Bug Fixes in This Version

4980835

Description: Attempting to modify the SRv6 Segments Left field using a direct Set action returned an unclear error instead of indicating that the action is unsupported.

Keyword: SRv6; DOCA Flow; error reporting; unsupported operation

Detected in version: 3.5.0

5149843

5165212

5165213

Description: OVS-DOCA may fail to attach a virtual-function representor to OVS, disrupting OVN-Kubernetes traffic.

Keyword: OVS-DOCA; VF representor; traffic disruption

Detected in version: 3.5.0

5230613

Description: On B300 NetPlugin systems, the bundled target maps may not represent supported multi-rail configurations, preventing a usable deployment plan.

Keyword: DMS; B300; NetPlugin; target map

Detected in version: 3.5.0

5230252

Description: When no stage is specified for a DMS single-plane plan, the plan may default to configuration and leave the host partially configured without warning.

Keyword: DMS; single-plane; deployment plan; partial configuration

Detected in version: 3.5.0

5229962

Description: On GB200 systems, planning a single-plane deployment with the bundled target map may fail.

Keyword: DMS; GB200; single-plane; target map

Detected in version: 3.5.0

5227828

Description: On single-plane deployments, the DMS configuration stage may fail while bringing up network interfaces.

Keyword: DMS; single-plane; configuration; network interfaces

Detected in version: 3.5.0

5227092

Description: On single-plane deployments, DMS may leave network and RDMA interfaces with their default names instead of the expected rail-based names.

Keyword: DMS; single-plane; interface naming; RDMA

Detected in version: 3.5.0

5223152

5228590

Description: In NetPlugin deployments, DMS may apply an incomplete RoCE traffic-class configuration to physical interfaces.

Keyword: DMS; NetPlugin; RoCE; traffic class

Detected in version: 3.5.0

5221883

Description: In dual-plane NetPlugin deployments, DMS may apply inconsistent RDMA interface names, leaving some interfaces with their default names.

Keyword: DMS; NetPlugin; RDMA; interface naming

Detected in version: 3.5.0

5221880

Description: In dual-plane NetPlugin deployments, DMS may assign an RDMA name to the wrong device, causing the preparation stage to fail.

Keyword: DMS; NetPlugin; RDMA; preparation

Detected in version: 3.5.0

5221409

Description: DO-SPCX deployment tasks may fail when the DMS command-line tool location is not set in the environment.

Keyword: DMS; DO-SPCX; command-line tool; deployment

Detected in version: 3.5.0

5217068

Description: DMS verification may report MTU, virtual-function, and huge-page mismatches even when the host is configured correctly.

Keyword: DMS; verification; false mismatch; host configuration

Detected in version: 3.5.0

5217061

Description: On systems with multiple NUMA nodes, DMS may reduce an existing huge-page allocation or leave fewer huge pages than configured.

Keyword: DMS; huge pages; NUMA; memory allocation

Detected in version: 3.5.0

5211680

Description: On GB300 dual-plane systems, DMS may apply network and congestion-control settings that do not match the deployment guide.

Keyword: DMS; GB300; dual-plane; network configuration

Detected in version: 3.5.0

5203065

Description: After a DMS installation fails, the status API may incorrectly report the plan as completed.

Keyword: DMS; status API; installation; status reporting

Detected in version: 3.5.0

5201742

5201640

Description: During DO-SPCX configuration, the DMS configuration stage may remain stuck at 13 of 14 phases.

Keyword: DMS; DO-SPCX; configuration; stalled deployment

Detected in version: 3.5.0

5201632

Description: During a DO-SPCX deployment, incorrect interface naming may cause the DMS preparation stage to fail after reboot.

Keyword: DMS; DO-SPCX; preparation; interface naming

Detected in version: 3.5.0

5199086

Description: In NetPlugin deployments, DMS configuration may fail while bringing up network interfaces.

Keyword: DMS; NetPlugin; network interfaces; configuration

Detected in version: 3.5.0

5197210 / 5174929

Description: The DMS Verify API may return error values instead of configuration results for a DO-SPCX preparation plan.

Keyword: DMS; Verify API; DO-SPCX; validation

Detected in version: 3.5.0

5191384 / 5191380

Description: DMS may reject attempts to configure congestion-control counter groups or enable congestion-control tracing through its command-line interface.

Keyword: DMS; congestion control; counter groups; tracing

Detected in version: 3.5.0

5185448

Description: DMS status may remain incomplete even after all configuration phases finish successfully.

Keyword: DMS; status reporting; configuration completion

Detected in version: 3.5.0

5185431

Description: On multi-plane systems using per-rail bonding, DMS may fail to bring up RDMA links and may assign incorrect RDMA interface names.

Keyword: DMS; multi-plane; RDMA; link activation

Detected in version: 3.5.0

5185406

Description: In multi-plane deployments, DMS may select the wrong network interface during link configuration, causing configuration to fail.

Keyword: DMS; multi-plane; network interfaces; link configuration

Detected in version: 3.5.0

5180951

Description: During DO-SPCX configuration, DMS may report that programmable congestion control is not ready even when it is available.

Keyword: DMS; DO-SPCX; congestion control; readiness check

Detected in version: 3.5.0

5180863

Description: The DO-SPCX configuration stage may stall while initializing programmable congestion control.

Keyword: DMS; DO-SPCX; congestion control; configuration

Detected in version: 3.5.0

5179634

Description: In NetPlugin deployments, the DOCA SPCX congestion-control service may fail to start for RDMA devices.

Keyword: DMS; NetPlugin; congestion control; RDMA

Detected in version: 3.5.0

5179628

Description: In NetPlugin deployments with split ports, the DMS preparation stage may fail even after network settings have been applied.

Keyword: DMS; NetPlugin; preparation; split ports

Detected in version: 3.5.0

5179435

Description: In DOCA XPlane, log rotation may fail, allowing XPlane logs to accumulate.

Keyword: DOCA XPlane; log rotation; log files

Detected in version: 3.5.0

5179302

Description: When DOCA XPlane uses a nondefault container registry, DMS boot validation may reject the image and block the deployment.

Keyword: DMS; DOCA XPlane; container registry; boot validation

Detected in version: 3.5.0

5179253

Description: On four-rail, dual-plane deployments, DMS may create unused virtual functions and representors, leaving extra interfaces active but unconfigured.

Keyword: DMS; multi-plane; virtual functions; network interfaces

Detected in version: 3.5.0

5175085

Description: After a successful preparation stage, cleaning up the active DMS plan may prevent the configuration stage from starting.

Keyword: DMS; preparation; configuration; plan cleanup

Detected in version: 3.5.0

5175073

Description: During DO-SPCX configuration, DMS may fail to identify virtual-function representors, causing configuration to stop.

Keyword: DMS; DO-SPCX; VF representors; configuration

Detected in version: 3.5.0

5166426

Description: When an XPlane client requests initial state after a local link has already failed, the existing link-down event may be omitted.

Keyword: DOCA XPlane; initial state; link failure; event reporting

Detected in version: 3.5.0

5165499

Description: DMS status may remain in progress indefinitely when a plan service fails.

Keyword: DMS; status reporting; plan failure

Detected in version: 3.5.0

5160589

Description: The DMS command-line tool may fail when querying the pending link type.

Keyword: DMS; command-line tool; link type; query failure

Detected in version: 3.5.0

5160542

Description: DMS may fail to configure ECN receive parameters on a network device.

Keyword: DMS; ECN; congestion control; configuration

Detected in version: 3.5.0

5158946

Description: DMS may return an empty response when querying breakout lane assignments for a module port.

Keyword: DMS; port breakout; lane query

Detected in version: 3.5.0

5158192

Description: DMS blueprint verification may report error values for supported settings, causing valid profiles to appear mismatched.

Keyword: DMS; blueprint verification; configuration validation

Detected in version: 3.5.0

5156646

Description: Installing the RA2.2 software multi-plane preparation profile on ConnectX-8 may fail during configuration verification.

Keyword: DMS; DO-SPCX; profile installation; ConnectX-8

Detected in version: 3.5.0

5154317

Description: DO-SPCX planning may fail because required planner data is missing from the installed package.

Keyword: DMS; DO-SPCX; planning; packaging

Detected in version: 3.5.0

5153995

Description: DMS may be unable to locate required DO-SPCX blueprint data, preventing deployment planning.

Keyword: DMS; DO-SPCX; blueprints; planning

Detected in version: 3.5.0

5151565

Description: The DOCA XPlane Kubernetes container may report version 0.0.0 in startup logs instead of the installed version.

Keyword: DOCA XPlane; Kubernetes; version reporting; logs

Detected in version: 3.4.0

5139760

Description: DMS may fail to apply link-aggregation resource settings during preparation, blocking DO-SPCX bring-up.

Keyword: DMS; DO-SPCX; LAG; preparation

Detected in version: 3.5.0

5139743

Description: DMS may generate an invalid port-breakout configuration during preparation, blocking DO-SPCX and XPlane bring-up.

Keyword: DMS; DO-SPCX; port breakout; XPlane

Detected in version: 3.5.0

5131673

Description: DOCA DMS packages may fail to build for openEuler 22.03 and CTyunOS 23.01, blocking BFB image creation.

Keyword: DOCA DMS; openEuler; CTyunOS; packaging

Detected in version: 3.5.0

5128524

Description: On BlueField-2, DMS may fail to change interface MTU.

Keyword: DMS; BlueField-2; MTU; configuration

Detected in version: 3.5.0

5128496

Description: The DO-SPCX installer may complete without applying configuration values from the generated deployment plan.

Keyword: DMS; DO-SPCX; installer; deployment plan

Detected in version: 3.5.0

5107936

Description: OVS-DOCA offload may stall after an asynchronous flow-insertion failure, which can also prevent Kubernetes pods from being created.

Keyword: OVS-DOCA; flow offload; deployment stall; Kubernetes

Detected in version: 3.2.2

5098855

Description: The DOCA Flow Tune package may be unavailable for Anolis 8.6, blocking installation and BFB creation.

Keyword: DOCA Flow Tune; Anolis 8.6; packaging; BFB

Detected in version: 3.5.0

5089659

Description: On Debian BFB images, DOCA Bench may report supported device capabilities as unavailable and fail to create benchmark pipeline steps.

Keyword: DOCA Bench; Debian; device capabilities; packaging

Detected in version: 3.4.0

5082213

Description: DMS operating-system verification may fail when required network tooling is unavailable.

Keyword: DMS; OS verification; network tools

Detected in version: 3.5.0

5042755

Description: DOCA Flow Performance may fail to start in DOCA mode, reporting that no queues are available.

Keyword: DOCA Flow Performance; queue allocation; application startup

Detected in version: 3.5.0

5011919

Description: DOCA PSP Gateway may stop forwarding encrypted traffic received on a physical function managed by the embedded switch.

Keyword: DOCA PSP Gateway; encryption; eSwitch; traffic forwarding

Detected in version: 3.5.0

4991583

Description: OVS may crash during startup after a DPU reboot or driver restart when DOCA initialization is enabled manually but not in the system configuration.

Keyword: Open vSwitch; service crash; DOCA initialization; configuration

Detected in version: 3.4.0

4989366

Description: DOCA Flow control-pipe entries may fail to match metadata tags, preventing the expected traffic rules from taking effect.

Keyword: DOCA Flow; metadata; control pipe; traffic matching

Detected in version: 3.5.0

4980744

Description: DOCA Argus may omit container-termination events, resulting in incomplete container lifecycle reporting.

Keyword: DOCA Argus; container events; lifecycle reporting

Detected in version: 3.4.0

4979016

Description: DOCA Bench throughput may drop significantly when DMA job sizes are not powers of two.

Keyword: DOCA Bench; DMA; performance; job size

Detected in version: 3.4.0

4958775

Description: NAT64 communication may fail across mixed-version deployments because its Unix-domain socket interface is not backward compatible.

Keyword: NAT64; backward compatibility; UDS; mixed versions

Detected in version: 3.3.0

4882015

Description: The DMS API may report a pending BlueField mode change as active before the required power cycle instead of returning the current mode.

Keyword: DMS; BlueField mode; state reporting; API

Detected in version: 3.3.0

4877227

Description: DOCA Flow may fail to create a flooding pipe containing a single entry.

Keyword: DOCA Flow; flooding pipe; flow entry

Detected in version: 3.3.0

4868003

Description: OVS-DOCA may lose flow context while processing hardware-offloaded traffic, resulting in packet-processing errors.

Keyword: OVS-DOCA; flow context; hardware offload; packet processing

Detected in version: 3.2.1

4784797

Description: Rapidly creating DOCA Flow pipes until resources are exhausted may trigger a device firmware reset instead of returning a resource-exhaustion error.

Keyword: DOCA Flow; resource exhaustion; firmware reset; pipe allocation

Detected in version: 3.0.0

4771333

Description: DOCA Bench may fail to run DOCA Comch workloads when the sweep option is used.

Keyword: DOCA Bench; DOCA Comch; parameter sweep; benchmarking

Detected in version: 3.2.0

4686141

Description: DOCA DPU-GPU remote offload may report errors while establishing communication between the server and orchestrator.

Keyword: DPU-GPU remote offload; server; orchestrator; communication

Detected in version: 3.2.0

5218407 / 5220065 / NVbug 6583798

Description: Resolved a rare issue where mlx5_alloc_dbrec() in rdma-core could return an invalid non-NULL pointer instead of NULL when doorbell-record allocation failed due to an out-of-memory condition. As a result, the caller could incorrectly treat the allocation failure as a success.

Keyword: RDMA

Detected in version: 3.3.0

5106990 / 5126719 / NVbug 6494049

Description: Resolved an issue where an IPoIB interface could hang when a fabric change, multicast membership refresh, and administrative teardown occurred simultaneously.

In this scenario, the interface could remain administratively down while still appearing operationally up inside the driver. As a result, the pending “mark link usable” operation did not complete, while the teardown flow waited for it to finish, causing network configuration on the host to freeze.

Keyword: IPoIB interface

Detected in version: 3.1.0

4870459 / 4409167 / 4722911 / 4837736 / 4877800 / 4930375 / 49889678 / NVbug 6494839

Description: Resolved an issue where configuring SFs on controller 1 could cause the mlxfwreset flow to time out and fail. The underlying queue pairs did not properly transition to a reset state, creating a command resource leak that blocked the reset flow.

Keyword: mlxfwreset; SF; QPs

Detected in version: 3.2.0

5005721 / 5069001 / 5219847 / 5220048 / 5220049 / 5220050 / NVbug 6552636

Description: Resolved a truncation issue for block sizes of 4GB or larger. This issue could occur when a DPDK program used an RDMA memory region above 4GB, where the driver might not map the memory correctly. As a result, the translation table could be filled with repeated addresses instead of advancing addresses.

Keyword: RDMA memory, DPDK

Detected in version: 3.4.0

4994944 / 4999889 / 4999890 / 5000477 / NVbug 6494527

Description: Resolved an issue that could occur when ovs-vswitchd queries netdev statistics over Netlink during NIC bring-up, potentially leading to a NULL pointer dereference.
Unable to handle kernel NULL ...

Keyword: ovs-vswitchd queries

Detected in version: 3.4.0

5119365

Description: Resolved a potential kernel panic that could occur when switching eSwitch modes between legacy and switchdev while accessing mlx5 hairpin settings under /sys/class/net/<interface>/.

Keyword: kernel panic, eSwitch

Detected in version: 3.4.0

5001191 / 3573149 / 5004447 / NVbug 6494498

Description: Resolved an issue where, on an Arm VM, performing an FLR (PCI) reset while simultaneously sending commands through cmdif could cause the VM to crash.

Keyword: Arm VM, FLR

Detected in version: 3.4.0

5105915 / 5191705 / NVbug 6548013

Description: Resolved an issue where missing coordination between eSwitch reconfiguration, including LAG creation and representor registration/loading, could lead to incorrect behavior.

Keyword: eSwitch, LAG

Detected in version: 3.4.0

5145461 / 5154015 / 5189266 / 5189267 / NVbug 6435606

Description: Resolved a race condition between uplink close and eswitch devcom pair/unpair events during the teardown of representor send-to-vport rules, which could result in unexpected behavior in some cases.

Keyword: mlx5 driver

Detected in version: 3.4.0

5106086 / NVbug 6494055

Description: Resolved an issue where a memory leak could occur if device registration with mlxdevm failed.

Keyword: Memory leak, mlxdevm

Detected in version: 3.4.0

5057168 / 5057748 / 5066051 / NVBug 6494265

Description: Fixed an issue where rdma_resolve_addr() returned an ENODEV error on an IPoIB interface when the destination was an IPv4 broadcast address.

Keyword: RDMA

Detected in version: 3.2.0

5158982

Description: Resolved an issue where DOCA-Host 3.4.0 kernel drivers could fail to build with the following RHEL errata kernels and later:

• RHEL 9.8: kernel 5.14.0-687.17.1.el9_8 and later

• RHEL 10.2: kernel 6.12.0-211.20.1.el10_2 and later

Users running these or newer errata kernels should use the packages for their RHEL version from the DOCA-Host 3.4.1 repository, which adds support for the updated RHEL errata kernels and includes an updated kernel-driver source package that resolves the build failure.

Keyword: RHEL errata kernel compatibility

Detected in version: 3.4.0

5036525 / 5042525 / NVbug 6188726

Description: Resolved an issue on RPM-based systems with DKMS version 3.4.0 or later where DOCA-Host DKMS packages could fail to automatically run DKMS commands in post-install scripts because the dkms command path changed from /usr/sbin/dkms to /usr/bin/dkms.

As a result, DOCA-Host kernel drivers, as well as nvidia_peermem from nvidia-kernel-open-dkms, could become unavailable.

Keyword: DOCA-Host DKMS packages

Detected in version: 3.4.0

5125190

Description: On BlueField-3 systems running the DOCA 3.4 Ubuntu 24.04 BFB, apt update may fail, preventing packages from being updated or installed.

Keywords: APT; Ubuntu; package management

Detected in version: 4.15.0

5180884

Description: On RHEL 9 and RHEL 10, DOCA Installer may fail to start and report a missing library.

Keywords: DOCA Installer; RHEL; missing library

Detected in version: 4.15.0

5172294

Description: On JBOF systems, BFB installation may target an NVMe device other than the intended BlueField NVMe.

Keywords: BFB installation; NVMe; JBOF

Detected in version: 4.15.0

5172025

Description: After installing a RHEL 10.2 BFB, the BlueField-3 DPU may stop at the UEFI menu instead of booting.

Keywords: BFB installation; RHEL; UEFI

Detected in version: 4.15.0

4989952

Description: On certain BlueField-3 devices upgraded from firmware 4.7.1.4 to 4.13.0, the configured 5200 MT/s DDR speed may not take effect after the first reboot.

Keywords: DDR; firmware upgrade

Detected in version: 4.13.0.13932

4986356

Description: On BlueField-2 systems running DOCA 3.3 or 3.4, dmidecode may reverse the reported sizes of the L1 data and instruction caches.

Keywords: dmidecode; cache information; BlueField-2

Detected in version: DOCA 3.4.0, DOCA 3.3.0

4966842

Description: A latent Day 1 bug in the UEFI RTC driver causes the month to be calculated incorrectly (off by one). This results in unexpected date jumps and inaccurate system time reporting.

Keywords: real-time clock; system date

Detected in version: 4.15.0

4867794

Description: On RHCOS 4.20, /etc/mlnx-release may show a release name and date that do not match the installed BFB.

Keywords: RHCOS; release metadata; BFB

Detected in version: 4.15.0

4742314

4546931

4822210

Description: Running mlxfwreset on BlueField-3 with Ubuntu 24.04 may trigger an ext4 journal warning call trace (ext4_journal_check_start+0xc8/0x120); no functional impact observed, safe to ignore.

Keywords: mlxfwreset; ext4; kernel warning

Detected in version: 4.15.0

4234285

Description: During BlueField boot, the BlueField BIOS/UEFI may fail to acquire a DHCP IP address from the BMC for Redfish communication. This would cause the BIOS to bypass the Redfish step and to ignore any configuration changes in Redfish.

Keywords: DHCP; UEFI; Redfish

Detected in version: 4.15.0

5149126

Description: During a golden-image BMC firmware update, activation may fail and leave the Redfish task stuck at 0%, preventing subsequent updates.

Reported in version: 26.04

5107116

Description: In rare cases, the PLDM firmware update process in NIC mode may fail during the activation phase due to a transient communication error on the device.

Reported in version: 26.04

4966486

Description: On BlueField-3 systems, requests to /redfish/v1/Managers/Bluefield_BMC may take several minutes and return an internal error or timeout. Other Redfish resources may remain responsive, and the UEFI System Configuration page may also load slowly.

Reported in version: 26.04

5159819

Description: After a watchdog-triggered DPU reset, the reset reason may be missing from the BMC system event log.

Reported in version: 26.04

5071652

Description: In rare cases, the CEC/ERoT firmware version reported through Redfish may appear incomplete or contain blank characters.

Reported in version: 26.04

5071052

Description: During a DPU OS reboot, the BMC system event log may incorrectly report a P1 link configuration error even when the link is healthy.

Reported in version: 26.04

5044522

Description: A timing side channel in HMAC-SHA256-128 integrity verification may allow an adjacent-network attacker to infer authentication data.

Reported in version: 26.04

4833493

Description: When the BMC event log is full, rebooting the BMC may cause existing events to be lost.

Reported in version: 26.04

5174504

Description: Resolved an issue where intensive parallel QP INIT2RTR / DESTROY operations could incur high operation latency and reduce connection establishment rates when PCC was enabled.

Keywords: PCC, INIT2RTR, DOCA PCC

Detected in version:

32.49.1014

Fixed in Release:

32.50.1002

4947006 / 4980956 / 4982511

Description: Resolved two rare race conditions in the PCC steering extension when the feature is enabled through path migration, empty RTT, or a custom header.
Under rare timing conditions during QP creation or destruction, these issues could cause traffic disruption or resource leaks.

Keywords: PCC steering extension

Detected in version:

32.49.1014

Fixed in Release:

32.50.1002

5166648

Description: Resolved an issue involving an uninitialized stack read and an inverted EtherType check in the RX parser.

Keywords: RX parser

Detected in version:

32.47.1026

Fixed in Release:

32.50.1002

4937640

Description: Resolved an issue where a PPCC mlxreg read operation could succeed when executed on a PF that is not the port owner, which could have security implications.

Keywords: PCC, PPCC

Discovered in Version: 

32.49.1014

Fixed in Release:

32.50.1002

4923806 / 4867111 / 4927875

Description: Resolved an issue that could cause RoCE communication failures between specific NIC pairs after repeated queue pair (QP) create/destroy cycles. In the affected cases, traffic could enter a bad state and fail with transport retry counter exceeded errors. This fix prevents the NIC from entering that condition.

Keywords: RDMA; connection failure

Discovered in Version: 

32.46.1006

Fixed in Release:

32.50.1002

5093695

Description: Resolved an issue where using mlxconfig to configure all modules with the 0xff module split value could reset the modules to the default configuration.

Keywords: mlxconfig

Discovered in Version: 

32.48.1000

Fixed in Release:

32.50.1002

5090639

Description: Resolved an issue in single-domain mode where an FLR on a GVMI did not clear the per-domain usage bits in SP_core_diag_multi_domain->pcie_link_latency_max_read and SP_core_diag_multi_domain->pcie_link_latency_min_read. As a result, when the next GVMI took ownership and attempted to configure these counters, it could receive DIAG_DATA_LIST_WRITE_STATUS_NO_RESOURCES.

Keywords: PCIe Link Latency Arch counters

Discovered in Version: 

32.49.1014

Fixed in Release:

32.50.1002

5077326 / 4902171

Description: Resulved an issue where a slow module response to the NCSI GetModuleSerialData command could incorrectly trigger a dead IRISC watchdog assert. The firmware now handles this waiting state correctly, so slow module responses no longer cause a false watchdog hang indication.

Keywords: NCSI

Detected in version:

32.43.1014

Fixed in Release:

32.50.1002

4987961

Description: Resolved an issue where older versions failed to configure an odd num_msix value for a VF. After upgrading to the new version, odd num_msix values can now be configured for virtio VFs.

Keywords: num_msix, virtio VF

Discovered in Version: 

32.48.1000

Fixed in Release:

32.50.1002

4925548 / 4901065

Description: Resolved an issue where HIDE_PORT2_PF could be configured in non-DPU mode. Since HIDE_PORT2_PF is only relevant in DPU mode, firmware now reverts it to the valid/default configuration when set in non-DPU mode.

Keywords: DPU NIC mode

Discovered in Version: 

32.49.1014

Fixed in Release:

32.50.1002

4918436

Description: Added VSEC FLR support for ECPF and other previously skipped functions.

Keywords: VSEC FLR support for ECPF

Discovered in Version: 

32.48.1000

Fixed in Release:

32.50.1002

4976266

Description: Resolved an issue with legacy queue affinity where traffic only used the bandwidth of one physical port at all times.

Keywords: Queue affinity

Discovered in Version: 

32.48.1000

Fixed in Release:

32.50.1002

4763579 / 4877802

Description: Resolved an issue on BlueField-3 Socket Direct configurations where an mlxfwreset PCI reset could time out during MGIR polling because the SD links reset outside the firmware sync window. The firmware PCI link-down/sync window was enlarged to prevent the timeout.

Keywords: Socket Direct, mlxfwreset PCI reset

Discovered in Version: 

32.47.1026

Fixed in Release:

32.50.1002