linux 2.6.16.38 w/ vs2.0.3-rc1
[linux-2.6.git] / arch / x86_64 / kernel / traps.c
index 6d38a2d..dd879d5 100644 (file)
@@ -6,12 +6,15 @@
  *
  *  Pentium III FXSR, SSE support
  *     Gareth Hughes <gareth@valinux.com>, May 2000
+ *
+ *  $Id: traps.c,v 1.36 2002/03/24 11:09:10 ak Exp $
  */
 
 /*
  * 'Traps.c' handles hardware traps and faults after we have saved some
  * state in 'entry.S'.
  */
+#include <linux/config.h>
 #include <linux/sched.h>
 #include <linux/kernel.h>
 #include <linux/string.h>
@@ -28,7 +31,6 @@
 #include <linux/nmi.h>
 #include <linux/kprobes.h>
 #include <linux/kexec.h>
-#include <linux/unwind.h>
 
 #include <asm/system.h>
 #include <asm/uaccess.h>
 #include <asm/i387.h>
 #include <asm/kdebug.h>
 #include <asm/processor.h>
-#include <asm/unwind.h>
+
 #include <asm/smp.h>
 #include <asm/pgalloc.h>
 #include <asm/pda.h>
 #include <asm/proto.h>
 #include <asm/nmi.h>
-#include <asm/stacktrace.h>
+
+extern struct gate_struct idt_table[256]; 
 
 asmlinkage void divide_error(void);
 asmlinkage void debug(void);
@@ -69,23 +72,18 @@ asmlinkage void alignment_check(void);
 asmlinkage void machine_check(void);
 asmlinkage void spurious_interrupt_bug(void);
 
-ATOMIC_NOTIFIER_HEAD(die_chain);
-EXPORT_SYMBOL(die_chain);
-
-extern char last_sysfs_file[];
+struct notifier_block *die_chain;
+static DEFINE_SPINLOCK(die_notifier_lock);
 
 int register_die_notifier(struct notifier_block *nb)
 {
-       vmalloc_sync_all();
-       return atomic_notifier_chain_register(&die_chain, nb);
-}
-EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
-
-int unregister_die_notifier(struct notifier_block *nb)
-{
-       return atomic_notifier_chain_unregister(&die_chain, nb);
+       int err = 0;
+       unsigned long flags;
+       spin_lock_irqsave(&die_notifier_lock, flags);
+       err = notifier_chain_register(&die_chain, nb);
+       spin_unlock_irqrestore(&die_notifier_lock, flags);
+       return err;
 }
-EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
 
 static inline void conditional_sti(struct pt_regs *regs)
 {
@@ -104,48 +102,38 @@ static inline void preempt_conditional_cli(struct pt_regs *regs)
 {
        if (regs->eflags & X86_EFLAGS_IF)
                local_irq_disable();
-       /* Make sure to not schedule here because we could be running
-          on an exception stack. */
        preempt_enable_no_resched();
 }
 
-static int kstack_depth_to_print = 12;
-#ifdef CONFIG_STACK_UNWIND
-static int call_trace = 1;
-#else
-#define call_trace (-1)
-#endif
+static int kstack_depth_to_print = 10;
 
 #ifdef CONFIG_KALLSYMS
-# include <linux/kallsyms.h>
-void printk_address(unsigned long address)
-{
+#include <linux/kallsyms.h> 
+int printk_address(unsigned long address)
+{ 
        unsigned long offset = 0, symsize;
        const char *symname;
        char *modname;
-       char *delim = ":";
+       char *delim = ":"; 
        char namebuf[128];
 
-       symname = kallsyms_lookup(address, &symsize, &offset,
-                                       &modname, namebuf);
-       if (!symname) {
-               printk(" [<%016lx>]\n", address);
-               return;
-       }
-       if (!modname)
+       symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf); 
+       if (!symname) 
+               return printk("[<%016lx>]", address);
+       if (!modname) 
                modname = delim = "";           
-       printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
-               address, delim, modname, delim, symname, offset, symsize);
-}
+        return printk("<%016lx>{%s%s%s%s%+ld}",
+                     address,delim,modname,delim,symname,offset); 
+} 
 #else
-void printk_address(unsigned long address)
-{
-       printk(" [<%016lx>]\n", address);
-}
+int printk_address(unsigned long address)
+{ 
+       return printk("[<%016lx>]", address);
+} 
 #endif
 
 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
-                                       unsigned *usedp, char **idp)
+                                       unsigned *usedp, const char **idp)
 {
        static char ids[][8] = {
                [DEBUG_STACK - 1] = "#DB",
@@ -159,68 +147,32 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
        };
        unsigned k;
 
-       /*
-        * Iterate over all exception stacks, and figure out whether
-        * 'stack' is in one of them:
-        */
        for (k = 0; k < N_EXCEPTION_STACKS; k++) {
                unsigned long end;
 
-               /*
-                * set 'end' to the end of the exception stack.
-                */
                switch (k + 1) {
-               /*
-                * TODO: this block is not needed i think, because
-                * setup64.c:cpu_init() sets up t->ist[DEBUG_STACK]
-                * properly too.
-                */
 #if DEBUG_STKSZ > EXCEPTION_STKSZ
                case DEBUG_STACK:
                        end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
                        break;
 #endif
                default:
-                       end = per_cpu(orig_ist, cpu).ist[k];
+                       end = per_cpu(init_tss, cpu).ist[k];
                        break;
                }
-               /*
-                * Is 'stack' above this exception frame's end?
-                * If yes then skip to the next frame.
-                */
                if (stack >= end)
                        continue;
-               /*
-                * Is 'stack' above this exception frame's start address?
-                * If yes then we found the right frame.
-                */
                if (stack >= end - EXCEPTION_STKSZ) {
-                       /*
-                        * Make sure we only iterate through an exception
-                        * stack once. If it comes up for the second time
-                        * then there's something wrong going on - just
-                        * break out and return NULL:
-                        */
                        if (*usedp & (1U << k))
                                break;
                        *usedp |= 1U << k;
                        *idp = ids[k];
                        return (unsigned long *)end;
                }
-               /*
-                * If this is a debug stack, and if it has a larger size than
-                * the usual exception stacks, then 'stack' might still
-                * be within the lower portion of the debug stack:
-                */
 #if DEBUG_STKSZ > EXCEPTION_STKSZ
                if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
                        unsigned j = N_EXCEPTION_STACKS - 1;
 
-                       /*
-                        * Black magic. A large debug stack is composed of
-                        * multiple exception stack entries, which we
-                        * iterate through now. Dont look:
-                        */
                        do {
                                ++j;
                                end -= EXCEPTION_STKSZ;
@@ -237,25 +189,6 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
        return NULL;
 }
 
-struct ops_and_data {
-       struct stacktrace_ops *ops;
-       void *data;
-};
-
-static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
-{
-       struct ops_and_data *oad = (struct ops_and_data *)context;
-       int n = 0;
-
-       while (unwind(info) == 0 && UNW_PC(info)) {
-               n++;
-               oad->ops->address(oad->data, UNW_PC(info));
-               if (arch_unw_user_mode(info))
-                       break;
-       }
-       return n;
-}
-
 /*
  * x86-64 can have upto three kernel stacks: 
  * process stack
@@ -263,64 +196,25 @@ static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
  */
 
-void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
-               struct stacktrace_ops *ops, void *data)
+void show_trace(unsigned long *stack)
 {
        const unsigned cpu = safe_smp_processor_id();
        unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
+       int i;
        unsigned used = 0;
 
-       if (!tsk)
-               tsk = current;
-
-       if (call_trace >= 0) {
-               int unw_ret = 0;
-               struct unwind_frame_info info;
-               struct ops_and_data oad = { .ops = ops, .data = data };
-
-               if (regs) {
-                       if (unwind_init_frame_info(&info, tsk, regs) == 0)
-                               unw_ret = dump_trace_unwind(&info, &oad);
-               } else if (tsk == current)
-                       unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
-               else {
-                       if (unwind_init_blocked(&info, tsk) == 0)
-                               unw_ret = dump_trace_unwind(&info, &oad);
-               }
-               if (unw_ret > 0) {
-                       if (call_trace == 1 && !arch_unw_user_mode(&info)) {
-                               ops->warning_symbol(data, "DWARF2 unwinder stuck at %s",
-                                            UNW_PC(&info));
-                               if ((long)UNW_SP(&info) < 0) {
-                                       ops->warning(data, "Leftover inexact backtrace:");
-                                       stack = (unsigned long *)UNW_SP(&info);
-                                       if (!stack)
-                                               return;
-                               } else
-                                       ops->warning(data, "Full inexact backtrace again:");
-                       } else if (call_trace >= 1)
-                               return;
-                       else
-                               ops->warning(data, "Full inexact backtrace again:");
-               } else
-                       ops->warning(data, "Inexact backtrace:");
-       }
-       if (!stack) {
-               unsigned long dummy;
-               stack = &dummy;
-               if (tsk && tsk != current)
-                       stack = (unsigned long *)tsk->thread.rsp;
-       }
+       printk("\nCall Trace:");
 
-       /*
-        * Print function call entries within a stack. 'cond' is the
-        * "end of stackframe" condition, that the 'stack++'
-        * iteration will eventually trigger.
-        */
 #define HANDLE_STACK(cond) \
        do while (cond) { \
                unsigned long addr = *stack++; \
                if (kernel_text_address(addr)) { \
+                       if (i > 50) { \
+                               printk("\n       "); \
+                               i = 0; \
+                       } \
+                       else \
+                               i += printk(" "); \
                        /* \
                         * If the address is either in the text segment of the \
                         * kernel, or in the region which contains vmalloc'ed \
@@ -329,31 +223,20 @@ void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * s
                         * down the cause of the crash will be able to figure \
                         * out the call path that was taken. \
                         */ \
-                       ops->address(data, addr);   \
+                       i += printk_address(addr); \
                } \
        } while (0)
 
-       /*
-        * Print function call entries in all stacks, starting at the
-        * current stack address. If the stacks consist of nested
-        * exceptions
-        */
-       for (;;) {
-               char *id;
+       for(i = 11; ; ) {
+               const char *id;
                unsigned long *estack_end;
                estack_end = in_exception_stack(cpu, (unsigned long)stack,
                                                &used, &id);
 
                if (estack_end) {
-                       if (ops->stack(data, id) < 0)
-                               break;
+                       i += printk(" <%s>", id);
                        HANDLE_STACK (stack < estack_end);
-                       ops->stack(data, "<EOE>");
-                       /*
-                        * We link to the next stack via the
-                        * second-to-last pointer (index -2 to end) in the
-                        * exception stack:
-                        */
+                       i += printk(" <EOE>");
                        stack = (unsigned long *) estack_end[-2];
                        continue;
                }
@@ -363,71 +246,23 @@ void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * s
                                (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 
                        if (stack >= irqstack && stack < irqstack_end) {
-                               if (ops->stack(data, "IRQ") < 0)
-                                       break;
+                               i += printk(" <IRQ>");
                                HANDLE_STACK (stack < irqstack_end);
-                               /*
-                                * We link to the next stack (which would be
-                                * the process stack normally) the last
-                                * pointer (index -1 to end) in the IRQ stack:
-                                */
                                stack = (unsigned long *) (irqstack_end[-1]);
                                irqstack_end = NULL;
-                               ops->stack(data, "EOI");
+                               i += printk(" <EOI>");
                                continue;
                        }
                }
                break;
        }
 
-       /*
-        * This handles the process stack:
-        */
        HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 #undef HANDLE_STACK
-}
-EXPORT_SYMBOL(dump_trace);
-
-static void
-print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
-{
-       print_symbol(msg, symbol);
-       printk("\n");
-}
-
-static void print_trace_warning(void *data, char *msg)
-{
-       printk("%s\n", msg);
-}
-
-static int print_trace_stack(void *data, char *name)
-{
-       printk(" <%s> ", name);
-       return 0;
-}
-
-static void print_trace_address(void *data, unsigned long addr)
-{
-       printk_address(addr);
-}
-
-static struct stacktrace_ops print_trace_ops = {
-       .warning = print_trace_warning,
-       .warning_symbol = print_trace_warning_symbol,
-       .stack = print_trace_stack,
-       .address = print_trace_address,
-};
-
-void
-show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
-{
-       printk("\nCall Trace:\n");
-       dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
        printk("\n");
 }
 
-static void
-_show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
+void show_stack(struct task_struct *tsk, unsigned long * rsp)
 {
        unsigned long *stack;
        int i;
@@ -457,16 +292,11 @@ _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
                        break;
                }
                if (i && ((i % 4) == 0))
-                       printk("\n");
-               printk(" %016lx", *stack++);
+                       printk("\n       ");
+               printk("%016lx ", *stack++);
                touch_nmi_watchdog();
        }
-       show_trace(tsk, regs, rsp);
-}
-
-void show_stack(struct task_struct *tsk, unsigned long * rsp)
-{
-       _show_stack(tsk, NULL, rsp);
+       show_trace((unsigned long *)rsp);
 }
 
 /*
@@ -475,7 +305,7 @@ void show_stack(struct task_struct *tsk, unsigned long * rsp)
 void dump_stack(void)
 {
        unsigned long dummy;
-       show_trace(NULL, NULL, &dummy);
+       show_trace(&dummy);
 }
 
 EXPORT_SYMBOL(dump_stack);
@@ -503,15 +333,16 @@ void show_registers(struct pt_regs *regs)
        if (in_kernel) {
 
                printk("Stack: ");
-               _show_stack(NULL, regs, (unsigned long*)rsp);
+               show_stack(NULL, (unsigned long*)rsp);
 
                printk("\nCode: ");
-               if (regs->rip < PAGE_OFFSET)
+               if(regs->rip < PAGE_OFFSET)
                        goto bad;
 
-               for (i=0; i<20; i++) {
+               for(i=0;i<20;i++)
+               {
                        unsigned char c;
-                       if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
+                       if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 bad:
                                printk(" Bad RIP value.");
                                break;
@@ -552,12 +383,10 @@ void out_of_line_bug(void)
 { 
        BUG(); 
 } 
-EXPORT_SYMBOL(out_of_line_bug);
 #endif
 
 static DEFINE_SPINLOCK(die_lock);
 static int die_owner = -1;
-static unsigned int die_nest_count;
 
 unsigned __kprobes long oops_begin(void)
 {
@@ -572,7 +401,6 @@ unsigned __kprobes long oops_begin(void)
                else
                        spin_lock(&die_lock);
        }
-       die_nest_count++;
        die_owner = cpu;
        console_verbose();
        bust_spinlocks(1);
@@ -583,15 +411,9 @@ void __kprobes oops_end(unsigned long flags)
 { 
        die_owner = -1;
        bust_spinlocks(0);
-       die_nest_count--;
-       if (die_nest_count)
-               /* We still own the lock */
-               local_irq_restore(flags);
-       else
-               /* Nest count reaches zero, release the lock. */
-               spin_unlock_irqrestore(&die_lock, flags);
+       spin_unlock_irqrestore(&die_lock, flags);
        if (panic_on_oops)
-               panic("Fatal exception");
+               panic("Oops");
 }
 
 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
@@ -608,9 +430,6 @@ void __kprobes __die(const char * str, struct pt_regs * regs, long err)
        printk("DEBUG_PAGEALLOC");
 #endif
        printk("\n");
-#ifdef CONFIG_SYSFS
-       printk(KERN_ALERT "last sysfs file: %s\n", last_sysfs_file);
-#endif
        notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
        show_registers(regs);
        /* Executive summary in case the oops scrolled away */
@@ -647,8 +466,6 @@ void __kprobes die_nmi(char *str, struct pt_regs *regs)
                panic("nmi watchdog");
        printk("console shuts up ...\n");
        oops_end(flags);
-       nmi_exit();
-       local_irq_enable();
        do_exit(SIGSEGV);
 }
 
@@ -658,6 +475,8 @@ static void __kprobes do_trap(int trapnr, int signr, char *str,
 {
        struct task_struct *tsk = current;
 
+       conditional_sti(regs);
+
        tsk->thread.error_code = error_code;
        tsk->thread.trap_no = trapnr;
 
@@ -666,7 +485,7 @@ static void __kprobes do_trap(int trapnr, int signr, char *str,
                        printk(KERN_INFO
                               "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
                               tsk->comm, tsk->pid, str,
-                              regs->rip, regs->rsp, error_code); 
+                              regs->rip,regs->rsp,error_code); 
 
                if (info)
                        force_sig_info(signr, info, tsk);
@@ -680,9 +499,9 @@ static void __kprobes do_trap(int trapnr, int signr, char *str,
        {            
                const struct exception_table_entry *fixup;
                fixup = search_exception_tables(regs->rip);
-               if (fixup)
+               if (fixup) {
                        regs->rip = fixup->fixup;
-               else    
+               } else  
                        die(str, regs, error_code);
                return;
        }
@@ -694,7 +513,6 @@ asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
        if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
                                                        == NOTIFY_STOP) \
                return; \
-       conditional_sti(regs);                                          \
        do_trap(trapnr, signr, str, regs, error_code, NULL); \
 }
 
@@ -709,7 +527,6 @@ asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
        if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
                                                        == NOTIFY_STOP) \
                return; \
-       conditional_sti(regs);                                          \
        do_trap(trapnr, signr, str, regs, error_code, &info); \
 }
 
@@ -723,17 +540,7 @@ DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 DO_ERROR(18, SIGSEGV, "reserved", reserved)
-
-/* Runs on IST stack */
-asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
-{
-       if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
-                       12, SIGBUS) == NOTIFY_STOP)
-               return;
-       preempt_conditional_sti(regs);
-       do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
-       preempt_conditional_cli(regs);
-}
+DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
 
 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 {
@@ -767,7 +574,7 @@ asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
                        printk(KERN_INFO
                       "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
                               tsk->comm, tsk->pid,
-                              regs->rip, regs->rsp, error_code); 
+                              regs->rip,regs->rsp,error_code); 
 
                force_sig(SIGSEGV, tsk);
                return;
@@ -867,9 +674,8 @@ asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
        if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
                return;
        }
-       preempt_conditional_sti(regs);
        do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
-       preempt_conditional_cli(regs);
+       return;
 }
 
 /* Help handler running on IST stack to switch back to user stack
@@ -1174,29 +980,14 @@ void __init trap_init(void)
 static int __init oops_dummy(char *s)
 { 
        panic_on_oops = 1;
-       return 1;
+       return -1; 
 } 
 __setup("oops=", oops_dummy); 
 
 static int __init kstack_setup(char *s)
 {
        kstack_depth_to_print = simple_strtoul(s,NULL,0);
-       return 1;
+       return 0;
 }
 __setup("kstack=", kstack_setup);
 
-#ifdef CONFIG_STACK_UNWIND
-static int __init call_trace_setup(char *s)
-{
-       if (strcmp(s, "old") == 0)
-               call_trace = -1;
-       else if (strcmp(s, "both") == 0)
-               call_trace = 0;
-       else if (strcmp(s, "newfallback") == 0)
-               call_trace = 1;
-       else if (strcmp(s, "new") == 0)
-               call_trace = 2;
-       return 1;
-}
-__setup("call_trace=", call_trace_setup);
-#endif