Oddities with recent LFS as a build system

 I'm back on my "build machine" taking a look at what impact
make-4.0 is likely to have.  I got this machine nearly 18 months ago
(AMD phenom, 4 real CPUs, 8GB of memory less the built-in graphics)
and at that time it had a habit of losing its lunch [ Interal
Compiler Errors ] if I ran make -j4 (kernel compiles almost always ok
but other things tended to fall flat on their face), so that I had to
drop the caches and limit myself to 3 jobs.

 Also, at some point the kernel cpufreq for the phenom changed to
acpi and it took me a while to tune my initscript so that it was
again faster than my intel SandyBridge.  But earlier this year things
were looking good and I was able to use make -j4 without worrying.

 Now, I'm building *from* LFS-7.4.  Wednesday I was building last
month's LFS-svn (my error, I apparently updated my repo without
rebuilding the local copy of the book) on linux-3.11.1 with
make-4.0.  Perhaps 3.11.1 was a bad kernel, but things fell over
with ICEs in chroot and I had to drop back to make -j3, and then at
the end of chapter 6 (when I realised I was using an out of date set
of packages) I got a segfault unmounting /mnt/lfs and lost the box
when I tried to drop back to a tty.  Turned out one of the ICE
segfaults was associated with a bad page table, and the box had
BUG:'d.

 Thursday, I've built 3.11.4 (hopefully, better - no bugs or oopses in
the kernel logs so far) but I've still had to drop back to make -j3
in chroot because of ICEs.  Also had an ICE with Python-2.7.5 (one of
the things I build before booting) and make -j3, but that completed ok
after I dropped the caches and retried.

 Now that it has booted, it seems to be ok.  I'm still running with
make -j4 for the moment.  But what really concerns me is the SBU
time: the only real changes since LFS-7.4 are make-4.0 and
gmp-5.1.3.  One of the things I normally do in a new system is to
rerun binutils pass 1 [ -j1 ] so that I can recalculate the SBU [
building from an old system will always be quicker ].  On 7.4 it
built itself in 208.542 seconds.  For this build, binutils pass 1
took 156.559 seconds (seems unrealistically quick) but rebuilding
itself after booting took 253.243 seconds which is a lot longer than
I was expecting.  I saw that mandb was running for some minutes
after I booted, but with 4 CPUs and nearly 8GB of RAM that didn't
ought to make a lot of difference to the timings.

 Is it just me and my not quite as good as expected hardware, or are
things getting flakey ?

ĸen
-- 
das eine Mal als Tragödie, dieses Mal als Farce
-- 
http://linuxfromscratch.org/mailman/listinfo/lfs-chat
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page