Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / arch / i386 / kernel / traps.c
index d4fd1c5..6dfbd46 100644 (file)
@@ -27,6 +27,7 @@
 #include <linux/ptrace.h>
 #include <linux/utsname.h>
 #include <linux/kprobes.h>
+#include <linux/kexec.h>
 
 #ifdef CONFIG_EISA
 #include <linux/ioport.h>
@@ -51,7 +52,6 @@
 #include <asm/arch_hooks.h>
 #include <asm/kdebug.h>
 
-#include <linux/irq.h>
 #include <linux/module.h>
 #include <linux/vserver/debug.h>
 
@@ -93,18 +93,22 @@ asmlinkage void spurious_interrupt_bug(void);
 asmlinkage void machine_check(void);
 
 static int kstack_depth_to_print = 24;
-struct notifier_block *i386die_chain;
-static DEFINE_SPINLOCK(die_notifier_lock);
+ATOMIC_NOTIFIER_HEAD(i386die_chain);
+
+extern char last_sysfs_file[];
 
 int register_die_notifier(struct notifier_block *nb)
 {
-       int err = 0;
-       unsigned long flags;
-       spin_lock_irqsave(&die_notifier_lock, flags);
-       err = notifier_chain_register(&i386die_chain, nb);
-       spin_unlock_irqrestore(&die_notifier_lock, flags);
-       return err;
+       vmalloc_sync_all();
+       return atomic_notifier_chain_register(&i386die_chain, nb);
 }
+EXPORT_SYMBOL(register_die_notifier);
+
+int unregister_die_notifier(struct notifier_block *nb)
+{
+       return atomic_notifier_chain_unregister(&i386die_chain, nb);
+}
+EXPORT_SYMBOL(unregister_die_notifier);
 
 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
 {
@@ -112,33 +116,59 @@ static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
                p < (void *)tinfo + THREAD_SIZE - 3;
 }
 
+/*
+ * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
+ */
+static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
+                                       int printed)
+{
+       if (!printed)
+               printk(log_lvl);
+
+#if CONFIG_STACK_BACKTRACE_COLS == 1
+       printk(" [<%08lx>] ", addr);
+#else
+       printk(" <%08lx> ", addr);
+#endif
+       print_symbol("%s", addr);
+
+       printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
+       if (printed)
+               printk(" ");
+       else
+               printk("\n");
+
+       return printed;
+}
+
 static inline unsigned long print_context_stack(struct thread_info *tinfo,
-                               unsigned long *stack, unsigned long ebp)
+                               unsigned long *stack, unsigned long ebp,
+                               char *log_lvl)
 {
        unsigned long addr;
+       int printed = 0; /* nr of entries already printed on current line */
 
 #ifdef CONFIG_FRAME_POINTER
        while (valid_stack_ptr(tinfo, (void *)ebp)) {
                addr = *(unsigned long *)(ebp + 4);
-               printk(" [<%08lx>] ", addr);
-               print_symbol("%s", addr);
-               printk("\n");
+               printed = print_addr_and_symbol(addr, log_lvl, printed);
                ebp = *(unsigned long *)ebp;
        }
 #else
        while (valid_stack_ptr(tinfo, stack)) {
                addr = *stack++;
-               if (__kernel_text_address(addr)) {
-                       printk(" [<%08lx>]", addr);
-                       print_symbol(" %s", addr);
-                       printk("\n");
-               }
+               if (__kernel_text_address(addr))
+                       printed = print_addr_and_symbol(addr, log_lvl, printed);
        }
 #endif
+       if (printed)
+               printk("\n");
+
        return ebp;
 }
 
-void show_trace(struct task_struct *task, unsigned long * stack)
+static void show_trace_log_lvl(struct task_struct *task,
+                              unsigned long *stack, char *log_lvl)
 {
        unsigned long ebp;
 
@@ -157,15 +187,21 @@ void show_trace(struct task_struct *task, unsigned long * stack)
                struct thread_info *context;
                context = (struct thread_info *)
                        ((unsigned long)stack & (~(THREAD_SIZE - 1)));
-               ebp = print_context_stack(context, stack, ebp);
+               ebp = print_context_stack(context, stack, ebp, log_lvl);
                stack = (unsigned long*)context->previous_esp;
                if (!stack)
                        break;
-               printk(" =======================\n");
+               printk("%s =======================\n", log_lvl);
        }
 }
 
-void show_stack(struct task_struct *task, unsigned long *esp)
+void show_trace(struct task_struct *task, unsigned long * stack)
+{
+       show_trace_log_lvl(task, stack, "");
+}
+
+static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
+                              char *log_lvl)
 {
        unsigned long *stack;
        int i;
@@ -182,11 +218,17 @@ void show_stack(struct task_struct *task, unsigned long *esp)
                if (kstack_end(stack))
                        break;
                if (i && ((i % 8) == 0))
-                       printk("\n       ");
+                       printk("\n%s       ", log_lvl);
                printk("%08lx ", *stack++);
        }
-       printk("\nCall Trace:\n");
-       show_trace(task, esp);
+       printk("\n%sCall Trace:\n", log_lvl);
+       show_trace_log_lvl(task, esp, log_lvl);
+}
+
+void show_stack(struct task_struct *task, unsigned long *esp)
+{
+       printk("       ");
+       show_stack_log_lvl(task, esp, "");
 }
 
 /*
@@ -209,47 +251,50 @@ void show_registers(struct pt_regs *regs)
        unsigned short ss;
 
        esp = (unsigned long) (&regs->esp);
-       ss = __KERNEL_DS;
-       if (regs->xcs & 3) {
+       savesegment(ss, ss);
+       if (user_mode_vm(regs)) {
                in_kernel = 0;
                esp = regs->esp;
                ss = regs->xss & 0xffff;
        }
        print_modules();
-       printk("CPU:    %d\nEIP:    %04x:[<%08lx>]    %s VLI\nEFLAGS: %08lx"
-                       "   (%s) \n",
+       printk(KERN_EMERG "CPU:    %d\nEIP:    %04x:[<%08lx>]    %s VLI\n"
+                       "EFLAGS: %08lx   (%s %.*s) \n",
                smp_processor_id(), 0xffff & regs->xcs, regs->eip,
-               print_tainted(), regs->eflags, system_utsname.release);
-       print_symbol("EIP is at %s\n", regs->eip);
-       printk("eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
+               print_tainted(), regs->eflags, system_utsname.release,
+               (int)strcspn(system_utsname.version, " "),
+               system_utsname.version);
+       print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
+       printk(KERN_EMERG "eax: %08lx   ebx: %08lx   ecx: %08lx   edx: %08lx\n",
                regs->eax, regs->ebx, regs->ecx, regs->edx);
-       printk("esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
+       printk(KERN_EMERG "esi: %08lx   edi: %08lx   ebp: %08lx   esp: %08lx\n",
                regs->esi, regs->edi, regs->ebp, esp);
-       printk("ds: %04x   es: %04x   ss: %04x\n",
+       printk(KERN_EMERG "ds: %04x   es: %04x   ss: %04x\n",
                regs->xds & 0xffff, regs->xes & 0xffff, ss);
-       printk("Process %s (pid: %d, threadinfo=%p task=%p)",
-               current->comm, current->pid, current_thread_info(), current);
+       printk(KERN_EMERG "Process %s (pid: %d[#%u], threadinfo=%p task=%p)",
+               current->comm, current->pid, current->xid,
+               current_thread_info(), current);
        /*
         * When in-kernel, we also print out the stack and code at the
         * time of the fault..
         */
        if (in_kernel) {
-               u8 *eip;
+               u8 __user *eip;
 
-               printk("\nStack: ");
-               show_stack(NULL, (unsigned long*)esp);
+               printk("\n" KERN_EMERG "Stack: ");
+               show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
 
-               printk("Code: ");
+               printk(KERN_EMERG "Code: ");
 
-               eip = (u8 *)regs->eip - 43;
+               eip = (u8 __user *)regs->eip - 43;
                for (i = 0; i < 64; i++, eip++) {
                        unsigned char c;
 
-                       if (eip < (u8 *)PAGE_OFFSET || __get_user(c, eip)) {
+                       if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
                                printk(" Bad EIP value.");
                                break;
                        }
-                       if (eip == (u8 *)regs->eip)
+                       if (eip == (u8 __user *)regs->eip)
                                printk("<%02x> ", c);
                        else
                                printk("%02x ", c);
@@ -266,34 +311,34 @@ static void handle_BUG(struct pt_regs *regs)
        char c;
        unsigned long eip;
 
-       if (regs->xcs & 3)
-               goto no_bug;            /* Not in kernel */
-
        eip = regs->eip;
 
        if (eip < PAGE_OFFSET)
                goto no_bug;
-       if (__get_user(ud2, (unsigned short *)eip))
+       if (__get_user(ud2, (unsigned short __user *)eip))
                goto no_bug;
        if (ud2 != 0x0b0f)
                goto no_bug;
-       if (__get_user(line, (unsigned short *)(eip + 2)))
+       if (__get_user(line, (unsigned short __user *)(eip + 2)))
                goto bug;
-       if (__get_user(file, (char **)(eip + 4)) ||
+       if (__get_user(file, (char * __user *)(eip + 4)) ||
                (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
                file = "<bad filename>";
 
-       printk("------------[ cut here ]------------\n");
-       printk(KERN_ALERT "kernel BUG at %s:%d!\n", file, line);
+       printk(KERN_EMERG "------------[ cut here ]------------\n");
+       printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
 
 no_bug:
        return;
 
        /* Here we know it was a BUG but file-n-line is unavailable */
 bug:
-       printk("Kernel BUG\n");
+       printk(KERN_EMERG "Kernel BUG\n");
 }
 
+/* This is gone through when something in the kernel
+ * has done something bad and is about to be terminated.
+*/
 void die(const char * str, struct pt_regs * regs, long err)
 {
        static struct {
@@ -306,43 +351,79 @@ void die(const char * str, struct pt_regs * regs, long err)
                .lock_owner_depth =     0
        };
        static int die_counter;
+       unsigned long flags;
+
+       oops_enter();
 
        vxh_throw_oops();
-       if (die.lock_owner != _smp_processor_id()) {
+
+       if (die.lock_owner != raw_smp_processor_id()) {
                console_verbose();
-               spin_lock_irq(&die.lock);
+               spin_lock_irqsave(&die.lock, flags);
                die.lock_owner = smp_processor_id();
                die.lock_owner_depth = 0;
                bust_spinlocks(1);
        }
+       else
+               local_save_flags(flags);
 
        if (++die.lock_owner_depth < 3) {
                int nl = 0;
+               unsigned long esp;
+               unsigned short ss;
+
                handle_BUG(regs);
-               printk(KERN_ALERT "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
+               printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
 #ifdef CONFIG_PREEMPT
-               printk("PREEMPT ");
+               printk(KERN_EMERG "PREEMPT ");
                nl = 1;
 #endif
 #ifdef CONFIG_SMP
+               if (!nl)
+                       printk(KERN_EMERG);
                printk("SMP ");
                nl = 1;
 #endif
 #ifdef CONFIG_DEBUG_PAGEALLOC
+               if (!nl)
+                       printk(KERN_EMERG);
                printk("DEBUG_PAGEALLOC");
                nl = 1;
 #endif
                if (nl)
                        printk("\n");
-       notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
-               show_registers(regs);
+#ifdef CONFIG_SYSFS
+               printk(KERN_ALERT "last sysfs file: %s\n", last_sysfs_file);
+#endif
+               if (notify_die(DIE_OOPS, str, regs, err,
+                       current->thread.trap_no, SIGSEGV) != NOTIFY_STOP) {
+                       show_registers(regs);
+                       vxh_dump_history();
+                       /* Executive summary in case the oops scrolled away */
+                       esp = (unsigned long) (&regs->esp);
+                       savesegment(ss, ss);
+                       if (user_mode(regs)) {
+                               esp = regs->esp;
+                               ss = regs->xss & 0xffff;
+                       }
+                       printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
+                       print_symbol("%s", regs->eip);
+                       printk(" SS:ESP %04x:%08lx\n", ss, esp);
+               } else
+                       regs = NULL;
        } else
-               printk(KERN_ERR "Recursive die() failure, output suppressed\n");
+               printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
 
        bust_spinlocks(0);
        die.lock_owner = -1;
-       spin_unlock_irq(&die.lock);
-       vxh_dump_history();
+       spin_unlock_irqrestore(&die.lock, flags);
+
+       if (!regs)
+               return;
+
+       if (kexec_should_crash(current))
+               crash_kexec(regs);
+
        if (in_interrupt())
                panic("Fatal exception in interrupt");
 
@@ -351,31 +432,34 @@ void die(const char * str, struct pt_regs * regs, long err)
                ssleep(5);
                panic("Fatal exception");
        }
+       oops_exit();
        do_exit(SIGSEGV);
 }
 
 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 {
-       if (!(regs->eflags & VM_MASK) && !(3 & regs->xcs))
+       if (!user_mode_vm(regs))
                die(str, regs, err);
 }
 
-static void do_trap(int trapnr, int signr, char *str, int vm86,
-                          struct pt_regs * regs, long error_code, siginfo_t *info)
+static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
+                             struct pt_regs * regs, long error_code,
+                             siginfo_t *info)
 {
+       struct task_struct *tsk = current;
+       tsk->thread.error_code = error_code;
+       tsk->thread.trap_no = trapnr;
+
        if (regs->eflags & VM_MASK) {
                if (vm86)
                        goto vm86_trap;
                goto trap_signal;
        }
 
-       if (!(regs->xcs & 3))
+       if (!user_mode(regs))
                goto kernel_trap;
 
        trap_signal: {
-               struct task_struct *tsk = current;
-               tsk->thread.error_code = error_code;
-               tsk->thread.trap_no = trapnr;
                if (info)
                        force_sig_info(signr, info, tsk);
                else
@@ -448,19 +532,96 @@ DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
 #endif
 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
-DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->eip)
+DO_ERROR_INFO( 6, SIGILL,  "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
-DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
 
-fastcall void do_general_protection(struct pt_regs * regs, long error_code)
+
+/*
+ * lazy-check for CS validity on exec-shield binaries:
+ *
+ * the original non-exec stack patch was written by
+ * Solar Designer <solar at openwall.com>. Thanks!
+ */
+static int
+check_lazy_exec_limit(int cpu, struct pt_regs *regs, long error_code)
+{
+       struct desc_struct *desc1, *desc2;
+       struct vm_area_struct *vma;
+       unsigned long limit;
+
+       if (current->mm == NULL)
+               return 0;
+
+       limit = -1UL;
+       if (current->mm->context.exec_limit != -1UL) {
+               limit = PAGE_SIZE;
+               spin_lock(&current->mm->page_table_lock);
+               for (vma = current->mm->mmap; vma; vma = vma->vm_next)
+                       if ((vma->vm_flags & VM_EXEC) && (vma->vm_end > limit))
+                               limit = vma->vm_end;
+               spin_unlock(&current->mm->page_table_lock);
+               if (limit >= TASK_SIZE)
+                       limit = -1UL;
+               current->mm->context.exec_limit = limit;
+       }
+       set_user_cs(&current->mm->context.user_cs, limit);
+
+       desc1 = &current->mm->context.user_cs;
+       desc2 = get_cpu_gdt_table(cpu) + GDT_ENTRY_DEFAULT_USER_CS;
+
+       if (desc1->a != desc2->a || desc1->b != desc2->b) {
+               /*
+                * The CS was not in sync - reload it and retry the
+                * instruction. If the instruction still faults then
+                * we won't hit this branch next time around.
+                */
+               if (print_fatal_signals >= 2) {
+                       printk("#GPF fixup (%ld[seg:%lx]) at %08lx, CPU#%d.\n", error_code, error_code/8, regs->eip, smp_processor_id());
+                       printk(" exec_limit: %08lx, user_cs: %08lx/%08lx, CPU_cs: %08lx/%08lx.\n", current->mm->context.exec_limit, desc1->a, desc1->b, desc2->a, desc2->b);
+               }
+               load_user_cs_desc(cpu, current->mm);
+               return 1;
+       }
+
+       return 0;
+}
+
+/*
+ * The fixup code for errors in iret jumps to here (iret_exc).  It loses
+ * the original trap number and error code.  The bogus trap 32 and error
+ * code 0 are what the vanilla kernel delivers via:
+ * DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
+ *
+ * In case of a general protection fault in the iret instruction, we
+ * need to check for a lazy CS update for exec-shield.
+ */
+fastcall void do_iret_error(struct pt_regs *regs, long error_code)
+{
+       int ok = check_lazy_exec_limit(get_cpu(), regs, error_code);
+       put_cpu();
+       if (!ok && notify_die(DIE_TRAP, "iret exception", regs,
+                             error_code, 32, SIGSEGV) != NOTIFY_STOP) {
+               siginfo_t info;
+               info.si_signo = SIGSEGV;
+               info.si_errno = 0;
+               info.si_code = ILL_BADSTK;
+               info.si_addr = 0;
+               do_trap(32, SIGSEGV, "iret exception", 0, regs, error_code,
+                       &info);
+       }
+}
+
+fastcall void __kprobes do_general_protection(struct pt_regs * regs,
+                                             long error_code)
 {
        int cpu = get_cpu();
        struct tss_struct *tss = &per_cpu(init_tss, cpu);
        struct thread_struct *thread = &current->thread;
+       int ok;
 
        /*
         * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
@@ -483,28 +644,45 @@ fastcall void do_general_protection(struct pt_regs * regs, long error_code)
                                tss->io_bitmap_max - thread->io_bitmap_max);
                tss->io_bitmap_max = thread->io_bitmap_max;
                tss->io_bitmap_base = IO_BITMAP_OFFSET;
+               tss->io_bitmap_owner = thread;
                put_cpu();
                return;
        }
-       put_cpu();
+
+       current->thread.error_code = error_code;
+       current->thread.trap_no = 13;
 
        if (regs->eflags & VM_MASK)
                goto gp_in_vm86;
 
-       if (!(regs->xcs & 3))
+       if (!user_mode(regs))
                goto gp_in_kernel;
 
+       ok = check_lazy_exec_limit(cpu, regs, error_code);
+
+       put_cpu();
+
+       if (ok)
+               return;
+
+       if (print_fatal_signals) {
+               printk("#GPF(%ld[seg:%lx]) at %08lx, CPU#%d.\n", error_code, error_code/8, regs->eip, smp_processor_id());
+               printk(" exec_limit: %08lx, user_cs: %08lx/%08lx.\n", current->mm->context.exec_limit, current->mm->context.user_cs.a, current->mm->context.user_cs.b);
+       }
+
        current->thread.error_code = error_code;
        current->thread.trap_no = 13;
        force_sig(SIGSEGV, current);
        return;
 
 gp_in_vm86:
+       put_cpu();
        local_irq_enable();
        handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
        return;
 
 gp_in_kernel:
+       put_cpu();
        if (!fixup_exception(regs)) {
                if (notify_die(DIE_GPF, "general protection fault", regs,
                                error_code, 13, SIGSEGV) == NOTIFY_STOP)
@@ -515,8 +693,10 @@ gp_in_kernel:
 
 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
 {
-       printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
-       printk("You probably have a hardware problem with your RAM chips\n");
+       printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
+                       "to continue\n");
+       printk(KERN_EMERG "You probably have a hardware problem with your RAM "
+                       "chips\n");
 
        /* Clear and disable the memory parity error line. */
        clear_mem_error(reason);
@@ -526,7 +706,7 @@ static void io_check_error(unsigned char reason, struct pt_regs * regs)
 {
        unsigned long i;
 
-       printk("NMI: IOCK error (debug interrupt?)\n");
+       printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
        show_registers(regs);
 
        /* Re-enable the IOCK line, wait for a few seconds */
@@ -558,20 +738,33 @@ static DEFINE_SPINLOCK(nmi_print_lock);
 
 void die_nmi (struct pt_regs *regs, const char *msg)
 {
+       if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
+           NOTIFY_STOP)
+               return;
+
        spin_lock(&nmi_print_lock);
        /*
        * We are in trouble anyway, lets at least try
        * to get a message out.
        */
        bust_spinlocks(1);
-       printk(msg);
+       printk(KERN_EMERG "%s", msg);
        printk(" on CPU%d, eip %08lx, registers:\n",
                smp_processor_id(), regs->eip);
        show_registers(regs);
-       printk("console shuts up ...\n");
+       printk(KERN_EMERG "console shuts up ...\n");
        console_silent();
        spin_unlock(&nmi_print_lock);
        bust_spinlocks(0);
+
+       /* If we are in kernel we are probably nested up pretty bad
+        * and might aswell get out now while we still can.
+       */
+       if (!user_mode_vm(regs)) {
+               current->thread.trap_no = 2;
+               crash_kexec(regs);
+       }
+
        do_exit(SIGSEGV);
 }
 
@@ -584,7 +777,7 @@ static void default_do_nmi(struct pt_regs * regs)
                reason = get_nmi_reason();
  
        if (!(reason & 0xc0)) {
-               if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
+               if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
                                                        == NOTIFY_STOP)
                        return;
 #ifdef CONFIG_X86_LOCAL_APIC
@@ -600,7 +793,7 @@ static void default_do_nmi(struct pt_regs * regs)
                unknown_nmi_error(reason, regs);
                return;
        }
-       if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
+       if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
                return;
        if (reason & 0x80)
                mem_parity_error(reason, regs);
@@ -627,9 +820,10 @@ fastcall void do_nmi(struct pt_regs * regs, long error_code)
        nmi_enter();
 
        cpu = smp_processor_id();
+
        ++nmi_count(cpu);
 
-       if (!nmi_callback(regs, cpu))
+       if (!rcu_dereference(nmi_callback)(regs, cpu))
                default_do_nmi(regs);
 
        nmi_exit();
@@ -637,16 +831,19 @@ fastcall void do_nmi(struct pt_regs * regs, long error_code)
 
 void set_nmi_callback(nmi_callback_t callback)
 {
-       nmi_callback = callback;
+       vmalloc_sync_all();
+       rcu_assign_pointer(nmi_callback, callback);
 }
+EXPORT_SYMBOL_GPL(set_nmi_callback);
 
 void unset_nmi_callback(void)
 {
        nmi_callback = dummy_nmi_callback;
 }
+EXPORT_SYMBOL_GPL(unset_nmi_callback);
 
 #ifdef CONFIG_KPROBES
-fastcall void do_int3(struct pt_regs *regs, long error_code)
+fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
 {
        if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
                        == NOTIFY_STOP)
@@ -680,12 +877,12 @@ fastcall void do_int3(struct pt_regs *regs, long error_code)
  * find every occurrence of the TF bit that could be saved away even
  * by user code)
  */
-fastcall void do_debug(struct pt_regs * regs, long error_code)
+fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
 {
        unsigned int condition;
        struct task_struct *tsk = current;
 
-       __asm__ __volatile__("movl %%db6,%0" : "=r" (condition));
+       get_debugreg(condition, 6);
 
        if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
                                        SIGTRAP) == NOTIFY_STOP)
@@ -716,7 +913,7 @@ fastcall void do_debug(struct pt_regs * regs, long error_code)
                 * check for kernel mode by just checking the CPL
                 * of CS.
                 */
-               if ((regs->xcs & 3) == 0)
+               if (!user_mode(regs))
                        goto clear_TF_reenable;
        }
 
@@ -727,9 +924,7 @@ fastcall void do_debug(struct pt_regs * regs, long error_code)
         * the signal is delivered.
         */
 clear_dr7:
-       __asm__("movl %0,%%db7"
-               : /* no output */
-               : "r" (0));
+       set_debugreg(0, 7);
        return;
 
 debug_vm86:
@@ -776,15 +971,18 @@ void math_error(void __user *eip)
         */
        cwd = get_fpu_cwd(task);
        swd = get_fpu_swd(task);
-       switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
-               case 0x000:
-               default:
+       switch (swd & ~cwd & 0x3f) {
+               case 0x000: /* No unmasked exception */
+                       return;
+               default:    /* Multiple exceptions */
                        break;
                case 0x001: /* Invalid Op */
-               case 0x041: /* Stack Fault */
-               case 0x241: /* Stack Fault | Direction */
+                       /*
+                        * swd & 0x240 == 0x040: Stack Underflow
+                        * swd & 0x240 == 0x240: Stack Overflow
+                        * User must clear the SF bit (0x40) if set
+                        */
                        info.si_code = FPE_FLTINV;
-                       /* Should we clear the SF or let user space do it ???? */
                        break;
                case 0x002: /* Denormalize */
                case 0x010: /* Underflow */
@@ -874,9 +1072,9 @@ fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
                                          error_code);
                        return;
                }
-               die_if_kernel("cache flush denied", regs, error_code);
                current->thread.trap_no = 19;
                current->thread.error_code = error_code;
+               die_if_kernel("cache flush denied", regs, error_code);
                force_sig(SIGSEGV, current);
        }
 }
@@ -961,8 +1159,8 @@ asmlinkage void math_state_restore(struct pt_regs regs)
 
 asmlinkage void math_emulate(long arg)
 {
-       printk("math-emulation not enabled and no coprocessor found.\n");
-       printk("killing %s.\n",current->comm);
+       printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
+       printk(KERN_EMERG "killing %s.\n",current->comm);
        force_sig(SIGFPE,current);
        schedule();
 }
@@ -979,7 +1177,7 @@ void __init trap_init_f00f_bug(void)
         * it uses the read-only mapped virtual address.
         */
        idt_descr.address = fix_to_virt(FIX_F00F_IDT);
-       __asm__ __volatile__("lidt %0" : : "m" (idt_descr));
+       load_idt(&idt_descr);
 }
 #endif
 
@@ -1049,9 +1247,9 @@ void __init trap_init(void)
        set_trap_gate(0,&divide_error);
        set_intr_gate(1,&debug);
        set_intr_gate(2,&nmi);
-       set_system_intr_gate(3, &int3); /* int3-5 can be called from all */
+       set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
        set_system_gate(4,&overflow);
-       set_system_gate(5,&bounds);
+       set_trap_gate(5,&bounds);
        set_trap_gate(6,&invalid_op);
        set_trap_gate(7,&device_not_available);
        set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
@@ -1069,6 +1267,28 @@ void __init trap_init(void)
 #endif
        set_trap_gate(19,&simd_coprocessor_error);
 
+       if (cpu_has_fxsr) {
+               /*
+                * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
+                * Generates a compile-time "error: zero width for bit-field" if
+                * the alignment is wrong.
+                */
+               struct fxsrAlignAssert {
+                       int _:!(offsetof(struct task_struct,
+                                       thread.i387.fxsave) & 15);
+               };
+
+               printk(KERN_INFO "Enabling fast FPU save and restore... ");
+               set_in_cr4(X86_CR4_OSFXSR);
+               printk("done.\n");
+       }
+       if (cpu_has_xmm) {
+               printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
+                               "support... ");
+               set_in_cr4(X86_CR4_OSXMMEXCPT);
+               printk("done.\n");
+       }
+
        set_system_gate(SYSCALL_VECTOR,&system_call);
 
        /*
@@ -1082,6 +1302,6 @@ void __init trap_init(void)
 static int __init kstack_setup(char *s)
 {
        kstack_depth_to_print = simple_strtoul(s, NULL, 0);
-       return 0;
+       return 1;
 }
 __setup("kstack=", kstack_setup);