[oe] Tosa kernel bloat

Richard Purdie rpurdie at rpsys.net
Mon Dec 31 23:18:59 UTC 2007


On Mon, 2007-12-31 at 21:10 +0100, Michael 'Mickey' Lauer wrote:
> >> Tosa kernel is bloated with the recent patch and change to
> >> defconfig-tosa
> >>
> >> |`/home/gordon/oe/build/tmp/work/tosa-angstrom-linux-gnueabi/linux-rp-2
> >> |.6.23-r14/linux-2.6.23' Size is 1296464
> >> | FATAL: This kernel is too big for your PXA Zaurus and will destroy
> >>
> >> data if you flash it. Please reduce the size of the kernel by making
> >> more of it modular.
> 
> > Same on Poodle:
> > | make[1]: Leaving directory
> > `/home/hrw/devel/build/angstrom/tmp/work/poodle-angstrom-linux-gnueabi/linux-rp-2.6.23-r14/linux-2.6.23'
> > | Size is 1306384
> > | FATAL: This kernel is too big for your PXA Zaurus and will destroy data
> > if you flash it. Please reduce the size of the kernel by making more of
> > it modular.
> 
> I vaguely remember that the host toolset has an influence on that. Do
> you remember where the same kernel compiled ok for Richard but other
> guys got the too-big error? Did we ever found the actual cause of
> that?

I don't think we did ever get to the bottom of that. It was a 64 bit
machine compiling for 32 bit that showed the problem.

I have to admit I haven't tested the size of the kernels, I have that
check disabled locally as I've been pushing kexec and relying on that. I
pushed a load of defconfig fixes for poodle and its size should be under
control now, I'm not sure about tosa.

We do have a problem with kernel sizes in recent kernels, one we can't
really afford to ignore any longer. I made a post to LKML about some
size tests I made. My conclusion is that the networking stack is the
point of bloat in the zaurus kernels now and its the area we need to
address. It seems its impossible to compile half the networking stack as
a module and I'd like to see that changed. Any ideas or help here
welcome!

Cheers,

Richard






More information about the Openembedded-devel mailing list