The virtio spec already supports the virtio queue reset function. This patch set
is to add this function to the kernel. The relevant virtio spec information is
here:

    https://github.com/oasis-tcs/virtio-spec/issues/124

Also regarding MMIO support for queue reset, I plan to support it after this
patch is passed.

Please review. Thanks.

v4:
  1. just the code of virtio, without virtio-net
  2. Performing reset on a queue is divided into these steps:
    1. reset_vq: reset one vq
    2. recycle the buffer from vq by virtqueue_detach_unused_buf()
    3. release the ring of the vq by vring_release_virtqueue()
    4. enable_reset_vq: re-enable the reset queue
  3. Simplify the parameters of enable_reset_vq()
  4. add container structures for virtio_pci_common_cfg

v3:
  1. keep vq, irq unreleased

Xuan Zhuo (14):
  virtio_pci: struct virtio_pci_common_cfg add queue_notify_data
  virtio: queue_reset: add VIRTIO_F_RING_RESET
  virtio_ring: queue_reset: add function vring_setup_virtqueue()
  virtio_ring: queue_reset: split: add __vring_init_virtqueue()
  virtio_ring: queue_reset: split: support enable reset queue
  virtio_ring: queue_reset: packed: support enable reset queue
  virtio_ring: queue_reset: extract the release function of the vq ring
  virtio_ring: queue_reset: add vring_release_virtqueue()
  virtio: queue_reset: struct virtio_config_ops add callbacks for
    queue_reset
  virtio_pci: queue_reset: update struct virtio_pci_common_cfg and
    option functions
  virtio_pci: queue_reset: release vq by vp_dev->vqs
  virtio_pci: queue_reset: setup_vq() support vring_setup_virtqueue()
  virtio_pci: queue_reset: vp_setup_vq() support ring_num
  virtio_pci: queue_reset: support VIRTIO_F_RING_RESET

 drivers/virtio/virtio_pci_common.c     |  65 +++++++---
 drivers/virtio/virtio_pci_common.h     |  11 +-
 drivers/virtio/virtio_pci_legacy.c     |   5 +-
 drivers/virtio/virtio_pci_modern.c     |  99 ++++++++++++--
 drivers/virtio/virtio_pci_modern_dev.c |  36 +++++
 drivers/virtio/virtio_ring.c           | 173 ++++++++++++++++++-------
 include/linux/virtio.h                 |   6 +
 include/linux/virtio_config.h          |  13 ++
 include/linux/virtio_pci_modern.h      |   2 +
 include/linux/virtio_ring.h            |  37 ++++--
 include/uapi/linux/virtio_config.h     |   7 +-
 include/uapi/linux/virtio_pci.h        |  14 ++
 12 files changed, 375 insertions(+), 93 deletions(-)

--
2.31.0

On Wed, Feb 9, 2022 at 8:29 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
>
> The virtio spec already supports the virtio queue reset function. This patch set
> is to add this function to the kernel. The relevant virtio spec information is
> here:
>
>     https://github.com/oasis-tcs/virtio-spec/issues/124
>
> Also regarding MMIO support for queue reset, I plan to support it after this
> patch is passed.

So I had an idea, we can implement ethtool_set_ringparam() in this
series to get one real users.

But this came into another question: it looks to me current virito-net
just uses the maximum ring size, so it basically means we just can
decrease the number from startup, so I wonder how much value if we
don't limit the startup queue size to a dedicated value.

Thanks

>
> Please review. Thanks.
>
> v4:
>   1. just the code of virtio, without virtio-net
>   2. Performing reset on a queue is divided into these steps:
>     1. reset_vq: reset one vq
>     2. recycle the buffer from vq by virtqueue_detach_unused_buf()
>     3. release the ring of the vq by vring_release_virtqueue()
>     4. enable_reset_vq: re-enable the reset queue
>   3. Simplify the parameters of enable_reset_vq()
>   4. add container structures for virtio_pci_common_cfg
>
> v3:
>   1. keep vq, irq unreleased
>
> Xuan Zhuo (14):
>   virtio_pci: struct virtio_pci_common_cfg add queue_notify_data
>   virtio: queue_reset: add VIRTIO_F_RING_RESET
>   virtio_ring: queue_reset: add function vring_setup_virtqueue()
>   virtio_ring: queue_reset: split: add __vring_init_virtqueue()
>   virtio_ring: queue_reset: split: support enable reset queue
>   virtio_ring: queue_reset: packed: support enable reset queue
>   virtio_ring: queue_reset: extract the release function of the vq ring
>   virtio_ring: queue_reset: add vring_release_virtqueue()
>   virtio: queue_reset: struct virtio_config_ops add callbacks for
>     queue_reset
>   virtio_pci: queue_reset: update struct virtio_pci_common_cfg and
>     option functions
>   virtio_pci: queue_reset: release vq by vp_dev->vqs
>   virtio_pci: queue_reset: setup_vq() support vring_setup_virtqueue()
>   virtio_pci: queue_reset: vp_setup_vq() support ring_num
>   virtio_pci: queue_reset: support VIRTIO_F_RING_RESET
>
>  drivers/virtio/virtio_pci_common.c     |  65 +++++++---
>  drivers/virtio/virtio_pci_common.h     |  11 +-
>  drivers/virtio/virtio_pci_legacy.c     |   5 +-
>  drivers/virtio/virtio_pci_modern.c     |  99 ++++++++++++--
>  drivers/virtio/virtio_pci_modern_dev.c |  36 +++++
>  drivers/virtio/virtio_ring.c           | 173 ++++++++++++++++++-------
>  include/linux/virtio.h                 |   6 +
>  include/linux/virtio_config.h          |  13 ++
>  include/linux/virtio_pci_modern.h      |   2 +
>  include/linux/virtio_ring.h            |  37 ++++--
>  include/uapi/linux/virtio_config.h     |   7 +-
>  include/uapi/linux/virtio_pci.h        |  14 ++
>  12 files changed, 375 insertions(+), 93 deletions(-)
>
> --
> 2.31.0
>

On Fri, 11 Feb 2022 13:40:52 +0800, Jason Wang <jasowang@redhat.com> wrote:
> On Wed, Feb 9, 2022 at 8:29 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> >
> > The virtio spec already supports the virtio queue reset function. This patch set
> > is to add this function to the kernel. The relevant virtio spec information is
> > here:
> >
> >     https://github.com/oasis-tcs/virtio-spec/issues/124
> >
> > Also regarding MMIO support for queue reset, I plan to support it after this
> > patch is passed.
>
> So I had an idea, we can implement ethtool_set_ringparam() in this
> series to get one real users.

I agree, in fact my local test is using this feature.

>
> But this came into another question: it looks to me current virito-net
> just uses the maximum ring size, so it basically means we just can
> decrease the number from startup, so I wonder how much value if we
> don't limit the startup queue size to a dedicated value.

I also have this consideration, so I want to add a virtio-net module parameter
to specify an initial value.

This initial value also has another meaning. In order to achieve high
performance, the backend can provide a large ring size capability, but we also
hope that the ring size can only be increased through ethtool -G when the user
needs it.

To implement this function, we need to add a new parameter to find_vqs().

If there is no problem, I will bring this function in the next version.

Thannks.


>
> Thanks
>
> >
> > Please review. Thanks.
> >
> > v4:
> >   1. just the code of virtio, without virtio-net
> >   2. Performing reset on a queue is divided into these steps:
> >     1. reset_vq: reset one vq
> >     2. recycle the buffer from vq by virtqueue_detach_unused_buf()
> >     3. release the ring of the vq by vring_release_virtqueue()
> >     4. enable_reset_vq: re-enable the reset queue
> >   3. Simplify the parameters of enable_reset_vq()
> >   4. add container structures for virtio_pci_common_cfg
> >
> > v3:
> >   1. keep vq, irq unreleased
> >
> > Xuan Zhuo (14):
> >   virtio_pci: struct virtio_pci_common_cfg add queue_notify_data
> >   virtio: queue_reset: add VIRTIO_F_RING_RESET
> >   virtio_ring: queue_reset: add function vring_setup_virtqueue()
> >   virtio_ring: queue_reset: split: add __vring_init_virtqueue()
> >   virtio_ring: queue_reset: split: support enable reset queue
> >   virtio_ring: queue_reset: packed: support enable reset queue
> >   virtio_ring: queue_reset: extract the release function of the vq ring
> >   virtio_ring: queue_reset: add vring_release_virtqueue()
> >   virtio: queue_reset: struct virtio_config_ops add callbacks for
> >     queue_reset
> >   virtio_pci: queue_reset: update struct virtio_pci_common_cfg and
> >     option functions
> >   virtio_pci: queue_reset: release vq by vp_dev->vqs
> >   virtio_pci: queue_reset: setup_vq() support vring_setup_virtqueue()
> >   virtio_pci: queue_reset: vp_setup_vq() support ring_num
> >   virtio_pci: queue_reset: support VIRTIO_F_RING_RESET
> >
> >  drivers/virtio/virtio_pci_common.c     |  65 +++++++---
> >  drivers/virtio/virtio_pci_common.h     |  11 +-
> >  drivers/virtio/virtio_pci_legacy.c     |   5 +-
> >  drivers/virtio/virtio_pci_modern.c     |  99 ++++++++++++--
> >  drivers/virtio/virtio_pci_modern_dev.c |  36 +++++
> >  drivers/virtio/virtio_ring.c           | 173 ++++++++++++++++++-------
> >  include/linux/virtio.h                 |   6 +
> >  include/linux/virtio_config.h          |  13 ++
> >  include/linux/virtio_pci_modern.h      |   2 +
> >  include/linux/virtio_ring.h            |  37 ++++--
> >  include/uapi/linux/virtio_config.h     |   7 +-
> >  include/uapi/linux/virtio_pci.h        |  14 ++
> >  12 files changed, 375 insertions(+), 93 deletions(-)
> >
> > --
> > 2.31.0
> >
>

On Fri, Feb 11, 2022 at 2:34 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
>
> On Fri, 11 Feb 2022 13:40:52 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > On Wed, Feb 9, 2022 at 8:29 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> > >
> > > The virtio spec already supports the virtio queue reset function. This patch set
> > > is to add this function to the kernel. The relevant virtio spec information is
> > > here:
> > >
> > >     https://github.com/oasis-tcs/virtio-spec/issues/124
> > >
> > > Also regarding MMIO support for queue reset, I plan to support it after this
> > > patch is passed.
> >
> > So I had an idea, we can implement ethtool_set_ringparam() in this
> > series to get one real users.
>
> I agree, in fact my local test is using this feature.

Right.

>
> >
> > But this came into another question: it looks to me current virito-net
> > just uses the maximum ring size, so it basically means we just can
> > decrease the number from startup, so I wonder how much value if we
> > don't limit the startup queue size to a dedicated value.
>
> I also have this consideration, so I want to add a virtio-net module parameter
> to specify an initial value.

But module parameters are not user friendly. Let it be changed via
ethtool should be sufficient.

>
> This initial value also has another meaning. In order to achieve high
> performance, the backend can provide a large ring size capability, but we also
> hope that the ring size can only be increased through ethtool -G when the user
> needs it.
>
> To implement this function, we need to add a new parameter to find_vqs().
>
> If there is no problem, I will bring this function in the next version.

I think so, and we need to choose a sane default value. I guess
something up to 1024 should be sufficient.

Thanks

>
> Thannks.
>
>
> >
> > Thanks
> >
> > >
> > > Please review. Thanks.
> > >
> > > v4:
> > >   1. just the code of virtio, without virtio-net
> > >   2. Performing reset on a queue is divided into these steps:
> > >     1. reset_vq: reset one vq
> > >     2. recycle the buffer from vq by virtqueue_detach_unused_buf()
> > >     3. release the ring of the vq by vring_release_virtqueue()
> > >     4. enable_reset_vq: re-enable the reset queue
> > >   3. Simplify the parameters of enable_reset_vq()
> > >   4. add container structures for virtio_pci_common_cfg
> > >
> > > v3:
> > >   1. keep vq, irq unreleased
> > >
> > > Xuan Zhuo (14):
> > >   virtio_pci: struct virtio_pci_common_cfg add queue_notify_data
> > >   virtio: queue_reset: add VIRTIO_F_RING_RESET
> > >   virtio_ring: queue_reset: add function vring_setup_virtqueue()
> > >   virtio_ring: queue_reset: split: add __vring_init_virtqueue()
> > >   virtio_ring: queue_reset: split: support enable reset queue
> > >   virtio_ring: queue_reset: packed: support enable reset queue
> > >   virtio_ring: queue_reset: extract the release function of the vq ring
> > >   virtio_ring: queue_reset: add vring_release_virtqueue()
> > >   virtio: queue_reset: struct virtio_config_ops add callbacks for
> > >     queue_reset
> > >   virtio_pci: queue_reset: update struct virtio_pci_common_cfg and
> > >     option functions
> > >   virtio_pci: queue_reset: release vq by vp_dev->vqs
> > >   virtio_pci: queue_reset: setup_vq() support vring_setup_virtqueue()
> > >   virtio_pci: queue_reset: vp_setup_vq() support ring_num
> > >   virtio_pci: queue_reset: support VIRTIO_F_RING_RESET
> > >
> > >  drivers/virtio/virtio_pci_common.c     |  65 +++++++---
> > >  drivers/virtio/virtio_pci_common.h     |  11 +-
> > >  drivers/virtio/virtio_pci_legacy.c     |   5 +-
> > >  drivers/virtio/virtio_pci_modern.c     |  99 ++++++++++++--
> > >  drivers/virtio/virtio_pci_modern_dev.c |  36 +++++
> > >  drivers/virtio/virtio_ring.c           | 173 ++++++++++++++++++-------
> > >  include/linux/virtio.h                 |   6 +
> > >  include/linux/virtio_config.h          |  13 ++
> > >  include/linux/virtio_pci_modern.h      |   2 +
> > >  include/linux/virtio_ring.h            |  37 ++++--
> > >  include/uapi/linux/virtio_config.h     |   7 +-
> > >  include/uapi/linux/virtio_pci.h        |  14 ++
> > >  12 files changed, 375 insertions(+), 93 deletions(-)
> > >
> > > --
> > > 2.31.0
> > >
> >
>