[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [virtio-dev] [PATCH v2] virtio-net: support device stats
在 2021/8/23 下午4:27, Xuan Zhuo 写道:
This patch allows the driver to obtain some statistics from the device. In the back-end implementation, we can count a lot of such information, which can be used for debugging and judging the running status of the back-end. We hope to directly display it to the user through ethtool. Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> --- v2: All keys define by this patch. Not defined by each backend. content.tex | 117 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 117 insertions(+) diff --git a/content.tex b/content.tex index 70a9765..b9fd10a 100644 --- a/content.tex +++ b/content.tex @@ -2978,6 +2978,9 @@ \subsection{Feature bits}\label{sec:Device Types / Network Device / Feature bits \item[VIRTIO_NET_F_CTRL_MAC_ADDR(23)] Set MAC address through control channel. +\item[VIRTIO_NET_F_CTRL_STATS(55)] Device can provide device-level statistics + to the driver through the control channel. + \item[VIRTIO_NET_F_HOST_USO (56)] Device can receive USO packets. Unlike UFO (fragmenting the packet) the USO splits large UDP packet to several segments when each of these smaller packets has UDP header. @@ -3023,6 +3026,7 @@ \subsubsection{Feature bit requirements}\label{sec:Device Types / Network Device \item[VIRTIO_NET_F_GUEST_ANNOUNCE] Requires VIRTIO_NET_F_CTRL_VQ. \item[VIRTIO_NET_F_MQ] Requires VIRTIO_NET_F_CTRL_VQ. \item[VIRTIO_NET_F_CTRL_MAC_ADDR] Requires VIRTIO_NET_F_CTRL_VQ. +\item[VIRTIO_NET_F_CTRL_STATS] Requires VIRTIO_NET_F_CTRL_VQ. \item[VIRTIO_NET_F_RSC_EXT] Requires VIRTIO_NET_F_HOST_TSO4 or VIRTIO_NET_F_HOST_TSO6. \item[VIRTIO_NET_F_RSS] Requires VIRTIO_NET_F_CTRL_VQ. \end{description} @@ -3901,6 +3905,7 @@ \subsubsection{Control Virtqueue}\label{sec:Device Types / Network Device / Devi u8 command; u8 command-specific-data[]; u8 ack; + u8 command-specific-data-reply[]; }; /* ack values */ @@ -3912,6 +3917,9 @@ \subsubsection{Control Virtqueue}\label{sec:Device Types / Network Device / Devi driver, and the device sets the \field{ack} byte. There is little it can do except issue a diagnostic if \field{ack} is not VIRTIO_NET_OK. +\field{command-specific-data-reply} is alloced by driver, then pass to device. +It is filled by the device. It is optional. These commands need to get some +information from the device. \paragraph{Packet Receive Filtering}\label{sec:Device Types / Network Device / Device Operation / Control Virtqueue / Packet Receive Filtering} \label{sec:Device Types / Network Device / Device Operation / Control Virtqueue / Setting Promiscuous Mode}%old label for latexdiff @@ -4390,6 +4398,115 @@ \subsubsection{Legacy Interface: Framing Requirements}\label{sec:Device See \ref{sec:Basic Facilities of a Virtio Device / Virtqueues / Message Framing}. +\paragraph{Device stats}\label{sec:Device Types / Network Device / Device Operation / Control Virtqueue / Device stats} + +If the VIRTIO_NET_F_CTRL_STATS feature is negotiated, the driver can +get device stats from the device by \field{command-specific-data-reply}. + +\subparagraph{Device stats keys}\label{sec:Device Types / Network Device / Device Operation / Control Virtqueue / Device stats keys / Device stats keys} + +The keys of the device stats are defined as follows. When the device fills in +\field{command-specific-data-reply}, it MUST be written in the following order. +Subsequent newly added keys MUST be added at the end of each category. +These keys will eventually be shown to the user. + +All stats are divided into three categories: + +\begin{itemize} + \item dev stats: the stats of the device + \begin{itemize} + \item dev_freeze: The number of device freezes. + \item dev_restore: The number of device restore.
Have we defined freezes/restore function int the spec? And I think the key should be defined like a C macro as others in the spec.
+ \end{itemize} + + \item rx stats: the stats of the rx queue + \begin{itemize} + \item rx[N]_inter: The number of interrupts sent by the rx queue. + \item rx[N]_drop: The number of packets dropped by the rx queue. Contains all kinds of packet loss. + \item rx[N]_drop_overruns: The number of packets dropped by the rx queue when no more avail desc. + \item rx[N]_csum_bad: The number of packets with abnormal csum in the rx queue. + \item rx[N]_lro_packets: The number of lro packets received by rx. + \item rx[N]_lro_bytes: The number of lro bytes received by rx.
Let's use gso instead of lro since lro is never defined in the spec.
+ \item rx[N]_oversize_pkts: The number of oversized packets received by the rx queue. + \end{itemize} + + \item tx stats: the stats of the tx queue + \begin{itemize} + \item tx[N]_inter: The number of interrupts sent by the tx queue. + \item tx[N]_drop: The number of packets dropped by the tx queue. Contains all kinds of packet loss. + \item tx[N]_csum: The number of packets that completes csum by the device. + \item tx[N]_tso_packets: The number of TSO packets transmitted.
Let's use gso instead.
+ \item tx[N]_tso_bytes: The number of TSO bytes transmitted. + \end{itemize} +\end{itemize} + +\subparagraph{Device stats get}\label{sec:Device Types / Network Device / Device Operation / Control Virtqueue / Device stats keys / Device stats get} + +To get the stats, the following definitions are used: +\begin{lstlisting} +#define VIRTIO_NET_CTRL_STATS 6 +#define VIRTIO_NET_CTRL_STATS_GET 0 +\end{lstlisting} + +The following layout structure are used: + +\field{command-specific-data} +\begin{lstlisting} +le64 version; +le64 dev_stats_num; +le64 rx_num; +le64 tx_num; +le64 rx_stats_num; +le64 tx_stats_num; +\end{lstlisting} + +\field{command-specific-data-reply} +\begin{lstlisting} +le64 dev_stats_num; +le64 rx_num; +le64 tx_num;
Any reason for having this, can we simply use max_virtqueue_pairs?
+le64 rx_stats_num; +le64 tx_stats_num; +le64 dev_stats[dev_stats_num]; +le64 rx_stats[rx_stats_num][rx_num]; +le64 tx_stats[tx_stats_num][tx_num];
I guess the version implies rx_stats_num?
+\end{lstlisting} + +The command VIRTIO_NET_CTRL_STATS_GET is used to obtain this information. + +Regarding the variables in \field{command-specific-data}: +\begin{itemize} + \item \field{version}: This is used to specify the version of the layout used. The current value MUST been 0.
I wonder if it's better to just extend the command instead of using version here. E.g VIRTIO_NET_CTRL_STATS_GET_EXTRA etc.
+ \item \field{dev_stats_num}: Indicates the number of dev stats supported by the driver. + \item \field{rx_num}: Indicates how many rx queue information the driver wants to receive. + \item \field{tx_num}: Indicates how many tx queue information the driver wants to receive. + \item \field{rx_stats_num}: Indicates the number of stats information the driver wants for each rx queue. + \item \field{tx_stats_num}: Indicates the number of stats information the driver wants for each tx queue. +\end{itemize} + +When the driver allocates \field{command-specific-data-reply}, it should set the +size of \field{command-specific-data-reply} based on the above values. + +\begin{lstlisting} +size = 5 * 8 + 8 * dev_stats_num + 8 * rx_num * rx_stats_num + 8 * tx_num * tx_stats_num; +\end{lstlisting} + +Regarding the variables in \field{command-specific-data-reply}, +these variables(\field{dev_stats_num}, \field{rx_num}, \field{tx_num}, +\field{rx_stats_num}, \field{tx_stats_num}) are set by the device and MUST be +less than or equal to the variables with the same name in +\field{command-specific-data}. These values indicate the amount actually filled +in. And the number of these implementations in +\field{command-specific-data-reply} is used as the size of the array dev_stats +and rx_stats and tx_stats. + +\field{command-specific-data-reply} is meaningful only when \field{ack} is equal to VIRTIO_NET_OK. + +\subparagraph{Legacy Interface: Device stats}\label{sec:Device Types / Network Device / Device Operation / Control Virtqueue / Device stats / Legacy Interface: Device stats} +When using the legacy interface, transitional devices and drivers +MUST format the all variables according to the native endian of the guest rather +than (necessarily when not using the legacy interface) little-endian.
I'm not sure it's worth to add the support for legacy driver. Since it can't support VIRTIO_NET_F_CTRL_STATS because it has only 32 bit features.
Thanks
+ \section{Block Device}\label{sec:Device Types / Block Device} The virtio block device is a simple virtual block device (ie. -- 2.31.0 --------------------------------------------------------------------- To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]