release_16_04.rst revision 97f17497
1DPDK Release 16.04
2==================
3
4
5New Features
6------------
7
8* **Added function to check primary process state.**
9
10  A new function ``rte_eal_primary_proc_alive()`` has been added
11  to allow the user to detect if a primary process is running.
12  Use cases for this feature include fault detection, and monitoring
13  using secondary processes.
14
15* **Enabled bulk allocation of mbufs.**
16
17  A new function ``rte_pktmbuf_alloc_bulk()`` has been added to allow the user
18  to bulk allocate mbufs.
19
20* **Added device link speed capabilities.**
21
22  The structure ``rte_eth_dev_info`` now has a ``speed_capa`` bitmap, which
23  allows the application to determine the supported speeds of each device.
24
25* **Added bitmap of link speeds to advertise.**
26
27  Added a feature to allow the definition of a set of advertised speeds for auto-negotiation,
28  explicitly disabling link auto-negotiation (single speed)
29  and full auto-negotiation.
30
31* **Added new poll-mode driver for Amazon Elastic Network Adapters (ENA).**
32
33  The driver operates for a variety of ENA adapters through feature negotiation
34  with the adapter and upgradable commands set.
35  The ENA driver handles PCI Physical and Virtual ENA functions.
36
37* **Restored vmxnet3 TX data ring.**
38
39  TX data ring has been shown to improve small packet forwarding performance
40  on the vSphere environment.
41
42* **Added vmxnet3 TX L4 checksum offload.**
43
44  Added support for TCP/UDP checksum offload to vmxnet3.
45
46* **Added vmxnet3 TSO support.**
47
48  Added support for TSO to vmxnet3.
49
50* **Added vmxnet3 support for jumbo frames.**
51
52  Added support for linking multi-segment buffers together to
53  handle Jumbo packets.
54
55* **Enabled Virtio 1.0 support.**
56
57  Enabled Virtio 1.0 support for Virtio pmd driver.
58
59* **Supported Virtio for ARM.**
60
61  Enabled Virtio support for ARMv7/v8. Tested for ARM64.
62  Virtio for ARM supports VFIO-noiommu mode only.
63  Virtio can work with other non-x86 architectures as well, like PowerPC.
64
65* **Supported Virtio offload in vhost-user.**
66
67  Added the offload and negotiation of checksum and TSO between vhost-user and
68  vanilla Linux Virtio guest.
69
70* **Added vhost-user live migration support.**
71
72* **Added vhost driver.**
73
74  Added a virtual PMD that wraps ``librte_vhost``.
75
76* **Added multicast promiscuous mode support on VF for ixgbe.**
77
78  Added multicast promiscuous mode support for the ixgbe VF driver so all VFs
79  can receive the multicast packets.
80
81  Please note if you want to use this promiscuous mode, you need both PF and VF
82  driver to support it. The reason is that this VF feature is configured in the PF.
83  If you use kernel PF driver and the dpdk VF driver, make sure the kernel PF driver supports
84  VF multicast promiscuous mode. If you use dpdk PF and  dpdk VF ensure the PF
85  driver is the same version as the VF.
86
87* **Added support for E-tag on X550.**
88
89  E-tag is defined in `802.1BR - Bridge Port Extension <http://www.ieee802.org/1/pages/802.1br.html>`_.
90
91  This feature is for the VF, but the settings are on the PF. It means
92  the CLIs should be used on the PF, but some of their effects will be shown on the VF.
93  The forwarding of E-tag packets based on GRP and E-CID_base will have an effect
94  on the PF. Theoretically, the E-tag packets can be forwarded to any pool/queue
95  but normally we'd like to forward the packets to the pools/queues belonging
96  to the VFs. And E-tag insertion and stripping will have an effect on VFs. When
97  a VF receives E-tag packets it should strip the E-tag. When the VF transmits
98  packets, it should insert the E-tag. Both actions can be offloaded.
99
100  When we want to use this E-tag support feature, the forwarding should be
101  enabled to forward the packets received by the PF to the indicated VFs. And insertion
102  and stripping should be enabled for VFs to offload the effort to hardware.
103
104  Features added:
105
106  * Support E-tag offloading of insertion and stripping.
107  * Support Forwarding E-tag packets to pools based on
108    GRP and E-CID_base.
109
110* **Added support for VxLAN and NVGRE checksum off-load on X550.**
111
112  * Added support for VxLAN and NVGRE RX/TX checksum off-load on
113    X550. RX/TX checksum off-load is provided on both inner and
114    outer IP header and TCP header.
115  * Added functions to support VxLAN port configuration. The
116    default VxLAN port number is 4789 but this can be updated
117    programmatically.
118
119* **Added support for new X550EM_a devices.**
120
121  Added support for new X550EM_a devices and their MAC types, X550EM_a and X550EM_a_vf.
122  Updated the relevant PMD to use the new devices and MAC types.
123
124* **Added x550em_x V2 device support.**
125
126  Added support for x550em_x V2 device. Only x550em_x V1 was supported before.
127  A mask for V1 and V2 is defined and used to support both.
128
129* **Supported link speed auto-negotiation on X550EM_X**
130
131  Normally the auto-negotiation is supported by firmware and software doesn't care about
132  it. But on x550em_x, firmware doesn't support auto-negotiation. As the ports of x550em_x
133  are 10GbE, if we connect the port with a peer which is 1GbE, the link will always
134  be down.
135  We added the support for auto-negotiation by software to avoid this link down issue.
136
137* **Added software-firmware sync on X550EM_a.**
138
139  Added support for software-firmware sync for resource sharing.
140  Use the PHY token, shared between software-firmware for PHY access on X550EM_a.
141
142* **Updated the i40e base driver.**
143
144  The i40e base driver was updated with changes including the
145  following:
146
147  * Use RX control AQ commands to read/write RX control registers.
148  * Add new X722 device IDs, and removed X710 one was never used.
149  * Expose registers for HASH/FD input set configuring.
150
151* **Enabled PCI extended tag for i40e.**
152
153  Enabled extended tag for i40e by checking and writing corresponding PCI config
154  space bytes, to boost the performance.
155  The legacy method of reading/writing sysfile supported by kernel module igb_uio
156  is now deprecated.
157
158* **Added i40e support for setting mac addresses.**
159
160* **Added dump of i40e registers and EEPROM.**
161
162* **Supported ether type setting of single and double VLAN for i40e**
163
164* **Added VMDQ DCB mode in i40e.**
165
166  Added support for DCB in VMDQ mode to i40e driver.
167
168* **Added i40e VEB switching support.**
169
170* **Added Flow director enhancements in i40e.**
171
172* **Added PF reset event reporting in i40e VF driver.**
173
174* **Added fm10k RX interrupt support.**
175
176* **Optimized fm10k TX.**
177
178  Optimized fm10k TX by freeing multiple mbufs at a time.
179
180* **Handled error flags in fm10k vector RX.**
181
182  Parse error flags in RX descriptor and set error bits in mbuf with vector instructions.
183
184* **Added fm10k FTAG based forwarding support.**
185
186* **Added mlx5 flow director support.**
187
188  Added flow director support (``RTE_FDIR_MODE_PERFECT`` and
189  ``RTE_FDIR_MODE_PERFECT_MAC_VLAN``).
190
191  Only available with Mellanox OFED >= 3.2.
192
193* **Added mlx5 RX VLAN stripping support.**
194
195  Added support for RX VLAN stripping.
196
197  Only available with Mellanox OFED >= 3.2.
198
199* **Added mlx5 link up/down callbacks.**
200
201  Implemented callbacks to bring link up and down.
202
203* **Added mlx5 support for operation in secondary processes.**
204
205  Implemented TX support in secondary processes (like mlx4).
206
207* **Added mlx5 RX CRC stripping configuration.**
208
209  Until now, CRC was always stripped. It can now be configured.
210
211  Only available with Mellanox OFED >= 3.2.
212
213* **Added mlx5 optional packet padding by HW.**
214
215  Added an option to make PCI bus transactions rounded to a multiple of a
216  cache line size for better alignment.
217
218  Only available with Mellanox OFED >= 3.2.
219
220* **Added mlx5 TX VLAN insertion support.**
221
222  Added support for TX VLAN insertion.
223
224  Only available with Mellanox OFED >= 3.2.
225
226* **Changed szedata2 driver type from vdev to pdev.**
227
228  Previously szedata2 device had to be added by ``--vdev`` option.
229  Now szedata2 PMD recognizes the device automatically during EAL
230  initialization.
231
232* **Added szedata2 functions for setting link up/down.**
233
234* **Added szedata2 promiscuous and allmulticast modes.**
235
236* **Added af_packet dynamic removal function.**
237
238  An af_packet device can now be detached using the API, like other PMD devices.
239
240* **Increased number of next hops for LPM IPv4 to 2^24.**
241
242  The ``next_hop`` field has been extended from 8 bits to 24 bits for IPv4.
243
244* **Added support of SNOW 3G (UEA2 and UIA2) for Intel Quick Assist devices.**
245
246  Enabled support for the SNOW 3G wireless algorithm for Intel Quick Assist devices.
247  Support for cipher-only and  hash-only is also provided
248  along with algorithm-chaining operations.
249
250* **Added SNOW3G SW PMD.**
251
252  A new Crypto PMD has been added, which provides SNOW 3G UEA2 ciphering
253  and SNOW3G UIA2 hashing.
254
255* **Added AES GCM PMD.**
256
257  Added new Crypto PMD to support AES-GCM authenticated encryption and
258  authenticated decryption in software.
259
260* **Added NULL Crypto PMD**
261
262  Added new Crypto PMD to support null crypto operations in software.
263
264* **Improved IP Pipeline Application.**
265
266  The following features have been added to ip_pipeline application;
267
268  * Added CPU utilization measurement and idle cycle rate computation.
269  * Added link identification support through existing port-mask option or by
270    specifying PCI device in every LINK section in the configuration file.
271  * Added load balancing support in passthrough pipeline.
272
273* **Added IPsec security gateway example.**
274
275  Added a new application implementing an IPsec Security Gateway.
276
277
278Resolved Issues
279---------------
280
281Drivers
282~~~~~~~
283
284* **ethdev: Fixed overflow for 100Gbps.**
285
286  100Gbps in Mbps (100000) was exceeding the 16-bit max value of ``link_speed``
287  in ``rte_eth_link``.
288
289* **ethdev: Fixed byte order consistency between fdir flow and mask.**
290
291  Fixed issue in ethdev library where the structure for setting
292  fdir's mask and flow entry was not consistent in byte ordering.
293
294* **cxgbe: Fixed crash due to incorrect size allocated for RSS table.**
295
296  Fixed a segfault that occurs when accessing part of port 0's RSS
297  table that gets overwritten by subsequent port 1's part of the RSS
298  table due to incorrect size allocated for each entry in the table.
299
300* **cxgbe: Fixed setting wrong device MTU.**
301
302  Fixed an incorrect device MTU being set due to the Ethernet header and
303  CRC lengths being added twice.
304
305* **ixgbe: Fixed zeroed VF mac address.**
306
307  Resolved an issue where the VF MAC address is zeroed out in cases where the VF
308  driver is loaded while the PF interface is down.
309  The solution is to only set it when we get an ACK from the PF.
310
311* **ixgbe: Fixed setting flow director flag twice.**
312
313  Resolved an issue where packets were being dropped when switching to perfect
314  filters mode.
315
316* **ixgbe: Set MDIO speed after MAC reset.**
317
318  The MDIO clock speed must be reconfigured after the MAC reset. The MDIO clock
319  speed becomes invalid, therefore the driver reads invalid PHY register values.
320  The driver now set the MDIO clock speed prior to initializing PHY ops and
321  again after the MAC reset.
322
323* **ixgbe: Fixed maximum number of available TX queues.**
324
325  In IXGBE, the maximum number of TX queues varies depending on the NIC operating
326  mode. This was not being updated in the device information, providing
327  an incorrect number in some cases.
328
329* **i40e: Generated MAC address for each VFs.**
330
331  It generates a MAC address for each VFs during PF host initialization,
332  and keeps the VF MAC address the same among different VF launch.
333
334* **i40e: Fixed failure of reading/writing RX control registers.**
335
336  Fixed i40e issue of failing to read/write rx control registers when
337  under stress with traffic, which might result in application launch
338  failure.
339
340* **i40e: Enabled vector driver by default.**
341
342  Previously, vector driver was disabled by default as it couldn't fill packet type
343  info for l3fwd to work well. Now there is an option for l3fwd to analyze
344  the packet type so the vector driver is enabled by default.
345
346* **i40e: Fixed link info of VF.**
347
348  Previously, the VF's link speed stayed at 10GbE and status always was up.
349  It did not change even when the physical link's status changed.
350  Now this issue is fixed to make VF's link info consistent with physical link.
351
352* **mlx5: Fixed possible crash during initialization.**
353
354  A crash could occur when failing to allocate private device context.
355
356* **mlx5: Added port type check.**
357
358  Added port type check to prevent port initialization on non-Ethernet link layers and
359  to report an error.
360
361* **mlx5: Applied VLAN filtering to broadcast and IPv6 multicast flows.**
362
363  Prevented reception of multicast frames outside of configured VLANs.
364
365* **mlx5: Fixed RX checksum offload in non L3/L4 packets.**
366
367  Fixed report of bad checksum for packets of unknown type.
368
369* **aesni_mb: Fixed wrong return value when creating a device.**
370
371  The ``cryptodev_aesni_mb_init()`` function was returning the device id of the device created,
372  instead of 0 (on success) that ``rte_eal_vdev_init()`` expects.
373  This made it impossible to create more than one aesni_mb device
374  from the command line.
375
376* **qat: Fixed AES GCM decryption.**
377
378  Allowed AES GCM on the cryptodev API, but in some cases gave invalid results
379  due to incorrect IV setting.
380
381
382Libraries
383~~~~~~~~~
384
385* **hash: Fixed CRC32c hash computation for non multiple of 4 bytes sizes.**
386
387  Fix crc32c hash functions to return a valid crc32c value for data lengths
388  not a multiple of 4 bytes.
389
390* **hash: Fixed hash library to support multi-process mode.**
391
392  Fix hash library to support multi-process mode, using a jump table,
393  instead of storing a function pointer to the key compare function.
394  Multi-process mode only works with the built-in compare functions,
395  however a custom compare function (not in the jump table) can only
396  be used in single-process mode.
397
398* **hash: Fixed return value when allocating an existing hash table.**
399
400  Changed the ``rte_hash*_create()`` functions to return ``NULL`` and set
401  ``rte_errno`` to ``EEXIST`` when the object name already exists. This is
402  the behavior described in the API documentation in the header file.
403  The previous behavior was to return a pointer to the existing object in
404  that case, preventing the caller from knowing if the object had to be freed
405  or not.
406
407* **lpm: Fixed return value when allocating an existing object.**
408
409  Changed the ``rte_lpm*_create()`` functions to return ``NULL`` and set
410  ``rte_errno`` to ``EEXIST`` when the object name already exists. This is
411  the behavior described in the API documentation in the header file.
412  The previous behavior was to return a pointer to the existing object in
413  that case, preventing the caller from knowing if the object had to be freed
414  or not.
415
416* **librte_port: Fixed segmentation fault for ring and ethdev writer nodrop.**
417
418  Fixed core dump issue on txq and swq when dropless is set to yes.
419
420
421Examples
422~~~~~~~~
423
424* **l3fwd-power: Fixed memory leak for non-IP packet.**
425
426  Fixed issue in l3fwd-power where, on receiving packets of types
427  other than IPv4 or IPv6, the mbuf was not released, and caused
428  a memory leak.
429
430* **l3fwd: Fixed using packet type blindly.**
431
432  l3fwd makes use of packet type information without querying if devices or PMDs
433  really set it. For those devices that don't set ptypes, add an option to parse it.
434
435* **examples/vhost: Fixed frequent mbuf allocation failure.**
436
437  The vhost-switch often fails to allocate mbuf when dequeue from vring because it
438  wrongly calculates the number of mbufs needed.
439
440
441API Changes
442-----------
443
444* The ethdev statistics counter ``imissed`` is considered to be independent of ``ierrors``.
445  All drivers are now counting the missed packets only once, i.e. drivers will
446  not increment ierrors anymore for missed packets.
447
448* The ethdev structure ``rte_eth_dev_info`` was changed to support device
449  speed capabilities.
450
451* The ethdev structures ``rte_eth_link`` and ``rte_eth_conf`` were changed to
452  support the new link API.
453
454* The functions ``rte_eth_dev_udp_tunnel_add`` and ``rte_eth_dev_udp_tunnel_delete``
455  have been renamed into ``rte_eth_dev_udp_tunnel_port_add`` and
456  ``rte_eth_dev_udp_tunnel_port_delete``.
457
458* The ``outer_mac`` and ``inner_mac`` fields in structure
459  ``rte_eth_tunnel_filter_conf`` are changed from pointer to struct in order
460  to keep code's readability.
461
462* The fields in ethdev structure ``rte_eth_fdir_masks`` were changed
463  to be in big endian.
464
465* A parameter ``vlan_type`` has been added to the function
466  ``rte_eth_dev_set_vlan_ether_type``.
467
468* The af_packet device init function is no longer public. The device should be attached
469  via the API.
470
471* The LPM ``next_hop`` field is extended from 8 bits to 24 bits for IPv4
472  while keeping ABI compatibility.
473
474* A new ``rte_lpm_config`` structure is used so the LPM library will allocate
475  exactly the amount of memory which is necessary to hold application’s rules.
476  The previous ABI is kept for compatibility.
477
478* The prototype for the pipeline input port, output port and table action
479  handlers are updated: the pipeline parameter is added,
480  the packets mask parameter has been either removed or made input-only.
481
482
483ABI Changes
484-----------
485
486* The RETA entry size in ``rte_eth_rss_reta_entry64`` has been increased
487  from 8-bit to 16-bit.
488
489* The ethdev flow director structure ``rte_eth_fdir_flow`` structure was
490  changed. New fields were added to extend flow director's input set.
491
492* The cmdline buffer size has been increase from 256 to 512.
493
494
495Shared Library Versions
496-----------------------
497
498The libraries prepended with a plus sign were incremented in this version.
499
500.. code-block:: diff
501
502   + libethdev.so.3
503     librte_acl.so.2
504     librte_cfgfile.so.2
505   + librte_cmdline.so.2
506     librte_distributor.so.1
507     librte_eal.so.2
508     librte_hash.so.2
509     librte_ip_frag.so.1
510     librte_ivshmem.so.1
511     librte_jobstats.so.1
512     librte_kni.so.2
513     librte_kvargs.so.1
514     librte_lpm.so.2
515     librte_mbuf.so.2
516     librte_mempool.so.1
517     librte_meter.so.1
518   + librte_pipeline.so.3
519     librte_pmd_bond.so.1
520     librte_pmd_ring.so.2
521     librte_port.so.2
522     librte_power.so.1
523     librte_reorder.so.1
524     librte_ring.so.1
525     librte_sched.so.1
526     librte_table.so.2
527     librte_timer.so.1
528     librte_vhost.so.2
529
530
531Tested Platforms
532----------------
533
534#. SuperMicro 1U
535
536   - BIOS: 1.0c
537   - Processor: Intel(R) Atom(TM) CPU C2758 @ 2.40GHz
538
539#. SuperMicro 1U
540
541   - BIOS: 1.0a
542   - Processor: Intel(R) Xeon(R) CPU D-1540 @ 2.00GHz
543   - Onboard NIC: Intel(R) X552/X557-AT (2x10G)
544
545     - Firmware-version: 0x800001cf
546     - Device ID (PF/VF): 8086:15ad /8086:15a8
547
548   - kernel driver version: 4.2.5 (ixgbe)
549
550#. SuperMicro 1U
551
552   - BIOS: 1.0a
553   - Processor: Intel(R) Xeon(R) CPU E5-4667 v3 @ 2.00GHz
554
555#. Intel(R) Server board S2600GZ
556
557   - BIOS: SE5C600.86B.02.02.0002.122320131210
558   - Processor: Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz
559
560#. Intel(R) Server board W2600CR
561
562   - BIOS: SE5C600.86B.02.01.0002.082220131453
563   - Processor: Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz
564
565#. Intel(R) Server board S2600CWT
566
567   - BIOS: SE5C610.86B.01.01.0009.060120151350
568   - Processor: Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz
569
570#. Intel(R) Server board S2600WTT
571
572   - BIOS: SE5C610.86B.01.01.0005.101720141054
573   - Processor: Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz
574
575#. Intel(R) Server board S2600WTT
576
577   - BIOS: SE5C610.86B.11.01.0044.090120151156
578   - Processor: Intel(R) Xeon(R) CPU E5-2695 v4 @ 2.10GHz
579
580
581Tested NICs
582-----------
583
584#. Intel(R) Ethernet Controller X540-AT2
585
586   - Firmware version: 0x80000389
587   - Device id (pf): 8086:1528
588   - Driver version: 3.23.2 (ixgbe)
589
590#. Intel(R) 82599ES 10 Gigabit Ethernet Controller
591
592   - Firmware version: 0x61bf0001
593   - Device id (pf/vf): 8086:10fb / 8086:10ed
594   - Driver version: 4.0.1-k (ixgbe)
595
596#. Intel(R) Corporation Ethernet Connection X552/X557-AT 10GBASE-T
597
598   - Firmware version: 0x800001cf
599   - Device id (pf/vf): 8086:15ad / 8086:15a8
600   - Driver version: 4.2.5 (ixgbe)
601
602#. Intel(R) Ethernet Converged Network Adapter X710-DA4 (4x10G)
603
604   - Firmware version: 5.02 0x80002284
605   - Device id (pf/vf): 8086:1572 / 8086:154c
606   - Driver version: 1.4.26 (i40e)
607
608#. Intel(R) Ethernet Converged Network Adapter X710-DA2 (2x10G)
609
610   - Firmware version: 5.02 0x80002282
611   - Device id (pf/vf): 8086:1572 / 8086:154c
612   - Driver version: 1.4.25 (i40e)
613
614#. Intel(R) Ethernet Converged Network Adapter XL710-QDA1 (1x40G)
615
616   - Firmware version: 5.02 0x80002281
617   - Device id (pf/vf): 8086:1584 / 8086:154c
618   - Driver version: 1.4.25 (i40e)
619
620#. Intel(R) Ethernet Converged Network Adapter XL710-QDA2 (2X40G)
621
622   - Firmware version: 5.02 0x80002285
623   - Device id (pf/vf): 8086:1583 / 8086:154c
624   - Driver version: 1.4.25 (i40e)
625
626#. Intel(R) 82576EB Gigabit Ethernet Controller
627
628   - Firmware version: 1.2.1
629   - Device id (pf): 8086:1526
630   - Driver version: 5.2.13-k (igb)
631
632#. Intel(R) Ethernet Controller I210
633
634   - Firmware version: 3.16, 0x80000500, 1.304.0
635   - Device id (pf): 8086:1533
636   - Driver version: 5.2.13-k (igb)
637
638#. Intel(R) Corporation I350 Gigabit Network Connection
639
640   - Firmware version: 1.48, 0x800006e7
641   - Device id (pf/vf): 8086:1521 / 8086:1520
642   - Driver version: 5.2.13-k (igb)
643
644
645#. Intel(R) Ethernet Multi-host Controller FM10000
646
647   - Firmware version: N/A
648   - Device id (pf/vf): 8086:15d0
649   - Driver version: 0.17.0.9 (fm10k)
650