[Intel-gfx] [RFC 04/10] drm/i915: Expose a PMU interface for perf queries

[Rogozhkin, Dmitry V]

On Tue, 2017-08-29 at 21:21 +0200, Peter Zijlstra wrote:
> On Tue, Aug 29, 2017 at 07:16:31PM +0000, Rogozhkin, Dmitry V wrote:
> > > Pretty strict, people tend to get fairly upset every time we leak stuff.
> > > In fact Debian and Android carry a perf_event_paranoid patch that
> > > default disables _everything_ :-(
> > 
> > Can you say more on that for Debian and Android? What exactly they do?
> > What is the value of perf_event_paranoid there? They disable everything
> > even for root and CAP_SYS_ADMIN? But still they don't remove this from
> > kernel on compilation stage, right? So users can explicitly change
> > perf_event_paranoid to the desired value?
> 
> They introduce (and default to) perf_event_paranoid = 3. Which
> disallows everything for unpriv user, root can still do things IIRC, I'd
> have to dig out the patch.
> 
> This way apps have no access to the syscall, but you can enable it using
> ADB by lowering the setting. So developers still have access, but
> regular apps do not.
> 

Hi, Peter.

How you would feel about the following idea (or close to it):
1. We introduce one more level for perf_event_paranoid=4 (or =3, I am
not sure whether Debian/Android =3 is considered uAPI) which would mean:
"disallow kernel profiling for unpriv, but let individual kernel modules
to have their own settings".
2. We will have i915 PMU custom setting
"/sys/module/i915/parameters/perf_event_paranoid" which will be in
effect only if global perf_event_paranoid=4 (or =3) and prevail over a
global setting

Would anything like that be acceptable upstream? This would permit
customers to configure i915 PMU support for unpriv users separately from
the rest of PMU subsystem.

Regards,
Dmitry.