feedback and request for help with x86_64 preview

[Karen Spearel]

On Tue, 2003-12-30 at 03:01, Warren Togami wrote:

> 2) http://togami.com/~warren/archive/2003/rhel3amd64-panic.jpg
> Booting any AMD64 Linux kernel almost immediately causes a kernel panic 
> due to corruption of the RIP instruction register.  Whenever it comes 
> back from the HLT instruction, the top 32bits are zeroed out rather than 
> retaining the bits like they should.  Some kernel guys have said that 
> they ran into this issue with Asus nForce3 motherboard which was 
> resolved by a BIOS update.  MSI took nearly 1 month to respond to my 
> initial query about this problem, and

I have an Asus K8V with an Athlon64 3200+ that has been running Justin's
x84_64 FC1 preview 2.4.22-1.2129 for several weeks without any
problems.  The system is vanilla otherwise though, with an ATI 9000 agp,
PATA drives hung on the VIA controller and std PS/2 keyboard and mouse. 
I updated my son's system with an MSI K8T Neo-FIS2R/Athlon64 3000+ for
Christmas...I only had it overnight so I only got in a limited amount of
testing before I had to get it running XP for him (he's a diehard
gamer...what can I say?).  However, using an old 20 gb PATA drive, the
aforementioned ATI 9000 (Anaconda didn't find his new 9600) and PS/2
kbd/mouse, the system installed and ran FC1 and happily did an NFS
install of Justin's x86_64 preview which it ran all night without issue.
Although it didn't get stressed a great deal, it did get fiddled with
and was rebooted multiple times.  It was running 1.0 BIOS which didn't
get updated until the next day after XP was running.  At any rate, it
seemed to be entirely happy with both 2115 in 32 bit mode and 2129 in 64
bit mode.  Both systems have AMI Bios.  Removing the sata_via reference
in modules.conf takes care of the initial non-fatal kernel oops seen on
the initial installs of x86_64 2129.

Now, if we are talking 2.6, that is an entirely different story.  An
attempt to get 2.6 running failed just as it does on my K8V...slab
corruption as far as the eye can see.  This seems to be a
incompatibility between AMI Bios and the released 2.6.0 kernel...a patch
which fixed the problem in test releases seem to have gotten
dropped...still looking into that one.  The K8V will run 2.6...just none
of the later kernels.    

> 3) Booting just gets STUCK during various stages of POST, sometimes in 
> GRUB, or sometimes a random kernel crash during bootup.  Sometimes GRUB 
> says an error like "Invalid Geometry".  This problem seems to occur 
> randomly, sometimes refusing to boot for an entire day, and some other 
> days working completely with 100% stability without any problems.  I'm 
> totally confused by this behavior.
> 

The Asus board does seem extraordinarily picky about memory.  I have
matched a pair of 512 mb DDR 400 with known good ram part/batch numbers
that run just fine on several Athlon XP motherboards at 400 that won't
run reliably above 333 on this board regardless of bios timings.  The
problems really only show up when making a kernel but they are there.  I
tried to send a message to Asus' tech dept the other day but predictably
the message bounced as the supplied address was no good.  The MSI board
was only running 1 512mb DIMM and doesn't seem to mind running it at
400...no problems reported after a week of very intense use.

I must say I was impressed with the 3000+ compared to my 3200+...there
just didn't seem to be any real speed difference although I didn't have
time to compare kernel builds between the two...would have required some
memory juggling that I didn't want to do anyhow. I'd also say that my
impressions of the MSI board were favorable...so I wonder if this board
wasn't a later rev than Warren's...the first pass always seems to have
hardware bugs....at these clock speeds, rerouting even one trace can
make a big difference.

Karen