Re: [PATCH v3 00/28] Add VDUSE support to Vhost library

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi David,

On 5/26/23 11:14, David Marchand wrote:
> On Thu, May 25, 2023 at 6:25 PM Maxime Coquelin
> <maxime.coquelin@redhat.com> wrote:
>>
>> Note: v2 is identical to v3, it is just a resend because
>> of an issue when posting v2 breaking the series in patchwork.
>>
>> This series introduces a new type of backend, VDUSE,
>> to the Vhost library.
>>
>> VDUSE stands for vDPA device in Userspace, it enables
>> implementing a Virtio device in userspace and have it
>> attached to the Kernel vDPA bus.
>>
>> Once attached to the vDPA bus, it can be used either by
>> Kernel Virtio drivers, like virtio-net in our case, via
>> the virtio-vdpa driver. Doing that, the device is visible
>> to the Kernel networking stack and is exposed to userspace
>> as a regular netdev.
>>
>> It can also be exposed to userspace thanks to the
>> vhost-vdpa driver, via a vhost-vdpa chardev that can be
>> passed to QEMU or Virtio-user PMD.
>>
>> While VDUSE support is already available in upstream
>> Kernel, a couple of patches are required to support
>> network device type:
>>
>> https://gitlab.com/mcoquelin/linux/-/tree/vduse_networking_rfc
>>
>> In order to attach the created VDUSE device to the vDPA
>> bus, a recent iproute2 version containing the vdpa tool is
>> required.
>>
>> Benchmark results:
>> ==================
>>
>> On this v2, PVP reference benchmark has been run & compared with
>> Vhost-user.
>>
>> When doing macswap forwarding in the worload, no difference is seen.
>> When doing io forwarding in the workload, we see 4% performance
>> degradation with VDUSE, comapred to Vhost-user/Virtio-user. It is
>> explained by the use of the IOTLB layer in the Vhost-library when using
>> VDUSE, whereas Vhost-user/Virtio-user does not make use of it.
>>
>> Usage:
>> ======
>>
>> 1. Probe required Kernel modules
>> # modprobe vdpa
>> # modprobe vduse
>> # modprobe virtio-vdpa
>>
>> 2. Build (require vduse kernel headers to be available)
>> # meson build
>> # ninja -C build
>>
>> 3. Create a VDUSE device (vduse0) using Vhost PMD with
>> testpmd (with 4 queue pairs in this example)
>> # ./build/app/dpdk-testpmd --no-pci --vdev=net_vhost0,iface=/dev/vduse/vduse0,queues=4 --log-level=*:9  -- -i --txq=4 --rxq=4
>>
>> 4. Attach the VDUSE device to the vDPA bus
>> # vdpa dev add name vduse0 mgmtdev vduse
>> => The virtio-net netdev shows up (eth0 here)
>> # ip l show eth0
>> 21: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP mode DEFAULT group default qlen 1000
>>      link/ether c2:73:ea:a7:68:6d brd ff:ff:ff:ff:ff:ff
>>
>> 5. Start/stop traffic in testpmd
>> testpmd> start
>> testpmd> show port stats 0
>>    ######################## NIC statistics for port 0  ########################
>>    RX-packets: 11         RX-missed: 0          RX-bytes:  1482
>>    RX-errors: 0
>>    RX-nombuf:  0
>>    TX-packets: 1          TX-errors: 0          TX-bytes:  62
>>
>>    Throughput (since last show)
>>    Rx-pps:            0          Rx-bps:            0
>>    Tx-pps:            0          Tx-bps:            0
>>    ############################################################################
>> testpmd> stop
>>
>> 6. Detach the VDUSE device from the vDPA bus
>> # vdpa dev del vduse0
>>
>> 7. Quit testpmd
>> testpmd> quit
>>
>> Known issues & remaining work:
>> ==============================
>> - Fix issue in FD manager (still polling while FD has been removed)
>> - Add Netlink support in Vhost library
>> - Support device reconnection
>>   -> a temporary patch to support reconnection via a tmpfs file is available,
>>      upstream solution would be in-kernel and is being developed.
>>   -> https://gitlab.com/mcoquelin/dpdk-next-virtio/-/commit/5ad06ce14159a9ce36ee168dd13ef389cec91137
>> - Support packed ring
>> - Provide more performance benchmark results
>>
>> Changes in v2/v3:
>> =================
>> - Fixed mem_set_dump() parameter (patch 4)
>> - Fixed accidental comment change (patch 7, Chenbo)
>> - Change from __builtin_ctz to __builtin_ctzll (patch 9, Chenbo)
>> - move change from patch 12 to 13 (Chenbo)
>> - Enable locks annotation for control queue (Patch 17)
>> - Send control queue notification when used descriptors enqueued (Patch 17)
>> - Lock control queue IOTLB lock (Patch 17)
>> - Fix error path in virtio_net_ctrl_pop() (Patch 17, Chenbo)
>> - Set VDUSE dev FD as NONBLOCK (Patch 18)
>> - Enable more Virtio features (Patch 18)
>> - Remove calls to pthread_setcancelstate() (Patch 22)
>> - Add calls to fdset_pipe_notify() when adding and deleting FDs from a set (Patch 22)
>> - Use RTE_DIM() to get requests string array size (Patch 22)
>> - Set reply result for IOTLB update message (Patch 25, Chenbo)
>> - Fix queues enablement with multiqueue (Patch 26)
>> - Move kickfd creation for better logging (Patch 26)
>> - Improve logging (Patch 26)
>> - Uninstall cvq kickfd in case of handler installation failure (Patch 27)
>> - Enable CVQ notifications once handler is installed (Patch 27)
>> - Don't advertise multiqueue and control queue if app only request single queue pair (Patch 27)
>> - Add release notes
>>
>> Maxime Coquelin (28):
>>    vhost: fix missing guest notif stat increment
>>    vhost: fix invalid call FD handling
>>    vhost: fix IOTLB entries overlap check with previous entry
>>    vhost: add helper of IOTLB entries coredump
>>    vhost: add helper for IOTLB entries shared page check
>>    vhost: don't dump unneeded pages with IOTLB
>>    vhost: change to single IOTLB cache per device
>>    vhost: add offset field to IOTLB entries
>>    vhost: add page size info to IOTLB entry
>>    vhost: retry translating IOVA after IOTLB miss
>>    vhost: introduce backend ops
>>    vhost: add IOTLB cache entry removal callback
>>    vhost: add helper for IOTLB misses
>>    vhost: add helper for interrupt injection
>>    vhost: add API to set max queue pairs
>>    net/vhost: use API to set max queue pairs
>>    vhost: add control virtqueue support
>>    vhost: add VDUSE device creation and destruction
>>    vhost: add VDUSE callback for IOTLB miss
>>    vhost: add VDUSE callback for IOTLB entry removal
>>    vhost: add VDUSE callback for IRQ injection
>>    vhost: add VDUSE events handler
>>    vhost: add support for virtqueue state get event
>>    vhost: add support for VDUSE status set event
>>    vhost: add support for VDUSE IOTLB update event
>>    vhost: add VDUSE device startup
>>    vhost: add multiqueue support to VDUSE
>>    vhost: add VDUSE device stop
>>
>>   doc/guides/prog_guide/vhost_lib.rst    |   4 +
>>   doc/guides/rel_notes/release_23_07.rst |  11 +
>>   drivers/net/vhost/rte_eth_vhost.c      |   3 +
>>   lib/vhost/iotlb.c                      | 333 +++++++------
>>   lib/vhost/iotlb.h                      |  45 +-
>>   lib/vhost/meson.build                  |   5 +
>>   lib/vhost/rte_vhost.h                  |  17 +
>>   lib/vhost/socket.c                     |  72 ++-
>>   lib/vhost/vduse.c                      | 646 +++++++++++++++++++++++++
>>   lib/vhost/vduse.h                      |  33 ++
>>   lib/vhost/version.map                  |   3 +
>>   lib/vhost/vhost.c                      |  51 +-
>>   lib/vhost/vhost.h                      |  90 ++--
>>   lib/vhost/vhost_user.c                 |  51 +-
>>   lib/vhost/vhost_user.h                 |   2 +-
>>   lib/vhost/virtio_net_ctrl.c            | 286 +++++++++++
>>   lib/vhost/virtio_net_ctrl.h            |  10 +
>>   17 files changed, 1424 insertions(+), 238 deletions(-)
>>   create mode 100644 lib/vhost/vduse.c
>>   create mode 100644 lib/vhost/vduse.h
>>   create mode 100644 lib/vhost/virtio_net_ctrl.c
>>   create mode 100644 lib/vhost/virtio_net_ctrl.h
> 
> I did not do a in-depth review, but overall, the series lgtm (and per
> patch compilation looks fine).
> 
> A few comments though:
> - patch 2 is the same as
> https://patchwork.dpdk.org/project/dpdk/patch/168431454344.558450.2397970324914136724.stgit@ebuild.local/
> It would be cool to report the review tags in the first series that
> gets applied.

Done

> - there may be a bug in patch 5, see comment on patch,
> - patch 4, 5 and 6 go together, with patch 6 being the fix itself. I
> understand it was easier to review as splitted patches, but maybe it
> would be simpler to squash them to make the backport trivial.

If possible I would prefer to keep them as separate patches, it will be
much easier to understand the code in the future if a regression
happened.

I'll help the LTS maintainer with backporting it (i.e. request to also
pick patch 5 and 5).

Does that work for you?

> - patch 7 (and some other patches in the series) will increase the
> virtio_net structure but we are not gaining anything on the
> vhost_virtqueue size, so a device + vqs memory footprint will slightly
> increase. This is not a problem afaics?

I have not noticed performance degradation being introduced, but it may 
be good to revisit it.

> - patch 15 breaks the doc, format is incorrect but the CI reported it
> so you will notice it before merging :-),
> 

Fixed!

Thanks,
Maxime