[PATCH] ARM: bcm: Don't build brcmstb SMP code when building for V6

[PATCH] ARM: bcm: Don't build brcmstb SMP code when building for V6

[Mark Brown]

From: Mark Brown <broonie@linaro.org>

The brcmstb SMP code contains isb() and dsb() which are not supported when
building for -march=armv6 as happens when building allmodconfig. Since I
have not been able to identify a way to override the compiler architecture
selection per file add a Kconfig dependency to ensure this code is not
built if V6 support is enabled.

There is doubtless a better solution.

Signed-off-by: Mark Brown <broonie@linaro.org>
---
 arch/arm/mach-bcm/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm/mach-bcm/Kconfig b/arch/arm/mach-bcm/Kconfig
index fc93800..372132d 100644
--- a/arch/arm/mach-bcm/Kconfig
+++ b/arch/arm/mach-bcm/Kconfig
@@ -102,6 +102,7 @@ config ARCH_BCM_5301X
 config ARCH_BRCMSTB
 	bool "Broadcom BCM7XXX based boards" if ARCH_MULTI_V7
 	depends on MMU
+	depends on !CPU_V6
 	select ARM_GIC
 	select MIGHT_HAVE_PCI
 	select HAVE_SMP
-- 
2.1.0.rc1

[PATCH] ARM: bcm: Don't build brcmstb SMP code when building for V6

[Brian Norris]

+ Russell

On Thu, Aug 21, 2014 at 09:06:34PM -0500, Mark Brown wrote:
> From: Mark Brown <broonie@linaro.org>
> 
> The brcmstb SMP code contains isb() and dsb() which are not supported when

The problem is not with isb() and dsb() (which have different
definitions when v6 is enabled), but with v7_exit_coherency_flush().

> building for -march=armv6 as happens when building allmodconfig. Since I
> have not been able to identify a way to override the compiler architecture
> selection per file add a Kconfig dependency to ensure this code is not
> built if V6 support is enabled.

I already proposed this solution to override the compiler architecture,
in the same pattern as other similar code in the same directory.

  http://article.gmane.org/gmane.linux.drivers.devicetree/84517

My using 'CFLAGS_<obj> += <flags>', you can compile in v7.

> There is doubtless a better solution.

I think mine is better? Not sure. Comments are welcome.

But anyway, this file *should* be dropped for 3.17 (resolving the
current problem at hand), if somebody will eventually pick up this
patch:

  http://article.gmane.org/gmane.linux.kernel/1767679

Hello, anybody? Should Matt be picking this up, or for post-rc1, do
Arnd/Olof take these patches directly?

> Signed-off-by: Mark Brown <broonie@linaro.org>
> ---
>  arch/arm/mach-bcm/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/arm/mach-bcm/Kconfig b/arch/arm/mach-bcm/Kconfig
> index fc93800..372132d 100644
> --- a/arch/arm/mach-bcm/Kconfig
> +++ b/arch/arm/mach-bcm/Kconfig
> @@ -102,6 +102,7 @@ config ARCH_BCM_5301X
>  config ARCH_BRCMSTB
>  	bool "Broadcom BCM7XXX based boards" if ARCH_MULTI_V7
>  	depends on MMU
> +	depends on !CPU_V6
>  	select ARM_GIC
>  	select MIGHT_HAVE_PCI
>  	select HAVE_SMP

Regards,
Brian

[PATCH] ARM: bcm: Don't build brcmstb SMP code when building for V6

[Mark Brown]

On Thu, Aug 21, 2014 at 10:09:07PM -0700, Brian Norris wrote:
> On Thu, Aug 21, 2014 at 09:06:34PM -0500, Mark Brown wrote:
> > From: Mark Brown <broonie@linaro.org>

> > building for -march=armv6 as happens when building allmodconfig. Since I
> > have not been able to identify a way to override the compiler architecture
> > selection per file add a Kconfig dependency to ensure this code is not
> > built if V6 support is enabled.

> I already proposed this solution to override the compiler architecture,
> in the same pattern as other similar code in the same directory.

>   http://article.gmane.org/gmane.linux.drivers.devicetree/84517

> My using 'CFLAGS_<obj> += <flags>', you can compile in v7.

I tried that but it wasn't doing the right thing for me - it appeared
since the compiler ends up getting both -march=armv7-a and -march=armv6
it complained that it couldn't satisfy both options.  It might be that
this is version dependant or the actual cause wsa something else, I
didn't investigate too hard.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20140822/274418c1/attachment.sig>

[PATCH] ARM: bcm: Don't build brcmstb SMP code when building for V6

[Brian Norris]

On Fri, Aug 22, 2014 at 09:44:03AM -0500, Mark Brown wrote:
> On Thu, Aug 21, 2014 at 10:09:07PM -0700, Brian Norris wrote:
> > On Thu, Aug 21, 2014 at 09:06:34PM -0500, Mark Brown wrote:
> > > From: Mark Brown <broonie@linaro.org>
> 
> > > building for -march=armv6 as happens when building allmodconfig. Since I
> > > have not been able to identify a way to override the compiler architecture
> > > selection per file add a Kconfig dependency to ensure this code is not
> > > built if V6 support is enabled.
> 
> > I already proposed this solution to override the compiler architecture,
> > in the same pattern as other similar code in the same directory.
> 
> >   http://article.gmane.org/gmane.linux.drivers.devicetree/84517
> 
> > My using 'CFLAGS_<obj> += <flags>', you can compile in v7.
> 
> I tried that but it wasn't doing the right thing for me - it appeared
> since the compiler ends up getting both -march=armv7-a and -march=armv6
> it complained that it couldn't satisfy both options.  It might be that
> this is version dependant or the actual cause wsa something else, I
> didn't investigate too hard.

If you ever see revisit this or see this again, I'd be interested in
seeing the gcc args (make V=1) and version that failed. My fix worked
for me, and I'll likely use it for the next revision if I don't find a
good reason not to.

Thanks,
Brian

[PATCH] ARM: bcm: Don't build brcmstb SMP code when building for V6

[Arnd Bergmann]

On Monday 25 August 2014 10:27:13 Brian Norris wrote:
> On Fri, Aug 22, 2014 at 09:44:03AM -0500, Mark Brown wrote:
> > On Thu, Aug 21, 2014 at 10:09:07PM -0700, Brian Norris wrote:
> > > On Thu, Aug 21, 2014 at 09:06:34PM -0500, Mark Brown wrote:
> > > > From: Mark Brown <broonie@linaro.org>
> > 
> > > > building for -march=armv6 as happens when building allmodconfig. Since I
> > > > have not been able to identify a way to override the compiler architecture
> > > > selection per file add a Kconfig dependency to ensure this code is not
> > > > built if V6 support is enabled.
> > 
> > > I already proposed this solution to override the compiler architecture,
> > > in the same pattern as other similar code in the same directory.
> > 
> > >   http://article.gmane.org/gmane.linux.drivers.devicetree/84517
> > 
> > > My using 'CFLAGS_<obj> += <flags>', you can compile in v7.
> > 
> > I tried that but it wasn't doing the right thing for me - it appeared
> > since the compiler ends up getting both -march=armv7-a and -march=armv6
> > it complained that it couldn't satisfy both options.  It might be that
> > this is version dependant or the actual cause wsa something else, I
> > didn't investigate too hard.

I've never seen that happen. Normally, the last option just overrides
all previous conflicting ones.

> If you ever see revisit this or see this again, I'd be interested in
> seeing the gcc args (make V=1) and version that failed. My fix worked
> for me, and I'll likely use it for the next revision if I don't find a
> good reason not to.

FWIW, I haven't seen any problems with the version below:

diff --git a/arch/arm/mach-bcm/Makefile b/arch/arm/mach-bcm/Makefile
index 67c492aabf4d..546d3d3022a4 100644
--- a/arch/arm/mach-bcm/Makefile
+++ b/arch/arm/mach-bcm/Makefile
@@ -38,3 +38,5 @@ ifeq ($(CONFIG_ARCH_BRCMSTB),y)
 obj-y				+= brcmstb.o
 obj-$(CONFIG_SMP)		+= headsmp-brcmstb.o platsmp-brcmstb.o
 endif
+
+CFLAGS_platsmp-brcmstb.o	+= -march=armv7-a


	Arnd