[Info-vax] OpenVMS I64 V8.1 "Evaluation Release"?

[glen herrmannsfeldt]

Johnny Billquist <bqt at softjar.se> wrote:

(snip)
>>> I'm still ready to claim that no machine produced today can connect 2^64
>>> bytes of ram to it. I just don't see it as even probable that anyone
>>> would be putting out 64 address pins.

>> Yes. It would be easy enough to add them when they are needed.
>> The architecture should be designed to allow for them, and even
>> should check that they are zero for now, but no need to implement
>> them yet.

> Maybe, maybe not. Look at the VAX. They had to introduce an incompatible 
> change to the PTE to go beyond 30 bit physical addresses, and then they 
> went to 34. There is absolutely no need, and no point in having much 
> relationship between how a virtual address looks like, and a physical.

Yes. But the page table entries could be designed for it, even if
the physical hardware doesn't (yet) implement them.

Looking at the previously noted PDF, it seems that page table
entries are 64 bits. Some flag bits are in the low 12 bits, where
the page offset goes in actual use. 

> Even more, for the VAX, the actual memory limit was actually 2^29, 
> if I remember right, since half the space was defined/reserved 
> for the I/O space.

Yes, VAX didn't plan ahead as well as they should have.

To start, the 512 byte page was too small. And they did waste some
of the high address bits that later on could have been useful.

(snip)
>> There should at least be the possibility. One could have alternate
>> forms for page tables, and use a smaller, more efficient form for now.

> Possibly, but it all depends on how the architecture have defined them. 
> Sometimes it's not easy to extend. And this I definitely do not know for 
> the s360 in 64 bit mode. Which is why I asked... :-)

They have been pretty good about alternate forms. Starting from
the beginning of S/370, both 2K and 4K pages were allowed, and
either 1M or 64K segment. Some of those options went away on
later processors, though.

Also, later on different forms of CCW to allow I/O addresses 
to extend to more bits. (I/O is physically addressed.)

>>> It would be interesting to hear if anyone with deep knowledge about the
>>> architecture could comment on what the theoretical limits are on how
>>> large physical addresses can be formed.

>> The z/ Principles of Operation is on the IBM web site. Last I knew,
>> they were reasonably priced, for the size, if you wanted to buy one.
>> Way over 1000 pages long.

It seems that paper manuals are not reasonably priced anymore.
That one is over $1000 in paper, but free in PDF form. Also,
almost 1500 pages long.

> Lets see if someone who knows speaks up...

-- glen