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 asmlinkage int system_call(void);
  67
  68 /* Do we ignore FPU interrupts ? */
  69 char ignore_fpu_irq = 0;
  70
  71 /*
  72  * The IDT has to be page-aligned to simplify the Pentium
  73  * F0 0F bug workaround.. We have a special link segment
  74  * for this.
  75  */
  76 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
  77
  78 asmlinkage void divide_error(void);
  79 asmlinkage void debug(void);
  80 asmlinkage void nmi(void);
  81 asmlinkage void int3(void);
  82 asmlinkage void overflow(void);
  83 asmlinkage void bounds(void);
  84 asmlinkage void invalid_op(void);
  85 asmlinkage void device_not_available(void);
  86 asmlinkage void coprocessor_segment_overrun(void);
  87 asmlinkage void invalid_TSS(void);
  88 asmlinkage void segment_not_present(void);
  89 asmlinkage void stack_segment(void);
  90 asmlinkage void general_protection(void);
  91 asmlinkage void page_fault(void);
  92 asmlinkage void coprocessor_error(void);
  93 asmlinkage void simd_coprocessor_error(void);
  94 asmlinkage void alignment_check(void);
  95 asmlinkage void spurious_interrupt_bug(void);
  96 asmlinkage void machine_check(void);
  97
  98 int kstack_depth_to_print = 24;
  99 static unsigned int code_bytes = 64;
 100
 101 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
 102 {
 103         return  p > (void *)tinfo &&
 104                 p <= (void *)tinfo + THREAD_SIZE - size;
 105 }
 106
 107 /* The form of the top of the frame on the stack */
 108 struct stack_frame {
 109         struct stack_frame *next_frame;
 110         unsigned long return_address;
 111 };
 112
 113 static inline unsigned long print_context_stack(struct thread_info *tinfo,
 114                                 unsigned long *stack, unsigned long ebp,
 115                                 struct stacktrace_ops *ops, void *data)
 116 {
 117 #ifdef  CONFIG_FRAME_POINTER
 118         struct stack_frame *frame = (struct stack_frame *)ebp;
 119         while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
 120                 struct stack_frame *next;
 121                 unsigned long addr;
 122
 123                 addr = frame->return_address;
 124                 ops->address(data, addr);
 125                 /*
 126                  * break out of recursive entries (such as
 127                  * end_of_stack_stop_unwind_function). Also,
 128                  * we can never allow a frame pointer to
 129                  * move downwards!
 130                  */
 131                 next = frame->next_frame;
 132                 if (next <= frame)
 133                         break;
 134                 frame = next;
 135         }
 136 #else
 137         while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
 138                 unsigned long addr;
 139
 140                 addr = *stack++;
 141                 if (__kernel_text_address(addr))
 142                         ops->address(data, addr);
 143         }
 144 #endif
 145         return ebp;
 146 }
 147
 148 #define MSG(msg) ops->warning(data, msg)
 149
 150 void dump_trace(struct task_struct *task, struct pt_regs *regs,
 151                 unsigned long *stack,
 152                 struct stacktrace_ops *ops, void *data)
 153 {
 154         unsigned long ebp = 0;
 155
 156         if (!task)
 157                 task = current;
 158
 159         if (!stack) {
 160                 unsigned long dummy;
 161                 stack = &dummy;
 162                 if (task != current)
 163                         stack = (unsigned long *)task->thread.esp;
 164         }
 165
 166 #ifdef CONFIG_FRAME_POINTER
 167         if (!ebp) {
 168                 if (task == current) {
 169                         /* Grab ebp right from our regs */
 170                         asm ("movl %%ebp, %0" : "=r" (ebp) : );
 171                 } else {
 172                         /* ebp is the last reg pushed by switch_to */
 173                         ebp = *(unsigned long *) task->thread.esp;
 174                 }
 175         }
 176 #endif
 177
 178         while (1) {
 179                 struct thread_info *context;
 180                 context = (struct thread_info *)
 181                         ((unsigned long)stack & (~(THREAD_SIZE - 1)));
 182                 ebp = print_context_stack(context, stack, ebp, ops, data);
 183                 /* Should be after the line below, but somewhere
 184                    in early boot context comes out corrupted and we
 185                    can't reference it -AK */
 186                 if (ops->stack(data, "IRQ") < 0)
 187                         break;
 188                 stack = (unsigned long*)context->previous_esp;
 189                 if (!stack)
 190                         break;
 191                 touch_nmi_watchdog();
 192         }
 193 }
 194 EXPORT_SYMBOL(dump_trace);
 195
 196 static void
 197 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 198 {
 199         printk(data);
 200         print_symbol(msg, symbol);
 201         printk("\n");
 202 }
 203
 204 static void print_trace_warning(void *data, char *msg)
 205 {
 206         printk("%s%s\n", (char *)data, msg);
 207 }
 208
 209 static int print_trace_stack(void *data, char *name)
 210 {
 211         return 0;
 212 }
 213
 214 /*
 215  * Print one address/symbol entries per line.
 216  */
 217 static void print_trace_address(void *data, unsigned long addr)
 218 {
 219         printk("%s [<%08lx>] ", (char *)data, addr);
 220         print_symbol("%s\n", addr);
 221         touch_nmi_watchdog();
 222 }
 223
 224 static struct stacktrace_ops print_trace_ops = {
 225         .warning = print_trace_warning,
 226         .warning_symbol = print_trace_warning_symbol,
 227         .stack = print_trace_stack,
 228         .address = print_trace_address,
 229 };
 230
 231 static void
 232 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 233                    unsigned long * stack, char *log_lvl)
 234 {
 235         dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
 236         printk("%s =======================\n", log_lvl);
 237 }
 238
 239 void show_trace(struct task_struct *task, struct pt_regs *regs,
 240                 unsigned long * stack)
 241 {
 242         show_trace_log_lvl(task, regs, stack, "");
 243 }
 244
 245 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
 246                                unsigned long *esp, char *log_lvl)
 247 {
 248         unsigned long *stack;
 249         int i;
 250
 251         if (esp == NULL) {
 252                 if (task)
 253                         esp = (unsigned long*)task->thread.esp;
 254                 else
 255                         esp = (unsigned long *)&esp;
 256         }
 257
 258         stack = esp;
 259         for(i = 0; i < kstack_depth_to_print; i++) {
 260                 if (kstack_end(stack))
 261                         break;
 262                 if (i && ((i % 8) == 0))
 263                         printk("\n%s       ", log_lvl);
 264                 printk("%08lx ", *stack++);
 265         }
 266         printk("\n%sCall Trace:\n", log_lvl);
 267         show_trace_log_lvl(task, regs, esp, log_lvl);
 268 }
 269
 270 void show_stack(struct task_struct *task, unsigned long *esp)
 271 {
 272         printk("       ");
 273         show_stack_log_lvl(task, NULL, esp, "");
 274 }
 275
 276 /*
 277  * The architecture-independent dump_stack generator
 278  */
 279 void dump_stack(void)
 280 {
 281         unsigned long stack;
 282
 283         show_trace(current, NULL, &stack);
 284 }
 285
 286 EXPORT_SYMBOL(dump_stack);
 287
 288 void show_registers(struct pt_regs *regs)
 289 {
 290         int i;
 291         int in_kernel = 1;
 292         unsigned long esp;
 293         unsigned short ss, gs;
 294
 295         esp = (unsigned long) (&regs->esp);
 296         savesegment(ss, ss);
 297         savesegment(gs, gs);
 298         if (user_mode_vm(regs)) {
 299                 in_kernel = 0;
 300                 esp = regs->esp;
 301                 ss = regs->xss & 0xffff;
 302         }
 303         print_modules();
 304         printk(KERN_EMERG "CPU:    %d\n"
 305                 KERN_EMERG "EIP:    %04x:[<%08lx>]    %s VLI\n"
 306                 KERN_EMERG "EFLAGS: %08lx   (%s %.*s)\n",
 307                 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
 308                 print_tainted(), regs->eflags, init_utsname()->release,
 309                 (int)strcspn(init_utsname()->version, " "),
 310                 init_utsname()->version);
 311         print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
 312         printk(KERN_EMERG "eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
 313                 regs->eax, regs->ebx, regs->ecx, regs->edx);
 314         printk(KERN_EMERG "esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
 315                 regs->esi, regs->edi, regs->ebp, esp);
 316         printk(KERN_EMERG "ds: %04x   es: %04x   fs: %04x  gs: %04x  ss: %04x\n",
 317                regs->xds & 0xffff, regs->xes & 0xffff, regs->xfs & 0xffff, gs, ss);
 318         printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
 319                 TASK_COMM_LEN, current->comm, current->pid,
 320                 current_thread_info(), current, task_thread_info(current));
 321         /*
 322          * When in-kernel, we also print out the stack and code at the
 323          * time of the fault..
 324          */
 325         if (in_kernel) {
 326                 u8 *eip;
 327                 unsigned int code_prologue = code_bytes * 43 / 64;
 328                 unsigned int code_len = code_bytes;
 329                 unsigned char c;
 330
 331                 printk("\n" KERN_EMERG "Stack: ");
 332                 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
 333
 334                 printk(KERN_EMERG "Code: ");
 335
 336                 eip = (u8 *)regs->eip - code_prologue;
 337                 if (eip < (u8 *)PAGE_OFFSET ||
 338                         probe_kernel_address(eip, c)) {
 339                         /* try starting at EIP */
 340                         eip = (u8 *)regs->eip;
 341                         code_len = code_len - code_prologue + 1;
 342                 }
 343                 for (i = 0; i < code_len; i++, eip++) {
 344                         if (eip < (u8 *)PAGE_OFFSET ||
 345                                 probe_kernel_address(eip, c)) {
 346                                 printk(" Bad EIP value.");
 347                                 break;
 348                         }
 349                         if (eip == (u8 *)regs->eip)
 350                                 printk("<%02x> ", c);
 351                         else
 352                                 printk("%02x ", c);
 353                 }
 354         }
 355         printk("\n");
 356 }
 357
 358 int is_valid_bugaddr(unsigned long eip)
 359 {
 360         unsigned short ud2;
 361
 362         if (eip < PAGE_OFFSET)
 363                 return 0;
 364         if (probe_kernel_address((unsigned short *)eip, ud2))
 365                 return 0;
 366
 367         return ud2 == 0x0b0f;
 368 }
 369
 370 /*
 371  * This is gone through when something in the kernel has done something bad and
 372  * is about to be terminated.
 373  */
 374 void die(const char * str, struct pt_regs * regs, long err)
 375 {
 376         static struct {
 377                 spinlock_t lock;
 378                 u32 lock_owner;
 379                 int lock_owner_depth;
 380         } die = {
 381                 .lock =                 __SPIN_LOCK_UNLOCKED(die.lock),
 382                 .lock_owner =           -1,
 383                 .lock_owner_depth =     0
 384         };
 385         static int die_counter;
 386         unsigned long flags;
 387
 388         oops_enter();
 389
 390         if (die.lock_owner != raw_smp_processor_id()) {
 391                 console_verbose();
 392                 spin_lock_irqsave(&die.lock, flags);
 393                 die.lock_owner = smp_processor_id();
 394                 die.lock_owner_depth = 0;
 395                 bust_spinlocks(1);
 396         }
 397         else
 398                 local_save_flags(flags);
 399
 400         if (++die.lock_owner_depth < 3) {
 401                 int nl = 0;
 402                 unsigned long esp;
 403                 unsigned short ss;
 404
 405                 report_bug(regs->eip, regs);
 406
 407                 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
 408 #ifdef CONFIG_PREEMPT
 409                 printk(KERN_EMERG "PREEMPT ");
 410                 nl = 1;
 411 #endif
 412 #ifdef CONFIG_SMP
 413                 if (!nl)
 414                         printk(KERN_EMERG);
 415                 printk("SMP ");
 416                 nl = 1;
 417 #endif
 418 #ifdef CONFIG_DEBUG_PAGEALLOC
 419                 if (!nl)
 420                         printk(KERN_EMERG);
 421                 printk("DEBUG_PAGEALLOC");
 422                 nl = 1;
 423 #endif
 424                 if (nl)
 425                         printk("\n");
 426                 if (notify_die(DIE_OOPS, str, regs, err,
 427                                         current->thread.trap_no, SIGSEGV) !=
 428                                 NOTIFY_STOP) {
 429                         show_registers(regs);
 430                         /* Executive summary in case the oops scrolled away */
 431                         esp = (unsigned long) (&regs->esp);
 432                         savesegment(ss, ss);
 433                         if (user_mode(regs)) {
 434                                 esp = regs->esp;
 435                                 ss = regs->xss & 0xffff;
 436                         }
 437                         printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
 438                         print_symbol("%s", regs->eip);
 439                         printk(" SS:ESP %04x:%08lx\n", ss, esp);
 440                 }
 441                 else
 442                         regs = NULL;
 443         } else
 444                 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
 445
 446         bust_spinlocks(0);
 447         die.lock_owner = -1;
 448         add_taint(TAINT_DIE);
 449         spin_unlock_irqrestore(&die.lock, flags);
 450
 451         if (!regs)
 452                 return;
 453
 454         if (kexec_should_crash(current))
 455                 crash_kexec(regs);
 456
 457         if (in_interrupt())
 458                 panic("Fatal exception in interrupt");
 459
 460         if (panic_on_oops)
 461                 panic("Fatal exception");
 462
 463         oops_exit();
 464         do_exit(SIGSEGV);
 465 }
 466
 467 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 468 {
 469         if (!user_mode_vm(regs))
 470                 die(str, regs, err);
 471 }
 472
 473 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
 474                               struct pt_regs * regs, long error_code,
 475                               siginfo_t *info)
 476 {
 477         struct task_struct *tsk = current;
 478
 479         if (regs->eflags & VM_MASK) {
 480                 if (vm86)
 481                         goto vm86_trap;
 482                 goto trap_signal;
 483         }
 484
 485         if (!user_mode(regs))
 486                 goto kernel_trap;
 487
 488         trap_signal: {
 489                 /*
 490                  * We want error_code and trap_no set for userspace faults and
 491                  * kernelspace faults which result in die(), but not
 492                  * kernelspace faults which are fixed up.  die() gives the
 493                  * process no chance to handle the signal and notice the
 494                  * kernel fault information, so that won't result in polluting
 495                  * the information about previously queued, but not yet
 496                  * delivered, faults.  See also do_general_protection below.
 497                  */
 498                 tsk->thread.error_code = error_code;
 499                 tsk->thread.trap_no = trapnr;
 500
 501                 if (info)
 502                         force_sig_info(signr, info, tsk);
 503                 else
 504                         force_sig(signr, tsk);
 505                 return;
 506         }
 507
 508         kernel_trap: {
 509                 if (!fixup_exception(regs)) {
 510                         tsk->thread.error_code = error_code;
 511                         tsk->thread.trap_no = trapnr;
 512                         die(str, regs, error_code);
 513                 }
 514                 return;
 515         }
 516
 517         vm86_trap: {
 518                 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
 519                 if (ret) goto trap_signal;
 520                 return;
 521         }
 522 }
 523
 524 #define DO_ERROR(trapnr, signr, str, name) \
 525 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 526 { \
 527         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 528                                                 == NOTIFY_STOP) \
 529                 return; \
 530         do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
 531 }
 532
 533 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
 534 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 535 { \
 536         siginfo_t info; \
 537         if (irq) \
 538                 local_irq_enable(); \
 539         info.si_signo = signr; \
 540         info.si_errno = 0; \
 541         info.si_code = sicode; \
 542         info.si_addr = (void __user *)siaddr; \
 543         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 544                                                 == NOTIFY_STOP) \
 545                 return; \
 546         do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
 547 }
 548
 549 #define DO_VM86_ERROR(trapnr, signr, str, name) \
 550 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 551 { \
 552         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 553                                                 == NOTIFY_STOP) \
 554                 return; \
 555         do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
 556 }
 557
 558 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 559 fastcall void do_##name(struct pt_regs * regs, long error_code) \
 560 { \
 561         siginfo_t info; \
 562         info.si_signo = signr; \
 563         info.si_errno = 0; \
 564         info.si_code = sicode; \
 565         info.si_addr = (void __user *)siaddr; \
 566         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 567                                                 == NOTIFY_STOP) \
 568                 return; \
 569         do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
 570 }
 571
 572 DO_VM86_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->eip)
 573 #ifndef CONFIG_KPROBES
 574 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
 575 #endif
 576 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
 577 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
 578 DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip, 0)
 579 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 580 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 581 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 582 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 583 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
 584 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
 585
 586 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
 587                                               long error_code)
 588 {
 589         int cpu = get_cpu();
 590         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 591         struct thread_struct *thread = &current->thread;
 592
 593         /*
 594          * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 595          * invalid offset set (the LAZY one) and the faulting thread has
 596          * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
 597          * and we set the offset field correctly. Then we let the CPU to
 598          * restart the faulting instruction.
 599          */
 600         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 601             thread->io_bitmap_ptr) {
 602                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 603                        thread->io_bitmap_max);
 604                 /*
 605                  * If the previously set map was extending to higher ports
 606                  * than the current one, pad extra space with 0xff (no access).
 607                  */
 608                 if (thread->io_bitmap_max < tss->io_bitmap_max)
 609                         memset((char *) tss->io_bitmap +
 610                                 thread->io_bitmap_max, 0xff,
 611                                 tss->io_bitmap_max - thread->io_bitmap_max);
 612                 tss->io_bitmap_max = thread->io_bitmap_max;
 613                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 614                 tss->io_bitmap_owner = thread;
 615                 put_cpu();
 616                 return;
 617         }
 618         put_cpu();
 619
 620         if (regs->eflags & VM_MASK)
 621                 goto gp_in_vm86;
 622
 623         if (!user_mode(regs))
 624                 goto gp_in_kernel;
 625
 626         current->thread.error_code = error_code;
 627         current->thread.trap_no = 13;
 628         if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
 629             printk_ratelimit())
 630                 printk(KERN_INFO
 631                     "%s[%d] general protection eip:%lx esp:%lx error:%lx\n",
 632                     current->comm, current->pid,
 633                     regs->eip, regs->esp, error_code);
 634
 635         force_sig(SIGSEGV, current);
 636         return;
 637
 638 gp_in_vm86:
 639         local_irq_enable();
 640         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 641         return;
 642
 643 gp_in_kernel:
 644         if (!fixup_exception(regs)) {
 645                 current->thread.error_code = error_code;
 646                 current->thread.trap_no = 13;
 647                 if (notify_die(DIE_GPF, "general protection fault", regs,
 648                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 649                         return;
 650                 die("general protection fault", regs, error_code);
 651         }
 652 }
 653
 654 static __kprobes void
 655 mem_parity_error(unsigned char reason, struct pt_regs * regs)
 656 {
 657         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
 658                 "CPU %d.\n", reason, smp_processor_id());
 659         printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 660
 661 #if defined(CONFIG_EDAC)
 662         if(edac_handler_set()) {
 663                 edac_atomic_assert_error();
 664                 return;
 665         }
 666 #endif
 667
 668         if (panic_on_unrecovered_nmi)
 669                 panic("NMI: Not continuing");
 670
 671         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 672
 673         /* Clear and disable the memory parity error line. */
 674         clear_mem_error(reason);
 675 }
 676
 677 static __kprobes void
 678 io_check_error(unsigned char reason, struct pt_regs * regs)
 679 {
 680         unsigned long i;
 681
 682         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 683         show_registers(regs);
 684
 685         /* Re-enable the IOCK line, wait for a few seconds */
 686         reason = (reason & 0xf) | 8;
 687         outb(reason, 0x61);
 688         i = 2000;
 689         while (--i) udelay(1000);
 690         reason &= ~8;
 691         outb(reason, 0x61);
 692 }
 693
 694 static __kprobes void
 695 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 696 {
 697 #ifdef CONFIG_MCA
 698         /* Might actually be able to figure out what the guilty party
 699         * is. */
 700         if( MCA_bus ) {
 701                 mca_handle_nmi();
 702                 return;
 703         }
 704 #endif
 705         printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
 706                 "CPU %d.\n", reason, smp_processor_id());
 707         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 708         if (panic_on_unrecovered_nmi)
 709                 panic("NMI: Not continuing");
 710
 711         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 712 }
 713
 714 static DEFINE_SPINLOCK(nmi_print_lock);
 715
 716 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
 717 {
 718         if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
 719             NOTIFY_STOP)
 720                 return;
 721
 722         spin_lock(&nmi_print_lock);
 723         /*
 724         * We are in trouble anyway, lets at least try
 725         * to get a message out.
 726         */
 727         bust_spinlocks(1);
 728         printk(KERN_EMERG "%s", msg);
 729         printk(" on CPU%d, eip %08lx, registers:\n",
 730                 smp_processor_id(), regs->eip);
 731         show_registers(regs);
 732         console_silent();
 733         spin_unlock(&nmi_print_lock);
 734         bust_spinlocks(0);
 735
 736         /* If we are in kernel we are probably nested up pretty bad
 737          * and might aswell get out now while we still can.
 738         */
 739         if (!user_mode_vm(regs)) {
 740                 current->thread.trap_no = 2;
 741                 crash_kexec(regs);
 742         }
 743
 744         do_exit(SIGSEGV);
 745 }
 746
 747 static __kprobes void default_do_nmi(struct pt_regs * regs)
 748 {
 749         unsigned char reason = 0;
 750
 751         /* Only the BSP gets external NMIs from the system.  */
 752         if (!smp_processor_id())
 753                 reason = get_nmi_reason();
 754
 755         if (!(reason & 0xc0)) {
 756                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 757                                                         == NOTIFY_STOP)
 758                         return;
 759 #ifdef CONFIG_X86_LOCAL_APIC
 760                 /*
 761                  * Ok, so this is none of the documented NMI sources,
 762                  * so it must be the NMI watchdog.
 763                  */
 764                 if (nmi_watchdog_tick(regs, reason))
 765                         return;
 766                 if (!do_nmi_callback(regs, smp_processor_id()))
 767 #endif
 768                         unknown_nmi_error(reason, regs);
 769
 770                 return;
 771         }
 772         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 773                 return;
 774         if (reason & 0x80)
 775                 mem_parity_error(reason, regs);
 776         if (reason & 0x40)
 777                 io_check_error(reason, regs);
 778         /*
 779          * Reassert NMI in case it became active meanwhile
 780          * as it's edge-triggered.
 781          */
 782         reassert_nmi();
 783 }
 784
 785 static int ignore_nmis;
 786
 787 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
 788 {
 789         int cpu;
 790
 791         nmi_enter();
 792
 793         cpu = smp_processor_id();
 794
 795         ++nmi_count(cpu);
 796
 797         if (!ignore_nmis)
 798                 default_do_nmi(regs);
 799
 800         nmi_exit();
 801 }
 802
 803 void stop_nmi(void)
 804 {
 805         acpi_nmi_disable();
 806         ignore_nmis++;
 807 }
 808
 809 void restart_nmi(void)
 810 {
 811         ignore_nmis--;
 812         acpi_nmi_enable();
 813 }
 814
 815 #ifdef CONFIG_KPROBES
 816 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
 817 {
 818         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 819                         == NOTIFY_STOP)
 820                 return;
 821         /* This is an interrupt gate, because kprobes wants interrupts
 822         disabled.  Normal trap handlers don't. */
 823         restore_interrupts(regs);
 824         do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
 825 }
 826 #endif
 827
 828 /*
 829  * Our handling of the processor debug registers is non-trivial.
 830  * We do not clear them on entry and exit from the kernel. Therefore
 831  * it is possible to get a watchpoint trap here from inside the kernel.
 832  * However, the code in ./ptrace.c has ensured that the user can
 833  * only set watchpoints on userspace addresses. Therefore the in-kernel
 834  * watchpoint trap can only occur in code which is reading/writing
 835  * from user space. Such code must not hold kernel locks (since it
 836  * can equally take a page fault), therefore it is safe to call
 837  * force_sig_info even though that claims and releases locks.
 838  *
 839  * Code in ./signal.c ensures that the debug control register
 840  * is restored before we deliver any signal, and therefore that
 841  * user code runs with the correct debug control register even though
 842  * we clear it here.
 843  *
 844  * Being careful here means that we don't have to be as careful in a
 845  * lot of more complicated places (task switching can be a bit lazy
 846  * about restoring all the debug state, and ptrace doesn't have to
 847  * find every occurrence of the TF bit that could be saved away even
 848  * by user code)
 849  */
 850 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
 851 {
 852         unsigned int condition;
 853         struct task_struct *tsk = current;
 854
 855         get_debugreg(condition, 6);
 856
 857         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 858                                         SIGTRAP) == NOTIFY_STOP)
 859                 return;
 860         /* It's safe to allow irq's after DR6 has been saved */
 861         if (regs->eflags & X86_EFLAGS_IF)
 862                 local_irq_enable();
 863
 864         /* Mask out spurious debug traps due to lazy DR7 setting */
 865         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 866                 if (!tsk->thread.debugreg[7])
 867                         goto clear_dr7;
 868         }
 869
 870         if (regs->eflags & VM_MASK)
 871                 goto debug_vm86;
 872
 873         /* Save debug status register where ptrace can see it */
 874         tsk->thread.debugreg[6] = condition;
 875
 876         /*
 877          * Single-stepping through TF: make sure we ignore any events in
 878          * kernel space (but re-enable TF when returning to user mode).
 879          */
 880         if (condition & DR_STEP) {
 881                 /*
 882                  * We already checked v86 mode above, so we can
 883                  * check for kernel mode by just checking the CPL
 884                  * of CS.
 885                  */
 886                 if (!user_mode(regs))
 887                         goto clear_TF_reenable;
 888         }
 889
 890         /* Ok, finally something we can handle */
 891         send_sigtrap(tsk, regs, error_code);
 892
 893         /* Disable additional traps. They'll be re-enabled when
 894          * the signal is delivered.
 895          */
 896 clear_dr7:
 897         set_debugreg(0, 7);
 898         return;
 899
 900 debug_vm86:
 901         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 902         return;
 903
 904 clear_TF_reenable:
 905         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 906         regs->eflags &= ~TF_MASK;
 907         return;
 908 }
 909
 910 /*
 911  * Note that we play around with the 'TS' bit in an attempt to get
 912  * the correct behaviour even in the presence of the asynchronous
 913  * IRQ13 behaviour
 914  */
 915 void math_error(void __user *eip)
 916 {
 917         struct task_struct * task;
 918         siginfo_t info;
 919         unsigned short cwd, swd;
 920
 921         /*
 922          * Save the info for the exception handler and clear the error.
 923          */
 924         task = current;
 925         save_init_fpu(task);
 926         task->thread.trap_no = 16;
 927         task->thread.error_code = 0;
 928         info.si_signo = SIGFPE;
 929         info.si_errno = 0;
 930         info.si_code = __SI_FAULT;
 931         info.si_addr = eip;
 932         /*
 933          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 934          * status.  0x3f is the exception bits in these regs, 0x200 is the
 935          * C1 reg you need in case of a stack fault, 0x040 is the stack
 936          * fault bit.  We should only be taking one exception at a time,
 937          * so if this combination doesn't produce any single exception,
 938          * then we have a bad program that isn't syncronizing its FPU usage
 939          * and it will suffer the consequences since we won't be able to
 940          * fully reproduce the context of the exception
 941          */
 942         cwd = get_fpu_cwd(task);
 943         swd = get_fpu_swd(task);
 944         switch (swd & ~cwd & 0x3f) {
 945                 case 0x000: /* No unmasked exception */
 946                         return;
 947                 default:    /* Multiple exceptions */
 948                         break;
 949                 case 0x001: /* Invalid Op */
 950                         /*
 951                          * swd & 0x240 == 0x040: Stack Underflow
 952                          * swd & 0x240 == 0x240: Stack Overflow
 953                          * User must clear the SF bit (0x40) if set
 954                          */
 955                         info.si_code = FPE_FLTINV;
 956                         break;
 957                 case 0x002: /* Denormalize */
 958                 case 0x010: /* Underflow */
 959                         info.si_code = FPE_FLTUND;
 960                         break;
 961                 case 0x004: /* Zero Divide */
 962                         info.si_code = FPE_FLTDIV;
 963                         break;
 964                 case 0x008: /* Overflow */
 965                         info.si_code = FPE_FLTOVF;
 966                         break;
 967                 case 0x020: /* Precision */
 968                         info.si_code = FPE_FLTRES;
 969                         break;
 970         }
 971         force_sig_info(SIGFPE, &info, task);
 972 }
 973
 974 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
 975 {
 976         ignore_fpu_irq = 1;
 977         math_error((void __user *)regs->eip);
 978 }
 979
 980 static void simd_math_error(void __user *eip)
 981 {
 982         struct task_struct * task;
 983         siginfo_t info;
 984         unsigned short mxcsr;
 985
 986         /*
 987          * Save the info for the exception handler and clear the error.
 988          */
 989         task = current;
 990         save_init_fpu(task);
 991         task->thread.trap_no = 19;
 992         task->thread.error_code = 0;
 993         info.si_signo = SIGFPE;
 994         info.si_errno = 0;
 995         info.si_code = __SI_FAULT;
 996         info.si_addr = eip;
 997         /*
 998          * The SIMD FPU exceptions are handled a little differently, as there
 999          * is only a single status/control register.  Thus, to determine which
1000          * unmasked exception was caught we must mask the exception mask bits
1001          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1002          */
1003         mxcsr = get_fpu_mxcsr(task);
1004         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1005                 case 0x000:
1006                 default:
1007                         break;
1008                 case 0x001: /* Invalid Op */
1009                         info.si_code = FPE_FLTINV;
1010                         break;
1011                 case 0x002: /* Denormalize */
1012                 case 0x010: /* Underflow */
1013                         info.si_code = FPE_FLTUND;
1014                         break;
1015                 case 0x004: /* Zero Divide */
1016                         info.si_code = FPE_FLTDIV;
1017                         break;
1018                 case 0x008: /* Overflow */
1019                         info.si_code = FPE_FLTOVF;
1020                         break;
1021                 case 0x020: /* Precision */
1022                         info.si_code = FPE_FLTRES;
1023                         break;
1024         }
1025         force_sig_info(SIGFPE, &info, task);
1026 }
1027
1028 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1029                                           long error_code)
1030 {
1031         if (cpu_has_xmm) {
1032                 /* Handle SIMD FPU exceptions on PIII+ processors. */
1033                 ignore_fpu_irq = 1;
1034                 simd_math_error((void __user *)regs->eip);
1035         } else {
1036                 /*
1037                  * Handle strange cache flush from user space exception
1038                  * in all other cases.  This is undocumented behaviour.
1039                  */
1040                 if (regs->eflags & VM_MASK) {
1041                         handle_vm86_fault((struct kernel_vm86_regs *)regs,
1042                                           error_code);
1043                         return;
1044                 }
1045                 current->thread.trap_no = 19;
1046                 current->thread.error_code = error_code;
1047                 die_if_kernel("cache flush denied", regs, error_code);
1048                 force_sig(SIGSEGV, current);
1049         }
1050 }
1051
1052 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1053                                           long error_code)
1054 {
1055 #if 0
1056         /* No need to warn about this any longer. */
1057         printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1058 #endif
1059 }
1060
1061 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1062                                           unsigned long kesp)
1063 {
1064         struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1065         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1066         unsigned long new_kesp = kesp - base;
1067         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1068         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1069         /* Set up base for espfix segment */
1070         desc &= 0x00f0ff0000000000ULL;
1071         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1072                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1073                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1074                 (lim_pages & 0xffff);
1075         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1076         return new_kesp;
1077 }
1078
1079 /*
1080  *  'math_state_restore()' saves the current math information in the
1081  * old math state array, and gets the new ones from the current task
1082  *
1083  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1084  * Don't touch unless you *really* know how it works.
1085  *
1086  * Must be called with kernel preemption disabled (in this case,
1087  * local interrupts are disabled at the call-site in entry.S).
1088  */
1089 asmlinkage void math_state_restore(void)
1090 {
1091         struct thread_info *thread = current_thread_info();
1092         struct task_struct *tsk = thread->task;
1093
1094         clts();         /* Allow maths ops (or we recurse) */
1095         if (!tsk_used_math(tsk))
1096                 init_fpu(tsk);
1097         restore_fpu(tsk);
1098         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
1099         tsk->fpu_counter++;
1100 }
1101 EXPORT_SYMBOL_GPL(math_state_restore);
1102
1103 #ifndef CONFIG_MATH_EMULATION
1104
1105 asmlinkage void math_emulate(long arg)
1106 {
1107         printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1108         printk(KERN_EMERG "killing %s.\n",current->comm);
1109         force_sig(SIGFPE,current);
1110         schedule();
1111 }
1112
1113 #endif /* CONFIG_MATH_EMULATION */
1114
1115 #ifdef CONFIG_X86_F00F_BUG
1116 void __init trap_init_f00f_bug(void)
1117 {
1118         __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1119
1120         /*
1121          * Update the IDT descriptor and reload the IDT so that
1122          * it uses the read-only mapped virtual address.
1123          */
1124         idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1125         load_idt(&idt_descr);
1126 }
1127 #endif
1128
1129 /*
1130  * This needs to use 'idt_table' rather than 'idt', and
1131  * thus use the _nonmapped_ version of the IDT, as the
1132  * Pentium F0 0F bugfix can have resulted in the mapped
1133  * IDT being write-protected.
1134  */
1135 void set_intr_gate(unsigned int n, void *addr)
1136 {
1137         _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1138 }
1139
1140 /*
1141  * This routine sets up an interrupt gate at directory privilege level 3.
1142  */
1143 static inline void set_system_intr_gate(unsigned int n, void *addr)
1144 {
1145         _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1146 }
1147
1148 static void __init set_trap_gate(unsigned int n, void *addr)
1149 {
1150         _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1151 }
1152
1153 static void __init set_system_gate(unsigned int n, void *addr)
1154 {
1155         _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1156 }
1157
1158 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1159 {
1160         _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1161 }
1162
1163
1164 void __init trap_init(void)
1165 {
1166 #ifdef CONFIG_EISA
1167         void __iomem *p = ioremap(0x0FFFD9, 4);
1168         if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1169                 EISA_bus = 1;
1170         }
1171         iounmap(p);
1172 #endif
1173
1174 #ifdef CONFIG_X86_LOCAL_APIC
1175         init_apic_mappings();
1176 #endif
1177
1178         set_trap_gate(0,&divide_error);
1179         set_intr_gate(1,&debug);
1180         set_intr_gate(2,&nmi);
1181         set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1182         set_system_gate(4,&overflow);
1183         set_trap_gate(5,&bounds);
1184         set_trap_gate(6,&invalid_op);
1185         set_trap_gate(7,&device_not_available);
1186         set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1187         set_trap_gate(9,&coprocessor_segment_overrun);
1188         set_trap_gate(10,&invalid_TSS);
1189         set_trap_gate(11,&segment_not_present);
1190         set_trap_gate(12,&stack_segment);
1191         set_trap_gate(13,&general_protection);
1192         set_intr_gate(14,&page_fault);
1193         set_trap_gate(15,&spurious_interrupt_bug);
1194         set_trap_gate(16,&coprocessor_error);
1195         set_trap_gate(17,&alignment_check);
1196 #ifdef CONFIG_X86_MCE
1197         set_trap_gate(18,&machine_check);
1198 #endif
1199         set_trap_gate(19,&simd_coprocessor_error);
1200
1201         if (cpu_has_fxsr) {
1202                 /*
1203                  * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1204                  * Generates a compile-time "error: zero width for bit-field" if
1205                  * the alignment is wrong.
1206                  */
1207                 struct fxsrAlignAssert {
1208                         int _:!(offsetof(struct task_struct,
1209                                         thread.i387.fxsave) & 15);
1210                 };
1211
1212                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1213                 set_in_cr4(X86_CR4_OSFXSR);
1214                 printk("done.\n");
1215         }
1216         if (cpu_has_xmm) {
1217                 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1218                                 "support... ");
1219                 set_in_cr4(X86_CR4_OSXMMEXCPT);
1220                 printk("done.\n");
1221         }
1222
1223         set_system_gate(SYSCALL_VECTOR,&system_call);
1224
1225         /*
1226          * Should be a barrier for any external CPU state.
1227          */
1228         cpu_init();
1229
1230         trap_init_hook();
1231 }
1232
1233 static int __init kstack_setup(char *s)
1234 {
1235         kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1236         return 1;
1237 }
1238 __setup("kstack=", kstack_setup);
1239
1240 static int __init code_bytes_setup(char *s)
1241 {
1242         code_bytes = simple_strtoul(s, NULL, 0);
1243         if (code_bytes > 8192)
1244                 code_bytes = 8192;
1245
1246         return 1;
1247 }
1248 __setup("code_bytes=", code_bytes_setup);