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 asmlinkage void call_debug(void);
  74
  75 struct notifier_block *die_chain;
  76 static DEFINE_SPINLOCK(die_notifier_lock);
  77
  78 int register_die_notifier(struct notifier_block *nb)
  79 {
  80         int err = 0;
  81         unsigned long flags;
  82         spin_lock_irqsave(&die_notifier_lock, flags);
  83         err = notifier_chain_register(&die_chain, nb);
  84         spin_unlock_irqrestore(&die_notifier_lock, flags);
  85         return err;
  86 }
  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 int kstack_depth_to_print = 10;
  95
  96 #ifdef CONFIG_KALLSYMS
  97 #include <linux/kallsyms.h>
  98 int printk_address(unsigned long address)
  99 {
 100         unsigned long offset = 0, symsize;
 101         const char *symname;
 102         char *modname;
 103         char *delim = ":";
 104         char namebuf[128];
 105
 106         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 107         if (!symname)
 108                 return printk("[<%016lx>]", address);
 109         if (!modname)
 110                 modname = delim = "";
 111         return printk("<%016lx>{%s%s%s%s%+ld}",
 112                       address,delim,modname,delim,symname,offset);
 113 }
 114 #else
 115 int printk_address(unsigned long address)
 116 {
 117         return printk("[<%016lx>]", address);
 118 }
 119 #endif
 120
 121 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 122                                         unsigned *usedp, const char **idp)
 123 {
 124         static const char ids[N_EXCEPTION_STACKS][8] = {
 125                 [DEBUG_STACK - 1] = "#DB",
 126                 [NMI_STACK - 1] = "NMI",
 127                 [DOUBLEFAULT_STACK - 1] = "#DF",
 128                 [STACKFAULT_STACK - 1] = "#SS",
 129                 [MCE_STACK - 1] = "#MC",
 130         };
 131         unsigned k;
 132
 133         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 134                 unsigned long end;
 135
 136                 end = per_cpu(init_tss, cpu).ist[k];
 137                 if (stack >= end)
 138                         continue;
 139                 if (stack >= end - EXCEPTION_STKSZ) {
 140                         if (*usedp & (1U << k))
 141                                 break;
 142                         *usedp |= 1U << k;
 143                         *idp = ids[k];
 144                         return (unsigned long *)end;
 145                 }
 146         }
 147         return NULL;
 148 }
 149
 150 /*
 151  * x86-64 can have upto three kernel stacks:
 152  * process stack
 153  * interrupt stack
 154  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 155  */
 156
 157 void show_trace(unsigned long *stack)
 158 {
 159         unsigned long addr;
 160         const unsigned cpu = safe_smp_processor_id();
 161         unsigned long *irqstack_end = (unsigned long *)cpu_pda[cpu].irqstackptr;
 162         int i;
 163         unsigned used = 0;
 164
 165         printk("\nCall Trace:");
 166
 167 #define HANDLE_STACK(cond) \
 168         do while (cond) { \
 169                 addr = *stack++; \
 170                 if (kernel_text_address(addr)) { \
 171                         /* \
 172                          * If the address is either in the text segment of the \
 173                          * kernel, or in the region which contains vmalloc'ed \
 174                          * memory, it *may* be the address of a calling \
 175                          * routine; if so, print it so that someone tracing \
 176                          * down the cause of the crash will be able to figure \
 177                          * out the call path that was taken. \
 178                          */ \
 179                         i += printk_address(addr); \
 180                         if (i > 50) { \
 181                                 printk("\n       "); \
 182                                 i = 0; \
 183                         } \
 184                         else \
 185                                 i += printk(" "); \
 186                 } \
 187         } while (0)
 188
 189         for(i = 0; ; ) {
 190                 const char *id;
 191                 unsigned long *estack_end;
 192                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 193                                                 &used, &id);
 194
 195                 if (estack_end) {
 196                         i += printk(" <%s> ", id);
 197                         HANDLE_STACK (stack < estack_end);
 198                         i += printk(" <EOE> ");
 199                         stack = (unsigned long *) estack_end[-2];
 200                         continue;
 201                 }
 202                 if (irqstack_end) {
 203                         unsigned long *irqstack;
 204                         irqstack = irqstack_end -
 205                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 206
 207                         if (stack >= irqstack && stack < irqstack_end) {
 208                                 i += printk(" <IRQ> ");
 209                                 HANDLE_STACK (stack < irqstack_end);
 210                                 stack = (unsigned long *) (irqstack_end[-1]);
 211                                 irqstack_end = NULL;
 212                                 i += printk(" <EOI> ");
 213                                 continue;
 214                         }
 215                 }
 216                 break;
 217         }
 218
 219         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 220 #undef HANDLE_STACK
 221         printk("\n");
 222 }
 223
 224 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 225 {
 226         unsigned long *stack;
 227         int i;
 228         const int cpu = safe_smp_processor_id();
 229         unsigned long *irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 230         unsigned long *irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE);
 231
 232         // debugging aid: "show_stack(NULL, NULL);" prints the
 233         // back trace for this cpu.
 234
 235         if (rsp == NULL) {
 236                 if (tsk)
 237                         rsp = (unsigned long *)tsk->thread.rsp;
 238                 else
 239                         rsp = (unsigned long *)&rsp;
 240         }
 241
 242         stack = rsp;
 243         for(i=0; i < kstack_depth_to_print; i++) {
 244                 if (stack >= irqstack && stack <= irqstack_end) {
 245                         if (stack == irqstack_end) {
 246                                 stack = (unsigned long *) (irqstack_end[-1]);
 247                                 printk(" <EOI> ");
 248                         }
 249                 } else {
 250                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 251                         break;
 252                 }
 253                 if (i && ((i % 4) == 0))
 254                         printk("\n       ");
 255                 printk("%016lx ", *stack++);
 256                 touch_nmi_watchdog();
 257         }
 258         show_trace((unsigned long *)rsp);
 259 }
 260
 261 /*
 262  * The architecture-independent dump_stack generator
 263  */
 264 void dump_stack(void)
 265 {
 266         unsigned long dummy;
 267         show_trace(&dummy);
 268 }
 269
 270 EXPORT_SYMBOL(dump_stack);
 271
 272 void show_registers(struct pt_regs *regs)
 273 {
 274         int i;
 275         int in_kernel = !user_mode(regs);
 276         unsigned long rsp;
 277         const int cpu = safe_smp_processor_id();
 278         struct task_struct *cur = cpu_pda[cpu].pcurrent;
 279
 280                 rsp = regs->rsp;
 281
 282         printk("CPU %d ", cpu);
 283         __show_regs(regs);
 284         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 285                 cur->comm, cur->pid, cur->thread_info, cur);
 286
 287         /*
 288          * When in-kernel, we also print out the stack and code at the
 289          * time of the fault..
 290          */
 291         if (in_kernel) {
 292
 293                 printk("Stack: ");
 294                 show_stack(NULL, (unsigned long*)rsp);
 295
 296                 printk("\nCode: ");
 297                 if(regs->rip < PAGE_OFFSET)
 298                         goto bad;
 299
 300                 for(i=0;i<20;i++)
 301                 {
 302                         unsigned char c;
 303                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 304 bad:
 305                                 printk(" Bad RIP value.");
 306                                 break;
 307                         }
 308                         printk("%02x ", c);
 309                 }
 310         }
 311         printk("\n");
 312 }
 313
 314 void handle_BUG(struct pt_regs *regs)
 315 {
 316         struct bug_frame f;
 317         char tmp;
 318
 319         if (user_mode(regs))
 320                 return;
 321         if (__copy_from_user(&f, (struct bug_frame *) regs->rip,
 322                              sizeof(struct bug_frame)))
 323                 return;
 324         if (f.filename >= 0 ||
 325             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 326                 return;
 327         if (__get_user(tmp, (char *)(long)f.filename))
 328                 f.filename = (int)(long)"unmapped filename";
 329         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 330         printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", (char *)(long)f.filename, f.line);
 331 }
 332
 333 #ifdef CONFIG_BUG
 334 void out_of_line_bug(void)
 335 {
 336         BUG();
 337 }
 338 #endif
 339
 340 static DEFINE_SPINLOCK(die_lock);
 341 static int die_owner = -1;
 342
 343 unsigned long oops_begin(void)
 344 {
 345         int cpu = safe_smp_processor_id();
 346         unsigned long flags;
 347
 348         /* racy, but better than risking deadlock. */
 349         local_irq_save(flags);
 350         if (!spin_trylock(&die_lock)) {
 351                 if (cpu == die_owner)
 352                         /* nested oops. should stop eventually */;
 353                 else
 354                         spin_lock(&die_lock);
 355         }
 356         die_owner = cpu;
 357         console_verbose();
 358         bust_spinlocks(1);
 359         return flags;
 360 }
 361
 362 void oops_end(unsigned long flags)
 363 {
 364         die_owner = -1;
 365         bust_spinlocks(0);
 366         spin_unlock_irqrestore(&die_lock, flags);
 367         if (panic_on_oops)
 368                 panic("Oops");
 369 }
 370
 371 void __die(const char * str, struct pt_regs * regs, long err)
 372 {
 373         static int die_counter;
 374         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 375 #ifdef CONFIG_PREEMPT
 376         printk("PREEMPT ");
 377 #endif
 378 #ifdef CONFIG_SMP
 379         printk("SMP ");
 380 #endif
 381 #ifdef CONFIG_DEBUG_PAGEALLOC
 382         printk("DEBUG_PAGEALLOC");
 383 #endif
 384         printk("\n");
 385         notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
 386         show_registers(regs);
 387         /* Executive summary in case the oops scrolled away */
 388         printk(KERN_ALERT "RIP ");
 389         printk_address(regs->rip);
 390         printk(" RSP <%016lx>\n", regs->rsp);
 391 }
 392
 393 void die(const char * str, struct pt_regs * regs, long err)
 394 {
 395         unsigned long flags = oops_begin();
 396
 397         handle_BUG(regs);
 398         __die(str, regs, err);
 399         oops_end(flags);
 400         do_exit(SIGSEGV);
 401 }
 402 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 403 {
 404         if (!(regs->eflags & VM_MASK) && (regs->cs == __KERNEL_CS))
 405                 die(str, regs, err);
 406 }
 407
 408 void die_nmi(char *str, struct pt_regs *regs)
 409 {
 410         unsigned long flags = oops_begin();
 411
 412         /*
 413          * We are in trouble anyway, lets at least try
 414          * to get a message out.
 415          */
 416         printk(str, safe_smp_processor_id());
 417         show_registers(regs);
 418         if (panic_on_timeout || panic_on_oops)
 419                 panic("nmi watchdog");
 420         printk("console shuts up ...\n");
 421         oops_end(flags);
 422         do_exit(SIGSEGV);
 423 }
 424
 425 static void __kprobes do_trap(int trapnr, int signr, char *str,
 426                               struct pt_regs * regs, long error_code,
 427                               siginfo_t *info)
 428 {
 429         conditional_sti(regs);
 430
 431 #ifdef CONFIG_CHECKING
 432        {
 433                unsigned long gs;
 434                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 435                rdmsrl(MSR_GS_BASE, gs);
 436                if (gs != (unsigned long)pda) {
 437                        wrmsrl(MSR_GS_BASE, pda);
 438                        printk("%s: wrong gs %lx expected %p rip %lx\n", str, gs, pda,
 439                               regs->rip);
 440                }
 441        }
 442 #endif
 443
 444         if (user_mode(regs)) {
 445                 struct task_struct *tsk = current;
 446
 447                 if (exception_trace && unhandled_signal(tsk, signr))
 448                         printk(KERN_INFO
 449                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 450                                tsk->comm, tsk->pid, str,
 451                                regs->rip,regs->rsp,error_code);
 452
 453                 tsk->thread.error_code = error_code;
 454                 tsk->thread.trap_no = trapnr;
 455                 if (info)
 456                         force_sig_info(signr, info, tsk);
 457                 else
 458                         force_sig(signr, tsk);
 459                 return;
 460         }
 461
 462
 463         /* kernel trap */
 464         {
 465                 const struct exception_table_entry *fixup;
 466                 fixup = search_exception_tables(regs->rip);
 467                 if (fixup) {
 468                         regs->rip = fixup->fixup;
 469                 } else
 470                         die(str, regs, error_code);
 471                 return;
 472         }
 473 }
 474
 475 #define DO_ERROR(trapnr, signr, str, name) \
 476 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 477 { \
 478         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 479                                                         == NOTIFY_STOP) \
 480                 return; \
 481         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 482 }
 483
 484 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 485 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 486 { \
 487         siginfo_t info; \
 488         info.si_signo = signr; \
 489         info.si_errno = 0; \
 490         info.si_code = sicode; \
 491         info.si_addr = (void __user *)siaddr; \
 492         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 493                                                         == NOTIFY_STOP) \
 494                 return; \
 495         do_trap(trapnr, signr, str, regs, error_code, &info); \
 496 }
 497
 498 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 499 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 500 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 501 DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->rip)
 502 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 503 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 504 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 505 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 506 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 507 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 508 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 509 DO_ERROR( 8, SIGSEGV, "double fault", double_fault)
 510
 511 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 512                                                 long error_code)
 513 {
 514         conditional_sti(regs);
 515
 516 #ifdef CONFIG_CHECKING
 517        {
 518                unsigned long gs;
 519                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 520                rdmsrl(MSR_GS_BASE, gs);
 521                if (gs != (unsigned long)pda) {
 522                        wrmsrl(MSR_GS_BASE, pda);
 523                        oops_in_progress++;
 524                        printk("general protection handler: wrong gs %lx expected %p\n", gs, pda);
 525                        oops_in_progress--;
 526                }
 527        }
 528 #endif
 529
 530         if (user_mode(regs)) {
 531                 struct task_struct *tsk = current;
 532
 533                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 534                         printk(KERN_INFO
 535                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 536                                tsk->comm, tsk->pid,
 537                                regs->rip,regs->rsp,error_code);
 538
 539                 tsk->thread.error_code = error_code;
 540                 tsk->thread.trap_no = 13;
 541                 force_sig(SIGSEGV, tsk);
 542                 return;
 543         }
 544
 545         /* kernel gp */
 546         {
 547                 const struct exception_table_entry *fixup;
 548                 fixup = search_exception_tables(regs->rip);
 549                 if (fixup) {
 550                         regs->rip = fixup->fixup;
 551                         return;
 552                 }
 553                 if (notify_die(DIE_GPF, "general protection fault", regs,
 554                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 555                         return;
 556                 die("general protection fault", regs, error_code);
 557         }
 558 }
 559
 560 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
 561 {
 562         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 563         printk("You probably have a hardware problem with your RAM chips\n");
 564
 565         /* Clear and disable the memory parity error line. */
 566         reason = (reason & 0xf) | 4;
 567         outb(reason, 0x61);
 568 }
 569
 570 static void io_check_error(unsigned char reason, struct pt_regs * regs)
 571 {
 572         printk("NMI: IOCK error (debug interrupt?)\n");
 573         show_registers(regs);
 574
 575         /* Re-enable the IOCK line, wait for a few seconds */
 576         reason = (reason & 0xf) | 8;
 577         outb(reason, 0x61);
 578         mdelay(2000);
 579         reason &= ~8;
 580         outb(reason, 0x61);
 581 }
 582
 583 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 584 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 585         printk("Dazed and confused, but trying to continue\n");
 586         printk("Do you have a strange power saving mode enabled?\n");
 587 }
 588
 589 /* Runs on IST stack. This code must keep interrupts off all the time.
 590    Nested NMIs are prevented by the CPU. */
 591 asmlinkage void default_do_nmi(struct pt_regs *regs)
 592 {
 593         unsigned char reason = 0;
 594         int cpu;
 595
 596         cpu = smp_processor_id();
 597
 598         /* Only the BSP gets external NMIs from the system.  */
 599         if (!cpu)
 600                 reason = get_nmi_reason();
 601
 602         if (!(reason & 0xc0)) {
 603                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
 604                                                                 == NOTIFY_STOP)
 605                         return;
 606 #ifdef CONFIG_X86_LOCAL_APIC
 607                 /*
 608                  * Ok, so this is none of the documented NMI sources,
 609                  * so it must be the NMI watchdog.
 610                  */
 611                 if (nmi_watchdog > 0) {
 612                         nmi_watchdog_tick(regs,reason);
 613                         return;
 614                 }
 615 #endif
 616                 unknown_nmi_error(reason, regs);
 617                 return;
 618         }
 619         if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
 620                 return;
 621
 622         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 623
 624         if (reason & 0x80)
 625                 mem_parity_error(reason, regs);
 626         if (reason & 0x40)
 627                 io_check_error(reason, regs);
 628 }
 629
 630 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 631 {
 632         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 633                 return;
 634         }
 635         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 636         return;
 637 }
 638
 639 /* Help handler running on IST stack to switch back to user stack
 640    for scheduling or signal handling. The actual stack switch is done in
 641    entry.S */
 642 asmlinkage struct pt_regs *sync_regs(struct pt_regs *eregs)
 643 {
 644         struct pt_regs *regs = eregs;
 645         /* Did already sync */
 646         if (eregs == (struct pt_regs *)eregs->rsp)
 647                 ;
 648         /* Exception from user space */
 649         else if (user_mode(eregs))
 650                 regs = ((struct pt_regs *)current->thread.rsp0) - 1;
 651         /* Exception from kernel and interrupts are enabled. Move to
 652            kernel process stack. */
 653         else if (eregs->eflags & X86_EFLAGS_IF)
 654                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 655         if (eregs != regs)
 656                 *regs = *eregs;
 657         return regs;
 658 }
 659
 660 /* runs on IST stack. */
 661 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 662                                    unsigned long error_code)
 663 {
 664         unsigned long condition;
 665         struct task_struct *tsk = current;
 666         siginfo_t info;
 667
 668 #ifdef CONFIG_CHECKING
 669        {
 670                /* RED-PEN interaction with debugger - could destroy gs */
 671                unsigned long gs;
 672                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 673                rdmsrl(MSR_GS_BASE, gs);
 674                if (gs != (unsigned long)pda) {
 675                        wrmsrl(MSR_GS_BASE, pda);
 676                        printk("debug handler: wrong gs %lx expected %p\n", gs, pda);
 677                }
 678        }
 679 #endif
 680
 681         get_debugreg(condition, 6);
 682
 683         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 684                                                 SIGTRAP) == NOTIFY_STOP)
 685                 return;
 686
 687         conditional_sti(regs);
 688
 689         /* Mask out spurious debug traps due to lazy DR7 setting */
 690         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 691                 if (!tsk->thread.debugreg7) {
 692                         goto clear_dr7;
 693                 }
 694         }
 695
 696         tsk->thread.debugreg6 = condition;
 697
 698         /* Mask out spurious TF errors due to lazy TF clearing */
 699         if (condition & DR_STEP) {
 700                 /*
 701                  * The TF error should be masked out only if the current
 702                  * process is not traced and if the TRAP flag has been set
 703                  * previously by a tracing process (condition detected by
 704                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 705                  * can be modified by the process itself in user mode,
 706                  * allowing programs to debug themselves without the ptrace()
 707                  * interface.
 708                  */
 709                 if (!user_mode(regs))
 710                        goto clear_TF_reenable;
 711                 /*
 712                  * Was the TF flag set by a debugger? If so, clear it now,
 713                  * so that register information is correct.
 714                  */
 715                 if (tsk->ptrace & PT_DTRACE) {
 716                         regs->eflags &= ~TF_MASK;
 717                         tsk->ptrace &= ~PT_DTRACE;
 718                 }
 719         }
 720
 721         /* Ok, finally something we can handle */
 722         tsk->thread.trap_no = 1;
 723         tsk->thread.error_code = error_code;
 724         info.si_signo = SIGTRAP;
 725         info.si_errno = 0;
 726         info.si_code = TRAP_BRKPT;
 727         if (!user_mode(regs))
 728                 goto clear_dr7;
 729
 730         info.si_addr = (void __user *)regs->rip;
 731         force_sig_info(SIGTRAP, &info, tsk);
 732 clear_dr7:
 733         set_debugreg(0UL, 7);
 734         return;
 735
 736 clear_TF_reenable:
 737         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 738         regs->eflags &= ~TF_MASK;
 739 }
 740
 741 static int kernel_math_error(struct pt_regs *regs, char *str)
 742 {
 743         const struct exception_table_entry *fixup;
 744         fixup = search_exception_tables(regs->rip);
 745         if (fixup) {
 746                 regs->rip = fixup->fixup;
 747                 return 1;
 748         }
 749         notify_die(DIE_GPF, str, regs, 0, 16, SIGFPE);
 750         /* Illegal floating point operation in the kernel */
 751         die(str, regs, 0);
 752         return 0;
 753 }
 754
 755 /*
 756  * Note that we play around with the 'TS' bit in an attempt to get
 757  * the correct behaviour even in the presence of the asynchronous
 758  * IRQ13 behaviour
 759  */
 760 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 761 {
 762         void __user *rip = (void __user *)(regs->rip);
 763         struct task_struct * task;
 764         siginfo_t info;
 765         unsigned short cwd, swd;
 766
 767         conditional_sti(regs);
 768         if (!user_mode(regs) &&
 769             kernel_math_error(regs, "kernel x87 math error"))
 770                 return;
 771
 772         /*
 773          * Save the info for the exception handler and clear the error.
 774          */
 775         task = current;
 776         save_init_fpu(task);
 777         task->thread.trap_no = 16;
 778         task->thread.error_code = 0;
 779         info.si_signo = SIGFPE;
 780         info.si_errno = 0;
 781         info.si_code = __SI_FAULT;
 782         info.si_addr = rip;
 783         /*
 784          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 785          * status.  0x3f is the exception bits in these regs, 0x200 is the
 786          * C1 reg you need in case of a stack fault, 0x040 is the stack
 787          * fault bit.  We should only be taking one exception at a time,
 788          * so if this combination doesn't produce any single exception,
 789          * then we have a bad program that isn't synchronizing its FPU usage
 790          * and it will suffer the consequences since we won't be able to
 791          * fully reproduce the context of the exception
 792          */
 793         cwd = get_fpu_cwd(task);
 794         swd = get_fpu_swd(task);
 795         switch (swd & ~cwd & 0x3f) {
 796                 case 0x000:
 797                 default:
 798                         break;
 799                 case 0x001: /* Invalid Op */
 800                         /*
 801                          * swd & 0x240 == 0x040: Stack Underflow
 802                          * swd & 0x240 == 0x240: Stack Overflow
 803                          * User must clear the SF bit (0x40) if set
 804                          */
 805                         info.si_code = FPE_FLTINV;
 806                         break;
 807                 case 0x002: /* Denormalize */
 808                 case 0x010: /* Underflow */
 809                         info.si_code = FPE_FLTUND;
 810                         break;
 811                 case 0x004: /* Zero Divide */
 812                         info.si_code = FPE_FLTDIV;
 813                         break;
 814                 case 0x008: /* Overflow */
 815                         info.si_code = FPE_FLTOVF;
 816                         break;
 817                 case 0x020: /* Precision */
 818                         info.si_code = FPE_FLTRES;
 819                         break;
 820         }
 821         force_sig_info(SIGFPE, &info, task);
 822 }
 823
 824 asmlinkage void bad_intr(void)
 825 {
 826         printk("bad interrupt");
 827 }
 828
 829 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 830 {
 831         void __user *rip = (void __user *)(regs->rip);
 832         struct task_struct * task;
 833         siginfo_t info;
 834         unsigned short mxcsr;
 835
 836         conditional_sti(regs);
 837         if (!user_mode(regs) &&
 838                 kernel_math_error(regs, "kernel simd math error"))
 839                 return;
 840
 841         /*
 842          * Save the info for the exception handler and clear the error.
 843          */
 844         task = current;
 845         save_init_fpu(task);
 846         task->thread.trap_no = 19;
 847         task->thread.error_code = 0;
 848         info.si_signo = SIGFPE;
 849         info.si_errno = 0;
 850         info.si_code = __SI_FAULT;
 851         info.si_addr = rip;
 852         /*
 853          * The SIMD FPU exceptions are handled a little differently, as there
 854          * is only a single status/control register.  Thus, to determine which
 855          * unmasked exception was caught we must mask the exception mask bits
 856          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 857          */
 858         mxcsr = get_fpu_mxcsr(task);
 859         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 860                 case 0x000:
 861                 default:
 862                         break;
 863                 case 0x001: /* Invalid Op */
 864                         info.si_code = FPE_FLTINV;
 865                         break;
 866                 case 0x002: /* Denormalize */
 867                 case 0x010: /* Underflow */
 868                         info.si_code = FPE_FLTUND;
 869                         break;
 870                 case 0x004: /* Zero Divide */
 871                         info.si_code = FPE_FLTDIV;
 872                         break;
 873                 case 0x008: /* Overflow */
 874                         info.si_code = FPE_FLTOVF;
 875                         break;
 876                 case 0x020: /* Precision */
 877                         info.si_code = FPE_FLTRES;
 878                         break;
 879         }
 880         force_sig_info(SIGFPE, &info, task);
 881 }
 882
 883 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 884 {
 885 }
 886
 887 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 888 {
 889 }
 890
 891 /*
 892  *  'math_state_restore()' saves the current math information in the
 893  * old math state array, and gets the new ones from the current task
 894  *
 895  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 896  * Don't touch unless you *really* know how it works.
 897  */
 898 asmlinkage void math_state_restore(void)
 899 {
 900         struct task_struct *me = current;
 901         clts();                 /* Allow maths ops (or we recurse) */
 902
 903         if (!used_math())
 904                 init_fpu(me);
 905         restore_fpu_checking(&me->thread.i387.fxsave);
 906         me->thread_info->status |= TS_USEDFPU;
 907 }
 908
 909 void do_call_debug(struct pt_regs *regs)
 910 {
 911         notify_die(DIE_CALL, "debug call", regs, 0, 255, SIGINT);
 912 }
 913
 914 void __init trap_init(void)
 915 {
 916         set_intr_gate(0,&divide_error);
 917         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 918         set_intr_gate_ist(2,&nmi,NMI_STACK);
 919         set_system_gate(3,&int3);
 920         set_system_gate(4,&overflow);   /* int4-5 can be called from all */
 921         set_system_gate(5,&bounds);
 922         set_intr_gate(6,&invalid_op);
 923         set_intr_gate(7,&device_not_available);
 924         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 925         set_intr_gate(9,&coprocessor_segment_overrun);
 926         set_intr_gate(10,&invalid_TSS);
 927         set_intr_gate(11,&segment_not_present);
 928         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 929         set_intr_gate(13,&general_protection);
 930         set_intr_gate(14,&page_fault);
 931         set_intr_gate(15,&spurious_interrupt_bug);
 932         set_intr_gate(16,&coprocessor_error);
 933         set_intr_gate(17,&alignment_check);
 934 #ifdef CONFIG_X86_MCE
 935         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 936 #endif
 937         set_intr_gate(19,&simd_coprocessor_error);
 938
 939 #ifdef CONFIG_IA32_EMULATION
 940         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 941 #endif
 942
 943         set_intr_gate(KDB_VECTOR, call_debug);
 944
 945         /*
 946          * Should be a barrier for any external CPU state.
 947          */
 948         cpu_init();
 949 }
 950
 951
 952 /* Actual parsing is done early in setup.c. */
 953 static int __init oops_dummy(char *s)
 954 {
 955         panic_on_oops = 1;
 956         return -1;
 957 }
 958 __setup("oops=", oops_dummy);
 959
 960 static int __init kstack_setup(char *s)
 961 {
 962         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 963         return 0;
 964 }
 965 __setup("kstack=", kstack_setup);
 966