linux/arch/x86/kernel/traps_32.c
<<
>>
Prefs
   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
  64int panic_on_unrecovered_nmi;
  65
  66DECLARE_BITMAP(used_vectors, NR_VECTORS);
  67EXPORT_SYMBOL_GPL(used_vectors);
  68
  69asmlinkage int system_call(void);
  70
  71/* Do we ignore FPU interrupts ? */
  72char 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 */
  79gate_desc idt_table[256]
  80        __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  81
  82asmlinkage void divide_error(void);
  83asmlinkage void debug(void);
  84asmlinkage void nmi(void);
  85asmlinkage void int3(void);
  86asmlinkage void overflow(void);
  87asmlinkage void bounds(void);
  88asmlinkage void invalid_op(void);
  89asmlinkage void device_not_available(void);
  90asmlinkage void coprocessor_segment_overrun(void);
  91asmlinkage void invalid_TSS(void);
  92asmlinkage void segment_not_present(void);
  93asmlinkage void stack_segment(void);
  94asmlinkage void general_protection(void);
  95asmlinkage void page_fault(void);
  96asmlinkage void coprocessor_error(void);
  97asmlinkage void simd_coprocessor_error(void);
  98asmlinkage void alignment_check(void);
  99asmlinkage void spurious_interrupt_bug(void);
 100asmlinkage void machine_check(void);
 101
 102int kstack_depth_to_print = 24;
 103static unsigned int code_bytes = 64;
 104
 105void printk_address(unsigned long address, int reliable)
 106{
 107#ifdef CONFIG_KALLSYMS
 108        unsigned long offset = 0, symsize;
 109        const char *symname;
 110        char *modname;
 111        char *delim = ":";
 112        char namebuf[128];
 113        char reliab[4] = "";
 114
 115        symname = kallsyms_lookup(address, &symsize, &offset,
 116                                        &modname, namebuf);
 117        if (!symname) {
 118                printk(" [<%08lx>]\n", address);
 119                return;
 120        }
 121        if (!reliable)
 122                strcpy(reliab, "? ");
 123
 124        if (!modname)
 125                modname = delim = "";
 126        printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
 127                address, reliab, delim, modname, delim, symname, offset, symsize);
 128#else
 129        printk(" [<%08lx>]\n", address);
 130#endif
 131}
 132
 133static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
 134{
 135        return  p > (void *)tinfo &&
 136                p <= (void *)tinfo + THREAD_SIZE - size;
 137}
 138
 139/* The form of the top of the frame on the stack */
 140struct stack_frame {
 141        struct stack_frame *next_frame;
 142        unsigned long return_address;
 143};
 144
 145static inline unsigned long print_context_stack(struct thread_info *tinfo,
 146                                unsigned long *stack, unsigned long bp,
 147                                const struct stacktrace_ops *ops, void *data)
 148{
 149        struct stack_frame *frame = (struct stack_frame *)bp;
 150
 151        while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
 152                unsigned long addr;
 153
 154                addr = *stack;
 155                if (__kernel_text_address(addr)) {
 156                        if ((unsigned long) stack == bp + 4) {
 157                                ops->address(data, addr, 1);
 158                                frame = frame->next_frame;
 159                                bp = (unsigned long) frame;
 160                        } else {
 161                                ops->address(data, addr, bp == 0);
 162                        }
 163                }
 164                stack++;
 165        }
 166        return bp;
 167}
 168
 169#define MSG(msg) ops->warning(data, msg)
 170
 171void dump_trace(struct task_struct *task, struct pt_regs *regs,
 172                unsigned long *stack, unsigned long bp,
 173                const struct stacktrace_ops *ops, void *data)
 174{
 175        if (!task)
 176                task = current;
 177
 178        if (!stack) {
 179                unsigned long dummy;
 180                stack = &dummy;
 181                if (task != current)
 182                        stack = (unsigned long *)task->thread.sp;
 183        }
 184
 185#ifdef CONFIG_FRAME_POINTER
 186        if (!bp) {
 187                if (task == current) {
 188                        /* Grab bp right from our regs */
 189                        asm ("movl %%ebp, %0" : "=r" (bp) : );
 190                } else {
 191                        /* bp is the last reg pushed by switch_to */
 192                        bp = *(unsigned long *) task->thread.sp;
 193                }
 194        }
 195#endif
 196
 197        while (1) {
 198                struct thread_info *context;
 199                context = (struct thread_info *)
 200                        ((unsigned long)stack & (~(THREAD_SIZE - 1)));
 201                bp = print_context_stack(context, stack, bp, ops, data);
 202                /* Should be after the line below, but somewhere
 203                   in early boot context comes out corrupted and we
 204                   can't reference it -AK */
 205                if (ops->stack(data, "IRQ") < 0)
 206                        break;
 207                stack = (unsigned long*)context->previous_esp;
 208                if (!stack)
 209                        break;
 210                touch_nmi_watchdog();
 211        }
 212}
 213EXPORT_SYMBOL(dump_trace);
 214
 215static void
 216print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
 217{
 218        printk(data);
 219        print_symbol(msg, symbol);
 220        printk("\n");
 221}
 222
 223static void print_trace_warning(void *data, char *msg)
 224{
 225        printk("%s%s\n", (char *)data, msg);
 226}
 227
 228static int print_trace_stack(void *data, char *name)
 229{
 230        return 0;
 231}
 232
 233/*
 234 * Print one address/symbol entries per line.
 235 */
 236static void print_trace_address(void *data, unsigned long addr, int reliable)
 237{
 238        printk("%s [<%08lx>] ", (char *)data, addr);
 239        if (!reliable)
 240                printk("? ");
 241        print_symbol("%s\n", addr);
 242        touch_nmi_watchdog();
 243}
 244
 245static const struct stacktrace_ops print_trace_ops = {
 246        .warning = print_trace_warning,
 247        .warning_symbol = print_trace_warning_symbol,
 248        .stack = print_trace_stack,
 249        .address = print_trace_address,
 250};
 251
 252static void
 253show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 254                unsigned long *stack, unsigned long bp, char *log_lvl)
 255{
 256        dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
 257        printk("%s =======================\n", log_lvl);
 258}
 259
 260void show_trace(struct task_struct *task, struct pt_regs *regs,
 261                unsigned long *stack, unsigned long bp)
 262{
 263        show_trace_log_lvl(task, regs, stack, bp, "");
 264}
 265
 266static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
 267                       unsigned long *sp, unsigned long bp, char *log_lvl)
 268{
 269        unsigned long *stack;
 270        int i;
 271
 272        if (sp == NULL) {
 273                if (task)
 274                        sp = (unsigned long*)task->thread.sp;
 275                else
 276                        sp = (unsigned long *)&sp;
 277        }
 278
 279        stack = sp;
 280        for(i = 0; i < kstack_depth_to_print; i++) {
 281                if (kstack_end(stack))
 282                        break;
 283                if (i && ((i % 8) == 0))
 284                        printk("\n%s       ", log_lvl);
 285                printk("%08lx ", *stack++);
 286        }
 287        printk("\n%sCall Trace:\n", log_lvl);
 288        show_trace_log_lvl(task, regs, sp, bp, log_lvl);
 289}
 290
 291void show_stack(struct task_struct *task, unsigned long *sp)
 292{
 293        printk("       ");
 294        show_stack_log_lvl(task, NULL, sp, 0, "");
 295}
 296
 297/*
 298 * The architecture-independent dump_stack generator
 299 */
 300void dump_stack(void)
 301{
 302        unsigned long stack;
 303        unsigned long bp = 0;
 304
 305#ifdef CONFIG_FRAME_POINTER
 306        if (!bp)
 307                asm("movl %%ebp, %0" : "=r" (bp):);
 308#endif
 309
 310        printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 311                current->pid, current->comm, print_tainted(),
 312                init_utsname()->release,
 313                (int)strcspn(init_utsname()->version, " "),
 314                init_utsname()->version);
 315        show_trace(current, NULL, &stack, bp);
 316}
 317
 318EXPORT_SYMBOL(dump_stack);
 319
 320void show_registers(struct pt_regs *regs)
 321{
 322        int i;
 323
 324        print_modules();
 325        __show_registers(regs, 0);
 326        printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
 327                TASK_COMM_LEN, current->comm, task_pid_nr(current),
 328                current_thread_info(), current, task_thread_info(current));
 329        /*
 330         * When in-kernel, we also print out the stack and code at the
 331         * time of the fault..
 332         */
 333        if (!user_mode_vm(regs)) {
 334                u8 *ip;
 335                unsigned int code_prologue = code_bytes * 43 / 64;
 336                unsigned int code_len = code_bytes;
 337                unsigned char c;
 338
 339                printk("\n" KERN_EMERG "Stack: ");
 340                show_stack_log_lvl(NULL, regs, &regs->sp, 0, KERN_EMERG);
 341
 342                printk(KERN_EMERG "Code: ");
 343
 344                ip = (u8 *)regs->ip - code_prologue;
 345                if (ip < (u8 *)PAGE_OFFSET ||
 346                        probe_kernel_address(ip, c)) {
 347                        /* try starting at EIP */
 348                        ip = (u8 *)regs->ip;
 349                        code_len = code_len - code_prologue + 1;
 350                }
 351                for (i = 0; i < code_len; i++, ip++) {
 352                        if (ip < (u8 *)PAGE_OFFSET ||
 353                                probe_kernel_address(ip, c)) {
 354                                printk(" Bad EIP value.");
 355                                break;
 356                        }
 357                        if (ip == (u8 *)regs->ip)
 358                                printk("<%02x> ", c);
 359                        else
 360                                printk("%02x ", c);
 361                }
 362        }
 363        printk("\n");
 364}       
 365
 366int is_valid_bugaddr(unsigned long ip)
 367{
 368        unsigned short ud2;
 369
 370        if (ip < PAGE_OFFSET)
 371                return 0;
 372        if (probe_kernel_address((unsigned short *)ip, ud2))
 373                return 0;
 374
 375        return ud2 == 0x0b0f;
 376}
 377
 378static int die_counter;
 379
 380int __kprobes __die(const char * str, struct pt_regs * regs, long err)
 381{
 382        unsigned long sp;
 383        unsigned short ss;
 384
 385        printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
 386#ifdef CONFIG_PREEMPT
 387        printk("PREEMPT ");
 388#endif
 389#ifdef CONFIG_SMP
 390        printk("SMP ");
 391#endif
 392#ifdef CONFIG_DEBUG_PAGEALLOC
 393        printk("DEBUG_PAGEALLOC");
 394#endif
 395        printk("\n");
 396
 397        if (notify_die(DIE_OOPS, str, regs, err,
 398                                current->thread.trap_no, SIGSEGV) !=
 399                        NOTIFY_STOP) {
 400                show_registers(regs);
 401                /* Executive summary in case the oops scrolled away */
 402                sp = (unsigned long) (&regs->sp);
 403                savesegment(ss, ss);
 404                if (user_mode(regs)) {
 405                        sp = regs->sp;
 406                        ss = regs->ss & 0xffff;
 407                }
 408                printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
 409                print_symbol("%s", regs->ip);
 410                printk(" SS:ESP %04x:%08lx\n", ss, sp);
 411                return 0;
 412        } else {
 413                return 1;
 414        }
 415}
 416
 417/*
 418 * This is gone through when something in the kernel has done something bad and
 419 * is about to be terminated.
 420 */
 421void die(const char * str, struct pt_regs * regs, long err)
 422{
 423        static struct {
 424                raw_spinlock_t lock;
 425                u32 lock_owner;
 426                int lock_owner_depth;
 427        } die = {
 428                .lock =                 __RAW_SPIN_LOCK_UNLOCKED,
 429                .lock_owner =           -1,
 430                .lock_owner_depth =     0
 431        };
 432        unsigned long flags;
 433
 434        oops_enter();
 435
 436        if (die.lock_owner != raw_smp_processor_id()) {
 437                console_verbose();
 438                raw_local_irq_save(flags);
 439                __raw_spin_lock(&die.lock);
 440                die.lock_owner = smp_processor_id();
 441                die.lock_owner_depth = 0;
 442                bust_spinlocks(1);
 443        } else
 444                raw_local_irq_save(flags);
 445
 446        if (++die.lock_owner_depth < 3) {
 447                report_bug(regs->ip, regs);
 448
 449                if (__die(str, regs, err))
 450                        regs = NULL;
 451        } else {
 452                printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
 453        }
 454
 455        bust_spinlocks(0);
 456        die.lock_owner = -1;
 457        add_taint(TAINT_DIE);
 458        __raw_spin_unlock(&die.lock);
 459        raw_local_irq_restore(flags);
 460
 461        if (!regs)
 462                return;
 463
 464        if (kexec_should_crash(current))
 465                crash_kexec(regs);
 466
 467        if (in_interrupt())
 468                panic("Fatal exception in interrupt");
 469
 470        if (panic_on_oops)
 471                panic("Fatal exception");
 472
 473        oops_exit();
 474        do_exit(SIGSEGV);
 475}
 476
 477static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 478{
 479        if (!user_mode_vm(regs))
 480                die(str, regs, err);
 481}
 482
 483static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
 484                              struct pt_regs * regs, long error_code,
 485                              siginfo_t *info)
 486{
 487        struct task_struct *tsk = current;
 488
 489        if (regs->flags & VM_MASK) {
 490                if (vm86)
 491                        goto vm86_trap;
 492                goto trap_signal;
 493        }
 494
 495        if (!user_mode(regs))
 496                goto kernel_trap;
 497
 498        trap_signal: {
 499                /*
 500                 * We want error_code and trap_no set for userspace faults and
 501                 * kernelspace faults which result in die(), but not
 502                 * kernelspace faults which are fixed up.  die() gives the
 503                 * process no chance to handle the signal and notice the
 504                 * kernel fault information, so that won't result in polluting
 505                 * the information about previously queued, but not yet
 506                 * delivered, faults.  See also do_general_protection below.
 507                 */
 508                tsk->thread.error_code = error_code;
 509                tsk->thread.trap_no = trapnr;
 510
 511                if (info)
 512                        force_sig_info(signr, info, tsk);
 513                else
 514                        force_sig(signr, tsk);
 515                return;
 516        }
 517
 518        kernel_trap: {
 519                if (!fixup_exception(regs)) {
 520                        tsk->thread.error_code = error_code;
 521                        tsk->thread.trap_no = trapnr;
 522                        die(str, regs, error_code);
 523                }
 524                return;
 525        }
 526
 527        vm86_trap: {
 528                int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
 529                if (ret) goto trap_signal;
 530                return;
 531        }
 532}
 533
 534#define DO_ERROR(trapnr, signr, str, name) \
 535void do_##name(struct pt_regs * regs, long error_code) \
 536{ \
 537        if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 538                                                == NOTIFY_STOP) \
 539                return; \
 540        do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
 541}
 542
 543#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
 544void do_##name(struct pt_regs * regs, long error_code) \
 545{ \
 546        siginfo_t info; \
 547        if (irq) \
 548                local_irq_enable(); \
 549        info.si_signo = signr; \
 550        info.si_errno = 0; \
 551        info.si_code = sicode; \
 552        info.si_addr = (void __user *)siaddr; \
 553        if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 554                                                == NOTIFY_STOP) \
 555                return; \
 556        do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
 557}
 558
 559#define DO_VM86_ERROR(trapnr, signr, str, name) \
 560void do_##name(struct pt_regs * regs, long error_code) \
 561{ \
 562        if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 563                                                == NOTIFY_STOP) \
 564                return; \
 565        do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
 566}
 567
 568#define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 569void do_##name(struct pt_regs * regs, long error_code) \
 570{ \
 571        siginfo_t info; \
 572        info.si_signo = signr; \
 573        info.si_errno = 0; \
 574        info.si_code = sicode; \
 575        info.si_addr = (void __user *)siaddr; \
 576        trace_hardirqs_fixup(); \
 577        if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 578                                                == NOTIFY_STOP) \
 579                return; \
 580        do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
 581}
 582
 583DO_VM86_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->ip)
 584#ifndef CONFIG_KPROBES
 585DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
 586#endif
 587DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
 588DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
 589DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
 590DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 591DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 592DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 593DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 594DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
 595DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
 596
 597void __kprobes do_general_protection(struct pt_regs * regs,
 598                                              long error_code)
 599{
 600        int cpu = get_cpu();
 601        struct tss_struct *tss = &per_cpu(init_tss, cpu);
 602        struct thread_struct *thread = &current->thread;
 603
 604        /*
 605         * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 606         * invalid offset set (the LAZY one) and the faulting thread has
 607         * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
 608         * and we set the offset field correctly. Then we let the CPU to
 609         * restart the faulting instruction.
 610         */
 611        if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 612            thread->io_bitmap_ptr) {
 613                memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 614                       thread->io_bitmap_max);
 615                /*
 616                 * If the previously set map was extending to higher ports
 617                 * than the current one, pad extra space with 0xff (no access).
 618                 */
 619                if (thread->io_bitmap_max < tss->io_bitmap_max)
 620                        memset((char *) tss->io_bitmap +
 621                                thread->io_bitmap_max, 0xff,
 622                                tss->io_bitmap_max - thread->io_bitmap_max);
 623                tss->io_bitmap_max = thread->io_bitmap_max;
 624                tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 625                tss->io_bitmap_owner = thread;
 626                put_cpu();
 627                return;
 628        }
 629        put_cpu();
 630
 631        if (regs->flags & VM_MASK)
 632                goto gp_in_vm86;
 633
 634        if (!user_mode(regs))
 635                goto gp_in_kernel;
 636
 637        current->thread.error_code = error_code;
 638        current->thread.trap_no = 13;
 639        if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
 640            printk_ratelimit()) {
 641                printk(KERN_INFO
 642                    "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 643                    current->comm, task_pid_nr(current),
 644                    regs->ip, regs->sp, error_code);
 645                print_vma_addr(" in ", regs->ip);
 646                printk("\n");
 647        }
 648
 649        force_sig(SIGSEGV, current);
 650        return;
 651
 652gp_in_vm86:
 653        local_irq_enable();
 654        handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 655        return;
 656
 657gp_in_kernel:
 658        if (!fixup_exception(regs)) {
 659                current->thread.error_code = error_code;
 660                current->thread.trap_no = 13;
 661                if (notify_die(DIE_GPF, "general protection fault", regs,
 662                                error_code, 13, SIGSEGV) == NOTIFY_STOP)
 663                        return;
 664                die("general protection fault", regs, error_code);
 665        }
 666}
 667
 668static __kprobes void
 669mem_parity_error(unsigned char reason, struct pt_regs * regs)
 670{
 671        printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
 672                "CPU %d.\n", reason, smp_processor_id());
 673        printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
 674
 675#if defined(CONFIG_EDAC)
 676        if(edac_handler_set()) {
 677                edac_atomic_assert_error();
 678                return;
 679        }
 680#endif
 681
 682        if (panic_on_unrecovered_nmi)
 683                panic("NMI: Not continuing");
 684
 685        printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 686
 687        /* Clear and disable the memory parity error line. */
 688        clear_mem_error(reason);
 689}
 690
 691static __kprobes void
 692io_check_error(unsigned char reason, struct pt_regs * regs)
 693{
 694        unsigned long i;
 695
 696        printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 697        show_registers(regs);
 698
 699        /* Re-enable the IOCK line, wait for a few seconds */
 700        reason = (reason & 0xf) | 8;
 701        outb(reason, 0x61);
 702        i = 2000;
 703        while (--i) udelay(1000);
 704        reason &= ~8;
 705        outb(reason, 0x61);
 706}
 707
 708static __kprobes void
 709unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 710{
 711#ifdef CONFIG_MCA
 712        /* Might actually be able to figure out what the guilty party
 713        * is. */
 714        if( MCA_bus ) {
 715                mca_handle_nmi();
 716                return;
 717        }
 718#endif
 719        printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
 720                "CPU %d.\n", reason, smp_processor_id());
 721        printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 722        if (panic_on_unrecovered_nmi)
 723                panic("NMI: Not continuing");
 724
 725        printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 726}
 727
 728static DEFINE_SPINLOCK(nmi_print_lock);
 729
 730void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
 731{
 732        if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
 733            NOTIFY_STOP)
 734                return;
 735
 736        spin_lock(&nmi_print_lock);
 737        /*
 738        * We are in trouble anyway, lets at least try
 739        * to get a message out.
 740        */
 741        bust_spinlocks(1);
 742        printk(KERN_EMERG "%s", msg);
 743        printk(" on CPU%d, ip %08lx, registers:\n",
 744                smp_processor_id(), regs->ip);
 745        show_registers(regs);
 746        console_silent();
 747        spin_unlock(&nmi_print_lock);
 748        bust_spinlocks(0);
 749
 750        /* If we are in kernel we are probably nested up pretty bad
 751         * and might aswell get out now while we still can.
 752        */
 753        if (!user_mode_vm(regs)) {
 754                current->thread.trap_no = 2;
 755                crash_kexec(regs);
 756        }
 757
 758        do_exit(SIGSEGV);
 759}
 760
 761static __kprobes void default_do_nmi(struct pt_regs * regs)
 762{
 763        unsigned char reason = 0;
 764
 765        /* Only the BSP gets external NMIs from the system.  */
 766        if (!smp_processor_id())
 767                reason = get_nmi_reason();
 768 
 769        if (!(reason & 0xc0)) {
 770                if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 771                                                        == NOTIFY_STOP)
 772                        return;
 773#ifdef CONFIG_X86_LOCAL_APIC
 774                /*
 775                 * Ok, so this is none of the documented NMI sources,
 776                 * so it must be the NMI watchdog.
 777                 */
 778                if (nmi_watchdog_tick(regs, reason))
 779                        return;
 780                if (!do_nmi_callback(regs, smp_processor_id()))
 781#endif
 782                        unknown_nmi_error(reason, regs);
 783
 784                return;
 785        }
 786        if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 787                return;
 788        if (reason & 0x80)
 789                mem_parity_error(reason, regs);
 790        if (reason & 0x40)
 791                io_check_error(reason, regs);
 792        /*
 793         * Reassert NMI in case it became active meanwhile
 794         * as it's edge-triggered.
 795         */
 796        reassert_nmi();
 797}
 798
 799static int ignore_nmis;
 800
 801__kprobes void do_nmi(struct pt_regs * regs, long error_code)
 802{
 803        int cpu;
 804
 805        nmi_enter();
 806
 807        cpu = smp_processor_id();
 808
 809        ++nmi_count(cpu);
 810
 811        if (!ignore_nmis)
 812                default_do_nmi(regs);
 813
 814        nmi_exit();
 815}
 816
 817void stop_nmi(void)
 818{
 819        acpi_nmi_disable();
 820        ignore_nmis++;
 821}
 822
 823void restart_nmi(void)
 824{
 825        ignore_nmis--;
 826        acpi_nmi_enable();
 827}
 828
 829#ifdef CONFIG_KPROBES
 830void __kprobes do_int3(struct pt_regs *regs, long error_code)
 831{
 832        trace_hardirqs_fixup();
 833
 834        if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 835                        == NOTIFY_STOP)
 836                return;
 837        /* This is an interrupt gate, because kprobes wants interrupts
 838        disabled.  Normal trap handlers don't. */
 839        restore_interrupts(regs);
 840        do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
 841}
 842#endif
 843
 844/*
 845 * Our handling of the processor debug registers is non-trivial.
 846 * We do not clear them on entry and exit from the kernel. Therefore
 847 * it is possible to get a watchpoint trap here from inside the kernel.
 848 * However, the code in ./ptrace.c has ensured that the user can
 849 * only set watchpoints on userspace addresses. Therefore the in-kernel
 850 * watchpoint trap can only occur in code which is reading/writing
 851 * from user space. Such code must not hold kernel locks (since it
 852 * can equally take a page fault), therefore it is safe to call
 853 * force_sig_info even though that claims and releases locks.
 854 * 
 855 * Code in ./signal.c ensures that the debug control register
 856 * is restored before we deliver any signal, and therefore that
 857 * user code runs with the correct debug control register even though
 858 * we clear it here.
 859 *
 860 * Being careful here means that we don't have to be as careful in a
 861 * lot of more complicated places (task switching can be a bit lazy
 862 * about restoring all the debug state, and ptrace doesn't have to
 863 * find every occurrence of the TF bit that could be saved away even
 864 * by user code)
 865 */
 866void __kprobes do_debug(struct pt_regs * regs, long error_code)
 867{
 868        unsigned int condition;
 869        struct task_struct *tsk = current;
 870
 871        trace_hardirqs_fixup();
 872
 873        get_debugreg(condition, 6);
 874
 875        /*
 876         * The processor cleared BTF, so don't mark that we need it set.
 877         */
 878        clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 879        tsk->thread.debugctlmsr = 0;
 880
 881        if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 882                                        SIGTRAP) == NOTIFY_STOP)
 883                return;
 884        /* It's safe to allow irq's after DR6 has been saved */
 885        if (regs->flags & X86_EFLAGS_IF)
 886                local_irq_enable();
 887
 888        /* Mask out spurious debug traps due to lazy DR7 setting */
 889        if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 890                if (!tsk->thread.debugreg7)
 891                        goto clear_dr7;
 892        }
 893
 894        if (regs->flags & VM_MASK)
 895                goto debug_vm86;
 896
 897        /* Save debug status register where ptrace can see it */
 898        tsk->thread.debugreg6 = condition;
 899
 900        /*
 901         * Single-stepping through TF: make sure we ignore any events in
 902         * kernel space (but re-enable TF when returning to user mode).
 903         */
 904        if (condition & DR_STEP) {
 905                /*
 906                 * We already checked v86 mode above, so we can
 907                 * check for kernel mode by just checking the CPL
 908                 * of CS.
 909                 */
 910                if (!user_mode(regs))
 911                        goto clear_TF_reenable;
 912        }
 913
 914        /* Ok, finally something we can handle */
 915        send_sigtrap(tsk, regs, error_code);
 916
 917        /* Disable additional traps. They'll be re-enabled when
 918         * the signal is delivered.
 919         */
 920clear_dr7:
 921        set_debugreg(0, 7);
 922        return;
 923
 924debug_vm86:
 925        handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 926        return;
 927
 928clear_TF_reenable:
 929        set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 930        regs->flags &= ~TF_MASK;
 931        return;
 932}
 933
 934/*
 935 * Note that we play around with the 'TS' bit in an attempt to get
 936 * the correct behaviour even in the presence of the asynchronous
 937 * IRQ13 behaviour
 938 */
 939void math_error(void __user *ip)
 940{
 941        struct task_struct * task;
 942        siginfo_t info;
 943        unsigned short cwd, swd;
 944
 945        /*
 946         * Save the info for the exception handler and clear the error.
 947         */
 948        task = current;
 949        save_init_fpu(task);
 950        task->thread.trap_no = 16;
 951        task->thread.error_code = 0;
 952        info.si_signo = SIGFPE;
 953        info.si_errno = 0;
 954        info.si_code = __SI_FAULT;
 955        info.si_addr = ip;
 956        /*
 957         * (~cwd & swd) will mask out exceptions that are not set to unmasked
 958         * status.  0x3f is the exception bits in these regs, 0x200 is the
 959         * C1 reg you need in case of a stack fault, 0x040 is the stack
 960         * fault bit.  We should only be taking one exception at a time,
 961         * so if this combination doesn't produce any single exception,
 962         * then we have a bad program that isn't syncronizing its FPU usage
 963         * and it will suffer the consequences since we won't be able to
 964         * fully reproduce the context of the exception
 965         */
 966        cwd = get_fpu_cwd(task);
 967        swd = get_fpu_swd(task);
 968        switch (swd & ~cwd & 0x3f) {
 969                case 0x000: /* No unmasked exception */
 970                        return;
 971                default:    /* Multiple exceptions */
 972                        break;
 973                case 0x001: /* Invalid Op */
 974                        /*
 975                         * swd & 0x240 == 0x040: Stack Underflow
 976                         * swd & 0x240 == 0x240: Stack Overflow
 977                         * User must clear the SF bit (0x40) if set
 978                         */
 979                        info.si_code = FPE_FLTINV;
 980                        break;
 981                case 0x002: /* Denormalize */
 982                case 0x010: /* Underflow */
 983                        info.si_code = FPE_FLTUND;
 984                        break;
 985                case 0x004: /* Zero Divide */
 986                        info.si_code = FPE_FLTDIV;
 987                        break;
 988                case 0x008: /* Overflow */
 989                        info.si_code = FPE_FLTOVF;
 990                        break;
 991                case 0x020: /* Precision */
 992                        info.si_code = FPE_FLTRES;
 993                        break;
 994        }
 995        force_sig_info(SIGFPE, &info, task);
 996}
 997
 998void do_coprocessor_error(struct pt_regs * regs, long error_code)
 999{
1000        ignore_fpu_irq = 1;
1001        math_error((void __user *)regs->ip);
1002}
1003
1004static void simd_math_error(void __user *ip)
1005{
1006        struct task_struct * task;
1007        siginfo_t info;
1008        unsigned short mxcsr;
1009
1010        /*
1011         * Save the info for the exception handler and clear the error.
1012         */
1013        task = current;
1014        save_init_fpu(task);
1015        task->thread.trap_no = 19;
1016        task->thread.error_code = 0;
1017        info.si_signo = SIGFPE;
1018        info.si_errno = 0;
1019        info.si_code = __SI_FAULT;
1020        info.si_addr = ip;
1021        /*
1022         * The SIMD FPU exceptions are handled a little differently, as there
1023         * is only a single status/control register.  Thus, to determine which
1024         * unmasked exception was caught we must mask the exception mask bits
1025         * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1026         */
1027        mxcsr = get_fpu_mxcsr(task);
1028        switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1029                case 0x000:
1030                default:
1031                        break;
1032                case 0x001: /* Invalid Op */
1033                        info.si_code = FPE_FLTINV;
1034                        break;
1035                case 0x002: /* Denormalize */
1036                case 0x010: /* Underflow */
1037                        info.si_code = FPE_FLTUND;
1038                        break;
1039                case 0x004: /* Zero Divide */
1040                        info.si_code = FPE_FLTDIV;
1041                        break;
1042                case 0x008: /* Overflow */
1043                        info.si_code = FPE_FLTOVF;
1044                        break;
1045                case 0x020: /* Precision */
1046                        info.si_code = FPE_FLTRES;
1047                        break;
1048        }
1049        force_sig_info(SIGFPE, &info, task);
1050}
1051
1052void do_simd_coprocessor_error(struct pt_regs * regs,
1053                                          long error_code)
1054{
1055        if (cpu_has_xmm) {
1056                /* Handle SIMD FPU exceptions on PIII+ processors. */
1057                ignore_fpu_irq = 1;
1058                simd_math_error((void __user *)regs->ip);
1059        } else {
1060                /*
1061                 * Handle strange cache flush from user space exception
1062                 * in all other cases.  This is undocumented behaviour.
1063                 */
1064                if (regs->flags & VM_MASK) {
1065                        handle_vm86_fault((struct kernel_vm86_regs *)regs,
1066                                          error_code);
1067                        return;
1068                }
1069                current->thread.trap_no = 19;
1070                current->thread.error_code = error_code;
1071                die_if_kernel("cache flush denied", regs, error_code);
1072                force_sig(SIGSEGV, current);
1073        }
1074}
1075
1076void do_spurious_interrupt_bug(struct pt_regs * regs,
1077                                          long error_code)
1078{
1079#if 0
1080        /* No need to warn about this any longer. */
1081        printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1082#endif
1083}
1084
1085unsigned long patch_espfix_desc(unsigned long uesp,
1086                                          unsigned long kesp)
1087{
1088        struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1089        unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1090        unsigned long new_kesp = kesp - base;
1091        unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1092        __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1093        /* Set up base for espfix segment */
1094        desc &= 0x00f0ff0000000000ULL;
1095        desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1096                ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1097                ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1098                (lim_pages & 0xffff);
1099        *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1100        return new_kesp;
1101}
1102
1103/*
1104 *  'math_state_restore()' saves the current math information in the
1105 * old math state array, and gets the new ones from the current task
1106 *
1107 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1108 * Don't touch unless you *really* know how it works.
1109 *
1110 * Must be called with kernel preemption disabled (in this case,
1111 * local interrupts are disabled at the call-site in entry.S).
1112 */
1113asmlinkage void math_state_restore(void)
1114{
1115        struct thread_info *thread = current_thread_info();
1116        struct task_struct *tsk = thread->task;
1117
1118        clts();         /* Allow maths ops (or we recurse) */
1119        if (!tsk_used_math(tsk))
1120                init_fpu(tsk);
1121        restore_fpu(tsk);
1122        thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
1123        tsk->fpu_counter++;
1124}
1125EXPORT_SYMBOL_GPL(math_state_restore);
1126
1127#ifndef CONFIG_MATH_EMULATION
1128
1129asmlinkage void math_emulate(long arg)
1130{
1131        printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1132        printk(KERN_EMERG "killing %s.\n",current->comm);
1133        force_sig(SIGFPE,current);
1134        schedule();
1135}
1136
1137#endif /* CONFIG_MATH_EMULATION */
1138
1139
1140void __init trap_init(void)
1141{
1142        int i;
1143
1144#ifdef CONFIG_EISA
1145        void __iomem *p = early_ioremap(0x0FFFD9, 4);
1146        if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1147                EISA_bus = 1;
1148        }
1149        early_iounmap(p, 4);
1150#endif
1151
1152#ifdef CONFIG_X86_LOCAL_APIC
1153        init_apic_mappings();
1154#endif
1155
1156        set_trap_gate(0,&divide_error);
1157        set_intr_gate(1,&debug);
1158        set_intr_gate(2,&nmi);
1159        set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1160        set_system_gate(4,&overflow);
1161        set_trap_gate(5,&bounds);
1162        set_trap_gate(6,&invalid_op);
1163        set_trap_gate(7,&device_not_available);
1164        set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1165        set_trap_gate(9,&coprocessor_segment_overrun);
1166        set_trap_gate(10,&invalid_TSS);
1167        set_trap_gate(11,&segment_not_present);
1168        set_trap_gate(12,&stack_segment);
1169        set_trap_gate(13,&general_protection);
1170        set_intr_gate(14,&page_fault);
1171        set_trap_gate(15,&spurious_interrupt_bug);
1172        set_trap_gate(16,&coprocessor_error);
1173        set_trap_gate(17,&alignment_check);
1174#ifdef CONFIG_X86_MCE
1175        set_trap_gate(18,&machine_check);
1176#endif
1177        set_trap_gate(19,&simd_coprocessor_error);
1178
1179        /*
1180         * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1181         * Generate a build-time error if the alignment is wrong.
1182         */
1183        BUILD_BUG_ON(offsetof(struct task_struct, thread.i387.fxsave) & 15);
1184        if (cpu_has_fxsr) {
1185                printk(KERN_INFO "Enabling fast FPU save and restore... ");
1186                set_in_cr4(X86_CR4_OSFXSR);
1187                printk("done.\n");
1188        }
1189        if (cpu_has_xmm) {
1190                printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1191                                "support... ");
1192                set_in_cr4(X86_CR4_OSXMMEXCPT);
1193                printk("done.\n");
1194        }
1195
1196        set_system_gate(SYSCALL_VECTOR,&system_call);
1197
1198        /* Reserve all the builtin and the syscall vector. */
1199        for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1200                set_bit(i, used_vectors);
1201        set_bit(SYSCALL_VECTOR, used_vectors);
1202
1203        /*
1204         * Should be a barrier for any external CPU state.
1205         */
1206        cpu_init();
1207
1208        trap_init_hook();
1209}
1210
1211static int __init kstack_setup(char *s)
1212{
1213        kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1214        return 1;
1215}
1216__setup("kstack=", kstack_setup);
1217
1218static int __init code_bytes_setup(char *s)
1219{
1220        code_bytes = simple_strtoul(s, NULL, 0);
1221        if (code_bytes > 8192)
1222                code_bytes = 8192;
1223
1224        return 1;
1225}
1226__setup("code_bytes=", code_bytes_setup);
1227
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.