On Tuesday, 03/28/2006 at 01:26 EST, Jim Bohnsack <[EMAIL PROTECTED]> wrote: > Ed--Thanks for opening up this subject train, because we will probably be > going to a z/890 next January. I'm on z/VM 4.4 which I (think I) know is > 31 bit/64 bit capable. I also "think" that the z series is 64 bit > architecture only that, as I understand, will tolerate 31 bit code, but at > the very basic level when CP ipl's, it is in a 64 bit box, so to come up on > the z/890, I'll have to be ipling the 64 bit nuc that gets built. Right > now, I just throw away the CPLOAD64 MODULE that gets built when I build a > new nuc. Then I copy CPLOAD32 to my parm disks as CPLOAD. I guess I'll > need to start keeping the bigger CPLOAD MODULE. If I am correct, the nuc > will come up in 64 bit mode but it will appear to my 31 bit code that > nothing has changed--right? I hope that there are no gotcha's in the 64 > bit code path because I won't have any way to test it until the new (old) > box rolls in the door. My co-worker was just in my office and insists that > the z-hardware initially ipls in 31 bit mode. If that is really the case, > I shouldn't even have to change from the CPLOAD32 MODULE I'm running.
Your co-worker is correct: Processors IPL in S/390 architecture mode and the operating system instructs the processor to switch to z/Architecture mode using the SIGP instruction. This is also why you can't run a 64-bit guest when using a 31-bit CP. If the real CPU is not in z/Architecture mode, then the guest can't get its virtual CPU into z/Architecture mode. The "problem" with new machines is typically the I/O subsystem. There are new chpid types and multiple subchannel sets. Dynamic I/O functions are, as a consequence, different as well. Operating systems that don't have "toleration" support can stumble and fall if they misinterpret their new surroundings. (We recently saw an older level of z/VM die when we attempted to IPL it on a z9-109.) Fixing these systems' world view is what the "toleration" PTFs (usually to 3.1 and 4.4) allow. Alan Altmark z/VM Development IBM Endicott