NVIDIA BlueField DPU BSP v4.0.3
1.0

Intelligent Platform Management Interface

NVIDIA® BlueField® DPU® software will respond to Intelligent Platform Management Bus (IPMB) commands sent from the BMC via its Arm I2C bus.

Warning

The BlueField ipmb_dev_int driver is registered at the 7-bit I2C address 0x30 by default. The I2C address of the BlueField can be changed in the file /usr/bin/set_emu_param.sh.

  • BlueField Controller cards provide connection from the host server BMC to BlueField Arm I2C bus

  • BlueField DPUs provide connection from the host server BMC to the BlueField NC-SI port

  • BlueField Reference Platforms provide connection from its on-board BMC to BlueField Arm I2C bus

List of IPMI Supported Sensors

Sensor

ID

Description

bluefield_temp

0

Support NIC monitoring of BlueField’s temperature

ddr0_0_temp(a)

1

Support monitoring of DDR0 temp (on memory controller 0)

ddr0_1_temp(a)

2

Support monitoring of DDR1 temp (on memory controller 0)

ddr1_0_temp(a)

3

Support monitoring of DDR0 temp (on memory controller 1)

ddr1_1_temp(a)

4

Support monitoring of DDR1 temp (on memory controller 1)

p0_temp

5

Port 0 temperature

p1_temp

6

Port 1 temperature

p0_link

7

Port0 link status

p1_link

8

Port1 link status

Warning

(a) These sensors are not available, and hence are not populated, on BlueField DPUs.

On BlueField-2 based boards, DDR sensors and FRUs are not supported. They will appear as no reading.


List of IPMI Supported FRUs

FRU

ID

Description

update_timer

0

set_emu_param.service is responsible for collecting data on sensors and FRUs every 3 seconds. This regular update is required for sensors but not for FRUs whose content is less susceptible to change. update_timer is used to sample the FRUs every hour instead. Users may need this timer in the case where they are issuing several raw IPMItool FRU read commands. This helps in assessing how much time users have to retrieve large FRU data before the next FRU update.

update_timer is a hexadecimal number.

fw_info

1

NVIDIA® ConnectX® firmware information, Arm firmware version, and MLNX_OFED version.

The fw_info is in ASCII format.

nic_pci_dev_info

2

NIC vendor ID, device ID, subsystem vendor ID, and subsystem device ID.

The nic_pci_dev_info is in ASCII format.

cpuinfo

3

CPU information reported in lscpu and /proc/cpuinfo.

The cpuinfo is in ASCII format.

ddr0_0_spd(a)

4

FRU for SPD MC0 DIMM 0 (MC = memory controller).

The ddr0_0_spd is in binary format.

ddr0_1_spd(a)

5

FRU for SPD MC0 DIMM1.

The ddr0_1_spd is in binary format.

ddr1_0_spd(a)

6

FRU for SPD MC1 DIMM0.

The ddr1_0_spd is in binary format.

ddr1_1_spd(a)

7

FRU for SPD MC1 DIMM1.

The ddr1_1_spd is in binary format.

emmc_info

8

eMMC size, list of its partitions, and partitions usage (in ASCII format).

eMMC CID, CSD, and extended CSD registers (in binary format).

The ASCII data is separated from the binary data with "StartBinary" marker.

qsfp0_eeprom

9

FRU for QSFP 0 EEPROM page 0 content (256 bytes in binary format)

qsfp1_eeprom

10

FRU for QSFP 1 EEPROM page 0 content (256 bytes in binary format)

ip_addresses

11

This FRU file can be used to write the BMC port 0 and port 1 IP addresses to the BlueField. It is empty to begin with.

The file passed through the "ipmitool fru write 11 <file>" command must have the following format:

Copy
Copied!
            

BMC: XXX.XXX.XXX.XXX P0: XXX.XXX.XXX.XXX P1: XXX.XXX.XXX.XXX

The size of the written file should be exactly 61 bytes.

dimms_ce_ue

12

FRU reporting the number of correctable and uncorrectable errors in the DIMMs.

This FRU is updated once every 3 seconds.

eth0

13

Network interface 0 information. Updated once every minute.

eth1

14

Network interface 1 information. Updated once every minute.

bf_uid

15

BlueField UID

eth_hw_counters

16

List of ConnectX interface hardware counters

Warning

(a) On BlueField-2 based boards, DDR sensors and FRUs are not supported. They will appear as no reading.


Supported IPMI Commands

The table below provides a list of supported IPMItool command arguments.

They can be issued from the BMC in the following format:

Copy
Copied!
            

ipmitool -I ipmb <ipmitool_command_argument>

BlueField software responds to IPMItool commands issued on BlueField console. IPMItool commands on Bluefield console are supported regardless if a host server BMC is connected to the Arm I2C bus on BlueField.

The format for these commands is as follows:

Copy
Copied!
            

$ ipmitool -U ADMIN -P ADMIN -p 9001 -H localhost <ipmitool_command_argument>

Command Description

IPMItool Command

Relevant IPMI 2.0
Rev 1.1 Spec Section

Get device ID

mc info

20.1

Broadcast “Get Device ID”

Part of "mc info"

20.9

Get BMC global enables

mc getenables

22.2

Get device SDR info

sdr info

35.2

Get device SDR

"sdr get", "sdr list" or
"sdr elist"

35.3

Get sensor hysteresis

sdr get <sensor-id>

35.7

Set sensor threshold

sensor thresh <sensor-id> <threshold> <setting>

  • sensor-id – name of the sensor for which a threshold is to be set

  • threshold – which threshold to set

    • ucr – upper critical

    • unc – upper non-critical

    • lnc – lower non-critical

    • lcr – lower critical

  • setting – the value to set the threshold to

To configure all lower thresholds, use : sensor thresh <sensor-id> lower <lnr> <lcr> <lnc>

Warning

The lower non-recoverable <lnr> option is not supported

To configure all upper thresholds, use: sensor thresh <sensor-id> upper <unc> <ucr> <unr>

Warning

The upper non-recoverable <unr> option is not supported

35.8

Get sensor threshold

sdr get <sensor-id>

35.9

Get sensor event enable

sdr get <sensor-id>

35.11

Get sensor reading

sensor reading <sensor-id>

35.14

Get sensor type

sdr type <type>

35.16

Read FRU data

fru read <fru-number> <file-to-write-to>

34.2

Get SDR repository info

sdr info

33.9

Get SEL info

"sel" or "sel info"

40.2

Get SEL allocation info

"sel" or "sel info"

40.3

Get SEL entry

"sel list" or "sel elist"

40.5

Add SEL entry

sel add <filename>

40.6

Delete SEL entry

sel delete <id>

40.8

Clear SEL

sel clear

40.9

Get SEL time

sel time get

40.1

Set SEL time

sel time set "MM/DD/YYYY HH:M:SS"

40.11

  1. Load the BlueField CentOS image.

    Warning

    The following steps are performed from the BlueField CentOS prompt. The BlueField is running CentOS 7.6 with kernel 5.4. The CentOS installation was done using the CentOS everything ISO image.

    The following drivers need to be loaded on the BlueField running CentOS:

    – jc42.ko
    – ee1004.ko
    – at24.ko
    – eeprom.ko
    – i2c-dev.ko

    Example of loading ee1004.ko, at24.ko, and eeprom.ko:

    Copy
    Copied!
                

    modprobe ee1004 modprobe at24 modprobe eeprom

    The i2c-dev module is built into the kernel 5.4.60 on CentOS 7.6.

  2. Optional: Update the i2c-mlx driver if the installed version is older than version i2c-mlx-1.0-0.gab579c6.src.rpm.

    1. Re-compile i2c-mlx. Run:

      Copy
      Copied!
                  

      $ yum remove -y kmod-i2c-mlx $ modprobe -rv i2c-mlx

    2. Transfer the i2c-mlx RPM from the BlueField software tarball under distro/SRPM onto the Arm. Run:

      Copy
      Copied!
                  

      $ rpmbuild --rebuild /root/i2c-mlx-1.0-0.g422740c.src.rpm $ yum install -y /root/rpmbuild/RPMS/aarch64/i2c-mlx-1.0-0.g422740c_5.4.17_mlnx.9.ga0bea68.aarch64.rpm $ ls -l /lib/modules/$(uname -r)/extra/i2c-mlx/i2c-mlx.ko

    3. Load i2c-mlx. Run:

      Copy
      Copied!
                  

      $ modprobe i2c-mlx

  3. Install the following packages:

    Copy
    Copied!
                

    $ yum install ipmitool lm_sensors

    If the above operation fails for IPMItool, run the following to install it:

    Copy
    Copied!
                

    wget http://sourceforge.net/projects/ipmitool/files/ipmitool/1.8.18/ipmitool-1.8.18.tar.gz tar -xvzf ipmitool-1.8.18.tar.gz cd ipmitool-1.8.18 ./bootstrap ./configure make make install DESTDIR=/tmp/package-ipmitool

  4. The i2c-tools package is also required, but the version contained in the CentOS Yum repository is old and does not work with BlueField. Therefore, please download i2c-tools version 4.1, and then build and install it.

    Copy
    Copied!
                

    # Build i2c-tools from a newer source wget http://mirrors.edge.kernel.org/pub/software/utils/i2c-tools/i2c-tools-4.1.tar.gz tar -xvzf i2c-tools-4.1.tar.gz cd i2c-tools-4.1 make make install PREFIX=/usr   # create a link to the libraries ln -sfn /usr/lib/libi2c.so.0.1.1 /lib64/libi2c.so ln -sfn /usr/lib/libi2c.so.0.1.1 /lib64/libi2c.so.0

  5. Generate an RPM binary from the BlueField's mlx-OpenIPMI-2.0.25 source RPM.

    The following packages might be needed to build the binary RPM depending on which version of CentOS you are using.

    Copy
    Copied!
                

    $ yum install libtool rpm-devel rpmdevtools rpmlint wget ncurses-devel automake $ rpmbuild --rebuild mlx-OpenIPMI-2.0.25-0.g581ebbb.src.rpm

    Warning

    You may obtain this rpm file by means of scp from the server host's Bluefield Distribution folder. For example:

    Copy
    Copied!
                

    $ scp <BF_INST_DIR>/distro/SRPMS/mlx-OpenIPMI-2.0.25-0.g4fdc53d.src.rpm <ip-address>:/<target_directory>/

    If there are issues with building the OpenIPMI RPM, verify that the swig package is not installed.

    Copy
    Copied!
                

    $ yum remove -y swig

  6. Generate a binary RPM from the ipmb-dev-int source RPM and install it. Run:

    Copy
    Copied!
                

    $ rpmbuild --rebuild ipmb-dev-int-1.0-0.g304ea0c.src.rpm

  7. Generate a binary RPM from the ipmb-host source RPM and install it. Run:

    Copy
    Copied!
                

    $ rpmbuild --rebuild ipmb-host-1.0-0.g304ea0c.src.rpm

  8. Load OpenIPMI, ipmb-host, and ipmb-dev-int RPM packages. Run:

    Copy
    Copied!
                

    $ yum install -y /root/rpmbuild/RPMS/aarch64/mlx-OpenIPMI-2.0.25-0.g581ebbb_5.4.0_49.el7a.aarch64.aarch64.rpm $ yum install -y /root/rpmbuild/RPMS/aarch64/ipmb-dev-int-1.0-0.g304ea0c_5.4.0_49.el7a.aarch64.aarch64.rpm $ yum install -y /root/rpmbuild/RPMS/aarch64/ipmb-host-1.0-0.g304ea0c_5.4.0_49.el7a.aarch64.aarch64.rpm

  9. Load the IPMB driver. Run:

    Copy
    Copied!
                

    $ modprobe ipmb-dev-int

  10. Install and start rasdaemon package. Run:

    Copy
    Copied!
                

    yum install rasdaemon systemctl enable rasdaemon systemctl start rasdaemon

  11. Start the IPMI daemon. Run:

    Copy
    Copied!
                

    $ systemctl enable mlx_ipmid $ systemctl start mlx_ipmid $ systemctl enable set_emu_param $ systemctl start set_emu_param

  12. Test if the IPMI daemon responds on the BlueField. For example, run:

    Copy
    Copied!
                

    $ ipmitool -U ADMIN -P ADMIN -p 9001 -H localhost mc info

  13. From the BMC, run:

    Copy
    Copied!
                

    $ ipmitool -I ipmb mc info

  14. Test that the BlueField can send requests to the BMC. Run:

    Copy
    Copied!
                

    $ ipmitool mc info

It is possible for the external host to retrieve IPMI data via the OOB interface or the ConnectX interfaces.

To do that, set the network interface address properly in progconf. For example, if the OOB ip address is 192.168.101.2, edit the OOB_IP variable in the /etc/ipmi/progconf file as follows:

Copy
Copied!
            

root@localhost:~# cat /etc/ipmi/progconf SUPPORT_IPMB="NONE" LOOP_PERIOD=3 BF_FAMILY=$(/usr/bin/bffamily | tr -d '[:space:]') OOB_IP="192.168.101.2"

Then reboot or restart the ipmi service as follows:

Copy
Copied!
            

systemctl restart mlx_ipmid

BlueField has 2 IPMB modes. It can be used as a responder but also as a requester.

  • Responder Mode
    When used as a responder, the BlueField receives IPMB request messages from the BMC on SMBus 2. It then, processes the message and sends a response back to the BMC. In this case, the BlueField needs to load the ipmb_dev_int driver.

    Copy
    Copied!
                

    BMC (requester) ----IPMB/SMBus 2----> BlueField (responder)

  • Requester Mode
    When used as a requester, the BlueField sends IPMB request messages to the BMC via SMBus 2. The BMC then, processes the request and sends a message back to the BlueField. So the BlueField needs to load the ipmb_host driver when the BMC is up. If the BMC is not up, ipmb_host will fail to load because it has to execute a handshake with the other end before loading.

    Copy
    Copied!
                

    BlueField (requester) ----IPMB/SMBus 2----> BMC (responder)

Both modes are enabled automatically at boot time on Yocto.

Warning

Once the set_emu_param.service is started, it will try to load the ipmb_host drivers. If the BMC is down or not responsive when BlueField tries to load the ipmb_host driver, the latter will not load successfully. In that case, make sure the BMC is up and operational, and run the following from BlueField's console:

Copy
Copied!
            

echo 0x1011 > /sys/bus/i2c/devices/i2c-2/delete_device rmmod ipmb_host

The set_emu_param.service script will try to load the driver again.

BlueField and BMC I2C Addresses on BlueField Reference Platform

BlueField in Responder Mode

Device

I2C Address

BlueField ipmb_dev_int

0x30

BMC ipmb_host

0x20

BlueField in Requester Mode

Device

I2C Address

BlueField ipmb_host

0x11

BMC ipmb_dev_int

0x10

To use a different BlueField or BMC I2C address, you must make changes to the following files' variables.

Filename Path

Parameter Change

/usr/bin/set_emu_param.sh

The ipmb_dev_int and ipmb_host drivers are registered at the following I2C addresses:
IPMB_DEV_INT_ADD=<BlueField I2C Address 1>
IPMB_HOST_ADD=<BlueField I2C Address 2>
These addresses must be different from one another. Otherwise, one of the drives will fail to register.

To change the BMC I2C address:
IPMB_HOST_CLIENTADDR=<BMC I2C Address>

<I2C address> must be equal to: 0x1000+<7-bit I2C address>

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