[U-Boot] [PATCH V7 5/5] omap-common: fixes BSS overwriting problem

[Stefano Babic]

On 31/10/2011 17:23, Simon Schwarz wrote:
> From: Simon Schwarz <simonschwarzcor at googlemail.com>
> 
> spl_nand overwrote BSS section because it reads a whole block everytime. Now
> loads the block to spare area and just copy the needed junk to destination.
> Whole block read is necessary for ecc check!
> 
> Signed-off-by: Simon Schwarz <simonschwarzcor at gmail.com>
> ---

> @@ -71,7 +71,8 @@ void spl_nand_load_image(void)
>  			CONFIG_SYS_NAND_PAGE_SIZE, (void *)header);
>  		spl_parse_image_header(header);
>  		nand_spl_load_image(CONFIG_SYS_NAND_SPL_KERNEL_OFFS,
> -			spl_image.size, (void *)spl_image.load_addr);
> +			spl_image.size,
> +			(void *)spl_image.load_addr - sizeof(header));
                                                             ^---


Do you mean maybe sizeof(*header) ?

However, spl_image.load_addr was already decremented in
spl_parse_image_header() and correctly set to 64 byte before the load
address. Do we really need it ?

I found the readon of the kernel corrupt image. We are setting a very
hard address in /nand_spl_simple.c:

ecc_calc = (u_char *)(CONFIG_SYS_SDRAM_BASE + 0x10000);

Because the image for a TI SOC is loaded at 0x80008000, we have a
conflict and the image is corrupted where the ECC is computed.

It is not a really good idea to fix in this way where to compute the
ECC. Should be not better to put it in the CONFIG_SYS_INIT_RAM_ADDR area ?

Best regards,
Stefano Babic

-- 
=====================================================================
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80  Email: office at denx.de
=====================================================================