On Fri, 2022-03-04 at 15:11 +0100, Andrew Lunn wrote: > On Fri, Mar 04, 2022 at 10:58:47AM -0300, Fabio Estevam wrote: > > Hi Martyn, > > > > On Thu, Mar 3, 2022 at 10:14 AM Fabio Estevam <festevam@xxxxxxxxx> > > wrote: > > > > > > Hi, > > > > > > On a imx8mm-iotgateway board running kernel 5.17-rc6 I am getting > > > the following > > > smsc95xx errors after a 'reboot' command: > > > > > > ci_hdrc ci_hdrc.1: remove, state 1 > > > usb usb2: USB disconnect, device number 1 > > > usb 2-1: USB disconnect, device number 2 > > > usb 2-1.1: USB disconnect, device number 3 > > > smsc95xx 2-1.1:1.0 eth1: unregister 'smsc95xx' usb-ci_hdrc.1-1.1, > > > smsc95xx USB 2.0 Ethernet > > > smsc95xx 2-1.1:1.0 eth1: Failed to read reg index 0x00000114: -19 > > > smsc95xx 2-1.1:1.0 eth1: Error reading MII_ACCESS > > > smsc95xx 2-1.1:1.0 eth1: __smsc95xx_mdio_read: MII is busy > > > smsc95xx 2-1.1:1.0 eth1: Failed to read reg index 0x00000114: -19 > > > smsc95xx 2-1.1:1.0 eth1: Error reading MII_ACCESS > > > smsc95xx 2-1.1:1.0 eth1: __smsc95xx_mdio_read: MII is busy > > > smsc95xx 2-1.1:1.0 eth1: hardware isn't capable of remote wakeup > > > > The same error has been previously reported by Ferry Toth on Intel > > Edison-Arduino: > > https://lore.kernel.org/netdev/5cd6fc87-0f8d-0b9b-42be-8180540a94e7@xxxxxxxxx/ > > > > It happens after commit a049a30fc27c ("net: usb: Correct PHY > > handling > > of smsc95xx") > > > > A simple way to reproduce it: > > > > echo -n "2-1" > /sys/bus/usb/drivers/usb/unbind > > > > It is not clear to me why register 0x114 (MII_ADDR) is accessed > > after > > smsc95xx unregistration. > > I suggest you look at the call sequences. When does phy_stop() and > phy_disconnect() get called? phylib has a work queue which runs once > per second to poll the PHY and get its status. That polling is > stopped > by calling phy_stop(). It is also assumed you can still talk to the > PHY while performing phy_stop(). > > I was just trying to remember the details from when I looked at this driver. IIRC, that's broadly the conclusion I came to and couldn't really see a way around it. Access to the phy was being lost before phy_stop() was called. Martyn > Andrew
