MQPRIO(8) Linux MQPRIO(8)
NAME
MQPRIO - Multiqueue Priority Qdisc (Offloaded Hardware QOS)
SYNOPSIS
tc qdisc ... dev dev ( parent classid | root) [ handle major: ] mqprio
[ num_tc tcs ] [ map P0 P1 P2... ] [ queues count1@offset1
count2@offset2 ... ]
[ hw 1|0 ] [ mode dcb|channel ] [ shaper dcb|bw_rlimit ]
[ min_rate min_rate1 min_rate2 ... ] [ max_rate max_rate1
max_rate2 ... ]
[ fp FP0 FP1 FP2 ... ]
DESCRIPTION
The MQPRIO qdisc is a simple queuing discipline that allows mapping
traffic flows to hardware queue ranges using priorities and a
configurable priority to traffic class mapping. A traffic class in this
context is a set of contiguous qdisc classes which map 1:1 to a set of
hardware exposed queues.
By default the qdisc allocates a pfifo qdisc (packet limited first in,
first out queue) per TX queue exposed by the lower layer device. Other
queuing disciplines may be added subsequently. Packets are enqueued
using the map parameter and hashed across the indicated queues in the
offset and count. By default these parameters are configured by the
hardware driver to match the hardware QOS structures.
Channel mode supports full offload of the mqprio options, the traffic
classes, the queue configurations and QOS attributes to the hardware.
Enabled hardware can provide hardware QOS with the ability to steer
traffic flows to designated traffic classes provided by this qdisc.
Hardware based QOS is configured using the shaper parameter. bw_rlimit
with minimum and maximum bandwidth rates can be used for setting
transmission rates on each traffic class. Also further qdiscs may be
added to the classes of MQPRIO to create more complex configurations.
ALGORITHM
On creation with 'tc qdisc add', eight traffic classes are created
mapping priorities 0..7 to traffic classes 0..7 and priorities greater
than 7 to traffic class 0. This requires base driver support and the
creation will fail on devices that do not support hardware QOS schemes.
These defaults can be overridden using the qdisc parameters. Providing
the 'hw 0' flag allows software to run without hardware coordination.
If hardware coordination is being used and arguments are provided that
the hardware can not support then an error is returned. For many users
hardware defaults should work reasonably well.
As one specific example numerous Ethernet cards support the 802.1Q link
strict priority transmission selection algorithm (TSA). MQPRIO enabled
hardware in conjunction with the classification methods below can
provide hardware offloaded support for this TSA.
CLASSIFICATION
Multiple methods are available to set the SKB priority which MQPRIO
uses to select which traffic class to enqueue the packet.
From user space
A process with sufficient privileges can encode the destination
class directly with SO_PRIORITY, see socket(7).
with iptables/nftables
An iptables/nftables rule can be created to match traffic flows
and set the priority. iptables(8)
with net_prio cgroups
The net_prio cgroup can be used to set the priority of all
sockets belong to an application. See kernel and cgroup
documentation for details.
QDISC PARAMETERS
num_tc Number of traffic classes to use. Up to 16 classes supported.
You cannot have more classes than queues
map The priority to traffic class map. Maps priorities 0..15 to a
specified traffic class.
queues Provide count and offset of queue range for each traffic class.
In the format, count@offset. Queue ranges for each traffic
classes cannot overlap and must be a contiguous range of queues.
hw Set to 1 to support hardware offload. Set to 0 to configure user
specified values in software only. The default value of this
parameter is 1
mode Set to channel for full use of the mqprio options. Use dcb to
offload only TC values and use hardware QOS defaults. Supported
with 'hw' set to 1 only.
shaper Use bw_rlimit to set bandwidth rate limits for a traffic class.
Use dcb for hardware QOS defaults. Supported with 'hw' set to 1
only.
min_rate
Minimum value of bandwidth rate limit for a traffic class.
Supported only when the 'shaper' argument is set to 'bw_rlimit'.
max_rate
Maximum value of bandwidth rate limit for a traffic class.
Supported only when the 'shaper' argument is set to 'bw_rlimit'.
fp Selects whether traffic classes are express (deliver packets via
the eMAC) or preemptible (deliver packets via the pMAC),
according to IEEE 802.1Q-2018 clause 6.7.2 Frame preemption.
Takes the form of an array (one element per traffic class) with
values being 'E' (for express) or 'P' (for preemptible).
Multiple priorities which map to the same traffic class, as well
as multiple TXQs which map to the same traffic class, must have
the same FP attributes. To interpret the FP as an attribute per
priority, the 'map' argument can be used for translation. To
interpret FP as an attribute per TXQ, the 'queues' argument can
be used for translation.
Traffic classes are express by default. The argument is
supported only with 'hw' set to 1. Preemptible traffic classes
are accepted only if the device has a MAC Merge layer
configurable through ethtool(8).
SEE ALSO
ethtool(8)
EXAMPLE
The following example shows how to attach priorities to 4 traffic
classes ("num_tc 4"), and then how to pair these traffic classes with 4
hardware queues with mqprio, with hardware coordination ("hw 1", or
does not specified, because 1 is the default value). Traffic class 0
(tc0) is mapped to hardware queue 0 (q0), tc1 is mapped to q1, tc2 is
mapped to q2, and tc3 is mapped q3.
# tc qdisc add dev eth0 root mqprio num_tc 4 map 0 0 0 0 1 1 1 1 2 2 2 2 3 3 3 3 queues 1@0 1@1 1@2 1@3 hw 1
The next example shows how to attach priorities to 3 traffic classes
("num_tc 3"), and how to pair these traffic classes with 4 queues,
without hardware coordination ("hw 0"). Traffic class 0 (tc0) is
mapped to hardware queue 0 (q0), tc1 is mapped to q1, tc2 and is mapped
to q2 and q3, where the queue selection between these two queues is
somewhat randomly decided.
# tc qdisc add dev eth0 root mqprio num_tc 3 map 0 0 0 0 1 1 1 1 2 2 2 2 2 2 2 2 queues 1@0 1@1 2@2 hw 0
In both cases from above the priority values from 0 to 3 (prio0-3) are
mapped to tc0, prio4-7 are mapped to tc1, and the prio8-11 are mapped
to tc2 ("map" attribute). The last four priority values (prio12-15) are
mapped in different ways in the two examples. They are mapped to tc3
in the first example and mapped to tc2 in the second example. The
values of these two examples are the following:
┌────┬────┬───────┐ ┌────┬────┬────────┐
│Prio│ tc │ queue │ │Prio│ tc │ queue │
├────┼────┼───────┤ ├────┼────┼────────┤
│ 0 │ 0 │ 0 │ │ 0 │ 0 │ 0 │
│ 1 │ 0 │ 0 │ │ 1 │ 0 │ 0 │
│ 2 │ 0 │ 0 │ │ 2 │ 0 │ 0 │
│ 3 │ 0 │ 0 │ │ 3 │ 0 │ 0 │
│ 4 │ 1 │ 1 │ │ 4 │ 1 │ 1 │
│ 5 │ 1 │ 1 │ │ 5 │ 1 │ 1 │
│ 6 │ 1 │ 1 │ │ 6 │ 1 │ 1 │
│ 7 │ 1 │ 1 │ │ 7 │ 1 │ 1 │
│ 8 │ 2 │ 2 │ │ 8 │ 2 │ 2 or 3 │
│ 9 │ 2 │ 2 │ │ 9 │ 2 │ 2 or 3 │
│ 10 │ 2 │ 2 │ │ 10 │ 2 │ 2 or 3 │
│ 11 │ 2 │ 2 │ │ 11 │ 2 │ 2 or 3 │
│ 12 │ 3 │ 3 │ │ 12 │ 2 │ 2 or 3 │
│ 13 │ 3 │ 3 │ │ 13 │ 2 │ 2 or 3 │
│ 14 │ 3 │ 3 │ │ 14 │ 2 │ 2 or 3 │
│ 15 │ 3 │ 3 │ │ 15 │ 2 │ 2 or 3 │
└────┴────┴───────┘ └────┴────┴────────┘
example1 example2
Another example of queue mapping is the following. There are 5 traffic
classes, and there are 8 hardware queues.
# tc qdisc add dev eth0 root mqprio num_tc 5 map 0 0 0 1 1 1 1 2 2 3 3 4 4 4 4 4 queues 1@0 2@1 1@3 1@4 3@5
The value mapping is the following for this example:
┌───────┐
tc0────┤Queue 0│◄────1@0
├───────┤
┌─┤Queue 1│◄────2@1
tc1──┤ ├───────┤
└─┤Queue 2│
├───────┤
tc2────┤Queue 3│◄────1@3
├───────┤
tc3────┤Queue 4│◄────1@4
├───────┤
┌─┤Queue 5│◄────3@5
│ ├───────┤
tc4──┼─┤Queue 6│
│ ├───────┤
└─┤Queue 7│
└───────┘
AUTHORS
John Fastabend, <john.r.fastabend@intel.com>
iproute2 24 Sept 2013 MQPRIO(8)
MQPRIO \- Multiqueue Priority Qdisc (Offloaded Hardware QOS)