nmxcli Management Tools | ESXi Driver User Manual

Overview

nmlxcli is the command-line management and diagnostics utility for NVIDIA native ESXi NIC drivers (nmlx5). It talks directly to the loaded kernel driver and lets the user:

Discover Mellanox uplinks and read their general, link, and status information
Read private (driver-level) statistics: rings, RX queues, flows, FW pages, HW tracer
Inspect RSS engines, DVX VFs, VPORTs, and connection-tracking objects
Configure ECN / RoCE congestion control
Enable/disable the port sniffer (including RDMA traffic)
Collect debug dumps: steering tables, FW resource dumps, CR dumps, FW tracer logs
Apply non-volatile device configuration (nvconfig) profiles
Set the E-Switch admin state for all SR-IOV VFs

The tool is installed at /opt/mellanox/bin/nmlxcli and is also registered as an esxcli plugin under the mellanox namespace (not all nmlxclioperations are available on esxcli).

Requirements: an ESXi host with the nmlx5 native driver loaded, and root privileges. If the driver isn't loaded or its version doesn't match the tool, nmlxcli reports a driver-compatibility error (and logs it to syslog).

Executing nmlxcli commands

esxcli mellanox <namespace/command>...

This is the supported, user-facing interface. The output is formatted, there’s built-in help (--help on any namespace), and input validation.

esxcli command reference

`esxcli mellanox uplink`

Command	Purpose	Required options
`list`	List all Mellanox uplinks with PCI, MAC, driver/FW version, WoL support	—
`info`	General information about one uplink	`-u <vmnicX>`
`state`	Uplink state (enabled/disabled/promisc/…)	`-u`
`timeouts`	Uplink timeout configuration, in milliseconds	`-u`

`esxcli mellanox uplink link`

Command	Purpose
`info`	Link status, duplex, speed, port type, auto-negotiation, transceiver, advertised link modes

`esxcli mellanox uplink coalesce`

Command	Purpose
`get`	Rx/Tx usecs and frames, adaptive Rx/Tx state, sample interval
`adaptive get`	Low/high packet-rate thresholds and their Rx/Tx usecs and frames

`esxcli mellanox uplink stats`

Command	Purpose
`all`	All private ring statistics
`activeRx`	Active RX queue-oriented statistics

`esxcli mellanox uplink flows stat`

Command	Purpose
`query`	Flow counters: created / destroyed / in-use. `-u` optional (all uplinks if omitted)
`clean`	Reset the flow counters. Note: results may become momentarily inconsistent

`esxcli mellanox uplink ecn`

ECN (congestion control for RoCE) is managed per protocol. Two protocols exist:

rRoceRp — Reaction Point (the sender that reduces its rate on congestion notification)
rRoceNp — Notification Point (the receiver that generates CNPs)

Each has the same verbs:

Command	Purpose
`enable` / `disable`	Turn the protocol on or off
`query`	Show whether the protocol is enabled
`get`	Show all tuning parameters
`set`	Change one or more tuning parameters

Plus:

Command	Purpose
`statistics get`	ECN counters
`statistics clear`	Reset ECN counters

rRoceRp set parameters (all optional, all take integers) control the rate-reduction/recovery algorithm: --clamp-tgt-rate, --clamp-tgt-rate-after, --rpg-time-reset, --rpg-byte-reset, --rpg-threshold, --rpg-max-rate, --rpg-ai-rate, --rpg-hai-rate, --rpg-gd, --rpg-min-dec-fac, --rpg-min-rate, --rate-first-cnp, --dce-tcp-g, --dce-tcp-rtt, --rate-reduce-period, --init-alpha.

rRoceNp set parameters control CNP generation: --min-time-cnp, --cnp-dscp (0–63), --cnp-prio-mode (0 = use configured 802.1p priority, 1 = copy from received packet), --cnp-802-p-prio (0–7).

Run esxcli mellanox uplink ecn rRoceRp set --help for the per-parameter descriptions and valid ranges.

`esxcli mellanox uplink sniffer`

Command	Purpose
`enable`	Enable the sniffer, including RDMA traffic
`disable`	Disable the sniffer
`query`	Show current sniffer state

`esxcli mellanox uplink fw`

Command	Purpose
`tracer enable` / `disable`	Turn FW tracing on/off for an uplink
`tracer state`	Tracer state plus counters (printed/excluded logs, events, errors, lost events, buffer overruns). `-u` optional
`resource menu`	List the available FW resource-dump segments. `-i <index>` narrows to one entry
`resource dump`	Dump a FW resource segment to a file

resource dump requires -u, -s <segment-type>, and -f <file-path>. Optional narrowing: -i/-j (indices), -o/-p (object counts), -v (VHCA ID).

`esxcli mellanox uplink rss`

Command	Purpose
`list`	All RSS engines on the uplink
`priQ`	One RSS engine, selected by primary OS/VMK queue ID (`-q`)

`esxcli mellanox uplink dvx`

Command	Purpose
`list`	All DVX VFs on the uplink

`esxcli mellanox uplink vfs`

Command	Purpose
`setAdminState`	Set the E-Switch admin state for all VF vports (`-z 0` = UP, `-z 1` = AUTO). The PF is left unchanged

`esxcli mellanox uplink conntrack`

Command	Purpose
`dumpVmk`	Connection-tracking objects in VMkernel format
`dumpHw`	Connection-tracking objects in hardware format

Both dump everything by default; -i <hw-id> dumps a single object.

`esxcli mellanox nvconfig`

Command	Purpose
`profile`	Apply a non-volatile configuration profile to a device

Requires -x <sbdf> (PCI address, function must be 0) and -w <profile>:

0 — DEFAULT: reset all configuration to factory defaults
1 — DPU: configure both host and embedded CPU on BlueField devices
2 — BF_AS_CX: configure a BlueField device to act as a plain NIC
3 — CX: configure a ConnectX device

Some profiles can only configure the currently-enabled PFs; in that case the tool tells you to re-apply the profile after a reboot to cover the remaining PFs.

Last updated: August 04, 2026