Am Sonntag, den 21.02.2021, 11:20 +0100 schrieb Hans de Goede: > Hi, Hi, > > On 2/21/21 9:51 AM, Hui Wang wrote: > > Once pre_reset() or post_reset() returns non-zero, the disconnect() > > and probe() of the usb_driver will be called. In the disconnect(), > > the scsi_host will be removed and be freed after scsi_host_put(), in > > the probe(), the new scsi_host and uas_dev_info will be created. > > > > If the usb_reset_device() is triggered by eh_device_reset_handler(), > > and pre_reset()/post_reset() returns non-zero, the disconnect() and > > probe() will be called, then returns to the eh_device_reset_handler(), > > it still accesses old scsi related variables and uas_dev_info, and so > > do its caller functions. > > > > Here change the pre_reset() and post_reset() to let them only return > > 0, after this change, the usb_reset_device() will only reset this > > usb devcie from its hub port, will not execute unbind and rebind > > usb_driver during reset. > > We only return non 0 from the pre/post reset handles if we failed > to ensure the device is in a known state. correct. Technically it is a bit unfortunate that UAS devices react a bit different to other SCSI devices, but we definitely cannot hide a failure. Arguably we should go into OFFLINE state. But that needs a good reason beyond theoretical considerations. Regards Oliver
