Re: Changing flash-kernel behaviour (Was: Support for sunxi-based ARM systems in d-i)
On Thu, 2014-05-01 at 16:28 +0200, Karsten Merker wrote:
> On Thu, May 01, 2014 at 09:43:14AM +0100, Ian Campbell wrote:
> > On Thu, 2014-05-01 at 08:36 +0100, Ian Campbell wrote:
> > >
> > > I think it would be better to put the actual kernel/initrd path (i.e.
> > > with the version suffix) directly into the boot.scr rather than
> > > creating
> > > a link to the kernel just to launch it with.
> > >
> > > I'd also like to see boot.scr-$version as the actual file and boot.scr
> > > as a link to the latest (with f-k creating boot.scr-$version for all
> > > installed versions). That's probably a separate project though
> Do all platforms flash-kernel cares for have their boot scripts
> on a filesystem that supports symlinks?
Ah yes. Well, that means it has to be a copy for those then I guess.
But for systems which are capable of booting from a normal extfs /boot I
think we should be doing away with this idea of mounting a magic
partition and just setting things up in the actual /boot properly.
> > BTW, the reason for this is that it would make it somewhat easier to
> > fallback to an older kernel on error, since you could just load and
> > source boot.scr-$version.
> That would indeed be nice, but it would come with a price. Would
> we be willing to drop support for an existing (although not
> officially Debian-supported) platform to achieve this? Copying
> instead of symlinking could of course be an alternative option -
> not particularly elegant but not dependent on filesystem
I think copying is fine given the constraints, so you would end up with
boot.scr-$version1 boot.scr-$version2 and boot.scr which was a copy of
the latest one.
I'd also be fine with only giving systems which can boot from the
actual /boot the benefit of this sort of flexibility.
> > BTW, the u-boot guys seem to want to converg on using either the
> > extlinux config file format or the BootloaderSpec as the standard
> > mechanism for configuring which kernel to use. THe former would probably
> > be easier to support (since we could just refactor update-extlinux out
> > of the existing x86 only package).
> >  http://www.freedesktop.org/wiki/Specifications/BootLoaderSpec/
> Hm, this spec mandates that the /boot partiton must be FAT.
> When looking at the discussion about Raspbian using FAT on /boot
> I doubt that Debian would implement that spec.
I think that is derived from the use of FAT for the system boot
partition on EFI. I suppose it is supportable using the same technique
as flash-kernel uses of copying stuff to a FAT partition which is
separate to /boot.