release_2_2.rst revision 97f17497
1DPDK Release 2.2
2================
3
4New Features
5------------
6
7* **Introduce ARMv7 and ARMv8 architectures.**
8
9  * It is now possible to build DPDK for the ARMv7 and ARMv8 platforms.
10  * ARMv7 can be tested with virtual PMD drivers.
11  * ARMv8 can be tested with virtual and physical PMD drivers.
12
13* **Enabled freeing of ring.**
14
15  A new function ``rte_ring_free()`` has been added to allow the user to free
16  a ring if it was created with ``rte_ring_create()``.
17
18* **Added keepalive support to EAL and example application.**
19
20* **Added experimental cryptodev API**
21
22  The cryptographic processing of packets is provided as a preview
23  with two drivers for:
24
25  * Intel QuickAssist devices
26  * Intel AES-NI multi-buffer library
27
28  Due to its experimental state, the API may change without prior notice.
29
30* **Added ethdev APIs for additional IEEE1588 support.**
31
32  Added functions to read, write and adjust system time in the NIC.
33  Added client slave sample application to demonstrate the IEEE1588
34  functionality.
35
36* **Extended Statistics.**
37
38  Defined an extended statistics naming scheme to store metadata in the name
39  string of each statistic. Refer to the Extended Statistics section of the
40  Programmers Guide for more details.
41
42  Implemented the extended statistics API for the following PMDs:
43
44  * ``igb``
45  * ``igbvf``
46  * ``i40e``
47  * ``i40evf``
48  * ``fm10k``
49  * ``virtio``
50
51* **Added API in ethdev to retrieve RX/TX queue information.**
52
53  *  Added the ability for the upper layer to query RX/TX queue information.
54  *  Added new fields in ``rte_eth_dev_info`` to represent information about
55     RX/TX descriptors min/max/align numbers, per queue, for the device.
56
57* **Added RSS dynamic configuration to bonding.**
58
59* **Updated the e1000 base driver.**
60
61  The e1000 base driver was updated with several features including the
62  following:
63
64  * Added new i218 devices
65  * Allowed both ULP and EEE in Sx state
66  * Initialized 88E1543 (Marvell 1543) PHY
67  * Added flags to set EEE advertisement modes
68  * Supported inverted format ETrackId
69  * Added bit to disable packetbuffer read
70  * Added defaults for i210 RX/TX PBSIZE
71  * Check more errors for ESB2 init and reset
72  * Check more NVM read errors
73  * Return code after setting receive address register
74  * Removed all NAHUM6LP_HW tags
75
76* **Added e1000 RX interrupt support.**
77
78* **Added igb TSO support for both PF and VF.**
79
80* **Added RSS enhancements to Intel x550 NIC.**
81
82  * Added support for 512 entry RSS redirection table.
83  * Added support for per VF RSS redirection table.
84
85* **Added Flow director enhancements on Intel x550 NIC.**
86
87  * Added 2 new flow director modes on x550.
88    One is MAC VLAN mode, the other is tunnel mode.
89
90* **Updated the i40e base driver.**
91
92  The i40e base driver was updated with several changes including the
93  following:
94
95  *  Added promiscuous on VLAN support
96  *  Added a workaround to drop all flow control frames
97  *  Added VF capabilities to virtual channel interface
98  *  Added TX Scheduling related AQ commands
99  *  Added additional PCTYPES supported for FortPark RSS
100  *  Added parsing for CEE DCBX TLVs
101  *  Added FortPark specific registers
102  *  Added AQ functions to handle RSS Key and LUT programming
103  *  Increased PF reset max loop limit
104
105* **Added i40e vector RX/TX.**
106
107* **Added i40e RX interrupt support.**
108
109* **Added i40e flow control support.**
110
111* **Added DCB support to i40e PF driver.**
112
113* **Added RSS/FD input set granularity on Intel X710/XL710.**
114
115* **Added different GRE key length for input set on Intel X710/XL710.**
116
117* **Added flow director support in i40e VF.**
118
119* **Added i40e support of early X722 series.**
120
121  Added early X722 support, for evaluation only, as the hardware is alpha.
122
123* **Added fm10k vector RX/TX.**
124
125* **Added fm10k TSO support for both PF and VF.**
126
127* **Added fm10k VMDQ support.**
128
129* **New NIC Boulder Rapid support.**
130
131  Added support for the Boulder Rapid variant of Intel's fm10k NIC family.
132
133* **Enhanced support for the Chelsio CXGBE driver.**
134
135  *  Added support for Jumbo Frames.
136  *  Optimized forwarding performance for Chelsio T5 40GbE cards.
137
138* **Improved enic TX packet rate.**
139
140  Reduced frequency of TX tail pointer updates to the NIC.
141
142* **Added support for link status interrupts in mlx4.**
143
144* **Added partial support (TX only) for secondary processes in mlx4.**
145
146* **Added support for Mellanox ConnectX-4 adapters (mlx5).**
147
148  The mlx5 poll-mode driver implements support for Mellanox ConnectX-4 EN
149  and Mellanox ConnectX-4 Lx EN families of 10/25/40/50/100 Gb/s adapters.
150
151  Like mlx4, this PMD is only available for Linux and is disabled by default
152  due to external dependencies (libibverbs and libmlx5).
153
154* **Added driver for Netronome nfp-6xxx card.**
155
156  Support for using Netronome nfp-6xxx with PCI VFs.
157
158* **Added virtual szedata2 driver for COMBO cards.**
159
160  Added virtual PMD for COMBO-100G and COMBO-80G cards.
161  PMD is disabled in default configuration.
162
163* **Enhanced support for virtio driver.**
164
165  * Virtio ring layout optimization (fixed avail ring)
166  * Vector RX
167  * Simple TX
168
169* **Added vhost-user multiple queue support.**
170
171* **Added port hotplug support to vmxnet3.**
172
173* **Added port hotplug support to xenvirt.**
174
175* **Added ethtool shim and sample application.**
176
177* **Added experimental performance thread example application.**
178
179  The new sample application demonstrates L3 forwarding with different threading
180  models: pthreads, cgroups, or lightweight threads. The example includes
181  a simple cooperative scheduler.
182
183  Due to its experimental state this application may change without notice.
184  The application is supported only for Linux x86_64.
185
186* **Enhancements to the IP pipeline application.**
187
188  The following features have been added to the ``ip_pipeline``
189  application;
190
191  * Added Multiple Producers/Multiple Consumers (MPSC)
192    and fragmentation/reassembly support to software rings.
193
194  * Added a dynamic pipeline reconfiguration feature that
195    allows binding a pipeline to other threads at runtime
196    using CLI commands.
197
198  * Added enable/disable of ``promisc`` mode from ``ip_pipeline``
199    configuration file.
200
201  * Added check on RX queues and TX queues of each link
202    whether they are used correctly in the ``ip_pipeline``
203    configuration file.
204
205  * Added flow id parameters to the flow-classification
206    table entries.
207
208  * Added more functions to the routing pipeline:
209    ARP table enable/disable, Q-in-Q and MPLS encapsulation,
210    add color (traffic-class for QoS) to the MPLS tag.
211
212  * Added flow-actions pipeline for traffic metering/marking
213    (for e.g. Two Rate Three Color Marker (trTCM)), policer etc.
214
215  * Modified the pass-through pipeline's actions-handler to
216    implement a generic approach to extract fields from the
217    packet's header and copy them to packet metadata.
218
219
220Resolved Issues
221---------------
222
223EAL
224~~~
225
226* **eal/linux: Fixed epoll timeout.**
227
228  Fixed issue where the ``rte_epoll_wait()`` function didn't return when the
229  underlying call to ``epoll_wait()`` timed out.
230
231
232Drivers
233~~~~~~~
234
235* **e1000/base: Synchronize PHY interface on non-ME systems.**
236
237  On power up, the MAC - PHY interface needs to be set to PCIe, even if the
238  cable is disconnected. In ME systems, the ME handles this on exit from the
239  Sx (Sticky mode) state. In non-ME, the driver handles it. Added a check for
240  non-ME system to the driver code that handles it.
241
242* **e1000/base: Increased timeout of reset check.**
243
244  Previously, in ``check_reset_block`` RSPCIPHY was polled for 100 ms before
245  determining that the ME veto was set. This was not enough and it was
246  increased to 300 ms.
247
248* **e1000/base: Disabled IPv6 extension header parsing on 82575.**
249
250  Disabled IPv6 options as per hardware limitation.
251
252* **e1000/base: Prevent ULP flow if cable connected.**
253
254  Enabling ULP on link down when the cable is connected caused an infinite
255  loop of link up/down indications in the NDIS driver.
256  The driver now enables ULP only when the cable is disconnected.
257
258* **e1000/base: Support different EEARBC for i210.**
259
260  EEARBC has changed on i210. It means EEARBC has a different address on
261  i210 than on other NICs. So, add a new entity named EEARBC_I210 to the
262  register list and make sure the right one is being used on i210.
263
264* **e1000/base: Fix K1 configuration.**
265
266  Added fix for the following updates to the K1 configurations:
267  TX idle period for entering K1 should be 128 ns.
268  Minimum TX idle period in K1 should be 256 ns.
269
270* **e1000/base: Fix link detect flow.**
271
272  Fix link detect flow in case where auto-negotiate is not enabled, by calling
273  ``e1000_setup_copper_link_generic`` instead of ``e1000_phy_setup_autoneg``.
274
275* **e1000/base: Fix link check for i354 M88E1112 PHY.**
276
277  The ``e1000_check_for_link_media_swap()`` function is supposed to check PHY
278  page 0 for copper and PHY page 1 for "other" (fiber) links. The driver
279  switched back from page 1 to page 0 too soon, before
280  ``e1000_check_for_link_82575()`` is executed and was never finding the link
281  on the fiber (other).
282
283  If the link is copper, as the M88E1112 page address is set to 1, it should be
284  set back to 0 before checking this link.
285
286* **e1000/base: Fix beacon duration for i217.**
287
288  Fix for I217 Packet Loss issue - The Management Engine sets the FEXTNVM4
289  Beacon Duration incorrectly. This fix ensures that the correct value will
290  always be set. Correct value for this field is 8 usec.
291
292* **e1000/base: Fix TIPG for non 10 half duplex mode.**
293
294  TIPG value is increased when setting speed to 10 half duplex to prevent
295  packet loss. However, it was never decreased again when speed
296  changed. This caused performance issues in the NDIS driver.
297  Fix this to restore TIPG to default value on non 10 half duplex.
298
299* **e1000/base: Fix reset of DH89XXCC SGMII.**
300
301  For DH89XXCC_SGMII, a write flush leaves registers of this device trashed
302  (0xFFFFFFFF). Add check for this device.
303
304  Also, after both Port SW Reset and Device Reset case, the platform should
305  wait at least 3ms before reading any registers. Remove this condition since
306  waiting is conditionally executed only for Device Reset.
307
308* **e1000/base: Fix redundant PHY power down for i210.**
309
310  Bit 11 of PHYREG 0 is used to power down PHY. The use of PHYREG 16 is
311  no longer necessary.
312
313* **e1000/base: fix jumbo frame CRC failures.**
314
315  Change the value of register 776.20[11:2] for jumbo mode from 0x1A to 0x1F.
316  This is to enlarge the gap between read and write pointers in the TX FIFO.
317
318* **e1000/base: Fix link flap on 82579.**
319
320  Several customers have reported a link flap issue on 82579. The symptoms
321  are random and intermittent link losses when 82579 is connected to specific
322  switches. the Issue was root caused as an inter-operability problem between
323  the NIC and at least some Broadcom PHYs in the Energy Efficient Ethernet
324  wake mechanism.
325
326  To fix the issue, we are disabling the Phase Locked Loop shutdown in 100M
327  Low Power Idle. This solution will cause an increase of power in 100M EEE
328  link. It may cost an additional 28mW in this specific mode.
329
330* **igb: Fixed IEEE1588 frame identification in I210.**
331
332  Fixed issue where the flag ``PKT_RX_IEEE1588_PTP`` was not being set
333  in the Intel I210 NIC, as the EtherType in RX descriptor is in bits 8:10 of
334  Packet Type and not in the default bits 0:2.
335
336* **igb: Fixed VF start with PF stopped.**
337
338  VF needs the PF interrupt support initialized even if not started.
339
340* **igb: Fixed VF MAC address when using with DPDK PF.**
341
342  Assign a random MAC address in VF when not assigned by PF.
343
344* **igb: Removed CRC bytes from byte counter statistics.**
345
346* **ixgbe: Fixed issue with X550 DCB.**
347
348  Fixed a DCB issue with x550 where for 8 TCs (Traffic Classes), if a packet
349  with user priority 6 or 7 was injected to the NIC, then the NIC would only
350  put 3 packets into the queue. There was also a similar issue for 4 TCs.
351
352* **ixgbe: Removed burst size restriction of vector RX.**
353
354  Fixed issue where a burst size less than 32 didn't receive anything.
355
356* **ixgbe: Fixed VF start with PF stopped.**
357
358  VF needs the PF interrupt support initialized even if not started.
359
360* **ixgbe: Fixed TX hang when RS distance exceeds HW limit.**
361
362  Fixed an issue where the TX queue can hang when a lot of highly fragmented
363  packets have to be sent. As part of that fix, ``tx_rs_thresh`` for ixgbe PMD
364  is not allowed to be greater then to 32 to comply with HW restrictions.
365
366* **ixgbe: Fixed rx error statistic counter.**
367
368  Fixed an issue that the rx error counter of ixgbe was not accurate. The
369  mac short packet discard count (mspdc) was added to the counter. Mac local
370  faults and mac remote faults are removed as they do not count packets but
371  errors, and jabber errors were removed as they are already accounted for
372  by the CRC error counter. Finally the XEC (l3 / l4 checksum error) counter
373  was removed due to errata, see commit 256ff05a9cae for details.
374
375* **ixgbe: Removed CRC bytes from byte counter statistics.**
376
377* **i40e: Fixed base driver allocation when not using first numa node.**
378
379  Fixed i40e issue that occurred when a DPDK application didn't initialize
380  ports if memory wasn't available on socket 0.
381
382* **i40e: Fixed maximum of 64 queues per port.**
383
384  Fixed an issue in i40e where it would not support more than 64 queues per
385  port, even though the hardware actually supports it. The real number of
386  queues may vary, as long as the total number of queues used in PF, VFs, VMDq
387  and FD does not exceeds the hardware maximum.
388
389* **i40e: Fixed statistics of packets.**
390
391  Added discarding packets on VSI to the stats and rectify the old statistics.
392
393* **i40e: Fixed issue of not freeing memzone.**
394
395  Fixed an issue of not freeing a memzone in the call to free the memory for
396  adminq DMA.
397
398* **i40e: Removed CRC bytes from byte counter statistics.**
399
400* **mlx: Fixed driver loading.**
401
402  The mlx drivers were unable to load when built as a shared library,
403  due to a missing symbol in the mempool library.
404
405* **mlx4: Performance improvements.**
406
407  Fixed bugs in TX and RX flows that improves mlx4 performance.
408
409* **mlx4: Fixed TX loss after initialization.**
410
411* **mlx4: Fixed scattered TX with too many segments.**
412
413* **mlx4: Fixed memory registration for indirect mbuf data.**
414
415* **vhost: Fixed Qemu shutdown.**
416
417  Fixed issue with libvirt ``virsh destroy`` not killing the VM.
418
419* **virtio: Fixed crash after changing link state.**
420
421  Fixed IO permission in the interrupt handler.
422
423* **virtio: Fixed crash when releasing queue.**
424
425  Fixed issue when releasing null control queue.
426
427
428Libraries
429~~~~~~~~~
430
431* **hash: Fixed memory allocation of Cuckoo Hash key table.**
432
433  Fixed issue where an incorrect Cuckoo Hash key table size could be
434  calculated limiting the size to 4GB.
435
436* **hash: Fixed incorrect lookup if key is all zero.**
437
438  Fixed issue in hash library that occurred if an all zero
439  key was not added to the table and the key was looked up,
440  resulting in an incorrect hit.
441
442* **hash: Fixed thread scaling by reducing contention.**
443
444  Fixed issue in the hash library where, using multiple cores with
445  hardware transactional memory support, thread scaling did not work,
446  due to the global ring that is shared by all cores.
447
448
449Examples
450~~~~~~~~
451
452* **l3fwd: Fixed crash with IPv6.**
453
454* **vhost_xen: Fixed compile error.**
455
456
457Other
458~~~~~
459
460* This release drops compatibility with Linux kernel 2.6.33. The minimum
461  kernel requirement is now 2.6.34.
462
463
464Known Issues
465------------
466
467* Some drivers do not fill in the packet type when receiving.
468  As the l3fwd example application requires this info, the i40e vector
469  driver must be disabled to benefit of the packet type with i40e.
470
471* Some (possibly all) VF drivers (e.g. i40evf) do not handle any PF reset
472  events/requests in the VF driver. This means that the VF driver may not work
473  after a PF reset in the host side. The workaround is to avoid triggering any
474  PF reset events/requests on the host side.
475
476* 100G link report support is missing.
477
478* **Mellanox PMDs (mlx4 & mlx5):**
479
480  * PMDs do not support CONFIG_RTE_BUILD_COMBINE_LIBS and
481    CONFIG_RTE_BUILD_SHARED_LIB simultaneously.
482
483  * There is performance degradation for small packets when the PMD is
484    compiled with ``SGE_WR_N = 4`` compared to the performance when ``SGE_WR_N
485    = 1``. If scattered packets are not used it is recommended to compile the
486    PMD with ``SGE_WR_N = 1``.
487
488  * When a Multicast or Broadcast packet is sent to the SR-IOV mlx4 VF,
489    it is returned back to the port.
490
491  * PMDs report "bad" L4 checksum when IP packet is received.
492
493  * mlx5 PMD reports "bad" checksum although the packet has "good" checksum.
494    Will be fixed in upcoming MLNX_OFED release.
495
496
497API Changes
498-----------
499
500* The deprecated flow director API is removed.
501  It was replaced by ``rte_eth_dev_filter_ctrl()``.
502
503* The ``dcb_queue`` is renamed to ``dcb_tc`` in following dcb configuration
504  structures: ``rte_eth_dcb_rx_conf``, ``rte_eth_dcb_tx_conf``,
505  ``rte_eth_vmdq_dcb_conf``, ``rte_eth_vmdq_dcb_tx_conf``.
506
507* The ``rte_eth_rx_queue_count()`` function now returns "int" instead of
508  "uint32_t" to allow the use of negative values as error codes on return.
509
510* The function ``rte_eal_pci_close_one()`` is removed.
511  It was replaced by ``rte_eal_pci_detach()``.
512
513* The deprecated ACL API ``ipv4vlan`` is removed.
514
515* The deprecated hash function ``rte_jhash2()`` is removed.
516  It was replaced by ``rte_jhash_32b()``.
517
518* The deprecated KNI functions are removed:
519  ``rte_kni_create()``, ``rte_kni_get_port_id()`` and ``rte_kni_info_get()``.
520
521* The deprecated ring PMD functions are removed:
522  ``rte_eth_ring_pair_create()`` and ``rte_eth_ring_pair_attach()``.
523
524* The devargs union field ``virtual`` is renamed to ``virt`` for C++
525  compatibility.
526
527
528ABI Changes
529-----------
530
531* The EAL and ethdev structures ``rte_intr_handle`` and ``rte_eth_conf`` were
532  changed to support RX interrupt. This was already included in 2.1 under the
533  ``CONFIG_RTE_NEXT_ABI`` #define.
534
535* The ethdev flow director entries for SCTP were changed.
536  This was already included in 2.1 under the ``CONFIG_RTE_NEXT_ABI`` #define.
537
538* The ethdev flow director structure ``rte_eth_fdir_flow_ext`` structure was
539  changed. New fields were added to support flow director filtering in VF.
540
541* The size of the ethdev structure ``rte_eth_hash_filter_info`` is changed
542  by adding a new element ``rte_eth_input_set_conf`` in a union.
543
544* New fields ``rx_desc_lim`` and ``tx_desc_lim`` are added into
545  ``rte_eth_dev_info`` structure.
546
547* For debug builds, the functions ``rte_eth_rx_burst()``, ``rte_eth_tx_burst()``
548  ``rte_eth_rx_descriptor_done()`` and ``rte_eth_rx_queue_count()`` will
549  no longer be separate functions in the DPDK libraries. Instead, they will
550  only be present in the ``rte_ethdev.h`` header file.
551
552* The maximum number of queues per port ``CONFIG_RTE_MAX_QUEUES_PER_PORT`` is
553  increased to 1024.
554
555* The mbuf structure was changed to support the unified packet type.
556  This was already included in 2.1 under the ``CONFIG_RTE_NEXT_ABI`` #define.
557
558* The dummy malloc library is removed. The content was moved into EAL in 2.1.
559
560* The LPM structure is changed. The deprecated field ``mem_location`` is
561  removed.
562
563* librte_table LPM: A new parameter to hold the table name will be added to
564  the LPM table parameter structure.
565
566* librte_table hash: The key mask parameter is added to the hash table
567  parameter structure for 8-byte key and 16-byte key extendable bucket
568  and LRU tables.
569
570* librte_port: Macros to access the packet meta-data stored within the packet
571  buffer has been adjusted to cover the packet mbuf structure.
572
573* librte_cfgfile: Allow longer names and values by increasing the constants
574  ``CFG_NAME_LEN`` and ``CFG_VALUE_LEN`` to 64 and 256 respectively.
575
576* vhost: a new field enabled is added to the ``vhost_virtqueue`` structure.
577
578* vhost: a new field ``virt_qp_nb`` is added to ``virtio_net`` structure, and
579  the ``virtqueue`` field is moved to the end of virtio_net structure.
580
581* vhost: a new operation ``vring_state_changed`` is added to
582  ``virtio_net_device_ops`` structure.
583
584* vhost: a few spaces are reserved both at ``vhost_virtqueue`` and
585  ``virtio_net`` structure for future extension.
586
587
588Shared Library Versions
589-----------------------
590
591The libraries prepended with a plus sign were incremented in this version.
592
593.. code-block:: diff
594
595   + libethdev.so.2
596   + librte_acl.so.2
597   + librte_cfgfile.so.2
598     librte_cmdline.so.1
599     librte_distributor.so.1
600   + librte_eal.so.2
601   + librte_hash.so.2
602     librte_ip_frag.so.1
603     librte_ivshmem.so.1
604     librte_jobstats.so.1
605   + librte_kni.so.2
606     librte_kvargs.so.1
607   + librte_lpm.so.2
608   + librte_mbuf.so.2
609     librte_mempool.so.1
610     librte_meter.so.1
611   + librte_pipeline.so.2
612     librte_pmd_bond.so.1
613   + librte_pmd_ring.so.2
614   + librte_port.so.2
615     librte_power.so.1
616     librte_reorder.so.1
617     librte_ring.so.1
618     librte_sched.so.1
619   + librte_table.so.2
620     librte_timer.so.1
621   + librte_vhost.so.2
622