On 2015/10/1 1:36, Borislav Petkov wrote:
On Thu, Oct 01, 2015 at 01:00:44AM +0800, Jiang Liu wrote:
Thanks Joerg, that makes sense. If some driver tries to binding to the IOMMU device, it will trigger the scenario as you described. For example, Xen backend driver will try to probe all PCI devices if enabled. I will do more investigation tomorrow.
Right, so this fixes the issue on my box, courtesy of Joerg. WE basically don't disable the IRQ on MSI-enabled devices. The AMD IOMMU uses a barebones PCI device but not a PCI driver, which would be an overkill.
diff --git a/arch/x86/pci/common.c b/arch/x86/pci/common.c index 09d3afc..29ec2eb 100644 --- a/arch/x86/pci/common.c +++ b/arch/x86/pci/common.c @@ -674,12 +674,15 @@ int pcibios_add_device(struct pci_dev *dev)
int pcibios_alloc_irq(struct pci_dev *dev) {
- if (pci_dev_msi_enabled(dev))
return 0;
We may return -EBUSY here to reject the probe operation. It doesn't make sense to continue the probe if MSI is already enabled, tt also helps to avoid calling pcibios_free_irq() in function pci_device_probe().
- return pcibios_enable_irq(dev);
}
void pcibios_free_irq(struct pci_dev *dev) {
- if (pcibios_disable_irq)
- if (!pci_dev_msi_enabled(dev) && pcibios_disable_irq)
The above change is not needed, pcibios_disable_irq() will first check !pci_has_managed_irq(dev) before actually freeing PCI irq. pci_has_managed_irq(dev) only returns true if pcibios_alloc_irq() succeeds.
So to summary, I think we only need following change to fix the regression: int pcibios_alloc_irq(struct pci_dev *dev) { + if (pci_dev_msi_enabled(dev)) + return -EBUSY;
What do you think? Thanks! Gerry
pcibios_disable_irq(dev);}