Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Tue, 3 Mar 2020 15:56:05 -0800
From:   Alexei Starovoitov <alexei.starovoitov@gmail.com>
To:     KP Singh <kpsingh@chromium.org>
Cc:     Andrii Nakryiko <andrii.nakryiko@gmail.com>,
        open list <linux-kernel@vger.kernel.org>,
        bpf <bpf@vger.kernel.org>, Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Paul Turner <pjt@google.com>,
        Florent Revest <revest@chromium.org>,
        Brendan Jackman <jackmanb@chromium.org>
Subject: Re: [PATCH bpf-next 2/7] bpf: JIT helpers for fmod_ret progs
Message-ID: <20200303235604.mdlamwx4z2ws3fzy@ast-mbp>
References: <20200303140950.6355-1-kpsingh@chromium.org>
 <20200303140950.6355-3-kpsingh@chromium.org>
 <CAEf4BzZJ2E2rmyz7k4F7s=EXPbaAX7XncvUcHukX_FYDWeD7BA@mail.gmail.com>
 <20200303222812.GA5265@chromium.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200303222812.GA5265@chromium.org>
User-Agent: NeoMutt/20180223
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Mar 03, 2020 at 11:28:12PM +0100, KP Singh wrote:
> > > +static void align16_branch_target(u8 **pprog)
> > > +{
> > > +       u8 *target, *prog = *pprog;
> > > +
> > > +       target = PTR_ALIGN(prog, 16);
> > > +       if (target != prog)
> > > +               emit_nops(&prog, target - prog);
> > > +       if (target != prog)
> > > +               pr_err("calcultion error\n");
> > 
> > this wasn't in the original code, do you feel like it's more important
> > to check this and print error?
> > 
> > also typo: calculation error, but then it's a bit brief and
> > uninformative message. So I don't know, maybe just drop it?
> 
> Ah, good catch! this is deinitely not intended to be here.
> It's a debug artifact and needs to dropped indeed.

That spurious pr_err() caught my attention as well.
After further analysis there is a bug here.
The function is missing last line:
        *pprog = prog;
Without it the nop insertion is actually not happenning.
Nops are being written, but next insns will overwrite them.
When I noticed it by code review I applied the patches to my tree
and run the tests and, as expected, all tests passed.
The existing test_xdp_veth.sh emits the most amount of unaligned
branches. Since then I've been thinking whether we could add a test
to catch things like this and couldn't come up with a way to test it
without burning a lot of code on it. So let's fix it and move on.
Could you rename this helper? May be emit_align() and pass 16 into it?
The code is not branch target specific. It's aligning the start
of the next instruction.
Also could you add a comment to:
        align16_branch_target(&prog);
        for (i = 0; i < fmod_ret->nr_progs; i++)
                emit_cond_near_jump(&branches[i], prog, branches[i],
                                    X86_JNE);
        kfree(branches);
to say that the loop is updating prior location to jump to aligned
branch target ?