On Mon, Jun 24, 2013 at 8:08 PM, Oliver Neukum [off-list ref] wrote:

On Monday 24 June 2013 19:16:43 Ming Lei wrote:

quoted

On Mon, Jun 24, 2013 at 6:28 PM, Oliver Neukum [off-list ref] wrote:

quoted

On Monday 24 June 2013 17:42:04 Ming Lei wrote:

quoted

This patch improves this above situation, and the qh will wait for 5
milliseconds before being unlinked from hardware, if one URB is submitted
during the period, the qh is move out of unlink wait list and the
interrupt transfer can be scheduled immediately, not like before: the
transfer is only linked to hardware until previous unlink is completed.

It seems to me that this logic should be used only if the URB completed
without error.

The current completed URB with error  doesn't mean the later submitted
URB will complete with error, so I don't think the logic is related with URB
completion error.

But

a) it is likelier
b) the driver will start error handling

And what about the bandwidth? At least if you unlink the URB, its bandwidth
should be given back immediately.

Yes, the patch doesn't affect unlinking intr URB because the
behavior of start_unlink_intr() isn't changed basically.

Thanks,
--
Ming Lei