Re: [PATCH v3] virtio-net: support distinguishing between partial and full checksum

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Nov 14, 2023 at 04:49:45PM +0800, Heng Qi wrote:
> virtio-net works in a virtualized system and is somewhat different from
> physical nics. One of the differences is that to save virtio device
> resources, rx may receive packets with partial checksum. However, XDP may
> cause partially checksummed packets to be dropped. So XDP loading conflicts
> with the feature VIRTIO_NET_F_GUEST_CSUM.
> 
> This patch lets the device to supply fully checksummed packets to the driver.
> Then XDP can coexist with VIRTIO_NET_F_GUEST_CSUM to enjoy the benefits of
> device verification checksum.
> 
> In addition, implementation of some performant devices do not generate
> partially checksummed packets, but the standard driver still need to clear
> VIRTIO_NET_F_GUEST_CSUM when loading XDP. If these devices enable the
> full checksum offloading, then the driver can load XDP without clearing
> VIRTIO_NET_F_GUEST_CSUM.
> 
> A new feature bit VIRTIO_NET_F_GUEST_FULL_CSUM is added to solve the above
> situation, which provides the driver with configurable receive full checksum
> offload. If the offload is enabled, then the device must supply fully
> checksummed packets to the driver.
> 
> Use case example:
> If VIRTIO_NET_F_GUEST_FULL_CSUM is negotiated and receive full checksum
> offload is enabled, after XDP processes a packet with full checksum, the
> VIRTIO_NET_HDR_F_DATA_VALID bit is still retained, resulting in the stack
> not needing to validate the checksum again. This is useful for guests:
>   1. Bring the driver advantages such as cpu savings.
>   2. For devices that do not generate partially checksummed packets themselves,
>      XDP can be loaded in the driver without modifying the hardware behavior.
> 
> Several solutions have been discussed in the previous proposal[1].
> After historical discussion, we have tried the method proposed by Jason[2],
> but some complex scenarios and challenges are difficult to deal with.
> We now return to the method suggested in [1].
> 
> [1] https://lists.oasis-open.org/archives/virtio-dev/202305/msg00291.html
> [2] https://lore.kernel.org/all/20230628030506.2213-1-hengqi@linux.alibaba.com/
> 
> Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
> Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
> ---
> v2->v3:
> - Add a section named "Driver Handles Fully Checksummed Packets"
>   and more descriptions. @Michael
> 
> v1->v2:
> - Modify full checksum functionality as a configurable offload
>   that is initially turned off. @Jason
> 
>  device-types/net/description.tex        | 73 +++++++++++++++++++++++--
>  device-types/net/device-conformance.tex |  1 +
>  device-types/net/driver-conformance.tex |  1 +
>  3 files changed, 69 insertions(+), 6 deletions(-)
> 
> diff --git a/device-types/net/description.tex b/device-types/net/description.tex
> index aff5e08..6937a2f 100644
> --- a/device-types/net/description.tex
> +++ b/device-types/net/description.tex
> @@ -122,6 +122,8 @@ \subsection{Feature bits}\label{sec:Device Types / Network Device / Feature bits
>      device with the same MAC address.
>  
>  \item[VIRTIO_NET_F_SPEED_DUPLEX(63)] Device reports speed and duplex.
> +
> +\item[VIRTIO_NET_F_GUEST_FULL_CSUM (64)] Driver handles packets with full checksum.
>  \end{description}
>  
>  \subsubsection{Feature bit requirements}\label{sec:Device Types / Network Device / Feature bits / Feature bit requirements}
> @@ -136,6 +138,7 @@ \subsubsection{Feature bit requirements}\label{sec:Device Types / Network Device
>  \item[VIRTIO_NET_F_GUEST_UFO] Requires VIRTIO_NET_F_GUEST_CSUM.
>  \item[VIRTIO_NET_F_GUEST_USO4] Requires VIRTIO_NET_F_GUEST_CSUM.
>  \item[VIRTIO_NET_F_GUEST_USO6] Requires VIRTIO_NET_F_GUEST_CSUM.
> +\item[VIRTIO_NET_F_GUEST_FULL_CSUM] Requires VIRTIO_NET_F_GUEST_CSUM and VIRTIO_NET_F_CTRL_GUEST_OFFLOADS.
>  
>  \item[VIRTIO_NET_F_HOST_TSO4] Requires VIRTIO_NET_F_CSUM.
>  \item[VIRTIO_NET_F_HOST_TSO6] Requires VIRTIO_NET_F_CSUM.



Apparently this is just to make the patch a bit smaller so you do not
have to find all instances of VIRTIO_NET_F_GUEST_CSUM and replace them
with "VIRTIO_NET_F_GUEST_CSUM or VIRTIO_NET_F_GUEST_FULL_CSUM".

But, this is actually a problem : we have places in spec which only
say VIRTIO_NET_F_GUEST_CSUM but actually mean "VIRTIO_NET_F_GUEST_CSUM
negotiated and not disabled by VIRTIO_NET_F_CTRL_GUEST_OFFLOADS".

So this just makes no sense to me. We have:
\item[VIRTIO_NET_F_GUEST_CSUM (1)] Driver handles packets with partial checksum.


and here apparently when you have driver that handles packets with partial checksum
*and* packets with full checksum then this means that no, it does not
handle packets with partial checksum.
It might look ok when you just look at the patch but when people read
the full spec this is just confusing.



Let me try: in fact VIRTIO_NET_F_GUEST_CSUM now means "driver handles
packets with checksum (partial or full)"? Even with no features checksum can cover
all of the packet nothing prevents that.

So now we have a new flag that means
"driver can only handle fully checksummed packets".






> @@ -398,6 +401,58 @@ \subsection{Device Initialization}\label{sec:Device Types / Network Device / Dev
>  A truly minimal driver would only accept VIRTIO_NET_F_MAC and ignore
>  everything else.
>  
> +\subsubsection{Driver Handles Fully Checksummed Packets}\label{sec:Device Types / Network Device / Device Initialization / Driver Handles Fully Checksummed Packets}
> +
> +The VIRTIO_NET_F_GUEST_CSUM feature indicates that the driver can handle
> +partially or fully checksummed packets from the device. When the
> +driver only expects fully checksummed packets, the VIRTIO_NET_F_GUEST_FULL_CSUM
> +feature can be negotiated if the device offers it.
> +Then the driver only handles packets with full checksum.
> +
> +By negotiating the VIRTIO_NET_F_GUEST_FULL_CSUM feature, the driver can
> +benefit, for example, from the device's ability to calculate and validate the checksum
> +in scenarios where partially checksummed packets are not compatible.
> +
> +Delivering fully checksummed packets rather than partially
> +checksummed packets incurs additional overhead for the device.
> +As a result, receive full-checksum offload (meaning the driver only handles
> +packets with full checksum) is disabled by default \ref{sec:Device Types / Network Device
> +/ Device Operation / Control Virtqueue / Offloads State Configuration}.
> +
> +Receive full-checksum offload can be enabled if the driver successfully
> +sends the VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET command with the
> +VIRTIO_NET_F_GUEST_FULL_CSUM bit set.

But how much is this "additional overhead" and how does driver know when
this should be enabled as opposed to just disabling checksum offload
completely?


> +
> +\drivernormative{\subsubsection}{Driver Handles Fully Checksummed Packets}{sec:Device Types / Network Device / Device Initialization / Driver Handles Fully Checksummed Packets}
> +
> +The driver MUST NOT enable receive full-checksum offload for which
> +VIRTIO_NET_F_GUEST_FULL_CSUM has not been negotiated.

I don't understand what this is saying.  We need to come up with a way
to document this all without inventing terms like "full-checksum offload".

IIUC all this does is basically require that checksum covers all of the
packet.

-- 
MST