Hi Alan, Rafael, Greg, long story short: I want to disable a usb-port completely during suspend because it isn't needed and we need to save energy, because is a 32bit ARM (OF-based) handheld device. I use the port to connect a usb-ethernet dongle (all needed drivers are builtin no modules) which is needed for the NFS. The usb-ethernet dongle supports the persist setting because it does a hw-reset during resume anyway. So what I did is: 1) Set the persist bit for the usb device 2) Set the control to auto for the usb device 3) Unset the pm_qos_no_power_off flag for the usb-port But the port gets not disabled. I debugged it and found a problem in usb_port_suspend() logic [1] and the generic PM-framework more precisely the dpm mechanism. The usbcore does the correct pm_runtime counting but the call [2] don't trigger the usb_port_runtime_suspend() [3] because the dpm enables all runtime-pm device before the shutdown is executed. IMHO both subsystem behaviours are correct and I don't know the _correct_ fix, therefore I wrote this email. As far as I understood it all non-ACPI platforms are affected. Regards, Marco [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/usb/core/hub.c?h=v5.5#n3238 [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/usb/core/hub.c?h=v5.5#n3328 [3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/usb/core/port.c?h=v5.5#n247
