Re: [RFC] ftrace: Add support to keep some functions out of ftrace

[Alexei Starovoitov]

On Mon, Aug 15, 2022 at 7:33 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, Aug 15, 2022 at 07:25:24AM -0700, Alexei Starovoitov wrote:
> > On Mon, Aug 15, 2022 at 5:37 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > >
> > > On Mon, Aug 15, 2022 at 11:44:32AM +0200, Jiri Olsa wrote:
> > > > On Mon, Aug 15, 2022 at 10:03:17AM +0200, Peter Zijlstra wrote:
> > > > > On Sat, Aug 13, 2022 at 03:02:52PM -0400, Steven Rostedt wrote:
> > > > > > On Fri, 12 Aug 2022 23:18:15 +0200
> > > > > > Jiri Olsa <olsajiri@xxxxxxxxx> wrote:
> > > > > >
> > > > > > > the patch below moves the bpf function into sepatate object and switches
> > > > > > > off the -mrecord-mcount for it.. so the function gets profile call
> > > > > > > generated but it's not visible to ftrace
> > > > >
> > > > > Why ?!?
> > > >
> > > > there's bpf dispatcher code that updates bpf_dispatcher_xdp_func
> > > > function with bpf_arch_text_poke and that can race with ftrace update
> > > > if the function is traced
> > >
> > > I thought bpf_arch_text_poke() wasn't allowed to touch kernel code and
> > > ftrace is in full control of it ?
> >
> > ftrace is not in "full control" of nop5 and must not be.
>
> It is in full control of the 'call __fentry__'. Absolute full NAK on you
> trying to make it otherwise.

Don't mix 'call fentry' generated by the compiler with nop5 inserted
by macroses or JITs.

> > Soon we will have nop5 in the middle of the function.
> > ftrace must not touch it.
>
> How are you generating that NOP and what for?

We're generating nop5-s in JITed code to further
attach to. For example when one bpf prog is being replaced by another.
Currently it's in the func prologue only.
In the future it will be anywhere in the body.