On Fri, Jun 03, 2022 at 10:29:58PM +0200, Michael Grzeschik wrote: > On Fri, Jun 03, 2022 at 03:37:44PM -0400, Alan Stern wrote: > > On Fri, Jun 03, 2022 at 07:22:09PM +0200, Michael Grzeschik wrote: > > > Also, I just found out that just parsing power_bits is not enough. > > > > > > E.g. when we use other tools to set clear PORT_POWER on the hub like > > > uhubctl to disable a port. The value does not represent the real state > > > of the port. > > > > The value in power_bits is always supposed to match the real state of > > the port. How does uhubctl manage to get them to disagree? > > https://github.com/mvp/uhubctl/blob/master/uhubctl.c#L1082 I see. It relies on the fact that the hub-specific requests have their recipient field set to Device or Other, not Interface (which in my opinion it should be). This means we can't intercept these requests easily. > It just calls a direct control transfer with rather CLEAR or SET_FEATURE > set. So this will be transfered completely passing the kernel usb core > layer. I actually would expect that the hubs interrupt worker would > trigger. I will have to check if this is the case. Regardless, the hub driver won't be aware that the port's power state has changed, so it won't update the power_bits value. > > > I think it is better to use hub_port_status and port_is_power_on from > > > hub.c to test the real state by sending a GET_STATUS command. > > > > Maybe. But if power_bits is working properly, this should not be > > needed. It would be better to fix the value stored in power_bits. > > I don't know if this is trivial. If it is not, I would prefer to > trigger the GET_STATUS in disable_show for now. Yes, I think there is no choice. Alan Stern
