On Tue, Aug 30, 2016 at 11:00:38PM +0200, Daniel Borkmann wrote: > On 08/30/2016 10:48 PM, Alexei Starovoitov wrote: > >On Tue, Aug 30, 2016 at 10:22:46PM +0200, Jakub Kicinski wrote: > >>On Tue, 30 Aug 2016 21:07:50 +0200, Daniel Borkmann wrote: > >>>>Having two modes seems more straight forward and I think we would only > >>>>need to pay attention in the LD_IMM64 case, I don't think I've seen > >>>>LLVM generating XORs, it's just the cBPF -> eBPF conversion. > >>> > >>>Okay, though, I think that the cBPF to eBPF migration wouldn't even > >>>pass through the bpf_parse() handling, since verifier is not aware on > >>>some of their aspects such as emitting calls directly (w/o *proto) or > >>>arg mappings. Probably make sense to reject these (bpf_prog_was_classic()) > >>>if they cannot be handled anyway? > >> > >>TBH again I only use cBPF for testing. It's a convenient way of > >>generating certain instruction sequences. I can probably just drop > >>it completely but the XOR patch is just 3 lines of code so not a huge > >>cost either... I'll keep patch 6 in my tree for now. > > > >if xor matching is only need for classic, I would drop that patch > >just to avoid unnecessary state collection. The number of lines > >is not a concern, but extra state for state prunning is. > > > >>Alternatively - is there any eBPF assembler out there? Something > >>converting verifier output back into ELF would be quite cool. > > > >would certainly be nice. I don't think there is anything standalone. > >btw llvm can be made to work as assembler only, but simple flex/bison > >is probably better. > > Never tried it out, but seems llvm backend doesn't have asm parser > implemented? > > $ clang -target bpf -O2 -c foo.c -S -o foo.S > $ llvm-mc -arch bpf foo.S -filetype=obj -o foo.o > llvm-mc: error: this target does not support assembly parsing. > > LLVM IR might work, but maybe too high level(?); alternatively, we could > make bpf_asm from tools/net/ eBPF aware for debugging purposes. If you > have a toolchain supporting libbfd et al, you could probably make use > of bpf_jit_dump() (like JITs do) and then bpf_jit_disasm tool (from > same dir as bpf_asm).
yes. llvm-based bpf asm is not complete. It's straightforward to add though. It won't be going through IR. Only 'mc' (machine instruciton) layer.
