On Sat, Mar 19, 2016 at 02:41:56AM -0400, [email protected] wrote: > I have, in the past, produced working LFS systems on P3 and Pentium D > hardware. For three months I have tried to compile LFS/CLFS systemd on a > G4500 Pentium skylake system. I have concluded after 13 attempts that the > "skylake prime number hardware bug" makes it impossible. > I had forgotten about the reports of that bug - but a quick look shows that in complex mathematical calculations the processor may hang or cause unpredictable system behaviour.
Presumably it can be fixed by loading the latest microcode http://www.linuxfromscratch.org/blfs/view/stable/postlfs/firmware.html For that I point to the stable book because we have not changed that page - for everything else the development BLFS book is likely to be better. > My new hardware is well behaved as long as Xorg is used, not wayland. All > distros tried work well as long as a more current kernel is used. > However, I think you have different problems (plural). It us known that gmp tries to use all available CPU instructions. This has caused problems in the past when somebody builds on one machine and then tries to run on a less-modern machine. In your case, I suspect a similar problem - although your machine is a skylake, the Pentium variant has the minimal specification - I will not be surprised if some new instructions were omitted. > Tried CLFS the first eleven times (most to completion) on Debian-based, > Ubuntu, openSUSE and Arch-based. Switched to LFS with the same results: > Kernel messages regularly popping-up everywhere and an ethernet interface not > discovered by dbus. > If you can report some of the kernel messages, there might be a common pattern - or they might be unimportant and you just need to either remove something from your .config (e.g. a driver which reports it cannot find what it needs), or perhaps change the LOGLEVEL (no, I do not know how to change that in systemd systems). For the undiscovered ethernet - is there anything related to the network in dmesg or the kernel log ? Are you sure that the booted version of systemd has the same idea about the interface name as the version on the host system ? > My last attempt using 7.9 RC2 with openSUSE Leap failed testing in Chapter > 6.14, GMP-6.1.0. I ignored the failed tests last time through and produced > the same bad results as with CLFS. > > The output of the failure is: > > make check-TESTS > make[4]: Entering directory '/sources/gmp-6.1.0/tests/mpn' > make[5]: Entering directory '/sources/gmp-6.1.0/tests/mpn' > PASS: t-asmtype > PASS: t-aors_1 > PASS: t-divrem_1 > PASS: t-mod_1 > ../../test-driver: line 107: 8187 Illegal instruction "$@" > $log_file > 2>&1 And here, we see that whatever instruction gmp used is not acceptable to your CPU. If I am right, it is a bug in how gmp's configure script tests for a feature (or even in how the kernel sets the cpu info). I suggest that you recompile gmp after copying configfsf.guess and configfsf.sub over config.guess and config.sub. ĸen -- This email was written using 100% recycled letters. -- http://lists.linuxfromscratch.org/listinfo/lfs-support FAQ: http://www.linuxfromscratch.org/blfs/faq.html Unsubscribe: See the above information page Do not top post on this list. A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing? A: Top-posting. Q: What is the most annoying thing in e-mail? http://en.wikipedia.org/wiki/Posting_style
