Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / kernel / debug / debug_core.c
diff --git a/kernel/kernel/debug/debug_core.c b/kernel/kernel/debug/debug_core.c
new file mode 100644 (file)
index 0000000..0874e2e
--- /dev/null
@@ -0,0 +1,1088 @@
+/*
+ * Kernel Debug Core
+ *
+ * Maintainer: Jason Wessel <jason.wessel@windriver.com>
+ *
+ * Copyright (C) 2000-2001 VERITAS Software Corporation.
+ * Copyright (C) 2002-2004 Timesys Corporation
+ * Copyright (C) 2003-2004 Amit S. Kale <amitkale@linsyssoft.com>
+ * Copyright (C) 2004 Pavel Machek <pavel@ucw.cz>
+ * Copyright (C) 2004-2006 Tom Rini <trini@kernel.crashing.org>
+ * Copyright (C) 2004-2006 LinSysSoft Technologies Pvt. Ltd.
+ * Copyright (C) 2005-2009 Wind River Systems, Inc.
+ * Copyright (C) 2007 MontaVista Software, Inc.
+ * Copyright (C) 2008 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
+ *
+ * Contributors at various stages not listed above:
+ *  Jason Wessel ( jason.wessel@windriver.com )
+ *  George Anzinger <george@mvista.com>
+ *  Anurekh Saxena (anurekh.saxena@timesys.com)
+ *  Lake Stevens Instrument Division (Glenn Engel)
+ *  Jim Kingdon, Cygnus Support.
+ *
+ * Original KGDB stub: David Grothe <dave@gcom.com>,
+ * Tigran Aivazian <tigran@sco.com>
+ *
+ * This file is licensed under the terms of the GNU General Public License
+ * version 2. This program is licensed "as is" without any warranty of any
+ * kind, whether express or implied.
+ */
+
+#define pr_fmt(fmt) "KGDB: " fmt
+
+#include <linux/pid_namespace.h>
+#include <linux/clocksource.h>
+#include <linux/serial_core.h>
+#include <linux/interrupt.h>
+#include <linux/spinlock.h>
+#include <linux/console.h>
+#include <linux/threads.h>
+#include <linux/uaccess.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/ptrace.h>
+#include <linux/string.h>
+#include <linux/delay.h>
+#include <linux/sched.h>
+#include <linux/sysrq.h>
+#include <linux/reboot.h>
+#include <linux/init.h>
+#include <linux/kgdb.h>
+#include <linux/kdb.h>
+#include <linux/pid.h>
+#include <linux/smp.h>
+#include <linux/mm.h>
+#include <linux/vmacache.h>
+#include <linux/rcupdate.h>
+
+#include <asm/cacheflush.h>
+#include <asm/byteorder.h>
+#include <linux/atomic.h>
+
+#include "debug_core.h"
+
+static int kgdb_break_asap;
+
+struct debuggerinfo_struct kgdb_info[NR_CPUS];
+
+/**
+ * kgdb_connected - Is a host GDB connected to us?
+ */
+int                            kgdb_connected;
+EXPORT_SYMBOL_GPL(kgdb_connected);
+
+/* All the KGDB handlers are installed */
+int                    kgdb_io_module_registered;
+
+/* Guard for recursive entry */
+static int                     exception_level;
+
+struct kgdb_io         *dbg_io_ops;
+static DEFINE_SPINLOCK(kgdb_registration_lock);
+
+/* Action for the reboot notifiter, a global allow kdb to change it */
+static int kgdbreboot;
+/* kgdb console driver is loaded */
+static int kgdb_con_registered;
+/* determine if kgdb console output should be used */
+static int kgdb_use_con;
+/* Flag for alternate operations for early debugging */
+bool dbg_is_early = true;
+/* Next cpu to become the master debug core */
+int dbg_switch_cpu;
+
+/* Use kdb or gdbserver mode */
+int dbg_kdb_mode = 1;
+
+static int __init opt_kgdb_con(char *str)
+{
+       kgdb_use_con = 1;
+       return 0;
+}
+
+early_param("kgdbcon", opt_kgdb_con);
+
+module_param(kgdb_use_con, int, 0644);
+module_param(kgdbreboot, int, 0644);
+
+/*
+ * Holds information about breakpoints in a kernel. These breakpoints are
+ * added and removed by gdb.
+ */
+static struct kgdb_bkpt                kgdb_break[KGDB_MAX_BREAKPOINTS] = {
+       [0 ... KGDB_MAX_BREAKPOINTS-1] = { .state = BP_UNDEFINED }
+};
+
+/*
+ * The CPU# of the active CPU, or -1 if none:
+ */
+atomic_t                       kgdb_active = ATOMIC_INIT(-1);
+EXPORT_SYMBOL_GPL(kgdb_active);
+static DEFINE_RAW_SPINLOCK(dbg_master_lock);
+static DEFINE_RAW_SPINLOCK(dbg_slave_lock);
+
+/*
+ * We use NR_CPUs not PERCPU, in case kgdb is used to debug early
+ * bootup code (which might not have percpu set up yet):
+ */
+static atomic_t                        masters_in_kgdb;
+static atomic_t                        slaves_in_kgdb;
+static atomic_t                        kgdb_break_tasklet_var;
+atomic_t                       kgdb_setting_breakpoint;
+
+struct task_struct             *kgdb_usethread;
+struct task_struct             *kgdb_contthread;
+
+int                            kgdb_single_step;
+static pid_t                   kgdb_sstep_pid;
+
+/* to keep track of the CPU which is doing the single stepping*/
+atomic_t                       kgdb_cpu_doing_single_step = ATOMIC_INIT(-1);
+
+/*
+ * If you are debugging a problem where roundup (the collection of
+ * all other CPUs) is a problem [this should be extremely rare],
+ * then use the nokgdbroundup option to avoid roundup. In that case
+ * the other CPUs might interfere with your debugging context, so
+ * use this with care:
+ */
+static int kgdb_do_roundup = 1;
+
+static int __init opt_nokgdbroundup(char *str)
+{
+       kgdb_do_roundup = 0;
+
+       return 0;
+}
+
+early_param("nokgdbroundup", opt_nokgdbroundup);
+
+/*
+ * Finally, some KGDB code :-)
+ */
+
+/*
+ * Weak aliases for breakpoint management,
+ * can be overriden by architectures when needed:
+ */
+int __weak kgdb_arch_set_breakpoint(struct kgdb_bkpt *bpt)
+{
+       int err;
+
+       err = probe_kernel_read(bpt->saved_instr, (char *)bpt->bpt_addr,
+                               BREAK_INSTR_SIZE);
+       if (err)
+               return err;
+       err = probe_kernel_write((char *)bpt->bpt_addr,
+                                arch_kgdb_ops.gdb_bpt_instr, BREAK_INSTR_SIZE);
+       return err;
+}
+
+int __weak kgdb_arch_remove_breakpoint(struct kgdb_bkpt *bpt)
+{
+       return probe_kernel_write((char *)bpt->bpt_addr,
+                                 (char *)bpt->saved_instr, BREAK_INSTR_SIZE);
+}
+
+int __weak kgdb_validate_break_address(unsigned long addr)
+{
+       struct kgdb_bkpt tmp;
+       int err;
+       /* Validate setting the breakpoint and then removing it.  If the
+        * remove fails, the kernel needs to emit a bad message because we
+        * are deep trouble not being able to put things back the way we
+        * found them.
+        */
+       tmp.bpt_addr = addr;
+       err = kgdb_arch_set_breakpoint(&tmp);
+       if (err)
+               return err;
+       err = kgdb_arch_remove_breakpoint(&tmp);
+       if (err)
+               pr_err("Critical breakpoint error, kernel memory destroyed at: %lx\n",
+                      addr);
+       return err;
+}
+
+unsigned long __weak kgdb_arch_pc(int exception, struct pt_regs *regs)
+{
+       return instruction_pointer(regs);
+}
+
+int __weak kgdb_arch_init(void)
+{
+       return 0;
+}
+
+int __weak kgdb_skipexception(int exception, struct pt_regs *regs)
+{
+       return 0;
+}
+
+/*
+ * Some architectures need cache flushes when we set/clear a
+ * breakpoint:
+ */
+static void kgdb_flush_swbreak_addr(unsigned long addr)
+{
+       if (!CACHE_FLUSH_IS_SAFE)
+               return;
+
+       if (current->mm) {
+               int i;
+
+               for (i = 0; i < VMACACHE_SIZE; i++) {
+                       if (!current->vmacache[i])
+                               continue;
+                       flush_cache_range(current->vmacache[i],
+                                         addr, addr + BREAK_INSTR_SIZE);
+               }
+       }
+
+       /* Force flush instruction cache if it was outside the mm */
+       flush_icache_range(addr, addr + BREAK_INSTR_SIZE);
+}
+
+/*
+ * SW breakpoint management:
+ */
+int dbg_activate_sw_breakpoints(void)
+{
+       int error;
+       int ret = 0;
+       int i;
+
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if (kgdb_break[i].state != BP_SET)
+                       continue;
+
+               error = kgdb_arch_set_breakpoint(&kgdb_break[i]);
+               if (error) {
+                       ret = error;
+                       pr_info("BP install failed: %lx\n",
+                               kgdb_break[i].bpt_addr);
+                       continue;
+               }
+
+               kgdb_flush_swbreak_addr(kgdb_break[i].bpt_addr);
+               kgdb_break[i].state = BP_ACTIVE;
+       }
+       return ret;
+}
+
+int dbg_set_sw_break(unsigned long addr)
+{
+       int err = kgdb_validate_break_address(addr);
+       int breakno = -1;
+       int i;
+
+       if (err)
+               return err;
+
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if ((kgdb_break[i].state == BP_SET) &&
+                                       (kgdb_break[i].bpt_addr == addr))
+                       return -EEXIST;
+       }
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if (kgdb_break[i].state == BP_REMOVED &&
+                                       kgdb_break[i].bpt_addr == addr) {
+                       breakno = i;
+                       break;
+               }
+       }
+
+       if (breakno == -1) {
+               for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+                       if (kgdb_break[i].state == BP_UNDEFINED) {
+                               breakno = i;
+                               break;
+                       }
+               }
+       }
+
+       if (breakno == -1)
+               return -E2BIG;
+
+       kgdb_break[breakno].state = BP_SET;
+       kgdb_break[breakno].type = BP_BREAKPOINT;
+       kgdb_break[breakno].bpt_addr = addr;
+
+       return 0;
+}
+
+int dbg_deactivate_sw_breakpoints(void)
+{
+       int error;
+       int ret = 0;
+       int i;
+
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if (kgdb_break[i].state != BP_ACTIVE)
+                       continue;
+               error = kgdb_arch_remove_breakpoint(&kgdb_break[i]);
+               if (error) {
+                       pr_info("BP remove failed: %lx\n",
+                               kgdb_break[i].bpt_addr);
+                       ret = error;
+               }
+
+               kgdb_flush_swbreak_addr(kgdb_break[i].bpt_addr);
+               kgdb_break[i].state = BP_SET;
+       }
+       return ret;
+}
+
+int dbg_remove_sw_break(unsigned long addr)
+{
+       int i;
+
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if ((kgdb_break[i].state == BP_SET) &&
+                               (kgdb_break[i].bpt_addr == addr)) {
+                       kgdb_break[i].state = BP_REMOVED;
+                       return 0;
+               }
+       }
+       return -ENOENT;
+}
+
+int kgdb_isremovedbreak(unsigned long addr)
+{
+       int i;
+
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if ((kgdb_break[i].state == BP_REMOVED) &&
+                                       (kgdb_break[i].bpt_addr == addr))
+                       return 1;
+       }
+       return 0;
+}
+
+int dbg_remove_all_break(void)
+{
+       int error;
+       int i;
+
+       /* Clear memory breakpoints. */
+       for (i = 0; i < KGDB_MAX_BREAKPOINTS; i++) {
+               if (kgdb_break[i].state != BP_ACTIVE)
+                       goto setundefined;
+               error = kgdb_arch_remove_breakpoint(&kgdb_break[i]);
+               if (error)
+                       pr_err("breakpoint remove failed: %lx\n",
+                              kgdb_break[i].bpt_addr);
+setundefined:
+               kgdb_break[i].state = BP_UNDEFINED;
+       }
+
+       /* Clear hardware breakpoints. */
+       if (arch_kgdb_ops.remove_all_hw_break)
+               arch_kgdb_ops.remove_all_hw_break();
+
+       return 0;
+}
+
+/*
+ * Return true if there is a valid kgdb I/O module.  Also if no
+ * debugger is attached a message can be printed to the console about
+ * waiting for the debugger to attach.
+ *
+ * The print_wait argument is only to be true when called from inside
+ * the core kgdb_handle_exception, because it will wait for the
+ * debugger to attach.
+ */
+static int kgdb_io_ready(int print_wait)
+{
+       if (!dbg_io_ops)
+               return 0;
+       if (kgdb_connected)
+               return 1;
+       if (atomic_read(&kgdb_setting_breakpoint))
+               return 1;
+       if (print_wait) {
+#ifdef CONFIG_KGDB_KDB
+               if (!dbg_kdb_mode)
+                       pr_crit("waiting... or $3#33 for KDB\n");
+#else
+               pr_crit("Waiting for remote debugger\n");
+#endif
+       }
+       return 1;
+}
+
+static int kgdb_reenter_check(struct kgdb_state *ks)
+{
+       unsigned long addr;
+
+       if (atomic_read(&kgdb_active) != raw_smp_processor_id())
+               return 0;
+
+       /* Panic on recursive debugger calls: */
+       exception_level++;
+       addr = kgdb_arch_pc(ks->ex_vector, ks->linux_regs);
+       dbg_deactivate_sw_breakpoints();
+
+       /*
+        * If the break point removed ok at the place exception
+        * occurred, try to recover and print a warning to the end
+        * user because the user planted a breakpoint in a place that
+        * KGDB needs in order to function.
+        */
+       if (dbg_remove_sw_break(addr) == 0) {
+               exception_level = 0;
+               kgdb_skipexception(ks->ex_vector, ks->linux_regs);
+               dbg_activate_sw_breakpoints();
+               pr_crit("re-enter error: breakpoint removed %lx\n", addr);
+               WARN_ON_ONCE(1);
+
+               return 1;
+       }
+       dbg_remove_all_break();
+       kgdb_skipexception(ks->ex_vector, ks->linux_regs);
+
+       if (exception_level > 1) {
+               dump_stack();
+               panic("Recursive entry to debugger");
+       }
+
+       pr_crit("re-enter exception: ALL breakpoints killed\n");
+#ifdef CONFIG_KGDB_KDB
+       /* Allow kdb to debug itself one level */
+       return 0;
+#endif
+       dump_stack();
+       panic("Recursive entry to debugger");
+
+       return 1;
+}
+
+static void dbg_touch_watchdogs(void)
+{
+       touch_softlockup_watchdog_sync();
+       clocksource_touch_watchdog();
+       rcu_cpu_stall_reset();
+}
+
+static int kgdb_cpu_enter(struct kgdb_state *ks, struct pt_regs *regs,
+               int exception_state)
+{
+       unsigned long flags;
+       int sstep_tries = 100;
+       int error;
+       int cpu;
+       int trace_on = 0;
+       int online_cpus = num_online_cpus();
+       u64 time_left;
+
+       kgdb_info[ks->cpu].enter_kgdb++;
+       kgdb_info[ks->cpu].exception_state |= exception_state;
+
+       if (exception_state == DCPU_WANT_MASTER)
+               atomic_inc(&masters_in_kgdb);
+       else
+               atomic_inc(&slaves_in_kgdb);
+
+       if (arch_kgdb_ops.disable_hw_break)
+               arch_kgdb_ops.disable_hw_break(regs);
+
+acquirelock:
+       /*
+        * Interrupts will be restored by the 'trap return' code, except when
+        * single stepping.
+        */
+       local_irq_save(flags);
+
+       cpu = ks->cpu;
+       kgdb_info[cpu].debuggerinfo = regs;
+       kgdb_info[cpu].task = current;
+       kgdb_info[cpu].ret_state = 0;
+       kgdb_info[cpu].irq_depth = hardirq_count() >> HARDIRQ_SHIFT;
+
+       /* Make sure the above info reaches the primary CPU */
+       smp_mb();
+
+       if (exception_level == 1) {
+               if (raw_spin_trylock(&dbg_master_lock))
+                       atomic_xchg(&kgdb_active, cpu);
+               goto cpu_master_loop;
+       }
+
+       /*
+        * CPU will loop if it is a slave or request to become a kgdb
+        * master cpu and acquire the kgdb_active lock:
+        */
+       while (1) {
+cpu_loop:
+               if (kgdb_info[cpu].exception_state & DCPU_NEXT_MASTER) {
+                       kgdb_info[cpu].exception_state &= ~DCPU_NEXT_MASTER;
+                       goto cpu_master_loop;
+               } else if (kgdb_info[cpu].exception_state & DCPU_WANT_MASTER) {
+                       if (raw_spin_trylock(&dbg_master_lock)) {
+                               atomic_xchg(&kgdb_active, cpu);
+                               break;
+                       }
+               } else if (kgdb_info[cpu].exception_state & DCPU_IS_SLAVE) {
+                       if (!raw_spin_is_locked(&dbg_slave_lock))
+                               goto return_normal;
+               } else {
+return_normal:
+                       /* Return to normal operation by executing any
+                        * hw breakpoint fixup.
+                        */
+                       if (arch_kgdb_ops.correct_hw_break)
+                               arch_kgdb_ops.correct_hw_break();
+                       if (trace_on)
+                               tracing_on();
+                       kgdb_info[cpu].exception_state &=
+                               ~(DCPU_WANT_MASTER | DCPU_IS_SLAVE);
+                       kgdb_info[cpu].enter_kgdb--;
+                       smp_mb__before_atomic();
+                       atomic_dec(&slaves_in_kgdb);
+                       dbg_touch_watchdogs();
+                       local_irq_restore(flags);
+                       return 0;
+               }
+               cpu_relax();
+       }
+
+       /*
+        * For single stepping, try to only enter on the processor
+        * that was single stepping.  To guard against a deadlock, the
+        * kernel will only try for the value of sstep_tries before
+        * giving up and continuing on.
+        */
+       if (atomic_read(&kgdb_cpu_doing_single_step) != -1 &&
+           (kgdb_info[cpu].task &&
+            kgdb_info[cpu].task->pid != kgdb_sstep_pid) && --sstep_tries) {
+               atomic_set(&kgdb_active, -1);
+               raw_spin_unlock(&dbg_master_lock);
+               dbg_touch_watchdogs();
+               local_irq_restore(flags);
+
+               goto acquirelock;
+       }
+
+       if (!kgdb_io_ready(1)) {
+               kgdb_info[cpu].ret_state = 1;
+               goto kgdb_restore; /* No I/O connection, resume the system */
+       }
+
+       /*
+        * Don't enter if we have hit a removed breakpoint.
+        */
+       if (kgdb_skipexception(ks->ex_vector, ks->linux_regs))
+               goto kgdb_restore;
+
+       /* Call the I/O driver's pre_exception routine */
+       if (dbg_io_ops->pre_exception)
+               dbg_io_ops->pre_exception();
+
+       /*
+        * Get the passive CPU lock which will hold all the non-primary
+        * CPU in a spin state while the debugger is active
+        */
+       if (!kgdb_single_step)
+               raw_spin_lock(&dbg_slave_lock);
+
+#ifdef CONFIG_SMP
+       /* If send_ready set, slaves are already waiting */
+       if (ks->send_ready)
+               atomic_set(ks->send_ready, 1);
+
+       /* Signal the other CPUs to enter kgdb_wait() */
+       else if ((!kgdb_single_step) && kgdb_do_roundup)
+               kgdb_roundup_cpus(flags);
+#endif
+
+       /*
+        * Wait for the other CPUs to be notified and be waiting for us:
+        */
+       time_left = loops_per_jiffy * HZ;
+       while (kgdb_do_roundup && --time_left &&
+              (atomic_read(&masters_in_kgdb) + atomic_read(&slaves_in_kgdb)) !=
+                  online_cpus)
+               cpu_relax();
+       if (!time_left)
+               pr_crit("Timed out waiting for secondary CPUs.\n");
+
+       /*
+        * At this point the primary processor is completely
+        * in the debugger and all secondary CPUs are quiescent
+        */
+       dbg_deactivate_sw_breakpoints();
+       kgdb_single_step = 0;
+       kgdb_contthread = current;
+       exception_level = 0;
+       trace_on = tracing_is_on();
+       if (trace_on)
+               tracing_off();
+
+       while (1) {
+cpu_master_loop:
+               if (dbg_kdb_mode) {
+                       kgdb_connected = 1;
+                       error = kdb_stub(ks);
+                       if (error == -1)
+                               continue;
+                       kgdb_connected = 0;
+               } else {
+                       error = gdb_serial_stub(ks);
+               }
+
+               if (error == DBG_PASS_EVENT) {
+                       dbg_kdb_mode = !dbg_kdb_mode;
+               } else if (error == DBG_SWITCH_CPU_EVENT) {
+                       kgdb_info[dbg_switch_cpu].exception_state |=
+                               DCPU_NEXT_MASTER;
+                       goto cpu_loop;
+               } else {
+                       kgdb_info[cpu].ret_state = error;
+                       break;
+               }
+       }
+
+       /* Call the I/O driver's post_exception routine */
+       if (dbg_io_ops->post_exception)
+               dbg_io_ops->post_exception();
+
+       if (!kgdb_single_step) {
+               raw_spin_unlock(&dbg_slave_lock);
+               /* Wait till all the CPUs have quit from the debugger. */
+               while (kgdb_do_roundup && atomic_read(&slaves_in_kgdb))
+                       cpu_relax();
+       }
+
+kgdb_restore:
+       if (atomic_read(&kgdb_cpu_doing_single_step) != -1) {
+               int sstep_cpu = atomic_read(&kgdb_cpu_doing_single_step);
+               if (kgdb_info[sstep_cpu].task)
+                       kgdb_sstep_pid = kgdb_info[sstep_cpu].task->pid;
+               else
+                       kgdb_sstep_pid = 0;
+       }
+       if (arch_kgdb_ops.correct_hw_break)
+               arch_kgdb_ops.correct_hw_break();
+       if (trace_on)
+               tracing_on();
+
+       kgdb_info[cpu].exception_state &=
+               ~(DCPU_WANT_MASTER | DCPU_IS_SLAVE);
+       kgdb_info[cpu].enter_kgdb--;
+       smp_mb__before_atomic();
+       atomic_dec(&masters_in_kgdb);
+       /* Free kgdb_active */
+       atomic_set(&kgdb_active, -1);
+       raw_spin_unlock(&dbg_master_lock);
+       dbg_touch_watchdogs();
+       local_irq_restore(flags);
+
+       return kgdb_info[cpu].ret_state;
+}
+
+/*
+ * kgdb_handle_exception() - main entry point from a kernel exception
+ *
+ * Locking hierarchy:
+ *     interface locks, if any (begin_session)
+ *     kgdb lock (kgdb_active)
+ */
+int
+kgdb_handle_exception(int evector, int signo, int ecode, struct pt_regs *regs)
+{
+       struct kgdb_state kgdb_var;
+       struct kgdb_state *ks = &kgdb_var;
+       int ret = 0;
+
+       if (arch_kgdb_ops.enable_nmi)
+               arch_kgdb_ops.enable_nmi(0);
+       /*
+        * Avoid entering the debugger if we were triggered due to an oops
+        * but panic_timeout indicates the system should automatically
+        * reboot on panic. We don't want to get stuck waiting for input
+        * on such systems, especially if its "just" an oops.
+        */
+       if (signo != SIGTRAP && panic_timeout)
+               return 1;
+
+       memset(ks, 0, sizeof(struct kgdb_state));
+       ks->cpu                 = raw_smp_processor_id();
+       ks->ex_vector           = evector;
+       ks->signo               = signo;
+       ks->err_code            = ecode;
+       ks->linux_regs          = regs;
+
+       if (kgdb_reenter_check(ks))
+               goto out; /* Ouch, double exception ! */
+       if (kgdb_info[ks->cpu].enter_kgdb != 0)
+               goto out;
+
+       ret = kgdb_cpu_enter(ks, regs, DCPU_WANT_MASTER);
+out:
+       if (arch_kgdb_ops.enable_nmi)
+               arch_kgdb_ops.enable_nmi(1);
+       return ret;
+}
+
+/*
+ * GDB places a breakpoint at this function to know dynamically
+ * loaded objects. It's not defined static so that only one instance with this
+ * name exists in the kernel.
+ */
+
+static int module_event(struct notifier_block *self, unsigned long val,
+       void *data)
+{
+       return 0;
+}
+
+static struct notifier_block dbg_module_load_nb = {
+       .notifier_call  = module_event,
+};
+
+int kgdb_nmicallback(int cpu, void *regs)
+{
+#ifdef CONFIG_SMP
+       struct kgdb_state kgdb_var;
+       struct kgdb_state *ks = &kgdb_var;
+
+       memset(ks, 0, sizeof(struct kgdb_state));
+       ks->cpu                 = cpu;
+       ks->linux_regs          = regs;
+
+       if (kgdb_info[ks->cpu].enter_kgdb == 0 &&
+                       raw_spin_is_locked(&dbg_master_lock)) {
+               kgdb_cpu_enter(ks, regs, DCPU_IS_SLAVE);
+               return 0;
+       }
+#endif
+       return 1;
+}
+
+int kgdb_nmicallin(int cpu, int trapnr, void *regs, int err_code,
+                                                       atomic_t *send_ready)
+{
+#ifdef CONFIG_SMP
+       if (!kgdb_io_ready(0) || !send_ready)
+               return 1;
+
+       if (kgdb_info[cpu].enter_kgdb == 0) {
+               struct kgdb_state kgdb_var;
+               struct kgdb_state *ks = &kgdb_var;
+
+               memset(ks, 0, sizeof(struct kgdb_state));
+               ks->cpu                 = cpu;
+               ks->ex_vector           = trapnr;
+               ks->signo               = SIGTRAP;
+               ks->err_code            = err_code;
+               ks->linux_regs          = regs;
+               ks->send_ready          = send_ready;
+               kgdb_cpu_enter(ks, regs, DCPU_WANT_MASTER);
+               return 0;
+       }
+#endif
+       return 1;
+}
+
+static void kgdb_console_write(struct console *co, const char *s,
+   unsigned count)
+{
+       unsigned long flags;
+
+       /* If we're debugging, or KGDB has not connected, don't try
+        * and print. */
+       if (!kgdb_connected || atomic_read(&kgdb_active) != -1 || dbg_kdb_mode)
+               return;
+
+       local_irq_save(flags);
+       gdbstub_msg_write(s, count);
+       local_irq_restore(flags);
+}
+
+static struct console kgdbcons = {
+       .name           = "kgdb",
+       .write          = kgdb_console_write,
+       .flags          = CON_PRINTBUFFER | CON_ENABLED,
+       .index          = -1,
+};
+
+#ifdef CONFIG_MAGIC_SYSRQ
+static void sysrq_handle_dbg(int key)
+{
+       if (!dbg_io_ops) {
+               pr_crit("ERROR: No KGDB I/O module available\n");
+               return;
+       }
+       if (!kgdb_connected) {
+#ifdef CONFIG_KGDB_KDB
+               if (!dbg_kdb_mode)
+                       pr_crit("KGDB or $3#33 for KDB\n");
+#else
+               pr_crit("Entering KGDB\n");
+#endif
+       }
+
+       kgdb_breakpoint();
+}
+
+static struct sysrq_key_op sysrq_dbg_op = {
+       .handler        = sysrq_handle_dbg,
+       .help_msg       = "debug(g)",
+       .action_msg     = "DEBUG",
+};
+#endif
+
+static int kgdb_panic_event(struct notifier_block *self,
+                           unsigned long val,
+                           void *data)
+{
+       /*
+        * Avoid entering the debugger if we were triggered due to a panic
+        * We don't want to get stuck waiting for input from user in such case.
+        * panic_timeout indicates the system should automatically
+        * reboot on panic.
+        */
+       if (panic_timeout)
+               return NOTIFY_DONE;
+
+       if (dbg_kdb_mode)
+               kdb_printf("PANIC: %s\n", (char *)data);
+       kgdb_breakpoint();
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block kgdb_panic_event_nb = {
+       .notifier_call  = kgdb_panic_event,
+       .priority       = INT_MAX,
+};
+
+void __weak kgdb_arch_late(void)
+{
+}
+
+void __init dbg_late_init(void)
+{
+       dbg_is_early = false;
+       if (kgdb_io_module_registered)
+               kgdb_arch_late();
+       kdb_init(KDB_INIT_FULL);
+}
+
+static int
+dbg_notify_reboot(struct notifier_block *this, unsigned long code, void *x)
+{
+       /*
+        * Take the following action on reboot notify depending on value:
+        *    1 == Enter debugger
+        *    0 == [the default] detatch debug client
+        *   -1 == Do nothing... and use this until the board resets
+        */
+       switch (kgdbreboot) {
+       case 1:
+               kgdb_breakpoint();
+       case -1:
+               goto done;
+       }
+       if (!dbg_kdb_mode)
+               gdbstub_exit(code);
+done:
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block dbg_reboot_notifier = {
+       .notifier_call          = dbg_notify_reboot,
+       .next                   = NULL,
+       .priority               = INT_MAX,
+};
+
+static void kgdb_register_callbacks(void)
+{
+       if (!kgdb_io_module_registered) {
+               kgdb_io_module_registered = 1;
+               kgdb_arch_init();
+               if (!dbg_is_early)
+                       kgdb_arch_late();
+               register_module_notifier(&dbg_module_load_nb);
+               register_reboot_notifier(&dbg_reboot_notifier);
+               atomic_notifier_chain_register(&panic_notifier_list,
+                                              &kgdb_panic_event_nb);
+#ifdef CONFIG_MAGIC_SYSRQ
+               register_sysrq_key('g', &sysrq_dbg_op);
+#endif
+               if (kgdb_use_con && !kgdb_con_registered) {
+                       register_console(&kgdbcons);
+                       kgdb_con_registered = 1;
+               }
+       }
+}
+
+static void kgdb_unregister_callbacks(void)
+{
+       /*
+        * When this routine is called KGDB should unregister from the
+        * panic handler and clean up, making sure it is not handling any
+        * break exceptions at the time.
+        */
+       if (kgdb_io_module_registered) {
+               kgdb_io_module_registered = 0;
+               unregister_reboot_notifier(&dbg_reboot_notifier);
+               unregister_module_notifier(&dbg_module_load_nb);
+               atomic_notifier_chain_unregister(&panic_notifier_list,
+                                              &kgdb_panic_event_nb);
+               kgdb_arch_exit();
+#ifdef CONFIG_MAGIC_SYSRQ
+               unregister_sysrq_key('g', &sysrq_dbg_op);
+#endif
+               if (kgdb_con_registered) {
+                       unregister_console(&kgdbcons);
+                       kgdb_con_registered = 0;
+               }
+       }
+}
+
+/*
+ * There are times a tasklet needs to be used vs a compiled in
+ * break point so as to cause an exception outside a kgdb I/O module,
+ * such as is the case with kgdboe, where calling a breakpoint in the
+ * I/O driver itself would be fatal.
+ */
+static void kgdb_tasklet_bpt(unsigned long ing)
+{
+       kgdb_breakpoint();
+       atomic_set(&kgdb_break_tasklet_var, 0);
+}
+
+static DECLARE_TASKLET(kgdb_tasklet_breakpoint, kgdb_tasklet_bpt, 0);
+
+void kgdb_schedule_breakpoint(void)
+{
+       if (atomic_read(&kgdb_break_tasklet_var) ||
+               atomic_read(&kgdb_active) != -1 ||
+               atomic_read(&kgdb_setting_breakpoint))
+               return;
+       atomic_inc(&kgdb_break_tasklet_var);
+       tasklet_schedule(&kgdb_tasklet_breakpoint);
+}
+EXPORT_SYMBOL_GPL(kgdb_schedule_breakpoint);
+
+static void kgdb_initial_breakpoint(void)
+{
+       kgdb_break_asap = 0;
+
+       pr_crit("Waiting for connection from remote gdb...\n");
+       kgdb_breakpoint();
+}
+
+/**
+ *     kgdb_register_io_module - register KGDB IO module
+ *     @new_dbg_io_ops: the io ops vector
+ *
+ *     Register it with the KGDB core.
+ */
+int kgdb_register_io_module(struct kgdb_io *new_dbg_io_ops)
+{
+       int err;
+
+       spin_lock(&kgdb_registration_lock);
+
+       if (dbg_io_ops) {
+               spin_unlock(&kgdb_registration_lock);
+
+               pr_err("Another I/O driver is already registered with KGDB\n");
+               return -EBUSY;
+       }
+
+       if (new_dbg_io_ops->init) {
+               err = new_dbg_io_ops->init();
+               if (err) {
+                       spin_unlock(&kgdb_registration_lock);
+                       return err;
+               }
+       }
+
+       dbg_io_ops = new_dbg_io_ops;
+
+       spin_unlock(&kgdb_registration_lock);
+
+       pr_info("Registered I/O driver %s\n", new_dbg_io_ops->name);
+
+       /* Arm KGDB now. */
+       kgdb_register_callbacks();
+
+       if (kgdb_break_asap)
+               kgdb_initial_breakpoint();
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(kgdb_register_io_module);
+
+/**
+ *     kkgdb_unregister_io_module - unregister KGDB IO module
+ *     @old_dbg_io_ops: the io ops vector
+ *
+ *     Unregister it with the KGDB core.
+ */
+void kgdb_unregister_io_module(struct kgdb_io *old_dbg_io_ops)
+{
+       BUG_ON(kgdb_connected);
+
+       /*
+        * KGDB is no longer able to communicate out, so
+        * unregister our callbacks and reset state.
+        */
+       kgdb_unregister_callbacks();
+
+       spin_lock(&kgdb_registration_lock);
+
+       WARN_ON_ONCE(dbg_io_ops != old_dbg_io_ops);
+       dbg_io_ops = NULL;
+
+       spin_unlock(&kgdb_registration_lock);
+
+       pr_info("Unregistered I/O driver %s, debugger disabled\n",
+               old_dbg_io_ops->name);
+}
+EXPORT_SYMBOL_GPL(kgdb_unregister_io_module);
+
+int dbg_io_get_char(void)
+{
+       int ret = dbg_io_ops->read_char();
+       if (ret == NO_POLL_CHAR)
+               return -1;
+       if (!dbg_kdb_mode)
+               return ret;
+       if (ret == 127)
+               return 8;
+       return ret;
+}
+
+/**
+ * kgdb_breakpoint - generate breakpoint exception
+ *
+ * This function will generate a breakpoint exception.  It is used at the
+ * beginning of a program to sync up with a debugger and can be used
+ * otherwise as a quick means to stop program execution and "break" into
+ * the debugger.
+ */
+noinline void kgdb_breakpoint(void)
+{
+       atomic_inc(&kgdb_setting_breakpoint);
+       wmb(); /* Sync point before breakpoint */
+       arch_kgdb_breakpoint();
+       wmb(); /* Sync point after breakpoint */
+       atomic_dec(&kgdb_setting_breakpoint);
+}
+EXPORT_SYMBOL_GPL(kgdb_breakpoint);
+
+static int __init opt_kgdb_wait(char *str)
+{
+       kgdb_break_asap = 1;
+
+       kdb_init(KDB_INIT_EARLY);
+       if (kgdb_io_module_registered)
+               kgdb_initial_breakpoint();
+
+       return 0;
+}
+
+early_param("kgdbwait", opt_kgdb_wait);