Patchwork [U-Boot] Makefile: Introduce ARCH_PLATFORM_LIBGCC variable

login
register
mail settings
Submitter Tom Rini
Date Oct. 11, 2013, 8:13 p.m.
Message ID <1381522383-25070-1-git-send-email-trini@ti.com>
Download mbox | patch
Permalink /patch/282928/
State Changes Requested
Delegated to: Albert ARIBAUD
Headers show

Comments

Tom Rini - Oct. 11, 2013, 8:13 p.m.
In some cases, such as arm multi-lib hardfloat (hf) toolchains, we will
have multiple libgcc.a's available, and the arch needs to provide
additional logic to determine the right file to use
(-print-libgcc-file-name contains no CFLAG parsing logic).

Cc: Albert Aribaud <albert.u.boot@aribaud.net>
Signed-off-by: Tom Rini <trini@ti.com>
---
 Makefile           |    4 ++++
 arch/arm/config.mk |   11 +++++++++++
 2 files changed, 15 insertions(+)
Tom Rini - Oct. 11, 2013, 9:01 p.m.
On Fri, Oct 11, 2013 at 04:13:03PM -0400, Tom Rini wrote:

> In some cases, such as arm multi-lib hardfloat (hf) toolchains, we will
> have multiple libgcc.a's available, and the arch needs to provide
> additional logic to determine the right file to use
> (-print-libgcc-file-name contains no CFLAG parsing logic).
> 
> Cc: Albert Aribaud <albert.u.boot@aribaud.net>
> Signed-off-by: Tom Rini <trini@ti.com>

For the record, build-tested with ELDKs for arm, powerpc and mips,
and other toolchains for nds32 m68k x86 and sandbox.
Albert ARIBAUD - Oct. 12, 2013, 7:43 a.m.
Hi Tom,

On Fri, 11 Oct 2013 16:13:03 -0400, Tom Rini <trini@ti.com> wrote:

> In some cases, such as arm multi-lib hardfloat (hf) toolchains, we will
> have multiple libgcc.a's available, and the arch needs to provide
> additional logic to determine the right file to use
> (-print-libgcc-file-name contains no CFLAG parsing logic).
> 
> Cc: Albert Aribaud <albert.u.boot@aribaud.net>
> Signed-off-by: Tom Rini <trini@ti.com>
> ---
>  Makefile           |    4 ++++
>  arch/arm/config.mk |   11 +++++++++++
>  2 files changed, 15 insertions(+)
> 
> diff --git a/Makefile b/Makefile
> index b09bfcc..13b396f 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -357,8 +357,12 @@ else
>  PLATFORM_LIBGCC = -L $(USE_PRIVATE_LIBGCC) -lgcc
>  endif
>  else
> +ifneq ("$(ARCH_PLATFORM_LIBGCC)","")
> +PLATFORM_LIBGCC := -L $(shell dirname $(ARCH_PLATFORM_LIBGCC)) -lgcc
> +else
>  PLATFORM_LIBGCC := -L $(shell dirname `$(CC) $(CFLAGS) -print-libgcc-file-name`) -lgcc
>  endif
> +endif
>  PLATFORM_LIBS += $(PLATFORM_LIBGCC)
>  export PLATFORM_LIBS
>  
> diff --git a/arch/arm/config.mk b/arch/arm/config.mk
> index d0cf43f..e1d0dec 100644
> --- a/arch/arm/config.mk
> +++ b/arch/arm/config.mk
> @@ -19,6 +19,17 @@ LDFLAGS_FINAL += --gc-sections
>  PLATFORM_RELFLAGS += -ffunction-sections -fdata-sections \
>                       -fno-common -ffixed-r9 -msoft-float
>  
> +#
> +# When we use a hardfp toolchain if there are both 'libgcc.a' (hardfp) and
> +# 'arm-linux-gnueabi/libgcc.a' (softfp) we need to use the latter.  We
> +# cannot always build with a hardfp-only toolchain.

What does this last sentence mean exactly?

> +ARCH_PLATFORM_LIBGCC := $(shell \
> +	X=`$(CC) -print-file-name=arm-linux-gnueabi/libgcc.a`; \
> +	if [ -f $$X ]; then echo $$X ; \
> +	else $(CC) -print-file-name=libgcc.a ; \
> +	fi)
> +
>  # Support generic board on ARM
>  __HAVE_ARCH_GENERIC_BOARD := y

Amicalement,
Tom Rini - Oct. 14, 2013, 3:16 p.m.
On Sat, Oct 12, 2013 at 09:43:33AM +0200, Albert ARIBAUD wrote:
> Hi Tom,
> 
> On Fri, 11 Oct 2013 16:13:03 -0400, Tom Rini <trini@ti.com> wrote:
> 
> > In some cases, such as arm multi-lib hardfloat (hf) toolchains, we will
> > have multiple libgcc.a's available, and the arch needs to provide
> > additional logic to determine the right file to use
> > (-print-libgcc-file-name contains no CFLAG parsing logic).
> > 
> > Cc: Albert Aribaud <albert.u.boot@aribaud.net>
> > Signed-off-by: Tom Rini <trini@ti.com>
> > ---
> >  Makefile           |    4 ++++
> >  arch/arm/config.mk |   11 +++++++++++
> >  2 files changed, 15 insertions(+)
> > 
> > diff --git a/Makefile b/Makefile
> > index b09bfcc..13b396f 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -357,8 +357,12 @@ else
> >  PLATFORM_LIBGCC = -L $(USE_PRIVATE_LIBGCC) -lgcc
> >  endif
> >  else
> > +ifneq ("$(ARCH_PLATFORM_LIBGCC)","")
> > +PLATFORM_LIBGCC := -L $(shell dirname $(ARCH_PLATFORM_LIBGCC)) -lgcc
> > +else
> >  PLATFORM_LIBGCC := -L $(shell dirname `$(CC) $(CFLAGS) -print-libgcc-file-name`) -lgcc
> >  endif
> > +endif
> >  PLATFORM_LIBS += $(PLATFORM_LIBGCC)
> >  export PLATFORM_LIBS
> >  
> > diff --git a/arch/arm/config.mk b/arch/arm/config.mk
> > index d0cf43f..e1d0dec 100644
> > --- a/arch/arm/config.mk
> > +++ b/arch/arm/config.mk
> > @@ -19,6 +19,17 @@ LDFLAGS_FINAL += --gc-sections
> >  PLATFORM_RELFLAGS += -ffunction-sections -fdata-sections \
> >                       -fno-common -ffixed-r9 -msoft-float
> >  
> > +#
> > +# When we use a hardfp toolchain if there are both 'libgcc.a' (hardfp) and
> > +# 'arm-linux-gnueabi/libgcc.a' (softfp) we need to use the latter.  We
> > +# cannot always build with a hardfp-only toolchain.
> 
> What does this last sentence mean exactly?

What it says.  If a hardfp only toolchain is used certain targets shall
not link.  I had Marek confirm this as well.  I don't have the exact
logs handy, but today there's a few targets that will fail, and the nand
flash patchset introduces another because we enforce -msoft-float on
U-Boot, and rely on libgcc from the toolchain for certain functions
(__udivdi3 for example).

I don't know how you weren't making this fail on ELDK 5.x armv7a-fp, but
I was and the Linaro 2013.03 toolchain also will show this problem.
Once I push the nand PR (shortly), building for am3517_crane for example
will fail without this patch.
Albert ARIBAUD - Oct. 14, 2013, 5:06 p.m.
Hi Tom,

On Mon, 14 Oct 2013 11:16:37 -0400, Tom Rini <trini@ti.com> wrote:

> On Sat, Oct 12, 2013 at 09:43:33AM +0200, Albert ARIBAUD wrote:
> > Hi Tom,
> > 
> > On Fri, 11 Oct 2013 16:13:03 -0400, Tom Rini <trini@ti.com> wrote:
> > 
> > > In some cases, such as arm multi-lib hardfloat (hf) toolchains, we will
> > > have multiple libgcc.a's available, and the arch needs to provide
> > > additional logic to determine the right file to use
> > > (-print-libgcc-file-name contains no CFLAG parsing logic).
> > > 
> > > Cc: Albert Aribaud <albert.u.boot@aribaud.net>
> > > Signed-off-by: Tom Rini <trini@ti.com>
> > > ---
> > >  Makefile           |    4 ++++
> > >  arch/arm/config.mk |   11 +++++++++++
> > >  2 files changed, 15 insertions(+)
> > > 
> > > diff --git a/Makefile b/Makefile
> > > index b09bfcc..13b396f 100644
> > > --- a/Makefile
> > > +++ b/Makefile
> > > @@ -357,8 +357,12 @@ else
> > >  PLATFORM_LIBGCC = -L $(USE_PRIVATE_LIBGCC) -lgcc
> > >  endif
> > >  else
> > > +ifneq ("$(ARCH_PLATFORM_LIBGCC)","")
> > > +PLATFORM_LIBGCC := -L $(shell dirname $(ARCH_PLATFORM_LIBGCC)) -lgcc
> > > +else
> > >  PLATFORM_LIBGCC := -L $(shell dirname `$(CC) $(CFLAGS) -print-libgcc-file-name`) -lgcc
> > >  endif
> > > +endif
> > >  PLATFORM_LIBS += $(PLATFORM_LIBGCC)
> > >  export PLATFORM_LIBS
> > >  
> > > diff --git a/arch/arm/config.mk b/arch/arm/config.mk
> > > index d0cf43f..e1d0dec 100644
> > > --- a/arch/arm/config.mk
> > > +++ b/arch/arm/config.mk
> > > @@ -19,6 +19,17 @@ LDFLAGS_FINAL += --gc-sections
> > >  PLATFORM_RELFLAGS += -ffunction-sections -fdata-sections \
> > >                       -fno-common -ffixed-r9 -msoft-float
> > >  
> > > +#
> > > +# When we use a hardfp toolchain if there are both 'libgcc.a' (hardfp) and
> > > +# 'arm-linux-gnueabi/libgcc.a' (softfp) we need to use the latter.  We
> > > +# cannot always build with a hardfp-only toolchain.
> > 
> > What does this last sentence mean exactly?
> 
> What it says.  If a hardfp only toolchain is used certain targets shall
> not link.  I had Marek confirm this as well.  I don't have the exact
> logs handy, but today there's a few targets that will fail, and the nand
> flash patchset introduces another because we enforce -msoft-float on
> U-Boot, and rely on libgcc from the toolchain for certain functions
> (__udivdi3 for example).

I was wondering if the sentence meant this, or if it meant "there is no
way we can ensure that we'll always build with a hardfp-only
toolchain". 

Maybe the ambiguity would be lifted by saying "Building with a
hardfp-only toolchain will fail [for some targets]"?

> I don't know how you weren't making this fail on ELDK 5.x armv7a-fp, but
> I was and the Linaro 2013.03 toolchain also will show this problem.
> Once I push the nand PR (shortly), building for am3517_crane for example
> will fail without this patch.

That may be because I have several toolchains installed, one of which
is in the standard path (/usr/bin, /usr/lib, etc).

Amicalement,

Patch

diff --git a/Makefile b/Makefile
index b09bfcc..13b396f 100644
--- a/Makefile
+++ b/Makefile
@@ -357,8 +357,12 @@  else
 PLATFORM_LIBGCC = -L $(USE_PRIVATE_LIBGCC) -lgcc
 endif
 else
+ifneq ("$(ARCH_PLATFORM_LIBGCC)","")
+PLATFORM_LIBGCC := -L $(shell dirname $(ARCH_PLATFORM_LIBGCC)) -lgcc
+else
 PLATFORM_LIBGCC := -L $(shell dirname `$(CC) $(CFLAGS) -print-libgcc-file-name`) -lgcc
 endif
+endif
 PLATFORM_LIBS += $(PLATFORM_LIBGCC)
 export PLATFORM_LIBS
 
diff --git a/arch/arm/config.mk b/arch/arm/config.mk
index d0cf43f..e1d0dec 100644
--- a/arch/arm/config.mk
+++ b/arch/arm/config.mk
@@ -19,6 +19,17 @@  LDFLAGS_FINAL += --gc-sections
 PLATFORM_RELFLAGS += -ffunction-sections -fdata-sections \
                      -fno-common -ffixed-r9 -msoft-float
 
+#
+# When we use a hardfp toolchain if there are both 'libgcc.a' (hardfp) and
+# 'arm-linux-gnueabi/libgcc.a' (softfp) we need to use the latter.  We
+# cannot always build with a hardfp-only toolchain.
+#
+ARCH_PLATFORM_LIBGCC := $(shell \
+	X=`$(CC) -print-file-name=arm-linux-gnueabi/libgcc.a`; \
+	if [ -f $$X ]; then echo $$X ; \
+	else $(CC) -print-file-name=libgcc.a ; \
+	fi)
+
 # Support generic board on ARM
 __HAVE_ARCH_GENERIC_BOARD := y