Re: [PATCH] infiniband: i40iw: replace atomic_add_return()

[Jason Gunthorpe]

On Tue, Dec 01, 2020 at 02:42:25PM +0000, Saleem, Shiraz wrote:

> >  	case I40IW_CM_STATE_ESTABLISHED:
> >  	case I40IW_CM_STATE_SYN_RCVD:
> > @@ -3020,7 +3020,7 @@ static int i40iw_cm_reject(struct i40iw_cm_node
> > *cm_node, const void *pdata, u8
> >  	i40iw_cleanup_retrans_entry(cm_node);
> > 
> >  	if (!loopback) {
> > -		passive_state = atomic_add_return(1, &cm_node->passive_state);
> > +		passive_state = atomic_inc_return(&cm_node->passive_state);
> 
> Fine with it as its consistent across i40iw. But aren't there many
> more instances of this across the tree?  Isn't this a choice best
> left to the developer?

I don't think it is a style choice, the narrowest applicable atomic
should always be used as a matter of clarity and performance.

Jason