[RFC] powerpc: set_dma_ops for pci hotplug

[RFC] powerpc: set_dma_ops for pci hotplug

[Hiroo Matsumoto]

[-- Attachment #1: Type: text/plain, Size: 759 bytes --]

Hi


I'm trying to use PCI Express Hot Plug on powerpc platform.
But PCI driver returns error when hotplug.
Error log is as below.
http://www.spinics.net/lists/linux-pci/msg14534.html

Some of PCI driver needs dma_ops.
On x86 platform, dma_ops is getting from external variable.
On powerpc platform, dma_ops is getting from archdata.dma_ops in struct
device.
There is a problem that archdata.dma_ops is set only when boot with
pcibios_setup_bus_devices but not set when hotplug.
So when hotplug, PCI driver's probe will return error.

I add code of checking and setting dma_ops in pcibios_enable_device.
It is called from pci_enable_device_xxx in PCI driver's probe before
checking dma_ops.
And PCI driver works good when hotplug.


Regards.

Hiroo MATSUMOTO

[-- Attachment #2: ppc-set_dma_ops-for-pcihp.patch --]
[-- Type: text/plain, Size: 1562 bytes --]

Signed-off-by: Hiroo MATSUMOTO <matsumoto.hiroo@jp.fujitsu.com>
diff --git a/arch/powerpc/kernel/pci-common.c b/arch/powerpc/kernel/pci-common.c
index 32656f1..080ff1d 100644
--- a/arch/powerpc/kernel/pci-common.c
+++ b/arch/powerpc/kernel/pci-common.c
@@ -111,6 +111,17 @@ static resource_size_t pcibios_io_size(const struct pci_controller *hose)
 #endif
 }
 
+static inline void pcibios_set_dma_ops(struct pci_dev *dev)
+{
+	/* Hook up default DMA ops */
+	set_dma_ops(&dev->dev, pci_dma_ops);
+	set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
+
+	/* Additional platform DMA/iommu setup */
+	if (ppc_md.pci_dma_dev_setup)
+		ppc_md.pci_dma_dev_setup(dev);
+}
+
 int pcibios_vaddr_is_ioport(void __iomem *address)
 {
 	int ret = 0;
@@ -1102,13 +1113,7 @@ void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
 		 */
 		set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
 
-		/* Hook up default DMA ops */
-		set_dma_ops(&dev->dev, pci_dma_ops);
-		set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
-
-		/* Additional platform DMA/iommu setup */
-		if (ppc_md.pci_dma_dev_setup)
-			ppc_md.pci_dma_dev_setup(dev);
+		pcibios_set_dma_ops(dev);
 
 		/* Read default IRQs and fixup if necessary */
 		pci_read_irq_line(dev);
@@ -1547,6 +1552,10 @@ EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
 
 int pcibios_enable_device(struct pci_dev *dev, int mask)
 {
+	/* dma_ops is NULL from hotplug */
+	if (!get_dma_ops(&dev->dev))
+		pcibios_set_dma_ops(dev);
+
 	if (ppc_md.pcibios_enable_device_hook)
 		if (ppc_md.pcibios_enable_device_hook(dev))
 			return -EINVAL;

Re: [RFC] powerpc: set_dma_ops for pci hotplug

[Hiroo Matsumoto]

Hi


I'm so sorry, but there is a better way than this way with
pcibios_enable_device() for issue.
I will write another code soon, so please wait.


Regards.

Hiroo MATSUMOTO

> Hi
> 
> 
> I'm trying to use PCI Express Hot Plug on powerpc platform.
> But PCI driver returns error when hotplug.
> Error log is as below.
> http://www.spinics.net/lists/linux-pci/msg14534.html
> 
> Some of PCI driver needs dma_ops.
> On x86 platform, dma_ops is getting from external variable.
> On powerpc platform, dma_ops is getting from archdata.dma_ops in struct
> device.
> There is a problem that archdata.dma_ops is set only when boot with
> pcibios_setup_bus_devices but not set when hotplug.
> So when hotplug, PCI driver's probe will return error.
> 
> I add code of checking and setting dma_ops in pcibios_enable_device.
> It is called from pci_enable_device_xxx in PCI driver's probe before
> checking dma_ops.
> And PCI driver works good when hotplug.
> 
> 
> Regards.
> 
> Hiroo MATSUMOTO
>