Known issues: MLNX_EN v5.8-1.1.2.1 LTS (2023)

Workaround:Remove (and clean: apt purge / dpkg --purge) the openvswitch-switch package before installing, or if you encounter this error message, remove the remaining file and continue with the installation:
rm-Datei /etc/systemd/system/openvswitch-switch.service.requires/ovs-record-hostname.service
apt install -f

version fixed: 5,8-1.0.1.1

Workaround:Set eSwitch mode to Legacy before enabling SR-IOV or reload mlx5_core to change eSwitch mode to Legacy.

version fixed: 5,8-1.0.1.1

3199628

version fixed: 5,8-1.0.1.1

Workaround:Set the jumbo MTU (9216) on all interfaces related to the uplink.

version fixed: 5,8-1.0.1.1

Workaround:N / D

version fixed: 5,8-1.0.1.1

Workaround:Remove the original mlnx-ofed-all-5.* package and recreate the repository with: createrepo RPMS/

version fixed: 5,8-1.0.1.1

Description:The neohost-backend package (included in MLNX_OFED) strictly depends on Python 2.7 and the existence of Python/usr/bin/python. This dependency is due to a pre-install test (which is not a standard practice) so /usr/bin/python fails to install if Python 2.7 is not present.

For this reason, the default installation has been disabled on newer systems that do not have Python 2 installed by default.

If there is an explicit requirement for this installation, use the command line option--con-back-end from neohost, this sanity check will be overridden and an attempt will be made to install it independently. May not be on newer systems/usr/bin/pythoneven if Python 2 is installed; Therefore, your installation fails.

Workaround:If neohost-backend is needed on a newer system, install Python 2 in advance and create the symlink/usr/bin/python-> Python2.

version fixed: 5,8-1.0.1.1

Workaround:N / D

version fixed: 5,8-1.0.1.1

Workaround:Install OFED with--add-kernel-supportBandera.

version fixed: 5,8-1.0.1.1

Workaround:Choose a name that meets the 16-character limit.

version fixed: 5,8-1.0.1.1

Workaround:N / D

version fixed: 5,8-1.0.1.1

Workaround:N / D

version fixed: 5,8-1.0.1.1

3114823

Discovered at startup: 5.7-1.0.2.0

Workaround:Installing chkconfig before OFED.

Discovered at startup: 5.7-1.0.2.0

This was caused by a bug in the scripts that create the Requires and Provides headers, which get confused by dependencies between different modules in the same external package.

Workaround:dnf install kernel-modules-<kernel-version> # if not the latest.

Discovered at startup: 5.7-1.0.2.0

Workaround:N / D

Discovered at startup: 5.7-1.0.2.0

Workaround:Install Python before starting the build.

Discovered at startup: 5.7-1.0.2.0

Workaround:Build and install MOFED with KMP disabled (no --kmp flag).

Discovered at startup: 5.7-1.0.2.0

Workaround:Build and install MOFED with KMP disabled (no --kmp flag).

Discovered at startup: 5.7-1.0.2.0

Workaround:N / D

Discovered at startup: 5.7-1.0.2.0

Workaround:N / D

Discovered at startup: 5.7-1.0.2.0

Workaround:N / D

Discovered at startup: 5.7-1.0.2.0

This is known to work fine and not fail on the RHEL7.x kernel, regardless of the userland package used (including the RHEL8.x userland).

Workaround:N / D

Discovered at startup: 5.6-1.0.3.5

1. When offloading is enabled, removing a physical port from the ovs-dpdk bridge requires a restart of the OVS service. Otherwise, the data path rules will be misconfigured.
2. When offloading is enabled, the physical port must be connected to a bridge.

Discovered at startup: 5.6-1.0.3.5

Workaround:N / D

Discovered at startup: 5.6-1.0.3.5

Workaround:Download the mlx5_ib module before going into switchdev mode.

Discovered at startup: 5.6-1.0.3.5

2811957

Workaround:N / D

Discovered at startup: 5.6-1.0.3.5

2979243

Workaround:Use the kernel support plugin.

Discovered at startup: 5.6-1.0.3.5

3011440

Workaround:Also, install mstflint in the exact same version (e.g. apt install mlnx-ofed-all-exact mstflint=4.16.0-1.56xxxx).

Discovered at startup: 5.6-1.0.3.5

3024520

Workaround:If necessary, delete files.

Discovered at startup: 5.6-1.0.3.5

3046601

Discovered at startup: 5.6-1.0.3.5

3046655

Discovered at startup: 5.6-1.0.3.5

3048411

Discovered at startup: 5.6-1.0.3.5

3048444

Discovered at startup: 5.6-1.0.3.5

3015210

Discovered at startup: 5.6-1.0.3.5

3028300

Discovered at startup: 5.6-1.0.3.5

3044255

Discovered at startup: 5.6-1.0.3.5

3046456

Discovered at startup: 5.6-1.0.3.5

2934149

Discovered at startup: 5.6-1.0.3.5

2901514

Discovered at startup: 5.6-1.0.3.5

2688191

Discovered at startup:5.4-1.0.3.0

2870299

Discovered at startup:5.5-1.0.3.2

2869722

Discovered at startup:5.5-1.0.3.2

2851639

Discovered at startup:5.5-1.0.3.2

2851639

Discovered at startup:5.5-1.0.3.2

2860855

Discovered at startup:5.5-1.0.3.2

2865983

Discovered at startup:5.5-1.0.3.2

2658644

Discovered at startup:5.4-1.0.3.0

2706345

Discovered at startup:5.4-1.0.3.0

2836997

To choose the best configuration parameters for a flow meter, run a few test measurements with different burst size values ​​versus the expected traffic load and average the results over a long period of time (tens of minutes).

Discovered at startup:5.4-1.0.3.0

2863456

Discovered at startup:5.4-0.5.1.1

2657392

Discovered at startup:5.4-0.5.1.1

2800993

2399503

2657392

Discovered at startup:5.4-0.5.1.1

2782406

Discovered at startup:5.4-3.0.3.0

2755632

Discovered at startup:5.4-3.0.3.0

2780436

Discovered at startup:5.4-3.0.3.0

2726021

Discovered at startup:5.4-3.0.3.0

2750653

Discovered at startup:5.4-1.0.3.0

2629375

Discovered at startup:5.4-1.0.3.0

2707997

Discovered at startup:5.4-1.0.3.0

2708531

Discovered at startup:5.4-1.0.3.0

2703043

Discovered at startup:5.4-1.0.3.0

2676405

Discovered at startup:5.4-1.0.3.0

2687943

Description:Offloading rules that redirect from VFs on one PF to VFs on the second PF are not supported on directly attached devices.

Keywords:as soon as possible², direct socket

Discovered at startup:5.4-1.0.3.0

2678672

Description:Disabling switchdev mode keeps the qdisc on the tunnel device from being destroyed and OVS keeps calling mlx5e_stats_flower() resulting in a NULL pointer panic and a memory leak.

Keywords:SwitchDev, mlx5, Tunnelverkehr

Discovered at startup:5.4-1.0.3.0

2566548

Description:When EEH is enabled on PPC systems, when performing the FW sync reset (using mlxfwreset with the --sync flag 1 or using the devlink dev fw_activate reload action), the EEH can capture the PCI reset and start the transfer in take possession. In the case of multiple executions in a row, the EEH can also decide to deactivate the device.

Keywords:PCP, EEH

Discovered at startup:5.4-1.0.3.0

2617950

Description:The TX port timestamp feature is supported in kernel versions 3.15 and later. In earlier kernel versions, the function is not supported and the ptp_tx<X>_* counters are not increased.

Keywords:Ettool

Discovered at startup:5.4-1.0.3.0

2390731

Description:Ethtool doesn't show advertised port speeds/capacity over 100Gb/s over and under 5.0 kernels even if it's supported.

Keywords:Ethtool port speed

Discovered at startup:5.4-1.0.3.0

2585575

Description:After disabling sync reset by setting enable_remote_dev_reset to false, performing firmware sync reset can sometimes cause a General Protection Fault and the system may hang.

Keywords:Firmware Update

Discovered at startup:5.3-1.0.0.1

2582565

Description:Performing a firmware reset or unbinding PF in switchdev mode can cause a kernel crash.

Keywords:SwitchDev, as soon as possible², Unpair, reset firmware

Discovered at startup:5.3-1.0.0.1

2587802

Description:PTP synchronization can be lost when using the private flag tx_port_ts.

Keywords:PTP-Synchronisation

Discovered at startup:5.3-1.0.0.1

2574943

Discovered at startup: 5.3-1.0.0.1

2568417

Discovered at startup: 5.3-1.0.0.1

2506425

Discovered at startup: 5.3-1.0.0.1

2492509

Discovered at startup: 5.3-1.0.0.1

2479327

Discovered at startup: 5.3-1.0.0.1

2584441

Discovered at startup: 5.3-1.0.0.1

2460865

Discovered at startup: 5.3-1.0.0.1

2383318

Discovered at startup: 5.3-1.0.0.1

2575647

Discovered at startup: 5.3-1.0.0.1

2395082

Discovered at startup: 5.2-1.0.4.0

2209987

Discovered at startup: 5.1-1.0.4.0

2248996

Discovered at startup:5.1-1.0.4.0

2175930

Discovered at startup:5.1-1.0.4.0

2198764

Discovered at startup:5.1-1.0.4.0

2237134

Workaround:Configure OVS to use a single controller thread:

#ovs-vsctl setze Open_vSwitch. other_config:n-handler-threads=1

Discovered at startup:5.1-1.0.4.0

2239894

Discovered at startup:5.1-1.0.4.0

2240671

Discovered at startup:5.1-1.0.4.0

2242546

Discovered at startup:5.1-1.0.4.0

2143007

Discovered at startup:5.1-1.0.4.0

2225952

Discovered at startup:5.1-1.0.4.0

2216521

Discovered at startup:5.1-1.0.4.0

2202520

Discovered at startup:5.1-1.0.4.0

2210752

Discovered at startup:5.1-1.0.4.0

2125036/2125031

Discovered at startup:5.1-1.0.4.0

2105447

Discovered at startup:5.1-1.0.4.0

2112251

Description:In kernel 4.10-4.14, when the far end of the Geneve tunnel is defined with IPv6, packets larger than MTU are not fragmented, resulting in no traffic being sent.

Discovered at startup:5.1-1.0.4.0

2102902

Discovered at startup:5.1-1.0.4.0

2111534

Discovered at startup:5.1-1.0.4.0

2094176

Discovered at startup:5.0-1.0.0.0

2044544

Discovered at startup:5.0-1.0.0.0

Discovered at startup:5.0-1.0.0.0

Discovered at startup:5.0-1.0.0.0

2072568

Discovered at startup:5.0-1.0.0.0

2093698

Discovered at startup:5.0-1.0.0.0

2093746

Discovered at startup:5.0-1.0.0.0

2083427

Discovered at startup:5.0-1.0.0.0

2067012

Discovered at startup:5.0-1.0.0.0

2036572

Discovered at startup:5.0-1.0.0.0

1997230

1955352

1979958

1991710

1967866

Description:MLNX_EN does not support XDP functionality in the RedHat 7 operating system, despite RedHat's declared support.

1753629

Description:A binding bug found in the 4.12 and 4.13 kernels can cause a slave to become permanently stuckBOND_LINK_FAILTo express. As a result, the following message may appear in the dmesg:

link: link status for interface eth1 failed, failed in 100ms

1712068

Therefore, with MLNX_EN v4.6-1.0.0.0 and later, it is recommended to use MFT v4.12.0 and later.

1730840

1717428

581631

description: GID entries referenced by a specific User Application cannot be removed while that User Application is running.

alternative solution: N/T

keywords: RoCE, GID

Discovered at startup: 4,5-1.0.1.0

1403313

description: Attempting to allocate an excessive amount of VF per PF on operating systems with kernel versions earlier than v4.15 may fail due to a known kernel issue.

alternative solution: Make sure to update your kernel version to v4.15 or later.

keywords: VF, PF, IOMMU, Núcleo, SO

Discovered at startup: 4,5-1.0.1.0

1521877

descriptionNote: On SLES 12 SP1 operating systems, a kernel trackpoint issue can cause undefined behavior when inserting a kernel module with an incorrect parameter.

alternative solution: N/T

keywords: mlx5-Controller, SLES 12 SP1

Discovered at startup: 4,5-1.0.1.0

504073

descriptionNote: When using ConnectX-5 with LRO systems over PPC, the HCA may experience a back pressure due to delayed PCI writes. In this case, bandwidth can drop from line rate to ~35 Gb/s. Packet loss or dropped frames can also be observed.

alternative solution: Look for a PCI backpressure indicator ("outbound_pci_stalled_wr" counter in ethtools advance). Disabling LRO helps reduce back pressure and its effects.

keywords: Flusskontrolle, LRO

Discovered at startup: 4.4-1.0.0.0

1424233

description: On RHEL v7.3, 7.4 and 7.5 operating systems, configuring IPv4 IP forwarding will disable LRO on existing interfaces. Manually re-enabling LRO with ethtool and adding a VLAN interface can result in warning call tracking.

alternative solution- Make sure that IPv4 IP forwarding and LRO are not enabled at the same time.

keywords: IPv4 forwarding, LRO

Discovered at startup: 4.4-1.0.1.0

1442507

description: Retpoline support in GCC causes an increase in CPU usage, resulting in a 15% performance drop in IP forwarding.

alternative solution: N/T

keywords: Retpoline, GCC, CPU, IP forwarding, Specter attack

Discovered at startup: 4.4-1.0.1.0

1425129

description: MLNX_EN cannot be installed on SLES 15 operating systems using the Zypper repository.

alternative solution: Install MLNX_EN using the default installation script instead of the zypper repository.

keywords: Installation, SLES, Zypper

Discovered at startup: 4.4-1.0.1.0

1241056

description: When working with ConnectX-4/ConnectX-5 HCA on PPC systems with hardware LRO and Adaptive Rx support, bandwidth drops from Full Wire Rate (FWS) to ~60 Gb/s.

alternative solution: Be sure to disable Adaptive Rx when enabling Hardware LRO:ethtool -C <interfaz> adaptive-rx disabled

ethtool -C <interfaz> rx-usecs 8 rx-frames 128

keywords: Hardware-LRO, Adaptive Rx, PPC

Discovered at startup: 4.3-1.0.1.0

1090612

description: The NVMEoF protocol does not support LBA format with non-zero metadata length. Therefore, NVMe namespaces configured in LBA format with a metadata length greater than 0 result in Enhanced Error Handling (EEH) on PowerPC systems.

alternative solution: Configure the NVMe namespace to use the LBA format with zero-length metadata.

keywords: NVMEoF, PowerPC, EEH

Discovered at startup: 4.3-1.0.1.0

1309621

description: In the default switchdev mode configuration, stateless addresses/downloads based on internal headers are not supported.

alternative solution: To enable stateless direction/downloads based on inner headers, disable encap by running:

devlink dev eswitch show pci/0000:83:00.1 encap deshabilitar

Or, if the kernel doesn't support devlink, run:

echo nenhum > /sys/kernel/debug/mlx5/<BDF>/compat/encap

monitoring: This is a hardware limitation.

keywords: switchdev, stateless dump, address

Discovered at startup: 4.3-1.0.1.0

1275082

descriptionNote: When configuring a non-default IPv6 link-local address or an address that is not based on the MAC device, the connection establishment via RoCEv2 may fail.

alternative solution: N/T

keywords: IPV6, RoCE, Link-Local-Adresse

Discovered at startup: 4.3-1.0.1.0

1307336

description: In RoCE LAG mode, while runningibdev2netdev -v, the status of the second port of the mlx4_0 IB device is "NA" because this IB device does not have a second port.

alternative solution: N/T

keywords: mlx4, RoCE LAG, ibdev2netdev, enlace

Discovered at startup: 4.3-1.0.1.0

1296355

description: The total number of MSI-Xs that can be allocated to VF and PF is limited to 2300 on Power9 platforms.

alternative solution: N/T

keywords: MSI-X, VF, PF, PPC, SR-IOV

Discovered at startup: 4.3-1.0.1.0

1259293

description: On Fedora 20 operating systems, loading the driver fails with an error message like the following: "[185.262460] kmem_cache_sanity_check (fs_ftes_0000:00:06.0): cache name already exists.”

This is because SLUB assignments bundle multiple kmem_cache_create boards into one board cache alias to save memory and increase cache performance. This causes the tile name to become obsolete.

alternative solution: Update the kernel version to kernel-3.19.8-100.fc20.x86_64.

Note that after rebooting you will need to rebuild with the new kernel
MLNX_EN against the new kernel version.

keywords: Fedora, load drivers

Discovered at startup: 4.3-1.0.1.0

1264359

description: Running perftest (ib_send_bw, ib_write_bw, etc.) in rdma-cm mode may increase the resp_cqe_error counter in /sys/class/infiniband/mlx5_0/ports/1/hw_counters/resp_cqe_error. This behavior is expected and is the result of receiving WQEs that have not been consumed.

alternative solution: N/T

keywords: most perfect, RDMA CM, mlx5

Discovered at startup: 4.3-1.0.1.0

1264956

descriptionNote: Configuring SR-IOV after disabling RoCE LAG via sysfs (/sys/bus/pci/drivers/mlx5_core/<bdf>/roce_lag_enable) may result in RoCE LAG being re-enabled when SR-IOV -Configuration fails.

alternative solution: Be sure to disable RoCE LAG again.

keywords: RoCE LAG, SR-IOV

Discovered at startup: 4.3-1.0.1.0

1263043

description: On RHEL7.4, due to an OS issue introduced in kmod package version 20-15.el7_4.6, parsing of depmod configuration files fails, resulting in one of the following issues:

• Resetting the controller fails and shows an error message like: "ERROR: module mlx5_core belongs to core and is not part of MLNX_EN, Ignore...”
• Malfunction of the nvmet_rdma kernel module despite installing MLNX_EN with the "--with-nvmf" option. An error message like: "nvmet_rdma: Unknown parameter "offload_mem_start" ignoredwill be featured in itdmesgSalida

alternative solution: Go toRedHat pageto update the kmod package version.

keywords: reset drivers, kmod, kmp, nvmf, nvmet_rdma

Discovered at startup: 4.2-1.2.0.0

-

description: Packet size limitation (actual MTU of the packet) for IPsec offload on Innova IPsec adapter cards: The current offload implementation does not support IP fragmentation. The size of the original packet must be such that it does not exceed the MTU size of the interface after ESP transformation (encryption of the original IP packet, which increases its length) and added headers (outer IP header):

• Internal IP packet size <= I/F MTU - ESP adds (20) - External_IP (20) - Fragmentation problem Reserved length (56)
• Interne IP-Paketgröße <= I/F MTU - 96

This mainly affects traffic forwarded with a lower MTU and UDP traffic. By default, TCP performs PMTU detection and sets the MSS accordingly.

alternative solution: N/T

keywords: Innova IPsec, MTU

Discovered at startup: 4.2-1.0.1.0

-

description: LLC/SNAP is not compatible with Innova IPsec adapter cards.

alternative solution: N/T

keywords: Innova IPsec, LLC/SNAP

Discovered at startup: 4.2-1.0.1.0

-

description: FEC is not supported on Innova IPsec adapter cards. When using switches, it may be necessary to change their settings.

alternative solution: N/T

keywords: Innova IPsec, FEC

Discovered at startup: 4.2-1.0.1.0

955929

description: Heavy traffic can cause SYN floods when using Innova IPsec adapter cards.

alternative solution: N/T

keywords: Innova IPsec, SYN-Flut

Discovered at startup: 4.2-1.0.1.0

-

description: Priority-based flow control is not supported on Innova IPsec adapter cards.

alternative solution: N/T

keywords: Innova IPsec, priority-based flow control

Discovered at startup: 4.2-1.0.1.0

-

description: Stop configuration is not supported when using Innova IPsec adapter cards. The default pause is the global pause (enabled).

alternative solution: N/T

keywords: Renew IPsec, global pause

Discovered at startup: 4.2-1.0.1.0

1045097

description: Plugging and unplugging a cable multiple times can cause connection errors when using Innova IPsec adapter cards.

alternative solution: N/T

keywords: Innova IPsec, cable, connection

Discovered at startup: 4.2-1.0.1.0

-

description: On Innova IPsec adapter cards, the supported MTU is between 512 and 2012 bytes. Setting MTU values ​​outside of this range may fail or result in traffic loss.

alternative solution: Set the MTU between 512 and 2012 bytes.

keywords: Innova IPsec, MTU

Discovered at startup: 4.2-1.0.1.0

1125184

description: On older kernel versions like Ubuntu 14.04 and RedHat 7.1, the VXLAN interface does not respond to ARP requests for an existing MAC address in its own ARP table. This issue has been fixed in the following latest kernel versions: Ubuntu 16.04 and RedHat 7.3.

alternative solution: N/T

keywords: ARP, VXLAN

Discovered at startup: 4.2-1.0.1.0

1134323

descriptionNote: Using kernel versions prior to 4.7 with IOMMU enabled may experience performance degradation and logical issues (e.g. temporary crashes) under heavy traffic loads. This is due to the fact that IOMMU-IOVA allocations are centralized, which requires many synchronization operations and high blocking overhead between CPUs.

alternative solution: Use kernel v4.7 or later or a supported kernel that includes the following patches:

• 2aac630429d9 iommu/vt-d: altere intel-iommu para usar números de quadros IOVA
• 9257b4a206fc iommu/iova: Introduce cache per CPU for iova allocation
• 22e2f9fa63b0 iommu/vt-d: Use IOVA cache for CPU

keywords: IOMMU, bloqueo polite

Discovered at startup: 4.2-1.0.1.0

1135738

description: With a page size setting of 64 KB, the DMA memory may run out when trying to increase the ring size/number of channels.

alternative solution: Reduce ringtone size/number of channels.

keywords: DMA, 64K page

Discovered at startup: 4.2-1.0.1.0

1159650

description: When configuring VF VST, outgoing VLAN tagged packets are dropped in case of ConnectX-4 HCA. In the case of ConnectX-5 HCAs, an additional VLAN tag is inserted for outgoing packets with VLAN tags.

alternative solution: N/T

keywords: VST

Discovered at startup: 4.2-1.0.1.0

1157770

description: On relatively old passthrough/VM machines with QEMU and libvirtd,

CMD may time out after loading the driver.

After the timeout, no more commands are executed and all controller operations are blocked.

alternative solution: Update QEMU and libvirtd on the KVM server.

Tested with (Ubuntu 16.10) are the following versions:

• libvirt 2.1.0
• QEMU 2.6.1

keywords: QEMU

Discovered at startup: 4.2-1.0.1.0

1147703

description: Using dm-multipath for high availability over NVMEoF block devices must be done with the "directio" path verifier.

alternative solution: N/T

keywords: NVMEoF

Discovered at startup: 4.2-1.0.1.0

1152408

description: RedHat v7.3 PPCLE and v7.4 PPCLE operating systems do not support qemu KVM by default. Running it gives the following error messagevirt-installTo create new virtual machines:

I can't find the qemu-kvm package to install

alternative solution: Purchase the following beta RPMs from 7.4ALT to 7.3/7.4 PPCLE (in the same order):

• qemu-img-.el7a.ppc64le.rpm
• qemu-kvm-common-.el7a.ppc64le.rpm
• qemu-kvm-.el7a.ppc64le.rpm

keywords: Virtualization, PPC, Power8, KVM, RedHat, PPC64LE

Discovered at startup: 4.2-1.0.1.0

1012719

description: At very high load of GSI QP (RDMA-CM applications) there may be a slight hang in the CQ polling stream. This is a temporary situation from which the driver will later recover.

alternative solution: N/T

keywords: RDMA-CM, GSI-QP, CQ

Discovered at startup: 4.2-1.0.1.0

1078630

descriptionNote: When working on RoCE LAG on Kernel v3.10, the kernel may fail to download the driver while Network Manager is running.

alternative solution: Please exit Network Manager before downloading the driver and restart it after the driver download is complete.

keywords: RoCE LAG, Administrator de red

Discovered at startup: 4.2-1.0.1.0

1149557

descriptionNote: When configuring VGT+, the maximum number of VLAN IDs allowed to appear in sysfs is 813 (up to the first 813).

alternative solution: N/T

keywords: VGT+

Discovered at startup: 4.2-1.0.1.0

995665/1165919

description: In kernels prior to v4.13, the connection between the NVMEoF host and the target fails on a hyperthreaded system with more than 1 socket.

alternative solution: Connect to the NVMEoF subsystem on the host side--nr-io-queues <number_queues>Bandera.

look at thatnumber_queuesmust be less than or equal tonum_socketsmultiplied by num_cores_per_socket.

keywords: NVMEoF

1039346

description: Enabling multiple namespaces per subsystem when using NVMEoF target offload is not supported.

alternative solution: To enable more than one namespace, create a subsystem for each.

keywords: NVMEoF target swap, namespace

1030301

description: Creating virtual roles on a device in LAG mode destroys the LAG configuration. Device binding through the Ethernet NICs continues to work as expected.

alternative solution: N/T

keywords: LAG, SR-IOV

1047616

descriptionNote: If a device's node GUID is set to zero (0000:0000:0000:0000), the RDMA_CM userland application may crash.

alternative solution: Set the GUID of the node to a non-zero value.

keywords: RDMA_CM

1051701

descriptionNote: New versions of iproute that support new kernel features may work poorly on old kernels that do not support these new features.

alternative solution: N/T

keywords: iproute

1007830

description: If you are working on the Xenserver hypervisor with SR-IOV enabled, make sure the following statements apply:

1. Immediately after enabling SR-IOV, unplug all virtual function controller instances from their PCI slots.
2. It is not allowed to detach the instance from the PF controller while VFs are active.

alternative solution: N/T

keywords: SR-IOV

1005786

description: When using ConnectX-5 adapter cards, the following error may be reported to dmesg indicating a temporary lack of DMA pages:

"mlx5_core... give_pages:289:(pid x): The allocation time for page Y exceeded the maximum allowed time

mlx5_core ... page_notify_fail:263:(pid x): Page allocation failure notification sent to fw in func_id(z).

mlx5_core ... pages_work_handler:471:(pid x): dar falha -12”

example: This can happen when trying to open more than 64 VFs per port.

alternative solution: N/T

keywords: mlx5_core, DMA

1008066/1009004

Description: Performing some user-side operations during reboot may lead to call tracking/panic due to bugs found in the Linux kernel.

For example: Runget_vf_stats(via iptool) during reboot.

Problemumgehung: N/A

Keywords: mlx5_core, reboot

1009488

description: Mounting MLNX_EN in a path that contains special characters such as brackets or spaces is not supported. For example, if you mount MLNX_EN at /media/CDROM(vcd)/, the installation fails with the following error message:

# cd /medios/CDROM\(vcd\)/

# ./Install on PC

sh: 1: syntax error: "(" unexpected

alternative solution: N/T

keywords: built-in

982144

description- When Download Traffic Sniffer is enabled, your bandwidth can decrease by up to 50%.

alternative solution: N/T

keywords: Download traffic tracker

981362

description: Setting a TC number via the tc tool is not supported on many operating systems.

alternative solution: Set the TC number via the sysfs file /sys/class/net/<interface>/qos/tc_num.

keywords: Ethernet, CT

979457

description: Setting IOMMU=ON can cause severe performance degradation due to a bug in IOMMU.

alternative solution: Make sure the following patches are in your kernel:

• iommu/vt-d: correct mapping of PASID table
• iommu/vt-d: IOMMU search solution for SR-IOV virtual functions

monitoring: These patches are already available on Ubuntu 16.04.02 and 17.04 operating systems.

keywords: performance, IOMMU