Re: [PATCH] Add linuxefi module

[Colin Watson]

On Mon, Jan 20, 2014 at 08:46:48PM -0500, SevenBits wrote:
> On Monday, January 20, 2014, Lubomir Rintel <address@hidden> wrote:
> > From: Matthew Garrett <address@hidden>
> >
> > This adds linuxefi module that provides a way to load Linux kernel
> > and RAM disk image via EFI services with linuxefi and initrdefi
> > commands, analogous to linux and initrd commands.
> 
> Why? What's wrong with the standard linux and initrd commands? They work
> just fine under UEFI.

The background to this is that if conditions permit it's helpful to hand
over to the kernel without calling ExitBootServices first, because it
allows the kernel to do more of its own quirks handling.  If shim is
present then it's used for signature verification first, since UEFI
Secure Boot forbids executing unsigned code before ExitBootServices;
although this patch is configured such that if shim is missing then no
signature check is performed (which is probably reasonable for
upstreaming).

We're carrying this patch in Debian/Ubuntu too, although I had to
disable it on i386_efi - I think it failed tests there.  It's a while
since I checked, and that patch might be obsolete now.  I also have an
additional fairly trivial patch to add more debugging printfs to
linuxefi, which I could apply if this is accepted.

I would be inclined to say that linuxefi should be essentially an
internal implementation detail, and that linux should forward to
linuxefi if appropriate.  I was never a particular fan of Matthew's
approach of adding an entirely new set of commands for it, and we don't
expose those in the configuration we generate in the Debian/Ubuntu
packaging.

-- 
Colin Watson                                       address@hidden