On Wed, 21 Jun 2017 19:58:24 +0200, bgoswami@codeaurora.org wrote:
Hi,
Current ALSA core implementation has PM QoS vote/unvote calls in snd_pcm_hw_params() and snd_pcm_hw_free().
However, this might have an implication in terms of power, as PM QoS vote would still be valid for that substream, even when the substream is paused. Especially for those cases, where, to avoid missing the QoS, one or more CPU cores might be prevented from entering into lower power mode (or power-collapse), as it might take longer to bring those cores out of power collapse. This might be specially more prominent when the period size is smaller.
So, in an effort to optimize power (by whatever smaller amount), I was wondering, if it is better to remove the PM QoS vote (pm_qos_remove_request) when Pause (SNDRV_PCM_TRIGGER_PAUSE_PUSH) is called from user space (using IOCTL) for a particular substream. Similarly, PM QoS vote can be added (pm_qos_add_request) back, when user space (using IOCTL) calls Resume (SNDRV_PCM_TRIGGER_PAUSE_RELEASE) for that same substream.
It's a good question. The current calls in hw_params and hw_free with a naive assumption about the usage pattern.
I am looking for opinions from subject matter experts, to see if the proposed approach might have any unintended side-effects or any known limitation that I might be missing.
I don't see a bit side effect, but I'd like to look at the real gain by such a chain. Could you test and measure? If we can see a significant difference, we should try harder to narrow the window, indeed.
thanks,
Takashi