Re: [PATCH 5/5] virtio: Add bounce DMA ops

[Lu Baolu]

On 2020/4/29 14:50, Michael S. Tsirkin wrote:
> On Wed, Apr 29, 2020 at 01:42:13PM +0800, Lu Baolu wrote:
> > On 2020/4/29 12:57, Michael S. Tsirkin wrote:
> > > On Wed, Apr 29, 2020 at 10:22:32AM +0800, Lu Baolu wrote:
> > > > On 2020/4/29 4:41, Michael S. Tsirkin wrote:
> > > > > On Tue, Apr 28, 2020 at 11:19:52PM +0530, Srivatsa Vaddagiri wrote:
> > > > > > * Michael S. Tsirkin<mst@xxxxxxxxxx>   [2020-04-28 12:17:57]:
> > > > > >
> > > > > > > Okay, but how is all this virtio specific?  For example, why not allow
> > > > > > > separate swiotlbs for any type of device?
> > > > > > > For example, this might make sense if a given device is from a
> > > > > > > different, less trusted vendor.
> > > > > > Is swiotlb commonly used for multiple devices that may be on different trust
> > > > > > boundaries (and not behind a hardware iommu)?
> > > > > Even a hardware iommu does not imply a 100% security from malicious
> > > > > hardware. First lots of people use iommu=pt for performance reasons.
> > > > > Second even without pt, unmaps are often batched, and sub-page buffers
> > > > > might be used for DMA, so we are not 100% protected at all times.
> > > > For untrusted devices, IOMMU is forced on even iommu=pt is used;
> > > I think you are talking about untrusted*drivers*  like with VFIO.
> > No. I am talking about untrusted devices like thunderbolt peripherals.
> > We always trust drivers hosted in kernel and the DMA APIs are designed
> > for them, right?
> >
> > Please refer to this series.
> >
> > https://lkml.org/lkml/2019/9/6/39
> >
> > Best regards,
> > baolu
> Oh, thanks for that! I didn't realize Linux is doing this.
>
> So it seems that with modern Linux, all one needs
> to do on x86 is mark the device as untrusted.
> It's already possible to do this with ACPI and with OF - would that be
> sufficient for achieving what this patchset is trying to do?

Yes.

> Adding more ways to mark a device as untrusted, and adding
> support for more platforms to use bounce buffers
> sounds like a reasonable thing to do.

Agreed.

Best regards,
baolu