[PATCH] nvmet-rdma: Suppress a class of lockdep complaints

[Max Gurtovoy]

On 10/05/2023 2:13, Bart Van Assche wrote:
> On 5/9/23 15:55, Max Gurtovoy wrote:
>> Can we fix the lockdep not to get confused
> 
> Lockdep uses static keys by default. Associating a single static key 
> with multiple synchronization object instances can cause false 
> positives. This is why I introduced support for dynamic keys in lockdep 
> about four years ago. I don't think that it is possible to make dynamic 
> keys the default without changing all synchronization primitives.
> 
>> or ignore it if its not a real issue ?
> 
> Most lockdep reports indicate a real problem so every lockdep report 
> should be reviewed. Having to review false positive complaints every 
> time blocktests is run is tedious. I think it is much better to suppress 
> false positive complaints (in a safe way) rather than asking everyone 
> who runs blktests to review all false positive lockdep complaints every 
> time blktests is run.

Yes, I understand the importance of that for sure.
But the concept of "dirtying" the base code only for the sake of the 
above reason sounds like a bad practice in general.

Maybe for that we can have some mechanism of side patches that one apply 
before running the blktests ?
This will leave the base code clean but will need to maintain these 
patches to be applied cleanly on the base code.
But I guess we can have some automation for that..

> 
> Bart.