xref: /f-stack/dpdk/doc/guides/nics/mvpp2.rst (revision 2d9fd380)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright(c) 2017 Marvell International Ltd.
3    Copyright(c) 2017 Semihalf.
4
5.. _mvpp2_poll_mode_driver:
6
7MVPP2 Poll Mode Driver
8======================
9
10The MVPP2 PMD (**librte_net_mvpp2**) provides poll mode driver support
11for the Marvell PPv2 (Packet Processor v2) 1/10 Gbps adapter.
12
13Detailed information about SoCs that use PPv2 can be obtained here:
14
15* https://www.marvell.com/embedded-processors/armada-70xx/
16* https://www.marvell.com/embedded-processors/armada-80xx/
17
18
19Features
20--------
21
22Features of the MVPP2 PMD are:
23
24- Speed capabilities
25- Link status
26- Tx Queue start/stop
27- MTU update
28- Jumbo frame
29- Promiscuous mode
30- Allmulticast mode
31- Unicast MAC filter
32- Multicast MAC filter
33- RSS hash
34- VLAN filter
35- CRC offload
36- L3 checksum offload
37- L4 checksum offload
38- Packet type parsing
39- Basic stats
40- :ref:`Extended stats <extstats>`
41- RX flow control
42- Scattered TX frames
43- :ref:`QoS <qossupport>`
44- :ref:`Flow API <flowapi>`
45- :ref:`Traffic metering and policing <mtrapi>`
46- :ref:`Traffic Management API <tmapi>`
47
48Limitations
49-----------
50
51- Number of lcores is limited to 9 by MUSDK internal design. If more lcores
52  need to be allocated, locking will have to be considered. Number of available
53  lcores can be changed via ``MRVL_MUSDK_HIFS_RESERVED`` define in
54  ``mrvl_ethdev.c`` source file.
55
56- Flushing vlans added for filtering is not possible due to MUSDK missing
57  functionality. Current workaround is to reset board so that PPv2 has a
58  chance to start in a sane state.
59
60- MUSDK architecture does not support changing configuration in run time.
61  All necessary configurations should be done before first dev_start().
62
63- RX queue start/stop is not supported.
64
65- Current implementation does not support replacement of buffers in the HW buffer pool
66  at run time, so it is responsibility of the application to ensure that MTU does not exceed the configured buffer size.
67
68- Configuring TX flow control currently is not supported.
69
70- In current implementation, mechanism for acknowledging transmitted packets (``tx_done_cleanup``) is not supported.
71
72- Running more than one DPDK-MUSDK application simultaneously is not supported.
73
74
75Prerequisites
76-------------
77
78- Custom Linux Kernel sources
79
80  .. code-block:: console
81
82     git clone https://github.com/MarvellEmbeddedProcessors/linux-marvell.git -b linux-4.4.120-armada-18.09
83
84- Out of tree `mvpp2x_sysfs` kernel module sources
85
86  .. code-block:: console
87
88     git clone https://github.com/MarvellEmbeddedProcessors/mvpp2x-marvell.git -b mvpp2x-armada-18.09
89
90- MUSDK (Marvell User-Space SDK) sources
91
92  .. code-block:: console
93
94     git clone https://github.com/MarvellEmbeddedProcessors/musdk-marvell.git -b musdk-armada-18.09
95
96  MUSDK is a light-weight library that provides direct access to Marvell's
97  PPv2 (Packet Processor v2). Alternatively prebuilt MUSDK library can be
98  requested from `Marvell Extranet <https://extranet.marvell.com>`_. Once
99  approval has been granted, library can be found by typing ``musdk`` in
100  the search box.
101
102  To get better understanding of the library one can consult documentation
103  available in the ``doc`` top level directory of the MUSDK sources.
104
105- DPDK environment
106
107  Follow the DPDK :ref:`Getting Started Guide for Linux <linux_gsg>` to setup
108  DPDK environment.
109
110
111Building DPDK
112-------------
113
114Driver needs precompiled MUSDK library during compilation.
115
116.. code-block:: console
117
118   export CROSS_COMPILE=<toolchain>/bin/aarch64-linux-gnu-
119   ./bootstrap
120   ./configure --host=aarch64-linux-gnu
121   make install
122
123MUSDK will be installed to `usr/local` under current directory.
124For the detailed build instructions please consult ``doc/musdk_get_started.txt``.
125
126The path to the MUSDK installation directory needs to set in meson, shown in the
127command below.
128
129For additional instructions regarding DPDK cross compilation please refer to :doc:`Cross compile DPDK for ARM64 <../linux_gsg/cross_build_dpdk_for_arm64>`.
130
131.. code-block:: console
132
133   meson -Dlib_musdk_dir=/path/to/musdk build ninja -C build
134
135
136Usage Example
137-------------
138
139MVPP2 PMD requires extra out of tree kernel modules to function properly.
140`musdk_cma` sources are part of the MUSDK. Please consult
141``doc/musdk_get_started.txt`` for the detailed build instructions.
142For `mvpp2x_sysfs` please consult ``Documentation/pp22_sysfs.txt`` for the
143detailed build instructions.
144
145.. code-block:: console
146
147   insmod musdk_cma.ko
148   insmod mvpp2x_sysfs.ko
149
150Additionally interfaces used by DPDK application need to be put up:
151
152.. code-block:: console
153
154   ip link set eth0 up
155   ip link set eth2 up
156
157In order to run testpmd example application following command can be used:
158
159.. code-block:: console
160
161   ./dpdk-testpmd --vdev=eth_mvpp2,iface=eth0,iface=eth2 -c 7 -- \
162     --burst=128 --txd=2048 --rxd=1024 --rxq=2 --txq=2  --nb-cores=2 \
163     -i -a --rss-udp
164
165.. _extstats:
166
167Extended stats
168--------------
169
170MVPP2 PMD supports the following extended statistics:
171
172	- ``rx_bytes``:	number of RX bytes
173	- ``rx_packets``: number of RX packets
174	- ``rx_unicast_packets``: number of RX unicast packets
175	- ``rx_errors``: number of RX MAC errors
176	- ``rx_fullq_dropped``: number of RX packets dropped due to full RX queue
177	- ``rx_bm_dropped``: number of RX packets dropped due to no available buffers in the HW pool
178	- ``rx_early_dropped``: number of RX packets that were early dropped
179	- ``rx_fifo_dropped``: number of RX packets dropped due to RX fifo overrun
180	- ``rx_cls_dropped``: number of RX packets dropped by classifier
181	- ``tx_bytes``: number of TX bytes
182	- ``tx_packets``: number of TX packets
183	- ``tx_unicast_packets``: number of TX unicast packets
184	- ``tx_errors``: number of TX MAC errors
185
186
187.. _qossupport:
188
189QoS Configuration
190-----------------
191
192QoS configuration is done through external configuration file. Path to the
193file must be given as `cfg` in driver's vdev parameter list.
194
195Configuration syntax
196~~~~~~~~~~~~~~~~~~~~
197
198.. code-block:: console
199
200   [policer <policer_id>]
201   token_unit = <token_unit>
202   color = <color_mode>
203   cir = <cir>
204   ebs = <ebs>
205   cbs = <cbs>
206
207   [port <portnum> default]
208   default_tc = <default_tc>
209   mapping_priority = <mapping_priority>
210
211   rate_limit_enable = <rate_limit_enable>
212   rate_limit = <rate_limit>
213   burst_size = <burst_size>
214
215   default_policer = <policer_id>
216
217   [port <portnum> tc <traffic_class>]
218   rxq = <rx_queue_list>
219   pcp = <pcp_list>
220   dscp = <dscp_list>
221   default_color = <default_color>
222
223   [port <portnum> tc <traffic_class>]
224   rxq = <rx_queue_list>
225   pcp = <pcp_list>
226   dscp = <dscp_list>
227
228   [port <portnum> txq <txqnum>]
229   sched_mode = <sched_mode>
230   wrr_weight = <wrr_weight>
231
232   rate_limit_enable = <rate_limit_enable>
233   rate_limit = <rate_limit>
234   burst_size = <burst_size>
235
236Where:
237
238- ``<portnum>``: DPDK Port number (0..n).
239
240- ``<default_tc>``: Default traffic class (e.g. 0)
241
242- ``<mapping_priority>``: QoS priority for mapping (`ip`, `vlan`, `ip/vlan` or `vlan/ip`).
243
244- ``<traffic_class>``: Traffic Class to be configured.
245
246- ``<rx_queue_list>``: List of DPDK RX queues (e.g. 0 1 3-4)
247
248- ``<pcp_list>``: List of PCP values to handle in particular TC (e.g. 0 1 3-4 7).
249
250- ``<dscp_list>``: List of DSCP values to handle in particular TC (e.g. 0-12 32-48 63).
251
252- ``<default_policer>``: Id of the policer configuration section to be used as default.
253
254- ``<policer_id>``: Id of the policer configuration section (0..31).
255
256- ``<token_unit>``: Policer token unit (`bytes` or `packets`).
257
258- ``<color_mode>``: Policer color mode (`aware` or `blind`).
259
260- ``<cir>``: Committed information rate in unit of kilo bits per second (data rate) or packets per second.
261
262- ``<cbs>``: Committed burst size in unit of kilo bytes or number of packets.
263
264- ``<ebs>``: Excess burst size in unit of kilo bytes or number of packets.
265
266- ``<default_color>``: Default color for specific tc.
267
268- ``<rate_limit_enable>``: Enables per port or per txq rate limiting (`0`/`1` to disable/enable).
269
270- ``<rate_limit>``: Committed information rate, in kilo bits per second.
271
272- ``<burst_size>``: Committed burst size, in kilo bytes.
273
274- ``<sched_mode>``: Egress scheduler mode (`wrr` or `sp`).
275
276- ``<wrr_weight>``: Txq weight.
277
278Setting PCP/DSCP values for the default TC is not required. All PCP/DSCP
279values not assigned explicitly to particular TC will be handled by the
280default TC.
281
282Configuration file example
283^^^^^^^^^^^^^^^^^^^^^^^^^^
284
285.. code-block:: console
286
287   [policer 0]
288   token_unit = bytes
289   color = blind
290   cir = 100000
291   ebs = 64
292   cbs = 64
293
294   [port 0 default]
295   default_tc = 0
296   mapping_priority = ip
297
298   rate_limit_enable = 1
299   rate_limit = 1000
300   burst_size = 2000
301
302   [port 0 tc 0]
303   rxq = 0 1
304
305   [port 0 txq 0]
306   sched_mode = wrr
307   wrr_weight = 10
308
309   [port 0 txq 1]
310   sched_mode = wrr
311   wrr_weight = 100
312
313   [port 0 txq 2]
314   sched_mode = sp
315
316   [port 0 tc 1]
317   rxq = 2
318   pcp = 5 6 7
319   dscp = 26-38
320
321   [port 1 default]
322   default_tc = 0
323   mapping_priority = vlan/ip
324
325   default_policer = 0
326
327   [port 1 tc 0]
328   rxq = 0
329   dscp = 10
330
331   [port 1 tc 1]
332   rxq = 1
333   dscp = 11-20
334
335   [port 1 tc 2]
336   rxq = 2
337   dscp = 30
338
339   [port 1 txq 0]
340   rate_limit_enable = 1
341   rate_limit = 10000
342   burst_size = 2000
343
344Usage example
345^^^^^^^^^^^^^
346
347.. code-block:: console
348
349   ./dpdk-testpmd --vdev=eth_mvpp2,iface=eth0,iface=eth2,cfg=/home/user/mrvl.conf \
350     -c 7 -- -i -a --disable-hw-vlan-strip --rxq=3 --txq=3
351
352.. _flowapi:
353
354Flow API
355--------
356
357PPv2 offers packet classification capabilities via classifier engine which
358can be configured via generic flow API offered by DPDK.
359
360The :ref:`flow_isolated_mode` is supported.
361
362For an additional description please refer to DPDK :doc:`../prog_guide/rte_flow`.
363
364Supported flow actions
365~~~~~~~~~~~~~~~~~~~~~~
366
367Following flow action items are supported by the driver:
368
369* DROP
370* QUEUE
371
372Supported flow items
373~~~~~~~~~~~~~~~~~~~~
374
375Following flow items and their respective fields are supported by the driver:
376
377* ETH
378
379  * source MAC
380  * destination MAC
381  * ethertype
382
383* VLAN
384
385  * PCP
386  * VID
387
388* IPV4
389
390  * DSCP
391  * protocol
392  * source address
393  * destination address
394
395* IPV6
396
397  * flow label
398  * next header
399  * source address
400  * destination address
401
402* UDP
403
404  * source port
405  * destination port
406
407* TCP
408
409  * source port
410  * destination port
411
412Classifier match engine
413~~~~~~~~~~~~~~~~~~~~~~~
414
415Classifier has an internal match engine which can be configured to
416operate in either exact or maskable mode.
417
418Mode is selected upon creation of the first unique flow rule as follows:
419
420* maskable, if key size is up to 8 bytes.
421* exact, otherwise, i.e for keys bigger than 8 bytes.
422
423Where the key size equals the number of bytes of all fields specified
424in the flow items.
425
426.. table:: Examples of key size calculation
427
428   +----------------------------------------------------------------------------+-------------------+-------------+
429   | Flow pattern                                                               | Key size in bytes | Used engine |
430   +============================================================================+===================+=============+
431   | ETH (destination MAC) / VLAN (VID)                                         | 6 + 2 = 8         | Maskable    |
432   +----------------------------------------------------------------------------+-------------------+-------------+
433   | VLAN (VID) / IPV4 (source address)                                         | 2 + 4 = 6         | Maskable    |
434   +----------------------------------------------------------------------------+-------------------+-------------+
435   | TCP (source port, destination port)                                        | 2 + 2 = 4         | Maskable    |
436   +----------------------------------------------------------------------------+-------------------+-------------+
437   | VLAN (priority) / IPV4 (source address)                                    | 1 + 4 = 5         | Maskable    |
438   +----------------------------------------------------------------------------+-------------------+-------------+
439   | IPV4 (destination address) / UDP (source port, destination port)           | 6 + 2 + 2 = 10    | Exact       |
440   +----------------------------------------------------------------------------+-------------------+-------------+
441   | VLAN (VID) / IPV6 (flow label, destination address)                        | 2 + 3 + 16 = 21   | Exact       |
442   +----------------------------------------------------------------------------+-------------------+-------------+
443   | IPV4 (DSCP, source address, destination address)                           | 1 + 4 + 4 = 9     | Exact       |
444   +----------------------------------------------------------------------------+-------------------+-------------+
445   | IPV6 (flow label, source address, destination address)                     | 3 + 16 + 16 = 35  | Exact       |
446   +----------------------------------------------------------------------------+-------------------+-------------+
447
448From the user perspective maskable mode means that masks specified
449via flow rules are respected. In case of exact match mode, masks
450which do not provide exact matching (all bits masked) are ignored.
451
452If the flow matches more than one classifier rule the first
453(with the lowest index) matched takes precedence.
454
455Flow rules usage example
456~~~~~~~~~~~~~~~~~~~~~~~~
457
458Before proceeding run testpmd user application:
459
460.. code-block:: console
461
462   ./dpdk-testpmd --vdev=eth_mvpp2,iface=eth0,iface=eth2 -c 3 -- -i --p 3 -a --disable-hw-vlan-strip
463
464Example #1
465^^^^^^^^^^
466
467.. code-block:: console
468
469   testpmd> flow create 0 ingress pattern eth src is 10:11:12:13:14:15 / end actions drop / end
470
471In this case key size is 6 bytes thus maskable type is selected. Testpmd
472will set mask to ff:ff:ff:ff:ff:ff i.e traffic explicitly matching
473above rule will be dropped.
474
475Example #2
476^^^^^^^^^^
477
478.. code-block:: console
479
480   testpmd> flow create 0 ingress pattern ipv4 src spec 10.10.10.0 src mask 255.255.255.0 / tcp src spec 0x10 src mask 0x10 / end action drop / end
481
482In this case key size is 8 bytes thus maskable type is selected.
483Flows which have IPv4 source addresses ranging from 10.10.10.0 to 10.10.10.255
484and tcp source port set to 16 will be dropped.
485
486Example #3
487^^^^^^^^^^
488
489.. code-block:: console
490
491   testpmd> flow create 0 ingress pattern vlan vid spec 0x10 vid mask 0x10 / ipv4 src spec 10.10.1.1 src mask 255.255.0.0 dst spec 11.11.11.1 dst mask 255.255.255.0 / end actions drop / end
492
493In this case key size is 10 bytes thus exact type is selected.
494Even though each item has partial mask set, masks will be ignored.
495As a result only flows with VID set to 16 and IPv4 source and destination
496addresses set to 10.10.1.1 and 11.11.11.1 respectively will be dropped.
497
498Limitations
499~~~~~~~~~~~
500
501Following limitations need to be taken into account while creating flow rules:
502
503* For IPv4 exact match type the key size must be up to 12 bytes.
504* For IPv6 exact match type the key size must be up to 36 bytes.
505* Following fields cannot be partially masked (all masks are treated as
506  if they were exact):
507
508  * ETH: ethertype
509  * VLAN: PCP, VID
510  * IPv4: protocol
511  * IPv6: next header
512  * TCP/UDP: source port, destination port
513
514* Only one classifier table can be created thus all rules in the table
515  have to match table format. Table format is set during creation of
516  the first unique flow rule.
517* Up to 5 fields can be specified per flow rule.
518* Up to 20 flow rules can be added.
519
520For additional information about classifier please consult
521``doc/musdk_cls_user_guide.txt``.
522
523.. _mtrapi:
524
525Traffic metering and policing
526-----------------------------
527
528MVPP2 PMD supports DPDK traffic metering and policing that allows the following:
529
5301. Meter ingress traffic.
5312. Do policing.
5323. Gather statistics.
533
534For an additional description please refer to DPDK :doc:`Traffic Metering and Policing API <../prog_guide/traffic_metering_and_policing>`.
535
536The policer objects defined by this feature can work with the default policer defined via config file as described in :ref:`QoS Support <qossupport>`.
537
538Limitations
539~~~~~~~~~~~
540
541The following capabilities are not supported:
542
543- MTR object meter DSCP table update
544- MTR object policer action update
545- MTR object enabled statistics
546
547Usage example
548~~~~~~~~~~~~~
549
5501. Run testpmd user app:
551
552   .. code-block:: console
553
554		./dpdk-testpmd --vdev=eth_mvpp2,iface=eth0,iface=eth2 -c 6 -- -i -p 3 -a --txd 1024 --rxd 1024
555
5562. Create meter profile:
557
558   .. code-block:: console
559
560		testpmd> add port meter profile 0 0 srtcm_rfc2697 2000 256 256
561
5623. Create meter:
563
564   .. code-block:: console
565
566		testpmd> create port meter 0 0 0 yes d d d 0 1 0
567
5684. Create flow rule witch meter attached:
569
570   .. code-block:: console
571
572		testpmd> flow create 0 ingress pattern ipv4 src is 10.10.10.1 / end actions meter mtr_id 0 / end
573
574For a detailed usage description please refer to "Traffic Metering and Policing" section in DPDK :doc:`Testpmd Runtime Functions <../testpmd_app_ug/testpmd_funcs>`.
575
576
577
578.. _tmapi:
579
580Traffic Management API
581----------------------
582
583MVPP2 PMD supports generic DPDK Traffic Management API which allows to
584configure the following features:
585
5861. Hierarchical scheduling
5872. Traffic shaping
5883. Congestion management
5894. Packet marking
590
591Internally TM is represented by a hierarchy (tree) of nodes.
592Node which has a parent is called a leaf whereas node without
593parent is called a non-leaf (root).
594MVPP2 PMD supports two level hierarchy where level 0 represents ports and level 1 represents tx queues of a given port.
595
596.. figure:: img/mvpp2_tm.*
597
598Nodes hold following types of settings:
599
600- for egress scheduler configuration: weight
601- for egress rate limiter: private shaper
602- bitmask indicating which statistics counters will be read
603
604Hierarchy is always constructed from the top, i.e first a root node is added
605then some number of leaf nodes. Number of leaf nodes cannot exceed number
606of configured tx queues.
607
608After hierarchy is complete it can be committed.
609
610
611For an additional description please refer to DPDK :doc:`Traffic Management API <../prog_guide/traffic_management>`.
612
613Limitations
614~~~~~~~~~~~
615
616The following capabilities are not supported:
617
618- Traffic manager WRED profile and WRED context
619- Traffic manager shared shaper update
620- Traffic manager packet marking
621- Maximum number of levels in hierarchy is 2
622- Currently dynamic change of a hierarchy is not supported
623
624Usage example
625~~~~~~~~~~~~~
626
627For a detailed usage description please refer to "Traffic Management" section in DPDK :doc:`Testpmd Runtime Functions <../testpmd_app_ug/testpmd_funcs>`.
628
6291. Run testpmd as follows:
630
631   .. code-block:: console
632
633		./dpdk-testpmd --vdev=net_mrvl,iface=eth0,iface=eth2,cfg=./qos_config -c 7 -- \
634		-i -p 3 --disable-hw-vlan-strip --rxq 3 --txq 3 --txd 1024 --rxd 1024
635
6362. Stop all ports:
637
638   .. code-block:: console
639
640		testpmd> port stop all
641
6423. Add shaper profile:
643
644   .. code-block:: console
645
646		testpmd> add port tm node shaper profile 0 0 900000 70000 0
647
648   Parameters have following meaning::
649
650		0       - Id of a port.
651		0       - Id of a new shaper profile.
652		900000  - Shaper rate in bytes/s.
653		70000   - Bucket size in bytes.
654		0       - Packet length adjustment - ignored.
655
6564. Add non-leaf node for port 0:
657
658   .. code-block:: console
659
660		testpmd> add port tm nonleaf node 0 3 -1 0 0 0 0 0 1 3 0
661
662   Parameters have following meaning::
663
664		 0  - Id of a port
665		 3  - Id of a new node.
666		-1  - Indicate that root does not have a parent.
667		 0  - Priority of the node.
668		 0  - Weight of the node.
669		 0  - Id of a level. Since this is a root 0 is passed.
670		 0  - Id of the shaper profile.
671		 0  - Number of SP priorities.
672		 3  - Enable statistics for both number of transmitted packets and bytes.
673		 0  - Number of shared shapers.
674
6755. Add leaf node for tx queue 0:
676
677   .. code-block:: console
678
679		testpmd> add port tm leaf node 0 0 3 0 30 1 -1 0 0 1 0
680
681   Parameters have following meaning::
682
683		 0  - Id of a port.
684		 0  - Id of a new node.
685		 3  - Id of the parent node.
686		 0  - Priority of a node.
687		 30 - WRR weight.
688		 1  - Id of a level. Since this is a leaf node 1 is passed.
689		-1  - Id of a shaper. -1 indicates that shaper is not attached.
690		 0  - Congestion management is not supported.
691		 0  - Congestion management is not supported.
692		 1  - Enable statistics counter for number of transmitted packets.
693		 0  - Number of shared shapers.
694
6956. Add leaf node for tx queue 1:
696
697   .. code-block:: console
698
699	testpmd> add port tm leaf node 0 1 3 0 60 1 -1 0 0 1 0
700
701   Parameters have following meaning::
702
703		 0  - Id of a port.
704		 1  - Id of a new node.
705		 3  - Id of the parent node.
706		 0  - Priority of a node.
707		 60 - WRR weight.
708		 1  - Id of a level. Since this is a leaf node 1 is passed.
709		-1  - Id of a shaper. -1 indicates that shaper is not attached.
710		 0  - Congestion management is not supported.
711		 0  - Congestion management is not supported.
712		 1  - Enable statistics counter for number of transmitted packets.
713		 0  - Number of shared shapers.
714
7157. Add leaf node for tx queue 2:
716
717   .. code-block:: console
718
719		testpmd> add port tm leaf node 0 2 3 0 99 1 -1 0 0 1 0
720
721   Parameters have following meaning::
722
723		 0  - Id of a port.
724		 2  - Id of a new node.
725		 3  - Id of the parent node.
726		 0  - Priority of a node.
727		 99 - WRR weight.
728		 1  - Id of a level. Since this is a leaf node 1 is passed.
729		-1  - Id of a shaper. -1 indicates that shaper is not attached.
730		 0  - Congestion management is not supported.
731		 0  - Congestion management is not supported.
732		 1  - Enable statistics counter for number of transmitted packets.
733		 0  - Number of shared shapers.
734
7358. Commit hierarchy:
736
737   .. code-block:: console
738
739		testpmd> port tm hierarchy commit 0 no
740
741  Parameters have following meaning::
742
743		0  - Id of a port.
744		no - Do not flush TM hierarchy if commit fails.
745
7469. Start all ports
747
748   .. code-block:: console
749
750		testpmd> port start all
751
752
753
75410. Enable forwarding
755
756   .. code-block:: console
757
758		testpmd> start
759