[Intel-gfx] [PATCH 1/1] drm/i915: Tracepoints for profiling DC9 entry and exit programming

Daniel Vetter daniel at ffwll.ch
Mon Jun 22 05:56:50 PDT 2015


On Sun, Jun 21, 2015 at 10:20:09AM +0530, Sagar Arun Kamble wrote:
> On Wed, 2015-06-17 at 13:30 +0200, Daniel Vetter wrote:
> > On Wed, Jun 17, 2015 at 04:21:10PM +0530, Sagar Arun Kamble wrote:
> > > DC9 entry and exit programming is critical part of suspend/resume
> > > sequences. This patch adds tracepoints that can help analyze time
> > > taken using analyze_suspend.py/FTrace.
> > > 
> > > Change-Id: I22fca5313c4349f8937eeb5a1c441c8ef76e5f4e
> > > Signed-off-by: Sagar Arun Kamble <sagar.a.kamble at intel.com>
> > 
> > Tracepoints are for system profiling when benchmarking and all that. What
> > exactly is this here for?
> Hi Daniel,
> These trace points are added so that we can know how much time is spent
> in entry/exit programming of DC9 state which is done during
> runtime/system suspend. Analyzing these times is critical for resume
> time KPIs.

What's a KPI? Also if we want to instrument our runtime pm interface then
I think we should do this across the driver (probably at the power well
interface). Also you seem to instrument only suspend resume and not rpm
here.

> > Also note that tracepoints are pretty much ABI, so high standards apply
> > here.
> Already these tracepoints are present at PM Core level.

You're adding new labels and stuff and (see above) it looks not all that
well thought out yet.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch


More information about the Intel-gfx mailing list