mbox series

[V1,0/5] vfio virtual address update

Message ID 1609861013-129801-1-git-send-email-steven.sistare@oracle.com (mailing list archive)
Headers show
Series vfio virtual address update | expand

Message

Steve Sistare Jan. 5, 2021, 3:36 p.m. UTC
Add interfaces that allow the underlying memory object of an iova range
to be mapped to a new virtual address in the host process:

  - VFIO_DMA_UNMAP_FLAG_SUSPEND for VFIO_IOMMU_UNMAP_DMA
  - VFIO_DMA_MAP_FLAG_RESUME flag for VFIO_IOMMU_MAP_DMA
  - VFIO_SUSPEND extension for VFIO_CHECK_EXTENSION

The suspend interface blocks vfio translation of host virtual addresses in
a range, but DMA to already-mapped pages continues.  The resume interface
records the new base VA and resumes translation.  The implementation
supports iommu type1 and mediated devices.

This functionality is necessary for live update, in which a host process
such as qemu exec's an updated version of itself, while preserving its
guest and vfio devices.  The process suspends vfio VA translation, exec's
its new self, mmap's the memory object(s) underlying vfio object, and
resumes VA translation.  For a working example that uses these new
interfaces, see the QEMU patch series "[PATCH V2] Live Update".

Patch 1 modifies the iova rbtree to allow iteration over ranges with gaps,
  without deleting each entry.  This is required by patch 4.
Patch 2 adds an option to unmap all ranges, which simplifies userland code.
Patch 3 adds an interface to detect if an iommu_group has a valid container,
  which patch 5 uses to release a blocked thread if a container is closed.
Patch 4 implements the new ioctl's.
Patch 5 adds blocking to complete the implementation .

Steve Sistare (5):
  vfio: maintain dma_list order
  vfio: option to unmap all
  vfio: detect closed container
  vfio: VA suspend interface
  vfio: block during VA suspend

 drivers/vfio/vfio.c             |  12 ++++
 drivers/vfio/vfio_iommu_type1.c | 122 ++++++++++++++++++++++++++++++++++------
 include/linux/vfio.h            |   1 +
 include/uapi/linux/vfio.h       |  19 ++++++-
 4 files changed, 135 insertions(+), 19 deletions(-)

Comments

Steve Sistare Jan. 12, 2021, 9:52 p.m. UTC | #1
On 1/5/2021 10:36 AM, Steve Sistare wrote:
> Add interfaces that allow the underlying memory object of an iova range
> to be mapped to a new virtual address in the host process:
> 
>   - VFIO_DMA_UNMAP_FLAG_SUSPEND for VFIO_IOMMU_UNMAP_DMA
>   - VFIO_DMA_MAP_FLAG_RESUME flag for VFIO_IOMMU_MAP_DMA
>   - VFIO_SUSPEND extension for VFIO_CHECK_EXTENSION
> 
> The suspend interface blocks vfio translation of host virtual addresses in
> a range, but DMA to already-mapped pages continues.  The resume interface
> records the new base VA and resumes translation.  The implementation
> supports iommu type1 and mediated devices.
> 
> This functionality is necessary for live update, in which a host process
> such as qemu exec's an updated version of itself, while preserving its
> guest and vfio devices.  The process suspends vfio VA translation, exec's
> its new self, mmap's the memory object(s) underlying vfio object, and
> resumes VA translation.  For a working example that uses these new
> interfaces, see the QEMU patch series "[PATCH V2] Live Update".
> 
> Patch 1 modifies the iova rbtree to allow iteration over ranges with gaps,
>   without deleting each entry.  This is required by patch 4.
> Patch 2 adds an option to unmap all ranges, which simplifies userland code.
> Patch 3 adds an interface to detect if an iommu_group has a valid container,
>   which patch 5 uses to release a blocked thread if a container is closed.
> Patch 4 implements the new ioctl's.
> Patch 5 adds blocking to complete the implementation .
> 
> Steve Sistare (5):
>   vfio: maintain dma_list order
>   vfio: option to unmap all
>   vfio: detect closed container
>   vfio: VA suspend interface
>   vfio: block during VA suspend
> 
>  drivers/vfio/vfio.c             |  12 ++++
>  drivers/vfio/vfio_iommu_type1.c | 122 ++++++++++++++++++++++++++++++++++------
>  include/linux/vfio.h            |   1 +
>  include/uapi/linux/vfio.h       |  19 ++++++-
>  4 files changed, 135 insertions(+), 19 deletions(-)
> 

Hi Alex,
  I can send a patch V2 for review if you weigh in on the following:

* preferred interface for unmap-all (patch 2)
* name of the suspend and resume flags (patch 4)
* is vfio_iommu_contained() acceptable, or is a new backend interface needed? (patch 5)

- Steve