On Dec 11, 2008, at 12:50 AM, Eygene Ryabinkin wrote:
> Marcel, Boris, good day.
> Wed, Dec 10, 2008 at 05:00:24PM -0800, Marcel Moolenaar wrote:
>>> Seems that just the same card should work:
>>> I've added some diagnistic. But 'rid' is not what you want, I guess:
>> The RID is fine. It should always be 0.
> Seems like a dumb question, but nevertheless.
> What I don't understand is the following: BAR to port mapping for
> the Timedia is tricky, it mixes BARs and offsets for the different
> ports (you should know this, since you wrote the support ;)).
Correct (on both accounts :-)
> But in uart_bus_probe you're passing rid = 0 and it is used for
> allocation and consequently the same rid is used for all ports (at
> I read the code in this way). But puc_get_bar() uses calculated rid
> values, but does essentially the same thing: resource allocation via
> bus_alloc_resource(). And sc->sc_bas is initialized from the obtained
> sc->sc_rres (inside uart_bus_probe) and it is subsequently used for
> ns8250_probe() that is failing.
> I see that uart_bus_pci.c calls uart_bus_probe() with the actual rid.
> It does not mean that puc code should do the same, but ...
It could have been done that way, but such is not necessary.
It would not have been a problem for uart to do it, as can
be seen from uart_bus_pci.c, but it would have introduced
some complexity for sio(4). We needed to support sio(4) at
that time and I didn't want to touch sio(4) at all. Since
puc(4) needs to maintain a mapping from the child's device_t
to some internal data structure, it was trivial to have the
child use RID 0 in all cases and have that mapped to the
right bus tag and handle pair...