[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [RFC] kernel-headers and /lib/modules/*/build -> ?



Manoj Srivastava wrote:
> 
> >>"Michel" == Michel Dänzer <michdaen@iiic.ethz.ch> writes:
> 
>  Michel> The goal of my proposal is to enable the user to install a
>  Michel> kernel-image and the accompanying kernel-headers and when
>  Michel> running that kernel, just invoke make to build modules for
>  Michel> it.
> 
>  Michel> Now before I get flamed for the technical changes this would
>  Michel> involve, I'd like to ask if that's a worthwhile goal at
>  Michel> all. :)
> 
>         A initial cut at the changes involved:
>  1) the official kernel-image packages do not ship the symbolic link
>     /lib/modules/$(uname -r)/build  installed by the kernel makefiles
>  2) the kernel image package prerm removes the symlink, so that
>     official kernel image packages are removable (and do not leave
>     cruft dirs with a symlink around)
>  3) Instead of shipping a tarball of kernel headers, we ship unpacked
>     kernel headers. Looking at the Packages files today, installed
>     sizes of kernel-header packages range from 20MB to 32MB.
>  4) the kernel header postinst is enhanced to look for the link
>     /lib/modules/$VERSION/build, and, if the dir exists but the
>     symbolic link does not exist, create the symlink, pointing to the
>     unpacked kernel headers package.
> 
>         As far as I can see, this benefits people who
>  a) are running the official kerel image

I don't see the need for discrimination between un/official kernel packages in
this regard.

>  b) who want to build unpackaged kernel modules
>  c) who do not have the real kernel sources around

For example, I (like a lot of powerpc users) am following an rsync tree to
build my kernels, so I don't have the source for any but the current kernel.

>  d) are disinclined to unpack the kernel headers tarball, and use a -I
>     directive while compiling their kernel module.

It's a matter of convenience for the experienced user, but more importantly, a
matter of being or not being able to do it for the newbie. Of course, the
question remains if the latter does (or even should ;) use Debian.


>         I personally do not know anyone apart from Michel who actually
>  fits these constraints (I mean, I have 5 kernel source dirs on my
>  machine at the moment -- who needs kernel headers? And I rarely, if
>  ever, run tye kitchen sunk and everything in an initrd kernel images
>  we are officially producing now)
> 
>         I think that the ROI, given the niche use, may not justify the
>  changes involved; but I am willing to be convinced otherwise.

I hope this helps to convince you a bit that it's not that much of a niche,
unless you consider the powerpc architecture a niche. :)


-- 
Earthling Michel Dänzer (MrCooper)    \   Debian GNU/Linux (powerpc) developer
CS student, Free Software enthusiast   \        XFree86 and DRI project member



Reply to: