On Thu, Jan 17, 2019 at 07:31:26PM +0000, Saleem, Shiraz wrote:
> >Subject: [PATCH for-next 1/1] IB/{hw,sw}: remove 'uobject->context' dependency in
> >object creation APIs
> >
> >Now when we have the udata passed to all the ib_xxx object creation APIs and the
> >additional function 'rdma_get_ucontext' to get the ib_ucontext from ib_udata stored
> >in uverbs_attr_bundle, we can finally start to remove the dependency of the drivers
> >in the ib_xxx->uobject->context.
> >
> >Signed-off-by: Shamir Rabinovitch <shamir.rabinovitch@xxxxxxxxxx>
> >---
> > drivers/infiniband/hw/bnxt_re/ib_verbs.c | 33 +++++--
> > drivers/infiniband/hw/cxgb3/iwch_provider.c | 4 +-
> > drivers/infiniband/hw/cxgb4/qp.c | 8 +-
> > drivers/infiniband/hw/hns/hns_roce_qp.c | 15 +++-
> > drivers/infiniband/hw/i40iw/i40iw_verbs.c | 15 +++-
> > drivers/infiniband/hw/mlx4/doorbell.c | 6 ++
> > drivers/infiniband/hw/mlx4/mr.c | 9 +-
> > drivers/infiniband/hw/mlx4/qp.c | 92 +++++++++++++-------
> > drivers/infiniband/hw/mlx4/srq.c | 13 ++-
> > drivers/infiniband/hw/mlx5/qp.c | 68 +++++++++++----
> > drivers/infiniband/hw/mlx5/srq.c | 13 ++-
> > drivers/infiniband/hw/mthca/mthca_provider.c | 28 ++++--
> > drivers/infiniband/hw/mthca/mthca_qp.c | 19 ++--
> > drivers/infiniband/hw/mthca/mthca_srq.c | 25 ++++--
> > drivers/infiniband/hw/nes/nes_verbs.c | 13 ++-
> > drivers/infiniband/hw/qedr/verbs.c | 8 +-
> > drivers/infiniband/hw/usnic/usnic_ib_verbs.c | 7 +-
> > drivers/infiniband/sw/rdmavt/qp.c | 10 ++-
> > drivers/infiniband/sw/rdmavt/srq.c | 10 ++-
> > drivers/infiniband/sw/rxe/rxe_qp.c | 5 +-
> > drivers/infiniband/sw/rxe/rxe_verbs.c | 5 +-
> > 21 files changed, 287 insertions(+), 119 deletions(-)
> >
> [....]
>
> >diff --git a/drivers/infiniband/hw/i40iw/i40iw_verbs.c
> >b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
> >index 12b31a8440be..194cd911c9de 100644
> >--- a/drivers/infiniband/hw/i40iw/i40iw_verbs.c
> >+++ b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
> >@@ -580,11 +580,16 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
> > struct i40iw_create_qp_info *qp_info;
> > struct i40iw_cqp_request *cqp_request;
> > struct cqp_commands_info *cqp_info;
> >+ struct ib_ucontext *ib_ucontext;
> >
> > struct i40iw_qp_host_ctx_info *ctx_info;
> > struct i40iwarp_offload_info *iwarp_info;
> > unsigned long flags;
> >
> >+ ib_ucontext = rdma_get_ucontext(udata);
> >+ if (udata && IS_ERR(ib_ucontext))
> >+ return ERR_CAST(ib_ucontext);
> >+
> > if (iwdev->closing)
> > return ERR_PTR(-ENODEV);
> >
> >@@ -674,7 +679,7 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
> > }
> > iwqp->ctx_info.qp_compl_ctx = req.user_compl_ctx;
> > iwqp->user_mode = 1;
> >- ucontext = to_ucontext(ibpd->uobject->context);
> >+ ucontext = to_ucontext(ib_ucontext);
> >
> > if (req.user_wqe_buffers) {
> > struct i40iw_pbl *iwpbl;
> >@@ -1832,6 +1837,7 @@ static struct ib_mr *i40iw_reg_user_mr(struct ib_pd *pd,
> > struct i40iw_pd *iwpd = to_iwpd(pd);
> > struct i40iw_device *iwdev = to_iwdev(pd->device);
> > struct i40iw_ucontext *ucontext;
> >+ struct ib_ucontext *ib_ucontext;
> > struct i40iw_pble_alloc *palloc;
> > struct i40iw_pbl *iwpbl;
> > struct i40iw_mr *iwmr;
> >@@ -1847,6 +1853,12 @@ static struct ib_mr *i40iw_reg_user_mr(struct ib_pd *pd,
> > int ret;
> > int pg_shift;
> >
> >+ ib_ucontext = rdma_get_ucontext(udata);
> >+ if (IS_ERR(ib_ucontext))
> >+ return ERR_CAST(ib_ucontext);
> >+
> >+ ucontext = to_ucontext(ib_ucontext);
> >+
> > if (iwdev->closing)
> > return ERR_PTR(-ENODEV);
> >
> >@@ -1872,7 +1884,6 @@ static struct ib_mr *i40iw_reg_user_mr(struct ib_pd *pd,
> > iwmr->region = region;
> > iwmr->ibmr.pd = pd;
> > iwmr->ibmr.device = pd->device;
> >- ucontext = to_ucontext(pd->uobject->context);
> >
> > iwmr->page_size = PAGE_SIZE;
> > iwmr->page_msk = PAGE_MASK;
> QP.
> > */
> > if (new_state == IB_QPS_RESET) {
> >- if (!ibuobject) {
> >+ if (!udata) {
> > mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
> > ibsrq ? to_msrq(ibsrq) : NULL);
> > if (send_cq != recv_cq)
>
> I think you missed one in i40iw_dereg_mr?
Thanks for the review Shiraz.
This patch (see description) only try to solve the problem with regard
to APIs that *create* objects. The reason is that Jason said in previous
iterations that the APIs that *destroy* objects should not assume the
existence of ib_udata. So the i40iw_dereg_mr will be handled in the next
patch and it will not have same solution as you see here.
>
> https://elixir.bootlin.com/linux/v5.0-rc2/source/drivers/infiniband/hw/i40iw/i40iw_verbs.c#L2097
>
> Shiraz
