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/module.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/nmi.h>
  29 #include <linux/kprobes.h>
  30 #include <linux/kexec.h>
  31 #include <linux/unwind.h>
  32
  33 #include <asm/system.h>
  34 #include <asm/uaccess.h>
  35 #include <asm/io.h>
  36 #include <asm/atomic.h>
  37 #include <asm/debugreg.h>
  38 #include <asm/desc.h>
  39 #include <asm/i387.h>
  40 #include <asm/kdebug.h>
  41 #include <asm/processor.h>
  42 #include <asm/unwind.h>
  43 #include <asm/smp.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/pda.h>
  46 #include <asm/proto.h>
  47 #include <asm/nmi.h>
  48 #include <asm/stacktrace.h>
  49
  50 asmlinkage void divide_error(void);
  51 asmlinkage void debug(void);
  52 asmlinkage void nmi(void);
  53 asmlinkage void int3(void);
  54 asmlinkage void overflow(void);
  55 asmlinkage void bounds(void);
  56 asmlinkage void invalid_op(void);
  57 asmlinkage void device_not_available(void);
  58 asmlinkage void double_fault(void);
  59 asmlinkage void coprocessor_segment_overrun(void);
  60 asmlinkage void invalid_TSS(void);
  61 asmlinkage void segment_not_present(void);
  62 asmlinkage void stack_segment(void);
  63 asmlinkage void general_protection(void);
  64 asmlinkage void page_fault(void);
  65 asmlinkage void coprocessor_error(void);
  66 asmlinkage void simd_coprocessor_error(void);
  67 asmlinkage void reserved(void);
  68 asmlinkage void alignment_check(void);
  69 asmlinkage void machine_check(void);
  70 asmlinkage void spurious_interrupt_bug(void);
  71
  72 ATOMIC_NOTIFIER_HEAD(die_chain);
  73 EXPORT_SYMBOL(die_chain);
  74
  75 int register_die_notifier(struct notifier_block *nb)
  76 {
  77         vmalloc_sync_all();
  78         return atomic_notifier_chain_register(&die_chain, nb);
  79 }
  80 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
  81
  82 int unregister_die_notifier(struct notifier_block *nb)
  83 {
  84         return atomic_notifier_chain_unregister(&die_chain, nb);
  85 }
  86 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
  87
  88 static inline void conditional_sti(struct pt_regs *regs)
  89 {
  90         if (regs->eflags & X86_EFLAGS_IF)
  91                 local_irq_enable();
  92 }
  93
  94 static inline void preempt_conditional_sti(struct pt_regs *regs)
  95 {
  96         preempt_disable();
  97         if (regs->eflags & X86_EFLAGS_IF)
  98                 local_irq_enable();
  99 }
 100
 101 static inline void preempt_conditional_cli(struct pt_regs *regs)
 102 {
 103         if (regs->eflags & X86_EFLAGS_IF)
 104                 local_irq_disable();
 105         /* Make sure to not schedule here because we could be running
 106            on an exception stack. */
 107         preempt_enable_no_resched();
 108 }
 109
 110 static int kstack_depth_to_print = 12;
 111 #ifdef CONFIG_STACK_UNWIND
 112 static int call_trace = 1;
 113 #else
 114 #define call_trace (-1)
 115 #endif
 116
 117 #ifdef CONFIG_KALLSYMS
 118 # include <linux/kallsyms.h>
 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;
 166
 167                 /*
 168                  * set 'end' to the end of the exception stack.
 169                  */
 170                 switch (k + 1) {
 171                 /*
 172                  * TODO: this block is not needed i think, because
 173                  * setup64.c:cpu_init() sets up t->ist[DEBUG_STACK]
 174                  * properly too.
 175                  */
 176 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 177                 case DEBUG_STACK:
 178                         end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
 179                         break;
 180 #endif
 181                 default:
 182                         end = per_cpu(orig_ist, cpu).ist[k];
 183                         break;
 184                 }
 185                 /*
 186                  * Is 'stack' above this exception frame's end?
 187                  * If yes then skip to the next frame.
 188                  */
 189                 if (stack >= end)
 190                         continue;
 191                 /*
 192                  * Is 'stack' above this exception frame's start address?
 193                  * If yes then we found the right frame.
 194                  */
 195                 if (stack >= end - EXCEPTION_STKSZ) {
 196                         /*
 197                          * Make sure we only iterate through an exception
 198                          * stack once. If it comes up for the second time
 199                          * then there's something wrong going on - just
 200                          * break out and return NULL:
 201                          */
 202                         if (*usedp & (1U << k))
 203                                 break;
 204                         *usedp |= 1U << k;
 205                         *idp = ids[k];
 206                         return (unsigned long *)end;
 207                 }
 208                 /*
 209                  * If this is a debug stack, and if it has a larger size than
 210                  * the usual exception stacks, then 'stack' might still
 211                  * be within the lower portion of the debug stack:
 212                  */
 213 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 214                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 215                         unsigned j = N_EXCEPTION_STACKS - 1;
 216
 217                         /*
 218                          * Black magic. A large debug stack is composed of
 219                          * multiple exception stack entries, which we
 220                          * iterate through now. Dont look:
 221                          */
 222                         do {
 223                                 ++j;
 224                                 end -= EXCEPTION_STKSZ;
 225                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 226                         } while (stack < end - EXCEPTION_STKSZ);
 227                         if (*usedp & (1U << j))
 228                                 break;
 229                         *usedp |= 1U << j;
 230                         *idp = ids[j];
 231                         return (unsigned long *)end;
 232                 }
 233 #endif
 234         }
 235         return NULL;
 236 }
 237
 238 struct ops_and_data {
 239         struct stacktrace_ops *ops;
 240         void *data;
 241 };
 242
 243 static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
 244 {
 245         struct ops_and_data *oad = (struct ops_and_data *)context;
 246         int n = 0;
 247
 248         while (unwind(info) == 0 && UNW_PC(info)) {
 249                 n++;
 250                 oad->ops->address(oad->data, UNW_PC(info));
 251                 if (arch_unw_user_mode(info))
 252                         break;
 253         }
 254         return n;
 255 }
 256
 257 /*
 258  * x86-64 can have upto three kernel stacks:
 259  * process stack
 260  * interrupt stack
 261  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 262  */
 263
 264 void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
 265                 struct stacktrace_ops *ops, void *data)
 266 {
 267         const unsigned cpu = safe_smp_processor_id();
 268         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 269         unsigned used = 0;
 270
 271         if (!tsk)
 272                 tsk = current;
 273
 274         if (call_trace >= 0) {
 275                 int unw_ret = 0;
 276                 struct unwind_frame_info info;
 277                 struct ops_and_data oad = { .ops = ops, .data = data };
 278
 279                 if (regs) {
 280                         if (unwind_init_frame_info(&info, tsk, regs) == 0)
 281                                 unw_ret = dump_trace_unwind(&info, &oad);
 282                 } else if (tsk == current)
 283                         unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
 284                 else {
 285                         if (unwind_init_blocked(&info, tsk) == 0)
 286                                 unw_ret = dump_trace_unwind(&info, &oad);
 287                 }
 288                 if (unw_ret > 0) {
 289                         if (call_trace == 1 && !arch_unw_user_mode(&info)) {
 290                                 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
 291                                              UNW_PC(&info));
 292                                 if ((long)UNW_SP(&info) < 0) {
 293                                         ops->warning(data, "Leftover inexact backtrace:\n");
 294                                         stack = (unsigned long *)UNW_SP(&info);
 295                                         if (!stack)
 296                                                 return;
 297                                 } else
 298                                         ops->warning(data, "Full inexact backtrace again:\n");
 299                         } else if (call_trace >= 1)
 300                                 return;
 301                         else
 302                                 ops->warning(data, "Full inexact backtrace again:\n");
 303                 } else
 304                         ops->warning(data, "Inexact backtrace:\n");
 305         }
 306         if (!stack) {
 307                 unsigned long dummy;
 308                 stack = &dummy;
 309                 if (tsk && tsk != current)
 310                         stack = (unsigned long *)tsk->thread.rsp;
 311         }
 312
 313         /*
 314          * Print function call entries within a stack. 'cond' is the
 315          * "end of stackframe" condition, that the 'stack++'
 316          * iteration will eventually trigger.
 317          */
 318 #define HANDLE_STACK(cond) \
 319         do while (cond) { \
 320                 unsigned long addr = *stack++; \
 321                 if (kernel_text_address(addr)) { \
 322                         /* \
 323                          * If the address is either in the text segment of the \
 324                          * kernel, or in the region which contains vmalloc'ed \
 325                          * memory, it *may* be the address of a calling \
 326                          * routine; if so, print it so that someone tracing \
 327                          * down the cause of the crash will be able to figure \
 328                          * out the call path that was taken. \
 329                          */ \
 330                         ops->address(data, addr);   \
 331                 } \
 332         } while (0)
 333
 334         /*
 335          * Print function call entries in all stacks, starting at the
 336          * current stack address. If the stacks consist of nested
 337          * exceptions
 338          */
 339         for (;;) {
 340                 char *id;
 341                 unsigned long *estack_end;
 342                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 343                                                 &used, &id);
 344
 345                 if (estack_end) {
 346                         if (ops->stack(data, id) < 0)
 347                                 break;
 348                         HANDLE_STACK (stack < estack_end);
 349                         ops->stack(data, "<EOE>");
 350                         /*
 351                          * We link to the next stack via the
 352                          * second-to-last pointer (index -2 to end) in the
 353                          * exception stack:
 354                          */
 355                         stack = (unsigned long *) estack_end[-2];
 356                         continue;
 357                 }
 358                 if (irqstack_end) {
 359                         unsigned long *irqstack;
 360                         irqstack = irqstack_end -
 361                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 362
 363                         if (stack >= irqstack && stack < irqstack_end) {
 364                                 if (ops->stack(data, "IRQ") < 0)
 365                                         break;
 366                                 HANDLE_STACK (stack < irqstack_end);
 367                                 /*
 368                                  * We link to the next stack (which would be
 369                                  * the process stack normally) the last
 370                                  * pointer (index -1 to end) in the IRQ stack:
 371                                  */
 372                                 stack = (unsigned long *) (irqstack_end[-1]);
 373                                 irqstack_end = NULL;
 374                                 ops->stack(data, "EOI");
 375                                 continue;
 376                         }
 377                 }
 378                 break;
 379         }
 380
 381         /*
 382          * This handles the process stack:
 383          */
 384         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 385 #undef HANDLE_STACK
 386 }
 387 EXPORT_SYMBOL(dump_trace);
 388
 389 static void
 390 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 391 {
 392         print_symbol(msg, symbol);
 393         printk("\n");
 394 }
 395
 396 static void print_trace_warning(void *data, char *msg)
 397 {
 398         printk("%s\n", msg);
 399 }
 400
 401 static int print_trace_stack(void *data, char *name)
 402 {
 403         printk(" <%s> ", name);
 404         return 0;
 405 }
 406
 407 static void print_trace_address(void *data, unsigned long addr)
 408 {
 409         printk_address(addr);
 410 }
 411
 412 static struct stacktrace_ops print_trace_ops = {
 413         .warning = print_trace_warning,
 414         .warning_symbol = print_trace_warning_symbol,
 415         .stack = print_trace_stack,
 416         .address = print_trace_address,
 417 };
 418
 419 void
 420 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
 421 {
 422         printk("\nCall Trace:\n");
 423         dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
 424         printk("\n");
 425 }
 426
 427 static void
 428 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
 429 {
 430         unsigned long *stack;
 431         int i;
 432         const int cpu = safe_smp_processor_id();
 433         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 434         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 435
 436         // debugging aid: "show_stack(NULL, NULL);" prints the
 437         // back trace for this cpu.
 438
 439         if (rsp == NULL) {
 440                 if (tsk)
 441                         rsp = (unsigned long *)tsk->thread.rsp;
 442                 else
 443                         rsp = (unsigned long *)&rsp;
 444         }
 445
 446         stack = rsp;
 447         for(i=0; i < kstack_depth_to_print; i++) {
 448                 if (stack >= irqstack && stack <= irqstack_end) {
 449                         if (stack == irqstack_end) {
 450                                 stack = (unsigned long *) (irqstack_end[-1]);
 451                                 printk(" <EOI> ");
 452                         }
 453                 } else {
 454                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 455                         break;
 456                 }
 457                 if (i && ((i % 4) == 0))
 458                         printk("\n");
 459                 printk(" %016lx", *stack++);
 460                 touch_nmi_watchdog();
 461         }
 462         show_trace(tsk, regs, rsp);
 463 }
 464
 465 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 466 {
 467         _show_stack(tsk, NULL, rsp);
 468 }
 469
 470 /*
 471  * The architecture-independent dump_stack generator
 472  */
 473 void dump_stack(void)
 474 {
 475         unsigned long dummy;
 476         show_trace(NULL, NULL, &dummy);
 477 }
 478
 479 EXPORT_SYMBOL(dump_stack);
 480
 481 void show_registers(struct pt_regs *regs)
 482 {
 483         int i;
 484         int in_kernel = !user_mode(regs);
 485         unsigned long rsp;
 486         const int cpu = safe_smp_processor_id();
 487         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 488
 489                 rsp = regs->rsp;
 490
 491         printk("CPU %d ", cpu);
 492         __show_regs(regs);
 493         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 494                 cur->comm, cur->pid, task_thread_info(cur), cur);
 495
 496         /*
 497          * When in-kernel, we also print out the stack and code at the
 498          * time of the fault..
 499          */
 500         if (in_kernel) {
 501
 502                 printk("Stack: ");
 503                 _show_stack(NULL, regs, (unsigned long*)rsp);
 504
 505                 printk("\nCode: ");
 506                 if (regs->rip < PAGE_OFFSET)
 507                         goto bad;
 508
 509                 for (i=0; i<20; i++) {
 510                         unsigned char c;
 511                         if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
 512 bad:
 513                                 printk(" Bad RIP value.");
 514                                 break;
 515                         }
 516                         printk("%02x ", c);
 517                 }
 518         }
 519         printk("\n");
 520 }
 521
 522 void handle_BUG(struct pt_regs *regs)
 523 {
 524         struct bug_frame f;
 525         long len;
 526         const char *prefix = "";
 527
 528         if (user_mode(regs))
 529                 return;
 530         if (__copy_from_user(&f, (const void __user *) regs->rip,
 531                              sizeof(struct bug_frame)))
 532                 return;
 533         if (f.filename >= 0 ||
 534             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 535                 return;
 536         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 537         if (len < 0 || len >= PATH_MAX)
 538                 f.filename = (int)(long)"unmapped filename";
 539         else if (len > 50) {
 540                 f.filename += len - 50;
 541                 prefix = "...";
 542         }
 543         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 544         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 545 }
 546
 547 #ifdef CONFIG_BUG
 548 void out_of_line_bug(void)
 549 {
 550         BUG();
 551 }
 552 EXPORT_SYMBOL(out_of_line_bug);
 553 #endif
 554
 555 static DEFINE_SPINLOCK(die_lock);
 556 static int die_owner = -1;
 557 static unsigned int die_nest_count;
 558
 559 unsigned __kprobes long oops_begin(void)
 560 {
 561         int cpu = safe_smp_processor_id();
 562         unsigned long flags;
 563
 564         oops_enter();
 565
 566         /* racy, but better than risking deadlock. */
 567         local_irq_save(flags);
 568         if (!spin_trylock(&die_lock)) {
 569                 if (cpu == die_owner)
 570                         /* nested oops. should stop eventually */;
 571                 else
 572                         spin_lock(&die_lock);
 573         }
 574         die_nest_count++;
 575         die_owner = cpu;
 576         console_verbose();
 577         bust_spinlocks(1);
 578         return flags;
 579 }
 580
 581 void __kprobes oops_end(unsigned long flags)
 582 {
 583         die_owner = -1;
 584         bust_spinlocks(0);
 585         die_nest_count--;
 586         if (die_nest_count)
 587                 /* We still own the lock */
 588                 local_irq_restore(flags);
 589         else
 590                 /* Nest count reaches zero, release the lock. */
 591                 spin_unlock_irqrestore(&die_lock, flags);
 592         if (panic_on_oops)
 593                 panic("Fatal exception");
 594         oops_exit();
 595 }
 596
 597 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 598 {
 599         static int die_counter;
 600         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 601 #ifdef CONFIG_PREEMPT
 602         printk("PREEMPT ");
 603 #endif
 604 #ifdef CONFIG_SMP
 605         printk("SMP ");
 606 #endif
 607 #ifdef CONFIG_DEBUG_PAGEALLOC
 608         printk("DEBUG_PAGEALLOC");
 609 #endif
 610         printk("\n");
 611         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 612         show_registers(regs);
 613         /* Executive summary in case the oops scrolled away */
 614         printk(KERN_ALERT "RIP ");
 615         printk_address(regs->rip);
 616         printk(" RSP <%016lx>\n", regs->rsp);
 617         if (kexec_should_crash(current))
 618                 crash_kexec(regs);
 619 }
 620
 621 void die(const char * str, struct pt_regs * regs, long err)
 622 {
 623         unsigned long flags = oops_begin();
 624
 625         handle_BUG(regs);
 626         __die(str, regs, err);
 627         oops_end(flags);
 628         do_exit(SIGSEGV);
 629 }
 630
 631 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 632 {
 633         unsigned long flags = oops_begin();
 634
 635         /*
 636          * We are in trouble anyway, lets at least try
 637          * to get a message out.
 638          */
 639         printk(str, safe_smp_processor_id());
 640         show_registers(regs);
 641         if (kexec_should_crash(current))
 642                 crash_kexec(regs);
 643         if (do_panic || panic_on_oops)
 644                 panic("Non maskable interrupt");
 645         oops_end(flags);
 646         nmi_exit();
 647         local_irq_enable();
 648         do_exit(SIGSEGV);
 649 }
 650
 651 static void __kprobes do_trap(int trapnr, int signr, char *str,
 652                               struct pt_regs * regs, long error_code,
 653                               siginfo_t *info)
 654 {
 655         struct task_struct *tsk = current;
 656
 657         tsk->thread.error_code = error_code;
 658         tsk->thread.trap_no = trapnr;
 659
 660         if (user_mode(regs)) {
 661                 if (exception_trace && unhandled_signal(tsk, signr))
 662                         printk(KERN_INFO
 663                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 664                                tsk->comm, tsk->pid, str,
 665                                regs->rip, regs->rsp, error_code);
 666
 667                 if (info)
 668                         force_sig_info(signr, info, tsk);
 669                 else
 670                         force_sig(signr, tsk);
 671                 return;
 672         }
 673
 674
 675         /* kernel trap */
 676         {
 677                 const struct exception_table_entry *fixup;
 678                 fixup = search_exception_tables(regs->rip);
 679                 if (fixup)
 680                         regs->rip = fixup->fixup;
 681                 else
 682                         die(str, regs, error_code);
 683                 return;
 684         }
 685 }
 686
 687 #define DO_ERROR(trapnr, signr, str, name) \
 688 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 689 { \
 690         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 691                                                         == NOTIFY_STOP) \
 692                 return; \
 693         conditional_sti(regs);                                          \
 694         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 695 }
 696
 697 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 698 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 699 { \
 700         siginfo_t info; \
 701         info.si_signo = signr; \
 702         info.si_errno = 0; \
 703         info.si_code = sicode; \
 704         info.si_addr = (void __user *)siaddr; \
 705         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 706                                                         == NOTIFY_STOP) \
 707                 return; \
 708         conditional_sti(regs);                                          \
 709         do_trap(trapnr, signr, str, regs, error_code, &info); \
 710 }
 711
 712 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 713 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 714 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 715 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 716 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 717 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 718 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 719 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 720 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 721 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 722
 723 /* Runs on IST stack */
 724 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 725 {
 726         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 727                         12, SIGBUS) == NOTIFY_STOP)
 728                 return;
 729         preempt_conditional_sti(regs);
 730         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 731         preempt_conditional_cli(regs);
 732 }
 733
 734 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 735 {
 736         static const char str[] = "double fault";
 737         struct task_struct *tsk = current;
 738
 739         /* Return not checked because double check cannot be ignored */
 740         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 741
 742         tsk->thread.error_code = error_code;
 743         tsk->thread.trap_no = 8;
 744
 745         /* This is always a kernel trap and never fixable (and thus must
 746            never return). */
 747         for (;;)
 748                 die(str, regs, error_code);
 749 }
 750
 751 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 752                                                 long error_code)
 753 {
 754         struct task_struct *tsk = current;
 755
 756         conditional_sti(regs);
 757
 758         tsk->thread.error_code = error_code;
 759         tsk->thread.trap_no = 13;
 760
 761         if (user_mode(regs)) {
 762                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 763                         printk(KERN_INFO
 764                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 765                                tsk->comm, tsk->pid,
 766                                regs->rip, regs->rsp, error_code);
 767
 768                 force_sig(SIGSEGV, tsk);
 769                 return;
 770         }
 771
 772         /* kernel gp */
 773         {
 774                 const struct exception_table_entry *fixup;
 775                 fixup = search_exception_tables(regs->rip);
 776                 if (fixup) {
 777                         regs->rip = fixup->fixup;
 778                         return;
 779                 }
 780                 if (notify_die(DIE_GPF, "general protection fault", regs,
 781                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 782                         return;
 783                 die("general protection fault", regs, error_code);
 784         }
 785 }
 786
 787 static __kprobes void
 788 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 789 {
 790         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 791                 reason);
 792         printk(KERN_EMERG "You probably have a hardware problem with your "
 793                 "RAM chips\n");
 794
 795         if (panic_on_unrecovered_nmi)
 796                 panic("NMI: Not continuing");
 797
 798         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 799
 800         /* Clear and disable the memory parity error line. */
 801         reason = (reason & 0xf) | 4;
 802         outb(reason, 0x61);
 803 }
 804
 805 static __kprobes void
 806 io_check_error(unsigned char reason, struct pt_regs * regs)
 807 {
 808         printk("NMI: IOCK error (debug interrupt?)\n");
 809         show_registers(regs);
 810
 811         /* Re-enable the IOCK line, wait for a few seconds */
 812         reason = (reason & 0xf) | 8;
 813         outb(reason, 0x61);
 814         mdelay(2000);
 815         reason &= ~8;
 816         outb(reason, 0x61);
 817 }
 818
 819 static __kprobes void
 820 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 821 {
 822         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
 823                 reason);
 824         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 825
 826         if (panic_on_unrecovered_nmi)
 827                 panic("NMI: Not continuing");
 828
 829         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 830 }
 831
 832 /* Runs on IST stack. This code must keep interrupts off all the time.
 833    Nested NMIs are prevented by the CPU. */
 834 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 835 {
 836         unsigned char reason = 0;
 837         int cpu;
 838
 839         cpu = smp_processor_id();
 840
 841         /* Only the BSP gets external NMIs from the system.  */
 842         if (!cpu)
 843                 reason = get_nmi_reason();
 844
 845         if (!(reason & 0xc0)) {
 846                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 847                                                                 == NOTIFY_STOP)
 848                         return;
 849                 /*
 850                  * Ok, so this is none of the documented NMI sources,
 851                  * so it must be the NMI watchdog.
 852                  */
 853                 if (nmi_watchdog_tick(regs,reason))
 854                         return;
 855                 if (!do_nmi_callback(regs,cpu))
 856                         unknown_nmi_error(reason, regs);
 857
 858                 return;
 859         }
 860         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 861                 return;
 862
 863         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 864
 865         if (reason & 0x80)
 866                 mem_parity_error(reason, regs);
 867         if (reason & 0x40)
 868                 io_check_error(reason, regs);
 869 }
 870
 871 /* runs on IST stack. */
 872 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 873 {
 874         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 875                 return;
 876         }
 877         preempt_conditional_sti(regs);
 878         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 879         preempt_conditional_cli(regs);
 880 }
 881
 882 /* Help handler running on IST stack to switch back to user stack
 883    for scheduling or signal handling. The actual stack switch is done in
 884    entry.S */
 885 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 886 {
 887         struct pt_regs *regs = eregs;
 888         /* Did already sync */
 889         if (eregs == (struct pt_regs *)eregs->rsp)
 890                 ;
 891         /* Exception from user space */
 892         else if (user_mode(eregs))
 893                 regs = task_pt_regs(current);
 894         /* Exception from kernel and interrupts are enabled. Move to
 895            kernel process stack. */
 896         else if (eregs->eflags & X86_EFLAGS_IF)
 897                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 898         if (eregs != regs)
 899                 *regs = *eregs;
 900         return regs;
 901 }
 902
 903 /* runs on IST stack. */
 904 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 905                                    unsigned long error_code)
 906 {
 907         unsigned long condition;
 908         struct task_struct *tsk = current;
 909         siginfo_t info;
 910
 911         get_debugreg(condition, 6);
 912
 913         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 914                                                 SIGTRAP) == NOTIFY_STOP)
 915                 return;
 916
 917         preempt_conditional_sti(regs);
 918
 919         /* Mask out spurious debug traps due to lazy DR7 setting */
 920         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 921                 if (!tsk->thread.debugreg7) {
 922                         goto clear_dr7;
 923                 }
 924         }
 925
 926         tsk->thread.debugreg6 = condition;
 927
 928         /* Mask out spurious TF errors due to lazy TF clearing */
 929         if (condition & DR_STEP) {
 930                 /*
 931                  * The TF error should be masked out only if the current
 932                  * process is not traced and if the TRAP flag has been set
 933                  * previously by a tracing process (condition detected by
 934                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 935                  * can be modified by the process itself in user mode,
 936                  * allowing programs to debug themselves without the ptrace()
 937                  * interface.
 938                  */
 939                 if (!user_mode(regs))
 940                        goto clear_TF_reenable;
 941                 /*
 942                  * Was the TF flag set by a debugger? If so, clear it now,
 943                  * so that register information is correct.
 944                  */
 945                 if (tsk->ptrace & PT_DTRACE) {
 946                         regs->eflags &= ~TF_MASK;
 947                         tsk->ptrace &= ~PT_DTRACE;
 948                 }
 949         }
 950
 951         /* Ok, finally something we can handle */
 952         tsk->thread.trap_no = 1;
 953         tsk->thread.error_code = error_code;
 954         info.si_signo = SIGTRAP;
 955         info.si_errno = 0;
 956         info.si_code = TRAP_BRKPT;
 957         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 958         force_sig_info(SIGTRAP, &info, tsk);
 959
 960 clear_dr7:
 961         set_debugreg(0UL, 7);
 962         preempt_conditional_cli(regs);
 963         return;
 964
 965 clear_TF_reenable:
 966         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 967         regs->eflags &= ~TF_MASK;
 968         preempt_conditional_cli(regs);
 969 }
 970
 971 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 972 {
 973         const struct exception_table_entry *fixup;
 974         fixup = search_exception_tables(regs->rip);
 975         if (fixup) {
 976                 regs->rip = fixup->fixup;
 977                 return 1;
 978         }
 979         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 980         /* Illegal floating point operation in the kernel */
 981         current->thread.trap_no = trapnr;
 982         die(str, regs, 0);
 983         return 0;
 984 }
 985
 986 /*
 987  * Note that we play around with the 'TS' bit in an attempt to get
 988  * the correct behaviour even in the presence of the asynchronous
 989  * IRQ13 behaviour
 990  */
 991 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 992 {
 993         void __user *rip = (void __user *)(regs->rip);
 994         struct task_struct * task;
 995         siginfo_t info;
 996         unsigned short cwd, swd;
 997
 998         conditional_sti(regs);
 999         if (!user_mode(regs) &&
1000             kernel_math_error(regs, "kernel x87 math error", 16))
1001                 return;
1002
1003         /*
1004          * Save the info for the exception handler and clear the error.
1005          */
1006         task = current;
1007         save_init_fpu(task);
1008         task->thread.trap_no = 16;
1009         task->thread.error_code = 0;
1010         info.si_signo = SIGFPE;
1011         info.si_errno = 0;
1012         info.si_code = __SI_FAULT;
1013         info.si_addr = rip;
1014         /*
1015          * (~cwd & swd) will mask out exceptions that are not set to unmasked
1016          * status.  0x3f is the exception bits in these regs, 0x200 is the
1017          * C1 reg you need in case of a stack fault, 0x040 is the stack
1018          * fault bit.  We should only be taking one exception at a time,
1019          * so if this combination doesn't produce any single exception,
1020          * then we have a bad program that isn't synchronizing its FPU usage
1021          * and it will suffer the consequences since we won't be able to
1022          * fully reproduce the context of the exception
1023          */
1024         cwd = get_fpu_cwd(task);
1025         swd = get_fpu_swd(task);
1026         switch (swd & ~cwd & 0x3f) {
1027                 case 0x000:
1028                 default:
1029                         break;
1030                 case 0x001: /* Invalid Op */
1031                         /*
1032                          * swd & 0x240 == 0x040: Stack Underflow
1033                          * swd & 0x240 == 0x240: Stack Overflow
1034                          * User must clear the SF bit (0x40) if set
1035                          */
1036                         info.si_code = FPE_FLTINV;
1037                         break;
1038                 case 0x002: /* Denormalize */
1039                 case 0x010: /* Underflow */
1040                         info.si_code = FPE_FLTUND;
1041                         break;
1042                 case 0x004: /* Zero Divide */
1043                         info.si_code = FPE_FLTDIV;
1044                         break;
1045                 case 0x008: /* Overflow */
1046                         info.si_code = FPE_FLTOVF;
1047                         break;
1048                 case 0x020: /* Precision */
1049                         info.si_code = FPE_FLTRES;
1050                         break;
1051         }
1052         force_sig_info(SIGFPE, &info, task);
1053 }
1054
1055 asmlinkage void bad_intr(void)
1056 {
1057         printk("bad interrupt");
1058 }
1059
1060 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1061 {
1062         void __user *rip = (void __user *)(regs->rip);
1063         struct task_struct * task;
1064         siginfo_t info;
1065         unsigned short mxcsr;
1066
1067         conditional_sti(regs);
1068         if (!user_mode(regs) &&
1069                 kernel_math_error(regs, "kernel simd math error", 19))
1070                 return;
1071
1072         /*
1073          * Save the info for the exception handler and clear the error.
1074          */
1075         task = current;
1076         save_init_fpu(task);
1077         task->thread.trap_no = 19;
1078         task->thread.error_code = 0;
1079         info.si_signo = SIGFPE;
1080         info.si_errno = 0;
1081         info.si_code = __SI_FAULT;
1082         info.si_addr = rip;
1083         /*
1084          * The SIMD FPU exceptions are handled a little differently, as there
1085          * is only a single status/control register.  Thus, to determine which
1086          * unmasked exception was caught we must mask the exception mask bits
1087          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1088          */
1089         mxcsr = get_fpu_mxcsr(task);
1090         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1091                 case 0x000:
1092                 default:
1093                         break;
1094                 case 0x001: /* Invalid Op */
1095                         info.si_code = FPE_FLTINV;
1096                         break;
1097                 case 0x002: /* Denormalize */
1098                 case 0x010: /* Underflow */
1099                         info.si_code = FPE_FLTUND;
1100                         break;
1101                 case 0x004: /* Zero Divide */
1102                         info.si_code = FPE_FLTDIV;
1103                         break;
1104                 case 0x008: /* Overflow */
1105                         info.si_code = FPE_FLTOVF;
1106                         break;
1107                 case 0x020: /* Precision */
1108                         info.si_code = FPE_FLTRES;
1109                         break;
1110         }
1111         force_sig_info(SIGFPE, &info, task);
1112 }
1113
1114 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1115 {
1116 }
1117
1118 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1119 {
1120 }
1121
1122 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1123 {
1124 }
1125
1126 /*
1127  *  'math_state_restore()' saves the current math information in the
1128  * old math state array, and gets the new ones from the current task
1129  *
1130  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1131  * Don't touch unless you *really* know how it works.
1132  */
1133 asmlinkage void math_state_restore(void)
1134 {
1135         struct task_struct *me = current;
1136         clts();                 /* Allow maths ops (or we recurse) */
1137
1138         if (!used_math())
1139                 init_fpu(me);
1140         restore_fpu_checking(&me->thread.i387.fxsave);
1141         task_thread_info(me)->status |= TS_USEDFPU;
1142         me->fpu_counter++;
1143 }
1144
1145 void __init trap_init(void)
1146 {
1147         set_intr_gate(0,&divide_error);
1148         set_intr_gate_ist(1,&debug,DEBUG_STACK);
1149         set_intr_gate_ist(2,&nmi,NMI_STACK);
1150         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1151         set_system_gate(4,&overflow);   /* int4 can be called from all */
1152         set_intr_gate(5,&bounds);
1153         set_intr_gate(6,&invalid_op);
1154         set_intr_gate(7,&device_not_available);
1155         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1156         set_intr_gate(9,&coprocessor_segment_overrun);
1157         set_intr_gate(10,&invalid_TSS);
1158         set_intr_gate(11,&segment_not_present);
1159         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1160         set_intr_gate(13,&general_protection);
1161         set_intr_gate(14,&page_fault);
1162         set_intr_gate(15,&spurious_interrupt_bug);
1163         set_intr_gate(16,&coprocessor_error);
1164         set_intr_gate(17,&alignment_check);
1165 #ifdef CONFIG_X86_MCE
1166         set_intr_gate_ist(18,&machine_check, MCE_STACK);
1167 #endif
1168         set_intr_gate(19,&simd_coprocessor_error);
1169
1170 #ifdef CONFIG_IA32_EMULATION
1171         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1172 #endif
1173
1174         /*
1175          * Should be a barrier for any external CPU state.
1176          */
1177         cpu_init();
1178 }
1179
1180
1181 static int __init oops_setup(char *s)
1182 {
1183         if (!s)
1184                 return -EINVAL;
1185         if (!strcmp(s, "panic"))
1186                 panic_on_oops = 1;
1187         return 0;
1188 }
1189 early_param("oops", oops_setup);
1190
1191 static int __init kstack_setup(char *s)
1192 {
1193         if (!s)
1194                 return -EINVAL;
1195         kstack_depth_to_print = simple_strtoul(s,NULL,0);
1196         return 0;
1197 }
1198 early_param("kstack", kstack_setup);
1199
1200 #ifdef CONFIG_STACK_UNWIND
1201 static int __init call_trace_setup(char *s)
1202 {
1203         if (!s)
1204                 return -EINVAL;
1205         if (strcmp(s, "old") == 0)
1206                 call_trace = -1;
1207         else if (strcmp(s, "both") == 0)
1208                 call_trace = 0;
1209         else if (strcmp(s, "newfallback") == 0)
1210                 call_trace = 1;
1211         else if (strcmp(s, "new") == 0)
1212                 call_trace = 2;
1213         return 0;
1214 }
1215 early_param("call_trace", call_trace_setup);
1216 #endif