Commit f4dc714aaa2d ("arm64: Turn u-boot.bin back into an ELF file after relocate-rela") introduce REMAKE_ELF option to recreate u-boot.elf from u-boot -> u-boot.bin + DT -> u-boot.elf.
The best is to ilustrate it from make V=1 output cat u-boot-nodtb.bin dts/dt.dtb > u-boot-dtb.bin cp u-boot-dtb.bin u-boot.bin aarch64-linux-gnu-objcopy -I binary -B aarch64 -O elf64-littleaarch64 u-boot.bin u-boot-elf.o aarch64-linux-gnu-ld.bfd u-boot-elf.o -o u-boot.elf --defsym="_start"=0x8000000 -Ttext=0x8000000
Last command has no explicit linker script passed that's why toolchain internal linker script is used. In Binutils 2.32 case it contains SIZEOF_HEADERS symbol which has changed behavior by commit https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=64029e93683a266c38... which result in situation that program headers has changed from (xilinx_zynqmp_mini_defconfig)
Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000010000 0x00000000fffc0000 0x00000000fffc0000 0x0000000000018918 0x0000000000018918 RW 0x10000
to
Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000000000 0x00000000fffb0000 0x00000000fffb0000 0x0000000000028918 0x0000000000028918 RW 0x10000
Xilinx tools like XSDB or Bootgen are using program headers for loading ELF to the right location and by above binutils change ELF is loaded to incorrect location.
The patch is explicitly use u-boot-elf.lds (just cat now) for u-boot.elf recreation which is called when REMAKE_ELF is setup. By purpose u-boot-elf.lds doesn't contain OUTPUT_FORMAT/OUTPUT_ARCH to be able to use by all archs.
Signed-off-by: Michal Simek michal.simek@xilinx.com ---
Changes in v1: - Compare to RFC - MIPS has different elf entry point that's why I replaced simple cat with sed to setup proper entry point
Sadanand, Mahesh: Please correct my description if I am wrong based on our internal discussion about this issue.
Tom: I have been told that it is GCC issue but I found that commit in binutils. I am not doing these makefile stuff that's why I expect this should be changed a little bit. Also arch/u-boot-elf.lds is likely incorrect location but didn't find any better.
--- Makefile | 11 +++++++++-- arch/u-boot-elf.lds | 9 +++++++++ 2 files changed, 18 insertions(+), 2 deletions(-) create mode 100644 arch/u-boot-elf.lds
diff --git a/Makefile b/Makefile index be1513227361..7aebfc3a2de1 100644 --- a/Makefile +++ b/Makefile @@ -1654,12 +1654,19 @@ ifndef PLATFORM_ELFENTRY endif quiet_cmd_u-boot-elf ?= LD $@ cmd_u-boot-elf ?= $(LD) u-boot-elf.o -o $@ \ - --defsym=$(PLATFORM_ELFENTRY)=$(CONFIG_SYS_TEXT_BASE) \ + -T u-boot-elf.lds --defsym=$(PLATFORM_ELFENTRY)=$(CONFIG_SYS_TEXT_BASE) \ -Ttext=$(CONFIG_SYS_TEXT_BASE) -u-boot.elf: u-boot.bin +u-boot.elf: u-boot.bin u-boot-elf.lds $(Q)$(OBJCOPY) -I binary $(PLATFORM_ELFFLAGS) $< u-boot-elf.o $(call if_changed,u-boot-elf)
+quiet_sed_elfentry = SED_ELF $@ +cmd_sed_elfentry = \ + sed "s@PLATFORM_ELFENTRY@$(PLATFORM_ELFENTRY)@g" $< > $@ + +u-boot-elf.lds: arch/u-boot-elf.lds prepare FORCE + $(call if_changed,sed_elfentry) + # MediaTek's ARM-based u-boot needs a header to contains its load address # which is parsed by the BootROM. # If the SPL build is enabled, the header will be added to the spl binary, diff --git a/arch/u-boot-elf.lds b/arch/u-boot-elf.lds new file mode 100644 index 000000000000..d3eb6f3fc1db --- /dev/null +++ b/arch/u-boot-elf.lds @@ -0,0 +1,9 @@ +ENTRY(PLATFORM_ELFENTRY) +SECTIONS +{ + . = PLATFORM_ELFENTRY; + + .data : { + *(.data*) + } +}