[debian] Re: How stupid I was to choose vfat for /boot

jidanni at jidanni.org jidanni at jidanni.org
Thu Nov 20 21:00:51 CET 2008

[Quoting more than needed so the smartphones-userland dudes see this
as Andy in not subscribed so his reply never got there.]

>>>>> "AG" == Andy Green <andy at openmoko.com> writes:

AG> Somebody in the thread at some point said:

AG> | I get it, Qi expects the kernel to be on the same partition as the
AG> | root file system, unlike that created by the Debian install.sh script.

AG> Yes exactly.

AG> | Conclusions: Debian team: notice that you'll have to ln or cp to
AG> | uImage-GTA02.bin and not just uImage.bin to match Qi, and of course it
AG> | should all be on the same partition... root and boot.

AG> There's another issue here to take note... the reason that Qi now has
AG> the device-specific kernel name is that there is an increasing number of
AG> devices that are interested in these same rootfs now.  GTA01, GTA02,
AG> GTA03 and eventually Glofiish and whatever else turns up can actually
AG> all use the same stock armv4 rootfs eventually, with kernels prebuilt
AG> into it for all supported devices: the bootloader will take the right
AG> kernel.

AG> So it means you can aim for one rootfs image for all devices, and even
AG> move an SD Card with the image between say GTA02 and GTA03.

AG> | As for booting into Debian from NOR, hmmm, well if Qi skips vfat
AG> | partitions, then we can still have that too.

AG> Yes.

(Users could copy kernels by hand to a now not usually mounted vfat
partition 1 as a backup kernel for booting from NOR.)

AG> | Several misconceptions on http://wiki.openmoko.org/wiki/Qi , "I'll try
AG> | to clear them up".

AG> Yes, thanks for cleaning it.

AG> | And Qi team: please turn on the verbosity, starting at the purple
AG> | screen can come after you've sold your first gold or platinum record
AG> | album of a telephone.
AG> |
AG> | I don't suppose Qi is flexible like grub2 or whatever, so we will just
AG> | have to cater to where it puts things.

AG> The U-Boot environment did not seem a good thing to me after GTA02
AG> experience, so I deliberately removed all persistent state private to
AG> the bootloader, its behaviour is deterministic then.  We literally had
AG> two devices from same factory with same nuked bootloader, kernel and
AG> rootfs that acted different by different U-Boot env, that is not what we
AG> need.

AG> I have patches above what's in git that introduce a "UI key" concept
AG> that you can control whether a "debugging" kernel commandline is used at
AG> boottime by holding down Aux, but unfortunately Qi dies if you use Aux
AG> right now with or without the patches, so solving the need for feedback
AG> about panic on LCM is "coming".

Let's think of ways the user can influence Qi's behavior while not
complicating Qi. (And not needing to remember to hold buttons.)
Perhaps if qi notices a file exists in the same boot directory,
qi_debug, it could turn on debugging. I assume qi cannot read files,
just detect their presence.

Also maybe different variations of the file name to make qi do
different things... of course then one wants to work in the factor of
upward compatibility... So perhaps QI_FFFF where FFFF are a hex bit
mask of debugging or other levels. And maybe this needs a -GTA02
appended too.

Anyway, I timed a Qi boot and a vfat+ext2 NOR boot, and they took
exactly the same time. 56 seconds to X windows appears, and at 69
seconds all is finished and we can type on the xvkbd. (Note I don't
have anything that uses its four antennas installed.) The only
difference is there is no messages with qi (bad).

More information about the openmoko-kernel mailing list