[lfs-support] Invalid instruction compiling kernel

Paul Rogers paulgrogers at fastmail.fm
Tue Aug 2 00:30:38 PDT 2016


Real Progress Report  8-)

Firstly, since that recompile of the four "gcc" packages with 
-march=i686 did no good, I restored everything back to my base level,
on the i7 and Tualatin.  (Never lose sight of one's base!)

Secondly, because all I'd actually tried to compile on the Tualatin was
the kernel, I tried to recompile gmp there to see if it was the kernel
source or the compiler.  I quickly got invalid instruction errors.  So
the compiler is broken.

Thirdly, since all I'd actually compiled here on the Conroe was the
kernel, and gmp is everybody's favorite culprit, I "collected" the four
package-management files for my base gmp package here into a tarball for
safekeeping, then rebuilt it specifically trying to give it no wiggle
room, i.e. CFLAGS="-march=i686 -mtune=i686" ... --host="i686-pc-linux-
gnu"!  That worked, establishing that indeed the compiler does work on
the Conroe.

Then I took the tarball of those as-builts to the Tualatin and replaced
what was there from the base build.  Next tried recompiling the gmp over
there and that proceeded well bast where it had blown up before!  YAY!
^C out of that and go for the kernel.  That's still compiling, FS at the
moment--it is just a single core 1.4GHz Pentium /// after all.

So, it appears that's a solution to building gmp so it works at some
base level of CPU compatibility, just using configure parameters.

>Again, in gmp rename the configfsf files to the corresponding
>config file.

I intend to try that later this morning, after I get some sleep.  I
think I prefer the method above, if this really is a robust solution, as
it involves no changes to the package source.  But right now all I have
to show as success is a still running kernel compilation.

-- 
Paul Rogers
paulgrogers at fastmail.fm
Rogers' Second Law: "Everything you do communicates."
(I do not personally endorse any additions after this line. TANSTAAFL
:-)

-- 
http://www.fastmail.com - The way an email service should be



More information about the lfs-support mailing list