On Wed, Dec 13, 2017 at 12:32:47PM +0100, Ladislav Michl wrote: > On Wed, Dec 13, 2017 at 12:16:05PM +0100, Johan Hovold wrote: > > On Mon, Dec 11, 2017 at 01:44:57PM +0100, Ladislav Michl wrote: > > > On Mon, Dec 11, 2017 at 12:52:46PM +0100, Johan Hovold wrote: > > > > On Mon, Dec 11, 2017 at 12:32:49PM +0100, Ladislav Michl wrote: > > > > > On Mon, Dec 11, 2017 at 11:10:35AM +0100, Johan Hovold wrote: > > > [snip] > > > > > > I'm afraid I don't consider this an improvement. I prefer using gotos > > > > > > for error paths, while keeping the success path out of the status > > > > > > switch. > > > > > > > > > > > > Furthermore, this isn't functionally equivalent as we'd not longer log > > > > > > an error for -EPIPE. > > > > > > > > > > Yes, you are right... Now, shouldn't we react somehow to stalled endpoint? > > > > > Tty side seems to be unaware of it. > > > > > > > > Recovering from a stalled endpoint is a bit involved, so for now we > > > > typically just log an error an bail out (forcing the user to reopen the > > > > port). This seems to work well enough as this condition should be rare. > > > > > > I just do not see this in code. I would expect pending tty I/O operation > > > would fail once USB device errors out with -EPIPE, so tty side consumer gets > > > notified about error. Either it is not there or I did not look hard enough :) > > > > No, we do not provide any error notification besides logging an error > > when an endpoint has been stalled (i.e. in case of a read, you would not > > receive any more data before the halt condition has been cleared). > > Hmm, could we make 1aba579f3cf5 a bit more generic then? I'm not sure this is worth the complexity it adds, or that it really solves anything as the fundamental problem remains; simply clearing the Halt is not sufficient unless the underlying reason for the condition triggering the stall has first been removed (and that is going to be device specific). > > Given that there's been no reports about this being an issue (for the > > past ten years and that I can recall), this crude handling appears to > > suffice. > > Because embedded people tend to power cycle stuck device on timeout. > That involves project specific hacks in userspace, which could be > probably handled better. I haven't seen any reports of this being an issue (and your stalled CDC device was really due to a hardware/power issue IIRC) so again, I'm not convinced this gives us anything. Johan -- 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
