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 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
 246 {
 247         void *t = (void *)tinfo;
 248         return p > t && p < t + THREAD_SIZE - 3;
 249 }
 250
 251 void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
 252                 struct stacktrace_ops *ops, void *data)
 253 {
 254         const unsigned cpu = smp_processor_id();
 255         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 256         unsigned used = 0;
 257         struct thread_info *tinfo;
 258
 259         if (!tsk)
 260                 tsk = current;
 261
 262         if (call_trace >= 0) {
 263                 int unw_ret = 0;
 264                 struct unwind_frame_info info;
 265                 struct ops_and_data oad = { .ops = ops, .data = data };
 266
 267                 if (regs) {
 268                         if (unwind_init_frame_info(&info, tsk, regs) == 0)
 269                                 unw_ret = dump_trace_unwind(&info, &oad);
 270                 } else if (tsk == current)
 271                         unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
 272                 else {
 273                         if (unwind_init_blocked(&info, tsk) == 0)
 274                                 unw_ret = dump_trace_unwind(&info, &oad);
 275                 }
 276                 if (unw_ret > 0) {
 277                         if (call_trace == 1 && !arch_unw_user_mode(&info)) {
 278                                 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
 279                                              UNW_PC(&info));
 280                                 if ((long)UNW_SP(&info) < 0) {
 281                                         ops->warning(data, "Leftover inexact backtrace:\n");
 282                                         stack = (unsigned long *)UNW_SP(&info);
 283                                         if (!stack)
 284                                                 return;
 285                                 } else
 286                                         ops->warning(data, "Full inexact backtrace again:\n");
 287                         } else if (call_trace >= 1)
 288                                 return;
 289                         else
 290                                 ops->warning(data, "Full inexact backtrace again:\n");
 291                 } else
 292                         ops->warning(data, "Inexact backtrace:\n");
 293         }
 294         if (!stack) {
 295                 unsigned long dummy;
 296                 stack = &dummy;
 297                 if (tsk && tsk != current)
 298                         stack = (unsigned long *)tsk->thread.rsp;
 299         }
 300         /*
 301          * Align the stack pointer on word boundary, later loops
 302          * rely on that (and corruption / debug info bugs can cause
 303          * unaligned values here):
 304          */
 305         stack = (unsigned long *)((unsigned long)stack & ~(sizeof(long)-1));
 306
 307         /*
 308          * Print function call entries within a stack. 'cond' is the
 309          * "end of stackframe" condition, that the 'stack++'
 310          * iteration will eventually trigger.
 311          */
 312 #define HANDLE_STACK(cond) \
 313         do while (cond) { \
 314                 unsigned long addr = *stack++; \
 315                 if (oops_in_progress ?          \
 316                         __kernel_text_address(addr) : \
 317                         kernel_text_address(addr)) { \
 318                         /* \
 319                          * If the address is either in the text segment of the \
 320                          * kernel, or in the region which contains vmalloc'ed \
 321                          * memory, it *may* be the address of a calling \
 322                          * routine; if so, print it so that someone tracing \
 323                          * down the cause of the crash will be able to figure \
 324                          * out the call path that was taken. \
 325                          */ \
 326                         ops->address(data, addr);   \
 327                 } \
 328         } while (0)
 329
 330         /*
 331          * Print function call entries in all stacks, starting at the
 332          * current stack address. If the stacks consist of nested
 333          * exceptions
 334          */
 335         for (;;) {
 336                 char *id;
 337                 unsigned long *estack_end;
 338                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 339                                                 &used, &id);
 340
 341                 if (estack_end) {
 342                         if (ops->stack(data, id) < 0)
 343                                 break;
 344                         HANDLE_STACK (stack < estack_end);
 345                         ops->stack(data, "<EOE>");
 346                         /*
 347                          * We link to the next stack via the
 348                          * second-to-last pointer (index -2 to end) in the
 349                          * exception stack:
 350                          */
 351                         stack = (unsigned long *) estack_end[-2];
 352                         continue;
 353                 }
 354                 if (irqstack_end) {
 355                         unsigned long *irqstack;
 356                         irqstack = irqstack_end -
 357                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 358
 359                         if (stack >= irqstack && stack < irqstack_end) {
 360                                 if (ops->stack(data, "IRQ") < 0)
 361                                         break;
 362                                 HANDLE_STACK (stack < irqstack_end);
 363                                 /*
 364                                  * We link to the next stack (which would be
 365                                  * the process stack normally) the last
 366                                  * pointer (index -1 to end) in the IRQ stack:
 367                                  */
 368                                 stack = (unsigned long *) (irqstack_end[-1]);
 369                                 irqstack_end = NULL;
 370                                 ops->stack(data, "EOI");
 371                                 continue;
 372                         }
 373                 }
 374                 break;
 375         }
 376
 377         /*
 378          * This handles the process stack:
 379          */
 380         tinfo = current_thread_info();
 381         HANDLE_STACK (valid_stack_ptr(tinfo, stack));
 382 #undef HANDLE_STACK
 383 }
 384 EXPORT_SYMBOL(dump_trace);
 385
 386 static void
 387 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 388 {
 389         print_symbol(msg, symbol);
 390         printk("\n");
 391 }
 392
 393 static void print_trace_warning(void *data, char *msg)
 394 {
 395         printk("%s\n", msg);
 396 }
 397
 398 static int print_trace_stack(void *data, char *name)
 399 {
 400         printk(" <%s> ", name);
 401         return 0;
 402 }
 403
 404 static void print_trace_address(void *data, unsigned long addr)
 405 {
 406         printk_address(addr);
 407 }
 408
 409 static struct stacktrace_ops print_trace_ops = {
 410         .warning = print_trace_warning,
 411         .warning_symbol = print_trace_warning_symbol,
 412         .stack = print_trace_stack,
 413         .address = print_trace_address,
 414 };
 415
 416 void
 417 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 418 {
 419         printk("\nCall Trace:\n");
 420         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 421         printk("\n");
 422 }
 423
 424 static void
 425 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 426 {
 427         unsigned long *stack;
 428         int i;
 429         const int cpu = smp_processor_id();
 430         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 431         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 432
 433         // debugging aid: "show_stack(NULL, NULL);" prints the
 434         // back trace for this cpu.
 435
 436         if (rsp == NULL) {
 437                 if (tsk)
 438                         rsp = (unsigned long *)tsk->thread.rsp;
 439                 else
 440                         rsp = (unsigned long *)&rsp;
 441         }
 442
 443         stack = rsp;
 444         for(i=0; i < kstack_depth_to_print; i++) {
 445                 if (stack >= irqstack && stack <= irqstack_end) {
 446                         if (stack == irqstack_end) {
 447                                 stack = (unsigned long *) (irqstack_end[-1]);
 448                                 printk(" <EOI> ");
 449                         }
 450                 } else {
 451                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 452                         break;
 453                 }
 454                 if (i && ((i % 4) == 0))
 455                         printk("\n");
 456                 printk(" %016lx", *stack++);
 457                 touch_nmi_watchdog();
 458         }
 459         show_trace(tsk, regs, rsp);
 460 }
 461
 462 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 463 {
 464         _show_stack(tsk, NULL, rsp);
 465 }
 466
 467 /*
 468  * The architecture-independent dump_stack generator
 469  */
 470 void dump_stack(void)
 471 {
 472         unsigned long dummy;
 473         show_trace(NULL, NULL, &dummy);
 474 }
 475
 476 EXPORT_SYMBOL(dump_stack);
 477
 478 void show_registers(struct pt_regs *regs)
 479 {
 480         int i;
 481         int in_kernel = !user_mode(regs);
 482         unsigned long rsp;
 483         const int cpu = smp_processor_id();
 484         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 485
 486                 rsp = regs->rsp;
 487
 488         printk("CPU %d ", cpu);
 489         __show_regs(regs);
 490         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 491                 cur->comm, cur->pid, task_thread_info(cur), cur);
 492
 493         /*
 494          * When in-kernel, we also print out the stack and code at the
 495          * time of the fault..
 496          */
 497         if (in_kernel) {
 498
 499                 printk("Stack: ");
 500                 _show_stack(NULL, regs, (unsigned long*)rsp);
 501
 502                 printk("\nCode: ");
 503                 if (regs->rip < PAGE_OFFSET)
 504                         goto bad;
 505
 506                 for (i=0; i<20; i++) {
 507                         unsigned char c;
 508                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 509 bad:
 510                                 printk(" Bad RIP value.");
 511                                 break;
 512                         }
 513                         printk("%02x ", c);
 514                 }
 515         }
 516         printk("\n");
 517 }
 518
 519 void handle_BUG(struct pt_regs *regs)
 520 {
 521         struct bug_frame f;
 522         long len;
 523         const char *prefix = "";
 524
 525         if (user_mode(regs))
 526                 return;
 527         if (__copy_from_user(&f, (const void __user *) regs->rip,
 528                              sizeof(struct bug_frame)))
 529                 return;
 530         if (f.filename >= 0 ||
 531             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 532                 return;
 533         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 534         if (len < 0 || len >= PATH_MAX)
 535                 f.filename = (int)(long)"unmapped filename";
 536         else if (len > 50) {
 537                 f.filename += len - 50;
 538                 prefix = "...";
 539         }
 540         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 541         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 542 }
 543
 544 #ifdef CONFIG_BUG
 545 void out_of_line_bug(void)
 546 {
 547         BUG();
 548 }
 549 EXPORT_SYMBOL(out_of_line_bug);
 550 #endif
 551
 552 static DEFINE_SPINLOCK(die_lock);
 553 static int die_owner = -1;
 554 static unsigned int die_nest_count;
 555
 556 unsigned __kprobes long oops_begin(void)
 557 {
 558         int cpu = smp_processor_id();
 559         unsigned long flags;
 560
 561         oops_enter();
 562
 563         /* racy, but better than risking deadlock. */
 564         local_irq_save(flags);
 565         if (!spin_trylock(&die_lock)) {
 566                 if (cpu == die_owner)
 567                         /* nested oops. should stop eventually */;
 568                 else
 569                         spin_lock(&die_lock);
 570         }
 571         die_nest_count++;
 572         die_owner = cpu;
 573         console_verbose();
 574         bust_spinlocks(1);
 575         return flags;
 576 }
 577
 578 void __kprobes oops_end(unsigned long flags)
 579 {
 580         die_owner = -1;
 581         bust_spinlocks(0);
 582         die_nest_count--;
 583         if (die_nest_count)
 584                 /* We still own the lock */
 585                 local_irq_restore(flags);
 586         else
 587                 /* Nest count reaches zero, release the lock. */
 588                 spin_unlock_irqrestore(&die_lock, flags);
 589         if (panic_on_oops)
 590                 panic("Fatal exception");
 591         oops_exit();
 592 }
 593
 594 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 595 {
 596         static int die_counter;
 597         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 598 #ifdef CONFIG_PREEMPT
 599         printk("PREEMPT ");
 600 #endif
 601 #ifdef CONFIG_SMP
 602         printk("SMP ");
 603 #endif
 604 #ifdef CONFIG_DEBUG_PAGEALLOC
 605         printk("DEBUG_PAGEALLOC");
 606 #endif
 607         printk("\n");
 608         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 609         show_registers(regs);
 610         /* Executive summary in case the oops scrolled away */
 611         printk(KERN_ALERT "RIP ");
 612         printk_address(regs->rip);
 613         printk(" RSP <%016lx>\n", regs->rsp);
 614         if (kexec_should_crash(current))
 615                 crash_kexec(regs);
 616 }
 617
 618 void die(const char * str, struct pt_regs * regs, long err)
 619 {
 620         unsigned long flags = oops_begin();
 621
 622         handle_BUG(regs);
 623         __die(str, regs, err);
 624         oops_end(flags);
 625         do_exit(SIGSEGV);
 626 }
 627
 628 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 629 {
 630         unsigned long flags = oops_begin();
 631
 632         /*
 633          * We are in trouble anyway, lets at least try
 634          * to get a message out.
 635          */
 636         printk(str, smp_processor_id());
 637         show_registers(regs);
 638         if (kexec_should_crash(current))
 639                 crash_kexec(regs);
 640         if (do_panic || panic_on_oops)
 641                 panic("Non maskable interrupt");
 642         oops_end(flags);
 643         nmi_exit();
 644         local_irq_enable();
 645         do_exit(SIGSEGV);
 646 }
 647
 648 static void __kprobes do_trap(int trapnr, int signr, char *str,
 649                               struct pt_regs * regs, long error_code,
 650                               siginfo_t *info)
 651 {
 652         struct task_struct *tsk = current;
 653
 654         tsk->thread.error_code = error_code;
 655         tsk->thread.trap_no = trapnr;
 656
 657         if (user_mode(regs)) {
 658                 if (exception_trace && unhandled_signal(tsk, signr))
 659                         printk(KERN_INFO
 660                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 661                                tsk->comm, tsk->pid, str,
 662                                regs->rip, regs->rsp, error_code);
 663
 664                 if (info)
 665                         force_sig_info(signr, info, tsk);
 666                 else
 667                         force_sig(signr, tsk);
 668                 return;
 669         }
 670
 671
 672         /* kernel trap */
 673         {
 674                 const struct exception_table_entry *fixup;
 675                 fixup = search_exception_tables(regs->rip);
 676                 if (fixup)
 677                         regs->rip = fixup->fixup;
 678                 else
 679                         die(str, regs, error_code);
 680                 return;
 681         }
 682 }
 683
 684 #define DO_ERROR(trapnr, signr, str, name) \
 685 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 686 { \
 687         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 688                                                         == NOTIFY_STOP) \
 689                 return; \
 690         conditional_sti(regs);                                          \
 691         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 692 }
 693
 694 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 695 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 696 { \
 697         siginfo_t info; \
 698         info.si_signo = signr; \
 699         info.si_errno = 0; \
 700         info.si_code = sicode; \
 701         info.si_addr = (void __user *)siaddr; \
 702         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 703                                                         == NOTIFY_STOP) \
 704                 return; \
 705         conditional_sti(regs);                                          \
 706         do_trap(trapnr, signr, str, regs, error_code, &info); \
 707 }
 708
 709 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 710 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 711 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 712 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 713 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 714 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 715 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 716 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 717 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 718 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 719
 720 /* Runs on IST stack */
 721 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 722 {
 723         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 724                         12, SIGBUS) == NOTIFY_STOP)
 725                 return;
 726         preempt_conditional_sti(regs);
 727         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 728         preempt_conditional_cli(regs);
 729 }
 730
 731 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 732 {
 733         static const char str[] = "double fault";
 734         struct task_struct *tsk = current;
 735
 736         /* Return not checked because double check cannot be ignored */
 737         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 738
 739         tsk->thread.error_code = error_code;
 740         tsk->thread.trap_no = 8;
 741
 742         /* This is always a kernel trap and never fixable (and thus must
 743            never return). */
 744         for (;;)
 745                 die(str, regs, error_code);
 746 }
 747
 748 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 749                                                 long error_code)
 750 {
 751         struct task_struct *tsk = current;
 752
 753         conditional_sti(regs);
 754
 755         tsk->thread.error_code = error_code;
 756         tsk->thread.trap_no = 13;
 757
 758         if (user_mode(regs)) {
 759                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 760                         printk(KERN_INFO
 761                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 762                                tsk->comm, tsk->pid,
 763                                regs->rip, regs->rsp, error_code);
 764
 765                 force_sig(SIGSEGV, tsk);
 766                 return;
 767         }
 768
 769         /* kernel gp */
 770         {
 771                 const struct exception_table_entry *fixup;
 772                 fixup = search_exception_tables(regs->rip);
 773                 if (fixup) {
 774                         regs->rip = fixup->fixup;
 775                         return;
 776                 }
 777                 if (notify_die(DIE_GPF, "general protection fault", regs,
 778                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 779                         return;
 780                 die("general protection fault", regs, error_code);
 781         }
 782 }
 783
 784 static __kprobes void
 785 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 786 {
 787         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 788                 reason);
 789         printk(KERN_EMERG "You probably have a hardware problem with your "
 790                 "RAM chips\n");
 791
 792         if (panic_on_unrecovered_nmi)
 793                 panic("NMI: Not continuing");
 794
 795         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 796
 797         /* Clear and disable the memory parity error line. */
 798         reason = (reason & 0xf) | 4;
 799         outb(reason, 0x61);
 800 }
 801
 802 static __kprobes void
 803 io_check_error(unsigned char reason, struct pt_regs * regs)
 804 {
 805         printk("NMI: IOCK error (debug interrupt?)\n");
 806         show_registers(regs);
 807
 808         /* Re-enable the IOCK line, wait for a few seconds */
 809         reason = (reason & 0xf) | 8;
 810         outb(reason, 0x61);
 811         mdelay(2000);
 812         reason &= ~8;
 813         outb(reason, 0x61);
 814 }
 815
 816 static __kprobes void
 817 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 818 {
 819         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 820                 reason);
 821         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 822
 823         if (panic_on_unrecovered_nmi)
 824                 panic("NMI: Not continuing");
 825
 826         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 827 }
 828
 829 /* Runs on IST stack. This code must keep interrupts off all the time.
 830    Nested NMIs are prevented by the CPU. */
 831 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 832 {
 833         unsigned char reason = 0;
 834         int cpu;
 835
 836         cpu = smp_processor_id();
 837
 838         /* Only the BSP gets external NMIs from the system.  */
 839         if (!cpu)
 840                 reason = get_nmi_reason();
 841
 842         if (!(reason & 0xc0)) {
 843                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 844                                                                 == NOTIFY_STOP)
 845                         return;
 846                 /*
 847                  * Ok, so this is none of the documented NMI sources,
 848                  * so it must be the NMI watchdog.
 849                  */
 850                 if (nmi_watchdog_tick(regs,reason))
 851                         return;
 852                 if (!do_nmi_callback(regs,cpu))
 853                         unknown_nmi_error(reason, regs);
 854
 855                 return;
 856         }
 857         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 858                 return;
 859
 860         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 861
 862         if (reason & 0x80)
 863                 mem_parity_error(reason, regs);
 864         if (reason & 0x40)
 865                 io_check_error(reason, regs);
 866 }
 867
 868 /* runs on IST stack. */
 869 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 870 {
 871         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 872                 return;
 873         }
 874         preempt_conditional_sti(regs);
 875         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 876         preempt_conditional_cli(regs);
 877 }
 878
 879 /* Help handler running on IST stack to switch back to user stack
 880    for scheduling or signal handling. The actual stack switch is done in
 881    entry.S */
 882 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 883 {
 884         struct pt_regs *regs = eregs;
 885         /* Did already sync */
 886         if (eregs == (struct pt_regs *)eregs->rsp)
 887                 ;
 888         /* Exception from user space */
 889         else if (user_mode(eregs))
 890                 regs = task_pt_regs(current);
 891         /* Exception from kernel and interrupts are enabled. Move to
 892            kernel process stack. */
 893         else if (eregs->eflags & X86_EFLAGS_IF)
 894                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 895         if (eregs != regs)
 896                 *regs = *eregs;
 897         return regs;
 898 }
 899
 900 /* runs on IST stack. */
 901 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 902                                    unsigned long error_code)
 903 {
 904         unsigned long condition;
 905         struct task_struct *tsk = current;
 906         siginfo_t info;
 907
 908         get_debugreg(condition, 6);
 909
 910         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 911                                                 SIGTRAP) == NOTIFY_STOP)
 912                 return;
 913
 914         preempt_conditional_sti(regs);
 915
 916         /* Mask out spurious debug traps due to lazy DR7 setting */
 917         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 918                 if (!tsk->thread.debugreg7) {
 919                         goto clear_dr7;
 920                 }
 921         }
 922
 923         tsk->thread.debugreg6 = condition;
 924
 925         /* Mask out spurious TF errors due to lazy TF clearing */
 926         if (condition & DR_STEP) {
 927                 /*
 928                  * The TF error should be masked out only if the current
 929                  * process is not traced and if the TRAP flag has been set
 930                  * previously by a tracing process (condition detected by
 931                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 932                  * can be modified by the process itself in user mode,
 933                  * allowing programs to debug themselves without the ptrace()
 934                  * interface.
 935                  */
 936                 if (!user_mode(regs))
 937                        goto clear_TF_reenable;
 938                 /*
 939                  * Was the TF flag set by a debugger? If so, clear it now,
 940                  * so that register information is correct.
 941                  */
 942                 if (tsk->ptrace & PT_DTRACE) {
 943                         regs->eflags &= ~TF_MASK;
 944                         tsk->ptrace &= ~PT_DTRACE;
 945                 }
 946         }
 947
 948         /* Ok, finally something we can handle */
 949         tsk->thread.trap_no = 1;
 950         tsk->thread.error_code = error_code;
 951         info.si_signo = SIGTRAP;
 952         info.si_errno = 0;
 953         info.si_code = TRAP_BRKPT;
 954         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 955         force_sig_info(SIGTRAP, &info, tsk);
 956
 957 clear_dr7:
 958         set_debugreg(0UL, 7);
 959         preempt_conditional_cli(regs);
 960         return;
 961
 962 clear_TF_reenable:
 963         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 964         regs->eflags &= ~TF_MASK;
 965         preempt_conditional_cli(regs);
 966 }
 967
 968 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 969 {
 970         const struct exception_table_entry *fixup;
 971         fixup = search_exception_tables(regs->rip);
 972         if (fixup) {
 973                 regs->rip = fixup->fixup;
 974                 return 1;
 975         }
 976         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 977         /* Illegal floating point operation in the kernel */
 978         current->thread.trap_no = trapnr;
 979         die(str, regs, 0);
 980         return 0;
 981 }
 982
 983 /*
 984  * Note that we play around with the 'TS' bit in an attempt to get
 985  * the correct behaviour even in the presence of the asynchronous
 986  * IRQ13 behaviour
 987  */
 988 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 989 {
 990         void __user *rip = (void __user *)(regs->rip);
 991         struct task_struct * task;
 992         siginfo_t info;
 993         unsigned short cwd, swd;
 994
 995         conditional_sti(regs);
 996         if (!user_mode(regs) &&
 997             kernel_math_error(regs, "kernel x87 math error", 16))
 998                 return;
 999
1000         /*
1001          * Save the info for the exception handler and clear the error.
1002          */
1003         task = current;
1004         save_init_fpu(task);
1005         task->thread.trap_no = 16;
1006         task->thread.error_code = 0;
1007         info.si_signo = SIGFPE;
1008         info.si_errno = 0;
1009         info.si_code = __SI_FAULT;
1010         info.si_addr = rip;
1011         /*
1012          * (~cwd & swd) will mask out exceptions that are not set to unmasked
1013          * status.  0x3f is the exception bits in these regs, 0x200 is the
1014          * C1 reg you need in case of a stack fault, 0x040 is the stack
1015          * fault bit.  We should only be taking one exception at a time,
1016          * so if this combination doesn't produce any single exception,
1017          * then we have a bad program that isn't synchronizing its FPU usage
1018          * and it will suffer the consequences since we won't be able to
1019          * fully reproduce the context of the exception
1020          */
1021         cwd = get_fpu_cwd(task);
1022         swd = get_fpu_swd(task);
1023         switch (swd & ~cwd & 0x3f) {
1024                 case 0x000:
1025                 default:
1026                         break;
1027                 case 0x001: /* Invalid Op */
1028                         /*
1029                          * swd & 0x240 == 0x040: Stack Underflow
1030                          * swd & 0x240 == 0x240: Stack Overflow
1031                          * User must clear the SF bit (0x40) if set
1032                          */
1033                         info.si_code = FPE_FLTINV;
1034                         break;
1035                 case 0x002: /* Denormalize */
1036                 case 0x010: /* Underflow */
1037                         info.si_code = FPE_FLTUND;
1038                         break;
1039                 case 0x004: /* Zero Divide */
1040                         info.si_code = FPE_FLTDIV;
1041                         break;
1042                 case 0x008: /* Overflow */
1043                         info.si_code = FPE_FLTOVF;
1044                         break;
1045                 case 0x020: /* Precision */
1046                         info.si_code = FPE_FLTRES;
1047                         break;
1048         }
1049         force_sig_info(SIGFPE, &info, task);
1050 }
1051
1052 asmlinkage void bad_intr(void)
1053 {
1054         printk("bad interrupt");
1055 }
1056
1057 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1058 {
1059         void __user *rip = (void __user *)(regs->rip);
1060         struct task_struct * task;
1061         siginfo_t info;
1062         unsigned short mxcsr;
1063
1064         conditional_sti(regs);
1065         if (!user_mode(regs) &&
1066                 kernel_math_error(regs, "kernel simd math error", 19))
1067                 return;
1068
1069         /*
1070          * Save the info for the exception handler and clear the error.
1071          */
1072         task = current;
1073         save_init_fpu(task);
1074         task->thread.trap_no = 19;
1075         task->thread.error_code = 0;
1076         info.si_signo = SIGFPE;
1077         info.si_errno = 0;
1078         info.si_code = __SI_FAULT;
1079         info.si_addr = rip;
1080         /*
1081          * The SIMD FPU exceptions are handled a little differently, as there
1082          * is only a single status/control register.  Thus, to determine which
1083          * unmasked exception was caught we must mask the exception mask bits
1084          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1085          */
1086         mxcsr = get_fpu_mxcsr(task);
1087         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1088                 case 0x000:
1089                 default:
1090                         break;
1091                 case 0x001: /* Invalid Op */
1092                         info.si_code = FPE_FLTINV;
1093                         break;
1094                 case 0x002: /* Denormalize */
1095                 case 0x010: /* Underflow */
1096                         info.si_code = FPE_FLTUND;
1097                         break;
1098                 case 0x004: /* Zero Divide */
1099                         info.si_code = FPE_FLTDIV;
1100                         break;
1101                 case 0x008: /* Overflow */
1102                         info.si_code = FPE_FLTOVF;
1103                         break;
1104                 case 0x020: /* Precision */
1105                         info.si_code = FPE_FLTRES;
1106                         break;
1107         }
1108         force_sig_info(SIGFPE, &info, task);
1109 }
1110
1111 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1112 {
1113 }
1114
1115 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1116 {
1117 }
1118
1119 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1120 {
1121 }
1122
1123 /*
1124  *  'math_state_restore()' saves the current math information in the
1125  * old math state array, and gets the new ones from the current task
1126  *
1127  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1128  * Don't touch unless you *really* know how it works.
1129  */
1130 asmlinkage void math_state_restore(void)
1131 {
1132         struct task_struct *me = current;
1133         clts();                 /* Allow maths ops (or we recurse) */
1134
1135         if (!used_math())
1136                 init_fpu(me);
1137         restore_fpu_checking(&me->thread.i387.fxsave);
1138         task_thread_info(me)->status |= TS_USEDFPU;
1139         me->fpu_counter++;
1140 }
1141
1142 void __init trap_init(void)
1143 {
1144         set_intr_gate(0,&divide_error);
1145         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1146         set_intr_gate_ist(2,&nmi,NMI_STACK);
1147         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1148         set_system_gate(4,&overflow);   /* int4 can be called from all */
1149         set_intr_gate(5,&bounds);
1150         set_intr_gate(6,&invalid_op);
1151         set_intr_gate(7,&device_not_available);
1152         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1153         set_intr_gate(9,&coprocessor_segment_overrun);
1154         set_intr_gate(10,&invalid_TSS);
1155         set_intr_gate(11,&segment_not_present);
1156         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1157         set_intr_gate(13,&general_protection);
1158         set_intr_gate(14,&page_fault);
1159         set_intr_gate(15,&spurious_interrupt_bug);
1160         set_intr_gate(16,&coprocessor_error);
1161         set_intr_gate(17,&alignment_check);
1162 #ifdef CONFIG_X86_MCE
1163         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1164 #endif
1165         set_intr_gate(19,&simd_coprocessor_error);
1166
1167 #ifdef CONFIG_IA32_EMULATION
1168         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1169 #endif
1170
1171         /*
1172          * Should be a barrier for any external CPU state.
1173          */
1174         cpu_init();
1175 }
1176
1177
1178 static int __init oops_setup(char *s)
1179 {
1180         if (!s)
1181                 return -EINVAL;
1182         if (!strcmp(s, "panic"))
1183                 panic_on_oops = 1;
1184         return 0;
1185 }
1186 early_param("oops", oops_setup);
1187
1188 static int __init kstack_setup(char *s)
1189 {
1190         if (!s)
1191                 return -EINVAL;
1192         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1193         return 0;
1194 }
1195 early_param("kstack", kstack_setup);
1196
1197 #ifdef CONFIG_STACK_UNWIND
1198 static int __init call_trace_setup(char *s)
1199 {
1200         if (!s)
1201                 return -EINVAL;
1202         if (strcmp(s, "old") == 0)
1203                 call_trace = -1;
1204         else if (strcmp(s, "both") == 0)
1205                 call_trace = 0;
1206         else if (strcmp(s, "newfallback") == 0)
1207                 call_trace = 1;
1208         else if (strcmp(s, "new") == 0)
1209                 call_trace = 2;
1210         return 0;
1211 }
1212 early_param("call_trace", call_trace_setup);
1213 #endif