[RFC PATCH v3 02/16] ARM: davinci: move private EDMA API to arm/common

[Sekhar Nori]

Matt,

Sorry about the late reply. I noticed this mail only after I started to
look at v5 of your series :(

On 1/11/2013 5:21 AM, Matt Porter wrote:
> On Sun, Oct 28, 2012 at 01:47:09PM +0530, Sekhar Nori wrote:
>> On 10/18/2012 6:56 PM, Matt Porter wrote:
>>> Move mach-davinci/dma.c to common/edma.c so it can be used
>>> by OMAP (specifically AM33xx) as well. This just moves the
>>> private EDMA API but does not support OMAP.
>>>
>>> Signed-off-by: Matt Porter <mporter at ti.com>
>>> ---
>>
>>> diff --git a/arch/arm/mach-davinci/devices.c b/arch/arm/mach-davinci/devices.c
>>> index 4c48a36..f45d591 100644
>>> --- a/arch/arm/mach-davinci/devices.c
>>> +++ b/arch/arm/mach-davinci/devices.c
>>> @@ -19,9 +19,10 @@
>>>  #include <mach/irqs.h>
>>>  #include <mach/cputype.h>
>>>  #include <mach/mux.h>
>>> -#include <mach/edma.h>
>>>  #include <linux/platform_data/mmc-davinci.h>
>>>  #include <mach/time.h>
>>> +#include <linux/platform_data/edma.h>
>>
>> Can you please introduce a patch to clean this mixture of linux/ and
>> mach/ includes?
> 
> Yeah, are you ok with a follow on series to clean all these files? I
> don't want to make this series longer with something not dma related
> that's purely cleanup from the great header reorg of 2012.

Okay. I agree doing this sort of change now will cause merge issues.

> 
>>> +
>>>  
>>>  #include "davinci.h"
>>>  #include "clock.h"
>>> @@ -141,10 +142,10 @@ static struct resource mmcsd0_resources[] = {
>>>  	},
>>>  	/* DMA channels: RX, then TX */
>>>  	{
>>> -		.start = EDMA_CTLR_CHAN(0, DAVINCI_DMA_MMCRXEVT),
>>> +		.start = EDMA_CTLR_CHAN(0, 26),
>>
>> Instead of just replacing the event #defines with plain numbers, can you
>> introduce a mach-davinci local edma.h which can then be included in the
>> davinci platform files which refer to edma channel numbers?
> 
> Ok, so when I removed the old edma.h it was full of unused defines and
> we are left with just this one. If I introduced something like that it
> would be used in just one place here. and only for these two values.
> How about we just add a comment?

I prefer adding a local #define in this file. This will keep it
consistent with what is done in arch/arm/mach-davinci/dm365.c for
example. Same for the change in sound/soc/davinci/davinci-sffsdr.c

> 
>>
>>> diff --git a/arch/arm/plat-omap/Kconfig b/arch/arm/plat-omap/Kconfig
>>> index 7cd56ed..153fab8 100644
>>> --- a/arch/arm/plat-omap/Kconfig
>>> +++ b/arch/arm/plat-omap/Kconfig
>>> @@ -28,6 +28,7 @@ config ARCH_OMAP2PLUS
>>>  	select OMAP_DM_TIMER
>>>  	select PROC_DEVICETREE if PROC_FS
>>>  	select SPARSE_IRQ
>>> +	select TI_PRIV_EDMA
>>
>> This hunk does not seem to belong to subject of this patch.
> 
> Let me reword the subject/description. The idea of this logical
> chunk was to put everything in place to make it build on OMAP,
> with the expectation that the build fails without the next patch
> in the series.

But this will break bisect. With just this patch applied, omap build
breaks. Looks like it will be more logical to merge this hunk with 3/14
since that's when you really start using the private DMA support on OMAP
platforms.

This issue is a must fix IMO.

I am ok with the patch otherwise. With these changes done, you can add:

Acked-by: Sekhar Nori <nsekhar at ti.com>

Thanks,
Sekhar