[PATCH] Calgary IOMMU: fix error path memleak in calgary_free_tar
[safe/jmp/linux-2.6] / arch / x86_64 / kernel / traps.c
index 05dbbf4..56d7ff0 100644 (file)
@@ -6,15 +6,12 @@
  *
  *  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>
@@ -30,6 +27,8 @@
 #include <linux/moduleparam.h>
 #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>
 
-extern struct gate_struct idt_table[256]; 
-
 asmlinkage void divide_error(void);
 asmlinkage void debug(void);
 asmlinkage void nmi(void);
@@ -71,20 +68,21 @@ asmlinkage void alignment_check(void);
 asmlinkage void machine_check(void);
 asmlinkage void spurious_interrupt_bug(void);
 
-struct notifier_block *die_chain;
-static DEFINE_SPINLOCK(die_notifier_lock);
+ATOMIC_NOTIFIER_HEAD(die_chain);
+EXPORT_SYMBOL(die_chain);
 
 int register_die_notifier(struct notifier_block *nb)
 {
-       int err = 0;
-       unsigned long flags;
-
        vmalloc_sync_all();
-       spin_lock_irqsave(&die_notifier_lock, flags);
-       err = notifier_chain_register(&die_chain, nb);
-       spin_unlock_irqrestore(&die_notifier_lock, flags);
-       return err;
+       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);
+}
+EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
 
 static inline void conditional_sti(struct pt_regs *regs)
 {
@@ -103,34 +101,44 @@ 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 = 10;
+static int kstack_depth_to_print = 12;
+#ifdef CONFIG_STACK_UNWIND
+static int call_trace = 1;
+#else
+#define call_trace (-1)
+#endif
 
 #ifdef CONFIG_KALLSYMS
-#include <linux/kallsyms.h> 
-int printk_address(unsigned long address)
-{ 
+# include <linux/kallsyms.h>
+void 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) 
-               return printk("[<%016lx>]", address);
-       if (!modname) 
+       symname = kallsyms_lookup(address, &symsize, &offset,
+                                       &modname, namebuf);
+       if (!symname) {
+               printk(" [<%016lx>]\n", address);
+               return;
+       }
+       if (!modname)
                modname = delim = "";           
-        return printk("<%016lx>{%s%s%s%s%+ld}",
-                     address, delim, modname, delim, symname, offset); 
-} 
+       printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
+               address, delim, modname, delim, symname, offset, symsize);
+}
 #else
-int printk_address(unsigned long address)
-{ 
-       return printk("[<%016lx>]", address);
-} 
+void printk_address(unsigned long address)
+{
+       printk(" [<%016lx>]\n", address);
+}
 #endif
 
 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
@@ -148,32 +156,68 @@ 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(init_tss, cpu).ist[k];
+                       end = per_cpu(orig_ist, 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;
@@ -190,6 +234,19 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
        return NULL;
 }
 
+static int show_trace_unwind(struct unwind_frame_info *info, void *context)
+{
+       int n = 0;
+
+       while (unwind(info) == 0 && UNW_PC(info)) {
+               n++;
+               printk_address(UNW_PC(info));
+               if (arch_unw_user_mode(info))
+                       break;
+       }
+       return n;
+}
+
 /*
  * x86-64 can have upto three kernel stacks: 
  * process stack
@@ -197,25 +254,56 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
  */
 
-void show_trace(unsigned long *stack)
+void show_trace(struct task_struct *tsk, struct pt_regs *regs, 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;
 
-       printk("\nCall Trace:");
+       printk("\nCall Trace:\n");
 
+       if (!tsk)
+               tsk = current;
+
+       if (call_trace >= 0) {
+               int unw_ret = 0;
+               struct unwind_frame_info info;
+
+               if (regs) {
+                       if (unwind_init_frame_info(&info, tsk, regs) == 0)
+                               unw_ret = show_trace_unwind(&info, NULL);
+               } else if (tsk == current)
+                       unw_ret = unwind_init_running(&info, show_trace_unwind, NULL);
+               else {
+                       if (unwind_init_blocked(&info, tsk) == 0)
+                               unw_ret = show_trace_unwind(&info, NULL);
+               }
+               if (unw_ret > 0) {
+                       if (call_trace == 1 && !arch_unw_user_mode(&info)) {
+                               print_symbol("DWARF2 unwinder stuck at %s\n",
+                                            UNW_PC(&info));
+                               if ((long)UNW_SP(&info) < 0) {
+                                       printk("Leftover inexact backtrace:\n");
+                                       stack = (unsigned long *)UNW_SP(&info);
+                               } else
+                                       printk("Full inexact backtrace again:\n");
+                       } else if (call_trace >= 1)
+                               return;
+                       else
+                               printk("Full inexact backtrace again:\n");
+               } else
+                       printk("Inexact backtrace:\n");
+       }
+
+       /*
+        * 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 \
@@ -224,20 +312,30 @@ void show_trace(unsigned long *stack)
                         * down the cause of the crash will be able to figure \
                         * out the call path that was taken. \
                         */ \
-                       i += printk_address(addr); \
+                       printk_address(addr); \
                } \
        } while (0)
 
-       for(i = 11; ; ) {
+       /*
+        * Print function call entries in all stacks, starting at the
+        * current stack address. If the stacks consist of nested
+        * exceptions
+        */
+       for ( ; ; ) {
                const char *id;
                unsigned long *estack_end;
                estack_end = in_exception_stack(cpu, (unsigned long)stack,
                                                &used, &id);
 
                if (estack_end) {
-                       i += printk(" <%s>", id);
+                       printk(" <%s>", id);
                        HANDLE_STACK (stack < estack_end);
-                       i += printk(" <EOE>");
+                       printk(" <EOE>");
+                       /*
+                        * We link to the next stack via the
+                        * second-to-last pointer (index -2 to end) in the
+                        * exception stack:
+                        */
                        stack = (unsigned long *) estack_end[-2];
                        continue;
                }
@@ -247,23 +345,32 @@ void show_trace(unsigned long *stack)
                                (IRQSTACKSIZE - 64) / sizeof(*irqstack);
 
                        if (stack >= irqstack && stack < irqstack_end) {
-                               i += printk(" <IRQ>");
+                               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;
-                               i += printk(" <EOI>");
+                               printk(" <EOI>");
                                continue;
                        }
                }
                break;
        }
 
+       /*
+        * This prints the process stack:
+        */
        HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
 #undef HANDLE_STACK
+
        printk("\n");
 }
 
-void show_stack(struct task_struct *tsk, unsigned long * rsp)
+static void _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long * rsp)
 {
        unsigned long *stack;
        int i;
@@ -293,11 +400,16 @@ void show_stack(struct task_struct *tsk, unsigned long * rsp)
                        break;
                }
                if (i && ((i % 4) == 0))
-                       printk("\n       ");
-               printk("%016lx ", *stack++);
+                       printk("\n");
+               printk(" %016lx", *stack++);
                touch_nmi_watchdog();
        }
-       show_trace((unsigned long *)rsp);
+       show_trace(tsk, regs, rsp);
+}
+
+void show_stack(struct task_struct *tsk, unsigned long * rsp)
+{
+       _show_stack(tsk, NULL, rsp);
 }
 
 /*
@@ -306,7 +418,7 @@ void show_stack(struct task_struct *tsk, unsigned long * rsp)
 void dump_stack(void)
 {
        unsigned long dummy;
-       show_trace(&dummy);
+       show_trace(NULL, NULL, &dummy);
 }
 
 EXPORT_SYMBOL(dump_stack);
@@ -333,7 +445,7 @@ void show_registers(struct pt_regs *regs)
        if (in_kernel) {
 
                printk("Stack: ");
-               show_stack(NULL, (unsigned long*)rsp);
+               _show_stack(NULL, regs, (unsigned long*)rsp);
 
                printk("\nCode: ");
                if (regs->rip < PAGE_OFFSET)
@@ -382,10 +494,12 @@ 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)
 {
@@ -400,6 +514,7 @@ unsigned __kprobes long oops_begin(void)
                else
                        spin_lock(&die_lock);
        }
+       die_nest_count++;
        die_owner = cpu;
        console_verbose();
        bust_spinlocks(1);
@@ -410,9 +525,15 @@ void __kprobes oops_end(unsigned long flags)
 { 
        die_owner = -1;
        bust_spinlocks(0);
-       spin_unlock_irqrestore(&die_lock, flags);
+       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);
        if (panic_on_oops)
-               panic("Oops");
+               panic("Fatal exception");
 }
 
 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
@@ -435,6 +556,8 @@ void __kprobes __die(const char * str, struct pt_regs * regs, long err)
        printk(KERN_ALERT "RIP ");
        printk_address(regs->rip); 
        printk(" RSP <%016lx>\n", regs->rsp); 
+       if (kexec_should_crash(current))
+               crash_kexec(regs);
 }
 
 void die(const char * str, struct pt_regs * regs, long err)
@@ -447,7 +570,7 @@ void die(const char * str, struct pt_regs * regs, long err)
        do_exit(SIGSEGV); 
 }
 
-void __kprobes die_nmi(char *str, struct pt_regs *regs)
+void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
 {
        unsigned long flags = oops_begin();
 
@@ -457,10 +580,13 @@ void __kprobes die_nmi(char *str, struct pt_regs *regs)
         */
        printk(str, safe_smp_processor_id());
        show_registers(regs);
-       if (panic_on_timeout || panic_on_oops)
-               panic("nmi watchdog");
-       printk("console shuts up ...\n");
+       if (kexec_should_crash(current))
+               crash_kexec(regs);
+       if (do_panic || panic_on_oops)
+               panic("Non maskable interrupt");
        oops_end(flags);
+       nmi_exit();
+       local_irq_enable();
        do_exit(SIGSEGV);
 }
 
@@ -470,8 +596,6 @@ 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;
 
@@ -508,6 +632,7 @@ 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); \
 }
 
@@ -522,6 +647,7 @@ 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); \
 }
 
@@ -535,7 +661,17 @@ 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)
-DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
+
+/* 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);
+}
 
 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
 {
@@ -593,8 +729,15 @@ asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
 static __kprobes 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 for unknown reason %02x.\n",
+               reason);
+       printk(KERN_EMERG "You probably have a hardware problem with your "
+               "RAM chips\n");
+
+       if (panic_on_unrecovered_nmi)
+               panic("NMI: Not continuing");
+
+       printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 
        /* Clear and disable the memory parity error line. */
        reason = (reason & 0xf) | 4;
@@ -617,9 +760,15 @@ io_check_error(unsigned char reason, struct pt_regs * regs)
 
 static __kprobes void
 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
-{      printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
-       printk("Dazed and confused, but trying to continue\n");
-       printk("Do you have a strange power saving mode enabled?\n");
+{
+       printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
+               reason);
+       printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
+
+       if (panic_on_unrecovered_nmi)
+               panic("NMI: Not continuing");
+
+       printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 }
 
 /* Runs on IST stack. This code must keep interrupts off all the time.
@@ -639,17 +788,15 @@ asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
                if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
                                                                == NOTIFY_STOP)
                        return;
-#ifdef CONFIG_X86_LOCAL_APIC
                /*
                 * Ok, so this is none of the documented NMI sources,
                 * so it must be the NMI watchdog.
                 */
-               if (nmi_watchdog > 0) {
-                       nmi_watchdog_tick(regs,reason);
+               if (nmi_watchdog_tick(regs,reason))
                        return;
-               }
-#endif
-               unknown_nmi_error(reason, regs);
+               if (!do_nmi_callback(regs,cpu))
+                       unknown_nmi_error(reason, regs);
+
                return;
        }
        if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
@@ -669,8 +816,9 @@ 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);
-       return;
+       preempt_conditional_cli(regs);
 }
 
 /* Help handler running on IST stack to switch back to user stack
@@ -975,14 +1123,29 @@ 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 0;
+       return 1;
 }
 __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