On 02/18/2013 11:39 PM, Ville Syrjälä wrote:
On Mon, Feb 18, 2013 at 02:50:49PM +0100, David Henningsson wrote:
On 02/18/2013 02:35 PM, Takashi Iwai wrote:
At Mon, 18 Feb 2013 14:11:10 +0100, David Henningsson wrote:
I'm experimenting with using ELD information in PulseAudio and found that it didn't work very well with hotplugging. Neither the proc file nor the ELD kcontrol was cleared correctly after unplug, and there was no way (except polling) to find out when ELD has become available.
Did you try Xingchao's patch for drm/i915? It's queued in drm-intel-next:
commit 7b9f35a6dd72f89452c58bbdbaf063027bf857ec Author: Wang Xingchao <xingchao.wang@intel.com> Date: Tue Jan 22 23:25:25 2013 +0800 drm/i915: HDMI/DP - ELD info refresh support for HaswellThanks for the hint. I'm currently developing not on a Haswell machine, but an (I think) Arrandale that I bought a few years ago. So I guess this patch does not have any effect on my hw?
While reviewing that patch I asked whether we need something like that on other platforms. Unfortunately I didn't get a real answer. The logic in the patch would seem to apply equally to all platforms, not just HSW. We also seem to be doing silly things like flipping the ELD valid bit in the register on and off a few times when we update the ELD. But I'm not all that familiar with the topic so take my ramblings with a pinch of salt.
But in any case, if you're seeing a problem with the behaviour of i915 you should definitely report it.
Looking at the HDA spec [1], e g "Figure 72: PD and ELDV unsolicited responses flow for digital display codecs" it looks like it's not valid to first send the PD bit and then ELDV a while later. So then I guess it's an issue that this actually seem to happen on the machine I use for testing this code.
From the same figure, it also looks like ELDV can be flipped back and forth (due to mode changes) when PD is true; so making patches that deal with PD and ELDV independently seem to be the right thing nonetheless.