Hi Adam, (CC'ing Sakari) On Wed, Nov 24, 2021 at 03:16:57PM -0600, Adam Ford wrote: > I am trying to use an OV5640 camera sensor that I've used on both an > i.MX6Q and an i.MX8M Mini (with good success) on an RZ/G2[MNH] board > connected to the 2-lane CSI interface. > > I can get the media-ctl to show the routings, and sometimes I can get > streaming. Often, I get a timeout: > > rcar-csi2 fea80000.csi2: Timeout waiting for LP-11 state > > Looking at the various mailing list e-mails for the LP-11, it's > unclear to me if the timeout is caused by the sensor not doing > something correctly or the CSI2 misbehaving. Before transitioning to the high speed (HS) mode, the D-PHY transmitter must drive the lane in the LP-11 state. This is the idle state of the lane when powered up and when not in the ultra low-power state (ULPS). The transition to HS mode on the receiver side involves observing the LP-11 state. Many D-PHY RX require configuring the PHY when the lane is in LP-11 state, and only then starting the sensor to transition to HS. This requires powering up the D-PHY TX and going to idle mode, which most sensors support. As we're deprecating the .s_power() subdev operation, however, we have no way to power up the sensor without starting it before the D-PHY RX gets configured. In some cases, the D-PHY RX can handle the power up sequence automatically. They can be fully configured (from a software point of view) while the lane is in the power down state LP-00, and they then handle the transition to the stop state LP-11 and to the HS mode automatically. This isn't true for all receivers, some need software configuration after the data lane reaches the LP-11 state and before it transitions to HS mode. According to the documentation, the R-Car CSI-2 receiver requires software intervention between LP-11 and HS mode at least on V3M and E3. There's also a software configuration step on H3, M3N, V3H and V3U, but there's a chance that one could possibly be bypassed. > I was hoping someone might have some suggestions of things I can try. I would first try to power up the sensor at probe time and keep it power forever, to see if it solves your issue. If it does, then introducing a CSI-2-specific subdev operation to power up the sensor (or officially de-deprecating .s_power() for this use case) could be an option to fix the issue properly. -- Regards, Laurent Pinchart
