Re: Cardbus bridge fixup for SPARCle
Julian Coleman <[email protected]>
| Newsgroups | gmane.os.netbsd.ports.sparc64 |
|---|---|
| Message-ID | <[email protected]> |
Hi,
> What's the best way to handle this? I assume that the bus-range property
> needs fixing up. It is currently:
>
> bus-range 00000000 000000ff ........ ........ ........
>
> Is the way to fix this to go through the OFW tree and set all the bus ranges
> correctly for all the PCI busses, which will allow the code in
> pccbb_attach_hook() to correctly set the secondary bus number?
The attached patch fixes up the bus-range correctly for psycho0. It starts
with a bus-range of 0 -> 0, checks for any PCI-PCI bridges and expands the
bus-range if any are found. On the SPARCle, there are no PCI-PCI bridges,
so we end up with a final range of 0 -> 0. With this patch, cardbus0 at
cbb0 attaches OK (*). Note, that the patch only alters the bus-range of
the root bridge and any bridges below that, but does not alter the businfo
register on any bridges.
Thanks,
J
(*) The slot doesn't power up any inserted cards, but I think that that is
caused by us not having a power-management driver for the Tadpole PMU.
--
NetBSD: simple; works; documented / Sailing at Newbiggin
http://www.netbsd.org/ / http://www.newbigginsailingclub.org/
psycho.c.diffs
(text/plain, 2.2 KB)
Index: psycho.c
===================================================================
RCS file: /cvsroot/src/sys/arch/sparc64/dev/psycho.c,v
retrieving revision 1.112
diff -u -r1.112 psycho.c
--- psycho.c 27 Jan 2012 18:53:03 -0000 1.112
+++ psycho.c 12 Oct 2012 17:05:34 -0000
@@ -108,6 +108,7 @@
static struct extent *psycho_alloc_extent(struct psycho_pbm *, int, int,
const char *);
static void psycho_get_bus_range(int, int *);
+static void psycho_fixup_bus_range(int, int *);
static void psycho_get_ranges(int, struct psycho_ranges **, int *);
static void psycho_set_intr(struct psycho_softc *, int, void *, uint64_t *,
uint64_t *);
@@ -468,6 +469,10 @@
pba.pba_bus = psycho_br[0];
pba.pba_bridgetag = NULL;
+ /* Fix up invalid 0x00-0xff bus-range, as found on SPARCle */
+ if (psycho_br[0] == 0 && psycho_br[1] == 0xff)
+ psycho_fixup_bus_range(sc->sc_node, psycho_br);
+
aprint_normal("bus range %u to %u", psycho_br[0], psycho_br[1]);
aprint_normal("; PCI bus %d", psycho_br[0]);
@@ -859,6 +864,44 @@
}
static void
+psycho_fixup_bus_range(int node0, int *brp0)
+{
+ int node;
+ int len, busrange[2], *brp;
+
+ DPRINTF(PDB_PROM,
+ ("psycho debug: fixing up `bus-range' for node %08x: %u - %u\n",
+ node0, brp0[0], brp0[1]));
+
+ /*
+ * Check all nodes under this one and increase the bus range to
+ * match. Recurse through PCI-PCI bridges. Cardbus bridges are
+ * fixed up in pccbb_attach_hook(). Assumes that "bus-range" for
+ * PCI-PCI bridges apart from this one is correct.
+ */
+ brp0[1] = brp0[0];
+ node = prom_firstchild(node0);
+ for (node = ((node)); node; node = prom_nextsibling(node)) {
+ len = 2;
+ brp = busrange;
+ if (prom_getprop(node, "bus-range", sizeof(*brp),
+ &len, &brp) != 0)
+ break;
+ if (len != 2)
+ break;
+ psycho_fixup_bus_range(node, busrange);
+ if (brp0[0] > busrange[0] && busrange[0] >= 0)
+ brp0[0] = busrange[0];
+ if (brp0[1] < busrange[1] && busrange[1] < 256)
+ brp0[1] = busrange[1];
+ }
+
+ DPRINTF(PDB_PROM,
+ ("psycho debug: fixed up `bus-range' for node %08x: %u - %u\n",
+ node0, brp[0], brp[1]));
+}
+
+static void
psycho_get_ranges(int node, struct psycho_ranges **rp, int *np)
{