5.4-RELEASE Athlon64 E3/E4 core support

[Scott Long]

Ryan R. wrote:

> First off, some info about my setup:
> 
> Motherboard: DFI Lanparty SLI-DR, nForce4, socket 939
> CPU: Athlon64 3700+ "San Diego" core, revision E4, socket 939, 1mb L2 cache
> Memory: 2x 512mb OCZ Value VX
> I've run prime95 in Windows for just over 84 hours, with zero errors. 
> I've run memtest86+ for about 12 hours, with zero errors.
> 
> Now, the problem is that I recently upgraded to this system from a socket
> 754 "Clawhammer" (C0 revision) setup, which worked perfectly in
> 5.3-RELEASE/amd64.  Before I upgraded the hardware, I didn't think to
> rebuild the kernel/world with a clean make.conf and kernel config.  When I
> booted the new machine, I got a kernel panic (fatal trap 9) on boot,
> between the detection of my drives and mounting root.  I tried a few
> things to get it working, but to no avail.  I then decided it has to be my
> old world/kernel, so I did a binary upgrade to 5.4-RELEASE/amd64.
> 
> After doing the binary upgrade, the system would boot again (at first,
> anyways).  Once the system had booted, I figured I should rebuild my
> ports.  While I was watching the build, the kernel paniced again (I forgot
> the exact message, and I didn't copy it down).  I once again tried a few
> things to get it to work, but since I had spent so much time trying to get
> it to work before the binary upgrade, I figured it'd be easier to just
> wipe the drive and start over fresh, that way I have a known good system
> to start with if any problems pop up.
> 
> So, I wiped the drive, and installed 5.4-RELEASE/amd64 on it.  Install
> went fine, system booted, but then the panics came back.  It usually
> happened while building some ports, but it happened in a different spot
> each time, so it's not like one specific port was causing me problems.  It
> also didn't happen only under load.
> 
> Now, I have the problem narrowed down to one of two things:
> 1) The hardware is bad.  This lead me to run the memtest/prime95 tests,
> but they revealed no problems.  This leads me to option 2.
> 2) The amd64 build of FreeBSD.
> 
> So, I once again wipe the drive, but this time I install 5.4-RELEASE/i386.
>  It installs fine, and boots fine.  One thing I notice during boot, is
> that SSE3 isn't detected by the kernel as a CPU feature anymore (amd64
> build did detect it).  I begin building all my ports, and I was able to
> get my system up and running without even a hint of trouble.
> 
> 
> So, I'm curious to know if there are any problems with the amd64 release
> with these E3/E4 revision CPUs, maybe relating to the SSE3 instructions? 
> Is there anything I can try to narrow down what exactly is causing my
> hardware to panic?  I know I should build a kernel with debugging support
> (I tried it before installing the i386 release, but got a kernel panic...
> go figure), but I probably wouldn't be able to see the problem in a trace
> if I tripped over it.
> 
> Any help would be appreciated.  Thanks.

It's pretty hard to make an nForce chipset work under FreeBSD 5.x.  The
biggest problem seems to be with how the legacy system clocks are 
implemented, which FreeBSD 5.x and prior relies on.  FreeBSD 6-CURRENT 
uses a completely different method that is compatible with the nForce,
so I'd recommend trying that instead.  I use it on my desktop nForce4
machine without any problems (except for the buggy nve ethernet driver).

Scott