Lenovo YUM Repository
Release and Change History
Lenovo YUM Repository
Release and Change History
| Bundle Release Information: |
| Targeted Server Family (Machine Type): ThinkSystem ST58 V3 Server (7DF5) |
| Targeted Operating System: SLES15SP5 |
| YUM Repository Build Date: 2024_12_17 |
| Device: |
| ThinkSystem Intel X710-T4L 10GBASE-T 4-Port PCIe Ethernet Adapter | Part Number: 4XC7A79699 Feature Code: AUZW |
||||
|---|---|---|---|---|---|
| PCIe Sub Vendor ID: 0x8086 PCIe Sub Device ID: 0x4019 |
| Release History: |
| Continued Support in Owasco - Latest Release | Change History Documents |
| Release ST45V3 Change History Documents |
===========================================================================
Intel v29.1 Network FW Update Release for Windows
Release date = 08/01/2024
===========================================================================
CONTENTS
________
1.0 Change history
2.0 Web information and support phone Number
3.0 Notices and trademarks
4.0 Disclaimer
1.0 Change history
___________________
This release includes the following updates:
============================================
Versions:
1) The following X710images are at NVM 9.50 level.
------------------------------------------------------------------------------------------------
Device Boot ROM (Option ROM) Firmware (Etrack ID)
------------------------------------------------------------------------------------------------
ThinkSystem X710-T2L PCIe 1.3616.0 8000F48E
ThinkSystem X710-T2L OCP 1.3616.0 8000F2A9
ThinkSystem X710-T4L PCIe 1.3616.0 8000F48D
ThinkSystem X710-T4L OCP 1.3616.0 8000F2AB
ThinkSystem X710-DA2 PCIe 1.3616.0 8000F2A8
2) NVMUPDATE tool updated to 1.41.11.4 to support new config version.
This update supports Lenovo ThinkSystem and ThinkAgile products.
Important Note: If you choose to install the nvmupdate64e utility manually, please ensure that you install
version 1.41.11.4 or newer.
Earlier versions of nvmupdate64e tool may not be not compatible or include all latest supported features.
New features:
- Enable E700 X722 Tx signal to "always on".
- Enable E700 to store events details on the NVM in Recovery Mode.When
entering the Recovery Mode, X710 will store the details of the event in
the NVM.
Bug fixes:
- Resolved an issue where POST to the
NetworkAdapter.ResetSettingsToDefault looked successful from postman,
but after reboot the adapter's Port 1 LLDP setting did not reset to
default.
- Previously, there were cases of Fortville disappearing from the PCIe
bus during reboot stress testing reported on Icelake-based platforms.
These platforms performed a PCIe Hot Reset before asserting PERST# as
part of the warm reboot cycle. PERST# is asserted at approximately the
end of the Hot Reset, but due mostly to the Fortville LCB
implementation, there was a variation in the relative timing of about
65 us, so many edge cases could occur.
- Previously, after configuring "ifconfig ethX down" command on Marvell
devices series 88E1512/88E1514, the peer side (loopback configuration)
was still reporting link up.
- Fixed an issue. Unexpected EMP Reset occurred in a rarely BMC request
combination in a very low probability, which may cause: 'tx_timeout' or
'ARQ: Unknown event 0x0000 ignored' or 'bond down' if bonding is
configured.
- IPS: 00879846 - Intel XXV710-DA2 PCIe 25Gb 2-Port SFP28 Adapter link
speed and status incorrect
Description:
Link event is reported only once at when the machine is powered on,
after that no link event will be generated any more even after the
network cable is inserted and removed, this is incorrect, link event
should be reported when we plugin or remove the AOC cable.
Fix:
Firmware reported link speed value based on advertised capability value
using DNL declaration, changes to read link speed from MAC
- IPS: 00852370 - IEPD Hot Issue: SR645v3 X710 T2L PCIe bandwidth degraded
Description:
X710 PCIe may enter x1 on the AMD Genoa platform in reboot and AC
process.
Customer updated system firmware XCC: KAX318V >> KAX322V UEFI: KAE112O
>> KAE116K Following event occurs intermittently, but high fail rate
PCIe Link Width has degraded from x8 to x1 in physical slot number 3.
Slot 3 is Intel X710-T2L 10GBASE-T 2-Port PCIe Ethernet Adapter,
customer had tried new NIC FW, replace NIC but no help. this system no
re-timer.
Fix:
The firmware changes to enter the PCIE detection phase when PERST
becomes de-asserted
Release net-9.30-6.2-1.3450.0-0 includes the following updates:
===============================================================
New features:
- Support RHEL8.8/9.2, SLES 15.5
- Get Link Status AQC response when SFP module is not initialized.
Description:
If the Get Link Status AQC is sent when the SFP module is not
initialized yet, the link can be lost.
When the SFP module is not initialized yet, FW will try again in a
reasonable timeout.
Bug fixes:
- IPS: 00840914 - [PA_IO Agile-Sprint202310_BDC_Storage_flex]SRIOV
enable/disable both can do SRIOV features
Description:
OS can still create SRIOV VFs when the adapter HII already disables SRIOV
in the BIOS
IPS: 00841528 - [PA_IO Agile-Sprint202310_BDC_Storage_SD530] Intel X722
LOM test SR-IOV function cannot enable VF under ESXi8.0u2 OS
Description:
Open to enter ESXi Host Client-Manage-Hardware-PCI Devices-Select Port
"Intel X722 10GB SFP+" Configure SR-IOV 2.UEFI has set SR-IOV function
3.ESXi8.0u2 Unable to enable VF redhat 9.2 can work well.
Fix:
Correctly set SR-IOV Control Capable bit for Lenovo images,
- IPS: 00839473 - [PA_IO Agile-Sprint202209_BDC_oceancat_Sprint202310]X722
lom WOL fail in SD650.
Description:
WOL function UEIF page show gray, OS can not enable WOL function
Fix:
Enable force link down requests and map channel number in a descending
order. Enable AltRAM implementation then reset corrupted memory block and
recalculate MSI_X VF vectors.
- IPS: 00831248 - XXV710 VPD info question
Description:
lspci -s BUS:Device.Function -vvv command to check VPD info, but it
can't show out the VPD info
Fix:
Enable VPD bit in NVM image
- IPS: 00769725 - [Hakuba] Abnormal cable plug (unplug) event on port 2
and 3 with Intel X710-T4L
Description:
Port2 and port3 of X710-T4L PCIE cannot display event in audit log and
link status/speed in inventory when plug/unplug the cable.
Fix:
There are missing MCTP link status changes as mixed PF no. with port
no. in the MCTP Link Status change handler, and overflowing PLDM queue
with RDE events that were classified as Big Event.
Map Port No. to the corresponding PF and continue, add additional
condition to disable unplanned eventReceiver physical address change.
Disable RDE events in the presence of asynchronous MCTP implementation
without polling for Big Events
Note: it may still occur in loopback mode.
- IPS: 00725539 - Link down on x722 4x10Gb SFP+ PHY card with 27.3 code
level (800039xx)
Description:
After upgrading the FW, all ports are down.
Fix:
Force OCP PnP Autodetection flow after NVM update.
- Resolved an issue where a rare PCI speed degradation with Gen3 to Gen2
was seen during extensive link tests after a reset is caused by a core
reset on an ICX-D platform.
- Resolved an issue where during a write, with http PATCH method
triggered, the LLDP Disable Valid bit was ignored, which resulted in a
skip of write in all scenarios instead of just for the same LLLDP
Enabled value.
- Resolved an issue where if BMC bonding was disabled, NCSI mode was set
to a Manual Switch Mode (depending on the platform), and BMC share port
has changed before doing a power cycle, recovery mode would occur.
Other prerequisites include disabled WOL, disabled LLDP, enabled PXE no
drop, and disabled auto power-on after power loss (in BIOS).
- Fixed an issue where the processing port id was always the same as the
mac address
- Previously, the addition of PCIe re-timers added to the total channel
latency causing replay timer timeout correctable errors.
- Resolved an issue where PLDM type MCTP packets were not accepted due to
missing configuration for PCIe VDM
Release net-9.2-6.2-1.3357.0-2 includes the following updates:
==============================================================
New features:
- Defines the IID (Instance IDs) field in the NC-SI packet's header,
which is used to match responses to requests and to differentiate
between new and retried commands.
Description:
According to the spec, If the IID of the current command is the same as
the previous command's IID, the command is a "Retried Command" and the
FW shall return the response of previous command (should be done
without re-executing the command). When the IID is different than from
the previous command, the command is executed.
The NC-SI spec is unclear on if the incoming commands IID is compared
to the IID of the previous command received on the same channel,
package or device. This feature defines a transfer to maintaining a
single buffer for the last NC-SI command received on the device per
transmitting interface.
Bug fixes:
- XXV710DA2 adapter intermittently lose link when doing DC power cycle test
Description:
XXV710DA2 adapter intermittently lose link when doing DC power cycle
test, this can be reproduced in about 200 DC cycle.
Fix:
When the FW starts to run, the gearbox has to be initialized and
configuration needs to be downloaded from the NVM and applied. If one
of the operation fails the FW will try again (several times). In this
case the configuration failed with the error which was not covered.
Error handling extended in the code to cover all cases
- Package Port Control command with Port Control equals 0x1 works like
reset if link is already up
Description:
In Campbell Pond, send Package Port Control Command to disable the
port, send command to enable it, then send command to enable it again.
The port is down instead of up.
Fix:
Added link status check in package command.
- Link Status Change AEN event is not send after link down
Description:
Sending AEN Enable command with Link Status Change equals true, then
sending Port Control Command with Port Control field equals 0x0 -
disable. There is no event after link down.
Fix:
Changed event rejection condition to response.
- After running reboot 123 times with qtest, slot4 (X710-T2L) shows an error
Description:
Used qtest ro run reboot 123 times, xcc web appears: PCIe Correctable
Error Threshold limit has been exceeded at Segment 0x0000 Bus 0xA8
Device 0x00 Function 0x01.
Fix:
Adjust the replay timer timeout value for different scenarios.
Release net-9.0-6.0-1.3266.0-6 includes the following updates:
==============================================================
New features:
- Windows 10,11 support for X710-T2L/T4L and XXV710
- Support RHEL8.6/9.0, SLES 15SP4
Bug fixes:
- IPS: 00685894 - I210 Option Rom Request
Description:
Lenovo required support for specific 4-part ID, additional HII fields
in "NIC Configuration" menu for I210 AT, remove "Device Level
Configuration" and "iSCSI Configuration" fields
Fix:
Added and removed corresponding fields.
- IPS: 00729600 - XXV710 Link issues
Description:
While DC power cycle, the port maybe lose the link
Fix:
When the FW starts to run, retry to configure the Retimer in case of
error.
1) With v27.1 release, add support for new SKUs:
- ThinkSystem Intel X710-T2L 10GBASE-T 2-Port PCIe Ethernet Adapter
- ThinkSystem Intel X710-T4L 10GBASE-T 4-Port PCIe Ethernet Adapter
- ThinkSystem Intel X710-T4L 10GBASE-T 4-Port OCP Ethernet Adapter
- ThinkSystem Intel X710-T2L 10GBASE-T 2-port Internal Lock OCP Ethernet Adapter
- ThinkSystem Intel I350 1GbE RJ45 4-port Internal Lock OCP Ethernet Adapter
2) With v27.3 release:
- FW/NVM updates for X722, X710, XXV710 and X550 based products.
Add support for new SKUs:
- Intel(R) Ethernet I210-T1 LOM
Note on Firmware Recovery Mode:
When a device is in Firmware Recovery mode it will not pass traffic or allow any configuration; you can only
attempt to recover the device's firmware. A device will enter Firmware Recovery mode if it detects a problem
that requires the firmware to be reprogrammed.
Resolving Firmware Recovery Mode Issues:
If your device is in Firmware Recovery mode you can restore it to factory defaults using the latest NVM
Update Package. Download the latest NVM Update Package from your vendor's support website and follow
the instructions in it.
*** NOTES:
- Before initiating device recovery, the integrity of the host operating system, device drivers
and firmware utilities must be verified and reinstalled if necessary. Fully functional operating
system, device drivers and tools are required for device recovery.
- AC power cycle your system even if the instructions do not indicate that an AC power cycle is required.
A power cycle is required to enable the device to restore and reload critical device settings to enable recovery.
To perform an AC power cycle, do the following:
a. Shut down the system if it is powered up.
b. Unplug all AC power cords from the system.
c. Leave the AC power cords unplugged for 15 seconds to allow the system power supply to discharge completely.
d. Plug in AC power cords to the system.
- Firmware Recovery Mode Errors and Messages:
When a device is in Firmware Recovery mode, the device drivers and preboot software and utilities may log or
display messages such as the following:
a. Firmware recovery mode detected. Limiting functionality.
Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.
b. Firmware recovery mode detected. Initialization failed.
c. Firmware recovery mode detected. Limiting functionality.
d. PXE-E0C: Firmware recovery mode detected. Initialization failed.
2.0 Web Sites and Support Phone Number
______________________________________
o You can download new and updated files for Lenovo products from the Lenovo
Web site:
http://datacentersupport.lenovo.com/
o With the original purchase of an Lenovo hardware product, you have access to
extensive support coverage. During the Lenovo hardware product warranty
period, you may call the Lenovo HelpCenter for hardware product assistance
covered under the terms of the Lenovo hardware warranty.
3.0 Trademarks and Notices
__________________________
Lenovo is a registered trademark of Lenovo Corporation in the United States
and other countries.
Intel is a registered trademark of Intel Corporation in the United States
and other countries.
Linux is a registered trademark of Linus Torvalds in the United States and
other countries.
Other company, product, and service names may be trademarks or service marks
of others.
4.0 Disclaimer
______________
THIS DOCUMENT IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND.
Lenovo DISCLAIMS ALL WARRANTIES, WHETHER EXPRESS OR IMPLIED,
INCLUDING WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF FITNESS
FOR A PARTICULAR PURPOSE AND MERCHANTABILITY WITH RESPECT TO THE
INFORMATION IN THIS DOCUMENT. BY FURNISHING THIS DOCUMENT, Lenovo
GRANTS NO LICENSES TO ANY PATENTS OR COPYRIGHTS.
Note to U.S. Government Users -- Documentation related to
restricted rights -- Use, duplication or disclosure is subject
to restrictions set forth in GSA ADP Schedule Contract with
Lenovo Corporation.
Lenovo Data Center Group Linux OS Support Home Page |
| linux.lenovo.com |
© 2018-2024 Lenovo. All rights reserved |
===============================================================================
Intel Linux device drivers package update
===============================================================================
Driver Version
------ -------
ice 1.14.9
i40e 2.25.7
ixgbe 5.20.3
igb 5.16.3
irdma 1.14.17
ice:
----
- Add Switch Recipe resources reuse
- Support FW Debug Dump
- Support internal temperature reading
- Support RoCEv2 Active/Active LAG
- Add delay of SW initialization during PHY FW upload
- Add temperature sensor readouts
- Add "No FEC" mode to allow the LESM's "Auto FEC" mode to use "No FEC"
in link establishment process
- Add support of ethtool flow-type ether
- Rx timestamp functionality for VFs
- Traffic mirroring via tc-mirred
- FW debug logs support in debugfs
- Introduce flexible VF loopback pacing configuration to improve
VF-to-VF throughput
- Devlink health support for Tx hang debugging
- Add PHY statistics dump
- LLDP support on VF
- DMA_buf w/o ODP support for Linux RDMA
- Print NIC FW version
- Multicast promiscuous mode doesn't work after PF-to-VF reset.
- Fixed issue of unexpected FW logs during some special environments,
such as tx_timeout or NVM update.
- Ice driver doesn't work when link-down-on-close is on.
- Input/output error shown after "link-down-on-close" priv-flags enabled.
- NULL pointer dereference happens in the ICE out-of-tree ICE driver.
Determine if PF Reset is in progress before accessing the coalesce
settings.
- It was observed that in some cases the TC rule is not removed when
adapter works in the switchdev mode, what was caused by improper
algorithm of TC rule removal method.
- "Scalable IOV" help text in Device Level Configuration is not
expected.
- Traffic statistics reported by ice driver have been changed to also
include RDMA traffic.
- Out-of-tree drivers don't support XDP redirect on v6.3. Report XDP
functionality through XDP features flag.
- DVM mode is not configured in parser library for raw packet filter.
Align the DVM configuration between kernel driver and hardware/DDP.
- Ice driver misc interrupt not getting generated for a interface.
Re-enable timestamp at the end of the BH of the interrupt.
i40e:
-----
- Add PHY debug register dump
- Add mdd-auto-reset-vf private flag
- Add core reset trigger to ethtool
- Add restore default speed when changed module does not supports
current speed settings
- Add tracepoints for nvmupdate troubleshooting
- Add tracepoints for pci reset callbacks for troubleshooting
- Add support for reading Trace Buffer
- Add ability to reset vf for tx mdd event also
- Add trace events related to SFP module IOCTLs
- Fix 32bit FW gtime wrapping issue for nvmupdate
- Fix err -110 after DCB ETS configuration
- Fix i40e_setup_misc_vector() error handling
- Fix missing rtnl locking when setting up pf switch
- Fix accessing vsi->active_filters without holding lock
- Fix SW DCB configuration when ADQ is set
- Fix PTP pins verification
- Fix passing tainted expression to kzalloc()
- Fix error by one for traffic class number
- mq core affinity reset on interface link lost. Removes
irq_set_affinity_hint() calls to make driver honor user affinity
settings
- I40e driver report "vf may be used uninitialized in this function"
during make process.
irdma:
------
- Support CVL 4.50, FVL 9.50
- Support RoCEv2 Active/Active LAG
- dma_buf w/o ODP support
ixgbe:
------
- Fix waiting for flash completion ACK
- Fix debugfs directory double creation
- Fix terminate probing right after wrong API detected
- mq core affinity reset on interface link lost/link state. Remove
irq_set_affinity_hint() calls to make driver honor user affinity
settings
igb:
----
- Fix array size overrun
- Fix adjfreq compatibility
Version [28.2]
===============================================================================
Driver Version
------ -------
ice 1.12.7
i40e 2.23.17
ixgbe 5.19.6
igb 5.14.16
irdma 1.12.55
New OS Support:
- SLES 15.5
ice:
----
- Large VF support (up tp 256 Qs)
- User can reassign PF LUT for VF usage
- Devlink port split support
- Tx rate limit support
- Dynamic msi-x vector allocations to various device functions via
devlink API
- PTP fixes for Timestamp timeout error.
- Resolved an issue where when using certain DDP package versions,
802.1ad type VLANs might not be correctly enabled on SIOV or SR-IOV
interfaces.
- Resolved an issue where running traffic via Open vSwitch (when an
SR-IOV VF have specific ipv6 address assigned), works for 20min
(until the neighbor table is clean on the switch/router) then it
stops until IPv6 from VF is deleted.
- Previously, the Switchdev feature's VF to VF communication was not
functional in the ice driver versions 1.10.1.2, 1.10.1.2.2, and
1.11.14.
- Resolved an issue where the NVM update process would need Power On
Reset on PTP devices unless the user manually turned off PTP activity
before the NVM update process was started.
- Resolved an issue where a performance issue caused the GNSS write
commands to be delayed and not delivered until the next commands
appeared.
- Resolved an issue where modifying inner or outer VLAN offload caps
while the link is down might break communications when the link is
later brought back up.
- Resolved an issue for Linux ice driver v1.7.16+ on E810 four-port
SKUs where VF could not be created on certain high-core count
platform because PF had exhausted all the MSIX interrupts.
- Resolved an issue where repeatedly adding/deleting a VF from a
namespace while also repeatedly changing its trust mode status could
result in call trace after a significant number of iterations.
- Previously an issue existed in the IAVF driver where the VF couldn't
communicate on inner or outer VLANs if the number of rx queues was
modified after switching off VLAN offloads.
- Resolved an issue when spoofchk was turned on, the VF device driver
would have pending DMA allocations while it was released from the
device.
- Previously a UEFI PXE installation of Red Hat Enterprise Linux 8.4 on
a local disk resulted with the system failing to boot.
- Resolved an issue where auxiliary module conflicts would be reported
between i40e and ice during the generation and installation of Linux
rpm packages.
- Resolved an issue where "ethtool -L/G" made E810 link down.
i40e:
-----
- Add PHY debug register dump
- Add mdd-auto-reset-vf private flag
- Fix FW v8.15 double vlan issue
- Fix 32bit FW gtime wrapping issue for nvmupdate
- Fix err -110 after DCB ETS configuration
- Fix pointer dereference after free
- Fix i40e_setup_misc_vector() error handling
- Fix missing rtnl locking when setting up pf switch
- Fix accessing vsi->active_filters without holding lock
- Fix SW DCB configuration when ADQ is set
- Fix PTP pins verification
- Fix passing tainted expression to kzalloc()
- Fix error by one for traffic class number
- Fix compilation issue with gcc older than 5.0.0
irdma:
------
- Support CVL 4.30, FVL 9.30
- Add readable descriptions to the printed AE codes in irdma to help
with debug and provide a better insight on what is going on.
- Fix the fixed value for ret code, make a vchnl re-negotiation
ixgbe:
------
- Add xdp frags support to ndo_xdp_xmit
- Fix kernel compilation issue related to u64_stats_fetch_ begin/retry
igb:
----
- Fix bitmaps declarations
- Fix array size overrun
- Fix adjfreq compatibility
- Fix e1000_read_nvm() error handling
- Fix always true if condition
- Fix netif_napi_add compilation for igb
- Fix warnings about ignored return value
Version [28.0]
===============================================================================
Driver Version
------ -------
ice 1.11.14
i40e 2.22.18
ixgbe 5.18.11
igb 5.13.16
irdma 1.11.58
ice:
----
- Add support for double VLAN in switchdev
- Add support for VLAN TPID filters in switchdev
- Add support for PPPoE hardware offload
- Add i2c write command
- Switch: dynamically add VLAN headers to dummy packets
- Add EXTTS feature to the feature bitmap
- Add mpls+tso support
- Expose RSS indirection tables for queue groups via ethtool
- Received packets with multicast MAC addresses are now redirected to
Flow Director and RSS queues
- Allocates transmit/receive queue pairs to be equal to the number of
local node CPU threads with the following constraints:
o The driver will allocate a minimum of 8 queue pairs, or the total
number of CPUs, whichever is lower.
o The driver will allocate a maximum of 64 queue pairs.
- LAG Active-Backup for RDMA RoCEv2
- Support for VXLAN/GRE in switchdev mode
- FEC auto detect in SR/LR aka No-FEC
- Support for GTP via switchdev and TC
- CRC aka FCS strip control via ethtool
- Accelerated Receive Flow Steering (aRFS) might not work correctly,
causing traffic to go to unexpected RX queues
- Fix DMA mappings leak
- Fix ice_xdp_xmit() when XDP TX queue number is not sufficient
- config netdev tc before setting queues number
- Fix VSI rebuild WARN_ON check for VF
- Fix call trace with null VSI during VF reset
- Fix tunnel checksum offload with fragmented traffic
- Fix crash by keep old cfg when update TCs more than queues
- Fix switchdev rules book keeping
- ethtool: Prohibit improper channel config for DCB
- TC filters do not function correctly when the GTP encryption is
enabled in the TC filter add command (enc_key_id 1)
- The sense of PFC is lost through a link bounce because the number of TCs is 1.
- Adding 16 MAC addresses to an untrusted VF that is attached to a VM,
then changing the VF's trust status to on and off again, could cause
the added MAC addresses to become non-functional.
- Change the number of reserved RX/TX queue pairs per PF
i40e:
-----
- Added support for the vf-source-pruning private flag
- Add VF VLAN pruning
- Add support for ethtool -s speed
- ptp: introduce helpers to adjust by scaled parts per million
- Add appropriate error message logged for incorrect duplex setting
- Fix dropped jumbo frames statistics
- Fix VF's MAC Address change on VM
- Fix to stop tx_timeout recovery if GLOBR fails
- Fix incorrect address type for IPv6 flow rules
- Fix ADQ rate limiting for PF
- Fix VF set max MTU size
- Fix set max_tx_rate when it is lower than 1 Mbps
- Fix DMA mappings leak
- Fix ethtool rx-flow-hash setting for X722
- Fix VF hang when reset is triggered on another VF
- Fix no connection when disable vlan strip before adding vlan
- Fix memory leak when driver_register() fails
- Fix delete if NULL check before dev_kfree_skb
- Fix convert to new udp_tunnel infrastructure
- Fix reading WoL settings from the adapter
- Fix disallow ip4 and ip6 l4_4_bytes
- Fix fd rule(src-ip) creation for flow type ip6
- Fix error message for hw limitation for TC VSI
irdma:
------
- Previous to this release, the ToS-to-user priority mappings were
hard-coded in Linux to use only 0, 2, 4, 6 VLAN PFC priority values
with RDMA traffic. Now in order to use other priorities with RDMA
traffic (that is, 1, 3, 5, 7), a VLAN is required to be setup using
the egress-qos-map option. For example, to map all priority 0 as
priority 3, one can use: ip link add link name
type vlan id egress-qos-map 0:3 1:0
- An rping can fail after a system reboot.
Version [27.6]
===============================================================================
Driver Version
------ -------
ice 1.9.11
i40e 2.20.12
ixgbe 5.16.5
igb 5.11.4
irdma 1.9.30
ice:
----
- ADQ support for 256 queues pairs
- ADQ acceleration using independent pollers
- ADQ auto configuration script
- ADQ per-tc inline Flow Director
- Support for Selectable Scheduler Topology
- Support for tc forward to queue
- Initial support for S-IOV
- Due to the previous bugs in PF-to-port mapping in both NVM and UEFI
Driver, old NVMs are not compatible with the new UEFI driver. As it
pertains to HII, NVMs can still be updated via FMP.
- When a VF interface is set as 'up' and assigned to a namespace, and the
namespace is then deleted, the dmesg log may show the error Failed to set
LAN Tx queue context, error: ICE_ERR_PARAM followed by error codes from
the ice and iavf drivers.
- Repeatedly assigning a VF interface to a network namespace then deleting
that namespace might result in an unexpected error message and might
possibly result in a call trace on the host system.
- When the queue settings of a port are modified using the ethtool -L ethx
combined XX command, the Interrupt Moderation settings reset to default.
- When the maximum allowed number of VLAN filters are created on a trusted
VF, and the VF is then set to untrusted and the VM is rebooted, the iavf
driver may not load correctly in the VM and may show errors in the VM
dmesg log.
- If trusted mode is enabled for a VF while promiscuous mode is disabled
and multicast promiscuous mode is enabled, unicast packets may be visible
on the VF and multicast packets may not be visible on the VF.
Alternatively, if promiscuous mode is enabled and multicast promiscuous
mode is disabled, then both unicast and multicast packets may not be
visible on the VF interface.
- A VF may incorrectly receive additional packets when trusted mode is
disabled but promiscuous mode is enabled.
- Receive hashing might not be enabled by default on Virtual Functions when
using an older iavf driver in combination with a newer PF driver version.
- Linux sysctl commands, or any automated scripting that alerts or sets
/proc/sys/ attributes using sysctl, might encounter a system crash that
includes irdma_net_event in the dmesg stack trace.
Workaround: With OOT irdma-1.8.X installed on the system, avoid running
sysctl while drivers are being loaded or unloaded.
- irdma stops working if the number of ice driver queues are changed
(ethtool -L) while the irdma driver is loaded. As a workaround, remove
(if previously loaded) and reload irdma after changing the number of
queues.
i40e:
-----
- Add support for ethtool -s speed
- VLAN tagged traffic sent from other VFs attached to NIC can be seen on
other VFs. This is due to issue with NIC settings and offload settings of
VFs. As a workaround, disable TX VLAN offload on VFs:
ethtool -K tx-vlan-offload off
This does not affect egress traffic from outside of the NIC
igb:
----
- Fix dropped PTP timestamps on i210
Version [27.3]
===============================================================================
Driver Version
------ -------
ice 1.8.8
i40e 2.19.3
ixgbe 5.14.6
igb 5.9.3
irdma 1.8.46
Lenovo Data Center Group Linux OS Support Home Page |
| linux.lenovo.com |
© 2018-2024 Lenovo. All rights reserved |
Release: 29.1 08/01/2024
============================
New OS Support:
- Support RHEL8.9/9.3
E810
-------------
New features:
- Get SFF Info NC-SI OEM Command
- Disable FW LLDP agent in and allow LLDP frames to be sent and received
in the VF.
- Supports Log of MNG packets for debugging purposes.
- Supports PLDM Type 4s for FRU data.
- The E810-CQDA2T adapter now supports 2x2x25 Gbps configuration.
- Supports reduced PTP jitter by improved timestamp accuracy and Port
timer compensation.
- E810-XXVDA4T and E810-CQDA2T now support DPLL updates.
- Supports shorter DPLL lock time
- Force FW LOG to uART even if PF Driver disables
- Option to disable PLDM FRU backup data response when NVM FRU
manufacturing data is not available
Bug fixes:
- IPS: 00788617 - [PA_SI Agile 23-1_USI_SDV_Pass 1_HW_SE450] Qtester
shows "VPD FW check failed (Error)
Description:
1. Power on system and boot into WS2022.
2. Setup Qtester to run 72hrs AC cycle test.
after AC power on, the BMC software calls PLDM_GetFirmwareParameters
and it sometimes (may be once after 300 AC power cycles) returns
Completion Code = 0x01(PLDM Base: Error).
Fix:
Firmware now returns PLDM_ERROR_NOT_READY Completion Code instead of
generic 0x01(PLDM Base: Error) when firmware is not fully ready after
platform reboot.
- When plugging DAC cables into two ports at the same time on Clifton
Channel, it is possible for one port to be unable to establish link.
- To have the metrics data shown on the respective port, map the port to
MAC address before accessing counter registers.
- Resolved an issue where 0x5089 was returned for command
get_sff_temperature when only port1 had optics of E810.
- PLDM T2 NumericSensors didn't send async event when any threshold was crossed.
- NCSI1.2 command Get Transceiver Temperature(0x4A) returned incorrect
temperatures due to wrong endianness.
- Module temp thresholds check running too early causes 0 return, which
is taken as valid and shuts down port.
- GetSensorReading command returns error code 0x1 in case when plug
temperature is higher than 78C (0x4E).
- E810 doesn't support non-contiguous TCs within the LLDP configuration.
It may result in a hang situation if receiving such LLDP packets during
power-on stage. A timeout was added to the firmware polling loop to
trigger a core reset which clears the hang and allows the device to
continue functioning.
- E810 4.3 NVMs use a common map, which introduces a change (new entries)
to an immediate TLV in PFA. FW fails to interpret the new entries in
this specific TLV, which results in adding those new entries again and
again upon every update from the former 4.3 image to 4.3 and above.
Eventually, due to the lack of space in that TLV, any normal attempt to
update will fail.
- NetworkDeviceFunction.Ethernet.PermanentMACaddress should stay with
factory setting if it has been programmed. Read relevant section from
Factory Settings.
- PLDM Type4 showing incorrect default Part Number
- NIC disappeared after NVM signature corruption
- Incorrect log entry format for timestamp value
- Get Link Status command does not return External Device Power Class
value
- Previously, the PLDM Type 5 Command "Activate Firmware" could
potentially cause a subsequent PLDM Type 2 event.
- Resolved an issue where during an ActiveFirmware request coming from
Update Agent (BMC) PLDM Type2, a PlatformEventMessage was generated by
the Firmware Device (Network Adapter). As a result, the response did
not happen, and Update Agent (BMC) treated it as Update Failed.
- Fixed an issue where the UEFI HII interface could not save the setting
of virtualization mode.
- Previously, in PLDM Platform Monitoring and Control, the
sensor_data_size field of SFP thermal sensor PDR was containing wrong
value: unsigned instead of signed char/byte.
- Previously, during ActiveFirmware request coming from Update Agent
(BMC) PLDM Type2 PlatformEventMessage was generated by our Firmware
Device (Network Adapter). As a result, response did not happen and
Update Agent (BMC) treated it as Update Failed.
- Firmware now acquires semaphore before "Load Factory Settings to PFA"
and releases it immediately after.
- Previously, when BMC subscribed to PLDM events, the NIC issued two
equivalent redfish events for the different origins (Network Adapter
and Port) caused by the same trigger (link state change for given
port). This issue caused unnecessary redfish event flood. The
resolution was to change firmware design in order to avoid sending
multiple redfish events for the same trigger.
- Previously, in Network Controller Sideband Interface after
EnableChannel command invoked, the OEM command IANA=0x02A2 GetLldp
occasionally returned LLDP TLVs with missing mandatory values for
chassis_id and port_id.
- Set packageID SDP to correct values for E810-CQDA1.
- Now the patched property of NetworkDeviceFunction.Ethernet.MACAddress
is persistent during reboot. Users can restore to PermanentMACAddress
with ResetSettingsToDefaults.
FVL/SGVL
--------
New features:
- Enable E700 X722 Tx signal to "always on".
- Enable E700 to store events details on the NVM in Recovery Mode.When
entering the Recovery Mode, X710 will store the details of the event in
the NVM.
Bug fixes:
- Resolved an issue where POST to the
NetworkAdapter.ResetSettingsToDefault looked successful from postman,
but after reboot the adapter's Port 1 LLDP setting did not reset to
default.
- Previously, there were cases of Fortville disappearing from the PCIe
bus during reboot stress testing reported on Icelake-based platforms.
These platforms performed a PCIe Hot Reset before asserting PERST# as
part of the warm reboot cycle. PERST# is asserted at approximately the
end of the Hot Reset, but due mostly to the Fortville LCB
implementation, there was a variation in the relative timing of about
65 us, so many edge cases could occur.
- Previously, after configuring "ifconfig ethX down" command on Marvell
devices series 88E1512/88E1514, the peer side (loopback configuration)
was still reporting link up.
- Fixed an issue. Unexpected EMP Reset occurred in a rarely BMC request
combination in a very low probability, which may cause: 'tx_timeout' or
'ARQ: Unknown event 0x0000 ignored' or 'bond down' if bonding is
configured.
- IPS: 00879846 - Intel XXV710-DA2 PCIe 25Gb 2-Port SFP28 Adapter link
speed and status incorrect
Description:
Link event is reported only once at when the machine is powered on,
after that no link event will be generated any more even after the
network cable is inserted and removed, this is incorrect, link event
should be reported when we plugin or remove the AOC cable.
Fix:
Firmware reported link speed value based on advertised capability value
using DNL declaration, changes to read link speed from MAC
- IPS: 00852370 - IEPD Hot Issue: SR645v3 X710 T2L PCIe bandwidth degraded
Description:
X710 PCIe may enter x1 on the AMD Genoa platform in reboot and AC
process.
Customer updated system firmware XCC: KAX318V >> KAX322V UEFI: KAE112O
>> KAE116K Following event occurs intermittently, but high fail rate
PCIe Link Width has degraded from x8 to x1 in physical slot number 3.
Slot 3 is Intel X710-T2L 10GBASE-T 2-Port PCIe Ethernet Adapter,
customer had tried new NIC FW, replace NIC but no help. this system no
re-timer.
Fix:
The firmware changes to enter the PCIE detection phase when PERST
becomes de-asserted
Release: 28.2 08/01/2023
========================
New OS Support:
- Support RHEL8.8/9.2, SLES 15.5
E810
-------------
New features:
- Support NC-SI 1.2 thermal monitoring commands.
Bug fixes:
- IPS: 00833367 - E810-DA2 10/25GbE SFP28 2-Port OCP Ethernet Adapter cannot
enable SRIOV in UEFI
Description:
SRIOV can not be enabled in UEFI
Fix:
Added logic to support on for all PFs
- Resolved an issue were an incorrectly interpreted BDF during MCTP
Endpoint discovery led to sending PLDM events to a non-existing device.
- Resolved an issue where the id field of Event BEJ was changed from the
originating resource Id as the resource number to a deferred binding
resource Id.
- A "Cannot initialize port" failure occurs when running nvmupdate
1.39.32.6 in SLES15SP3, because the tool doesn't support ioctl.
- A SyncE-related transaction between driver and device may cause a PHY
loading failure, during NVM upgrade. Subsequently, PHY errors are
reported to UEFI driver during reboot, resulting in a BIOS hang.
- Resolved an issue where E810-XXV-4T didn't expose one of the On Chip
thermal sensors via PLDM. The sensor was not visible via BMC.
- Resolved an issue where CVL4.1 and CVL4.2 had a problem with Recovery
Clock configuration on port 4, in the case of fully reversed PF<>MAC
mapping. Other configurations were not impacted.
- Resolved an issue where the NC-SI "Reset Channel" (0x05) command was
not reverting the port link state change made by for Real Time
Comprehensive Embedded Management "Network Port Control' (0x34)
command.
- Resolved an issue where, during communication via PLDM Type 6, FW did
not move the operation to the failed state and did not return
INVALID_DATA. Instead, it returned UNSUPPORTED in the case of an
invalid payload.
- Reverted an incorrectly added strict handling of the payload-less action.
- Resolved an issue where the Network Device and Port Metrics are
disabled due to an BMC issue and missing support on BMC side.
- Resolved an issue where in order to change the Forward Error Correction
setting in the UEFI HII the user needs to change the Media detection
setting, save the changes, and then change the FEC setting.
- Previously, promiscuous mode did not see all packets; it only saw those
packets arriving over the wire (that is, not sent from the same
physical function (PF) but a different virtual function (VF).
- Resolved an issue where during the second time of a POST operation,
NetworkAdapter.ResetSettingsToDefault failed with the BMC reporting
"Internal Server Error."
- Resolved an issue when both RMII-based transport (RBT) and Management
Component Transport Protocol (MCTP) were enabled, and PT was enabled on
RBT, then OEM command 0x26 (config LLDP) was being rejected from NC-SI
over MCTP with reason code 0x5081.
- Corrected an issue where updating an NVM image of E810 could result in
a "PHY NVM is not supported in devlink" warning message.
- Previously, E8102CQDA2 adapter showed Linkcap width as x16.
FVL/SGVL
--------
New features:
- Get Link Status AQC response when SFP module is not initialized.
Description:
If the Get Link Status AQC is sent when the SFP module is not
initialized yet, the link can be lost.
When the SFP module is not initialized yet, FW will try again in a
reasonable timeout.
Bug fixes:
- IPS: 00819239 - [Kahauale_BBFV_HCT]X710-T4L NIC adapter can't saved the
"Active Physical Functions" option on uEFI menu.
IPS: 00831264/00844425 - XXV710-DA2 PCIe Adapter NIC port didn't
disable after changed the "Active physical functions"
Description:
Intel XXV710-DA2 PCIe 25Gb 2-Port SFP28 Adapter NIC port didn't disable
after changed the "Active Physical" setting on uEFI menu.
Fix:
Reset the length of the PCIe Alt Autoload section during image
downgrade to a version earlier than FVL9.2
- IPS: 00840914 - [PA_IO Agile-Sprint202310_BDC_Storage_flex]SRIOV
enable/disable both can do SRIOV features
Description:
OS can still create SRIOV VFs when the adapter HII already disables SRIOV
in the BIOS
IPS: 00841528 - [PA_IO Agile-Sprint202310_BDC_Storage_SD530] Intel X722
LOM test SR-IOV function cannot enable VF under ESXi8.0u2 OS
Description:
Open to enter ESXi Host Client-Manage-Hardware-PCI Devices-Select Port
"Intel X722 10GB SFP+" Configure SR-IOV 2.UEFI has set SR-IOV function
3.ESXi8.0u2 Unable to enable VF redhat 9.2 can work well.
Fix:
Correctly set SR-IOV Control Capable bit for Lenovo images,
- IPS: 00839473 - [PA_IO Agile-Sprint202209_BDC_oceancat_Sprint202310]X722
lom WOL fail in SD650.
Description:
WOL function UEIF page show gray, OS can not enable WOL function
Fix:
Enable force link down requests and map channel number in a descending
order. Enable AltRAM implementation then reset corrupted memory block and
recalculate MSI_X VF vectors.
- IPS: 00831248 - XXV710 VPD info question
Description:
lspci -s BUS:Device.Function -vvv command to check VPD info, but it
can't show out the VPD info
Fix:
Enable VPD bit in NVM image
- IPS: 00769725 - [Hakuba] Abnormal cable plug (unplug) event on port 2
and 3 with Intel X710-T4L
Description:
Port2 and port3 of X710-T4L PCIE cannot display event in audit log and
link status/speed in inventory when plug/unplug the cable.
Fix:
There are missing MCTP link status changes as mixed PF no. with port
no. in the MCTP Link Status change handler, and overflowing PLDM queue
with RDE events that were classified as Big Event.
Map Port No. to the corresponding PF and continue, add additional
condition to disable unplanned eventReceiver physical address change.
Disable RDE events in the presence of asynchronous MCTP implementation
without polling for Big Events
Note: it may still occur in loopback mode.
- IPS: 00725539 - Link down on x722 4x10Gb SFP+ PHY card with 27.3 code
level (800039xx)
Description:
After upgrading the FW, all ports are down.
Fix:
Force OCP PnP Autodetection flow after NVM update.
- Resolved an issue where a rare PCI speed degradation with Gen3 to Gen2
was seen during extensive link tests after a reset is caused by a core
reset on an ICX-D platform.
- Resolved an issue where during a write, with http PATCH method
triggered, the LLDP Disable Valid bit was ignored, which resulted in a
skip of write in all scenarios instead of just for the same LLLDP
Enabled value.
- Resolved an issue where if BMC bonding was disabled, NCSI mode was set
to a Manual Switch Mode (depending on the platform), and BMC share port
has changed before doing a power cycle, recovery mode would occur.
Other prerequisites include disabled WOL, disabled LLDP, enabled PXE no
drop, and disabled auto power-on after power loss (in BIOS).
- Fixed an issue where the processing port id was always the same as the
mac address
- Previously, the addition of PCIe re-timers added to the total channel
latency causing replay timer timeout correctable errors.
- Resolved an issue where PLDM type MCTP packets were not accepted due to
missing configuration for PCIe VDM
I350/340/210
------------
New features:
- IPS: 00699338 - Name of I350 interlock in XCC is inconsistent with V1
Description:
Card Name field shall be taken from V1 VPD field instead of "large
resource string". This change applies to 1G Lenovo devices with enabled
AIM support and existing V1 VPD field.
Release: 28.0 02/16/2023
========================
E810
-------------
New features:
- Support RHEL8.7/9.1, ESXi 8.0
- OROM update support in Recovery mode
Description:
With this implementation, the NVM host update flow in recovery mode
should be the same as in the operational mode.
- Defines the IID (Instance IDs) field in the NC-SI packet's header,
which is used to match responses to requests and to differentiate
between new and retried commands.
Description:
According to the spec, If the IID of the current command is the same as
the previous command's IID, the command is a "Retried Command" and the
FW shall return the response of previous command (should be done
without re-executing the command). When the IID is different than from
the previous command, the command is executed.
The NC-SI spec is unclear on if the incoming commands IID is compared
to the IID of the previous command received on the same channel,
package or device. This feature defines a transfer to maintaining a
single buffer for the last NC-SI command received on the device per
transmitting interface.
- Add FC-FEC(CL74) support with optical transceiver 25G-SR
- NVMUpdate - only access devices from config file
Description:
The affected component shall only access devices (read/write flash or
registers) listed in config file that match the following criteria:
* if device in config file contains: DEVICE and VENDOR all devices
with specified device ID and vendor ID will be accessed;
* if device in config file contains: DEVICE, VENDOR, SUBDEVICE,
SUBVENDOR all devices with specified 4partID will be accessed,
* if device in config file contains: DEVICE, VENDOR, SUBDEVICE,
SUBVENDOR, REVISION all devices with specified 4partID and
specific revision will be accessed.
Device not listed in config file (do not match the criteria above)
shall be marked as "Update not available' in interactive mode.
- NVMUpdate - OROM update in recovery
Description:
If the device is in recovery mode and has the capability of supporting
for OROM update in recovery mode. it can be updated OROM either from
NVM image or from separate file (FLB).
- RoCEv2 LAG
- FEC auto-detect for CVL LR/SR mode
Description:
Only valid FEC modes will be attempted based on the media and PHY type.
- Enable hardware arbitration
- IPS: 00675913 - FEC Configuration through UEFI HII for 25G CVL adapters
Bug fixes:
- IPS: 00796141 - E810 fails to link with 10/25Gb SFP28 connect to
adapter and 10Gb SFP+ connected
Description:
Connect Lenovo 10/25Gb SFP28 to E810, Lenovo 10Gb SFP+ to switch (CISCO 9K)
Link fails to come up.
Fix:
Change the preserved "LinkDefaultOverrides" fields in the NVM image so
that the settings in the NVM images make the firmware to try all the
FEC modes during the link training.
- IPS: 00736066 - [PA_IO Agile-Sprint202209_BDC_Storage_SR655] E810-DA2
OCP ShareNIC cannot obtain DHCP IP under SR655
Description:
BMC failed getting an IP address on the sharenic port of E810-xxvDA2
OCP via DHCP
Fix:
Use proper packageID SDP settings for SDP0 and SDP1
- IPS: 00735574 - [PA_QUR 22C_BDC_SDV_FW_AMD1P_UEFI][RMK-5883]E810-DA4
OCP Ethernet Adapter cannot obtain DHCP IP
Destination:
E810-DA4 OCP Ethernet Adapter cannot obtain DHCP IP.
Fix:
In implementation, use different variables to identify VSI in different
context.
- Both ports are loosing link randomly after dcb-map change
Destination:
The E810 adapter might experience an adapter-wide reset (all ports)
when the DCBx willing configuration change (in firmware managed mode)
propagated from the switch removes a TC that was RDMA enabled. This
typically occurs when removing a TC associated with UP0 because it's
the default UP that RDMA based it's configuration on. The reset results
in a temporary loss in connectivity as the adapter re-initializes.
Fix:
FW changes along with changes in SW to allow for RDMA to disable
QueueSets before suspending the removed TCs from the scheduler tree.
FVL/SGVL
--------
New features:
- Defines the IID (Instance IDs) field in the NC-SI packet's header,
which is used to match responses to requests and to differentiate
between new and retried commands.
Description:
According to the spec, If the IID of the current command is the same as
the previous command's IID, the command is a "Retried Command" and the
FW shall return the response of previous command (should be done
without re-executing the command). When the IID is different than from
the previous command, the command is executed.
The NC-SI spec is unclear on if the incoming commands IID is compared
to the IID of the previous command received on the same channel,
package or device. This feature defines a transfer to maintaining a
single buffer for the last NC-SI command received on the device per
transmitting interface.
Bug fixes:
- XXV710DA2 adapter intermittently lose link when doing DC power cycle test
Description:
XXV710DA2 adapter intermittently lose link when doing DC power cycle
test, this can be reproduced in about 200 DC cycle.
Fix:
When the FW starts to run, the gearbox has to be initialized and
configuration needs to be downloaded from the NVM and applied. If one
of the operation fails the FW will try again (several times). In this
case the configuration failed with the error which was not covered.
Error handling extended in the code to cover all cases
- Package Port Control command with Port Control equals 0x1 works like
reset if link is already up
Description:
In Campbell Pond, send Package Port Control Command to disable the
port, send command to enable it, then send command to enable it again.
The port is down instead of up.
Fix:
Added link status check in package command.
- Link Status Change AEN event is not send after link down
Description:
Sending AEN Enable command with Link Status Change equals true, then
sending Port Control Command with Port Control field equals 0x0 -
disable. There is no event after link down.
Fix:
Changed event rejection condition to response.
- After running reboot 123 times with qtest, slot4 (X710-T2L) shows an error
Description:
Used qtest ro run reboot 123 times, xcc web appears: PCIe Correctable
Error Threshold limit has been exceeded at Segment 0x0000 Bus 0xA8
Device 0x00 Function 0x01.
Fix:
Adjust the replay timer timeout value for different scenarios.
Lenovo Data Center Group Linux OS Support Home Page |
| linux.lenovo.com |
© 2018-2024 Lenovo. All rights reserved |