xref: /dpdk/doc/guides/nics/ixgbe.rst (revision 68a03efeed657e6e05f281479b33b51102797e15)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright(c) 2010-2016 Intel Corporation.
3
4IXGBE Driver
5============
6
7Vector PMD for IXGBE
8--------------------
9
10Vector PMD uses Intel® SIMD instructions to optimize packet I/O.
11It improves load/store bandwidth efficiency of L1 data cache by using a wider SSE/AVX register 1 (1).
12The wider register gives space to hold multiple packet buffers so as to save instruction number when processing bulk of packets.
13
14There is no change to PMD API. The RX/TX handler are the only two entries for vPMD packet I/O.
15They are transparently registered at runtime RX/TX execution if all condition checks pass.
16
171.  To date, only an SSE version of IX GBE vPMD is available.
18
19Some constraints apply as pre-conditions for specific optimizations on bulk packet transfers.
20The following sections explain RX and TX constraints in the vPMD.
21
22RX Constraints
23~~~~~~~~~~~~~~
24
25Prerequisites and Pre-conditions
26^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
27
28The following prerequisites apply:
29
30*   To enable vPMD to work for RX, bulk allocation for Rx must be allowed.
31
32Ensure that the following pre-conditions are satisfied:
33
34*   rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
35
36*   rxq->rx_free_thresh < rxq->nb_rx_desc
37
38*   (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
39
40*   rxq->nb_rx_desc  < (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST)
41
42These conditions are checked in the code.
43
44Scattered packets are not supported in this mode.
45If an incoming packet is greater than the maximum acceptable length of one "mbuf" data size (by default, the size is 2 KB),
46vPMD for RX would be disabled.
47
48By default, IXGBE_MAX_RING_DESC is set to 4096 and RTE_PMD_IXGBE_RX_MAX_BURST is set to 32.
49
50Feature not Supported by RX Vector PMD
51^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
52
53Some features are not supported when trying to increase the throughput in vPMD.
54They are:
55
56*   IEEE1588
57
58*   FDIR
59
60*   Header split
61
62*   RX checksum off load
63
64Other features are supported using optional MACRO configuration. They include:
65
66*   HW VLAN strip
67
68*   HW extend dual VLAN
69
70To guarantee the constraint, capabilities in dev_conf.rxmode.offloads will be checked:
71
72*   DEV_RX_OFFLOAD_VLAN_STRIP
73
74*   DEV_RX_OFFLOAD_VLAN_EXTEND
75
76*   DEV_RX_OFFLOAD_CHECKSUM
77
78*   DEV_RX_OFFLOAD_HEADER_SPLIT
79
80*   dev_conf
81
82fdir_conf->mode will also be checked.
83
84VF Runtime Options
85^^^^^^^^^^^^^^^^^^
86
87The following ``devargs`` options can be enabled at runtime. They must
88be passed as part of EAL arguments. For example,
89
90.. code-block:: console
91
92   dpdk-testpmd -a af:10.0,pflink_fullchk=1 -- -i
93
94- ``pflink_fullchk`` (default **0**)
95
96  When calling ``rte_eth_link_get_nowait()`` to get VF link status,
97  this option is used to control how VF synchronizes its status with
98  PF's. If set, VF will not only check the PF's physical link status
99  by reading related register, but also check the mailbox status. We
100  call this behavior as fully checking. And checking mailbox will
101  trigger PF's mailbox interrupt generation. If unset, the application
102  can get the VF's link status quickly by just reading the PF's link
103  status register, this will avoid the whole system's mailbox interrupt
104  generation.
105
106  ``rte_eth_link_get()`` will still use the mailbox method regardless
107  of the pflink_fullchk setting.
108
109RX Burst Size
110^^^^^^^^^^^^^
111
112As vPMD is focused on high throughput, it assumes that the RX burst size is equal to or greater than 32 per burst.
113It returns zero if using nb_pkt < 32 as the expected packet number in the receive handler.
114
115TX Constraint
116~~~~~~~~~~~~~
117
118Prerequisite
119^^^^^^^^^^^^
120
121The only prerequisite is related to tx_rs_thresh.
122The tx_rs_thresh value must be greater than or equal to RTE_PMD_IXGBE_TX_MAX_BURST,
123but less or equal to RTE_IXGBE_TX_MAX_FREE_BUF_SZ.
124Consequently, by default the tx_rs_thresh value is in the range 32 to 64.
125
126Feature not Supported by TX Vector PMD
127^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
128
129TX vPMD only works when offloads is set to 0
130
131This means that it does not support any TX offload.
132
133Application Programming Interface
134---------------------------------
135
136In DPDK release v16.11 an API for ixgbe specific functions has been added to the ixgbe PMD.
137The declarations for the API functions are in the header ``rte_pmd_ixgbe.h``.
138
139Sample Application Notes
140------------------------
141
142l3fwd
143~~~~~
144
145When running l3fwd with vPMD, there is one thing to note.
146In the configuration, ensure that DEV_RX_OFFLOAD_CHECKSUM in port_conf.rxmode.offloads is NOT set.
147Otherwise, by default, RX vPMD is disabled.
148
149load_balancer
150~~~~~~~~~~~~~
151
152As in the case of l3fwd, to enable vPMD, do NOT set DEV_RX_OFFLOAD_CHECKSUM in port_conf.rxmode.offloads.
153In addition, for improved performance, use -bsz "(32,32),(64,64),(32,32)" in load_balancer to avoid using the default burst size of 144.
154
155
156Limitations or Known issues
157---------------------------
158
159Malicious Driver Detection not Supported
160~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
161
162The Intel x550 series NICs support a feature called MDD (Malicious
163Driver Detection) which checks the behavior of the VF driver.
164If this feature is enabled, the VF must use the advanced context descriptor
165correctly and set the CC (Check Context) bit.
166DPDK PF doesn't support MDD, but kernel PF does. We may hit problem in this
167scenario kernel PF + DPDK VF. If user enables MDD in kernel PF, DPDK VF will
168not work. Because kernel PF thinks the VF is malicious. But actually it's not.
169The only reason is the VF doesn't act as MDD required.
170There's significant performance impact to support MDD. DPDK should check if
171the advanced context descriptor should be set and set it. And DPDK has to ask
172the info about the header length from the upper layer, because parsing the
173packet itself is not acceptable. So, it's too expensive to support MDD.
174When using kernel PF + DPDK VF on x550, please make sure to use a kernel
175PF driver that disables MDD or can disable MDD.
176
177Some kernel drivers already disable MDD by default while some kernels can use
178the command ``insmod ixgbe.ko MDD=0,0`` to disable MDD. Each "0" in the
179command refers to a port. For example, if there are 6 ixgbe ports, the command
180should be changed to ``insmod ixgbe.ko MDD=0,0,0,0,0,0``.
181
182
183Statistics
184~~~~~~~~~~
185
186The statistics of ixgbe hardware must be polled regularly in order for it to
187remain consistent. Running a DPDK application without polling the statistics will
188cause registers on hardware to count to the maximum value, and "stick" at
189that value.
190
191In order to avoid statistic registers every reaching the maximum value,
192read the statistics from the hardware using ``rte_eth_stats_get()`` or
193``rte_eth_xstats_get()``.
194
195The maximum time between statistics polls that ensures consistent results can
196be calculated as follows:
197
198.. code-block:: c
199
200  max_read_interval = UINT_MAX / max_packets_per_second
201  max_read_interval = 4294967295 / 14880952
202  max_read_interval = 288.6218096127183 (seconds)
203  max_read_interval = ~4 mins 48 sec.
204
205In order to ensure valid results, it is recommended to poll every 4 minutes.
206
207MTU setting
208~~~~~~~~~~~
209
210Although the user can set the MTU separately on PF and VF ports, the ixgbe NIC
211only supports one global MTU per physical port.
212So when the user sets different MTUs on PF and VF ports in one physical port,
213the real MTU for all these PF and VF ports is the largest value set.
214This behavior is based on the kernel driver behavior.
215
216VF MAC address setting
217~~~~~~~~~~~~~~~~~~~~~~
218
219On ixgbe, the concept of "pool" can be used for different things depending on
220the mode. In VMDq mode, "pool" means a VMDq pool. In IOV mode, "pool" means a
221VF.
222
223There is no RTE API to add a VF's MAC address from the PF. On ixgbe, the
224``rte_eth_dev_mac_addr_add()`` function can be used to add a VF's MAC address,
225as a workaround.
226
227X550 does not support legacy interrupt mode
228~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
229
230Description
231^^^^^^^^^^^
232X550 cannot get interrupts if using ``uio_pci_generic`` module or using legacy
233interrupt mode of ``igb_uio`` or ``vfio``. Because the errata of X550 states
234that the Interrupt Status bit is not implemented. The errata is the item #22
235from `X550 spec update <https://www.intel.com/content/dam/www/public/us/en/
236documents/specification-updates/ethernet-x550-spec-update.pdf>`_
237
238Implication
239^^^^^^^^^^^
240When using ``uio_pci_generic`` module or using legacy interrupt mode of
241``igb_uio`` or ``vfio``, the Interrupt Status bit would be checked if the
242interrupt is coming. Since the bit is not implemented in X550, the irq cannot
243be handled correctly and cannot report the event fd to DPDK apps. Then apps
244cannot get interrupts and ``dmesg`` will show messages like ``irq #No.: ``
245``nobody cared.``
246
247Workaround
248^^^^^^^^^^
249Do not bind the ``uio_pci_generic`` module in X550 NICs.
250Do not bind ``igb_uio`` with legacy mode in X550 NICs.
251Before binding ``vfio`` with legacy mode in X550 NICs, use ``modprobe vfio ``
252``nointxmask=1`` to load ``vfio`` module if the intx is not shared with other
253devices.
254
255RSS isn't supported when QinQ is enabled
256~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
257
258Due to FW limitation, IXGBE doesn't support RSS when QinQ is enabled currently.
259
260UDP with zero checksum is reported as error
261~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
262
263Intel 82599 10 Gigabit Ethernet Controller Specification Update (Revision 2.87)
264Errata: 44 Integrity Error Reported for IPv4/UDP Packets With Zero Checksum
265
266To support UDP zero checksum, the zero and bad UDP checksum packet is marked as
267PKT_RX_L4_CKSUM_UNKNOWN, so the application needs to recompute the checksum to
268validate it.
269
270Inline crypto processing support
271--------------------------------
272
273Inline IPsec processing is supported for ``RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO``
274mode for ESP packets only:
275
276- ESP authentication only: AES-128-GMAC (128-bit key)
277- ESP encryption and authentication: AES-128-GCM (128-bit key)
278
279IPsec Security Gateway Sample Application supports inline IPsec processing for
280ixgbe PMD.
281
282For more details see the IPsec Security Gateway Sample Application and Security
283library documentation.
284
285
286Virtual Function Port Representors
287----------------------------------
288The IXGBE PF PMD supports the creation of VF port representors for the control
289and monitoring of IXGBE virtual function devices. Each port representor
290corresponds to a single virtual function of that device. Using the ``devargs``
291option ``representor`` the user can specify which virtual functions to create
292port representors for on initialization of the PF PMD by passing the VF IDs of
293the VFs which are required.::
294
295  -a DBDF,representor=[0,1,4]
296
297Currently hot-plugging of representor ports is not supported so all required
298representors must be specified on the creation of the PF.
299
300Supported Chipsets and NICs
301---------------------------
302
303- Intel 82599EB 10 Gigabit Ethernet Controller
304- Intel 82598EB 10 Gigabit Ethernet Controller
305- Intel 82599ES 10 Gigabit Ethernet Controller
306- Intel 82599EN 10 Gigabit Ethernet Controller
307- Intel Ethernet Controller X540-AT2
308- Intel Ethernet Controller X550-BT2
309- Intel Ethernet Controller X550-AT2
310- Intel Ethernet Controller X550-AT
311- Intel Ethernet Converged Network Adapter X520-SR1
312- Intel Ethernet Converged Network Adapter X520-SR2
313- Intel Ethernet Converged Network Adapter X520-LR1
314- Intel Ethernet Converged Network Adapter X520-DA1
315- Intel Ethernet Converged Network Adapter X520-DA2
316- Intel Ethernet Converged Network Adapter X520-DA4
317- Intel Ethernet Converged Network Adapter X520-QDA1
318- Intel Ethernet Converged Network Adapter X520-T2
319- Intel 10 Gigabit AF DA Dual Port Server Adapter
320- Intel 10 Gigabit AT Server Adapter
321- Intel 10 Gigabit AT2 Server Adapter
322- Intel 10 Gigabit CX4 Dual Port Server Adapter
323- Intel 10 Gigabit XF LR Server Adapter
324- Intel 10 Gigabit XF SR Dual Port Server Adapter
325- Intel 10 Gigabit XF SR Server Adapter
326- Intel Ethernet Converged Network Adapter X540-T1
327- Intel Ethernet Converged Network Adapter X540-T2
328- Intel Ethernet Converged Network Adapter X550-T1
329- Intel Ethernet Converged Network Adapter X550-T2
330