[PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Paul Gortmaker]

Using the 64 bit toolchains from kernel.org[1], one finds that
you can't build the a500_defconfig, because the Makefile will
stomp whatever value you have in your env. for CROSS_COMPILE.

This shows up since the kernel.org toolchains for parisc-64
do not have the "-gnu" prefix, and so you run into it always
saying hppa64-linux-gnu-gcc not found, regardless of the fact
you've not got "-gnu" put anywhere into CROSS_COMPILE.

[1] ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
index 55cca1d..cefff6d 100644
--- a/arch/parisc/Makefile
+++ b/arch/parisc/Makefile
@@ -31,7 +31,9 @@ ifdef CONFIG_64BIT
 UTS_MACHINE	:= parisc64
 CHECKFLAGS	+= -D__LP64__=1 -m64
 WIDTH		:= 64
-CROSS_COMPILE	:= hppa64-linux-gnu-
+ifndef CROSS_COMPILE
+ CROSS_COMPILE	:= hppa64-linux-gnu-
+endif
 else # 32-bit
 WIDTH		:=
 endif
-- 
1.7.9.1

Re: [PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 322 bytes --]

On Sunday 26 February 2012 18:26:53 Paul Gortmaker wrote:
> --- a/arch/parisc/Makefile
> +++ b/arch/parisc/Makefile
>
> -CROSS_COMPILE	:= hppa64-linux-gnu-
> +ifndef CROSS_COMPILE
> + CROSS_COMPILE	:= hppa64-linux-gnu-
> +endif

this should accomplish the same thing in practice:
	CROSS_COMPILE ?= hppa64-linux-gnu-
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Paul Gortmaker]

[Re: [PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.] On 26/02/2012 (Sun 19:23) Mike Frysinger wrote:

> On Sunday 26 February 2012 18:26:53 Paul Gortmaker wrote:
> > --- a/arch/parisc/Makefile
> > +++ b/arch/parisc/Makefile
> >
> > -CROSS_COMPILE	:= hppa64-linux-gnu-
> > +ifndef CROSS_COMPILE
> > + CROSS_COMPILE	:= hppa64-linux-gnu-
> > +endif
> 
> this should accomplish the same thing in practice:
> 	CROSS_COMPILE ?= hppa64-linux-gnu-

Good point, I'll wait for a bit in case I was missing something else
obvious, and then resend.

THanks,
P.

> -mike

Re: [PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[John David Anglin]

On 2/26/2012 6:26 PM, Paul Gortmaker wrote:
> Using the 64 bit toolchains from kernel.org[1], one finds that
> you can't build the a500_defconfig, because the Makefile will
> stomp whatever value you have in your env. for CROSS_COMPILE.
>
> This shows up since the kernel.org toolchains for parisc-64
> do not have the "-gnu" prefix, and so you run into it always
> saying hppa64-linux-gnu-gcc not found, regardless of the fact
> you've not got "-gnu" put anywhere into CROSS_COMPILE.
>
> [1] ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/
>
> Signed-off-by: Paul Gortmaker<paul.gortmaker@windriver.com>
>
> diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
> index 55cca1d..cefff6d 100644
> --- a/arch/parisc/Makefile
> +++ b/arch/parisc/Makefile
> @@ -31,7 +31,9 @@ ifdef CONFIG_64BIT
>   UTS_MACHINE	:= parisc64
>   CHECKFLAGS	+= -D__LP64__=1 -m64
>   WIDTH		:= 64
> -CROSS_COMPILE	:= hppa64-linux-gnu-
> +ifndef CROSS_COMPILE
> + CROSS_COMPILE	:= hppa64-linux-gnu-
> +endif
>   else # 32-bit
>   WIDTH		:=
>   endif
This change breaks native builds.  So, I assume CROSS_COMPILE must be
previously defined.

I couldn't figure out where it is defined.  CONFIG_CROSS_COMPILE in my
.config was an empty string  and CROSS_COMPILE was not defined in my
build environment.

The default toolchain prefix on Debian systems contains "gnu-".

Given that users can use CONFIG_CROSS_COMPILE or set CROSS_COMPILE
in their build env, maybe setting CROSS_COMPILE in the Makefile is a 
mistake.

Dave

-- 
John David Anglin    dave.anglin@bell.net

Re: [PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Paul Gortmaker]

On 12-02-27 10:14 AM, John David Anglin wrote:
> On 2/26/2012 6:26 PM, Paul Gortmaker wrote:
>> Using the 64 bit toolchains from kernel.org[1], one finds that
>> you can't build the a500_defconfig, because the Makefile will
>> stomp whatever value you have in your env. for CROSS_COMPILE.
>>
>> This shows up since the kernel.org toolchains for parisc-64
>> do not have the "-gnu" prefix, and so you run into it always
>> saying hppa64-linux-gnu-gcc not found, regardless of the fact
>> you've not got "-gnu" put anywhere into CROSS_COMPILE.
>>
>> [1] ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/
>>
>> Signed-off-by: Paul Gortmaker<paul.gortmaker@windriver.com>
>>
>> diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
>> index 55cca1d..cefff6d 100644
>> --- a/arch/parisc/Makefile
>> +++ b/arch/parisc/Makefile
>> @@ -31,7 +31,9 @@ ifdef CONFIG_64BIT
>>   UTS_MACHINE	:= parisc64
>>   CHECKFLAGS	+= -D__LP64__=1 -m64
>>   WIDTH		:= 64
>> -CROSS_COMPILE	:= hppa64-linux-gnu-
>> +ifndef CROSS_COMPILE
>> + CROSS_COMPILE	:= hppa64-linux-gnu-
>> +endif
>>   else # 32-bit
>>   WIDTH		:=
>>   endif
> This change breaks native builds.  So, I assume CROSS_COMPILE must be
> previously defined.

You lost me here.  As in a native 32bit env. doing a build for the
64 bit case is setting CROSS_COMPILE?

> 
> I couldn't figure out where it is defined.  CONFIG_CROSS_COMPILE in my
> .config was an empty string  and CROSS_COMPILE was not defined in my
> build environment.
> 
> The default toolchain prefix on Debian systems contains "gnu-".
> 
> Given that users can use CONFIG_CROSS_COMPILE or set CROSS_COMPILE
> in their build env, maybe setting CROSS_COMPILE in the Makefile is a 
> mistake.

Yeah, I'd support a complete nuke-and-pave of the whole Makefile
chunk as well.  It does seem redundant and causes more issues than
it solves.

Paul.

> 
> Dave
> 

Re: [PATCH] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[John David Anglin]

On 2/27/2012 11:28 AM, Paul Gortmaker wrote:
> On 12-02-27 10:14 AM, John David Anglin wrote:
>> >  On 2/26/2012 6:26 PM, Paul Gortmaker wrote:
>>> >>  Using the 64 bit toolchains from kernel.org[1], one finds that
>>> >>  you can't build the a500_defconfig, because the Makefile will
>>> >>  stomp whatever value you have in your env. for CROSS_COMPILE.
>>> >>
>>> >>  This shows up since the kernel.org toolchains for parisc-64
>>> >>  do not have the "-gnu" prefix, and so you run into it always
>>> >>  saying hppa64-linux-gnu-gcc not found, regardless of the fact
>>> >>  you've not got "-gnu" put anywhere into CROSS_COMPILE.
>>> >>
>>> >>  [1]ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/
>>> >>
>>> >>  Signed-off-by: Paul Gortmaker<paul.gortmaker@windriver.com>
>>> >>
>>> >>  diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
>>> >>  index 55cca1d..cefff6d 100644
>>> >>  --- a/arch/parisc/Makefile
>>> >>  +++ b/arch/parisc/Makefile
>>> >>  @@ -31,7 +31,9 @@ ifdef CONFIG_64BIT
>>> >>     UTS_MACHINE	:= parisc64
>>> >>     CHECKFLAGS	+= -D__LP64__=1 -m64
>>> >>     WIDTH		:= 64
>>> >>  -CROSS_COMPILE	:= hppa64-linux-gnu-
>>> >>  +ifndef CROSS_COMPILE
>>> >>  + CROSS_COMPILE	:= hppa64-linux-gnu-
>>> >>  +endif
>>> >>     else # 32-bit
>>> >>     WIDTH		:=
>>> >>     endif
>> >  This change breaks native builds.  So, I assume CROSS_COMPILE must be
>> >  previously defined.
> You lost me here.  As in a native 32bit env. doing a build for the
> 64 bit case is setting CROSS_COMPILE?
>
Yes, native hppa is 32 bit .  Building a 64-bit kernel with CONFIG_64BIT 
causes the
setting of CROSS_COMPILE and the selection of hppa64-linux-gnu-gcc, etc.
If CROSS_COMPILE wasn't previously defined, your change wouldn't have have
affected the 64-bit build.

Dave

-- 
John David Anglin    dave.anglin@bell.net

[PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Paul Gortmaker]

Using the 64 bit toolchains from kernel.org[1], one finds that
you can't build the a500_defconfig, because the Makefile will
stomp whatever value you have in your env. for CROSS_COMPILE.

This shows up since the kernel.org toolchains for parisc-64
do not have the "-gnu" prefix, and so you run into it always
saying hppa64-linux-gnu-gcc not found, regardless of the fact
you've not got "-gnu" put anywhere into CROSS_COMPILE.

Since you can set CROSS_COMPILE from the environment and/or set
it in your .config file, there really is no need for it to be
poked at by the Makefile at all.  So just delete it.

[1] ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
---

[v2: no ifdef/endif -- just delete the offending line.]

diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
index 55cca1d..a533367 100644
--- a/arch/parisc/Makefile
+++ b/arch/parisc/Makefile
@@ -31,7 +31,6 @@ ifdef CONFIG_64BIT
 UTS_MACHINE	:= parisc64
 CHECKFLAGS	+= -D__LP64__=1 -m64
 WIDTH		:= 64
-CROSS_COMPILE	:= hppa64-linux-gnu-
 else # 32-bit
 WIDTH		:=
 endif
-- 
1.7.9.1

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[James Bottomley]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 2149 bytes --]

On Mon, 2012-02-27 at 15:59 -0500, Paul Gortmaker wrote:
> Using the 64 bit toolchains from kernel.org[1], one finds that
> you can't build the a500_defconfig, because the Makefile will
> stomp whatever value you have in your env. for CROSS_COMPILE.
> 
> This shows up since the kernel.org toolchains for parisc-64
> do not have the "-gnu" prefix, and so you run into it always
> saying hppa64-linux-gnu-gcc not found, regardless of the fact
> you've not got "-gnu" put anywhere into CROSS_COMPILE.
> 
> Since you can set CROSS_COMPILE from the environment and/or set
> it in your .config file, there really is no need for it to be
> poked at by the Makefile at all.  So just delete it.
> 
> [1] ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/
> 
> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
> ---
> 
> [v2: no ifdef/endif -- just delete the offending line.]
> 
> diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
> index 55cca1d..a533367 100644
> --- a/arch/parisc/Makefile
> +++ b/arch/parisc/Makefile
> @@ -31,7 +31,6 @@ ifdef CONFIG_64BIT
>  UTS_MACHINE	:= parisc64
>  CHECKFLAGS	+= -D__LP64__=1 -m64
>  WIDTH		:= 64
> -CROSS_COMPILE	:= hppa64-linux-gnu-
>  else # 32-bit
>  WIDTH		:=
>  endif

We can't do that ... it crashes a standard build almost immediately
because it doesn't pick the 64 bit compiler (and we have no biarch
solution).

We need make on a 64 bit config to work out of the box.  It breaks
with ?= as JDA pointed out because CROSS_COMPILE is always defined.
Does this satisfy everyone?

James

---

diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
index 55cca1d..19ab7b2 100644
--- a/arch/parisc/Makefile
+++ b/arch/parisc/Makefile
@@ -31,7 +31,11 @@ ifdef CONFIG_64BIT
 UTS_MACHINE	:= parisc64
 CHECKFLAGS	+= -D__LP64__=1 -m64
 WIDTH		:= 64
+
+# FIXME: if no default set, should really try to locate dynamically
+ifeq ($(CROSS_COMPILE),)
 CROSS_COMPILE	:= hppa64-linux-gnu-
+endif
 else # 32-bit
 WIDTH		:=
 endif

ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[James Bottomley]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 3482 bytes --]

On Mon, 2012-02-27 at 21:41 +0000, James Bottomley wrote:
> On Mon, 2012-02-27 at 15:59 -0500, Paul Gortmaker wrote:
> > Using the 64 bit toolchains from kernel.org[1], one finds that
> > you can't build the a500_defconfig, because the Makefile will
> > stomp whatever value you have in your env. for CROSS_COMPILE.
> > 
> > This shows up since the kernel.org toolchains for parisc-64
> > do not have the "-gnu" prefix, and so you run into it always
> > saying hppa64-linux-gnu-gcc not found, regardless of the fact
> > you've not got "-gnu" put anywhere into CROSS_COMPILE.
> > 
> > Since you can set CROSS_COMPILE from the environment and/or set
> > it in your .config file, there really is no need for it to be
> > poked at by the Makefile at all.  So just delete it.
> > 
> > [1] ftp://ftp.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.5.2/
> > 
> > Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
> > ---
> > 
> > [v2: no ifdef/endif -- just delete the offending line.]
> > 
> > diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
> > index 55cca1d..a533367 100644
> > --- a/arch/parisc/Makefile
> > +++ b/arch/parisc/Makefile
> > @@ -31,7 +31,6 @@ ifdef CONFIG_64BIT
> >  UTS_MACHINE	:= parisc64
> >  CHECKFLAGS	+= -D__LP64__=1 -m64
> >  WIDTH		:= 64
> > -CROSS_COMPILE	:= hppa64-linux-gnu-
> >  else # 32-bit
> >  WIDTH		:=
> >  endif
> 
> We can't do that ... it crashes a standard build almost immediately
> because it doesn't pick the 64 bit compiler (and we have no biarch
> solution).
> 
> We need make on a 64 bit config to work out of the box.  It breaks
> with ?= as JDA pointed out because CROSS_COMPILE is always defined.
> Does this satisfy everyone?
> 
> James
> 
> ---
> 
> diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
> index 55cca1d..19ab7b2 100644
> --- a/arch/parisc/Makefile
> +++ b/arch/parisc/Makefile
> @@ -31,7 +31,11 @@ ifdef CONFIG_64BIT
>  UTS_MACHINE	:= parisc64
>  CHECKFLAGS	+= -D__LP64__=1 -m64
>  WIDTH		:= 64
> +
> +# FIXME: if no default set, should really try to locate dynamically
> +ifeq ($(CROSS_COMPILE),)
>  CROSS_COMPILE	:= hppa64-linux-gnu-
> +endif
>  else # 32-bit
>  WIDTH		:=
>  endif

There's an awful lot more than just this wrong with that Makefile.  For
instance, NATIVE can never be set because ifeq() isn't globbing in make
scripts ...  So I think a final (and dynamically choosing if unset)
solution is this one.

I'll stop now before I find something else wrong ...

James

---

diff --git a/arch/parisc/Makefile b/arch/parisc/Makefile
index 55cca1d..c823f4a 100644
--- a/arch/parisc/Makefile
+++ b/arch/parisc/Makefile
@@ -23,24 +23,20 @@ NM		= sh $(srctree)/arch/parisc/nm
 CHECKFLAGS	+= -D__hppa__=1
 
 MACHINE		:= $(shell uname -m)
-ifeq ($(MACHINE),parisc*)
-NATIVE		:= 1
-endif
 
 ifdef CONFIG_64BIT
 UTS_MACHINE	:= parisc64
 CHECKFLAGS	+= -D__LP64__=1 -m64
 WIDTH		:= 64
-CROSS_COMPILE	:= hppa64-linux-gnu-
+
+# if no default set, try to find the corresponding 64 bit compiler
+ifeq ($(CROSS_COMPILE),)
+CROSS_COMPILE	:= $(shell $(CC) -dumpmachine | sed s/hppa-\\\(.*\\\)/hppa64-\\1-/)
+endif
 else # 32-bit
 WIDTH		:=
 endif
 
-# attempt to help out folks who are cross-compiling
-ifeq ($(NATIVE),1)
-CROSS_COMPILE	:= hppa$(WIDTH)-linux-
-endif
-
 OBJCOPY_FLAGS =-O binary -R .note -R .comment -S
 
 cflags-y	:= -pipe



ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 532 bytes --]

On Monday 27 February 2012 17:03:10 James Bottomley wrote:
> --- a/arch/parisc/Makefile
> +++ b/arch/parisc/Makefile
> 
> -CROSS_COMPILE	:= hppa64-linux-gnu-
> +
> +# if no default set, try to find the corresponding 64 bit compiler
> +ifeq ($(CROSS_COMPILE),)
> +CROSS_COMPILE	:= $(shell $(CC) -dumpmachine | sed s/hppa-\\\(.*\\\)/hppa64-\\1-/)
> +endif

fails when using "hppa2.0-xxx".  perhaps:
CROSS_COMPILE ?= $(shell $(CC) -dumpmachine | sed 's/^hppa[^-]*-/hppa64-/')

don't even need the ifeq check then
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[James Bottomley]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 895 bytes --]

On Mon, 2012-02-27 at 20:04 -0500, Mike Frysinger wrote:
> On Monday 27 February 2012 17:03:10 James Bottomley wrote:
> > --- a/arch/parisc/Makefile
> > +++ b/arch/parisc/Makefile
> > 
> > -CROSS_COMPILE	:= hppa64-linux-gnu-
> > +
> > +# if no default set, try to find the corresponding 64 bit compiler
> > +ifeq ($(CROSS_COMPILE),)
> > +CROSS_COMPILE	:= $(shell $(CC) -dumpmachine | sed s/hppa-\\\(.*\\\)/hppa64-\\1-/)
> > +endif
> 
> fails when using "hppa2.0-xxx".  perhaps:
> CROSS_COMPILE ?= $(shell $(CC) -dumpmachine | sed 's/^hppa[^-]*-/hppa64-/')
> 
> don't even need the ifeq check then

How is that different from Carlos' suggestion?  The reason it doesn't
work is the same ... CROSS_COMPILE is defined to be empty.

James

ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 1211 bytes --]

On Monday 27 February 2012 22:17:01 James Bottomley wrote:
> On Mon, 2012-02-27 at 20:04 -0500, Mike Frysinger wrote:
> > On Monday 27 February 2012 17:03:10 James Bottomley wrote:
> > > --- a/arch/parisc/Makefile
> > > +++ b/arch/parisc/Makefile
> > > 
> > > -CROSS_COMPILE	:= hppa64-linux-gnu-
> > > +
> > > +# if no default set, try to find the corresponding 64 bit compiler
> > > +ifeq ($(CROSS_COMPILE),)
> > > +CROSS_COMPILE	:= $(shell $(CC) -dumpmachine | sed
> > > s/hppa-\\\(.*\\\)/hppa64-\\1-/) +endif
> > 
> > fails when using "hppa2.0-xxx".  perhaps:
> > CROSS_COMPILE ?= $(shell $(CC) -dumpmachine | sed
> > 's/^hppa[^-]*-/hppa64-/')
> > 
> > don't even need the ifeq check then
> 
> How is that different from Carlos' suggestion?

sorry, but i'm not seeing any e-mails from Carlos in this thread.  what is his 
suggestion ?

> The reason it doesn't work is the same ... CROSS_COMPILE is defined to be
> empty.

my suggestion was purely an incremental improvement on the patch you posted.  
it improves two things:
	- detection when the default tuple is hppa2.0-xxx rather than hppa-xxx
	- one line for set-variable-only-when-its-unset rather than three lines
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[James Bottomley]

On Tue, 2012-02-28 at 01:06 -0500, Mike Frysinger wrote:
> On Monday 27 February 2012 22:17:01 James Bottomley wrote:
> > On Mon, 2012-02-27 at 20:04 -0500, Mike Frysinger wrote:
> > > On Monday 27 February 2012 17:03:10 James Bottomley wrote:
> > > > --- a/arch/parisc/Makefile
> > > > +++ b/arch/parisc/Makefile
> > > > 
> > > > -CROSS_COMPILE	:= hppa64-linux-gnu-
> > > > +
> > > > +# if no default set, try to find the corresponding 64 bit compiler
> > > > +ifeq ($(CROSS_COMPILE),)
> > > > +CROSS_COMPILE	:= $(shell $(CC) -dumpmachine | sed
> > > > s/hppa-\\\(.*\\\)/hppa64-\\1-/) +endif
> > > 
> > > fails when using "hppa2.0-xxx".  perhaps:

Is that a valid prefix?  I was thinking hppa32 might be, but hppa2.0 is
supposed to be hppa-xxx with -mpa-risc-2-0.

But anyway, I think being clever and removing unused stuff is likely
orthogonal to the first patch which is just making it all work.

> > > CROSS_COMPILE ?= $(shell $(CC) -dumpmachine | sed
> > > 's/^hppa[^-]*-/hppa64-/')
> > > 
> > > don't even need the ifeq check then
> > 
> > How is that different from Carlos' suggestion?
> 
> sorry, but i'm not seeing any e-mails from Carlos in this thread.  what is his 
> suggestion ?

Actually, I misspoke, it was your suggestion in the first thread.

> > The reason it doesn't work is the same ... CROSS_COMPILE is defined to be
> > empty.
> 
> my suggestion was purely an incremental improvement on the patch you posted.  
> it improves two things:
> 	- detection when the default tuple is hppa2.0-xxx rather than hppa-xxx
> 	- one line for set-variable-only-when-its-unset rather than three lines
> -mike

It can't be done ... defined but empty is different from not defined in
makefiles (ifdef or ?= only checks for not defined).  CROSS_COMPILE gets
automatically defined by the top level Makefile ... that's why we need a
check for empty not a check for not defined.

James

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 2230 bytes --]

On Tuesday 28 February 2012 09:10:45 James Bottomley wrote:
> On Tue, 2012-02-28 at 01:06 -0500, Mike Frysinger wrote:
> > On Monday 27 February 2012 22:17:01 James Bottomley wrote:
> > > On Mon, 2012-02-27 at 20:04 -0500, Mike Frysinger wrote:
> > > > On Monday 27 February 2012 17:03:10 James Bottomley wrote:
> > > > > --- a/arch/parisc/Makefile
> > > > > +++ b/arch/parisc/Makefile
> > > > > 
> > > > > -CROSS_COMPILE	:= hppa64-linux-gnu-
> > > > > +
> > > > > +# if no default set, try to find the corresponding 64 bit compiler
> > > > > +ifeq ($(CROSS_COMPILE),)
> > > > > +CROSS_COMPILE	:= $(shell $(CC) -dumpmachine | sed
> > > > > s/hppa-\\\(.*\\\)/hppa64-\\1-/) +endif
> > > > 
> > > > fails when using "hppa2.0-xxx".  perhaps:
> 
> Is that a valid prefix?  I was thinking hppa32 might be, but hppa2.0 is
> supposed to be hppa-xxx with -mpa-risc-2-0.

we've been using hppa2.0 for years.  gnuconfig certainly recognizes it:
$ grep hppa /usr/share/gnuconfig/config.sub 
        | h8300 | h8500 | hppa | hppa1.[01] | hppa2.0 | hppa2.0[nw] | hppa64 \
        | hppa-* | hppa1.[01]-* | hppa2.0-* | hppa2.0[nw]-* | hppa64-* \

and i've seen diff projects key off of the hppa2.0* to enable newer features by 
default.

> But anyway, I think being clever and removing unused stuff is likely
> orthogonal to the first patch which is just making it all work.

sure, but your new patch added the sed :)

> > > The reason it doesn't work is the same ... CROSS_COMPILE is defined to
> > > be empty.
> > 
> > my suggestion was purely an incremental improvement on the patch you
> > posted.
> > 
> > it improves two things:
> > 	- detection when the default tuple is hppa2.0-xxx rather than hppa-xxx
> > 	- one line for set-variable-only-when-its-unset rather than three lines
> 
> It can't be done ... defined but empty is different from not defined in
> makefiles (ifdef or ?= only checks for not defined).  CROSS_COMPILE gets
> automatically defined by the top level Makefile ... that's why we need a
> check for empty not a check for not defined.

yeah, OK ... this behavior changed from what i was remembering.  ?= used to 
work until the top level Makefile was improved.
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[John David Anglin]

On 2/28/2012 10:21 AM, Mike Frysinger wrote:
> we've been using hppa2.0 for years.  gnuconfig certainly recognizes it:
> $ grep hppa /usr/share/gnuconfig/config.sub
>          | h8300 | h8500 | hppa | hppa1.[01] | hppa2.0 | hppa2.0[nw] | hppa64 \
>          | hppa-* | hppa1.[01]-* | hppa2.0-* | hppa2.0[nw]-* | hppa64-* \
>
> and i've seen diff projects key off of the hppa2.0* to enable newer features by
> default.
Some of the above are specific to HP-UX, but hppa1.1, hppa2.0, hppa and 
hppa64 are
valid for Linux based on config.guess.

Dave

-- 
John David Anglin    dave.anglin@bell.net

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 1477 bytes --]

On Tuesday 28 February 2012 13:22:50 Grant Grundler wrote:
> On Tue, Feb 28, 2012 at 7:33 AM, John David Anglin wrote:
> > On 2/28/2012 10:21 AM, Mike Frysinger wrote:
> >> we've been using hppa2.0 for years.  gnuconfig certainly recognizes it:
> >> $ grep hppa /usr/share/gnuconfig/config.**sub
> >> 
> >>         | h8300 | h8500 | hppa | hppa1.[01] | hppa2.0 | hppa2.0[nw] |
> >> 
> >> hppa64 \
> >> 
> >>         | hppa-* | hppa1.[01]-* | hppa2.0-* | hppa2.0[nw]-* | hppa64-* \
> >> 
> >> and i've seen diff projects key off of the hppa2.0* to enable newer
> >> features by
> >> default.
> > 
> > Some of the above are specific to HP-UX, but hppa1.1, hppa2.0, hppa and
> > hppa64 are
> > valid for Linux based on config.guess.
> 
> Mike, et al,
> I'd suggest NOT using hppa2.0 as a surrogate for 64-bit build. I'd prefer
> hppa64 if 64-bit is meant.
> 
> hppa2.0 can also mean PA2.0 compliant binary compiled using ILP32 model.
> right?

i don't think that's what the proposed change does.  James' patch runs $(CC) 
to find the active tuple, then replaces the first part with "hppa64" to get a 
64bit toolchain automatically.  but that magic only works when the tuple is 
"hppa-xxx" and so will fail when the system is using "hppa2.0-xxx".

short story: the proposed change will additionally support turning "hppa2.0-
linux-gnu" into "hppa64-linux-gnu" rather than the existing "only turns hppa-
linux-gnu" into "hppa64-linux-gnu".
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH v2] parisc: dont unconditionally override CROSS_COMPILE for 64 bit.

[Sam Ravnborg]

On Tue, Feb 28, 2012 at 01:37:10PM -0500, Mike Frysinger wrote:
> On Tuesday 28 February 2012 13:22:50 Grant Grundler wrote:
> > On Tue, Feb 28, 2012 at 7:33 AM, John David Anglin wrote:
> > > On 2/28/2012 10:21 AM, Mike Frysinger wrote:
> > >> we've been using hppa2.0 for years.  gnuconfig certainly recognizes it:
> > >> $ grep hppa /usr/share/gnuconfig/config.**sub
> > >> 
> > >>         | h8300 | h8500 | hppa | hppa1.[01] | hppa2.0 | hppa2.0[nw] |
> > >> 
> > >> hppa64 \
> > >> 
> > >>         | hppa-* | hppa1.[01]-* | hppa2.0-* | hppa2.0[nw]-* | hppa64-* \
> > >> 
> > >> and i've seen diff projects key off of the hppa2.0* to enable newer
> > >> features by
> > >> default.
> > > 
> > > Some of the above are specific to HP-UX, but hppa1.1, hppa2.0, hppa and
> > > hppa64 are
> > > valid for Linux based on config.guess.
> > 
> > Mike, et al,
> > I'd suggest NOT using hppa2.0 as a surrogate for 64-bit build. I'd prefer
> > hppa64 if 64-bit is meant.
> > 
> > hppa2.0 can also mean PA2.0 compliant binary compiled using ILP32 model.
> > right?
> 
> i don't think that's what the proposed change does.  James' patch runs $(CC) 
> to find the active tuple, then replaces the first part with "hppa64" to get a 
> 64bit toolchain automatically.  but that magic only works when the tuple is 
> "hppa-xxx" and so will fail when the system is using "hppa2.0-xxx".
> 
> short story: the proposed change will additionally support turning "hppa2.0-
> linux-gnu" into "hppa64-linux-gnu" rather than the existing "only turns hppa-
> linux-gnu" into "hppa64-linux-gnu".

There is a small helper:

    cc-cross-prefix

Which may be used ifyou have two or mroe candidated for CROSS_COMPILE.

>From mips:
 CROSS_COMPILE := $(call cc-cross-prefix, $(tool-archpref)-linux-  $(tool-archpref)-linux-gnu-  $(tool-archpref)-unknown-linux-gnu-)

Dunno if this would help parisc in this case..

	Sam