meta-linaro and toolchain-layer

meta-linaro and toolchain-layer

[Khem Raj]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi,

Ken Werner @ Linaro has created a meta-linaro layer. I have created a
github repo of same here https://github.com/kraj/meta-linaro

Moving forward soon gcc 4.7 will land in OE-Core and 4.6 will be
more or less retired. In natural progression it would move to
toolchain-layer if there are consumers of it.

linaro is also working on gcc-4.7 improvements in  same manner as
4.6  and there could be more tools e.g. gdb binutils etc. that could
be useful for armv7+ from that layer.

I would like to propose few things we could do.

Use gcc-4.6 from meta-linaro and retire it from toolchain-layer
which means toolchain layer only has gcc-4.5 and eventually
This also means that we don't move gcc-4.6 into toolchain-layer
when its relegated from OE-Core

In similar fashion add recipes for gcc-linaro-4.7 whenever linaro
makes first 4.7 based release

Does anybody see downsides to this approach ?

This could mean that distro may have multiple compilers for different
architectures e.g. linaro toolchain for arm
architecture and may be OE-Core provided toolchain for rest of them

another option is that we do the same thing where we cherry-pick
linaro patches into toolchain-layer and append them on top gcc from
OE-Core which we have been doing with 4.6 it does mean extra work
validation and maintenance that I would like to avoid.

Thanks

- -Khem
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk+ArNYACgkQuwUzVZGdMxQ1gACeNBUacwHJrvYuzpStJKeVZHb6
4qUAniohGmPBRhVJJtjEpGlG/JyKdCvt
=fnnP
-----END PGP SIGNATURE-----

Re: meta-linaro and toolchain-layer

[Denys Dmytriyenko]

On Sat, Apr 07, 2012 at 02:08:38PM -0700, Khem Raj wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hi,
> 
> Ken Werner @ Linaro has created a meta-linaro layer. I have created a
> github repo of same here https://github.com/kraj/meta-linaro

FWIW, I've been closely monitoring the progress of meta-linaro layer for past 
month and its status as applicable to meta-ti:

https://lists.yoctoproject.org/pipermail/meta-ti/2012-March/000747.html

When I tried it last time to build some of our machines in meta-ti, there were 
issues, but I didn't have time to follow up or investigate and I was busy with 
the BSP presentation at the time...

As a parallel effort, I was looking into working on the prebuilt toolchain 
support - i.e. recipes to use binary Linaro toolchain in OE/Yocto. I grabbed 
the latest 2012.03 release before leaving for the BSP Summit and Linux Collab 
and will give it a try ASAP.


> Moving forward soon gcc 4.7 will land in OE-Core and 4.6 will be
> more or less retired. In natural progression it would move to
> toolchain-layer if there are consumers of it.
> 
> linaro is also working on gcc-4.7 improvements in  same manner as
> 4.6  and there could be more tools e.g. gdb binutils etc. that could
> be useful for armv7+ from that layer.
> 
> I would like to propose few things we could do.
> 
> Use gcc-4.6 from meta-linaro and retire it from toolchain-layer
> which means toolchain layer only has gcc-4.5 and eventually
> This also means that we don't move gcc-4.6 into toolchain-layer
> when its relegated from OE-Core
> 
> In similar fashion add recipes for gcc-linaro-4.7 whenever linaro
> makes first 4.7 based release
> 
> Does anybody see downsides to this approach ?

Sounds like a decent plan, but we need to validate the newest toolchain 
versions with the real hardware before making the switch and dropping the old 
versions - 4.5 is still the most stable build for me and the broad range of 
architectures and platforms I work with at TI from ARM9 to Cortex-A15...


> This could mean that distro may have multiple compilers for different
> architectures e.g. linaro toolchain for arm
> architecture and may be OE-Core provided toolchain for rest of them
> 
> another option is that we do the same thing where we cherry-pick
> linaro patches into toolchain-layer and append them on top gcc from
> OE-Core which we have been doing with 4.6 it does mean extra work
> validation and maintenance that I would like to avoid.

-- 
Denys

Re: meta-linaro and toolchain-layer

[Khem Raj]

On Sun, Apr 8, 2012 at 5:43 PM, Denys Dmytriyenko <denis@denix.org> wrote:
> On Sat, Apr 07, 2012 at 02:08:38PM -0700, Khem Raj wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> Hi,
>>
>> Ken Werner @ Linaro has created a meta-linaro layer. I have created a
>> github repo of same here https://github.com/kraj/meta-linaro
>
> FWIW, I've been closely monitoring the progress of meta-linaro layer for past
> month and its status as applicable to meta-ti:
>
> https://lists.yoctoproject.org/pipermail/meta-ti/2012-March/000747.html
>
> When I tried it last time to build some of our machines in meta-ti, there were
> issues, but I didn't have time to follow up or investigate and I was busy with
> the BSP presentation at the time...

yes and I have tried to resolve those kinks. I am now able to build
systemd-image using
angstrom distro and the resulting image boots fine on qemuarm. It
needs a patch in
OE-Core first. which I will propose soon.

>
> As a parallel effort, I was looking into working on the prebuilt toolchain
> support - i.e. recipes to use binary Linaro toolchain in OE/Yocto. I grabbed
> the latest 2012.03 release before leaving for the BSP Summit and Linux Collab
> and will give it a try ASAP.

yes thats in there it just needs some testing

>
>
>> Moving forward soon gcc 4.7 will land in OE-Core and 4.6 will be
>> more or less retired. In natural progression it would move to
>> toolchain-layer if there are consumers of it.
>>
>> linaro is also working on gcc-4.7 improvements in  same manner as
>> 4.6  and there could be more tools e.g. gdb binutils etc. that could
>> be useful for armv7+ from that layer.
>>
>> I would like to propose few things we could do.
>>
>> Use gcc-4.6 from meta-linaro and retire it from toolchain-layer
>> which means toolchain layer only has gcc-4.5 and eventually
>> This also means that we don't move gcc-4.6 into toolchain-layer
>> when its relegated from OE-Core
>>
>> In similar fashion add recipes for gcc-linaro-4.7 whenever linaro
>> makes first 4.7 based release
>>
>> Does anybody see downsides to this approach ?
>
> Sounds like a decent plan, but we need to validate the newest toolchain
> versions with the real hardware before making the switch and dropping the old
> versions - 4.5 is still the most stable build for me and the broad range of
> architectures and platforms I work with at TI from ARM9 to Cortex-A15...
>

I dont plan to touch 4.5 recipes for now. It will remain in
toolchain-layer as it is.
its 4.6 onwards that I want us to be getting from meta-linaro.

>
>> This could mean that distro may have multiple compilers for different
>> architectures e.g. linaro toolchain for arm
>> architecture and may be OE-Core provided toolchain for rest of them
>>
>> another option is that we do the same thing where we cherry-pick
>> linaro patches into toolchain-layer and append them on top gcc from
>> OE-Core which we have been doing with 4.6 it does mean extra work
>> validation and maintenance that I would like to avoid.
>
> --
> Denys
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: meta-linaro and toolchain-layer

[Martin Jansa]

[-- Attachment #1: Type: text/plain, Size: 1573 bytes --]

On Sat, Apr 07, 2012 at 02:08:38PM -0700, Khem Raj wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hi,
> 
> Ken Werner @ Linaro has created a meta-linaro layer. I have created a
> github repo of same here https://github.com/kraj/meta-linaro
> 
> Moving forward soon gcc 4.7 will land in OE-Core and 4.6 will be
> more or less retired. In natural progression it would move to
> toolchain-layer if there are consumers of it.
> 
> linaro is also working on gcc-4.7 improvements in  same manner as
> 4.6  and there could be more tools e.g. gdb binutils etc. that could
> be useful for armv7+ from that layer.
> 
> I would like to propose few things we could do.
> 
> Use gcc-4.6 from meta-linaro and retire it from toolchain-layer
> which means toolchain layer only has gcc-4.5 and eventually
> This also means that we don't move gcc-4.6 into toolchain-layer
> when its relegated from OE-Core

I haven't tried gcc-4.7 on my armv4t machines yet, because I fear
similar issues like with binutils gold (which are still not resolved),
because nobody uses armv4t much nowadays :/.

The same I fear from 4.6 from meta-linaro, but maybe they will be
responsive as you were, when I reported issues on armv4t with some new
patch added by linaro..

So using meta-linaro for gcc-4.6 and gcc-4.7 is fine, but please keep
current "snapshot" of gcc-4.6 (oe-core+toolchain-layer patches)
somewhere as safe version before meta-linaro 4.6 is well tested on all
devices.

Cheers,

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 205 bytes --]

Re: meta-linaro and toolchain-layer

[Denys Dmytriyenko]

On Tue, Apr 10, 2012 at 08:41:31AM +0200, Martin Jansa wrote:
> On Sat, Apr 07, 2012 at 02:08:38PM -0700, Khem Raj wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> > 
> > Hi,
> > 
> > Ken Werner @ Linaro has created a meta-linaro layer. I have created a
> > github repo of same here https://github.com/kraj/meta-linaro
> > 
> > Moving forward soon gcc 4.7 will land in OE-Core and 4.6 will be
> > more or less retired. In natural progression it would move to
> > toolchain-layer if there are consumers of it.
> > 
> > linaro is also working on gcc-4.7 improvements in  same manner as
> > 4.6  and there could be more tools e.g. gdb binutils etc. that could
> > be useful for armv7+ from that layer.
> > 
> > I would like to propose few things we could do.
> > 
> > Use gcc-4.6 from meta-linaro and retire it from toolchain-layer
> > which means toolchain layer only has gcc-4.5 and eventually
> > This also means that we don't move gcc-4.6 into toolchain-layer
> > when its relegated from OE-Core
> 
> I haven't tried gcc-4.7 on my armv4t machines yet, because I fear
> similar issues like with binutils gold (which are still not resolved),
> because nobody uses armv4t much nowadays :/.
> 
> The same I fear from 4.6 from meta-linaro, but maybe they will be
> responsive as you were, when I reported issues on armv4t with some new
> patch added by linaro..
> 
> So using meta-linaro for gcc-4.6 and gcc-4.7 is fine, but please keep
> current "snapshot" of gcc-4.6 (oe-core+toolchain-layer patches)
> somewhere as safe version before meta-linaro 4.6 is well tested on all
> devices.

FWIW, Linaro only cares about actively supporting Cortex-A8 and up (armv7). 
While they promise to not break any older archs intentionally, they are not 
even testing them regularly. But if something gets accidentally broken, they 
will probably accept the fix...

-- 
Denys

Re: meta-linaro and toolchain-layer

[Khem Raj]

On Mon, Apr 9, 2012 at 11:41 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
>
> I haven't tried gcc-4.7 on my armv4t machines yet, because I fear
> similar issues like with binutils gold (which are still not resolved),
> because nobody uses armv4t much nowadays :/.

yes I have tried qemuarm/armv4t build along with qemuarm/armv5t build
as a test on 4.7
it boots systemd-image fine. But I would expect others to chime in and
provide some
testing help on their platforms/machines, I am still actively fixing
and rebasing
the 4.7 patches so expect a bit of churn there

>
> The same I fear from 4.6 from meta-linaro, but maybe they will be
> responsive as you were, when I reported issues on armv4t with some new
> patch added by linaro..

I will make sure that happens and I will be maintaining the layer at
above location
and see how the interaction goes and will help in maintaining the layer

>
> So using meta-linaro for gcc-4.6 and gcc-4.7 is fine, but please keep
> current "snapshot" of gcc-4.6 (oe-core+toolchain-layer patches)
> somewhere as safe version before meta-linaro 4.6 is well tested on all
> devices.

yes essentially we can recover is any time and in worst case I have to maintain
meta-linaro in which case we can merge both into one and maintain it
but I will actively
seek linaro's help in maintaining the layer.

>
> Cheers,

Re: meta-linaro and toolchain-layer

[Denys Dmytriyenko]

On Tue, Apr 10, 2012 at 03:48:26PM -0700, Khem Raj wrote:
> On Mon, Apr 9, 2012 at 11:41 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
> >
> > I haven't tried gcc-4.7 on my armv4t machines yet, because I fear
> > similar issues like with binutils gold (which are still not resolved),
> > because nobody uses armv4t much nowadays :/.
> 
> yes I have tried qemuarm/armv4t build along with qemuarm/armv5t build
> as a test on 4.7
> it boots systemd-image fine. But I would expect others to chime in and
> provide some
> testing help on their platforms/machines, I am still actively fixing
> and rebasing
> the 4.7 patches so expect a bit of churn there
> 
> >
> > The same I fear from 4.6 from meta-linaro, but maybe they will be
> > responsive as you were, when I reported issues on armv4t with some new
> > patch added by linaro..
> 
> I will make sure that happens and I will be maintaining the layer at
> above location

Why? Do we need another fork? Can you upstream your fixes?


> and see how the interaction goes and will help in maintaining the layer
> > So using meta-linaro for gcc-4.6 and gcc-4.7 is fine, but please keep
> > current "snapshot" of gcc-4.6 (oe-core+toolchain-layer patches)
> > somewhere as safe version before meta-linaro 4.6 is well tested on all
> > devices.
> 
> yes essentially we can recover is any time and in worst case I have to maintain
> meta-linaro in which case we can merge both into one and maintain it
> but I will actively
> seek linaro's help in maintaining the layer.

-- 
Denys

Re: meta-linaro and toolchain-layer

[Khem Raj]

On Wed, Apr 11, 2012 at 10:24 AM, Denys Dmytriyenko <denis@denix.org> wrote:
>
>
> Why? Do we need another fork? Can you upstream your fixes?

its not the intention. infact all changes will be upstreamed to one
place whatever that is
its a placeholder for development since the patch flow process for maintaining
this layer is still nascent.

Re: meta-linaro and toolchain-layer

[McClintock Matthew-B29882]

On Sat, Apr 7, 2012 at 4:08 PM, Khem Raj <raj.khem@gmail.com> wrote:
>
> Use gcc-4.6 from meta-linaro and retire it from toolchain-layer
> which means toolchain layer only has gcc-4.5 and eventually
> This also means that we don't move gcc-4.6 into toolchain-layer
> when its relegated from OE-Core

What is the harm in saving off the 4.6 toolchain recipes off in a
layer that we other users could later add back in if they needed it? I
for see issues where folks will insist on using the same toolchain
they have been using for years to come.

-M

Re: meta-linaro and toolchain-layer

[Khem Raj]

On Wed, Apr 11, 2012 at 8:59 AM, McClintock Matthew-B29882
<B29882@freescale.com> wrote:
>
> What is the harm in saving off the 4.6 toolchain recipes off in a
> layer that we other users could later add back in if they needed it? I
> for see issues where folks will insist on using the same toolchain
> they have been using for years to come.

usually the users who want that toolchain would than use a given
release of OE-Core
keeping many versions functional is a lot of work. Someone has to
constantly make
sure that metadata/recipes are working with the old toolchain all the
time master
changes.