1 .. SPDX-License-Identifier: GPL-2.0+
3 =================================================================
4 Linux Base Driver for the Intel(R) Ethernet Controller 800 Series
5 =================================================================
7 Intel ice Linux driver.
8 Copyright(c) 2018-2021 Intel Corporation.
14 - Identifying Your Adapter
16 - Additional Features & Configurations
17 - Performance Optimization
20 The associated Virtual Function (VF) driver for this driver is iavf.
22 Driver information can be obtained using ethtool and lspci.
24 For questions related to hardware requirements, refer to the documentation
25 supplied with your Intel adapter. All hardware requirements listed apply to use
28 This driver supports XDP (Express Data Path) and AF_XDP zero-copy. Note that
29 XDP is blocked for frame sizes larger than 3KB.
32 Identifying Your Adapter
33 ========================
34 For information on how to identify your adapter, and for the latest Intel
35 network drivers, refer to the Intel Support website:
36 https://www.intel.com/support
42 Packet drops may occur under receive stress
43 -------------------------------------------
44 Devices based on the Intel(R) Ethernet Controller 800 Series are designed to
45 tolerate a limited amount of system latency during PCIe and DMA transactions.
46 If these transactions take longer than the tolerated latency, it can impact the
47 length of time the packets are buffered in the device and associated memory,
48 which may result in dropped packets. These packets drops typically do not have
49 a noticeable impact on throughput and performance under standard workloads.
51 If these packet drops appear to affect your workload, the following may improve
54 1) Make sure that your system's physical memory is in a high-performance
55 configuration, as recommended by the platform vendor. A common
56 recommendation is for all channels to be populated with a single DIMM
58 2) In your system's BIOS/UEFI settings, select the "Performance" profile.
59 3) Your distribution may provide tools like "tuned," which can help tweak
60 kernel settings to achieve better standard settings for different workloads.
63 Configuring SR-IOV for improved network security
64 ------------------------------------------------
65 In a virtualized environment, on Intel(R) Ethernet Network Adapters that
66 support SR-IOV, the virtual function (VF) may be subject to malicious behavior.
67 Software-generated layer two frames, like IEEE 802.3x (link flow control), IEEE
68 802.1Qbb (priority based flow-control), and others of this type, are not
69 expected and can throttle traffic between the host and the virtual switch,
70 reducing performance. To resolve this issue, and to ensure isolation from
71 unintended traffic streams, configure all SR-IOV enabled ports for VLAN tagging
72 from the administrative interface on the PF. This configuration allows
73 unexpected, and potentially malicious, frames to be dropped.
75 See "Configuring VLAN Tagging on SR-IOV Enabled Adapter Ports" later in this
76 README for configuration instructions.
79 Do not unload port driver if VF with active VM is bound to it
80 -------------------------------------------------------------
81 Do not unload a port's driver if a Virtual Function (VF) with an active Virtual
82 Machine (VM) is bound to it. Doing so will cause the port to appear to hang.
83 Once the VM shuts down, or otherwise releases the VF, the command will
87 Important notes for SR-IOV and Link Aggregation
88 -----------------------------------------------
89 Link Aggregation is mutually exclusive with SR-IOV.
91 - If Link Aggregation is active, SR-IOV VFs cannot be created on the PF.
92 - If SR-IOV is active, you cannot set up Link Aggregation on the interface.
94 Bridging and MACVLAN are also affected by this. If you wish to use bridging or
95 MACVLAN with SR-IOV, you must set up bridging or MACVLAN before enabling
96 SR-IOV. If you are using bridging or MACVLAN in conjunction with SR-IOV, and
97 you want to remove the interface from the bridge or MACVLAN, you must follow
100 1. Destroy SR-IOV VFs if they exist
101 2. Remove the interface from the bridge or MACVLAN
102 3. Recreate SRIOV VFs as needed
105 Additional Features and Configurations
106 ======================================
110 The driver utilizes the ethtool interface for driver configuration and
111 diagnostics, as well as displaying statistical information. The latest ethtool
112 version is required for this functionality. Download it at:
113 https://kernel.org/pub/software/network/ethtool/
115 NOTE: The rx_bytes value of ethtool does not match the rx_bytes value of
116 Netdev, due to the 4-byte CRC being stripped by the device. The difference
117 between the two rx_bytes values will be 4 x the number of Rx packets. For
118 example, if Rx packets are 10 and Netdev (software statistics) displays
119 rx_bytes as "X", then ethtool (hardware statistics) will display rx_bytes as
120 "X+40" (4 bytes CRC x 10 packets).
123 Viewing Link Messages
124 ---------------------
125 Link messages will not be displayed to the console if the distribution is
126 restricting system messages. In order to see network driver link messages on
127 your console, set dmesg to eight by entering the following::
131 NOTE: This setting is not saved across reboots.
134 Dynamic Device Personalization
135 ------------------------------
136 Dynamic Device Personalization (DDP) allows you to change the packet processing
137 pipeline of a device by applying a profile package to the device at runtime.
138 Profiles can be used to, for example, add support for new protocols, change
139 existing protocols, or change default settings. DDP profiles can also be rolled
140 back without rebooting the system.
142 The DDP package loads during device initialization. The driver looks for
143 ``intel/ice/ddp/ice.pkg`` in your firmware root (typically ``/lib/firmware/``
144 or ``/lib/firmware/updates/``) and checks that it contains a valid DDP package
147 NOTE: Your distribution should likely have provided the latest DDP file, but if
148 ice.pkg is missing, you can find it in the linux-firmware repository or from
151 If the driver is unable to load the DDP package, the device will enter Safe
152 Mode. Safe Mode disables advanced and performance features and supports only
153 basic traffic and minimal functionality, such as updating the NVM or
154 downloading a new driver or DDP package. Safe Mode only applies to the affected
155 physical function and does not impact any other PFs. See the "Intel(R) Ethernet
156 Adapters and Devices User Guide" for more details on DDP and Safe Mode.
160 - If you encounter issues with the DDP package file, you may need to download
161 an updated driver or DDP package file. See the log messages for more
164 - The ice.pkg file is a symbolic link to the default DDP package file.
166 - You cannot update the DDP package if any PF drivers are already loaded. To
167 overwrite a package, unload all PFs and then reload the driver with the new
170 - Only the first loaded PF per device can download a package for that device.
172 You can install specific DDP package files for different physical devices in
173 the same system. To install a specific DDP package file:
175 1. Download the DDP package file you want for your device.
177 2. Rename the file ice-xxxxxxxxxxxxxxxx.pkg, where 'xxxxxxxxxxxxxxxx' is the
178 unique 64-bit PCI Express device serial number (in hex) of the device you
179 want the package downloaded on. The filename must include the complete
180 serial number (including leading zeros) and be all lowercase. For example,
181 if the 64-bit serial number is b887a3ffffca0568, then the file name would be
182 ice-b887a3ffffca0568.pkg.
184 To find the serial number from the PCI bus address, you can use the
187 # lspci -vv -s af:00.0 | grep -i Serial
188 Capabilities: [150 v1] Device Serial Number b8-87-a3-ff-ff-ca-05-68
190 You can use the following command to format the serial number without the
193 # lspci -vv -s af:00.0 | grep -i Serial | awk '{print $7}' | sed s/-//g
196 3. Copy the renamed DDP package file to
197 ``/lib/firmware/updates/intel/ice/ddp/``. If the directory does not yet
198 exist, create it before copying the file.
200 4. Unload all of the PFs on the device.
202 5. Reload the driver with the new package.
204 NOTE: The presence of a device-specific DDP package file overrides the loading
205 of the default DDP package file (ice.pkg).
208 Intel(R) Ethernet Flow Director
209 -------------------------------
210 The Intel Ethernet Flow Director performs the following tasks:
212 - Directs receive packets according to their flows to different queues
213 - Enables tight control on routing a flow in the platform
214 - Matches flows and CPU cores for flow affinity
216 NOTE: This driver supports the following flow types:
227 Each flow type supports valid combinations of IP addresses (source or
228 destination) and UDP/TCP/SCTP ports (source and destination). You can supply
229 only a source IP address, a source IP address and a destination port, or any
230 combination of one or more of these four parameters.
232 NOTE: This driver allows you to filter traffic based on a user-defined flexible
233 two-byte pattern and offset by using the ethtool user-def and mask fields. Only
234 L3 and L4 flow types are supported for user-defined flexible filters. For a
235 given flow type, you must clear all Intel Ethernet Flow Director filters before
236 changing the input set (for that flow type).
239 Flow Director Filters
240 ---------------------
241 Flow Director filters are used to direct traffic that matches specified
242 characteristics. They are enabled through ethtool's ntuple interface. To enable
243 or disable the Intel Ethernet Flow Director and these filters::
245 # ethtool -K <ethX> ntuple <off|on>
247 NOTE: When you disable ntuple filters, all the user programmed filters are
248 flushed from the driver cache and hardware. All needed filters must be re-added
249 when ntuple is re-enabled.
251 To display all of the active filters::
255 To add a new filter::
257 # ethtool -U <ethX> flow-type <type> src-ip <ip> [m <ip_mask>] dst-ip <ip>
258 [m <ip_mask>] src-port <port> [m <port_mask>] dst-port <port> [m <port_mask>]
262 <ethX> - the Ethernet device to program
263 <type> - can be ip4, tcp4, udp4, sctp4, ip6, tcp6, udp6, sctp6
264 <ip> - the IP address to match on
265 <ip_mask> - the IPv4 address to mask on
266 NOTE: These filters use inverted masks.
267 <port> - the port number to match on
268 <port_mask> - the 16-bit integer for masking
269 NOTE: These filters use inverted masks.
270 <queue> - the queue to direct traffic toward (-1 discards the
275 # ethtool -U <ethX> delete <N>
277 Where <N> is the filter ID displayed when printing all the active filters,
278 and may also have been specified using "loc <N>" when adding the filter.
282 To add a filter that directs packet to queue 2::
284 # ethtool -U <ethX> flow-type tcp4 src-ip 192.168.10.1 dst-ip \
285 192.168.10.2 src-port 2000 dst-port 2001 action 2 [loc 1]
287 To set a filter using only the source and destination IP address::
289 # ethtool -U <ethX> flow-type tcp4 src-ip 192.168.10.1 dst-ip \
290 192.168.10.2 action 2 [loc 1]
292 To set a filter based on a user-defined pattern and offset::
294 # ethtool -U <ethX> flow-type tcp4 src-ip 192.168.10.1 dst-ip \
295 192.168.10.2 user-def 0x4FFFF action 2 [loc 1]
297 where the value of the user-def field contains the offset (4 bytes) and
298 the pattern (0xffff).
300 To match TCP traffic sent from 192.168.0.1, port 5300, directed to 192.168.0.5,
301 port 80, and then send it to queue 7::
303 # ethtool -U enp130s0 flow-type tcp4 src-ip 192.168.0.1 dst-ip 192.168.0.5
304 src-port 5300 dst-port 80 action 7
306 To add a TCPv4 filter with a partial mask for a source IP subnet::
308 # ethtool -U <ethX> flow-type tcp4 src-ip 192.168.0.0 m 0.255.255.255 dst-ip
309 192.168.5.12 src-port 12600 dst-port 31 action 12
313 For each flow-type, the programmed filters must all have the same matching
314 input set. For example, issuing the following two commands is acceptable::
316 # ethtool -U enp130s0 flow-type ip4 src-ip 192.168.0.1 src-port 5300 action 7
317 # ethtool -U enp130s0 flow-type ip4 src-ip 192.168.0.5 src-port 55 action 10
319 Issuing the next two commands, however, is not acceptable, since the first
320 specifies src-ip and the second specifies dst-ip::
322 # ethtool -U enp130s0 flow-type ip4 src-ip 192.168.0.1 src-port 5300 action 7
323 # ethtool -U enp130s0 flow-type ip4 dst-ip 192.168.0.5 src-port 55 action 10
325 The second command will fail with an error. You may program multiple filters
326 with the same fields, using different values, but, on one device, you may not
327 program two tcp4 filters with different matching fields.
329 The ice driver does not support matching on a subportion of a field, thus
330 partial mask fields are not supported.
333 Flex Byte Flow Director Filters
334 -------------------------------
335 The driver also supports matching user-defined data within the packet payload.
336 This flexible data is specified using the "user-def" field of the ethtool
337 command in the following way:
341 ============================== ============================
342 ``31 28 24 20 16`` ``15 12 8 4 0``
343 ``offset into packet payload`` ``2 bytes of flexible data``
344 ============================== ============================
350 ... user-def 0x4FFFF ...
352 tells the filter to look 4 bytes into the payload and match that value against
353 0xFFFF. The offset is based on the beginning of the payload, and not the
354 beginning of the packet. Thus
358 flow-type tcp4 ... user-def 0x8BEAF ...
360 would match TCP/IPv4 packets which have the value 0xBEAF 8 bytes into the
363 Note that ICMP headers are parsed as 4 bytes of header and 4 bytes of payload.
364 Thus to match the first byte of the payload, you must actually add 4 bytes to
365 the offset. Also note that ip4 filters match both ICMP frames as well as raw
366 (unknown) ip4 frames, where the payload will be the L3 payload of the IP4
369 The maximum offset is 64. The hardware will only read up to 64 bytes of data
370 from the payload. The offset must be even because the flexible data is 2 bytes
371 long and must be aligned to byte 0 of the packet payload.
373 The user-defined flexible offset is also considered part of the input set and
374 cannot be programmed separately for multiple filters of the same type. However,
375 the flexible data is not part of the input set and multiple filters may use the
376 same offset but match against different data.
381 Allows you to set the hash bytes per flow type and any combination of one or
382 more options for Receive Side Scaling (RSS) hash byte configuration.
386 # ethtool -N <ethX> rx-flow-hash <type> <option>
389 tcp4 signifying TCP over IPv4
390 udp4 signifying UDP over IPv4
391 tcp6 signifying TCP over IPv6
392 udp6 signifying UDP over IPv6
393 And <option> is one or more of:
394 s Hash on the IP source address of the Rx packet.
395 d Hash on the IP destination address of the Rx packet.
396 f Hash on bytes 0 and 1 of the Layer 4 header of the Rx packet.
397 n Hash on bytes 2 and 3 of the Layer 4 header of the Rx packet.
400 Accelerated Receive Flow Steering (aRFS)
401 ----------------------------------------
402 Devices based on the Intel(R) Ethernet Controller 800 Series support
403 Accelerated Receive Flow Steering (aRFS) on the PF. aRFS is a load-balancing
404 mechanism that allows you to direct packets to the same CPU where an
405 application is running or consuming the packets in that flow.
409 - aRFS requires that ntuple filtering is enabled via ethtool.
410 - aRFS support is limited to the following packet types:
412 - TCP over IPv4 and IPv6
413 - UDP over IPv4 and IPv6
414 - Nonfragmented packets
416 - aRFS only supports Flow Director filters, which consist of the
417 source/destination IP addresses and source/destination ports.
418 - aRFS and ethtool's ntuple interface both use the device's Flow Director. aRFS
419 and ntuple features can coexist, but you may encounter unexpected results if
420 there's a conflict between aRFS and ntuple requests. See "Intel(R) Ethernet
421 Flow Director" for additional information.
425 1. Enable the Intel Ethernet Flow Director and ntuple filters using ethtool.
429 # ethtool -K <ethX> ntuple on
431 2. Set up the number of entries in the global flow table. For example:
435 # NUM_RPS_ENTRIES=16384
436 # echo $NUM_RPS_ENTRIES > /proc/sys/net/core/rps_sock_flow_entries
438 3. Set up the number of entries in the per-queue flow table. For example:
443 # for file in /sys/class/net/$IFACE/queues/rx-*/rps_flow_cnt; do
444 # echo $(($NUM_RPS_ENTRIES/$NUM_RX_QUEUES)) > $file;
447 4. Disable the IRQ balance daemon (this is only a temporary stop of the service
448 until the next reboot).
452 # systemctl stop irqbalance
454 5. Configure the interrupt affinity.
456 See ``/Documentation/core-api/irq/irq-affinity.rst``
459 To disable aRFS using ethtool::
461 # ethtool -K <ethX> ntuple off
463 NOTE: This command will disable ntuple filters and clear any aRFS filters in
464 software and hardware.
468 1. Set the server application on the desired CPU (e.g., CPU 4).
472 # taskset -c 4 netserver
474 2. Use netperf to route traffic from the client to CPU 4 on the server with
475 aRFS configured. This example uses TCP over IPv4.
479 # netperf -H <Host IPv4 Address> -t TCP_STREAM
482 Enabling Virtual Functions (VFs)
483 --------------------------------
484 Use sysfs to enable virtual functions (VF).
486 For example, you can create 4 VFs as follows::
488 # echo 4 > /sys/class/net/<ethX>/device/sriov_numvfs
490 To disable VFs, write 0 to the same file::
492 # echo 0 > /sys/class/net/<ethX>/device/sriov_numvfs
494 The maximum number of VFs for the ice driver is 256 total (all ports). To check
495 how many VFs each PF supports, use the following command::
497 # cat /sys/class/net/<ethX>/device/sriov_totalvfs
499 Note: You cannot use SR-IOV when link aggregation (LAG)/bonding is active, and
500 vice versa. To enforce this, the driver checks for this mutual exclusion.
503 Displaying VF Statistics on the PF
504 ----------------------------------
505 Use the following command to display the statistics for the PF and its VFs::
507 # ip -s link show dev <ethX>
509 NOTE: The output of this command can be very large due to the maximum number of
512 The PF driver will display a subset of the statistics for the PF and for all
513 VFs that are configured. The PF will always print a statistics block for each
514 of the possible VFs, and it will show zero for all unconfigured VFs.
517 Configuring VLAN Tagging on SR-IOV Enabled Adapter Ports
518 --------------------------------------------------------
519 To configure VLAN tagging for the ports on an SR-IOV enabled adapter, use the
520 following command. The VLAN configuration should be done before the VF driver
521 is loaded or the VM is booted. The VF is not aware of the VLAN tag being
522 inserted on transmit and removed on received frames (sometimes called "port
527 # ip link set dev <ethX> vf <id> vlan <vlan id>
529 For example, the following will configure PF eth0 and the first VF on VLAN 10::
531 # ip link set dev eth0 vf 0 vlan 10
534 Enabling a VF link if the port is disconnected
535 ----------------------------------------------
536 If the physical function (PF) link is down, you can force link up (from the
537 host PF) on any virtual functions (VF) bound to the PF.
539 For example, to force link up on VF 0 bound to PF eth0::
541 # ip link set eth0 vf 0 state enable
543 Note: If the command does not work, it may not be supported by your system.
546 Setting the MAC Address for a VF
547 --------------------------------
548 To change the MAC address for the specified VF::
550 # ip link set <ethX> vf 0 mac <address>
554 # ip link set <ethX> vf 0 mac 00:01:02:03:04:05
556 This setting lasts until the PF is reloaded.
558 NOTE: Assigning a MAC address for a VF from the host will disable any
559 subsequent requests to change the MAC address from within the VM. This is a
560 security feature. The VM is not aware of this restriction, so if this is
561 attempted in the VM, it will trigger MDD events.
564 Trusted VFs and VF Promiscuous Mode
565 -----------------------------------
566 This feature allows you to designate a particular VF as trusted and allows that
567 trusted VF to request selective promiscuous mode on the Physical Function (PF).
569 To set a VF as trusted or untrusted, enter the following command in the
572 # ip link set dev <ethX> vf 1 trust [on|off]
574 NOTE: It's important to set the VF to trusted before setting promiscuous mode.
575 If the VM is not trusted, the PF will ignore promiscuous mode requests from the
576 VF. If the VM becomes trusted after the VF driver is loaded, you must make a
577 new request to set the VF to promiscuous.
579 Once the VF is designated as trusted, use the following commands in the VM to
580 set the VF to promiscuous mode.
582 For promiscuous all::
584 # ip link set <ethX> promisc on
585 Where <ethX> is a VF interface in the VM
587 For promiscuous Multicast::
589 # ip link set <ethX> allmulticast on
590 Where <ethX> is a VF interface in the VM
592 NOTE: By default, the ethtool private flag vf-true-promisc-support is set to
593 "off," meaning that promiscuous mode for the VF will be limited. To set the
594 promiscuous mode for the VF to true promiscuous and allow the VF to see all
595 ingress traffic, use the following command::
597 # ethtool --set-priv-flags <ethX> vf-true-promisc-support on
599 The vf-true-promisc-support private flag does not enable promiscuous mode;
600 rather, it designates which type of promiscuous mode (limited or true) you will
601 get when you enable promiscuous mode using the ip link commands above. Note
602 that this is a global setting that affects the entire device. However, the
603 vf-true-promisc-support private flag is only exposed to the first PF of the
604 device. The PF remains in limited promiscuous mode regardless of the
605 vf-true-promisc-support setting.
607 Next, add a VLAN interface on the VF interface. For example::
609 # ip link add link eth2 name eth2.100 type vlan id 100
611 Note that the order in which you set the VF to promiscuous mode and add the
612 VLAN interface does not matter (you can do either first). The result in this
613 example is that the VF will get all traffic that is tagged with VLAN 100.
616 Malicious Driver Detection (MDD) for VFs
617 ----------------------------------------
618 Some Intel Ethernet devices use Malicious Driver Detection (MDD) to detect
619 malicious traffic from the VF and disable Tx/Rx queues or drop the offending
620 packet until a VF driver reset occurs. You can view MDD messages in the PF's
621 system log using the dmesg command.
623 - If the PF driver logs MDD events from the VF, confirm that the correct VF
625 - To restore functionality, you can manually reload the VF or VM or enable
627 - When automatic VF resets are enabled, the PF driver will immediately reset
628 the VF and reenable queues when it detects MDD events on the receive path.
629 - If automatic VF resets are disabled, the PF will not automatically reset the
630 VF when it detects MDD events.
632 To enable or disable automatic VF resets, use the following command::
634 # ethtool --set-priv-flags <ethX> mdd-auto-reset-vf on|off
637 MAC and VLAN Anti-Spoofing Feature for VFs
638 ------------------------------------------
639 When a malicious driver on a Virtual Function (VF) interface attempts to send a
640 spoofed packet, it is dropped by the hardware and not transmitted.
642 NOTE: This feature can be disabled for a specific VF::
644 # ip link set <ethX> vf <vf id> spoofchk {off|on}
649 Jumbo Frames support is enabled by changing the Maximum Transmission Unit (MTU)
650 to a value larger than the default value of 1500.
652 Use the ifconfig command to increase the MTU size. For example, enter the
653 following where <ethX> is the interface number::
655 # ifconfig <ethX> mtu 9000 up
657 Alternatively, you can use the ip command as follows::
659 # ip link set mtu 9000 dev <ethX>
660 # ip link set up dev <ethX>
662 This setting is not saved across reboots.
665 NOTE: The maximum MTU setting for jumbo frames is 9702. This corresponds to the
666 maximum jumbo frame size of 9728 bytes.
668 NOTE: This driver will attempt to use multiple page sized buffers to receive
669 each jumbo packet. This should help to avoid buffer starvation issues when
670 allocating receive packets.
672 NOTE: Packet loss may have a greater impact on throughput when you use jumbo
673 frames. If you observe a drop in performance after enabling jumbo frames,
674 enabling flow control may mitigate the issue.
677 Speed and Duplex Configuration
678 ------------------------------
679 In addressing speed and duplex configuration issues, you need to distinguish
680 between copper-based adapters and fiber-based adapters.
682 In the default mode, an Intel(R) Ethernet Network Adapter using copper
683 connections will attempt to auto-negotiate with its link partner to determine
684 the best setting. If the adapter cannot establish link with the link partner
685 using auto-negotiation, you may need to manually configure the adapter and link
686 partner to identical settings to establish link and pass packets. This should
687 only be needed when attempting to link with an older switch that does not
688 support auto-negotiation or one that has been forced to a specific speed or
689 duplex mode. Your link partner must match the setting you choose. 1 Gbps speeds
690 and higher cannot be forced. Use the autonegotiation advertising setting to
691 manually set devices for 1 Gbps and higher.
693 Speed, duplex, and autonegotiation advertising are configured through the
694 ethtool utility. For the latest version, download and install ethtool from the
697 https://kernel.org/pub/software/network/ethtool/
699 To see the speed configurations your device supports, run the following::
703 Caution: Only experienced network administrators should force speed and duplex
704 or change autonegotiation advertising manually. The settings at the switch must
705 always match the adapter settings. Adapter performance may suffer or your
706 adapter may not operate if you configure the adapter differently from your
710 Data Center Bridging (DCB)
711 --------------------------
712 NOTE: The kernel assumes that TC0 is available, and will disable Priority Flow
713 Control (PFC) on the device if TC0 is not available. To fix this, ensure TC0 is
714 enabled when setting up DCB on your switch.
716 DCB is a configuration Quality of Service implementation in hardware. It uses
717 the VLAN priority tag (802.1p) to filter traffic. That means that there are 8
718 different priorities that traffic can be filtered into. It also enables
719 priority flow control (802.1Qbb) which can limit or eliminate the number of
720 dropped packets during network stress. Bandwidth can be allocated to each of
721 these priorities, which is enforced at the hardware level (802.1Qaz).
723 DCB is normally configured on the network using the DCBX protocol (802.1Qaz), a
724 specialization of LLDP (802.1AB). The ice driver supports the following
725 mutually exclusive variants of DCBX support:
727 1) Firmware-based LLDP Agent
728 2) Software-based LLDP Agent
730 In firmware-based mode, firmware intercepts all LLDP traffic and handles DCBX
731 negotiation transparently for the user. In this mode, the adapter operates in
732 "willing" DCBX mode, receiving DCB settings from the link partner (typically a
733 switch). The local user can only query the negotiated DCB configuration. For
734 information on configuring DCBX parameters on a switch, please consult the
735 switch manufacturer's documentation.
737 In software-based mode, LLDP traffic is forwarded to the network stack and user
738 space, where a software agent can handle it. In this mode, the adapter can
739 operate in either "willing" or "nonwilling" DCBX mode and DCB configuration can
740 be both queried and set locally. This mode requires the FW-based LLDP Agent to
745 - You can enable and disable the firmware-based LLDP Agent using an ethtool
746 private flag. Refer to the "FW-LLDP (Firmware Link Layer Discovery Protocol)"
747 section in this README for more information.
748 - In software-based DCBX mode, you can configure DCB parameters using software
749 LLDP/DCBX agents that interface with the Linux kernel's DCB Netlink API. We
750 recommend using OpenLLDP as the DCBX agent when running in software mode. For
751 more information, see the OpenLLDP man pages and
752 https://github.com/intel/openlldp.
753 - The driver implements the DCB netlink interface layer to allow the user space
754 to communicate with the driver and query DCB configuration for the port.
755 - iSCSI with DCB is not supported.
758 FW-LLDP (Firmware Link Layer Discovery Protocol)
759 ------------------------------------------------
760 Use ethtool to change FW-LLDP settings. The FW-LLDP setting is per port and
761 persists across boots.
765 # ethtool --set-priv-flags <ethX> fw-lldp-agent on
769 # ethtool --set-priv-flags <ethX> fw-lldp-agent off
771 To check the current LLDP setting::
773 # ethtool --show-priv-flags <ethX>
775 NOTE: You must enable the UEFI HII "LLDP Agent" attribute for this setting to
776 take effect. If "LLDP AGENT" is set to disabled, you cannot enable it from the
782 Ethernet Flow Control (IEEE 802.3x) can be configured with ethtool to enable
783 receiving and transmitting pause frames for ice. When transmit is enabled,
784 pause frames are generated when the receive packet buffer crosses a predefined
785 threshold. When receive is enabled, the transmit unit will halt for the time
786 delay specified when a pause frame is received.
788 NOTE: You must have a flow control capable link partner.
790 Flow Control is disabled by default.
792 Use ethtool to change the flow control settings.
794 To enable or disable Rx or Tx Flow Control::
796 # ethtool -A <ethX> rx <on|off> tx <on|off>
798 Note: This command only enables or disables Flow Control if auto-negotiation is
799 disabled. If auto-negotiation is enabled, this command changes the parameters
800 used for auto-negotiation with the link partner.
802 Note: Flow Control auto-negotiation is part of link auto-negotiation. Depending
803 on your device, you may not be able to change the auto-negotiation setting.
807 - The ice driver requires flow control on both the port and link partner. If
808 flow control is disabled on one of the sides, the port may appear to hang on
810 - You may encounter issues with link-level flow control (LFC) after disabling
811 DCB. The LFC status may show as enabled but traffic is not paused. To resolve
812 this issue, disable and reenable LFC using ethtool::
814 # ethtool -A <ethX> rx off tx off
815 # ethtool -A <ethX> rx on tx on
820 This driver supports NAPI (Rx polling mode).
821 For more information on NAPI, see
822 https://www.linuxfoundation.org/collaborate/workgroups/networking/napi
827 This driver supports MACVLAN. Kernel support for MACVLAN can be tested by
828 checking if the MACVLAN driver is loaded. You can run 'lsmod | grep macvlan' to
829 see if the MACVLAN driver is loaded or run 'modprobe macvlan' to try to load
834 - In passthru mode, you can only set up one MACVLAN device. It will inherit the
835 MAC address of the underlying PF (Physical Function) device.
838 IEEE 802.1ad (QinQ) Support
839 ---------------------------
840 The IEEE 802.1ad standard, informally known as QinQ, allows for multiple VLAN
841 IDs within a single Ethernet frame. VLAN IDs are sometimes referred to as
842 "tags," and multiple VLAN IDs are thus referred to as a "tag stack." Tag stacks
843 allow L2 tunneling and the ability to segregate traffic within a particular
844 VLAN ID, among other uses.
848 - Receive checksum offloads and VLAN acceleration are not supported for 802.1ad
851 - 0x88A8 traffic will not be received unless VLAN stripping is disabled with
852 the following command::
854 # ethtool -K <ethX> rxvlan off
856 - 0x88A8/0x8100 double VLANs cannot be used with 0x8100 or 0x8100/0x8100 VLANS
857 configured on the same port. 0x88a8/0x8100 traffic will not be received if
858 0x8100 VLANs are configured.
860 - The VF can only transmit 0x88A8/0x8100 (i.e., 802.1ad/802.1Q) traffic if:
862 1) The VF is not assigned a port VLAN.
863 2) spoofchk is disabled from the PF. If you enable spoofchk, the VF will
864 not transmit 0x88A8/0x8100 traffic.
866 - The VF may not receive all network traffic based on the Inner VLAN header
867 when VF true promiscuous mode (vf-true-promisc-support) and double VLANs are
868 enabled in SR-IOV mode.
870 The following are examples of how to configure 802.1ad (QinQ)::
872 # ip link add link eth0 eth0.24 type vlan proto 802.1ad id 24
873 # ip link add link eth0.24 eth0.24.371 type vlan proto 802.1Q id 371
875 Where "24" and "371" are example VLAN IDs.
878 Tunnel/Overlay Stateless Offloads
879 ---------------------------------
880 Supported tunnels and overlays include VXLAN, GENEVE, and others depending on
881 hardware and software configuration. Stateless offloads are enabled by default.
883 To view the current state of all offloads::
888 UDP Segmentation Offload
889 ------------------------
890 Allows the adapter to offload transmit segmentation of UDP packets with
891 payloads up to 64K into valid Ethernet frames. Because the adapter hardware is
892 able to complete data segmentation much faster than operating system software,
893 this feature may improve transmission performance.
894 In addition, the adapter may use fewer CPU resources.
898 - The application sending UDP packets must support UDP segmentation offload.
900 To enable/disable UDP Segmentation Offload, issue the following command::
902 # ethtool -K <ethX> tx-udp-segmentation [off|on]
905 Performance Optimization
906 ========================
907 Driver defaults are meant to fit a wide variety of workloads, but if further
908 optimization is required, we recommend experimenting with the following
912 Rx Descriptor Ring Size
913 -----------------------
914 To reduce the number of Rx packet discards, increase the number of Rx
915 descriptors for each Rx ring using ethtool.
917 Check if the interface is dropping Rx packets due to buffers being full
918 (rx_dropped.nic can mean that there is no PCIe bandwidth)::
920 # ethtool -S <ethX> | grep "rx_dropped"
922 If the previous command shows drops on queues, it may help to increase
923 the number of descriptors using 'ethtool -G'::
925 # ethtool -G <ethX> rx <N>
926 Where <N> is the desired number of ring entries/descriptors
928 This can provide temporary buffering for issues that create latency while
929 the CPUs process descriptors.
932 Interrupt Rate Limiting
933 -----------------------
934 This driver supports an adaptive interrupt throttle rate (ITR) mechanism that
935 is tuned for general workloads. The user can customize the interrupt rate
936 control for specific workloads, via ethtool, adjusting the number of
937 microseconds between interrupts.
939 To set the interrupt rate manually, you must disable adaptive mode::
941 # ethtool -C <ethX> adaptive-rx off adaptive-tx off
943 For lower CPU utilization:
945 Disable adaptive ITR and lower Rx and Tx interrupts. The examples below
946 affect every queue of the specified interface.
948 Setting rx-usecs and tx-usecs to 80 will limit interrupts to about
949 12,500 interrupts per second per queue::
951 # ethtool -C <ethX> adaptive-rx off adaptive-tx off rx-usecs 80 tx-usecs 80
955 Disable adaptive ITR and ITR by setting rx-usecs and tx-usecs to 0
958 # ethtool -C <ethX> adaptive-rx off adaptive-tx off rx-usecs 0 tx-usecs 0
960 Per-queue interrupt rate settings:
962 The following examples are for queues 1 and 3, but you can adjust other
965 To disable Rx adaptive ITR and set static Rx ITR to 10 microseconds or
966 about 100,000 interrupts/second, for queues 1 and 3::
968 # ethtool --per-queue <ethX> queue_mask 0xa --coalesce adaptive-rx off
971 To show the current coalesce settings for queues 1 and 3::
973 # ethtool --per-queue <ethX> queue_mask 0xa --show-coalesce
975 Bounding interrupt rates using rx-usecs-high:
977 :Valid Range: 0-236 (0=no limit)
979 The range of 0-236 microseconds provides an effective range of 4,237 to
980 250,000 interrupts per second. The value of rx-usecs-high can be set
981 independently of rx-usecs and tx-usecs in the same ethtool command, and is
982 also independent of the adaptive interrupt moderation algorithm. The
983 underlying hardware supports granularity in 4-microsecond intervals, so
984 adjacent values may result in the same interrupt rate.
986 The following command would disable adaptive interrupt moderation, and allow
987 a maximum of 5 microseconds before indicating a receive or transmit was
988 complete. However, instead of resulting in as many as 200,000 interrupts per
989 second, it limits total interrupts per second to 50,000 via the rx-usecs-high
994 # ethtool -C <ethX> adaptive-rx off adaptive-tx off rx-usecs-high 20
995 rx-usecs 5 tx-usecs 5
998 Virtualized Environments
999 ------------------------
1000 In addition to the other suggestions in this section, the following may be
1001 helpful to optimize performance in VMs.
1003 Using the appropriate mechanism (vcpupin) in the VM, pin the CPUs to
1004 individual LCPUs, making sure to use a set of CPUs included in the
1005 device's local_cpulist: ``/sys/class/net/<ethX>/device/local_cpulist``.
1007 Configure as many Rx/Tx queues in the VM as available. (See the iavf driver
1008 documentation for the number of queues supported.) For example::
1010 # ethtool -L <virt_interface> rx <max> tx <max>
1015 For general information, go to the Intel support website at:
1016 https://www.intel.com/support/
1018 or the Intel Wired Networking project hosted by Sourceforge at:
1019 https://sourceforge.net/projects/e1000
1021 If an issue is identified with the released source code on a supported kernel
1022 with a supported adapter, email the specific information related to the issue
1023 to e1000-devel@lists.sf.net.
1028 Intel is a trademark or registered trademark of Intel Corporation or its
1029 subsidiaries in the United States and/or other countries.
1031 * Other names and brands may be claimed as the property of others.