[U-Boot] [PATCH v4 1/7] arm: add MMU/D-Cache support for Faraday cores

[Kuo-Jung Su]

2013/6/11 Albert ARIBAUD <albert.u.boot at aribaud.net>:
> Hi Kuo-Jung,
>
> On Tue,  7 May 2013 14:25:07 +0800, Kuo-Jung Su <dantesu at gmail.com>
> wrote:
>
>> diff --git a/arch/arm/include/asm/dma-mapping.h b/arch/arm/include/asm/dma-mapping.h
>> index 5bbb0a0..5a13af5 100644
>> --- a/arch/arm/include/asm/dma-mapping.h
>> +++ b/arch/arm/include/asm/dma-mapping.h
>> @@ -3,6 +3,9 @@
>>   * Stelian Pop <stelian at popies.net>
>>   * Lead Tech Design <www.leadtechdesign.com>
>>   *
>> + * (C) Copyright 2010
>> + * Dante Su <dantesu at faraday-tech.com>
>
> Fix Copyright notices (dates) throughout the patch (and series as
> needed).
>

Got it, thanks.

>>   * See file CREDITS for list of people who contributed to this
>>   * project.
>>   *
>> @@ -24,22 +27,76 @@
>>  #ifndef __ASM_ARM_DMA_MAPPING_H
>>  #define __ASM_ARM_DMA_MAPPING_H
>>
>> +#if defined(CONFIG_FARADAY) && !defined(CONFIG_SYS_DCACHE_OFF)
>> +#include <asm/u-boot.h>
>> +#include <asm/global_data.h>
>> +#include <asm/io.h>
>> +#include <malloc.h>
>> +
>> +DECLARE_GLOBAL_DATA_PTR;
>> +#endif /* CONFIG_FARADAY && !CONFIG_SYS_DCACHE_OFF */
>> +
>>  enum dma_data_direction {
>>       DMA_BIDIRECTIONAL       = 0,
>>       DMA_TO_DEVICE           = 1,
>>       DMA_FROM_DEVICE         = 2,
>>  };
>>
>> -static void *dma_alloc_coherent(size_t len, unsigned long *handle)
>> +static inline void *dma_alloc_coherent(size_t len, unsigned long *handle)
>>  {
>> -     *handle = (unsigned long)malloc(len);
>> +#if defined(CONFIG_FARADAY) && !defined(CONFIG_SYS_DCACHE_OFF)
>> +     void *map, *va = memalign(ARCH_DMA_MINALIGN, len);
>> +
>> +     if (va && gd->arch.cpu_mmu) {
>> +             invalidate_dcache_range((ulong)va, (ulong)va + len);
>> +             map = map_physmem((phys_addr_t)va, len, MAP_NOCACHE);
>> +             if (!map)
>> +                     free(va);
>> +             va = map;
>> +     }
>> +
>> +     if (handle)
>> +             *handle = virt_to_phys(va);
>> +
>> +     return va;
>> +#else  /* CONFIG_FARADAY && !CONFIG_SYS_DCACHE_OFF */
>> +     *handle = (unsigned long)memalign(ARCH_DMA_MINALIGN, len);
>>       return (void *)*handle;
>
> This is not identical to what the code was before the patch. Why the
> difference?
>

Yes, it's not identical to what the code was.

It was:
    *handle = (unsigned long)malloc(len);
But I think it should be
    *handle = (unsigned long)memalign(ARCH_DMA_MINALIGN, len);

Because even though the MMU/D-cache is off, some DMA engines still
requires strict address alignment.

For example, the Faraday FTMAC110 & FTGMAC100 ether-net controllers
expect the descriptors are always aligned to 16-bytes boundary.

--
Best wishes,
Kuo-Jung Su