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  *  $Id: traps.c,v 1.36 2002/03/24 11:09:10 ak Exp $
  11  */
  12
  13 /*
  14  * 'Traps.c' handles hardware traps and faults after we have saved some
  15  * state in 'entry.S'.
  16  */
  17 #include <linux/config.h>
  18 #include <linux/sched.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/timer.h>
  24 #include <linux/mm.h>
  25 #include <linux/init.h>
  26 #include <linux/delay.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/module.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/nmi.h>
  32 #include <linux/kprobes.h>
  33
  34 #include <asm/system.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/io.h>
  37 #include <asm/atomic.h>
  38 #include <asm/debugreg.h>
  39 #include <asm/desc.h>
  40 #include <asm/i387.h>
  41 #include <asm/kdebug.h>
  42 #include <asm/processor.h>
  43
  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
  50 extern struct gate_struct idt_table[256];
  51
  52 asmlinkage void divide_error(void);
  53 asmlinkage void debug(void);
  54 asmlinkage void nmi(void);
  55 asmlinkage void int3(void);
  56 asmlinkage void overflow(void);
  57 asmlinkage void bounds(void);
  58 asmlinkage void invalid_op(void);
  59 asmlinkage void device_not_available(void);
  60 asmlinkage void double_fault(void);
  61 asmlinkage void coprocessor_segment_overrun(void);
  62 asmlinkage void invalid_TSS(void);
  63 asmlinkage void segment_not_present(void);
  64 asmlinkage void stack_segment(void);
  65 asmlinkage void general_protection(void);
  66 asmlinkage void page_fault(void);
  67 asmlinkage void coprocessor_error(void);
  68 asmlinkage void simd_coprocessor_error(void);
  69 asmlinkage void reserved(void);
  70 asmlinkage void alignment_check(void);
  71 asmlinkage void machine_check(void);
  72 asmlinkage void spurious_interrupt_bug(void);
  73
  74 struct notifier_block *die_chain;
  75 static DEFINE_SPINLOCK(die_notifier_lock);
  76
  77 int register_die_notifier(struct notifier_block *nb)
  78 {
  79         int err = 0;
  80         unsigned long flags;
  81
  82         vmalloc_sync_all();
  83         spin_lock_irqsave(&die_notifier_lock, flags);
  84         err = notifier_chain_register(&die_chain, nb);
  85         spin_unlock_irqrestore(&die_notifier_lock, flags);
  86         return err;
  87 }
  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         preempt_enable_no_resched();
 107 }
 108
 109 static int kstack_depth_to_print = 10;
 110
 111 #ifdef CONFIG_KALLSYMS
 112 #include <linux/kallsyms.h>
 113 int printk_address(unsigned long address)
 114 {
 115         unsigned long offset = 0, symsize;
 116         const char *symname;
 117         char *modname;
 118         char *delim = ":";
 119         char namebuf[128];
 120
 121         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 122         if (!symname)
 123                 return printk("[<%016lx>]", address);
 124         if (!modname)
 125                 modname = delim = "";
 126         return printk("<%016lx>{%s%s%s%s%+ld}",
 127                       address,delim,modname,delim,symname,offset);
 128 }
 129 #else
 130 int printk_address(unsigned long address)
 131 {
 132         return printk("[<%016lx>]", address);
 133 }
 134 #endif
 135
 136 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 137                                         unsigned *usedp, const char **idp)
 138 {
 139         static char ids[][8] = {
 140                 [DEBUG_STACK - 1] = "#DB",
 141                 [NMI_STACK - 1] = "NMI",
 142                 [DOUBLEFAULT_STACK - 1] = "#DF",
 143                 [STACKFAULT_STACK - 1] = "#SS",
 144                 [MCE_STACK - 1] = "#MC",
 145 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 146                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 147 #endif
 148         };
 149         unsigned k;
 150
 151         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 152                 unsigned long end;
 153
 154                 switch (k + 1) {
 155 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 156                 case DEBUG_STACK:
 157                         end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
 158                         break;
 159 #endif
 160                 default:
 161                         end = per_cpu(init_tss, cpu).ist[k];
 162                         break;
 163                 }
 164                 if (stack >= end)
 165                         continue;
 166                 if (stack >= end - EXCEPTION_STKSZ) {
 167                         if (*usedp & (1U << k))
 168                                 break;
 169                         *usedp |= 1U << k;
 170                         *idp = ids[k];
 171                         return (unsigned long *)end;
 172                 }
 173 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 174                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 175                         unsigned j = N_EXCEPTION_STACKS - 1;
 176
 177                         do {
 178                                 ++j;
 179                                 end -= EXCEPTION_STKSZ;
 180                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 181                         } while (stack < end - EXCEPTION_STKSZ);
 182                         if (*usedp & (1U << j))
 183                                 break;
 184                         *usedp |= 1U << j;
 185                         *idp = ids[j];
 186                         return (unsigned long *)end;
 187                 }
 188 #endif
 189         }
 190         return NULL;
 191 }
 192
 193 /*
 194  * x86-64 can have upto three kernel stacks:
 195  * process stack
 196  * interrupt stack
 197  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 198  */
 199
 200 void show_trace(unsigned long *stack)
 201 {
 202         const unsigned cpu = safe_smp_processor_id();
 203         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 204         int i;
 205         unsigned used = 0;
 206
 207         printk("\nCall Trace:");
 208
 209 #define HANDLE_STACK(cond) \
 210         do while (cond) { \
 211                 unsigned long addr = *stack++; \
 212                 if (kernel_text_address(addr)) { \
 213                         if (i > 50) { \
 214                                 printk("\n       "); \
 215                                 i = 0; \
 216                         } \
 217                         else \
 218                                 i += printk(" "); \
 219                         /* \
 220                          * If the address is either in the text segment of the \
 221                          * kernel, or in the region which contains vmalloc'ed \
 222                          * memory, it *may* be the address of a calling \
 223                          * routine; if so, print it so that someone tracing \
 224                          * down the cause of the crash will be able to figure \
 225                          * out the call path that was taken. \
 226                          */ \
 227                         i += printk_address(addr); \
 228                 } \
 229         } while (0)
 230
 231         for(i = 11; ; ) {
 232                 const char *id;
 233                 unsigned long *estack_end;
 234                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 235                                                 &used, &id);
 236
 237                 if (estack_end) {
 238                         i += printk(" <%s>", id);
 239                         HANDLE_STACK (stack < estack_end);
 240                         i += printk(" <EOE>");
 241                         stack = (unsigned long *) estack_end[-2];
 242                         continue;
 243                 }
 244                 if (irqstack_end) {
 245                         unsigned long *irqstack;
 246                         irqstack = irqstack_end -
 247                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 248
 249                         if (stack >= irqstack && stack < irqstack_end) {
 250                                 i += printk(" <IRQ>");
 251                                 HANDLE_STACK (stack < irqstack_end);
 252                                 stack = (unsigned long *) (irqstack_end[-1]);
 253                                 irqstack_end = NULL;
 254                                 i += printk(" <EOI>");
 255                                 continue;
 256                         }
 257                 }
 258                 break;
 259         }
 260
 261         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 262 #undef HANDLE_STACK
 263         printk("\n");
 264 }
 265
 266 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 267 {
 268         unsigned long *stack;
 269         int i;
 270         const int cpu = safe_smp_processor_id();
 271         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 272         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 273
 274         // debugging aid: "show_stack(NULL, NULL);" prints the
 275         // back trace for this cpu.
 276
 277         if (rsp == NULL) {
 278                 if (tsk)
 279                         rsp = (unsigned long *)tsk->thread.rsp;
 280                 else
 281                         rsp = (unsigned long *)&rsp;
 282         }
 283
 284         stack = rsp;
 285         for(i=0; i < kstack_depth_to_print; i++) {
 286                 if (stack >= irqstack && stack <= irqstack_end) {
 287                         if (stack == irqstack_end) {
 288                                 stack = (unsigned long *) (irqstack_end[-1]);
 289                                 printk(" <EOI> ");
 290                         }
 291                 } else {
 292                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 293                         break;
 294                 }
 295                 if (i && ((i % 4) == 0))
 296                         printk("\n       ");
 297                 printk("%016lx ", *stack++);
 298                 touch_nmi_watchdog();
 299         }
 300         show_trace((unsigned long *)rsp);
 301 }
 302
 303 /*
 304  * The architecture-independent dump_stack generator
 305  */
 306 void dump_stack(void)
 307 {
 308         unsigned long dummy;
 309         show_trace(&dummy);
 310 }
 311
 312 EXPORT_SYMBOL(dump_stack);
 313
 314 void show_registers(struct pt_regs *regs)
 315 {
 316         int i;
 317         int in_kernel = !user_mode(regs);
 318         unsigned long rsp;
 319         const int cpu = safe_smp_processor_id();
 320         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 321
 322                 rsp = regs->rsp;
 323
 324         printk("CPU %d ", cpu);
 325         __show_regs(regs);
 326         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 327                 cur->comm, cur->pid, task_thread_info(cur), cur);
 328
 329         /*
 330          * When in-kernel, we also print out the stack and code at the
 331          * time of the fault..
 332          */
 333         if (in_kernel) {
 334
 335                 printk("Stack: ");
 336                 show_stack(NULL, (unsigned long*)rsp);
 337
 338                 printk("\nCode: ");
 339                 if(regs->rip < PAGE_OFFSET)
 340                         goto bad;
 341
 342                 for(i=0;i<20;i++)
 343                 {
 344                         unsigned char c;
 345                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 346 bad:
 347                                 printk(" Bad RIP value.");
 348                                 break;
 349                         }
 350                         printk("%02x ", c);
 351                 }
 352         }
 353         printk("\n");
 354 }
 355
 356 void handle_BUG(struct pt_regs *regs)
 357 {
 358         struct bug_frame f;
 359         long len;
 360         const char *prefix = "";
 361
 362         if (user_mode(regs))
 363                 return;
 364         if (__copy_from_user(&f, (const void __user *) regs->rip,
 365                              sizeof(struct bug_frame)))
 366                 return;
 367         if (f.filename >= 0 ||
 368             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 369                 return;
 370         len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
 371         if (len < 0 || len >= PATH_MAX)
 372                 f.filename = (int)(long)"unmapped filename";
 373         else if (len > 50) {
 374                 f.filename += len - 50;
 375                 prefix = "...";
 376         }
 377         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 378         printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
 379 }
 380
 381 #ifdef CONFIG_BUG
 382 void out_of_line_bug(void)
 383 {
 384         BUG();
 385 }
 386 #endif
 387
 388 static DEFINE_SPINLOCK(die_lock);
 389 static int die_owner = -1;
 390
 391 unsigned __kprobes long oops_begin(void)
 392 {
 393         int cpu = safe_smp_processor_id();
 394         unsigned long flags;
 395
 396         /* racy, but better than risking deadlock. */
 397         local_irq_save(flags);
 398         if (!spin_trylock(&die_lock)) {
 399                 if (cpu == die_owner)
 400                         /* nested oops. should stop eventually */;
 401                 else
 402                         spin_lock(&die_lock);
 403         }
 404         die_owner = cpu;
 405         console_verbose();
 406         bust_spinlocks(1);
 407         return flags;
 408 }
 409
 410 void __kprobes oops_end(unsigned long flags)
 411 {
 412         die_owner = -1;
 413         bust_spinlocks(0);
 414         spin_unlock_irqrestore(&die_lock, flags);
 415         if (panic_on_oops)
 416                 panic("Oops");
 417 }
 418
 419 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
 420 {
 421         static int die_counter;
 422         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 423 #ifdef CONFIG_PREEMPT
 424         printk("PREEMPT ");
 425 #endif
 426 #ifdef CONFIG_SMP
 427         printk("SMP ");
 428 #endif
 429 #ifdef CONFIG_DEBUG_PAGEALLOC
 430         printk("DEBUG_PAGEALLOC");
 431 #endif
 432         printk("\n");
 433         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 434         show_registers(regs);
 435         /* Executive summary in case the oops scrolled away */
 436         printk(KERN_ALERT "RIP ");
 437         printk_address(regs->rip);
 438         printk(" RSP <%016lx>\n", regs->rsp);
 439 }
 440
 441 void die(const char * str, struct pt_regs * regs, long err)
 442 {
 443         unsigned long flags = oops_begin();
 444
 445         handle_BUG(regs);
 446         __die(str, regs, err);
 447         oops_end(flags);
 448         do_exit(SIGSEGV);
 449 }
 450
 451 void __kprobes die_nmi(char *str, struct pt_regs *regs)
 452 {
 453         unsigned long flags = oops_begin();
 454
 455         /*
 456          * We are in trouble anyway, lets at least try
 457          * to get a message out.
 458          */
 459         printk(str, safe_smp_processor_id());
 460         show_registers(regs);
 461         if (panic_on_timeout || panic_on_oops)
 462                 panic("nmi watchdog");
 463         printk("console shuts up ...\n");
 464         oops_end(flags);
 465         do_exit(SIGSEGV);
 466 }
 467
 468 static void __kprobes do_trap(int trapnr, int signr, char *str,
 469                               struct pt_regs * regs, long error_code,
 470                               siginfo_t *info)
 471 {
 472         struct task_struct *tsk = current;
 473
 474         conditional_sti(regs);
 475
 476         tsk->thread.error_code = error_code;
 477         tsk->thread.trap_no = trapnr;
 478
 479         if (user_mode(regs)) {
 480                 if (exception_trace && unhandled_signal(tsk, signr))
 481                         printk(KERN_INFO
 482                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 483                                tsk->comm, tsk->pid, str,
 484                                regs->rip,regs->rsp,error_code);
 485
 486                 if (info)
 487                         force_sig_info(signr, info, tsk);
 488                 else
 489                         force_sig(signr, tsk);
 490                 return;
 491         }
 492
 493
 494         /* kernel trap */
 495         {
 496                 const struct exception_table_entry *fixup;
 497                 fixup = search_exception_tables(regs->rip);
 498                 if (fixup) {
 499                         regs->rip = fixup->fixup;
 500                 } else
 501                         die(str, regs, error_code);
 502                 return;
 503         }
 504 }
 505
 506 #define DO_ERROR(trapnr, signr, str, name) \
 507 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 508 { \
 509         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 510                                                         == NOTIFY_STOP) \
 511                 return; \
 512         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 513 }
 514
 515 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 516 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 517 { \
 518         siginfo_t info; \
 519         info.si_signo = signr; \
 520         info.si_errno = 0; \
 521         info.si_code = sicode; \
 522         info.si_addr = (void __user *)siaddr; \
 523         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 524                                                         == NOTIFY_STOP) \
 525                 return; \
 526         do_trap(trapnr, signr, str, regs, error_code, &info); \
 527 }
 528
 529 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 530 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 531 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 532 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 533 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 534 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 535 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 536 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 537 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 538 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 539 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 540
 541 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 542 {
 543         static const char str[] = "double fault";
 544         struct task_struct *tsk = current;
 545
 546         /* Return not checked because double check cannot be ignored */
 547         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 548
 549         tsk->thread.error_code = error_code;
 550         tsk->thread.trap_no = 8;
 551
 552         /* This is always a kernel trap and never fixable (and thus must
 553            never return). */
 554         for (;;)
 555                 die(str, regs, error_code);
 556 }
 557
 558 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 559                                                 long error_code)
 560 {
 561         struct task_struct *tsk = current;
 562
 563         conditional_sti(regs);
 564
 565         tsk->thread.error_code = error_code;
 566         tsk->thread.trap_no = 13;
 567
 568         if (user_mode(regs)) {
 569                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 570                         printk(KERN_INFO
 571                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 572                                tsk->comm, tsk->pid,
 573                                regs->rip,regs->rsp,error_code);
 574
 575                 force_sig(SIGSEGV, tsk);
 576                 return;
 577         }
 578
 579         /* kernel gp */
 580         {
 581                 const struct exception_table_entry *fixup;
 582                 fixup = search_exception_tables(regs->rip);
 583                 if (fixup) {
 584                         regs->rip = fixup->fixup;
 585                         return;
 586                 }
 587                 if (notify_die(DIE_GPF, "general protection fault", regs,
 588                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 589                         return;
 590                 die("general protection fault", regs, error_code);
 591         }
 592 }
 593
 594 static __kprobes void
 595 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 596 {
 597         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 598         printk("You probably have a hardware problem with your RAM chips\n");
 599
 600         /* Clear and disable the memory parity error line. */
 601         reason = (reason & 0xf) | 4;
 602         outb(reason, 0x61);
 603 }
 604
 605 static __kprobes void
 606 io_check_error(unsigned char reason, struct pt_regs * regs)
 607 {
 608         printk("NMI: IOCK error (debug interrupt?)\n");
 609         show_registers(regs);
 610
 611         /* Re-enable the IOCK line, wait for a few seconds */
 612         reason = (reason & 0xf) | 8;
 613         outb(reason, 0x61);
 614         mdelay(2000);
 615         reason &= ~8;
 616         outb(reason, 0x61);
 617 }
 618
 619 static __kprobes void
 620 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 621 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 622         printk("Dazed and confused, but trying to continue\n");
 623         printk("Do you have a strange power saving mode enabled?\n");
 624 }
 625
 626 /* Runs on IST stack. This code must keep interrupts off all the time.
 627    Nested NMIs are prevented by the CPU. */
 628 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
 629 {
 630         unsigned char reason = 0;
 631         int cpu;
 632
 633         cpu = smp_processor_id();
 634
 635         /* Only the BSP gets external NMIs from the system.  */
 636         if (!cpu)
 637                 reason = get_nmi_reason();
 638
 639         if (!(reason & 0xc0)) {
 640                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 641                                                                 == NOTIFY_STOP)
 642                         return;
 643 #ifdef CONFIG_X86_LOCAL_APIC
 644                 /*
 645                  * Ok, so this is none of the documented NMI sources,
 646                  * so it must be the NMI watchdog.
 647                  */
 648                 if (nmi_watchdog > 0) {
 649                         nmi_watchdog_tick(regs,reason);
 650                         return;
 651                 }
 652 #endif
 653                 unknown_nmi_error(reason, regs);
 654                 return;
 655         }
 656         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 657                 return;
 658
 659         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 660
 661         if (reason & 0x80)
 662                 mem_parity_error(reason, regs);
 663         if (reason & 0x40)
 664                 io_check_error(reason, regs);
 665 }
 666
 667 /* runs on IST stack. */
 668 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 669 {
 670         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 671                 return;
 672         }
 673         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 674         return;
 675 }
 676
 677 /* Help handler running on IST stack to switch back to user stack
 678    for scheduling or signal handling. The actual stack switch is done in
 679    entry.S */
 680 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 681 {
 682         struct pt_regs *regs = eregs;
 683         /* Did already sync */
 684         if (eregs == (struct pt_regs *)eregs->rsp)
 685                 ;
 686         /* Exception from user space */
 687         else if (user_mode(eregs))
 688                 regs = task_pt_regs(current);
 689         /* Exception from kernel and interrupts are enabled. Move to
 690            kernel process stack. */
 691         else if (eregs->eflags & X86_EFLAGS_IF)
 692                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 693         if (eregs != regs)
 694                 *regs = *eregs;
 695         return regs;
 696 }
 697
 698 /* runs on IST stack. */
 699 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 700                                    unsigned long error_code)
 701 {
 702         unsigned long condition;
 703         struct task_struct *tsk = current;
 704         siginfo_t info;
 705
 706         get_debugreg(condition, 6);
 707
 708         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 709                                                 SIGTRAP) == NOTIFY_STOP)
 710                 return;
 711
 712         preempt_conditional_sti(regs);
 713
 714         /* Mask out spurious debug traps due to lazy DR7 setting */
 715         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 716                 if (!tsk->thread.debugreg7) {
 717                         goto clear_dr7;
 718                 }
 719         }
 720
 721         tsk->thread.debugreg6 = condition;
 722
 723         /* Mask out spurious TF errors due to lazy TF clearing */
 724         if (condition & DR_STEP) {
 725                 /*
 726                  * The TF error should be masked out only if the current
 727                  * process is not traced and if the TRAP flag has been set
 728                  * previously by a tracing process (condition detected by
 729                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 730                  * can be modified by the process itself in user mode,
 731                  * allowing programs to debug themselves without the ptrace()
 732                  * interface.
 733                  */
 734                 if (!user_mode(regs))
 735                        goto clear_TF_reenable;
 736                 /*
 737                  * Was the TF flag set by a debugger? If so, clear it now,
 738                  * so that register information is correct.
 739                  */
 740                 if (tsk->ptrace & PT_DTRACE) {
 741                         regs->eflags &= ~TF_MASK;
 742                         tsk->ptrace &= ~PT_DTRACE;
 743                 }
 744         }
 745
 746         /* Ok, finally something we can handle */
 747         tsk->thread.trap_no = 1;
 748         tsk->thread.error_code = error_code;
 749         info.si_signo = SIGTRAP;
 750         info.si_errno = 0;
 751         info.si_code = TRAP_BRKPT;
 752         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 753         force_sig_info(SIGTRAP, &info, tsk);
 754
 755 clear_dr7:
 756         set_debugreg(0UL, 7);
 757         preempt_conditional_cli(regs);
 758         return;
 759
 760 clear_TF_reenable:
 761         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 762         regs->eflags &= ~TF_MASK;
 763         preempt_conditional_cli(regs);
 764 }
 765
 766 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 767 {
 768         const struct exception_table_entry *fixup;
 769         fixup = search_exception_tables(regs->rip);
 770         if (fixup) {
 771                 regs->rip = fixup->fixup;
 772                 return 1;
 773         }
 774         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 775         /* Illegal floating point operation in the kernel */
 776         current->thread.trap_no = trapnr;
 777         die(str, regs, 0);
 778         return 0;
 779 }
 780
 781 /*
 782  * Note that we play around with the 'TS' bit in an attempt to get
 783  * the correct behaviour even in the presence of the asynchronous
 784  * IRQ13 behaviour
 785  */
 786 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 787 {
 788         void __user *rip = (void __user *)(regs->rip);
 789         struct task_struct * task;
 790         siginfo_t info;
 791         unsigned short cwd, swd;
 792
 793         conditional_sti(regs);
 794         if (!user_mode(regs) &&
 795             kernel_math_error(regs, "kernel x87 math error", 16))
 796                 return;
 797
 798         /*
 799          * Save the info for the exception handler and clear the error.
 800          */
 801         task = current;
 802         save_init_fpu(task);
 803         task->thread.trap_no = 16;
 804         task->thread.error_code = 0;
 805         info.si_signo = SIGFPE;
 806         info.si_errno = 0;
 807         info.si_code = __SI_FAULT;
 808         info.si_addr = rip;
 809         /*
 810          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 811          * status.  0x3f is the exception bits in these regs, 0x200 is the
 812          * C1 reg you need in case of a stack fault, 0x040 is the stack
 813          * fault bit.  We should only be taking one exception at a time,
 814          * so if this combination doesn't produce any single exception,
 815          * then we have a bad program that isn't synchronizing its FPU usage
 816          * and it will suffer the consequences since we won't be able to
 817          * fully reproduce the context of the exception
 818          */
 819         cwd = get_fpu_cwd(task);
 820         swd = get_fpu_swd(task);
 821         switch (swd & ~cwd & 0x3f) {
 822                 case 0x000:
 823                 default:
 824                         break;
 825                 case 0x001: /* Invalid Op */
 826                         /*
 827                          * swd & 0x240 == 0x040: Stack Underflow
 828                          * swd & 0x240 == 0x240: Stack Overflow
 829                          * User must clear the SF bit (0x40) if set
 830                          */
 831                         info.si_code = FPE_FLTINV;
 832                         break;
 833                 case 0x002: /* Denormalize */
 834                 case 0x010: /* Underflow */
 835                         info.si_code = FPE_FLTUND;
 836                         break;
 837                 case 0x004: /* Zero Divide */
 838                         info.si_code = FPE_FLTDIV;
 839                         break;
 840                 case 0x008: /* Overflow */
 841                         info.si_code = FPE_FLTOVF;
 842                         break;
 843                 case 0x020: /* Precision */
 844                         info.si_code = FPE_FLTRES;
 845                         break;
 846         }
 847         force_sig_info(SIGFPE, &info, task);
 848 }
 849
 850 asmlinkage void bad_intr(void)
 851 {
 852         printk("bad interrupt");
 853 }
 854
 855 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 856 {
 857         void __user *rip = (void __user *)(regs->rip);
 858         struct task_struct * task;
 859         siginfo_t info;
 860         unsigned short mxcsr;
 861
 862         conditional_sti(regs);
 863         if (!user_mode(regs) &&
 864                 kernel_math_error(regs, "kernel simd math error", 19))
 865                 return;
 866
 867         /*
 868          * Save the info for the exception handler and clear the error.
 869          */
 870         task = current;
 871         save_init_fpu(task);
 872         task->thread.trap_no = 19;
 873         task->thread.error_code = 0;
 874         info.si_signo = SIGFPE;
 875         info.si_errno = 0;
 876         info.si_code = __SI_FAULT;
 877         info.si_addr = rip;
 878         /*
 879          * The SIMD FPU exceptions are handled a little differently, as there
 880          * is only a single status/control register.  Thus, to determine which
 881          * unmasked exception was caught we must mask the exception mask bits
 882          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 883          */
 884         mxcsr = get_fpu_mxcsr(task);
 885         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 886                 case 0x000:
 887                 default:
 888                         break;
 889                 case 0x001: /* Invalid Op */
 890                         info.si_code = FPE_FLTINV;
 891                         break;
 892                 case 0x002: /* Denormalize */
 893                 case 0x010: /* Underflow */
 894                         info.si_code = FPE_FLTUND;
 895                         break;
 896                 case 0x004: /* Zero Divide */
 897                         info.si_code = FPE_FLTDIV;
 898                         break;
 899                 case 0x008: /* Overflow */
 900                         info.si_code = FPE_FLTOVF;
 901                         break;
 902                 case 0x020: /* Precision */
 903                         info.si_code = FPE_FLTRES;
 904                         break;
 905         }
 906         force_sig_info(SIGFPE, &info, task);
 907 }
 908
 909 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 910 {
 911 }
 912
 913 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 914 {
 915 }
 916
 917 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 918 {
 919 }
 920
 921 /*
 922  *  'math_state_restore()' saves the current math information in the
 923  * old math state array, and gets the new ones from the current task
 924  *
 925  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 926  * Don't touch unless you *really* know how it works.
 927  */
 928 asmlinkage void math_state_restore(void)
 929 {
 930         struct task_struct *me = current;
 931         clts();                 /* Allow maths ops (or we recurse) */
 932
 933         if (!used_math())
 934                 init_fpu(me);
 935         restore_fpu_checking(&me->thread.i387.fxsave);
 936         task_thread_info(me)->status |= TS_USEDFPU;
 937 }
 938
 939 void __init trap_init(void)
 940 {
 941         set_intr_gate(0,&divide_error);
 942         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 943         set_intr_gate_ist(2,&nmi,NMI_STACK);
 944         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
 945         set_system_gate(4,&overflow);   /* int4 can be called from all */
 946         set_intr_gate(5,&bounds);
 947         set_intr_gate(6,&invalid_op);
 948         set_intr_gate(7,&device_not_available);
 949         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 950         set_intr_gate(9,&coprocessor_segment_overrun);
 951         set_intr_gate(10,&invalid_TSS);
 952         set_intr_gate(11,&segment_not_present);
 953         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 954         set_intr_gate(13,&general_protection);
 955         set_intr_gate(14,&page_fault);
 956         set_intr_gate(15,&spurious_interrupt_bug);
 957         set_intr_gate(16,&coprocessor_error);
 958         set_intr_gate(17,&alignment_check);
 959 #ifdef CONFIG_X86_MCE
 960         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 961 #endif
 962         set_intr_gate(19,&simd_coprocessor_error);
 963
 964 #ifdef CONFIG_IA32_EMULATION
 965         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 966 #endif
 967
 968         /*
 969          * Should be a barrier for any external CPU state.
 970          */
 971         cpu_init();
 972 }
 973
 974
 975 /* Actual parsing is done early in setup.c. */
 976 static int __init oops_dummy(char *s)
 977 {
 978         panic_on_oops = 1;
 979         return -1;
 980 }
 981 __setup("oops=", oops_dummy);
 982
 983 static int __init kstack_setup(char *s)
 984 {
 985         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 986         return 0;
 987 }
 988 __setup("kstack=", kstack_setup);
 989