Re: i5 and invalid instructions

[Jaap Boender <jaapb%kerguelen.org@localhost>]

On Tuesday 13 Nov 2012 13:20:47 you wrote:
> On Tue, Nov 13, 2012 at 01:12:09PM +0000, Jaap Boender wrote:
> > Compiling with -march=core2 -msse3 -mfpmath=sse works fine, so it's
> > probably SSE4 or something like that.
> 
> Stupid question: are you talking about compiling the kernel with that
> flags, or compiling (parts/all of) userland?

What I do is the following:
- put the standard cpuflags include in /etc/mk.conf
- patch cpuflags to output -march=core2 instead of -march=native (*)
- run ./build.sh -j 5 tools
- run ./build.sh -j 5 kernel
- install the new kernel and reboot (everything works fine up to here)
- run ./build.sh -j 5 distribution
- run ./build.sh -j 5 install=/

If I put in the (*) step, everything works fine, if I don't, I get the 
segmentation faults. It would seem that the userland causes the problems, 
therefore. (Not sure which options cpuflags sets where, but I'm just changing 
the detection part, not the option-setting part, so it should just be the 
standard behaviour)

  Jaap
-- 
"Then I'll tell the truth. We're allowed to do that in emergencies."