3.8.3 build error, compiler segfault

[Nils Kassube]

Gene Heskett wrote:
> On Wednesday 27 March 2013 17:57:43 Nils Kassube did opine:
> > Gene Heskett wrote:
> > > Adding a few more echo's to the script, and verifying that every
> > > line
> > > in it was terminated with " && \\r"  I reran it, and see an error
> > > that should have stopped it right there, but did not:
> > > ==========================
> > > 
> > >   CC      net/ipv6/icmp.o
> > > 
> > > net/ipv4/icmp.c:1115: internal compiler error: Segmentation fault

> > Anyway, it is a strange fault if the compiler segfaults. Is it
> > repeatedly at that particular file? If it isn't, you should probably
> > check your RAM.
> > 
> > And if it is always the same file, you could try a newer version of
> > the compiler. If there is none for 10.04 you could use a newer
> > *Ubuntu version, possibly in a virtual machine.

> Gee Nils, that was a week or more ago,

Sorry for beeing too late, but I didn't see an expiry date for your 
inquiry. And it seems you're living much too fast - IIRC, four days is 
much less than a week, not more than a week. :)

> and yes it was repeatable on
> the same file, in the bluetooth tree IIRC.

net/ipv4/icmp.c doesn't look like bluetooth to me, more like the 
standard network stack. But the compiler crash ist still strange.

> So I backtracked to 3.0.19, and had that work, once.  But now I cannot
> make it work again because every time I touch the .config file with a
> make xconfig, the SOB nukes any and all mention of support for / on
> an ext4 formatted drive that needs 'sata_nv' built in as a starter. 

Are you sure that it needs 'sata_nv' built in? Wouldn't it be sufficient 
to add it to /etc/modules and run 

sudo update-initramfs -c -k $(uname -r)

afterwards? Or if want to use it with another kernel than the currently 
running kernel, use the appropriate kernel name instead of the "$(uname 
-r)". As I understand it, the module sata_nv would then be loaded from 
the initrd before the disks are accessed by the kernel.

> Even a make oldconfdig using this 2.6.32 kernels .config as a starter
> is a non-starter, falling over because it can't find the root drive's
> UUID about 4.9 seconds into the boot.  Same old same old for 2 newer
> kernels in between.  Its plain to me that something in the "make
> *config" barn is broken.
> 
> And its hell to try and get rid of the modules I don't need, but which
> take another hour to build, with just a 'gedit .config' session.

Who cares about the extra modules? They will only waste some MB on your 
disk, not in memory (if you don't load them). For me it takes more time 
to remove modules from the configuration than to build them.

> So ATM I am back on the stock (for linuxcnc) kernel, and swearing at
> its memory collection because with 4G of dram its 500 megs into swap
> in 3 or 4 days, running like a plenarium worm at the 25th mile of a
> marathon, so I have to do yet another *^%#$#@ reboot, so my head
> doesn't hurt so bad.

I don't think another kernel would reduce the amount of swap used. That 
is usually a matter of userland applications.


Nils