Change Details

Hi, I noticed that the ATF for some SoCs is not linked to an aligned address. E.g. ZYNQ-US. There is a linker flag for on GNU-LD "--nmagic" which disables page alignment. The problem does not affect the section table but the program headers. Without `--nmagic` the Program Headers (PHDRS) look like this: $ aarch64-linux-gnu-readelf -l bl31.elf ``` Elf file type is EXEC (Executable file) Entry point 0xfffea000 There are 2 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000000000 0x00000000fffe0000 0x00000000fffe0000 0x0000000000016a4d 0x000000000001f000 RWE 0x10000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10 Section to Segment mapping: Segment Sections... 00 .text .rodata .data stacks .bss xlat_table coherent_ram 01 ``` while with `--nmagic` the program headers look like this: $ aarch64-linux-gnu-readelf -l bl31.elf ``` Elf file type is EXEC (Executable file) Entry point 0xfffea000 There are 2 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000001000 0x00000000fffea000 0x00000000fffea000 0x000000000000ca4d 0x0000000000015000 RWE 0x1000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10 Section to Segment mapping: Segment Sections... 00 .text .rodata .data stacks .bss xlat_table coherent_ram 01 ``` . The BIG difference is that that without the PHDRS basically say to load the file copy (<start of elf file>+0x0)++0x16a4d to 0xFFFE0000 on the platform . This includes the ELF header by the way + overwrites for ZYNQ-US the FSBL-Handoff table when using Xilinx FSBL. When I link with page alignment disabled the PHDRS basically say to load the file copy (<start of elf file>+0x1000)++0xca4d to 0xFFFEA000 on the platform . This time the FSBL-Handoff table gets not overwritten on ZYNQ-US. Some debug solutions on the market perfer to use the PHDRS (e.g. to load code for a foreign architecture) over the section table. Thus it destroys the in my situation the FSBL handoff. Is there a reason why ATF does not link with --nmagic? Is there another reason why the debug symbols are not generated when compiling in release?

Hi, I noticed that the ATF for some SoCs is not linked to an aligned address. E.g. ZYNQ-US. There is a linker flag for on GNU-LD "--nmagic" which disables page alignment. The problem does not affect the section table but the program headers. Without `--nmagic` the Program Headers (PHDRS) look like this: $ aarch64-linux-gnu-readelf -l bl31.elf ``` Elf file type is EXEC (Executable file) Entry point 0xfffea000 There are 2 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000000000 0x00000000fffe0000 0x00000000fffe0000 0x0000000000016a4d 0x000000000001f000 RWE 0x10000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10 Section to Segment mapping: Segment Sections... 00 .text .rodata .data stacks .bss xlat_table coherent_ram 01 ``` while with `--nmagic` the program headers look like this: $ aarch64-linux-gnu-readelf -l bl31.elf ``` Elf file type is EXEC (Executable file) Entry point 0xfffea000 There are 2 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000001000 0x00000000fffea000 0x00000000fffea000 0x000000000000ca4d 0x0000000000015000 RWE 0x1000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10 Section to Segment mapping: Segment Sections... 00 .text .rodata .data stacks .bss xlat_table coherent_ram 01 ``` . The BIG difference is that that without `--nmagic` the PHDRS basically say: to load the file copy (<start of elf file>+0x0)++0x16a4d to 0xFFFE0000 on the platform . This includes the ELF header by the way + overwrites for ZYNQ-US the FSBL-Handoff table when using Xilinx FSBL. When I link with page alignment disabled (with `--nmagic`) the PHDRS basically say: to load the file copy (<start of elf file>+0x1000)++0xca4d to 0xFFFEA000 on the platform . This time the FSBL-Handoff table gets not overwritten on ZYNQ-US. Some debug solutions on the market perfer to use the PHDRS (e.g. to load code for a foreign architecture) over the section table. Thus it destroys the in my situation the FSBL handoff. Is there a reason why ATF does not link with --nmagic? Is there another reason why the debug symbols are not generated when compiling in release?

Hi, I noticed that the ATF for some SoCs is not linked to an aligned address. E.g. ZYNQ-US. There is a linker flag for on GNU-LD "--nmagic" which disables page alignment. The problem does not affect the section table but the program headers. Without `--nmagic` the Program Headers (PHDRS) look like this: $ aarch64-linux-gnu-readelf -l bl31.elf ``` Elf file type is EXEC (Executable file) Entry point 0xfffea000 There are 2 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000000000 0x00000000fffe0000 0x00000000fffe0000 0x0000000000016a4d 0x000000000001f000 RWE 0x10000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10 Section to Segment mapping: Segment Sections... 00 .text .rodata .data stacks .bss xlat_table coherent_ram 01 ``` while with `--nmagic` the program headers look like this: $ aarch64-linux-gnu-readelf -l bl31.elf ``` Elf file type is EXEC (Executable file) Entry point 0xfffea000 There are 2 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align LOAD 0x0000000000001000 0x00000000fffea000 0x00000000fffea000 0x000000000000ca4d 0x0000000000015000 RWE 0x1000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10 Section to Segment mapping: Segment Sections... 00 .text .rodata .data stacks .bss xlat_table coherent_ram 01 ``` . The BIG difference is that that without `--nmagic` the PHDRS basically say: to load the file copy copy (<start of elf file>+0x0)++0x16a4d to 0xFFFE0000 on the platform . This includes the ELF header by the way + overwrites for ZYNQ-US the FSBL-Handoff table when using Xilinx FSBL. When I link with page alignment disabled (with `--nmagic`) the PHDRS basically say: to load the file copy copy (<start of elf file>+0x1000)++0xca4d to 0xFFFEA000 on the platform . This time the FSBL-Handoff table gets not overwritten on ZYNQ-US. Some debug solutions on the market perfer to use the PHDRS (e.g. to load code for a foreign architecture) over the section table. Thus it destroys the in my situation the FSBL handoff. Is there a reason why ATF does not link with --nmagic? Is there another reason why the debug symbols are not generated when compiling in release?