From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Sender: List-Post: List-Help: List-Unsubscribe: List-Subscribe: Received: from lists.oasis-open.org (oasis-open.org [10.110.1.242]) by lists.oasis-open.org (Postfix) with ESMTP id 1C6D69861B0 for ; Tue, 31 May 2022 05:32:02 +0000 (UTC) Date: Tue, 31 May 2022 01:31:53 -0400 From: "Michael S. Tsirkin" Message-ID: <20220531012552-mutt-send-email-mst@kernel.org> References: <20220507071533.104000-1-xuanzhuo@linux.alibaba.com> <2a597a0d-29b2-adc9-75e2-cc1c552384f4@redhat.com> <1653544991.258558-1-xuanzhuo@linux.alibaba.com> <20220530141708-mutt-send-email-mst@kernel.org> MIME-Version: 1.0 In-Reply-To: Subject: Re: [virtio-dev] Re: [PATCH v2] virtio_net: support split header Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: quoted-printable To: Jason Wang Cc: Xuan Zhuo , Virtio-Dev List-ID: On Tue, May 31, 2022 at 12:38:14PM +0800, Jason Wang wrote: > On Tue, May 31, 2022 at 2:23 AM Michael S. Tsirkin wrote= : > > > > On Mon, May 30, 2022 at 01:56:06PM +0800, Jason Wang wrote: > > > On Thu, May 26, 2022 at 2:31 PM Xuan Zhuo wrote: > > > > > > > > On Mon, 9 May 2022 16:41:11 +0800, Jason Wang = wrote: > > > > > > > > > > =E5=9C=A8 2022/5/7 15:15, Xuan Zhuo =E5=86=99=E9=81=93: > > > > > > The purpose of this feature is to split the header and the payl= oad of > > > > > > the packet. > > > > > > > > > > > > | receive buffer = | > > > > > > | 0th descriptor | 1th descri= ptor | > > > > > > | virtnet hdr | mac | ip hdr | tcp hdr|<-- hold -->| payload = | > > > > > > > > > > > > We can use a buffer plus a separate page when allocating the re= ceive > > > > > > buffer. In this way, we can ensure that all payloads can be > > > > > > independently in a page, which is very beneficial for the zeroc= opy > > > > > > implemented by the upper layer. > > > > > > > > > > > > Signed-off-by: Xuan Zhuo > > > > > > --- > > > > > > conformance.tex | 2 ++ > > > > > > content.tex | 72 ++++++++++++++++++++++++++++++++++++++++= +++++++++ > > > > > > 2 files changed, 74 insertions(+) > > > > > > > > > > > > diff --git a/conformance.tex b/conformance.tex > > > > > > index 663e7c3..6f561fb 100644 > > > > > > --- a/conformance.tex > > > > > > +++ b/conformance.tex > > > > > > @@ -149,6 +149,7 @@ \section{Conformance Targets}\label{sec:Con= formance / Conformance Targets} > > > > > > \item \ref{drivernormative:Device Types / Network Device / De= vice Operation / Control Virtqueue / Automatic receive steering in multique= ue mode} > > > > > > \item \ref{drivernormative:Device Types / Network Device / De= vice Operation / Control Virtqueue / Offloads State Configuration / Setting= Offloads State} > > > > > > \item \ref{drivernormative:Device Types / Network Device / De= vice Operation / Control Virtqueue / Receive-side scaling (RSS) } > > > > > > +\item \ref{drivernormative:Device Types / Network Device / Dev= ice Operation / Control Virtqueue / Split Header} > > > > > > \end{itemize} > > > > > > > > > > > > \conformance{\subsection}{Block Driver Conformance}\label{sec= :Conformance / Driver Conformance / Block Driver Conformance} > > > > > > @@ -411,6 +412,7 @@ \section{Conformance Targets}\label{sec:Con= formance / Conformance Targets} > > > > > > \item \ref{devicenormative:Device Types / Network Device / De= vice Operation / Control Virtqueue / Gratuitous Packet Sending} > > > > > > \item \ref{devicenormative:Device Types / Network Device / De= vice Operation / Control Virtqueue / Automatic receive steering in multique= ue mode} > > > > > > \item \ref{devicenormative:Device Types / Network Device / De= vice Operation / Control Virtqueue / Receive-side scaling (RSS) / RSS proce= ssing} > > > > > > +\item \ref{devicenormative:Device Types / Network Device / Dev= ice Operation / Control Virtqueue / Split Header} > > > > > > \end{itemize} > > > > > > > > > > > > \conformance{\subsection}{Block Device Conformance}\label{sec= :Conformance / Device Conformance / Block Device Conformance} > > > > > > diff --git a/content.tex b/content.tex > > > > > > index 060bdab..3340402 100644 > > > > > > --- a/content.tex > > > > > > +++ b/content.tex > > > > > > @@ -3092,6 +3092,9 @@ \subsection{Feature bits}\label{sec:Devic= e Types / Network Device / Feature bits > > > > > > \item[VIRTIO_NET_F_CTRL_MAC_ADDR(23)] Set MAC address through= control > > > > > > channel. > > > > > > > > > > > > +\item[VIRTIO_NET_F_SPLIT_HEADER (55)] Device supports to split= the header and > > > > > > + the payload. > > > > > > + > > > > > > \item[VIRTIO_NET_F_HOST_USO (56)] Device can receive USO pack= ets. Unlike UFO > > > > > > (fragmenting the packet) the USO splits large UDP packet > > > > > > to several segments when each of these smaller packets has U= DP header. > > > > > > @@ -3139,6 +3142,7 @@ \subsubsection{Feature bit requirements}\= label{sec:Device Types / Network Device > > > > > > \item[VIRTIO_NET_F_CTRL_MAC_ADDR] Requires VIRTIO_NET_F_CTRL_= VQ. > > > > > > \item[VIRTIO_NET_F_RSC_EXT] Requires VIRTIO_NET_F_HOST_TSO4 o= r VIRTIO_NET_F_HOST_TSO6. > > > > > > \item[VIRTIO_NET_F_RSS] Requires VIRTIO_NET_F_CTRL_VQ. > > > > > > +\item[VIRTIO_NET_F_SPLIT_HEADER] Requires VIRTIO_NET_F_CTRL_VQ= . > > > > > > \end{description} > > > > > > > > > > > > \subsubsection{Legacy Interface: Feature bits}\label{sec:Devi= ce Types / Network Device / Feature bits / Legacy Interface: Feature bits} > > > > > > @@ -3370,6 +3374,7 @@ \subsection{Device Operation}\label{sec:D= evice Types / Network Device / Device O > > > > > > #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 > > > > > > #define VIRTIO_NET_HDR_F_DATA_VALID 2 > > > > > > #define VIRTIO_NET_HDR_F_RSC_INFO 4 > > > > > > +#define VIRTIO_NET_HDR_F_SPLIT_HEADER 8 > > > > > > u8 flags; > > > > > > #define VIRTIO_NET_HDR_GSO_NONE 0 > > > > > > #define VIRTIO_NET_HDR_GSO_TCPV4 1 > > > > > > @@ -4471,6 +4476,73 @@ \subsubsection{Control Virtqueue}\label{= sec:Device Types / Network Device / Devi > > > > > > according to the native endian of the guest rather than > > > > > > (necessarily when not using the legacy interface) little-endi= an. > > > > > > > > > > > > +\paragraph{Split Header}\label{sec:Device Types / Network Devi= ce / Device Operation / Control Virtqueue / Split Header} > > > > > > + > > > > > > +If the VIRTIO_NET_F_SPLIT_HEADER feature is negotiated, > > > > > > +the device supports to split the header and the payload. > > > > > > +The header and payload will be separated into different buffer= s. > > > > > > > > > > > > > > > I think you meant "descriptors" instead of "buffers". > > > > > > > > Will fix. > > > > No, actually, I think this is exactly buffers. > > device is always in terms of buffers, descriptors is > > a means to implement buffers. > > > > And the feature should > > depend on mergeable buffers. Without mergeable buffers you do not > > need these tricks, just add two descriptors and be done > > with it. >=20 > How do we know the length of a header in advance to allocate the exact > descriptor properly in this case? >=20 > Thanks We don't, but with mergeable buffers we don't have to. So basically, the feature says that instead of using all of 1st buffer then proceeding to the next one, device will use the 1st one just for the headers. - without split header =09buf1: hdr1 + data1part1 =09buf2: data1part2 - with split header =09buf1: hdr1 =09buf1: data1part1 =09buf2: data1part2 how do buffers map to descriptors isn't that relevant from this POV. > > > > Right? > > > > > > > > > > > > > > > > > > > > > > + > > > > > > +\subparagraph{Split Header}\label{sec:Device Types / Network D= evice / Device Operation / Control Virtqueue / Split Header / Setting Split= Header} > > > > > > + > > > > > > +To configure the split header, the following layout structure = and definitions > > > > > > +are used: > > > > > > + > > > > > > +\begin{lstlisting} > > > > > > +struct virtio_net_split_header_config { > > > > > > +#define VIRTIO_NET_SPLIT_HEADER_TYPE_TCPv4 1 > > > > > > > > > > > > > > > I think it's better to capitalize: TCPV4 or TCP4. > > > > > > > > OK. > > > > > > > > > > > > > > > > > > > > +#define VIRTIO_NET_SPLIT_HEADER_TYPE_TCPv6 2 > > > > > > +#define VIRTIO_NET_SPLIT_HEADER_TYPE_UDPv4 4 > > > > > > +#define VIRTIO_NET_SPLIT_HEADER_TYPE_UDPv6 8 > > > > > > + le64 type; > > > > > > +}; > > > > > > + > > > > > > +#define VIRTIO_NET_CTRL_SPLIT_HEADER 6 > > > > > > + #define VIRTIO_NET_CTRL_SPLIT_HEADER_SET 0 > > > > > > +\end{lstlisting} > > > > > > + > > > > > > +The class VIRTIO_NET_CTRL_SPLIT_HEADER has one command: > > > > > > +VIRTIO_NET_CTRL_SPLIT_HEADER_SET applies the new split header = configuration. > > > > > > + > > > > > > +\field{type} passed as command data is a bitmask, bits set def= ine > > > > > > +packet types to split header, bits cleared - split header to b= e disabled. > > > > > > > > > > > > > > > (Not a native speaker, try my best to help) > > > > > > > > > > This sentence might need some tweaks: > > > > > > > > > > The driver can enable or disable the split by setting or clearing > > > > > corresponding bits in \field{type}. > > > > > > > > Thanks. > > > > > > > > > > > > > > > > > > > > + > > > > > > +The header contains the struct virtio_net_hdr and the header o= f the package. > > > > > > +Such as \field{VIRTIO_NET_SPLIT_HEADER_TYPE_TCPv4} specified h= eader contains > > > > > > +virtio_net_hdr, MAC header, IPv4 header (including IPv4 option= s), TCP header > > > > > > +(include TCP options). The back part is the payload. > > > > > > > > > > > > > > > I think it's better to either explain each type one by one instea= d of > > > > > only describing the TCPv4. Or we can simply say the head contains= all > > > > > headers before level 4 payload? > > > > > > > > OK. Thanks. > > > > > > > So this always splits out at protocol level? > > What if device wants to align e.g. IP protocol header? > > Ethernet? > > > > > > Also, what about aligning at an offset? > > E.g. NET_IP_ALIGN? > > > > > > > > > > > > > > > > > > > > > > > > + > > > > > > +\devicenormative{\subparagraph}{Setting Split Header}{Device T= ypes / Network Device / Device Operation / Control Virtqueue / Split Header= } > > > > > > + > > > > > > +Split header MUST be disabled after device initialization. > > > > > > > > > > > > > > > I think you actually mean "upon device reset"? > > > > > > > > Yes. > > > > > > > > > > > > > > > > > > > > > > > > + > > > > > > +A device MUST NOT perform split header in the following cases: > > > > > > +\begin{itemize} > > > > > > + \item device does not recognize protocol of the packet. > > > > > > + \item \field{type} does not include the protocol of the pa= cket. > > > > > > + \item the packet is a IP fragmentation. > > > > > > > > > > > > > > > What's the reason for this limit? > > > > > > > > Rethinking this problem, some implementation difficulties that I or= iginally > > > > thought can be solved. So I think this limitation can be removed. > > > > > > > > > > > > > > > > > > > > + \item the receive buffer consists of only one descriptor. > > > > > > > > > > > > > > > descriptor actually. > > > > > > > > Will fix. > > > > > > > > > > > > > > > > > > > > + \item the header exceeds the size of the 0th descriptor. > > > > > > > > > > > > > > > It looks to me the spec count from the first. > > > > > > > > Will fix. > > > > > > > > > > > > > > > > > > > > + \item If VIRTIO_NET_F_MRG_RXBUF is not negotiated and the = size of the > > > > > > + payload is greater than the total size of the 1th\ldot= s Nth descriptor. > > > > > > > > > > > > > > > Maybe something like the following is better: > > > > > > > > > > The size of the payload exceeds the length of the descriptor or b= uffer > > > > > chain starting from the 2nd descriptor. > > > > > > > > OK. > > > > > > > > > > > > > > > > > > > > +\end{itemize} > > > > > > + > > > > > > +If the split header completed, > > > > > > > > > > > > > > > If the header is split by the device? > > > > > > > > Yes. > > > > > > > > > > > > > > > > > > > > then the \field{flags} of virtnet hdr MUST > > > > > > > > > > > > > > > We can drop "then" here and it's better to use structure virtio_n= et_hdr > > > > > > > > Will fix. > > > > > > > > > > > > > > > > > > > > +contains VIRTIO_NET_HDR_F_SPLIT_HEADER. The header MUST is on = the buffer of the > > > > > > +0th descriptor, > > > > > > > > > > > > > > > first and second should be used here. > > > > > > > > OK > > > > > > > > > > > > > > > > > > > > and the payload MUST starts from the buffer of the 1th descri= ptor. > > > > > > +The device MUST set \field{hdr_len} of virtnet hdr. > > > > > > > > > > > > > > > To what value? > > > > > > > > I'll make it clear in the next version. > > > > > > > > > > > > > > > > > > > > + > > > > > > +If VIRTIO_NET_F_MRG_RXBUF is negotiated and the device is to u= se multiple > > > > > > +receive buffers, each subsequent receive buffer MUST skip the = 0th descriptor. > > > > > > > > > > > > > > > I don't see why we need this. > > > > > > > > If used to implement zerocopy, the second desc points to a page-ali= gned buffer. > > > > The first desc points to a small buffer for saving the header. The = first buffer > > > > and subsequent buffers are generally discontinuous. So giving up th= e first > > > > buffer and directly guaranteeing the data to the page-aligned buffe= r can achieve > > > > better performance. And the driver can reuse the first buffer. > > > > > > I see, it might be better if we can explain this here. > > > > > > Thanks > > > > > > > > > > > > > > > > > > > > > > > + > > > > > > +\drivernormative{\subparagraph}{Setting Split Header}{Device T= ypes / Network Device / Device Operation / Control Virtqueue / Split Header= } > > > > > > + > > > > > > +If VIRTIO_NET_HDR_F_SPLIT_HEADER bit in \field{flags} is set, = the driver MUST > > > > > > +believe \field{hdr_len}, the length of the header in the 0th d= escriptor is equal > > > > > > +to the length of struct virtio_net_hdr plus \field{hdr_len}. > > > > > > > > > > > > > > > See above we can reuse the device normative an drop the second pa= rt > > > > > starting from " the length of the header ..." > > > > > > > > OK > > > > > > > > > > > > > > > > > > > > + > > > > > > +If the split header function is enable, the buffers submitted = by the driver > > > > > > +SHOULD at least be composed of two descriptors. > > > > > > > > > > > > > > > Do we need to mention it's only used for RX? (I think we don't ne= ed this > > > > > for TX). > > > > > > > > Yes. > > > > > > > > > > > > > > > > > > > > The buffer specified by the 0th > > > > > > +descriptor SHOULD be able to accommodate the header. > > > > > > > > > > > > > > > We probably need to clarify that the header here means both the v= irtio > > > > > net header and the protocol header. > > > > > > > > Will fix. > > > > > > > > Thanks. > > > > > > > > > > > > > > Thanks > > > > > > > > > > > > > > > > > > > > > > \subsubsection{Legacy Interface: Framing Requirements}\label{= sec:Device > > > > > > Types / Network Device / Legacy Interface: Framing Requiremen= ts} > > > > > > > > > > > > > > > -----------------------------------------------------------------= ---- > > > > > To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.o= rg > > > > > For additional commands, e-mail: virtio-dev-help@lists.oasis-open= .org > > > > > > > > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org