powerpc/powermac: Add missing of_node_put

arch/powerpc/platforms/powermac/pci.c +1 -0

index bcf50d7..800fcce 100644
@@ -661,6 +661,7 @@ static void __init init_second_ohare(void)
 			pci_find_hose_for_OF_device(np);
 		if (!hose) {
 			printk(KERN_ERR "Can't find PCI hose for OHare2 !\n");
+			of_node_put(np);
 			return;
 		}
 		early_read_config_word(hose, bus, devfn, PCI_COMMAND, &cmd);

arch/powerpc/platforms/powermac/time.c +0 -0

index 59eb840..394593c 100644
@@ -265,12 +265,14 @@ int __init via_calibrate_decr(void)
 	struct resource rsrc;

I think this particular patch is also required against Linus' kernel (not just linux-next).

Hi Nicolas,

Quoted 3 lines · On Wed, 3 Dec 2008 10:19:01 +1100 Stephen Rothwell <[email protected]> wrote:

I think this particular patch is also required against Linus' kernel (not just linux-next).

In fact, I think all these patches can apply to Linus' current tree.

arch/powerpc/platforms/powermac/time.c +0 -0

@@ -265,12 +265,14 @@ int __init via_calibrate_decr(void)
 	struct resource rsrc;

This still misses a path - if that `return 0' is taken, we still leak the reference.

This is reason #345 why sprinkling return statements all over your code is bad.

I fixed it up thusly. Please check.

I'm really in two minds about applying any of the of_node_put patches that only affect powermacs. The reference counts only matter on platforms where we update the OF device tree at runtime, which is currently only IBM pSeries machines. Since we don't have any hotplug on powermacs, and never will have, the OF device tree is completely static and we don't actually need refcounts on the nodes at all, so who cares if they're a bit higher than they might be?

In particular, the VIA whose node we're looking for here is built-in on the motherboard, and there can never be more than one, and it can never be removed.

Hi Paul,

Quoted 9 lines · On Sun, 7 Dec 2008 13:31:00 +1100 Paul Mackerras <[email protected]> wrote:

I'm really in two minds about applying any of the of_node_put patches that only affect powermacs. The reference counts only matter on platforms where we update the OF device tree at runtime, which is currently only IBM pSeries machines. Since we don't have any hotplug on powermacs, and never will have, the OF device tree is completely static and we don't actually need refcounts on the nodes at all, so who cares if they're a bit higher than they might be?

Quoted 3 lines

In particular, the VIA whose node we're looking for here is built-in on the motherboard, and there can never be more than one, and it can never be removed.

I my mind it is about consistent use of the API and good examples for people to copy. Also, in about a year you will be presented with the same set of patches when a new pair of eyes looks at the same code and notices the discrepancy ...

Hi Paul,

Quoted 9 lines · On Sun, 7 Dec 2008 13:31:00 +1100 Paul Mackerras <[email protected]> wrote:

I'm really in two minds about applying any of the of_node_put patches that only affect powermacs. The reference counts only matter on platforms where we update the OF device tree at runtime, which is currently only IBM pSeries machines. Since we don't have any hotplug on powermacs, and never will have, the OF device tree is completely static and we don't actually need refcounts on the nodes at all, so who cares if they're a bit higher than they might be?

Quoted 3 lines

In particular, the VIA whose node we're looking for here is built-in on the motherboard, and there can never be more than one, and it can never be removed.

I my mind it is about consistent use of the API and good examples for people to copy. Also, in about a year you will be presented with the same set of patches when a new pair of eyes looks at the same code and notices the discrepancy ...