On Tue, Jan 08, 2019 at 10:53:56AM +0200, Leon Romanovsky wrote:
> On Tue, Jan 08, 2019 at 03:24:54PM +0800, Lijun Ou wrote:
> > When flush cqe, it needs to get the pointer of rq and sq from
> > db address space of user and update it into qp context by
> > modified qp. if rq is not exist, it will not get the value
> > from db addresss space of user.
> >
> > Signed-off-by: Lijun Ou <oulijun@xxxxxxxxxx>
> > V1->V2:
> > 1. regenerate this patch in order to resolve the conflict
> > drivers/infiniband/hw/hns/hns_roce_qp.c | 16 ++++++++--------
> > 1 file changed, 8 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
> > index 54031c5..0d06bd8 100644
> > +++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
> > @@ -675,6 +675,10 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
> > dev_err(dev, "rq record doorbell map failed!\n");
> > goto err_sq_dbmap;
> > }
> > +
> > + /* indicate kernel supports rq record db */
> > + resp.cap_flags |= HNS_ROCE_SUPPORT_RQ_RECORD_DB;
> > + hr_qp->rdb_en = 1;
> > }
> > } else {
> > if (init_attr->create_flags &
> > @@ -783,16 +787,10 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
> > else
> > hr_qp->doorbell_qpn = cpu_to_le64(hr_qp->qpn);
> >
> > - if (udata && (udata->outlen >= sizeof(resp)) &&
> > - (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB)) {
> > -
> > - /* indicate kernel supports rq record db */
> > - resp.cap_flags |= HNS_ROCE_SUPPORT_RQ_RECORD_DB;
> > + if (udata && (udata->outlen >= sizeof(resp))) {
>
> I wonder if this check is correct, don't you suppose to do?
> "ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)));"
> and remove "(udata->outlen >= sizeof(resp))"
yes
It should also zero fill beyond, but all our drivers are broken this
way today.
Jason
