Bug#858308: flash-kernel: documentation for upgrading Odroid-U2/U3 to flash-kernel
On Tue, Mar 21, 2017 at 06:45:30AM +0100, Geert Stappers wrote:
> On Tue, Mar 21, 2017 at 12:31:46AM +0100, Adam Borowski wrote:
> > If you're booting from a SD card, skip until running sd-fusing.sh
> > Edit sd_fusing.sh, change sector numbers to those listed in
> s/sd_fusing.sh/sd-fusing.sh/
> > ./sd-fusing.sh /dev/mmcblk0boot0
> s/sd-fusing.sh/sd_fusing.sh/
>
> Hyphen or underscore
D'oh!
> > what doesn't work
> > -----------------
> >
> > * rebooting: the machine gets stuck and needs to be physically power-cycled
BTW, Vagrant reports that rebooting _usually_ works on his U3, it rebooted
correctly only once on my U2. Never had such a problem with the vendor
kernel before.
> > * audio
> >
> > * the kernel whines about broken HDMI, video capture -- I can't even test
> > those but I don't care
>
> Please add 'Whoever can test HDMI, please report your milage'
Okay.
> > * the machine gets very hot; with 3.8 it never was even noticeably warm
>
> What does mean: Kernel 3.8 is good or bad? Please remove ambigous mean
I dropped this entire line: that "very hot" is up to 77⁰ (at 70⁰ right now
in the middle of a libreoffice build) and never tripped, thus I have nothing
but subjective feelings from touching it by hand. It's not unlikely the
difference comes from me having moved the machine, it was AFAIR resting on a
big whole-metal router before which could have drained heat well.
The fun thing is that all those years before I bought a fan for this box,
but because of delusions of replacing my main machine with this non-noisy
thing never actually mounted it. Could do it now and play with overclocking
to 2GHz as Samsung used to advertise the U2 can do...
New version attached.
Also, I'm not married to being the sole editor of this file, please feel
free to make any edits you want...
--
⢀⣴⠾⠻⢶⣦⠀ Meow!
⣾⠁⢠⠒⠀⣿⡁
⢿⡄⠘⠷⠚⠋⠀ Collisions shmolisions, let's see them find a collision or second
⠈⠳⣄⠀⠀⠀⠀ preimage for double rot13!
How to upgrade an Odroid-U2/U3 to the modern kernel.
====================================================
requirements
------------
While strictly speaking, with perfect luck you don't need any extra bits,
in reality you need at least:
* an eMMC reader (if you bought your eMMC from Hardkernel, the included
adapter works well in _some_ SD readers)
* a serial console (UART). If you don't have one yet, with shipping it
costs its weight in gold -- but fortunately it's not too heavy.
u-boot and other pre-kernel gubbins
-----------------------------------
apt install u-boot-exynos
wget http://odroid.in/guides/ubuntu-lfs/boot.tar.gz
Replace u-boot.bin from that tarball (2013 Samsung-private build) with
/usr/lib/u-boot/odroid/u-boot.bin (2016 vanilla), the other pieces are ok.
If you're booting from a SD card, skip until running sd_fusing.sh
On eMMC:
Edit sd_fusing.sh, change sector numbers to those listed in
/usr/share/doc/u-boot-exynos/README.odroid.gz :
.----
signed_bl1_position=0
bl2_position=30
uboot_position=62
tzsw_position=2110
`----
You may notice that these positions look wrong: on your eMMC (as originally
shipped) you have all those pieces in "SD" positions. Tʜᴏꜱᴇ ᴀʀᴇ ᴅᴇᴄᴏʏꜱ ᴍᴀᴅᴇ
ᴛᴏ ᴡᴀꜱᴛᴇ ʏᴏᴜʀ ᴛɪᴍᴇ! Replacing them does exactly nothing, the real bits are
on hidden parts of the eMMC.
There are rumours that it is possible to access them from a SD reader by a
"partition switch", but despite hours of searching I found no information
how to do that. The only way seems to be using the U2 itself. Which is a
dangerous thing as if anything goes wrong you won't be able to boot.
Fortunately, recovery is possible with a micro-USB cable and a SD card with
a special image: http://forum.odroid.com/viewtopic.php?f=53&t=969 -- I did
not have the need to test that, though.
On the U2, you need to unlock the secret partition first:
https://www.kernel.org/doc/Documentation/mmc/mmc-dev-parts.txt
.----
echo 0 > /sys/block/mmcblk0boot0/force_ro
`----
The device name does vary: it's /dev/mmcblk0 on some kernels (vendor and
Debian's 4.10-trunk-armmp), /dev/mmcblk1 on others (Debian 4.9 and vanilla
4.11-rc2). Only boot0 has anything, boot1 and rpmb are full of zeroes.
Run:
.----
./sd_fusing.sh /dev/mmcblk0boot0
`----
Watch the output -- it has no error detection at all and will lie that all
went ok even if it didn't!
Now pray and reboot.
u-boot console
--------------
Fortunately, unlike some other SoCs, modern u-boot _can_ boot both vendor
and mainline kernels. Be prepared to do some manual loading, though. For
convenience, here are relevant bits of old /boot/boot.scr:
.----
setenv bootcmd "fatload mmc 0:1 0x40008000 zImage; fatload mmc 0:1 0x42000000 uInitrd; bootm 0x40008000 0x42000000"
setenv bootargs "console=tty1 console=ttySAC1,115200n8 root=UUID=e139ce78-9841-40fe-8823-96a304a09859 rootwait ro mem=2047M"
`----
If boot fails, type the commands inside bootcmd manually, perhaps replacing
"fatload" with "ext4load" (also helpful "ext4ls mmc 0:1" and so on). The
"bootm" command is wrong: you need to replace it with "bootz".
fat -> ext4
-----------
Debian's kernel installation is heavily married to POSIX features that are
unavailable on fat. And the Odroid image uses fat for /boot (and so does
every single SoC image I've seen). Changing ln to cp and so on is not
enough, I've gave up and reformatted. Tar your /boot up, unmount, mkfs.ext4
/dev/mmcblk0p1, untar, recheck that the files are in /boot rather than
/boot/boot :þ then edit /etc/fstab (survivable with sysvinit, you're screwed
with systemd -- you can recover by editing the card via a reader). Reboot,
do the manual u-boot ritual.
Obviously, don't delete that tarball yet.
kernel
------
As kernel hooks fail if flash-kernel is unhappy, you need to unpack the
kernel image first. Grabbing linux-image-armmp from the Debian archive lets
you boot but nothing more: I didn't manage to get even ethernet to work, and
without network getting other images is unpleasant. The in-archive 4.9 also
sucks speed-wise compared to both vendor 3.8 and to linus/master. Thus,
let's take a kernel from that Linus guy. 4.11-rc2 works fine for me:
CPU test: gcc -O2 for(int i=0;i<1048576*256;i++) rand();
I/O test: dd if=/dev/mmcblk0 of=/dev/null bs=1048576 count=20000
CPU I/O
vendor 3.8 15s 51MB/s
Debian 4.9 18.8s 71MB/s
vanilla 4.11-rc2 15s 84MB/s
git clone https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
make exynos_defconfig
make menuconfig # edit to taste
make -j4 bindeb-pkg
dpkg -i just the image
flash-kernel
------------
apt install flash-kernel
When asked by debconf about kernel cmdline, copy the bootargs:
.----
console=tty1 console=ttySAC1,115200n8 root=UUID=e139ce78-9841-40fe-8823-96a304a09859 rootwait ro mem=2047M
`----
(with whatever extras you want).
Autodetection will fail, you need to:
.----
echo "Hardkernel ODROID-U3 board based on Exynos4412" >/etc/flash-kernel/machine
`----
(Both U2 and U3 use the same DT ("device tree" -- a machine description).)
flash-kernel, reboot. If it fails and you can't get it up via u-boot
console, use the SD/eMMC reader on your big box to restore boot.scr, zImage
and uInitrd from the /boot tarball you saved.
networking
----------
I didn't manage to get the network driver, smsc95xx over usbnet, to work on
Debian-shipped kernels. It works on vanilla 4.11-rc2.
Also it seems you need to change "allow-hotplug eth0" to "auto eth0" in
/etc/network/interfaces.
what doesn't work
-----------------
* rebooting: the machine gets stuck and needs to be physically power-cycled
* audio
* the kernel whines about broken HDMI, video capture -- I can't test those
(none of my monitors talk to Mali400 with any kernels), if someone can
test, please report
Reply to: