Re: [PATCH rdma-next v12 5/6] RDMA/bnxt_re: Support dmabuf for CQ rings

[Leon Romanovsky <leon@kernel.org>]

On Fri, Feb 13, 2026 at 10:54:25AM -0400, Jason Gunthorpe wrote:
> On Fri, Feb 13, 2026 at 01:12:56PM +0200, Leon Romanovsky wrote:
> > > +int bnxt_re_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
> > > +		      struct uverbs_attr_bundle *attrs)
> > > +{
> > > +	return bnxt_re_create_cq_umem(ibcq, attr, NULL, attrs);
> > > +}
> > 
> > Please don't mix create_cq and create_cq_umem.
> > https://lore.kernel.org/linux-rdma/20260213-refactor-umem-v1-15-f3be85847922@nvidia.com/T/#u
> 
> Either we drop this one patch and put those 50 ahead of it, or we just
> take this one and rebase the above.. The above has the advantage that
> it enables all drivers to support cq dmabuf in one giant shot.
> 
> However, frankly I'm getting tired of looking at this bnxt_re stuff so
> I'd like to just see it done.

In addition, push them to create 2 separate functions.
One is .create_cq_umem() for uverbs flow and another .create_cq()
variant for kernel flow.

bnxt_re_create_cq()
 { 
  if (udata)
     return bnxt_re_create_cq_umem()

  .... <kernel CQ>
  }

It will allow me to rebase my series more easily.

Thanks

> 
> Jason