https://sourceware.org/bugzilla/show_bug.cgi?id=27180
--- Comment #2 from Nelson Chu <nelsonc1225 at sourceware dot org> --- Hi Julius, Thanks for reporting this. Your assumption is correct, the PCREL relocs are converted to the directly access relocs, but we don't update them to the relocation table, so we will get segment fault when the --emit-relocs is set. Attached is my proposed solution, without fully testing. My idea is that try to update the PCREL relocs and their referenced symbols to the corresponding HI20/ LO12_I/S relocs and correct target symbols. But there is a problem for the R_RISCV_GOT_HI20 + PCREL_LO12. If we convert it to HI20 + LO12, then we should update the symbol to the got entry (maybe need to add one, but seems hard to get the dynindx or ...). Or alternatively, we probably can encode the symbol value to the r_addend directly, without referring to a symbol, like what RELATIVE relocs do. I don't know which one is better, so I just report the dangerous relocs for those changed R_RISCV_GOT_HI20, rather than report segment fault. -- You are receiving this mail because: You are on the CC list for the bug.
