> The ia64 would need to "jump through the hoop" of a call to
> unw_init_running() before it calls HYPERVISOR_shutdown()

Is the attached patch what you want?
 

Not quite -- it leaves out the most important part.  Your
patch will cause a switch_stack register dump to be laid down,
but the address of the switch_stack needs to be stored in the
task_struct->thread.ksp of the panicking task:
 

diff -r 114c8cc94466 -r 262eb4ba1212 linux-2.6-xen-sparse/arch/ia64/kernel/setup.c
--- a/linux-2.6-xen-sparse/arch/ia64/kernel/setup.c     Thu Dec 21 17:47:31 2006 +0900
+++ b/linux-2.6-xen-sparse/arch/ia64/kernel/setup.c     Fri Dec 22 12:03:18 2006 +0900
@@ -78,10 +78,17 @@ EXPORT_SYMBOL(__per_cpu_offset);
 #endif
 
 #ifdef CONFIG_XEN
+static void
+xen_panic_hypercall(struct unw_frame_info *info, void *arg)
+{
+       HYPERVISOR_shutdown(SHUTDOWN_crash);
+       /* we're never actually going to get here... */
+}
+
 static int
 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
 {
-       HYPERVISOR_shutdown(SHUTDOWN_crash);
+        unw_init_running(xen_panic_hypercall, NULL);
        /* we're never actually going to get here... */
        return NOTIFY_DONE;
 }

So the xen_panic_hypercall() should look like this:

static void
xen_panic_hypercall(struct unw_frame_info *info, void *arg)
{
       current->thread.ksp = (__u64)info->sw - 16;
       HYPERVISOR_shutdown(SHUTDOWN_crash);
       /* we're never actually going to get here... */
}

Doing the store to the current->thread.ksp as above mimics
the behavior of all of the other tasks on the system.
 
Thanks,
  Dave