Hi Tomi,
On 5/16/2012 11:08 AM, Tomi Valkeinen wrote:
On Tue, 2012-05-15 at 11:55 -0600, Paul Walmsley wrote:
Hello Tomi,
On Mon, 14 May 2012, Tomi Valkeinen wrote:
I've been doing testing to understand the problem, but so far I don't
have any idea why things go wrong. I haven't found out any logic in
which configuration works and which doesn't. Looks to me that for some
reason the PM prevents DSS from getting data fast enough with certain
fifo thresholds.
I have two ways to avoid the problem, but I've been reluctant to make
patches for those because I feel it's just hiding the problem. One way
is to change DISPC SIDLEMODE or MIDLEMODE to disallow idle/standby. The
other is to use certain fifo threshold values, which just seem to work
for unknown reasons.
Considering that we already have a SIDLEMODE hack in DSS for omap3 when
using DSI, I wonder if the omap3 PM + DSS combination is just plain
broken, and we should disallow idle. I'm not quite sure what are the
implications of that.
I'd appreciate comments from the PM people =).
This may be caused by one of the DPLLs going into autoidle. This can
involve a significant wakeup latency. I'd suggest looking at DPLL3, which
provides the DSS interface clock, and DPLL4, which can provide the DSS
functional clock.
Could you try:
1. applying something like the patch at the bottom of this message and
seeing if it makes any difference?
2. if #1 does not work, changing the "dpll3" in the patch to "dpll4" ?
3. if #2 does not work, disabling autoidle on both dpll3 and dpll4?
Disabling DPLL3 autoidle fixes the problem. Disabling DPLL4 autoidle
doesn't affect the problem.
The issues your are facing seems to be the well known DSS low power
refresh mode we've been trying to use since OMAP2 :-).
DPLL3 = CORE DPLL, whereas DPLL4 = PER DPLL, so that's normal that only
the DPLL3 change is affecting the issue.
But... Isn't the iface clock only needed to program DSS? When the DSS is
just running independently, plain func clock should be enough, right? Or
is iface clock used for the communication between DSS and
CORE/DPLL/wakeup/something?
No, the DSS does have two ports, the slave one for the register access
and a master one for the DMA inside the DSS to fetch the data from the
memory.
So basically, what is happening is that as soon as you allow MSTANDBY to
happen, the DSS will allow the iclk to be gated, and assuming the
interconnect is not used by any other initiator (like USB), the CORE
DPLL (DPLL3) can be gated.
The issue is that when the DSS will reach the low threshold, it will
de-assert the MSTANDBY to be able to get some fresh data. But since the
DPLL3 was in idle, it will take several tens of us before the clock is
re-enabled and thus before the interconnect can be used by the DSS to
get the data. Meanwhile the DSS FIFO is still getting low.
If the latency to re-enable the DPLL + interconnect is above the
duration that the FIFO DSS can support, you do have some FIFO underflow.
I also suspect that this could be just a plain DSS bug. The default FIFO
low/high thresholds are 960/1023 bytes (i.e. DSS starts refilling the
FIFO when there are 960 or less bytes in the fifo, and stops at 1023.
The fifo is 1024 bytes). The values are calculated with fifo_size -
burst_size and fifo_size - 1.
We are now using FIFO merge features, which combines multiple fifos into
one when possible, making the fifo size 1024*3 = 3072. Using the same
low threshold and increasing the high threshold to 960/3071 works fine.
Changing the high threshold to 3008 causes underflows. Increasing the
low threshold to ~1600 makes DSS work again.
That's weird, in theory what should matter is only the diff between the
high and low. Well the low value should be as high as possible as well
to support the wakeup latency.
So I think that the high thresholds of 3071 and 3008 are so close to
each other that there shouldn't be any real difference in practice,
presuming everything works. But, for whatever reason, fetching of the
pixels becomes much more inefficient or with much higher start latency,
causing the underflows.
I guess we'd need HW people to debug this, but their interest in OMAP3
is probably long gone. So I think I'll just have to use values that seem
to work for the use cases we test.
Anyway, in this case, a PM QoS constraint should be set on the
interconnect to ensure that potentially the DPLL3 will not be autogated
as soon as the iclk is gated.
The point is that whereas you set the FIFO to support the wakeup latency
or you prevent the idle mode because you conside the DSS cannot support
that latency because FIFO cannot be merged.
Except that I'm not sure the PM QoS OMAP is in mainline yet :-(
Regards,
Benoit
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
