Re: [Xen-devel] [RFC 6/6] acpi:arm64: Add support for parsing IORT table

[Robin Murphy <robin.murphy@xxxxxxx>]

On 12/06/17 14:44, Jan Beulich wrote:
>>>> On 12.06.17 at 15:36, <julien.grall@xxxxxxx> wrote:
>> On 09/06/17 12:15, Robin Murphy wrote:
>>> On 08/06/17 20:30, Sameer Goel wrote:
>>> [...]
>>>>  /**
>>>> - * iort_iommu_configure - Set-up IOMMU configuration for a device.
>>>> + * iort_iommu_configure - Set-up IOMMU configuration for a device. This
>>>> + * function sets up the fwspec as needed for a given device. Only PCI
>>>> + * devices are supported for now.
>>>>   *
>>>>   * @dev: device to configure
>>>>   *
>>>> - * Returns: iommu_ops pointer on configuration success
>>>> - *          NULL on configuration failure
>>>> + * Returns: Appropriate acpi_status
>>>>   */
>>>> -const struct iommu_ops *iort_iommu_configure(struct device *dev)
>>>> +acpi_status iort_iommu_configure(struct device *dev)
>>>>  {
>>>>    struct acpi_iort_node *node, *parent;
>>>> -  const struct iommu_ops *ops = NULL;
>>>>    u32 streamid = 0;
>>>> +  acpi_status status = AE_OK;
>>>>
>>>>    if (dev_is_pci(dev)) {
>>>> -          struct pci_bus *bus = to_pci_dev(dev)->bus;
>>>> +          struct pci_dev *pci_device = to_pci_dev(dev);
>>>>            u32 rid;
>>>>
>>>> -          pci_for_each_dma_alias(to_pci_dev(dev), __get_pci_rid,
>>>> -                                 &rid);
>>>> +          rid = PCI_BDF2(pci_device->bus,pci_device->devfn);
>>>
>>> Beware that the Linux code isn't actually correct to begin with[1]. I
>>> don't know how much Xen deals with PCI bridges and quirks, but as it
>>> stands you should be able to trivially expose the flaw here by plugging
>>> in a Marvell 88SE912x-based SATA card and watching either DMA or MSIs
>>> (or even both) kick up stream table faults.
>>>
>>> Robin.
>>>
>>> [1]:http://www.spinics.net/lists/linux-acpi/msg74844.html 
>>
>> I am not sure how x86 is handling that. The closest I can find would be 
>> domain_context_mapping. I have CCed x86 folks to get more feedback here.
> 
> I'm lacking enough context to know whether this is the issue
> with what we call phantom devices, or something else.

Phantom functions, PCIe-PCI bridges, basically anything that can result
in traffic from a single endpoint appearing on multiple RIDs, or a RID
other than the device's BDF, at the root complex (and IOMMUs/MSI
doorbells beyond).

Robin.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel