On Thu, Nov 01, 2018 at 11:03:25AM -0500, Shiraz Saleem wrote:
> On Wed, Oct 31, 2018 at 02:42:41PM +0200, Shamir Rabinovitch wrote:
> > Next patch will replace usage of pd->uobject with rdma_is_user_pd
> > function. As suggested by Jason, some of the code was miss using
> > the pd->uobject pointer in places it should have not been used and
> > better alternative exists. Fix this just before we do the replace.
> >
> > Signed-off-by: Shamir Rabinovitch <shamir.rabinovitch@xxxxxxxxxx>
> > drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 2 +-
> > drivers/infiniband/hw/i40iw/i40iw_verbs.c | 42 ++++++++++++++----------------
> > drivers/infiniband/hw/mlx4/qp.c | 6 ++---
> > drivers/infiniband/hw/mlx5/qp.c | 4 +--
> > 4 files changed, 26 insertions(+), 28 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/i40iw/i40iw_verbs.c b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
> > index 1028758..9d3178d 100644
> > +++ b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
> > @@ -673,28 +673,26 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
> > goto error;
> > }
> > iwqp->ctx_info.qp_compl_ctx = req.user_compl_ctx;
> > - if (ibpd->uobject && ibpd->uobject->context) {
>
> if (udata) sufficiently guarantees ucontext?
Yes, the core code can't invoke these methods with a !NULL udata
against a PD that is not also a user PD.
> > - iwqp->user_mode = 1;
> > - ucontext = to_ucontext(ibpd->uobject->context);
> > -
> [..]
>
> > + iwqp->user_mode = 1;
> > + ucontext = to_ucontext(ibpd->uobject->context);
>
> Sorry I am confused. If your intent was to remove pd->uobject usage, why isnt this replaced?
The next patch in the series replaces all the ibpd->uobject->context
stuff with something like get_ucontext_from_udata(udata)
Jason
