On Wed, 2021-01-13 at 11:47 -0600, Matthew Thode wrote:
> On 21-01-13 15:35:15, Michał Górny wrote:
> > Support dracut's uefi=yes configuration option that creates a combined
> > UEFI stub, kernel and initramfs in a single UEFI executable. If such
> > an output is detected, install it in place of the actual kernel image
> > and stub out the duplicate initrd to save space.
> >
> > Signed-off-by: Michał Górny <[email protected]>
> > ---
> > eclass/dist-kernel-utils.eclass | 31 ++++++++++++++++++++++++++++++-
> > 1 file changed, 30 insertions(+), 1 deletion(-)
> >
> > diff --git a/eclass/dist-kernel-utils.eclass
> > b/eclass/dist-kernel-utils.eclass
> > index d65dc0924b40..50ad15f8b1fd 100644
> > --- a/eclass/dist-kernel-utils.eclass
> > +++ b/eclass/dist-kernel-utils.eclass
> > @@ -41,8 +41,20 @@ dist-kernel_build_initramfs() {
> > local output=${1}
> > local version=${2}
> >
> >
> >
> >
> > + local rel_image_path=$(dist-kernel_get_image_path)
> > + local image=${output%/*}/${rel_image_path##*/}
> > +
> > + local args=(
> > + --force
> > + # if uefi=yes is used, dracut needs to locate the kernel image
> > + --kernel-image "${image}"
> > +
> > + # positional arguments
> > + "${output}" "${version}"
> > + )
> > +
> > ebegin "Building initramfs via dracut"
> > - dracut --force "${output}" "${version}"
> > + dracut "${args[@]}"
> > eend ${?} || die -n "Building initramfs failed"
> > }
> >
> >
> >
> >
> > @@ -85,6 +97,23 @@ dist-kernel_install_kernel() {
> > local image=${2}
> > local map=${3}
> >
> >
> >
> >
> > + # if dracut is used in eufi=yes mode, initrd will actually
> > + # be a combined kernel+initramfs UEFI executable. we can easily
> > + # recognize it by PE magic (vs cpio for a regular initramfs)
> > + local initrd=${image%/*}/initrd
> > + local magic
> > + [[ -s ${initrd} ]] && read -n 2 magic < "${initrd}"
> > + if [[ ${magic} == MZ ]]; then
>
> This magic header is matched by both the kernel and the new uefi
> executible. I think a better check would be...
>
> if file "${image_path}" | grep -q 'EFI application'; then
...but we're not testing image but initrd which can't be the kernel
otherwise.
>
> > + einfo "Combined UEFI kernel+initramfs executable found"
> > + # install the combined executable in place of kernel
> > + image=${initrd}.uefi
> > + mv "${initrd}" "${image}" || die
> > + # put an empty file in place of initrd. installing a duplicate
> > + # file would waste disk space, and removing it entirely provokes
> > + # kernel-install to regenerate it via dracut.
> > + > "${initrd}"
> > + fi
> > +
> > ebegin "Installing the kernel via installkernel"
> > # note: .config is taken relatively to System.map;
> > # initrd relatively to bzImage
> > --
> > 2.30.0
>
> I want to make sure you are not overwriting the original kernel image,
> if you do then subsiquent --config runs will include the kernel image,
> making the final image grow.
>
> In order to avoid that I did some file juggling, but it looks like it's
> cleaner here.
I still don't understand why it has been overwriting the kernel for you.
>
> https://gist.github.com/prometheanfire/aeffa1c3f92d3d2af312b3b6915051fb
> 'worked' but was kinda unclean.
>
>
--
Best regards,
Michał Górny
