Drop the redundant port open flag and corresponding checks. USB serial
core will not call any of these driver callbacks for a closed port, and the
write URBs are stopped at close().

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

Drop read-urb check which is always false from completion the callback.

The driver read-urb pointer is set at every open and is never cleared.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

Drop the custom port driver data accessors and paranoid sanity checks.
The driver data is not cleared until the driver is unbound.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

Drop the helper used to retrieve the serial struct pointer from the port
structure.

Note that this helper was only used when the serial structure was
actually not needed.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

Drop the likewise paranoid serial structure sanity checks.

USB serial core sets the serial pointer in the port structures at
initialisation and it is never cleared, and similar for the serial
structure type.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

Drop the paranoid port structure sanity checks which are confusing at
best.

The driver data port pointer is set at port probe and never cleared,
while USB serial core sets the tty driver data at install and won't call
any driver functions with a NULL port pointer.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

The bulk-in URB context is set to the driver-data struct at open and is
never cleared so drop the redundant check in the completion callback.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

The interrupt handling is completely broken and has in fact never been
been enabled due to an inverted test for an interrupt endpoint in
open() that prevented the interrupt URB from being submitted.

Other highlights include missing interrupt URB resubmission (had it
ever been submitted), missing locking when managing the open-port count,
and NULL-pointer dereferences that could have been triggered by a
malicious device.

Rip out this broken crap which is beyond repair instead of pretending
we support this feature.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

The driver would return success and leave the port structures
half-initialised if any of the register accesses during probe fails.

This would specifically leave the port control urb unallocated,
something which could trigger a NULL pointer dereference on interrupt
events.

Fortunately the interrupt implementation is completely broken and has
never even been enabled...

Note that the zero-length-enable register write used to set the zle-flag
for all ports is moved to attach.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>

Document the MCS7810 detection hack which relies on having the GPO and
GPI pins connected as recommended by ASIX.

Note that GPO (pin 42) is really RTS of the third port which will be
toggled for the corresponding physical port on two- and four-port
devices.

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Johan Hovold <johan@kernel.org>