On Tue, 24 Sep 2019, David Heinzelmann wrote: > > I really don't understand this. > > > > Your patch involves the case where there was a connect-change event but > > the port is still enabled. Now maybe I've forgotten about one of the > > pathways, but it seems like that combination should never occur. > > > > Certainly it shouldn't occur in your case. The device disconnects and > > then reconnects with a new set of descriptors. The disconnect should > > cause the port to be disabled, and the port should remain disabled > > after the reconnect occurs. So how can your new code run in the first > > place? > > > > Alan Stern > > > > Hi, > > I have a log with two devices which are connected to a hub and the hub is plugged in. > > The device which is not working in this log: > > Sep 24 08:32:21 kernel: usb 2-6-port1: status 0203 change 0011 > Sep 24 08:32:21 kernel: usb 2-6.1: new SuperSpeed Gen 1 USB device number 65 using xhci_hcd Ah, SuperSpeed. You're using a USB-3 device. That does make a difference. > Sep 24 08:32:21 kernel: usb 2-6.1: udev 65, busnum 2, minor = 192 > Sep 24 08:32:21 kernel: usb 2-6.1: New USB device found, idVendor=1409, idProduct=3240, bcdDevice= 0.00 > Sep 24 08:32:21 kernel: usb 2-6.1: New USB device strings: Mfr=1, Product=2, SerialNumber=0 > Sep 24 08:32:21 kernel: usb 2-6.1: Product: USB 3.0 Camera > Sep 24 08:32:21 kernel: usb 2-6.1: Manufacturer: Camera Manufacturer > > Now the firmware download happens and the device is re-enumerating and a disconnect/connect should occur. > But the only change which is seen is the following output: > > Sep 24 08:32:23 kernel: usb 2-6-port1: link state change > Sep 24 08:32:23 kernel: usb 2-6-port1: status 0203, change 0041, 5.0 Gb/s > > Now the resuscitation is happening but from my understanding this is not correct as in the reality there was a > reconnect from the device. So I tried to initiate a device reconnect if the device descriptor changed. > > It also seems to me that the enumeration from the second device (usb 2-6-port1) is blocking > the port change event and so the actual disconnect is missed. Now it all makes sense. Yes, I agree that your patch is the appropriate thing to do -- except that it contains at least one logic error: It doesn't handle the return code from usb_get_device_descriptor() properly. Also, I think you should expand the immediately preceding comment. Explain that it is indeed possible for the port to be enabled at this point, because USB-3 connections are initialized automatically by the host controller hardware when the connection is detected. Alan Stern
