On 1 March 2013 13:07, Jim Mulder <[email protected]> wrote: > We have gotten to the point where Mr. Relson and myself > are among the remaining old-timers in MVS development in Poughkeepsie,
Sort of like when you look around a hospital ER, and all the doctors - not just the interns, but even the residents - look to be about 14 years old. :-) > and we have only been here since around 1979. "in the beginning, > as soon as the REFR attribute was available" is considerably > before that time. My understanding from folklore is that the REFR > attribute predates MVS, and its purpose was to designate modules > for which a new copy of the module could be safely loaded (refreshing) > after a storage related machine check. It was certainly documented that way for (at least) MVT, and perhaps earlier versions of OS/360. Whether the necessary combination of the hardware truly being in a defined state after such a machine check, and the software support needed, ever really worked is another matter. In my experience, machine checks on S/360s were almost always followed shortly by the need to IPL. > So if you are asking why the designers at that time did not > choose to also use the REFR attribute for write protection > purposes, I can only speculate. Since that predated > address spaces, the only choice for write protection would be > using storage key 0. I don't know whether or not there was > any means at that time for providing key 0 storage > within a region. But to allow the code to be fetched while > executing in the region's key, key 0 storage would have to be > non-fetch protected. That would allow the code to be viewed > by programs running in other regions, and that may have been > considered to be undesirable for security purposes. It may be worth pointing out that MVT, up to its last days, did not offer any supervisor support for fetch protection. None at all, though the 360 hardware had had it for years. At one point I looked at as part of an ill-conceived project to run work of different classifications (not military - university undergrad and university admin work) on the same MVT image, and quickly concluded that the MVT components' own fetch references to key zero storage were so sloppy (as to PSW key) that there was not much hope. Perhaps someone else did it, and other work to harden MVT, but by the time this was a hot topic MVS was already available - albeit non performing. Tony H. ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
