arch/x86_64/kernel/traps.c

   1 /*
   2  *  linux/arch/x86-64/traps.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   6  *
   7  *  Pentium III FXSR, SSE support
   8  *      Gareth Hughes <gareth@valinux.com>, May 2000
   9  */
  10
  11 /*
  12  * 'Traps.c' handles hardware traps and faults after we have saved some
  13  * state in 'entry.S'.
  14  */
  15 #include <linux/sched.h>
  16 #include <linux/kernel.h>
  17 #include <linux/string.h>
  18 #include <linux/errno.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/timer.h>
  21 #include <linux/mm.h>
  22 #include <linux/init.h>
  23 #include <linux/delay.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/kallsyms.h>
  27 #include <linux/module.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/nmi.h>
  30 #include <linux/kprobes.h>
  31 #include <linux/kexec.h>
  32 #include <linux/unwind.h>
  33 #include <linux/uaccess.h>
  34
  35 #include <asm/system.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 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 #define MSG(txt) ops->warning(data, txt)
 239
 240 /*
 241  * x86-64 can have upto three kernel stacks:
 242  * process stack
 243  * interrupt stack
 244  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 245  */
 246
 247 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
 248 {
 249         void *t = (void *)tinfo;
 250         return p > t && p < t + THREAD_SIZE - 3;
 251 }
 252
 253 void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
 254                 unsigned long *stack,
 255                 struct stacktrace_ops *ops, void *data)
 256 {
 257         const unsigned cpu = get_cpu();
 258         unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
 259         unsigned used = 0;
 260         struct thread_info *tinfo;
 261
 262         if (!tsk)
 263                 tsk = current;
 264
 265         if (call_trace >= 0) {
 266                 int unw_ret = 0;
 267                 struct unwind_frame_info info;
 268                 struct ops_and_data oad = { .ops = ops, .data = data };
 269
 270                 if (regs) {
 271                         if (unwind_init_frame_info(&info, tsk, regs) == 0)
 272                                 unw_ret = dump_trace_unwind(&info, &oad);
 273                 } else if (tsk == current)
 274                         unw_ret = unwind_init_running(&info, dump_trace_unwind,
 275                                                       &oad);
 276                 else {
 277                         if (unwind_init_blocked(&info, tsk) == 0)
 278                                 unw_ret = dump_trace_unwind(&info, &oad);
 279                 }
 280                 if (unw_ret > 0) {
 281                         if (call_trace == 1 && !arch_unw_user_mode(&info)) {
 282                                 ops->warning_symbol(data,
 283                                              "DWARF2 unwinder stuck at %s",
 284                                              UNW_PC(&info));
 285                                 if ((long)UNW_SP(&info) < 0) {
 286                                         MSG("Leftover inexact backtrace:");
 287                                         stack = (unsigned long *)UNW_SP(&info);
 288                                         if (!stack)
 289                                                 goto out;
 290                                 } else
 291                                         MSG("Full inexact backtrace again:");
 292                         } else if (call_trace >= 1)
 293                                 goto out;
 294                         else
 295                                 MSG("Full inexact backtrace again:");
 296                 } else
 297                         MSG("Inexact backtrace:");
 298         }
 299         if (!stack) {
 300                 unsigned long dummy;
 301                 stack = &dummy;
 302                 if (tsk && tsk != current)
 303                         stack = (unsigned long *)tsk->thread.rsp;
 304         }
 305         /*
 306          * Align the stack pointer on word boundary, later loops
 307          * rely on that (and corruption / debug info bugs can cause
 308          * unaligned values here):
 309          */
 310         stack = (unsigned long *)((unsigned long)stack & ~(sizeof(long)-1));
 311
 312         /*
 313          * Print function call entries within a stack. 'cond' is the
 314          * "end of stackframe" condition, that the 'stack++'
 315          * iteration will eventually trigger.
 316          */
 317 #define HANDLE_STACK(cond) \
 318         do while (cond) { \
 319                 unsigned long addr = *stack++; \
 320                 if (oops_in_progress ?          \
 321                         __kernel_text_address(addr) : \
 322                         kernel_text_address(addr)) { \
 323                         /* \
 324                          * If the address is either in the text segment of the \
 325                          * kernel, or in the region which contains vmalloc'ed \
 326                          * memory, it *may* be the address of a calling \
 327                          * routine; if so, print it so that someone tracing \
 328                          * down the cause of the crash will be able to figure \
 329                          * out the call path that was taken. \
 330                          */ \
 331                         ops->address(data, addr);   \
 332                 } \
 333         } while (0)
 334
 335         /*
 336          * Print function call entries in all stacks, starting at the
 337          * current stack address. If the stacks consist of nested
 338          * exceptions
 339          */
 340         for (;;) {
 341                 char *id;
 342                 unsigned long *estack_end;
 343                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 344                                                 &used, &id);
 345
 346                 if (estack_end) {
 347                         if (ops->stack(data, id) < 0)
 348                                 break;
 349                         HANDLE_STACK (stack < estack_end);
 350                         ops->stack(data, "<EOE>");
 351                         /*
 352                          * We link to the next stack via the
 353                          * second-to-last pointer (index -2 to end) in the
 354                          * exception stack:
 355                          */
 356                         stack = (unsigned long *) estack_end[-2];
 357                         continue;
 358                 }
 359                 if (irqstack_end) {
 360                         unsigned long *irqstack;
 361                         irqstack = irqstack_end -
 362                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 363
 364                         if (stack >= irqstack && stack < irqstack_end) {
 365                                 if (ops->stack(data, "IRQ") < 0)
 366                                         break;
 367                                 HANDLE_STACK (stack < irqstack_end);
 368                                 /*
 369                                  * We link to the next stack (which would be
 370                                  * the process stack normally) the last
 371                                  * pointer (index -1 to end) in the IRQ stack:
 372                                  */
 373                                 stack = (unsigned long *) (irqstack_end[-1]);
 374                                 irqstack_end = NULL;
 375                                 ops->stack(data, "EOI");
 376                                 continue;
 377                         }
 378                 }
 379                 break;
 380         }
 381
 382         /*
 383          * This handles the process stack:
 384          */
 385         tinfo = current_thread_info();
 386         HANDLE_STACK (valid_stack_ptr(tinfo, stack));
 387 #undef HANDLE_STACK
 388 out:
 389         put_cpu();
 390 }
 391 EXPORT_SYMBOL(dump_trace);
 392
 393 static void
 394 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 395 {
 396         print_symbol(msg, symbol);
 397         printk("\n");
 398 }
 399
 400 static void print_trace_warning(void *data, char *msg)
 401 {
 402         printk("%s\n", msg);
 403 }
 404
 405 static int print_trace_stack(void *data, char *name)
 406 {
 407         printk(" <%s> ", name);
 408         return 0;
 409 }
 410
 411 static void print_trace_address(void *data, unsigned long addr)
 412 {
 413         printk_address(addr);
 414 }
 415
 416 static struct stacktrace_ops print_trace_ops = {
 417         .warning = print_trace_warning,
 418         .warning_symbol = print_trace_warning_symbol,
 419         .stack = print_trace_stack,
 420         .address = print_trace_address,
 421 };
 422
 423 void
 424 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 425 {
 426         printk("\nCall Trace:\n");
 427         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 428         printk("\n");
 429 }
 430
 431 static void
 432 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 433 {
 434         unsigned long *stack;
 435         int i;
 436         const int cpu = smp_processor_id();
 437         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 438         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 439
 440         // debugging aid: "show_stack(NULL, NULL);" prints the
 441         // back trace for this cpu.
 442
 443         if (rsp == NULL) {
 444                 if (tsk)
 445                         rsp = (unsigned long *)tsk->thread.rsp;
 446                 else
 447                         rsp = (unsigned long *)&rsp;
 448         }
 449
 450         stack = rsp;
 451         for(i=0; i < kstack_depth_to_print; i++) {
 452                 if (stack >= irqstack && stack <= irqstack_end) {
 453                         if (stack == irqstack_end) {
 454                                 stack = (unsigned long *) (irqstack_end[-1]);
 455                                 printk(" <EOI> ");
 456                         }
 457                 } else {
 458                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 459                         break;
 460                 }
 461                 if (i && ((i % 4) == 0))
 462                         printk("\n");
 463                 printk(" %016lx", *stack++);
 464                 touch_nmi_watchdog();
 465         }
 466         show_trace(tsk, regs, rsp);
 467 }
 468
 469 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 470 {
 471         _show_stack(tsk, NULL, rsp);
 472 }
 473
 474 /*
 475  * The architecture-independent dump_stack generator
 476  */
 477 void dump_stack(void)
 478 {
 479         unsigned long dummy;
 480         show_trace(NULL, NULL, &dummy);
 481 }
 482
 483 EXPORT_SYMBOL(dump_stack);
 484
 485 void show_registers(struct pt_regs *regs)
 486 {
 487         int i;
 488         int in_kernel = !user_mode(regs);
 489         unsigned long rsp;
 490         const int cpu = smp_processor_id();
 491         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 492
 493                 rsp = regs->rsp;
 494
 495         printk("CPU %d ", cpu);
 496         __show_regs(regs);
 497         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 498                 cur->comm, cur->pid, task_thread_info(cur), cur);
 499
 500         /*
 501          * When in-kernel, we also print out the stack and code at the
 502          * time of the fault..
 503          */
 504         if (in_kernel) {
 505
 506                 printk("Stack: ");
 507                 _show_stack(NULL, regs, (unsigned long*)rsp);
 508
 509                 printk("\nCode: ");
 510                 if (regs->rip < PAGE_OFFSET)
 511                         goto bad;
 512
 513                 for (i=0; i<20; i++) {
 514                         unsigned char c;
 515                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 516 bad:
 517                                 printk(" Bad RIP value.");
 518                                 break;
 519                         }
 520                         printk("%02x ", c);
 521                 }
 522         }
 523         printk("\n");
 524 }
 525
 526 void handle_BUG(struct pt_regs *regs)
 527 {
 528         struct bug_frame f;
 529         long len;
 530         const char *prefix = "";
 531
 532         if (user_mode(regs))
 533                 return;
 534         if (__copy_from_user(&f, (const void __user *) regs->rip,
 535                              sizeof(struct bug_frame)))
 536                 return;
 537         if (f.filename >= 0 ||
 538             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 539                 return;
 540         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 541         if (len < 0 || len >= PATH_MAX)
 542                 f.filename = (int)(long)"unmapped filename";
 543         else if (len > 50) {
 544                 f.filename += len - 50;
 545                 prefix = "...";
 546         }
 547         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 548         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 549 }
 550
 551 #ifdef CONFIG_BUG
 552 void out_of_line_bug(void)
 553 {
 554         BUG();
 555 }
 556 EXPORT_SYMBOL(out_of_line_bug);
 557 #endif
 558
 559 static DEFINE_SPINLOCK(die_lock);
 560 static int die_owner = -1;
 561 static unsigned int die_nest_count;
 562
 563 unsigned __kprobes long oops_begin(void)
 564 {
 565         int cpu = smp_processor_id();
 566         unsigned long flags;
 567
 568         oops_enter();
 569
 570         /* racy, but better than risking deadlock. */
 571         local_irq_save(flags);
 572         if (!spin_trylock(&die_lock)) {
 573                 if (cpu == die_owner)
 574                         /* nested oops. should stop eventually */;
 575                 else
 576                         spin_lock(&die_lock);
 577         }
 578         die_nest_count++;
 579         die_owner = cpu;
 580         console_verbose();
 581         bust_spinlocks(1);
 582         return flags;
 583 }
 584
 585 void __kprobes oops_end(unsigned long flags)
 586 {
 587         die_owner = -1;
 588         bust_spinlocks(0);
 589         die_nest_count--;
 590         if (die_nest_count)
 591                 /* We still own the lock */
 592                 local_irq_restore(flags);
 593         else
 594                 /* Nest count reaches zero, release the lock. */
 595                 spin_unlock_irqrestore(&die_lock, flags);
 596         if (panic_on_oops)
 597                 panic("Fatal exception");
 598         oops_exit();
 599 }
 600
 601 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 602 {
 603         static int die_counter;
 604         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 605 #ifdef CONFIG_PREEMPT
 606         printk("PREEMPT ");
 607 #endif
 608 #ifdef CONFIG_SMP
 609         printk("SMP ");
 610 #endif
 611 #ifdef CONFIG_DEBUG_PAGEALLOC
 612         printk("DEBUG_PAGEALLOC");
 613 #endif
 614         printk("\n");
 615         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 616         show_registers(regs);
 617         /* Executive summary in case the oops scrolled away */
 618         printk(KERN_ALERT "RIP ");
 619         printk_address(regs->rip);
 620         printk(" RSP <%016lx>\n", regs->rsp);
 621         if (kexec_should_crash(current))
 622                 crash_kexec(regs);
 623 }
 624
 625 void die(const char * str, struct pt_regs * regs, long err)
 626 {
 627         unsigned long flags = oops_begin();
 628
 629         handle_BUG(regs);
 630         __die(str, regs, err);
 631         oops_end(flags);
 632         do_exit(SIGSEGV);
 633 }
 634
 635 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 636 {
 637         unsigned long flags = oops_begin();
 638
 639         /*
 640          * We are in trouble anyway, lets at least try
 641          * to get a message out.
 642          */
 643         printk(str, smp_processor_id());
 644         show_registers(regs);
 645         if (kexec_should_crash(current))
 646                 crash_kexec(regs);
 647         if (do_panic || panic_on_oops)
 648                 panic("Non maskable interrupt");
 649         oops_end(flags);
 650         nmi_exit();
 651         local_irq_enable();
 652         do_exit(SIGSEGV);
 653 }
 654
 655 static void __kprobes do_trap(int trapnr, int signr, char *str,
 656                               struct pt_regs * regs, long error_code,
 657                               siginfo_t *info)
 658 {
 659         struct task_struct *tsk = current;
 660
 661         tsk->thread.error_code = error_code;
 662         tsk->thread.trap_no = trapnr;
 663
 664         if (user_mode(regs)) {
 665                 if (exception_trace && unhandled_signal(tsk, signr))
 666                         printk(KERN_INFO
 667                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 668                                tsk->comm, tsk->pid, str,
 669                                regs->rip, regs->rsp, error_code);
 670
 671                 if (info)
 672                         force_sig_info(signr, info, tsk);
 673                 else
 674                         force_sig(signr, tsk);
 675                 return;
 676         }
 677
 678
 679         /* kernel trap */
 680         {
 681                 const struct exception_table_entry *fixup;
 682                 fixup = search_exception_tables(regs->rip);
 683                 if (fixup)
 684                         regs->rip = fixup->fixup;
 685                 else
 686                         die(str, regs, error_code);
 687                 return;
 688         }
 689 }
 690
 691 #define DO_ERROR(trapnr, signr, str, name) \
 692 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 693 { \
 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, NULL); \
 699 }
 700
 701 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 702 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 703 { \
 704         siginfo_t info; \
 705         info.si_signo = signr; \
 706         info.si_errno = 0; \
 707         info.si_code = sicode; \
 708         info.si_addr = (void __user *)siaddr; \
 709         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 710                                                         == NOTIFY_STOP) \
 711                 return; \
 712         conditional_sti(regs);                                          \
 713         do_trap(trapnr, signr, str, regs, error_code, &info); \
 714 }
 715
 716 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 717 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 718 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 719 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 720 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 721 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 722 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 723 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 724 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 725 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 726
 727 /* Runs on IST stack */
 728 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 729 {
 730         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 731                         12, SIGBUS) == NOTIFY_STOP)
 732                 return;
 733         preempt_conditional_sti(regs);
 734         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 735         preempt_conditional_cli(regs);
 736 }
 737
 738 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 739 {
 740         static const char str[] = "double fault";
 741         struct task_struct *tsk = current;
 742
 743         /* Return not checked because double check cannot be ignored */
 744         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 745
 746         tsk->thread.error_code = error_code;
 747         tsk->thread.trap_no = 8;
 748
 749         /* This is always a kernel trap and never fixable (and thus must
 750            never return). */
 751         for (;;)
 752                 die(str, regs, error_code);
 753 }
 754
 755 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 756                                                 long error_code)
 757 {
 758         struct task_struct *tsk = current;
 759
 760         conditional_sti(regs);
 761
 762         tsk->thread.error_code = error_code;
 763         tsk->thread.trap_no = 13;
 764
 765         if (user_mode(regs)) {
 766                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 767                         printk(KERN_INFO
 768                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 769                                tsk->comm, tsk->pid,
 770                                regs->rip, regs->rsp, error_code);
 771
 772                 force_sig(SIGSEGV, tsk);
 773                 return;
 774         }
 775
 776         /* kernel gp */
 777         {
 778                 const struct exception_table_entry *fixup;
 779                 fixup = search_exception_tables(regs->rip);
 780                 if (fixup) {
 781                         regs->rip = fixup->fixup;
 782                         return;
 783                 }
 784                 if (notify_die(DIE_GPF, "general protection fault", regs,
 785                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 786                         return;
 787                 die("general protection fault", regs, error_code);
 788         }
 789 }
 790
 791 static __kprobes void
 792 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 793 {
 794         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 795                 reason);
 796         printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 797
 798         if (panic_on_unrecovered_nmi)
 799                 panic("NMI: Not continuing");
 800
 801         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 802
 803         /* Clear and disable the memory parity error line. */
 804         reason = (reason & 0xf) | 4;
 805         outb(reason, 0x61);
 806 }
 807
 808 static __kprobes void
 809 io_check_error(unsigned char reason, struct pt_regs * regs)
 810 {
 811         printk("NMI: IOCK error (debug interrupt?)\n");
 812         show_registers(regs);
 813
 814         /* Re-enable the IOCK line, wait for a few seconds */
 815         reason = (reason & 0xf) | 8;
 816         outb(reason, 0x61);
 817         mdelay(2000);
 818         reason &= ~8;
 819         outb(reason, 0x61);
 820 }
 821
 822 static __kprobes void
 823 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 824 {
 825         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 826                 reason);
 827         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 828
 829         if (panic_on_unrecovered_nmi)
 830                 panic("NMI: Not continuing");
 831
 832         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 833 }
 834
 835 /* Runs on IST stack. This code must keep interrupts off all the time.
 836    Nested NMIs are prevented by the CPU. */
 837 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 838 {
 839         unsigned char reason = 0;
 840         int cpu;
 841
 842         cpu = smp_processor_id();
 843
 844         /* Only the BSP gets external NMIs from the system.  */
 845         if (!cpu)
 846                 reason = get_nmi_reason();
 847
 848         if (!(reason & 0xc0)) {
 849                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 850                                                                 == NOTIFY_STOP)
 851                         return;
 852                 /*
 853                  * Ok, so this is none of the documented NMI sources,
 854                  * so it must be the NMI watchdog.
 855                  */
 856                 if (nmi_watchdog_tick(regs,reason))
 857                         return;
 858                 if (!do_nmi_callback(regs,cpu))
 859                         unknown_nmi_error(reason, regs);
 860
 861                 return;
 862         }
 863         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 864                 return;
 865
 866         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 867
 868         if (reason & 0x80)
 869                 mem_parity_error(reason, regs);
 870         if (reason & 0x40)
 871                 io_check_error(reason, regs);
 872 }
 873
 874 /* runs on IST stack. */
 875 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 876 {
 877         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 878                 return;
 879         }
 880         preempt_conditional_sti(regs);
 881         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 882         preempt_conditional_cli(regs);
 883 }
 884
 885 /* Help handler running on IST stack to switch back to user stack
 886    for scheduling or signal handling. The actual stack switch is done in
 887    entry.S */
 888 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 889 {
 890         struct pt_regs *regs = eregs;
 891         /* Did already sync */
 892         if (eregs == (struct pt_regs *)eregs->rsp)
 893                 ;
 894         /* Exception from user space */
 895         else if (user_mode(eregs))
 896                 regs = task_pt_regs(current);
 897         /* Exception from kernel and interrupts are enabled. Move to
 898            kernel process stack. */
 899         else if (eregs->eflags & X86_EFLAGS_IF)
 900                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 901         if (eregs != regs)
 902                 *regs = *eregs;
 903         return regs;
 904 }
 905
 906 /* runs on IST stack. */
 907 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 908                                    unsigned long error_code)
 909 {
 910         unsigned long condition;
 911         struct task_struct *tsk = current;
 912         siginfo_t info;
 913
 914         get_debugreg(condition, 6);
 915
 916         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 917                                                 SIGTRAP) == NOTIFY_STOP)
 918                 return;
 919
 920         preempt_conditional_sti(regs);
 921
 922         /* Mask out spurious debug traps due to lazy DR7 setting */
 923         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 924                 if (!tsk->thread.debugreg7) {
 925                         goto clear_dr7;
 926                 }
 927         }
 928
 929         tsk->thread.debugreg6 = condition;
 930
 931         /* Mask out spurious TF errors due to lazy TF clearing */
 932         if (condition & DR_STEP) {
 933                 /*
 934                  * The TF error should be masked out only if the current
 935                  * process is not traced and if the TRAP flag has been set
 936                  * previously by a tracing process (condition detected by
 937                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 938                  * can be modified by the process itself in user mode,
 939                  * allowing programs to debug themselves without the ptrace()
 940                  * interface.
 941                  */
 942                 if (!user_mode(regs))
 943                        goto clear_TF_reenable;
 944                 /*
 945                  * Was the TF flag set by a debugger? If so, clear it now,
 946                  * so that register information is correct.
 947                  */
 948                 if (tsk->ptrace & PT_DTRACE) {
 949                         regs->eflags &= ~TF_MASK;
 950                         tsk->ptrace &= ~PT_DTRACE;
 951                 }
 952         }
 953
 954         /* Ok, finally something we can handle */
 955         tsk->thread.trap_no = 1;
 956         tsk->thread.error_code = error_code;
 957         info.si_signo = SIGTRAP;
 958         info.si_errno = 0;
 959         info.si_code = TRAP_BRKPT;
 960         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 961         force_sig_info(SIGTRAP, &info, tsk);
 962
 963 clear_dr7:
 964         set_debugreg(0UL, 7);
 965         preempt_conditional_cli(regs);
 966         return;
 967
 968 clear_TF_reenable:
 969         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 970         regs->eflags &= ~TF_MASK;
 971         preempt_conditional_cli(regs);
 972 }
 973
 974 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 975 {
 976         const struct exception_table_entry *fixup;
 977         fixup = search_exception_tables(regs->rip);
 978         if (fixup) {
 979                 regs->rip = fixup->fixup;
 980                 return 1;
 981         }
 982         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 983         /* Illegal floating point operation in the kernel */
 984         current->thread.trap_no = trapnr;
 985         die(str, regs, 0);
 986         return 0;
 987 }
 988
 989 /*
 990  * Note that we play around with the 'TS' bit in an attempt to get
 991  * the correct behaviour even in the presence of the asynchronous
 992  * IRQ13 behaviour
 993  */
 994 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 995 {
 996         void __user *rip = (void __user *)(regs->rip);
 997         struct task_struct * task;
 998         siginfo_t info;
 999         unsigned short cwd, swd;
1000
1001         conditional_sti(regs);
1002         if (!user_mode(regs) &&
1003             kernel_math_error(regs, "kernel x87 math error", 16))
1004                 return;
1005
1006         /*
1007          * Save the info for the exception handler and clear the error.
1008          */
1009         task = current;
1010         save_init_fpu(task);
1011         task->thread.trap_no = 16;
1012         task->thread.error_code = 0;
1013         info.si_signo = SIGFPE;
1014         info.si_errno = 0;
1015         info.si_code = __SI_FAULT;
1016         info.si_addr = rip;
1017         /*
1018          * (~cwd & swd) will mask out exceptions that are not set to unmasked
1019          * status.  0x3f is the exception bits in these regs, 0x200 is the
1020          * C1 reg you need in case of a stack fault, 0x040 is the stack
1021          * fault bit.  We should only be taking one exception at a time,
1022          * so if this combination doesn't produce any single exception,
1023          * then we have a bad program that isn't synchronizing its FPU usage
1024          * and it will suffer the consequences since we won't be able to
1025          * fully reproduce the context of the exception
1026          */
1027         cwd = get_fpu_cwd(task);
1028         swd = get_fpu_swd(task);
1029         switch (swd & ~cwd & 0x3f) {
1030                 case 0x000:
1031                 default:
1032                         break;
1033                 case 0x001: /* Invalid Op */
1034                         /*
1035                          * swd & 0x240 == 0x040: Stack Underflow
1036                          * swd & 0x240 == 0x240: Stack Overflow
1037                          * User must clear the SF bit (0x40) if set
1038                          */
1039                         info.si_code = FPE_FLTINV;
1040                         break;
1041                 case 0x002: /* Denormalize */
1042                 case 0x010: /* Underflow */
1043                         info.si_code = FPE_FLTUND;
1044                         break;
1045                 case 0x004: /* Zero Divide */
1046                         info.si_code = FPE_FLTDIV;
1047                         break;
1048                 case 0x008: /* Overflow */
1049                         info.si_code = FPE_FLTOVF;
1050                         break;
1051                 case 0x020: /* Precision */
1052                         info.si_code = FPE_FLTRES;
1053                         break;
1054         }
1055         force_sig_info(SIGFPE, &info, task);
1056 }
1057
1058 asmlinkage void bad_intr(void)
1059 {
1060         printk("bad interrupt");
1061 }
1062
1063 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1064 {
1065         void __user *rip = (void __user *)(regs->rip);
1066         struct task_struct * task;
1067         siginfo_t info;
1068         unsigned short mxcsr;
1069
1070         conditional_sti(regs);
1071         if (!user_mode(regs) &&
1072                 kernel_math_error(regs, "kernel simd math error", 19))
1073                 return;
1074
1075         /*
1076          * Save the info for the exception handler and clear the error.
1077          */
1078         task = current;
1079         save_init_fpu(task);
1080         task->thread.trap_no = 19;
1081         task->thread.error_code = 0;
1082         info.si_signo = SIGFPE;
1083         info.si_errno = 0;
1084         info.si_code = __SI_FAULT;
1085         info.si_addr = rip;
1086         /*
1087          * The SIMD FPU exceptions are handled a little differently, as there
1088          * is only a single status/control register.  Thus, to determine which
1089          * unmasked exception was caught we must mask the exception mask bits
1090          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1091          */
1092         mxcsr = get_fpu_mxcsr(task);
1093         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1094                 case 0x000:
1095                 default:
1096                         break;
1097                 case 0x001: /* Invalid Op */
1098                         info.si_code = FPE_FLTINV;
1099                         break;
1100                 case 0x002: /* Denormalize */
1101                 case 0x010: /* Underflow */
1102                         info.si_code = FPE_FLTUND;
1103                         break;
1104                 case 0x004: /* Zero Divide */
1105                         info.si_code = FPE_FLTDIV;
1106                         break;
1107                 case 0x008: /* Overflow */
1108                         info.si_code = FPE_FLTOVF;
1109                         break;
1110                 case 0x020: /* Precision */
1111                         info.si_code = FPE_FLTRES;
1112                         break;
1113         }
1114         force_sig_info(SIGFPE, &info, task);
1115 }
1116
1117 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1118 {
1119 }
1120
1121 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1122 {
1123 }
1124
1125 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1126 {
1127 }
1128
1129 /*
1130  *  'math_state_restore()' saves the current math information in the
1131  * old math state array, and gets the new ones from the current task
1132  *
1133  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1134  * Don't touch unless you *really* know how it works.
1135  */
1136 asmlinkage void math_state_restore(void)
1137 {
1138         struct task_struct *me = current;
1139         clts();                 /* Allow maths ops (or we recurse) */
1140
1141         if (!used_math())
1142                 init_fpu(me);
1143         restore_fpu_checking(&me->thread.i387.fxsave);
1144         task_thread_info(me)->status |= TS_USEDFPU;
1145         me->fpu_counter++;
1146 }
1147
1148 void __init trap_init(void)
1149 {
1150         set_intr_gate(0,&divide_error);
1151         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1152         set_intr_gate_ist(2,&nmi,NMI_STACK);
1153         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1154         set_system_gate(4,&overflow);   /* int4 can be called from all */
1155         set_intr_gate(5,&bounds);
1156         set_intr_gate(6,&invalid_op);
1157         set_intr_gate(7,&device_not_available);
1158         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1159         set_intr_gate(9,&coprocessor_segment_overrun);
1160         set_intr_gate(10,&invalid_TSS);
1161         set_intr_gate(11,&segment_not_present);
1162         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1163         set_intr_gate(13,&general_protection);
1164         set_intr_gate(14,&page_fault);
1165         set_intr_gate(15,&spurious_interrupt_bug);
1166         set_intr_gate(16,&coprocessor_error);
1167         set_intr_gate(17,&alignment_check);
1168 #ifdef CONFIG_X86_MCE
1169         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1170 #endif
1171         set_intr_gate(19,&simd_coprocessor_error);
1172
1173 #ifdef CONFIG_IA32_EMULATION
1174         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1175 #endif
1176
1177         /*
1178          * Should be a barrier for any external CPU state.
1179          */
1180         cpu_init();
1181 }
1182
1183
1184 static int __init oops_setup(char *s)
1185 {
1186         if (!s)
1187                 return -EINVAL;
1188         if (!strcmp(s, "panic"))
1189                 panic_on_oops = 1;
1190         return 0;
1191 }
1192 early_param("oops", oops_setup);
1193
1194 static int __init kstack_setup(char *s)
1195 {
1196         if (!s)
1197                 return -EINVAL;
1198         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1199         return 0;
1200 }
1201 early_param("kstack", kstack_setup);
1202
1203 #ifdef CONFIG_STACK_UNWIND
1204 static int __init call_trace_setup(char *s)
1205 {
1206         if (!s)
1207                 return -EINVAL;
1208         if (strcmp(s, "old") == 0)
1209                 call_trace = -1;
1210         else if (strcmp(s, "both") == 0)
1211                 call_trace = 0;
1212         else if (strcmp(s, "newfallback") == 0)
1213                 call_trace = 1;
1214         else if (strcmp(s, "new") == 0)
1215                 call_trace = 2;
1216         return 0;
1217 }
1218 early_param("call_trace", call_trace_setup);
1219 #endif