Re: [PATCH 1/7] rockchip: mkimage: Split size_and_off and size_and_nimage

Hi Alexey,

On 7/13/26 12:02 PM, Alexey Charkov wrote:
> From: Jonas Karlman <[email protected]>
> 
> Split 32-bit size_and_off and size_and_nimage fields of the v2 image
> format header into their own 16-bit size, offset and num_images fields.
> 
> Set num_images based on number of images passed by the datafile
> parameter and size based on the offset to the hash field to fix using a
> single init data file and no boot data file for the v2 image format.
> 

When you need to start listing things you do in a commit, it means it 
needs to be split into multiple individual commits doing one thing at a 
time. I appreciate this is taken from Jonas's tree but it's fine to 
modify (well, from a maintainer's perspective at least :) ). Please list 
between Jonas's and your Signed-off-by what you changed since Jonas's 
version, in square brackets.

> Signed-off-by: Jonas Karlman <[email protected]>
> Signed-off-by: Alexey Charkov <[email protected]>
> ---
>   tools/rkcommon.c | 44 ++++++++++++++++++++++++--------------------
>   1 file changed, 24 insertions(+), 20 deletions(-)
> 
> diff --git a/tools/rkcommon.c b/tools/rkcommon.c
> index b39777fc0607..034896d57f80 100644
> --- a/tools/rkcommon.c
> +++ b/tools/rkcommon.c
> @@ -34,15 +34,16 @@ enum hash_type {
>   /**
>    * struct image_entry
>    *
> - * @size_and_off:	[31:16]image size;[15:0]image offset
> - * @address:	default as 0xFFFFFFFF
> + * @offset:	image offset (unit as 512 byte blocks)
> + * @size:	image size (unit as 512 byte blocks)
> + * @address:	load address (default as 0xFFFFFFFF)

Can it be anything but 0xFFFFFFFF? It's not even configurable currently.

>    * @flag:	no use
>    * @counter:	no use
>    * @hash:	hash of image
> - *
>    */
>   struct image_entry {
> -	uint32_t size_and_off;
> +	uint16_t offset;
> +	uint16_t size;

Do we need to start __attribute__ ((__packed__))'ing the structure to 
make sure there's no padding involved ever?

>   	uint32_t address;
>   	uint32_t flag;
>   	uint32_t counter;
> @@ -56,16 +57,17 @@ struct image_entry {
>    * This is stored at SD card block 64 (where each block is 512 bytes)
>    *
>    * @magic:	Magic (must be RK_MAGIC_V2)
> - * @size_and_nimage:	[31:16]number of images;[15:0]
> - *			offset to hash field of header(unit as 4Byte)
> - * @boot_flag:	[3:0]hash type(0:none,1:sha256,2:sha512)
> - * @signature:	hash or signature for header info
> - *
> + * @size:	offset to hash field of header (unit as 4 bytes)

I'm assuming we mean "unit as multiples of 4 bytes"? Is that correct? 
Can we say that instead, I find it clearer.

> + * @num_images:	number of images

Can we improve this documentation as well? I'm assuming this is the 
number of images stored in idbloader.img (so typically SPL + optionally 
TPL + optionally VPL/boost?)

> + * @boot_flag:	[3:0] hash type (0:none, 1:sha256, 2:sha512)
> + * @images:	images

Not sure this is helping, either specify what those are or simply don't 
comment.

> + * @hash:	hash or signature for header info
>    */
>   struct header0_info_v2 {
>   	uint32_t magic;
>   	uint8_t reserved[4];
> -	uint32_t size_and_nimage;
> +	uint16_t size;
> +	uint16_t num_images;

Do we need to start __attribute__ ((__packed__))'ing the structure to 
make sure there's no padding involved ever?

Side question, should we start using the proper expected endianness 
here? e.g. __le16/__le32 instead of uint32_t and uint16_t? What do you 
think?

>   	uint32_t boot_flag;
>   	uint8_t reserved1[104];
>   	struct image_entry images[4];
> @@ -351,17 +353,18 @@ static void rkcommon_set_header0_v2(void *buf, struct image_tool_params *params)
>   	printf("Image Type:   Rockchip %s boot image\n",
>   		rkcommon_get_spl_hdr(params));
>   	memset(buf, '\0', RK_INIT_OFFSET * RK_BLK_SIZE);
> -	hdr->magic   = cpu_to_le32(RK_MAGIC_V2);
> -	hdr->size_and_nimage = cpu_to_le32((2 << 16) + 384);
> +	hdr->magic = cpu_to_le32(RK_MAGIC_V2);

Please don't mix cosmetic changes as those and logic changes, in the 
same commit. The whitespaces removal makes it harder to identify quickly 
what is actually changed.

>   	hdr->boot_flag = cpu_to_le32(HASH_SHA256);
>   	sector_offset = 4;
>   	image_size_array[0] = spl_params.init_size;
>   	image_size_array[1] = spl_params.boot_size;
>   
>   	for (i = 0; i < 2; i++) {
> +		if (!image_size_array[i])
> +			break;
>   		image_sector_count = image_size_array[i] / RK_BLK_SIZE;
> -		hdr->images[i].size_and_off = cpu_to_le32((image_sector_count
> -							<< 16) + sector_offset);
> +		hdr->images[i].offset = cpu_to_le16(sector_offset);
> +		hdr->images[i].size = cpu_to_le16(image_sector_count);
>   		hdr->images[i].address = 0xFFFFFFFF;
>   		hdr->images[i].counter = cpu_to_le32(i + 1);
>   		image_ptr = buf + sector_offset * RK_BLK_SIZE;
> @@ -370,6 +373,8 @@ static void rkcommon_set_header0_v2(void *buf, struct image_tool_params *params)
>   		sector_offset = sector_offset + image_sector_count;
>   	}
>   
> +	hdr->num_images = cpu_to_le16(i);

Is it though? if image_size_array[i] is 0, it means nothing was passed 
as the i+1th image to mkimage -T rksd/rkspi -d no?

> +	hdr->size = cpu_to_le16(offsetof(typeof(*hdr), hash) / sizeof(uint32_t));
>   	do_sha256_hash(buf, (void *)hdr->hash - buf, hdr->hash);
>   }
>   
> @@ -516,10 +521,8 @@ void rkcommon_print_header(const void *buf, struct image_tool_params *params)
>   			return;
>   		}
>   
> -		init_size = header0_v2.images[0].size_and_off >> 16;
> -		init_size = init_size * RK_BLK_SIZE;
> -		boot_size = header0_v2.images[1].size_and_off >> 16;
> -		boot_size = boot_size * RK_BLK_SIZE;
> +		init_size = le16_to_cpu(header0_v2.images[0].size) * RK_BLK_SIZE;
> +		boot_size = le16_to_cpu(header0_v2.images[1].size) * RK_BLK_SIZE;
>   	} else {
>   		ret = rkcommon_parse_header(buf, &header0, &spl_info);
>   
> @@ -533,8 +536,9 @@ void rkcommon_print_header(const void *buf, struct image_tool_params *params)
>   		}
>   
>   		image_type = ret;
> -		init_size = header0.init_size * RK_BLK_SIZE;
> -		boot_size = header0.init_boot_size * RK_BLK_SIZE - init_size;
> +		init_size = le16_to_cpu(header0.init_size) * RK_BLK_SIZE;
> +		boot_size = le16_to_cpu(header0.init_boot_size) * RK_BLK_SIZE -
> +			    init_size;
>   

I could identify multiple commits to be split out of this one:
- update documentation,
- whitespace around ->magic assignment,
- break the loop when image_size_array[i] is 0,
- properly translate init_size/boot_size into CPU endianness,
- split size_and_nimage u32 into two u16,
- split size_and_off u32 into two u16,
- programmatically set ->size,
- programmatically set ->num_images,

Cheers,
Quentin