| Message ID | 20181002075212.20729-1-kamalheib1@gmail.com (mailing list archive) |
|---|---|
| State | Rejected |
| Headers | show |
| Series | [rdma-next] RDMA: Adding a new device managed gids cap | expand |
On Tue, Oct 02, 2018 at 10:52:12AM +0300, Kamal Heib wrote: > To avoid checking if add_gid() and del_gid() are implemented adding a new > device cap flag that will indicate whether a specific device does > support managing his RoCE GIDs table or not. > Why? What will it give us? Thanks
On Tue, Oct 02, 2018 at 11:07:38AM +0300, Leon Romanovsky wrote: > On Tue, Oct 02, 2018 at 10:52:12AM +0300, Kamal Heib wrote: > > To avoid checking if add_gid() and del_gid() are implemented adding a new > > device cap flag that will indicate whether a specific device does > > support managing his RoCE GIDs table or not. > > > > Why? What will it give us? > > Thanks I plan to work on something similar to "verbs_context_ops" that was introduced in rdma-core, so, this change will drop the dependency on checking if the providers implemented the call-backs or not. Thanks, Kamal
On Tue, Oct 02, 2018 at 11:39:43AM +0300, Kamal Heib wrote: > On Tue, Oct 02, 2018 at 11:07:38AM +0300, Leon Romanovsky wrote: > > On Tue, Oct 02, 2018 at 10:52:12AM +0300, Kamal Heib wrote: > > > To avoid checking if add_gid() and del_gid() are implemented adding a new > > > device cap flag that will indicate whether a specific device does > > > support managing his RoCE GIDs table or not. > > > > > > > Why? What will it give us? > > > > Thanks > > I plan to work on something similar to "verbs_context_ops" that was introduced in rdma-core, > so, this change will drop the dependency on checking if the providers implemented the call-backs or not. I would be happy to see this patch as part of whole feature, because this specific patch gives nothing to us in this current form. Thanks > > Thanks, > Kamal >
On 10/2/2018 4:54 AM, Leon Romanovsky wrote: > On Tue, Oct 02, 2018 at 11:39:43AM +0300, Kamal Heib wrote: >> On Tue, Oct 02, 2018 at 11:07:38AM +0300, Leon Romanovsky wrote: >>> On Tue, Oct 02, 2018 at 10:52:12AM +0300, Kamal Heib wrote: >>>> To avoid checking if add_gid() and del_gid() are implemented adding a new >>>> device cap flag that will indicate whether a specific device does >>>> support managing his RoCE GIDs table or not. >>>> >>> >>> Why? What will it give us? >>> >>> Thanks >> >> I plan to work on something similar to "verbs_context_ops" that was introduced in rdma-core, >> so, this change will drop the dependency on checking if the providers implemented the call-backs or not. > > I would be happy to see this patch as part of whole feature, because > this specific patch gives nothing to us in this current form. > > Thanks > >> >> Thanks, >> Kamal >> Agree with Leon here. Patch has to stand on it's own, or at least come along with something that makes it relevant. -Denny
On Wed, Oct 03, 2018 at 11:29:32AM -0400, Dennis Dalessandro wrote: > On 10/2/2018 4:54 AM, Leon Romanovsky wrote: > > On Tue, Oct 02, 2018 at 11:39:43AM +0300, Kamal Heib wrote: > > > On Tue, Oct 02, 2018 at 11:07:38AM +0300, Leon Romanovsky wrote: > > > > On Tue, Oct 02, 2018 at 10:52:12AM +0300, Kamal Heib wrote: > > > > > To avoid checking if add_gid() and del_gid() are implemented adding a new > > > > > device cap flag that will indicate whether a specific device does > > > > > support managing his RoCE GIDs table or not. > > > > > > > > > > > > > Why? What will it give us? > > > > > > > > Thanks > > > > > > I plan to work on something similar to "verbs_context_ops" that > > > was introduced in rdma-core, so, this change will drop the > > > dependency on checking if the providers implemented the > > > call-backs or not. > > > > I would be happy to see this patch as part of whole feature, because > > this specific patch gives nothing to us in this current form. > > > Agree with Leon here. Patch has to stand on it's own, or at least come along > with something that makes it relevant. I also generally dislike the idea to test a bit to check if the function you are about to invoke is NULL or not. ops are to be NULL if they are not supported. Jason
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index bc2b9e038439..50d77ca23d76 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -165,7 +165,8 @@ int bnxt_re_query_device(struct ib_device *ibdev, | IB_DEVICE_N_NOTIFY_CQ | IB_DEVICE_MEM_WINDOW | IB_DEVICE_MEM_WINDOW_TYPE_2B - | IB_DEVICE_MEM_MGT_EXTENSIONS; + | IB_DEVICE_MEM_MGT_EXTENSIONS + | IB_DEVICE_MANAGED_GIDS; ib_attr->max_send_sge = dev_attr->max_qp_sges; ib_attr->max_recv_sge = dev_attr->max_qp_sges; ib_attr->max_sge_rd = dev_attr->max_qp_sges; diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c index 8c5160ec3a4d..63d64e1166a5 100644 --- a/drivers/infiniband/hw/hns/hns_roce_main.c +++ b/drivers/infiniband/hw/hns/hns_roce_main.c @@ -205,7 +205,8 @@ static int hns_roce_query_device(struct ib_device *ib_dev, props->max_qp = hr_dev->caps.num_qps; props->max_qp_wr = hr_dev->caps.max_wqes; props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT | - IB_DEVICE_RC_RNR_NAK_GEN; + IB_DEVICE_RC_RNR_NAK_GEN | + IB_DEVICE_MANAGED_GIDS; props->max_send_sge = hr_dev->caps.max_sq_sg; props->max_recv_sge = hr_dev->caps.max_rq_sg; props->max_sge_rd = 1; diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index fa5d20eccc21..546028129fef 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -471,7 +471,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev, IB_DEVICE_PORT_ACTIVE_EVENT | IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_RC_RNR_NAK_GEN | - IB_DEVICE_BLOCK_MULTICAST_LOOPBACK; + IB_DEVICE_BLOCK_MULTICAST_LOOPBACK | + IB_DEVICE_MANAGED_GIDS; if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR) props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR; if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR) diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 597cd3c171c9..814a17e229bf 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -765,7 +765,8 @@ static int mlx5_ib_query_device(struct ib_device *ibdev, props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT | IB_DEVICE_PORT_ACTIVE_EVENT | IB_DEVICE_SYS_IMAGE_GUID | - IB_DEVICE_RC_RNR_NAK_GEN; + IB_DEVICE_RC_RNR_NAK_GEN | + IB_DEVICE_MANAGED_GIDS; if (MLX5_CAP_GEN(mdev, pkv)) props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR; diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c index b65d10b0a875..792edfb796ad 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c @@ -113,7 +113,8 @@ int pvrdma_query_device(struct ib_device *ibdev, } props->device_cap_flags |= IB_DEVICE_PORT_ACTIVE_EVENT | - IB_DEVICE_RC_RNR_NAK_GEN; + IB_DEVICE_RC_RNR_NAK_GEN | + IB_DEVICE_MANAGED_GIDS; return 0; } diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 9897d2329f2c..402320958172 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -241,6 +241,7 @@ enum ib_device_cap_flags { IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), /* The device supports padding incoming writes to cacheline. */ IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), + IB_DEVICE_MANAGED_GIDS = (1ULL << 37), }; enum ib_signature_prot_cap { @@ -3094,7 +3095,7 @@ static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, u8 port_num) { return rdma_protocol_roce(device, port_num) && - device->add_gid && device->del_gid; + (device->attrs.device_cap_flags & IB_DEVICE_MANAGED_GIDS); } /*
To avoid checking if add_gid() and del_gid() are implemented adding a new device cap flag that will indicate whether a specific device does support managing his RoCE GIDs table or not. Signed-off-by: Kamal Heib <kamalheib1@gmail.com> --- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 3 ++- drivers/infiniband/hw/hns/hns_roce_main.c | 3 ++- drivers/infiniband/hw/mlx4/main.c | 3 ++- drivers/infiniband/hw/mlx5/main.c | 3 ++- drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c | 3 ++- include/rdma/ib_verbs.h | 3 ++- 6 files changed, 12 insertions(+), 6 deletions(-)