Messages in this thread | | | Date | Mon, 18 May 2015 11:32:07 -0700 | From | Alexei Starovoitov <> | Subject | Re: [RFC PATCH v3 16/37] bpf tools: Collect eBPF programs from their own sections |
| |
On 5/18/15 5:47 AM, Wangnan (F) wrote: > > >>> + prog->insns_cnt = size / sizeof(struct bpf_insn); >>> + memcpy(prog->insns, data, >>> + prog->insns_cnt * sizeof(struct bpf_insn)); >> Doesn't the data need to be swapped? >> >> Thanks, >> Namhyung >> > > I'm not very sure, since they are instructions. Byte order of > instructions and byte order of data > are not always same. Think about ARM. Therefore another choice is to > swap them in kernel, > keep user-kernel interface clean. > > Alexei Starovoitov, do you think we should use uniformed instruction > byte order in both big and little > endian kernel on user-kernel interface, or let userspace feed swapped > instructions to kernel if > endianess is not match?
see my reply to the other patch. endianness is the job of llvm. Let's fix it there.
| |