workqueue: WQ_MEM_RECLAIM nvmet-wq:nvmet_rdma_release_queue_work [nvmet_rdma] is flushing !WQ_MEM_RECLAIM irdma-cleanup-wq:irdma_flush_worker [irdma]

[Zhu Yanjun]

在 2024/12/14 3:26, Honggang LI 写道:
> On Fri, Dec 13, 2024 at 08:30:01PM +0100, Zhu Yanjun wrote:
>> I delved into this problem. It seems that it is a known problem.
>> Can you apply the following to make tests again?
>>
>> diff --git a/drivers/infiniband/hw/irdma/hw.c
>> b/drivers/infiniband/hw/irdma/hw.c
>> index ad50b77282f8..31501ff9f282 100644
>> --- a/drivers/infiniband/hw/irdma/hw.c
>> +++ b/drivers/infiniband/hw/irdma/hw.c
>> @@ -1872,7 +1872,7 @@ int irdma_rt_init_hw(struct irdma_device *iwdev,
>>                   * free cq bufs
>>                   */
>>                  iwdev->cleanup_wq = alloc_workqueue("irdma-cleanup-wq",
>> -                                       WQ_UNBOUND, WQ_UNBOUND_MAX_ACTIVE);
>> +                                       WQ_UNBOUND|WQ_MEM_RECLAIM,
> 
> After add flag WQ_MEM_RECLAIM, the warning message is gone. However,
> it may raise similar issue fixed by commit 2cc7d150550 again.
> 
> thanks
> 
> commit 2cc7d150550cc981aceedf008f5459193282425c
> Author: Sindhu Devale <sindhu.devale at intel.com>
> Date:   Tue Apr 23 11:27:17 2024 -0700
> 
>      i40e: Do not use WQ_MEM_RECLAIM flag for workqueue

I read the commit log carefully. If I understand the commit log 
correctly, the flag WQ_MEM_RECLAIM is used in i40e while it is
not used in i40iw.
The fix is to remove the flag WQ_MEM_RECLAIM from i40e.

"
     Issue reported by customer during SRIOV testing, call trace:
     When both i40e and the i40iw driver are loaded, a warning
     in check_flush_dependency is being triggered. This seems
     to be because of the i40e driver workqueue is allocated with
     the WQ_MEM_RECLAIM flag, and the i40iw one is not.

     Similar error was encountered on ice too and it was fixed by
     removing the flag. Do the same for i40e too.
"
I do not have E810 device and i40e device and can not read the issue 
reported by customer during SRIOV testing.
Thus, let Intel engineers continue to handle this problem.
@Mustafa Ismail

Zhu Yanjun

>