[PATCH] Time: i386 Conversion - part 2: Rework TSC Support
[safe/jmp/linux-2.6] / arch / i386 / kernel / traps.c
index 431a551..dcc1447 100644 (file)
@@ -52,7 +52,6 @@
 #include <asm/arch_hooks.h>
 #include <asm/kdebug.h>
 
-#include <linux/irq.h>
 #include <linux/module.h>
 
 #include "mach_traps.h"
@@ -93,53 +92,86 @@ 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);
 
 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)
 {
        return  p > (void *)tinfo &&
                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);
+               /*
+                * break out of recursive entries (such as
+                * end_of_stack_stop_unwind_function):
+                */
+               if (ebp == *(unsigned long *)ebp)
+                       break;
                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;
 
@@ -158,15 +190,21 @@ 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;
@@ -183,11 +221,17 @@ 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, "");
 }
 
 /*
@@ -211,25 +255,28 @@ void show_registers(struct pt_regs *regs)
 
        esp = (unsigned long) (&regs->esp);
        savesegment(ss, ss);
-       if (user_mode(regs)) {
+       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, ti=%p task=%p task.ti=%p)",
+               TASK_COMM_LEN, current->comm, current->pid,
+               current_thread_info(), current, current->thread_info);
        /*
         * When in-kernel, we also print out the stack and code at the
         * time of the fault..
@@ -237,10 +284,10 @@ void show_registers(struct pt_regs *regs)
        if (in_kernel) {
                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 __user *)regs->eip - 43;
                for (i = 0; i < 64; i++, eip++) {
@@ -281,15 +328,15 @@ static void handle_BUG(struct pt_regs *regs)
                (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
@@ -307,41 +354,71 @@ void die(const char * str, struct pt_regs * regs, long err)
                .lock_owner_depth =     0
        };
        static int die_counter;
+       unsigned long flags;
+
+       oops_enter();
 
        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);
+               if (notify_die(DIE_OOPS, str, regs, err,
+                                       current->thread.trap_no, SIGSEGV) !=
+                               NOTIFY_STOP) {
+                       show_registers(regs);
+                       /* 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);
+       spin_unlock_irqrestore(&die.lock, flags);
+
+       if (!regs)
+               return;
 
        if (kexec_should_crash(current))
                crash_kexec(regs);
@@ -354,6 +431,7 @@ void die(const char * str, struct pt_regs * regs, long err)
                ssleep(5);
                panic("Fatal exception");
        }
+       oops_exit();
        do_exit(SIGSEGV);
 }
 
@@ -453,7 +531,7 @@ 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)
@@ -489,6 +567,7 @@ fastcall void __kprobes do_general_protection(struct pt_regs * regs,
                                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;
        }
@@ -524,8 +603,10 @@ 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);
@@ -535,7 +616,7 @@ 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 */
@@ -567,7 +648,7 @@ static DEFINE_SPINLOCK(nmi_print_lock);
 
 void die_nmi (struct pt_regs *regs, const char *msg)
 {
-       if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 0, SIGINT) ==
+       if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
            NOTIFY_STOP)
                return;
 
@@ -577,11 +658,11 @@ void die_nmi (struct pt_regs *regs, const char *msg)
        * 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);
@@ -589,7 +670,7 @@ void die_nmi (struct pt_regs *regs, const char *msg)
        /* 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(regs)) {
+       if (!user_mode_vm(regs)) {
                current->thread.trap_no = 2;
                crash_kexec(regs);
        }
@@ -606,7 +687,7 @@ 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
@@ -622,7 +703,7 @@ 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);
@@ -650,13 +731,6 @@ fastcall void do_nmi(struct pt_regs * regs, long error_code)
 
        cpu = smp_processor_id();
 
-#ifdef CONFIG_HOTPLUG_CPU
-       if (!cpu_online(cpu)) {
-               nmi_exit();
-               return;
-       }
-#endif
-
        ++nmi_count(cpu);
 
        if (!rcu_dereference(nmi_callback)(regs, cpu))
@@ -667,6 +741,7 @@ fastcall void do_nmi(struct pt_regs * regs, long error_code)
 
 void set_nmi_callback(nmi_callback_t callback)
 {
+       vmalloc_sync_all();
        rcu_assign_pointer(nmi_callback, callback);
 }
 EXPORT_SYMBOL_GPL(set_nmi_callback);
@@ -994,8 +1069,8 @@ 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();
 }
@@ -1082,9 +1157,9 @@ 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);
@@ -1102,6 +1177,28 @@ 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);
 
        /*
@@ -1115,6 +1212,6 @@ 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);