Re: [PATCH v4 1/5] perf/core: Add PERF_FORMAT_DROPPED

[Namhyung Kim <namhyung@kernel.org>]

Hello,

On Wed, Oct 23, 2024 at 10:05:32PM +1100, Michael Ellerman wrote:
> Namhyung Kim <namhyung@kernel.org> writes:
> > When a perf_event is dropped due to some kind of (SW-based) filter, it
> > won't generate sample data.  For example, software events drops samples
> > when it doesn't match to privilege from exclude_{user,kernel}.
> >
> > In order to account such dropped samples, add a new counter in the
> > perf_event, and let users can read(2) the number with the new
> > PERF_FORMAT_DROPPED like the lost sample count.
> 
> Are we sure there's no scenario where exposing the dropped event count
> gives an unprivileged user a way to probe what's happening in the
> kernel, which is supposed to be prevented by exclude_kernel?
> 
> Clearly it provides an attacker with some information, ie. the event
> fired in the kernel and was dropped.
> 
> For most events that's not very interesting, but for some maybe it could
> be a useful signal?

Hmm.. good point.  It'd give some information to users.  I'm not sure
how much impact it'd have, but there are some folks who want to know
exact number of samples including dropped ones to reconstruct total
period for the monitoring session.

> 
> On the other hand most CPU PMUs implement filtering in hardware, which
> this won't affect, so maybe I'm being too paranoid.

Right, it might be possible to estimate some numbers by comparing with
similar events in the core PMU that implements HW filtering even without
this interface IMHO.

Thanks,
Namhyung