Hi Zoie, On Tue, Mar 04, 2025 at 08:37:27AM +0000, Zoie Lin (林禹妡) wrote: > > ... > > On Thu, 2025-02-13 at 12:57 +0100, AngeloGioacchino Del Regno wrote: > > > > > > Turning on and off regulators at start of transfer and end of > > transfer respectively > > is very expensive and, while it makes sense for power efficiency of > > the controller, > > it doesn't make sense for: > > > > 1. Responsiveness (latency); and > > 2. Platform power efficiency at a whole. > > > > As a start, just set the autosuspend delay to 250ms; this gives you > > at least time > > to bring up clocks and regulators and usually finish a transfer at > > 400KHz, giving > > you the chance to also get some more requests before autosuspend > > decides to, well, > > auto..suspend the device. > > > > The right way of choosing an autosuspend delay a bit more precisely, > > though, for > > this device, would be to check the bus speed and calculate the > > autosuspend time > > accordingly. > > We understand the importance of balancing power efficiency and > responsiveness. Given the diversity of I2C devices connected > under each platform, it is indeed challenging to estimate an > appropriate delay time for testing. > > Therefore, we would like to proceed with your initial suggestion > and set the autosuspend delay to 250ms. Based on our preliminary > tests, this adjustment appears to be a practical solution across > our existing platforms. Am I understanding correctly that you are planning a v2? Andi
