From: Artemy Kovalyov <artemyko@xxxxxxxxxxxx>
Handle ODP atomic operations. When initiator of RDMA atomic
operation use ODP MR to provide source data handle pagefault properly.
Signed-off-by: Artemy Kovalyov <artemyko@xxxxxxxxxxxx>
Signed-off-by: Leon Romanovsky <leon@xxxxxxxxxx>
Signed-off-by: Saeed Mahameed <saeedm@xxxxxxxxxxxx>
---
drivers/infiniband/hw/mlx5/odp.c | 88 +++++++++++++++++++++++-----------------
include/linux/mlx5/mlx5_ifc.h | 2 +-
include/linux/mlx5/qp.h | 18 ++++++++
3 files changed, 69 insertions(+), 39 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 26f96c7..971b288 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -144,6 +144,9 @@ void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.read))
caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_READ;
+ if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.atomic))
+ caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC;
+
return;
}
@@ -386,6 +389,17 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev,
return ret < 0 ? ret : npages;
}
+static const u32 mlx5_ib_odp_opcode_cap[] = {
+ [MLX5_OPCODE_SEND] = IB_ODP_SUPPORT_SEND,
+ [MLX5_OPCODE_SEND_IMM] = IB_ODP_SUPPORT_SEND,
+ [MLX5_OPCODE_SEND_INVAL] = IB_ODP_SUPPORT_SEND,
+ [MLX5_OPCODE_RDMA_WRITE] = IB_ODP_SUPPORT_WRITE,
+ [MLX5_OPCODE_RDMA_WRITE_IMM] = IB_ODP_SUPPORT_WRITE,
+ [MLX5_OPCODE_RDMA_READ] = IB_ODP_SUPPORT_READ,
+ [MLX5_OPCODE_ATOMIC_CS] = IB_ODP_SUPPORT_ATOMIC,
+ [MLX5_OPCODE_ATOMIC_FA] = IB_ODP_SUPPORT_ATOMIC,
+};
+
/*
* Parse initiator WQE. Advances the wqe pointer to point at the
* scatter-gather list, and set wqe_end to the end of the WQE.
@@ -396,6 +410,8 @@ static int mlx5_ib_mr_initiator_pfault_handler(
{
struct mlx5_wqe_ctrl_seg *ctrl = *wqe;
u16 wqe_index = pfault->wqe.wqe_index;
+ u32 transport_caps;
+ struct mlx5_base_av *av;
unsigned ds, opcode;
#if defined(DEBUG)
u32 ctrl_wqe_index, ctrl_qpn;
@@ -441,53 +457,49 @@ static int mlx5_ib_mr_initiator_pfault_handler(
opcode = be32_to_cpu(ctrl->opmod_idx_opcode) &
MLX5_WQE_CTRL_OPCODE_MASK;
+
switch (qp->ibqp.qp_type) {
case IB_QPT_RC:
- switch (opcode) {
- case MLX5_OPCODE_SEND:
- case MLX5_OPCODE_SEND_IMM:
- case MLX5_OPCODE_SEND_INVAL:
- if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
- IB_ODP_SUPPORT_SEND))
- goto invalid_transport_or_opcode;
- break;
- case MLX5_OPCODE_RDMA_WRITE:
- case MLX5_OPCODE_RDMA_WRITE_IMM:
- if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
- IB_ODP_SUPPORT_WRITE))
- goto invalid_transport_or_opcode;
- *wqe += sizeof(struct mlx5_wqe_raddr_seg);
- break;
- case MLX5_OPCODE_RDMA_READ:
- if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
- IB_ODP_SUPPORT_READ))
- goto invalid_transport_or_opcode;
- *wqe += sizeof(struct mlx5_wqe_raddr_seg);
- break;
- default:
- goto invalid_transport_or_opcode;
- }
+ transport_caps = dev->odp_caps.per_transport_caps.rc_odp_caps;
break;
case IB_QPT_UD:
- switch (opcode) {
- case MLX5_OPCODE_SEND:
- case MLX5_OPCODE_SEND_IMM:
- if (!(dev->odp_caps.per_transport_caps.ud_odp_caps &
- IB_ODP_SUPPORT_SEND))
- goto invalid_transport_or_opcode;
- *wqe += sizeof(struct mlx5_wqe_datagram_seg);
- break;
- default:
- goto invalid_transport_or_opcode;
- }
+ transport_caps = dev->odp_caps.per_transport_caps.ud_odp_caps;
break;
default:
-invalid_transport_or_opcode:
- mlx5_ib_err(dev, "ODP fault on QP of an unsupported opcode or transport. transport: 0x%x opcode: 0x%x.\n",
- qp->ibqp.qp_type, opcode);
+ mlx5_ib_err(dev, "ODP fault on QP of an unsupported transport 0x%x\n",
+ qp->ibqp.qp_type);
+ return -EFAULT;
+ }
+
+ if (unlikely(opcode >= sizeof(mlx5_ib_odp_opcode_cap) /
+ sizeof(mlx5_ib_odp_opcode_cap[0]) ||
+ !(transport_caps & mlx5_ib_odp_opcode_cap[opcode]))) {
+ mlx5_ib_err(dev, "ODP fault on QP of an unsupported opcode 0x%x\n",
+ opcode);
return -EFAULT;
}
+ if (qp->ibqp.qp_type != IB_QPT_RC) {
+ av = *wqe;
+ if (av->dqp_dct & be32_to_cpu(MLX5_WQE_AV_EXT))
+ *wqe += sizeof(struct mlx5_av);
+ else
+ *wqe += sizeof(struct mlx5_base_av);
+ }
+
+ switch (opcode) {
+ case MLX5_OPCODE_RDMA_WRITE:
+ case MLX5_OPCODE_RDMA_WRITE_IMM:
+ case MLX5_OPCODE_RDMA_READ:
+ *wqe += sizeof(struct mlx5_wqe_raddr_seg);
+ break;
+ case MLX5_OPCODE_ATOMIC_CS:
+ case MLX5_OPCODE_ATOMIC_FA:
+ *wqe += sizeof(struct mlx5_wqe_raddr_seg);
+ *wqe += sizeof(struct mlx5_wqe_atomic_seg);
+ break;
+ }
+
return 0;
}
diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index 608dc98..15f8967 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -328,7 +328,7 @@ struct mlx5_ifc_odp_per_transport_service_cap_bits {
u8 receive[0x1];
u8 write[0x1];
u8 read[0x1];
- u8 reserved_at_4[0x1];
+ u8 atomic[0x1];
u8 srq_receive[0x1];
u8 reserved_at_6[0x1a];
};
diff --git a/include/linux/mlx5/qp.h b/include/linux/mlx5/qp.h
index 9ed775f..219c699 100644
--- a/include/linux/mlx5/qp.h
+++ b/include/linux/mlx5/qp.h
@@ -212,6 +212,7 @@ struct mlx5_wqe_ctrl_seg {
#define MLX5_WQE_CTRL_OPCODE_MASK 0xff
#define MLX5_WQE_CTRL_WQE_INDEX_MASK 0x00ffff00
#define MLX5_WQE_CTRL_WQE_INDEX_SHIFT 8
+#define MLX5_WQE_AV_EXT 0x80000000
enum {
MLX5_ETH_WQE_L3_INNER_CSUM = 1 << 4,
@@ -242,6 +243,23 @@ struct mlx5_wqe_masked_atomic_seg {
__be64 compare_mask;
};
+struct mlx5_base_av {
+ union {
+ struct {
+ __be32 qkey;
+ __be32 reserved;
+ } qkey;
+ __be64 dc_key;
+ } key;
+ __be32 dqp_dct;
+ u8 stat_rate_sl;
+ u8 fl_mlid;
+ union {
+ __be16 rlid;
+ __be16 udp_sport;
+ };
+};
+
struct mlx5_av {
union {
struct {
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
