On Sat, 26 Apr 2014 23:17:57 +0200
Luc Verhaegen <[email protected]> wrote:
> >
> > If there is any undocumented knob to increase the DEBE DMA burst size,
> > that would very likely resolve the problem. Another interesting thing
> > is that PRECHARGE/ACTIVATE latency is critical. Higher DRAM clock speed
> > only helps indirectly by reducing this latency.
> >
> > And by the way, the DEBE in Allwinner A20 does not seem to be affected.
> > Only Allwinner A10 suffers from these FullHD troubles.
>
> Some obvious questions...
>
> Were you testing this with the scaler (DEFE) enabled,
No, this applies to DEBE only. If DEFE is doing the framebuffer
scanout, then everything is (somewhat) better.
> or were you using layers directly?
Just an ordinary framebuffer, which means a single primary layer and
nothing else.
> Are you seeing this over VGA as well?
Yes. And you should be also seeing this over either HDMI or VGA on the
Allwinner A10 hardware if you try to reproduce the problem.
I have spent some time during the last few weeks experimenting
with the DRAM controller. And I think that I know the DDR3 kung
fu a little bit better now :-) This makes me believe that the DEBE
scanout issue at 1920x1080-32@60Hz on A10 is really unfixable
by simply increasing the DRAM clock speed because we are crossing
a certain threshold. DEBE is sucking the memory bandwidth dry by
filling it with the page miss bubbles. At the 60Hz refresh rate,
the gap between these bubbles becomes so small, that a page miss
triggered by a CPU memory access does not fully fit this gap.
As a result, the next DEBE memory access gets postponed a bit,
and we cant send enough data over HDMI in time. Because page miss
penalties are determined by the tRP/tRCD/CL delays (in nanoseconds)
from the DDR3 datasheets, this becomes a hard limit.
Now an interesting question is indeed whether the 1920x1080-32@60Hz
scanout done by DEFE can be really made reliable by increasing the DRAM
clock speed and optimizing the timings. At the 480MHz DRAM clock speed
we were not quite there yet:
http://ssvb.github.io/2013/06/27/fullhd-x11-desktop-performance-of-the-allwinner-a10.html
The sudden memset performance drop and the HDMI signal disruption
are both different symptoms of the same problem.
And again, this all applies to A10 only. The A20 users should not worry.
--
Best regards,
Siarhei Siamashka
--
You received this message because you are subscribed to the Google Groups
"linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
