Re: [PATCH v3 bpf-next] RV32G eBPF JIT

From: Luke Nelson <lukenels@cs.washington.edu>
To: "Björn Töpel" <bjorn.topel@gmail.com>
Cc: bpf <bpf@vger.kernel.org>, Jiong Wang <jiong.wang@netronome.com>,
	Xi Wang <xi.wang@gmail.com>,
	Paul Walmsley <paul.walmsley@sifive.com>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Albert Ou <aou@eecs.berkeley.edu>,
	Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Martin KaFai Lau <kafai@fb.com>, Song Liu <songliubraving@fb.com>,
	Yonghong Song <yhs@fb.com>, Andrii Nakryiko <andriin@fb.com>,
	linux-riscv@lists.infradead.org,
	LKML <linux-kernel@vger.kernel.org>,
	Netdev <netdev@vger.kernel.org>
Subject: Re: [PATCH v3 bpf-next] RV32G eBPF JIT
Date: Fri, 28 Feb 2020 14:20:11 -0800	[thread overview]
Message-ID: <CADasFoAB8PNRTQifs6-dapGWY=0J2usYNCYqg5g--jYG37rybQ@mail.gmail.com> (raw)
In-Reply-To: <CAJ+HfNiOoLWpQAPhKL6cUVTZ0vTwuSabZzypzAmbRThD3ChGzA@mail.gmail.com>

On Fri, Feb 28, 2020 at 6:25 AM Björn Töpel <bjorn.topel@gmail.com> wrote:
>
> Luke/Xi, apologies for the slow reponse. (All my RV work is done on
> non-payed time, so that's that. :-)) Very nice that you're still
> working on it!

No worries, thanks for reviewing!

> >
> > - ALU64 DIV/MOD: Requires loops to implement on 32-bit hardware.
>
> Even though it requires loops, JIT support would be nice. OTOH, arm
> doesn't support that either...

We could probably implement this by calling into a helper function
for BPF_ALU64 | BPF_DIV/BPF_MOD, but none of the other JITs (e.g.,
arm,x86) for 32-bit architectures do this. We could add support to
rv32 and other architectures in a future patch.

> > - BPF_XADD | BPF_DW: Requires either an 8-byte atomic instruction
> >   in the target (which doesn't exist in RV32), or acqusition of
> >   locks in generated code.
> >
>
> Any ideas how this could be addressed for RV32G?

I don't believe there is a simple way to correctly implement BPF_XADD
| BPF_DW without hardware support for 64-bit atomic operations,
like for other 32-bit JITs.

> In general I agree with Song; It would be good if the 64/32 bit
> variants would share more code. RISC-V 64/32 *are* very similar, and
> we should be able to benefit from that codewise.
>
> Pull out all functions are that common -- most of the emit_*, the
> parts of the registers, the branch relaxation, and context
> structs. Hopefully, the acutal RV32/64 specfic parts will be pretty
> small.

Thanks for the suggestion. I'll factor out the common functionality
into a header for the next revision.

> Finally; There are some checkpatch issues: run 'checkpatch.pl --strict'.
> [...]

I'll fix the rest of the issues you found in the next revision.

> Thanks for the hard work! I'll take it for a spin, with help from the
> guide above, this weekend!

Thanks again!
- Luke