Subject: Re: kernel panic!
To: Riccardo Mottola <rollei@tiscalinet.it>
From: Timo Schoeler <timo.schoeler@macfinity.net>
List: port-macppc
Date: 06/27/2005 10:01:32
thus Riccardo Mottola <rollei@tiscalinet.it> spake on Mon, 27 Jun 2005 00:01:22 +0200:

> Hey,
> 
> 
> > options ALTIVEC sets it if I remember correctly. But all it does is to
> > enable C-stubs, ABI-support and assembler mnemonics for AltiVec
> > instructions.
> 
> I enabled those all because Timo said that enabling at least one of
> those optimiwzation features would have the bug go away.
> 
> 
> > > the kernel is unstable as ever! slight compilations, use, cvs work...
> > > may crash the box withn 5-10 minutes!!!
> > 
> > Does this happen with L2-cache disabled too? If not the L2 vs. CPU clock
> > ratio may be wrong. Some mainboards don't like to run the CPU bus at
> > 50MHz. If you have MacOS somewhere around I'd strongly recommend to get
> > NewerTech's CacheProfiler ( or whatever they call it these days. It
> > supports many CPU cards from other vendors as well, like my Phase 5 G3 )
> > and play with it until you find something that works, then change the
> > NetBSD kernel accordingly.
> > The G4 card almost certainly tries to run the bus at 50MHz, some even
> > try to overclock it slightly (cpu0: 409.10 MHz sounds a bit odd). Some
> > boards have problems at 50MHz, even more at >50MHz. Memory may be an
> > issue too - I have 2 32MB modules here that worked fine with two 604e
> > cards running the bus at 43.something MHz, but went bananas with the G3
> > running it at 50MHz - there are plenty of memory testers for MacOS,
> > can't hurt to try one of them. ( I recommend to do that in MacOS not
> > because I like MacOS so much ( I don't ) - but MacOS  is the OS where
> > you'll have an utility to fiddle with CPU/cache settings on the fly and
> > it should be a little more robust against flaky memory ).
> 
> http://www.everymac.com/upgrade_cards/newertech/maxpowr_g4/maxpowr_g4_400.html
> 
> the cache specs look to be correct: 2:1m 1Mbyte... 400Mhz
> 
> netbsd reports the card as 400mhz , apple system profiler as 446
> 
> I didn't find the newetech sofware you mention, I found "powerlogix
> cpudirector" but the data it gives is pretty suspicious. When I start it
> up, it reports:
> 
> 400 Mhz
> 1mb cache running 1:3 and bus speed of 44Mhz...
> speculative access off (and how does netbsd behve here? I left it
> disabled since it says to do so on OldWorld powermacs or it would cause
> corruption of filesystem and other problms) and "dynamic power
> managment" enabled.
> 
> I set it 1:2 .
> 
> then I run some heavy-duty apps I stil have on disk. Months I didn't use
> teh, 3d modelling, rendering... damn is this box fast. I didn't remember
> macos 8.6 was so quick :) running strata studio pro blitz is amazing.
> The matrox surely makes its part, 320mb of ram and g4/400... I did run
> these apps with my 604/120! the 604e/200 was something I tried under
> netbsd only. This things runs raytracing as wild, has incredible speed.
> MacOS had some incredible 3d apps at the time, really.
> 
> I run povray and rendered as wild and I left all applications open, just
> to fill up ram :) Povray did a scene in only 5 minutes, cool :) I loaded
> the box as high as possible, running a render in PoVray and one in
> form-z.
> NetBSD would have long crashed. I then had an app crash, but amapi 5 was
> always a crasher and evidently time didn't cure it :) "es" exited it...
> I played soundtracker modules and could do real-time wave preview... ah
> teh guys from the future crew... Early ninties were cool times for
> computers, software and electronic music! And of course open photoshop
> and run the (useless) gaussian blur filter. Ahah!
> 
> ok, just to be sure it is netbsd's kernel fault. And damn, you can say
> waht you want about unix kernels, but I can measure a higher disk
> thoughput here and my s3m module tracker files decoded perfectly and
> didn't click.
> 
> but yes, it is macos, no prompt. I am almost lost using classic now,
> incredible.
> 
> thus, I seek further help. Maybe build a kernel with less optimizations?
> hmmm btw, currently even building a kernel is problematic.
> 
> cheers
> 
>    Riccardo

for me it sounds much more like NetBSD setting up L2 cache not correctly (too high clock speed, maybe kicking one or the other waitstate, etc.) so earlier or later a data error occurs (no ECC here!) and crashes the machine?

that's quite normal when kicking it way out out it's specs ;)

would be best to investigate this under NetBSD, fiddling the knobs to turn the card to a

* correct CPU:L2 Cache ratio

* correct bus speed (start with 40MHz and then turn up, not 50MHz and then down! ;)

* NetBSD using the right /size/ of cache(s)...

just an idea,

timo

-- 
Timo Schoeler | http://macfinity.net/~tis | timo.schoeler@macfinity.net
//macfinity -- finest IT services | http://macfinity.net
Key fingerprint = F844 51BE C22C F6BD 1196  90B2 EF68 C851 6E12 2D8A

There are 10 types of people in the world. Those who understand binary
and those who don't.