portsorch: initial support for link-training and remote advertisement #2149

ds952811 · 2022-02-18T02:21:33Z

Signed-off-by: Dante Su [email protected]

What I did

Add Link-Training support to portsorch, while Gearbox is not in the scope
Add support for remote speed advertisement
Add support for capability checker

Why I did it

In the case of DAC, static pre-calibrated pre-emphasis is rarely available on SONIC, as most of the ODM are expecting this to be done dynamically at runtime via link-training, hence we'll need this feature to improve the link quality
Help users identify the cause of autoneg link failure
Avoid un-necessary SAI calls for autoneg and link-training controls if this feature is not supported

How I verified it

Manual test
Ran the Unit-tests to the corresponding changes

Details if related

HLD: sonic-net/SONiC#924, sonic-net/SONiC#925

dgsudharsan · 2022-04-01T03:51:23Z

orchagent/portsorch.cpp

@@ -296,9 +317,11 @@ static bool isValidPortTypeForLagMember(const Port& port)
 PortsOrch::PortsOrch(DBConnector *db, DBConnector *stateDb, vector<table_name_with_pri_t> &tableNames, DBConnector *chassisAppDb) :
        Orch(db, tableNames),
        m_portStateTable(stateDb, STATE_PORT_TABLE_NAME),
+        m_stateLinkTrainingTable(stateDb, STATE_LINK_TRAINING_TABLE_NAME),


Do we need a separate link training table? Why can't we add two fields to existing port table in State DB?

It's now merged into 'link_training_status' of PORT_TABLE in STATE_DB

dgsudharsan · 2022-04-01T03:59:38Z

orchagent/portsorch.cpp

+
+    auto executor = new ExecutableTimer(m_timer, this, "PORT_STATUS_POLL");
+    Orch::addExecutor(executor);
+    m_timer->start();


This timer shouldn't be enabled if the platform doesn't support link training or there is not link training config present.

This will be migrated to use the implementation in #2215, which is for AutoNeg, and it will only activate the poll when the port is admin up, link_training feature is available and is enabled.

dgsudharsan · 2022-04-01T04:02:36Z

orchagent/portsorch.cpp

+    sai_attribute_t attr;
+
+    attr.id = SAI_PORT_ATTR_SUPPORTED_LINK_TRAINING_MODE;
+    status = sai_port_api->set_port_attribute(port.m_port_id, &attr);


I don't get this logic. This is not the way to query and understand capabilities. We have separate API sai_query_attribute_capability. Please use that API.

As discussed in #2215, sai_query_attribute_capability does not support per-port basis capability query, while the AN/LT may only be available on certain ports on the switch ASIC

I believe the API used should have been get_port_attribute instead of set.

Thanks, will fix it

Still not fixed

Done, sorry about that, I thought I did this, but it's on the AutoNeg PR.

dgsudharsan · 2022-04-01T04:04:34Z

orchagent/portsorch.cpp

+{
+    string supp;
+
+    if (m_stateLinkTrainingTable.hget(port.m_alias, "supported", supp))


Please use internal Port data structure to cache the capability rather than writing to DB and reading from it within the same process

This is now cached as member field of the internal Port data structure

dgsudharsan · 2022-04-01T04:13:12Z

orchagent/portsorch.cpp

@@ -112,6 +113,26 @@ static map<string, int> autoneg_mode_map =
    { "off", 0 }
 };

+static map<string, int> link_training_mode_map =
+{
+    { "on", 1 },


Can't we use a bool instead of int? This will make code much simpler

In order to distinguish 'NOT CONFIURED' and 'OFF', we need -1 for this, and this is the same approach for the AutoNeg in the current implementation of SONiC

dgsudharsan · 2022-04-01T04:14:13Z

orchagent/portsorch.cpp

+                            }
+                            continue;
+                        }
+                        m_stateLinkTrainingTable.hset(p.m_alias, "status", lt > 0 ? "on" : "off");


Can't we use lt_str as such in status?

Thanks, will do.

dgsudharsan · 2022-04-01T04:18:21Z

tests/test_port_lt.py

+        fvs = swsscommon.FieldValuePairs([("link_training","on")])
+        tbl.set("Ethernet4", fvs)
+
+        # validate if autoneg false is pushed to asic db when set first time


Typo in comments, should be link-training instead of autoneg

Thanks, will fix it

dgsudharsan · 2022-04-01T04:18:51Z

tests/test_port_lt.py

+        expected_fields = {"SAI_PORT_ATTR_LINK_TRAINING_ENABLE":"false"}
+        adb.wait_for_field_match("ASIC_STATE:SAI_OBJECT_TYPE_PORT", port_oid, expected_fields)
+
+        # validate if autoneg true is pushed to asic db when set first time


Typo in comments, should be link-training instead of autoneg

Thanks, will fix it

dgsudharsan · 2022-04-20T00:53:41Z

orchagent/portsorch.cpp

+    sai_attribute_t attr;
+
+    attr.id = SAI_PORT_ATTR_SUPPORTED_LINK_TRAINING_MODE;
+    status = sai_port_api->set_port_attribute(port.m_port_id, &attr);


I believe the API used should have been get_port_attribute instead of set.

dgsudharsan · 2022-04-20T00:53:51Z

orchagent/portsorch.cpp

+    sai_status_t status;
+    sai_attribute_t attr;
+
+#ifdef SAI_PORT_ATTR_SUPPORTED_LINK_TRAINING_MODE


Why is this ifdef needed?

This is for the build failure in the PR and also the backward compatibility of the legacy SAI implementation, where SAI_PORT_ATTR_SUPPORTED_LINK_TRAINING_MODE is not implemented, while the other control attributes are supported

We cannot have #ifdef in the code.

dgsudharsan · 2022-04-20T01:04:18Z

orchagent/portsorch.cpp

@@ -61,6 +61,7 @@ extern string gMyAsicName;
 #define DEFAULT_VLAN_ID     1
 #define MAX_VALID_VLAN_ID   4094

+#define PORT_STAT_POLLING_SEC                             3


I believe 1 second polling is aggressive for this scenario. We should change it to 5 or 10 seconds. @prsunny What is your opinion?

Please also rename this to PORT_STATE_POLLING_SEC or similar? STAT means port statistics (like rx packets/bytes ) which is a different thing altogether

Sure, it's done

dgsudharsan · 2022-04-21T01:05:58Z

orchagent/portsorch.cpp

@@ -2351,6 +2426,9 @@ bool PortsOrch::initPort(const string &alias, const string &role, const int inde
                /* Create associated Gearbox lane mapping */
                initGearboxPort(p);

+                /* Initialize port capabilities */
+                initPortCapLinkTraining(p);


This function loops through all the ports and get capability. This would increase the port initialization time which is a bottleneck in fastboot scenarios. If the user has no LT configs this operation is not needed.
Instead init the capability only when the user configures LT and capability not in cache.

Noted, and it's done

Junchao-Mellanox · 2022-04-27T02:11:28Z

orchagent/portsorch.cpp

@@ -298,7 +319,8 @@ PortsOrch::PortsOrch(DBConnector *db, DBConnector *stateDb, vector<table_name_wi
        m_portStateTable(stateDb, STATE_PORT_TABLE_NAME),
        port_stat_manager(PORT_STAT_COUNTER_FLEX_COUNTER_GROUP, StatsMode::READ, PORT_STAT_FLEX_COUNTER_POLLING_INTERVAL_MS, false),
        port_buffer_drop_stat_manager(PORT_BUFFER_DROP_STAT_FLEX_COUNTER_GROUP, StatsMode::READ, PORT_BUFFER_DROP_STAT_POLLING_INTERVAL_MS, false),
-        queue_stat_manager(QUEUE_STAT_COUNTER_FLEX_COUNTER_GROUP, StatsMode::READ, QUEUE_STAT_FLEX_COUNTER_POLLING_INTERVAL_MS, false)
+        queue_stat_manager(QUEUE_STAT_COUNTER_FLEX_COUNTER_GROUP, StatsMode::READ, QUEUE_STAT_FLEX_COUNTER_POLLING_INTERVAL_MS, false),
+        m_timer(new SelectableTimer(timespec { .tv_sec = PORT_STATE_POLLING_SEC, .tv_nsec = 0 }))


Please rename m_timer to something more readable.

Junchao-Mellanox · 2022-04-27T02:13:43Z

orchagent/portsorch.cpp

+    sai_attribute_t attr;
+
+    attr.id = SAI_PORT_ATTR_SUPPORTED_LINK_TRAINING_MODE;
+    status = sai_port_api->set_port_attribute(port.m_port_id, &attr);


Still not fixed

Junchao-Mellanox · 2022-04-27T02:19:02Z

orchagent/portsorch.cpp

+    }
+    else
+    {
+        port.m_cap_lt = 1; /* Default to 1 for vstest */


I am not sure about this. It means that m_cap_lt will be 1 even if SAI return error. I would suggest to implement unit test if VS test does not support link training.

This is an issue in VS, do you know where is the source of this SAI model? SAI/bm/p4-sai?
Could we this VS enhancement deal in a separate PR? We'll need to do this for both AutoNeg and LinkTraining

Junchao-Mellanox · 2022-04-27T02:21:01Z

orchagent/portsorch.cpp

+    sai_status_t status = sai_port_api->set_port_attribute(port.m_port_id, &attr);
+    if (status != SAI_STATUS_SUCCESS)
+    {
+        SWSS_LOG_ERROR("Failed to set LT %s to port pid:%" PRIx64,


Better to print the port alias here

Junchao-Mellanox · 2022-04-27T02:21:32Z

orchagent/portsorch.cpp

+        return handleSaiSetStatus(SAI_API_PORT, status);
+    }
+
+    SWSS_LOG_INFO("Set LT %s to port pid:%" PRIx64, op.c_str(), port.m_port_id);


Better to print the port alias here, and maybe it should be NOTICE instead of INFO

Junchao-Mellanox · 2022-04-27T02:30:24Z

orchagent/portsorch.cpp

+
+    sai_attribute_t attr;
+    attr.id = SAI_PORT_ATTR_LINK_TRAINING_FAILURE_STATUS;
+    attr.value.u32 = 0;


Yes, this is redundant, will have it removed

Junchao-Mellanox · 2022-04-27T02:38:01Z

orchagent/portsorch.cpp

+    {
+        if ((it->second == 0) || !getPort(it->first, port))
+        {
+            m_port_state_poll.erase(it);


it will invalid the iterator and cause unpredictable behavior

Junchao-Mellanox · 2022-04-27T02:40:28Z

orchagent/portsorch.h

+
+    typedef enum {
+        PORT_STATE_POLL_NONE = 0,
+        PORT_STATE_POLL_AN   = 0x00000001, /* Auto Negotiation */


never used?

It's part of the review for AutoNeg, we're having 2 PRs in review at the same time, and they're using the same port state poller.

Junchao-Mellanox · 2022-04-27T02:43:59Z

orchagent/portsorch.cpp

+    }
+    else
+    {
+        m_port_state_poll[port.m_alias] &= ~type;


check if it should be remove from m_port_state_poll and stop timer if need

The stop() will only take place inside the poller thread to reduce the race conditions if any.

Junchao-Mellanox · 2022-04-27T02:44:11Z

orchagent/portsorch.cpp

+    if (active)
+    {
+        m_port_state_poll[port.m_alias] |= type;
+        m_timer->start();


what if the timer is already started?

This is now addressed in the PR below, issue start() to a running poller is no longer having the expiration timer reset.
sonic-net/sonic-swss-common#614

zhangyanzhao · 2022-05-16T21:56:17Z

/azpw run Azure sonic-swss

zhangyanzhao · 2022-05-16T21:56:55Z

/azpw run Azure.sonic-swss

ds952811 · 2022-05-17T00:18:00Z

/azpw run Azure.sonic-swss

mssonicbld · 2022-05-17T00:18:01Z

/AzurePipelines run Azure.sonic-swss

azure-pipelines · 2022-05-17T00:18:11Z

Azure Pipelines successfully started running 1 pipeline(s).

dgsudharsan · 2022-05-17T00:38:04Z

orchagent/portsorch.cpp

+
+                if (!lt_str.empty() && (p.m_type == Port::PHY))
+                {
+                    if (link_training_mode_map.find(lt_str) == link_training_mode_map.end())


Similar to AN, I don't see a lt capability query here. Please add

This is intentionally removed at the earlier commit

LT is similar to AN, it's a port capability available on a per-port basis, not system-wide on a single switch ASIC

For a port capability available on a per-port basis, SAI_PORT_ATTR_SUPPORTED_XXXX should be introduced into saiport.h

As per the latest comments from Rita, the SAI spec. was finalized around March time frame, it's too late for 202205 right now

Since this is a port capability available on a per-port basis, using SAI_PORT_ATTR_LINK_TRAINING_ENABLE in sai_query_attribute_capability is not going to work, and during the SAI subgroup meeting a few months back, there was a conclusion that per-port capability query should be tracked in a separated PR, as all the existing attributes should all be migrated, not just the new attribute for LT.

For your reference, the function prototype of sai_query_attribute_capability is as follows, it does not work on a per-port basis
https://github.com/opencomputeproject/SAI/blob/master/inc/saiobject.h#L221

sai_status_t sai_query_attribute_capability( _In_ sai_object_id_t switch_id, _In_ sai_object_type_t object_type, _In_ sai_attr_id_t attr_id, _Out_ sai_attr_capability_t *attr_capability);

Can you please add a comment to make sure once SAI is there we revisit and add the check? It is important that we ensure it so that devices and specific ports where it is not supported will have check in place to avoid SAI failure thereby prevent orchagent crash

dgsudharsan · 2022-05-17T00:42:54Z

orchagent/portsorch.cpp

+        }
+        if (port.m_admin_state_up && port.m_link_training > 0)
+        {
+            updatePortStateLinkTraining(port);


Should this move inside the below if condition?

We already have the if-statement for the port admin state and LT enablement, may I have more details of the additional checker that's missing? Thank you.

Similar to autoneg should updatePortStateLinkTraining be done only when status == SAI_PORT_OPER_STATUS_UP? Else it will be called redundantly during down as updateStatePoll will take care of invoking this.

In the case of AN

Link Up: Refresh AN states

Link Down: Clear the remote advertisement

In the case of LT

Link Up: Refresh LT states

Link Down: Refresh LT states

Hence the updatePortStateLinkTraining() is invoked before checking the operation link status

dgsudharsan · 2022-05-17T00:43:55Z

orchagent/portsorch.cpp

@@ -6883,3 +7132,115 @@ std::unordered_set<std::string> PortsOrch::generateCounterStats(const string& ty
    }
    return counter_stats;
 }
+
+void PortsOrch::updatePortStateAutoNeg(const Port &port)


The function name is misleading. Should it be updatePortRemoveAdvSpeeds?

This is a placeholder for the operational AN state, not specific to remote speed only, there are more upcoming in the PRs in the furture (e.g. tx/rx pause, FEC ....)

dgsudharsan · 2022-05-17T00:48:08Z

tests/test_port_an.py

+        fvs = swsscommon.FieldValuePairs([("autoneg","on"),("admin_status","up")])
+        ctbl.set("Ethernet0", fvs)
+
+        time.sleep(10)


Why do we need 10 second sleep?

Yep, you're right now, I'll remove this, thank you!

Allow me to rephrase, 10 sec is necessary.
Please note the poll interval is 5 sec, and in a worst-case scenario, the remote advertisement will not be available until 2nd poll.

prgeor · 2022-05-17T05:41:59Z

orchagent/portsorch.cpp

@@ -2128,6 +2211,32 @@ task_process_status PortsOrch::setPortAutoNeg(sai_object_id_t id, int an)
    return task_success;
 }

+task_process_status PortsOrch::setPortLinkTraining(const Port &port, bool state)


how is it ensured, LT is not enabled if its not supported on this port?

The LT capability checker is now restored, such a request will be blocked in doPortTask()

prgeor · 2022-05-17T05:47:11Z

orchagent/portsorch.cpp

+                    lt = link_training_mode_map[lt_str];
+                    if (lt != p.m_link_training)
+                    {
+                        auto status = setPortLinkTraining(p, lt > 0 ? true : false);


are we ensuring if port is admin down then we are not enabling LT?

The LT could be enabled/disable regardless of the port admin state, it's a set of control registers on the SerDes

prgeor · 2022-05-17T05:52:08Z

orchagent/portsorch.cpp

+    }
+}
+
+void PortsOrch::updatePortStateLinkTraining(const Port &port)


brief comment what this function does?

prgeor · 2022-05-17T05:52:28Z

orchagent/portsorch.cpp

@@ -6883,3 +7132,115 @@ std::unordered_set<std::string> PortsOrch::generateCounterStats(const string& ty
    }
    return counter_stats;
 }
+
+void PortsOrch::updatePortStateAutoNeg(const Port &port)


brief comment what this func does

dgsudharsan · 2022-05-19T01:38:20Z

orchagent/portsorch.cpp

+        }
+        if (port.m_admin_state_up && port.m_link_training > 0)
+        {
+            updatePortStateLinkTraining(port);


Similar to autoneg should updatePortStateLinkTraining be done only when status == SAI_PORT_OPER_STATUS_UP? Else it will be called redundantly during down as updateStatePoll will take care of invoking this.

dgsudharsan · 2022-05-19T01:41:47Z

orchagent/portsorch.cpp

+    sai_status_t status;
+    sai_attribute_t attr;
+
+    attr.id = SAI_PORT_ATTR_SUPPORTED_AUTO_NEG_MODE;


I have one major concern here. Since SAI enforcing this involves SAI implementing this attribute, do we know whether all vendors support it? If not it might break autoneg in their platforms

Yes, it will cause breakage, hence the AN capability will be updated to use 1(supported) by default

Sure. Please use this capability to a warning log in case if SAI returns not supported for the attribute. SAI vendors can then mitigate to implement this attribute without the functional impact.

dgsudharsan · 2022-05-21T14:58:24Z

orchagent/portsorch.cpp

+    // it's not available in SAI v1.10 paried with SONiC.202205.
+    // And given that LT is part of AN, it should be okay to use
+    // SAI_PORT_ATTR_SUPPORTED_AUTO_NEG_MODE as a fallback plan.
+    attr.id = SAI_PORT_ATTR_SUPPORTED_AUTO_NEG_MODE;


I prefer not to use a wrong attribute. Instead add a comment as todo and leave without using capability.

dgsudharsan · 2022-05-21T15:00:35Z

orchagent/portsorch.cpp

-        SWSS_LOG_NOTICE("Unable to get %s AN capability", port.m_alias.c_str());
+        // To avoid breakage on the existing platforms, AN should be 1 by default
+        port.m_cap_an = 1;
+        SWSS_LOG_WARN("Unable to get %s AN capability, assumming it's supported",


I prefer you just add "Unable to get AN support capability" in the log. Please remove "assuming it's supported"

dgsudharsan · 2022-06-01T19:56:03Z

orchagent/portsorch.cpp

-        m_portStateTable.hset(port.m_alias, "rmt_adv_speeds", adv_speeds);
-    }
-    else
+    if (port.m_admin_state_up)


Previously the get remote adv speeds is done even when admin state is down and this new logic will restrict it to only when admin up.
This means polling becomes irrelevant. Can you explain the change for the reason?

When the port is admin down, the SerDes is stopped, and it's impossible to get a correct remote advertisement, hence this redundant operation should be skipped if the port is disabled.

Then why do we need the poller? Can't we just update the remote advertised speeds on link up? The entire polling logic was introduced to get remote advertisement when link was down. Should that be removed?

Sorry for the confusion, but that's a different story

MAC disabled + link down
This is a case we have port.m_admin_state_up == false, while the link is surely down, and the remote advertisement is not available since the MAC/SerDes is disabled

MAC enabled + link down
This is a case we have port.m_admin_state_up == true, while the link is operationally down, and the remote advertisement could be available as the MAC/SerDes is enabled, frame exchanges are happening on the PHY layer (i.e. It's the internal PHY, in this case. Precisely speaking, it's the TSC drivers on the Broadcom chips)

MAC enabled + link up
This is a general case of port.m_admin_state_up == true and link is operationally up.

Conclusion:
While the remote advertisement could be available in case 2 and 3, it's not available in case 1, hence the poller should only fetching the status when MAC/SerDes is enabled.

prgeor · 2022-06-20T17:38:58Z

"/azp run"

prgeor · 2022-06-22T21:23:28Z

@ds952811 can you rebase with latest swss master to see if the build test passes?

prsunny · 2022-06-22T21:26:00Z

/azp run

azure-pipelines · 2022-06-22T21:26:09Z

Azure Pipelines successfully started running 1 pipeline(s).

commit 18632a3 Author: Dante Su <[email protected]> Date: Mon May 23 12:22:49 2022 +0000 optimize port state refresh logic Signed-off-by: Dante Su <[email protected]> commit 081d491 Author: ds952811 <[email protected]> Date: Mon May 23 02:33:56 2022 +0000 address review comments Signed-off-by: ds952811 <[email protected]> commit 84bdde4 Author: Dante Su <[email protected]> Date: Fri May 20 02:15:59 2022 +0000 update the default LT capability upon get failures Signed-off-by: Dante Su <[email protected]> commit 0f73666 Author: Dante Su <[email protected]> Date: Thu May 19 11:28:38 2022 +0000 Rename updatePortStatesXX as refreshPortStatesXX Signed-off-by: Dante Su <[email protected]> commit ddd57fe Author: Dante Su <[email protected]> Date: Thu May 19 04:03:13 2022 +0000 Have AN cap defaults to 1, and use AN attr for LT cap query Signed-off-by: Dante Su <[email protected]> commit 876e605 Author: Dante Su <[email protected]> Date: Fri May 13 11:15:12 2022 +0000 drop LT capability query Signed-off-by: Dante Su <[email protected]> commit 55ced7d Author: Dante Su <[email protected]> Date: Fri Apr 29 13:53:17 2022 +0000 incorporate autoneg support from PR#2215 Signed-off-by: Dante Su <[email protected]> commit a04594e Author: Dante Su <[email protected]> Date: Thu Apr 28 16:33:14 2022 +0000 address review comments Signed-off-by: Dante Su <[email protected]> commit e9eeb9a Author: Dante Su <[email protected]> Date: Thu Apr 28 15:00:04 2022 +0000 address review comments Signed-off-by: Dante Su <[email protected]> commit 4ff604d Author: Dante Su <[email protected]> Date: Fri Apr 22 03:51:56 2022 +0000 Stop the port state poll by default Signed-off-by: Dante Su <[email protected]> commit bdfb8d8 Author: Dante Su <[email protected]> Date: Fri Apr 22 03:48:07 2022 +0000 address review comments Signed-off-by: Dante Su <[email protected]> commit 1c6bda8 Author: Dante Su <[email protected]> Date: Mon Apr 18 08:46:21 2022 +0000 Restore pre-emphasis when LT is transitioned from ON to OFF Signed-off-by: Dante Su <[email protected]> commit 09a9b33 Author: Dante Su <[email protected]> Date: Mon Apr 18 02:33:11 2022 +0000 fix build failure due to SAI_PORT_ATTR_SUPPORTED_LINK_TRAINING_MODE Signed-off-by: Dante Su <[email protected]> commit b0bee3e Author: Dante Su <[email protected]> Date: Thu Apr 14 07:54:14 2022 +0000 address review comments Signed-off-by: Dante Su <[email protected]> commit c4345ef Author: Dante Su <[email protected]> Date: Fri Mar 25 02:26:05 2022 +0000 portsorch: initial support for link-training - What I did Add Link-Training support to portsorch, while Gearbox is not in the scope - Why I did it In the case of DAC, static pre-calibrated pre-emphasis is rarely available on SONIC, as most of the ODM are expecting this to be done dynamically at runtime via link-training, hence we'll need this feature to improve the link quality - How I verified it Manual test Ran the Unit-tests to the corresponding changes Signed-off-by: Dante Su <[email protected]>

ds952811 requested a review from prsunny as a code owner February 18, 2022 02:21

ds952811 mentioned this pull request Feb 18, 2022

[link-training] initial support sonic-net/SONiC#925

Merged

ds952811 mentioned this pull request Mar 25, 2022

schema.h: add STATE_LINK_TRAINING_TABLE_NAME sonic-net/sonic-swss-common#592

Closed

ds952811 force-pushed the link-training branch from 37887a2 to 0983f5a Compare March 25, 2022 03:09

dgsudharsan requested changes Apr 1, 2022

View reviewed changes

dgsudharsan requested changes Apr 20, 2022

View reviewed changes

dgsudharsan requested changes Apr 21, 2022

View reviewed changes

ds952811 force-pushed the link-training branch from 64020f7 to 5cf3f0e Compare April 22, 2022 03:48

Junchao-Mellanox suggested changes Apr 27, 2022

View reviewed changes

prsunny requested a review from prgeor April 28, 2022 17:30

ds952811 changed the title ~~portsorch: initial support for link-training~~ portsorch: initial support for link-training and remote advertisement May 4, 2022

This was referenced May 4, 2022

[autoneg] add support for remote speed advertisement and clarify the expected autoneg behaviors sonic-net/SONiC#924

Merged

[autoneg] add support for capability checker and remote speed advertisement #2215

Closed

dgsudharsan requested changes May 17, 2022

View reviewed changes

prgeor reviewed May 18, 2022

View reviewed changes

dgsudharsan requested changes May 19, 2022

View reviewed changes

ds952811 force-pushed the link-training branch from f932c23 to ddd57fe Compare May 19, 2022 07:14

dgsudharsan requested changes May 21, 2022

View reviewed changes

dgsudharsan requested changes Jun 1, 2022

View reviewed changes

ds952811 force-pushed the link-training branch from 18632a3 to d9a1f03 Compare June 14, 2022 00:44

Junchao-Mellanox previously approved these changes Jun 23, 2022

View reviewed changes

ds952811 dismissed Junchao-Mellanox’s stale review via 4de2dbc June 23, 2022 01:48

ds952811 force-pushed the link-training branch from d9a1f03 to 4de2dbc Compare June 23, 2022 01:48

dgsudharsan approved these changes Jun 23, 2022

View reviewed changes

ds952811 closed this Jun 28, 2022

ds952811 deleted the link-training branch June 28, 2022 09:03

ds952811 mentioned this pull request Jun 28, 2022

portsorch: initial support for link-training #2359

Merged

portsorch: initial support for link-training and remote advertisement #2149

portsorch: initial support for link-training and remote advertisement #2149

Conversation

ds952811 commented Feb 18, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ds952811 Apr 14, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dgsudharsan Apr 20, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhangyanzhao commented May 16, 2022

zhangyanzhao commented May 16, 2022

ds952811 commented May 17, 2022

mssonicbld commented May 17, 2022

azure-pipelines bot commented May 17, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ds952811 May 19, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ds952811 commented Feb 18, 2022 •

edited

Loading

ds952811 Apr 14, 2022 •

edited

Loading

dgsudharsan Apr 20, 2022 •

edited

Loading

ds952811 May 19, 2022 •

edited

Loading