Bug Fixes

1.0

This table lists the bugs fixed in this release.
For the list of old bug fixes, please refer to MLNX_OFED Archived Bug Fixes file at:
http://www.mellanox.com/pdf/prod_software/MLNX_OFED_Archived_Bug_Fixes.pdf

Internal Reference Number

Description

2396956

Description: Fixed an issue were device under massive load may hit iommu allocation failures. For more information see "RX Page Cache Size Limit" section in the user manual.

Keywords: Legacy libibverbs

Discovered in Release: 4.9-2.2.4.0

Fixed in Release: 4.9-3.1.5.0

2434638

Description: Fixed an issue where "ibv_devinfo -v" command did not print some of the MEM_WINDOW capabilities, even though they were supported.

Keywords: Legacy libibverbs

Discovered in Release: 4.9-2.2.4.0

Fixed in Release: 4.9-3.1.5.0

2292762

Description: Fixed a kernel panic scenario that may have taken place when using sysfs to cancel the probing of VFs and performing reboot while the VFs are still managed by the mlx5 driver.

Keywords: Proved VFs

Discovered in Release: 5.1-2.3.7.1

Fixed in Release: 5.1-2.5.8.0

2265055

Description: Added missing release of the lock held in the traffic class error flow.

Keywords: mutex_unlock

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2245228

Description: Fixed an issue of a crash when attempting to access roce_enable sysfs in unprobed VFs.

Keywords: roce_enable, unprobed VFs

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2211311

Description: Fixed an issue where Rx port buffers cell size was wrong, leading to wrong buffers size reported by mlnx_qos/netdev qos/buffer_size sysfs.

Keywords: mlx5e, RX buffers, mlnx_qos

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2143067

Description: If Openibd was configured to enable the SRP daemon, it now also enables srp_daemon from rdma-core.

Keywords: Openibd, SRP daemon, srp_daemon, rdma-core

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2192791

Description: Fixed the issue where packages neohost-backend and neohost-sdk were not properly removed by the uninstallation procedure and may have required manual removal before re-installing or upgrading the MLNX_OFED driver.

Keywords: NEO-Host, SDK

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2226715

Description: Fixed an issue where bringing up PF interface failed when using SR-IOV and configuring RoCE mode for v2 only.

Keywords: PF, SR-IOV, RoCE v2

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2242041

Description: Fixed leak of memory pages when using ODP.

Keywords: ODP

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2249090

Description: Fixed an issue where IBV_EXP_ACCESS_TUNNELED_ATOMIC capability did not work for ibv_exp_reg_mr experimental verb.

Keywords: ibv_exp_reg_mr

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2178677

Description: Marked "ibacm-devel" as a package name explicitly to avoid accidentally including a symlink to it in the UPSTREAM_LIBS version of MLNX_OFED.

Keywords: ibacm-devel, UPSTREAM_LIBS

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2255829

Description: Fixed an issue with metadata packages generation in the eth-only directory. This allows using the directory as a repository for package managers.

Keywords: Metadata packages

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2328754

Description: Fixed an issue that may have caused panic during peer memory invalidation flow.

Keywords: GPUDirect

Discovered in Release: 4.9-0.1.7.0

Fixed in Release: 4.9-2.2.4.0

2119017

Description: Fixed the issue where injecting EEH may cause extra Kernel prints, such as: “EEH: Might be infinite loop in mlx5_core driver”.

Keywords: EEH, kernel

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2076546

Description: Fixed the issue where in RPM-based OSs with non-default kernels, using repositories after re-creating the installer (using --add-kernel-support) would result in improper installation of the drivers.

Keywords: Installation, OS

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2114957

Description: Fixed the issue where MLNX_OFED installation may have depended on python2 package even when attempting to install it on OSs whose default package is python3.

Keywords: Installation, python

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2143258

Description: Fixed a typo in perftest package where help messages wrongly displayed the conversion result between Gb/s and MB/s (20^2 instead of 2^20).

Keywords: perftest

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2094216

Description: Fixed the issue of when one of the LAG slaves went down, LAG deactivation failed, ultimately causing bandwidth degradation.

Keywords: RoCE LAG

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2133778

Description: The mlx5 driver maintains a subdirectory for every open eth port in /sys/kernel/debug/. For the default network namespace, the sub-directory name is the name of the interface, like "eth8". The new convention for the network interfaces moved to the non-default network namespaces is the interfaces name followed by "@" and the port's PCI ID. For example: "eth8@0000:af:00.3".

Keywords: Namespace

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2122684

Description: Fixed the issue where OFED uninstallation resulted in the removal of dependency packages, such as qemu-system-* (qemu-system-x86).

Keywords: Uninstallation, dependency, qemu-system-x86

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2135476

Description: Added KMP ability to install MLNX_OFED Kernel modules on SLES12 SP5 and SLES15 kernel maintenance updates.

Keywords: KMP, SLES, kernel

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2149577

Description: Fixed the issue where openibd script load used to fail when esp6_offload module did not load successfully.

Keywords: openibd, esp6_offload

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2163879

Description: Added dependency of package mpi-selectors on perl-Getopt-Long system package. On minimal installs of RPM-based OSs, installing mpi-selectros will also install the required system package perl-Getopt-Long.

Keywords: Dependency, perl-Getopt-Long

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2107532

Description: Fixed the issue where in certain rare scenarios, due to Rx page not being replenished, the same page fragment mistakenly became assigned to two different Rx descriptors.

Keywords: Memory corruption, Rx page recycle

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2116234

Description: Fixed the issue where ibsim was missing after OFED installation.

Keywords: ibsim, installation

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2116233

Description: Fixed an issue where ucx-kmem was missing after OFED installation.

Keywords: ucx-kmem, installation

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2107776

Description: Fixed a driver load issue with Errata-kernel on SLES15 SP1.

Keywords: Load, SLES, Errata

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2105536

Description: Fixed an issue in the Hairpin feature which prevented adding hairpin flows using TC tool.

Keywords: Hairpin, TC

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2090321

Description: Fixed the issue where WQ queue flushing was not handled properly in the event of EEH.

Keywords: WQ, EEH

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2076311

Description: Fixed a rare kernel crash scenario when exiting an application that uses RMPP mads intensively.

Keywords: MAD RMPP

Discovered in Release: 4.0-1.0.1.0

Fixed in Release: 4.9-0.1.7.0

2096998

Description: Fixed the issue where NEO-Host could not be installed from the MLNX_OFED package when working on Ubuntu and Debian OSs.

Keywords: NEO-Host, Ubuntu, Debian

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2057076

Description: Fixed the issue where installing MLNX_OFED using --add-kernel-support option did not work over RHEL 8 OSs.

Keywords: --add-kernel-support, installation, RHEL

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2090186

Description: Fixed a possible kernel crash scenario when AER/slot reset in done in parallel to user space commands execution.

Keywords: mlx5_core, AER, slot reset

Discovered in Release: 4.3-1.0.1.0

Fixed in Release: 4.9-0.1.7.0

2093410

Description: Added missing ECN configuration under sysfs for PFs in SwitchDev mode.

Keywords: sysfs, ASAP, SwitchDev, ECN

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

1916029

Description: Fixed the issue of when firmware response time to commands became very long, some commands failed upon timeout. The driver may have then triggered a timeout completion on the wrong entry, leading to a NULL pointer call trace.

Keywords: Firmware, timeout, NULL

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2036394

Description: Added driver support for kernels with the old XDP_REDIRECT infrastructure that uses the following NetDev operations: .ndo_xdp_flush and .ndo_xdp_xmit.

Keywords: XDP_REDIRECT, Soft lockup

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2072871

Description: Fixed an issue where the usage of --excludedocs Open MPI RPM option resulted in the removal of non-documentation related files.

Keywords: --excludedocs, Open MPI, RPM

Discovered in Release: 4.5-1.0.1.0

Fixed in Release: 4.9-0.1.7.0

2072884

Description: Removed all cases of automated loading of MLNX_OFED kernel modules outside of openibd to preserve the startup process of previous MLNX_OFED versions. These loads conflict with openibd, which has its own logic to overcome issues. Such issues can be inbox driver load instead of MLNX_OFED, or module load with wrong parameter value. They might also load modules while openibd is trying to unload the driver stack.

Keywords: Installation, openibd

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2052037

Description: Disabled automated loading of some modules through udev triggers to preserve the startup process of previous MLNX_OFED versions.

Keywords: Installation, udev

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2022634

Description: Fixed a typo in the packages build command line which could cause the installation of MLNX_OFED on SLES OSs to fail when using the option --without-depcheck.

Keywords: Installation, SLES

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2022619

Description: Fixed the issue where uninstallation of MLNX_OFED would hang due to a bug in the package dependency check.

Keywords: Uninstallation, dependency

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2047221

Description: Reference count (refcount) for RDMA connection ID (cm_id) was not incremented in rdma_resolve_addr() function, resulting in a cm_id use-after-free access.

A fix was applied to increment the cm_id refcount.

Keywords: rdma_resolve_addr(), cm_id

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.9-0.1.7.0

2045181

Description: Fixed a race condition which caused kernel panic when moving two ports to SwitchDev mode at the same time.

Keywords: ASAP, SwitchDev, race

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2004488

Description: Allowed accessing sysfs hardware counters in SwitchDev mode.

Keywords: ASAP, hardware counters, sysfs, SwitchDev

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2030943

Description: Function smp_processor_id() is called in the RX page recycle flow to determine the core to run on. This is intended to run in NAPI context. However, due to a bug in backporting, the RX page recycle was mistakenly called also in the RQ close flow when not needed.

Keywords: Rx page recycle, smp_processor_id

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.9-0.1.7.0

2074487

Description: Fixed an issue where port link state was automatically changed (without admin state involvement) to "UP" after reboot.

Keywords: Link state, UP

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2022618

Description: Fixed a hang with ConnectX-3 adapter cards when running over SLES 11 OS.

Keywords: OS, SLES, ConnectX-3

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.9-0.1.7.0

2064711

Description: Fixed an issue where RDMA CM connection failed when port space was small.

Keywords: RDMA CM

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

2076424

Description: Traffic mirroring with OVS offload and non-offload over VxLAN interface is now supported.

Note: For kernel 4.9, make sure to use a dedicated OVS version.

Keywords: VxLAN, OVS

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

1828321

Description: Fixed the issue of when working with VF LAG while the bond device is in active-active mode, running fwreset would result in unequal traffic on both PFs, and PFs would not reach line rate.

Keywords: VF LAG, bonding, PF

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.9-0.1.7.0

1975293

Description: Installing OFED with --with-openvswitch flag no longer requires manual removal of the existing Open vSwitch.

Keywords: OVS, Open vSwitch, openvswitch

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

1939719

Description: Fixed an issue of when running openibd restart after the installation of MLNX_OFED on SLES12 SP5 and SLES15 SP1 OSs with the latest Kernel (v4.12.14) resulted in an error that the modules did not belong to that Kernel. This was due to the fact that the module installed by MLNX_OFED was incompatible with new Kernel's module.

Keywords: SLES, operating system, OS, installation, Kernel, module

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

2001966

Description: Fixed an issue of when bond was created over VF netdevices in SwitchDev mode, the VF netdevice would be treated as representor netdevice. This caused the mlx5_core driver to crash in case it received netdevice events related to bond device.

Keywords: PF, VF, SwitchDev, netdevice, bonding

Discovered in Release: 4.7-3.2.9.0

Fixed in Release: 4.9-0.1.7.0

1816629

Description: Fixed an issue where following a bad affinity occurrence in VF LAG mode, traffic was sent after the port went up/down in the switch.

Keywords: Traffic, VF LAG

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.9-0.1.7.0

1718531

Description: Added support for VLAN header rewrite on CentOS 7.2 OS.

Keywords: VLAN, ASAP, switchdev, CentOS 7.2

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.9-0.1.7.0

1556337

Description: Fixed the issue where adding VxLAN decapsulation rule with enc_tos and enc_ttl failed.

Keywords: VxLAN, decapsulation

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.9-0.1.7.0

1949260

Description: Fixed a race condition that resulted in kernel panic when running IPoIB traffic in Connected mode.

Keywords: IPoIB

Discovered in Release: 4.5-1.0.1.0

Fixed in Release: 4.7-3.2.9.0

1973828

Description: Fixed wrong EEPROM length for small form factor (SFF) 8472 from 256 to 512 bytes.

Keywords: EEPROM, SFF

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.7-3.2.9.0

1915553

Description: Fixed the issue where errno field was not sent in all error flows of ibv_reg_mr API.

Keywords: ibv_reg_mr

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-3.2.9.0

1970901

Description: Fixed the issue where mlx5 IRQ name did not change to express the state of the interface.

Keywords: Ethernet, PCIe, IRQ

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.7-3.2.9.0

1915587

Description: Udaddy application is now functional in Legacy mode.

Keywords: Udaddy, MLNX_OFED legacy, RDMA-CM

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.7-3.2.9.0

1931421

Description: Added support for E-Switch (SR-IOV Legacy) mode in RHEL 7.7 OSs.

Keywords: E-Switch, SR-IOV, RHEL, RedHat

Discovered in Release: 4.7-1.0.0.1

Fixed in Release: 4.7-3.2.9.0

1945411/1839353

Description: Fixed the issue of when XDP_REDIRECT fails, pages got double-freed due to a bug in the refcnt_bias feature.

Keywords: XDP, XDP_REDIRECT, refcnt_bias

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-3.2.9.0

1547200

Description: Fixed an issue where IPoIB Tx queue may get stuck, leading to timeout warnings in dmesg.

Keywords: IPoIB

Discovered in Release: 4.5-1.0.1.0

Fixed in Release: 4.7-1.0.0.1

1817636

Description: Fixed the issue of when disabling one port on the Server side, VF-LAG Tx Affinity would not work on the Client side.

Keywords: VF-LAG, Tx Affinity

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-1.0.0.1

1800525

Description: When configuring the Time-stamping feature, CQE compression will be disabled. This fix entails the removal of a warning message that appeared upon attempting to disable CQE compression when it has already been disabled.

Keywords: Time-stamping, CQE compression

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-1.0.0.1

1431282

Description: Fixed the issue where software reset may have resulted in an order inversion of interface names.

Keywords: Software reset

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.7-1.0.0.1

1843020

Description: Server reboot may result in a system crash.

Keywords: reboot, crash

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.7-1.0.0.1

1734102

Description: Fixed the issue where Ubuntu v16.04.05 and v16.04.05 OSs could not be used with their native kernels.

Keywords: Ubuntu, Kernel, OS

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-1.0.0.1

1811973

Description: VF mirroring offload is now supported.

Keywords: ASAP2, VF mirroring

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-1.0.0.1

1841634

Description: The number of guaranteed counters per VF is now calculated based on the number of ports mapped to that VF. This allows more VFs to have counters allocated.

Keywords: Counters, VF

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.7-1.0.0.1

1758983

Description: Installing MLNX_OFED on RHEL 7.6 OSs platform x86_64 and RHEL 7.6 ALT OSs platform PPCLE using YUM is now supported.

Keywords: RHEL, RedHat, YUM, OS, operating system

Discovered in Release: 4.6-1.0.1.1

Fixed in Release: 4.7-1.0.0.1

1523548

Description: Fixed the issue where RDMA connection persisted even after dropping the network interface.

Keywords: Network interface, RDMA

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.6-1.0.1.1

1712870

Description: Fixed the issue where small packets with non-zero padding were wrongly reported as "checksum complete" even though the padding was not covered by the csum calculation. These packets now report "checksum unnecessary".

In addition, an ethtool private flag has been introduced to control the "checksum complete" feature: ethtool --set-priv-flags eth1 rx_no_csum_complete on/off

Keywords: csum error, checksum, mlx5_core

Discovered in Release: 4.5-1.0.1.0

Fixed in Release: 4.6-1.0.1.1

1648597

Description: Fixed the wrong wording in the FW tracer ownership startup message (from "FW Tracer Owner" to "FWTracer: Ownership granted and active").

Keywords: FW Tracer

Discovered in Release: 4.5-1.0.1.0

Fixed in Release: 4.6-1.0.1.1

1581631

Description: Fixed the issue where GID entries referenced to by a certain user application could not be deleted while that user application was running.

Keywords: RoCE, GID

Discovered in Release: 4.5-1.0.1.0

Fixed in Release: 4.6-1.0.1.1

1368390

Description: Fixed the issue where MLNX_OFED could not be installed on RHEL 7.x Alt OSs using YUM repository.

Keywords: Installation, YUM, RHEL

Discovered in Release: 4.3-3.0.2.1

Fixed in Release: 4.6-1.0.1.1

1531817

Description: Fixed an issue of when the number of channels configured was less than the number of CPUs available, part of the CPUs would not be used by Tx queues.

Keywords: Performance, Tx, CPU

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.5-1.0.1.0

1571977

Description: Fixed an issue of when the same CQ is connected to some QPs with SRQ and some without, wrong wr_id might be reported by ibv_poll_cq .

Keywords: libmlx5, wr_id

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.5-1.0.1.0

1380135

Description: Fixed the issue where IB port link used to flap due to MAD heartbeat response delay when using new CQ API.

Keywords: IB port link, CQ API, MAD heartbeat

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.5-1.0.1.0

1498931

Description: Fixed the issue where establishing TCP connection took too long due to failure of SA PathRecord query callback handler.

Keywords: TCP, SA PathRecord

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.5-1.0.1.0

1514096

Description: Fixed the issue where lack of high order allocations caused driver load failure. All high order allocations are now changed to order-0 allocations.

Keywords: mlx5, high order allocation

Discovered in Release: 4.0-2.0.2.0

Fixed in Release: 4.5-1.0.1.0

1524932

Description: Fixed a backport issue on some OSs, such as RHEL v7.x, where mlx5 driver would support ip link set DEVICE vf NUM rate TXRATE old command, instead of ip link set DEVICE vf NUM max_tx_rate TXRATE min_tx_rate TXRATE new command.

Keywords: mlx5 driver

Discovered in Release: 4.0-2.0.2.0

Fixed in Release: 4.5-1.0.1.0

1498585

Description: Fixed the issue of when performing configuration changes, mlx5e counters values were reset.

Keywords: Ethernet counters

Discovered in Release: 4.0-2.0.2.0

Fixed in Release: 4.5-1.0.1.0

1484603

Description: Fixed the issue of when using ibv_exp_cqe_ts_to_ns verb to convert a packet's hardware timestamp to UTC time in nanoseconds, the result may appear backwards compared to the converted time of a previous packet.

Keywords: libibverbs

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.5-1.0.1.0

1425027

Description: Fixed the issue where attempting to establish a RoCE connection on the default GID or on IPv6 link-local address might have failed when two or more netdevices that belong to HCA ports were slaves under a bonding master.

This might also have resulted in the following error message in the kernel log: “ __ib_cache_gid_add: unable to add gid fe80:0000:0000:0000:f652:14ff:fe46:7391 error=-28 ”.

Keywords: RoCE, bonding

Discovered in Release: 4.4-1.0.0.0

Fixed in Release: 4.5-1.0.1.0

1480206

Description: Modified mlx5_ib SRQs behavior. Now the SRQs are allocated to “order 1” pages instead of contiguous ones to lower the probability of out-of-memory scenarios.

Keywords: SRQ, mlx5_ib

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.4-2.0.7.0

1363375

Description: Modified mlx5_ib QPs behavior. Now the QPs are allocated to “order 1” pages instead of contiguous ones to lower the probability of out-of-memory scenarios.

Keywords: IPoIB, mlx5_ib

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.4-2.0.7.0

1332080

Description: Modified mlx4_ib QPs behavior. Now the QPs are allocated to “order 1” pages instead of contiguous ones to lower the probability of out-of-memory scenarios.

Keywords: IPoIB, mlx4_ib

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.4-1.0.0.0

1412468

Description: Added support for multi-host connection on mstflint’s mstfwreset.

Keywords: mstfwreset, mstflint, MFT, multi-host

Discovered in Release: 4.3-1.0.1.0

Fixed in Release: 4.4-1.0.0.0

1423319

Description: Removed the following prints on server shutdown: mlx5_core 0005:81:00.1: mlx5_enter_error_state:96:(pid1): start mlx5_core 0005:81:00.1: mlx5_enter_error_state:109:(pid1): end

Keywords: mlx5, fast shutdown

Discovered in Release: 4.3-1.0.1.0

Fixed in Release: 4.4-1.0.0.0

1433092

Description: Fixed an issue of when querying for IBV_EXP_VALUES_HW_CLOCK_NS (using ibv_exp_query_values function) without querying for IBV_EXP_VALUES_HW_CLOCK, 0 value was returned.

Keywords: mlx5, CQE time-stamping

Discovered in Release: 4.3-1.0.1.0

Fixed in Release: 4.4-1.0.0.0

1318251

Description: Fixed the issue of when bringing mlx4/mlx5 devices up or down, a call trace in

nvme_rdma_remove_one or nvmet_rdma_remove_one may occur.

Keywords: NVMEoF, mlx4, mlx5, call trace

Discovered in Release: 4.3-1.0.1.0

Fixed in Release: 4.4-1.0.0.0

1181815

Description: Fixed an issue where 4K UD packets were dropped when working with 4K MTU on mlx4 devices.

Keywords: mlx4, 4K MTU, UD

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.3-1.0.1.0

1247458

Description: Added support for VLAN Tag (VST) creation on RedHat v7.4 with new iproute2 packages (iptool).

Keywords: SR-IOV, VST, RedHat

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.3-1.0.1.0

1229554

Description: Enabled RDMA CM to honor incoming requests coming from ports of different devices.

Keywords: RDMA CM

Discovered in Release: 4.2-1.0.0.0

Fixed in Release: 4.3-1.0.1.0

1262257

Description: Fixed an issue where sending Work Requests (WRs) with multiple entries where the first entry is less than 18 bytes used to fail.

Keywords: ConnectX-5; libibverbs; Raw QP

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.3-1.0.1.0

1249358/1261023

Description: Fixed the issue of when the interface was down, ethtool counters ceased to increase. As a result, RoCE traffic counters were not always counted.

Keywords: Ethtool counters, mlx5

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.3-1.0.1.0

1244509

Description: Fixed compilation errors of MLNX_OFED over kernel when CONFIG_PTP_1588_CLOCK parameter was not set.

Keywords: PTP, mlx5e

Discovered in Release: 4.2-1.2.0.0

Fixed in Release: 4.3-1.0.1.0

1266802

Description: Fixed an issue where the system used to hang when trying to allocate multiple device memory buffers from different processes simultaneously.

Keywords: Device memory programming

Discovered in Release: 4.2-1.0.0.0

Fixed in Release: 4.3-1.0.1.0

1120424

Description: Fixed incorrect SGE number of RSS QP.

Keywords: RSS, SGE

Discovered in Release: 4.1-1.0.2.0

Fixed in Release: 4.2-1.0.0.0

1078887

Description: Fixed an issue where post_list and CQ_mod features in perftest did not function when running the --run_infinitely flag.

Keywords: perftest, --run_infinitely

Discovered in Release: 4.2-1.0.1.0

Fixed in Release: 4.2-1.2.0.0

1186260

Description: Fixed the issue where CNP counters exposed under /sys/class/infiniband/mlx5_bond_0/ports/1/hw_counters/ did not aggregate both physical functions when working in RoCE LAG mode.

Keywords: RoCE, LAG, ECN, Congestion Counters

Discovered in Release: 4.2-1.0.1.0

Fixed in Release: 4.2-1.2.0.0

1178129

Description: Fixed an issue that prevented Windows virtual machines running over MLNX_OFED Linux hypervisors from operating ConnectX-3 IB ports.

When such failures occurred, the following message (or similar) appeared in the Linux HV message log when users attempted to start up a Windows VM running a ConnectX-3 VF:

mlx4_core 0000:81:00.0: vhcr command 0x1a slave:1 in_param 0x793000 in_mod=0x210 op_mod=0x0 failed with error:0, status -22

Keywords: SR-IOV, RDMA, VM, KVM, Windows

Discovered in Release: 4.2-1.0.1.0

Fixed in Release: 4.2-1.2.0.0

1192374

Description: Fixed wrong calculation of max_device_ctx capability in ConnectX-4, ConnectX-4 Lx, and ConnectX-5 HCAs.

Keywords: ibv_exp_query_device, max_device_ctx mlx5

Discovered in Release: 4.2-1.0.1.0

Fixed in Release: 4.2-1.2.0.0

1084791

Description: Fixed the issue where occasionally, after reboot, rpm commands used to fail and create a core file, with messages such as “Bus error (core dumped)”, causing the openibd service to fail to start.

Keywords: rpm, openibd

Discovered in Release: 3.4-2.0.0.0

Fixed in Release: 4.2-1.0.0.0

960642/960653

Description: Added support for min_tx_rate and max_tx_rate limit per virtual function ConnectX-5 and ConnectX-5 Ex adapter cards.

Keywords: SR-IOV, mlx5

Discovered in Release: 4.0-1.0.1.0

Fixed in Release: 4.2-1.0.0.0

866072/869183

Description: Fixed the issue where RoCE v2 multicast traffic using RDMA-CM with IPv4 address was not received.

Keywords: RoCE

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.2-1.0.0.0

1163835

Description: Fixed an issue where ethtool -P output was 00:00:00:00:00:00 when using old kernels.

Keywords: ethtool, Permanent MAC address, mlx4, mlx5

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.2-1.0.0.0

1067158

Description: Replaced a few “GPL only” legacy libibverbs functions with upstream implementation that conforms with libibverbs GPL/BSD dual license model.

Keywords: libibverbs, license

Discovered in Release: 4.1-1.0.2.0

Fixed in Release: 4.2-1.0.0.0

1119377

Description: Fixed an issue where ACCESS_REG command failure used to appear upon RoCE Multihost driver restart in dmesg. Such an error message looked as follows:

mlx5_core 0000:01:00.0: mlx5_cmd_check:705:(pid 20037): ACCESS_REG(0x805) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0x15c356)

Keywords: RoCE, multihost, mlx5

Discovered in Release: 4.1-1.0.2.0

Fixed in Release: 4.2-1.0.0.0

1122937

Description: Fixed an issue where concurrent client requests got corrupted when working in persistent server mode due to a race condition on the server side.

Keywords: librdmacm, rping

Discovered in Release: 4.1-1.0.2.0

Fixed in Release: 4.2-1.0.0.0

1102158

Description: Fixed an issue where client side did not exit gracefully in RTT mode when the server side was not reachable.

Keywords: librdmacm, rping

Discovered in Release: 4.1-1.0.2.0

Fixed in Release: 4.2-1.0.0.0

1038933

Description: Fixed a backport issue where IPv6 procedures were called while they were not supported in the underlying kernel.

Keywords: iw_cm

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1064722

Description: Added log debug prints when changing HW configuration via DCB. To enable log debug prints, run: ethtool -s <devname> msglvl hw on/off

Keywords: DCB, msglvl

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1013076

Description: Fixed the issue where reassembly of packets larger than 64k might have failed when ipfrag threshold was low. This issue was present only on RHEL 6.3, 6.4, 6.5, and Ubuntu 12.04.

This packet drop could be seen from the netstat tool, indicated by the “packet reassembles failed” counter.

Keywords: IPoIB, Packet Fragmentation

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1022251

Description: Fixed SKB memory leak issue that was introduced in kernel 4.11, and added warning messages to the Soft RoCE driver for easy detection of future SKB leaks.

Keywords: Soft RoCE

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1044546

Description: Fixed the issue where a kernel crash used to occur when RXe device was coupled with a virtual (dummy) device.

Keywords: Soft RoCE

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1047617

Description: Fixed the issue where a race condition in the RoCE GID cache used to cause for the loss of IP-based GIDs.

Keywords: RoCE, GID

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1006768

Description: Fixed the issue where an rdma_cm connection between a client and a server that were on the same host was not possible when working over VLAN interfaces.

Keywords: RDMACM

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

801807

Description: Fixed an issue where RDMACM connection used to fail upon high connection rate accompanied with the error message: RDMA_CM_EVENT_UNREACHABLE .

Keywords: RDMACM

Discovered in Release: 3.0-2.0.1

Fixed in Release: 4.1-1.0.2.0

869768

Description: Fixed the issue where SR-IOV was not supported in systems with a page size greater than 16KB.

Keywords: SR-IOV, mlx5, PPC

Discovered in Release: 4.0-2.0.0.1

Fixed in Release: 4.1-1.0.2.0

1155972

Description: Fixed mlx4 kernel crash upon server shutdown due to NULL pointer dereference.

Keywords: mlx4, shutdown

Discovered in Release: 3.3-1.0.4.0

Fixed in Release: 4.0-2.0.0.1

919545

Description: Fixed the issue of when the Kernel becomes out of memory upon driver start, it could crash on SLES 12 SP2.

Keywords: mlx_5 Eth Driver

Discovered in Release: 3.4-2.0.0.0

Fixed in Release: 4.0-2.0.0.1

970668

Description: Fixed the issue where very high stress on DC QP transport might have triggered NMI messages on specific servers.

Keywords: mlx5 Driver

Discovered in Release: 4.0-1.0.1.0

Fixed in Release: 4.0-2.0.0.1

966134

Description: Allowed Ethernet VFs to open Raw Ethernet QPs even if RoCE is not supported for the VF.

Keywords: mlx4_ib

Discovered in Release: 3.0-1.0.1

Fixed in Release: 4.0-2.0.0.1

864063

Description: Fixed the issue of when Spoof-check may have been turned on for MAC address 00:00:00:00:00:00.

Keywords: mlx4

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-2.0.0.1

869209

Description: Fixed an issue that caused TCP packets to be received in an out of order manner when Large Receive Offload (LRO) is on.

Keywords: mlx5_en

Discovered in Release: 3.3-1.0.0.0

Fixed in Release: 4.0-2.0.0.1

913319

Description: Fixed the issue of low performance when creating many address handles.

Keywords: libibverbs

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-1.0.1.0

912897

Description: Added debug prints to ib_umem_get function to fix lack of error indication when this function fails.

Keywords: InfiniBand

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-1.0.1.0

945887

Description: [ConnectX-3] Fixed the issue where multicast traffic over Raw Ethernet QP on virtual functions were received on the same QP (loopback).

Keywords: SR-IOV

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-1.0.1.0

920292

Description: Fixed three issues in libmlx5 that were found by NVIDIA in the patches that are part of MLNX_OFED v3.4:

  1. mlx5_exp_peer_commit_qp returns number of entries = 4 instead of 3.

  2. Peer capability check is wrong - should fail the check when there is neither NOR nor GEQ support.

  3. Missing break in mlx5_exp_peer_peek_cq. There is now fallthrough in the IBV_EXP_PEER_PEEK_ABSOLUTE case.

Keywords: libmlx5

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-1.0.1.0

890285

Description: Fixed the issue where memory allocation for CQ buffers used to fail when increasing the RX ring size.

Keywords: mlx5_core

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-1.0.1.0

867094

Description: Fixed the issue where MLNX_OFED used to fail to load on 4K page Arm architecture.

Keywords: Arm

Discovered in Release: 3.4-1.0.0.0

Fixed in Release: 4.0-1.0.1.0

© Copyright 2023, NVIDIA. Last updated on Oct 23, 2023.