[OE-core] [PATCH] wic: Generate startup.nsh for EFI cases if none found

[Tom Rini]

On Fri, Sep 29, 2017 at 02:27:57PM +0300, Ed Bartosh wrote:
> On Thu, Sep 28, 2017 at 01:44:29PM -0400, Tom Rini wrote:
> > On Thu, Sep 28, 2017 at 06:47:07PM +0300, Ed Bartosh wrote:
> > > On Wed, Sep 20, 2017 at 12:03:27PM -0400, Tom Rini wrote:
> > > > In the case of non-wic images there is logic today to generate a
> > > > startup.nsh file that will be executed by EFI to run the loader that the
> > > > image contains.  In the WIC case is currently depends on that file being
> > > > generated elsewhere and placed in DEPLOY_DIR_IMAGE and only used if
> > > > present there.
> > > 
> > > What's wrong with this approach?
> > 
> > No one ever provides a startup.nsh and everyone that wants one creates
> > the same one line trivial example.  The end result is that no WIC images
> > are Just Bootable on UEFI systems unless you first go and spell that out
> > as the desired booting device.  This isn't an awesome workflow which is
> > why the non-WIC cases make the required startup.nsh :)
> 
> Would it be better if EFI providers create this file?
> 
> I still believe that wic should't hack the filesystem content unless
> it's really unavoidable. So far I know only one exception: updating
> /etc/fstab. And even that is not always needed (see --no-update-fstab
> patchset for further details.)

Well, it depends on your view of who is supposed to do what.  Today, in
wic BootimgEFIPlugin mirrors the efi_populate() function of
systemd-boot/grub-efi.bbclass.  That's where startup.nsh is made because
it needs to know the name of the EFI application (also technically the
path, but EFI\BOOT\ is spec mandatated I believe).  So we can't
easily make the deploy functions create startup.nsh without duplicating
logic from the bbclasss.

> > > I'd be happy to make wic to do only partitioning and assembling the
> > > image and avoiding to modify image content as much as possible.
> > > That would make wic design much more clear and let us to remove
> > > a lot of duplication between wic and meta/classes code.
> > > 
> > > Regarding boot partition content, I think preparing it from bootfs
> > > directory the same way as it's done for root partition is the way to go.
> > > You can find more details about it here:
> > > https://bugzilla.yoctoproject.org/show_bug.cgi?id=10073
> > 
> > I don't conceptually see a problem with going that route.  But today WIC
> > images aren't nearly as useful as they could be, with a rather tiny
> > change.
> 
> If we agree that wic should avoid modifying content then the obvious way
> to solve this is to provide required content (startup.nsh in this case)
> either by EFI related recipes or core classes.

Maybe I have to change my mind after thinking harder :)  Where's the
logic that creates the boot partition now?

> > My patch is also a regression-fix, I believe, in that at some point in
> > the past, when Christopher's patch went in, things were laid out such
> > that startup.nsh was often/always generated by another class and placed
> > where WIC would find it and copy it in.  At some point that was
> > broken/changed, and no one noticed / was interested enough to fix it.
> 
> If this functionality is covered by wic test suite this wouldn't
> happen.

Once we agree on what the fix looks like, I'll see if I can figure out
how to add in another test. :)

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20170929/975c1f8e/attachment-0002.sig>