Angstrom, ep93xx, broken rootfs

Angstrom, ep93xx, broken rootfs

[Alex Williams]

Hello!
I decided to update my OpenEmbedded from git today. Result: broken
rootfs, many applications lost such as busybox and many other files,
symlinks. Tried to build minimal-image... One week before update I had
perfect rootfs and everything work fine... Resetting many commits
since last week did not helped me.
Details here: http://pastebin.com/m7ddd3941

Re: Angstrom, ep93xx, broken rootfs

[Koen Kooi]

On 22-02-09 19:34, Alex Williams wrote:
> Hello!
> I decided to update my OpenEmbedded from git today. Result: broken
> rootfs, many applications lost such as busybox and many other files,
> symlinks. Tried to build minimal-image... One week before update I had
> perfect rootfs and everything work fine... Resetting many commits
> since last week did not helped me.
> Details here: http://pastebin.com/m7ddd3941

That should be solved now, thanks for spotting!

/me grumbles at += vs _append " "

Re: Angstrom, ep93xx, broken rootfs

[Alex Williams]

> That should be solved now, thanks for spotting!
> /me grumbles at += vs _append " "

Now everything builds fine! Thank you for lightspeed reaction on this
problem! =)

ep93xx, ts72xx, gesbc-9302 and more (was Re: Angstrom, ep93xx, broken rootfs)

[Christian Gagneraud]

[this is a not so short email about EP-93xx CPU, EP-93xx based 
machine, and a request to have them officially supported by OE/Angstrom]

Alex Williams wrote:
> Hello!
> I decided to update my OpenEmbedded from git today. Result: broken
> rootfs, many applications lost such as busybox and many other files,
> symlinks. Tried to build minimal-image... One week before update I had
> perfect rootfs and everything work fine... Resetting many commits
> since last week did not helped me.
> Details here: http://pastebin.com/m7ddd3941

I'm just curious, which board do you use?

I would like to highlight an "issue" to oe developpers: The machine 
name "ep93xx" is completely misleading, because it refers to a CPU 
name not a board/machine name, the conf/machine/ep93xx.conf file says 
that this is for the "Cirrus Logic EP-93xx Development Platform", 
which is the vendor development board for the EP-93xx CPU, so the name 
should be something like "ep93xxevb" or "devkit-ep93xx", or whatever 
saying it refers to a dev board not to all EP-93xx based machine.

As well, there is a conf/machine/include/tune-ep9312.inc which is not 
used by any machine conf file, does anyone know why?
That's  a pity because this serie of CPU (EP93xx) has FP support 
support through maverick crunch. Which is supported by linux and gcc4, 
and this is exactly the purpose of this tune-ep9312.inc file (i did 
not test it yet, see below).

By greping "ep93" in conf/machine and packages/linux, it comes:
Known machine that use the EP-93xx CPU:
- ep93xx (the Cirrus Logic dev board)
- ts72xx (Technologic systems board)
- gesbc-9302 (
Kernel declared as EP93xx compatible (via COMPATIBLE_MACHINE = "ep93xx"):
- ep93xx-kernel_2.6.17+2.6.18-rc1.bb
- ep93xx-kernel_2.6.19+2.6.20-rc7.bb
Kernel compatible with EP93xx (via patch whose name contains "ep93"):
- linux_2.6.22+2.6.23-rc5
- linux_2.6.22.6.bb
- linux_2.6.24.bb

I'm currently working with a TS-7260 and a TS-7800 boards. I would 
like to contribute to oe:
- add machine conf file for ts7800 (based on marvell orion CPU)
- add a kernel "linux-ts" (vendor provided kernel, currently 
linux-ts-2.6.21 which support both ts72xx series and ts78xx series)
- add the latest kernel patch published on the ts7xxx mailing list, 
which bring support for kernel 2.6.28 on ts72xx boards, i was told 
that the this vanilla kernel fully support the ts7800 board, still 
need to test
- add the possibility to the user to select which kernel to use in 
local.conf, by changing the ts72xx.conf file with something like this:
PREFERRED_PROVIDER_virtual/kernel ?= "linux"
PREFERRED_VERSION_linux-ts        ?= "2.6.21"
PREFERRED_VERSION_linux           ?= "2.6.28"

With the above modification, I am currently able to build rootfs and 
boot into it with both kernels (linux-ts-2.6.21 and linux-2.6.28), and 
with both angstrom stable (2007.1) and dev (2009.X) branches

Later on, i would like to experiment with maverick crunch, to check if 
it is usable and really brings noticable FP performance. If so, then:
- rename the conf/machine/include/tune-ep9312.inc to 
conf/machine/include/tune-ep93xx.inc
- make the machine conf files use tune-ep93xx.inc instead of 
tune-arm920t.inc

Of course, i don't want to break anything, hence this email.
What people thinks about this?

Finally I would like to see the machines ts72xx and ts7800 officially 
supported by openembedded and/or angstrom (so that we can use 
Narcissus for example!;))

I'm willing to be a machine maintainer/mentor for the ts72xx and the 
ts7800. At work we use them in lot of our products, and we decided to 
switch to OE/Angstron distro (after encouraging test results).
What's the procedure for this exactly?

Regards,
Chris.

> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: ep93xx, ts72xx, gesbc-9302 and more (was Re: Angstrom, ep93xx, broken rootfs)

[Christian Gagneraud]

Christian Gagneraud wrote:
> [this is a not so short email about EP-93xx CPU, EP-93xx based machine, 
> and a request to have them officially supported by OE/Angstrom]
> 
> Alex Williams wrote:
>> Hello!
>> I decided to update my OpenEmbedded from git today. Result: broken
>> rootfs, many applications lost such as busybox and many other files,
>> symlinks. Tried to build minimal-image... One week before update I had
>> perfect rootfs and everything work fine... Resetting many commits
>> since last week did not helped me.
>> Details here: http://pastebin.com/m7ddd3941
> 
> I'm just curious, which board do you use?
> 
> I would like to highlight an "issue" to oe developpers: The machine name 
> "ep93xx" is completely misleading, because it refers to a CPU name not a 
> board/machine name, the conf/machine/ep93xx.conf file says that this is 
> for the "Cirrus Logic EP-93xx Development Platform", which is the vendor 
> development board for the EP-93xx CPU, so the name should be something 
> like "ep93xxevb" or "devkit-ep93xx", or whatever saying it refers to a 
> dev board not to all EP-93xx based machine.
> 
> As well, there is a conf/machine/include/tune-ep9312.inc which is not 
> used by any machine conf file, does anyone know why?

OK, found why! There's lot of trouble with maverick crunch, it seems 
to be buggy, and for now, nobody came with a solution to have it 
working correctly.

For those who are interested, a very good start page to understand the 
issues is  http://wiki.debian.org/ArmEabiMaverickCrunch.
A toolchain can be found here: http://simplemachines.it/tools/ (it 
seems to be the most active for the moment)

Regards,
Chris

> That's  a pity because this serie of CPU (EP93xx) has FP support support 
> through maverick crunch. Which is supported by linux and gcc4, and this 
> is exactly the purpose of this tune-ep9312.inc file (i did not test it 
> yet, see below).
> 
> By greping "ep93" in conf/machine and packages/linux, it comes:
> Known machine that use the EP-93xx CPU:
> - ep93xx (the Cirrus Logic dev board)
> - ts72xx (Technologic systems board)
> - gesbc-9302 (
> Kernel declared as EP93xx compatible (via COMPATIBLE_MACHINE = "ep93xx"):
> - ep93xx-kernel_2.6.17+2.6.18-rc1.bb
> - ep93xx-kernel_2.6.19+2.6.20-rc7.bb
> Kernel compatible with EP93xx (via patch whose name contains "ep93"):
> - linux_2.6.22+2.6.23-rc5
> - linux_2.6.22.6.bb
> - linux_2.6.24.bb
> 
> I'm currently working with a TS-7260 and a TS-7800 boards. I would like 
> to contribute to oe:
> - add machine conf file for ts7800 (based on marvell orion CPU)
> - add a kernel "linux-ts" (vendor provided kernel, currently 
> linux-ts-2.6.21 which support both ts72xx series and ts78xx series)
> - add the latest kernel patch published on the ts7xxx mailing list, 
> which bring support for kernel 2.6.28 on ts72xx boards, i was told that 
> the this vanilla kernel fully support the ts7800 board, still need to test
> - add the possibility to the user to select which kernel to use in 
> local.conf, by changing the ts72xx.conf file with something like this:
> PREFERRED_PROVIDER_virtual/kernel ?= "linux"
> PREFERRED_VERSION_linux-ts        ?= "2.6.21"
> PREFERRED_VERSION_linux           ?= "2.6.28"
> 
> With the above modification, I am currently able to build rootfs and 
> boot into it with both kernels (linux-ts-2.6.21 and linux-2.6.28), and 
> with both angstrom stable (2007.1) and dev (2009.X) branches
> 
> Later on, i would like to experiment with maverick crunch, to check if 
> it is usable and really brings noticable FP performance. If so, then:
> - rename the conf/machine/include/tune-ep9312.inc to 
> conf/machine/include/tune-ep93xx.inc
> - make the machine conf files use tune-ep93xx.inc instead of 
> tune-arm920t.inc
> 
> Of course, i don't want to break anything, hence this email.
> What people thinks about this?
> 
> Finally I would like to see the machines ts72xx and ts7800 officially 
> supported by openembedded and/or angstrom (so that we can use Narcissus 
> for example!;))
> 
> I'm willing to be a machine maintainer/mentor for the ts72xx and the 
> ts7800. At work we use them in lot of our products, and we decided to 
> switch to OE/Angstron distro (after encouraging test results).
> What's the procedure for this exactly?
> 
> Regards,
> Chris.
> 
>>
>> _______________________________________________
>> Openembedded-devel mailing list
>> Openembedded-devel@lists.openembedded.org
>> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
> 
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: ep93xx, ts72xx, gesbc-9302 and more (was Re: Angstrom, ep93xx, broken rootfs)

[Theodore A. Roth]

Hi Chris,

On Sun, Feb 22, 2009 at 1:43 PM, Christian Gagneraud
<cgagneraud@techworks.ie> wrote:

>
> Finally I would like to see the machines ts72xx and ts7800 officially
> supported by openembedded and/or angstrom (so that we can use Narcissus for
> example!;))
>
> I'm willing to be a machine maintainer/mentor for the ts72xx and the ts7800.
> At work we use them in lot of our products, and we decided to switch to
> OE/Angstron distro (after encouraging test results).
> What's the procedure for this exactly?
>
> Regards,
> Chris.
>

I too am working on adding support for the TS7800 to OE. I have only
been focusing on using vanilla linux kernel 2.6.28 for now. I have put
my branch of the OE sources up on github. I would love to collaborate
with you on getting this into the upstream OE tree. I also have a
TS7390 board I would like to contribute support in the future.

So far, I have been working on getting a very minimal initramfs rootfs
running off the SD card for the ts7800. I have been successful in
getting to a shell prompt using a uClibc Angstrom 2009 dev build.
There were a few problems I had to figure out to get that to build
though.

My OE git tree is available here:

  http://github.com/troth/oe-ts7xxx/tree/ts7800

I have a separate git tree for my local setup here:

  http://github.com/troth/oe-setup/tree/master

Ted Roth

Re: ep93xx, ts72xx, gesbc-9302 and more (was Re: Angstrom, ep93xx, broken rootfs)

[Marcin Juszkiewicz]

Dnia niedziela, 22 lutego 2009 o 22:43:34 Christian Gagneraud 
napisał(a):

> I would like to highlight an "issue" to oe developpers: The machine
> name "ep93xx" is completely misleading, because it refers to a CPU
> name not a board/machine name, the conf/machine/ep93xx.conf file says
> that this is for the "Cirrus Logic EP-93xx Development Platform",
> which is the vendor development board for the EP-93xx CPU, so the
> name should be something like "ep93xxevb" or "devkit-ep93xx", or
> whatever saying it refers to a dev board not to all EP-93xx based
> machine.

I use EDB9301 here. 2.6.29 kernel from OE should works on all EDB93xx 
boards made by Cirrus Logic (I enabled their support).

> As well, there is a conf/machine/include/tune-ep9312.inc which is not
> used by any machine conf file, does anyone know why?
> That's  a pity because this serie of CPU (EP93xx) has FP support
> support through maverick crunch. Which is supported by linux and
> gcc4, and this is exactly the purpose of this tune-ep9312.inc file (i
> did not test it yet, see below).

As you noted in next email the problem is with toolchain. The good thing 
is that Martin Guy and Hasjim Williams got it sorted. I merged their 
work recently into OpenEmbedded. By default "ep93xx" machine do not use 
it because cpu cycles are then wasted in context switches and there is 
no big speedup in normal applications (those which use heavy FP can be 
compiled for Maverick optimisations).

> I'm currently working with a TS-7260 and a TS-7800 boards.

> Later on, i would like to experiment with maverick crunch, to check
> if it is usable and really brings noticable FP performance. If so,
> then: - rename the conf/machine/include/tune-ep9312.inc to
> conf/machine/include/tune-ep93xx.inc
> - make the machine conf files use tune-ep93xx.inc instead of
> tune-arm920t.inc

tune-ep3912 is named like this because -march=ep3912 which it uses. 
Please update to recent OE and test MaverickCrunch on your board.


Regards, 
-- 
JID:      hrw@jabber.org
Website:  http://marcin.juszkiewicz.com.pl/
LinkedIn: http://www.linkedin.com/in/marcinjuszkiewicz