When IOMMU and vhost are enabled together, QEMU tracks IOTLB or
Device-TLB unmap events depending on whether Device-TLB is enabled. But
even if Device-TLB and PCI ATS is enabled, the guest can reject to use
it. For example, this situation appears when Windows Server 2022 is
running with intel-iommu with device-iotlb=on and virtio-net-pci with
vhost=on. The guest implies that no address translation info cached in
device IOTLB and doesn't send device IOTLB invalidation commands. So,
it leads to irrelevant address translations in vhost-net in the host
kernel. Therefore network frames from the guest in host tap interface
contains wrong payload data.

This series adds checking of ATS state for proper unmap flag register
(IOMMU_NOTIFIER_UNMAP or IOMMU_NOTIFIER_DEVIOTLB_UNMAP).

Tested on Windows Server 2022, Windows 11 and Fedora guests with
 -device virtio-net-pci,bus=pci.3,netdev=nd0,iommu_platform=on,ats=on
 -netdev tap,id=nd0,ifname=tap1,script=no,downscript=no,vhost=on
 -device intel-iommu,intremap=on,eim=on,device-iotlb=on/off
Tested on Fedora guest with
 -device virtio-iommu

Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2001312

v4: call vhost_toggle_device_iotlb regardless of vhost backend,
    move vhost_started check to generic part
v3: call virtio_pci_ats_ctrl_trigger directly, remove
    IOMMU_NOTIFIER_UNMAP fallbacks
v2: remove memory_region_iommu_notify_flags_changed, move trigger to
    VirtioDeviceClass, use vhost_ops, use device_iotlb name

Viktor Prutyanov (2):
  vhost: register and change IOMMU flag depending on Device-TLB state
  virtio-net: pass Device-TLB enable/disable events to vhost

 hw/net/virtio-net.c       |  1 +
 hw/virtio/vhost.c         | 38 ++++++++++++++++++++++++++------------
 include/hw/virtio/vhost.h |  1 +
 3 files changed, 28 insertions(+), 12 deletions(-)

-- 
2.21.0

On Thu, May 25, 2023 at 03:57:40PM +0300, Viktor Prutyanov wrote:
> When IOMMU and vhost are enabled together, QEMU tracks IOTLB or
> Device-TLB unmap events depending on whether Device-TLB is enabled. But
> even if Device-TLB and PCI ATS is enabled, the guest can reject to use
> it. For example, this situation appears when Windows Server 2022 is
> running with intel-iommu with device-iotlb=on and virtio-net-pci with
> vhost=on. The guest implies that no address translation info cached in
> device IOTLB and doesn't send device IOTLB invalidation commands. So,
> it leads to irrelevant address translations in vhost-net in the host
> kernel. Therefore network frames from the guest in host tap interface
> contains wrong payload data.
> 
> This series adds checking of ATS state for proper unmap flag register
> (IOMMU_NOTIFIER_UNMAP or IOMMU_NOTIFIER_DEVIOTLB_UNMAP).
> 
> Tested on Windows Server 2022, Windows 11 and Fedora guests with
>  -device virtio-net-pci,bus=pci.3,netdev=nd0,iommu_platform=on,ats=on
>  -netdev tap,id=nd0,ifname=tap1,script=no,downscript=no,vhost=on
>  -device intel-iommu,intremap=on,eim=on,device-iotlb=on/off
> Tested on Fedora guest with
>  -device virtio-iommu

Fails build on windows hosts:

https://gitlab.com/mstredhat/qemu/-/pipelines/909063579/failures

C:\GitLab-Runner\builds\mstredhat\qemu\output/../hw/net/virtio-net.c:3954: undefined reference to `vhost_toggle_device_iotlb'




> Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2001312
> 
> v4: call vhost_toggle_device_iotlb regardless of vhost backend,
>     move vhost_started check to generic part
> v3: call virtio_pci_ats_ctrl_trigger directly, remove
>     IOMMU_NOTIFIER_UNMAP fallbacks
> v2: remove memory_region_iommu_notify_flags_changed, move trigger to
>     VirtioDeviceClass, use vhost_ops, use device_iotlb name
> 
> Viktor Prutyanov (2):
>   vhost: register and change IOMMU flag depending on Device-TLB state
>   virtio-net: pass Device-TLB enable/disable events to vhost
> 
>  hw/net/virtio-net.c       |  1 +
>  hw/virtio/vhost.c         | 38 ++++++++++++++++++++++++++------------
>  include/hw/virtio/vhost.h |  1 +
>  3 files changed, 28 insertions(+), 12 deletions(-)
> 
> -- 
> 2.21.0