jalr IDL and description is unclear and potentially invalid

[ThePuzzlemaker]

The ISA manual and Sail code specify that jalr clears the LSB of the calculated target address before jumping. The IDL code and description in this database do not specify this (not even in the jump function, to my knowledge), potentially making these entries invalid.

Current value and location

jump(X[rs1] + imm)

https://github.com/riscv-software-src/riscv-unified-db/blob/main/arch/inst/I/jalr.yaml#L30

Expected value
If I understand IDL correctly, it should be something vaguely like this:

jump({(X[rs1] + imm)[XLEN - 1:1], 1'b0})

The description should also be updated in my opinion to better clarify this behaviour.

[ThinkOpenly]

Good catch!

You are correct about the jump function. It does not clear any bits. It will, however, raise an exception of the target address has low-order bits set (https://github.com/riscv-software-src/riscv-unified-db/blob/4d8b28a63a34be7262c0537962c0f3e8ef88bf43/arch/isa/globals.isa#L703C1-L710C1):

    # raise a misaligned exception if address is not aligned to IALIGN
    if ((ialign() == 16) && ((target_addr & 0x1) != 0)) {
      # the target PC is not halfword-aligned
      raise(ExceptionCode::InstructionAddressMisaligned, mode(), target_addr);
    } else if ((target_addr & 0x3) != 0) {
      # the target PC is not word-aligned
      raise(ExceptionCode::InstructionAddressMisaligned, mode(), target_addr);
    }

Another approach with IDL (to my understanding of IDL):

jump((X[rs1] + imm) & ~1)

[dhower-qc]

Yep, good catch. Either fix (Wren's or Paul's) should fix it. @ThePuzzlemaker, would you be able to make a PR for the fix?