On 3/31/20 7:12 AM, Hans de Goede wrote:
Hi,
On 3/31/20 12:54 PM, Pierre-Louis Bossart wrote:
Don't believe WoV for WPT has been added for SST linux so .ignore_suspend=1 achieves nothing. The offload part is probably just limited to bigger buffer size compared to system pin than anything else. So, nothing prevents from removing .ignore_suspend for SST solutions until WoV is verified. Don't know about SOF side. Pierre, what's your take on this?
I think on Baytrail and Broadwell we will most likely never enable hotwording on the Intel DSP, and S0i1-Playback isn't planned either.
However hotwording is enabled on the RT5677 and there may be clocking dependencies so that the RT5677 remains operational - at least the mclk needs to be on, so the bdw-rt5677 case probably needs more work.
While I am at it, I think the notion of .ignore_suspend has an assumption baked in. It will work if the only suspend mode is S0i1. If the platform can choose between S0i1 and S3, then we would still want to suspend all paths in S3, which currently isn't really possible.
You can check if S0i1 will be used by using:
pm_suspend_default_s2idle()
If this returns true then S0i1 will be used, otherwise S3.
This is defined in kernel/power/suspend.c and always build, so it should be ok to also use this in non x86 specific code-paths (it will simply always return false there I believe).
Yes, but what I meant is that when the target is S3, we have no way of disabling those .ignore_suspend that were added for S0i1 usages.