[Info-vax] Why it is a good idea that OpenVMS isn't on x86-64 just yet

[Stephen Hoffman]

On 2016-01-15 12:16:32 +0000, Neil Rieck said:

> Why it is a good idea that OpenVMS isn't on x86-64 just yet

This is the sort of stuff that the compiler and OS folks deal with.

If you want a better article on this particular topic, start with 
http://danluu.com/cpu-bugs/

All processors have errata.   Intel is famous for their no-fix errata, 
for instance.   Another recent example: various Haswell have had 
Transactional Synchronization eXtensions (TSX) disabled, which was (is) 
a really interesting feature, but it didn't work correctly on at least 
the early Haswell processors.   (Telling which Haswell has working TSX 
and which doesn't can be interesting, too.)

There were bugs in various VAX designs that had workarounds in VMS.  
The VAX-11/750 had a firmware fix — IIRC, that fix required the 
presence of or the addition of a control store card, which not all 
VAX-11/750 boxes had at the time, too.   Various VAX systems had 
mechanisms that could be used to patch the instructions — the control 
store — and various VAX systems had vectors that allowed changes to the 
console programs, for instance.   Alpha and Itanium have analogous 
mechanisms, as do most (all?) other modern processors.

Alpha microprocessors had bugs in certain passes, too.   Among those: 
http://labs.hoffmanlabs.com/node/374

Intel processors ship with what are called chicken bits or kill bits; 
with flags that disable certain features.

Now as to how long Intel can keep dragging along all that accumulated 
compatibility within the x86 design, that is an open discussion.

That difficulty may be what gets RISC-V some additional interest, 
beyond its cost advantages over the ARM license costs.  RISC-V is a 
non-starter for Microsoft Windows and RT on ARM didn't work out for 
Redmond, but ARM already has BSD and Linux use, and RISC-V looks 
interesting to HPE and Oracle, among others.    Whether that's enough?

Operating systems have bugs and errata, too.  One recent one in OpenVMS 
is still open: http://www.eight-cubed.com/blog/archives/001289.html

It's bugs all the way down, unfortunately.


-- 
Pure Personal Opinion | HoffmanLabs LLC