numam-dpdk/doc/guides/nics/netvsc.rst
Long Li 096b31fc0d net/netvsc: control use of external mbuf on Rx
When receiving packets, netvsp puts data in a buffer mapped through UIO.
Depending on packet size, netvsc may attach the buffer as an external
mbuf. This is not a problem if this mbuf is consumed in the application,
and the application can correctly read data out of an external mbuf.

However, there are two problems with data in an external mbuf.
1. Due to the limitation of the kernel UIO implementation, physical
   address of this external buffer is not exposed to the user-mode. If
   this mbuf is passed to another driver, the other driver is unable to
   map this buffer to iova.
2. Some DPDK applications are not aware of external mbuf, and may bug
   when they receive an mbuf with external buffer attached.

Introduce a driver parameter "rx_extmbuf_enable" to control if netvsc
should use external mbuf for receiving packets. The default value is 0.
(netvsc doesn't use external mbuf, it always allocates mbuf and copy
data to mbuf) A non-zero value tells netvsc to attach external buffers
to mbuf on receiving packets, thus avoid copying memory.

Signed-off-by: Long Li <longli@microsoft.com>
2020-11-03 23:35:07 +01:00

144 lines
5.6 KiB
ReStructuredText

.. SPDX-License-Identifier: BSD-3-Clause
Copyright(c) Microsoft Corporation.
Netvsc poll mode driver
=======================
The Netvsc Poll Mode driver (PMD) provides support for the paravirtualized
network device for Microsoft Hyper-V. It can be used with
Window Server 2008/2012/2016, Windows 10.
The device offers multi-queue support (if kernel and host support it),
checksum and segmentation offloads.
Features and Limitations of Hyper-V PMD
---------------------------------------
In this release, the hyper PMD driver provides the basic functionality of packet reception and transmission.
* It supports merge-able buffers per packet when receiving packets and scattered buffer per packet
when transmitting packets. The packet size supported is from 64 to 65536.
* The PMD supports multicast packets and promiscuous mode subject to restrictions on the host.
In order to this to work, the guest network configuration on Hyper-V must be configured to allow MAC address
spoofing.
* The device has only a single MAC address.
Hyper-V driver does not support MAC or VLAN filtering because the Hyper-V host does not support it.
* VLAN tags are always stripped and presented in mbuf tci field.
* The Hyper-V driver does not use or support interrupts. Link state change
callback is done via change events in the packet ring.
* The maximum number of queues is limited by the host (currently 64).
When used with 4.16 kernel only a single queue is available.
* This driver supports SR-IOV network acceleration.
If SR-IOV is enabled then the driver will transparently manage the interface,
and send and receive packets using the VF path.
The VDEV_NETVSC and FAILSAFE drivers are *not* used when using netvsc PMD.
Installation
------------
The Netvsc PMD is a standalone driver, similar to virtio and vmxnet3.
Using Netvsc PMD requires that the associated VMBUS device be bound to the userspace
I/O device driver for Hyper-V (uio_hv_generic). By default, all netvsc devices
will be bound to the Linux kernel driver; in order to use netvsc PMD the
device must first be overridden.
The first step is to identify the network device to override.
VMBUS uses Universal Unique Identifiers
(`UUID`_) to identify devices on the bus similar to how PCI uses Domain:Bus:Function.
The UUID associated with a Linux kernel network device can be determined
by looking at the sysfs information. To find the UUID for eth1 and
store it in a shell variable:
.. code-block:: console
DEV_UUID=$(basename $(readlink /sys/class/net/eth1/device))
.. _`UUID`: https://en.wikipedia.org/wiki/Universally_unique_identifier
There are several possible ways to assign the uio device driver for a device.
The easiest way (but only on 4.18 or later)
is to use the `driverctl Device Driver control utility`_ to override
the normal kernel device.
.. code-block:: console
driverctl -b vmbus set-override $DEV_UUID uio_hv_generic
.. _`driverctl Device Driver control utility`: https://gitlab.com/driverctl/driverctl
Any settings done with driverctl are by default persistent and will be reapplied
on reboot.
On older kernels, the same effect can be had by manual sysfs bind and unbind
operations:
.. code-block:: console
NET_UUID="f8615163-df3e-46c5-913f-f2d2f965ed0e"
modprobe uio_hv_generic
echo $NET_UUID > /sys/bus/vmbus/drivers/uio_hv_generic/new_id
echo $DEV_UUID > /sys/bus/vmbus/drivers/hv_netvsc/unbind
echo $DEV_UUID > /sys/bus/vmbus/drivers/uio_hv_generic/bind
.. Note::
The dpdk-devbind.py script can not be used since it only handles PCI devices.
Prerequisites
-------------
The following prerequisites apply:
* Linux kernel support for UIO on vmbus is done with the uio_hv_generic driver.
Full support of multiple queues requires the 4.17 kernel. It is possible
to use the netvsc PMD with 4.16 kernel but it is limited to a single queue.
Netvsc PMD arguments
--------------------
The user can specify below argument in devargs.
#. ``latency``:
A netvsc device uses a mailbox page to indicate to the host that there
is something in the transmit queue. The host scans this page at a
periodic interval. This parameter allows adjusting the value that
is used by the host. Smaller values improve transmit latency, and larger
values save CPU cycles. This parameter is in microseconds.
If the value is too large or too small it will be
ignored by the host. (Default: 50)
#. ``rx_copybreak``:
The rx_copybreak sets the threshold where the driver uses an external
mbuf to avoid having to copy data. Setting 0 for copybreak will cause
driver to always create an external mbuf. Setting a value greater than
the MTU would prevent it from ever making an external mbuf and always
copy. The default value is 256 (bytes).
#. ``tx_copybreak``:
The tx_copybreak sets the threshold where the driver aggregates
multiple small packets into one request. If tx_copybreak is 0 then
each packet goes as a VMBus request (no copying). If tx_copybreak is
set larger than the MTU, then all packets smaller than the chunk size
of the VMBus send buffer will be copied; larger packets always have to
go as a single direct request. The default value is 512 (bytes).
#. ``rx_extmbuf_enable``:
The rx_extmbuf_enable is used to control if netvsc should use external
mbuf for receiving packets. The default value is 0. (netvsc doesn't use
external mbuf, it always allocates mbuf and copy received data to mbuf)
A non-zero value tells netvsc to attach external buffers to mbuf on
receiving packets, thus avoid copying memory. Use of external buffers
requires the application is able to read data from external mbuf.