- Why I did it
To include the fix for the issue of Modification of shared headroom on the fly can get to negative occupancy that leads to PFC been sent from the switch continuously.
- How I did it
Updated submodule pointer and version in relevant Makefile.
- How to verify it
Build an image and run tests from sonic-mgmt.
Signed-off-by: Volodymyr Samotiy <volodymyrs@nvidia.com>
- Why I did it
Optimize thermal control policies to simplify the logic and add more protection code in policies to make sure it works even if kernel algorithm does not work.
- How I did it
Reduce unused thermal policies
Add timely ASIC temperature check in thermal policy to make sure ASIC temperature and fan speed is coordinated
Minimum allowed fan speed now is calculated by max of the expected fan speed among all policies
Move some logic from fan.py to thermal.py to make it more readable
- How to verify it
1. Manual test
2. Regression
Backport #9258 to 201911
Why I did it
When PSU is powered off, the PSU is still on the switch and the air flow is still the same. In this case, it is not necessary to set FAN speed to 100%.
How I did it
When PSU is powered of, don't treat it as absent.
How to verify it
Adjust existing unit test case
Add new case in sonic-mgmt
Conflicts:
platform/mellanox/mlnx-platform-api/sonic_platform/thermal_infos.py
- Why I did it
To include latest fixes.
1. On CMIS modules, after low power configuration, the firmware waited for the module state to be ModuleReady instead of ModuleLowPower causing delays.
2. When connecting Spectrum devices with optical transceivers that support RXLOS, remote side port down might cause the switch firmware to get stuck and cause unexpected switch behavior.
3. On rare occasions, when working with port rates of 1GbE or 10GbE and congestion occurs, packets may get stuck in the chip and may cause switch to hang.
4. When ECMP has high amount of next-hops based on VLAN interfaces, in some rare cases, packets will get a wrong VLAN tag and will be dropped.
5. Using SN4600C with copper or optics loopback cables in NRZ speeds, link may raise in long link up times ( up to 70 seconds).
6. When connecting SN4600C to SN4600C after Fastboot in 50GbE No_FEC mode with a copper cable, the link up time may take ~20 seconds.
- How I did it
Updated SDK submodule and relevant makefiles with the required versions.
- How to verify it
Build an image and run tests from "soni-mgmt".
Signed-off-by: Volodymyr Samotiy <volodymyrs@nvidia.com>
#### Why I did it
Upgrade Mellanox-SAI to 1.19.3 to support reclaiming reserved buffer on admin down ports
#### How I did it
To support reclaiming reserved buffer on admin down ports.
#### How to verify it
Regression test and manual test.
- Why I did it
Update SDK\FW version to 4.4.3326\2008.3326. This version contains:
New Features:
1. Add support for Fast Boot for SN3800
Bug Fixing:
1. In some cases, when the total number of allocations exceeds the resource limit, an error can occur due to incorrect resource release procedure. This issue is most likely to affect the following resources: flow counters, ACL actions, PBS, WJH filter, Tunnels, ECMP containers, MC (L2 &L3)
2. On Spectrum systems, when using Async Router API with IPV6, an error message in the log regarding failing to remove ECMP container may show up. This error is not functional and can be safely ignored.
3. On Spectrum-2 systems and above, when using warm boot, setting max_bridge_num to a value greater than 1968 will cause an error and potential crash.
4. Some Molex cables do not support speed after reboot
- How I did it
- How to verify it
Was verified by running regression tests that includes complete sonic-mgmt tests supported
Signed-off-by: Dror Prital <drorp@nvidia.com>
* [Mellanox] Update FW version to 2008.3218 (#8079)
Update FW version to 2008.3218, fixing the following issues:
- 50G/100G links that are operationally down before warm-reboot are not coming up after warm-reboot
- 50G/100G links with admin shut / no shut commands are not coming up after warm-reboot
- Why I did it
Update FW version to 2008_3110 fixing SN3800 specific warm boot scenario:
1. Disable interface
2. Warm Boot
3. Enable Interface --> link will remain down.
- How I did it
Use new FW that contains the fix for the problem mentioned above
- How to verify it
Run the scenario mentioned above and make sure that the link is up after warm boot
Signed-off-by: Dror Prital <drorp@nvidia.com>
- Why I did it
Updated FW to xx.2008.2526 version.
Fixed issues:
1. Spectrum-2, Spectrum-3 | sFlow | High CPU load and high on fully loaded switch.
2. Spectrum-2, Spectrum-3 | Fine grain LAG | in rare cases doesn’t update the right entry
- How I did it
Updated submodule pointer and version in a Makefile.
- How to verify it
Full regression and bugs validation
Signed-off-by: Shlomi Bitton <shlomibi@nvidia.com>
New features and fixes in the new SDK/FW:
SN4600C | AN/LT support
SN2700 | AN/LT bugs fixes
WJH | FID_MISS support
Signed-off-by: Kebo Liu <kebol@nvidia.com>
Fix the following issues:
Spectrum-2, Spectrum-3 | Port | Fix link issue when using 25 GbE rate between two ports while one is on Spectrum-2-based system and the other is on Spectrum-3-based system
All | warmboot | fail to upgrade from earlier SONiC versions with official SDK/FW 4.4.2306 (was on SONiC 201911)
All | What-Just-Happened | When enabling or disabling WJH under high traffic load to the host CPU, in very specific and low probability conditions, an error could occur, that may result in loss of data, channel failure or in extreme cases SW failure
Signed-off-by: Volodymyr Samotiy <volodymyrs@nvidia.com>
Bug fixes
-Removing critical thermal zones to prevent unexpected software system shutdown:
Kernel 4.9 -0071-mlxsw-core-Remove-critical-trip-point-from-thermal-z.patch
Kernel 4.19 -076-mlxsw-core-Remove-critical-trip-point-from-thermal-z.patch
- hw-mgmt: thermal: Add hardcoded critical trip point
- Removing redundant link for cpld3 for fixed systems (SN2100, SN2010).
- Fix an issue with a missed attribute for cpld3 (port CPLD) for SN2700, SN2410.
Signed-off-by: Stephen Sun <stephens@nvidia.com>
- Why I did it
To pick up new features and fix from SDK/FW and SAI
SDK/FW new Feature:
All | Added support for multiple modules and cable types. For full list contact Nvidia networking support
Spectrum-3 | SN46000C | Added support for up to 5W on ports 49 to 64 .
SDK/FW bugs' fix:
All | fast reboot | fast boot failure from latest 201811 to 201911 and above
Spectrum | 10GbE/1GbE Transceiver (FTLX8574D3BCV) stopped working after firmware upgrade
Spectrum-2 | When device is rebooted with locked Optical Transceivers in split mode, the firmware may get stuck
Spectrum-2 | SN3700 | When connecting at 200GbE to Ixia K400, Ixia receives CRC errors
Spectrum-2 | SN3800 | On rare occasions packets loss may be experienced due to signal integrity issues
Spectrum-2 | When the port is a member of a LAG, after a warmboot and port toggle on the peer-side, the port remains down
Spectrum-3 | SN4700 | While using Optic cable in Split 4x1 mode in PAM4, when two first ports are toggled, the other 2 ports go down
Spectrum-3 | SN4700 | When working in 400GbE, deleting the headroom configuration (changing buffer size to zero) on the fly may cause continual packet drops
SAI
All | Counters | Update tunnel decap counter to capture VNI miss
- How I did it
Update the related version number in the make files and update the submodule pointer accordingly.
- How to verify it
Run regression test and everything works good.
In rare case can see that xcvrd failed due to "UnboundLocalError: local variable 'label_port' referenced before assignment"
Init "label_port" as None at the beginning of the function, to avoid the case that "label_port" not assigned.
Bring in a fix for thermal shutdown observed while executing warm-reboot:
- All | prevent FW access during ISSU
Signed-off-by: Stepan Blyschak <stepanb@nvidia.com>
To have the following fixes:
* All | Port status remains down after warm boot and flapping the port on peer side
* All | LAG HASH | IPv6 SRC_IP is not accounted in LAG hashing [
* All | ASIC driver | Kernel crash observed when driver reload is initiated before it fully loaded
* Spectrum-3 | Buffer | In lossless configuration, headroom is been evicted only when the shared buffers is free
Signed-off-by: Volodymyr Samotiy <volodymyrs@nvidia.com>
Include SAI bug fixes:
Apply device MAC on port host interface when port is removed from LAG.
[Shared Headroom]: fixed watermark handling for SHP flow
Decrease verbosity of policer unbind message when no policer is attached
Signed-off-by: Stepan Blyschak <stepanb@nvidia.com>
** Why I did it **
Disable SDK extended dump due to issue found
** How I did it **
Update SAI submodule
** How to verify it **
Verify the SDK extended dump is not called.
Signed-off-by: Eran Dahan <erand@nvidia.com>
when parallel build is enabled, both docker-fpm-frr and docker-syncd-brcm
is built at the same time, docker-fpm-frr requires swss which requires to
install libsaivs-dev. docker-syncd-brcm requires syncd package which requires
to install libsaibcm-dev.
since libsaivs-dev and libsaibcm-dev install the sai header in the same
location, these two packages cannot be installed at the same time. Therefore,
we need to serialize the build between these two packages. Simply uninstall
the conflict package is not enough to solve this issue. The correct solution
is to have one package wait for another package to be uninstalled.
For example, if syncd is built first, then it will install libsaibcm-dev.
Meanwhile, if the swss build job starts and tries to install libsaivs-dev,
it will first try to query if libsaibcm-dev is installed or not. if it is
installed, then it will wait until libsaibcm-dev is uninstalled. After syncd
job is finished, it will uninstall libsaibcm-dev and swss build job will be
unblocked.
To solve this issue, _UNINSTALLS is introduced to uninstall a package that
is no longer needed and to allow blocked job to continue.
Signed-off-by: Guohan Lu <lguohan@gmail.com>
backport c4b5b002c3
make swss build depends only on libsairedis instead of syncd. This allows to build swss without depending
on vendor sai library.
Currently, libsairedis build also buils syncd which requires vendor SAI lib. This makes difficult to build
swss docker in buster while still keeping syncd docker in stretch, as swss requires libsairedis which also
build syncd and requires vendor to provide SAI for buster. As swss docker does not really contain syncd
binary, so it is not necessary to build syncd for swss docker.
[submodule]: update sonic-sairedis
1e42517996bfe41ac58d4c25ee3f93502befcb9d (HEAD -> 201911) [build]: add option to build without syncd
Signed-off-by: Guohan Lu <lguohan@gmail.com>
In order to prevent "mlxsw_minimal" driver accessing ASIC during in
service firmware upgrade flow, SDK will raise "OFFLINE" 'udev' event
at early beginning of such flow. When this event is received,
hw-managemnet will remove "mlxsw_minimal" driver.
There is no need to implement opposite "ONLINE" event, since this flow
is ended up with "kexec".
Signed-off-by: Kebo Liu <kebol@nvidia.com>
Bugs fixes:
All | Kernel | During system reload when CPU is loaded with heavy traffic, a Kernel Panic may occur.
All | Modules, Port split | FW stuck when device rebooted with locked Optical Transceivers in split mode
Spectrum-3 | PFC | On Spectrum-3 systems, slow reaction time to Rx pause packets on 40GbE ports may lead to buffer overflow on servers.
Spectrum-3 | SN4700, Port Split | On rare occasion SN4700, conducting 100G split (4x25G) in NRZ when splitter port 1 or 2 are down, ports 3 and 4 will also go down.
Enahncments:
All | Kernel | new notification on ISSU start, so other kernel drivers can disable any interface to ASIC
Signed-off-by: Kebo Liu <kebol@nvidia.com>
- Why I did it
Fix setting PSU led to 'green' or 'red' states.
Fix return False if unsupported color request.
Remove 'off' option for PSU led API since it is not supported in Mellanox.
- How I did it
Fix import missing information.
Return 'False' when unsupported led color is requested, preventing an exception.
- How to verify it
Try to set PSU LED to different status with Mellanox platform device.
Try to set PSU LED color to unsupported color with Mellanox platform device.
EEPROM cache file is not refreshed after install a new ONIE version even if the eeprom data is updated. The current Eeprom class always try to read from the cache file when the file exists. The PR is aimed to fix it.
As new hw-mgmt expose the sysfs for PSU fan max speed, we need support max/min speed for PSU fan in mellanox platform API.
Conflicts:
platform/mellanox/mlnx-platform-api/sonic_platform/fan.py
