[U-Boot] Commit "ARM: CPU: arm926ejs: Consolidate cache routines to common file" breaks u-boot on Dreamplug

[Leigh Brown]

Hi Adam,

Thanks very much for your response.

On 2019-02-27 12:50, Adam Ford wrote:
> On Wed, Feb 27, 2019 at 5:32 AM Leigh Brown <leigh at solinno.co.uk> 
> wrote:
>> 
>> Hello,
>> 
>> Vagrant Cascadian asked for people to test the version of u-boot
>> packaged
>> for Debian Buster.  I tested u-boot on my Dreamplug and found it was 
>> not
>> working correctly.  I raised a bug for Debian[1] but I have also 
>> tested
>> with the mainline version of u-boot and found the same issues.
>> 
>> The first issue is that the following commit caused u-boot to no 
>> longer
>> be able to access usb storage on the Dreamplug:
>> 
>> commit 93b283d49f933f95f3a6f40762936f454ac655a8
>> Author: Adam Ford <aford173 at gmail.com>
>> Date:   Thu Aug 16 13:23:11 2018 -0500
>> 
> 
> Sorry about that.
> 
>>      ARM: CPU: arm926ejs: Consolidate cache routines to common file
>> 
>>      Four different boards had different options for enabling cache
>>      that were virtually all the same.  This consolidates these
>>      common functions into arch/arm/cpu/arm926ejs/cache.c
>> 
>>      This also has the positive side-effect of enabling cache on
>>      the Davinci (da850) boards.
>> 
>>      Signed-off-by: Adam Ford <aford173 at gmail.com>
>>      [trini: Add mach-at91 to the list of consolidations]
>>      Signed-off-by: Tom Rini <trini at konsulko.com>
>> 
>> I don't have much knowledge of ARM caching, but the following patch
>> makes
>> it work again on my Dreamplug.
> 
> I am not that familiar with the ARM caching either, I was just hoping
> to enable it on the da850 board and compared the various code chunks
> between ARM 926 boards and noticed a common thread.  Tom took it and
> added some more.

Okay.  Hopefully Tom can comment on my proposed fixes.

> 
>> 
>> diff --git a/arch/arm/mach-kirkwood/cpu.c 
>> b/arch/arm/mach-kirkwood/cpu.c
>> index d54de53f31..8a065d73ae 100644
>> --- a/arch/arm/mach-kirkwood/cpu.c
>> +++ b/arch/arm/mach-kirkwood/cpu.c
>> @@ -291,7 +291,6 @@ int arch_misc_init(void)
>>         temp |= (1 << 22);
>>         writefr_extra_feature_reg(temp);
>> 
> #ifndef CONFIG_SYS_ICACHE_OFF
>> -       icache_enable();
> #endif
> 
> Instead of commenting out that line, try defining
> CONFIG_SYS_ICACHE_OFF in your header like you did for the
> CONFIG_SYS_DCACHE_OFF and encapsulate the function call with ifdef's
> so any other kirkwood processors can enable/disable it independently.

The reason I removed that line is because icache_enable() is called from
enable_caches() which is called from initr_caches() which is in the list
of functions in init_sequence_r[] prior to arch_misc_init().  In other
words, it will already have been called by then if CONFIG_SYS_ICACHE_OFF
is not set.  Does that make sense?

>>         /* Change reset vector to address 0x0 */
>>         temp = get_cr();
>>         set_cr(temp & ~CR_V);
>> diff --git a/include/configs/dreamplug.h b/include/configs/dreamplug.h
>> index f4d717213c..6348935c68 100644
>> --- a/include/configs/dreamplug.h
>> +++ b/include/configs/dreamplug.h
>> @@ -79,4 +79,6 @@
>>   #define CONFIG_SYS_ATA_IDE0_OFFSET    MV_SATA_PORT0_OFFSET
>>   #endif /*CONFIG_MVSATA_IDE*/
>> 
>> +#define CONFIG_SYS_DCACHE_OFF
> #define CONFIG_SYS_ICACHE_OFF

The reason I have not done this is because the Kirkwood arch_misc_init()
function was already unconditionally enabling the instruction cache, so
we want to retain that behaviour - I think.  I hope that makes sense.

>> +
>>   #endif /* _CONFIG_DREAMPLUG_H */
>> 
>> I'd be grateful if someone could take a look.  If you need me to do 
>> any
>> testing  or provide any more information, please let me know.
>> 
>> I found another issue (documented in the same bug).  I'll send a
>> separate
>> email about that.
>> 
>> Regards,
>> 
>> Leigh.
>> 
>> --
>> [1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=923379

Regards,

Leigh.