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         spin_lock_irqsave(&die_notifier_lock, flags);
  82         err = notifier_chain_register(&die_chain, nb);
  83         spin_unlock_irqrestore(&die_notifier_lock, flags);
  84         return err;
  85 }
  86
  87 static inline void conditional_sti(struct pt_regs *regs)
  88 {
  89         if (regs->eflags & X86_EFLAGS_IF)
  90                 local_irq_enable();
  91 }
  92
  93 static int kstack_depth_to_print = 10;
  94
  95 #ifdef CONFIG_KALLSYMS
  96 #include <linux/kallsyms.h>
  97 int printk_address(unsigned long address)
  98 {
  99         unsigned long offset = 0, symsize;
 100         const char *symname;
 101         char *modname;
 102         char *delim = ":";
 103         char namebuf[128];
 104
 105         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 106         if (!symname)
 107                 return printk("[<%016lx>]", address);
 108         if (!modname)
 109                 modname = delim = "";
 110         return printk("<%016lx>{%s%s%s%s%+ld}",
 111                       address,delim,modname,delim,symname,offset);
 112 }
 113 #else
 114 int printk_address(unsigned long address)
 115 {
 116         return printk("[<%016lx>]", address);
 117 }
 118 #endif
 119
 120 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 121                                         unsigned *usedp, const char **idp)
 122 {
 123         static char ids[][8] = {
 124                 [DEBUG_STACK - 1] = "#DB",
 125                 [NMI_STACK - 1] = "NMI",
 126                 [DOUBLEFAULT_STACK - 1] = "#DF",
 127                 [STACKFAULT_STACK - 1] = "#SS",
 128                 [MCE_STACK - 1] = "#MC",
 129 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 130                 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 131 #endif
 132         };
 133         unsigned k;
 134
 135         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 136                 unsigned long end;
 137
 138                 switch (k + 1) {
 139 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 140                 case DEBUG_STACK:
 141                         end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
 142                         break;
 143 #endif
 144                 default:
 145                         end = per_cpu(init_tss, cpu).ist[k];
 146                         break;
 147                 }
 148                 if (stack >= end)
 149                         continue;
 150                 if (stack >= end - EXCEPTION_STKSZ) {
 151                         if (*usedp & (1U << k))
 152                                 break;
 153                         *usedp |= 1U << k;
 154                         *idp = ids[k];
 155                         return (unsigned long *)end;
 156                 }
 157 #if DEBUG_STKSZ > EXCEPTION_STKSZ
 158                 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
 159                         unsigned j = N_EXCEPTION_STACKS - 1;
 160
 161                         do {
 162                                 ++j;
 163                                 end -= EXCEPTION_STKSZ;
 164                                 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
 165                         } while (stack < end - EXCEPTION_STKSZ);
 166                         if (*usedp & (1U << j))
 167                                 break;
 168                         *usedp |= 1U << j;
 169                         *idp = ids[j];
 170                         return (unsigned long *)end;
 171                 }
 172 #endif
 173         }
 174         return NULL;
 175 }
 176
 177 /*
 178  * x86-64 can have upto three kernel stacks:
 179  * process stack
 180  * interrupt stack
 181  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
 182  */
 183
 184 void show_trace(unsigned long *stack)
 185 {
 186         unsigned long addr;
 187         const unsigned cpu = safe_smp_processor_id();
 188         unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
 189         int i;
 190         unsigned used = 0;
 191
 192         printk("\nCall Trace:");
 193
 194 #define HANDLE_STACK(cond) \
 195         do while (cond) { \
 196                 addr = *stack++; \
 197                 if (kernel_text_address(addr)) { \
 198                         /* \
 199                          * If the address is either in the text segment of the \
 200                          * kernel, or in the region which contains vmalloc'ed \
 201                          * memory, it *may* be the address of a calling \
 202                          * routine; if so, print it so that someone tracing \
 203                          * down the cause of the crash will be able to figure \
 204                          * out the call path that was taken. \
 205                          */ \
 206                         i += printk_address(addr); \
 207                         if (i > 50) { \
 208                                 printk("\n       "); \
 209                                 i = 0; \
 210                         } \
 211                         else \
 212                                 i += printk(" "); \
 213                 } \
 214         } while (0)
 215
 216         for(i = 0; ; ) {
 217                 const char *id;
 218                 unsigned long *estack_end;
 219                 estack_end = in_exception_stack(cpu, (unsigned long)stack,
 220                                                 &used, &id);
 221
 222                 if (estack_end) {
 223                         i += printk(" <%s> ", id);
 224                         HANDLE_STACK (stack < estack_end);
 225                         i += printk(" <EOE> ");
 226                         stack = (unsigned long *) estack_end[-2];
 227                         continue;
 228                 }
 229                 if (irqstack_end) {
 230                         unsigned long *irqstack;
 231                         irqstack = irqstack_end -
 232                                 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 233
 234                         if (stack >= irqstack && stack < irqstack_end) {
 235                                 i += printk(" <IRQ> ");
 236                                 HANDLE_STACK (stack < irqstack_end);
 237                                 stack = (unsigned long *) (irqstack_end[-1]);
 238                                 irqstack_end = NULL;
 239                                 i += printk(" <EOI> ");
 240                                 continue;
 241                         }
 242                 }
 243                 break;
 244         }
 245
 246         HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 247 #undef HANDLE_STACK
 248         printk("\n");
 249 }
 250
 251 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 252 {
 253         unsigned long *stack;
 254         int i;
 255         const int cpu = safe_smp_processor_id();
 256         unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
 257         unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
 258
 259         // debugging aid: "show_stack(NULL, NULL);" prints the
 260         // back trace for this cpu.
 261
 262         if (rsp == NULL) {
 263                 if (tsk)
 264                         rsp = (unsigned long *)tsk->thread.rsp;
 265                 else
 266                         rsp = (unsigned long *)&rsp;
 267         }
 268
 269         stack = rsp;
 270         for(i=0; i < kstack_depth_to_print; i++) {
 271                 if (stack >= irqstack && stack <= irqstack_end) {
 272                         if (stack == irqstack_end) {
 273                                 stack = (unsigned long *) (irqstack_end[-1]);
 274                                 printk(" <EOI> ");
 275                         }
 276                 } else {
 277                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 278                         break;
 279                 }
 280                 if (i && ((i % 4) == 0))
 281                         printk("\n       ");
 282                 printk("%016lx ", *stack++);
 283                 touch_nmi_watchdog();
 284         }
 285         show_trace((unsigned long *)rsp);
 286 }
 287
 288 /*
 289  * The architecture-independent dump_stack generator
 290  */
 291 void dump_stack(void)
 292 {
 293         unsigned long dummy;
 294         show_trace(&dummy);
 295 }
 296
 297 EXPORT_SYMBOL(dump_stack);
 298
 299 void show_registers(struct pt_regs *regs)
 300 {
 301         int i;
 302         int in_kernel = !user_mode(regs);
 303         unsigned long rsp;
 304         const int cpu = safe_smp_processor_id();
 305         struct task_struct *cur = cpu_pda(cpu)->pcurrent;
 306
 307                 rsp = regs->rsp;
 308
 309         printk("CPU %d ", cpu);
 310         __show_regs(regs);
 311         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 312                 cur->comm, cur->pid, cur->thread_info, cur);
 313
 314         /*
 315          * When in-kernel, we also print out the stack and code at the
 316          * time of the fault..
 317          */
 318         if (in_kernel) {
 319
 320                 printk("Stack: ");
 321                 show_stack(NULL, (unsigned long*)rsp);
 322
 323                 printk("\nCode: ");
 324                 if(regs->rip < PAGE_OFFSET)
 325                         goto bad;
 326
 327                 for(i=0;i<20;i++)
 328                 {
 329                         unsigned char c;
 330                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 331 bad:
 332                                 printk(" Bad RIP value.");
 333                                 break;
 334                         }
 335                         printk("%02x ", c);
 336                 }
 337         }
 338         printk("\n");
 339 }
 340
 341 void handle_BUG(struct pt_regs *regs)
 342 {
 343         struct bug_frame f;
 344         char tmp;
 345
 346         if (user_mode(regs))
 347                 return;
 348         if (__copy_from_user(&f, (const void __user *) regs->rip,
 349                              sizeof(struct bug_frame)))
 350                 return;
 351         if (f.filename >= 0 ||
 352             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 353                 return;
 354         if (__get_user(tmp, (char *)(long)f.filename))
 355                 f.filename = (int)(long)"unmapped filename";
 356         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 357         printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", (char *)(long)f.filename, f.line);
 358 }
 359
 360 #ifdef CONFIG_BUG
 361 void out_of_line_bug(void)
 362 {
 363         BUG();
 364 }
 365 #endif
 366
 367 static DEFINE_SPINLOCK(die_lock);
 368 static int die_owner = -1;
 369
 370 unsigned long oops_begin(void)
 371 {
 372         int cpu = safe_smp_processor_id();
 373         unsigned long flags;
 374
 375         /* racy, but better than risking deadlock. */
 376         local_irq_save(flags);
 377         if (!spin_trylock(&die_lock)) {
 378                 if (cpu == die_owner)
 379                         /* nested oops. should stop eventually */;
 380                 else
 381                         spin_lock(&die_lock);
 382         }
 383         die_owner = cpu;
 384         console_verbose();
 385         bust_spinlocks(1);
 386         return flags;
 387 }
 388
 389 void oops_end(unsigned long flags)
 390 {
 391         die_owner = -1;
 392         bust_spinlocks(0);
 393         spin_unlock_irqrestore(&die_lock, flags);
 394         if (panic_on_oops)
 395                 panic("Oops");
 396 }
 397
 398 void __die(const char * str, struct pt_regs * regs, long err)
 399 {
 400         static int die_counter;
 401         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 402 #ifdef CONFIG_PREEMPT
 403         printk("PREEMPT ");
 404 #endif
 405 #ifdef CONFIG_SMP
 406         printk("SMP ");
 407 #endif
 408 #ifdef CONFIG_DEBUG_PAGEALLOC
 409         printk("DEBUG_PAGEALLOC");
 410 #endif
 411         printk("\n");
 412         notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
 413         show_registers(regs);
 414         /* Executive summary in case the oops scrolled away */
 415         printk(KERN_ALERT "RIP ");
 416         printk_address(regs->rip);
 417         printk(" RSP <%016lx>\n", regs->rsp);
 418 }
 419
 420 void die(const char * str, struct pt_regs * regs, long err)
 421 {
 422         unsigned long flags = oops_begin();
 423
 424         handle_BUG(regs);
 425         __die(str, regs, err);
 426         oops_end(flags);
 427         do_exit(SIGSEGV);
 428 }
 429
 430 void die_nmi(char *str, struct pt_regs *regs)
 431 {
 432         unsigned long flags = oops_begin();
 433
 434         /*
 435          * We are in trouble anyway, lets at least try
 436          * to get a message out.
 437          */
 438         printk(str, safe_smp_processor_id());
 439         show_registers(regs);
 440         if (panic_on_timeout || panic_on_oops)
 441                 panic("nmi watchdog");
 442         printk("console shuts up ...\n");
 443         oops_end(flags);
 444         do_exit(SIGSEGV);
 445 }
 446
 447 static void __kprobes do_trap(int trapnr, int signr, char *str,
 448                               struct pt_regs * regs, long error_code,
 449                               siginfo_t *info)
 450 {
 451         struct task_struct *tsk = current;
 452
 453         conditional_sti(regs);
 454
 455         tsk->thread.error_code = error_code;
 456         tsk->thread.trap_no = trapnr;
 457
 458         if (user_mode(regs)) {
 459                 if (exception_trace && unhandled_signal(tsk, signr))
 460                         printk(KERN_INFO
 461                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 462                                tsk->comm, tsk->pid, str,
 463                                regs->rip,regs->rsp,error_code);
 464
 465                 if (info)
 466                         force_sig_info(signr, info, tsk);
 467                 else
 468                         force_sig(signr, tsk);
 469                 return;
 470         }
 471
 472
 473         /* kernel trap */
 474         {
 475                 const struct exception_table_entry *fixup;
 476                 fixup = search_exception_tables(regs->rip);
 477                 if (fixup) {
 478                         regs->rip = fixup->fixup;
 479                 } else
 480                         die(str, regs, error_code);
 481                 return;
 482         }
 483 }
 484
 485 #define DO_ERROR(trapnr, signr, str, name) \
 486 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 487 { \
 488         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 489                                                         == NOTIFY_STOP) \
 490                 return; \
 491         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 492 }
 493
 494 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 495 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 496 { \
 497         siginfo_t info; \
 498         info.si_signo = signr; \
 499         info.si_errno = 0; \
 500         info.si_code = sicode; \
 501         info.si_addr = (void __user *)siaddr; \
 502         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 503                                                         == NOTIFY_STOP) \
 504                 return; \
 505         do_trap(trapnr, signr, str, regs, error_code, &info); \
 506 }
 507
 508 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 509 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 510 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 511 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
 512 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 513 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 514 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 515 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 516 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 517 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 518 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 519
 520 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 521 {
 522         static const char str[] = "double fault";
 523         struct task_struct *tsk = current;
 524
 525         /* Return not checked because double check cannot be ignored */
 526         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 527
 528         tsk->thread.error_code = error_code;
 529         tsk->thread.trap_no = 8;
 530
 531         /* This is always a kernel trap and never fixable (and thus must
 532            never return). */
 533         for (;;)
 534                 die(str, regs, error_code);
 535 }
 536
 537 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 538                                                 long error_code)
 539 {
 540         struct task_struct *tsk = current;
 541
 542         conditional_sti(regs);
 543
 544         tsk->thread.error_code = error_code;
 545         tsk->thread.trap_no = 13;
 546
 547         if (user_mode(regs)) {
 548                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 549                         printk(KERN_INFO
 550                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 551                                tsk->comm, tsk->pid,
 552                                regs->rip,regs->rsp,error_code);
 553
 554                 force_sig(SIGSEGV, tsk);
 555                 return;
 556         }
 557
 558         /* kernel gp */
 559         {
 560                 const struct exception_table_entry *fixup;
 561                 fixup = search_exception_tables(regs->rip);
 562                 if (fixup) {
 563                         regs->rip = fixup->fixup;
 564                         return;
 565                 }
 566                 if (notify_die(DIE_GPF, "general protection fault", regs,
 567                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 568                         return;
 569                 die("general protection fault", regs, error_code);
 570         }
 571 }
 572
 573 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
 574 {
 575         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 576         printk("You probably have a hardware problem with your RAM chips\n");
 577
 578         /* Clear and disable the memory parity error line. */
 579         reason = (reason & 0xf) | 4;
 580         outb(reason, 0x61);
 581 }
 582
 583 static void io_check_error(unsigned char reason, struct pt_regs * regs)
 584 {
 585         printk("NMI: IOCK error (debug interrupt?)\n");
 586         show_registers(regs);
 587
 588         /* Re-enable the IOCK line, wait for a few seconds */
 589         reason = (reason & 0xf) | 8;
 590         outb(reason, 0x61);
 591         mdelay(2000);
 592         reason &= ~8;
 593         outb(reason, 0x61);
 594 }
 595
 596 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 597 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 598         printk("Dazed and confused, but trying to continue\n");
 599         printk("Do you have a strange power saving mode enabled?\n");
 600 }
 601
 602 /* Runs on IST stack. This code must keep interrupts off all the time.
 603    Nested NMIs are prevented by the CPU. */
 604 asmlinkage void default_do_nmi(struct pt_regs *regs)
 605 {
 606         unsigned char reason = 0;
 607         int cpu;
 608
 609         cpu = smp_processor_id();
 610
 611         /* Only the BSP gets external NMIs from the system.  */
 612         if (!cpu)
 613                 reason = get_nmi_reason();
 614
 615         if (!(reason & 0xc0)) {
 616                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 617                                                                 == NOTIFY_STOP)
 618                         return;
 619 #ifdef CONFIG_X86_LOCAL_APIC
 620                 /*
 621                  * Ok, so this is none of the documented NMI sources,
 622                  * so it must be the NMI watchdog.
 623                  */
 624                 if (nmi_watchdog > 0) {
 625                         nmi_watchdog_tick(regs,reason);
 626                         return;
 627                 }
 628 #endif
 629                 unknown_nmi_error(reason, regs);
 630                 return;
 631         }
 632         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 633                 return;
 634
 635         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 636
 637         if (reason & 0x80)
 638                 mem_parity_error(reason, regs);
 639         if (reason & 0x40)
 640                 io_check_error(reason, regs);
 641 }
 642
 643 /* runs on IST stack. */
 644 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
 645 {
 646         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 647                 return;
 648         }
 649         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 650         return;
 651 }
 652
 653 /* Help handler running on IST stack to switch back to user stack
 654    for scheduling or signal handling. The actual stack switch is done in
 655    entry.S */
 656 asmlinkage struct pt_regs *sync_regs(struct pt_regs *eregs)
 657 {
 658         struct pt_regs *regs = eregs;
 659         /* Did already sync */
 660         if (eregs == (struct pt_regs *)eregs->rsp)
 661                 ;
 662         /* Exception from user space */
 663         else if (user_mode(eregs))
 664                 regs = ((struct pt_regs *)current->thread.rsp0) - 1;
 665         /* Exception from kernel and interrupts are enabled. Move to
 666            kernel process stack. */
 667         else if (eregs->eflags & X86_EFLAGS_IF)
 668                 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
 669         if (eregs != regs)
 670                 *regs = *eregs;
 671         return regs;
 672 }
 673
 674 /* runs on IST stack. */
 675 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
 676                                    unsigned long error_code)
 677 {
 678         unsigned long condition;
 679         struct task_struct *tsk = current;
 680         siginfo_t info;
 681
 682         get_debugreg(condition, 6);
 683
 684         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 685                                                 SIGTRAP) == NOTIFY_STOP)
 686                 return;
 687
 688         conditional_sti(regs);
 689
 690         /* Mask out spurious debug traps due to lazy DR7 setting */
 691         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 692                 if (!tsk->thread.debugreg7) {
 693                         goto clear_dr7;
 694                 }
 695         }
 696
 697         tsk->thread.debugreg6 = condition;
 698
 699         /* Mask out spurious TF errors due to lazy TF clearing */
 700         if (condition & DR_STEP) {
 701                 /*
 702                  * The TF error should be masked out only if the current
 703                  * process is not traced and if the TRAP flag has been set
 704                  * previously by a tracing process (condition detected by
 705                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 706                  * can be modified by the process itself in user mode,
 707                  * allowing programs to debug themselves without the ptrace()
 708                  * interface.
 709                  */
 710                 if (!user_mode(regs))
 711                        goto clear_TF_reenable;
 712                 /*
 713                  * Was the TF flag set by a debugger? If so, clear it now,
 714                  * so that register information is correct.
 715                  */
 716                 if (tsk->ptrace & PT_DTRACE) {
 717                         regs->eflags &= ~TF_MASK;
 718                         tsk->ptrace &= ~PT_DTRACE;
 719                 }
 720         }
 721
 722         /* Ok, finally something we can handle */
 723         tsk->thread.trap_no = 1;
 724         tsk->thread.error_code = error_code;
 725         info.si_signo = SIGTRAP;
 726         info.si_errno = 0;
 727         info.si_code = TRAP_BRKPT;
 728         info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
 729         force_sig_info(SIGTRAP, &info, tsk);
 730
 731 clear_dr7:
 732         set_debugreg(0UL, 7);
 733         return;
 734
 735 clear_TF_reenable:
 736         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 737         regs->eflags &= ~TF_MASK;
 738 }
 739
 740 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 741 {
 742         const struct exception_table_entry *fixup;
 743         fixup = search_exception_tables(regs->rip);
 744         if (fixup) {
 745                 regs->rip = fixup->fixup;
 746                 return 1;
 747         }
 748         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 749         /* Illegal floating point operation in the kernel */
 750         current->thread.trap_no = trapnr;
 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", 16))
 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", 19))
 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 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 892 {
 893 }
 894
 895 /*
 896  *  'math_state_restore()' saves the current math information in the
 897  * old math state array, and gets the new ones from the current task
 898  *
 899  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 900  * Don't touch unless you *really* know how it works.
 901  */
 902 asmlinkage void math_state_restore(void)
 903 {
 904         struct task_struct *me = current;
 905         clts();                 /* Allow maths ops (or we recurse) */
 906
 907         if (!used_math())
 908                 init_fpu(me);
 909         restore_fpu_checking(&me->thread.i387.fxsave);
 910         me->thread_info->status |= TS_USEDFPU;
 911 }
 912
 913 void __init trap_init(void)
 914 {
 915         set_intr_gate(0,&divide_error);
 916         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 917         set_intr_gate_ist(2,&nmi,NMI_STACK);
 918         set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
 919         set_system_gate(4,&overflow);   /* int4 can be called from all */
 920         set_intr_gate(5,&bounds);
 921         set_intr_gate(6,&invalid_op);
 922         set_intr_gate(7,&device_not_available);
 923         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 924         set_intr_gate(9,&coprocessor_segment_overrun);
 925         set_intr_gate(10,&invalid_TSS);
 926         set_intr_gate(11,&segment_not_present);
 927         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 928         set_intr_gate(13,&general_protection);
 929         set_intr_gate(14,&page_fault);
 930         set_intr_gate(15,&spurious_interrupt_bug);
 931         set_intr_gate(16,&coprocessor_error);
 932         set_intr_gate(17,&alignment_check);
 933 #ifdef CONFIG_X86_MCE
 934         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 935 #endif
 936         set_intr_gate(19,&simd_coprocessor_error);
 937
 938 #ifdef CONFIG_IA32_EMULATION
 939         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 940 #endif
 941
 942         /*
 943          * Should be a barrier for any external CPU state.
 944          */
 945         cpu_init();
 946 }
 947
 948
 949 /* Actual parsing is done early in setup.c. */
 950 static int __init oops_dummy(char *s)
 951 {
 952         panic_on_oops = 1;
 953         return -1;
 954 }
 955 __setup("oops=", oops_dummy);
 956
 957 static int __init kstack_setup(char *s)
 958 {
 959         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 960         return 0;
 961 }
 962 __setup("kstack=", kstack_setup);
 963