Bug 210097
| Summary: | 5.9.6 hard lockup with infiniband | ||
|---|---|---|---|
| Product: | File System | Reporter: | SimplyCorbett (thezombiehunter) |
| Component: | NFS | Assignee: | Tejun Heo (tj) |
| Status: | NEW --- | ||
| Severity: | high | CC: | cel |
| Priority: | P1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Kernel Version: | 5.9.6 | Subsystem: | |
| Regression: | No | Bisected commit-id: | |
| Attachments: |
lshw
Kernel log Kernel config |
||
Created attachment 293535 [details]
Kernel log
Created attachment 293537 [details]
Kernel config
As of 5.9.11 the kernel is no longer hard crashing and files successfully transfer. However, whatever is transferring the files becomes unresponsive until the transfer is finished which did not happen in kernel 5.8. |
Created attachment 293533 [details] lshw Upgraded from 5.8.18 to 5.9.6 and I'm now getting hard lockups. Setup: 3900x 128GB RAM 2080 ti Several SATA devices with both f2fs and ext4, all mounted with cryptsetup ConnectX-3 infiniband FDR I have two connectx-3 cards connected together directly with opensm, no switch. On 5.8.18 this works fine. On 5.9.6 I am getting hard lockups. Replication: CLIENT: /etc/fstab 172.16.1.20:/mnt/Heavy-1 /mnt/Ares-3TB-1 nfs4 _netdev,auto,proto=rdma,port=20049,hard,intr,rsize=65536,wsize=65536,noatime 0 0 SERVER: echo rdma 20049 > /proc/fs/nfsd/portlist BOTH: /etc/init.d/opensm stop && modprobe ib_ipoib && modprobe ib_umad && modprobe ib_ipoib && modprobe ib_srp && modprobe ib_uverbs && modprobe rdma_ucm && modprobe mlx4_ib && modprobe ib_core && modprobe svcrdma && modprobe xprtrdma && /etc/init.d/net.ib0 start && /etc/init.d/net.ib1 start && /etc/init.d/opensm start How to replicate: Transfer 30GB .img file from server to client, wait five to ten seconds, watch kernel go nope and crash. Logs, lshw attached.