How do I select the version of gcc to use for a build?

I'm having some issues building a working kernel for the RPi under Denzil, which is using gcc 4.6.3. The 'official' kernel uses gcc 4.5.1.

I thought it may be worth trying gcc-csl-arm-2008q1, but I can't work out how to select this. I've tried setting

PREFERRED_VERSION_gcc = "csl-arm-2008q1"

in my machine.conf, but the build is still using 4.6.3.

Chris Tapp

opensource@...
www.keylevel.com

On 25 Jun 2012, at 20:56, Chris Tapp wrote:

How do I select the version of gcc to use for a build?

I'm having some issues building a working kernel for the RPi under Denzil, which is using gcc 4.6.3. The 'official' kernel uses gcc 4.5.1.

I thought it may be worth trying gcc-csl-arm-2008q1, but I can't work out how to select this. I've tried setting

PREFERRED_VERSION_gcc = "csl-arm-2008q1"

in my machine.conf, but the build is still using 4.6.3.

Got it - that should have been GCCVERSION= ...

However, that still doesn't give me a working toolchain as there are no gcc-runtime or libgcc versions to go with it.

So, the real question should be "Is there another toolchain I can use for an arm6? If there is, how do I select it?"

Chris Tapp

opensource@...
www.keylevel.com

On Mon, Jun 25, 2012 at 12:56 PM, Chris Tapp <opensource@...> wrote:

I'm having some issues building a working kernel for the RPi under Denzil, which is using gcc 4.6.3. The 'official' kernel uses gcc 4.5.1.

Do you have issues building it with 4.6.3 ?

Hi Khem,

On 25 Jun 2012, at 22:16, Khem Raj wrote:

On Mon, Jun 25, 2012 at 12:56 PM, Chris Tapp <opensource@...> wrote:

I'm having some issues building a working kernel for the RPi under Denzil, which is using gcc 4.6.3. The 'official' kernel uses gcc 4.5.1.

Do you have issues building it with 4.6.3 ?

Not sure. All I know is some of the kernel SRVREVs run with 4.6.3, some don't. The ones that don't will run if I build them will run using the download of the corresponding kernel image from the repository - these were built from the same source under 4.5.1. I thought it would be worth trying with another version.

4.5.1 would be the best bet, but that's not in Denzil. The BSP doesn't work with Edison, which does have 4.5.1. Not sure if the best bet is to modify the BSP or migrate 4.5.1 to Denzil (unfortunately not as easy as just copying the files).

Chris Tapp

opensource@...
www.keylevel.com

On Mon, Jun 25, 2012 at 2:24 PM, Chris Tapp <opensource@...> wrote:

On 25 Jun 2012, at 22:16, Khem Raj wrote:

On Mon, Jun 25, 2012 at 12:56 PM, Chris Tapp <opensource@...> wrote:

I'm having some issues building a working kernel for the RPi under Denzil, which is using gcc 4.6.3. The 'official' kernel uses gcc 4.5.1.

Do you have issues building it with 4.6.3 ?

Not sure. All I know is some of the kernel SRVREVs run with 4.6.3, some don't. The ones that don't will run if I build them will run using the download of the corresponding kernel image from the repository - these were built from the same source under 4.5.1. I thought it would be worth trying with another version.

4.5.1 would be the best bet, but that's not in Denzil. The BSP doesn't work with Edison, which does have 4.5.1. Not sure if the best bet is to modify the BSP or migrate 4.5.1 to Denzil (unfortunately not as easy as just copying the files).

well either way you have work to do. I think making kernel work with
4.6+ is forward looking and will be beneficial in future too. It would
be nice to know
whats the gcc 4.5 dependency, which version of kernel are we talking about ?

On 25 Jun 2012, at 22:29, Khem Raj wrote:

On Mon, Jun 25, 2012 at 2:24 PM, Chris Tapp <opensource@...> wrote:

On 25 Jun 2012, at 22:16, Khem Raj wrote:

On Mon, Jun 25, 2012 at 12:56 PM, Chris Tapp <opensource@...> wrote:

I'm having some issues building a working kernel for the RPi under Denzil, which is using gcc 4.6.3. The 'official' kernel uses gcc 4.5.1.

Do you have issues building it with 4.6.3 ?

Not sure. All I know is some of the kernel SRVREVs run with 4.6.3, some don't. The ones that don't will run if I build them will run using the download of the corresponding kernel image from the repository - these were built from the same source under 4.5.1. I thought it would be worth trying with another version.

4.5.1 would be the best bet, but that's not in Denzil. The BSP doesn't work with Edison, which does have 4.5.1. Not sure if the best bet is to modify the BSP or migrate 4.5.1 to Denzil (unfortunately not as easy as just copying the files).

well either way you have work to do. I think making kernel work with
4.6+ is forward looking and will be beneficial in future too. It would
be nice to know
whats the gcc 4.5 dependency, which version of kernel are we talking about ?

It's the rpi-patches branch at https://github.com/raspberrypi/linux.

Chris Tapp

opensource@...
www.keylevel.com

On Mon, Jun 25, 2012 at 2:37 PM, Chris Tapp <opensource@...> wrote:

It's the rpi-patches branch at https://github.com/raspberrypi/linux.

looks like 3.1.x
should be OK to compile with 4.6 I guess ..

On 25 Jun 2012, at 22:41, Khem Raj wrote:

On Mon, Jun 25, 2012 at 2:37 PM, Chris Tapp <opensource@...> wrote:

It's the rpi-patches branch at https://github.com/raspberrypi/linux.

looks like 3.1.x
should be OK to compile with 4.6 I guess ..

Yes, it's 3.1.9 and it does compile. But the latest commits don't run when I build them (system displays a 'failed to execute kernel' error screen). Paul Eggleton has failed to get any to work when built using 4.7.0 (current master), so it does look as if this is compiler-related.

None of the commits appear to be likely to cause a change from 'good' to 'bad'.

Chris Tapp

opensource@...
www.keylevel.com

On Mon, Jun 25, 2012 at 2:48 PM, Chris Tapp <opensource@...> wrote:

Yes, it's 3.1.9 and it does compile. But the latest commits don't run when I build them (system displays a 'failed to execute kernel' error screen). Paul Eggleton has failed to get any to work when built using 4.7.0 (current master), so it does look as if this is compiler-related.

None of the commits appear to be likely to cause a change from 'good' to 'bad'.

can you tell me which commit is last good ?
yes there are interaction between kernel and compiler
sometimes kernel code has wrong expectation of gcc and when it gets
fixed in gcc things dont work so in such cases kernel is fixed in some
case kernel may expose real gcc bug too. Cant tell much unless its narrowed down

in some case

On 25 Jun 2012, at 22:52, Khem Raj wrote:

On Mon, Jun 25, 2012 at 2:48 PM, Chris Tapp <opensource@...> wrote:

Yes, it's 3.1.9 and it does compile. But the latest commits don't run when I build them (system displays a 'failed to execute kernel' error screen). Paul Eggleton has failed to get any to work when built using 4.7.0 (current master), so it does look as if this is compiler-related.

None of the commits appear to be likely to cause a change from 'good' to 'bad'.

can you tell me which commit is last good ?
yes there are interaction between kernel and compiler
sometimes kernel code has wrong expectation of gcc and when it gets
fixed in gcc things dont work so in such cases kernel is fixed in some
case kernel may expose real gcc bug too. Cant tell much unless its narrowed down

Last one I've got to work is https://github.com/raspberrypi/linux/commit/3fc65f9f291fb19cad434968a4a0f66ab749e637.

I've opened an issue at https://github.com/raspberrypi/linux/commit/3fc65f9f291fb19cad434968a4a0f66ab749e637 which also shows the firmware version that I know works, but that's probably not important here.

I think I've got it building under 4.5.1 (Edison, as that was easier to do for a quick test).

Chris Tapp

opensource@...
www.keylevel.com

On Mon, Jun 25, 2012 at 2:57 PM, Chris Tapp <opensource@...> wrote:

On 25 Jun 2012, at 22:52, Khem Raj wrote:

On Mon, Jun 25, 2012 at 2:48 PM, Chris Tapp <opensource@...> wrote:

Yes, it's 3.1.9 and it does compile. But the latest commits don't run when I build them (system displays a 'failed to execute kernel' error screen). Paul Eggleton has failed to get any to work when built using 4.7.0 (current master), so it does look as if this is compiler-related.

None of the commits appear to be likely to cause a change from 'good' to 'bad'.

can you tell me which commit is last good ?
yes there are interaction between kernel and compiler
sometimes kernel code has wrong expectation of gcc and when it gets
fixed in gcc things dont work so in such cases kernel is fixed in some
case kernel may expose real gcc bug too. Cant tell much unless its narrowed down

Last one I've got to work is https://github.com/raspberrypi/linux/commit/3fc65f9f291fb19cad434968a4a0f66ab749e637.

I've opened an issue at https://github.com/raspberrypi/linux/commit/3fc65f9f291fb19cad434968a4a0f66ab749e637 which also shows the firmware version that I know works, but that's probably not important here.

well there are options in .config being changed that can mean that now
you could be waking up the sleeping dinosaurs.

I think I've got it building under 4.5.1 (Edison, as that was easier to do for a quick test).

OK

On 25 Jun 2012, at 23:05, Khem Raj wrote:

On Mon, Jun 25, 2012 at 2:57 PM, Chris Tapp <opensource@...> wrote:

On 25 Jun 2012, at 22:52, Khem Raj wrote:

On Mon, Jun 25, 2012 at 2:48 PM, Chris Tapp <opensource@...> wrote:

Yes, it's 3.1.9 and it does compile. But the latest commits don't run when I build them (system displays a 'failed to execute kernel' error screen). Paul Eggleton has failed to get any to work when built using 4.7.0 (current master), so it does look as if this is compiler-related.

None of the commits appear to be likely to cause a change from 'good' to 'bad'.

can you tell me which commit is last good ?
yes there are interaction between kernel and compiler
sometimes kernel code has wrong expectation of gcc and when it gets
fixed in gcc things dont work so in such cases kernel is fixed in some
case kernel may expose real gcc bug too. Cant tell much unless its narrowed down

Last one I've got to work is https://github.com/raspberrypi/linux/commit/3fc65f9f291fb19cad434968a4a0f66ab749e637.

I've opened an issue at https://github.com/raspberrypi/linux/commit/3fc65f9f291fb19cad434968a4a0f66ab749e637 which also shows the firmware version that I know works, but that's probably not important here.

well there are options in .config being changed that can mean that now
you could be waking up the sleeping dinosaurs.

Scary! (As my daughters would say).

I think I've got it building under 4.5.1 (Edison, as that was easier to do for a quick test).

OK

Well, that proves it's not the compiler. Still fails to boot, even though the toolchain that's supposed to work is 4.5.1. Does 32-bit / 64-bit host make a difference?

Chris Tapp

opensource@...
www.keylevel.com

On Mon, Jun 25, 2012 at 3:46 PM, Chris Tapp <opensource@...> wrote:

Well, that proves it's not the compiler. Still fails to boot, even though the toolchain that's supposed to work is 4.5.1. Does 32-bit / 64-bit host make a difference?

host should not matter although from my compiler experience I have
seen few cases where hosts injects bugs into cross compiler which
showed up on the target in compiled code but those cases were rare.

You should make sure that the compiler if the guys (who claim it works
) doesnt carry patches that are special.

On 26 Jun 2012, at 02:36, Khem Raj wrote:

On Mon, Jun 25, 2012 at 3:46 PM, Chris Tapp <opensource@...> wrote:

Well, that proves it's not the compiler. Still fails to boot, even though the toolchain that's supposed to work is 4.5.1. Does 32-bit / 64-bit host make a difference?

host should not matter although from my compiler experience I have
seen few cases where hosts injects bugs into cross compiler which
showed up on the target in compiled code but those cases were rare.

Yes, I've not really seen any issues in this area before.

You should make sure that the compiler if the guys (who claim it works
) doesnt carry patches that are special.

The tree that holds the tools (https://github.com/raspberrypi/tools) doesn't seem to have any source, so I'm assuming it's unmodified. However, I've asked the question just incase...

One thing I have noticed is the kernel images I build are less than half the size of the 'official' ones.

Thanks for all the help you've given me on this :-)

Chris Tapp

opensource@...
www.keylevel.com