On 1/10/25 10:49 AM, Peng Fan wrote:
On Sun, Jan 05, 2025 at 10:56:29PM +0100, Marek Vasut wrote:
On 1/3/25 7:45 AM, Alice Guo wrote:
From: Ye Li ye.li@nxp.com
When SPL loading image to secure region, for example, ATF and tee to DDR secure region. Because the USDHC controller is non-secure master, it can't access this region and will cause loading issue.
So override h_spl_load_read to use a trampoline buffer in nonsecure region, then use CPU to copy the image from trampoline buffer to destination secure region.
Signed-off-by: Ye Li ye.li@nxp.com Signed-off-by: Alice Guo alice.guo@nxp.com Reviewed-by: Peng Fan peng.fan@nxp.com
arch/arm/mach-imx/imx9/scmi/soc.c | 40 +++++++++++++++++++++++++++++++++++++++ common/spl/spl_mmc.c | 2 +- 2 files changed, 41 insertions(+), 1 deletion(-)
diff --git a/arch/arm/mach-imx/imx9/scmi/soc.c b/arch/arm/mach-imx/imx9/scmi/soc.c index fefb1a6f4ca58722a5d93c39e4f6a7aaf85aa177..8a7df24afecc1d6d77b7162be639b56af95fd975 100644 --- a/arch/arm/mach-imx/imx9/scmi/soc.c +++ b/arch/arm/mach-imx/imx9/scmi/soc.c @@ -41,6 +41,8 @@ #include <scmi_agent.h> #include <scmi_protocols.h> #endif +#include <spl.h> +#include <mmc.h> DECLARE_GLOBAL_DATA_PTR; @@ -830,3 +832,41 @@ enum boot_device get_boot_device(void) return boot_dev; } #endif
+ulong h_spl_load_read(struct spl_load_info *load, ulong off,
ulong size, void *buf)Can you simply define spl_get_load_buffer() instead to make sure SPL loads your stuff into the correct location ?
spl_get_load_buffer does not help here, unless do a big change to spl_mmc.c h_spl_load_read is set to info->read, info->read is not just for reading headers, it also needs to read container image data.
This hack in generic code is a NAK.
This is set it as __weak, and overwrite in soc code, there are many similar code
The more I am looking at this, the more this simply does not work. Consider these use cases for example where this fails:
- Board boots from SPI NOR - Board boots from filesystem on SD/eMMC
So what about taking a different approach. First, use the SPL loader as-is to load whatever blobs you need to memory locations into which you can load. Second, implement board or arch specific spl_perform_fixups(), which is already empty weak function, and make it move the blobs into the correct location. Would that work ?