[rescue] Ultra1 OBP firmware message on installation boot

[Curtis H. Wilbar Jr.]

As I recall it has something to do with the Ultra 1 processor when executing
a particular sequence of 64 bit instructions.  It is more related to the
hardware than the software.  Software wise as I recall that sequence would
never really occur.... but if you specifically make an assembly program to
make it occur, there is no realy way to protect against it (except run
the kernel in 32 bit not 64).

Wish I could give you a URL to more data, but I'd have to go searching
and I'd imagine this shouldn't be too hard to find in google... try
"Ultra 1" +"64 bit" +"32 bit", and possibly +"bug" +"halt", etc...

-- Curt

>Date: Thu, 13 Feb 2003 20:00:22 +1100
>From: Sid Odgers <sunrescue at dysphoria.eu.org>
>To: "Curtis H. Wilbar Jr." <rescue at hawkmountain.net>, The Rescue List 
<rescue at sunhelp.org>
>Subject: Re: [rescue] Ultra1 OBP firmware message on installation boot
>User-Agent: Mutt/1.4i
>
>Is this problem similar to the intel F0/0F and cyrix 'Coma' bug from
>a few years back?  Or is it a Solaris bug?  I hadn't heard anything
>about it, and I've got a few multi-user Ultra1s running NetBSD/64.
>
>On Thu, Feb 13, 2003 at 03:50:03AM -0500, Curtis H. Wilbar Jr. wrote:
>> Keep in mind there is an issue on Ultra-1 machines in 64 bit mode
>> where a userland assembly language program can do something bad
>> to the machine (I think it can halt the cpu).  This problem is
>> not present when running in 32 bit mode.
>> 
>> This is documented out there on the web somewhere.  If users will not
>> be allowed logins to the box (i.e. a web server only allowing http and
>> maybe ftp) then that won't be a problem.
>> 
>> -- Curt
>> 
>> >Date: Thu, 13 Feb 2003 19:48:01 +1100
>> >From: Sid Odgers <sunrescue at dysphoria.eu.org>
>> >To: patrick at zill.net, The Rescue List <rescue at sunhelp.org>
>> >Subject: Re: [rescue] Ultra1 OBP firmware message on installation boot
>> >User-Agent: Mutt/1.4i
>> >
>> >Yea,
>> >
>> >In terms of performance, unless you're doing heavily memory bound
>> >jobs (ie processes which require more than 4GB of VM) or compute
>> >jobs which use 64 bit integers or floats, you'd be hard pressed
>> >to notice the difference.  If you ARE, then 64 bit computation
>> >will be orders of magnitude faster (1 64-bit instruction takes
>> >the place of 3 or 4 32-bit instructions which emulate the
>> >64-bit instructions for long double(?) and long long types),
>> >and large datasets will be unusable as a 64-bit memory pointer
>> >is required to use more than 4GB of VM on flat architectures (
>> >and it's usually not worth the performance hit on segmented
>> >architectures like the newer IA32 stuff which, from memory,
>> >has a 4 bit segment offset pointer in the LDT, allowing you
>> >to address more than 4GB of VM in 4GB chunks)
>> >
>> >Nonetheless, 64-bit mode is 'cooler', and the (minimal) performance
>> >hit you take for running 64 bit code instead of 32 bit code isn't
>> >going to be noticable either way on a 1/170. :)
>> >
>> >On Thu, Feb 13, 2003 at 02:25:53AM -0500, Patrick Giagnocavo +1.717.201.3366 
>> wrote:
>> >> I am loading Solaris 9 on an Ultra1/170 with OBP 3.5 .
>> >> 
>> >> On boot, the installer gives a message about using the 32bit OS
>> >> instead of 64bit OS.
>> >> 
>> >> I googled and found out how to change it, but my question is, is there
>> >> a performance difference between the 2 kernels?  
>> >> 
>> >> I will be doing some PostgreSQL and web serving and some Perl (spit).
>> >> 
>> >> --Patrick
>> >> _______________________________________________
>> >> rescue list - http://www.sunhelp.org/mailman/listinfo/rescue
>> >_______________________________________________
>> >rescue list - http://www.sunhelp.org/mailman/listinfo/rescue
>> _______________________________________________
>> rescue list - http://www.sunhelp.org/mailman/listinfo/rescue
>_______________________________________________
>rescue list - http://www.sunhelp.org/mailman/listinfo/rescue