On Tue, 2019-04-09 at 10:44 -0400, Alan Stern wrote: +AD4 On Mon, 8 Apr 2019, Martin K. Petersen wrote: +AD4 +AD4 +AD4 +AD4 +AD4 Alan, +AD4 +AD4 +AD4 +AD4 +AD4 So it looks as though the SCSI subsystem doesn't like to have a reset +AD4 +AD4 +AD4 handler call scsi+AF8-remove+AF8-host. +AD4 +AD4 +AD4 +AD4 Are you talking about a PCI device removal handler or a SCSI error +AD4 +AD4 handler? +AD4 +AD4 The context of this discussion is a USB mass-storage device where the +AD4 device's port on its upstream hub has been powered off. The +AD4 powered-off port causes an executing command to time out. As a result +AD4 the SCSI error handler runs and calls the USB reset routine, but the +AD4 reset fails because the kernel is unable to communicate with the device +AD4 through the powered-off port. This causes the USB reset routine to +AD4 unbind the device from its USB driver, which in turn calls +AD4 scsi+AF8-remove+AF8-host -- while the error handler is still running. >From which context does that unbind happen? From inside a SCSI EH callback or from the context of a workqueue? I think the former is not allowed but that the latter is allowed. The SRP initiator driver (ib+AF8-srp.c) follows the latter approach. See also srp+AF8-queue+AF8-remove+AF8-work(). Bart.
