[PATCH 14/16] Resource assignments for PCI devices on Celleb

[PATCH 14/16] Resource assignments for PCI devices on Celleb

[Ishizaki Kou]

Celleb has multiple PCI buses. Devices on some of those PCI buses require to
assign resources by Linux. 
This patch is for such buses not to claim resources, because claimed resource
does not become a target for resource assign.

Signed-off-by: Kou Ishizaki <kou.ishizaki@toshiba.co.jp>
---

Index: linux-2.6.19/arch/powerpc/kernel/pci_64.c
diff -u linux-2.6.19/arch/powerpc/kernel/pci_64.c:1.1.1.2 linux-2.6.19/arch/powerpc/kernel/pci_64.c:1.3
--- linux-2.6.19/arch/powerpc/kernel/pci_64.c:1.1.1.2	Tue Oct 24 13:35:53 2006
+++ linux-2.6.19/arch/powerpc/kernel/pci_64.c	Mon Nov  6 22:04:23 2006
@@ -247,8 +247,16 @@
 	if (firmware_has_feature(FW_FEATURE_ISERIES))
 		return;
 
+#ifdef CONFIG_PPC_MULTIPLATFORM
+	list_for_each_entry(b, &pci_root_buses, node) {
+		if (!machine_is(celleb) ||
+		    ppc_md.pci_probe_mode(b) == PCI_PROBE_DEVTREE)
+			pcibios_claim_one_bus(b);
+	}
+#else
 	list_for_each_entry(b, &pci_root_buses, node)
 		pcibios_claim_one_bus(b);
+#endif
 }
 
 #ifdef CONFIG_PPC_MULTIPLATFORM

Re: [PATCH 14/16] Resource assignments for PCI devices on Celleb

[Benjamin Herrenschmidt]

On Wed, 2006-11-15 at 18:58 +0900, Ishizaki Kou wrote:
> Celleb has multiple PCI buses. Devices on some of those PCI buses require to
> assign resources by Linux. 
> This patch is for such buses not to claim resources, because claimed resource
> does not become a target for resource assign.

I do not fully understand why you need that. You should be able to let
the PCI code claim all resources wether they are on real PCI or not.

Ben.

Re: [PATCH 14/16] Resource assignments for PCI devices on Celleb

[Ishizaki Kou]

Hi Ben-san,

>On Wed, 2006-11-15 at 18:58 +0900, Ishizaki Kou wrote:
>> Celleb has multiple PCI buses. Devices on some of those PCI buses require to
>> assign resources by Linux. 
>> This patch is for such buses not to claim resources, because claimed resource
>> does not become a target for resource assign.
>
>I do not fully understand why you need that. You should be able to let
>the PCI code claim all resources wether they are on real PCI or not.


This is a temporary workaround for following problem.
   
On Celleb, values of PCI base addresses of PCI devices which are not
described in device tree are not initialized before linux starts. So
we use pci_bus_assign_resources() later to assign resources for them 
at ppc_md.pcibios_fixup(). But by original code, when
ppc_md.pcibios_fixup() is called, resources are already claimed and
pci_bus_assign_resource() does not work to assign resources.

It may not be a good idea to call pci_bus_assign_resources() at the
timing of ppc_md.pcibios_fixup(), but we did not find an alternative
idea.


Thank you,
Kou Ishizaki
Toshiba

Re: [PATCH 14/16] Resource assignments for PCI devices on Celleb

[Benjamin Herrenschmidt]

\
> This is a temporary workaround for following problem.
>    
> On Celleb, values of PCI base addresses of PCI devices which are not
> described in device tree are not initialized before linux starts. So
> we use pci_bus_assign_resources() later to assign resources for them 
> at ppc_md.pcibios_fixup(). But by original code, when
> ppc_md.pcibios_fixup() is called, resources are already claimed and
> pci_bus_assign_resource() does not work to assign resources.
> 
> It may not be a good idea to call pci_bus_assign_resources() at the
> timing of ppc_md.pcibios_fixup(), but we did not find an alternative
> idea.

Hrm.. not everything should be claimed... there should be some space
left to assign remaining devices. I need to look into it. There are
several non resolved issues unfortunately with the ppc64 PCI code vs.
assigning resources. (Well, the issues are with the generic code in fact
which is what ppc64 uses. It tends to not work very well on setups where
some resources have already been assigned and some not).

Ben.