Errata 2 for the isp1760 explains that the chip sometimes does not issue
interrupts when an ATL (bulk or control) transfer is completed. There are
several issues with the current work-around (SOF interrupts) for this:

1) It seems the chip sometimes does not even set the done bit for a
   completed transfer, in which case SOF interrupts does not solve
   the problem since we still check the done map to find out which
   transfer descriptors to handle.

2) The above point seems to happen only when ATL and SOF interrupts
   are enabled at the same time. However, disabling ATL interrupts
   increases the latency between transfer completion and handling.
   This is very noticeable in the testusb suite, which take several
   minutes more to run with ATL interrupts disabled.

This patch removes the code to switch on SOF interrupts, and instead
use a kernel timer to periodically check for "old" descriptors that
have their VALID and ACTIVE flags unset, indicating completion, thus
avoiding the dependency on the chip's done map (and SOF interrupts)
to find transfers affected by this HW bug.

[bigeasy@linutronix: 80 lines limit]
Signed-off-by: NArvid Brodin <arvid.brodin@enea.com>
Signed-off-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: NGreg Kroah-Hartman <gregkh@suse.de>