Subject: Re: what's this machine check mean?
To: None <port-alpha@netbsd.org>
From: der Mouse <mouse@Rodents.Montreal.QC.CA>
List: port-alpha
Date: 04/17/2000 21:55:51
> Sadly, should your axppci33 actually work "perfectly", it's lack of
> speed will be so pronounced that the distinction between "perfect"
> and "broken" will be meaningless.

I think you've been spoiled by high-end machines.  My fastest home
machine at present is a 486DLC/33; the fastest of them which I can
count on for a "make build" without falling over is a SPARCstation IPX
(supposedly 40MHz, for what that's worth).

A 166MHz alpha, even without cache, will be a distinct step up for me.

> Those come with and without cache, BTW.

Mine has no cache, but does have sockets I could put cache in.

> Now, having discouraged you from even caring, if you do want to try a
> really deep hack you might be able to make a marginal kernel reliable
> with a custom set of bank timing registers.  The LCA has an
> integrated DRAM controller, see page 5-20 in ec-qc4ga-te,

I'm fetching it now.  I suspect this isn't it, though - see below.

> It's not a true fix, because (1) the SRM settings have to work well
> enough to get the kernel loaded

This has not proved to be a problem.  The only time I saw machine
checks before the kernel booted was when I had some otherwise untried
memory in there (not sure it was even as fast as 80ns).

> and (2) those registers are write-only (!@#!) so you don't get to
> tweak them, you can only just load really conservative values in.

(Eww.)  I'll keep it in mind...and whether I had it or not, I haven't
read ec-qc4ga-te.  I'm flipping through it now.

[re power supplies and grounding]
> 	* Don't worry about your setup too much .. the board is a world
> 	  unto itself, and doesn't really care where ground is until
> 	  you start connecting it to other things, which generally
> 	  supply a good ground or, like serial I/O, are somewhat
> 	  tolerant of bad ones.

That's more or less what I figured.  I've set up the system with
exactly four things connected off-board: (1) the power supply; (2) the
console serial line; (3) the (SCSI) disk, which is powered off the same
power supply as the motherboard; and (4) a twisted-pair Ethernet card
in one of the PCI slots.

As I remarked, there is no case to ground it to, so I haven't worried
too much about it.

[Wilko, on speed]
> NoName boards were sold in the Netherlands last November
> new-in-the-box for US$ 12 each.  At such a price one cannot complain
> about price-performance ratios.

Mine was US$ 12 less than that.  I'm not complaining. :-)

>> Those come with and without cache, BTW.  I'm not sure what the speed
>> contribution of the cache is, although it feels funny to even use
>> the word "speed" when referring to an LCA box. :-)
> Without cache life becomes rather miserable.

If I can make these bloody machine checks go away, I'll probably get
some cache to go in it.

Today I picked up five 4M sticks of 70ns-rated RAM and have tried the
machine with them.

It's no worse, but it's also no better.  I'm seeing exactly two
predictable aspects of these machine checks:

	- They always show a pc in lca_mem_read_1

	- They always occur approximately the same amount of time after
	   power-up, this time delay getting smaller with later
	   attempts.

In view of the latter, I suspect thermal problems - especially since
the 3.3V regulator heatsink gets hot enough to concern me.  I'm
pondering ways to cool things more effectively.

					der Mouse

			       mouse@rodents.montreal.qc.ca
		     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B