Raspberry Pi: still getting prefetch aborts

[George Mitchell]

On 01/15/14 08:56, Ian Lepore wrote:
> On Tue, 2014-01-14 at 19:06 -0500, George Mitchell wrote:
>> On 12/25/13 19:09, George Mitchell wrote:
>>> FreeBSD 10.0-PRERELEASE (RPI-B) #0 r259866M: Wed Dec 25 17:26:28 EST 2013
>>> (M because I selected serial output in RPI-B)
>>>
>>> /usr/ports is NFS mounted from elsewhere.
>>> [Long description of prefetch abort] [...]
>>
>> Well, that got a deafeningly silent response.  I finally copied my
>> /usr/ports onto the SD card instead of NFS mounting it, and now my
>> port build is chugging along happily so far.  So the problem,
>> whatever it is, is triggered by NFS.                    -- George
>
> I've been consistantly unable to reproduce the userland crash you see on
> rpi, even though the tracebacks and all make it look a lot like the
> wrong-endian kernel crashes I see on Wandboard (although my gut tells me
> it's not really the same problem).  My setup is a lot like yours, with
> nfs-mounted filesystems, but when I build ports that way it either works
> fine, or the port builds die for other reasons.
>
> -- Ian
>
Well, for me, I still get the abort if /usr/ports is NFS-mounted.  (To
be more precise, in case it is a factor, I'm running the automounter
and /usr/ports is a symbolic link to /host/sullivan/usr/local/pi/ports.)

Build version: FreeBSD 10.0-PRERELEASE #0 r260564M
svn version in /usr/ports: 337160

cd /usr/ports/ports-mgmt/pkg
make

gives an immediate prefetch abort 100% of the time.  Putting /usr/ports
on my SD card has gotten me a substantial part of the way through
installing print/cups, though I have seen one freeze-up with no debugger
prompt on the serial port and no response to a break.          -- George