[RFC PATCH] ARM: dma: default ZONE_DMA to y if ARM_LPAE

[RFC PATCH] ARM: dma: default ZONE_DMA to y if ARM_LPAE

[Will Deacon]

On systems running with LPAE enabled, DMA masters capable only of 32-bit
addressing require ZONE_DMA to be set. Given the proliferation of such
masters, and the tendency to re-use existing system IP with newer
processors, this patch defaults ZONE_DMA to y if ARM_LPAE, rather than
force each SoC to select it individually.

Cc: Olof Johansson <olof@lixom.net>
Signed-off-by: Will Deacon <will.deacon@arm.com>
---
 arch/arm/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 82f06982..7fede35 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -192,6 +192,7 @@ config ARCH_MAY_HAVE_PC_FDC
 
 config ZONE_DMA
 	bool
+	default y if ARM_LPAE
 
 config NEED_DMA_MAP_STATE
        def_bool y
-- 
1.8.2.2

[RFC PATCH] ARM: dma: default ZONE_DMA to y if ARM_LPAE

[Olof Johansson]

On Fri, Jul 26, 2013 at 02:05:51PM +0100, Will Deacon wrote:
> On systems running with LPAE enabled, DMA masters capable only of 32-bit
> addressing require ZONE_DMA to be set. Given the proliferation of such
> masters, and the tendency to re-use existing system IP with newer
> processors, this patch defaults ZONE_DMA to y if ARM_LPAE, rather than
> force each SoC to select it individually.
> 
> Cc: Olof Johansson <olof@lixom.net>
> Signed-off-by: Will Deacon <will.deacon@arm.com>
> ---
>  arch/arm/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index 82f06982..7fede35 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -192,6 +192,7 @@ config ARCH_MAY_HAVE_PC_FDC
>  
>  config ZONE_DMA
>  	bool
> +	default y if ARM_LPAE

I guess you could bikeshed over this vs selecting it from ARM_LPAE, but either
way gets the work done. I don't have a preference.

Acked-by: Olof Johansson <olof@lixom.net>

..for Russell's patch tracker?


-Olof

[RFC PATCH] ARM: dma: default ZONE_DMA to y if ARM_LPAE

[Russell King - ARM Linux]

On Sun, Aug 11, 2013 at 02:51:12PM -0700, Olof Johansson wrote:
> On Fri, Jul 26, 2013 at 02:05:51PM +0100, Will Deacon wrote:
> > On systems running with LPAE enabled, DMA masters capable only of 32-bit
> > addressing require ZONE_DMA to be set. Given the proliferation of such
> > masters, and the tendency to re-use existing system IP with newer
> > processors, this patch defaults ZONE_DMA to y if ARM_LPAE, rather than
> > force each SoC to select it individually.
> > 
> > Cc: Olof Johansson <olof@lixom.net>
> > Signed-off-by: Will Deacon <will.deacon@arm.com>
> > ---
> >  arch/arm/Kconfig | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> > index 82f06982..7fede35 100644
> > --- a/arch/arm/Kconfig
> > +++ b/arch/arm/Kconfig
> > @@ -192,6 +192,7 @@ config ARCH_MAY_HAVE_PC_FDC
> >  
> >  config ZONE_DMA
> >  	bool
> > +	default y if ARM_LPAE
> 
> I guess you could bikeshed over this vs selecting it from ARM_LPAE, but either
> way gets the work done. I don't have a preference.
> 
> Acked-by: Olof Johansson <olof@lixom.net>
> 
> ..for Russell's patch tracker?

No.  As I keep telling people, go look at my dma-mask patches and stop
papering over this problem.

[RFC PATCH] ARM: dma: default ZONE_DMA to y if ARM_LPAE

[Olof Johansson]

On Sun, Aug 11, 2013 at 3:05 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:

> No.  As I keep telling people, go look at my dma-mask patches and stop
> papering over this problem.

Ah, yes, I'm going through email backlog. The DMA mask series will
definitely cover this in a better way.

This is a minimal patch that could be suitable for -stable backport.
If we care about that -- I'm sure people can just amend their configs
where needed.


-Olof

[RFC PATCH] ARM: dma: default ZONE_DMA to y if ARM_LPAE

[Russell King - ARM Linux]

On Sun, Aug 11, 2013 at 03:14:26PM -0700, Olof Johansson wrote:
> On Sun, Aug 11, 2013 at 3:05 PM, Russell King - ARM Linux
> <linux@arm.linux.org.uk> wrote:
> 
> > No.  As I keep telling people, go look at my dma-mask patches and stop
> > papering over this problem.
> 
> Ah, yes, I'm going through email backlog. The DMA mask series will
> definitely cover this in a better way.
> 
> This is a minimal patch that could be suitable for -stable backport.
> If we care about that -- I'm sure people can just amend their configs
> where needed.

Well, it's not a regression, it's something which has never worked.  In
my mind, that isn't really -stable material - stable is for stuff that
was either working and has been broken or for security stuff.

Having all your system memory above 4G _and_ having 32-bit DMA masks is
something which never worked.