Question about PSR training and idle frames

Question about PSR training and idle frames

[Siva Chandra]

Hi,

I would like to know what "training" and "idle frames" mean wrt PSR. I
apologize if this is the wrong list for such questions.

I am looking at the i915 driver in 3.10 which sets idle_frames to 1
(function intel_edp_psr_enable_source in intel_dp.c) when enabling
PSR. What does this mean, and what should one observe if idle_frames
is set to a higher number?

Thanks,
Siva Chandra

Re: Question about PSR training and idle frames

[Ville Syrjälä]

On Tue, Feb 25, 2014 at 10:06:19AM -0800, Siva Chandra wrote:
> Hi,
> 
> I would like to know what "training" and "idle frames" mean wrt PSR. I
> apologize if this is the wrong list for such questions.

Training just refers to the DP link training that may be necessary
after exiting PSR.

> 
> I am looking at the i915 driver in 3.10 which sets idle_frames to 1
> (function intel_edp_psr_enable_source in intel_dp.c) when enabling
> PSR. What does this mean, and what should one observe if idle_frames
> is set to a higher number?

IIRC idle frames indicates how many consecutive static frames must be
displayed before the hardware will enter PSR. So the higher you make
it, the longer you burn power scanning out the same picture.

But there might be some tradeoff with the PSR exit latency vs.
interactive content, and maybe there's some power cost to entering/exiting
PSR so if you every time end up exiting PSR immediately after entering
it, it might make sense to not enter PSR in the first place. These
are just gut feeling type of things, I have no numbers to back up anything.

-- 
Ville Syrjälä
Intel OTC

Re: Question about PSR training and idle frames

[Siva Chandra]

On Tue, Feb 25, 2014 at 10:34 AM, Ville Syrjälä
<ville.syrjala@linux.intel.com> wrote:
>> I am looking at the i915 driver in 3.10 which sets idle_frames to 1
>> (function intel_edp_psr_enable_source in intel_dp.c) when enabling
>> PSR. What does this mean, and what should one observe if idle_frames
>> is set to a higher number?
>
> IIRC idle frames indicates how many consecutive static frames must be
> displayed before the hardware will enter PSR. So the higher you make
> it, the longer you burn power scanning out the same picture.

Lets say the sequence of frames is {F(1), F(2), ... F(j), F(j+1), ...}.
Then, does idle_frames=1 mean that the hardware will enter PSR
whenever it finds F(j+1) == F(j) for any j?

If that is indeed the case, and if I have a static display on screen
at 60fps, should the hardware enter PSR in ~1/30 secs after the
display went static?

I have a few follow up questions:

1. There is a param named EDP_PSR_PERF_CNT. What is this measuring,
and in what units?
2. I understand that PSR entry is managed by hardware. Is exit also
managed by hardware? AFAICT, it is managed by hardware; if F(j+1) !=
F(j), it should exit PSR.
3. Are there any interrupts that one can listen to to note PSR entry
and exit in software?

Thanks in advance,
Siva Chandra
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx