Hi,

after resuming from S3 (deep) suspend, both ports of my E810-C-Q2
are unusable: the VSI rebuild succeeds, but opening the VSI fails
because adding the VLAN 0 filter returns -EIO. Unloading and
reloading ice + irdma restores the card every time. The issue
reproduces reliably on every suspend/resume cycle.

Hardware / software
-------------------
NIC:      Intel E810-C-Q2 [8086:1592], subsystem [8086:0002], rev 02
          PBA K91258-011, 2x 100G QSFP28, RoCEv2 (irdma) in use
Board:    ASUS Pro WS WRX80E-SAGE SE WIFI II, BIOS 1702 (08/22/2025)
          AMD Threadripper (WRX80)
Kernel:   7.2.11-1-cachyos (in-tree ice and irdma)
NVM:      5.10 (0x80022601), fw.mgmt 8.1.1, netlist 4.4.2000-3.29.0,
          OROM 1.4113.0
          Also reproduced with NVM 4.90 (0x80020ef2, fw.mgmt 7.9.1)
DDP:      ICE OS Default Package 1.3.43.0
Sleep:    mem_sleep = deep (S3)
Config:   MTU 9000, DCB enabled, FW LLDP disabled (DCBx/LLDP in SW mode)

Steps to reproduce
------------------
1. Boot, link comes up at 100G, RDMA port ACTIVE
2. systemctl suspend
3. Resume

Expected: both ports operational after resume
Actual:   "Failed to open VSI", no traffic possible

Kernel log (resume)
-------------------
PM: suspend entry (deep)
ice 0000:61:00.0: PTP reset successful
ice 0000:61:00.0: VSI rebuilt. VSI index 0, type ICE_VSI_PF
ice 0000:61:00.0: VSI rebuilt. VSI index 1, type ICE_VSI_CTRL
ice 0000:61:00.1: PTP reset successful
ice 0000:61:00.1: VSI rebuilt. VSI index 0, type ICE_VSI_PF
ice 0000:61:00.1: VSI rebuilt. VSI index 1, type ICE_VSI_CTRL
PM: suspend exit
ice 0000:61:00.0: Failure Adding VLAN 0 on VSI 6, status -5
ice 0000:61:00.0 enp97s0f0np0: Failed to open VSI 0x0006 on switch 0x0001
ice 0000:61:00.1: Failure Adding VLAN 0 on VSI 8, status -5
ice 0000:61:00.1 enp97s0f1np1: Failed to open VSI 0x0008 on switch 0x0000

When unloading the driver afterwards:
ice 0000:61:00.1: Failed to remove Rx LLDP rule on VSI 8 error: -5
ice 0000:61:00.0: Failed to remove Rx LLDP rule on VSI 6 error: -5

Observation
-----------
During the NVM update to 5.10, the firmware was activated via EMP
reset while the driver was loaded. That rebuild succeeded, and the
log shows the DDP package being downloaded again:

ice 0000:61:00.0: The DDP package was successfully loaded: ICE OS
Default Package version 1.3.43.0
...
ice 0000:61:00.0: VSI rebuilt. VSI index 0, type ICE_VSI_PF

The rebuild after resume shows no DDP download at all, and then
the switch filter commands fail. My guess is that the device loses
power in S3, the DDP package is lost, and the resume path does not
reload it. I have not verified this in the code, so please treat it
as a hint only.

Possibly related: at every boot the kernel logs
pci_bus 0000:60: Some PCI device resources are unassigned, try
booting with pci=realloc
(the NIC is at 0000:61:00.x). In addition, every boot shows
"Failed to add generic LLDP Rx filter on VSI x error: -5, falling
back to specialized AQ control" on both ports.

Workaround
----------
modprobe -r irdma; modprobe -r ice; modprobe ice; modprobe irdma

I am happy to test patches, provide full logs, or reproduce on a
mainline kernel if that helps.

Thanks,
Timo

P.S. Sorry for the noise: my first mail on this topic was sent by
mistake due to a misconfiguration in my mail client. Please
disregard it and use this one instead.

Reply via email to