[PATCH 3/3] abootimg: Implement smart image load feature

[Mattijs Korpershoek]

Hi Roman,

Thank you for the patch.

On dim., mai 19, 2024 at 19:18, Roman Stratiienko <r.stratiienko at gmail.com> wrote:

> Load only part of the boot partition that contains valuable
> information, thus improving the boot time.
>
> Signed-off-by: Roman Stratiienko <r.stratiienko at gmail.com>
> ---
>  boot/image-android.c | 70 ++++++++++++++++++++++++++++++++++++++++++++
>  cmd/abootimg.c       | 40 ++++++++++++++++++++++---
>  include/image.h      | 12 ++++++++
>  3 files changed, 118 insertions(+), 4 deletions(-)
>
> diff --git a/boot/image-android.c b/boot/image-android.c
> index da8003f370..d00a896a40 100644
> --- a/boot/image-android.c
> +++ b/boot/image-android.c
> @@ -204,6 +204,76 @@ bool android_image_get_data(const void *boot_hdr, const void *vendor_boot_hdr,
>  	return true;
>  }
>  
> +/**
> + * android_image_get_valuable_size() - get the size of the android image
> + *
> + * This function checks if the image is Android boot image and returns the
> + * valuable size of the image.
> + *
> + * @hdr_addr: Boot image header address (boot or vendor_boot)
> + *
> + * @return size of the image on success, 0 on failure
> + */
> +size_t android_image_get_valuable_size(const void *hdr_addr)

Can't we use a ssize_t to return a negative error code?
0 as error seems odd and not what is used in most functions in boot/image-android.c

> +{
> +	int version, size;
> +
> +	if (is_android_boot_image_header(hdr_addr)) {
> +		const struct andr_boot_img_hdr_v0 *hdr = hdr_addr;
> +
> +		version = ((struct andr_boot_img_hdr_v0 *)hdr_addr)->header_version;
> +		if (version > 2) {
> +			const struct andr_boot_img_hdr_v3 *hdr = hdr_addr;
> +
> +			size = ALIGN(hdr->header_size, ANDR_GKI_PAGE_SIZE);
> +			size += ALIGN(hdr->kernel_size, ANDR_GKI_PAGE_SIZE);
> +			size += ALIGN(hdr->ramdisk_size, ANDR_GKI_PAGE_SIZE);
> +
> +			if (version > 3)
> +				size += ALIGN(hdr->signature_size, ANDR_GKI_PAGE_SIZE);
> +
> +			return size;
> +		}
> +
> +		size = hdr->page_size;
> +		size += ALIGN(hdr->kernel_size, hdr->page_size);
> +		size += ALIGN(hdr->ramdisk_size, hdr->page_size);
> +		size += ALIGN(hdr->second_size, hdr->page_size);
> +
> +		if (version > 0)
> +			size += ALIGN(hdr->recovery_dtbo_size, hdr->page_size);
> +
> +		if (version > 1)
> +			size += ALIGN(hdr->dtb_size, hdr->page_size);
> +
> +		return size;
> +	}
> +
> +	if (is_android_vendor_boot_image_header(hdr_addr)) {
> +		const struct andr_vnd_boot_img_hdr *hdr = hdr_addr;
> +
> +		version = ((struct andr_vnd_boot_img_hdr *)hdr_addr)->header_version;
> +		if (version < 3) {
> +			printf("Vendor boot image header version %d is not supported\n", version);
> +
> +			return 0;
> +		}
> +
> +		size = ALIGN(hdr->header_size, hdr->page_size);
> +		size += ALIGN(hdr->vendor_ramdisk_size, hdr->page_size);
> +		size += ALIGN(hdr->dtb_size, hdr->page_size);
> +
> +		if (version > 3) {
> +			size += ALIGN(hdr->vendor_ramdisk_table_size, hdr->page_size);
> +			size += ALIGN(hdr->bootconfig_size, hdr->page_size);
> +		}
> +
> +		return size;
> +	}
> +
> +	return 0;
> +}
> +
>  static ulong android_image_get_kernel_addr(struct andr_image_data *img_data)
>  {
>  	/*
> diff --git a/cmd/abootimg.c b/cmd/abootimg.c
> index 808c9c4941..fe7c5c5e2c 100644
> --- a/cmd/abootimg.c
> +++ b/cmd/abootimg.c
> @@ -182,6 +182,35 @@ static int abootimg_get_dtb(int argc, char *const argv[])
>  	return CMD_RET_USAGE;
>  }
>  
> +static int abootimg_smart_load(struct blk_desc *desc, struct disk_partition *info, void *addr)
> +{
> +	int ret, size;
> +
> +	ret = blk_dread(desc, info->start, 4096 / info->blksz, addr);

Why 4096? Can we use #define for this or reuse an exising definition?

> +	if (ret < 0) {
> +		printf("Error: Failed to read partition\n");
> +		return CMD_RET_FAILURE;
> +	}
> +
> +	size = android_image_get_valuable_size(addr);
> +	if (size == 0)
> +		return 0;
> +
> +	ret = blk_dread(desc, info->start, DIV_ROUND_UP(size, info->blksz), addr);
> +	if (ret < 0) {
> +		printf("Error: Failed to read partition\n");
> +		return CMD_RET_FAILURE;
> +	}
> +
> +	memset(addr + size, 0, info->size * info->blksz - size);
> +
> +	printf("Loaded Android boot image using smart load (%d/%d MB)\n",

Why only boot image? Should this be more generic be passing the
partition name instead?

> +	       (int)DIV_ROUND_UP(size, 1024 * 1024),
> +	       (int)DIV_ROUND_UP(info->size * info->blksz, 1024 * 1024));
> +
> +	return size;
> +}
> +
>  static int do_abootimg_addr(struct cmd_tbl *cmdtp, int flag, int argc,
>  			    char *const argv[])
>  {
> @@ -299,10 +328,13 @@ static int do_abootimg_load(struct cmd_tbl *cmdtp, int flag, int argc,
>  		return CMD_RET_FAILURE;
>  	}
>  
> -	ret = blk_dread(desc, info.start, info.size, addr);
> -	if (ret < 0) {
> -		printf("Error: Failed to read partition %s\n", buf);
> -		goto fail;
> +	ret = abootimg_smart_load(desc, &info, addr);
> +	if (ret <= 0) {
> +		ret = blk_dread(desc, info.start, info.size, addr);
> +		if (ret < 0) {
> +			printf("Error: Failed to read partition %s\n", buf);
> +			goto fail;
> +		}
>  	}
>  
>  	sprintf(buf, "abootimg_%s_ptr", argv[3]);
> diff --git a/include/image.h b/include/image.h
> index c5b288f62b..7d8ff40c3f 100644
> --- a/include/image.h
> +++ b/include/image.h
> @@ -1826,6 +1826,18 @@ struct andr_image_data;
>  bool android_image_get_data(const void *boot_hdr, const void *vendor_boot_hdr,
>  			    struct andr_image_data *data);
>  
> +/**
> + * android_image_get_valuable_size() - get the size of the android image
> + *
> + * This function checks if the image is Android boot image and returns the
> + * valuable size of the image.
> + *
> + * @hdr_addr: Boot image header address (boot or vendor_boot)
> + *
> + * @return size of the image on success, 0 on failure
> + */
> +size_t android_image_get_valuable_size(const void *hdr_addr);
> +
>  struct andr_boot_img_hdr_v0;
>  
>  /**
> -- 
> 2.40.1