[U-Boot] [PATCH 2/2] spl: socfpga: stratix10: add hex file output for spl image

[Dalon L Westergreen]

On Tue, 2018-08-28 at 01:54 +0200, Marek Vasut wrote:
> On 08/28/2018 12:03 AM, Dalon L Westergreen wrote:
> On Mon, 2018-08-27 at 21:03 +0200, Marek Vasut wrote:
> On 08/27/2018 05:30 PM, Dalon L Westergreen wrote:
> On Tue, 2018-08-21 at 05:52 +0200, Marek Vasut wrote:
> On 08/20/2018 11:04 PM, Dalon L Westergreen wrote:
> On Mon, 2018-08-20 at 20:33 +0200, Marek Vasut wrote:
> On 08/20/2018 03:54 PM, Dalon Westergreen wrote:
> Stratix10 requires a hex image of the spl for boot.  The hex
> image is added to the FPGA configuration image and loaded to
> the processor memory by the configuration engine.
> 
> Signed-off-by: Dalon Westergreen <dwesterg at gmail.com <mailto:dwesterg at gmail.com> <mailto:dwesterg at gmail.com <mailto:dwesterg at gmail.com>> <mailto:dwesterg at gmail.com <mailto:dwesterg at gmail.com> <mailto:dwesterg at gmail.com <mailto:dwesterg at gmail.com>>>>
> ---
>  scripts/Makefile.spl | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/scripts/Makefile.spl b/scripts/Makefile.spl
> index 76d08fd92b..c424f87e6e 100644
> --- a/scripts/Makefile.spl
> +++ b/scripts/Makefile.spl
> @@ -190,6 +190,7 @@ endif
>  ifdef CONFIG_ARCH_SOCFPGA
>  ALL-$(CONFIG_TARGET_SOCFPGA_GEN5)	+= $(obj)/$(SPL_BIN).sfp
>  ALL-$(CONFIG_TARGET_SOCFPGA_ARRIA10)	+= $(obj)/$(SPL_BIN).sfp
> +ALL-$(CONFIG_TARGET_SOCFPGA_STRATIX10)	+= $(obj)/$(SPL_BIN).hex
>  endif
>  
>  ifdef CONFIG_ARCH_SUNXI
> @@ -299,6 +300,15 @@ OBJCOPYFLAGS_u-boot-x86-16bit-spl.bin := -O binary -j .start16 -j .resetvec
>  $(obj)/u-boot-x86-16bit-spl.bin: $(obj)/u-boot-spl FORCE
>  	$(call if_changed,objcopy)
>  
> +ifdef CONFIG_TARGET_SOCFPGA_STRATIX10
> +OBJCOPYFLAGS_$(SPL_BIN).hex = -I binary -O ihex --change-addresses 0xffe00000
> 
> Why is this --change-address needed ? This smells like a hack of some
> sort ...
> 
> I believe the tool that uses this file expects this offset, that said i have not
> 
> tried using a hex file without this change address applied.  I will try without 
> 
> this, and see what happens.
> 
> You probably need to adjust the link address of the SPL or U-Boot too.
> 
> The link address appears to be correctly set to 0xffe00000 which is the location of the onchip ram.  
> 
> The quartus_cpf tool itself errors out when the --change-address option is not used.  The quartus_cpf 
> 
> checks that the address range of the hex is entirely within the onchip ram.  My suggestion would
> 
> be to leave the --change-address option as is.  
> 
> So what is _not_ in the onchip RAM then ? Is something sticking out of
> the OCRAM , that's why the quartus_cpf fails ? Maybe that's something
> you should fix.
> 
> Thanks Marek, I'm m not very familiar with the hex format
> 
> The intel ihex is terrible (sorry), it's really hard to decode. But
> here's some example:
> https://en.wikipedia.org/wiki/Intel_HEX
> 
> but there only 2 lines that differ when
> 
> using the change address option, the first and last, of the hex. 
> 
> 
> in the file with the change-address option :02000004FFE01B is prepended to the file, and 
> 
> :04000005FFE0000018 is added to the second to last line.  It is followed by :00000001FF (end of file).
> 
> 
> My belief is the change-address is simply specifying the start address of the hex data converted from
> 
> the binary.  This offset is required by the quartus_cpf tool as a form of validation.
> 
> This is what objcopy(1) says:
> 
>        --change-addresses incr
>        --adjust-vma incr
>            Change the VMA and LMA addresses of all sections, as well as
> the start address, by adding incr.  Some object file formats do not
> permit section addresses to be changed
>            arbitrarily.  Note that this does not relocate the sections;
> if the program expects sections to be loaded at a certain address, and
> this option is used to change the sections such
>            that they are loaded at a different address, the program may
> fail.
> 
> So I ran make V=1 , to see how the hex file is generated, and this is how:
>   aarch64-linux-gnu-objcopy  -j .text -j .secure_text -j .secure_data -j
> .rodata -j .data -j .u_boot_list -j .rela.dyn -j .got -j .got.plt -j
> .binman_sym_table -j .text_rest -j .dtb.init.rodata -j .efi_runtime -j
> .efi_runtime_rel -I binary -O ihex --change-addresses 0xffe00000
> spl/u-boot-spl.bin spl/u-boot-spl.hex
> 
> It's generated from u-boot-spl.bin , which is weird. The u-boot.hex is
> generated from u-boot (ELF binary). I think that's what the SPL hex
> should be generated from too, since the ELF binary contains all the
> relocation info.
> 
> So a target similar to u-boot.hex should do:
> OBJCOPYFLAGS_$(SPL_BIN).hex = -O ihex
> 
> $(obj)/$(SPL_BIN).hex: $(obj)/$(SPL_BIN) FORCE
>        $(call if_changed,objcopy)
> 
> Does that work for you ?
> 

Actually no, it is weird that it is using SPL_BIN, the intent was to use the binary with
the devicetree included.  That is the reason for generating from the binary.  I will update
the patch i sent to use $(SPL_BIN)-dtb.bin which was the intent unless there is another
way you would like the hex output to include the dtb?