Subject: Re: AMD64 stability (was: Quick build.sh type question)
To: Andy Ruhl <acruhl@gmail.com>
From: Richard Rauch <rkr@olib.org>
List: port-amd64
Date: 12/04/2004 07:28:01
On Sat, Dec 04, 2004 at 04:44:04AM -0700, Andy Ruhl wrote:
> On Fri, 3 Dec 2004 22:39:27 -0600, Richard Rauch <rkr@olib.org> wrote:
> > Your question is a bit vague, I guess.  (^&
> 
> Yeah, it was, and I have another day of progress to help out a little...
> 
> > Which version of NetBSD/amd64 are you looking at (well, you said the
> > "snapshot", but I haven't checked its version)?
> 
> I successfully built the 2.0 release branch yesterday and rebooted and
> this seems to have fixed things. Where things wouldn't build, now they
> do. This is cool. I'm much happier now.
> 
> > And "stable and usable" for what?
> 
> Desktop stuff mostly. I've noticed that I'm losing some of the stuff

Probably it is more than stable enough if you can get your applications
to run.


> I'd run under Linux emulation now. I guess I need to figure out how to
> get the 32 bit stuff working. I generally don't want to have a box

You can boot NetBSD/i386 if you want 32-bit support.  (^&


 [...]
> >  * Lack of support for some of my hardware at first.  E.g., no DMA
> >    support for the hard drive.  This wasn't fatal, but as I recall it
> >    was a whole lot faster to do *everything* via NFS from an i386 than
> >    to use the local hard drive.
> 
> I appear to have DMA working on an older ATA66 drive. Maybe this was fixed?

Yes.  Note, the "at first" qualifier.  (^&

It depends on the chipset, I'm sure, but mine has been doing DMA happily
for a long time.


> >  * I have had to turn off IOAPIC in my kernel config and build a custom
> >    kernel in order to keep my hardware all configuring and to keep it from
> >    dying (losing interrupts) during normal use.  I have never seen this
> >    officially fixed; the GNU/LINUX kernel, at least around January 2004
> >    or so, actually disabled IOAPIC if it detected this motherboard
> >    chipset that I'm using.  (nVidia nForce3.)
> 
> Hmm. This might be a sporadic problem with my Via based board too.
> I'll give this a try. Although I don't think I've seen this since
> going to the latest release branch.

It wouldn't hurt, but I've only heard of this in relation to nVidia
chipsets.


 [...]
> >  * X loses mouse button events if the mouse is moving while the button
> >    is clicked/released.  I believe that M. Bouyer(sp?) confirmed that
> >    he also saw this on the Alpha, so it is believed to be a general
> >    X problem, not a NetBSD or AMD64-specific issue.  (But it certainly
> >    does not manifest with my i386 boxes---same mice.)
> 
> I have not seen this yet.

Maybe it was fixed, then.


> >  * Similarly, I've noticed that occasionally CapsLock releases have been
> >    lost, though I can't correlate them to mouse motion.  I never had
> >    this with the i386, and have an i386 laptop that is running a not-too-
> >    distant -current.  (I can't get the laptop fully up to date due to
> >    another issue that I haven't the expertise to quickly solve and that I
> >    have not yet been able to even *slowly* solve.  (^&)
> 
> Haven't seen this yet either.

I don't see it too often, but when it happens it's annoying and hard to
get out of.


> > Presently, my system is pretty stable, though the particular revision
> > I am on (2.99.10 kernel or so) tends to sporadically seize up.  I did
> > not experience that with older kernels.  It has not got to the point that
> > I've bothered trying to build a new kernel in the hopes of fixing this
> > issue.
> 
> Haven't seen this since going to release. I'm guessing some hardware
> is particularly flaky. PCs just aren't too high quality anymore.

No.   It's not that.  And I should clarify: The seizure is not a
complete lockup.  Just things stop for half a second or a second,
at least in X.  It is not uncoming while moving the mouse out of
one window to see it freeze just at the last position it was in
the window.  I'm not sure what is happening during that time, but
in a second the mouse exits the window and life is good again.

This only started happening around 2.99.10, I think, so it probably
never affected any 2.0 release candidate, and has probably been
fixed.  (I seem to recall hearing a discussion of a similar bug)
The problem has not been quite annoying enough to reboot with a new
kernel...(^&


> > I use my AMD64 as my primary system.  The biggest annoyance I've had is
> > having to adjust to using links-gui as my graphical browser.  links has
> > its virtues, but I didn't like being forced to change.  (^&  (No other
> > graphical browser I've tried has worked for me under NetBSD/amd64.
> > See a recent thread on this list.)
> 
> I'm using Firefox right now as a matter of fact. It wouldn't build
> before I upgraded to the latest release branch, now it does. I can
> probably try making a package if you want to try it (after the weekend
> though).

No thanks.  I got tired of binary packages a long time ago; too many
headaches.  (^&  And that was on a Pentium II.

If, as you say, pkgsrc is now building firefox on the AMD64, it will
probably work for me, too.



> > I do not run any of the 2.0 release candidates on my AMD64 box,
> > however.  I expect that I'll continue to track -current for a while,
> > so my experiences may not be directly relavent to your question.
> > Still, it's an answer of sorts.  (^&
> 
> I thought I was going to go to current but I haven't seen a need so
> far. Maybe I'm missing something?

Not that I know of, but this machine has been fairly happy on -current
for a year.  I'll probably switch to 2.1 when it is out.

-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/