gdbsx: gdbserver for xen


Welcome to gdbsx. gdbsx is a gdbserver program to debug guest kernels and 
kernel modules. It runs on dom0 running on xen hypervisor and allows debug 
of 32 or 64bit PV or HVM elf guest binaries. It can also be run standalone, 
without remote gdb, to dump context of any/all VCPUs of any guest.

It is divided in two parts, gx and xg. The former interacts with remote gdb,
while latter interacts with xen and exports public APIs that can be used to 
create a plug in for any other debugger or binary type.


USAGE:
   USAGE 1:
   - dom0> gdbsx -c 1 64 : displays VCPU contexts for 64bit guest with domid 1

   USAGE 2:
   - dom0> gdbsx -a 2 64 9999
             connects to a 64bit guest with domid 2 and waits for gdb connection
   - now, connect to the above gdbsx from a remote system or dom0 as: 
      bash> gdb ./vmlinux             (exact matching vmlinux of guest kernel)
      (gdb) target remote dom0:9999 

   - Additionally, to debug 32bit loadable kernel modules, please do following:
      (gdb) p init_mm.pgd[3]
      $1 = {pgd = 0x1b874f027}
      (gdb) monitor pgd3 0x1b874f027  (Make sure value is in HEX)
      pgd3val set to: 0x1b874f027

   - use gdb as normal, breakpoints, single step, etc...
   - when need to break into gdb, instead of ctrl-c, just do "xm pause <domid>"
     on dom0 to pause the guest. this will break into gdb right away.
   - detach/quit from gdb (leave gdbsx alone) to gracefully exit.
   - if ctrl-c or core-dumped, make sure to do xm unpause if guest still paused.

   - multiple vcpus: 
	 o  gdb>set scheduler-locking on   : for single step of correct vcpu.

         o  since gdb is not kernel debugger, vcpus are emulated via threads
            Thus, gdb>info threads : will show all vcpus. Then, switch thread 
	    to get to another vcpu, etc...  Remember, gdb has it's own [thread]
	    id, off by 1. 

   - See below for some useful gdb macros. Please email me if you've more.


NOTES:
   - For now, it is not possible to run gdbsx on a guest and gdb inside 
     the same guest at the same time.
   - To report problems, please run gdbsx with -d and collect output.
   - VCPU offlining is not supported. Thus [0-NUMVCPUs] are all assumed active.

TIPS:
   - make sure firewall is disabled on dom0 if running gdb on a different host.
   - Must be at least gdb version 6.5-16.x to debug el5 kernels.


Mukesh Rathor
Oracle Corporation,
Redwood Shores,  CA  USA
mukesh[dot]rathor[at]oracle[dot]com


------------------------------------------------------------------------------

USEFUL gdb macros:

# Courtesy Zhigang W (http://10.182.120.78/tech/vt/ovm/debug/gdbinit.macros):

define ps
	dont-repeat
	set $tasks = (struct list_head *)init_task->tasks
	set $offset = (unsigned long)&init_task->tasks - (unsigned long)&init_task
	set $task = $tasks
	set $task_entry = (struct task_struct *)((unsigned long)$task - $offset)
	printf "Pointer       PID      Command\n"
	printf "%p %-9d%s\n", $task_entry, $task_entry->pid, $task_entry->comm
	set $task = $task->next
	while $task != $tasks
		set $task_entry = (struct task_struct *)((unsigned long)$task - $offset)
		if ($task_entry->pid) != 0
			printf "%p %-9d%s\n", $task_entry, $task_entry->pid, $task_entry->comm
		end
		set $task = $task->next
	end
end

document ps
Report a snapshot of the current processes.
end


define lsmod
	dont-repeat
	# 4 for 32bit kernels. 8 for 64bit kernels.
	set $sz = sizeof(long)     
	set $mod = (struct list_head *)modules
	printf "modptr      address     name\n"
	while 1
		set $mod_entry = (struct module *)((unsigned long)$mod - $sz)
		if ($sz == 4)
			printf "%08lx  %08lx  %s\n", $mod_entry,      \
			       $mod_entry->module_core, $mod_entry->name 
		else
			printf "%016lx  %016lx  %s\n", $mod_entry,      \
			       $mod_entry->module_core, $mod_entry->name 
		end
		set $mod = $mod->next
		if ($mod == &modules)
			loop_break
		end
	end
end

document lsmod
Show the list of modules loaded in the Linux kernel.
end

define log
	dont-repeat
	printf "%s", log_buf
end

document log
Dump system message buffer.
end

------------------------------------------------------------------------------