Re: [PATCH bpf-next 2/4] bpf: fix lockdep false positive in stackmap

[Eric Dumazet]

On 02/05/2019 07:30 PM, Alexei Starovoitov wrote:

> Thanks for the reminder :)
> 
> I've been waiting for Peter's direction on this one.
> Happy to fix it whichever way.
> 
> To recap:
> Approach 1:
> s/up_read/up_read_non_owner/ from irq_work + rwsem_release as Longman 
> proposed.
> 
> Apporach 2:
> introduce down_read_trylock_non_owner and pair it with up_read_non_owner
> in both irq_work and normal.
> 
> My preference is 1. I think Longman's as well.
> 


Unless we envision other uses for down_read_trylock_non_owner() and 
up_read_non_owner(),
I agree 1) seems good enough, at least for the short term.

Thanks !