[U-Boot] [RFC PATCH] SPL: FIT: Enable SPL_FIT_LOAD for sd bootmode for fat partions

[Michal Simek]

On 2.5.2016 06:06, Lokesh Vutla wrote:
> Hi Michal,
> 
> On Thursday 28 April 2016 03:01 PM, Michal Simek wrote:
>> Support U-Boot SPL to load FIT image from fat partition.
>> Fit image can be setup via CONFIG_SPL_FS_LOAD_KERNEL_NAME.
>> Falcon mode is not supported.
>>
>> Signed-off-by: Michal Simek <michal.simek at xilinx.com>
>> ---
>>
>>  common/spl/spl_fat.c | 50 ++++++++++++++++++++++++++++++++++++++++++++------
>>  fs/fat/fat.c         |  4 ++--
>>  2 files changed, 46 insertions(+), 8 deletions(-)
>>
>> diff --git a/common/spl/spl_fat.c b/common/spl/spl_fat.c
>> index d16cd540e38a..4e319c5fa470 100644
>> --- a/common/spl/spl_fat.c
>> +++ b/common/spl/spl_fat.c
>> @@ -13,6 +13,7 @@
>>  #include <spl.h>
>>  #include <asm/u-boot.h>
>>  #include <fat.h>
>> +#include <libfdt.h>
>>  #include <errno.h>
>>  #include <image.h>
>>  
>> @@ -39,6 +40,29 @@ static int spl_register_fat_device(struct blk_desc *block_dev, int partition)
>>  	return err;
>>  }
>>  
>> +#ifdef CONFIG_SPL_LOAD_FIT
>> +static ulong spl_fat_file_read(struct spl_load_info *load, ulong sector,
>> +			       ulong count, void *buf)
>> +{
>> +	int err;
>> +	loff_t actread;
>> +	char *filename = (char *)load->priv;
>> +
>> +	debug("%s: name %s, sector %lx, count %lx, buf %lx\n",
>> +	      __func__, filename,  sector, count, (ulong)buf);
>> +
>> +	err = file_fat_read_at(filename, sector, buf, count, &actread);
>> +	if (err < 0) {
>> +		printf("%s: error reading image %s, err - %d\n",
>> +		       __func__, filename, err);
>> +		return err;
>> +	}
>> +
>> +	debug("actread %lx\n", (ulong)actread);
>> +	return actread;
>> +}
>> +#endif
>> +
>>  int spl_load_image_fat(struct blk_desc *block_dev,
>>  						int partition,
>>  						const char *filename)
>> @@ -57,16 +81,29 @@ int spl_load_image_fat(struct blk_desc *block_dev,
>>  	if (err <= 0)
>>  		goto end;
>>  
>> -	spl_parse_image_header(header);
>> +	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT) &&
>> +	    image_get_magic(header) == FDT_MAGIC) {
>> +		struct spl_load_info load;
>> +
>> +		debug("Found FIT\n");
>> +		load.priv = (char *)filename;
>> +		load.bl_len = 1;
>> +		load.read = spl_fat_file_read;
>> +		spl_load_simple_fit(&load, 0, header);
>> +	} else {
>> +		debug("Legacy image\n");
>>  
>> -	err = file_fat_read(filename, (u8 *)(uintptr_t)spl_image.load_addr, 0);
>> +		spl_parse_image_header(header);
>>  
>> +		err = file_fat_read(filename,
>> +				    (u8 *)(uintptr_t)spl_image.load_addr, 0);
>>  end:
>>  #ifdef CONFIG_SPL_LIBCOMMON_SUPPORT
>> -	if (err <= 0)
>> -		printf("%s: error reading image %s, err - %d\n",
>> -		       __func__, filename, err);
>> +		if (err <= 0)
>> +			printf("%s: error reading image %s, err - %d\n",
>> +			       __func__, filename, err);
>>  #endif
>> +	}
>>  
>>  	return (err <= 0);
>>  }
>> @@ -81,6 +118,7 @@ int spl_load_image_fat_os(struct blk_desc *block_dev, int partition)
>>  	if (err)
>>  		return err;
>>  
>> +#if !defined(CONFIG_SPL_LOAD_FIT)
>>  #if defined(CONFIG_SPL_ENV_SUPPORT) && defined(CONFIG_SPL_OS_BOOT)
>>  	file = getenv("falcon_args_file");
>>  	if (file) {
>> @@ -116,7 +154,7 @@ defaults:
>>  #endif
>>  		return -1;
>>  	}
>> -
>> +#endif
>>  	return spl_load_image_fat(block_dev, partition,
>>  			CONFIG_SPL_FS_LOAD_KERNEL_NAME);
>>  }
>> diff --git a/fs/fat/fat.c b/fs/fat/fat.c
>> index 600a90e30922..0d987e0465ee 100644
>> --- a/fs/fat/fat.c
>> +++ b/fs/fat/fat.c
>> @@ -281,9 +281,9 @@ get_cluster(fsdata *mydata, __u32 clustnum, __u8 *buffer, unsigned long size)
>>  
>>  	if ((unsigned long)buffer & (ARCH_DMA_MINALIGN - 1)) {
>>  		ALLOC_CACHE_ALIGN_BUFFER(__u8, tmpbuf, mydata->sect_size);
>> -
>> +#if !defined(CONFIG_SPL_LOAD_FIT)
>>  		printf("FAT: Misaligned buffer address (%p)\n", buffer);
>> -
>> +#endif
> 
> IMO, this is a hack. Why should fs worry about if it as fit image or
> not. Also the read performance will be very slow if you do not pass the
> aligned buffer address.  I had a different approach[1] for this: first
> copy the image to aligned buffer and then do a memcpy to the proper
> destination(which showed a better performance). May be this is wrong.

I agree that's why this was RFC not regular patch.
I have looked at your solution and also Simon's comments and truth is
that your patch has a lot of duplicated stuff.
This solution is smaller.
Regarding buffer alignment. I think this can be simply added to read
function to keep it in the right place.

Thanks,
Michal