Re: [PATCH] perf/x86: Fix overlap counter scheduling bug

From: Peter Zijlstra
Date: Tue Nov 08 2016 - 13:27:47 EST


On Tue, Nov 08, 2016 at 05:25:34PM +0000, Liang, Kan wrote:

> > I think all the 0x3 mask need the overlap flag set, since they clearly overlap
> > with the 0x1 masks. That would improve the scheduling.
> >
>
> How much the overlap hint can improve the scheduling?
> Because there is not only snbep_uncore_cbox, but also other uncore events
> which have overlapping masks.


Hurm, not much. We're saved by the fact that we schedule from wwin to
wmax, which means that we first place the 0x01 events, and then try and
fit the 0x03 events on top. That should already be good.

/me ponders more..

The comment with EVENT_CONSTRAINT_OVERLAP states: "This is the case if
the counter mask of such an event is not a subset of any other counter
mask of a constraint with an equal or higher weight".

Esp. that latter part is of interest here I think, our overlapping mask
is 0x0e, that has 3 bits set and is the highest weight mask in on the
PMU, therefore it will be placed last. Can we still create a scenario
where we would need to rewind that?

The scenario for AMD Fam15h is we're having masks like:

0x3F -- 111111
0x38 -- 111000
0x07 -- 000111

0x09 -- 001001

And we mark 0x09 as overlapping, because it is not a direct subset of
0x38 or 0x07 and has less weight than either of those. This means we'll
first try and place the 0x09 event, then try and place 0x38/0x07 events.
Now imagine we have:

3 * 0x07 + 0x09

and the initial pick for the 0x09 event is counter 0, then we'll fail to
place all 0x07 events. So we'll pop back, try counter 4 for the 0x09
event, and then re-try all 0x07 events, which will now work.



But given, that in the uncore case, the overlapping event is the
heaviest mask, I don't think this can happen. Or did I overlook
something.... takes a bit to page all this back in.