Hello,
I am making a toy OS kernel on RISC-V platform. The kernel image I built is a ELF file that should be booted using U-Boot's bootelf command. However, I was getting such error when doing bootelf -p <memory address of my ELF file at memory>:
Unhandled exception: Store/AMO access fault
My kernel file is a 64-bit ELF file with the structure of: (the full readelf report is attached in the email)
``` Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flags Align RISCV_ATTRIBUT 0x000000000000c7a0 0x0000000000000000 0x0000000000000000 0x0000000000000039 0x0000000000000000 R 0x1 LOAD 0x0000000000001000 0x0000000080200000 0x0000000080200000 0x0000000000001000 0x0000000000001000 R 0x1000 LOAD 0x0000000000002000 0x0000000080201000 0x0000000080201000 0x000000000000a7a0 0x0000000000012890 RWE 0x1000 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 0x10
Section to Segment mapping: Segment Sections... 00 .riscv.attributes 01 .entry 02 .text .rodata .data .bss 03 ```
The function for loading x64 ELF file to the memory is "unsigned long load_elf64_image_phdr(unsigned long addr)" at lib/elf.c. As I looked into it, the function enumerate every segment and load the segment to the memory. The real problem is that not every segment has to be loaded into the memory. The segment should be loaded if and only if its type is "LOAD". In another words, the function should check "phdr->p_type" first before loading the segment.
I have also looked up codes in QEMU. QEMU has "-kernel" option to load image file directly to the memory, which supports ELF file loading. QEMU checks "phdr->p_type" before loading segments.
Regards, Chenxi Huang ------------
Background Information:
* Linux 5.19.0-35-generic #36~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Fri Feb 17 15:17:25 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
* RISC-V GNU Toolchain Version: v2023.01.31 65056bd@riscv-collab/riscv-gnu-toolchain * U-Boot f8f47e6f