[PATCH v4 2/2] PCI: quirks: Fix ThunderX2 dma alias handling

Tue Apr 11 08:34:07 PDT 2017

On 11/04/17 14:41, Bjorn Helgaas wrote:
> [+cc Joerg]
> 
> On Tue, Apr 11, 2017 at 07:10:48AM +0000, Jayachandran C wrote:
>> On Mon, Apr 10, 2017 at 08:28:47PM -0500, Bjorn Helgaas wrote:
>>> Hi Jayachandran,
>>>
>>> On Mon, Apr 03, 2017 at 01:15:04PM +0000, Jayachandran C wrote:
>>>> The Cavium ThunderX2 arm64 SoCs (called Broadcom Vulcan earlier), the PCI
>>>> topology is slightly unusual. For a multi-node system, it looks like:
>>>>
>>>> [node level PCI bridges - one per node]
>>>>     [SoC PCI devices with MSI-X but no IOMMU]
>>>>     [PCI-PCIe "glue" bridges - upto 14, one per real port below]
>>>>         [PCIe real root ports associated with IOMMU and GICv3 ITS]
>>>>             [External PCI devices connected to PCIe links]
>>>>
>>>> The top two levels of bridges should have introduced aliases since they
>>>> are PCI and PCI/PCIe bridges, but in the case of ThunderX2 they do not.
>>>> In the case of external PCIe devices, the "real" root ports are connected
>>>> to the SMMU and the GIC ITS, so PCI-PCIe bridge does not introduce an
>>>> alias. The SoC PCI devices are directly connected to the GIC ITS, so the
>>>> node level bridges do not introduce an alias either.
>>>>
>>>> To handle this quirk, we mark the real PCIe root ports and node level
>>>> PCI bridges with the flag PCI_DEV_FLAGS_BRIDGE_XLATE_ROOT.  With this,
>>>> pci_for_each_dma_alias() works correctly for external PCIe devices and
>>>> SoC PCI devices.
>>>>
>>>> For the current revision of Cavium ThunderX2, the VendorID and Device ID
>>>> are from Broadcom Vulcan (14e4:90XX).
>>>
>>> Can you supply some text here about why we want to apply this patch?
>>> E.g., does it avoid making unnecessary IOMMU mappings, improve
>>> performance, avoid a crash, etc?
>>
>> If this is for the commit message, I hope the following is ok:
>>
>> "With this change, both MSI-X and IO virtualization work correctly on
>> Cavium ThunderX2. The GIC ITS driver gets the correct device ID to
>> configure MSI-X, the SMMUv3 driver gets the correct Stream IDs for PCI
>> devices, and the IOMMU groups are setup correctly."
> 
> This doesn't get at what the actual problem is.  I'm hoping for
> something like "without this change, we set up an IOMMU mapping for
> requestor ID X, but device DMA uses requestor ID Y because ...., which
> results in an IOMMU fault"

The fact that certain bits of code only consider the last-seen alias RID
does mean we will install all mappings for a RID the IOMMU will never
see, causing subsequent attempt to access them (via the real RID) to
fault. Even with that fixed, considering the past-the-IOMMU-connection
bridges will cause IOMMU group assignment to believe the IOMMU can't
distinguish them and thus prevent individual devices being assignable to
different VMs.

> I've been puzzling over the fact that most of the callers of
> pci_for_each_dma_alias() don't seem to use it correctly.  For Intel
> IOMMUs, domain_context_mapping() uses it to add a mapping for every
> possible alias.  But most of the other callers only look at the last
> alias and ignore all the others.

As it happens, I've just been looking into this and reaching the same
conclusion (not least because a fair few of of the incorrect uses have
my fingerprints all over them). The of_iommu_configure() and
corresponding iort_iommu_configure() uses turn out to already be broken
WRT DMA phantom functions vs. MSIs, but pretty straightforward to fix
(and I now have one of the offending Marvell SATA cards to test things
with). The one in its_pci_msi_prepare() turns out to be totally
backwards, as it actually wants to iterate through every device which
may alias with the given device (any suggestions for the neatest way to
do that are most welcome).

The other uses in pci_msi_*() are a little trickier, as in general we're
assuming every device is associated with just a single ID (certainly for
ARM GICv3 this assumption runs quite strongly through the architecture)
- Marc and I have chucked some ideas around, but in the short term, it
might be easier to just not even pretend to support MSIs from behind
aliasing bridges for the DT/IORT cases. Either way, those are also
broken for the phantom function case (and the tentative fix I've written
will need rethinking in light of this discussion, oh well).

>  That might work most of the time,
> but:
> 
>   - There's no guarantee that pci_for_each_dma_alias() iterates in any
>     particular order, so relying on the current order is fragile,
> 
>   - The pci_add_dma_alias() interface allows an arbitrary number of
>     aliases (as long as they're all on the same bus), and some devices
>     do use more than one, e.g., quirk_dma_func0_alias(),
>     quirk_mic_x200_dma_alias(),
> 
>   - pci_for_each_dma_alias() translates the rules in the PCIe to
>     PCI/PCI-X Bridge spec, r1.0, sec 2.3, about taking ownership into
>     aliases.  I think it's important to pay attention to *every*
>     possible alias, not just the last one.

Ha, that exact page is still open on my desktop since the "oh crap!"
moment last Friday :)

If I've interpreted that spec correctly, and it definitely is the case
that a bridge may alias or not on a per-transaction basis, then that
does end up making matters simpler; I can remove all the attempts to
skip any IDs that the IOMMU is guaranteed never to see due to aliasing,
if that set is in fact empty.

> I suspect the reason this patch makes a difference is because the
> current pci_for_each_dma_alias() believes one of those top-level
> bridges is an alias, and the iterator produces it last, so that's the
> one you map.  The IOMMU is attached lower down, so that top-level
> bridge is not in fact an alias, but since you only look at the *last*
> one, you don't map the correct aliases from lower down in the tree.
> 
> Stopping the iterator earlier happens to make the last alias be one of
> the correct ones, but it doesn't solve the problems of quirked devices
> that can use multiple requester IDs, and it doesn't solve the problem
> of PCIe-to-PCI bridges that optionally take ownership of transactions.

Yes, that's pretty much the state of things, other than also solving the
legitimate problem of get_pci_alias_group() going too far and inferring
a false lack of isolation.

Robin.

>> I can send out a new patch if needed.
>>
>> The on chip SATA and USB use MSI-X, so this is needed for basic
>> functionality of the platform.
> 
> No need for a new patch; I can integrate something into the changelog.
> 
>>>> Signed-off-by: Jayachandran C <jnair at caviumnetworks.com>
>>>> ---
>>>>  drivers/pci/quirks.c | 14 ++++++++++++++
>>>>  1 file changed, 14 insertions(+)
>>>>
>>>> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
>>>> index 6736836..564a84a 100644
>>>> --- a/drivers/pci/quirks.c
>>>> +++ b/drivers/pci/quirks.c
>>>> @@ -3958,6 +3958,20 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x2260, quirk_mic_x200_dma_alias);
>>>>  DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x2264, quirk_mic_x200_dma_alias);
>>>>  
>>>>  /*
>>>> + * The IOMMU and interrupt controller on Broadcom Vulcan/Cavium ThunderX2 are
>>>> + * associated not at the root bus, but at a bridge below. This quirk flag
>>>> + * will ensure that the aliases are identified correctly.
>>>> + */
>>>> +static void quirk_bridge_cavm_thrx2_pcie_root(struct pci_dev *pdev)
>>>> +{
>>>> +	pdev->dev_flags |= PCI_DEV_FLAGS_BRIDGE_XLATE_ROOT;
>>>> +}
>>>> +DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_BROADCOM, 0x9000,
>>>> +				quirk_bridge_cavm_thrx2_pcie_root);
>>>> +DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_BROADCOM, 0x9084,
>>>> +				quirk_bridge_cavm_thrx2_pcie_root);
>>>> +
>>>> +/*
>>>>   * Intersil/Techwell TW686[4589]-based video capture cards have an empty (zero)
>>>>   * class code.  Fix it.
>>>>   */
>>
>> Thanks,
>> JC.
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> linux-arm-kernel at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel