arch/x86/kernel/traps_32.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *
   4  *  Pentium III FXSR, SSE support
   5  *      Gareth Hughes <gareth@valinux.com>, May 2000
   6  */
   7
   8 /*
   9  * 'Traps.c' handles hardware traps and faults after we have saved some
  10  * state in 'asm.s'.
  11  */
  12 #include <linux/sched.h>
  13 #include <linux/kernel.h>
  14 #include <linux/string.h>
  15 #include <linux/errno.h>
  16 #include <linux/timer.h>
  17 #include <linux/mm.h>
  18 #include <linux/init.h>
  19 #include <linux/delay.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/highmem.h>
  23 #include <linux/kallsyms.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/utsname.h>
  26 #include <linux/kprobes.h>
  27 #include <linux/kexec.h>
  28 #include <linux/unwind.h>
  29 #include <linux/uaccess.h>
  30 #include <linux/nmi.h>
  31 #include <linux/bug.h>
  32
  33 #ifdef CONFIG_EISA
  34 #include <linux/ioport.h>
  35 #include <linux/eisa.h>
  36 #endif
  37
  38 #ifdef CONFIG_MCA
  39 #include <linux/mca.h>
  40 #endif
  41
  42 #if defined(CONFIG_EDAC)
  43 #include <linux/edac.h>
  44 #endif
  45
  46 #include <asm/processor.h>
  47 #include <asm/system.h>
  48 #include <asm/io.h>
  49 #include <asm/atomic.h>
  50 #include <asm/debugreg.h>
  51 #include <asm/desc.h>
  52 #include <asm/i387.h>
  53 #include <asm/nmi.h>
  54 #include <asm/unwind.h>
  55 #include <asm/smp.h>
  56 #include <asm/arch_hooks.h>
  57 #include <linux/kdebug.h>
  58 #include <asm/stacktrace.h>
  59
  60 #include <linux/module.h>
  61
  62 #include "mach_traps.h"
  63
  64 int panic_on_unrecovered_nmi;
  65
  66 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  67 EXPORT_SYMBOL_GPL(used_vectors);
  68
  69 asmlinkage int system_call(void);
  70
  71 /* Do we ignore FPU interrupts ? */
  72 char ignore_fpu_irq = 0;
  73
  74 /*
  75  * The IDT has to be page-aligned to simplify the Pentium
  76  * F0 0F bug workaround.. We have a special link segment
  77  * for this.
  78  */
  79 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
  80
  81 asmlinkage void divide_error(void);
  82 asmlinkage void debug(void);
  83 asmlinkage void nmi(void);
  84 asmlinkage void int3(void);
  85 asmlinkage void overflow(void);
  86 asmlinkage void bounds(void);
  87 asmlinkage void invalid_op(void);
  88 asmlinkage void device_not_available(void);
  89 asmlinkage void coprocessor_segment_overrun(void);
  90 asmlinkage void invalid_TSS(void);
  91 asmlinkage void segment_not_present(void);
  92 asmlinkage void stack_segment(void);
  93 asmlinkage void general_protection(void);
  94 asmlinkage void page_fault(void);
  95 asmlinkage void coprocessor_error(void);
  96 asmlinkage void simd_coprocessor_error(void);
  97 asmlinkage void alignment_check(void);
  98 asmlinkage void spurious_interrupt_bug(void);
  99 asmlinkage void machine_check(void);
 100
 101 int kstack_depth_to_print = 24;
 102 static unsigned int code_bytes = 64;
 103
 104 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
 105 {
 106         return  p > (void *)tinfo &&
 107                 p <= (void *)tinfo + THREAD_SIZE - size;
 108 }
 109
 110 /* The form of the top of the frame on the stack */
 111 struct stack_frame {
 112         struct stack_frame *next_frame;
 113         unsigned long return_address;
 114 };
 115
 116 static inline unsigned long print_context_stack(struct thread_info *tinfo,
 117                                 unsigned long *stack, unsigned long ebp,
 118                                 const struct stacktrace_ops *ops, void *data)
 119 {
 120 #ifdef  CONFIG_FRAME_POINTER
 121         struct stack_frame *frame = (struct stack_frame *)ebp;
 122         while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
 123                 struct stack_frame *next;
 124                 unsigned long addr;
 125
 126                 addr = frame->return_address;
 127                 ops->address(data, addr);
 128                 /*
 129                  * break out of recursive entries (such as
 130                  * end_of_stack_stop_unwind_function). Also,
 131                  * we can never allow a frame pointer to
 132                  * move downwards!
 133                  */
 134                 next = frame->next_frame;
 135                 if (next <= frame)
 136                         break;
 137                 frame = next;
 138         }
 139 #else
 140         while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
 141                 unsigned long addr;
 142
 143                 addr = *stack++;
 144                 if (__kernel_text_address(addr))
 145                         ops->address(data, addr);
 146         }
 147 #endif
 148         return ebp;
 149 }
 150
 151 #define MSG(msg) ops->warning(data, msg)
 152
 153 void dump_trace(struct task_struct *task, struct pt_regs *regs,
 154                 unsigned long *stack,
 155                 const struct stacktrace_ops *ops, void *data)
 156 {
 157         unsigned long ebp = 0;
 158
 159         if (!task)
 160                 task = current;
 161
 162         if (!stack) {
 163                 unsigned long dummy;
 164                 stack = &dummy;
 165                 if (task != current)
 166                         stack = (unsigned long *)task->thread.esp;
 167         }
 168
 169 #ifdef CONFIG_FRAME_POINTER
 170         if (!ebp) {
 171                 if (task == current) {
 172                         /* Grab ebp right from our regs */
 173                         asm ("movl %%ebp, %0" : "=r" (ebp) : );
 174                 } else {
 175                         /* ebp is the last reg pushed by switch_to */
 176                         ebp = *(unsigned long *) task->thread.esp;
 177                 }
 178         }
 179 #endif
 180
 181         while (1) {
 182                 struct thread_info *context;
 183                 context = (struct thread_info *)
 184                         ((unsigned long)stack & (~(THREAD_SIZE - 1)));
 185                 ebp = print_context_stack(context, stack, ebp, ops, data);
 186                 /* Should be after the line below, but somewhere
 187                    in early boot context comes out corrupted and we
 188                    can't reference it -AK */
 189                 if (ops->stack(data, "IRQ") < 0)
 190                         break;
 191                 stack = (unsigned long*)context->previous_esp;
 192                 if (!stack)
 193                         break;
 194                 touch_nmi_watchdog();
 195         }
 196 }
 197 EXPORT_SYMBOL(dump_trace);
 198
 199 static void
 200 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 201 {
 202         printk(data);
 203         print_symbol(msg, symbol);
 204         printk("\n");
 205 }
 206
 207 static void print_trace_warning(void *data, char *msg)
 208 {
 209         printk("%s%s\n", (char *)data, msg);
 210 }
 211
 212 static int print_trace_stack(void *data, char *name)
 213 {
 214         return 0;
 215 }
 216
 217 /*
 218  * Print one address/symbol entries per line.
 219  */
 220 static void print_trace_address(void *data, unsigned long addr)
 221 {
 222         printk("%s [<%08lx>] ", (char *)data, addr);
 223         print_symbol("%s\n", addr);
 224         touch_nmi_watchdog();
 225 }
 226
 227 static const struct stacktrace_ops print_trace_ops = {
 228         .warning = print_trace_warning,
 229         .warning_symbol = print_trace_warning_symbol,
 230         .stack = print_trace_stack,
 231         .address = print_trace_address,
 232 };
 233
 234 static void
 235 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 236                    unsigned long * stack, char *log_lvl)
 237 {
 238         dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
 239         printk("%s =======================\n", log_lvl);
 240 }
 241
 242 void show_trace(struct task_struct *task, struct pt_regs *regs,
 243                 unsigned long * stack)
 244 {
 245         show_trace_log_lvl(task, regs, stack, "");
 246 }
 247
 248 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
 249                                unsigned long *esp, char *log_lvl)
 250 {
 251         unsigned long *stack;
 252         int i;
 253
 254         if (esp == NULL) {
 255                 if (task)
 256                         esp = (unsigned long*)task->thread.esp;
 257                 else
 258                         esp = (unsigned long *)&esp;
 259         }
 260
 261         stack = esp;
 262         for(i = 0; i < kstack_depth_to_print; i++) {
 263                 if (kstack_end(stack))
 264                         break;
 265                 if (i && ((i % 8) == 0))
 266                         printk("\n%s       ", log_lvl);
 267                 printk("%08lx ", *stack++);
 268         }
 269         printk("\n%sCall Trace:\n", log_lvl);
 270         show_trace_log_lvl(task, regs, esp, log_lvl);
 271 }
 272
 273 void show_stack(struct task_struct *task, unsigned long *esp)
 274 {
 275         printk("       ");
 276         show_stack_log_lvl(task, NULL, esp, "");
 277 }
 278
 279 /*
 280  * The architecture-independent dump_stack generator
 281  */
 282 void dump_stack(void)
 283 {
 284         unsigned long stack;
 285
 286         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 287                 current->pid, current->comm, print_tainted(),
 288                 init_utsname()->release,
 289                 (int)strcspn(init_utsname()->version, " "),
 290                 init_utsname()->version);
 291         show_trace(current, NULL, &stack);
 292 }
 293
 294 EXPORT_SYMBOL(dump_stack);
 295
 296 void show_registers(struct pt_regs *regs)
 297 {
 298         int i;
 299
 300         print_modules();
 301         __show_registers(regs, 0);
 302         printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
 303                 TASK_COMM_LEN, current->comm, task_pid_nr(current),
 304                 current_thread_info(), current, task_thread_info(current));
 305         /*
 306          * When in-kernel, we also print out the stack and code at the
 307          * time of the fault..
 308          */
 309         if (!user_mode_vm(regs)) {
 310                 u8 *eip;
 311                 unsigned int code_prologue = code_bytes * 43 / 64;
 312                 unsigned int code_len = code_bytes;
 313                 unsigned char c;
 314
 315                 printk("\n" KERN_EMERG "Stack: ");
 316                 show_stack_log_lvl(NULL, regs, &regs->esp, KERN_EMERG);
 317
 318                 printk(KERN_EMERG "Code: ");
 319
 320                 eip = (u8 *)regs->eip - code_prologue;
 321                 if (eip < (u8 *)PAGE_OFFSET ||
 322                         probe_kernel_address(eip, c)) {
 323                         /* try starting at EIP */
 324                         eip = (u8 *)regs->eip;
 325                         code_len = code_len - code_prologue + 1;
 326                 }
 327                 for (i = 0; i < code_len; i++, eip++) {
 328                         if (eip < (u8 *)PAGE_OFFSET ||
 329                                 probe_kernel_address(eip, c)) {
 330                                 printk(" Bad EIP value.");
 331                                 break;
 332                         }
 333                         if (eip == (u8 *)regs->eip)
 334                                 printk("<%02x> ", c);
 335                         else
 336                                 printk("%02x ", c);
 337                 }
 338         }
 339         printk("\n");
 340 }
 341
 342 int is_valid_bugaddr(unsigned long eip)
 343 {
 344         unsigned short ud2;
 345
 346         if (eip < PAGE_OFFSET)
 347                 return 0;
 348         if (probe_kernel_address((unsigned short *)eip, ud2))
 349                 return 0;
 350
 351         return ud2 == 0x0b0f;
 352 }
 353
 354 /*
 355  * This is gone through when something in the kernel has done something bad and
 356  * is about to be terminated.
 357  */
 358 void die(const char * str, struct pt_regs * regs, long err)
 359 {
 360         static struct {
 361                 raw_spinlock_t lock;
 362                 u32 lock_owner;
 363                 int lock_owner_depth;
 364         } die = {
 365                 .lock =                 __RAW_SPIN_LOCK_UNLOCKED,
 366                 .lock_owner =           -1,
 367                 .lock_owner_depth =     0
 368         };
 369         static int die_counter;
 370         unsigned long flags;
 371
 372         oops_enter();
 373
 374         if (die.lock_owner != raw_smp_processor_id()) {
 375                 console_verbose();
 376                 __raw_spin_lock(&die.lock);
 377                 raw_local_save_flags(flags);
 378                 die.lock_owner = smp_processor_id();
 379                 die.lock_owner_depth = 0;
 380                 bust_spinlocks(1);
 381         }
 382         else
 383                 raw_local_save_flags(flags);
 384
 385         if (++die.lock_owner_depth < 3) {
 386                 unsigned long esp;
 387                 unsigned short ss;
 388
 389                 report_bug(regs->eip, regs);
 390
 391                 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff,
 392                        ++die_counter);
 393 #ifdef CONFIG_PREEMPT
 394                 printk("PREEMPT ");
 395 #endif
 396 #ifdef CONFIG_SMP
 397                 printk("SMP ");
 398 #endif
 399 #ifdef CONFIG_DEBUG_PAGEALLOC
 400                 printk("DEBUG_PAGEALLOC");
 401 #endif
 402                 printk("\n");
 403
 404                 if (notify_die(DIE_OOPS, str, regs, err,
 405                                         current->thread.trap_no, SIGSEGV) !=
 406                                 NOTIFY_STOP) {
 407                         show_registers(regs);
 408                         /* Executive summary in case the oops scrolled away */
 409                         esp = (unsigned long) (&regs->esp);
 410                         savesegment(ss, ss);
 411                         if (user_mode(regs)) {
 412                                 esp = regs->esp;
 413                                 ss = regs->xss & 0xffff;
 414                         }
 415                         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
 416                         print_symbol("%s", regs->eip);
 417                         printk(" SS:ESP %04x:%08lx\n", ss, esp);
 418                 }
 419                 else
 420                         regs = NULL;
 421         } else
 422                 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
 423
 424         bust_spinlocks(0);
 425         die.lock_owner = -1;
 426         add_taint(TAINT_DIE);
 427         __raw_spin_unlock(&die.lock);
 428         raw_local_irq_restore(flags);
 429
 430         if (!regs)
 431                 return;
 432
 433         if (kexec_should_crash(current))
 434                 crash_kexec(regs);
 435
 436         if (in_interrupt())
 437                 panic("Fatal exception in interrupt");
 438
 439         if (panic_on_oops)
 440                 panic("Fatal exception");
 441
 442         oops_exit();
 443         do_exit(SIGSEGV);
 444 }
 445
 446 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 447 {
 448         if (!user_mode_vm(regs))
 449                 die(str, regs, err);
 450 }
 451
 452 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
 453                               struct pt_regs * regs, long error_code,
 454                               siginfo_t *info)
 455 {
 456         struct task_struct *tsk = current;
 457
 458         if (regs->eflags & VM_MASK) {
 459                 if (vm86)
 460                         goto vm86_trap;
 461                 goto trap_signal;
 462         }
 463
 464         if (!user_mode(regs))
 465                 goto kernel_trap;
 466
 467         trap_signal: {
 468                 /*
 469                  * We want error_code and trap_no set for userspace faults and
 470                  * kernelspace faults which result in die(), but not
 471                  * kernelspace faults which are fixed up.  die() gives the
 472                  * process no chance to handle the signal and notice the
 473                  * kernel fault information, so that won't result in polluting
 474                  * the information about previously queued, but not yet
 475                  * delivered, faults.  See also do_general_protection below.
 476                  */
 477                 tsk->thread.error_code = error_code;
 478                 tsk->thread.trap_no = trapnr;
 479
 480                 if (info)
 481                         force_sig_info(signr, info, tsk);
 482                 else
 483                         force_sig(signr, tsk);
 484                 return;
 485         }
 486
 487         kernel_trap: {
 488                 if (!fixup_exception(regs)) {
 489                         tsk->thread.error_code = error_code;
 490                         tsk->thread.trap_no = trapnr;
 491                         die(str, regs, error_code);
 492                 }
 493                 return;
 494         }
 495
 496         vm86_trap: {
 497                 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
 498                 if (ret) goto trap_signal;
 499                 return;
 500         }
 501 }
 502
 503 #define DO_ERROR(trapnr, signr, str, name) \
 504 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 505 { \
 506         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 507                                                 == NOTIFY_STOP) \
 508                 return; \
 509         do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
 510 }
 511
 512 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
 513 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 514 { \
 515         siginfo_t info; \
 516         if (irq) \
 517                 local_irq_enable(); \
 518         info.si_signo = signr; \
 519         info.si_errno = 0; \
 520         info.si_code = sicode; \
 521         info.si_addr = (void __user *)siaddr; \
 522         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 523                                                 == NOTIFY_STOP) \
 524                 return; \
 525         do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
 526 }
 527
 528 #define DO_VM86_ERROR(trapnr, signr, str, name) \
 529 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 530 { \
 531         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 532                                                 == NOTIFY_STOP) \
 533                 return; \
 534         do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
 535 }
 536
 537 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 538 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 539 { \
 540         siginfo_t info; \
 541         info.si_signo = signr; \
 542         info.si_errno = 0; \
 543         info.si_code = sicode; \
 544         info.si_addr = (void __user *)siaddr; \
 545         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 546                                                 == NOTIFY_STOP) \
 547                 return; \
 548         do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
 549 }
 550
 551 DO_VM86_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->eip)
 552 #ifndef CONFIG_KPROBES
 553 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
 554 #endif
 555 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
 556 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
 557 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip, 0)
 558 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 559 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 560 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 561 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 562 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
 563 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
 564
 565 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
 566                                               long error_code)
 567 {
 568         int cpu = get_cpu();
 569         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 570         struct thread_struct *thread = &current->thread;
 571
 572         /*
 573          * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 574          * invalid offset set (the LAZY one) and the faulting thread has
 575          * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
 576          * and we set the offset field correctly. Then we let the CPU to
 577          * restart the faulting instruction.
 578          */
 579         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 580             thread->io_bitmap_ptr) {
 581                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 582                        thread->io_bitmap_max);
 583                 /*
 584                  * If the previously set map was extending to higher ports
 585                  * than the current one, pad extra space with 0xff (no access).
 586                  */
 587                 if (thread->io_bitmap_max < tss->io_bitmap_max)
 588                         memset((char *) tss->io_bitmap +
 589                                 thread->io_bitmap_max, 0xff,
 590                                 tss->io_bitmap_max - thread->io_bitmap_max);
 591                 tss->io_bitmap_max = thread->io_bitmap_max;
 592                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 593                 tss->io_bitmap_owner = thread;
 594                 put_cpu();
 595                 return;
 596         }
 597         put_cpu();
 598
 599         if (regs->eflags & VM_MASK)
 600                 goto gp_in_vm86;
 601
 602         if (!user_mode(regs))
 603                 goto gp_in_kernel;
 604
 605         current->thread.error_code = error_code;
 606         current->thread.trap_no = 13;
 607         if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
 608             printk_ratelimit())
 609                 printk(KERN_INFO
 610                     "%s[%d] general protection eip:%lx esp:%lx error:%lx\n",
 611                     current->comm, task_pid_nr(current),
 612                     regs->eip, regs->esp, error_code);
 613
 614         force_sig(SIGSEGV, current);
 615         return;
 616
 617 gp_in_vm86:
 618         local_irq_enable();
 619         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 620         return;
 621
 622 gp_in_kernel:
 623         if (!fixup_exception(regs)) {
 624                 current->thread.error_code = error_code;
 625                 current->thread.trap_no = 13;
 626                 if (notify_die(DIE_GPF, "general protection fault", regs,
 627                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 628                         return;
 629                 die("general protection fault", regs, error_code);
 630         }
 631 }
 632
 633 static __kprobes void
 634 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 635 {
 636         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
 637                 "CPU %d.\n", reason, smp_processor_id());
 638         printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 639
 640 #if defined(CONFIG_EDAC)
 641         if(edac_handler_set()) {
 642                 edac_atomic_assert_error();
 643                 return;
 644         }
 645 #endif
 646
 647         if (panic_on_unrecovered_nmi)
 648                 panic("NMI: Not continuing");
 649
 650         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 651
 652         /* Clear and disable the memory parity error line. */
 653         clear_mem_error(reason);
 654 }
 655
 656 static __kprobes void
 657 io_check_error(unsigned char reason, struct pt_regs * regs)
 658 {
 659         unsigned long i;
 660
 661         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 662         show_registers(regs);
 663
 664         /* Re-enable the IOCK line, wait for a few seconds */
 665         reason = (reason & 0xf) | 8;
 666         outb(reason, 0x61);
 667         i = 2000;
 668         while (--i) udelay(1000);
 669         reason &= ~8;
 670         outb(reason, 0x61);
 671 }
 672
 673 static __kprobes void
 674 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 675 {
 676 #ifdef CONFIG_MCA
 677         /* Might actually be able to figure out what the guilty party
 678         * is. */
 679         if( MCA_bus ) {
 680                 mca_handle_nmi();
 681                 return;
 682         }
 683 #endif
 684         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
 685                 "CPU %d.\n", reason, smp_processor_id());
 686         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 687         if (panic_on_unrecovered_nmi)
 688                 panic("NMI: Not continuing");
 689
 690         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 691 }
 692
 693 static DEFINE_SPINLOCK(nmi_print_lock);
 694
 695 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
 696 {
 697         if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
 698             NOTIFY_STOP)
 699                 return;
 700
 701         spin_lock(&nmi_print_lock);
 702         /*
 703         * We are in trouble anyway, lets at least try
 704         * to get a message out.
 705         */
 706         bust_spinlocks(1);
 707         printk(KERN_EMERG "%s", msg);
 708         printk(" on CPU%d, eip %08lx, registers:\n",
 709                 smp_processor_id(), regs->eip);
 710         show_registers(regs);
 711         console_silent();
 712         spin_unlock(&nmi_print_lock);
 713         bust_spinlocks(0);
 714
 715         /* If we are in kernel we are probably nested up pretty bad
 716          * and might aswell get out now while we still can.
 717         */
 718         if (!user_mode_vm(regs)) {
 719                 current->thread.trap_no = 2;
 720                 crash_kexec(regs);
 721         }
 722
 723         do_exit(SIGSEGV);
 724 }
 725
 726 static __kprobes void default_do_nmi(struct pt_regs * regs)
 727 {
 728         unsigned char reason = 0;
 729
 730         /* Only the BSP gets external NMIs from the system.  */
 731         if (!smp_processor_id())
 732                 reason = get_nmi_reason();
 733
 734         if (!(reason & 0xc0)) {
 735                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 736                                                         == NOTIFY_STOP)
 737                         return;
 738 #ifdef CONFIG_X86_LOCAL_APIC
 739                 /*
 740                  * Ok, so this is none of the documented NMI sources,
 741                  * so it must be the NMI watchdog.
 742                  */
 743                 if (nmi_watchdog_tick(regs, reason))
 744                         return;
 745                 if (!do_nmi_callback(regs, smp_processor_id()))
 746 #endif
 747                         unknown_nmi_error(reason, regs);
 748
 749                 return;
 750         }
 751         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 752                 return;
 753         if (reason & 0x80)
 754                 mem_parity_error(reason, regs);
 755         if (reason & 0x40)
 756                 io_check_error(reason, regs);
 757         /*
 758          * Reassert NMI in case it became active meanwhile
 759          * as it's edge-triggered.
 760          */
 761         reassert_nmi();
 762 }
 763
 764 static int ignore_nmis;
 765
 766 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
 767 {
 768         int cpu;
 769
 770         nmi_enter();
 771
 772         cpu = smp_processor_id();
 773
 774         ++nmi_count(cpu);
 775
 776         if (!ignore_nmis)
 777                 default_do_nmi(regs);
 778
 779         nmi_exit();
 780 }
 781
 782 void stop_nmi(void)
 783 {
 784         acpi_nmi_disable();
 785         ignore_nmis++;
 786 }
 787
 788 void restart_nmi(void)
 789 {
 790         ignore_nmis--;
 791         acpi_nmi_enable();
 792 }
 793
 794 #ifdef CONFIG_KPROBES
 795 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
 796 {
 797         trace_hardirqs_fixup();
 798
 799         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 800                         == NOTIFY_STOP)
 801                 return;
 802         /* This is an interrupt gate, because kprobes wants interrupts
 803         disabled.  Normal trap handlers don't. */
 804         restore_interrupts(regs);
 805         do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
 806 }
 807 #endif
 808
 809 /*
 810  * Our handling of the processor debug registers is non-trivial.
 811  * We do not clear them on entry and exit from the kernel. Therefore
 812  * it is possible to get a watchpoint trap here from inside the kernel.
 813  * However, the code in ./ptrace.c has ensured that the user can
 814  * only set watchpoints on userspace addresses. Therefore the in-kernel
 815  * watchpoint trap can only occur in code which is reading/writing
 816  * from user space. Such code must not hold kernel locks (since it
 817  * can equally take a page fault), therefore it is safe to call
 818  * force_sig_info even though that claims and releases locks.
 819  *
 820  * Code in ./signal.c ensures that the debug control register
 821  * is restored before we deliver any signal, and therefore that
 822  * user code runs with the correct debug control register even though
 823  * we clear it here.
 824  *
 825  * Being careful here means that we don't have to be as careful in a
 826  * lot of more complicated places (task switching can be a bit lazy
 827  * about restoring all the debug state, and ptrace doesn't have to
 828  * find every occurrence of the TF bit that could be saved away even
 829  * by user code)
 830  */
 831 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
 832 {
 833         unsigned int condition;
 834         struct task_struct *tsk = current;
 835
 836         trace_hardirqs_fixup();
 837
 838         get_debugreg(condition, 6);
 839
 840         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 841                                         SIGTRAP) == NOTIFY_STOP)
 842                 return;
 843         /* It's safe to allow irq's after DR6 has been saved */
 844         if (regs->eflags & X86_EFLAGS_IF)
 845                 local_irq_enable();
 846
 847         /* Mask out spurious debug traps due to lazy DR7 setting */
 848         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 849                 if (!tsk->thread.debugreg[7])
 850                         goto clear_dr7;
 851         }
 852
 853         if (regs->eflags & VM_MASK)
 854                 goto debug_vm86;
 855
 856         /* Save debug status register where ptrace can see it */
 857         tsk->thread.debugreg[6] = condition;
 858
 859         /*
 860          * Single-stepping through TF: make sure we ignore any events in
 861          * kernel space (but re-enable TF when returning to user mode).
 862          */
 863         if (condition & DR_STEP) {
 864                 /*
 865                  * We already checked v86 mode above, so we can
 866                  * check for kernel mode by just checking the CPL
 867                  * of CS.
 868                  */
 869                 if (!user_mode(regs))
 870                         goto clear_TF_reenable;
 871         }
 872
 873         /* Ok, finally something we can handle */
 874         send_sigtrap(tsk, regs, error_code);
 875
 876         /* Disable additional traps. They'll be re-enabled when
 877          * the signal is delivered.
 878          */
 879 clear_dr7:
 880         set_debugreg(0, 7);
 881         return;
 882
 883 debug_vm86:
 884         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 885         return;
 886
 887 clear_TF_reenable:
 888         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 889         regs->eflags &= ~TF_MASK;
 890         return;
 891 }
 892
 893 /*
 894  * Note that we play around with the 'TS' bit in an attempt to get
 895  * the correct behaviour even in the presence of the asynchronous
 896  * IRQ13 behaviour
 897  */
 898 void math_error(void __user *eip)
 899 {
 900         struct task_struct * task;
 901         siginfo_t info;
 902         unsigned short cwd, swd;
 903
 904         /*
 905          * Save the info for the exception handler and clear the error.
 906          */
 907         task = current;
 908         save_init_fpu(task);
 909         task->thread.trap_no = 16;
 910         task->thread.error_code = 0;
 911         info.si_signo = SIGFPE;
 912         info.si_errno = 0;
 913         info.si_code = __SI_FAULT;
 914         info.si_addr = eip;
 915         /*
 916          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 917          * status.  0x3f is the exception bits in these regs, 0x200 is the
 918          * C1 reg you need in case of a stack fault, 0x040 is the stack
 919          * fault bit.  We should only be taking one exception at a time,
 920          * so if this combination doesn't produce any single exception,
 921          * then we have a bad program that isn't syncronizing its FPU usage
 922          * and it will suffer the consequences since we won't be able to
 923          * fully reproduce the context of the exception
 924          */
 925         cwd = get_fpu_cwd(task);
 926         swd = get_fpu_swd(task);
 927         switch (swd & ~cwd & 0x3f) {
 928                 case 0x000: /* No unmasked exception */
 929                         return;
 930                 default:    /* Multiple exceptions */
 931                         break;
 932                 case 0x001: /* Invalid Op */
 933                         /*
 934                          * swd & 0x240 == 0x040: Stack Underflow
 935                          * swd & 0x240 == 0x240: Stack Overflow
 936                          * User must clear the SF bit (0x40) if set
 937                          */
 938                         info.si_code = FPE_FLTINV;
 939                         break;
 940                 case 0x002: /* Denormalize */
 941                 case 0x010: /* Underflow */
 942                         info.si_code = FPE_FLTUND;
 943                         break;
 944                 case 0x004: /* Zero Divide */
 945                         info.si_code = FPE_FLTDIV;
 946                         break;
 947                 case 0x008: /* Overflow */
 948                         info.si_code = FPE_FLTOVF;
 949                         break;
 950                 case 0x020: /* Precision */
 951                         info.si_code = FPE_FLTRES;
 952                         break;
 953         }
 954         force_sig_info(SIGFPE, &info, task);
 955 }
 956
 957 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
 958 {
 959         ignore_fpu_irq = 1;
 960         math_error((void __user *)regs->eip);
 961 }
 962
 963 static void simd_math_error(void __user *eip)
 964 {
 965         struct task_struct * task;
 966         siginfo_t info;
 967         unsigned short mxcsr;
 968
 969         /*
 970          * Save the info for the exception handler and clear the error.
 971          */
 972         task = current;
 973         save_init_fpu(task);
 974         task->thread.trap_no = 19;
 975         task->thread.error_code = 0;
 976         info.si_signo = SIGFPE;
 977         info.si_errno = 0;
 978         info.si_code = __SI_FAULT;
 979         info.si_addr = eip;
 980         /*
 981          * The SIMD FPU exceptions are handled a little differently, as there
 982          * is only a single status/control register.  Thus, to determine which
 983          * unmasked exception was caught we must mask the exception mask bits
 984          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 985          */
 986         mxcsr = get_fpu_mxcsr(task);
 987         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 988                 case 0x000:
 989                 default:
 990                         break;
 991                 case 0x001: /* Invalid Op */
 992                         info.si_code = FPE_FLTINV;
 993                         break;
 994                 case 0x002: /* Denormalize */
 995                 case 0x010: /* Underflow */
 996                         info.si_code = FPE_FLTUND;
 997                         break;
 998                 case 0x004: /* Zero Divide */
 999                         info.si_code = FPE_FLTDIV;
1000                         break;
1001                 case 0x008: /* Overflow */
1002                         info.si_code = FPE_FLTOVF;
1003                         break;
1004                 case 0x020: /* Precision */
1005                         info.si_code = FPE_FLTRES;
1006                         break;
1007         }
1008         force_sig_info(SIGFPE, &info, task);
1009 }
1010
1011 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1012                                           long error_code)
1013 {
1014         if (cpu_has_xmm) {
1015                 /* Handle SIMD FPU exceptions on PIII+ processors. */
1016                 ignore_fpu_irq = 1;
1017                 simd_math_error((void __user *)regs->eip);
1018         } else {
1019                 /*
1020                  * Handle strange cache flush from user space exception
1021                  * in all other cases.  This is undocumented behaviour.
1022                  */
1023                 if (regs->eflags & VM_MASK) {
1024                         handle_vm86_fault((struct kernel_vm86_regs *)regs,
1025                                           error_code);
1026                         return;
1027                 }
1028                 current->thread.trap_no = 19;
1029                 current->thread.error_code = error_code;
1030                 die_if_kernel("cache flush denied", regs, error_code);
1031                 force_sig(SIGSEGV, current);
1032         }
1033 }
1034
1035 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1036                                           long error_code)
1037 {
1038 #if 0
1039         /* No need to warn about this any longer. */
1040         printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1041 #endif
1042 }
1043
1044 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1045                                           unsigned long kesp)
1046 {
1047         struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1048         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1049         unsigned long new_kesp = kesp - base;
1050         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1051         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1052         /* Set up base for espfix segment */
1053         desc &= 0x00f0ff0000000000ULL;
1054         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1055                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1056                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1057                 (lim_pages & 0xffff);
1058         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1059         return new_kesp;
1060 }
1061
1062 /*
1063  *  'math_state_restore()' saves the current math information in the
1064  * old math state array, and gets the new ones from the current task
1065  *
1066  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1067  * Don't touch unless you *really* know how it works.
1068  *
1069  * Must be called with kernel preemption disabled (in this case,
1070  * local interrupts are disabled at the call-site in entry.S).
1071  */
1072 asmlinkage void math_state_restore(void)
1073 {
1074         struct thread_info *thread = current_thread_info();
1075         struct task_struct *tsk = thread->task;
1076
1077         clts();         /* Allow maths ops (or we recurse) */
1078         if (!tsk_used_math(tsk))
1079                 init_fpu(tsk);
1080         restore_fpu(tsk);
1081         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
1082         tsk->fpu_counter++;
1083 }
1084 EXPORT_SYMBOL_GPL(math_state_restore);
1085
1086 #ifndef CONFIG_MATH_EMULATION
1087
1088 asmlinkage void math_emulate(long arg)
1089 {
1090         printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1091         printk(KERN_EMERG "killing %s.\n",current->comm);
1092         force_sig(SIGFPE,current);
1093         schedule();
1094 }
1095
1096 #endif /* CONFIG_MATH_EMULATION */
1097
1098 /*
1099  * This needs to use 'idt_table' rather than 'idt', and
1100  * thus use the _nonmapped_ version of the IDT, as the
1101  * Pentium F0 0F bugfix can have resulted in the mapped
1102  * IDT being write-protected.
1103  */
1104 void set_intr_gate(unsigned int n, void *addr)
1105 {
1106         _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1107 }
1108
1109 /*
1110  * This routine sets up an interrupt gate at directory privilege level 3.
1111  */
1112 static inline void set_system_intr_gate(unsigned int n, void *addr)
1113 {
1114         _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1115 }
1116
1117 static void __init set_trap_gate(unsigned int n, void *addr)
1118 {
1119         _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1120 }
1121
1122 static void __init set_system_gate(unsigned int n, void *addr)
1123 {
1124         _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1125 }
1126
1127 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1128 {
1129         _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1130 }
1131
1132
1133 void __init trap_init(void)
1134 {
1135         int i;
1136
1137 #ifdef CONFIG_EISA
1138         void __iomem *p = ioremap(0x0FFFD9, 4);
1139         if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1140                 EISA_bus = 1;
1141         }
1142         iounmap(p);
1143 #endif
1144
1145 #ifdef CONFIG_X86_LOCAL_APIC
1146         init_apic_mappings();
1147 #endif
1148
1149         set_trap_gate(0,&divide_error);
1150         set_intr_gate(1,&debug);
1151         set_intr_gate(2,&nmi);
1152         set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1153         set_system_gate(4,&overflow);
1154         set_trap_gate(5,&bounds);
1155         set_trap_gate(6,&invalid_op);
1156         set_trap_gate(7,&device_not_available);
1157         set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1158         set_trap_gate(9,&coprocessor_segment_overrun);
1159         set_trap_gate(10,&invalid_TSS);
1160         set_trap_gate(11,&segment_not_present);
1161         set_trap_gate(12,&stack_segment);
1162         set_trap_gate(13,&general_protection);
1163         set_intr_gate(14,&page_fault);
1164         set_trap_gate(15,&spurious_interrupt_bug);
1165         set_trap_gate(16,&coprocessor_error);
1166         set_trap_gate(17,&alignment_check);
1167 #ifdef CONFIG_X86_MCE
1168         set_trap_gate(18,&machine_check);
1169 #endif
1170         set_trap_gate(19,&simd_coprocessor_error);
1171
1172         if (cpu_has_fxsr) {
1173                 /*
1174                  * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1175                  * Generates a compile-time "error: zero width for bit-field" if
1176                  * the alignment is wrong.
1177                  */
1178                 struct fxsrAlignAssert {
1179                         int _:!(offsetof(struct task_struct,
1180                                         thread.i387.fxsave) & 15);
1181                 };
1182
1183                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1184                 set_in_cr4(X86_CR4_OSFXSR);
1185                 printk("done.\n");
1186         }
1187         if (cpu_has_xmm) {
1188                 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1189                                 "support... ");
1190                 set_in_cr4(X86_CR4_OSXMMEXCPT);
1191                 printk("done.\n");
1192         }
1193
1194         set_system_gate(SYSCALL_VECTOR,&system_call);
1195
1196         /* Reserve all the builtin and the syscall vector. */
1197         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1198                 set_bit(i, used_vectors);
1199         set_bit(SYSCALL_VECTOR, used_vectors);
1200
1201         /*
1202          * Should be a barrier for any external CPU state.
1203          */
1204         cpu_init();
1205
1206         trap_init_hook();
1207 }
1208
1209 static int __init kstack_setup(char *s)
1210 {
1211         kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1212         return 1;
1213 }
1214 __setup("kstack=", kstack_setup);
1215
1216 static int __init code_bytes_setup(char *s)
1217 {
1218         code_bytes = simple_strtoul(s, NULL, 0);
1219         if (code_bytes > 8192)
1220                 code_bytes = 8192;
1221
1222         return 1;
1223 }
1224 __setup("code_bytes=", code_bytes_setup);