On Fri, Jan 15, 2021 at 11:33:40AM -0800, Atish Patra wrote:
On Fri, Jan 15, 2021 at 8:00 AM Ilias Apalodimas ilias.apalodimas@linaro.org wrote:
Atish reports than on RISC-V, accessing the EFI variables causes a kernel panic. An objdump of the file verifies that, since the global pointer for efi_var_buf ends up in .GOT section which is not mapped in virtual address space for Linux.
<snip of efi_var_mem_find>
0000000000000084 <efi_var_mem_find>: 84: 715d addi sp,sp,-80
- objdump -dr
0000000000000086 <.LCFI2>: 86: e0a2 sd s0,64(sp) 88: fc26 sd s1,56(sp) 8a: e486 sd ra,72(sp) 8c: f84a sd s2,48(sp) 8e: f44e sd s3,40(sp) 90: f052 sd s4,32(sp) 92: ec56 sd s5,24(sp) 94: 00000497 auipc s1,0x0 94: R_RISCV_GOT_HI20 efi_var_buf 98: 0004b483 ld s1,0(s1) # 94 <.LCFI2+0xe> 98: R_RISCV_PCREL_LO12_I .L0 98: R_RISCV_RELAX *ABS*
- objdump -t
0000000000000084 g F .text.efi_runtime 00000000000000b8 efi_var_mem_find
With the patch applied:
- objdump -dr
0000000000000086 <.LCFI2>: 86: e0a2 sd s0,64(sp) 88: fc26 sd s1,56(sp) 8a: e486 sd ra,72(sp) 8c: f84a sd s2,48(sp) 8e: f44e sd s3,40(sp) 90: f052 sd s4,32(sp) 92: ec56 sd s5,24(sp) 94: 00000497 auipc s1,0x0 94: R_RISCV_PCREL_HI20 .LANCHOR0 94: R_RISCV_RELAX *ABS* 98: 00048493 mv s1,s1 98: R_RISCV_PCREL_LO12_I .L0 98: R_RISCV_RELAX *ABS*
- objdump -t
0000000000000008 l O .data.efi_runtime 0000000000000008 efi_var_buf
On arm64 this works, because there's no .GOT entries for this and everything is converted to relative references.
Just curious to know: Is it because of linker script magic or compiler optimization ? I might have missed something but I did not find anything relevant in the arm64 linker scripts.
I replied on the original thread regarding what happens in Arm and we get the feature working [1]
[1] https://lists.denx.de/pipermail/u-boot/2021-January/437484.html Cheers /Ilias