RE: [PATCH V10 08/11] iommu/vt-d: Add svm/sva invalidate function

From: Liu, Yi L
Date: Wed Apr 01 2020 - 02:57:51 EST


> From: Tian, Kevin <kevin.tian@xxxxxxxxx>
> Sent: Wednesday, April 1, 2020 2:24 PM
> To: Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>
> Subject: RE: [PATCH V10 08/11] iommu/vt-d: Add svm/sva invalidate function
>
> > From: Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>
> > Sent: Wednesday, April 1, 2020 2:14 AM
> >
> > On Sat, 28 Mar 2020 10:01:42 +0000
> > "Tian, Kevin" <kevin.tian@xxxxxxxxx> wrote:
> >
> > > > From: Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>
> > > > Sent: Saturday, March 21, 2020 7:28 AM
> > > >
> > > > When Shared Virtual Address (SVA) is enabled for a guest OS via
> > > > vIOMMU, we need to provide invalidation support at IOMMU API and
> > > > driver level. This patch adds Intel VT-d specific function to
> > > > implement iommu passdown invalidate API for shared virtual address.
> > > >
> > > > The use case is for supporting caching structure invalidation
> > > > of assigned SVM capable devices. Emulated IOMMU exposes queue
> > >
> > > emulated IOMMU -> vIOMMU, since virito-iommu could use the
> > > interface as well.
> > >
> > True, but it does not invalidate this statement about emulated IOMMU. I
> > will add another statement saying "the same interface can be used for
> > virtio-IOMMU as well". OK?
>
> sure
>
> >
> > > > invalidation capability and passes down all descriptors from the
> > > > guest to the physical IOMMU.
> > > >
> > > > The assumption is that guest to host device ID mapping should be
> > > > resolved prior to calling IOMMU driver. Based on the device handle,
> > > > host IOMMU driver can replace certain fields before submit to the
> > > > invalidation queue.
> > > >
> > > > ---
> > > > v7 review fixed in v10
> > > > ---
> > > >
> > > > Signed-off-by: Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>
> > > > Signed-off-by: Ashok Raj <ashok.raj@xxxxxxxxx>
> > > > Signed-off-by: Liu, Yi L <yi.l.liu@xxxxxxxxx>
> > > > ---
> > > > drivers/iommu/intel-iommu.c | 182
> > > > ++++++++++++++++++++++++++++++++++++++++++++
> > > > 1 file changed, 182 insertions(+)
> > > >
> > > > diff --git a/drivers/iommu/intel-iommu.c
> > > > b/drivers/iommu/intel-iommu.c index b1477cd423dd..a76afb0fd51a
> > > > 100644 --- a/drivers/iommu/intel-iommu.c
> > > > +++ b/drivers/iommu/intel-iommu.c
> > > > @@ -5619,6 +5619,187 @@ static void
> > > > intel_iommu_aux_detach_device(struct iommu_domain *domain,
> > > > aux_domain_remove_dev(to_dmar_domain(domain), dev);
> > > > }
> > > >
> > > > +/*
> > > > + * 2D array for converting and sanitizing IOMMU generic TLB
> > > > granularity to
> > > > + * VT-d granularity. Invalidation is typically included in the
> > > > unmap operation
> > > > + * as a result of DMA or VFIO unmap. However, for assigned devices
> > > > guest
> > > > + * owns the first level page tables. Invalidations of translation
> > > > caches in the
> > > > + * guest are trapped and passed down to the host.
> > > > + *
> > > > + * vIOMMU in the guest will only expose first level page tables,
> > > > therefore
> > > > + * we do not include IOTLB granularity for request without PASID
> > > > (second level).
> > >
> > > I would revise above as "We do not support IOTLB granularity for
> > > request without PASID (second level), therefore any vIOMMU
> > > implementation that exposes the SVA capability to the guest should
> > > only expose the first level page tables, implying all invalidation
> > > requests from the guest will include a valid PASID"
> > >
> > Sounds good.
> >
> > > > + *
> > > > + * For example, to find the VT-d granularity encoding for IOTLB
> > > > + * type and page selective granularity within PASID:
> > > > + * X: indexed by iommu cache type
> > > > + * Y: indexed by enum iommu_inv_granularity
> > > > + * [IOMMU_CACHE_INV_TYPE_IOTLB][IOMMU_INV_GRANU_ADDR]
> > > > + *
> > > > + * Granu_map array indicates validity of the table. 1: valid, 0:
> > > > invalid
> > > > + *
> > > > + */
> > > > +const static int
> > > >
> > inv_type_granu_map[IOMMU_CACHE_INV_TYPE_NR][IOMMU_INV_GRANU_
> > > > NR] = {
> > > > + /*
> > > > + * PASID based IOTLB invalidation: PASID selective (per
> > > > PASID),
> > > > + * page selective (address granularity)
> > > > + */
> > > > + {0, 1, 1},
> > > > + /* PASID based dev TLBs, only support all PASIDs or single
> > > > PASID */
> > > > + {1, 1, 0},
> > >
> > > Is this combination correct? when single PASID is being specified, it
> > > is essentially a page-selective invalidation since you need provide
> > > Address and Size.
> > >
> > This is for translation between generic UAPI granu to VT-d granu, it
> > has nothing to do with address and size.
>
> Generic UAPI defines three granularities: domain, pasid and addr.
> from the definition domain applies all entries related to did, pasid
> applies to all entries related to pasid, while addr is specific for a
> range.
>
> from what we just confirmed internally with VT-d spec owner, our
> PASID based dev TLB invalidation always requires addr and size,
> while current uAPI doesn't support multiple PASIDs based range
> invaliation. It sounds to me that you want to use domain to replace
> multiple PASIDs case (G=1), but it then changes the meaning of
> the domain granularity and easily lead to confusion.
>
> I feel Eric's proposal makes more sense. Here we'd better use {0, 0, 1}
> to indicate only addr range invalidation is allowed, matching the
> spec definition. We may use a special flag in iommu_inv_addr_info
> to indicate G=1 case, if necessary.

I agree. G=1 case should be supported. I think we had a flag for global
as there is GL bit in p_iotlb_inv_dsc (a.k.a ext_iotlb_inv_dsc), but it was
dropped as 3.0 spec dropped GL bit. Let's add it back as for DevTLB
flush case.

> > e.g.
> > If user passes IOMMU_INV_GRANU_PASID for the single PASID case as you
> > mentioned, this map table shows it is valid.
> >
> > Then the lookup result will get VT-d granu:
> > QI_DEV_IOTLB_GRAN_PASID_SEL, which means G=0.
> >
> >
> > > > + /* PASID cache */
> > >
> > > PASID cache is fully managed by the host. Guest PASID cache
> > > invalidation is interpreted by vIOMMU for bind and unbind operations.
> > > I don't think we should accept any PASID cache invalidation from
> > > userspace or guest.
> > >
> >
> > True for vIOMMU, this is here for completeness. Can be used by virtio
> > IOMMU, since PC flush is inclusive (IOTLB, devTLB), it is more
> > efficient.
>
> I think it is not correct in concept. We should not allow the userspace or
> guest to request an operation which is beyond its privilege (just because
> doing so may bring some performance benefit). You can always introduce
> new cmd for such purpose.

I guess it was added for the pasid table binding case? Now, our platform
doesn't support it. So I guess we can just make it as unsupported in the
2D table.

Regards,
Yi Liu