Re: [RFC PATCH 0/6] [RFC] Faultable tracepoints (v2)

[Michael Jeanson]

[ Adding Mathieu Desnoyers in CC ]

On 2021-02-23 21 h 16, Steven Rostedt wrote:
> On Thu, 18 Feb 2021 17:21:19 -0500
> Michael Jeanson <mjeanson@xxxxxxxxxxxx> wrote:
>
> > This series only implements the tracepoint infrastructure required to
> > allow tracers to handle page faults. Modifying each tracer to handle
> > those page faults would be a next step after we all agree on this piece
> > of instrumentation infrastructure.
>
> I started taking a quick look at this, and came up with the question: how
> do you allow preemption when dealing with per-cpu buffers or storage to
> record the data?
>
> That is, perf, bpf and ftrace are all using some kind of per-cpu data, and
> this is the reason for the need to disable preemption. What's the solution
> that LTTng is using for this? I know it has a per cpu buffers too, but does
> it have some kind of "per task" buffer that is being used to extract the
> data that can fault?
>
> -- Steve