[U-Boot] [PATCH 1/3] common: add ifdefs around bouncebuf.c body

[Stephen Warren]

On 11/07/2012 06:21 AM, Marek Vasut wrote:
> Dear Stephen Warren,
> 
>> On 11/06/2012 03:57 PM, Marek Vasut wrote:
>>> Dear Stephen Warren,
>>>
>>>> On 11/06/2012 03:43 PM, Marek Vasut wrote:
>>>>> Dear Stephen Warren,
>>>>>
>>>>>> On 11/05/2012 05:54 PM, Marek Vasut wrote:
>>>>>>> Dear Stephen Warren,
>>>>>>>
>>>>>>>> From: Stephen Warren <swarren at nvidia.com>
>>>>>>>>
>>>>>>>> If a U-Boot config file enables CONFIG_BOUNCE_BUFFER only for the
>>>>>>>> main U-Boot build and not for the SPL, then config.mk will contain
>>>>>>>> CONFIG_BOUNCE_BUFFER=y, so common/Makefile will build bouncebuf.c
>>>>>>>> for both the SPL and main U-Boot, but config.h won't set
>>>>>>>> CONFIG_BOUNCE_BUFFER for the SPL, so bouncebuf.h will provide static
>>>>>>>> inline functions, which will conflict with the compiled bouncebuf.c.
>>>>>>>> Solve this by guarding the body of bouncebuf.c with the ifdef to
>>>>>>>> avoid conflicts.
>>>>>>>
>>>>>>> Uh, don't you want the bounce buffer not compiled in for SPL? Then
>>>>>>> maybe add CONFIG_SPL_BOUNCE_BUFFER to force BB to be compiled into
>>>>>>> SPL or something ...
>>>>>>
>>>>>> Not compiling bouncebuf.c for SPL would solve this too. I have no idea
>>>>>> what build system contortions would be required to do this though. Do
>>>>>> you think the build system should be fixed first rather than taking
>>>>>> this series/patch?
>>>>>>
>>>>>> I guess we shouldn't need a separate CONFIG_SPL_BOUNCE_BUFFER option
>>>>>> though; we should rather simply set CONFIG_SPL_BOUNCE_BUFFER
>>>>>> appropriately for SPL and non-SPL, and have everything key off that
>>>>>> one variable, right?
>>>>>
>>>>> How will you be able to configure it separately for spl and non-spl ?
>>>>
>>>> For example,
>>>>
>>>> include/configs/trimslice.h contains:
>>>>
>>>> /* SD/MMC */
>>>> #define CONFIG_MMC
>>>> #define CONFIG_GENERIC_MMC
>>>> #define CONFIG_TEGRA_MMC
>>>> #define CONFIG_CMD_MMC
>>>>
>>>> However, we don't use MMC in our SPL, but don't want to pollute every
>>>> Tegra board file with ifdefs for SPL, so
>>>> include/configs/tegra-common-post.h (which is included at the end of
>>>> trimslice.h) contains:
>>>>
>>>> #ifdef CONFIG_SPL_BUILD
>>>> ...
>>>> /* remove MMC support */
>>>> #ifdef CONFIG_MMC
>>>> #undef CONFIG_MMC
>>>> #endif
>>>> #ifdef CONFIG_GENERIC_MMC
>>>> #undef CONFIG_GENERIC_MMC
>>>> #endif
>>>> #ifdef CONFIG_TEGRA_MMC
>>>> #undef CONFIG_TEGRA_MMC
>>>> #endif
>>>> #ifdef CONFIG_CMD_MMC
>>>> #undef CONFIG_CMD_MMC
>>>> #endif
>>>> ...
>>>> #endif
>>>>
>>>> And in the V1 patch I proposed to tegra-common-post.h, I added the
>>>> following at the end:
>>>>
>>>> #ifdef CONFIG_TEGRA_MMC
>>>> #define CONFIG_BOUNCE_BUFFER
>>>> #endif
>>>
>>> Yet, this doesn't solve the problem with SPL ... since for SPL, you'd
>>> have to do ifdef CONFIG_SPL, no?
>>
>> Sorry, what problem with the SPL is this not solving?
> 
> I think I was tired when replying (sorry, the conference is really heavy on me). 
> I though you wanted to disable BB only for SPL, but now I see BB being enabled 
> depends on tegra mmc.
> 
> btw. shouldn't --gc-sections remove BB code if it's not used at all?

Yes, I assume so. In v2 of the patch series I have simply enabled
CONFIG_BOUNCE_BUFFER unconditionally on Tegra, which removes the need
for any SPL-specific changes.

Before enabling it:

Configuring for trimslice board...
   text	   data	    bss	    dec	    hex	filename
 245625	   9304	 274968	 529897	  815e9	./u-boot
  14451	    208	     72	  14731	   398b	./spl/u-boot-spl

After enabling it:

Configuring for trimslice board...
   text	   data	    bss	    dec	    hex	filename
 245742	   9304	 274964	 530010	  8165a	./u-boot
  14451	    208	     72	  14731	   398b	./spl/u-boot-spl

SPL didn't change since, so this seems to be working fine.