diff mbox

[1/1] deb-pkg: fix cross-compile build

Message ID 1304488750-13606-1-git-send-email-subtil@gmail.com (mailing list archive)
State New, archived
Headers show

Commit Message

Nuno Subtil May 4, 2011, 5:59 a.m. UTC
When generating the header package, scripts/package/builddeb was inferring the
target architecture by looking at the output of dpkg --print-architecture. This
allows KBUILD_DEBARCH to override that.

Signed-off-by: Nuno Subtil <subtil@gmail.com>
---
 scripts/package/builddeb |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)

Comments

Michal Marek May 4, 2011, 8:48 p.m. UTC | #1
Adding Maks to CC.

On Tue, May 03, 2011 at 10:59:10PM -0700, Nuno Subtil wrote:
> When generating the header package, scripts/package/builddeb was inferring the
> target architecture by looking at the output of dpkg --print-architecture. This
> allows KBUILD_DEBARCH to override that.
> 
> Signed-off-by: Nuno Subtil <subtil@gmail.com>
> ---
>  scripts/package/builddeb |    7 ++++++-
>  1 files changed, 6 insertions(+), 1 deletions(-)
> 
> diff --git a/scripts/package/builddeb b/scripts/package/builddeb
> index f6cbc3d..9372145 100644
> --- a/scripts/package/builddeb
> +++ b/scripts/package/builddeb
> @@ -246,7 +246,12 @@ mkdir -p "$destdir"
>  (cd $srctree; tar -c -f - -T /tmp/files$$) | (cd $destdir; tar -xf -)
>  (cd $objtree; tar -c -f - -T /tmp/objfiles$$) | (cd $destdir; tar -xf -)
>  rm -f /tmp/files$$ /tmp/objfiles$$
> -arch=$(dpkg --print-architecture)
> +
> +if [ -n "$KBUILD_DEBARCH" ] ; then
> +	arch="$KBUILD_DEBARCH"
> +else
> +	arch=$(dpkg --print-architecture)
> +fi
>  
>  cat <<EOF >> debian/control
>  
> -- 
> 1.7.4.4
> 
> Please Cc me on any replies related to this patch.
> 
> Thanks,
> Nuno
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
maximilian attems May 5, 2011, 9:07 a.m. UTC | #2
On Wed, May 04, 2011 at 10:48:07PM +0200, Michal Marek wrote:
> Adding Maks to CC.

thanks, this one looks, good please disguard the other patch.
 
> On Tue, May 03, 2011 at 10:59:10PM -0700, Nuno Subtil wrote:
> > When generating the header package, scripts/package/builddeb was inferring the
> > target architecture by looking at the output of dpkg --print-architecture. This
> > allows KBUILD_DEBARCH to override that.
> > 
> > Signed-off-by: Nuno Subtil <subtil@gmail.com>
Acked-by: maximilian attems <max@stro.at>
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Robert May 5, 2011, 3:42 p.m. UTC | #3
On 05/05/2011 02:07 AM, maximilian attems wrote:
> On Wed, May 04, 2011 at 10:48:07PM +0200, Michal Marek wrote:
>> Adding Maks to CC.
> thanks, this one looks, good please disguard the other patch.
>  
>> On Tue, May 03, 2011 at 10:59:10PM -0700, Nuno Subtil wrote:
>>> When generating the header package, scripts/package/builddeb was inferring the
>>> target architecture by looking at the output of dpkg --print-architecture. This
>>> allows KBUILD_DEBARCH to override that.
>>>
>>> Signed-off-by: Nuno Subtil <subtil@gmail.com>
> Acked-by: maximilian attems <max@stro.at>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
Greetings,

Compile breaks when using the following command:
    fakeroot make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- deb-pkg

With 'set -x' in scripts/package/builddeb, the relevant output is:
    + '[' -n '' ']'
    ++ dpkg --print-architecture
    + arch=i386
This indicates that KBUILD_DEBARCH is not set.

The debian/control file is written with the following section:
    Package: linux-headers-2.6.39-rc5+
    Provides: linux-headers, linux-headers-2.6
    Architecture: i386
    Description: Linux kernel headers for 2.6.39-rc5+ on i386
     This package provides kernel header files for 2.6.39-rc5+ on i386

This produces the following error output:
    + forcearch=-DArchitecture=armel
    + dpkg-gencontrol -isp -DArchitecture=armel
-plinux-headers-2.6.39-rc5+ -P/opt/kernel-linus/linux-2.6/debian/hdrtmp
    dpkg-gencontrol: error: current host architecture 'armel' does not
appear in package's architecture list (i386)

Regards,
Robert

--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Nuno Subtil May 5, 2011, 5:10 p.m. UTC | #4
This patch is not meant to fix that case, only the case where
KBUILD_DEBARCH is used.

KBUILD_DEBARCH seems to be meant to be used as an override for what
Debian wants to see as the target architecture. Arguably, the right
thing to do here might be to do the same arch detection sequence
that's done in create_package(), which looks at UTS_MACHINE first and
the KBUILD_DEBARCH override after that (though I'm not entirely sure).

This patch only adds the KBUILD_DEBARCH override, but that is still
better than the current situation. I can easily get the rest in later
if needed.

Nuno

On Thu, May 5, 2011 at 08:42, Robert Gordon
<robert@greenroomsoftware.com> wrote:
> On 05/05/2011 02:07 AM, maximilian attems wrote:
>> On Wed, May 04, 2011 at 10:48:07PM +0200, Michal Marek wrote:
>>> Adding Maks to CC.
>> thanks, this one looks, good please disguard the other patch.
>>
>>> On Tue, May 03, 2011 at 10:59:10PM -0700, Nuno Subtil wrote:
>>>> When generating the header package, scripts/package/builddeb was inferring the
>>>> target architecture by looking at the output of dpkg --print-architecture. This
>>>> allows KBUILD_DEBARCH to override that.
>>>>
>>>> Signed-off-by: Nuno Subtil <subtil@gmail.com>
>> Acked-by: maximilian attems <max@stro.at>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Greetings,
>
> Compile breaks when using the following command:
>    fakeroot make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- deb-pkg
>
> With 'set -x' in scripts/package/builddeb, the relevant output is:
>    + '[' -n '' ']'
>    ++ dpkg --print-architecture
>    + arch=i386
> This indicates that KBUILD_DEBARCH is not set.
>
> The debian/control file is written with the following section:
>    Package: linux-headers-2.6.39-rc5+
>    Provides: linux-headers, linux-headers-2.6
>    Architecture: i386
>    Description: Linux kernel headers for 2.6.39-rc5+ on i386
>     This package provides kernel header files for 2.6.39-rc5+ on i386
>
> This produces the following error output:
>    + forcearch=-DArchitecture=armel
>    + dpkg-gencontrol -isp -DArchitecture=armel
> -plinux-headers-2.6.39-rc5+ -P/opt/kernel-linus/linux-2.6/debian/hdrtmp
>    dpkg-gencontrol: error: current host architecture 'armel' does not
> appear in package's architecture list (i386)
>
> Regards,
> Robert
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Robert May 5, 2011, 6:34 p.m. UTC | #5
On 5/5/11 11:23 AM, Robert Gordon wrote:
> On 5/5/11 10:10 AM, Nuno Subtil wrote:
>> This patch is not meant to fix that case, only the case where
>> KBUILD_DEBARCH is used.
>>
>> KBUILD_DEBARCH seems to be meant to be used as an override for what
>> Debian wants to see as the target architecture. Arguably, the right
>> thing to do here might be to do the same arch detection sequence
>> that's done in create_package(), which looks at UTS_MACHINE first and
>> the KBUILD_DEBARCH override after that (though I'm not entirely sure).
>>
>> This patch only adds the KBUILD_DEBARCH override, but that is still
>> better than the current situation. I can easily get the rest in later
>> if needed.
>>
>> Nuno
>>
>> On Thu, May 5, 2011 at 08:42, Robert Gordon
>> <robert@greenroomsoftware.com> wrote:
>>> On 05/05/2011 02:07 AM, maximilian attems wrote:
>>>> On Wed, May 04, 2011 at 10:48:07PM +0200, Michal Marek wrote:
>>>>> Adding Maks to CC.
>>>> thanks, this one looks, good please disguard the other patch.
>>>>
>>>>> On Tue, May 03, 2011 at 10:59:10PM -0700, Nuno Subtil wrote:
>>>>>> When generating the header package, scripts/package/builddeb was
>>>>>> inferring the
>>>>>> target architecture by looking at the output of dpkg
>>>>>> --print-architecture. This
>>>>>> allows KBUILD_DEBARCH to override that.
>>>>>>
>>>>>> Signed-off-by: Nuno Subtil<subtil@gmail.com>
>>>> Acked-by: maximilian attems<max@stro.at>
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe
>>>> linux-kbuild" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>> Greetings,
>>>
>>> Compile breaks when using the following command:
>>> fakeroot make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- deb-pkg
>>>
>>> With 'set -x' in scripts/package/builddeb, the relevant output is:
>>> + '[' -n '' ']'
>>> ++ dpkg --print-architecture
>>> + arch=i386
>>> This indicates that KBUILD_DEBARCH is not set.
>>>
>>> The debian/control file is written with the following section:
>>> Package: linux-headers-2.6.39-rc5+
>>> Provides: linux-headers, linux-headers-2.6
>>> Architecture: i386
>>> Description: Linux kernel headers for 2.6.39-rc5+ on i386
>>> This package provides kernel header files for 2.6.39-rc5+ on i386
>>>
>>> This produces the following error output:
>>> + forcearch=-DArchitecture=armel
>>> + dpkg-gencontrol -isp -DArchitecture=armel
>>> -plinux-headers-2.6.39-rc5+ -P/opt/kernel-linus/linux-2.6/debian/hdrtmp
>>> dpkg-gencontrol: error: current host architecture 'armel' does not
>>> appear in package's architecture list (i386)
>>>
>>> Regards,
>>> Robert
>>>
>>>
> Nuno,
> Thanks for the response. I am very new at this process, but I am trying
> to learn. Yesterday, I submitted a patch that was intended to address
> the same issue that you have addressed. I did what you are suggesting by
> mimicking what is done in create_package(), but leaving i386/x86_64
> untouched. Here is the subject of the patch I sent:
>
> [PATCH] kbuild, deb-pkg: set host machine $arch correctly when
> cross-compiling or not
>
> Please advise if I should resend.
>
> Regards,
> Robert
>
> -- *
> *
>

Apologies, resending because messages were rejected by the mailing lists 
(probably due to html) - Robert
Nuno Subtil May 10, 2011, 5:29 a.m. UTC | #6
It looks like your patch duplicates the arch detection logic that's
already in create_package, except not completely. I think this is not
desirable.

I modified your patch to isolate this logic into a separate function
and respect the KBUILD_DEBARCH override (which mostly means I moved
the original logic into a separate function). This should make your
case work as well, besides respecting the KBUILD_DEBARCH override.

Maks, Michal, can you please take a look at this new patch?

Thanks,
Nuno

On Thu, May 5, 2011 at 11:23, Robert Gordon
<robert@greenroomsoftware.com> wrote:
> On 5/5/11 10:10 AM, Nuno Subtil wrote:
>
> This patch is not meant to fix that case, only the case where
> KBUILD_DEBARCH is used.
>
> KBUILD_DEBARCH seems to be meant to be used as an override for what
> Debian wants to see as the target architecture. Arguably, the right
> thing to do here might be to do the same arch detection sequence
> that's done in create_package(), which looks at UTS_MACHINE first and
> the KBUILD_DEBARCH override after that (though I'm not entirely sure).
>
> This patch only adds the KBUILD_DEBARCH override, but that is still
> better than the current situation. I can easily get the rest in later
> if needed.
>
> Nuno
>
> On Thu, May 5, 2011 at 08:42, Robert Gordon
> <robert@greenroomsoftware.com> wrote:
>
> On 05/05/2011 02:07 AM, maximilian attems wrote:
>
> On Wed, May 04, 2011 at 10:48:07PM +0200, Michal Marek wrote:
>
> Adding Maks to CC.
>
> thanks, this one looks, good please disguard the other patch.
>
> On Tue, May 03, 2011 at 10:59:10PM -0700, Nuno Subtil wrote:
>
> When generating the header package, scripts/package/builddeb was inferring
> the
> target architecture by looking at the output of dpkg --print-architecture.
> This
> allows KBUILD_DEBARCH to override that.
>
> Signed-off-by: Nuno Subtil <subtil@gmail.com>
>
> Acked-by: maximilian attems <max@stro.at>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
> Greetings,
>
> Compile breaks when using the following command:
>    fakeroot make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- deb-pkg
>
> With 'set -x' in scripts/package/builddeb, the relevant output is:
>    + '[' -n '' ']'
>    ++ dpkg --print-architecture
>    + arch=i386
> This indicates that KBUILD_DEBARCH is not set.
>
> The debian/control file is written with the following section:
>    Package: linux-headers-2.6.39-rc5+
>    Provides: linux-headers, linux-headers-2.6
>    Architecture: i386
>    Description: Linux kernel headers for 2.6.39-rc5+ on i386
>     This package provides kernel header files for 2.6.39-rc5+ on i386
>
> This produces the following error output:
>    + forcearch=-DArchitecture=armel
>    + dpkg-gencontrol -isp -DArchitecture=armel
> -plinux-headers-2.6.39-rc5+ -P/opt/kernel-linus/linux-2.6/debian/hdrtmp
>    dpkg-gencontrol: error: current host architecture 'armel' does not
> appear in package's architecture list (i386)
>
> Regards,
> Robert
>
>
> Nuno,
>      Thanks for the response.  I am very new at this process, but I am
> trying to learn.  Yesterday, I submitted a patch that was intended to
> address the same issue that you have addressed.  I did what you are
> suggesting by mimicking what is done in create_package(), but leaving
> i386/x86_64 untouched.  Here is the subject of the patch I sent:
>
> [PATCH] kbuild, deb-pkg: set host machine $arch correctly when
> cross-compiling or not
>
>      Please advise if I should resend.
>
> Regards,
> Robert
>
> --
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
maximilian attems May 10, 2011, 12:41 p.m. UTC | #7
On Mon, May 09, 2011 at 10:29:22PM -0700, Nuno Subtil wrote:
> It looks like your patch duplicates the arch detection logic that's
> already in create_package, except not completely. I think this is not
> desirable.
> 
> I modified your patch to isolate this logic into a separate function
> and respect the KBUILD_DEBARCH override (which mostly means I moved
> the original logic into a separate function). This should make your
> case work as well, besides respecting the KBUILD_DEBARCH override.
> 
> Maks, Michal, can you please take a look at this new patch?

thanks doing right now.

Michal please hold off until I have time to give it some runs.
Robert May 12, 2011, 2:01 p.m. UTC | #8
On 05/10/2011 05:41 AM, maximilian attems wrote:
> On Mon, May 09, 2011 at 10:29:22PM -0700, Nuno Subtil wrote:
>> It looks like your patch duplicates the arch detection logic that's
>> already in create_package, except not completely. I think this is not
>> desirable.
>>
>> I modified your patch to isolate this logic into a separate function
>> and respect the KBUILD_DEBARCH override (which mostly means I moved
>> the original logic into a separate function). This should make your
>> case work as well, besides respecting the KBUILD_DEBARCH override.
>>
>> Maks, Michal, can you please take a look at this new patch?
> thanks doing right now.
>
> Michal please hold off until I have time to give it some runs.
>
I have tested this patch.  It builds fine for arm with the following
command:
fakeroot make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- deb-pkg

However x86_64 fails (when using an i386 machine).
fakeroot make ARCH=x86_64 deb-pkg

The error occurs in scripts/package/builddeb
+ forcearch=-DArchitecture=amd64
+ dpkg-gencontrol -isp -DArchitecture=amd64 -plinux-headers-2.6.39-rc7+
-P/opt/kernel-linus/linux-2.6/debian/hdrtmp
dpkg-gencontrol: error: current host architecture 'i386' does not appear
in package's architecture list (amd64)

The debian/control file is as follows:
cat debian/control
Source: linux-upstream
Section: kernel
Priority: optional
Maintainer: Anonymous <root@robert-VGN-Z690Y>
Standards-Version: 3.8.4
Homepage: http://www.kernel.org/

Package: linux-image-2.6.39-rc7+
Provides: linux-image, linux-image-2.6, linux-modules-2.6.39-rc7+
Suggests: linux-firmware-image
Architecture: any
Description: Linux kernel, version 2.6.39-rc7+
 This package contains the Linux kernel, modules and corresponding other
 files, version: 2.6.39-rc7+.

Package: linux-headers-2.6.39-rc7+
Provides: linux-headers, linux-headers-2.6
Architecture: amd64
Description: Linux kernel headers for 2.6.39-rc7+ on amd64
 This package provides kernel header files for 2.6.39-rc7+ on amd64
 .
 This is useful for people who need to build external modules



The debian/control file for an unpatched scripts/package/builddeb would
assign 'i386' instead of 'amd64'.

Regards,
Robert

--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Nuno Subtil May 12, 2011, 7:45 p.m. UTC | #9
> The debian/control file for an unpatched scripts/package/builddeb would
> assign 'i386' instead of 'amd64'.

That seems wrong to me. Were the resulting packages actually usable on amd64?

On Thu, May 12, 2011 at 07:01, Robert Gordon
<robert@greenroomsoftware.com> wrote:
> On 05/10/2011 05:41 AM, maximilian attems wrote:
>> On Mon, May 09, 2011 at 10:29:22PM -0700, Nuno Subtil wrote:
>>> It looks like your patch duplicates the arch detection logic that's
>>> already in create_package, except not completely. I think this is not
>>> desirable.
>>>
>>> I modified your patch to isolate this logic into a separate function
>>> and respect the KBUILD_DEBARCH override (which mostly means I moved
>>> the original logic into a separate function). This should make your
>>> case work as well, besides respecting the KBUILD_DEBARCH override.
>>>
>>> Maks, Michal, can you please take a look at this new patch?
>> thanks doing right now.
>>
>> Michal please hold off until I have time to give it some runs.
>>
> I have tested this patch.  It builds fine for arm with the following
> command:
> fakeroot make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- deb-pkg
>
> However x86_64 fails (when using an i386 machine).
> fakeroot make ARCH=x86_64 deb-pkg
>
> The error occurs in scripts/package/builddeb
> + forcearch=-DArchitecture=amd64
> + dpkg-gencontrol -isp -DArchitecture=amd64 -plinux-headers-2.6.39-rc7+
> -P/opt/kernel-linus/linux-2.6/debian/hdrtmp
> dpkg-gencontrol: error: current host architecture 'i386' does not appear
> in package's architecture list (amd64)
>
> The debian/control file is as follows:
> cat debian/control
> Source: linux-upstream
> Section: kernel
> Priority: optional
> Maintainer: Anonymous <root@robert-VGN-Z690Y>
> Standards-Version: 3.8.4
> Homepage: http://www.kernel.org/
>
> Package: linux-image-2.6.39-rc7+
> Provides: linux-image, linux-image-2.6, linux-modules-2.6.39-rc7+
> Suggests: linux-firmware-image
> Architecture: any
> Description: Linux kernel, version 2.6.39-rc7+
>  This package contains the Linux kernel, modules and corresponding other
>  files, version: 2.6.39-rc7+.
>
> Package: linux-headers-2.6.39-rc7+
> Provides: linux-headers, linux-headers-2.6
> Architecture: amd64
> Description: Linux kernel headers for 2.6.39-rc7+ on amd64
>  This package provides kernel header files for 2.6.39-rc7+ on amd64
>  .
>  This is useful for people who need to build external modules
>
>
>
> The debian/control file for an unpatched scripts/package/builddeb would
> assign 'i386' instead of 'amd64'.
>
> Regards,
> Robert
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Robert May 13, 2011, 2:50 a.m. UTC | #10
On 5/12/11 12:45 PM, Nuno Subtil wrote:
>> The debian/control file for an unpatched scripts/package/builddeb would
>> assign 'i386' instead of 'amd64'.
>
> That seems wrong to me. Were the resulting packages actually usable on amd64?

I modeled my patch after behavior I saw in the unpatched script.  I had 
also originally set x86_64 to amd64 with the exact results I get with 
your patch.  So, I looked at what the unpatched kernel was doing, which 
is putting i386 in those fields when building for x86_64 on an i386 
machine.  I don't know the results would actually run because I don't 
have an x86_64 myself.  It may not run, but it does build.

cheers!
diff mbox

Patch

diff --git a/scripts/package/builddeb b/scripts/package/builddeb
index f6cbc3d..9372145 100644
--- a/scripts/package/builddeb
+++ b/scripts/package/builddeb
@@ -246,7 +246,12 @@  mkdir -p "$destdir"
 (cd $srctree; tar -c -f - -T /tmp/files$$) | (cd $destdir; tar -xf -)
 (cd $objtree; tar -c -f - -T /tmp/objfiles$$) | (cd $destdir; tar -xf -)
 rm -f /tmp/files$$ /tmp/objfiles$$
-arch=$(dpkg --print-architecture)
+
+if [ -n "$KBUILD_DEBARCH" ] ; then
+	arch="$KBUILD_DEBARCH"
+else
+	arch=$(dpkg --print-architecture)
+fi
 
 cat <<EOF >> debian/control