[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH for 4.6 13/13] xen/iommu: smmu: Add Xen specific code to be able to use the driver

On 19/02/15 11:31 am, Julien Grall wrote:

On 19/02/2015 02:55, Manish wrote:

On 18/02/15 11:52 pm, Julien Grall wrote:

On 18/02/2015 17:30, Jaggi, Manish wrote:
[manish] There are general comments on the data structures
(a) I don't see a use case where for same domain (VM) there would be
different context banks , so linked list may not be required.

I guess you mean the list in arm_smmu_xen_domain? All the devices
pass-through to a domain may not be protected by the same SMMU.
Therefore the context banks are different.
you are right. For each smmu the context bank instance for a xen domain
is duplicated with just a change in context bank id. I was thinking can
it be minimized.

Also, for now a context is allocated per-device. It should be rework
to share the context between multiple device protected by the same
SMMU and pass-through to the same domain.
Yes, this is exactly I an doing at my end.

(b) Also iommu group may not be relevant for the same reason.
I am curious to find the use cases.

The iommu_group is used to store the configuration of the device
protected by an SMMU (i.e the stream ids associated to this domain).

I'm a bit surprised that you think they are not useful...
How do we create an iommu_group in xen ? AFAIK an iommu group is a vfio
group in linux which is assigned to a a VM.
lkvm run -m 512 -k home/Image ...  --vfio-groups=48, 51

> For Xen There are 2 ways of attaching devices using xl pci-attach or in
domU cfg file. Should they create different iommu_groups ?

Why are you talking about VFIO? The iommu_group is created by the SMMU when a new device is added (see arm_smmu_device). At the time we are speaking, Linux and Xen use one iommu group per-device. No matters what VFIO does.

Even though we have two ways to attach PCI on the toolstack, both are using the same hypercall at the end.


I think we can have a discussion over it, parallely I will send RFC patches on your current state of xen-unstable smmu code.

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.