I think I tracked down the problem I had on NetBSD-current/i386 with
pcvt and losing keyboard control to a fresh X server trying to grab the
graphics mode on the same virtual console as an existing server.  I
didn't have problems when I ran 'xinit -- :1' from another virtual
console; only when I ran it _from within an xterm_ did I have problems.

Here's a patch which seems to fix the immediate problem; it makes the
VT_SETMODE fail if something tries to set process mode on the same VT as
an active process.  However, this doesn't mean that "xinit -- vtY :1"
will start an X server on vtY--the context at the time of the VT_PROCESS
may still be the current virtual console if the existing X server
doesn't yield before the new one tries to set VT_PROCESS mode.  Maybe
there's some missing handshake on the console switch completing.

==John

*** 1.1	1994/08/05 01:39:53
--- pcvt_ext.c	1994/08/07 02:55:52
***************
*** 2297,2302 ****
--- 2297,2306 ----
  #endif /* PCVT_FAKE_SYSCONS10 */
  
  	case VT_SETMODE:
+ 		if (vsp->smode.mode == VT_PROCESS &&
+ 		    ((struct vt_mode *)data)->mode == VT_PROCESS &&
+ 		    vsp->proc != p)
+ 			return EBUSY;		/* already in use on this VT */
  		vsp->smode = *(struct vt_mode *)data;
  		if(vsp->smode.mode == VT_PROCESS) {
  			vsp->proc = p;


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