DPU modules take >16 minutes to come Online after chassis module startup on Cisco 8102-28FH-DPU-O

[vvolam]

Description

DPU modules (DPU0-DPU3) on Cisco 8102-28FH-DPU-O SmartSwitch take more than 16 minutes to transition from Offline to Online after config chassis module startup, far exceeding the expected ~6 minute window.

Steps to Reproduce

1. On Cisco 8102-28FH-DPU-O running SONiC build 20251110.16
2. Shut down DPU modules: config chassis module shutdown DPU0 (through DPU3)
3. Wait for all DPUs to reach Offline/down state (~30 seconds)
4. Start up DPU modules: config chassis module startup DPU0 (through DPU3)
5. DPUs remain Offline for >16 minutes before eventually coming Online

Expected Behavior

DPU modules should come Online within 6 minutes (360 seconds) after config chassis module startup.

Actual Behavior

• Startup command issued at ~23:41 UTC
• DPUs still Offline at ~23:53 UTC (after 12 minutes / 720 seconds of polling)
• DPUs eventually come Online around ~23:58 UTC (~17 minutes after startup)

All 4 configured DPUs (DPU0-DPU3) exhibit this behavior consistently across multiple test runs.

Platform Information

• Platform: x86_64-8102_28fh_dpu_o-r0
• HwSKU: Cisco-8102-28FH-DPU-O
• ASIC: cisco-8000
• DPU type: AMD Pensando DSC (arm64-elba-asic-flash128-r0)
• Build: 20251110.16 (Debian 13 Bookworm)

Impact

This causes test_reboot_cause and subsequent tests (test_system_health_state, test_dpu_console, test_npu_dpu_date) in sonic-mgmt SmartSwitch platform tests to fail due to DPU module status timeout.

Logs

DPU status during startup wait (every 30 seconds for 360s):

DPU0 is offline ...
DPU1 is offline ...
DPU2 is offline ...
DPU3 is offline ...
(repeats for 12 polling cycles)
check_dpu_module_status is still False after 360 seconds, exit with False

After test failure, during teardown (~17 min after startup):

DPU3  AMD Pensando DSC   N/A   Online   up
DPU2  AMD Pensando DSC   N/A   Online   up
DPU1  AMD Pensando DSC   N/A   Online   up
DPU0  AMD Pensando DSC   N/A   Online   up