feat(plat/st): improve FIP image loading from MMC

Instead of using a scratch buffer of 512 bytes, we can directly use the
image address and max size. The mmc_block_dev_spec struct info is then
overwritten for each image with this info, except FW_CONFIG and GPT
table which will still use the scratch buffer.
This allows using multiple blocks read on MMC, and so improves the boot
time.
A cache invalidate is required for the remaining data not used from the
first and last blocks read. It is not required for FW_CONFIG_ID,
as it is in scratch buffer in SYSRAM, and also because bl_mem_params
struct is overwritten in this case. This should also not be done if
the image is not found (OP-TEE extra binaries when using SP_min).

Change-Id: If3ecfdfe35bb9db66284036ca49c4bd1be4fd121
Signed-off-by: Yann Gautier <yann.gautier@foss.st.com>
2 files changed