[solved] Firefox and Seamonkey: internal compiler error

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

Hi there,

I'm using 2 computers:

Atom N270 Notebook
Xeon X5650 Workstation

I'm building big packages for my Atom notebook on my Xeon machine within a chroot environment. The Atom has the movbe instruction, which the Xeon doesn't understand. So I disabled that instruction for all concerning packages.

The last system-update had been done quite a long time ago. As I remember, everything worked fine. But this time I can't compile neither firefox nor seamonkey. The fail with the following error:

eccerr0r · Posted: Wed May 01, 2013 2:25 am Post subject:

This happens in the chroot?
Apparently that ld has some instruction the compiling machine doesn't understand, you may want to try rebuilding binutils with xeon instructions to see if it will go away (and hope itself won't crash during build). Since you're not building with the atom anyway, it's not like it will affect it that much...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Ant P. · Watchman Joined: 18 Apr 2009 Posts: 6920

"invalid opcode" points to using CFLAGS in the build system the host can't handle.

You might want to look into the emerge --root= option instead of an entire chroot, if you do all the building on the Xeon anyway.

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

I've got another suspicion:

As a part of the system-update, I upgraded gcc from 4.6.2 to 4.7.2.

I rebuilt libtool, binutils, glib, glibc. But that didn't help anything.

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

ok, the next step.

I ran locally on my Atom Notebook:

eccerr0r · Posted: Wed May 01, 2013 2:44 pm Post subject:

Yes, segfaults are usually hardware problems with gcc. but invalid instruction can also be hardware issues - if your ram/motherboard/cpu is corrupting instruction memory you'll get invalid instructions, but if corruption occurs in data memory, you'll tend to get segmentation faults or silent data corruption.

That is not to say gcc code writers are perfect coders and never write code with bugs... but unlikely...

(Incidentally, I make my Atom N270 netbook compile for itself (it runs Gentoo) but using other machines on the network for distcc. Other than disk space issues, it's not horribly bad. Then again I have 2GB RAM in this machine.)
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

Ok, to ensure, if it's really related to my chroot-environment, I tried to compile Firefox via distcc only on the Atom:

eccerr0r · Posted: Wed May 01, 2013 7:40 pm Post subject:

I'd say your Xeon may be having problems. If the Xeon is having segfault problems with the chroot, likely it will also have problems with distcc. If you ran without distcc on the Atom, perhaps the outcome would be different?

Does the Xeon compile its own stuff fine still? Are you using the same gcc for both machines?
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

eccerr0r · Posted: Thu May 02, 2013 9:20 pm Post subject:

I've got my toolchain, firefox, etc. all compiled just fine with gcc-4.6.3 with CFLAGS="-O2 -mtune=i686 -pipe".

It may well be that gcc-4.7.2 and its corresponding binutils that's broken, that's why it's marked unstable/testing. I don't know if there are any special CFLAGS being used, but this also tends to stress the compilers, possibly differently with different versions.

The compiler on the Xeon should be just fine emitting movbe instructions from the assembler, as long as it does not have to execute them...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

eccerr0r · Posted: Fri May 03, 2013 1:14 pm Post subject:

I was just pointing out: this seems to be wrong though it's not the problem at hand:

From your first post:

i686-pc-linux-gnu-g++ -o Stack.o -c -I./../../dist/system_wrappers_js -include /var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src/config/gcc_hidden.h -DENABLE_TYPEDARRAY_MOVE -DENABLE_YARR_JIT=1 -DMOZ_GLUE_IN_PROGRAM -DNO_NSPR_10_SUPPORT -DEXPORT_JS_API -DJS_HAS_CTYPES -DDLL_PREFIX=\"lib\" -DDLL_SUFFIX=\".so\" -DUSE_ZLIB -I/usr/lib/libffi-3.0.13/include -I. -I/var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src/../../mfbt/double-conversion -I/var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src -I. -I./../../dist/include -I/usr/include/nspr -I/var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src -I/var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src/assembler -I/var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src/yarr -fPIC -Wall -Wpointer-arith -Woverloaded-virtual -Werror=return-type -Wtype-limits -Wempty-body -Werror=conversion-null -Wno-ctor-dtor-privacy -Wno-overlength-strings -Wno-invalid-offsetof -Wno-variadic-macros -Wcast-align -march=atom -O2 -fomit-frame-pointer -pipe -mno-movbe -fno-rtti -ffunction-sections -fdata-sections -fno-exceptions -pthread -pipe -DNDEBUG -DTRIMMED -O2 -fomit-frame-pointer -DUSE_SYSTEM_MALLOC=1 -DENABLE_ASSEMBLER=1 -DENABLE_JIT=1 -DMOZILLA_CLIENT -include ./js-confdefs.h -MD -MF .deps/Stack.o.pp /var/tmp/portage/www-client/firefox-20.0.1/work/mozilla-release/js/src/vm/Stack.cpp

A bit contradictory I think... I would think either there's a problem the ebuild/makefile or something setup weird with your make.conf...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

eccerr0r · Posted: Fri May 03, 2013 5:16 pm Post subject:

Yes that's annoying that tools built by the packages themselves are using special instructions. That's why I tried to opt for using distcc whenever I can, which should avoid the problem and still use the special instructions tailored each individual CPU... except...

why I mostly just use -mtune=i686 nowadays. I got fed up with it when I had an emergency/temporary disk move when my athlonxp server's disk died. I put my p4's disk on it that was tuned for sse2, which ended up in more grief since now I had to rebuild the system while waiting to get a new disk (downtime). I would have to rebuild again when moving it back to the p4 if I didn't switch to i686...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

eccerr0r · Posted: Fri May 03, 2013 9:32 pm Post subject:

So you're actually physically transferring the disk over to do updates? I heard that NFS does not work for portage, wonder if there are any other network filesystems that do. Or how *are* you getting the images back and forth? Hopefully not rsync *shiver* ...

Really I'm lazy and the machines can go and build while I do something else, and it's still faster than local only when I use distcc for the most part (only exception is the linker phase in firefox because distcc cannot remote link). Distcc over wifi is pretty bad too, it is too slow and its latency bad - you have to use wired ethernet, preferably Gbit with small frames. 100Mbit is marginal but that's all I have on my Atom, and I have to make sure it's hooked. But yes my atom also has a disk problem which I end up just hooking up a USB HDD for just building firefox, pretty much all the other packages build just fine in 2GB RAM building in tmpfs.

I've found that qemu is pretty slow, but maybe not too bad (my test environment was the Maemo SDK/ubuntu-32 bit and running some contrived benchmark). On my Core2 Duo it about halves the speed (using KVM and VMX; it's unusable without VMX). I heard Virtualbox is better but I haven't tried it.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

musv · Advocate Joined: 01 Dec 2002 Posts: 3337 Location: de

Just to provide more information:

The problem is described in Bug #461694.

eccerr0r · Posted: Tue May 14, 2013 1:56 pm Post subject:

Heh. Another reason never to write cute code: