Paul,

- Added Rafael for the PM QoS discussion -

On Tue, Oct 30, 2012 at 10:04 AM, Paul Walmsley [off-list ref] wrote:

On Tue, 30 Oct 2012, Paul Walmsley wrote:

quoted

Based on a very quick look, I'd say the original patch 3db11fe is broken.
I don't see how it can ensure that its PM_QOS_CPU_DMA_LATENCY request is
honored when CONFIG_CPU_IDLE=n. CONFIG_CPU_IDLE=n is the default for
omap2plus_defconfig.

So in fact to follow up on this, looks like one of two changes are needed:

1. Revert 3db11fe

or

2. If CONFIG_CPU_IDLE=n, the driver needs to call disable_hlt() in place
of the pm_qos constraint add, and call enable_hlt() in place of the pm_qos
constraint remove.

I do not think this is correct. Using disable_hlt is a too radical
solution and will prevent the idle completely, this is not what we
want.

Rafael, what do you think?

Furthermore without the patch 3db11fe enable_hlt and disable_hlt are
not used in the driver so this change is not the real fix for the
issue. To me the cause is somewhere else. I was hoping Felipe's
ordering fix would do it...

Jean, could you please pick a solution and send a patch for the 3.7-rc
timeframe, to fix the bug in 3db11fe?
?

Let me try to reproduce the issue and hopefully investigate a bit more.
Feel free to revert the patch if you feel like it is the thing to do.

- Paul

Jean