RISCV64 driver does not have the proper EFI subsystem set about gnu-efi HOT 5 CLOSED

Yep, no 16bit REL
(https://github.com/bminor/binutils-gdb/blob/92f59183ab085f5762650c3242a33119d161e668/bfd/elfxx-riscv.c#L973)
RISCV64 will have to use generator mode (like fwupd-efi) and/or objcopy
It can't support delayed symbol mode

from gnu-efi.

Any chance we could just combine the 16-bit Subsystem and subsequent 16-bit DllCharacteristics into a 32-bit, since we have 32bit REL? DllCharacteristics is always zero for UEFI and we are dealing with little endian values, so, on paper, this looks like a simpler workaround.

However, while my testing doing just that compiles without throwing any error (and without the need for the #ifndef EFI_SUBSYSTEM), the resulting Subsystem and DllCharacteristics fields in the PE are always set to 0. Yet, if I change the name of the EFI_SUBSYSTEM variable in the .S to something else, the linker will throw an error about undefined reference for that other name, which tends to indicate that the linker does see the 32-bit EFI_SUBSYSTEM as a delayed symbol, but it somehow still fail to populate it with the value we pass in --defsym=EFI_SUBSYSTEM...

Am I missing something obvious? Or is just another RISCV64 toolchain limitation/bug?

from gnu-efi.

Yet, if I change the name of the EFI_SUBSYSTEM variable in the .S to something else, the linker will throw an error about undefined reference for that other name, which tends to indicate that the linker does see the 32-bit EFI_SUBSYSTEM as a delayed symbol, but it somehow still fail to populate it with the value we pass in --defsym=EFI_SUBSYSTEM...

That also seems weird 😕 , it should work like that.

Also DllCharacteristics is now populated by NX bit data for UEFI

from gnu-efi.

Hmm, further testing shows that the value we pass in --defsym=EFI_SUBSYSTEM ends up in the .rodata section rather than in the header...

from gnu-efi.

All things considered, I think our best course of action for now is to just add an extra step to our build that sets the SubSystem property after the objcopy step, and I have just proposed a patchset that does this.

from gnu-efi.