Disassembler not correctly representing negative offsets

[Oshawk]

An instruction such as:

79a6b0ff00000000

Is disassembled to:

ldxdw r6, [r10+0xffb0]

Whereas I believe it should be:

ldxdw r6, [r10-0x50]

It seems like this applies to all instructions involving offsets.

[Oshawk]

lddw is also incorrectly represented. It only takes one imm, not two.

[dmakarov]

I forgot rbpf has its own disassembler. Thanks for reporting these errors.

[dmakarov]

lddw is also incorrectly represented. It only takes one imm, not two.

I don't see two immediate operands in disassembler output.

[Oshawk]

I mean lddw is the only 16-byte instruction. It takes the imm from two consecutive 8-bytes (see https://github.com/solana-labs/rbpf/blob/main/src/ebpf.rs#L554). The disassembly only takes one imm.

[dmakarov]

I mean lddw is the only 16-byte instruction. It takes the imm from two consecutive 8-bytes (see https://github.com/solana-labs/rbpf/blob/main/src/ebpf.rs#L554). The disassembly only takes one imm.

yes, it's a 64-bit immediate value, and it is decoded from the 128-bits of the instruction code. There are no two immediates. Not all bits of the 128-bit instruction code are used. Can you give an example of code that was disassembled incorrectly?

[Oshawk]

Wait. Sorry, I'm stupid. I didn't realise that augment_lddw_unchecked had been applied prior to disassemble_instruction. I think it disassembles correctly.