diff options
author | David S. Miller <davem@ultra5.davemloft.net> | 2007-03-04 12:53:19 -0800 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2007-04-26 01:55:10 -0700 |
commit | 01f94c4a6ced476ce69b895426fc29bfc48c69bd (patch) | |
tree | 68866108644a1e90ae7688d2e35cb8462ab6a07c /arch/sparc64/kernel/of_device.c | |
parent | a378fd0ee8ea6af5dafd0ab3d634f22b926b5ac4 (diff) | |
download | blackbird-op-linux-01f94c4a6ced476ce69b895426fc29bfc48c69bd.tar.gz blackbird-op-linux-01f94c4a6ced476ce69b895426fc29bfc48c69bd.zip |
[SPARC64]: Fix sabre pci controllers with new probing scheme.
The SIMBA APB bridge is strange, it is a PCI bridge but it lacks
some standard OF properties, in particular it lacks a 'ranges'
property.
What you have to do is read the IO and MEM range registers in
the APB bridge to determine the ranges handled by each bridge.
So fill in the bus resources by doing that.
Since we now handle this quirk in the generic PCI and OF device
probing layers, we can flat out eliminate all of that code from
the sabre pci controller driver.
In fact we can thus eliminate completely another quirk of the sabre
driver. It tried to make the two APB bridges look like PBMs but that
makes zero sense now (and it's questionable whether it ever made sense).
So now just use pbm_A and probe the whole PCI hierarchy using that as
the root.
This simplification allows many future cleanups to occur.
Also, I've found yet another quirk that needs to be worked around
while testing this. You can't use the 'class-code' OF firmware
property, especially for IDE controllers. We have to read the value
out of PCI config space or else we'll see the value the device was
showing before it was programmed into native mode.
I'm starting to think it might be wise to just read all of the values
out of PCI config space instead of using the OF properties. :-/
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc64/kernel/of_device.c')
-rw-r--r-- | arch/sparc64/kernel/of_device.c | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/arch/sparc64/kernel/of_device.c b/arch/sparc64/kernel/of_device.c index 6b8798141f2c..8964eacb0048 100644 --- a/arch/sparc64/kernel/of_device.c +++ b/arch/sparc64/kernel/of_device.c @@ -317,6 +317,11 @@ static unsigned int of_bus_default_get_flags(const u32 *addr) static int of_bus_pci_match(struct device_node *np) { if (!strcmp(np->type, "pci") || !strcmp(np->type, "pciex")) { + char *model = of_get_property(np, "model", NULL); + + if (model && !strcmp(model, "SUNW,simba")) + return 0; + /* Do not do PCI specific frobbing if the * PCI bridge lacks a ranges property. We * want to pass it through up to the next @@ -332,6 +337,21 @@ static int of_bus_pci_match(struct device_node *np) return 0; } +static int of_bus_simba_match(struct device_node *np) +{ + char *model = of_get_property(np, "model", NULL); + + if (model && !strcmp(model, "SUNW,simba")) + return 1; + return 0; +} + +static int of_bus_simba_map(u32 *addr, const u32 *range, + int na, int ns, int pna) +{ + return 0; +} + static void of_bus_pci_count_cells(struct device_node *np, int *addrc, int *sizec) { @@ -436,6 +456,15 @@ static struct of_bus of_busses[] = { .map = of_bus_pci_map, .get_flags = of_bus_pci_get_flags, }, + /* SIMBA */ + { + .name = "simba", + .addr_prop_name = "assigned-addresses", + .match = of_bus_simba_match, + .count_cells = of_bus_pci_count_cells, + .map = of_bus_simba_map, + .get_flags = of_bus_pci_get_flags, + }, /* SBUS */ { .name = "sbus", |