Add Jani from Gfx driver team and audio team.
Hi Takashi,
-----Original Message----- From: Takashi Iwai [mailto:tiwai@suse.de] Sent: Friday, June 27, 2014 6:28 PM
At Fri, 27 Jun 2014 06:38:47 +0000, Lin, Mengdong wrote:
Hi Takashi,
Sorry. This patch cannot handle the following case: If only embedded DP is connected on boot, the gfx driver will turn off
the display power well on boot case before the audio driver probes and request this power.
So the M/V values programmed by BIOS in EM4/5 registers will be lost at
the very beginning. And so the audio driver will save invalid M/N values on suspend and restore these invalid values on resume.
One phenomenon is: if HDMI or DP monitor is connected after boot,
audio playback rate is ~20% faster than normal.
And I guess it's the patch cannot fix the Haswell issue https://bugzilla.kernel.org/show_bug.cgi?id=74861
There are two possible solutions: (1) Follows Windows model: Windows, gfx driver notifies BIOS on LPSP
(when only eDP is connected) exit and BIOS reprograms those EM4/5 registers.
There is no such implementation in Linux gfx driver now. We need
time to check the effort.
(2) Gfx driver notify audio driver on exiting LPSP and CDCLK value (by
reading GPU registers), so audio driver can reprograms EM4/5 registers.
The open source interface between audio/gfx driver is still not ready.
Would an private API to get CDCLK from gfx driver will be a possible workaround atm?
Could you share your opinion on this issue?
To me, the former sounds like an easier solution. In the latter case, it can be still an issue of module loading order.
Can we introduce a private i915 interface for the audio driver to get the CDCLK as a quick fix? Then the audio driver can reprogram the EM4/5 M/N values as per the CDCLK.
The interface can be like 'int i915_get_cdclk_freq(void)' and the audio driver will use symbol_request() to get this private API and avoid the issue of module loading order. We can put the relevant code in sound/pci/hda/ hda_i915.c
For Broadwell Jani has implemented a notification to BIOS, so BIOS can reprogram the EM4/5 value after restoring the power. http://patchwork.freedesktop.org/patch/28866/ http://patchwork.freedesktop.org/patch/28868/
But for Haswell, BIOS does not support the above notification and provide another heavy-weight notification which will enable/disable the HD-A controller at runtime and so need OS to re-enumerate the device stack. That causes system hang. And test on Haswell shows that reprogramming the EM4/5 as per the CDCLK is also enough to give good BCLK and audio rate. So we hope to avoid further dependency on the black-box BIOS.
Libin is working on the generic interface between audio and gfx now and we can clean up code in the future. But before the interface is ready, a private API to query CDCLK seems to be a quick and reliable fix for Haswell and Broadwell.
Thanks Mengdong