blob: 0a576166e9dd90d9ceda79f2bf154299500f1a2a [file] [log] [blame]
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +02001.. SPDX-License-Identifier: GPL-2.0
Peter P Waskiewicz Jra093bf02007-06-28 20:45:47 -07002
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +02003===========================================
4HOWTO for multiqueue network device support
5===========================================
Peter P Waskiewicz Jra093bf02007-06-28 20:45:47 -07006
7Section 1: Base driver requirements for implementing multiqueue support
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +02008=======================================================================
Peter P Waskiewicz Jra093bf02007-06-28 20:45:47 -07009
10Intro: Kernel support for multiqueue devices
11---------------------------------------------------------
12
David S. Millerb19fa1f2008-07-08 23:14:24 -070013Kernel support for multiqueue devices is always present.
Peter P Waskiewicz Jra093bf02007-06-28 20:45:47 -070014
Peter P Waskiewicz Jra093bf02007-06-28 20:45:47 -070015Base drivers are required to use the new alloc_etherdev_mq() or
16alloc_netdev_mq() functions to allocate the subqueues for the device. The
17underlying kernel API will take care of the allocation and deallocation of
18the subqueue memory, as well as netdev configuration of where the queues
19exist in memory.
20
21The base driver will also need to manage the queues as it does the global
22netdev->queue_lock today. Therefore base drivers should use the
23netif_{start|stop|wake}_subqueue() functions to manage each queue while the
24device is still operational. netdev->queue_lock is still used when the device
25comes online or when it's completely shut down (unregister_netdev(), etc.).
26
Alexander Duyck92651942008-09-12 16:29:34 -070027
28Section 2: Qdisc support for multiqueue devices
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020029===============================================
Alexander Duyck92651942008-09-12 16:29:34 -070030
Alexander Duyckf07d1502008-09-12 17:57:23 -070031Currently two qdiscs are optimized for multiqueue devices. The first is the
32default pfifo_fast qdisc. This qdisc supports one qdisc per hardware queue.
33A new round-robin qdisc, sch_multiq also supports multiple hardware queues. The
Alexander Duyck92651942008-09-12 16:29:34 -070034qdisc is responsible for classifying the skb's and then directing the skb's to
35bands and queues based on the value in skb->queue_mapping. Use this field in
36the base driver to determine which queue to send the skb to.
37
Alexander Duyckf07d1502008-09-12 17:57:23 -070038sch_multiq has been added for hardware that wishes to avoid head-of-line
39blocking. It will cycle though the bands and verify that the hardware queue
Alexander Duyck92651942008-09-12 16:29:34 -070040associated with the band is not stopped prior to dequeuing a packet.
41
42On qdisc load, the number of bands is based on the number of queues on the
43hardware. Once the association is made, any skb with skb->queue_mapping set,
44will be queued to the band associated with the hardware queue.
45
46
47Section 3: Brief howto using MULTIQ for multiqueue devices
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020048==========================================================
Alexander Duyck92651942008-09-12 16:29:34 -070049
50The userspace command 'tc,' part of the iproute2 package, is used to configure
51qdiscs. To add the MULTIQ qdisc to your network device, assuming the device
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020052is called eth0, run the following command::
Alexander Duyck92651942008-09-12 16:29:34 -070053
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020054 # tc qdisc add dev eth0 root handle 1: multiq
Alexander Duyck92651942008-09-12 16:29:34 -070055
56The qdisc will allocate the number of bands to equal the number of queues that
57the device reports, and bring the qdisc online. Assuming eth0 has 4 Tx
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020058queues, the band mapping would look like::
Alexander Duyck92651942008-09-12 16:29:34 -070059
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020060 band 0 => queue 0
61 band 1 => queue 1
62 band 2 => queue 2
63 band 3 => queue 3
Alexander Duyck92651942008-09-12 16:29:34 -070064
Alexander Duyckf07d1502008-09-12 17:57:23 -070065Traffic will begin flowing through each queue based on either the simple_tx_hash
66function or based on netdev->select_queue() if you have it defined.
Alexander Duyck92651942008-09-12 16:29:34 -070067
Alexander Duyckca9b0e22008-09-12 16:30:20 -070068The behavior of tc filters remains the same. However a new tc action,
69skbedit, has been added. Assuming you wanted to route all traffic to a
Alexander Duyck67333bb2008-09-12 17:56:50 -070070specific host, for example 192.168.0.3, through a specific queue you could use
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020071this action and establish a filter such as::
Alexander Duyckca9b0e22008-09-12 16:30:20 -070072
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020073 tc filter add dev eth0 parent 1: protocol ip prio 1 u32 \
74 match ip dst 192.168.0.3 \
75 action skbedit queue_mapping 3
Alexander Duyck92651942008-09-12 16:29:34 -070076
Mauro Carvalho Chehabe98aa682020-04-30 18:04:01 +020077:Author: Alexander Duyck <alexander.h.duyck@intel.com>
78:Original Author: Peter P. Waskiewicz Jr. <peter.p.waskiewicz.jr@intel.com>