Icetea: qsfp_service: DR2_4x400G,DR1_8x200G,DR1_8x100G optics support by joywu-coder · Pull Request #557 · facebook/fboss

joywu-coder · 2025-09-22T11:15:28Z

Description

Add codes to support the new DR2_4x400G, DR1_8x200G, and DR1_8x100G modes for Icetea QSFP service.

Add new profile PROFILE_100G_1_PAM4_RS544X2N_OPTICAL = 54 to switch_config.thrift.
Add codes to support the new DR2_4x400G, DR1_8x200G, and DR1_8x100G modes.
Change QSFP_HW_TEST codes to support the verification of the three new modes.
Change kUsecDatapathStateUpdateTime to 10s(actual test needs 7~8s) to release the lanes from DeInit, fix the error print: "E0919 17:16:58.348301 28340 CmisModule.cpp:3906] Transceiver 0: Datapath didn't come out of deactivated state even after waiting 5000000 uSec".

Motivation

According to the requirement change, Icetea QSFP service needs to support the new DR2_4x400G, DR1_8x200G, and DR1_8x100G modes.

Test Plan

Test cases will cover the following modes:

2 x 800G: 2 logical ports, each port contains 4 lanes and 200G per lane.
---->SMFMediaInterfaceCode: DR4_800G = 0x77
---->PROFILE_800G_4_PAM4_RS544X2N_OPTICAL = 52
2 x 400G: 2 logical ports, each port contains 4 lanes and 100G per lane.
---->SMFMediaInterfaceCode: DR4_400G = 0x1C
---->PROFILE_400G_4_PAM4_RS544X2N_OPTICAL = 38
4 x 400G: 4 logical ports, each port contains 2 lanes and 200G per lane.
---->SMFMediaInterfaceCode: DR2_400G = 0x75
---->PROFILE_400G_2_PAM4_RS544X2N_OPTICAL = 51
8 x 200G: 8 logical ports, each port contains 1 lane and 200G per lane.
---->SMFMediaInterfaceCode: DR1_200G = 0x73
---->PROFILE_200G_1_PAM4_RS544X2N_OPTICAL = 53
8 x 100G: 8 logical ports, each port contains 1 lane and 100G per lane.
---->SMFMediaInterfaceCode: DR1_100G = 0x14
---->PROFILE_100G_1_PAM4_RS544X2N_OPTICAL = 54

Test steps as follows:

Compilation has passed for qsfp_hw_test, qsfp_service, and fboss-platform-mapping-gen
Run "fboss-platform-mapping-gen --platform-name icetea800bc" can generate the platform_mapping configs with the new profile "PROFILE_100G_1_PAM4_RS544X2N_OPTICAL = 54".
icetea800bc_platform_mapping_53-54-51-38-52.json
Transceivers deploy in the front panel:
Ports 1-18: 18 Finisar optical modules.
Ports 19-32: 14 Terahop(Innolight) optical modules.
Port 33: 1 Innolight QSFP28.
QSFP Hardware Test Config:
Ports 1-4: PROFILE_200G_1_PAM4_RS544X2N_OPTICAL = 53.
Ports 5-8: PROFILE_100G_1_PAM4_RS544X2N_OPTICAL = 54.
Ports 9-12: PROFILE_400G_2_PAM4_RS544X2N_OPTICAL = 51.
Ports 13-16: PROFILE_400G_4_PAM4_RS544X2N_OPTICAL = 38.
Ports 17-32: PROFILE_800G_4_PAM4_RS544X2N_OPTICAL = 52.
Port 33(QSFP28): PROFILE_100G_4_NRZ_RS528_OPTICAL = 23.
qsfp_profile_53-54-51-38-52_all_ports_conf.txt
Run 0.7.2 t0_qsfp_hw_tests, the result passed:
./bin/run_test.py qsfp --qsfp-config ./qsfp_profile_53-54-51-38-52_all_ports.conf --platform_mapping_override_path ./icetea800bc_platform_mapping_53-54-51-38-52.json --filter_file=./share/hw_sanity_tests/t0_qsfp_hw_tests.conf --skip-known-bad-tests "icetea800bc/physdk-credo-0.7.2/credo-0.7.2"

Run 0.7.2 100% QSFP HW tests, the result passed:
./bin/run_test.py qsfp --qsfp-config ./qsfp_profile_53-54-51-38-52_all_ports.conf --platform_mapping_override_path ./icetea800bc_platform_mapping_53-54-51-38-52.json --skip-known-bad-tests "icetea800bc/physdk-credo-0.7.2/credo-0.7.2"

Run 0.9.0 t0_qsfp_hw_tests, the result passed:
./bin/run_test.py qsfp --qsfp-config ./qsfp_profile_53-54-51-38-52_all_ports.conf --platform_mapping_override_path ./icetea800bc_platform_mapping_53-54-51-38-52.json --filter_file=./share/hw_sanity_tests/t0_qsfp_hw_tests.conf --skip-known-bad-tests "icetea800bc/physdk-credo-0.9.0/credo-0.9.0"

Run 0.9.0 100% QSFP HW tests, the result passed:
./bin/run_test.py qsfp --qsfp-config ./qsfp_profile_53-54-51-38-52_all_ports.conf --platform_mapping_override_path ./icetea800bc_platform_mapping_53-54-51-38-52.json --skip-known-bad-tests "icetea800bc/physdk-credo-0.9.0/credo-0.9.0"

Please refer to the detailed logs in GDrive: https://drive.google.com/drive/folders/1tJns4Vmogfe0g7Fu5BgN66uTV47xWrAs

Test result for the commit: 920a04e

Run 0.7.2 t0_qsfp_hw_tests, the result passed:

2. Run 0.7.2 100% QSFP HW tests, the result passed:

Please refer to the new logs in GDrive: https://drive.google.com/drive/folders/1seDPes3BA5gnPuzrPwI65AKtpCEV68sX

harshitgulati18 · 2025-09-24T04:46:06Z

  PROFILE_400G_2_PAM4_RS544X2N_OPTICAL = 51,
  PROFILE_800G_4_PAM4_RS544X2N_OPTICAL = 52,
  PROFILE_200G_1_PAM4_RS544X2N_OPTICAL = 53,
+  PROFILE_100G_1_PAM4_RS544X2N_OPTICAL = 54,


Can we reuse PROFILE_100G_1_PAM4_RS544_OPTICAL instead of defining another new profile?

Can we reuse PROFILE_100G_1_PAM4_RS544_OPTICAL instead of defining another new profile?

Yes, using PROFILE_100G_1_PAM4_RS544_OPTICAL = 47 can work as well. We originally thought the "FEC RS544X2N" for 100G should be supported, too.

harshitgulati18 · 2025-09-24T04:49:08Z

+  DR2_400G = 27,
+  DR2_4x400G = 28
+  DR1_8x200G = 29,
+  DR1_100G = 30,
+  DR1_8x100G = 31,


We only need DR2_400G and DR1_100G here

DR2_4x400G is DR2_400G media interface code on lanes 0,2,4,6. Similarly for the other modes

We only need DR2_400G and DR1_100G here

DR2_4x400G is DR2_400G media interface code on lanes 0,2,4,6. Similarly for the other modes

Got it, thanks.

harshitgulati18 · 2025-09-24T04:52:32Z

            SMFMediaInterfaceCode::LR4_10_400G,
            SMFMediaInterfaceCode::LR4_10_400G,
        },
+        /* new requirement not supoort the two combinations


Please keep this section and not comment out. While not a supported mode on Icetea right now, we may still support it on other platforms

Please keep this section and not comment out. While not a supported mode on Icetea right now, we may still support it on other platforms

Got it, thanks.

harshitgulati18 · 2025-09-24T04:56:51Z

+    } else if (
+        smfCode == SMFMediaInterfaceCode::DR2_400G &&
+        firstModuleCapability->hostStartLanes.size() == 4) {
+      moduleMediaInterface = MediaInterfaceCode::DR2_4x400G;
+    } else if (
+        smfCode == SMFMediaInterfaceCode::DR1_200G &&
+        firstModuleCapability->hostStartLanes.size() == 8) {
+      moduleMediaInterface = MediaInterfaceCode::DR1_8x200G;
+    } else if (
+        smfCode == SMFMediaInterfaceCode::DR1_100G &&
+        firstModuleCapability->hostStartLanes.size() == 8) {
+      moduleMediaInterface = MediaInterfaceCode::DR1_8x100G;


We don't need these changes.

moduleMediaInterface corresponds to the SMF code corresponding to the first Application that module advertises and not the speed we configure on it. In the 2x800G-DR4 case, the module always advertises 800G-DR4 on lanes 0 and 2. The code in lines 2960-2962 already checks this.

We don't need these changes.

moduleMediaInterface corresponds to the SMF code corresponding to the first Application that module advertises and not the speed we configure on it. In the 2x800G-DR4 case, the module always advertises 800G-DR4 on lanes 0 and 2. The code in lines 2960-2962 already checks this.

Got it, thanks.

harshitgulati18 · 2025-09-24T04:59:27Z

        }
        break;
+      case MediaInterfaceCode::DR2_400G:
+        expectedMediaLanes = {0, 1};


expectedMediaLane depends on hostLane right?
If hostLane is 0,1 expectedMediaLanes = {0,1}
If hostLane is 2,3 expectedMediaLanes = {2,3}
If hostLane is 4,5 expectedMediaLanes = {4,5}
If hostLane is 6,7 expectedMediaLanes = {6,7}

expectedMediaLane depends on hostLane right? If hostLane is 0,1 expectedMediaLanes = {0,1} If hostLane is 2,3 expectedMediaLanes = {2,3} If hostLane is 4,5 expectedMediaLanes = {4,5} If hostLane is 6,7 expectedMediaLanes = {6,7}

Yes, I have handled the mapping in the switch case "MediaInterfaceCode::DR4_2x800G" and the sub switch case "cfg::PortProfileID::PROFILE_400G_2_PAM4_RS544X2N_OPTICAL".

harshitgulati18 · 2025-09-24T05:01:12Z

+            *mediaIntfCode == MediaInterfaceCode::DR2_4x400G ||
+            *mediaIntfCode == MediaInterfaceCode::DR1_8x200G ||
+            *mediaIntfCode == MediaInterfaceCode::DR1_8x100G ||


Not required, as this switch case is on moduleMediaInterface which is always DR4_2x800G

Not required, as this switch case is on moduleMediaInterface which is always DR4_2x800G

Got it, thanks.

harshitgulati18 · 2025-09-24T05:01:41Z

+             *mediaIntfCode == MediaInterfaceCode::DR2_4x400G ||
+             *mediaIntfCode == MediaInterfaceCode::DR1_8x200G ||
+             *mediaIntfCode == MediaInterfaceCode::DR1_8x100G ||


Not required, as this switch case is on moduleMediaInterface which is always DR4_2x800G

Not required, as this switch case is on moduleMediaInterface which is always DR4_2x800G

Got it, thanks.

shiva-menta · 2025-09-25T22:12:56Z

@joywu-coder Can you let us know once you've tested with these new code changes?

joywu-coder · 2025-09-26T00:37:00Z

@joywu-coder Can you let us know once you've tested with these new code changes?

@shiva-menta Yes, it has been verified. The new logs are attached at the end of the description. Please check. Thank you.

facebook-github-bot · 2025-09-26T14:45:28Z

@shiva-menta has imported this pull request. If you are a Meta employee, you can view this in D83347793.

facebook-github-bot · 2025-10-01T17:37:01Z

@shiva-menta merged this pull request in 49a2778.

Icetea: qsfp_service: DR2_4x400G,DR1_8x200G,DR1_8x100G optics support

be6478d

meta-cla Bot added the CLA Signed label Sep 22, 2025

harshitgulati18 requested changes Sep 24, 2025

View reviewed changes

joywu-coder added 3 commits September 25, 2025 17:07

Merge branch 'facebook:main' into icetea_qsfp_new_modes_support

547b5d4

Icetea: qsfp_service: revise the code based on the review comments

e4066dc

Icetea: qsfp_service: fix the clang-format issue

920a04e

meta-codesync Bot closed this in 49a2778 Oct 1, 2025

facebook-github-bot added the Merged label Oct 1, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Icetea: qsfp_service: DR2_4x400G,DR1_8x200G,DR1_8x100G optics support#557

Icetea: qsfp_service: DR2_4x400G,DR1_8x200G,DR1_8x100G optics support#557
joywu-coder wants to merge 4 commits into
facebook:mainfrom
joywu-coder:icetea_qsfp_new_modes_support

joywu-coder commented Sep 22, 2025 •

edited

Loading

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

harshitgulati18 Sep 24, 2025

joywu-coder Sep 25, 2025

shiva-menta commented Sep 25, 2025

joywu-coder commented Sep 26, 2025

facebook-github-bot commented Sep 26, 2025

facebook-github-bot commented Oct 1, 2025

Labels

4 participants

Uh oh!

Conversation

joywu-coder commented Sep 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Motivation

Test Plan

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

shiva-menta commented Sep 25, 2025

joywu-coder commented Sep 26, 2025

facebook-github-bot commented Sep 26, 2025

facebook-github-bot commented Oct 1, 2025

Labels

4 participants

joywu-coder commented Sep 22, 2025 •

edited

Loading