PowerPC Cross-compile of 2.6 kernels

PowerPC Cross-compile of 2.6 kernels

[Judith Lebzelter]

Hello,

In response to requests at OLS, we've added cross-compile
capability to the PLM, and the first architecture
implemented is PowerPC.  The powerpc code is
generated via a cross-compiler set up using Dan
Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
glibc-2.3.2 and built for the powerpc-750.

The filter run is the compile regress developed by John
Cherry at OSDL.  Refer to his prior mail on lkml for the
results of this filter on ia386 and IA64.

Look at
    http://www.osdl.org/plm-cgi/plm?module=search
and look up linux-2.6.0-test5 or any later kernels for the
results of this filter under 'PPC-Cross Compile Regress'.

Does anyone have any input regarding requests for
additional architectures or improvements to the
filters?  Please cc me in any responses to lkml as I do
not currently monitor this list, though other OSDL
employees do.

Thanks;

Judith Lebzelter
OSDL

Re: PowerPC Cross-compile of 2.6 kernels

[Tom Rini]

[-- Attachment #1: Type: text/plain, Size: 1227 bytes --]

On Wed, Sep 10, 2003 at 04:41:05PM -0700, Judith Lebzelter wrote:
> Hello,
> 
> In response to requests at OLS, we've added cross-compile
> capability to the PLM, and the first architecture
> implemented is PowerPC.  The powerpc code is
> generated via a cross-compiler set up using Dan
> Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> glibc-2.3.2 and built for the powerpc-750.
> 
> The filter run is the compile regress developed by John
> Cherry at OSDL.  Refer to his prior mail on lkml for the
> results of this filter on ia386 and IA64.
> 
> Look at
>     http://www.osdl.org/plm-cgi/plm?module=search
> and look up linux-2.6.0-test5 or any later kernels for the
> results of this filter under 'PPC-Cross Compile Regress'.
> 
> Does anyone have any input regarding requests for
> additional architectures or improvements to the
> filters?  Please cc me in any responses to lkml as I do
> not currently monitor this list, though other OSDL
> employees do.

Is there any way to seed the config so that it will override what
allmodconfig (or allyesconfig) sets?  Both of these targets by nature
pick an 'odd' machine to compile for.

-- 
Tom Rini
http://gate.crashing.org/~trini/

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: PowerPC Cross-compile of 2.6 kernels

[Judith Lebzelter]

On Fri, 12 Sep 2003, Tom Rini wrote:

> On Wed, Sep 10, 2003 at 04:41:05PM -0700, Judith Lebzelter wrote:
> > Hello,
> >
> > In response to requests at OLS, we've added cross-compile
> > capability to the PLM, and the first architecture
> > implemented is PowerPC.  The powerpc code is
> > generated via a cross-compiler set up using Dan
> > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> > glibc-2.3.2 and built for the powerpc-750.
> >
> > The filter run is the compile regress developed by John
> > Cherry at OSDL.  Refer to his prior mail on lkml for the
> > results of this filter on ia386 and IA64.
> >
> > Look at
> >     http://www.osdl.org/plm-cgi/plm?module=search
> > and look up linux-2.6.0-test5 or any later kernels for the
> > results of this filter under 'PPC-Cross Compile Regress'.
> >
> > Does anyone have any input regarding requests for
> > additional architectures or improvements to the
> > filters?  Please cc me in any responses to lkml as I do
> > not currently monitor this list, though other OSDL
> > employees do.
>
> Is there any way to seed the config so that it will override what
> allmodconfig (or allyesconfig) sets?  Both of these targets by nature
> pick an 'odd' machine to compile for.
>

The 'allmodconfig' is what we run, and it chose
#
# Platform support
#
CONFIG_PPC=y
CONFIG_PPC32=y
CONFIG_6xx=y

#
# IBM 4xx options
#
CONFIG_EMBEDDEDBOOT=y
CONFIG_8260=y
CONFIG_PPC_STD_MMU=y
CONFIG_SERIAL_CONSOLE=y
# CONFIG_EST8260 is not set
# CONFIG_SBS8260 is not set
# CONFIG_RPX6 is not set
# CONFIG_TQM8260 is not set
CONFIG_WILLOW_1=y

I had to change this to do TQM8260 because the header file willow.h is
missing and the number of errors for this one file made the compile look
far worse than it actually was.  If you suggest something, I will
change it.

I also realize that my output should include any changes I make to the
configurations and have that changed right away.

We are also adding the 'allyesconfig' and will apply the same platform
changes as to 'allmodconfig'.

Thanks;

Judith
>

Re: PowerPC Cross-compile of 2.6 kernels

[Tom Rini]

On Fri, Sep 12, 2003 at 09:52:23AM -0700, Judith Lebzelter wrote:

> On Fri, 12 Sep 2003, Tom Rini wrote:
> 
> > On Wed, Sep 10, 2003 at 04:41:05PM -0700, Judith Lebzelter wrote:
> > > Hello,
> > >
> > > In response to requests at OLS, we've added cross-compile
> > > capability to the PLM, and the first architecture
> > > implemented is PowerPC.  The powerpc code is
> > > generated via a cross-compiler set up using Dan
> > > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> > > glibc-2.3.2 and built for the powerpc-750.
> > >
> > > The filter run is the compile regress developed by John
> > > Cherry at OSDL.  Refer to his prior mail on lkml for the
> > > results of this filter on ia386 and IA64.
> > >
> > > Look at
> > >     http://www.osdl.org/plm-cgi/plm?module=search
> > > and look up linux-2.6.0-test5 or any later kernels for the
> > > results of this filter under 'PPC-Cross Compile Regress'.
> > >
> > > Does anyone have any input regarding requests for
> > > additional architectures or improvements to the
> > > filters?  Please cc me in any responses to lkml as I do
> > > not currently monitor this list, though other OSDL
> > > employees do.
> >
> > Is there any way to seed the config so that it will override what
> > allmodconfig (or allyesconfig) sets?  Both of these targets by nature
> > pick an 'odd' machine to compile for.
> >
> 
> The 'allmodconfig' is what we run, and it chose
> #
> # Platform support
> #
> CONFIG_PPC=y
> CONFIG_PPC32=y
> CONFIG_6xx=y
> 
> #
> # IBM 4xx options
> #
> CONFIG_EMBEDDEDBOOT=y
> CONFIG_8260=y
> CONFIG_PPC_STD_MMU=y
> CONFIG_SERIAL_CONSOLE=y
> # CONFIG_EST8260 is not set
> # CONFIG_SBS8260 is not set
> # CONFIG_RPX6 is not set
> # CONFIG_TQM8260 is not set
> CONFIG_WILLOW_1=y
> 
> I had to change this to do TQM8260 because the header file willow.h is
> missing and the number of errors for this one file made the compile look
> far worse than it actually was.  If you suggest something, I will
> change it.

Don't select 8260 at all, and then the default choice should be
CONFIG_PPC_MULTIPLATFORM, which is normally the best choice.

I suppose this is more incentive for removing the rather bogus
distinction made for 8260 boards.

> I also realize that my output should include any changes I make to the
> configurations and have that changed right away.
> 
> We are also adding the 'allyesconfig' and will apply the same platform
> changes as to 'allmodconfig'.

Okay, thanks.

-- 
Tom Rini
http://gate.crashing.org/~trini/

Re: PowerPC Cross-compile of 2.6 kernels

[Judith Lebzelter]

On Fri, 12 Sep 2003, Tom Rini wrote:

> On Fri, Sep 12, 2003 at 09:52:23AM -0700, Judith Lebzelter wrote:
>
> > On Fri, 12 Sep 2003, Tom Rini wrote:
> >
> > > On Wed, Sep 10, 2003 at 04:41:05PM -0700, Judith Lebzelter wrote:
> > > > Hello,
> > > >
> > > > In response to requests at OLS, we've added cross-compile
> > > > capability to the PLM, and the first architecture
> > > > implemented is PowerPC.  The powerpc code is
> > > > generated via a cross-compiler set up using Dan
> > > > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> > > > glibc-2.3.2 and built for the powerpc-750.
> > > >
> > > > The filter run is the compile regress developed by John
> > > > Cherry at OSDL.  Refer to his prior mail on lkml for the
> > > > results of this filter on ia386 and IA64.
> > > >
> > > > Look at
> > > >     http://www.osdl.org/plm-cgi/plm?module=search
> > > > and look up linux-2.6.0-test5 or any later kernels for the
> > > > results of this filter under 'PPC-Cross Compile Regress'.
> > > >
> > > > Does anyone have any input regarding requests for
> > > > additional architectures or improvements to the
> > > > filters?  Please cc me in any responses to lkml as I do
> > > > not currently monitor this list, though other OSDL
> > > > employees do.
> > >
> > > Is there any way to seed the config so that it will override what
> > > allmodconfig (or allyesconfig) sets?  Both of these targets by nature
> > > pick an 'odd' machine to compile for.
> > >
> >
> > The 'allmodconfig' is what we run, and it chose
> > #
> > # Platform support
> > #
> > CONFIG_PPC=y
> > CONFIG_PPC32=y
> > CONFIG_6xx=y
> >
> > #
> > # IBM 4xx options
> > #
> > CONFIG_EMBEDDEDBOOT=y
> > CONFIG_8260=y
> > CONFIG_PPC_STD_MMU=y
> > CONFIG_SERIAL_CONSOLE=y
> > # CONFIG_EST8260 is not set
> > # CONFIG_SBS8260 is not set
> > # CONFIG_RPX6 is not set
> > # CONFIG_TQM8260 is not set
> > CONFIG_WILLOW_1=y
> >
> > I had to change this to do TQM8260 because the header file willow.h is
> > missing and the number of errors for this one file made the compile look
> > far worse than it actually was.  If you suggest something, I will
> > change it.
>
> Don't select 8260 at all, and then the default choice should be
> CONFIG_PPC_MULTIPLATFORM, which is normally the best choice.
>

I have changed it.  Thanks for the input.

I've also started a web page to summarize the ppc filter output for
2.6 base and mm kernel releases:
	http://www.osdl.org/archive/plm/ppc

Judith

> I suppose this is more incentive for removing the rather bogus
> distinction made for 8260 boards.
>
> > I also realize that my output should include any changes I make to the
> > configurations and have that changed right away.
> >
> > We are also adding the 'allyesconfig' and will apply the same platform
> > changes as to 'allmodconfig'.
>
> Okay, thanks.
>
>

Re: PowerPC Cross-compile of 2.6 kernels

[Tom Rini]

On Fri, Sep 12, 2003 at 01:13:48PM -0700, Judith Lebzelter wrote:
> On Fri, 12 Sep 2003, Tom Rini wrote:
> 
> > On Fri, Sep 12, 2003 at 09:52:23AM -0700, Judith Lebzelter wrote:
> >
> > > On Fri, 12 Sep 2003, Tom Rini wrote:
> > >
> > > > On Wed, Sep 10, 2003 at 04:41:05PM -0700, Judith Lebzelter wrote:
> > > > > Hello,
> > > > >
> > > > > In response to requests at OLS, we've added cross-compile
> > > > > capability to the PLM, and the first architecture
> > > > > implemented is PowerPC.  The powerpc code is
> > > > > generated via a cross-compiler set up using Dan
> > > > > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> > > > > glibc-2.3.2 and built for the powerpc-750.
> > > > >
> > > > > The filter run is the compile regress developed by John
> > > > > Cherry at OSDL.  Refer to his prior mail on lkml for the
> > > > > results of this filter on ia386 and IA64.
> > > > >
> > > > > Look at
> > > > >     http://www.osdl.org/plm-cgi/plm?module=search
> > > > > and look up linux-2.6.0-test5 or any later kernels for the
> > > > > results of this filter under 'PPC-Cross Compile Regress'.
> > > > >
> > > > > Does anyone have any input regarding requests for
> > > > > additional architectures or improvements to the
> > > > > filters?  Please cc me in any responses to lkml as I do
> > > > > not currently monitor this list, though other OSDL
> > > > > employees do.
> > > >
> > > > Is there any way to seed the config so that it will override what
> > > > allmodconfig (or allyesconfig) sets?  Both of these targets by nature
> > > > pick an 'odd' machine to compile for.
> > > >
> > >
> > > The 'allmodconfig' is what we run, and it chose
> > > #
> > > # Platform support
> > > #
> > > CONFIG_PPC=y
> > > CONFIG_PPC32=y
> > > CONFIG_6xx=y
> > >
> > > #
> > > # IBM 4xx options
> > > #
> > > CONFIG_EMBEDDEDBOOT=y
> > > CONFIG_8260=y
> > > CONFIG_PPC_STD_MMU=y
> > > CONFIG_SERIAL_CONSOLE=y
> > > # CONFIG_EST8260 is not set
> > > # CONFIG_SBS8260 is not set
> > > # CONFIG_RPX6 is not set
> > > # CONFIG_TQM8260 is not set
> > > CONFIG_WILLOW_1=y
> > >
> > > I had to change this to do TQM8260 because the header file willow.h is
> > > missing and the number of errors for this one file made the compile look
> > > far worse than it actually was.  If you suggest something, I will
> > > change it.
> >
> > Don't select 8260 at all, and then the default choice should be
> > CONFIG_PPC_MULTIPLATFORM, which is normally the best choice.
> >
> 
> I have changed it.  Thanks for the input.

Okay.  And this motivated me to fix the root issue, and make 8260 be
just another classic PPC set of boards.  I'll let you know when this
gets into Linus' tree.

-- 
Tom Rini
http://gate.crashing.org/~trini/

Re: PowerPC Cross-compile of 2.6 kernels

[Cort Dougan]

It would be nice to see the 4xx and 8xx chips being tested.  There are a
lot of rarely tested configurations and targets in the PPC kernel.

} > > > > > In response to requests at OLS, we've added cross-compile
} > > > > > capability to the PLM, and the first architecture
} > > > > > implemented is PowerPC.  The powerpc code is
} > > > > > generated via a cross-compiler set up using Dan
} > > > > > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
} > > > > > glibc-2.3.2 and built for the powerpc-750.
} > > > > >
} > > > > > The filter run is the compile regress developed by John
} > > > > > Cherry at OSDL.  Refer to his prior mail on lkml for the
} > > > > > results of this filter on ia386 and IA64.
} > > > > >
} > > > > > Look at
} > > > > >     http://www.osdl.org/plm-cgi/plm?module=search
} > > > > > and look up linux-2.6.0-test5 or any later kernels for the
} > > > > > results of this filter under 'PPC-Cross Compile Regress'.
} > > > > >
} > > > > > Does anyone have any input regarding requests for
} > > > > > additional architectures or improvements to the
} > > > > > filters?  Please cc me in any responses to lkml as I do
} > > > > > not currently monitor this list, though other OSDL
} > > > > > employees do.

Re: PowerPC Cross-compile of 2.6 kernels

[Judith Lebzelter]

On Fri, 12 Sep 2003, Cort Dougan wrote:

> It would be nice to see the 4xx and 8xx chips being tested.  There are a
> lot of rarely tested configurations and targets in the PPC kernel.

I could run a simpler compile filter, one compile that counts and
prints all the errors and warnings, on these extra platforms.  Did you
have any specific platform in mind?  40x give me 'walnut' and 8xx gives
me 'rpx-lite' if I do 'make defconfig' followed by 'make oldconfig' where
I set the main platform (to 40x or 8xx) and accept all the default
options.

Judith


>
> } > > > > > In response to requests at OLS, we've added cross-compile
> } > > > > > capability to the PLM, and the first architecture
> } > > > > > implemented is PowerPC.  The powerpc code is
> } > > > > > generated via a cross-compiler set up using Dan
> } > > > > > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> } > > > > > glibc-2.3.2 and built for the powerpc-750.
> } > > > > >
> } > > > > > The filter run is the compile regress developed by John
> } > > > > > Cherry at OSDL.  Refer to his prior mail on lkml for the
> } > > > > > results of this filter on ia386 and IA64.
> } > > > > >
> } > > > > > Look at
> } > > > > >     http://www.osdl.org/plm-cgi/plm?module=search
> } > > > > > and look up linux-2.6.0-test5 or any later kernels for the
> } > > > > > results of this filter under 'PPC-Cross Compile Regress'.
> } > > > > >
> } > > > > > Does anyone have any input regarding requests for
> } > > > > > additional architectures or improvements to the
> } > > > > > filters?  Please cc me in any responses to lkml as I do
> } > > > > > not currently monitor this list, though other OSDL
> } > > > > > employees do.
>

Re: PowerPC Cross-compile of 2.6 kernels

[Tom Rini]

On Mon, Sep 15, 2003 at 02:12:26PM -0700, Judith Lebzelter wrote:
> On Fri, 12 Sep 2003, Cort Dougan wrote:
> 
> > It would be nice to see the 4xx and 8xx chips being tested.  There are a
> > lot of rarely tested configurations and targets in the PPC kernel.
> 
> I could run a simpler compile filter, one compile that counts and
> prints all the errors and warnings, on these extra platforms.  Did you
> have any specific platform in mind?  40x give me 'walnut' and 8xx gives
> me 'rpx-lite' if I do 'make defconfig' followed by 'make oldconfig' where
> I set the main platform (to 40x or 8xx) and accept all the default
> options.

'Walnut' is a good 40x platform.  I wouldn't worry about 8xx right now,
however.

-- 
Tom Rini
http://gate.crashing.org/~trini/

Re: PowerPC Cross-compile of 2.6 kernels

[Tom Rini]

On Fri, Sep 12, 2003 at 02:32:43PM -0700, Tom Rini wrote:
> On Fri, Sep 12, 2003 at 01:13:48PM -0700, Judith Lebzelter wrote:
> > On Fri, 12 Sep 2003, Tom Rini wrote:
> > 
> > > On Fri, Sep 12, 2003 at 09:52:23AM -0700, Judith Lebzelter wrote:
> > >
> > > > On Fri, 12 Sep 2003, Tom Rini wrote:
> > > >
> > > > > On Wed, Sep 10, 2003 at 04:41:05PM -0700, Judith Lebzelter wrote:
> > > > > > Hello,
> > > > > >
> > > > > > In response to requests at OLS, we've added cross-compile
> > > > > > capability to the PLM, and the first architecture
> > > > > > implemented is PowerPC.  The powerpc code is
> > > > > > generated via a cross-compiler set up using Dan
> > > > > > Kegels's crosstool-0.22 on an i386 host using gcc-3.3.1,
> > > > > > glibc-2.3.2 and built for the powerpc-750.
> > > > > >
> > > > > > The filter run is the compile regress developed by John
> > > > > > Cherry at OSDL.  Refer to his prior mail on lkml for the
> > > > > > results of this filter on ia386 and IA64.
> > > > > >
> > > > > > Look at
> > > > > >     http://www.osdl.org/plm-cgi/plm?module=search
> > > > > > and look up linux-2.6.0-test5 or any later kernels for the
> > > > > > results of this filter under 'PPC-Cross Compile Regress'.
> > > > > >
> > > > > > Does anyone have any input regarding requests for
> > > > > > additional architectures or improvements to the
> > > > > > filters?  Please cc me in any responses to lkml as I do
> > > > > > not currently monitor this list, though other OSDL
> > > > > > employees do.
> > > > >
> > > > > Is there any way to seed the config so that it will override what
> > > > > allmodconfig (or allyesconfig) sets?  Both of these targets by nature
> > > > > pick an 'odd' machine to compile for.
> > > > >
> > > >
> > > > The 'allmodconfig' is what we run, and it chose
> > > > #
> > > > # Platform support
> > > > #
> > > > CONFIG_PPC=y
> > > > CONFIG_PPC32=y
> > > > CONFIG_6xx=y
> > > >
> > > > #
> > > > # IBM 4xx options
> > > > #
> > > > CONFIG_EMBEDDEDBOOT=y
> > > > CONFIG_8260=y
> > > > CONFIG_PPC_STD_MMU=y
> > > > CONFIG_SERIAL_CONSOLE=y
> > > > # CONFIG_EST8260 is not set
> > > > # CONFIG_SBS8260 is not set
> > > > # CONFIG_RPX6 is not set
> > > > # CONFIG_TQM8260 is not set
> > > > CONFIG_WILLOW_1=y
> > > >
> > > > I had to change this to do TQM8260 because the header file willow.h is
> > > > missing and the number of errors for this one file made the compile look
> > > > far worse than it actually was.  If you suggest something, I will
> > > > change it.
> > >
> > > Don't select 8260 at all, and then the default choice should be
> > > CONFIG_PPC_MULTIPLATFORM, which is normally the best choice.
> > >
> > 
> > I have changed it.  Thanks for the input.
> 
> Okay.  And this motivated me to fix the root issue, and make 8260 be
> just another classic PPC set of boards.  I'll let you know when this
> gets into Linus' tree.

As of 2.6.0-test6, allyesconfig (with no fiddling of other options)
on PPC is as sane as it's going to be. :)

-- 
Tom Rini
http://gate.crashing.org/~trini/

Re: PowerPC Cross-compile of 2.6 kernels

[Judith Lebzelter]

On Mon, 29 Sep 2003, Tom Rini wrote:

>
> As of 2.6.0-test6, allyesconfig (with no fiddling of other options)
> on PPC is as sane as it's going to be. :)
>

Does this mean that the results of the compile are not useful or even
confusing, since they do not fuss with configuration options?  Would it be
better not to run the 'allyesconfig' option?

Here are the posted results:
         http://www.osdl.org/archive/plm/ppc/


Judith

>

Re: PowerPC Cross-compile of 2.6 kernels

[Tom Rini]

On Mon, Sep 29, 2003 at 01:21:53PM -0700, Judith Lebzelter wrote:
> On Mon, 29 Sep 2003, Tom Rini wrote:
> 
> >
> > As of 2.6.0-test6, allyesconfig (with no fiddling of other options)
> > on PPC is as sane as it's going to be. :)
> >
> 
> Does this mean that the results of the compile are not useful or even
> confusing, since they do not fuss with configuration options?  Would it be
> better not to run the 'allyesconfig' option?

Well, on PPC we get CONFIG_ISA, which is useful for a few drivers, but I
suspect that it will enable a number of 'other' drivers as well.  I'm
not sure what would be more useful however.

-- 
Tom Rini
http://gate.crashing.org/~trini/