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
  34 #include <asm/system.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/io.h>
  37 #include <asm/atomic.h>
  38 #include <asm/debugreg.h>
  39 #include <asm/desc.h>
  40 #include <asm/i387.h>
  41 #include <asm/kdebug.h>
  42 #include <asm/processor.h>
  43 #include <asm/unwind.h>
  44 #include <asm/smp.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/pda.h>
  47 #include <asm/proto.h>
  48 #include <asm/nmi.h>
  49 #include <asm/stacktrace.h>
  50
  51 asmlinkage void divide_error(void);
  52 asmlinkage void debug(void);
  53 asmlinkage void nmi(void);
  54 asmlinkage void int3(void);
  55 asmlinkage void overflow(void);
  56 asmlinkage void bounds(void);
  57 asmlinkage void invalid_op(void);
  58 asmlinkage void device_not_available(void);
  59 asmlinkage void double_fault(void);
  60 asmlinkage void coprocessor_segment_overrun(void);
  61 asmlinkage void invalid_TSS(void);
  62 asmlinkage void segment_not_present(void);
  63 asmlinkage void stack_segment(void);
  64 asmlinkage void general_protection(void);
  65 asmlinkage void page_fault(void);
  66 asmlinkage void coprocessor_error(void);
  67 asmlinkage void simd_coprocessor_error(void);
  68 asmlinkage void reserved(void);
  69 asmlinkage void alignment_check(void);
  70 asmlinkage void machine_check(void);
  71 asmlinkage void spurious_interrupt_bug(void);
  72
  73 ATOMIC_NOTIFIER_HEAD(die_chain);
  74 EXPORT_SYMBOL(die_chain);
  75
  76 int register_die_notifier(struct notifier_block *nb)
  77 {
  78         vmalloc_sync_all();
  79         return atomic_notifier_chain_register(&die_chain, nb);
  80 }
  81 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
  82
  83 int unregister_die_notifier(struct notifier_block *nb)
  84 {
  85         return atomic_notifier_chain_unregister(&die_chain, nb);
  86 }
  87 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
  88
  89 static inline void conditional_sti(struct pt_regs *regs)
  90 {
  91         if (regs->eflags & X86_EFLAGS_IF)
  92                 local_irq_enable();
  93 }
  94
  95 static inline void preempt_conditional_sti(struct pt_regs *regs)
  96 {
  97         preempt_disable();
  98         if (regs->eflags & X86_EFLAGS_IF)
  99                 local_irq_enable();
 100 }
 101
 102 static inline void preempt_conditional_cli(struct pt_regs *regs)
 103 {
 104         if (regs->eflags & X86_EFLAGS_IF)
 105                 local_irq_disable();
 106         /* Make sure to not schedule here because we could be running
 107            on an exception stack. */
 108         preempt_enable_no_resched();
 109 }
 110
 111 static int kstack_depth_to_print = 12;
 112 #ifdef CONFIG_STACK_UNWIND
 113 static int call_trace = 1;
 114 #else
 115 #define call_trace (-1)
 116 #endif
 117
 118 #ifdef CONFIG_KALLSYMS
 119 void printk_address(unsigned long address)
 120 {
 121         unsigned long offset = 0, symsize;
 122         const char *symname;
 123         char *modname;
 124         char *delim = ":";
 125         char namebuf[128];
 126
 127         symname = kallsyms_lookup(address, &symsize, &offset,
 128                                         &modname, namebuf);
 129         if (!symname) {
 130                 printk(" [<%016lx>]\n", address);
 131                 return;
 132         }
 133         if (!modname)
 134                 modname = delim = "";
 135         printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
 136                 address, delim, modname, delim, symname, offset, symsize);
 137 }
 138 #else
 139 void printk_address(unsigned long address)
 140 {
 141         printk(" [<%016lx>]\n", address);
 142 }
 143 #endif
 144
 145 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 146                                         unsigned *usedp, char **idp)
 147 {
 148         static char ids[][8] = {
 149                 [DEBUG_STACK - 1] = "#DB",
 150                 [NMI_STACK - 1] = "NMI",
 151                 [DOUBLEFAULT_STACK - 1] = "#DF",
 152                 [STACKFAULT_STACK - 1] = "#SS",
 153                 [MCE_STACK - 1] = "#MC",
 154 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 155                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 156 #endif
 157         };
 158         unsigned k;
 159
 160         /*
 161          * Iterate over all exception stacks, and figure out whether
 162          * 'stack' is in one of them:
 163          */
 164         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 165                 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
 166                 /*
 167                  * Is 'stack' above this exception frame's end?
 168                  * If yes then skip to the next frame.
 169                  */
 170                 if (stack >= end)
 171                         continue;
 172                 /*
 173                  * Is 'stack' above this exception frame's start address?
 174                  * If yes then we found the right frame.
 175                  */
 176                 if (stack >= end - EXCEPTION_STKSZ) {
 177                         /*
 178                          * Make sure we only iterate through an exception
 179                          * stack once. If it comes up for the second time
 180                          * then there's something wrong going on - just
 181                          * break out and return NULL:
 182                          */
 183                         if (*usedp & (1U << k))
 184                                 break;
 185                         *usedp |= 1U << k;
 186                         *idp = ids[k];
 187                         return (unsigned long *)end;
 188                 }
 189                 /*
 190                  * If this is a debug stack, and if it has a larger size than
 191                  * the usual exception stacks, then 'stack' might still
 192                  * be within the lower portion of the debug stack:
 193                  */
 194 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 195                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 196                         unsigned j = N_EXCEPTION_STACKS - 1;
 197
 198                         /*
 199                          * Black magic. A large debug stack is composed of
 200                          * multiple exception stack entries, which we
 201                          * iterate through now. Dont look:
 202                          */
 203                         do {
 204                                 ++j;
 205                                 end -= EXCEPTION_STKSZ;
 206                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 207                         } while (stack < end - EXCEPTION_STKSZ);
 208                         if (*usedp & (1U << j))
 209                                 break;
 210                         *usedp |= 1U << j;
 211                         *idp = ids[j];
 212                         return (unsigned long *)end;
 213                 }
 214 #endif
 215         }
 216         return NULL;
 217 }
 218
 219 struct ops_and_data {
 220         struct stacktrace_ops *ops;
 221         void *data;
 222 };
 223
 224 static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
 225 {
 226         struct ops_and_data *oad = (struct ops_and_data *)context;
 227         int n = 0;
 228
 229         while (unwind(info) == 0 && UNW_PC(info)) {
 230                 n++;
 231                 oad->ops->address(oad->data, UNW_PC(info));
 232                 if (arch_unw_user_mode(info))
 233                         break;
 234         }
 235         return n;
 236 }
 237
 238 /*
 239  * x86-64 can have upto three kernel stacks:
 240  * process stack
 241  * interrupt stack
 242  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 243  */
 244
 245 void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
 246                 struct stacktrace_ops *ops, void *data)
 247 {
 248         const unsigned cpu = smp_processor_id();
 249         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 250         unsigned used = 0;
 251
 252         if (!tsk)
 253                 tsk = current;
 254
 255         if (call_trace >= 0) {
 256                 int unw_ret = 0;
 257                 struct unwind_frame_info info;
 258                 struct ops_and_data oad = { .ops = ops, .data = data };
 259
 260                 if (regs) {
 261                         if (unwind_init_frame_info(&info, tsk, regs) == 0)
 262                                 unw_ret = dump_trace_unwind(&info, &oad);
 263                 } else if (tsk == current)
 264                         unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
 265                 else {
 266                         if (unwind_init_blocked(&info, tsk) == 0)
 267                                 unw_ret = dump_trace_unwind(&info, &oad);
 268                 }
 269                 if (unw_ret > 0) {
 270                         if (call_trace == 1 && !arch_unw_user_mode(&info)) {
 271                                 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
 272                                              UNW_PC(&info));
 273                                 if ((long)UNW_SP(&info) < 0) {
 274                                         ops->warning(data, "Leftover inexact backtrace:\n");
 275                                         stack = (unsigned long *)UNW_SP(&info);
 276                                         if (!stack)
 277                                                 return;
 278                                 } else
 279                                         ops->warning(data, "Full inexact backtrace again:\n");
 280                         } else if (call_trace >= 1)
 281                                 return;
 282                         else
 283                                 ops->warning(data, "Full inexact backtrace again:\n");
 284                 } else
 285                         ops->warning(data, "Inexact backtrace:\n");
 286         }
 287         if (!stack) {
 288                 unsigned long dummy;
 289                 stack = &dummy;
 290                 if (tsk && tsk != current)
 291                         stack = (unsigned long *)tsk->thread.rsp;
 292         }
 293         /*
 294          * Align the stack pointer on word boundary, later loops
 295          * rely on that (and corruption / debug info bugs can cause
 296          * unaligned values here):
 297          */
 298         stack = (unsigned long *)((unsigned long)stack & ~(sizeof(long)-1));
 299
 300         /*
 301          * Print function call entries within a stack. 'cond' is the
 302          * "end of stackframe" condition, that the 'stack++'
 303          * iteration will eventually trigger.
 304          */
 305 #define HANDLE_STACK(cond) \
 306         do while (cond) { \
 307                 unsigned long addr = *stack++; \
 308                 if (oops_in_progress ?          \
 309                         __kernel_text_address(addr) : \
 310                         kernel_text_address(addr)) { \
 311                         /* \
 312                          * If the address is either in the text segment of the \
 313                          * kernel, or in the region which contains vmalloc'ed \
 314                          * memory, it *may* be the address of a calling \
 315                          * routine; if so, print it so that someone tracing \
 316                          * down the cause of the crash will be able to figure \
 317                          * out the call path that was taken. \
 318                          */ \
 319                         ops->address(data, addr);   \
 320                 } \
 321         } while (0)
 322
 323         /*
 324          * Print function call entries in all stacks, starting at the
 325          * current stack address. If the stacks consist of nested
 326          * exceptions
 327          */
 328         for (;;) {
 329                 char *id;
 330                 unsigned long *estack_end;
 331                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 332                                                 &used, &id);
 333
 334                 if (estack_end) {
 335                         if (ops->stack(data, id) < 0)
 336                                 break;
 337                         HANDLE_STACK (stack < estack_end);
 338                         ops->stack(data, "<EOE>");
 339                         /*
 340                          * We link to the next stack via the
 341                          * second-to-last pointer (index -2 to end) in the
 342                          * exception stack:
 343                          */
 344                         stack = (unsigned long *) estack_end[-2];
 345                         continue;
 346                 }
 347                 if (irqstack_end) {
 348                         unsigned long *irqstack;
 349                         irqstack = irqstack_end -
 350                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 351
 352                         if (stack >= irqstack && stack < irqstack_end) {
 353                                 if (ops->stack(data, "IRQ") < 0)
 354                                         break;
 355                                 HANDLE_STACK (stack < irqstack_end);
 356                                 /*
 357                                  * We link to the next stack (which would be
 358                                  * the process stack normally) the last
 359                                  * pointer (index -1 to end) in the IRQ stack:
 360                                  */
 361                                 stack = (unsigned long *) (irqstack_end[-1]);
 362                                 irqstack_end = NULL;
 363                                 ops->stack(data, "EOI");
 364                                 continue;
 365                         }
 366                 }
 367                 break;
 368         }
 369
 370         /*
 371          * This handles the process stack:
 372          */
 373         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 374 #undef HANDLE_STACK
 375 }
 376 EXPORT_SYMBOL(dump_trace);
 377
 378 static void
 379 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 380 {
 381         print_symbol(msg, symbol);
 382         printk("\n");
 383 }
 384
 385 static void print_trace_warning(void *data, char *msg)
 386 {
 387         printk("%s\n", msg);
 388 }
 389
 390 static int print_trace_stack(void *data, char *name)
 391 {
 392         printk(" <%s> ", name);
 393         return 0;
 394 }
 395
 396 static void print_trace_address(void *data, unsigned long addr)
 397 {
 398         printk_address(addr);
 399 }
 400
 401 static struct stacktrace_ops print_trace_ops = {
 402         .warning = print_trace_warning,
 403         .warning_symbol = print_trace_warning_symbol,
 404         .stack = print_trace_stack,
 405         .address = print_trace_address,
 406 };
 407
 408 void
 409 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 410 {
 411         printk("\nCall Trace:\n");
 412         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 413         printk("\n");
 414 }
 415
 416 static void
 417 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 418 {
 419         unsigned long *stack;
 420         int i;
 421         const int cpu = smp_processor_id();
 422         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 423         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 424
 425         // debugging aid: "show_stack(NULL, NULL);" prints the
 426         // back trace for this cpu.
 427
 428         if (rsp == NULL) {
 429                 if (tsk)
 430                         rsp = (unsigned long *)tsk->thread.rsp;
 431                 else
 432                         rsp = (unsigned long *)&rsp;
 433         }
 434
 435         stack = rsp;
 436         for(i=0; i < kstack_depth_to_print; i++) {
 437                 if (stack >= irqstack && stack <= irqstack_end) {
 438                         if (stack == irqstack_end) {
 439                                 stack = (unsigned long *) (irqstack_end[-1]);
 440                                 printk(" <EOI> ");
 441                         }
 442                 } else {
 443                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 444                         break;
 445                 }
 446                 if (i && ((i % 4) == 0))
 447                         printk("\n");
 448                 printk(" %016lx", *stack++);
 449                 touch_nmi_watchdog();
 450         }
 451         show_trace(tsk, regs, rsp);
 452 }
 453
 454 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 455 {
 456         _show_stack(tsk, NULL, rsp);
 457 }
 458
 459 /*
 460  * The architecture-independent dump_stack generator
 461  */
 462 void dump_stack(void)
 463 {
 464         unsigned long dummy;
 465         show_trace(NULL, NULL, &dummy);
 466 }
 467
 468 EXPORT_SYMBOL(dump_stack);
 469
 470 void show_registers(struct pt_regs *regs)
 471 {
 472         int i;
 473         int in_kernel = !user_mode(regs);
 474         unsigned long rsp;
 475         const int cpu = smp_processor_id();
 476         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 477
 478                 rsp = regs->rsp;
 479
 480         printk("CPU %d ", cpu);
 481         __show_regs(regs);
 482         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 483                 cur->comm, cur->pid, task_thread_info(cur), cur);
 484
 485         /*
 486          * When in-kernel, we also print out the stack and code at the
 487          * time of the fault..
 488          */
 489         if (in_kernel) {
 490
 491                 printk("Stack: ");
 492                 _show_stack(NULL, regs, (unsigned long*)rsp);
 493
 494                 printk("\nCode: ");
 495                 if (regs->rip < PAGE_OFFSET)
 496                         goto bad;
 497
 498                 for (i=0; i<20; i++) {
 499                         unsigned char c;
 500                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 501 bad:
 502                                 printk(" Bad RIP value.");
 503                                 break;
 504                         }
 505                         printk("%02x ", c);
 506                 }
 507         }
 508         printk("\n");
 509 }
 510
 511 void handle_BUG(struct pt_regs *regs)
 512 {
 513         struct bug_frame f;
 514         long len;
 515         const char *prefix = "";
 516
 517         if (user_mode(regs))
 518                 return;
 519         if (__copy_from_user(&f, (const void __user *) regs->rip,
 520                              sizeof(struct bug_frame)))
 521                 return;
 522         if (f.filename >= 0 ||
 523             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 524                 return;
 525         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 526         if (len < 0 || len >= PATH_MAX)
 527                 f.filename = (int)(long)"unmapped filename";
 528         else if (len > 50) {
 529                 f.filename += len - 50;
 530                 prefix = "...";
 531         }
 532         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 533         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 534 }
 535
 536 #ifdef CONFIG_BUG
 537 void out_of_line_bug(void)
 538 {
 539         BUG();
 540 }
 541 EXPORT_SYMBOL(out_of_line_bug);
 542 #endif
 543
 544 static DEFINE_SPINLOCK(die_lock);
 545 static int die_owner = -1;
 546 static unsigned int die_nest_count;
 547
 548 unsigned __kprobes long oops_begin(void)
 549 {
 550         int cpu = smp_processor_id();
 551         unsigned long flags;
 552
 553         oops_enter();
 554
 555         /* racy, but better than risking deadlock. */
 556         local_irq_save(flags);
 557         if (!spin_trylock(&die_lock)) {
 558                 if (cpu == die_owner)
 559                         /* nested oops. should stop eventually */;
 560                 else
 561                         spin_lock(&die_lock);
 562         }
 563         die_nest_count++;
 564         die_owner = cpu;
 565         console_verbose();
 566         bust_spinlocks(1);
 567         return flags;
 568 }
 569
 570 void __kprobes oops_end(unsigned long flags)
 571 {
 572         die_owner = -1;
 573         bust_spinlocks(0);
 574         die_nest_count--;
 575         if (die_nest_count)
 576                 /* We still own the lock */
 577                 local_irq_restore(flags);
 578         else
 579                 /* Nest count reaches zero, release the lock. */
 580                 spin_unlock_irqrestore(&die_lock, flags);
 581         if (panic_on_oops)
 582                 panic("Fatal exception");
 583         oops_exit();
 584 }
 585
 586 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 587 {
 588         static int die_counter;
 589         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 590 #ifdef CONFIG_PREEMPT
 591         printk("PREEMPT ");
 592 #endif
 593 #ifdef CONFIG_SMP
 594         printk("SMP ");
 595 #endif
 596 #ifdef CONFIG_DEBUG_PAGEALLOC
 597         printk("DEBUG_PAGEALLOC");
 598 #endif
 599         printk("\n");
 600         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 601         show_registers(regs);
 602         /* Executive summary in case the oops scrolled away */
 603         printk(KERN_ALERT "RIP ");
 604         printk_address(regs->rip);
 605         printk(" RSP <%016lx>\n", regs->rsp);
 606         if (kexec_should_crash(current))
 607                 crash_kexec(regs);
 608 }
 609
 610 void die(const char * str, struct pt_regs * regs, long err)
 611 {
 612         unsigned long flags = oops_begin();
 613
 614         handle_BUG(regs);
 615         __die(str, regs, err);
 616         oops_end(flags);
 617         do_exit(SIGSEGV);
 618 }
 619
 620 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 621 {
 622         unsigned long flags = oops_begin();
 623
 624         /*
 625          * We are in trouble anyway, lets at least try
 626          * to get a message out.
 627          */
 628         printk(str, smp_processor_id());
 629         show_registers(regs);
 630         if (kexec_should_crash(current))
 631                 crash_kexec(regs);
 632         if (do_panic || panic_on_oops)
 633                 panic("Non maskable interrupt");
 634         oops_end(flags);
 635         nmi_exit();
 636         local_irq_enable();
 637         do_exit(SIGSEGV);
 638 }
 639
 640 static void __kprobes do_trap(int trapnr, int signr, char *str,
 641                               struct pt_regs * regs, long error_code,
 642                               siginfo_t *info)
 643 {
 644         struct task_struct *tsk = current;
 645
 646         tsk->thread.error_code = error_code;
 647         tsk->thread.trap_no = trapnr;
 648
 649         if (user_mode(regs)) {
 650                 if (exception_trace && unhandled_signal(tsk, signr))
 651                         printk(KERN_INFO
 652                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 653                                tsk->comm, tsk->pid, str,
 654                                regs->rip, regs->rsp, error_code);
 655
 656                 if (info)
 657                         force_sig_info(signr, info, tsk);
 658                 else
 659                         force_sig(signr, tsk);
 660                 return;
 661         }
 662
 663
 664         /* kernel trap */
 665         {
 666                 const struct exception_table_entry *fixup;
 667                 fixup = search_exception_tables(regs->rip);
 668                 if (fixup)
 669                         regs->rip = fixup->fixup;
 670                 else
 671                         die(str, regs, error_code);
 672                 return;
 673         }
 674 }
 675
 676 #define DO_ERROR(trapnr, signr, str, name) \
 677 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 678 { \
 679         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 680                                                         == NOTIFY_STOP) \
 681                 return; \
 682         conditional_sti(regs);                                          \
 683         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 684 }
 685
 686 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 687 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 688 { \
 689         siginfo_t info; \
 690         info.si_signo = signr; \
 691         info.si_errno = 0; \
 692         info.si_code = sicode; \
 693         info.si_addr = (void __user *)siaddr; \
 694         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 695                                                         == NOTIFY_STOP) \
 696                 return; \
 697         conditional_sti(regs);                                          \
 698         do_trap(trapnr, signr, str, regs, error_code, &info); \
 699 }
 700
 701 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 702 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 703 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 704 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 705 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 706 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 707 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 708 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 709 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 710 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 711
 712 /* Runs on IST stack */
 713 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 714 {
 715         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 716                         12, SIGBUS) == NOTIFY_STOP)
 717                 return;
 718         preempt_conditional_sti(regs);
 719         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 720         preempt_conditional_cli(regs);
 721 }
 722
 723 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 724 {
 725         static const char str[] = "double fault";
 726         struct task_struct *tsk = current;
 727
 728         /* Return not checked because double check cannot be ignored */
 729         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 730
 731         tsk->thread.error_code = error_code;
 732         tsk->thread.trap_no = 8;
 733
 734         /* This is always a kernel trap and never fixable (and thus must
 735            never return). */
 736         for (;;)
 737                 die(str, regs, error_code);
 738 }
 739
 740 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 741                                                 long error_code)
 742 {
 743         struct task_struct *tsk = current;
 744
 745         conditional_sti(regs);
 746
 747         tsk->thread.error_code = error_code;
 748         tsk->thread.trap_no = 13;
 749
 750         if (user_mode(regs)) {
 751                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 752                         printk(KERN_INFO
 753                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 754                                tsk->comm, tsk->pid,
 755                                regs->rip, regs->rsp, error_code);
 756
 757                 force_sig(SIGSEGV, tsk);
 758                 return;
 759         }
 760
 761         /* kernel gp */
 762         {
 763                 const struct exception_table_entry *fixup;
 764                 fixup = search_exception_tables(regs->rip);
 765                 if (fixup) {
 766                         regs->rip = fixup->fixup;
 767                         return;
 768                 }
 769                 if (notify_die(DIE_GPF, "general protection fault", regs,
 770                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 771                         return;
 772                 die("general protection fault", regs, error_code);
 773         }
 774 }
 775
 776 static __kprobes void
 777 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 778 {
 779         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 780                 reason);
 781         printk(KERN_EMERG "You probably have a hardware problem with your "
 782                 "RAM chips\n");
 783
 784         if (panic_on_unrecovered_nmi)
 785                 panic("NMI: Not continuing");
 786
 787         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 788
 789         /* Clear and disable the memory parity error line. */
 790         reason = (reason & 0xf) | 4;
 791         outb(reason, 0x61);
 792 }
 793
 794 static __kprobes void
 795 io_check_error(unsigned char reason, struct pt_regs * regs)
 796 {
 797         printk("NMI: IOCK error (debug interrupt?)\n");
 798         show_registers(regs);
 799
 800         /* Re-enable the IOCK line, wait for a few seconds */
 801         reason = (reason & 0xf) | 8;
 802         outb(reason, 0x61);
 803         mdelay(2000);
 804         reason &= ~8;
 805         outb(reason, 0x61);
 806 }
 807
 808 static __kprobes void
 809 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 810 {
 811         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 812                 reason);
 813         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 814
 815         if (panic_on_unrecovered_nmi)
 816                 panic("NMI: Not continuing");
 817
 818         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 819 }
 820
 821 /* Runs on IST stack. This code must keep interrupts off all the time.
 822    Nested NMIs are prevented by the CPU. */
 823 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 824 {
 825         unsigned char reason = 0;
 826         int cpu;
 827
 828         cpu = smp_processor_id();
 829
 830         /* Only the BSP gets external NMIs from the system.  */
 831         if (!cpu)
 832                 reason = get_nmi_reason();
 833
 834         if (!(reason & 0xc0)) {
 835                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 836                                                                 == NOTIFY_STOP)
 837                         return;
 838                 /*
 839                  * Ok, so this is none of the documented NMI sources,
 840                  * so it must be the NMI watchdog.
 841                  */
 842                 if (nmi_watchdog_tick(regs,reason))
 843                         return;
 844                 if (!do_nmi_callback(regs,cpu))
 845                         unknown_nmi_error(reason, regs);
 846
 847                 return;
 848         }
 849         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 850                 return;
 851
 852         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 853
 854         if (reason & 0x80)
 855                 mem_parity_error(reason, regs);
 856         if (reason & 0x40)
 857                 io_check_error(reason, regs);
 858 }
 859
 860 /* runs on IST stack. */
 861 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 862 {
 863         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 864                 return;
 865         }
 866         preempt_conditional_sti(regs);
 867         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 868         preempt_conditional_cli(regs);
 869 }
 870
 871 /* Help handler running on IST stack to switch back to user stack
 872    for scheduling or signal handling. The actual stack switch is done in
 873    entry.S */
 874 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 875 {
 876         struct pt_regs *regs = eregs;
 877         /* Did already sync */
 878         if (eregs == (struct pt_regs *)eregs->rsp)
 879                 ;
 880         /* Exception from user space */
 881         else if (user_mode(eregs))
 882                 regs = task_pt_regs(current);
 883         /* Exception from kernel and interrupts are enabled. Move to
 884            kernel process stack. */
 885         else if (eregs->eflags & X86_EFLAGS_IF)
 886                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 887         if (eregs != regs)
 888                 *regs = *eregs;
 889         return regs;
 890 }
 891
 892 /* runs on IST stack. */
 893 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 894                                    unsigned long error_code)
 895 {
 896         unsigned long condition;
 897         struct task_struct *tsk = current;
 898         siginfo_t info;
 899
 900         get_debugreg(condition, 6);
 901
 902         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 903                                                 SIGTRAP) == NOTIFY_STOP)
 904                 return;
 905
 906         preempt_conditional_sti(regs);
 907
 908         /* Mask out spurious debug traps due to lazy DR7 setting */
 909         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 910                 if (!tsk->thread.debugreg7) {
 911                         goto clear_dr7;
 912                 }
 913         }
 914
 915         tsk->thread.debugreg6 = condition;
 916
 917         /* Mask out spurious TF errors due to lazy TF clearing */
 918         if (condition & DR_STEP) {
 919                 /*
 920                  * The TF error should be masked out only if the current
 921                  * process is not traced and if the TRAP flag has been set
 922                  * previously by a tracing process (condition detected by
 923                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 924                  * can be modified by the process itself in user mode,
 925                  * allowing programs to debug themselves without the ptrace()
 926                  * interface.
 927                  */
 928                 if (!user_mode(regs))
 929                        goto clear_TF_reenable;
 930                 /*
 931                  * Was the TF flag set by a debugger? If so, clear it now,
 932                  * so that register information is correct.
 933                  */
 934                 if (tsk->ptrace & PT_DTRACE) {
 935                         regs->eflags &= ~TF_MASK;
 936                         tsk->ptrace &= ~PT_DTRACE;
 937                 }
 938         }
 939
 940         /* Ok, finally something we can handle */
 941         tsk->thread.trap_no = 1;
 942         tsk->thread.error_code = error_code;
 943         info.si_signo = SIGTRAP;
 944         info.si_errno = 0;
 945         info.si_code = TRAP_BRKPT;
 946         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 947         force_sig_info(SIGTRAP, &info, tsk);
 948
 949 clear_dr7:
 950         set_debugreg(0UL, 7);
 951         preempt_conditional_cli(regs);
 952         return;
 953
 954 clear_TF_reenable:
 955         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 956         regs->eflags &= ~TF_MASK;
 957         preempt_conditional_cli(regs);
 958 }
 959
 960 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 961 {
 962         const struct exception_table_entry *fixup;
 963         fixup = search_exception_tables(regs->rip);
 964         if (fixup) {
 965                 regs->rip = fixup->fixup;
 966                 return 1;
 967         }
 968         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 969         /* Illegal floating point operation in the kernel */
 970         current->thread.trap_no = trapnr;
 971         die(str, regs, 0);
 972         return 0;
 973 }
 974
 975 /*
 976  * Note that we play around with the 'TS' bit in an attempt to get
 977  * the correct behaviour even in the presence of the asynchronous
 978  * IRQ13 behaviour
 979  */
 980 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 981 {
 982         void __user *rip = (void __user *)(regs->rip);
 983         struct task_struct * task;
 984         siginfo_t info;
 985         unsigned short cwd, swd;
 986
 987         conditional_sti(regs);
 988         if (!user_mode(regs) &&
 989             kernel_math_error(regs, "kernel x87 math error", 16))
 990                 return;
 991
 992         /*
 993          * Save the info for the exception handler and clear the error.
 994          */
 995         task = current;
 996         save_init_fpu(task);
 997         task->thread.trap_no = 16;
 998         task->thread.error_code = 0;
 999         info.si_signo = SIGFPE;
1000         info.si_errno = 0;
1001         info.si_code = __SI_FAULT;
1002         info.si_addr = rip;
1003         /*
1004          * (~cwd & swd) will mask out exceptions that are not set to unmasked
1005          * status.  0x3f is the exception bits in these regs, 0x200 is the
1006          * C1 reg you need in case of a stack fault, 0x040 is the stack
1007          * fault bit.  We should only be taking one exception at a time,
1008          * so if this combination doesn't produce any single exception,
1009          * then we have a bad program that isn't synchronizing its FPU usage
1010          * and it will suffer the consequences since we won't be able to
1011          * fully reproduce the context of the exception
1012          */
1013         cwd = get_fpu_cwd(task);
1014         swd = get_fpu_swd(task);
1015         switch (swd & ~cwd & 0x3f) {
1016                 case 0x000:
1017                 default:
1018                         break;
1019                 case 0x001: /* Invalid Op */
1020                         /*
1021                          * swd & 0x240 == 0x040: Stack Underflow
1022                          * swd & 0x240 == 0x240: Stack Overflow
1023                          * User must clear the SF bit (0x40) if set
1024                          */
1025                         info.si_code = FPE_FLTINV;
1026                         break;
1027                 case 0x002: /* Denormalize */
1028                 case 0x010: /* Underflow */
1029                         info.si_code = FPE_FLTUND;
1030                         break;
1031                 case 0x004: /* Zero Divide */
1032                         info.si_code = FPE_FLTDIV;
1033                         break;
1034                 case 0x008: /* Overflow */
1035                         info.si_code = FPE_FLTOVF;
1036                         break;
1037                 case 0x020: /* Precision */
1038                         info.si_code = FPE_FLTRES;
1039                         break;
1040         }
1041         force_sig_info(SIGFPE, &info, task);
1042 }
1043
1044 asmlinkage void bad_intr(void)
1045 {
1046         printk("bad interrupt");
1047 }
1048
1049 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1050 {
1051         void __user *rip = (void __user *)(regs->rip);
1052         struct task_struct * task;
1053         siginfo_t info;
1054         unsigned short mxcsr;
1055
1056         conditional_sti(regs);
1057         if (!user_mode(regs) &&
1058                 kernel_math_error(regs, "kernel simd math error", 19))
1059                 return;
1060
1061         /*
1062          * Save the info for the exception handler and clear the error.
1063          */
1064         task = current;
1065         save_init_fpu(task);
1066         task->thread.trap_no = 19;
1067         task->thread.error_code = 0;
1068         info.si_signo = SIGFPE;
1069         info.si_errno = 0;
1070         info.si_code = __SI_FAULT;
1071         info.si_addr = rip;
1072         /*
1073          * The SIMD FPU exceptions are handled a little differently, as there
1074          * is only a single status/control register.  Thus, to determine which
1075          * unmasked exception was caught we must mask the exception mask bits
1076          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1077          */
1078         mxcsr = get_fpu_mxcsr(task);
1079         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1080                 case 0x000:
1081                 default:
1082                         break;
1083                 case 0x001: /* Invalid Op */
1084                         info.si_code = FPE_FLTINV;
1085                         break;
1086                 case 0x002: /* Denormalize */
1087                 case 0x010: /* Underflow */
1088                         info.si_code = FPE_FLTUND;
1089                         break;
1090                 case 0x004: /* Zero Divide */
1091                         info.si_code = FPE_FLTDIV;
1092                         break;
1093                 case 0x008: /* Overflow */
1094                         info.si_code = FPE_FLTOVF;
1095                         break;
1096                 case 0x020: /* Precision */
1097                         info.si_code = FPE_FLTRES;
1098                         break;
1099         }
1100         force_sig_info(SIGFPE, &info, task);
1101 }
1102
1103 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1104 {
1105 }
1106
1107 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1108 {
1109 }
1110
1111 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1112 {
1113 }
1114
1115 /*
1116  *  'math_state_restore()' saves the current math information in the
1117  * old math state array, and gets the new ones from the current task
1118  *
1119  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1120  * Don't touch unless you *really* know how it works.
1121  */
1122 asmlinkage void math_state_restore(void)
1123 {
1124         struct task_struct *me = current;
1125         clts();                 /* Allow maths ops (or we recurse) */
1126
1127         if (!used_math())
1128                 init_fpu(me);
1129         restore_fpu_checking(&me->thread.i387.fxsave);
1130         task_thread_info(me)->status |= TS_USEDFPU;
1131         me->fpu_counter++;
1132 }
1133
1134 void __init trap_init(void)
1135 {
1136         set_intr_gate(0,&divide_error);
1137         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1138         set_intr_gate_ist(2,&nmi,NMI_STACK);
1139         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1140         set_system_gate(4,&overflow);   /* int4 can be called from all */
1141         set_intr_gate(5,&bounds);
1142         set_intr_gate(6,&invalid_op);
1143         set_intr_gate(7,&device_not_available);
1144         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1145         set_intr_gate(9,&coprocessor_segment_overrun);
1146         set_intr_gate(10,&invalid_TSS);
1147         set_intr_gate(11,&segment_not_present);
1148         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1149         set_intr_gate(13,&general_protection);
1150         set_intr_gate(14,&page_fault);
1151         set_intr_gate(15,&spurious_interrupt_bug);
1152         set_intr_gate(16,&coprocessor_error);
1153         set_intr_gate(17,&alignment_check);
1154 #ifdef CONFIG_X86_MCE
1155         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1156 #endif
1157         set_intr_gate(19,&simd_coprocessor_error);
1158
1159 #ifdef CONFIG_IA32_EMULATION
1160         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1161 #endif
1162
1163         /*
1164          * Should be a barrier for any external CPU state.
1165          */
1166         cpu_init();
1167 }
1168
1169
1170 static int __init oops_setup(char *s)
1171 {
1172         if (!s)
1173                 return -EINVAL;
1174         if (!strcmp(s, "panic"))
1175                 panic_on_oops = 1;
1176         return 0;
1177 }
1178 early_param("oops", oops_setup);
1179
1180 static int __init kstack_setup(char *s)
1181 {
1182         if (!s)
1183                 return -EINVAL;
1184         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1185         return 0;
1186 }
1187 early_param("kstack", kstack_setup);
1188
1189 #ifdef CONFIG_STACK_UNWIND
1190 static int __init call_trace_setup(char *s)
1191 {
1192         if (!s)
1193                 return -EINVAL;
1194         if (strcmp(s, "old") == 0)
1195                 call_trace = -1;
1196         else if (strcmp(s, "both") == 0)
1197                 call_trace = 0;
1198         else if (strcmp(s, "newfallback") == 0)
1199                 call_trace = 1;
1200         else if (strcmp(s, "new") == 0)
1201                 call_trace = 2;
1202         return 0;
1203 }
1204 early_param("call_trace", call_trace_setup);
1205 #endif