On Thu, 2 Aug 2012, Alan Stern wrote: > On Thu, 2 Aug 2012, B, Ravi wrote: > > > Alan you may be correct. Let me explain, the issue occurs while > > unplugging the device during Tx transfer, > > 1) During Tx transfer, TX-DMA is programmed to transfer the data. > > 2) The TX-DMA completes the transfer and generates completion interrupt. > > 3) On tx completion interrupt context the URB is given back. > > In normal scenario, the sequence 1 to 3 occurs during tx transfers. > > Issue occurs when the device is unplugged while TX-DMA transfer is in > > progress. The DMA halts and never generates completion interrupt. In > > What host controller driver are you using? > > Why doesn't the controller hardware detect that the device fails to > send handshake packets and abort the transfer? According to the USB-2 > spec, after three failures the controller must end the transfer. This > sounds like a bug in the controller or its driver. > > > this scenario, this active request at DMA is given back when > > application calls dequeue or unlink the URB. Since in this scenario > > since there is no timeout occurs on the URB, we expect the scsi layer > > should unlink the active URB upon DISCONNECT of device. > > Not until the command times out. > > > In this scenario, when the DISCONNECT occur, scsi does not unlink the > > active URB, this patch fixing this issue. > > Can you point me which function unlink the active URB (which is not > > returned). > > Let me know if you have better solution or any suggestion. > > I would expect the command to time out and be unlinked by > command_abort() after 30 seconds or so, and I would expect > scsi_remove_host() to wait until this happens. If this doesn't happen, > it indicates a bug in the SCSI or block layers. I tried doing a similar experiment on my system. I got rid of the code that makes ehci-hcd stop retrying after a maximum number of transaction errors (so that it would continue to retry indefinitely), and then pulled out my USB flash drive while a program was reading from it. The result was exactly as predicted: After 30 seconds the current transfer was aborted, and scsi_remove_host() waited for this to happen. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
