si_errno = 0;
 952         info.si_code = __SI_FAULT;
 953         info.si_addr = rip;
 954         /*
 955          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 956          * status.  0x3f is the exception bits in these regs, 0x200 is the
 957          * C1 reg you need in case of a stack fault, 0x040 is the stack
 958          * fault bit.  We should only be taking one exception at a time,
 959          * so if this combination doesn't produce any single exception,
 960          * then we have a bad program that isn't synchronizing its FPU usage
 961          * and it will suffer the consequences since we won't be able to
 962          * fully reproduce the context of the exception
 963          */
 964         cwd = get_fpu_cwd(task);
 965         swd = get_fpu_swd(task);
 966         switch (swd & ~cwd & 0x3f) {
 967                 case 0x000:
 968                 default:
 969                         break;
 970                 case 0x001: /* Invalid Op */
 971                         /*
 972                          * swd & 0x240 == 0x040: Stack Underflow
 973                          * swd & 0x240 == 0x240: Stack Overflow
 974                          * User must clear the SF bit (0x40) if set
 975                          */
 976                         info.si_code = FPE_FLTINV;
 977                         break;
 978                 case 0x002: /* Denormalize */
 979                 case 0x010: /* Underflow */
 980                         info.si_code = FPE_FLTUND;
 981                         break;
 982                 case 0x004: /* Zero Divide */
 983                         info.si_code = FPE_FLTDIV;
 984                         break;
 985                 case 0x008: /* Overflow */
 986                         info.si_code = FPE_FLTOVF;
 987                         break;
 988                 case 0x020: /* Precision */
 989                         info.si_code = FPE_FLTRES;
 990                         break;
 991         }
 992         force_sig_info(SIGFPE, &info, task);
 993 }
 994
 995 asmlinkage void bad_intr(void)
 996 {
 997         printk("bad interrupt");
 998 }
 999
1000 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1001 {
1002         void __user *rip = (void __user *)(regs->rip);
1003         struct task_struct * task;
1004         siginfo_t info;
1005         unsigned short mxcsr;
1006
1007         conditional_sti(regs);
1008         if (!user_mode(regs) &&
1009                 kernel_math_error(regs, "kernel simd math error", 19))
1010                 return;
1011
1012         /*
1013          * Save the info for the exception handler and clear the error.
1014          */
1015         task = current;
1016         save_init_fpu(task);
1017         task->thread.trap_no = 19;
1018         task->thread.error_code = 0;
1019         info.si_signo = SIGFPE;
1020         info.si_errno = 0;
1021         info.si_code = __SI_FAULT;
1022         info.si_addr = rip;
1023         /*
1024          * The SIMD FPU exceptions are handled a little differently, as there
1025          * is only a single status/control register.  Thus, to determine which
1026          * unmasked exception was caught we must mask the exception mask bits
1027          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1028          */
1029         mxcsr = get_fpu_mxcsr(task);
1030         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1031                 case 0x000:
1032                 default:
1033                         break;
1034                 case 0x001: /* Invalid Op */
1035                         info.si_code = FPE_FLTINV;
1036                         break;
1037                 case 0x002: /* Denormalize */
1038                 case 0x010: /* Underflow */
1039                         info.si_code = FPE_FLTUND;
1040                         break;
1041                 case 0x004: /* Zero Divide */
1042                         info.si_code = FPE_FLTDIV;
1043                         break;
1044                 case 0x008: /* Overflow */
1045                         info.si_code = FPE_FLTOVF;
1046                         break;
1047                 case 0x020: /* Precision */
1048                         info.si_code = FPE_FLTRES;
1049                         break;
1050         }
1051         force_sig_info(SIGFPE, &info, task);
1052 }
1053
1054 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1055 {
1056 }
1057
1058 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1059 {
1060 }
1061
1062 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1063 {
1064 }
1065
1066 /*
1067  *  'math_state_restore()' saves the current math information in the
1068  * old math state array, and gets the new ones from the current task
1069  *
1070  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1071  * Don't touch unless you *really* know how it works.
1072  */
1073 asmlinkage void math_state_restore(void)
1074 {
1075         struct task_struct *me = current;
1076         clts();                 /* Allow maths ops (or we recurse) */
1077
1078         if (!used_math())
1079                 init_fpu(me);
1080         restore_fpu_checking(&me->thread.i387.fxsave);
1081         task_thread_info(me)->status |= TS_USEDFPU;
1082         me->fpu_counter++;
1083 }
1084
1085 void __init trap_init(void)
1086 {
1087         set_intr_gate(0,&divide_error);
1088         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1089         set_intr_gate_ist(2,&nmi,NMI_STACK);
1090         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1091         set_system_gate(4,&overflow);   /* int4 can be called from all */
1092         set_intr_gate(5,&bounds);
1093         set_intr_gate(6,&invalid_op);
1094         set_intr_gate(7,&device_not_available);
1095         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1096         set_intr_gate(9,&coprocessor_segment_overrun);
1097         set_intr_gate(10,&invalid_TSS);
1098         set_intr_gate(11,&segment_not_present);
1099         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1100         set_intr_gate(13,&general_protection);
1101         set_intr_gate(14,&page_fault);
1102         set_intr_gate(15,&spurious_interrupt_bug);
1103         set_intr_gate(16,&coprocessor_error);
1104         set_intr_gate(17,&alignment_check);
1105 #ifdef CONFIG_X86_MCE
1106         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1107 #endif
1108         set_intr_gate(19,&simd_coprocessor_error);
1109
1110 #ifdef CONFIG_IA32_EMULATION
1111         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1112 #endif
1113
1114         /*
1115          * Should be a barrier for any external CPU state.
1116          */
1117         cpu_init();
1118 }
1119
1120
1121 static int __init oops_setup(char *s)
1122 {
1123         if (!s)
1124                 return -EINVAL;
1125         if (!strcmp(s, "panic"))
1126                 panic_on_oops = 1;
1127         return 0;
1128 }
1129 early_param("oops", oops_setup);
1130
1131 static int __init kstack_setup(char *s)
1132 {
1133         if (!s)
1134                 return -EINVAL;
1135         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1136         return 0;
1137 }
1138 early_param("kstack", kstack_setup);