Re: Re: Wait State 00050064

On Fri, 20 Sep 2019 at 14:07, Robert Schreiber [email protected]
wrote:

Excellent.  Thanks for the clarification to what MVS is doing.
> Nonetheless, my use of LPARNUM 01 was incorrect and was ultimately the
> cause of the wait state. Just to be clear here this is a user error and NOT
> a Hercules bug. Perhaps it might have been nice for Hercules to flag the
> LPARNUM 01 since its inconsistent with ARCHLEVEL S/370?
>

I had meant to add a bit to that effect, but hit send too quickly. It may
well be a user error to specify LPARNUM at all on S/370, but IMHO it is a
Hercules bug to accept it for an architecture that has no concept of LPARs.
What is certainly a bug is for Hercules to then deliver an incorrect result
from the STIDP instruction.

I don't think this can be argued away by calling it retrofitting of new
instructions into S/370 as an extension.

It sounds as though you did a lot of detailed debugging to catch this. If
you look at that code that does the failing GETMAIN, I'll bet you'll find
the GETMAIN'd address is loaded into CR15, which is architected to point to
the MCEL.

Tony H.


On 9/20/2019 12:48 PM, Tony Harminc [email protected] [H390-MVS] wrote:

On Fri, 20 Sep 2019 at 11:44, Robert Schreiber [email protected] wrote:


I did a lot of tracing, etc and found that NIP (Nucleus Initialization
Program) was
getting an 80A abend by asking for a huge amount of memory. What was
going on is that
bytes 6-7 in the result of an STIDP instruction was being used to
calculate the amount of
memory get. I have no idea what MVS is try to accomplish here--this
has to be an MVS bug.

It's not.


Referring to SA22-7201-08, Pages 10-90/91:

That's your problem. You're looking at a post S/370 architecture PofO.
If you look at the S/370 Principles of Operation (e.g. the -10 at
Bitsavers), you'll see that the rightmost two bytes from STIDP contain
the maximum length of the Machine Check Extended Logout (MCEL) for
this machine. A perfectly reasonable thing to do a GETMAIN for.


"The format bit (F) in bit position 48 specifies the format of the first two
digits of the CPU identification number."

Not in S/370.

Tony H.