Calculate pool size in t1 as 24 * downlink port + 8 * uplink port
- Take both port and peer MTU into account when calculating headroom
- Worst case factor is decreased to 50%
- Mellanox-SN2700-C28D8 t0, assume 48 * 50G/5m + 8 * 100G/40m ports
- Mellanox-SN2700 (C32)
- t0: 16 * 100G/5m + 16 * 100G/40m
- t1: 16 * 100G/40m + 16 * 100G/300m
Signed-off-by: Stephen Sun <stephens@mellanox.com>
Co-authored-by: Stephen Sun <stephens@mellanox.com>
As part of consolidating all common Python-based functionality into the new sonic-py-common package, this pull request:
1. Redirects all Python applications/scripts in sonic-buildimage repo which previously imported sonic_device_util or sonic_daemon_base to instead import sonic-py-common, which was added in https://github.com/Azure/sonic-buildimage/pull/5003
2. Replaces all calls to `sonic_device_util.get_platform_info()` to instead call `sonic_py_common.get_platform()` and removes any calls to `sonic_device_util.get_machine_info()` which are no longer necessary (i.e., those which were only used to pass the results to `sonic_device_util.get_platform_info()`.
3. Removes unused imports to the now-deprecated sonic-daemon-base package and sonic_device_util.py module
This is the next step toward resolving https://github.com/Azure/sonic-buildimage/issues/4999
Also reverted my previous change in which device_info.get_platform() would first try obtaining the platform ID string from Config DB and fall back to gathering it from machine.conf upon failure because this function is called by sonic-cfggen before the data is in the DB, in which case, the db_connect() call will hang indefinitely, which was not the behavior I expected. As of now, the function will always reference machine.conf.
* Update the buffer size based on the latest excel
Signed-off-by: Stephen Sun <stephens@mellanox.com>
* Align the buffer configuration with the latest formula:
- reduce redundant "*2" in formula
- use port MTU for local sending the PFC frame and peer lossless MTU for peer sending lossless traffic
Buffer pool size updated accordingly.
Signed-off-by: Stephen Sun <stephens@mellanox.com>
Align SFP key names with new standard defined in https://github.com/Azure/sonic-platform-common/pull/97
- hardwarerev -> hardware_rev
- serialnum -> serial
- manufacturename -> manufacturer
- modelname -> model
- Connector -> connector
The new feature system health support a platform based configuration file. Add configuration files for all Mellanox platform.
Add a configuration file for SN2700, other platform will use a soft link to it.
* Change port index in port_config.ini to 1-based
* Add default port index to port_config.ini, change platform plugins to accept 1-based port index
* fix port index in sfp_event.py
* New SKU support for MSN3420
Signed-off-by: Shlomi Bitton <shlomibi@mellanox.com>
Conflicts:
device/mellanox/x86_64-mlnx_msn2700-r0/plugins/sfputil.py
* Add CPLD's
* Symlink fixes and semantics
* Adding new platform at end of lines
update FW to xx_2000_3298
update SAI to 1.16.0
update Spectrum-1 and Spectrum-2 buffer pool size according to the new SDK default config change.
modified: ../../device/mellanox/x86_64-mlnx_msn2700-r0/ACS-MSN2700/buffers_defaults_t0.j2
modified: ../../device/mellanox/x86_64-mlnx_msn2700-r0/ACS-MSN2700/buffers_defaults_t1.j2
modified: ../../device/mellanox/x86_64-mlnx_msn3700-r0/ACS-MSN3700/buffers_defaults_t0.j2
modified: ../../device/mellanox/x86_64-mlnx_msn3700-r0/ACS-MSN3700/buffers_defaults_t1.j2
modified: fw.mk
modified: mlnx-sai.mk
modified: mlnx-sai/SAI-Implementation
modified: sdk-src/sx-kernel/Switch-SDK-drivers
modified: sdk.mk
signed-off by kebol@mellanox.com
* [plugins]add fan functions, add voltage, current, power for psu
* [plugins]link fanutil.py and psuutil.py to those in 2700
* [plugin]add thermal
* [plugin]add symbol links for thermalutil for all SKUs
* [MultiDB] (except ./src and ./dockers dirs): replace redis-cli with sonic-db-cli and use new DBConnector
* update comment for a potential bug
* update comment
* add TODO maker as review reqirement
xcvrd is broken because a newly introduced interface get_transceiver_dom_threshold_info_dict in common code calls a unsupported interface _read_eeprom_specific_bytes.
Fix the issue by implement get_transceiver_dom_threshold_info_dict to avoid calling the unsupported interface.
* fix the issue Bug SW #1816356 which is due to failing to test whether dom supported prior to reading dom data
* use pre-defined variable to avoid magic number.
no need to read 16 bytes, 1 byte is enough since calibration and dom capability are all in bytes at offset 92
Variables SFP_VLOT_OFFSET and QSFP_VLOT_OFFSET containing the typo are originally defined in repo sonic-platform-common. The typo has been fixed in PR #33. However, some Mellanox-specific code hasn't updated correspondingly, which results in xcvrd fail to start.
This PR updates the variable name in Mellanox-specific code correspondingly to fix that.
* [sfputil]Remove the dependency on sysfs for sfputil, mainly get_presence and port_to_eeprom_mapping
Remove the dependency on sysfs, including:
1. rewrite get_presence by using ethtool;
2. remove interface port_to_eeprom_mapping which is no longer referenced;
3. remove code that references port_to_eeprom_mapping and _port_to_eeprom_mapping;
4. remove private member qsfp_sysfs_path which is no longer referenced.
* [sfputil.py]
minor adjustment: move the presence=False to the beginning of get_presence.
* [device/mellanox/x86_64-mlnx_msn2700-r0/plugins/sfputil.py]
purpose and restrictions
1. reading eeprom via ethtool.
2. avoid changing common codes shared by all the manufacture (sonic-platform-common), contrain all the modifications with Mellanox-specific code.
current implementation
A new class based on SfpUtilBase and a new method _read_eeprom_specific_bytes_via_ethtool have been introduced in order to change the way the eprom DOM data is read. Typically the best practice to do this kind of thing is to contrain the modification within the function which execute reading operations only and keep other stuffs (especially the interface) untouched. However, this can hardly be achieved since the original reading function takes the file object as input parameter to represent the port. It is done by having the file object to point to /var/run/hwmanagement files, which will not be maintained in the future. As a result, a new interface has to be introduced with a port number/name as input parameter in order to get rid of the dependency on the those files:
_read_eeprom_specific_bytes_via_ethtool
Since the interface changed, all methods that call the interface should also be overwritten in order to call the new interface, including:
_read_eeprom_devid
get_transceiver_info_dict
get_transceiver_dom_info_dict
Only interface used to read eeprom DOM has been replaced and the main logic has not been changed except the following mentioned.
1. reading DOM data for sfp port, which is implementioned in get_transceiver_dom_info_dict. In this case a "calibration" should be firstly read from eeprom before other values like temperature, voltage, rx/tx power, can be parsed. However, this has been ignored in the original code, resulting in that the data cann't be parsed.
2. In the original implemention the data area containing the data are read from DOM separatedly in order to avoid read uncessary data and achieve a better performance. Having used ethtool to read DOM data, the performance gap between reading all the area and reading the spot data separatedly has been narrowed to almost zero. To make the code neat and readable, we change the way to read this data.
* [sfputil] Returns dict with all data set to N/A for ports without dom support
Currently, the way in which dom data is read has been changed from using sysfs to using ethtool.
The ethtool returns None for ports without dom support, resulting in None being returned. However, this fails xcvrd to add the TRANSCEIVER_DOM_SENSOR table entry of associated port to CONFIG_DB and then causes SNMP fail.
To address this issue a default dict is initialized with all data set to 'N/A' and is returned is the above case.
BTW, in the original implementation which sysfs is used to read dom data, even though non-None data is returned for ports without dom support, it does not contain valid data. This can result in wrong data in TRANSCEIVER_DOM_SENSOR table.
* [sfputil]
removing unnecessary empty lines
removing redundent code
replacing hardcoding strings/numbers with predefined const variables
* Use default timeout value which will block the select function
* Submodule update for argument type issue in Select class
Submodule update sonic-swss-common:
e8caaea - Align the argument type with epoll_wait() (#255)
3ea133d - [selectable]: Update throw message (#253)
Signed-off-by: Kevin Wang <kevinw@mellanox.com>
* 1) DSCP 46 to 5; 2) ecn config for lossless traffic; 3) ecn on by default; 4) DWRR equal weight;
Signed-off-by: Wenda <wenni@microsoft.com>
* 1) link pg & queue 5 to lossy buffer profile; 2) ingress lossless alpha 1/8
Signed-off-by: Wenda <wenni@microsoft.com>
* Update the test case for qos & buffer json template
Signed-off-by: Wenda <wenni@microsoft.com>
* Migrate a7050-qx32 and s6000 to use pg_profile lookup architecture
Signed-off-by: Wenda <wenni@microsoft.com>
* Update pg headroom egress service pool for a7050-qx-32s, a7050-qx32, and s6000
Signed-off-by: Wenda <wenni@microsoft.com>
* Link queue 5 to lossy profile
Signed-off-by: Wenda <wenni@microsoft.com>
