lkml.org 
[lkml]   [2020]   [Apr]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v11 05/10] iommu/vt-d: Add bind guest PASID support
On Fri, 17 Apr 2020 09:46:55 +0200
Auger Eric <eric.auger@redhat.com> wrote:

> Hi Kevin,
> On 4/17/20 4:45 AM, Tian, Kevin wrote:
> >> From: Auger Eric
> >> Sent: Thursday, April 16, 2020 6:43 PM
> >>
> > [...]
> >>>>> + if (svm) {
> >>>>> + /*
> >>>>> + * If we found svm for the PASID, there must
> >>>>> be at
> >>>>> + * least one device bond, otherwise svm should
> >>>>> be freed.
> >>>>> + */
> >>>>> + if (WARN_ON(list_empty(&svm->devs))) {
> >>>>> + ret = -EINVAL;
> >>>>> + goto out;
> >>>>> + }
> >>>>> +
> >>>>> + for_each_svm_dev(sdev, svm, dev) {
> >>>>> + /* In case of multiple sub-devices of
> >>>>> the same pdev
> >>>>> + * assigned, we should allow multiple
> >>>>> bind calls with
> >>>>> + * the same PASID and pdev.
> >>>>> + */
> >>>>> + sdev->users++;
> >>>> What if this is not an mdev device. Is it also allowed?
> >>> Yes. IOMMU and VT-d driver is not mdev aware. Here mdev is just an
> >>> example of normal use case. You can bind the same PCI device (PF
> >>> or SRIOV VF) more than once to the same PASID. Just need to
> >>> unbind also.
> >>
> >> I don't get the point of binding a non mdev device several times
> >> with the same PASID. Do you intend to allow that at userspace
> >> level or prevent this from happening in VFIO?
> >
> > I feel it's better to prevent this from happening, otherwise VFIO
> > also needs to track the bind count and do multiple unbinds at
> > mm_exit. But it's not necessary to prevent it in VFIO. We can check
> > here upon whether aux_domain is valid, and if not return -EBUSY.
> Ah OK. So if we can detect the case here it is even better
>
I don't understand why VFIO cannot track, since it is mdev aware. if we
don;t refcount the users, one mdev unbind will result unbind for all
mdev under the same pdev. That may not be the right thing to do.

> Thanks
>
> Eric
> >
> >>
> >> Besides, the comment is a bit misleading as it gives the
> >> impression it is only true for mdev and there is no associated
> >> check.
> >
> > Thanks
> > Kevin
> >
>

[Jacob Pan]

\
 
 \ /
  Last update: 2020-04-17 17:23    [W:0.104 / U:0.220 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site