*
  * 2005-08-12 Keith Owens <kaos@sgi.com>
  *           Convert MCA/INIT handlers to use per event stacks and SAL/OS state.
+ *
+ * 2005-10-07 Keith Owens <kaos@sgi.com>
+ *           Add notify_die() hooks.
  */
 #include <linux/config.h>
 #include <linux/types.h>
 #include <linux/sched.h>
 #include <linux/interrupt.h>
 #include <linux/irq.h>
-#include <linux/kallsyms.h>
 #include <linux/smp_lock.h>
 #include <linux/bootmem.h>
 #include <linux/acpi.h>
 #include <linux/workqueue.h>
 
 #include <asm/delay.h>
+#include <asm/kdebug.h>
 #include <asm/machvec.h>
 #include <asm/meminit.h>
 #include <asm/page.h>
 
 static int mca_init;
 
+
+static void inline
+ia64_mca_spin(const char *func)
+{
+       printk(KERN_EMERG "%s: spinning here, not returning to SAL\n", func);
+       while (1)
+               cpu_relax();
+}
 /*
  * IA64_MCA log support
  */
  *  Outputs :   None
  */
 static irqreturn_t
-ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs)
+ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *regs)
 {
        unsigned long flags;
        int cpu = smp_processor_id();
 
        /* Mask all interrupts */
        local_irq_save(flags);
+       if (notify_die(DIE_MCA_RENDZVOUS_ENTER, "MCA", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
 
        ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE;
        /* Register with the SAL monarch that the slave has
         */
        ia64_sal_mc_rendez();
 
+       if (notify_die(DIE_MCA_RENDZVOUS_PROCESS, "MCA", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
+
        /* Wait for the monarch cpu to exit. */
        while (monarch_cpu != -1)
               cpu_relax();     /* spin until monarch leaves */
 
+       if (notify_die(DIE_MCA_RENDZVOUS_LEAVE, "MCA", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
+
        /* Enable all interrupts */
        local_irq_restore(flags);
        return IRQ_HANDLED;
        oops_in_progress = 1;   /* FIXME: make printk NMI/MCA/INIT safe */
        previous_current = ia64_mca_modify_original_stack(regs, sw, sos, "MCA");
        monarch_cpu = cpu;
+       if (notify_die(DIE_MCA_MONARCH_ENTER, "MCA", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
        ia64_wait_for_slaves(cpu);
 
        /* Wakeup all the processors which are spinning in the rendezvous loop.
         * spinning in SAL does not work.
         */
        ia64_mca_wakeup_all();
+       if (notify_die(DIE_MCA_MONARCH_PROCESS, "MCA", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
 
        /* Get the MCA error record and log it */
        ia64_mca_log_sal_error_record(SAL_INFO_TYPE_MCA);
                ia64_sal_clear_state_info(SAL_INFO_TYPE_MCA);
                sos->os_status = IA64_MCA_CORRECTED;
        }
+       if (notify_die(DIE_MCA_MONARCH_LEAVE, "MCA", regs, 0, 0, recover)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
 
        set_curr_task(cpu, previous_current);
        monarch_cpu = -1;
 
 #endif /* CONFIG_ACPI */
 
+static int
+default_monarch_init_process(struct notifier_block *self, unsigned long val, void *data)
+{
+       int c;
+       struct task_struct *g, *t;
+       if (val != DIE_INIT_MONARCH_PROCESS)
+               return NOTIFY_DONE;
+       printk(KERN_ERR "Processes interrupted by INIT -");
+       for_each_online_cpu(c) {
+               struct ia64_sal_os_state *s;
+               t = __va(__per_cpu_mca[c] + IA64_MCA_CPU_INIT_STACK_OFFSET);
+               s = (struct ia64_sal_os_state *)((char *)t + MCA_SOS_OFFSET);
+               g = s->prev_task;
+               if (g) {
+                       if (g->pid)
+                               printk(" %d", g->pid);
+                       else
+                               printk(" %d (cpu %d task 0x%p)", g->pid, task_cpu(g), g);
+               }
+       }
+       printk("\n\n");
+       if (read_trylock(&tasklist_lock)) {
+               do_each_thread (g, t) {
+                       printk("\nBacktrace of pid %d (%s)\n", t->pid, t->comm);
+                       show_stack(t, NULL);
+               } while_each_thread (g, t);
+               read_unlock(&tasklist_lock);
+       }
+       return NOTIFY_DONE;
+}
+
 /*
  * C portion of the OS INIT handler
  *
        static atomic_t slaves;
        static atomic_t monarchs;
        task_t *previous_current;
-       int cpu = smp_processor_id(), c;
-       struct task_struct *g, *t;
+       int cpu = smp_processor_id();
 
        oops_in_progress = 1;   /* FIXME: make printk NMI/MCA/INIT safe */
        console_loglevel = 15;  /* make sure printks make it to console */
                ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_INIT;
                while (monarch_cpu == -1)
                       cpu_relax();     /* spin until monarch enters */
+               if (notify_die(DIE_INIT_SLAVE_ENTER, "INIT", regs, 0, 0, 0)
+                               == NOTIFY_STOP)
+                       ia64_mca_spin(__FUNCTION__);
+               if (notify_die(DIE_INIT_SLAVE_PROCESS, "INIT", regs, 0, 0, 0)
+                               == NOTIFY_STOP)
+                       ia64_mca_spin(__FUNCTION__);
                while (monarch_cpu != -1)
                       cpu_relax();     /* spin until monarch leaves */
+               if (notify_die(DIE_INIT_SLAVE_LEAVE, "INIT", regs, 0, 0, 0)
+                               == NOTIFY_STOP)
+                       ia64_mca_spin(__FUNCTION__);
                printk("Slave on cpu %d returning to normal service.\n", cpu);
                set_curr_task(cpu, previous_current);
                ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
        }
 
        monarch_cpu = cpu;
+       if (notify_die(DIE_INIT_MONARCH_ENTER, "INIT", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
 
        /*
         * Wait for a bit.  On some machines (e.g., HP's zx2000 and zx6000, INIT can be
        printk("Delaying for 5 seconds...\n");
        udelay(5*1000000);
        ia64_wait_for_slaves(cpu);
-       printk(KERN_ERR "Processes interrupted by INIT -");
-       for_each_online_cpu(c) {
-               struct ia64_sal_os_state *s;
-               t = __va(__per_cpu_mca[c] + IA64_MCA_CPU_INIT_STACK_OFFSET);
-               s = (struct ia64_sal_os_state *)((char *)t + MCA_SOS_OFFSET);
-               g = s->prev_task;
-               if (g) {
-                       if (g->pid)
-                               printk(" %d", g->pid);
-                       else
-                               printk(" %d (cpu %d task 0x%p)", g->pid, task_cpu(g), g);
-               }
-       }
-       printk("\n\n");
-       if (read_trylock(&tasklist_lock)) {
-               do_each_thread (g, t) {
-                       printk("\nBacktrace of pid %d (%s)\n", t->pid, t->comm);
-                       show_stack(t, NULL);
-               } while_each_thread (g, t);
-               read_unlock(&tasklist_lock);
-       }
+       /* If nobody intercepts DIE_INIT_MONARCH_PROCESS then we drop through
+        * to default_monarch_init_process() above and just print all the
+        * tasks.
+        */
+       if (notify_die(DIE_INIT_MONARCH_PROCESS, "INIT", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
+       if (notify_die(DIE_INIT_MONARCH_LEAVE, "INIT", regs, 0, 0, 0)
+                       == NOTIFY_STOP)
+               ia64_mca_spin(__FUNCTION__);
        printk("\nINIT dump complete.  Monarch on cpu %d returning to normal service.\n", cpu);
        atomic_dec(&monarchs);
        set_curr_task(cpu, previous_current);
        s64 rc;
        struct ia64_sal_retval isrv;
        u64 timeout = IA64_MCA_RENDEZ_TIMEOUT;  /* platform specific */
+       static struct notifier_block default_init_monarch_nb = {
+               .notifier_call = default_monarch_init_process,
+               .priority = 0/* we need to notified last */
+       };
 
        IA64_MCA_DEBUG("%s: begin\n", __FUNCTION__);
 
                       "(status %ld)\n", rc);
                return;
        }
+       if (register_die_notifier(&default_init_monarch_nb)) {
+               printk(KERN_ERR "Failed to register default monarch INIT process\n");
+               return;
+       }
 
        IA64_MCA_DEBUG("%s: registered OS INIT handler with SAL\n", __FUNCTION__);
 
 
 EXPORT_SYMBOL(fpswa_interface);
 
 struct notifier_block *ia64die_chain;
-static DEFINE_SPINLOCK(die_notifier_lock);
 
-int register_die_notifier(struct notifier_block *nb)
+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(&ia64die_chain, nb);
-       spin_unlock_irqrestore(&die_notifier_lock, flags);
-       return err;
+       return notifier_chain_register(&ia64die_chain, nb);
 }
+EXPORT_SYMBOL_GPL(register_die_notifier);
+
+int
+unregister_die_notifier(struct notifier_block *nb)
+{
+       return notifier_chain_unregister(&ia64die_chain, nb);
+}
+EXPORT_SYMBOL_GPL(unregister_die_notifier);
 
 void __init
 trap_init (void)
        if (++die.lock_owner_depth < 3) {
                printk("%s[%d]: %s %ld [%d]\n",
                        current->comm, current->pid, str, err, ++die_counter);
+               (void) notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
                show_regs(regs);
        } else
                printk(KERN_ERR "Recursive die() failure, output suppressed\n");
        switch (break_num) {
              case 0: /* unknown error (used by GCC for __builtin_abort()) */
                if (notify_die(DIE_BREAK, "break 0", regs, break_num, TRAP_BRKPT, SIGTRAP)
-                               == NOTIFY_STOP) {
+                               == NOTIFY_STOP)
                        return;
-               }
                die_if_kernel("bugcheck!", regs, break_num);
                sig = SIGILL; code = ILL_ILLOPC;
                break;
                sig = SIGILL; code = __ILL_BNDMOD;
                break;
 
-             case 0x80200:
-             case 0x80300:
-               if (notify_die(DIE_BREAK, "kprobe", regs, break_num, TRAP_BRKPT, SIGTRAP)
-                               == NOTIFY_STOP) {
-                       return;
-               }
-               sig = SIGTRAP; code = TRAP_BRKPT;
-               break;
-
              default:
                if (break_num < 0x40000 || break_num > 0x100000)
                        die_if_kernel("Bad break", regs, break_num);
                if (break_num < 0x80000) {
                        sig = SIGILL; code = __ILL_BREAK;
                } else {
+                       if (notify_die(DIE_BREAK, "bad break", regs, break_num, TRAP_BRKPT, SIGTRAP)
+                                       == NOTIFY_STOP)
+                               return;
                        sig = SIGTRAP; code = TRAP_BRKPT;
                }
        }
 #endif
                        break;
                      case 35: siginfo.si_code = TRAP_BRANCH; ifa = 0; break;
-                     case 36:
-                             if (notify_die(DIE_SS, "ss", ®s, vector,
-                                            vector, SIGTRAP) == NOTIFY_STOP)
-                                     return;
-                             siginfo.si_code = TRAP_TRACE; ifa = 0; break;
+                     case 36: siginfo.si_code = TRAP_TRACE; ifa = 0; break;
                }
+               if (notify_die(DIE_FAULT, "ia64_fault", ®s, vector, siginfo.si_code, SIGTRAP)
+                               == NOTIFY_STOP)
+                       return;
                siginfo.si_signo = SIGTRAP;
                siginfo.si_errno = 0;
                siginfo.si_addr  = (void __user *) ifa;