Re: [PATCH bpf-next 4/5] bpf, x64: rework pro/epilogue and tailcall handling in JIT

From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
To: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Cc: "Daniel Borkmann" <daniel@iogearbox.net>,
	"Alexei Starovoitov" <ast@kernel.org>, bpf <bpf@vger.kernel.org>,
	"Network Development" <netdev@vger.kernel.org>,
	"Björn Töpel" <bjorn.topel@intel.com>,
	"Karlsson, Magnus" <magnus.karlsson@intel.com>
Subject: Re: [PATCH bpf-next 4/5] bpf, x64: rework pro/epilogue and tailcall handling in JIT
Date: Fri, 17 Jul 2020 09:16:27 -0700	[thread overview]
Message-ID: <CAADnVQLz0ojicbKS4LSjCCb5yaK9xKzB0MEJNYzb+Z6bVAFt4g@mail.gmail.com> (raw)
In-Reply-To: <20200717105744.GB11239@ranger.igk.intel.com>

On Fri, Jul 17, 2020 at 4:02 AM Maciej Fijalkowski
<maciej.fijalkowski@intel.com> wrote:
>
> On Thu, Jul 16, 2020 at 07:16:24PM -0700, Alexei Starovoitov wrote:
> > On Fri, Jul 17, 2020 at 01:06:07AM +0200, Daniel Borkmann wrote:
> > > > +                         ret = bpf_arch_text_poke(poke->tailcall_bypass,
> > > > +                                                  BPF_MOD_JUMP,
> > > > +                                                  NULL, bypass_addr);
> > > > +                         BUG_ON(ret < 0 && ret != -EINVAL);
> > > > +                         /* let other CPUs finish the execution of program
> > > > +                          * so that it will not possible to expose them
> > > > +                          * to invalid nop, stack unwind, nop state
> > > > +                          */
> > > > +                         synchronize_rcu();
> > >
> > > Very heavyweight that we need to potentially call this /multiple/ times for just a
> > > /single/ map update under poke mutex even ... but agree it's needed here to avoid
> > > racing. :(
> >
> > Yeah. I wasn't clear with my suggestion earlier.
> > I meant to say that synchronize_rcu() can be done between two loops.
> > list_for_each_entry(elem, &aux->poke_progs, list)
> >    for (i = 0; i < elem->aux->size_poke_tab; i++)
> >         bpf_arch_text_poke(poke->tailcall_bypass, ...
> > synchronize_rcu();
> > list_for_each_entry(elem, &aux->poke_progs, list)
> >    for (i = 0; i < elem->aux->size_poke_tab; i++)
> >         bpf_arch_text_poke(poke->poke->tailcall_target, ...
> >
> > Not sure how much better it will be though.
> > text_poke is heavy.
> > I think it's heavier than synchronize_rcu().
> > Long term we can do batch of text_poke-s.
>
> Yeah since we introduce another poke target we could come up with
> preparing the vector of pokes as you're saying?

yes. eventually. Pls keep it simple for now.
The logic is already quite complex.