On Sat, 4 Dec 1999, Andrew Gillham wrote:

> Windows NT has PALcode embedded in it somehow?

Windows NT has god only knows what embedded in it.  We all know where
Digital has been taking it from them since the beginning.  Alpha's even
have a PeeCee DRAM-refreshtimer-speakerbeeper in them, no?  i'm feeling
more wholesome not knowing what's going on inside that tent.

not that anyone is offering to tell me.

> "SRM is required for NetBSD/alpha" along the lines of "OpenFirmware
> is required for NetBSD/macppc." (e.g. to boot and get started)

notPC's don't use the same MAME-type arcade machine architecture as PC's.  
The OS generally stays on friendly terms with the firmware even after
boot.  This is good, because it allows some simple hardware abstraction to
take place so that motherboards don't need to be pin-for-pin,
trace-for-trace compatible with the golden original version of the Bionic
Commando board or the PC XT or whatever video game you happen to be
talking about.  Supporting a chip or a chip family is good enough.

The arcade machine philosophy is perhaps workable, for arcade machines.
I've been harping about the PS2 recently.  If you buy a Playstation, you
know your small investment is protected for at least three or four years.  
Every Playstation sold in that release cycle will be basically identical,
so you don't have to worry about Johnny down the street getting a faster
Playstation than you, at least not for a while.

Likewise, if your OS works on the first Playstation, it will almost
certainly keep working for a few years without further attention. When a
new one does come out, it will be a BFD, completely different, radical
architecture changes, new CPU, require all new MD code.  

But working this way is outrageous when you're talking about $1000-$3000
machines that change just very, very slightly on a monthly basis like
Alpha's, with a bunch of different CPU models and steppings and
architecture codenames. If you expect your code to be the only code
running in the system, you are doomed to the constant nuissance-level
hacks that make good i386 support such a challenge--obscure tweaks to the
OS to keep up with the latest buggy chipset used by some flash-in-the-pan
Chineese Slave Labour motherboard manufacturer.  That's what firmware is
supposed to help.

 OS Q.  Any SCSI stuff on this box?
 FW A.  There is an ISP-ish chip.
        The ISP is a (Revision xxxxx), in case you care.
        It is attached (here.)
        It is initialized and ready for your use.

as opposed to:
/* probe for ISP */
[...]
/* are we on ISA? */
[...]
/* don't touch 0x224 with an 0x8a or an 0x9f, because it might 
   be a soundblaster and lock up the bus */
[...]
#if 1
/* clever hack from frightwig@eroticom.com */
[...]
#endif
#ifdef __i386__
/* workaround for 440BX bug */
[...]
[...]
[...]
[...]
[...]
[...]
[...]
[...]
#else /* Digital host bridge */
[...]
#endif
/* great!  now, initialize the chip. */
#ifdef SUPPORT_ISP1040A
[...]
#else /* we are on a 1040B or later */
[...]
#endif
[...]
[...] /* grrmph..  eh, nevermind. */
[...]
[...]

My understanding is obviously minimal--the other posts here suggest that
the PALcode idea is a lot more complex and interesting than this.

PC's keep trying to do something vaguely like Firmware, but from reading
stuff on these lists it seems they still don't ``get it,'' and the BIOS
people who keep trying are all fired as far as NetBSD is concerned.  
Thanks to NT the Alpha seems to have inherited some of this insanity.  
The people who wrote MILO _wanted_ an arcade machine, and refused to trust
firmware, or even admit firmware's right to exist--which, is often the
right way to deal with bugridden PeeCee nonsense, but the Alpha is
hopefully not so badly off.

> I said it was a stupid question.

If it's so stupid, then why don't the Linux people understand it?

MILO spawned a whole mess of its own--a dozen forks and branches and
revisions, half of which violate the GPL, half of which don't work, and
half of which disappear from the site the moment you realize you need
them. Digital funded some development on it, which brought two branches
and who knows how many revisions of broken video card BIOS x86-emulator
code to the table.  I remember at one point having three video cards (one
of which counted for two, becuase it had a socketed ROM chip that I could
remove, and this changed things a bit), three versions of MILO, SRM and
ARC, and workingness defined as ``works in SRM, / works in MILO / works at
kernel boot / works in X'' which gets you a 96-cell 4-dimensional
matrix--i fell over laughing and plugged in a Mac SE serial console. MILO
inspired the NILO project, the Overly Complex 256kB proposed Linux
Netbooting Architecture complete with a scheduler and an interrupt
handler, intended to replace the sprightly BSD netboot code they took from
us. And then suddenly the ``firmware'' in the Corel Netwinder is a leprous
Linux kernel, just like MILO.  dumdumdum.Duh...  MILO, strikes again.

I tell you as one who used it--MILO is eeevil.  Stay away from it.

-- 
Miles Nordin / v:1-888-857-2723 fax:+1 530 579-8680
555 Bryant Street PMB 182 / Palo Alto, CA 94301-1700 / US