Commit d006199e72a9 ("serial: sh-sci: Regtype probing doesn't need to be
fatal.") made sci_init_single() return when sci_probe_regmap() succeeds,
although it should return when sci_probe_regmap() fails.  This causes
systems using the serial sh-sci driver to crash during boot.

Fix the problem by using the right return condition.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Presently these were all using the same static string with no regard to
dev_name() and the like. This implements a bit of rework to name the IRQ
dynamically, as it should have been doing all along anyways.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

Ultimately we want everything to be going through the clock framework and
runtime pm, so kill off the per-port callbacks that enabled ports to
bypass the common infrastructure.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

This was using a BUG_ON(), but it's not strictly necessary, so relax the
constraints a bit.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

For all ports with a valid SCLSR register we can use the generic FIFO
overrun detection logic. Test the validity of the SCLSR register rather
than depending explicitly on port type, which can be ambiguous for the
SCIFA/B types.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

This consolidates all of the TX/RX fill/room nonsense in to a single set
of fairly heavyweight definitions. The implementation goes in descending
order of complexity, testing the register map for capabilities until we
run out of options and do it the legacy SCI way. Masks are derived
directly from the per-port FIFO size, meaning that platforms with FIFO
sizes not matching the standard port types will still need to manually
fix them up.

This also fixes up a number of issues such as tx_empty being completely
bogus for SCI and IrDA ports, some ports using masks smaller or greater
than their FIFO size, and so forth.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

This takes a bit of a sledgehammer to the horribly CPU subtype
ifdef-ridden header and abstracts all of the different register layouts
in to distinct types which in turn can be overriden on a per-port basis,
or permitted to default to the map matching the port type at probe time.

In the process this ultimately fixes up inumerable bugs with mismatches
on various CPU types (particularly the legacy ones that were obviously
broken years ago and no one noticed) and provides a more tightly coupled
and consolidated platform for extending and implementing generic
features.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

Non-SCI parts do not have the special port reg necessary for cases where
the RX and SCI pins are muxed and need to be manually polled, so these
like always fall back on the normal FIFO processing paths. SH7760 is in a
class in and of itself with regards to mapping its SIM card interface via
the SCI port class despite not having any of the RXD lines wired up and
so implicitly behaving more like a SCIF in this regard. Out of the other
CPUs, some support the port check via the same block while others do it
through an external SuperI/O, so it's not even possible to perform the
check relative to the ioremapped cookie offset, so the separate read
semantics are preserved here, too.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

This consolidates all of the broken out overrun handling and ensures that
we have sensible defaults per-port type, in addition to making sure that
overruns are flagged appropriately in the error mask for parts that
haven't explicitly disabled support for it.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>