[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20160513095718.GB13689@arm.com>
Date: Fri, 13 May 2016 10:57:19 +0100
From: Will Deacon <will.deacon@....com>
To: Zi Shen Lim <zlim.lnx@...il.com>
Cc: Alexei Starovoitov <ast@...nel.org>,
"David S. Miller" <davem@...emloft.net>,
Catalin Marinas <catalin.marinas@....com>,
Daniel Borkmann <daniel@...earbox.net>,
Yang Shi <yang.shi@...aro.org>, netdev@...r.kernel.org,
linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH] arm64: bpf: jit JMP_JSET_{X,K}
On Thu, May 12, 2016 at 11:37:58PM -0700, Zi Shen Lim wrote:
> Original implementation commit e54bcde3d69d ("arm64: eBPF JIT compiler")
> had the relevant code paths, but due to an oversight always fail jiting.
>
> As a result, we had been falling back to BPF interpreter whenever a BPF
> program has JMP_JSET_{X,K} instructions.
>
> With this fix, we confirm that the corresponding tests in lib/test_bpf
> continue to pass, and also jited.
>
> ...
> [ 2.784553] test_bpf: #30 JSET jited:1 188 192 197 PASS
> [ 2.791373] test_bpf: #31 tcpdump port 22 jited:1 325 677 625 PASS
> [ 2.808800] test_bpf: #32 tcpdump complex jited:1 323 731 991 PASS
> ...
> [ 3.190759] test_bpf: #237 JMP_JSET_K: if (0x3 & 0x2) return 1 jited:1 110 PASS
> [ 3.192524] test_bpf: #238 JMP_JSET_K: if (0x3 & 0xffffffff) return 1 jited:1 98 PASS
> [ 3.211014] test_bpf: #249 JMP_JSET_X: if (0x3 & 0x2) return 1 jited:1 120 PASS
> [ 3.212973] test_bpf: #250 JMP_JSET_X: if (0x3 & 0xffffffff) return 1 jited:1 89 PASS
> ...
>
> Fixes: e54bcde3d69d ("arm64: eBPF JIT compiler")
> Signed-off-by: Zi Shen Lim <zlim.lnx@...il.com>
> ---
> arch/arm64/net/bpf_jit_comp.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
> index 031ed08..d0d5190 100644
> --- a/arch/arm64/net/bpf_jit_comp.c
> +++ b/arch/arm64/net/bpf_jit_comp.c
> @@ -478,6 +478,7 @@ emit_cond_jmp:
> case BPF_JGE:
> jmp_cond = A64_COND_CS;
> break;
> + case BPF_JSET:
> case BPF_JNE:
> jmp_cond = A64_COND_NE;
> break;
Are you sure about this? filter.txt says:
jne - Jump on K != A
...
jset - Jump on k & A
so it looks weird wiring them both to the same thing. I'm not sure you
can express this as a simple CMP + B.<cond>.
Will
Powered by blists - more mailing lists