diff options
Diffstat (limited to 'drivers/infiniband/hw/mlx5')
30 files changed, 3416 insertions, 920 deletions
diff --git a/drivers/infiniband/hw/mlx5/Makefile b/drivers/infiniband/hw/mlx5/Makefile index 72a526236c2e..11878ddf7cc7 100644 --- a/drivers/infiniband/hw/mlx5/Makefile +++ b/drivers/infiniband/hw/mlx5/Makefile @@ -6,8 +6,10 @@ mlx5_ib-y := ah.o \ cong.o \ counters.o \ cq.o \ + data_direct.o \ dm.o \ doorbell.o \ + fs.o \ gsi.o \ ib_virt.o \ mad.o \ @@ -25,7 +27,6 @@ mlx5_ib-y := ah.o \ mlx5_ib-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += odp.o mlx5_ib-$(CONFIG_MLX5_ESWITCH) += ib_rep.o mlx5_ib-$(CONFIG_INFINIBAND_USER_ACCESS) += devx.o \ - fs.o \ qos.o \ std_types.o mlx5_ib-$(CONFIG_MLX5_MACSEC) += macsec.o diff --git a/drivers/infiniband/hw/mlx5/ah.c b/drivers/infiniband/hw/mlx5/ah.c index 505bc47fd575..531a57f9ee7e 100644 --- a/drivers/infiniband/hw/mlx5/ah.c +++ b/drivers/infiniband/hw/mlx5/ah.c @@ -50,11 +50,12 @@ static __be16 mlx5_ah_get_udp_sport(const struct mlx5_ib_dev *dev, return sport; } -static void create_ib_ah(struct mlx5_ib_dev *dev, struct mlx5_ib_ah *ah, +static int create_ib_ah(struct mlx5_ib_dev *dev, struct mlx5_ib_ah *ah, struct rdma_ah_init_attr *init_attr) { struct rdma_ah_attr *ah_attr = init_attr->ah_attr; enum ib_gid_type gid_type; + int rate_val; if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) { const struct ib_global_route *grh = rdma_ah_read_grh(ah_attr); @@ -67,7 +68,10 @@ static void create_ib_ah(struct mlx5_ib_dev *dev, struct mlx5_ib_ah *ah, ah->av.tclass = grh->traffic_class; } - ah->av.stat_rate_sl = (rdma_ah_get_static_rate(ah_attr) << 4); + rate_val = mlx5r_ib_rate(dev, rdma_ah_get_static_rate(ah_attr)); + if (rate_val < 0) + return rate_val; + ah->av.stat_rate_sl = rate_val << 4; if (ah_attr->type == RDMA_AH_ATTR_TYPE_ROCE) { if (init_attr->xmit_slave) @@ -88,6 +92,8 @@ static void create_ib_ah(struct mlx5_ib_dev *dev, struct mlx5_ib_ah *ah, ah->av.fl_mlid = rdma_ah_get_path_bits(ah_attr) & 0x7f; ah->av.stat_rate_sl |= (rdma_ah_get_sl(ah_attr) & 0xf); } + + return 0; } int mlx5_ib_create_ah(struct ib_ah *ibah, struct rdma_ah_init_attr *init_attr, @@ -120,8 +126,7 @@ int mlx5_ib_create_ah(struct ib_ah *ibah, struct rdma_ah_init_attr *init_attr, return err; } - create_ib_ah(dev, ah, init_attr); - return 0; + return create_ib_ah(dev, ah, init_attr); } int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr) diff --git a/drivers/infiniband/hw/mlx5/cmd.c b/drivers/infiniband/hw/mlx5/cmd.c index 1d0c8d5e745b..7c08e3008927 100644 --- a/drivers/infiniband/hw/mlx5/cmd.c +++ b/drivers/infiniband/hw/mlx5/cmd.c @@ -177,7 +177,7 @@ int mlx5_cmd_xrcd_dealloc(struct mlx5_core_dev *dev, u32 xrcdn, u16 uid) return mlx5_cmd_exec_in(dev, dealloc_xrcd, in); } -int mlx5_cmd_mad_ifc(struct mlx5_core_dev *dev, const void *inb, void *outb, +int mlx5_cmd_mad_ifc(struct mlx5_ib_dev *dev, const void *inb, void *outb, u16 opmod, u8 port) { int outlen = MLX5_ST_SZ_BYTES(mad_ifc_out); @@ -195,12 +195,18 @@ int mlx5_cmd_mad_ifc(struct mlx5_core_dev *dev, const void *inb, void *outb, MLX5_SET(mad_ifc_in, in, opcode, MLX5_CMD_OP_MAD_IFC); MLX5_SET(mad_ifc_in, in, op_mod, opmod); - MLX5_SET(mad_ifc_in, in, port, port); + if (dev->ib_dev.type == RDMA_DEVICE_TYPE_SMI) { + MLX5_SET(mad_ifc_in, in, plane_index, port); + MLX5_SET(mad_ifc_in, in, port, + smi_to_native_portnum(dev, port)); + } else { + MLX5_SET(mad_ifc_in, in, port, port); + } data = MLX5_ADDR_OF(mad_ifc_in, in, mad); memcpy(data, inb, MLX5_FLD_SZ_BYTES(mad_ifc_in, mad)); - err = mlx5_cmd_exec_inout(dev, mad_ifc, in, out); + err = mlx5_cmd_exec_inout(dev->mdev, mad_ifc, in, out); if (err) goto out; @@ -239,3 +245,24 @@ int mlx5_cmd_uar_dealloc(struct mlx5_core_dev *dev, u32 uarn, u16 uid) MLX5_SET(dealloc_uar_in, in, uid, uid); return mlx5_cmd_exec_in(dev, dealloc_uar, in); } + +int mlx5_cmd_query_vuid(struct mlx5_core_dev *dev, bool data_direct, + char *out_vuid) +{ + u8 out[MLX5_ST_SZ_BYTES(query_vuid_out) + + MLX5_ST_SZ_BYTES(array1024_auto)] = {}; + u8 in[MLX5_ST_SZ_BYTES(query_vuid_in)] = {}; + char *vuid; + int err; + + MLX5_SET(query_vuid_in, in, opcode, MLX5_CMD_OPCODE_QUERY_VUID); + MLX5_SET(query_vuid_in, in, vhca_id, MLX5_CAP_GEN(dev, vhca_id)); + MLX5_SET(query_vuid_in, in, data_direct, data_direct); + err = mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out)); + if (err) + return err; + + vuid = MLX5_ADDR_OF(query_vuid_out, out, vuid); + memcpy(out_vuid, vuid, MLX5_ST_SZ_BYTES(array1024_auto)); + return 0; +} diff --git a/drivers/infiniband/hw/mlx5/cmd.h b/drivers/infiniband/hw/mlx5/cmd.h index 93a971a40d11..e6c88b6ebd0d 100644 --- a/drivers/infiniband/hw/mlx5/cmd.h +++ b/drivers/infiniband/hw/mlx5/cmd.h @@ -54,8 +54,10 @@ int mlx5_cmd_detach_mcg(struct mlx5_core_dev *dev, union ib_gid *mgid, u32 qpn, u16 uid); int mlx5_cmd_xrcd_alloc(struct mlx5_core_dev *dev, u32 *xrcdn, u16 uid); int mlx5_cmd_xrcd_dealloc(struct mlx5_core_dev *dev, u32 xrcdn, u16 uid); -int mlx5_cmd_mad_ifc(struct mlx5_core_dev *dev, const void *inb, void *outb, +int mlx5_cmd_mad_ifc(struct mlx5_ib_dev *dev, const void *inb, void *outb, u16 opmod, u8 port); int mlx5_cmd_uar_alloc(struct mlx5_core_dev *dev, u32 *uarn, u16 uid); int mlx5_cmd_uar_dealloc(struct mlx5_core_dev *dev, u32 uarn, u16 uid); +int mlx5_cmd_query_vuid(struct mlx5_core_dev *dev, bool data_direct, + char *out_vuid); #endif /* MLX5_IB_CMD_H */ diff --git a/drivers/infiniband/hw/mlx5/cong.c b/drivers/infiniband/hw/mlx5/cong.c index f87531318feb..a78a067e3ce7 100644 --- a/drivers/infiniband/hw/mlx5/cong.c +++ b/drivers/infiniband/hw/mlx5/cong.c @@ -458,6 +458,12 @@ void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num) dbg_cc_params->root = debugfs_create_dir("cc_params", mlx5_debugfs_get_dev_root(mdev)); for (i = 0; i < MLX5_IB_DBG_CC_MAX; i++) { + if ((i == MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID || + i == MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP)) + if (!MLX5_CAP_GEN(mdev, roce) || + !MLX5_CAP_ROCE(mdev, roce_cc_general)) + continue; + dbg_cc_params->params[i].offset = i; dbg_cc_params->params[i].dev = dev; dbg_cc_params->params[i].port_num = port_num; diff --git a/drivers/infiniband/hw/mlx5/counters.c b/drivers/infiniband/hw/mlx5/counters.c index 8300ce622835..b847084dcd99 100644 --- a/drivers/infiniband/hw/mlx5/counters.c +++ b/drivers/infiniband/hw/mlx5/counters.c @@ -83,6 +83,8 @@ static const struct mlx5_ib_counter extended_err_cnts[] = { INIT_Q_COUNTER(resp_remote_access_errors), INIT_Q_COUNTER(resp_cqe_flush_error), INIT_Q_COUNTER(req_cqe_flush_error), + INIT_Q_COUNTER(req_transport_retries_exceeded), + INIT_Q_COUNTER(req_rnr_retries_exceeded), }; static const struct mlx5_ib_counter roce_accl_cnts[] = { @@ -102,6 +104,8 @@ static const struct mlx5_ib_counter vport_extended_err_cnts[] = { INIT_VPORT_Q_COUNTER(resp_remote_access_errors), INIT_VPORT_Q_COUNTER(resp_cqe_flush_error), INIT_VPORT_Q_COUNTER(req_cqe_flush_error), + INIT_VPORT_Q_COUNTER(req_transport_retries_exceeded), + INIT_VPORT_Q_COUNTER(req_rnr_retries_exceeded), }; static const struct mlx5_ib_counter vport_roce_accl_cnts[] = { @@ -136,6 +140,13 @@ static const struct mlx5_ib_counter rdmatx_cnp_op_cnts[] = { INIT_OP_COUNTER(cc_tx_cnp_pkts, CC_TX_CNP_PKTS), }; +static const struct mlx5_ib_counter packets_op_cnts[] = { + INIT_OP_COUNTER(rdma_tx_packets, RDMA_TX_PACKETS), + INIT_OP_COUNTER(rdma_tx_bytes, RDMA_TX_BYTES), + INIT_OP_COUNTER(rdma_rx_packets, RDMA_RX_PACKETS), + INIT_OP_COUNTER(rdma_rx_bytes, RDMA_RX_BYTES), +}; + static int mlx5_ib_read_counters(struct ib_counters *counters, struct ib_counters_read_attr *read_attr, struct uverbs_attr_bundle *attrs) @@ -423,6 +434,52 @@ done: return num_counters; } +static bool is_rdma_bytes_counter(u32 type) +{ + if (type == MLX5_IB_OPCOUNTER_RDMA_TX_BYTES || + type == MLX5_IB_OPCOUNTER_RDMA_RX_BYTES || + type == MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP || + type == MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP) + return true; + + return false; +} + +static int do_per_qp_get_op_stat(struct rdma_counter *counter) +{ + struct mlx5_ib_dev *dev = to_mdev(counter->device); + const struct mlx5_ib_counters *cnts = get_counters(dev, counter->port); + struct mlx5_rdma_counter *mcounter = to_mcounter(counter); + int i, ret, index, num_hw_counters; + u64 packets = 0, bytes = 0; + + for (i = MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP; + i <= MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; i++) { + if (!mcounter->fc[i]) + continue; + + ret = mlx5_fc_query(dev->mdev, mcounter->fc[i], + &packets, &bytes); + if (ret) + return ret; + + num_hw_counters = cnts->num_q_counters + + cnts->num_cong_counters + + cnts->num_ext_ppcnt_counters; + + index = i - MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP + + num_hw_counters; + + if (is_rdma_bytes_counter(i)) + counter->stats->value[index] = bytes; + else + counter->stats->value[index] = packets; + + clear_bit(index, counter->stats->is_disabled); + } + return 0; +} + static int do_get_op_stat(struct ib_device *ibdev, struct rdma_hw_stats *stats, u32 port_num, int index) @@ -430,7 +487,7 @@ static int do_get_op_stat(struct ib_device *ibdev, struct mlx5_ib_dev *dev = to_mdev(ibdev); const struct mlx5_ib_counters *cnts; const struct mlx5_ib_op_fc *opfcs; - u64 packets = 0, bytes; + u64 packets, bytes; u32 type; int ret; @@ -449,8 +506,11 @@ static int do_get_op_stat(struct ib_device *ibdev, if (ret) return ret; + if (is_rdma_bytes_counter(type)) + stats->value[index] = bytes; + else + stats->value[index] = packets; out: - stats->value[index] = packets; return index; } @@ -519,19 +579,30 @@ static int mlx5_ib_counter_update_stats(struct rdma_counter *counter) { struct mlx5_ib_dev *dev = to_mdev(counter->device); const struct mlx5_ib_counters *cnts = get_counters(dev, counter->port); + int ret; + + ret = mlx5_ib_query_q_counters(dev->mdev, cnts, counter->stats, + counter->id); + if (ret) + return ret; + + if (!counter->mode.bind_opcnt) + return 0; - return mlx5_ib_query_q_counters(dev->mdev, cnts, - counter->stats, counter->id); + return do_per_qp_get_op_stat(counter); } static int mlx5_ib_counter_dealloc(struct rdma_counter *counter) { + struct mlx5_rdma_counter *mcounter = to_mcounter(counter); struct mlx5_ib_dev *dev = to_mdev(counter->device); u32 in[MLX5_ST_SZ_DW(dealloc_q_counter_in)] = {}; if (!counter->id) return 0; + WARN_ON(!xa_empty(&mcounter->qpn_opfc_xa)); + mlx5r_fs_destroy_fcs(dev, counter); MLX5_SET(dealloc_q_counter_in, in, opcode, MLX5_CMD_OP_DEALLOC_Q_COUNTER); MLX5_SET(dealloc_q_counter_in, in, counter_set_id, counter->id); @@ -539,9 +610,10 @@ static int mlx5_ib_counter_dealloc(struct rdma_counter *counter) } static int mlx5_ib_counter_bind_qp(struct rdma_counter *counter, - struct ib_qp *qp) + struct ib_qp *qp, u32 port) { struct mlx5_ib_dev *dev = to_mdev(qp->device); + bool new = false; int err; if (!counter->id) { @@ -556,24 +628,46 @@ static int mlx5_ib_counter_bind_qp(struct rdma_counter *counter, return err; counter->id = MLX5_GET(alloc_q_counter_out, out, counter_set_id); + new = true; } err = mlx5_ib_qp_set_counter(qp, counter); if (err) goto fail_set_counter; + err = mlx5r_fs_bind_op_fc(qp, counter, port); + if (err) + goto fail_bind_op_fc; + return 0; +fail_bind_op_fc: + mlx5_ib_qp_set_counter(qp, NULL); fail_set_counter: - mlx5_ib_counter_dealloc(counter); - counter->id = 0; + if (new) { + mlx5_ib_counter_dealloc(counter); + counter->id = 0; + } return err; } -static int mlx5_ib_counter_unbind_qp(struct ib_qp *qp) +static int mlx5_ib_counter_unbind_qp(struct ib_qp *qp, u32 port) { - return mlx5_ib_qp_set_counter(qp, NULL); + struct rdma_counter *counter = qp->counter; + int err; + + mlx5r_fs_unbind_op_fc(qp, counter); + + err = mlx5_ib_qp_set_counter(qp, NULL); + if (err) + goto fail_set_counter; + + return 0; + +fail_set_counter: + mlx5r_fs_bind_op_fc(qp, counter, port); + return err; } static void mlx5_ib_fill_counters(struct mlx5_ib_dev *dev, @@ -673,6 +767,12 @@ static void mlx5_ib_fill_counters(struct mlx5_ib_dev *dev, descs[j].priv = &rdmatx_cnp_op_cnts[i].type; } } + + for (i = 0; i < ARRAY_SIZE(packets_op_cnts); i++, j++) { + descs[j].name = packets_op_cnts[i].name; + descs[j].flags |= IB_STAT_FLAG_OPTIONAL; + descs[j].priv = &packets_op_cnts[i].type; + } } @@ -723,6 +823,8 @@ static int __mlx5_ib_alloc_counters(struct mlx5_ib_dev *dev, num_op_counters = ARRAY_SIZE(basic_op_cnts); + num_op_counters += ARRAY_SIZE(packets_op_cnts); + if (MLX5_CAP_FLOWTABLE(dev->mdev, ft_field_support_2_nic_receive_rdma.bth_opcode)) num_op_counters += ARRAY_SIZE(rdmarx_cnp_op_cnts); @@ -752,10 +854,58 @@ err: return -ENOMEM; } +/* + * Checks if the given flow counter type should be sharing the same flow counter + * with another type and if it should, checks if that other type flow counter + * was already created, if both conditions are met return true and the counter + * else return false. + */ +bool mlx5r_is_opfc_shared_and_in_use(struct mlx5_ib_op_fc *opfcs, u32 type, + struct mlx5_ib_op_fc **opfc) +{ + u32 shared_fc_type; + + switch (type) { + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_TX_BYTES; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_RX_BYTES; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP; + break; + default: + return false; + } + + *opfc = &opfcs[shared_fc_type]; + if (!(*opfc)->fc) + return false; + + return true; +} + static void mlx5_ib_dealloc_counters(struct mlx5_ib_dev *dev) { u32 in[MLX5_ST_SZ_DW(dealloc_q_counter_in)] = {}; int num_cnt_ports = dev->num_ports; + struct mlx5_ib_op_fc *in_use_opfc; int i, j; if (is_mdev_switchdev_mode(dev->mdev)) @@ -777,11 +927,15 @@ static void mlx5_ib_dealloc_counters(struct mlx5_ib_dev *dev) if (!dev->port[i].cnts.opfcs[j].fc) continue; - if (IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)) - mlx5_ib_fs_remove_op_fc(dev, - &dev->port[i].cnts.opfcs[j], j); + if (mlx5r_is_opfc_shared_and_in_use( + dev->port[i].cnts.opfcs, j, &in_use_opfc)) + goto skip; + + mlx5_ib_fs_remove_op_fc(dev, + &dev->port[i].cnts.opfcs[j], j); mlx5_fc_destroy(dev->mdev, dev->port[i].cnts.opfcs[j].fc); +skip: dev->port[i].cnts.opfcs[j].fc = NULL; } } @@ -975,8 +1129,8 @@ static int mlx5_ib_modify_stat(struct ib_device *device, u32 port, unsigned int index, bool enable) { struct mlx5_ib_dev *dev = to_mdev(device); + struct mlx5_ib_op_fc *opfc, *in_use_opfc; struct mlx5_ib_counters *cnts; - struct mlx5_ib_op_fc *opfc; u32 num_hw_counters, type; int ret; @@ -1000,6 +1154,13 @@ static int mlx5_ib_modify_stat(struct ib_device *device, u32 port, if (opfc->fc) return -EEXIST; + if (mlx5r_is_opfc_shared_and_in_use(cnts->opfcs, type, + &in_use_opfc)) { + opfc->fc = in_use_opfc->fc; + opfc->rule[0] = in_use_opfc->rule[0]; + return 0; + } + opfc->fc = mlx5_fc_create(dev->mdev, false); if (IS_ERR(opfc->fc)) return PTR_ERR(opfc->fc); @@ -1015,12 +1176,23 @@ static int mlx5_ib_modify_stat(struct ib_device *device, u32 port, if (!opfc->fc) return -EINVAL; + if (mlx5r_is_opfc_shared_and_in_use(cnts->opfcs, type, &in_use_opfc)) + goto out; + mlx5_ib_fs_remove_op_fc(dev, opfc, type); mlx5_fc_destroy(dev->mdev, opfc->fc); +out: opfc->fc = NULL; return 0; } +static void mlx5_ib_counter_init(struct rdma_counter *counter) +{ + struct mlx5_rdma_counter *mcounter = to_mcounter(counter); + + xa_init(&mcounter->qpn_opfc_xa); +} + static const struct ib_device_ops hw_stats_ops = { .alloc_hw_port_stats = mlx5_ib_alloc_hw_port_stats, .get_hw_stats = mlx5_ib_get_hw_stats, @@ -1029,8 +1201,10 @@ static const struct ib_device_ops hw_stats_ops = { .counter_dealloc = mlx5_ib_counter_dealloc, .counter_alloc_stats = mlx5_ib_counter_alloc_stats, .counter_update_stats = mlx5_ib_counter_update_stats, - .modify_hw_stat = IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) ? - mlx5_ib_modify_stat : NULL, + .modify_hw_stat = mlx5_ib_modify_stat, + .counter_init = mlx5_ib_counter_init, + + INIT_RDMA_OBJ_SIZE(rdma_counter, mlx5_rdma_counter, rdma_counter), }; static const struct ib_device_ops hw_switchdev_vport_op = { @@ -1045,6 +1219,9 @@ static const struct ib_device_ops hw_switchdev_stats_ops = { .counter_dealloc = mlx5_ib_counter_dealloc, .counter_alloc_stats = mlx5_ib_counter_alloc_stats, .counter_update_stats = mlx5_ib_counter_update_stats, + .counter_init = mlx5_ib_counter_init, + + INIT_RDMA_OBJ_SIZE(rdma_counter, mlx5_rdma_counter, rdma_counter), }; static const struct ib_device_ops counters_ops = { diff --git a/drivers/infiniband/hw/mlx5/counters.h b/drivers/infiniband/hw/mlx5/counters.h index 6bcaaa52e2b2..bd03cee42014 100644 --- a/drivers/infiniband/hw/mlx5/counters.h +++ b/drivers/infiniband/hw/mlx5/counters.h @@ -8,10 +8,25 @@ #include "mlx5_ib.h" +struct mlx5_rdma_counter { + struct rdma_counter rdma_counter; + + struct mlx5_fc *fc[MLX5_IB_OPCOUNTER_MAX]; + struct xarray qpn_opfc_xa; +}; + +static inline struct mlx5_rdma_counter * +to_mcounter(struct rdma_counter *counter) +{ + return container_of(counter, struct mlx5_rdma_counter, rdma_counter); +} + int mlx5_ib_counters_init(struct mlx5_ib_dev *dev); void mlx5_ib_counters_cleanup(struct mlx5_ib_dev *dev); void mlx5_ib_counters_clear_description(struct ib_counters *counters); int mlx5_ib_flow_counters_set_data(struct ib_counters *ibcounters, struct mlx5_ib_create_flow *ucmd); u16 mlx5_ib_get_counters_id(struct mlx5_ib_dev *dev, u32 port_num); +bool mlx5r_is_opfc_shared_and_in_use(struct mlx5_ib_op_fc *opfcs, u32 type, + struct mlx5_ib_op_fc **opfc); #endif /* _MLX5_IB_COUNTERS_H */ diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c index 9773d2a3d97f..1aa5311b03e9 100644 --- a/drivers/infiniband/hw/mlx5/cq.c +++ b/drivers/infiniband/hw/mlx5/cq.c @@ -38,6 +38,9 @@ #include "srq.h" #include "qp.h" +#define UVERBS_MODULE_NAME mlx5_ib +#include <rdma/uverbs_named_ioctl.h> + static void mlx5_ib_cq_comp(struct mlx5_core_cq *cq, struct mlx5_eqe *eqe) { struct ib_cq *ibcq = &to_mibcq(cq)->ibcq; @@ -487,7 +490,7 @@ repoll: } qpn = ntohl(cqe64->sop_drop_qpn) & 0xffffff; - if (!*cur_qp || (qpn != (*cur_qp)->ibqp.qp_num)) { + if (!*cur_qp || (qpn != (*cur_qp)->trans_qp.base.mqp.qpn)) { /* We do not have to take the QP table lock here, * because CQs will be locked while QPs are removed * from the table. @@ -714,7 +717,8 @@ static int mini_cqe_res_format_to_hw(struct mlx5_ib_dev *dev, u8 format) static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata, struct mlx5_ib_cq *cq, int entries, u32 **cqb, - int *cqe_size, int *index, int *inlen) + int *cqe_size, int *index, int *inlen, + struct uverbs_attr_bundle *attrs) { struct mlx5_ib_create_cq ucmd = {}; unsigned long page_size; @@ -788,7 +792,11 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata, order_base_2(page_size) - MLX5_ADAPTER_PAGE_SHIFT); MLX5_SET(cqc, cqc, page_offset, page_offset_quantized); - if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX) { + if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_CREATE_CQ_UAR_INDEX)) { + err = uverbs_copy_from(index, attrs, MLX5_IB_ATTR_CREATE_CQ_UAR_INDEX); + if (err) + goto err_cqb; + } else if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX) { *index = ucmd.uar_page_index; } else if (context->bfregi.lib_uar_dyn) { err = -EINVAL; @@ -942,8 +950,9 @@ static void notify_soft_wc_handler(struct work_struct *work) } int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, - struct ib_udata *udata) + struct uverbs_attr_bundle *attrs) { + struct ib_udata *udata = &attrs->driver_udata; struct ib_device *ibdev = ibcq->device; int entries = attr->cqe; int vector = attr->comp_vector; @@ -980,7 +989,7 @@ int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, if (udata) { err = create_cq_user(dev, udata, cq, entries, &cqb, &cqe_size, - &index, &inlen); + &index, &inlen, attrs); if (err) return err; } else { @@ -1442,3 +1451,17 @@ int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc) return 0; } + +ADD_UVERBS_ATTRIBUTES_SIMPLE( + mlx5_ib_cq_create, + UVERBS_OBJECT_CQ, + UVERBS_METHOD_CQ_CREATE, + UVERBS_ATTR_PTR_IN( + MLX5_IB_ATTR_CREATE_CQ_UAR_INDEX, + UVERBS_ATTR_TYPE(u32), + UA_OPTIONAL)); + +const struct uapi_definition mlx5_ib_create_cq_defs[] = { + UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_CQ, &mlx5_ib_cq_create), + {}, +}; diff --git a/drivers/infiniband/hw/mlx5/data_direct.c b/drivers/infiniband/hw/mlx5/data_direct.c new file mode 100644 index 000000000000..b9ba84afaae2 --- /dev/null +++ b/drivers/infiniband/hw/mlx5/data_direct.c @@ -0,0 +1,227 @@ +// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB +/* + * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved + */ + +#include "mlx5_ib.h" +#include "data_direct.h" + +static LIST_HEAD(mlx5_data_direct_dev_list); +static LIST_HEAD(mlx5_data_direct_reg_list); + +/* + * This mutex should be held when accessing either of the above lists + */ +static DEFINE_MUTEX(mlx5_data_direct_mutex); + +struct mlx5_data_direct_registration { + struct mlx5_ib_dev *ibdev; + char vuid[MLX5_ST_SZ_BYTES(array1024_auto) + 1]; + struct list_head list; +}; + +static const struct pci_device_id mlx5_data_direct_pci_table[] = { + { PCI_VDEVICE(MELLANOX, 0x2100) }, /* ConnectX-8 Data Direct */ + { 0, } +}; + +static int mlx5_data_direct_vpd_get_vuid(struct mlx5_data_direct_dev *dev) +{ + struct pci_dev *pdev = dev->pdev; + unsigned int vpd_size, kw_len; + u8 *vpd_data; + int start; + int ret; + + vpd_data = pci_vpd_alloc(pdev, &vpd_size); + if (IS_ERR(vpd_data)) { + pci_err(pdev, "Unable to read VPD, err=%ld\n", PTR_ERR(vpd_data)); + return PTR_ERR(vpd_data); + } + + start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, "VU", &kw_len); + if (start < 0) { + ret = start; + pci_err(pdev, "VU keyword not found, err=%d\n", ret); + goto end; + } + + dev->vuid = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL); + ret = dev->vuid ? 0 : -ENOMEM; + +end: + kfree(vpd_data); + return ret; +} + +static void mlx5_data_direct_shutdown(struct pci_dev *pdev) +{ + pci_disable_device(pdev); +} + +static int mlx5_data_direct_set_dma_caps(struct pci_dev *pdev) +{ + int err; + + err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); + if (err) { + dev_warn(&pdev->dev, + "Warning: couldn't set 64-bit PCI DMA mask, err=%d\n", err); + err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); + if (err) { + dev_err(&pdev->dev, "Can't set PCI DMA mask, err=%d\n", err); + return err; + } + } + + dma_set_max_seg_size(&pdev->dev, SZ_2G); + return 0; +} + +int mlx5_data_direct_ib_reg(struct mlx5_ib_dev *ibdev, char *vuid) +{ + struct mlx5_data_direct_registration *reg; + struct mlx5_data_direct_dev *dev; + + reg = kzalloc(sizeof(*reg), GFP_KERNEL); + if (!reg) + return -ENOMEM; + + reg->ibdev = ibdev; + strcpy(reg->vuid, vuid); + + mutex_lock(&mlx5_data_direct_mutex); + list_for_each_entry(dev, &mlx5_data_direct_dev_list, list) { + if (strcmp(dev->vuid, vuid) == 0) { + mlx5_ib_data_direct_bind(ibdev, dev); + break; + } + } + + /* Add the registration to its global list, to be used upon bind/unbind + * of its affiliated data direct device + */ + list_add_tail(®->list, &mlx5_data_direct_reg_list); + mutex_unlock(&mlx5_data_direct_mutex); + return 0; +} + +void mlx5_data_direct_ib_unreg(struct mlx5_ib_dev *ibdev) +{ + struct mlx5_data_direct_registration *reg; + + mutex_lock(&mlx5_data_direct_mutex); + list_for_each_entry(reg, &mlx5_data_direct_reg_list, list) { + if (reg->ibdev == ibdev) { + list_del(®->list); + kfree(reg); + goto end; + } + } + + WARN_ON(true); +end: + mutex_unlock(&mlx5_data_direct_mutex); +} + +static void mlx5_data_direct_dev_reg(struct mlx5_data_direct_dev *dev) +{ + struct mlx5_data_direct_registration *reg; + + mutex_lock(&mlx5_data_direct_mutex); + list_for_each_entry(reg, &mlx5_data_direct_reg_list, list) { + if (strcmp(dev->vuid, reg->vuid) == 0) + mlx5_ib_data_direct_bind(reg->ibdev, dev); + } + + /* Add the data direct device to the global list, further IB devices may + * use it later as well + */ + list_add_tail(&dev->list, &mlx5_data_direct_dev_list); + mutex_unlock(&mlx5_data_direct_mutex); +} + +static void mlx5_data_direct_dev_unreg(struct mlx5_data_direct_dev *dev) +{ + struct mlx5_data_direct_registration *reg; + + mutex_lock(&mlx5_data_direct_mutex); + /* Prevent any further affiliations */ + list_del(&dev->list); + list_for_each_entry(reg, &mlx5_data_direct_reg_list, list) { + if (strcmp(dev->vuid, reg->vuid) == 0) + mlx5_ib_data_direct_unbind(reg->ibdev); + } + mutex_unlock(&mlx5_data_direct_mutex); +} + +static int mlx5_data_direct_probe(struct pci_dev *pdev, const struct pci_device_id *id) +{ + struct mlx5_data_direct_dev *dev; + int err; + + dev = kzalloc(sizeof(*dev), GFP_KERNEL); + if (!dev) + return -ENOMEM; + + dev->device = &pdev->dev; + dev->pdev = pdev; + + pci_set_drvdata(dev->pdev, dev); + err = pci_enable_device(pdev); + if (err) { + dev_err(dev->device, "Cannot enable PCI device, err=%d\n", err); + goto err; + } + + pci_set_master(pdev); + err = mlx5_data_direct_set_dma_caps(pdev); + if (err) + goto err_disable; + + if (pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32) && + pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64) && + pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP128)) + dev_dbg(dev->device, "Enabling pci atomics failed\n"); + + err = mlx5_data_direct_vpd_get_vuid(dev); + if (err) + goto err_disable; + + mlx5_data_direct_dev_reg(dev); + return 0; + +err_disable: + pci_disable_device(pdev); +err: + kfree(dev); + return err; +} + +static void mlx5_data_direct_remove(struct pci_dev *pdev) +{ + struct mlx5_data_direct_dev *dev = pci_get_drvdata(pdev); + + mlx5_data_direct_dev_unreg(dev); + pci_disable_device(pdev); + kfree(dev->vuid); + kfree(dev); +} + +static struct pci_driver mlx5_data_direct_driver = { + .name = KBUILD_MODNAME, + .id_table = mlx5_data_direct_pci_table, + .probe = mlx5_data_direct_probe, + .remove = mlx5_data_direct_remove, + .shutdown = mlx5_data_direct_shutdown, +}; + +int mlx5_data_direct_driver_register(void) +{ + return pci_register_driver(&mlx5_data_direct_driver); +} + +void mlx5_data_direct_driver_unregister(void) +{ + pci_unregister_driver(&mlx5_data_direct_driver); +} diff --git a/drivers/infiniband/hw/mlx5/data_direct.h b/drivers/infiniband/hw/mlx5/data_direct.h new file mode 100644 index 000000000000..2fd2bdbe8f69 --- /dev/null +++ b/drivers/infiniband/hw/mlx5/data_direct.h @@ -0,0 +1,23 @@ +/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ +/* + * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved + */ + +#ifndef _MLX5_IB_DATA_DIRECT_H +#define _MLX5_IB_DATA_DIRECT_H + +struct mlx5_ib_dev; + +struct mlx5_data_direct_dev { + struct device *device; + struct pci_dev *pdev; + char *vuid; + struct list_head list; +}; + +int mlx5_data_direct_ib_reg(struct mlx5_ib_dev *ibdev, char *vuid); +void mlx5_data_direct_ib_unreg(struct mlx5_ib_dev *ibdev); +int mlx5_data_direct_driver_register(void); +void mlx5_data_direct_driver_unregister(void); + +#endif diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c index 869369cb5b5f..2479da8620ca 100644 --- a/drivers/infiniband/hw/mlx5/devx.c +++ b/drivers/infiniband/hw/mlx5/devx.c @@ -13,6 +13,7 @@ #include <rdma/uverbs_std_types.h> #include <linux/mlx5/driver.h> #include <linux/mlx5/fs.h> +#include <rdma/ib_ucaps.h> #include "mlx5_ib.h" #include "devx.h" #include "qp.h" @@ -27,6 +28,19 @@ enum devx_obj_flags { DEVX_OBJ_FLAGS_INDIRECT_MKEY = 1 << 0, DEVX_OBJ_FLAGS_DCT = 1 << 1, DEVX_OBJ_FLAGS_CQ = 1 << 2, + DEVX_OBJ_FLAGS_HW_FREED = 1 << 3, +}; + +#define MAX_ASYNC_CMDS 8 + +struct mlx5_async_cmd { + struct ib_uobject *uobject; + void *in; + int in_size; + u32 out[MLX5_ST_SZ_DW(general_obj_out_cmd_hdr)]; + int err; + struct mlx5_async_work cb_work; + struct completion comp; }; struct devx_async_data { @@ -109,7 +123,27 @@ devx_ufile2uctx(const struct uverbs_attr_bundle *attrs) return to_mucontext(ib_uverbs_get_ucontext(attrs)); } -int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user) +static int set_uctx_ucaps(struct mlx5_ib_dev *dev, u64 req_ucaps, u32 *cap) +{ + if (UCAP_ENABLED(req_ucaps, RDMA_UCAP_MLX5_CTRL_LOCAL)) { + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RDMA_CTRL) + *cap |= MLX5_UCTX_CAP_RDMA_CTRL; + else + return -EOPNOTSUPP; + } + + if (UCAP_ENABLED(req_ucaps, RDMA_UCAP_MLX5_CTRL_OTHER_VHCA)) { + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & + MLX5_UCTX_CAP_RDMA_CTRL_OTHER_VHCA) + *cap |= MLX5_UCTX_CAP_RDMA_CTRL_OTHER_VHCA; + else + return -EOPNOTSUPP; + } + + return 0; +} + +int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user, u64 req_ucaps) { u32 in[MLX5_ST_SZ_DW(create_uctx_in)] = {}; u32 out[MLX5_ST_SZ_DW(create_uctx_out)] = {}; @@ -123,14 +157,22 @@ int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user) return -EINVAL; uctx = MLX5_ADDR_OF(create_uctx_in, in, uctx); - if (is_user && capable(CAP_NET_RAW) && - (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RAW_TX)) + if (is_user && + (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RAW_TX) && + capable(CAP_NET_RAW)) cap |= MLX5_UCTX_CAP_RAW_TX; - if (is_user && capable(CAP_SYS_RAWIO) && + if (is_user && (MLX5_CAP_GEN(dev->mdev, uctx_cap) & - MLX5_UCTX_CAP_INTERNAL_DEV_RES)) + MLX5_UCTX_CAP_INTERNAL_DEV_RES) && + capable(CAP_SYS_RAWIO)) cap |= MLX5_UCTX_CAP_INTERNAL_DEV_RES; + if (req_ucaps) { + err = set_uctx_ucaps(dev, req_ucaps, &cap); + if (err) + return err; + } + MLX5_SET(create_uctx_in, in, opcode, MLX5_CMD_OP_CREATE_UCTX); MLX5_SET(uctx, uctx, cap, cap); @@ -1405,7 +1447,9 @@ static int devx_obj_cleanup(struct ib_uobject *uobject, */ mlx5r_deref_wait_odp_mkey(&obj->mkey); - if (obj->flags & DEVX_OBJ_FLAGS_DCT) + if (obj->flags & DEVX_OBJ_FLAGS_HW_FREED) + ret = 0; + else if (obj->flags & DEVX_OBJ_FLAGS_DCT) ret = mlx5_core_destroy_dct(obj->ib_dev, &obj->core_dct); else if (obj->flags & DEVX_OBJ_FLAGS_CQ) ret = mlx5_core_destroy_cq(obj->ib_dev->mdev, &obj->core_cq); @@ -2558,7 +2602,7 @@ int mlx5_ib_devx_init(struct mlx5_ib_dev *dev) struct mlx5_devx_event_table *table = &dev->devx_event_table; int uid; - uid = mlx5_ib_devx_create(dev, false); + uid = mlx5_ib_devx_create(dev, false, 0); if (uid > 0) { dev->devx_whitelist_uid = uid; xa_init(&table->event_xa); @@ -2595,6 +2639,82 @@ void mlx5_ib_devx_cleanup(struct mlx5_ib_dev *dev) } } +static void devx_async_destroy_cb(int status, struct mlx5_async_work *context) +{ + struct mlx5_async_cmd *devx_out = container_of(context, + struct mlx5_async_cmd, cb_work); + struct devx_obj *obj = devx_out->uobject->object; + + if (!status) + obj->flags |= DEVX_OBJ_FLAGS_HW_FREED; + + complete(&devx_out->comp); +} + +static void devx_async_destroy(struct mlx5_ib_dev *dev, + struct mlx5_async_cmd *cmd) +{ + init_completion(&cmd->comp); + cmd->err = mlx5_cmd_exec_cb(&dev->async_ctx, cmd->in, cmd->in_size, + &cmd->out, sizeof(cmd->out), + devx_async_destroy_cb, &cmd->cb_work); +} + +static void devx_wait_async_destroy(struct mlx5_async_cmd *cmd) +{ + if (!cmd->err) + wait_for_completion(&cmd->comp); + atomic_set(&cmd->uobject->usecnt, 0); +} + +void mlx5_ib_ufile_hw_cleanup(struct ib_uverbs_file *ufile) +{ + struct mlx5_async_cmd async_cmd[MAX_ASYNC_CMDS]; + struct ib_ucontext *ucontext = ufile->ucontext; + struct ib_device *device = ucontext->device; + struct mlx5_ib_dev *dev = to_mdev(device); + struct ib_uobject *uobject; + struct devx_obj *obj; + int head = 0; + int tail = 0; + + list_for_each_entry(uobject, &ufile->uobjects, list) { + WARN_ON(uverbs_try_lock_object(uobject, UVERBS_LOOKUP_WRITE)); + + /* + * Currently we only support QP destruction, if other objects + * are to be destroyed need to add type synchronization to the + * cleanup algorithm and handle pre/post FW cleanup for the + * new types if needed. + */ + if (uobj_get_object_id(uobject) != MLX5_IB_OBJECT_DEVX_OBJ || + (get_dec_obj_type(uobject->object, MLX5_EVENT_TYPE_MAX) != + MLX5_OBJ_TYPE_QP)) { + atomic_set(&uobject->usecnt, 0); + continue; + } + + obj = uobject->object; + + async_cmd[tail % MAX_ASYNC_CMDS].in = obj->dinbox; + async_cmd[tail % MAX_ASYNC_CMDS].in_size = obj->dinlen; + async_cmd[tail % MAX_ASYNC_CMDS].uobject = uobject; + + devx_async_destroy(dev, &async_cmd[tail % MAX_ASYNC_CMDS]); + tail++; + + if (tail - head == MAX_ASYNC_CMDS) { + devx_wait_async_destroy(&async_cmd[head % MAX_ASYNC_CMDS]); + head++; + } + } + + while (head != tail) { + devx_wait_async_destroy(&async_cmd[head % MAX_ASYNC_CMDS]); + head++; + } +} + static ssize_t devx_async_cmd_event_read(struct file *filp, char __user *buf, size_t count, loff_t *pos) { @@ -2673,7 +2793,6 @@ static const struct file_operations devx_async_cmd_event_fops = { .read = devx_async_cmd_event_read, .poll = devx_async_cmd_event_poll, .release = uverbs_uobject_fd_release, - .llseek = no_llseek, }; static ssize_t devx_async_event_read(struct file *filp, char __user *buf, @@ -2788,7 +2907,6 @@ static const struct file_operations devx_async_event_fops = { .read = devx_async_event_read, .poll = devx_async_event_poll, .release = uverbs_uobject_fd_release, - .llseek = no_llseek, }; static void devx_async_cmd_event_destroy_uobj(struct ib_uobject *uobj, @@ -2949,7 +3067,7 @@ DECLARE_UVERBS_NAMED_METHOD( MLX5_IB_METHOD_DEVX_OBJ_MODIFY, UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_MODIFY_HANDLE, UVERBS_IDR_ANY_OBJECT, - UVERBS_ACCESS_WRITE, + UVERBS_ACCESS_READ, UA_MANDATORY), UVERBS_ATTR_PTR_IN( MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_IN, diff --git a/drivers/infiniband/hw/mlx5/devx.h b/drivers/infiniband/hw/mlx5/devx.h index ee2213275fd6..ee9e7d3af93f 100644 --- a/drivers/infiniband/hw/mlx5/devx.h +++ b/drivers/infiniband/hw/mlx5/devx.h @@ -24,12 +24,14 @@ struct devx_obj { struct list_head event_sub; /* holds devx_event_subscription entries */ }; #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) -int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user); +int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user, u64 req_ucaps); void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid); int mlx5_ib_devx_init(struct mlx5_ib_dev *dev); void mlx5_ib_devx_cleanup(struct mlx5_ib_dev *dev); +void mlx5_ib_ufile_hw_cleanup(struct ib_uverbs_file *ufile); #else -static inline int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user) +static inline int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user, + u64 req_ucaps) { return -EOPNOTSUPP; } @@ -41,5 +43,8 @@ static inline int mlx5_ib_devx_init(struct mlx5_ib_dev *dev) static inline void mlx5_ib_devx_cleanup(struct mlx5_ib_dev *dev) { } +static inline void mlx5_ib_ufile_hw_cleanup(struct ib_uverbs_file *ufile) +{ +} #endif #endif /* _MLX5_IB_DEVX_H */ diff --git a/drivers/infiniband/hw/mlx5/fs.c b/drivers/infiniband/hw/mlx5/fs.c index 520034acf73a..680627f1de33 100644 --- a/drivers/infiniband/hw/mlx5/fs.c +++ b/drivers/infiniband/hw/mlx5/fs.c @@ -12,6 +12,7 @@ #include <rdma/mlx5_user_ioctl_verbs.h> #include <rdma/ib_hdrs.h> #include <rdma/ib_umem.h> +#include <rdma/ib_ucaps.h> #include <linux/mlx5/driver.h> #include <linux/mlx5/fs.h> #include <linux/mlx5/fs_helpers.h> @@ -32,6 +33,11 @@ enum { MATCH_CRITERIA_ENABLE_MISC2_BIT }; + +struct mlx5_per_qp_opfc { + struct mlx5_ib_op_fc opfcs[MLX5_IB_OPCOUNTER_MAX]; +}; + #define HEADER_IS_ZERO(match_criteria, headers) \ !(memchr_inv(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \ 0, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \ @@ -678,7 +684,7 @@ enum flow_table_type { #define MLX5_FS_MAX_TYPES 6 #define MLX5_FS_MAX_ENTRIES BIT(16) -static bool mlx5_ib_shared_ft_allowed(struct ib_device *device) +static bool __maybe_unused mlx5_ib_shared_ft_allowed(struct ib_device *device) { struct mlx5_ib_dev *dev = to_mdev(device); @@ -690,7 +696,7 @@ static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_ib_dev *dev, struct mlx5_ib_flow_prio *prio, int priority, int num_entries, int num_groups, - u32 flags) + u32 flags, u16 vport) { struct mlx5_flow_table_attr ft_attr = {}; struct mlx5_flow_table *ft; @@ -698,6 +704,7 @@ static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_ib_dev *dev, ft_attr.prio = priority; ft_attr.max_fte = num_entries; ft_attr.flags = flags; + ft_attr.vport = vport; ft_attr.autogroup.max_num_groups = num_groups; ft = mlx5_create_auto_grouped_flow_table(ns, &ft_attr); if (IS_ERR(ft)) @@ -792,18 +799,25 @@ static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev, ft = prio->flow_table; if (!ft) return _get_prio(dev, ns, prio, priority, max_table_size, - num_groups, flags); + num_groups, flags, 0); return prio; } enum { + RDMA_RX_ECN_OPCOUNTER_PER_QP_PRIO, + RDMA_RX_CNP_OPCOUNTER_PER_QP_PRIO, + RDMA_RX_PKTS_BYTES_OPCOUNTER_PER_QP_PRIO, RDMA_RX_ECN_OPCOUNTER_PRIO, RDMA_RX_CNP_OPCOUNTER_PRIO, + RDMA_RX_PKTS_BYTES_OPCOUNTER_PRIO, }; enum { + RDMA_TX_CNP_OPCOUNTER_PER_QP_PRIO, + RDMA_TX_PKTS_BYTES_OPCOUNTER_PER_QP_PRIO, RDMA_TX_CNP_OPCOUNTER_PRIO, + RDMA_TX_PKTS_BYTES_OPCOUNTER_PRIO, }; static int set_vhca_port_spec(struct mlx5_ib_dev *dev, u32 port_num, @@ -867,6 +881,344 @@ static int set_cnp_spec(struct mlx5_ib_dev *dev, u32 port_num, return 0; } +/* Returns the prio we should use for the given optional counter type, + * whereas for bytes type we use the packet type, since they share the same + * resources. + */ +static struct mlx5_ib_flow_prio *get_opfc_prio(struct mlx5_ib_dev *dev, + u32 type) +{ + u32 prio_type; + + switch (type) { + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES: + prio_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES: + prio_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP: + prio_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP: + prio_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP; + break; + default: + prio_type = type; + } + + return &dev->flow_db->opfcs[prio_type]; +} + +static void put_per_qp_prio(struct mlx5_ib_dev *dev, + enum mlx5_ib_optional_counter_type type) +{ + enum mlx5_ib_optional_counter_type per_qp_type; + struct mlx5_ib_flow_prio *prio; + + switch (type) { + case MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS: + per_qp_type = MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS: + per_qp_type = MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS: + per_qp_type = MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS: + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES: + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS: + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES: + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; + break; + default: + return; + } + + prio = get_opfc_prio(dev, per_qp_type); + put_flow_table(dev, prio, true); +} + +static int get_per_qp_prio(struct mlx5_ib_dev *dev, + enum mlx5_ib_optional_counter_type type) +{ + enum mlx5_ib_optional_counter_type per_qp_type; + enum mlx5_flow_namespace_type fn_type; + struct mlx5_flow_namespace *ns; + struct mlx5_ib_flow_prio *prio; + int priority; + + switch (type) { + case MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS; + priority = RDMA_RX_ECN_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS; + priority = RDMA_RX_CNP_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS; + priority = RDMA_TX_CNP_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS; + priority = RDMA_TX_PKTS_BYTES_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS; + priority = RDMA_TX_PKTS_BYTES_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS; + priority = RDMA_RX_PKTS_BYTES_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES: + fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS; + priority = RDMA_RX_PKTS_BYTES_OPCOUNTER_PER_QP_PRIO; + per_qp_type = MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; + break; + default: + return -EINVAL; + } + + ns = mlx5_get_flow_namespace(dev->mdev, fn_type); + if (!ns) + return -EOPNOTSUPP; + + prio = get_opfc_prio(dev, per_qp_type); + if (prio->flow_table) + return 0; + + prio = _get_prio(dev, ns, prio, priority, MLX5_FS_MAX_POOL_SIZE, 1, 0, 0); + if (IS_ERR(prio)) + return PTR_ERR(prio); + + prio->refcount = 1; + + return 0; +} + +static struct mlx5_per_qp_opfc * +get_per_qp_opfc(struct mlx5_rdma_counter *mcounter, u32 qp_num, bool *new) +{ + struct mlx5_per_qp_opfc *per_qp_opfc; + + *new = false; + + per_qp_opfc = xa_load(&mcounter->qpn_opfc_xa, qp_num); + if (per_qp_opfc) + return per_qp_opfc; + per_qp_opfc = kzalloc(sizeof(*per_qp_opfc), GFP_KERNEL); + + if (!per_qp_opfc) + return NULL; + + *new = true; + return per_qp_opfc; +} + +static int add_op_fc_rules(struct mlx5_ib_dev *dev, + struct mlx5_rdma_counter *mcounter, + struct mlx5_per_qp_opfc *per_qp_opfc, + struct mlx5_ib_flow_prio *prio, + enum mlx5_ib_optional_counter_type type, + u32 qp_num, u32 port_num) +{ + struct mlx5_ib_op_fc *opfc = &per_qp_opfc->opfcs[type], *in_use_opfc; + struct mlx5_flow_act flow_act = {}; + struct mlx5_flow_destination dst; + struct mlx5_flow_spec *spec; + int i, err, spec_num; + bool is_tx; + + if (opfc->fc) + return -EEXIST; + + if (mlx5r_is_opfc_shared_and_in_use(per_qp_opfc->opfcs, type, + &in_use_opfc)) { + opfc->fc = in_use_opfc->fc; + opfc->rule[0] = in_use_opfc->rule[0]; + return 0; + } + + opfc->fc = mcounter->fc[type]; + + spec = kcalloc(MAX_OPFC_RULES, sizeof(*spec), GFP_KERNEL); + if (!spec) { + err = -ENOMEM; + goto null_fc; + } + + switch (type) { + case MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP: + if (set_ecn_ce_spec(dev, port_num, &spec[0], + MLX5_FS_IPV4_VERSION) || + set_ecn_ce_spec(dev, port_num, &spec[1], + MLX5_FS_IPV6_VERSION)) { + err = -EOPNOTSUPP; + goto free_spec; + } + spec_num = 2; + is_tx = false; + + MLX5_SET_TO_ONES(fte_match_param, spec[1].match_criteria, + misc_parameters.bth_dst_qp); + MLX5_SET(fte_match_param, spec[1].match_value, + misc_parameters.bth_dst_qp, qp_num); + spec[1].match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS; + break; + case MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS_PER_QP: + if (!MLX5_CAP_FLOWTABLE( + dev->mdev, + ft_field_support_2_nic_receive_rdma.bth_opcode) || + set_cnp_spec(dev, port_num, &spec[0])) { + err = -EOPNOTSUPP; + goto free_spec; + } + spec_num = 1; + is_tx = false; + break; + case MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS_PER_QP: + if (!MLX5_CAP_FLOWTABLE( + dev->mdev, + ft_field_support_2_nic_transmit_rdma.bth_opcode) || + set_cnp_spec(dev, port_num, &spec[0])) { + err = -EOPNOTSUPP; + goto free_spec; + } + spec_num = 1; + is_tx = true; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP: + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP: + spec_num = 1; + is_tx = true; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP: + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP: + spec_num = 1; + is_tx = false; + break; + default: + err = -EINVAL; + goto free_spec; + } + + if (is_tx) { + MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria, + misc_parameters.source_sqn); + MLX5_SET(fte_match_param, spec->match_value, + misc_parameters.source_sqn, qp_num); + } else { + MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria, + misc_parameters.bth_dst_qp); + MLX5_SET(fte_match_param, spec->match_value, + misc_parameters.bth_dst_qp, qp_num); + } + + spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS; + + dst.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER; + dst.counter = opfc->fc; + + flow_act.action = + MLX5_FLOW_CONTEXT_ACTION_COUNT | MLX5_FLOW_CONTEXT_ACTION_ALLOW; + + for (i = 0; i < spec_num; i++) { + opfc->rule[i] = mlx5_add_flow_rules(prio->flow_table, &spec[i], + &flow_act, &dst, 1); + if (IS_ERR(opfc->rule[i])) { + err = PTR_ERR(opfc->rule[i]); + goto del_rules; + } + } + prio->refcount += spec_num; + + err = xa_err(xa_store(&mcounter->qpn_opfc_xa, qp_num, per_qp_opfc, + GFP_KERNEL)); + if (err) + goto del_rules; + + kfree(spec); + + return 0; + +del_rules: + while (i--) + mlx5_del_flow_rules(opfc->rule[i]); + put_flow_table(dev, prio, false); +free_spec: + kfree(spec); +null_fc: + opfc->fc = NULL; + return err; +} + +static bool is_fc_shared_and_in_use(struct mlx5_rdma_counter *mcounter, + u32 type, struct mlx5_fc **fc) +{ + u32 shared_fc_type; + + switch (type) { + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; + break; + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP: + shared_fc_type = MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP; + break; + default: + return false; + } + + *fc = mcounter->fc[shared_fc_type]; + if (!(*fc)) + return false; + + return true; +} + +void mlx5r_fs_destroy_fcs(struct mlx5_ib_dev *dev, + struct rdma_counter *counter) +{ + struct mlx5_rdma_counter *mcounter = to_mcounter(counter); + struct mlx5_fc *in_use_fc; + int i; + + for (i = MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP; + i <= MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; i++) { + if (!mcounter->fc[i]) + continue; + + if (is_fc_shared_and_in_use(mcounter, i, &in_use_fc)) { + mcounter->fc[i] = NULL; + continue; + } + + mlx5_fc_destroy(dev->mdev, mcounter->fc[i]); + mcounter->fc[i] = NULL; + } +} + int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num, struct mlx5_ib_op_fc *opfc, enum mlx5_ib_optional_counter_type type) @@ -921,6 +1273,20 @@ int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num, priority = RDMA_TX_CNP_OPCOUNTER_PRIO; break; + case MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS: + case MLX5_IB_OPCOUNTER_RDMA_TX_BYTES: + spec_num = 1; + fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS; + priority = RDMA_TX_PKTS_BYTES_OPCOUNTER_PRIO; + break; + + case MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS: + case MLX5_IB_OPCOUNTER_RDMA_RX_BYTES: + spec_num = 1; + fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS; + priority = RDMA_RX_PKTS_BYTES_OPCOUNTER_PRIO; + break; + default: err = -EOPNOTSUPP; goto free; @@ -932,18 +1298,22 @@ int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num, goto free; } - prio = &dev->flow_db->opfcs[type]; + prio = get_opfc_prio(dev, type); if (!prio->flow_table) { + err = get_per_qp_prio(dev, type); + if (err) + goto free; + prio = _get_prio(dev, ns, prio, priority, - dev->num_ports * MAX_OPFC_RULES, 1, 0); + dev->num_ports * MAX_OPFC_RULES, 1, 0, 0); if (IS_ERR(prio)) { err = PTR_ERR(prio); - goto free; + goto put_prio; } } dst.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER; - dst.counter_id = mlx5_fc_id(opfc->fc); + dst.counter = opfc->fc; flow_act.action = MLX5_FLOW_CONTEXT_ACTION_COUNT | MLX5_FLOW_CONTEXT_ACTION_ALLOW; @@ -965,6 +1335,8 @@ del_rules: for (i -= 1; i >= 0; i--) mlx5_del_flow_rules(opfc->rule[i]); put_flow_table(dev, prio, false); +put_prio: + put_per_qp_prio(dev, type); free: kfree(spec); return err; @@ -974,12 +1346,115 @@ void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev, struct mlx5_ib_op_fc *opfc, enum mlx5_ib_optional_counter_type type) { + struct mlx5_ib_flow_prio *prio; int i; + prio = get_opfc_prio(dev, type); + for (i = 0; i < MAX_OPFC_RULES && opfc->rule[i]; i++) { mlx5_del_flow_rules(opfc->rule[i]); - put_flow_table(dev, &dev->flow_db->opfcs[type], true); + put_flow_table(dev, prio, true); + } + + put_per_qp_prio(dev, type); +} + +void mlx5r_fs_unbind_op_fc(struct ib_qp *qp, struct rdma_counter *counter) +{ + struct mlx5_rdma_counter *mcounter = to_mcounter(counter); + struct mlx5_ib_dev *dev = to_mdev(counter->device); + struct mlx5_per_qp_opfc *per_qp_opfc; + struct mlx5_ib_op_fc *in_use_opfc; + struct mlx5_ib_flow_prio *prio; + int i, j; + + per_qp_opfc = xa_load(&mcounter->qpn_opfc_xa, qp->qp_num); + if (!per_qp_opfc) + return; + + for (i = MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP; + i <= MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP; i++) { + if (!per_qp_opfc->opfcs[i].fc) + continue; + + if (mlx5r_is_opfc_shared_and_in_use(per_qp_opfc->opfcs, i, + &in_use_opfc)) { + per_qp_opfc->opfcs[i].fc = NULL; + continue; + } + + for (j = 0; j < MAX_OPFC_RULES; j++) { + if (!per_qp_opfc->opfcs[i].rule[j]) + continue; + mlx5_del_flow_rules(per_qp_opfc->opfcs[i].rule[j]); + prio = get_opfc_prio(dev, i); + put_flow_table(dev, prio, true); + } + per_qp_opfc->opfcs[i].fc = NULL; + } + + kfree(per_qp_opfc); + xa_erase(&mcounter->qpn_opfc_xa, qp->qp_num); +} + +int mlx5r_fs_bind_op_fc(struct ib_qp *qp, struct rdma_counter *counter, + u32 port) +{ + struct mlx5_rdma_counter *mcounter = to_mcounter(counter); + struct mlx5_ib_dev *dev = to_mdev(qp->device); + struct mlx5_per_qp_opfc *per_qp_opfc; + struct mlx5_ib_flow_prio *prio; + struct mlx5_ib_counters *cnts; + struct mlx5_ib_op_fc *opfc; + struct mlx5_fc *in_use_fc; + int i, err, per_qp_type; + bool new; + + if (!counter->mode.bind_opcnt) + return 0; + + cnts = &dev->port[port - 1].cnts; + + for (i = 0; i <= MLX5_IB_OPCOUNTER_RDMA_RX_BYTES; i++) { + opfc = &cnts->opfcs[i]; + if (!opfc->fc) + continue; + + per_qp_type = i + MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP; + prio = get_opfc_prio(dev, per_qp_type); + WARN_ON(!prio->flow_table); + + if (is_fc_shared_and_in_use(mcounter, per_qp_type, &in_use_fc)) + mcounter->fc[per_qp_type] = in_use_fc; + + if (!mcounter->fc[per_qp_type]) { + mcounter->fc[per_qp_type] = mlx5_fc_create(dev->mdev, + false); + if (IS_ERR(mcounter->fc[per_qp_type])) + return PTR_ERR(mcounter->fc[per_qp_type]); + } + + per_qp_opfc = get_per_qp_opfc(mcounter, qp->qp_num, &new); + if (!per_qp_opfc) { + err = -ENOMEM; + goto free_fc; + } + err = add_op_fc_rules(dev, mcounter, per_qp_opfc, prio, + per_qp_type, qp->qp_num, port); + if (err) + goto del_rules; } + + return 0; + +del_rules: + mlx5r_fs_unbind_op_fc(qp, counter); + if (new) + kfree(per_qp_opfc); +free_fc: + if (xa_empty(&mcounter->qpn_opfc_xa)) + mlx5r_fs_destroy_fcs(dev, counter); + return err; } static void set_underlay_qp(struct mlx5_ib_dev *dev, @@ -1113,8 +1588,8 @@ static struct mlx5_ib_flow_handler *_create_flow_rule(struct mlx5_ib_dev *dev, handler->ibcounters = flow_act.counters; dest_arr[dest_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER; - dest_arr[dest_num].counter_id = - mlx5_fc_id(mcounters->hw_cntrs_hndl); + dest_arr[dest_num].counter = + mcounters->hw_cntrs_hndl; dest_num++; } @@ -1170,11 +1645,6 @@ static struct mlx5_ib_flow_handler *create_flow_rule(struct mlx5_ib_dev *dev, return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL); } -enum { - LEFTOVERS_MC, - LEFTOVERS_UC, -}; - static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *dev, struct mlx5_ib_flow_prio *ft_prio, struct ib_flow_attr *flow_attr, @@ -1184,43 +1654,32 @@ static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *de struct mlx5_ib_flow_handler *handler = NULL; static struct { - struct ib_flow_attr flow_attr; struct ib_flow_spec_eth eth_flow; - } leftovers_specs[] = { - [LEFTOVERS_MC] = { - .flow_attr = { - .num_of_specs = 1, - .size = sizeof(leftovers_specs[0]) - }, - .eth_flow = { - .type = IB_FLOW_SPEC_ETH, - .size = sizeof(struct ib_flow_spec_eth), - .mask = {.dst_mac = {0x1} }, - .val = {.dst_mac = {0x1} } - } - }, - [LEFTOVERS_UC] = { - .flow_attr = { - .num_of_specs = 1, - .size = sizeof(leftovers_specs[0]) - }, - .eth_flow = { - .type = IB_FLOW_SPEC_ETH, - .size = sizeof(struct ib_flow_spec_eth), - .mask = {.dst_mac = {0x1} }, - .val = {.dst_mac = {} } - } - } - }; + struct ib_flow_attr flow_attr; + } leftovers_wc = { .flow_attr = { .num_of_specs = 1, + .size = sizeof(leftovers_wc) }, + .eth_flow = { + .type = IB_FLOW_SPEC_ETH, + .size = sizeof(struct ib_flow_spec_eth), + .mask = { .dst_mac = { 0x1 } }, + .val = { .dst_mac = { 0x1 } } } }; - handler = create_flow_rule(dev, ft_prio, - &leftovers_specs[LEFTOVERS_MC].flow_attr, - dst); + static struct { + struct ib_flow_spec_eth eth_flow; + struct ib_flow_attr flow_attr; + } leftovers_uc = { .flow_attr = { .num_of_specs = 1, + .size = sizeof(leftovers_uc) }, + .eth_flow = { + .type = IB_FLOW_SPEC_ETH, + .size = sizeof(struct ib_flow_spec_eth), + .mask = { .dst_mac = { 0x1 } }, + .val = { .dst_mac = {} } } }; + + handler = create_flow_rule(dev, ft_prio, &leftovers_wc.flow_attr, dst); if (!IS_ERR(handler) && flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) { handler_ucast = create_flow_rule(dev, ft_prio, - &leftovers_specs[LEFTOVERS_UC].flow_attr, - dst); + &leftovers_uc.flow_attr, dst); if (IS_ERR(handler_ucast)) { mlx5_del_flow_rules(handler->rule); ft_prio->refcount--; @@ -1413,17 +1872,51 @@ free_ucmd: return ERR_PTR(err); } +static int mlx5_ib_fill_transport_ns_info(struct mlx5_ib_dev *dev, + enum mlx5_flow_namespace_type type, + u32 *flags, u16 *vport_idx, + u16 *vport, + struct mlx5_core_dev **ft_mdev, + u32 ib_port) +{ + struct mlx5_core_dev *esw_mdev; + + if (!is_mdev_switchdev_mode(dev->mdev)) + return 0; + + if (!MLX5_CAP_ADV_RDMA(dev->mdev, rdma_transport_manager)) + return -EOPNOTSUPP; + + if (!dev->port[ib_port - 1].rep) + return -EINVAL; + + esw_mdev = mlx5_eswitch_get_core_dev(dev->port[ib_port - 1].rep->esw); + if (esw_mdev != dev->mdev) + return -EOPNOTSUPP; + + *flags |= MLX5_FLOW_TABLE_OTHER_VPORT; + *ft_mdev = esw_mdev; + *vport = dev->port[ib_port - 1].rep->vport; + *vport_idx = dev->port[ib_port - 1].rep->vport_index; + + return 0; +} + static struct mlx5_ib_flow_prio * _get_flow_table(struct mlx5_ib_dev *dev, u16 user_priority, enum mlx5_flow_namespace_type ns_type, - bool mcast) + bool mcast, u32 ib_port) { + struct mlx5_core_dev *ft_mdev = dev->mdev; struct mlx5_flow_namespace *ns = NULL; struct mlx5_ib_flow_prio *prio = NULL; int max_table_size = 0; + u16 vport_idx = 0; bool esw_encap; u32 flags = 0; + u16 vport = 0; int priority; + int ret; if (mcast) priority = MLX5_IB_FLOW_MCAST_PRIO; @@ -1471,13 +1964,38 @@ _get_flow_table(struct mlx5_ib_dev *dev, u16 user_priority, MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev, log_max_ft_size)); priority = user_priority; break; + case MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX: + case MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX: + if (ib_port == 0 || user_priority > MLX5_RDMA_TRANSPORT_BYPASS_PRIO) + return ERR_PTR(-EINVAL); + ret = mlx5_ib_fill_transport_ns_info(dev, ns_type, &flags, + &vport_idx, &vport, + &ft_mdev, ib_port); + if (ret) + return ERR_PTR(ret); + + if (ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX) + max_table_size = + BIT(MLX5_CAP_FLOWTABLE_RDMA_TRANSPORT_RX( + ft_mdev, log_max_ft_size)); + else + max_table_size = + BIT(MLX5_CAP_FLOWTABLE_RDMA_TRANSPORT_TX( + ft_mdev, log_max_ft_size)); + priority = user_priority; + break; default: break; } max_table_size = min_t(int, max_table_size, MLX5_FS_MAX_ENTRIES); - ns = mlx5_get_flow_namespace(dev->mdev, ns_type); + if (ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX || + ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX) + ns = mlx5_get_flow_vport_namespace(ft_mdev, ns_type, vport_idx); + else + ns = mlx5_get_flow_namespace(ft_mdev, ns_type); + if (!ns) return ERR_PTR(-EOPNOTSUPP); @@ -1497,6 +2015,12 @@ _get_flow_table(struct mlx5_ib_dev *dev, u16 user_priority, case MLX5_FLOW_NAMESPACE_RDMA_TX: prio = &dev->flow_db->rdma_tx[priority]; break; + case MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX: + prio = &dev->flow_db->rdma_transport_rx[ib_port - 1]; + break; + case MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX: + prio = &dev->flow_db->rdma_transport_tx[ib_port - 1]; + break; default: return ERR_PTR(-EINVAL); } @@ -1507,7 +2031,7 @@ _get_flow_table(struct mlx5_ib_dev *dev, u16 user_priority, return prio; return _get_prio(dev, ns, prio, priority, max_table_size, - MLX5_FS_MAX_TYPES, flags); + MLX5_FS_MAX_TYPES, flags, vport); } static struct mlx5_ib_flow_handler * @@ -1603,7 +2127,7 @@ static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher, static struct mlx5_ib_flow_handler *raw_fs_rule_add( struct mlx5_ib_dev *dev, struct mlx5_ib_flow_matcher *fs_matcher, struct mlx5_flow_context *flow_context, struct mlx5_flow_act *flow_act, - u32 counter_id, void *cmd_in, int inlen, int dest_id, int dest_type) + struct mlx5_fc *counter, void *cmd_in, int inlen, int dest_id, int dest_type) { struct mlx5_flow_destination *dst; struct mlx5_ib_flow_prio *ft_prio; @@ -1626,7 +2150,8 @@ static struct mlx5_ib_flow_handler *raw_fs_rule_add( mutex_lock(&dev->flow_db->lock); ft_prio = _get_flow_table(dev, fs_matcher->priority, - fs_matcher->ns_type, mcast); + fs_matcher->ns_type, mcast, + fs_matcher->ib_port); if (IS_ERR(ft_prio)) { err = PTR_ERR(ft_prio); goto unlock; @@ -1652,8 +2177,12 @@ static struct mlx5_ib_flow_handler *raw_fs_rule_add( } if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) { + if (WARN_ON(!counter)) { + err = -EINVAL; + goto unlock; + } dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER; - dst[dst_num].counter_id = counter_id; + dst[dst_num].counter = counter; dst_num++; } @@ -1738,6 +2267,12 @@ mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type, case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TX: *namespace = MLX5_FLOW_NAMESPACE_RDMA_TX; break; + case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TRANSPORT_RX: + *namespace = MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX; + break; + case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TRANSPORT_TX: + *namespace = MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX; + break; default: return -EINVAL; } @@ -1827,7 +2362,8 @@ static int get_dests(struct uverbs_attr_bundle *attrs, return -EINVAL; /* Allow only DEVX object or QP as dest when inserting to RDMA_RX */ - if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) && + if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX || + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX) && ((!dest_devx && !dest_qp) || (dest_devx && dest_qp))) return -EINVAL; @@ -1844,7 +2380,8 @@ static int get_dests(struct uverbs_attr_bundle *attrs, return -EINVAL; /* Allow only flow table as dest when inserting to FDB or RDMA_RX */ if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS || - fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) && + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX || + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX) && *dest_type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) return -EINVAL; } else if (dest_qp) { @@ -1865,20 +2402,23 @@ static int get_dests(struct uverbs_attr_bundle *attrs, *dest_id = mqp->raw_packet_qp.rq.tirn; *dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR; } else if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS || - fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) && + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX || + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX) && !(*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)) { *dest_type = MLX5_FLOW_DESTINATION_TYPE_PORT; } if (*dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR && (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS || - fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX)) + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX || + fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX)) return -EINVAL; return 0; } -static bool is_flow_counter(void *obj, u32 offset, u32 *counter_id) +static bool +is_flow_counter(void *obj, u32 offset, u32 *counter_id, u32 *fc_bulk_size) { struct devx_obj *devx_obj = obj; u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode); @@ -1888,6 +2428,7 @@ static bool is_flow_counter(void *obj, u32 offset, u32 *counter_id) if (offset && offset >= devx_obj->flow_counter_bulk_size) return false; + *fc_bulk_size = devx_obj->flow_counter_bulk_size; *counter_id = MLX5_GET(dealloc_flow_counter_in, devx_obj->dinbox, flow_counter_id); @@ -1904,13 +2445,13 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)( { struct mlx5_flow_context flow_context = {.flow_tag = MLX5_FS_DEFAULT_FLOW_TAG}; - u32 *offset_attr, offset = 0, counter_id = 0; int dest_id, dest_type = -1, inlen, len, ret, i; struct mlx5_ib_flow_handler *flow_handler; struct mlx5_ib_flow_matcher *fs_matcher; struct ib_uobject **arr_flow_actions; struct ib_uflow_resources *uflow_res; struct mlx5_flow_act flow_act = {}; + struct mlx5_fc *counter = NULL; struct ib_qp *qp = NULL; void *devx_obj, *cmd_in; struct ib_uobject *uobj; @@ -1937,6 +2478,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)( len = uverbs_attr_get_uobjs_arr(attrs, MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX, &arr_flow_actions); if (len) { + u32 *offset_attr, fc_bulk_size, offset = 0, counter_id = 0; devx_obj = arr_flow_actions[0]->object; if (uverbs_attr_is_valid(attrs, @@ -1956,8 +2498,11 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)( offset = *offset_attr; } - if (!is_flow_counter(devx_obj, offset, &counter_id)) + if (!is_flow_counter(devx_obj, offset, &counter_id, &fc_bulk_size)) return -EINVAL; + counter = mlx5_fc_local_create(counter_id, offset, fc_bulk_size); + if (IS_ERR(counter)) + return PTR_ERR(counter); flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_COUNT; } @@ -1968,8 +2513,10 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)( MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE); uflow_res = flow_resources_alloc(MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS); - if (!uflow_res) - return -ENOMEM; + if (!uflow_res) { + ret = -ENOMEM; + goto destroy_counter; + } len = uverbs_attr_get_uobjs_arr(attrs, MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS, &arr_flow_actions); @@ -1996,7 +2543,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)( flow_handler = raw_fs_rule_add(dev, fs_matcher, &flow_context, &flow_act, - counter_id, cmd_in, inlen, dest_id, dest_type); + counter, cmd_in, inlen, dest_id, dest_type); if (IS_ERR(flow_handler)) { ret = PTR_ERR(flow_handler); goto err_out; @@ -2007,6 +2554,9 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)( return 0; err_out: ib_uverbs_flow_resources_free(uflow_res); +destroy_counter: + if (counter) + mlx5_fc_local_destroy(counter); return ret; } @@ -2338,6 +2888,15 @@ static int mlx5_ib_matcher_ns(struct uverbs_attr_bundle *attrs, return 0; } +static bool verify_context_caps(struct mlx5_ib_dev *dev, u64 enabled_caps) +{ + if (is_mdev_switchdev_mode(dev->mdev)) + return UCAP_ENABLED(enabled_caps, + RDMA_UCAP_MLX5_CTRL_OTHER_VHCA); + + return UCAP_ENABLED(enabled_caps, RDMA_UCAP_MLX5_CTRL_LOCAL); +} + static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)( struct uverbs_attr_bundle *attrs) { @@ -2386,6 +2945,26 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)( goto end; } + if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_IB_PORT)) { + err = uverbs_copy_from(&obj->ib_port, attrs, + MLX5_IB_ATTR_FLOW_MATCHER_IB_PORT); + if (err) + goto end; + if (!rdma_is_port_valid(&dev->ib_dev, obj->ib_port)) { + err = -EINVAL; + goto end; + } + if (obj->ns_type != MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX && + obj->ns_type != MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX) { + err = -EINVAL; + goto end; + } + if (!verify_context_caps(dev, uobj->context->enabled_caps)) { + err = -EOPNOTSUPP; + goto end; + } + } + uobj->object = obj; obj->mdev = dev->mdev; atomic_set(&obj->usecnt, 0); @@ -2433,7 +3012,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_STEERING_ANCHOR_CREATE)( mutex_lock(&dev->flow_db->lock); - ft_prio = _get_flow_table(dev, priority, ns_type, 0); + ft_prio = _get_flow_table(dev, priority, ns_type, 0, 0); if (IS_ERR(ft_prio)) { err = PTR_ERR(ft_prio); goto free_obj; @@ -2819,7 +3398,10 @@ DECLARE_UVERBS_NAMED_METHOD( UA_OPTIONAL), UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE, enum mlx5_ib_uapi_flow_table_type, - UA_OPTIONAL)); + UA_OPTIONAL), + UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_FLOW_MATCHER_IB_PORT, + UVERBS_ATTR_TYPE(u32), + UA_OPTIONAL)); DECLARE_UVERBS_NAMED_METHOD_DESTROY( MLX5_IB_METHOD_FLOW_MATCHER_DESTROY, @@ -2889,8 +3471,26 @@ int mlx5_ib_fs_init(struct mlx5_ib_dev *dev) if (!dev->flow_db) return -ENOMEM; + dev->flow_db->rdma_transport_rx = kcalloc(dev->num_ports, + sizeof(struct mlx5_ib_flow_prio), + GFP_KERNEL); + if (!dev->flow_db->rdma_transport_rx) + goto free_flow_db; + + dev->flow_db->rdma_transport_tx = kcalloc(dev->num_ports, + sizeof(struct mlx5_ib_flow_prio), + GFP_KERNEL); + if (!dev->flow_db->rdma_transport_tx) + goto free_rdma_transport_rx; + mutex_init(&dev->flow_db->lock); ib_set_device_ops(&dev->ib_dev, &flow_ops); return 0; + +free_rdma_transport_rx: + kfree(dev->flow_db->rdma_transport_rx); +free_flow_db: + kfree(dev->flow_db); + return -ENOMEM; } diff --git a/drivers/infiniband/hw/mlx5/fs.h b/drivers/infiniband/hw/mlx5/fs.h index b9734904f5f0..2ebe86e5be10 100644 --- a/drivers/infiniband/hw/mlx5/fs.h +++ b/drivers/infiniband/hw/mlx5/fs.h @@ -8,23 +8,8 @@ #include "mlx5_ib.h" -#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) int mlx5_ib_fs_init(struct mlx5_ib_dev *dev); void mlx5_ib_fs_cleanup_anchor(struct mlx5_ib_dev *dev); -#else -static inline int mlx5_ib_fs_init(struct mlx5_ib_dev *dev) -{ - dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL); - - if (!dev->flow_db) - return -ENOMEM; - - mutex_init(&dev->flow_db->lock); - return 0; -} - -inline void mlx5_ib_fs_cleanup_anchor(struct mlx5_ib_dev *dev) {} -#endif static inline void mlx5_ib_fs_cleanup(struct mlx5_ib_dev *dev) { @@ -40,6 +25,8 @@ static inline void mlx5_ib_fs_cleanup(struct mlx5_ib_dev *dev) * is a safe assumption that all references are gone. */ mlx5_ib_fs_cleanup_anchor(dev); + kfree(dev->flow_db->rdma_transport_tx); + kfree(dev->flow_db->rdma_transport_rx); kfree(dev->flow_db); } #endif /* _MLX5_IB_FS_H */ diff --git a/drivers/infiniband/hw/mlx5/ib_rep.c b/drivers/infiniband/hw/mlx5/ib_rep.c index c7a4ee896121..49af1cfbe6d1 100644 --- a/drivers/infiniband/hw/mlx5/ib_rep.c +++ b/drivers/infiniband/hw/mlx5/ib_rep.c @@ -13,6 +13,7 @@ mlx5_ib_set_vport_rep(struct mlx5_core_dev *dev, int vport_index) { struct mlx5_ib_dev *ibdev; + struct net_device *ndev; ibdev = mlx5_eswitch_uplink_get_proto_dev(dev->priv.eswitch, REP_IB); if (!ibdev) @@ -20,12 +21,9 @@ mlx5_ib_set_vport_rep(struct mlx5_core_dev *dev, ibdev->port[vport_index].rep = rep; rep->rep_data[REP_IB].priv = ibdev; - write_lock(&ibdev->port[vport_index].roce.netdev_lock); - ibdev->port[vport_index].roce.netdev = - mlx5_ib_get_rep_netdev(rep->esw, rep->vport); - write_unlock(&ibdev->port[vport_index].roce.netdev_lock); + ndev = mlx5_ib_get_rep_netdev(rep->esw, rep->vport); - return 0; + return ib_device_set_netdev(&ibdev->ib_dev, ndev, vport_index + 1); } static void mlx5_ib_register_peer_vport_reps(struct mlx5_core_dev *mdev); @@ -104,10 +102,15 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep) ibdev->is_rep = true; vport_index = rep->vport_index; ibdev->port[vport_index].rep = rep; - ibdev->port[vport_index].roce.netdev = - mlx5_ib_get_rep_netdev(lag_master->priv.eswitch, rep->vport); ibdev->mdev = lag_master; ibdev->num_ports = num_ports; + ibdev->ib_dev.phys_port_cnt = num_ports; + ret = ib_device_set_netdev(&ibdev->ib_dev, + mlx5_ib_get_rep_netdev(lag_master->priv.eswitch, + rep->vport), + vport_index + 1); + if (ret) + goto fail_add; ret = __mlx5_ib_add(ibdev, profile); if (ret) @@ -160,9 +163,8 @@ mlx5_ib_vport_rep_unload(struct mlx5_eswitch_rep *rep) } port = &dev->port[vport_index]; - write_lock(&port->roce.netdev_lock); - port->roce.netdev = NULL; - write_unlock(&port->roce.netdev_lock); + + ib_device_set_netdev(&dev->ib_dev, NULL, vport_index + 1); rep->rep_data[REP_IB].priv = NULL; port->rep = NULL; diff --git a/drivers/infiniband/hw/mlx5/mad.c b/drivers/infiniband/hw/mlx5/mad.c index 0c3c4e64812c..2453ae4384a7 100644 --- a/drivers/infiniband/hw/mlx5/mad.c +++ b/drivers/infiniband/hw/mlx5/mad.c @@ -69,7 +69,7 @@ static int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, if (ignore_bkey || !in_wc) op_modifier |= 0x2; - return mlx5_cmd_mad_ifc(dev->mdev, in_mad, response_mad, op_modifier, + return mlx5_cmd_mad_ifc(dev, in_mad, response_mad, op_modifier, port); } @@ -147,8 +147,39 @@ static void pma_cnt_assign(struct ib_pma_portcounters *pma_cnt, vl_15_dropped); } -static int query_ib_ppcnt(struct mlx5_core_dev *dev, u8 port_num, void *out, - size_t sz) +static void pma_cnt_ext_assign_ppcnt(struct ib_pma_portcounters_ext *cnt_ext, + void *out) +{ + void *out_pma = MLX5_ADDR_OF(ppcnt_reg, out, + counter_set); + +#define MLX5_GET_EXT_CNTR(counter_name) \ + MLX5_GET64(ib_ext_port_cntrs_grp_data_layout, \ + out_pma, counter_name##_high) + + cnt_ext->port_xmit_data = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_xmit_data) >> 2); + cnt_ext->port_rcv_data = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_rcv_data) >> 2); + + cnt_ext->port_xmit_packets = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_xmit_pkts)); + cnt_ext->port_rcv_packets = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_rcv_pkts)); + + cnt_ext->port_unicast_xmit_packets = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_unicast_xmit_pkts)); + cnt_ext->port_unicast_rcv_packets = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_unicast_rcv_pkts)); + + cnt_ext->port_multicast_xmit_packets = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_multicast_xmit_pkts)); + cnt_ext->port_multicast_rcv_packets = + cpu_to_be64(MLX5_GET_EXT_CNTR(port_multicast_rcv_pkts)); +} + +static int query_ib_ppcnt(struct mlx5_core_dev *dev, u8 port_num, u8 plane_num, + void *out, size_t sz, bool ext) { u32 *in; int err; @@ -160,8 +191,14 @@ static int query_ib_ppcnt(struct mlx5_core_dev *dev, u8 port_num, void *out, } MLX5_SET(ppcnt_reg, in, local_port, port_num); - - MLX5_SET(ppcnt_reg, in, grp, MLX5_INFINIBAND_PORT_COUNTERS_GROUP); + MLX5_SET(ppcnt_reg, in, plane_ind, plane_num); + + if (ext) + MLX5_SET(ppcnt_reg, in, grp, + MLX5_INFINIBAND_EXTENDED_PORT_COUNTERS_GROUP); + else + MLX5_SET(ppcnt_reg, in, grp, + MLX5_INFINIBAND_PORT_COUNTERS_GROUP); err = mlx5_core_access_reg(dev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); @@ -188,7 +225,9 @@ static int process_pma_cmd(struct mlx5_ib_dev *dev, u32 port_num, mdev = dev->mdev; mdev_port_num = 1; } - if (MLX5_CAP_GEN(dev->mdev, num_ports) == 1) { + if (MLX5_CAP_GEN(dev->mdev, num_ports) == 1 && + !mlx5_core_mp_enabled(mdev) && + dev->ib_dev.type != RDMA_DEVICE_TYPE_SMI) { /* set local port to one for Function-Per-Port HCA. */ mdev = dev->mdev; mdev_port_num = 1; @@ -207,7 +246,8 @@ static int process_pma_cmd(struct mlx5_ib_dev *dev, u32 port_num, if (in_mad->mad_hdr.attr_id == IB_PMA_PORT_COUNTERS_EXT) { struct ib_pma_portcounters_ext *pma_cnt_ext = (struct ib_pma_portcounters_ext *)(out_mad->data + 40); - int sz = MLX5_ST_SZ_BYTES(query_vport_counter_out); + int sz = max(MLX5_ST_SZ_BYTES(query_vport_counter_out), + MLX5_ST_SZ_BYTES(ppcnt_reg)); out_cnt = kvzalloc(sz, GFP_KERNEL); if (!out_cnt) { @@ -215,10 +255,18 @@ static int process_pma_cmd(struct mlx5_ib_dev *dev, u32 port_num, goto done; } - err = mlx5_core_query_vport_counter(mdev, 0, 0, mdev_port_num, - out_cnt); - if (!err) - pma_cnt_ext_assign(pma_cnt_ext, out_cnt); + if (dev->ib_dev.type == RDMA_DEVICE_TYPE_SMI) { + err = query_ib_ppcnt(mdev, mdev_port_num, + port_num, out_cnt, sz, 1); + if (!err) + pma_cnt_ext_assign_ppcnt(pma_cnt_ext, out_cnt); + } else { + err = mlx5_core_query_vport_counter(mdev, 0, 0, + mdev_port_num, + out_cnt); + if (!err) + pma_cnt_ext_assign(pma_cnt_ext, out_cnt); + } } else { struct ib_pma_portcounters *pma_cnt = (struct ib_pma_portcounters *)(out_mad->data + 40); @@ -230,7 +278,13 @@ static int process_pma_cmd(struct mlx5_ib_dev *dev, u32 port_num, goto done; } - err = query_ib_ppcnt(mdev, mdev_port_num, out_cnt, sz); + if (dev->ib_dev.type == RDMA_DEVICE_TYPE_SMI) + err = query_ib_ppcnt(mdev, mdev_port_num, port_num, + out_cnt, sz, 0); + else + err = query_ib_ppcnt(mdev, mdev_port_num, 0, + out_cnt, sz, 0); + if (!err) pma_cnt_assign(pma_cnt, out_cnt); } diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index c2b557e64290..ce7610740412 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -47,7 +47,9 @@ #include <rdma/uverbs_ioctl.h> #include <rdma/mlx5_user_ioctl_verbs.h> #include <rdma/mlx5_user_ioctl_cmds.h> +#include <rdma/ib_ucaps.h> #include "macsec.h" +#include "data_direct.h" #define UVERBS_MODULE_NAME mlx5_ib #include <rdma/uverbs_named_ioctl.h> @@ -146,16 +148,52 @@ static struct mlx5_roce *mlx5_get_rep_roce(struct mlx5_ib_dev *dev, if (upper && port->rep->vport == MLX5_VPORT_UPLINK) continue; - - read_lock(&port->roce.netdev_lock); - rep_ndev = mlx5_ib_get_rep_netdev(port->rep->esw, - port->rep->vport); - if (rep_ndev == ndev) { - read_unlock(&port->roce.netdev_lock); + rep_ndev = ib_device_get_netdev(&dev->ib_dev, i + 1); + if (rep_ndev && rep_ndev == ndev) { + dev_put(rep_ndev); *port_num = i + 1; return &port->roce; } - read_unlock(&port->roce.netdev_lock); + + dev_put(rep_ndev); + } + + return NULL; +} + +static bool mlx5_netdev_send_event(struct mlx5_ib_dev *dev, + struct net_device *ndev, + struct net_device *upper, + struct net_device *ib_ndev) +{ + if (!dev->ib_active) + return false; + + /* Event is about our upper device */ + if (upper == ndev) + return true; + + /* RDMA device is not in lag and not in switchdev */ + if (!dev->is_rep && !upper && ndev == ib_ndev) + return true; + + /* RDMA devie is in switchdev */ + if (dev->is_rep && ndev == ib_ndev) + return true; + + return false; +} + +static struct net_device *mlx5_ib_get_rep_uplink_netdev(struct mlx5_ib_dev *ibdev) +{ + struct mlx5_ib_port *port; + int i; + + for (i = 0; i < ibdev->num_ports; i++) { + port = &ibdev->port[i]; + if (port->rep && port->rep->vport == MLX5_VPORT_UPLINK) { + return ib_device_get_netdev(&ibdev->ib_dev, i + 1); + } } return NULL; @@ -167,6 +205,7 @@ static int mlx5_netdev_event(struct notifier_block *this, struct mlx5_roce *roce = container_of(this, struct mlx5_roce, nb); struct net_device *ndev = netdev_notifier_info_to_dev(ptr); u32 port_num = roce->native_port_num; + struct net_device *ib_ndev = NULL; struct mlx5_core_dev *mdev; struct mlx5_ib_dev *ibdev; @@ -180,47 +219,67 @@ static int mlx5_netdev_event(struct notifier_block *this, /* Should already be registered during the load */ if (ibdev->is_rep) break; - write_lock(&roce->netdev_lock); + + ib_ndev = ib_device_get_netdev(&ibdev->ib_dev, port_num); + /* Exit if already registered */ + if (ib_ndev) + goto put_ndev; + if (ndev->dev.parent == mdev->device) - roce->netdev = ndev; - write_unlock(&roce->netdev_lock); + ib_device_set_netdev(&ibdev->ib_dev, ndev, port_num); break; case NETDEV_UNREGISTER: /* In case of reps, ib device goes away before the netdevs */ - write_lock(&roce->netdev_lock); - if (roce->netdev == ndev) - roce->netdev = NULL; - write_unlock(&roce->netdev_lock); - break; + if (ibdev->is_rep) + break; + ib_ndev = ib_device_get_netdev(&ibdev->ib_dev, port_num); + if (ib_ndev == ndev) + ib_device_set_netdev(&ibdev->ib_dev, NULL, port_num); + goto put_ndev; case NETDEV_CHANGE: case NETDEV_UP: case NETDEV_DOWN: { - struct net_device *lag_ndev = mlx5_lag_get_roce_netdev(mdev); struct net_device *upper = NULL; - if (lag_ndev) { - upper = netdev_master_upper_dev_get(lag_ndev); - dev_put(lag_ndev); + if (!netif_is_lag_master(ndev) && !netif_is_lag_port(ndev) && + !mlx5_core_mp_enabled(mdev)) + return NOTIFY_DONE; + + if (mlx5_lag_is_roce(mdev) || mlx5_lag_is_sriov(mdev)) { + struct net_device *lag_ndev; + + if(mlx5_lag_is_roce(mdev)) + lag_ndev = ib_device_get_netdev(&ibdev->ib_dev, 1); + else /* sriov lag */ + lag_ndev = mlx5_ib_get_rep_uplink_netdev(ibdev); + + if (lag_ndev) { + upper = netdev_master_upper_dev_get(lag_ndev); + dev_put(lag_ndev); + } else { + goto done; + } } if (ibdev->is_rep) roce = mlx5_get_rep_roce(ibdev, ndev, upper, &port_num); if (!roce) return NOTIFY_DONE; - if ((upper == ndev || - ((!upper || ibdev->is_rep) && ndev == roce->netdev)) && - ibdev->ib_active) { + + ib_ndev = ib_device_get_netdev(&ibdev->ib_dev, port_num); + + if (mlx5_netdev_send_event(ibdev, ndev, upper, ib_ndev)) { struct ib_event ibev = { }; enum ib_port_state port_state; if (get_port_state(&ibdev->ib_dev, port_num, &port_state)) - goto done; + goto put_ndev; if (roce->last_port_state == port_state) - goto done; + goto put_ndev; roce->last_port_state = port_state; ibev.device = &ibdev->ib_dev; @@ -229,7 +288,7 @@ static int mlx5_netdev_event(struct notifier_block *this, else if (port_state == IB_PORT_ACTIVE) ibev.event = IB_EVENT_PORT_ACTIVE; else - goto done; + goto put_ndev; ibev.element.port_num = port_num; ib_dispatch_event(&ibev); @@ -240,39 +299,13 @@ static int mlx5_netdev_event(struct notifier_block *this, default: break; } +put_ndev: + dev_put(ib_ndev); done: mlx5_ib_put_native_port_mdev(ibdev, port_num); return NOTIFY_DONE; } -static struct net_device *mlx5_ib_get_netdev(struct ib_device *device, - u32 port_num) -{ - struct mlx5_ib_dev *ibdev = to_mdev(device); - struct net_device *ndev; - struct mlx5_core_dev *mdev; - - mdev = mlx5_ib_get_native_port_mdev(ibdev, port_num, NULL); - if (!mdev) - return NULL; - - ndev = mlx5_lag_get_roce_netdev(mdev); - if (ndev) - goto out; - - /* Ensure ndev does not disappear before we invoke dev_hold() - */ - read_lock(&ibdev->port[port_num - 1].roce.netdev_lock); - ndev = ibdev->port[port_num - 1].roce.netdev; - if (ndev) - dev_hold(ndev); - read_unlock(&ibdev->port[port_num - 1].roce.netdev_lock); - -out: - mlx5_ib_put_native_port_mdev(ibdev, port_num); - return ndev; -} - struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *ibdev, u32 ib_port_num, u32 *native_port_num) @@ -283,6 +316,14 @@ struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *ibdev, struct mlx5_ib_multiport_info *mpi; struct mlx5_ib_port *port; + if (ibdev->ib_dev.type == RDMA_DEVICE_TYPE_SMI) { + if (native_port_num) + *native_port_num = smi_to_native_portnum(ibdev, + ib_port_num); + return ibdev->mdev; + + } + if (!mlx5_core_mp_enabled(ibdev->mdev) || ll != IB_LINK_LAYER_ETHERNET) { if (native_port_num) @@ -444,6 +485,10 @@ static int translate_eth_ext_proto_oper(u32 eth_proto_oper, u16 *active_speed, *active_width = IB_WIDTH_2X; *active_speed = IB_SPEED_NDR; break; + case MLX5E_PROT_MASK(MLX5E_200GAUI_1_200GBASE_CR1_KR1): + *active_width = IB_WIDTH_1X; + *active_speed = IB_SPEED_XDR; + break; case MLX5E_PROT_MASK(MLX5E_400GAUI_8_400GBASE_CR8): *active_width = IB_WIDTH_8X; *active_speed = IB_SPEED_HDR; @@ -452,10 +497,18 @@ static int translate_eth_ext_proto_oper(u32 eth_proto_oper, u16 *active_speed, *active_width = IB_WIDTH_4X; *active_speed = IB_SPEED_NDR; break; + case MLX5E_PROT_MASK(MLX5E_400GAUI_2_400GBASE_CR2_KR2): + *active_width = IB_WIDTH_2X; + *active_speed = IB_SPEED_XDR; + break; case MLX5E_PROT_MASK(MLX5E_800GAUI_8_800GBASE_CR8_KR8): *active_width = IB_WIDTH_8X; *active_speed = IB_SPEED_NDR; break; + case MLX5E_PROT_MASK(MLX5E_800GAUI_4_800GBASE_CR4_KR4): + *active_width = IB_WIDTH_4X; + *active_speed = IB_SPEED_XDR; + break; default: return -EINVAL; } @@ -504,10 +557,10 @@ static int mlx5_query_port_roce(struct ib_device *device, u32 port_num, */ if (dev->is_rep) err = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, - 1); + 1, 0); else err = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, - mdev_port_num); + mdev_port_num, 0); if (err) goto out; ext = !!MLX5_GET_ETH_PROTO(ptys_reg, out, true, eth_proto_capability); @@ -539,11 +592,11 @@ static int mlx5_query_port_roce(struct ib_device *device, u32 port_num, if (!put_mdev) goto out; - ndev = mlx5_ib_get_netdev(device, port_num); + ndev = ib_device_get_netdev(device, port_num); if (!ndev) goto out; - if (dev->lag_active) { + if (mlx5_lag_is_roce(mdev) || mlx5_lag_is_sriov(mdev)) { rcu_read_lock(); upper = netdev_master_upper_dev_get_rcu(ndev); if (upper) { @@ -1146,6 +1199,14 @@ static int mlx5_ib_query_device(struct ib_device *ibdev, MLX5_IB_QUERY_DEV_RESP_PACKET_BASED_CREDIT_MODE; resp.flags |= MLX5_IB_QUERY_DEV_RESP_FLAGS_SCAT2CQE_DCT; + + if (MLX5_CAP_GEN_2(mdev, dp_ordering_force) && + (MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_xrc) || + MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_dc) || + MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_rc) || + MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_ud) || + MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_uc))) + resp.flags |= MLX5_IB_QUERY_DEV_RESP_FLAGS_OOO_DP; } if (offsetofend(typeof(resp), sw_parsing_caps) <= uhw_outlen) { @@ -1334,11 +1395,11 @@ static int mlx5_query_hca_port(struct ib_device *ibdev, u32 port, struct mlx5_ib_dev *dev = to_mdev(ibdev); struct mlx5_core_dev *mdev = dev->mdev; struct mlx5_hca_vport_context *rep; + u8 vl_hw_cap, plane_index = 0; u16 max_mtu; u16 oper_mtu; int err; u16 ib_link_width_oper; - u8 vl_hw_cap; rep = kzalloc(sizeof(*rep), GFP_KERNEL); if (!rep) { @@ -1348,6 +1409,11 @@ static int mlx5_query_hca_port(struct ib_device *ibdev, u32 port, /* props being zeroed by the caller, avoid zeroing it here */ + if (ibdev->type == RDMA_DEVICE_TYPE_SMI) { + plane_index = port; + port = smi_to_native_portnum(dev, port); + } + err = mlx5_query_hca_vport_context(mdev, 0, port, 0, rep); if (err) goto out; @@ -1358,7 +1424,14 @@ static int mlx5_query_hca_port(struct ib_device *ibdev, u32 port, props->sm_sl = rep->sm_sl; props->state = rep->vport_state; props->phys_state = rep->port_physical_state; - props->port_cap_flags = rep->cap_mask1; + + props->port_cap_flags = rep->cap_mask1; + if (dev->num_plane) { + props->port_cap_flags |= IB_PORT_SM_DISABLED; + props->port_cap_flags &= ~IB_PORT_SM; + } else if (ibdev->type == RDMA_DEVICE_TYPE_SMI) + props->port_cap_flags &= ~IB_PORT_CM_SUP; + props->gid_tbl_len = mlx5_get_gid_table_len(MLX5_CAP_GEN(mdev, gid_table_size)); props->max_msg_sz = 1 << MLX5_CAP_GEN(mdev, log_max_msg); props->pkey_tbl_len = mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(mdev, pkey_table_size)); @@ -1371,7 +1444,7 @@ static int mlx5_query_hca_port(struct ib_device *ibdev, u32 port, props->port_cap_flags2 = rep->cap_mask2; err = mlx5_query_ib_port_oper(mdev, &ib_link_width_oper, - &props->active_speed, port); + &props->active_speed, port, plane_index); if (err) goto out; @@ -1811,7 +1884,7 @@ static int set_ucontext_resp(struct ib_ucontext *uctx, } resp->qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp); - if (dev->wc_support) + if (mlx5_wc_support_get(dev->mdev)) resp->bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev, log_bf_reg_size); resp->cache_line_size = cache_line_size(); @@ -1874,6 +1947,12 @@ static int set_ucontext_resp(struct ib_ucontext *uctx, return 0; } +static bool uctx_rdma_ctrl_is_enabled(u64 enabled_caps) +{ + return UCAP_ENABLED(enabled_caps, RDMA_UCAP_MLX5_CTRL_LOCAL) || + UCAP_ENABLED(enabled_caps, RDMA_UCAP_MLX5_CTRL_OTHER_VHCA); +} + static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { @@ -1916,10 +1995,17 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx, return -EINVAL; if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) { - err = mlx5_ib_devx_create(dev, true); + err = mlx5_ib_devx_create(dev, true, uctx->enabled_caps); if (err < 0) goto out_ctx; context->devx_uid = err; + + if (uctx_rdma_ctrl_is_enabled(uctx->enabled_caps)) { + err = mlx5_cmd_add_privileged_uid(dev->mdev, + context->devx_uid); + if (err) + goto out_devx; + } } lib_uar_4k = req.lib_caps & MLX5_LIB_CAP_4K_UAR; @@ -1934,7 +2020,7 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx, /* updates req->total_num_bfregs */ err = calc_total_bfregs(dev, lib_uar_4k, &req, bfregi); if (err) - goto out_devx; + goto out_ucap; mutex_init(&bfregi->lock); bfregi->lib_uar_4k = lib_uar_4k; @@ -1942,7 +2028,7 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx, GFP_KERNEL); if (!bfregi->count) { err = -ENOMEM; - goto out_devx; + goto out_ucap; } bfregi->sys_pages = kcalloc(bfregi->num_sys_pages, @@ -2006,6 +2092,11 @@ out_sys_pages: out_count: kfree(bfregi->count); +out_ucap: + if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX && + uctx_rdma_ctrl_is_enabled(uctx->enabled_caps)) + mlx5_cmd_remove_privileged_uid(dev->mdev, context->devx_uid); + out_devx: if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) mlx5_ib_devx_destroy(dev, context->devx_uid); @@ -2050,8 +2141,12 @@ static void mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) kfree(bfregi->sys_pages); kfree(bfregi->count); - if (context->devx_uid) + if (context->devx_uid) { + if (uctx_rdma_ctrl_is_enabled(ibcontext->enabled_caps)) + mlx5_cmd_remove_privileged_uid(dev->mdev, + context->devx_uid); mlx5_ib_devx_destroy(dev, context->devx_uid); + } } static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev, @@ -2338,7 +2433,7 @@ static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vm switch (command) { case MLX5_IB_MMAP_WC_PAGE: case MLX5_IB_MMAP_ALLOC_WC: - if (!dev->wc_support) + if (!mlx5_wc_support_get(dev->mdev)) return -EPERM; fallthrough; case MLX5_IB_MMAP_NC_PAGE: @@ -2777,6 +2872,23 @@ static int mlx5_ib_event_slave_port(struct notifier_block *nb, return NOTIFY_OK; } +static int mlx5_ib_get_plane_num(struct mlx5_core_dev *mdev, u8 *num_plane) +{ + struct mlx5_hca_vport_context vport_ctx; + int err; + + *num_plane = 0; + if (!MLX5_CAP_GEN(mdev, ib_virt) || !MLX5_CAP_GEN_2(mdev, multiplane)) + return 0; + + err = mlx5_query_hca_vport_context(mdev, 0, 1, 0, &vport_ctx); + if (err) + return err; + + *num_plane = vport_ctx.num_plane; + return 0; +} + static int set_has_smi_cap(struct mlx5_ib_dev *dev) { struct mlx5_hca_vport_context vport_ctx; @@ -2787,10 +2899,15 @@ static int set_has_smi_cap(struct mlx5_ib_dev *dev) return 0; for (port = 1; port <= dev->num_ports; port++) { - if (!MLX5_CAP_GEN(dev->mdev, ib_virt)) { + if (dev->num_plane) { + dev->port_caps[port - 1].has_smi = false; + continue; + } else if (!MLX5_CAP_GEN(dev->mdev, ib_virt) || + dev->ib_dev.type == RDMA_DEVICE_TYPE_SMI) { dev->port_caps[port - 1].has_smi = true; continue; } + err = mlx5_query_hca_vport_context(dev->mdev, 0, port, 0, &vport_ctx); if (err) { @@ -2824,37 +2941,72 @@ static u8 mlx5_get_umr_fence(u8 umr_fence_cap) } } -static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev) +int mlx5_ib_dev_res_cq_init(struct mlx5_ib_dev *dev) { struct mlx5_ib_resources *devr = &dev->devr; - struct ib_srq_init_attr attr; - struct ib_device *ibdev; struct ib_cq_init_attr cq_attr = {.cqe = 1}; - int port; + struct ib_device *ibdev; + struct ib_pd *pd; + struct ib_cq *cq; int ret = 0; - ibdev = &dev->ib_dev; - if (!MLX5_CAP_GEN(dev->mdev, xrc)) - return -EOPNOTSUPP; + /* + * devr->c0 is set once, never changed until device unload. + * Avoid taking the mutex if initialization is already done. + */ + if (devr->c0) + return 0; - devr->p0 = ib_alloc_pd(ibdev, 0); - if (IS_ERR(devr->p0)) - return PTR_ERR(devr->p0); + mutex_lock(&devr->cq_lock); + if (devr->c0) + goto unlock; - devr->c0 = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr); - if (IS_ERR(devr->c0)) { - ret = PTR_ERR(devr->c0); - goto error1; + ibdev = &dev->ib_dev; + pd = ib_alloc_pd(ibdev, 0); + if (IS_ERR(pd)) { + ret = PTR_ERR(pd); + mlx5_ib_err(dev, "Couldn't allocate PD for res init, err=%d\n", ret); + goto unlock; } - ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn0, 0); - if (ret) - goto error2; + cq = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr); + if (IS_ERR(cq)) { + ret = PTR_ERR(cq); + mlx5_ib_err(dev, "Couldn't create CQ for res init, err=%d\n", ret); + ib_dealloc_pd(pd); + goto unlock; + } - ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn1, 0); + devr->p0 = pd; + devr->c0 = cq; + +unlock: + mutex_unlock(&devr->cq_lock); + return ret; +} + +int mlx5_ib_dev_res_srq_init(struct mlx5_ib_dev *dev) +{ + struct mlx5_ib_resources *devr = &dev->devr; + struct ib_srq_init_attr attr; + struct ib_srq *s0, *s1; + int ret = 0; + + /* + * devr->s1 is set once, never changed until device unload. + * Avoid taking the mutex if initialization is already done. + */ + if (devr->s1) + return 0; + + mutex_lock(&devr->srq_lock); + if (devr->s1) + goto unlock; + + ret = mlx5_ib_dev_res_cq_init(dev); if (ret) - goto error3; + goto unlock; memset(&attr, 0, sizeof(attr)); attr.attr.max_sge = 1; @@ -2862,10 +3014,11 @@ static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev) attr.srq_type = IB_SRQT_XRC; attr.ext.cq = devr->c0; - devr->s0 = ib_create_srq(devr->p0, &attr); - if (IS_ERR(devr->s0)) { - ret = PTR_ERR(devr->s0); - goto err_create; + s0 = ib_create_srq(devr->p0, &attr); + if (IS_ERR(s0)) { + ret = PTR_ERR(s0); + mlx5_ib_err(dev, "Couldn't create SRQ 0 for res init, err=%d\n", ret); + goto unlock; } memset(&attr, 0, sizeof(attr)); @@ -2873,51 +3026,116 @@ static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev) attr.attr.max_wr = 1; attr.srq_type = IB_SRQT_BASIC; - devr->s1 = ib_create_srq(devr->p0, &attr); - if (IS_ERR(devr->s1)) { - ret = PTR_ERR(devr->s1); - goto error6; + s1 = ib_create_srq(devr->p0, &attr); + if (IS_ERR(s1)) { + ret = PTR_ERR(s1); + mlx5_ib_err(dev, "Couldn't create SRQ 1 for res init, err=%d\n", ret); + ib_destroy_srq(s0); } - for (port = 0; port < ARRAY_SIZE(devr->ports); ++port) - INIT_WORK(&devr->ports[port].pkey_change_work, - pkey_change_handler); - - return 0; + devr->s0 = s0; + devr->s1 = s1; -error6: - ib_destroy_srq(devr->s0); -err_create: - mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn1, 0); -error3: - mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0); -error2: - ib_destroy_cq(devr->c0); -error1: - ib_dealloc_pd(devr->p0); +unlock: + mutex_unlock(&devr->srq_lock); return ret; } -static void mlx5_ib_dev_res_cleanup(struct mlx5_ib_dev *dev) +static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev) { struct mlx5_ib_resources *devr = &dev->devr; - int port; + int ret; - /* - * Make sure no change P_Key work items are still executing. - * - * At this stage, the mlx5_ib_event should be unregistered - * and it ensures that no new works are added. - */ - for (port = 0; port < ARRAY_SIZE(devr->ports); ++port) - cancel_work_sync(&devr->ports[port].pkey_change_work); + if (!MLX5_CAP_GEN(dev->mdev, xrc)) + return -EOPNOTSUPP; + + ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn0, 0); + if (ret) + return ret; + + ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn1, 0); + if (ret) { + mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0); + return ret; + } + + mutex_init(&devr->cq_lock); + mutex_init(&devr->srq_lock); - ib_destroy_srq(devr->s1); - ib_destroy_srq(devr->s0); + return 0; +} + +static void mlx5_ib_dev_res_cleanup(struct mlx5_ib_dev *dev) +{ + struct mlx5_ib_resources *devr = &dev->devr; + + /* After s0/s1 init, they are not unset during the device lifetime. */ + if (devr->s1) { + ib_destroy_srq(devr->s1); + ib_destroy_srq(devr->s0); + } mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn1, 0); mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0); - ib_destroy_cq(devr->c0); - ib_dealloc_pd(devr->p0); + /* After p0/c0 init, they are not unset during the device lifetime. */ + if (devr->c0) { + ib_destroy_cq(devr->c0); + ib_dealloc_pd(devr->p0); + } + mutex_destroy(&devr->cq_lock); + mutex_destroy(&devr->srq_lock); +} + +static int +mlx5_ib_create_data_direct_resources(struct mlx5_ib_dev *dev) +{ + int inlen = MLX5_ST_SZ_BYTES(create_mkey_in); + struct mlx5_core_dev *mdev = dev->mdev; + void *mkc; + u32 mkey; + u32 pdn; + u32 *in; + int err; + + err = mlx5_core_alloc_pd(mdev, &pdn); + if (err) + return err; + + in = kvzalloc(inlen, GFP_KERNEL); + if (!in) { + err = -ENOMEM; + goto err; + } + + MLX5_SET(create_mkey_in, in, data_direct, 1); + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_PA); + MLX5_SET(mkc, mkc, lw, 1); + MLX5_SET(mkc, mkc, lr, 1); + MLX5_SET(mkc, mkc, rw, 1); + MLX5_SET(mkc, mkc, rr, 1); + MLX5_SET(mkc, mkc, a, 1); + MLX5_SET(mkc, mkc, pd, pdn); + MLX5_SET(mkc, mkc, length64, 1); + MLX5_SET(mkc, mkc, qpn, 0xffffff); + err = mlx5_core_create_mkey(mdev, &mkey, in, inlen); + kvfree(in); + if (err) + goto err; + + dev->ddr.mkey = mkey; + dev->ddr.pdn = pdn; + return 0; + +err: + mlx5_core_dealloc_pd(mdev, pdn); + return err; +} + +static void +mlx5_ib_free_data_direct_resources(struct mlx5_ib_dev *dev) +{ + mlx5_core_destroy_mkey(dev->mdev, dev->ddr.mkey); + mlx5_core_dealloc_pd(dev->mdev, dev->ddr.pdn); } static u32 get_core_cap_flags(struct ib_device *ibdev, @@ -2933,6 +3151,13 @@ static u32 get_core_cap_flags(struct ib_device *ibdev, if (rep->grh_required) ret |= RDMA_CORE_CAP_IB_GRH_REQUIRED; + if (dev->num_plane) + return ret | RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_IB_MAD | + RDMA_CORE_CAP_IB_CM | RDMA_CORE_CAP_IB_SA | + RDMA_CORE_CAP_AF_IB; + else if (ibdev->type == RDMA_DEVICE_TYPE_SMI) + return ret | RDMA_CORE_CAP_IB_MAD | RDMA_CORE_CAP_IB_SMI; + if (ll == IB_LINK_LAYER_INFINIBAND) return ret | RDMA_CORE_PORT_IBA_IB; @@ -2968,6 +3193,9 @@ static int mlx5_port_immutable(struct ib_device *ibdev, u32 port_num, return err; if (ll == IB_LINK_LAYER_INFINIBAND) { + if (ibdev->type == RDMA_DEVICE_TYPE_SMI) + port_num = smi_to_native_portnum(dev, port_num); + err = mlx5_query_hca_vport_context(dev->mdev, 0, port_num, 0, &rep); if (err) @@ -3010,6 +3238,67 @@ static void get_dev_fw_str(struct ib_device *ibdev, char *str) fw_rev_sub(dev->mdev)); } +static int lag_event(struct notifier_block *nb, unsigned long event, void *data) +{ + struct mlx5_ib_dev *dev = container_of(nb, struct mlx5_ib_dev, + lag_events); + struct mlx5_core_dev *mdev = dev->mdev; + struct ib_device *ibdev = &dev->ib_dev; + struct net_device *old_ndev = NULL; + struct mlx5_ib_port *port; + struct net_device *ndev; + u32 portnum = 0; + int ret = 0; + int i; + + switch (event) { + case MLX5_DRIVER_EVENT_ACTIVE_BACKUP_LAG_CHANGE_LOWERSTATE: + ndev = data; + if (ndev) { + if (!mlx5_lag_is_roce(mdev)) { + // sriov lag + for (i = 0; i < dev->num_ports; i++) { + port = &dev->port[i]; + if (port->rep && port->rep->vport == + MLX5_VPORT_UPLINK) { + portnum = i; + break; + } + } + } + old_ndev = ib_device_get_netdev(ibdev, portnum + 1); + ret = ib_device_set_netdev(ibdev, ndev, portnum + 1); + if (ret) + goto out; + + if (old_ndev) + roce_del_all_netdev_gids(ibdev, portnum + 1, + old_ndev); + rdma_roce_rescan_port(ibdev, portnum + 1); + } + break; + default: + return NOTIFY_DONE; + } + +out: + dev_put(old_ndev); + return notifier_from_errno(ret); +} + +static void mlx5e_lag_event_register(struct mlx5_ib_dev *dev) +{ + dev->lag_events.notifier_call = lag_event; + blocking_notifier_chain_register(&dev->mdev->priv.lag_nh, + &dev->lag_events); +} + +static void mlx5e_lag_event_unregister(struct mlx5_ib_dev *dev) +{ + blocking_notifier_chain_unregister(&dev->mdev->priv.lag_nh, + &dev->lag_events); +} + static int mlx5_eth_lag_init(struct mlx5_ib_dev *dev) { struct mlx5_core_dev *mdev = dev->mdev; @@ -3031,6 +3320,7 @@ static int mlx5_eth_lag_init(struct mlx5_ib_dev *dev) goto err_destroy_vport_lag; } + mlx5e_lag_event_register(dev); dev->flow_db->lag_demux_ft = ft; dev->lag_ports = mlx5_lag_get_num_ports(mdev); dev->lag_active = true; @@ -3048,6 +3338,7 @@ static void mlx5_eth_lag_cleanup(struct mlx5_ib_dev *dev) if (dev->lag_active) { dev->lag_active = false; + mlx5e_lag_event_unregister(dev); mlx5_destroy_flow_table(dev->flow_db->lag_demux_ft); dev->flow_db->lag_demux_ft = NULL; @@ -3306,6 +3597,41 @@ unbind: return false; } +static int mlx5_ib_data_direct_init(struct mlx5_ib_dev *dev) +{ + char vuid[MLX5_ST_SZ_BYTES(array1024_auto) + 1] = {}; + int ret; + + if (!MLX5_CAP_GEN(dev->mdev, data_direct) || + !MLX5_CAP_GEN_2(dev->mdev, query_vuid)) + return 0; + + ret = mlx5_cmd_query_vuid(dev->mdev, true, vuid); + if (ret) + return ret; + + ret = mlx5_ib_create_data_direct_resources(dev); + if (ret) + return ret; + + INIT_LIST_HEAD(&dev->data_direct_mr_list); + ret = mlx5_data_direct_ib_reg(dev, vuid); + if (ret) + mlx5_ib_free_data_direct_resources(dev); + + return ret; +} + +static void mlx5_ib_data_direct_cleanup(struct mlx5_ib_dev *dev) +{ + if (!MLX5_CAP_GEN(dev->mdev, data_direct) || + !MLX5_CAP_GEN_2(dev->mdev, query_vuid)) + return; + + mlx5_data_direct_ib_unreg(dev); + mlx5_ib_free_data_direct_resources(dev); +} + static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev) { u32 port_num = mlx5_core_native_port_num(dev->mdev) - 1; @@ -3352,7 +3678,8 @@ static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev) list_for_each_entry(mpi, &mlx5_ib_unaffiliated_port_list, list) { if (dev->sys_image_guid == mpi->sys_image_guid && - (mlx5_core_native_port_num(mpi->mdev) - 1) == i) { + (mlx5_core_native_port_num(mpi->mdev) - 1) == i && + mlx5_core_same_coredev_type(dev->mdev, mpi->mdev)) { bound = mlx5_ib_bind_slave_port(dev, mpi); } @@ -3613,7 +3940,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_UAR_OBJ_ALLOC)( alloc_type != MLX5_IB_UAPI_UAR_ALLOC_TYPE_NC) return -EOPNOTSUPP; - if (!to_mdev(c->ibucontext.device)->wc_support && + if (!mlx5_wc_support_get(to_mdev(c->ibucontext.device)->mdev) && alloc_type == MLX5_IB_UAPI_UAR_ALLOC_TYPE_BF) return -EOPNOTSUPP; @@ -3682,14 +4009,24 @@ ADD_UVERBS_ATTRIBUTES_SIMPLE( dump_fill_mkey), UA_MANDATORY)); +ADD_UVERBS_ATTRIBUTES_SIMPLE( + mlx5_ib_reg_dmabuf_mr, + UVERBS_OBJECT_MR, + UVERBS_METHOD_REG_DMABUF_MR, + UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + enum mlx5_ib_uapi_reg_dmabuf_flags, + UA_OPTIONAL)); + static const struct uapi_definition mlx5_ib_defs[] = { UAPI_DEF_CHAIN(mlx5_ib_devx_defs), UAPI_DEF_CHAIN(mlx5_ib_flow_defs), UAPI_DEF_CHAIN(mlx5_ib_qos_defs), UAPI_DEF_CHAIN(mlx5_ib_std_types_defs), UAPI_DEF_CHAIN(mlx5_ib_dm_defs), + UAPI_DEF_CHAIN(mlx5_ib_create_cq_defs), UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_DEVICE, &mlx5_ib_query_context), + UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_MR, &mlx5_ib_reg_dmabuf_mr), UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_VAR, UAPI_DEF_IS_OBJ_SUPPORTED(var_is_supported)), UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_UAR), @@ -3698,6 +4035,7 @@ static const struct uapi_definition mlx5_ib_defs[] = { static void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev) { + mlx5_ib_data_direct_cleanup(dev); mlx5_ib_cleanup_multiport_master(dev); WARN_ON(!xa_empty(&dev->odp_mkeys)); mutex_destroy(&dev->cap_mask_mutex); @@ -3713,13 +4051,11 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) dev->ib_dev.node_type = RDMA_NODE_IB_CA; dev->ib_dev.local_dma_lkey = 0 /* not supported for now */; - dev->ib_dev.phys_port_cnt = dev->num_ports; dev->ib_dev.dev.parent = mdev->device; dev->ib_dev.lag_flags = RDMA_LAG_FLAGS_HASH_ALL_SLAVES; for (i = 0; i < dev->num_ports; i++) { spin_lock_init(&dev->port[i].mp.mpi_lock); - rwlock_init(&dev->port[i].roce.netdev_lock); dev->port[i].roce.dev = dev; dev->port[i].roce.native_port_num = i + 1; dev->port[i].roce.last_port_state = IB_PORT_DOWN; @@ -3751,6 +4087,7 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) dev->ib_dev.num_comp_vectors = mlx5_comp_vectors_max(mdev); mutex_init(&dev->cap_mask_mutex); + mutex_init(&dev->data_direct_lock); INIT_LIST_HEAD(&dev->qp_list); spin_lock_init(&dev->reset_flow_resource_lock); xa_init(&dev->odp_mkeys); @@ -3759,25 +4096,22 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) spin_lock_init(&dev->dm.lock); dev->dm.dev = mdev; + err = mlx5_ib_data_direct_init(dev); + if (err) + goto err_mp; + return 0; -err: - mlx5r_macsec_dealloc_gids(dev); err_mp: mlx5_ib_cleanup_multiport_master(dev); +err: + mlx5r_macsec_dealloc_gids(dev); return err; } -static int mlx5_ib_enable_driver(struct ib_device *dev) -{ - struct mlx5_ib_dev *mdev = to_mdev(dev); - int ret; - - ret = mlx5_ib_test_wc(mdev); - mlx5_ib_dbg(mdev, "Write-Combining %s", - mdev->wc_support ? "supported" : "not supported"); - - return ret; -} +static struct ib_device *mlx5_ib_add_sub_dev(struct ib_device *parent, + enum rdma_nl_dev_type type, + const char *name); +static void mlx5_ib_del_sub_dev(struct ib_device *sub_dev); static const struct ib_device_ops mlx5_ib_dev_ops = { .owner = THIS_MODULE, @@ -3785,6 +4119,7 @@ static const struct ib_device_ops mlx5_ib_dev_ops = { .uverbs_abi_ver = MLX5_IB_UVERBS_ABI_VERSION, .add_gid = mlx5_ib_add_gid, + .add_sub_dev = mlx5_ib_add_sub_dev, .alloc_mr = mlx5_ib_alloc_mr, .alloc_mr_integrity = mlx5_ib_alloc_mr_integrity, .alloc_pd = mlx5_ib_alloc_pd, @@ -3799,6 +4134,7 @@ static const struct ib_device_ops mlx5_ib_dev_ops = { .dealloc_pd = mlx5_ib_dealloc_pd, .dealloc_ucontext = mlx5_ib_dealloc_ucontext, .del_gid = mlx5_ib_del_gid, + .del_sub_dev = mlx5_ib_del_sub_dev, .dereg_mr = mlx5_ib_dereg_mr, .destroy_ah = mlx5_ib_destroy_ah, .destroy_cq = mlx5_ib_destroy_cq, @@ -3809,7 +4145,6 @@ static const struct ib_device_ops mlx5_ib_dev_ops = { .drain_rq = mlx5_ib_drain_rq, .drain_sq = mlx5_ib_drain_sq, .device_group = &mlx5_attr_group, - .enable_driver = mlx5_ib_enable_driver, .get_dev_fw_str = get_dev_fw_str, .get_dma_mr = mlx5_ib_get_dma_mr, .get_link_layer = mlx5_ib_port_link_layer, @@ -3839,6 +4174,7 @@ static const struct ib_device_ops mlx5_ib_dev_ops = { .req_notify_cq = mlx5_ib_arm_cq, .rereg_user_mr = mlx5_ib_rereg_user_mr, .resize_cq = mlx5_ib_resize_cq, + .ufile_hw_cleanup = mlx5_ib_ufile_hw_cleanup, INIT_RDMA_OBJ_SIZE(ib_ah, mlx5_ib_ah, ibah), INIT_RDMA_OBJ_SIZE(ib_counters, mlx5_ib_mcounters, ibcntrs), @@ -3900,8 +4236,47 @@ static int mlx5_ib_init_var_table(struct mlx5_ib_dev *dev) return (var_table->bitmap) ? 0 : -ENOMEM; } +static void mlx5_ib_cleanup_ucaps(struct mlx5_ib_dev *dev) +{ + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RDMA_CTRL) + ib_remove_ucap(RDMA_UCAP_MLX5_CTRL_LOCAL); + + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & + MLX5_UCTX_CAP_RDMA_CTRL_OTHER_VHCA) + ib_remove_ucap(RDMA_UCAP_MLX5_CTRL_OTHER_VHCA); +} + +static int mlx5_ib_init_ucaps(struct mlx5_ib_dev *dev) +{ + int ret; + + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RDMA_CTRL) { + ret = ib_create_ucap(RDMA_UCAP_MLX5_CTRL_LOCAL); + if (ret) + return ret; + } + + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & + MLX5_UCTX_CAP_RDMA_CTRL_OTHER_VHCA) { + ret = ib_create_ucap(RDMA_UCAP_MLX5_CTRL_OTHER_VHCA); + if (ret) + goto remove_local; + } + + return 0; + +remove_local: + if (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RDMA_CTRL) + ib_remove_ucap(RDMA_UCAP_MLX5_CTRL_LOCAL); + return ret; +} + static void mlx5_ib_stage_caps_cleanup(struct mlx5_ib_dev *dev) { + if (MLX5_CAP_GEN_2_64(dev->mdev, general_obj_types_127_64) & + MLX5_HCA_CAP_2_GENERAL_OBJECT_TYPES_RDMA_CTRL) + mlx5_ib_cleanup_ucaps(dev); + bitmap_free(dev->var_table.bitmap); } @@ -3952,6 +4327,13 @@ static int mlx5_ib_stage_caps_init(struct mlx5_ib_dev *dev) return err; } + if (MLX5_CAP_GEN_2_64(dev->mdev, general_obj_types_127_64) & + MLX5_HCA_CAP_2_GENERAL_OBJECT_TYPES_RDMA_CTRL) { + err = mlx5_ib_init_ucaps(dev); + if (err) + return err; + } + dev->ib_dev.use_cq_dim = true; return 0; @@ -3985,7 +4367,6 @@ static const struct ib_device_ops mlx5_ib_dev_common_roce_ops = { .create_wq = mlx5_ib_create_wq, .destroy_rwq_ind_table = mlx5_ib_destroy_rwq_ind_table, .destroy_wq = mlx5_ib_destroy_wq, - .get_netdev = mlx5_ib_get_netdev, .modify_wq = mlx5_ib_modify_wq, INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx5_ib_rwq_ind_table, @@ -4053,17 +4434,6 @@ static void mlx5_ib_stage_cong_debugfs_cleanup(struct mlx5_ib_dev *dev) mlx5_core_native_port_num(dev->mdev) - 1); } -static int mlx5_ib_stage_uar_init(struct mlx5_ib_dev *dev) -{ - dev->mdev->priv.uar = mlx5_get_uars_page(dev->mdev); - return PTR_ERR_OR_ZERO(dev->mdev->priv.uar); -} - -static void mlx5_ib_stage_uar_cleanup(struct mlx5_ib_dev *dev) -{ - mlx5_put_uars_page(dev->mdev, dev->mdev->priv.uar); -} - static int mlx5_ib_stage_bfrag_init(struct mlx5_ib_dev *dev) { int err; @@ -4089,7 +4459,10 @@ static int mlx5_ib_stage_ib_reg_init(struct mlx5_ib_dev *dev) { const char *name; - if (!mlx5_lag_is_active(dev->mdev)) + if (dev->sub_dev_name) { + name = dev->sub_dev_name; + ib_mark_name_assigned_by_user(&dev->ib_dev); + } else if (!mlx5_lag_is_active(dev->mdev)) name = "mlx5_%d"; else name = "mlx5_bond_%d"; @@ -4100,6 +4473,7 @@ static void mlx5_ib_stage_pre_ib_reg_umr_cleanup(struct mlx5_ib_dev *dev) { mlx5_mkey_cache_cleanup(dev); mlx5r_umr_resource_cleanup(dev); + mlx5r_umr_cleanup(dev); } static void mlx5_ib_stage_ib_reg_cleanup(struct mlx5_ib_dev *dev) @@ -4111,7 +4485,7 @@ static int mlx5_ib_stage_post_ib_reg_umr_init(struct mlx5_ib_dev *dev) { int ret; - ret = mlx5r_umr_resource_init(dev); + ret = mlx5r_umr_init(dev); if (ret) return ret; @@ -4166,6 +4540,13 @@ static void mlx5_ib_stage_delay_drop_cleanup(struct mlx5_ib_dev *dev) static int mlx5_ib_stage_dev_notifier_init(struct mlx5_ib_dev *dev) { + struct mlx5_ib_resources *devr = &dev->devr; + int port; + + for (port = 0; port < ARRAY_SIZE(devr->ports); ++port) + INIT_WORK(&devr->ports[port].pkey_change_work, + pkey_change_handler); + dev->mdev_events.notifier_call = mlx5_ib_event; mlx5_notifier_register(dev->mdev, &dev->mdev_events); @@ -4176,8 +4557,30 @@ static int mlx5_ib_stage_dev_notifier_init(struct mlx5_ib_dev *dev) static void mlx5_ib_stage_dev_notifier_cleanup(struct mlx5_ib_dev *dev) { + struct mlx5_ib_resources *devr = &dev->devr; + int port; + mlx5r_macsec_event_unregister(dev); mlx5_notifier_unregister(dev->mdev, &dev->mdev_events); + + for (port = 0; port < ARRAY_SIZE(devr->ports); ++port) + cancel_work_sync(&devr->ports[port].pkey_change_work); +} + +void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev, + struct mlx5_data_direct_dev *dev) +{ + mutex_lock(&ibdev->data_direct_lock); + ibdev->data_direct_dev = dev; + mutex_unlock(&ibdev->data_direct_lock); +} + +void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev) +{ + mutex_lock(&ibdev->data_direct_lock); + mlx5_ib_revoke_data_direct_mrs(ibdev); + ibdev->data_direct_dev = NULL; + mutex_unlock(&ibdev->data_direct_lock); } void __mlx5_ib_remove(struct mlx5_ib_dev *dev, @@ -4251,9 +4654,6 @@ static const struct mlx5_ib_profile pf_profile = { STAGE_CREATE(MLX5_IB_STAGE_DEVICE_RESOURCES, mlx5_ib_dev_res_init, mlx5_ib_dev_res_cleanup), - STAGE_CREATE(MLX5_IB_STAGE_DEVICE_NOTIFIER, - mlx5_ib_stage_dev_notifier_init, - mlx5_ib_stage_dev_notifier_cleanup), STAGE_CREATE(MLX5_IB_STAGE_ODP, mlx5_ib_odp_init_one, mlx5_ib_odp_cleanup_one), @@ -4263,9 +4663,6 @@ static const struct mlx5_ib_profile pf_profile = { STAGE_CREATE(MLX5_IB_STAGE_CONG_DEBUGFS, mlx5_ib_stage_cong_debugfs_init, mlx5_ib_stage_cong_debugfs_cleanup), - STAGE_CREATE(MLX5_IB_STAGE_UAR, - mlx5_ib_stage_uar_init, - mlx5_ib_stage_uar_cleanup), STAGE_CREATE(MLX5_IB_STAGE_BFREG, mlx5_ib_stage_bfrag_init, mlx5_ib_stage_bfrag_cleanup), @@ -4278,6 +4675,9 @@ static const struct mlx5_ib_profile pf_profile = { STAGE_CREATE(MLX5_IB_STAGE_IB_REG, mlx5_ib_stage_ib_reg_init, mlx5_ib_stage_ib_reg_cleanup), + STAGE_CREATE(MLX5_IB_STAGE_DEVICE_NOTIFIER, + mlx5_ib_stage_dev_notifier_init, + mlx5_ib_stage_dev_notifier_cleanup), STAGE_CREATE(MLX5_IB_STAGE_POST_IB_REG_UMR, mlx5_ib_stage_post_ib_reg_umr_init, NULL), @@ -4314,18 +4714,12 @@ const struct mlx5_ib_profile raw_eth_profile = { STAGE_CREATE(MLX5_IB_STAGE_DEVICE_RESOURCES, mlx5_ib_dev_res_init, mlx5_ib_dev_res_cleanup), - STAGE_CREATE(MLX5_IB_STAGE_DEVICE_NOTIFIER, - mlx5_ib_stage_dev_notifier_init, - mlx5_ib_stage_dev_notifier_cleanup), STAGE_CREATE(MLX5_IB_STAGE_COUNTERS, mlx5_ib_counters_init, mlx5_ib_counters_cleanup), STAGE_CREATE(MLX5_IB_STAGE_CONG_DEBUGFS, mlx5_ib_stage_cong_debugfs_init, mlx5_ib_stage_cong_debugfs_cleanup), - STAGE_CREATE(MLX5_IB_STAGE_UAR, - mlx5_ib_stage_uar_init, - mlx5_ib_stage_uar_cleanup), STAGE_CREATE(MLX5_IB_STAGE_BFREG, mlx5_ib_stage_bfrag_init, mlx5_ib_stage_bfrag_cleanup), @@ -4338,6 +4732,9 @@ const struct mlx5_ib_profile raw_eth_profile = { STAGE_CREATE(MLX5_IB_STAGE_IB_REG, mlx5_ib_stage_ib_reg_init, mlx5_ib_stage_ib_reg_cleanup), + STAGE_CREATE(MLX5_IB_STAGE_DEVICE_NOTIFIER, + mlx5_ib_stage_dev_notifier_init, + mlx5_ib_stage_dev_notifier_cleanup), STAGE_CREATE(MLX5_IB_STAGE_POST_IB_REG_UMR, mlx5_ib_stage_post_ib_reg_umr_init, NULL), @@ -4349,6 +4746,90 @@ const struct mlx5_ib_profile raw_eth_profile = { NULL), }; +static const struct mlx5_ib_profile plane_profile = { + STAGE_CREATE(MLX5_IB_STAGE_INIT, + mlx5_ib_stage_init_init, + mlx5_ib_stage_init_cleanup), + STAGE_CREATE(MLX5_IB_STAGE_CAPS, + mlx5_ib_stage_caps_init, + mlx5_ib_stage_caps_cleanup), + STAGE_CREATE(MLX5_IB_STAGE_NON_DEFAULT_CB, + mlx5_ib_stage_non_default_cb, + NULL), + STAGE_CREATE(MLX5_IB_STAGE_QP, + mlx5_init_qp_table, + mlx5_cleanup_qp_table), + STAGE_CREATE(MLX5_IB_STAGE_SRQ, + mlx5_init_srq_table, + mlx5_cleanup_srq_table), + STAGE_CREATE(MLX5_IB_STAGE_DEVICE_RESOURCES, + mlx5_ib_dev_res_init, + mlx5_ib_dev_res_cleanup), + STAGE_CREATE(MLX5_IB_STAGE_BFREG, + mlx5_ib_stage_bfrag_init, + mlx5_ib_stage_bfrag_cleanup), + STAGE_CREATE(MLX5_IB_STAGE_IB_REG, + mlx5_ib_stage_ib_reg_init, + mlx5_ib_stage_ib_reg_cleanup), +}; + +static struct ib_device *mlx5_ib_add_sub_dev(struct ib_device *parent, + enum rdma_nl_dev_type type, + const char *name) +{ + struct mlx5_ib_dev *mparent = to_mdev(parent), *mplane; + enum rdma_link_layer ll; + int ret; + + if (mparent->smi_dev) + return ERR_PTR(-EEXIST); + + ll = mlx5_port_type_cap_to_rdma_ll(MLX5_CAP_GEN(mparent->mdev, + port_type)); + if (type != RDMA_DEVICE_TYPE_SMI || !mparent->num_plane || + ll != IB_LINK_LAYER_INFINIBAND || + !MLX5_CAP_GEN_2(mparent->mdev, multiplane_qp_ud)) + return ERR_PTR(-EOPNOTSUPP); + + mplane = ib_alloc_device(mlx5_ib_dev, ib_dev); + if (!mplane) + return ERR_PTR(-ENOMEM); + + mplane->port = kcalloc(mparent->num_plane * mparent->num_ports, + sizeof(*mplane->port), GFP_KERNEL); + if (!mplane->port) { + ret = -ENOMEM; + goto fail_kcalloc; + } + + mplane->ib_dev.type = type; + mplane->mdev = mparent->mdev; + mplane->num_ports = mparent->num_plane; + mplane->sub_dev_name = name; + mplane->ib_dev.phys_port_cnt = mplane->num_ports; + + ret = __mlx5_ib_add(mplane, &plane_profile); + if (ret) + goto fail_ib_add; + + mparent->smi_dev = mplane; + return &mplane->ib_dev; + +fail_ib_add: + kfree(mplane->port); +fail_kcalloc: + ib_dealloc_device(&mplane->ib_dev); + return ERR_PTR(ret); +} + +static void mlx5_ib_del_sub_dev(struct ib_device *sub_dev) +{ + struct mlx5_ib_dev *mdev = to_mdev(sub_dev); + + to_mdev(sub_dev->parent)->smi_dev = NULL; + __mlx5_ib_remove(mdev, mdev->profile, MLX5_IB_STAGE_MAX); +} + static int mlx5r_mp_probe(struct auxiliary_device *adev, const struct auxiliary_device_id *id) { @@ -4373,7 +4854,8 @@ static int mlx5r_mp_probe(struct auxiliary_device *adev, mutex_lock(&mlx5_ib_multiport_mutex); list_for_each_entry(dev, &mlx5_ib_dev_list, ib_dev_list) { - if (dev->sys_image_guid == mpi->sys_image_guid) + if (dev->sys_image_guid == mpi->sys_image_guid && + mlx5_core_same_coredev_type(dev->mdev, mpi->mdev)) bound = mlx5_ib_bind_slave_port(dev, mpi); if (bound) { @@ -4426,15 +4908,23 @@ static int mlx5r_probe(struct auxiliary_device *adev, dev = ib_alloc_device(mlx5_ib_dev, ib_dev); if (!dev) return -ENOMEM; + + if (ll == IB_LINK_LAYER_INFINIBAND) { + ret = mlx5_ib_get_plane_num(mdev, &dev->num_plane); + if (ret) + goto fail; + } + dev->port = kcalloc(num_ports, sizeof(*dev->port), GFP_KERNEL); if (!dev->port) { - ib_dealloc_device(&dev->ib_dev); - return -ENOMEM; + ret = -ENOMEM; + goto fail; } dev->mdev = mdev; dev->num_ports = num_ports; + dev->ib_dev.phys_port_cnt = num_ports; if (ll == IB_LINK_LAYER_ETHERNET && !mlx5_get_roce_state(mdev)) profile = &raw_eth_profile; @@ -4442,14 +4932,17 @@ static int mlx5r_probe(struct auxiliary_device *adev, profile = &pf_profile; ret = __mlx5_ib_add(dev, profile); - if (ret) { - kfree(dev->port); - ib_dealloc_device(&dev->ib_dev); - return ret; - } + if (ret) + goto fail_ib_add; auxiliary_set_drvdata(adev, dev); return 0; + +fail_ib_add: + kfree(dev->port); +fail: + ib_dealloc_device(&dev->ib_dev); + return ret; } static void mlx5r_remove(struct auxiliary_device *adev) @@ -4509,17 +5002,23 @@ static int __init mlx5_ib_init(void) ret = mlx5r_rep_init(); if (ret) goto rep_err; + ret = mlx5_data_direct_driver_register(); + if (ret) + goto dd_err; ret = auxiliary_driver_register(&mlx5r_mp_driver); if (ret) goto mp_err; ret = auxiliary_driver_register(&mlx5r_driver); if (ret) goto drv_err; + return 0; drv_err: auxiliary_driver_unregister(&mlx5r_mp_driver); mp_err: + mlx5_data_direct_driver_unregister(); +dd_err: mlx5r_rep_cleanup(); rep_err: mlx5_ib_qp_event_cleanup(); @@ -4531,6 +5030,7 @@ qp_event_err: static void __exit mlx5_ib_cleanup(void) { + mlx5_data_direct_driver_unregister(); auxiliary_driver_unregister(&mlx5r_driver); auxiliary_driver_unregister(&mlx5r_mp_driver); mlx5r_rep_cleanup(); diff --git a/drivers/infiniband/hw/mlx5/mem.c b/drivers/infiniband/hw/mlx5/mem.c index 96ffbbaf0a73..af321f6ef7f5 100644 --- a/drivers/infiniband/hw/mlx5/mem.c +++ b/drivers/infiniband/hw/mlx5/mem.c @@ -32,7 +32,6 @@ #include <rdma/ib_umem_odp.h> #include "mlx5_ib.h" -#include <linux/jiffies.h> /* * Fill in a physical address list. ib_umem_num_dma_blocks() entries will be @@ -94,202 +93,3 @@ unsigned long __mlx5_umem_find_best_quantized_pgoff( return 0; return page_size; } - -#define WR_ID_BF 0xBF -#define WR_ID_END 0xBAD -#define TEST_WC_NUM_WQES 255 -#define TEST_WC_POLLING_MAX_TIME_JIFFIES msecs_to_jiffies(100) -static int post_send_nop(struct mlx5_ib_dev *dev, struct ib_qp *ibqp, u64 wr_id, - bool signaled) -{ - struct mlx5_ib_qp *qp = to_mqp(ibqp); - struct mlx5_wqe_ctrl_seg *ctrl; - struct mlx5_bf *bf = &qp->bf; - __be32 mmio_wqe[16] = {}; - unsigned long flags; - unsigned int idx; - int i; - - if (unlikely(dev->mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)) - return -EIO; - - spin_lock_irqsave(&qp->sq.lock, flags); - - idx = qp->sq.cur_post & (qp->sq.wqe_cnt - 1); - ctrl = mlx5_frag_buf_get_wqe(&qp->sq.fbc, idx); - - memset(ctrl, 0, sizeof(struct mlx5_wqe_ctrl_seg)); - ctrl->fm_ce_se = signaled ? MLX5_WQE_CTRL_CQ_UPDATE : 0; - ctrl->opmod_idx_opcode = - cpu_to_be32(((u32)(qp->sq.cur_post) << 8) | MLX5_OPCODE_NOP); - ctrl->qpn_ds = cpu_to_be32((sizeof(struct mlx5_wqe_ctrl_seg) / 16) | - (qp->trans_qp.base.mqp.qpn << 8)); - - qp->sq.wrid[idx] = wr_id; - qp->sq.w_list[idx].opcode = MLX5_OPCODE_NOP; - qp->sq.wqe_head[idx] = qp->sq.head + 1; - qp->sq.cur_post += DIV_ROUND_UP(sizeof(struct mlx5_wqe_ctrl_seg), - MLX5_SEND_WQE_BB); - qp->sq.w_list[idx].next = qp->sq.cur_post; - qp->sq.head++; - - memcpy(mmio_wqe, ctrl, sizeof(*ctrl)); - ((struct mlx5_wqe_ctrl_seg *)&mmio_wqe)->fm_ce_se |= - MLX5_WQE_CTRL_CQ_UPDATE; - - /* Make sure that descriptors are written before - * updating doorbell record and ringing the doorbell - */ - wmb(); - - qp->db.db[MLX5_SND_DBR] = cpu_to_be32(qp->sq.cur_post); - - /* Make sure doorbell record is visible to the HCA before - * we hit doorbell - */ - wmb(); - for (i = 0; i < 8; i++) - mlx5_write64(&mmio_wqe[i * 2], - bf->bfreg->map + bf->offset + i * 8); - io_stop_wc(); - - bf->offset ^= bf->buf_size; - - spin_unlock_irqrestore(&qp->sq.lock, flags); - - return 0; -} - -static int test_wc_poll_cq_result(struct mlx5_ib_dev *dev, struct ib_cq *cq) -{ - int ret; - struct ib_wc wc = {}; - unsigned long end = jiffies + TEST_WC_POLLING_MAX_TIME_JIFFIES; - - do { - ret = ib_poll_cq(cq, 1, &wc); - if (ret < 0 || wc.status) - return ret < 0 ? ret : -EINVAL; - if (ret) - break; - } while (!time_after(jiffies, end)); - - if (!ret) - return -ETIMEDOUT; - - if (wc.wr_id != WR_ID_BF) - ret = 0; - - return ret; -} - -static int test_wc_do_send(struct mlx5_ib_dev *dev, struct ib_qp *qp) -{ - int err, i; - - for (i = 0; i < TEST_WC_NUM_WQES; i++) { - err = post_send_nop(dev, qp, WR_ID_BF, false); - if (err) - return err; - } - - return post_send_nop(dev, qp, WR_ID_END, true); -} - -int mlx5_ib_test_wc(struct mlx5_ib_dev *dev) -{ - struct ib_cq_init_attr cq_attr = { .cqe = TEST_WC_NUM_WQES + 1 }; - int port_type_cap = MLX5_CAP_GEN(dev->mdev, port_type); - struct ib_qp_init_attr qp_init_attr = { - .cap = { .max_send_wr = TEST_WC_NUM_WQES }, - .qp_type = IB_QPT_UD, - .sq_sig_type = IB_SIGNAL_REQ_WR, - .create_flags = MLX5_IB_QP_CREATE_WC_TEST, - }; - struct ib_qp_attr qp_attr = { .port_num = 1 }; - struct ib_device *ibdev = &dev->ib_dev; - struct ib_qp *qp; - struct ib_cq *cq; - struct ib_pd *pd; - int ret; - - if (!MLX5_CAP_GEN(dev->mdev, bf)) - return 0; - - if (!dev->mdev->roce.roce_en && - port_type_cap == MLX5_CAP_PORT_TYPE_ETH) { - if (mlx5_core_is_pf(dev->mdev)) - dev->wc_support = arch_can_pci_mmap_wc(); - return 0; - } - - ret = mlx5_alloc_bfreg(dev->mdev, &dev->wc_bfreg, true, false); - if (ret) - goto print_err; - - if (!dev->wc_bfreg.wc) - goto out1; - - pd = ib_alloc_pd(ibdev, 0); - if (IS_ERR(pd)) { - ret = PTR_ERR(pd); - goto out1; - } - - cq = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr); - if (IS_ERR(cq)) { - ret = PTR_ERR(cq); - goto out2; - } - - qp_init_attr.recv_cq = cq; - qp_init_attr.send_cq = cq; - qp = ib_create_qp(pd, &qp_init_attr); - if (IS_ERR(qp)) { - ret = PTR_ERR(qp); - goto out3; - } - - qp_attr.qp_state = IB_QPS_INIT; - ret = ib_modify_qp(qp, &qp_attr, - IB_QP_STATE | IB_QP_PORT | IB_QP_PKEY_INDEX | - IB_QP_QKEY); - if (ret) - goto out4; - - qp_attr.qp_state = IB_QPS_RTR; - ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE); - if (ret) - goto out4; - - qp_attr.qp_state = IB_QPS_RTS; - ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN); - if (ret) - goto out4; - - ret = test_wc_do_send(dev, qp); - if (ret < 0) - goto out4; - - ret = test_wc_poll_cq_result(dev, cq); - if (ret > 0) { - dev->wc_support = true; - ret = 0; - } - -out4: - ib_destroy_qp(qp); -out3: - ib_destroy_cq(cq); -out2: - ib_dealloc_pd(pd); -out1: - mlx5_free_bfreg(dev->mdev, &dev->wc_bfreg); -print_err: - if (ret) - mlx5_ib_err( - dev, - "Error %d while trying to test write-combining support\n", - ret); - return ret; -} diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index bbe79b86c717..fde859d207ae 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -63,17 +63,6 @@ __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits, return GENMASK(largest_pg_shift, pgsz_shift); } -/* - * For mkc users, instead of a page_offset the command has a start_iova which - * specifies both the page_offset and the on-the-wire IOVA - */ -#define mlx5_umem_find_best_pgsz(umem, typ, log_pgsz_fld, pgsz_shift, iova) \ - ib_umem_find_best_pgsz(umem, \ - __mlx5_log_page_size_to_bitmap( \ - __mlx5_bit_sz(typ, log_pgsz_fld), \ - pgsz_shift), \ - iova) - static __always_inline unsigned long __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits, unsigned int offset_shift) @@ -115,6 +104,19 @@ unsigned long __mlx5_umem_find_best_quantized_pgoff( __mlx5_bit_sz(typ, page_offset_fld), 0, scale, \ page_offset_quantized) +static inline unsigned long +mlx5_umem_dmabuf_find_best_pgsz(struct ib_umem_dmabuf *umem_dmabuf) +{ + /* + * mkeys used for dmabuf are fixed at PAGE_SIZE because we must be able + * to hold any sgl after a move operation. Ideally the mkc page size + * could be changed at runtime to be optimal, but right now the driver + * cannot do that. + */ + return ib_umem_find_best_pgsz(&umem_dmabuf->umem, PAGE_SIZE, + umem_dmabuf->umem.iova); +} + enum { MLX5_IB_MMAP_OFFSET_START = 9, MLX5_IB_MMAP_OFFSET_END = 255, @@ -274,6 +276,7 @@ struct mlx5_ib_flow_matcher { struct mlx5_core_dev *mdev; atomic_t usecnt; u8 match_criteria_enable; + u32 ib_port; }; struct mlx5_ib_steering_anchor { @@ -291,6 +294,18 @@ enum mlx5_ib_optional_counter_type { MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS, MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS, MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS, + MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS, + MLX5_IB_OPCOUNTER_RDMA_TX_BYTES, + MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS, + MLX5_IB_OPCOUNTER_RDMA_RX_BYTES, + + MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP, + MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS_PER_QP, + MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS_PER_QP, + MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP, + MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP, + MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP, + MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP, MLX5_IB_OPCOUNTER_MAX, }; @@ -305,6 +320,8 @@ struct mlx5_ib_flow_db { struct mlx5_ib_flow_prio rdma_tx[MLX5_IB_NUM_FLOW_FT]; struct mlx5_ib_flow_prio opfcs[MLX5_IB_OPCOUNTER_MAX]; struct mlx5_flow_table *lag_demux_ft; + struct mlx5_ib_flow_prio *rdma_transport_rx; + struct mlx5_ib_flow_prio *rdma_transport_tx; /* Protect flow steering bypass flow tables * when add/del flow rules. * only single add/removal of flow steering rule could be done @@ -334,6 +351,7 @@ struct mlx5_ib_flow_db { #define MLX5_IB_UPD_XLT_PD BIT(4) #define MLX5_IB_UPD_XLT_ACCESS BIT(5) #define MLX5_IB_UPD_XLT_INDIRECT BIT(6) +#define MLX5_IB_UPD_XLT_DOWNGRADE BIT(7) /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags. * @@ -341,7 +359,6 @@ struct mlx5_ib_flow_db { * rely on the range reserved for that use in the ib_qp_create_flags enum. */ #define MLX5_IB_QP_CREATE_SQPN_QP1 IB_QP_CREATE_RESERVED_START -#define MLX5_IB_QP_CREATE_WC_TEST (IB_QP_CREATE_RESERVED_START << 1) struct wr_list { u16 opcode; @@ -520,6 +537,7 @@ struct mlx5_ib_qp { struct mlx5_bf bf; u8 has_rq:1; u8 is_rss:1; + u8 is_ooo_rq:1; /* only for user space QPs. For kernel * we have it from the bf object @@ -628,6 +646,8 @@ enum mlx5_mkey_type { MLX5_MKEY_MR = 1, MLX5_MKEY_MW, MLX5_MKEY_INDIRECT_DEVX, + MLX5_MKEY_NULL, + MLX5_MKEY_IMPLICIT_CHILD, }; struct mlx5r_cache_rb_key { @@ -643,9 +663,10 @@ struct mlx5_ib_mkey { unsigned int ndescs; struct wait_queue_head wait; refcount_t usecount; - /* User Mkey must hold either a rb_key or a cache_ent. */ + /* Cacheable user Mkey must hold either a rb_key or a cache_ent. */ struct mlx5r_cache_rb_key rb_key; struct mlx5_cache_ent *cache_ent; + u8 cacheable : 1; }; #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE) @@ -664,11 +685,19 @@ struct mlx5_ib_mkey { #define mlx5_update_odp_stats(mr, counter_name, value) \ atomic64_add(value, &((mr)->odp_stats.counter_name)) +#define mlx5_update_odp_stats_with_handled(mr, counter_name, value) \ + do { \ + mlx5_update_odp_stats(mr, counter_name, value); \ + atomic64_add(1, &((mr)->odp_stats.counter_name##_handled)); \ + } while (0) + struct mlx5_ib_mr { struct ib_mr ibmr; struct mlx5_ib_mkey mmkey; struct ib_umem *umem; + /* The mr is data direct related */ + u8 data_direct :1; union { /* Used only by kernel MRs (umem == NULL) */ @@ -706,6 +735,11 @@ struct mlx5_ib_mr { } odp_destroy; struct ib_odp_counters odp_stats; bool is_odp_implicit; + /* The affilated data direct crossed mr */ + struct mlx5_ib_mr *dd_crossed_mr; + struct list_head dd_node; + u8 revoked :1; + struct mlx5_ib_mkey null_mmkey; }; }; }; @@ -751,6 +785,8 @@ struct umr_common { */ struct mutex lock; unsigned int state; + /* Protects from repeat UMR QP creation */ + struct mutex init_lock; }; #define NUM_MKEYS_PER_PAGE \ @@ -781,6 +817,7 @@ struct mlx5_cache_ent { u8 is_tmp:1; u8 disabled:1; u8 fill_to_high_water:1; + u8 tmp_cleanup_scheduled:1; /* * - limit is the low water mark for stored mkeys, 2* limit is the @@ -812,7 +849,6 @@ struct mlx5_mkey_cache { struct mutex rb_lock; struct dentry *fs_root; unsigned long last_add; - struct delayed_work remove_ent_dwork; }; struct mlx5_ib_port_resources { @@ -820,13 +856,20 @@ struct mlx5_ib_port_resources { struct work_struct pkey_change_work; }; +struct mlx5_data_direct_resources { + u32 pdn; + u32 mkey; +}; + struct mlx5_ib_resources { struct ib_cq *c0; + struct mutex cq_lock; u32 xrcdn0; u32 xrcdn1; struct ib_pd *p0; struct ib_srq *s0; struct ib_srq *s1; + struct mutex srq_lock; struct mlx5_ib_port_resources ports[2]; }; @@ -856,6 +899,14 @@ void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev, struct mlx5_ib_op_fc *opfc, enum mlx5_ib_optional_counter_type type); +int mlx5r_fs_bind_op_fc(struct ib_qp *qp, struct rdma_counter *counter, + u32 port); + +void mlx5r_fs_unbind_op_fc(struct ib_qp *qp, struct rdma_counter *counter); + +void mlx5r_fs_destroy_fcs(struct mlx5_ib_dev *dev, + struct rdma_counter *counter); + struct mlx5_ib_multiport_info; struct mlx5_ib_multiport { @@ -868,8 +919,6 @@ struct mlx5_roce { /* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL * netdev pointer */ - rwlock_t netdev_lock; - struct net_device *netdev; struct notifier_block nb; struct netdev_net_notifier nn; struct notifier_block mdev_nb; @@ -954,15 +1003,14 @@ enum mlx5_ib_stages { MLX5_IB_STAGE_QP, MLX5_IB_STAGE_SRQ, MLX5_IB_STAGE_DEVICE_RESOURCES, - MLX5_IB_STAGE_DEVICE_NOTIFIER, MLX5_IB_STAGE_ODP, MLX5_IB_STAGE_COUNTERS, MLX5_IB_STAGE_CONG_DEBUGFS, - MLX5_IB_STAGE_UAR, MLX5_IB_STAGE_BFREG, MLX5_IB_STAGE_PRE_IB_REG_UMR, MLX5_IB_STAGE_WHITELIST_UID, MLX5_IB_STAGE_IB_REG, + MLX5_IB_STAGE_DEVICE_NOTIFIER, MLX5_IB_STAGE_POST_IB_REG_UMR, MLX5_IB_STAGE_DELAY_DROP, MLX5_IB_STAGE_RESTRACK, @@ -1114,7 +1162,11 @@ struct mlx5_macsec { struct mlx5_ib_dev { struct ib_device ib_dev; struct mlx5_core_dev *mdev; + struct mlx5_data_direct_dev *data_direct_dev; + /* protect accessing data_direct_dev */ + struct mutex data_direct_lock; struct notifier_block mdev_events; + struct notifier_block lag_events; int num_ports; /* serialize update of capability mask */ @@ -1122,7 +1174,6 @@ struct mlx5_ib_dev { u8 ib_active:1; u8 is_rep:1; u8 lag_active:1; - u8 wc_support:1; u8 fill_delay; struct umr_common umrc; /* sync used page count stats @@ -1145,10 +1196,10 @@ struct mlx5_ib_dev { /* protect resources needed as part of reset flow */ spinlock_t reset_flow_resource_lock; struct list_head qp_list; + struct list_head data_direct_mr_list; /* Array with num_ports elements */ struct mlx5_ib_port *port; struct mlx5_sq_bfreg bfreg; - struct mlx5_sq_bfreg wc_bfreg; struct mlx5_sq_bfreg fp_bfreg; struct mlx5_ib_delay_drop delay_drop; const struct mlx5_ib_profile *profile; @@ -1170,10 +1221,15 @@ struct mlx5_ib_dev { u16 pkey_table_len; u8 lag_ports; struct mlx5_special_mkeys mkeys; + struct mlx5_data_direct_resources ddr; #ifdef CONFIG_MLX5_MACSEC struct mlx5_macsec macsec; #endif + + u8 num_plane; + struct mlx5_ib_dev *smi_dev; + const char *sub_dev_name; }; static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq) @@ -1272,6 +1328,8 @@ to_mmmap(struct rdma_user_mmap_entry *rdma_entry) struct mlx5_user_mmap_entry, rdma_entry); } +int mlx5_ib_dev_res_cq_init(struct mlx5_ib_dev *dev); +int mlx5_ib_dev_res_srq_init(struct mlx5_ib_dev *dev); int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt, struct mlx5_db *db); void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db); @@ -1311,7 +1369,7 @@ int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer, int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer, size_t buflen, size_t *bc); int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, - struct ib_udata *udata); + struct uverbs_attr_bundle *attrs); int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata); int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc); int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags); @@ -1324,7 +1382,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int fd, int access_flags, - struct ib_udata *udata); + struct uverbs_attr_bundle *attrs); int mlx5_ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, @@ -1335,7 +1393,6 @@ int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata); int mlx5_ib_dealloc_mw(struct ib_mw *mw); struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, int access_flags); -void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr); void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr); struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, u64 length, u64 virt_addr, int access_flags, @@ -1377,7 +1434,6 @@ int mlx5_ib_query_port(struct ib_device *ibdev, u32 port, struct ib_port_attr *props); void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas, u64 access_flags); -void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num); int mlx5_ib_get_cqe_size(struct ib_cq *ibcq); int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev); void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev); @@ -1405,6 +1461,10 @@ int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm, struct ib_dm_mr_attr *attr, struct uverbs_attr_bundle *attrs); +void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev, + struct mlx5_data_direct_dev *dev); +void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev); +void mlx5_ib_revoke_data_direct_mrs(struct mlx5_ib_dev *dev); #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev); @@ -1413,8 +1473,8 @@ void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev); int __init mlx5_ib_odp_init(void); void mlx5_ib_odp_cleanup(void); int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev); -void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, - struct mlx5_ib_mr *mr, int flags); +int mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, + struct mlx5_ib_mr *mr, int flags); int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, @@ -1435,8 +1495,11 @@ static inline int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev) { return 0; } -static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, - struct mlx5_ib_mr *mr, int flags) {} +static inline int mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, + struct mlx5_ib_mr *mr, int flags) +{ + return -EOPNOTSUPP; +} static inline int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, @@ -1512,6 +1575,7 @@ extern const struct uapi_definition mlx5_ib_devx_defs[]; extern const struct uapi_definition mlx5_ib_flow_defs[]; extern const struct uapi_definition mlx5_ib_qos_defs[]; extern const struct uapi_definition mlx5_ib_std_types_defs[]; +extern const struct uapi_definition mlx5_ib_create_cq_defs[]; static inline int is_qp1(enum ib_qp_type qp_type) { @@ -1612,8 +1676,6 @@ static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_ib_mkey *mmkey) wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0); } -int mlx5_ib_test_wc(struct mlx5_ib_dev *dev); - static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev) { /* @@ -1680,4 +1742,26 @@ static inline bool mlx5_umem_needs_ats(struct mlx5_ib_dev *dev, int set_roce_addr(struct mlx5_ib_dev *dev, u32 port_num, unsigned int index, const union ib_gid *gid, const struct ib_gid_attr *attr); + +static inline u32 smi_to_native_portnum(struct mlx5_ib_dev *dev, u32 port) +{ + return (port - 1) / dev->num_ports + 1; +} + +/* + * For mkc users, instead of a page_offset the command has a start_iova which + * specifies both the page_offset and the on-the-wire IOVA + */ +static __always_inline unsigned long +mlx5_umem_mkc_find_best_pgsz(struct mlx5_ib_dev *dev, struct ib_umem *umem, + u64 iova) +{ + int page_size_bits = + MLX5_CAP_GEN_2(dev->mdev, umr_log_entity_size_5) ? 6 : 5; + unsigned long bitmap = + __mlx5_log_page_size_to_bitmap(page_size_bits, 0); + + return ib_umem_find_best_pgsz(umem, bitmap, iova); +} + #endif /* MLX5_IB_H */ diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index a8ee2ca1f4a1..57f9bc2a4a3a 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -43,18 +43,22 @@ #include "dm.h" #include "mlx5_ib.h" #include "umr.h" +#include "data_direct.h" enum { MAX_PENDING_REG_MR = 8, }; +#define MLX5_MR_CACHE_PERSISTENT_ENTRY_MIN_DESCS 4 #define MLX5_UMR_ALIGN 2048 static void create_mkey_callback(int status, struct mlx5_async_work *context); static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags, - unsigned int page_size, bool populate); + unsigned long page_size, bool populate, + int access_mode); +static int __mlx5_ib_dereg_mr(struct ib_mr *ibmr); static void set_mkc_access_pd_addr_fields(void *mkc, int acc, u64 start_addr, struct ib_pd *pd) @@ -211,9 +215,9 @@ static void create_mkey_callback(int status, struct mlx5_async_work *context) spin_lock_irqsave(&ent->mkeys_queue.lock, flags); push_mkey_locked(ent, mkey_out->mkey); + ent->pending--; /* If we are doing fill_to_high_water then keep going. */ queue_adjust_cache_locked(ent); - ent->pending--; spin_unlock_irqrestore(&ent->mkeys_queue.lock, flags); kfree(mkey_out); } @@ -246,6 +250,7 @@ static void set_cache_mkc(struct mlx5_cache_ent *ent, void *mkc) MLX5_SET(mkc, mkc, access_mode_1_0, ent->rb_key.access_mode & 0x3); MLX5_SET(mkc, mkc, access_mode_4_2, (ent->rb_key.access_mode >> 2) & 0x7); + MLX5_SET(mkc, mkc, ma_translation_mode, !!ent->rb_key.ats); MLX5_SET(mkc, mkc, translations_octword_size, get_mkc_octo_size(ent->rb_key.access_mode, @@ -520,12 +525,27 @@ static void queue_adjust_cache_locked(struct mlx5_cache_ent *ent) ent->fill_to_high_water = false; if (ent->pending) queue_delayed_work(ent->dev->cache.wq, &ent->dwork, - msecs_to_jiffies(1000)); + secs_to_jiffies(1)); else mod_delayed_work(ent->dev->cache.wq, &ent->dwork, 0); } } +static void clean_keys(struct mlx5_ib_dev *dev, struct mlx5_cache_ent *ent) +{ + u32 mkey; + + spin_lock_irq(&ent->mkeys_queue.lock); + while (ent->mkeys_queue.ci) { + mkey = pop_mkey_locked(ent); + spin_unlock_irq(&ent->mkeys_queue.lock); + mlx5_core_destroy_mkey(dev->mdev, mkey); + spin_lock_irq(&ent->mkeys_queue.lock); + } + ent->tmp_cleanup_scheduled = false; + spin_unlock_irq(&ent->mkeys_queue.lock); +} + static void __cache_work_func(struct mlx5_cache_ent *ent) { struct mlx5_ib_dev *dev = ent->dev; @@ -556,7 +576,7 @@ static void __cache_work_func(struct mlx5_cache_ent *ent) "add keys command failed, err %d\n", err); queue_delayed_work(cache->wq, &ent->dwork, - msecs_to_jiffies(1000)); + secs_to_jiffies(1)); } } } else if (ent->mkeys_queue.ci > 2 * ent->limit) { @@ -597,7 +617,11 @@ static void delayed_cache_work_func(struct work_struct *work) struct mlx5_cache_ent *ent; ent = container_of(work, struct mlx5_cache_ent, dwork.work); - __cache_work_func(ent); + /* temp entries are never filled, only cleaned */ + if (ent->is_tmp) + clean_keys(ent->dev, ent); + else + __cache_work_func(ent); } static int cache_ent_key_cmp(struct mlx5r_cache_rb_key key1, @@ -641,10 +665,8 @@ static int mlx5_cache_ent_insert(struct mlx5_mkey_cache *cache, new = &((*new)->rb_left); if (cmp < 0) new = &((*new)->rb_right); - if (cmp == 0) { - mutex_unlock(&cache->rb_lock); + if (cmp == 0) return -EEXIST; - } } /* Add new node and rebalance tree. */ @@ -660,6 +682,7 @@ mkey_cache_ent_from_rb_key(struct mlx5_ib_dev *dev, { struct rb_node *node = dev->cache.rb_root.rb_node; struct mlx5_cache_ent *cur, *smallest = NULL; + u64 ndescs_limit; int cmp; /* @@ -678,17 +701,24 @@ mkey_cache_ent_from_rb_key(struct mlx5_ib_dev *dev, return cur; } + /* + * Limit the usage of mkeys larger than twice the required size while + * also allowing the usage of smallest cache entry for small MRs. + */ + ndescs_limit = max_t(u64, rb_key.ndescs * 2, + MLX5_MR_CACHE_PERSISTENT_ENTRY_MIN_DESCS); + return (smallest && smallest->rb_key.access_mode == rb_key.access_mode && smallest->rb_key.access_flags == rb_key.access_flags && - smallest->rb_key.ats == rb_key.ats) ? + smallest->rb_key.ats == rb_key.ats && + smallest->rb_key.ndescs <= ndescs_limit) ? smallest : NULL; } static struct mlx5_ib_mr *_mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, - struct mlx5_cache_ent *ent, - int access_flags) + struct mlx5_cache_ent *ent) { struct mlx5_ib_mr *mr; int err; @@ -719,6 +749,8 @@ static struct mlx5_ib_mr *_mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, } mr->mmkey.cache_ent = ent; mr->mmkey.type = MLX5_MKEY_MR; + mr->mmkey.rb_key = ent->rb_key; + mr->mmkey.cacheable = true; init_waitqueue_head(&mr->mmkey.wait); return mr; } @@ -761,22 +793,7 @@ struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, if (!ent) return ERR_PTR(-EOPNOTSUPP); - return _mlx5_mr_cache_alloc(dev, ent, access_flags); -} - -static void clean_keys(struct mlx5_ib_dev *dev, struct mlx5_cache_ent *ent) -{ - u32 mkey; - - cancel_delayed_work(&ent->dwork); - spin_lock_irq(&ent->mkeys_queue.lock); - while (ent->mkeys_queue.ci) { - mkey = pop_mkey_locked(ent); - spin_unlock_irq(&ent->mkeys_queue.lock); - mlx5_core_destroy_mkey(dev->mdev, mkey); - spin_lock_irq(&ent->mkeys_queue.lock); - } - spin_unlock_irq(&ent->mkeys_queue.lock); + return _mlx5_mr_cache_alloc(dev, ent); } static void mlx5_mkey_cache_debugfs_cleanup(struct mlx5_ib_dev *dev) @@ -821,7 +838,7 @@ static void mlx5_mkey_cache_debugfs_init(struct mlx5_ib_dev *dev) static void delay_time_func(struct timer_list *t) { - struct mlx5_ib_dev *dev = from_timer(dev, t, delay_timer); + struct mlx5_ib_dev *dev = timer_container_of(dev, t, delay_timer); WRITE_ONCE(dev->fill_delay, 0); } @@ -891,10 +908,6 @@ mlx5r_cache_create_ent_locked(struct mlx5_ib_dev *dev, ent->limit = 0; mlx5_mkey_cache_debugfs_add_ent(dev, ent); - } else { - mod_delayed_work(ent->dev->cache.wq, - &ent->dev->cache.remove_ent_dwork, - msecs_to_jiffies(30 * 1000)); } return ent; @@ -905,33 +918,23 @@ mkeys_err: return ERR_PTR(ret); } -static void remove_ent_work_func(struct work_struct *work) +static void mlx5r_destroy_cache_entries(struct mlx5_ib_dev *dev) { - struct mlx5_mkey_cache *cache; + struct rb_root *root = &dev->cache.rb_root; struct mlx5_cache_ent *ent; - struct rb_node *cur; - - cache = container_of(work, struct mlx5_mkey_cache, - remove_ent_dwork.work); - mutex_lock(&cache->rb_lock); - cur = rb_last(&cache->rb_root); - while (cur) { - ent = rb_entry(cur, struct mlx5_cache_ent, node); - cur = rb_prev(cur); - mutex_unlock(&cache->rb_lock); - - spin_lock_irq(&ent->mkeys_queue.lock); - if (!ent->is_tmp) { - spin_unlock_irq(&ent->mkeys_queue.lock); - mutex_lock(&cache->rb_lock); - continue; - } - spin_unlock_irq(&ent->mkeys_queue.lock); + struct rb_node *node; - clean_keys(ent->dev, ent); - mutex_lock(&cache->rb_lock); + mutex_lock(&dev->cache.rb_lock); + node = rb_first(root); + while (node) { + ent = rb_entry(node, struct mlx5_cache_ent, node); + node = rb_next(node); + clean_keys(dev, ent); + rb_erase(&ent->node, root); + mlx5r_mkeys_uninit(ent); + kfree(ent); } - mutex_unlock(&cache->rb_lock); + mutex_unlock(&dev->cache.rb_lock); } int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev) @@ -949,7 +952,6 @@ int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev) mutex_init(&dev->slow_path_mutex); mutex_init(&dev->cache.rb_lock); dev->cache.rb_root = RB_ROOT; - INIT_DELAYED_WORK(&dev->cache.remove_ent_dwork, remove_ent_work_func); cache->wq = alloc_ordered_workqueue("mkey_cache", WQ_MEM_RECLAIM); if (!cache->wq) { mlx5_ib_warn(dev, "failed to create work queue\n"); @@ -961,7 +963,7 @@ int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev) mlx5_mkey_cache_debugfs_init(dev); mutex_lock(&cache->rb_lock); for (i = 0; i <= mkey_cache_max_order(dev); i++) { - rb_key.ndescs = 1 << (i + 2); + rb_key.ndescs = MLX5_MR_CACHE_PERSISTENT_ENTRY_MIN_DESCS << i; ent = mlx5r_cache_create_ent_locked(dev, rb_key, true); if (IS_ERR(ent)) { ret = PTR_ERR(ent); @@ -986,6 +988,8 @@ int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev) err: mutex_unlock(&cache->rb_lock); mlx5_mkey_cache_debugfs_cleanup(dev); + mlx5r_destroy_cache_entries(dev); + destroy_workqueue(cache->wq); mlx5_ib_warn(dev, "failed to create mkey cache entry\n"); return ret; } @@ -1000,7 +1004,6 @@ void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev) return; mutex_lock(&dev->cache.rb_lock); - cancel_delayed_work(&dev->cache.remove_ent_dwork); for (node = rb_first(root); node; node = rb_next(node)) { ent = rb_entry(node, struct mlx5_cache_ent, node); spin_lock_irq(&ent->mkeys_queue.lock); @@ -1020,20 +1023,10 @@ void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev) mlx5_cmd_cleanup_async_ctx(&dev->async_ctx); /* At this point all entries are disabled and have no concurrent work. */ - mutex_lock(&dev->cache.rb_lock); - node = rb_first(root); - while (node) { - ent = rb_entry(node, struct mlx5_cache_ent, node); - node = rb_next(node); - clean_keys(dev, ent); - rb_erase(&ent->node, root); - mlx5r_mkeys_uninit(ent); - kfree(ent); - } - mutex_unlock(&dev->cache.rb_lock); + mlx5r_destroy_cache_entries(dev); destroy_workqueue(dev->cache.wq); - del_timer_sync(&dev->delay_timer); + timer_delete_sync(&dev->delay_timer); } struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc) @@ -1061,6 +1054,7 @@ struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc) MLX5_SET(mkc, mkc, length64, 1); set_mkc_access_pd_addr_fields(mkc, acc | IB_ACCESS_RELAXED_ORDERING, 0, pd); + MLX5_SET(mkc, mkc, ma_translation_mode, MLX5_CAP_GEN(dev->mdev, ats)); err = mlx5_ib_create_mkey(dev, &mr->mmkey, in, inlen); if (err) @@ -1125,24 +1119,22 @@ static unsigned int mlx5_umem_dmabuf_default_pgsz(struct ib_umem *umem, static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, struct ib_umem *umem, u64 iova, - int access_flags) + int access_flags, int access_mode) { - struct mlx5r_cache_rb_key rb_key = { - .access_mode = MLX5_MKC_ACCESS_MODE_MTT, - }; struct mlx5_ib_dev *dev = to_mdev(pd->device); + struct mlx5r_cache_rb_key rb_key = {}; struct mlx5_cache_ent *ent; struct mlx5_ib_mr *mr; - unsigned int page_size; + unsigned long page_size; if (umem->is_dmabuf) page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova); else - page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, - 0, iova); + page_size = mlx5_umem_mkc_find_best_pgsz(dev, umem, iova); if (WARN_ON(!page_size)) return ERR_PTR(-EINVAL); + rb_key.access_mode = access_mode; rb_key.ndescs = ib_umem_num_dma_blocks(umem, page_size); rb_key.ats = mlx5_umem_needs_ats(dev, umem, access_flags); rb_key.access_flags = get_unchangeable_access_flags(dev, access_flags); @@ -1153,15 +1145,16 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, */ if (!ent) { mutex_lock(&dev->slow_path_mutex); - mr = reg_create(pd, umem, iova, access_flags, page_size, false); + mr = reg_create(pd, umem, iova, access_flags, page_size, false, access_mode); mutex_unlock(&dev->slow_path_mutex); if (IS_ERR(mr)) return mr; mr->mmkey.rb_key = rb_key; + mr->mmkey.cacheable = true; return mr; } - mr = _mlx5_mr_cache_alloc(dev, ent, access_flags); + mr = _mlx5_mr_cache_alloc(dev, ent); if (IS_ERR(mr)) return mr; @@ -1173,13 +1166,71 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, return mr; } +static struct ib_mr * +reg_create_crossing_vhca_mr(struct ib_pd *pd, u64 iova, u64 length, int access_flags, + u32 crossed_lkey) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + int access_mode = MLX5_MKC_ACCESS_MODE_CROSSING; + struct mlx5_ib_mr *mr; + void *mkc; + int inlen; + u32 *in; + int err; + + if (!MLX5_CAP_GEN(dev->mdev, crossing_vhca_mkey)) + return ERR_PTR(-EOPNOTSUPP); + + mr = kzalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) + return ERR_PTR(-ENOMEM); + + inlen = MLX5_ST_SZ_BYTES(create_mkey_in); + in = kvzalloc(inlen, GFP_KERNEL); + if (!in) { + err = -ENOMEM; + goto err_1; + } + + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + MLX5_SET(mkc, mkc, crossing_target_vhca_id, + MLX5_CAP_GEN(dev->mdev, vhca_id)); + MLX5_SET(mkc, mkc, translations_octword_size, crossed_lkey); + MLX5_SET(mkc, mkc, access_mode_1_0, access_mode & 0x3); + MLX5_SET(mkc, mkc, access_mode_4_2, (access_mode >> 2) & 0x7); + + /* for this crossing mkey IOVA should be 0 and len should be IOVA + len */ + set_mkc_access_pd_addr_fields(mkc, access_flags, 0, pd); + MLX5_SET64(mkc, mkc, len, iova + length); + + MLX5_SET(mkc, mkc, free, 0); + MLX5_SET(mkc, mkc, umr_en, 0); + err = mlx5_ib_create_mkey(dev, &mr->mmkey, in, inlen); + if (err) + goto err_2; + + mr->mmkey.type = MLX5_MKEY_MR; + set_mr_fields(dev, mr, length, access_flags, iova); + mr->ibmr.pd = pd; + kvfree(in); + mlx5_ib_dbg(dev, "crossing mkey = 0x%x\n", mr->mmkey.key); + + return &mr->ibmr; +err_2: + kvfree(in); +err_1: + kfree(mr); + return ERR_PTR(err); +} + /* * If ibmr is NULL it will be allocated by reg_create. * Else, the given ibmr will be used. */ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags, - unsigned int page_size, bool populate) + unsigned long page_size, bool populate, + int access_mode) { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_ib_mr *mr; @@ -1188,7 +1239,9 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, int inlen; u32 *in; int err; - bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg)); + bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg)) && + (access_mode == MLX5_MKC_ACCESS_MODE_MTT); + bool ksm_mode = (access_mode == MLX5_MKC_ACCESS_MODE_KSM); if (!page_size) return ERR_PTR(-EINVAL); @@ -1211,7 +1264,7 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, } pas = (__be64 *)MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt); if (populate) { - if (WARN_ON(access_flags & IB_ACCESS_ON_DEMAND)) { + if (WARN_ON(access_flags & IB_ACCESS_ON_DEMAND || ksm_mode)) { err = -EINVAL; goto err_2; } @@ -1227,14 +1280,22 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); set_mkc_access_pd_addr_fields(mkc, access_flags, iova, populate ? pd : dev->umrc.pd); + /* In case a data direct flow, overwrite the pdn field by its internal kernel PD */ + if (umem->is_dmabuf && ksm_mode) + MLX5_SET(mkc, mkc, pd, dev->ddr.pdn); + MLX5_SET(mkc, mkc, free, !populate); - MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT); + MLX5_SET(mkc, mkc, access_mode_1_0, access_mode); MLX5_SET(mkc, mkc, umr_en, 1); MLX5_SET64(mkc, mkc, len, umem->length); MLX5_SET(mkc, mkc, bsf_octword_size, 0); - MLX5_SET(mkc, mkc, translations_octword_size, - get_octo_len(iova, umem->length, mr->page_shift)); + if (ksm_mode) + MLX5_SET(mkc, mkc, translations_octword_size, + get_octo_len(iova, umem->length, mr->page_shift) * 2); + else + MLX5_SET(mkc, mkc, translations_octword_size, + get_octo_len(iova, umem->length, mr->page_shift)); MLX5_SET(mkc, mkc, log_page_size, mr->page_shift); if (mlx5_umem_needs_ats(dev, umem, access_flags)) MLX5_SET(mkc, mkc, ma_translation_mode, 1); @@ -1371,13 +1432,15 @@ static struct ib_mr *create_real_mr(struct ib_pd *pd, struct ib_umem *umem, xlt_with_umr = mlx5r_umr_can_load_pas(dev, umem->length); if (xlt_with_umr) { - mr = alloc_cacheable_mr(pd, umem, iova, access_flags); + mr = alloc_cacheable_mr(pd, umem, iova, access_flags, + MLX5_MKC_ACCESS_MODE_MTT); } else { - unsigned int page_size = mlx5_umem_find_best_pgsz( - umem, mkc, log_page_size, 0, iova); + unsigned long page_size = + mlx5_umem_mkc_find_best_pgsz(dev, umem, iova); mutex_lock(&dev->slow_path_mutex); - mr = reg_create(pd, umem, iova, access_flags, page_size, true); + mr = reg_create(pd, umem, iova, access_flags, page_size, + true, MLX5_MKC_ACCESS_MODE_MTT); mutex_unlock(&dev->slow_path_mutex); } if (IS_ERR(mr)) { @@ -1440,7 +1503,8 @@ static struct ib_mr *create_user_odp_mr(struct ib_pd *pd, u64 start, u64 length, if (IS_ERR(odp)) return ERR_CAST(odp); - mr = alloc_cacheable_mr(pd, &odp->umem, iova, access_flags); + mr = alloc_cacheable_mr(pd, &odp->umem, iova, access_flags, + MLX5_MKC_ACCESS_MODE_MTT); if (IS_ERR(mr)) { ib_umem_release(&odp->umem); return ERR_CAST(mr); @@ -1468,6 +1532,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct ib_umem *umem; + int err; if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM)) return ERR_PTR(-EOPNOTSUPP); @@ -1475,6 +1540,10 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mlx5_ib_dbg(dev, "start 0x%llx, iova 0x%llx, length 0x%llx, access_flags 0x%x\n", start, iova, length, access_flags); + err = mlx5r_umr_resource_init(dev); + if (err) + return ERR_PTR(err); + if (access_flags & IB_ACCESS_ON_DEMAND) return create_user_odp_mr(pd, start, length, iova, access_flags, udata); @@ -1491,7 +1560,7 @@ static void mlx5_ib_dmabuf_invalidate_cb(struct dma_buf_attachment *attach) dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv); - if (!umem_dmabuf->sgt) + if (!umem_dmabuf->sgt || !mr) return; mlx5r_umr_update_mr_pas(mr, MLX5_IB_UPD_XLT_ZAP); @@ -1503,31 +1572,31 @@ static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = { .move_notify = mlx5_ib_dmabuf_invalidate_cb, }; -struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, - u64 length, u64 virt_addr, - int fd, int access_flags, - struct ib_udata *udata) +static struct ib_mr * +reg_user_mr_dmabuf(struct ib_pd *pd, struct device *dma_device, + u64 offset, u64 length, u64 virt_addr, + int fd, int access_flags, int access_mode) { + bool pinned_mode = (access_mode == MLX5_MKC_ACCESS_MODE_KSM); struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_ib_mr *mr = NULL; struct ib_umem_dmabuf *umem_dmabuf; int err; - if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM) || - !IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) - return ERR_PTR(-EOPNOTSUPP); - - mlx5_ib_dbg(dev, - "offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x\n", - offset, virt_addr, length, fd, access_flags); + err = mlx5r_umr_resource_init(dev); + if (err) + return ERR_PTR(err); - /* dmabuf requires xlt update via umr to work. */ - if (!mlx5r_umr_can_load_pas(dev, length)) - return ERR_PTR(-EINVAL); + if (!pinned_mode) + umem_dmabuf = ib_umem_dmabuf_get(&dev->ib_dev, + offset, length, fd, + access_flags, + &mlx5_ib_dmabuf_attach_ops); + else + umem_dmabuf = ib_umem_dmabuf_get_pinned_with_dma_device(&dev->ib_dev, + dma_device, offset, length, + fd, access_flags); - umem_dmabuf = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, - access_flags, - &mlx5_ib_dmabuf_attach_ops); if (IS_ERR(umem_dmabuf)) { mlx5_ib_dbg(dev, "umem_dmabuf get failed (%ld)\n", PTR_ERR(umem_dmabuf)); @@ -1535,7 +1604,7 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, } mr = alloc_cacheable_mr(pd, &umem_dmabuf->umem, virt_addr, - access_flags); + access_flags, access_mode); if (IS_ERR(mr)) { ib_umem_release(&umem_dmabuf->umem); return ERR_CAST(mr); @@ -1545,9 +1614,13 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, atomic_add(ib_umem_num_pages(mr->umem), &dev->mdev->priv.reg_pages); umem_dmabuf->private = mr; - err = mlx5r_store_odp_mkey(dev, &mr->mmkey); - if (err) - goto err_dereg_mr; + if (!pinned_mode) { + err = mlx5r_store_odp_mkey(dev, &mr->mmkey); + if (err) + goto err_dereg_mr; + } else { + mr->data_direct = true; + } err = mlx5_ib_init_dmabuf_mr(mr); if (err) @@ -1555,10 +1628,101 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, return &mr->ibmr; err_dereg_mr: - mlx5_ib_dereg_mr(&mr->ibmr, NULL); + __mlx5_ib_dereg_mr(&mr->ibmr); return ERR_PTR(err); } +static struct ib_mr * +reg_user_mr_dmabuf_by_data_direct(struct ib_pd *pd, u64 offset, + u64 length, u64 virt_addr, + int fd, int access_flags) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + struct mlx5_data_direct_dev *data_direct_dev; + struct ib_mr *crossing_mr; + struct ib_mr *crossed_mr; + int ret = 0; + + /* As of HW behaviour the IOVA must be page aligned in KSM mode */ + if (!PAGE_ALIGNED(virt_addr) || (access_flags & IB_ACCESS_ON_DEMAND)) + return ERR_PTR(-EOPNOTSUPP); + + mutex_lock(&dev->data_direct_lock); + data_direct_dev = dev->data_direct_dev; + if (!data_direct_dev) { + ret = -EINVAL; + goto end; + } + + /* The device's 'data direct mkey' was created without RO flags to + * simplify things and allow for a single mkey per device. + * Since RO is not a must, mask it out accordingly. + */ + access_flags &= ~IB_ACCESS_RELAXED_ORDERING; + crossed_mr = reg_user_mr_dmabuf(pd, &data_direct_dev->pdev->dev, + offset, length, virt_addr, fd, + access_flags, MLX5_MKC_ACCESS_MODE_KSM); + if (IS_ERR(crossed_mr)) { + ret = PTR_ERR(crossed_mr); + goto end; + } + + mutex_lock(&dev->slow_path_mutex); + crossing_mr = reg_create_crossing_vhca_mr(pd, virt_addr, length, access_flags, + crossed_mr->lkey); + mutex_unlock(&dev->slow_path_mutex); + if (IS_ERR(crossing_mr)) { + __mlx5_ib_dereg_mr(crossed_mr); + ret = PTR_ERR(crossing_mr); + goto end; + } + + list_add_tail(&to_mmr(crossed_mr)->dd_node, &dev->data_direct_mr_list); + to_mmr(crossing_mr)->dd_crossed_mr = to_mmr(crossed_mr); + to_mmr(crossing_mr)->data_direct = true; +end: + mutex_unlock(&dev->data_direct_lock); + return ret ? ERR_PTR(ret) : crossing_mr; +} + +struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, + u64 length, u64 virt_addr, + int fd, int access_flags, + struct uverbs_attr_bundle *attrs) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + int mlx5_access_flags = 0; + int err; + + if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM) || + !IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) + return ERR_PTR(-EOPNOTSUPP); + + if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS)) { + err = uverbs_get_flags32(&mlx5_access_flags, attrs, + MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + MLX5_IB_UAPI_REG_DMABUF_ACCESS_DATA_DIRECT); + if (err) + return ERR_PTR(err); + } + + mlx5_ib_dbg(dev, + "offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x, mlx5_access_flags 0x%x\n", + offset, virt_addr, length, fd, access_flags, mlx5_access_flags); + + /* dmabuf requires xlt update via umr to work. */ + if (!mlx5r_umr_can_load_pas(dev, length)) + return ERR_PTR(-EINVAL); + + if (mlx5_access_flags & MLX5_IB_UAPI_REG_DMABUF_ACCESS_DATA_DIRECT) + return reg_user_mr_dmabuf_by_data_direct(pd, offset, length, virt_addr, + fd, access_flags); + + return reg_user_mr_dmabuf(pd, pd->device->dma_device, + offset, length, virt_addr, + fd, access_flags, MLX5_MKC_ACCESS_MODE_MTT); +} + /* * True if the change in access flags can be done via UMR, only some access * flags can be updated. @@ -1570,7 +1734,8 @@ static bool can_use_umr_rereg_access(struct mlx5_ib_dev *dev, unsigned int diffs = current_access_flags ^ target_access_flags; if (diffs & ~(IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | - IB_ACCESS_REMOTE_READ | IB_ACCESS_RELAXED_ORDERING)) + IB_ACCESS_REMOTE_READ | IB_ACCESS_RELAXED_ORDERING | + IB_ACCESS_REMOTE_ATOMIC)) return false; return mlx5r_umr_can_reconfig(dev, current_access_flags, target_access_flags); @@ -1589,8 +1754,7 @@ static bool can_use_umr_rereg_pas(struct mlx5_ib_mr *mr, if (!mlx5r_umr_can_load_pas(dev, new_umem->length)) return false; - *page_size = - mlx5_umem_find_best_pgsz(new_umem, mkc, log_page_size, 0, iova); + *page_size = mlx5_umem_mkc_find_best_pgsz(dev, new_umem, iova); if (WARN_ON(!*page_size)) return false; return (mr->mmkey.cache_ent->rb_key.ndescs) >= @@ -1653,7 +1817,7 @@ struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, struct mlx5_ib_mr *mr = to_mmr(ib_mr); int err; - if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM)) + if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM) || mr->data_direct) return ERR_PTR(-EOPNOTSUPP); mlx5_ib_dbg( @@ -1781,7 +1945,8 @@ err: static void mlx5_free_priv_descs(struct mlx5_ib_mr *mr) { - if (!mr->umem && mr->descs) { + if (!mr->umem && !mr->data_direct && + mr->ibmr.type != IB_MR_TYPE_DM && mr->descs) { struct ib_device *device = mr->ibmr.device; int size = mr->max_descs * mr->desc_size; struct mlx5_ib_dev *dev = to_mdev(device); @@ -1802,7 +1967,6 @@ static int cache_ent_find_and_store(struct mlx5_ib_dev *dev, if (mr->mmkey.cache_ent) { spin_lock_irq(&mr->mmkey.cache_ent->mkeys_queue.lock); - mr->mmkey.cache_ent->in_use--; goto end; } @@ -1835,7 +1999,89 @@ end: return ret; } -int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) +static int mlx5_ib_revoke_data_direct_mr(struct mlx5_ib_mr *mr) +{ + struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device); + struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem); + int err; + + lockdep_assert_held(&dev->data_direct_lock); + mr->revoked = true; + err = mlx5r_umr_revoke_mr(mr); + if (WARN_ON(err)) + return err; + + ib_umem_dmabuf_revoke(umem_dmabuf); + return 0; +} + +void mlx5_ib_revoke_data_direct_mrs(struct mlx5_ib_dev *dev) +{ + struct mlx5_ib_mr *mr, *next; + + lockdep_assert_held(&dev->data_direct_lock); + + list_for_each_entry_safe(mr, next, &dev->data_direct_mr_list, dd_node) { + list_del(&mr->dd_node); + mlx5_ib_revoke_data_direct_mr(mr); + } +} + +static int mlx5_revoke_mr(struct mlx5_ib_mr *mr) +{ + struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device); + struct mlx5_cache_ent *ent = mr->mmkey.cache_ent; + bool is_odp = is_odp_mr(mr); + bool is_odp_dma_buf = is_dmabuf_mr(mr) && + !to_ib_umem_dmabuf(mr->umem)->pinned; + bool from_cache = !!ent; + int ret = 0; + + if (is_odp) + mutex_lock(&to_ib_umem_odp(mr->umem)->umem_mutex); + + if (is_odp_dma_buf) + dma_resv_lock(to_ib_umem_dmabuf(mr->umem)->attach->dmabuf->resv, NULL); + + if (mr->mmkey.cacheable && !mlx5r_umr_revoke_mr(mr) && !cache_ent_find_and_store(dev, mr)) { + ent = mr->mmkey.cache_ent; + /* upon storing to a clean temp entry - schedule its cleanup */ + spin_lock_irq(&ent->mkeys_queue.lock); + if (from_cache) + ent->in_use--; + if (ent->is_tmp && !ent->tmp_cleanup_scheduled) { + mod_delayed_work(ent->dev->cache.wq, &ent->dwork, + secs_to_jiffies(30)); + ent->tmp_cleanup_scheduled = true; + } + spin_unlock_irq(&ent->mkeys_queue.lock); + goto out; + } + + if (ent) { + spin_lock_irq(&ent->mkeys_queue.lock); + ent->in_use--; + mr->mmkey.cache_ent = NULL; + spin_unlock_irq(&ent->mkeys_queue.lock); + } + ret = destroy_mkey(dev, mr); +out: + if (is_odp) { + if (!ret) + to_ib_umem_odp(mr->umem)->private = NULL; + mutex_unlock(&to_ib_umem_odp(mr->umem)->umem_mutex); + } + + if (is_odp_dma_buf) { + if (!ret) + to_ib_umem_dmabuf(mr->umem)->private = NULL; + dma_resv_unlock(to_ib_umem_dmabuf(mr->umem)->attach->dmabuf->resv); + } + + return ret; +} + +static int __mlx5_ib_dereg_mr(struct ib_mr *ibmr) { struct mlx5_ib_mr *mr = to_mmr(ibmr); struct mlx5_ib_dev *dev = to_mdev(ibmr->device); @@ -1880,16 +2126,9 @@ int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) } /* Stop DMA */ - if (mr->umem && mlx5r_umr_can_load_pas(dev, mr->umem->length)) - if (mlx5r_umr_revoke_mr(mr) || - cache_ent_find_and_store(dev, mr)) - mr->mmkey.cache_ent = NULL; - - if (!mr->mmkey.cache_ent) { - rc = destroy_mkey(to_mdev(mr->ibmr.device), mr); - if (rc) - return rc; - } + rc = mlx5_revoke_mr(mr); + if (rc) + return rc; if (mr->umem) { bool is_odp = is_odp_mr(mr); @@ -1909,9 +2148,40 @@ int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) return 0; } +static int dereg_crossing_data_direct_mr(struct mlx5_ib_dev *dev, + struct mlx5_ib_mr *mr) +{ + struct mlx5_ib_mr *dd_crossed_mr = mr->dd_crossed_mr; + int ret; + + ret = __mlx5_ib_dereg_mr(&mr->ibmr); + if (ret) + return ret; + + mutex_lock(&dev->data_direct_lock); + if (!dd_crossed_mr->revoked) + list_del(&dd_crossed_mr->dd_node); + + ret = __mlx5_ib_dereg_mr(&dd_crossed_mr->ibmr); + mutex_unlock(&dev->data_direct_lock); + return ret; +} + +int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) +{ + struct mlx5_ib_mr *mr = to_mmr(ibmr); + struct mlx5_ib_dev *dev = to_mdev(ibmr->device); + + if (mr->data_direct) + return dereg_crossing_data_direct_mr(dev, mr); + + return __mlx5_ib_dereg_mr(ibmr); +} + static void mlx5_set_umr_free_mkey(struct ib_pd *pd, u32 *in, int ndescs, int access_mode, int page_shift) { + struct mlx5_ib_dev *dev = to_mdev(pd->device); void *mkc; mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); @@ -1924,6 +2194,9 @@ static void mlx5_set_umr_free_mkey(struct ib_pd *pd, u32 *in, int ndescs, MLX5_SET(mkc, mkc, access_mode_4_2, (access_mode >> 2) & 0x7); MLX5_SET(mkc, mkc, umr_en, 1); MLX5_SET(mkc, mkc, log_page_size, page_shift); + if (access_mode == MLX5_MKC_ACCESS_MODE_PA || + access_mode == MLX5_MKC_ACCESS_MODE_MTT) + MLX5_SET(mkc, mkc, ma_translation_mode, MLX5_CAP_GEN(dev->mdev, ats)); } static int _mlx5_alloc_mkey_descs(struct ib_pd *pd, struct mlx5_ib_mr *mr, diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index 4a04cbc5b78a..eaa2f9f5f3a9 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -34,6 +34,9 @@ #include <linux/kernel.h> #include <linux/dma-buf.h> #include <linux/dma-resv.h> +#include <linux/hmm.h> +#include <linux/hmm-dma.h> +#include <linux/pci-p2pdma.h> #include "mlx5_ib.h" #include "cmd.h" @@ -45,7 +48,7 @@ /* Contains the details of a pagefault. */ struct mlx5_pagefault { u32 bytes_committed; - u32 token; + u64 token; u8 event_subtype; u8 type; union { @@ -74,6 +77,14 @@ struct mlx5_pagefault { u32 rdma_op_len; u64 rdma_va; } rdma; + struct { + u64 va; + u32 mkey; + u32 fault_byte_count; + u32 prefetch_before_byte_count; + u32 prefetch_after_byte_count; + u8 flags; + } memory; }; struct mlx5_ib_pf_eq *eq; @@ -99,13 +110,20 @@ static u64 mlx5_imr_ksm_entries; static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries, struct mlx5_ib_mr *imr, int flags) { + struct mlx5_core_dev *dev = mr_to_mdev(imr)->mdev; struct mlx5_klm *end = pklm + nentries; + int step = MLX5_CAP_ODP(dev, mem_page_fault) ? MLX5_IMR_MTT_SIZE : 0; + __be32 key = MLX5_CAP_ODP(dev, mem_page_fault) ? + cpu_to_be32(imr->null_mmkey.key) : + mr_to_mdev(imr)->mkeys.null_mkey; + u64 va = + MLX5_CAP_ODP(dev, mem_page_fault) ? idx * MLX5_IMR_MTT_SIZE : 0; if (flags & MLX5_IB_UPD_XLT_ZAP) { - for (; pklm != end; pklm++, idx++) { + for (; pklm != end; pklm++, idx++, va += step) { pklm->bcount = cpu_to_be32(MLX5_IMR_MTT_SIZE); - pklm->key = mr_to_mdev(imr)->mkeys.null_mkey; - pklm->va = 0; + pklm->key = key; + pklm->va = cpu_to_be64(va); } return; } @@ -129,7 +147,7 @@ static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries, */ lockdep_assert_held(&to_ib_umem_odp(imr->umem)->umem_mutex); - for (; pklm != end; pklm++, idx++) { + for (; pklm != end; pklm++, idx++, va += step) { struct mlx5_ib_mr *mtt = xa_load(&imr->implicit_children, idx); pklm->bcount = cpu_to_be32(MLX5_IMR_MTT_SIZE); @@ -137,47 +155,56 @@ static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries, pklm->key = cpu_to_be32(mtt->ibmr.lkey); pklm->va = cpu_to_be64(idx * MLX5_IMR_MTT_SIZE); } else { - pklm->key = mr_to_mdev(imr)->mkeys.null_mkey; - pklm->va = 0; + pklm->key = key; + pklm->va = cpu_to_be64(va); } } } -static u64 umem_dma_to_mtt(dma_addr_t umem_dma) -{ - u64 mtt_entry = umem_dma & ODP_DMA_ADDR_MASK; - - if (umem_dma & ODP_READ_ALLOWED_BIT) - mtt_entry |= MLX5_IB_MTT_READ; - if (umem_dma & ODP_WRITE_ALLOWED_BIT) - mtt_entry |= MLX5_IB_MTT_WRITE; - - return mtt_entry; -} - -static void populate_mtt(__be64 *pas, size_t idx, size_t nentries, - struct mlx5_ib_mr *mr, int flags) +static int populate_mtt(__be64 *pas, size_t start, size_t nentries, + struct mlx5_ib_mr *mr, int flags) { struct ib_umem_odp *odp = to_ib_umem_odp(mr->umem); - dma_addr_t pa; + bool downgrade = flags & MLX5_IB_UPD_XLT_DOWNGRADE; + struct pci_p2pdma_map_state p2pdma_state = {}; + struct ib_device *dev = odp->umem.ibdev; size_t i; if (flags & MLX5_IB_UPD_XLT_ZAP) - return; + return 0; for (i = 0; i < nentries; i++) { - pa = odp->dma_list[idx + i]; - pas[i] = cpu_to_be64(umem_dma_to_mtt(pa)); + unsigned long pfn = odp->map.pfn_list[start + i]; + dma_addr_t dma_addr; + + pfn = odp->map.pfn_list[start + i]; + if (!(pfn & HMM_PFN_VALID)) + /* ODP initialization */ + continue; + + dma_addr = hmm_dma_map_pfn(dev->dma_device, &odp->map, + start + i, &p2pdma_state); + if (ib_dma_mapping_error(dev, dma_addr)) + return -EFAULT; + + dma_addr |= MLX5_IB_MTT_READ; + if ((pfn & HMM_PFN_WRITE) && !downgrade) + dma_addr |= MLX5_IB_MTT_WRITE; + + pas[i] = cpu_to_be64(dma_addr); + odp->npages++; } + return 0; } -void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, - struct mlx5_ib_mr *mr, int flags) +int mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, + struct mlx5_ib_mr *mr, int flags) { if (flags & MLX5_IB_UPD_XLT_INDIRECT) { populate_klm(xlt, idx, nentries, mr, flags); + return 0; } else { - populate_mtt(xlt, idx, nentries, mr, flags); + return populate_mtt(xlt, idx, nentries, mr, flags); } } @@ -213,10 +240,28 @@ static void destroy_unused_implicit_child_mr(struct mlx5_ib_mr *mr) unsigned long idx = ib_umem_start(odp) >> MLX5_IMR_MTT_SHIFT; struct mlx5_ib_mr *imr = mr->parent; + /* + * If userspace is racing freeing the parent implicit ODP MR then we can + * loose the race with parent destruction. In this case + * mlx5_ib_free_odp_mr() will free everything in the implicit_children + * xarray so NOP is fine. This child MR cannot be destroyed here because + * we are under its umem_mutex. + */ if (!refcount_inc_not_zero(&imr->mmkey.usecount)) return; - xa_erase(&imr->implicit_children, idx); + xa_lock(&imr->implicit_children); + if (__xa_cmpxchg(&imr->implicit_children, idx, mr, NULL, GFP_KERNEL) != + mr) { + xa_unlock(&imr->implicit_children); + mlx5r_deref_odp_mkey(&imr->mmkey); + return; + } + + if (MLX5_CAP_ODP(mr_to_mdev(mr)->mdev, mem_page_fault)) + __xa_erase(&mr_to_mdev(mr)->odp_mkeys, + mlx5_base_mkey(mr->mmkey.key)); + xa_unlock(&imr->implicit_children); /* Freeing a MR is a sleeping operation, so bounce to a work queue */ INIT_WORK(&mr->odp_destroy.work, free_implicit_child_mr_work); @@ -250,6 +295,8 @@ static bool mlx5_ib_invalidate_range(struct mmu_interval_notifier *mni, if (!umem_odp->npages) goto out; mr = umem_odp->private; + if (!mr) + goto out; start = max_t(u64, ib_umem_start(umem_odp), range->start); end = min_t(u64, ib_umem_end(umem_odp), range->end); @@ -268,15 +315,11 @@ static bool mlx5_ib_invalidate_range(struct mmu_interval_notifier *mni, * estimate the cost of another UMR vs. the cost of bigger * UMR. */ - if (umem_odp->dma_list[idx] & - (ODP_READ_ALLOWED_BIT | ODP_WRITE_ALLOWED_BIT)) { + if (umem_odp->map.pfn_list[idx] & HMM_PFN_VALID) { if (!in_block) { blk_start_idx = idx; in_block = 1; } - - /* Count page invalidations */ - invalidations += idx - blk_start_idx + 1; } else { u64 umr_offset = idx & umr_block_mask; @@ -286,16 +329,21 @@ static bool mlx5_ib_invalidate_range(struct mmu_interval_notifier *mni, MLX5_IB_UPD_XLT_ZAP | MLX5_IB_UPD_XLT_ATOMIC); in_block = 0; + /* Count page invalidations */ + invalidations += idx - blk_start_idx + 1; } } } - if (in_block) + if (in_block) { mlx5r_umr_update_xlt(mr, blk_start_idx, idx - blk_start_idx + 1, 0, MLX5_IB_UPD_XLT_ZAP | MLX5_IB_UPD_XLT_ATOMIC); + /* Count page invalidations */ + invalidations += idx - blk_start_idx + 1; + } - mlx5_update_odp_stats(mr, invalidations, invalidations); + mlx5_update_odp_stats_with_handled(mr, invalidations, invalidations); /* * We are now sure that the device will not access the @@ -332,46 +380,46 @@ static void internal_fill_odp_caps(struct mlx5_ib_dev *dev) else dev->odp_max_size = BIT_ULL(MLX5_MAX_UMR_SHIFT + PAGE_SHIFT); - if (MLX5_CAP_ODP(dev->mdev, ud_odp_caps.send)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, ud_odp_caps.send)) caps->per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SEND; - if (MLX5_CAP_ODP(dev->mdev, ud_odp_caps.srq_receive)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, ud_odp_caps.srq_receive)) caps->per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV; - if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.send)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, rc_odp_caps.send)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SEND; - if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.receive)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, rc_odp_caps.receive)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_RECV; - if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.write)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, rc_odp_caps.write)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_WRITE; - if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.read)) + if (MLX5_CAP_ODP_SCHEME(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)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, rc_odp_caps.atomic)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC; - if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.srq_receive)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, rc_odp_caps.srq_receive)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV; - if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.send)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, xrc_odp_caps.send)) caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_SEND; - if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.receive)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, xrc_odp_caps.receive)) caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_RECV; - if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.write)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, xrc_odp_caps.write)) caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_WRITE; - if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.read)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, xrc_odp_caps.read)) caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_READ; - if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.atomic)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, xrc_odp_caps.atomic)) caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_ATOMIC; - if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.srq_receive)) + if (MLX5_CAP_ODP_SCHEME(dev->mdev, xrc_odp_caps.srq_receive)) caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV; if (MLX5_CAP_GEN(dev->mdev, fixed_buffer_size) && @@ -388,13 +436,29 @@ static void mlx5_ib_page_fault_resume(struct mlx5_ib_dev *dev, int wq_num = pfault->event_subtype == MLX5_PFAULT_SUBTYPE_WQE ? pfault->wqe.wq_num : pfault->token; u32 in[MLX5_ST_SZ_DW(page_fault_resume_in)] = {}; + void *info; int err; MLX5_SET(page_fault_resume_in, in, opcode, MLX5_CMD_OP_PAGE_FAULT_RESUME); - MLX5_SET(page_fault_resume_in, in, page_fault_type, pfault->type); - MLX5_SET(page_fault_resume_in, in, token, pfault->token); - MLX5_SET(page_fault_resume_in, in, wq_number, wq_num); - MLX5_SET(page_fault_resume_in, in, error, !!error); + + if (pfault->event_subtype == MLX5_PFAULT_SUBTYPE_MEMORY) { + info = MLX5_ADDR_OF(page_fault_resume_in, in, + page_fault_info.mem_page_fault_info); + MLX5_SET(mem_page_fault_info, info, fault_token_31_0, + pfault->token & 0xffffffff); + MLX5_SET(mem_page_fault_info, info, fault_token_47_32, + (pfault->token >> 32) & 0xffff); + MLX5_SET(mem_page_fault_info, info, error, !!error); + } else { + info = MLX5_ADDR_OF(page_fault_resume_in, in, + page_fault_info.trans_page_fault_info); + MLX5_SET(trans_page_fault_info, info, page_fault_type, + pfault->type); + MLX5_SET(trans_page_fault_info, info, fault_token, + pfault->token); + MLX5_SET(trans_page_fault_info, info, wq_number, wq_num); + MLX5_SET(trans_page_fault_info, info, error, !!error); + } err = mlx5_cmd_exec_in(dev->mdev, page_fault_resume, in); if (err) @@ -466,8 +530,18 @@ static struct mlx5_ib_mr *implicit_get_child_mr(struct mlx5_ib_mr *imr, refcount_inc(&ret->mmkey.usecount); goto out_lock; } - xa_unlock(&imr->implicit_children); + if (MLX5_CAP_ODP(dev->mdev, mem_page_fault)) { + ret = __xa_store(&dev->odp_mkeys, mlx5_base_mkey(mr->mmkey.key), + &mr->mmkey, GFP_KERNEL); + if (xa_is_err(ret)) { + ret = ERR_PTR(xa_err(ret)); + __xa_erase(&imr->implicit_children, idx); + goto out_lock; + } + mr->mmkey.type = MLX5_MKEY_IMPLICIT_CHILD; + } + xa_unlock(&imr->implicit_children); mlx5_ib_dbg(mr_to_mdev(imr), "key %x mr %p\n", mr->mmkey.key, mr); return mr; @@ -478,6 +552,57 @@ out_mr: return ret; } +/* + * When using memory scheme ODP, implicit MRs can't use the reserved null mkey + * and each implicit MR needs to assign a private null mkey to get the page + * faults on. + * The null mkey is created with the properties to enable getting the page + * fault for every time it is accessed and having all relevant access flags. + */ +static int alloc_implicit_mr_null_mkey(struct mlx5_ib_dev *dev, + struct mlx5_ib_mr *imr, + struct mlx5_ib_pd *pd) +{ + size_t inlen = MLX5_ST_SZ_BYTES(create_mkey_in) + 64; + void *mkc; + u32 *in; + int err; + + in = kzalloc(inlen, GFP_KERNEL); + if (!in) + return -ENOMEM; + + MLX5_SET(create_mkey_in, in, translations_octword_actual_size, 4); + MLX5_SET(create_mkey_in, in, pg_access, 1); + + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + MLX5_SET(mkc, mkc, a, 1); + MLX5_SET(mkc, mkc, rw, 1); + MLX5_SET(mkc, mkc, rr, 1); + MLX5_SET(mkc, mkc, lw, 1); + MLX5_SET(mkc, mkc, lr, 1); + MLX5_SET(mkc, mkc, free, 0); + MLX5_SET(mkc, mkc, umr_en, 0); + MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT); + + MLX5_SET(mkc, mkc, translations_octword_size, 4); + MLX5_SET(mkc, mkc, log_page_size, 61); + MLX5_SET(mkc, mkc, length64, 1); + MLX5_SET(mkc, mkc, pd, pd->pdn); + MLX5_SET64(mkc, mkc, start_addr, 0); + MLX5_SET(mkc, mkc, qpn, 0xffffff); + + err = mlx5_core_create_mkey(dev->mdev, &imr->null_mmkey.key, in, inlen); + if (err) + goto free_in; + + imr->null_mmkey.type = MLX5_MKEY_NULL; + +free_in: + kfree(in); + return err; +} + struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, int access_flags) { @@ -510,6 +635,16 @@ struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, imr->is_odp_implicit = true; xa_init(&imr->implicit_children); + if (MLX5_CAP_ODP(dev->mdev, mem_page_fault)) { + err = alloc_implicit_mr_null_mkey(dev, imr, pd); + if (err) + goto out_mr; + + err = mlx5r_store_odp_mkey(dev, &imr->null_mmkey); + if (err) + goto out_mr; + } + err = mlx5r_umr_update_xlt(imr, 0, mlx5_imr_ksm_entries, MLX5_KSM_PAGE_SHIFT, @@ -544,6 +679,14 @@ void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr) xa_erase(&mr->implicit_children, idx); mlx5_ib_dereg_mr(&mtt->ibmr, NULL); } + + if (mr->null_mmkey.key) { + xa_erase(&mr_to_mdev(mr)->odp_mkeys, + mlx5_base_mkey(mr->null_mmkey.key)); + + mlx5_core_destroy_mkey(mr_to_mdev(mr)->mdev, + mr->null_mmkey.key); + } } #define MLX5_PF_FLAGS_DOWNGRADE BIT(1) @@ -555,7 +698,7 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp, { int page_shift, ret, np; bool downgrade = flags & MLX5_PF_FLAGS_DOWNGRADE; - u64 access_mask; + u64 access_mask = 0; u64 start_idx; bool fault = !(flags & MLX5_PF_FLAGS_SNAPSHOT); u32 xlt_flags = MLX5_IB_UPD_XLT_ATOMIC; @@ -563,12 +706,14 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp, if (flags & MLX5_PF_FLAGS_ENABLE) xlt_flags |= MLX5_IB_UPD_XLT_ENABLE; + if (flags & MLX5_PF_FLAGS_DOWNGRADE) + xlt_flags |= MLX5_IB_UPD_XLT_DOWNGRADE; + page_shift = odp->page_shift; start_idx = (user_va - ib_umem_start(odp)) >> page_shift; - access_mask = ODP_READ_ALLOWED_BIT; if (odp->umem.writable && !downgrade) - access_mask |= ODP_WRITE_ALLOWED_BIT; + access_mask |= HMM_PFN_WRITE; np = ib_umem_odp_map_dma_and_lock(odp, user_va, bcnt, access_mask, fault); if (np < 0) @@ -693,7 +838,7 @@ static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt, struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem); u32 xlt_flags = 0; int err; - unsigned int page_size; + unsigned long page_size; if (flags & MLX5_PF_FLAGS_ENABLE) xlt_flags |= MLX5_IB_UPD_XLT_ENABLE; @@ -705,14 +850,15 @@ static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt, return err; } - page_size = mlx5_umem_find_best_pgsz(&umem_dmabuf->umem, mkc, - log_page_size, 0, - umem_dmabuf->umem.iova); - if (unlikely(page_size < PAGE_SIZE)) { + page_size = mlx5_umem_dmabuf_find_best_pgsz(umem_dmabuf); + if (!page_size) { ib_umem_dmabuf_unmap_pages(umem_dmabuf); err = -EINVAL; } else { - err = mlx5r_umr_update_mr_pas(mr, xlt_flags); + if (mr->data_direct) + err = mlx5r_umr_update_data_direct_ksm_pas(mr, xlt_flags); + else + err = mlx5r_umr_update_mr_pas(mr, xlt_flags); } dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv); @@ -735,24 +881,31 @@ static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt, * >0: Number of pages mapped */ static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt, - u32 *bytes_mapped, u32 flags) + u32 *bytes_mapped, u32 flags, bool permissive_fault) { struct ib_umem_odp *odp = to_ib_umem_odp(mr->umem); - if (unlikely(io_virt < mr->ibmr.iova)) + if (unlikely(io_virt < mr->ibmr.iova) && !permissive_fault) return -EFAULT; if (mr->umem->is_dmabuf) return pagefault_dmabuf_mr(mr, bcnt, bytes_mapped, flags); if (!odp->is_implicit_odp) { + u64 offset = io_virt < mr->ibmr.iova ? 0 : io_virt - mr->ibmr.iova; u64 user_va; - if (check_add_overflow(io_virt - mr->ibmr.iova, - (u64)odp->umem.address, &user_va)) + if (check_add_overflow(offset, (u64)odp->umem.address, + &user_va)) return -EFAULT; - if (unlikely(user_va >= ib_umem_end(odp) || - ib_umem_end(odp) - user_va < bcnt)) + + if (permissive_fault) { + if (user_va < ib_umem_start(odp)) + user_va = ib_umem_start(odp); + if ((user_va + bcnt) > ib_umem_end(odp)) + bcnt = ib_umem_end(odp) - user_va; + } else if (unlikely(user_va >= ib_umem_end(odp) || + ib_umem_end(odp) - user_va < bcnt)) return -EFAULT; return pagefault_real_mr(mr, odp, user_va, bcnt, bytes_mapped, flags); @@ -799,11 +952,31 @@ static bool mkey_is_eq(struct mlx5_ib_mkey *mmkey, u32 key) return mmkey->key == key; } +static struct mlx5_ib_mkey *find_odp_mkey(struct mlx5_ib_dev *dev, u32 key) +{ + struct mlx5_ib_mkey *mmkey; + + xa_lock(&dev->odp_mkeys); + mmkey = xa_load(&dev->odp_mkeys, mlx5_base_mkey(key)); + if (!mmkey) { + mmkey = ERR_PTR(-ENOENT); + goto out; + } + if (!mkey_is_eq(mmkey, key)) { + mmkey = ERR_PTR(-EFAULT); + goto out; + } + refcount_inc(&mmkey->usecount); +out: + xa_unlock(&dev->odp_mkeys); + + return mmkey; +} + /* * Handle a single data segment in a page-fault WQE or RDMA region. * - * Returns number of OS pages retrieved on success. The caller may continue to - * the next data segment. + * Returns zero on success. The caller may continue to the next data segment. * Can return the following error codes: * -EAGAIN to designate a temporary error. The caller will abort handling the * page fault and resolve it. @@ -816,7 +989,7 @@ static int pagefault_single_data_segment(struct mlx5_ib_dev *dev, u32 *bytes_committed, u32 *bytes_mapped) { - int npages = 0, ret, i, outlen, cur_outlen = 0, depth = 0; + int ret, i, outlen, cur_outlen = 0, depth = 0, pages_in_range; struct pf_frame *head = NULL, *frame; struct mlx5_ib_mkey *mmkey; struct mlx5_ib_mr *mr; @@ -826,32 +999,24 @@ static int pagefault_single_data_segment(struct mlx5_ib_dev *dev, io_virt += *bytes_committed; bcnt -= *bytes_committed; - next_mr: - xa_lock(&dev->odp_mkeys); - mmkey = xa_load(&dev->odp_mkeys, mlx5_base_mkey(key)); - if (!mmkey) { - xa_unlock(&dev->odp_mkeys); - mlx5_ib_dbg( - dev, - "skipping non ODP MR (lkey=0x%06x) in page fault handler.\n", - key); - if (bytes_mapped) - *bytes_mapped += bcnt; - /* - * The user could specify a SGL with multiple lkeys and only - * some of them are ODP. Treat the non-ODP ones as fully - * faulted. - */ - ret = 0; - goto end; - } - refcount_inc(&mmkey->usecount); - xa_unlock(&dev->odp_mkeys); - - if (!mkey_is_eq(mmkey, key)) { - mlx5_ib_dbg(dev, "failed to find mkey %x\n", key); - ret = -EFAULT; + mmkey = find_odp_mkey(dev, key); + if (IS_ERR(mmkey)) { + ret = PTR_ERR(mmkey); + if (ret == -ENOENT) { + mlx5_ib_dbg( + dev, + "skipping non ODP MR (lkey=0x%06x) in page fault handler.\n", + key); + if (bytes_mapped) + *bytes_mapped += bcnt; + /* + * The user could specify a SGL with multiple lkeys and + * only some of them are ODP. Treat the non-ODP ones as + * fully faulted. + */ + ret = 0; + } goto end; } @@ -859,13 +1024,20 @@ next_mr: case MLX5_MKEY_MR: mr = container_of(mmkey, struct mlx5_ib_mr, mmkey); - ret = pagefault_mr(mr, io_virt, bcnt, bytes_mapped, 0); + pages_in_range = (ALIGN(io_virt + bcnt, PAGE_SIZE) - + (io_virt & PAGE_MASK)) >> + PAGE_SHIFT; + ret = pagefault_mr(mr, io_virt, bcnt, bytes_mapped, 0, false); if (ret < 0) goto end; - mlx5_update_odp_stats(mr, faults, ret); + mlx5_update_odp_stats_with_handled(mr, faults, ret); + + if (ret < pages_in_range) { + ret = -EFAULT; + goto end; + } - npages += ret; ret = 0; break; @@ -946,7 +1118,7 @@ next_mr: } end: - if (mmkey) + if (!IS_ERR(mmkey)) mlx5r_deref_odp_mkey(mmkey); while (head) { frame = head; @@ -956,7 +1128,7 @@ end: kfree(out); *bytes_committed = 0; - return ret ? ret : npages; + return ret; } /* @@ -975,8 +1147,7 @@ end: * the committed bytes). * @receive_queue: receive WQE end of sg list * - * Returns the number of pages loaded if positive, zero for an empty WQE, or a - * negative error code. + * Returns zero for success or a negative error code. */ static int pagefault_data_segments(struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfault, @@ -984,7 +1155,7 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, void *wqe_end, u32 *bytes_mapped, u32 *total_wqe_bytes, bool receive_queue) { - int ret = 0, npages = 0; + int ret = 0; u64 io_virt; __be32 key; u32 byte_count; @@ -1041,10 +1212,9 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, bytes_mapped); if (ret < 0) break; - npages += ret; } - return ret < 0 ? ret : npages; + return ret; } /* @@ -1268,7 +1438,7 @@ read_user: if (ret) mlx5_ib_err( dev, - "Failed reading a WQE following page fault, error %d, wqe_index %x, qpn %x\n", + "Failed reading a WQE following page fault, error %d, wqe_index %x, qpn %llx\n", ret, wqe_index, pfault->token); resolve_page_fault: @@ -1280,12 +1450,6 @@ resolve_page_fault: free_page((unsigned long)wqe_start); } -static int pages_in_range(u64 address, u32 length) -{ - return (ALIGN(address + length, PAGE_SIZE) - - (address & PAGE_MASK)) >> PAGE_SHIFT; -} - static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfault) { @@ -1324,16 +1488,16 @@ static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_dev *dev, if (ret == -EAGAIN) { /* We're racing with an invalidation, don't prefetch */ prefetch_activated = 0; - } else if (ret < 0 || pages_in_range(address, length) > ret) { + } else if (ret < 0) { mlx5_ib_page_fault_resume(dev, pfault, 1); if (ret != -ENOENT) - mlx5_ib_dbg(dev, "PAGE FAULT error %d. QP 0x%x, type: 0x%x\n", + mlx5_ib_dbg(dev, "PAGE FAULT error %d. QP 0x%llx, type: 0x%x\n", ret, pfault->token, pfault->type); return; } mlx5_ib_page_fault_resume(dev, pfault, 0); - mlx5_ib_dbg(dev, "PAGE FAULT completed. QP 0x%x, type: 0x%x, prefetch_activated: %d\n", + mlx5_ib_dbg(dev, "PAGE FAULT completed. QP 0x%llx, type: 0x%x, prefetch_activated: %d\n", pfault->token, pfault->type, prefetch_activated); @@ -1349,12 +1513,80 @@ static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_dev *dev, prefetch_len, &bytes_committed, NULL); if (ret < 0 && ret != -EAGAIN) { - mlx5_ib_dbg(dev, "Prefetch failed. ret: %d, QP 0x%x, address: 0x%.16llx, length = 0x%.16x\n", + mlx5_ib_dbg(dev, "Prefetch failed. ret: %d, QP 0x%llx, address: 0x%.16llx, length = 0x%.16x\n", ret, pfault->token, address, prefetch_len); } } } +#define MLX5_MEMORY_PAGE_FAULT_FLAGS_LAST BIT(7) +static void mlx5_ib_mr_memory_pfault_handler(struct mlx5_ib_dev *dev, + struct mlx5_pagefault *pfault) +{ + u64 prefetch_va = + pfault->memory.va - pfault->memory.prefetch_before_byte_count; + size_t prefetch_size = pfault->memory.prefetch_before_byte_count + + pfault->memory.fault_byte_count + + pfault->memory.prefetch_after_byte_count; + struct mlx5_ib_mkey *mmkey; + struct mlx5_ib_mr *mr, *child_mr; + int ret = 0; + + mmkey = find_odp_mkey(dev, pfault->memory.mkey); + if (IS_ERR(mmkey)) + goto err; + + switch (mmkey->type) { + case MLX5_MKEY_IMPLICIT_CHILD: + child_mr = container_of(mmkey, struct mlx5_ib_mr, mmkey); + mr = child_mr->parent; + break; + case MLX5_MKEY_NULL: + mr = container_of(mmkey, struct mlx5_ib_mr, null_mmkey); + break; + default: + mr = container_of(mmkey, struct mlx5_ib_mr, mmkey); + break; + } + + /* If prefetch fails, handle only demanded page fault */ + ret = pagefault_mr(mr, prefetch_va, prefetch_size, NULL, 0, true); + if (ret < 0) { + ret = pagefault_mr(mr, pfault->memory.va, + pfault->memory.fault_byte_count, NULL, 0, + true); + if (ret < 0) + goto err; + } + + mlx5_update_odp_stats_with_handled(mr, faults, ret); + mlx5r_deref_odp_mkey(mmkey); + + if (pfault->memory.flags & MLX5_MEMORY_PAGE_FAULT_FLAGS_LAST) + mlx5_ib_page_fault_resume(dev, pfault, 0); + + mlx5_ib_dbg( + dev, + "PAGE FAULT completed %s. token 0x%llx, mkey: 0x%x, va: 0x%llx, byte_count: 0x%x\n", + pfault->memory.flags & MLX5_MEMORY_PAGE_FAULT_FLAGS_LAST ? + "" : + "without resume cmd", + pfault->token, pfault->memory.mkey, pfault->memory.va, + pfault->memory.fault_byte_count); + + return; + +err: + if (!IS_ERR(mmkey)) + mlx5r_deref_odp_mkey(mmkey); + mlx5_ib_page_fault_resume(dev, pfault, 1); + mlx5_ib_dbg( + dev, + "PAGE FAULT error. token 0x%llx, mkey: 0x%x, va: 0x%llx, byte_count: 0x%x, err: %d\n", + pfault->token, pfault->memory.mkey, pfault->memory.va, + pfault->memory.fault_byte_count, ret); +} + static void mlx5_ib_pfault(struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfault) { u8 event_subtype = pfault->event_subtype; @@ -1366,6 +1598,9 @@ static void mlx5_ib_pfault(struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfaul case MLX5_PFAULT_SUBTYPE_RDMA: mlx5_ib_mr_rdma_pfault_handler(dev, pfault); break; + case MLX5_PFAULT_SUBTYPE_MEMORY: + mlx5_ib_mr_memory_pfault_handler(dev, pfault); + break; default: mlx5_ib_err(dev, "Invalid page fault event subtype: 0x%x\n", event_subtype); @@ -1384,6 +1619,7 @@ static void mlx5_ib_eqe_pf_action(struct work_struct *work) mempool_free(pfault, eq->pool); } +#define MEMORY_SCHEME_PAGE_FAULT_GRANULARITY 4096 static void mlx5_ib_eq_pf_process(struct mlx5_ib_pf_eq *eq) { struct mlx5_eqe_page_fault *pf_eqe; @@ -1400,15 +1636,12 @@ static void mlx5_ib_eq_pf_process(struct mlx5_ib_pf_eq *eq) pf_eqe = &eqe->data.page_fault; pfault->event_subtype = eqe->sub_type; - pfault->bytes_committed = be32_to_cpu(pf_eqe->bytes_committed); - - mlx5_ib_dbg(eq->dev, - "PAGE_FAULT: subtype: 0x%02x, bytes_committed: 0x%06x\n", - eqe->sub_type, pfault->bytes_committed); switch (eqe->sub_type) { case MLX5_PFAULT_SUBTYPE_RDMA: /* RDMA based event */ + pfault->bytes_committed = + be32_to_cpu(pf_eqe->rdma.bytes_committed); pfault->type = be32_to_cpu(pf_eqe->rdma.pftype_token) >> 24; pfault->token = @@ -1422,10 +1655,12 @@ static void mlx5_ib_eq_pf_process(struct mlx5_ib_pf_eq *eq) be32_to_cpu(pf_eqe->rdma.rdma_op_len); pfault->rdma.rdma_va = be64_to_cpu(pf_eqe->rdma.rdma_va); - mlx5_ib_dbg(eq->dev, - "PAGE_FAULT: type:0x%x, token: 0x%06x, r_key: 0x%08x\n", - pfault->type, pfault->token, - pfault->rdma.r_key); + mlx5_ib_dbg( + eq->dev, + "PAGE_FAULT: subtype: 0x%02x, bytes_committed: 0x%06x, type:0x%x, token: 0x%06llx, r_key: 0x%08x\n", + eqe->sub_type, pfault->bytes_committed, + pfault->type, pfault->token, + pfault->rdma.r_key); mlx5_ib_dbg(eq->dev, "PAGE_FAULT: rdma_op_len: 0x%08x, rdma_va: 0x%016llx\n", pfault->rdma.rdma_op_len, @@ -1434,6 +1669,8 @@ static void mlx5_ib_eq_pf_process(struct mlx5_ib_pf_eq *eq) case MLX5_PFAULT_SUBTYPE_WQE: /* WQE based event */ + pfault->bytes_committed = + be32_to_cpu(pf_eqe->wqe.bytes_committed); pfault->type = (be32_to_cpu(pf_eqe->wqe.pftype_wq) >> 24) & 0x7; pfault->token = @@ -1445,11 +1682,47 @@ static void mlx5_ib_eq_pf_process(struct mlx5_ib_pf_eq *eq) be16_to_cpu(pf_eqe->wqe.wqe_index); pfault->wqe.packet_size = be16_to_cpu(pf_eqe->wqe.packet_length); - mlx5_ib_dbg(eq->dev, - "PAGE_FAULT: type:0x%x, token: 0x%06x, wq_num: 0x%06x, wqe_index: 0x%04x\n", - pfault->type, pfault->token, - pfault->wqe.wq_num, - pfault->wqe.wqe_index); + mlx5_ib_dbg( + eq->dev, + "PAGE_FAULT: subtype: 0x%02x, bytes_committed: 0x%06x, type:0x%x, token: 0x%06llx, wq_num: 0x%06x, wqe_index: 0x%04x\n", + eqe->sub_type, pfault->bytes_committed, + pfault->type, pfault->token, pfault->wqe.wq_num, + pfault->wqe.wqe_index); + break; + + case MLX5_PFAULT_SUBTYPE_MEMORY: + /* Memory based event */ + pfault->bytes_committed = 0; + pfault->token = + be32_to_cpu(pf_eqe->memory.token31_0) | + ((u64)be16_to_cpu(pf_eqe->memory.token47_32) + << 32); + pfault->memory.va = be64_to_cpu(pf_eqe->memory.va); + pfault->memory.mkey = be32_to_cpu(pf_eqe->memory.mkey); + pfault->memory.fault_byte_count = (be32_to_cpu( + pf_eqe->memory.demand_fault_pages) >> 12) * + MEMORY_SCHEME_PAGE_FAULT_GRANULARITY; + pfault->memory.prefetch_before_byte_count = + be16_to_cpu( + pf_eqe->memory.pre_demand_fault_pages) * + MEMORY_SCHEME_PAGE_FAULT_GRANULARITY; + pfault->memory.prefetch_after_byte_count = + be16_to_cpu( + pf_eqe->memory.post_demand_fault_pages) * + MEMORY_SCHEME_PAGE_FAULT_GRANULARITY; + pfault->memory.flags = pf_eqe->memory.flags; + mlx5_ib_dbg( + eq->dev, + "PAGE_FAULT: subtype: 0x%02x, token: 0x%06llx, mkey: 0x%06x, fault_byte_count: 0x%06x, va: 0x%016llx, flags: 0x%02x\n", + eqe->sub_type, pfault->token, + pfault->memory.mkey, + pfault->memory.fault_byte_count, + pfault->memory.va, pfault->memory.flags); + mlx5_ib_dbg( + eq->dev, + "PAGE_FAULT: prefetch size: before: 0x%06x, after 0x%06x\n", + pfault->memory.prefetch_before_byte_count, + pfault->memory.prefetch_after_byte_count); break; default: @@ -1712,7 +1985,7 @@ static void mlx5_ib_prefetch_mr_work(struct work_struct *w) for (i = 0; i < work->num_sge; ++i) { ret = pagefault_mr(work->frags[i].mr, work->frags[i].io_virt, work->frags[i].length, &bytes_mapped, - work->pf_flags); + work->pf_flags, false); if (ret <= 0) continue; mlx5_update_odp_stats(work->frags[i].mr, prefetch, ret); @@ -1763,7 +2036,7 @@ static int mlx5_ib_prefetch_sg_list(struct ib_pd *pd, if (IS_ERR(mr)) return PTR_ERR(mr); ret = pagefault_mr(mr, sg_list[i].addr, sg_list[i].length, - &bytes_mapped, pf_flags); + &bytes_mapped, pf_flags, false); if (ret < 0) { mlx5r_deref_odp_mkey(&mr->mmkey); return ret; diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index 8115ab107149..88724d15705d 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -1107,8 +1107,6 @@ static int _create_kernel_qp(struct mlx5_ib_dev *dev, if (init_attr->qp_type == MLX5_IB_QPT_REG_UMR) qp->bf.bfreg = &dev->fp_bfreg; - else if (qp->flags & MLX5_IB_QP_CREATE_WC_TEST) - qp->bf.bfreg = &dev->wc_bfreg; else qp->bf.bfreg = &dev->bfreg; @@ -1962,7 +1960,7 @@ static int atomic_size_to_mode(int size_mask) } static int get_atomic_mode(struct mlx5_ib_dev *dev, - enum ib_qp_type qp_type) + struct mlx5_ib_qp *qp) { u8 atomic_operations = MLX5_CAP_ATOMIC(dev->mdev, atomic_operations); u8 atomic = MLX5_CAP_GEN(dev->mdev, atomic); @@ -1972,7 +1970,7 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev, if (!atomic) return -EOPNOTSUPP; - if (qp_type == MLX5_IB_QPT_DCT) + if (qp->type == MLX5_IB_QPT_DCT) atomic_size_mask = MLX5_CAP_ATOMIC(dev->mdev, atomic_size_dc); else atomic_size_mask = MLX5_CAP_ATOMIC(dev->mdev, atomic_size_qp); @@ -1986,6 +1984,10 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev, atomic_operations & MLX5_ATOMIC_OPS_FETCH_ADD)) atomic_mode = MLX5_ATOMIC_MODE_IB_COMP; + /* OOO DP QPs do not support larger than 8-Bytes atomic operations */ + if (atomic_mode > MLX5_ATOMIC_MODE_8B && qp->is_ooo_rq) + atomic_mode = MLX5_ATOMIC_MODE_8B; + return atomic_mode; } @@ -2841,6 +2843,29 @@ static int check_valid_flow(struct mlx5_ib_dev *dev, struct ib_pd *pd, return 0; } +static bool get_dp_ooo_cap(struct mlx5_core_dev *mdev, enum ib_qp_type qp_type) +{ + if (!MLX5_CAP_GEN_2(mdev, dp_ordering_force)) + return false; + + switch (qp_type) { + case IB_QPT_RC: + return MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_rc); + case IB_QPT_XRC_INI: + case IB_QPT_XRC_TGT: + return MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_xrc); + case IB_QPT_UC: + return MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_uc); + case IB_QPT_UD: + return MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_ud); + case MLX5_IB_QPT_DCI: + case MLX5_IB_QPT_DCT: + return MLX5_CAP_GEN(mdev, dp_ordering_ooo_all_dc); + default: + return false; + } +} + static void process_vendor_flag(struct mlx5_ib_dev *dev, int *flags, int flag, bool cond, struct mlx5_ib_qp *qp) { @@ -2959,14 +2984,6 @@ static void process_create_flag(struct mlx5_ib_dev *dev, int *flags, int flag, return; } - if (flag == MLX5_IB_QP_CREATE_WC_TEST) { - /* - * Special case, if condition didn't meet, it won't be error, - * just different in-kernel flow. - */ - *flags &= ~MLX5_IB_QP_CREATE_WC_TEST; - return; - } mlx5_ib_dbg(dev, "Verbs create QP flag 0x%X is not supported\n", flag); } @@ -3027,8 +3044,6 @@ static int process_create_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, IB_QP_CREATE_PCI_WRITE_END_PADDING, MLX5_CAP_GEN(mdev, end_pad), qp); - process_create_flag(dev, &create_flags, MLX5_IB_QP_CREATE_WC_TEST, - qp_type != MLX5_IB_QPT_REG_UMR, qp); process_create_flag(dev, &create_flags, MLX5_IB_QP_CREATE_SQPN_QP1, true, qp); @@ -3097,7 +3112,6 @@ static int create_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd, switch (qp->type) { case MLX5_IB_QPT_DCT: err = create_dct(dev, pd, qp, params); - rdma_restrack_no_track(&qp->ibqp.res); break; case MLX5_IB_QPT_DCI: err = create_dci(dev, pd, qp, params); @@ -3109,9 +3123,9 @@ static int create_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd, err = mlx5_ib_create_gsi(pd, qp, params->attr); break; case MLX5_IB_QPT_HW_GSI: - case MLX5_IB_QPT_REG_UMR: rdma_restrack_no_track(&qp->ibqp.res); fallthrough; + case MLX5_IB_QPT_REG_UMR: default: if (params->udata) err = create_user_qp(dev, pd, qp, params); @@ -3247,6 +3261,10 @@ int mlx5_ib_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attr, enum ib_qp_type type; int err; + err = mlx5_ib_dev_res_srq_init(dev); + if (err) + return err; + err = check_qp_type(dev, attr, &type); if (err) return err; @@ -3374,7 +3392,7 @@ static int set_qpc_atomic_flags(struct mlx5_ib_qp *qp, if (access_flags & IB_ACCESS_REMOTE_ATOMIC) { int atomic_mode; - atomic_mode = get_atomic_mode(dev, qp->type); + atomic_mode = get_atomic_mode(dev, qp); if (atomic_mode < 0) return -EOPNOTSUPP; @@ -3429,11 +3447,11 @@ static int ib_to_mlx5_rate_map(u8 rate) return 0; } -static int ib_rate_to_mlx5(struct mlx5_ib_dev *dev, u8 rate) +int mlx5r_ib_rate(struct mlx5_ib_dev *dev, u8 rate) { u32 stat_rate_support; - if (rate == IB_RATE_PORT_CURRENT) + if (rate == IB_RATE_PORT_CURRENT || rate == IB_RATE_800_GBPS) return 0; if (rate < IB_RATE_2_5_GBPS || rate > IB_RATE_800_GBPS) @@ -3578,7 +3596,7 @@ static int mlx5_set_path(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, sizeof(grh->dgid.raw)); } - err = ib_rate_to_mlx5(dev, rdma_ah_get_static_rate(ah)); + err = mlx5r_ib_rate(dev, rdma_ah_get_static_rate(ah)); if (err < 0) return err; MLX5_SET(ads, path, stat_rate, err); @@ -4226,7 +4244,12 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, /* todo implement counter_index functionality */ - if (is_sqp(qp->type)) + if (dev->ib_dev.type == RDMA_DEVICE_TYPE_SMI && is_qp0(qp->type)) { + MLX5_SET(ads, pri_path, vhca_port_num, + smi_to_native_portnum(dev, qp->port)); + if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) + MLX5_SET(ads, pri_path, plane_index, qp->port); + } else if (is_sqp(qp->type)) MLX5_SET(ads, pri_path, vhca_port_num, qp->port); if (attr_mask & IB_QP_PORT) @@ -4272,14 +4295,14 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, MLX5_SET(qpc, qpc, retry_count, attr->retry_cnt); if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC && attr->max_rd_atomic) - MLX5_SET(qpc, qpc, log_sra_max, ilog2(attr->max_rd_atomic)); + MLX5_SET(qpc, qpc, log_sra_max, fls(attr->max_rd_atomic - 1)); if (attr_mask & IB_QP_SQ_PSN) MLX5_SET(qpc, qpc, next_send_psn, attr->sq_psn); if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC && attr->max_dest_rd_atomic) MLX5_SET(qpc, qpc, log_rra_max, - ilog2(attr->max_dest_rd_atomic)); + fls(attr->max_dest_rd_atomic - 1)); if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC)) { err = set_qpc_atomic_flags(qp, attr, attr_mask, qpc); @@ -4320,6 +4343,11 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, if (qp->flags & MLX5_IB_QP_CREATE_SQPN_QP1) MLX5_SET(qpc, qpc, deth_sqpn, 1); + if (qp->is_ooo_rq && cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) { + MLX5_SET(qpc, qpc, dp_ordering_1, 1); + MLX5_SET(qpc, qpc, dp_ordering_force, 1); + } + mlx5_cur = to_mlx5_state(cur_state); mlx5_new = to_mlx5_state(new_state); @@ -4535,7 +4563,7 @@ static int mlx5_ib_modify_dct(struct ib_qp *ibqp, struct ib_qp_attr *attr, if (attr->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC) { int atomic_mode; - atomic_mode = get_atomic_mode(dev, MLX5_IB_QPT_DCT); + atomic_mode = get_atomic_mode(dev, qp); if (atomic_mode < 0) return -EOPNOTSUPP; @@ -4551,6 +4579,8 @@ static int mlx5_ib_modify_dct(struct ib_qp *ibqp, struct ib_qp_attr *attr, set_id = mlx5_ib_get_counters_id(dev, attr->port_num - 1); MLX5_SET(dctc, dctc, counter_set_id, set_id); + + qp->port = attr->port_num; } else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) { struct mlx5_ib_modify_qp_resp resp = {}; u32 out[MLX5_ST_SZ_DW(create_dct_out)] = {}; @@ -4577,6 +4607,10 @@ static int mlx5_ib_modify_dct(struct ib_qp *ibqp, struct ib_qp_attr *attr, MLX5_SET(dctc, dctc, hop_limit, attr->ah_attr.grh.hop_limit); if (attr->ah_attr.type == RDMA_AH_ATTR_TYPE_ROCE) MLX5_SET(dctc, dctc, eth_prio, attr->ah_attr.sl & 0x7); + if (qp->is_ooo_rq) { + MLX5_SET(dctc, dctc, dp_ordering_1, 1); + MLX5_SET(dctc, dctc, dp_ordering_force, 1); + } err = mlx5_core_create_dct(dev, &qp->dct.mdct, qp->dct.in, MLX5_ST_SZ_BYTES(create_dct_in), out, @@ -4610,10 +4644,6 @@ static bool mlx5_ib_modify_qp_allowed(struct mlx5_ib_dev *dev, if (qp->type == IB_QPT_RAW_PACKET || qp->type == MLX5_IB_QPT_REG_UMR) return true; - /* Internal QP used for wc testing, with NOPs in wq */ - if (qp->flags & MLX5_IB_QP_CREATE_WC_TEST) - return true; - return false; } @@ -4684,11 +4714,16 @@ int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, min(udata->inlen, sizeof(ucmd)))) return -EFAULT; - if (ucmd.comp_mask || + if (ucmd.comp_mask & ~MLX5_IB_MODIFY_QP_OOO_DP || memchr_inv(&ucmd.burst_info.reserved, 0, sizeof(ucmd.burst_info.reserved))) return -EOPNOTSUPP; + if (ucmd.comp_mask & MLX5_IB_MODIFY_QP_OOO_DP) { + if (!get_dp_ooo_cap(dev->mdev, qp->type)) + return -EOPNOTSUPP; + qp->is_ooo_rq = 1; + } } if (qp->type == IB_QPT_GSI) @@ -5041,7 +5076,7 @@ static int mlx5_ib_dct_query_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *mqp, } if (qp_attr_mask & IB_QP_PORT) - qp_attr->port_num = MLX5_GET(dctc, dctc, port); + qp_attr->port_num = mqp->port; if (qp_attr_mask & IB_QP_MIN_RNR_TIMER) qp_attr->min_rnr_timer = MLX5_GET(dctc, dctc, min_rnr_nak); if (qp_attr_mask & IB_QP_AV) { diff --git a/drivers/infiniband/hw/mlx5/qp.h b/drivers/infiniband/hw/mlx5/qp.h index b6ee7c3ee1ca..2530e7730635 100644 --- a/drivers/infiniband/hw/mlx5/qp.h +++ b/drivers/infiniband/hw/mlx5/qp.h @@ -56,4 +56,5 @@ int mlx5_core_xrcd_dealloc(struct mlx5_ib_dev *dev, u32 xrcdn); int mlx5_ib_qp_set_counter(struct ib_qp *qp, struct rdma_counter *counter); int mlx5_ib_qp_event_init(void); void mlx5_ib_qp_event_cleanup(void); +int mlx5r_ib_rate(struct mlx5_ib_dev *dev, u8 rate); #endif /* _MLX5_IB_QP_H */ diff --git a/drivers/infiniband/hw/mlx5/qpc.c b/drivers/infiniband/hw/mlx5/qpc.c index d9cf6982d645..146d03ae40bd 100644 --- a/drivers/infiniband/hw/mlx5/qpc.c +++ b/drivers/infiniband/hw/mlx5/qpc.c @@ -21,8 +21,10 @@ mlx5_get_rsc(struct mlx5_qp_table *table, u32 rsn) spin_lock_irqsave(&table->lock, flags); common = radix_tree_lookup(&table->tree, rsn); - if (common) + if (common && !common->invalid) refcount_inc(&common->refcount); + else + common = NULL; spin_unlock_irqrestore(&table->lock, flags); @@ -178,6 +180,18 @@ static int create_resource_common(struct mlx5_ib_dev *dev, return 0; } +static void modify_resource_common_state(struct mlx5_ib_dev *dev, + struct mlx5_core_qp *qp, + bool invalid) +{ + struct mlx5_qp_table *table = &dev->qp_table; + unsigned long flags; + + spin_lock_irqsave(&table->lock, flags); + qp->common.invalid = invalid; + spin_unlock_irqrestore(&table->lock, flags); +} + static void destroy_resource_common(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp) { @@ -249,7 +263,8 @@ int mlx5_qpc_create_qp(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp, if (err) goto err_cmd; - mlx5_debug_qp_add(dev->mdev, qp); + if (dev->ib_dev.type != RDMA_DEVICE_TYPE_SMI) + mlx5_debug_qp_add(dev->mdev, qp); return 0; @@ -307,7 +322,8 @@ int mlx5_core_destroy_qp(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp) { u32 in[MLX5_ST_SZ_DW(destroy_qp_in)] = {}; - mlx5_debug_qp_remove(dev->mdev, qp); + if (dev->ib_dev.type != RDMA_DEVICE_TYPE_SMI) + mlx5_debug_qp_remove(dev->mdev, qp); destroy_resource_common(dev, qp); @@ -504,7 +520,9 @@ int mlx5_init_qp_table(struct mlx5_ib_dev *dev) spin_lock_init(&table->lock); INIT_RADIX_TREE(&table->tree, GFP_ATOMIC); xa_init(&table->dct_xa); - mlx5_qp_debugfs_init(dev->mdev); + + if (dev->ib_dev.type != RDMA_DEVICE_TYPE_SMI) + mlx5_qp_debugfs_init(dev->mdev); table->nb.notifier_call = rsc_event_notifier; mlx5_notifier_register(dev->mdev, &table->nb); @@ -517,7 +535,8 @@ void mlx5_cleanup_qp_table(struct mlx5_ib_dev *dev) struct mlx5_qp_table *table = &dev->qp_table; mlx5_notifier_unregister(dev->mdev, &table->nb); - mlx5_qp_debugfs_cleanup(dev->mdev); + if (dev->ib_dev.type != RDMA_DEVICE_TYPE_SMI) + mlx5_qp_debugfs_cleanup(dev->mdev); } int mlx5_core_qp_query(struct mlx5_ib_dev *dev, struct mlx5_core_qp *qp, @@ -604,8 +623,20 @@ err_destroy_rq: int mlx5_core_destroy_rq_tracked(struct mlx5_ib_dev *dev, struct mlx5_core_qp *rq) { + int ret; + + /* The rq destruction can be called again in case it fails, hence we + * mark the common resource as invalid and only once FW destruction + * is completed successfully we actually destroy the resources. + */ + modify_resource_common_state(dev, rq, true); + ret = destroy_rq_tracked(dev, rq->qpn, rq->uid); + if (ret) { + modify_resource_common_state(dev, rq, false); + return ret; + } destroy_resource_common(dev, rq); - return destroy_rq_tracked(dev, rq->qpn, rq->uid); + return 0; } static void destroy_sq_tracked(struct mlx5_ib_dev *dev, u32 sqn, u16 uid) diff --git a/drivers/infiniband/hw/mlx5/restrack.c b/drivers/infiniband/hw/mlx5/restrack.c index 4ac429e72004..67841922c7b8 100644 --- a/drivers/infiniband/hw/mlx5/restrack.c +++ b/drivers/infiniband/hw/mlx5/restrack.c @@ -96,9 +96,18 @@ static int fill_stat_mr_entry(struct sk_buff *msg, struct ib_mr *ibmr) atomic64_read(&mr->odp_stats.faults))) goto err_table; if (rdma_nl_stat_hwcounter_entry( + msg, "page_faults_handled", + atomic64_read(&mr->odp_stats.faults_handled))) + goto err_table; + if (rdma_nl_stat_hwcounter_entry( msg, "page_invalidations", atomic64_read(&mr->odp_stats.invalidations))) goto err_table; + if (rdma_nl_stat_hwcounter_entry( + msg, "page_invalidations_handled", + atomic64_read(&mr->odp_stats.invalidations_handled))) + goto err_table; + if (rdma_nl_stat_hwcounter_entry(msg, "page_prefetch", atomic64_read(&mr->odp_stats.prefetch))) goto err_table; @@ -156,6 +165,34 @@ static int fill_res_cq_entry_raw(struct sk_buff *msg, struct ib_cq *ibcq) return fill_res_raw(msg, dev, MLX5_SGMT_TYPE_PRM_QUERY_CQ, cq->mcq.cqn); } +static int fill_res_qp_entry(struct sk_buff *msg, struct ib_qp *ibqp) +{ + struct mlx5_ib_qp *qp = to_mqp(ibqp); + int ret; + + if (qp->type < IB_QPT_DRIVER) + return 0; + + switch (qp->type) { + case MLX5_IB_QPT_REG_UMR: + ret = nla_put_string(msg, RDMA_NLDEV_ATTR_RES_SUBTYPE, + "REG_UMR"); + break; + case MLX5_IB_QPT_DCT: + ret = nla_put_string(msg, RDMA_NLDEV_ATTR_RES_SUBTYPE, "DCT"); + break; + case MLX5_IB_QPT_DCI: + ret = nla_put_string(msg, RDMA_NLDEV_ATTR_RES_SUBTYPE, "DCI"); + break; + default: + return 0; + } + if (ret) + return ret; + + return nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, IB_QPT_DRIVER); +} + static int fill_res_qp_entry_raw(struct sk_buff *msg, struct ib_qp *ibqp) { struct mlx5_ib_dev *dev = to_mdev(ibqp->device); @@ -168,6 +205,7 @@ static const struct ib_device_ops restrack_ops = { .fill_res_cq_entry_raw = fill_res_cq_entry_raw, .fill_res_mr_entry = fill_res_mr_entry, .fill_res_mr_entry_raw = fill_res_mr_entry_raw, + .fill_res_qp_entry = fill_res_qp_entry, .fill_res_qp_entry_raw = fill_res_qp_entry_raw, .fill_stat_mr_entry = fill_stat_mr_entry, }; diff --git a/drivers/infiniband/hw/mlx5/srq.c b/drivers/infiniband/hw/mlx5/srq.c index a056ea835da5..bcb6b324af50 100644 --- a/drivers/infiniband/hw/mlx5/srq.c +++ b/drivers/infiniband/hw/mlx5/srq.c @@ -199,20 +199,27 @@ int mlx5_ib_create_srq(struct ib_srq *ib_srq, int err; struct mlx5_srq_attr in = {}; __u32 max_srq_wqes = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz); + __u32 max_sge_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_rq) / + sizeof(struct mlx5_wqe_data_seg); if (init_attr->srq_type != IB_SRQT_BASIC && init_attr->srq_type != IB_SRQT_XRC && init_attr->srq_type != IB_SRQT_TM) return -EOPNOTSUPP; - /* Sanity check SRQ size before proceeding */ - if (init_attr->attr.max_wr >= max_srq_wqes) { - mlx5_ib_dbg(dev, "max_wr %d, cap %d\n", - init_attr->attr.max_wr, - max_srq_wqes); + /* Sanity check SRQ and sge size before proceeding */ + if (init_attr->attr.max_wr >= max_srq_wqes || + init_attr->attr.max_sge > max_sge_sz) { + mlx5_ib_dbg(dev, "max_wr %d,wr_cap %d,max_sge %d, sge_cap:%d\n", + init_attr->attr.max_wr, max_srq_wqes, + init_attr->attr.max_sge, max_sge_sz); return -EINVAL; } + err = mlx5_ib_dev_res_cq_init(dev); + if (err) + return err; + mutex_init(&srq->mutex); spin_lock_init(&srq->lock); srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1); diff --git a/drivers/infiniband/hw/mlx5/std_types.c b/drivers/infiniband/hw/mlx5/std_types.c index bbfcce3bdc84..bdb568411091 100644 --- a/drivers/infiniband/hw/mlx5/std_types.c +++ b/drivers/infiniband/hw/mlx5/std_types.c @@ -10,6 +10,7 @@ #include <linux/mlx5/eswitch.h> #include <linux/mlx5/vport.h> #include "mlx5_ib.h" +#include "data_direct.h" #define UVERBS_MODULE_NAME mlx5_ib #include <rdma/uverbs_named_ioctl.h> @@ -111,6 +112,23 @@ out: return err; } +static int fill_multiport_info(struct mlx5_ib_dev *dev, u32 port_num, + struct mlx5_ib_uapi_query_port *info) +{ + struct mlx5_core_dev *mdev; + + mdev = mlx5_ib_get_native_port_mdev(dev, port_num, NULL); + if (!mdev) + return -EINVAL; + + info->vport_vhca_id = MLX5_CAP_GEN(mdev, vhca_id); + info->flags |= MLX5_IB_UAPI_QUERY_PORT_VPORT_VHCA_ID; + + mlx5_ib_put_native_port_mdev(dev, port_num); + + return 0; +} + static int fill_switchdev_info(struct mlx5_ib_dev *dev, u32 port_num, struct mlx5_ib_uapi_query_port *info) { @@ -177,12 +195,60 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_QUERY_PORT)( ret = fill_switchdev_info(dev, port_num, &info); if (ret) return ret; + } else if (mlx5_core_mp_enabled(dev->mdev)) { + ret = fill_multiport_info(dev, port_num, &info); + if (ret) + return ret; } return uverbs_copy_to_struct_or_zero(attrs, MLX5_IB_ATTR_QUERY_PORT, &info, sizeof(info)); } +static int UVERBS_HANDLER(MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH)( + struct uverbs_attr_bundle *attrs) +{ + struct mlx5_data_direct_dev *data_direct_dev; + struct mlx5_ib_ucontext *c; + struct mlx5_ib_dev *dev; + int out_len = uverbs_attr_get_len(attrs, + MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH); + u32 dev_path_len; + char *dev_path; + int ret; + + c = to_mucontext(ib_uverbs_get_ucontext(attrs)); + if (IS_ERR(c)) + return PTR_ERR(c); + dev = to_mdev(c->ibucontext.device); + mutex_lock(&dev->data_direct_lock); + data_direct_dev = dev->data_direct_dev; + if (!data_direct_dev) { + ret = -ENODEV; + goto end; + } + + dev_path = kobject_get_path(&data_direct_dev->device->kobj, GFP_KERNEL); + if (!dev_path) { + ret = -ENOMEM; + goto end; + } + + dev_path_len = strlen(dev_path) + 1; + if (dev_path_len > out_len) { + ret = -ENOSPC; + goto end; + } + + ret = uverbs_copy_to(attrs, MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH, dev_path, + dev_path_len); + kfree(dev_path); + +end: + mutex_unlock(&dev->data_direct_lock); + return ret; +} + DECLARE_UVERBS_NAMED_METHOD( MLX5_IB_METHOD_QUERY_PORT, UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_QUERY_PORT_PORT_NUM, @@ -193,9 +259,17 @@ DECLARE_UVERBS_NAMED_METHOD( reg_c0), UA_MANDATORY)); +DECLARE_UVERBS_NAMED_METHOD( + MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH, + UVERBS_ATTR_PTR_OUT( + MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH, + UVERBS_ATTR_MIN_SIZE(0), + UA_MANDATORY)); + ADD_UVERBS_METHODS(mlx5_ib_device, UVERBS_OBJECT_DEVICE, - &UVERBS_METHOD(MLX5_IB_METHOD_QUERY_PORT)); + &UVERBS_METHOD(MLX5_IB_METHOD_QUERY_PORT), + &UVERBS_METHOD(MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH)); DECLARE_UVERBS_NAMED_METHOD( MLX5_IB_METHOD_PD_QUERY, diff --git a/drivers/infiniband/hw/mlx5/umr.c b/drivers/infiniband/hw/mlx5/umr.c index e76142f6fa88..5be4426a2884 100644 --- a/drivers/infiniband/hw/mlx5/umr.c +++ b/drivers/infiniband/hw/mlx5/umr.c @@ -135,22 +135,28 @@ static int mlx5r_umr_qp_rst2rts(struct mlx5_ib_dev *dev, struct ib_qp *qp) int mlx5r_umr_resource_init(struct mlx5_ib_dev *dev) { struct ib_qp_init_attr init_attr = {}; - struct ib_pd *pd; struct ib_cq *cq; struct ib_qp *qp; - int ret; + int ret = 0; - pd = ib_alloc_pd(&dev->ib_dev, 0); - if (IS_ERR(pd)) { - mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n"); - return PTR_ERR(pd); - } + + /* + * UMR qp is set once, never changed until device unload. + * Avoid taking the mutex if initialization is already done. + */ + if (dev->umrc.qp) + return 0; + + mutex_lock(&dev->umrc.init_lock); + /* First user allocates the UMR resources. Skip if already allocated. */ + if (dev->umrc.qp) + goto unlock; cq = ib_alloc_cq(&dev->ib_dev, NULL, 128, 0, IB_POLL_SOFTIRQ); if (IS_ERR(cq)) { mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n"); ret = PTR_ERR(cq); - goto destroy_pd; + goto unlock; } init_attr.send_cq = cq; @@ -160,7 +166,7 @@ int mlx5r_umr_resource_init(struct mlx5_ib_dev *dev) init_attr.cap.max_send_sge = 1; init_attr.qp_type = MLX5_IB_QPT_REG_UMR; init_attr.port_num = 1; - qp = ib_create_qp(pd, &init_attr); + qp = ib_create_qp(dev->umrc.pd, &init_attr); if (IS_ERR(qp)) { mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n"); ret = PTR_ERR(qp); @@ -171,22 +177,22 @@ int mlx5r_umr_resource_init(struct mlx5_ib_dev *dev) if (ret) goto destroy_qp; - dev->umrc.qp = qp; dev->umrc.cq = cq; - dev->umrc.pd = pd; sema_init(&dev->umrc.sem, MAX_UMR_WR); mutex_init(&dev->umrc.lock); dev->umrc.state = MLX5_UMR_STATE_ACTIVE; + dev->umrc.qp = qp; + mutex_unlock(&dev->umrc.init_lock); return 0; destroy_qp: ib_destroy_qp(qp); destroy_cq: ib_free_cq(cq); -destroy_pd: - ib_dealloc_pd(pd); +unlock: + mutex_unlock(&dev->umrc.init_lock); return ret; } @@ -194,36 +200,38 @@ void mlx5r_umr_resource_cleanup(struct mlx5_ib_dev *dev) { if (dev->umrc.state == MLX5_UMR_STATE_UNINIT) return; + mutex_destroy(&dev->umrc.lock); + /* After device init, UMR cp/qp are not unset during the lifetime. */ ib_destroy_qp(dev->umrc.qp); ib_free_cq(dev->umrc.cq); - ib_dealloc_pd(dev->umrc.pd); } -static int mlx5r_umr_recover(struct mlx5_ib_dev *dev) +int mlx5r_umr_init(struct mlx5_ib_dev *dev) { - struct umr_common *umrc = &dev->umrc; - struct ib_qp_attr attr; - int err; + struct ib_pd *pd; - attr.qp_state = IB_QPS_RESET; - err = ib_modify_qp(umrc->qp, &attr, IB_QP_STATE); - if (err) { - mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n"); - goto err; + pd = ib_alloc_pd(&dev->ib_dev, 0); + if (IS_ERR(pd)) { + mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n"); + return PTR_ERR(pd); } + dev->umrc.pd = pd; - err = mlx5r_umr_qp_rst2rts(dev, umrc->qp); - if (err) - goto err; + mutex_init(&dev->umrc.init_lock); - umrc->state = MLX5_UMR_STATE_ACTIVE; return 0; +} -err: - umrc->state = MLX5_UMR_STATE_ERR; - return err; +void mlx5r_umr_cleanup(struct mlx5_ib_dev *dev) +{ + if (!dev->umrc.pd) + return; + + mutex_destroy(&dev->umrc.init_lock); + ib_dealloc_pd(dev->umrc.pd); } + static int mlx5r_umr_post_send(struct ib_qp *ibqp, u32 mkey, struct ib_cqe *cqe, struct mlx5r_umr_wqe *wqe, bool with_data) { @@ -270,6 +278,61 @@ out: return err; } +static int mlx5r_umr_recover(struct mlx5_ib_dev *dev, u32 mkey, + struct mlx5r_umr_context *umr_context, + struct mlx5r_umr_wqe *wqe, bool with_data) +{ + struct umr_common *umrc = &dev->umrc; + struct ib_qp_attr attr; + int err; + + mutex_lock(&umrc->lock); + /* Preventing any further WRs to be sent now */ + if (umrc->state != MLX5_UMR_STATE_RECOVER) { + mlx5_ib_warn(dev, "UMR recovery encountered an unexpected state=%d\n", + umrc->state); + umrc->state = MLX5_UMR_STATE_RECOVER; + } + mutex_unlock(&umrc->lock); + + /* Sending a final/barrier WR (the failed one) and wait for its completion. + * This will ensure that all the previous WRs got a completion before + * we set the QP state to RESET. + */ + err = mlx5r_umr_post_send(umrc->qp, mkey, &umr_context->cqe, wqe, + with_data); + if (err) { + mlx5_ib_warn(dev, "UMR recovery post send failed, err %d\n", err); + goto err; + } + + /* Since the QP is in an error state, it will only receive + * IB_WC_WR_FLUSH_ERR. However, as it serves only as a barrier + * we don't care about its status. + */ + wait_for_completion(&umr_context->done); + + attr.qp_state = IB_QPS_RESET; + err = ib_modify_qp(umrc->qp, &attr, IB_QP_STATE); + if (err) { + mlx5_ib_warn(dev, "Couldn't modify UMR QP to RESET, err=%d\n", err); + goto err; + } + + err = mlx5r_umr_qp_rst2rts(dev, umrc->qp); + if (err) { + mlx5_ib_warn(dev, "Couldn't modify UMR QP to RTS, err=%d\n", err); + goto err; + } + + umrc->state = MLX5_UMR_STATE_ACTIVE; + return 0; + +err: + umrc->state = MLX5_UMR_STATE_ERR; + return err; +} + static void mlx5r_umr_done(struct ib_cq *cq, struct ib_wc *wc) { struct mlx5_ib_umr_context *context = @@ -334,9 +397,7 @@ static int mlx5r_umr_post_send_wait(struct mlx5_ib_dev *dev, u32 mkey, mlx5_ib_warn(dev, "reg umr failed (%u). Trying to recover and resubmit the flushed WQEs, mkey = %u\n", umr_context.status, mkey); - mutex_lock(&umrc->lock); - err = mlx5r_umr_recover(dev); - mutex_unlock(&umrc->lock); + err = mlx5r_umr_recover(dev, mkey, &umr_context, wqe, with_data); if (err) mlx5_ib_warn(dev, "couldn't recover UMR, err %d\n", err); @@ -603,44 +664,47 @@ static void mlx5r_umr_final_update_xlt(struct mlx5_ib_dev *dev, wqe->data_seg.byte_count = cpu_to_be32(sg->length); } -/* - * Send the DMA list to the HW for a normal MR using UMR. - * Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP - * flag may be used. - */ -int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +static int +_mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags, bool dd) { + size_t ent_size = dd ? sizeof(struct mlx5_ksm) : sizeof(struct mlx5_mtt); struct mlx5_ib_dev *dev = mr_to_mdev(mr); struct device *ddev = &dev->mdev->pdev->dev; struct mlx5r_umr_wqe wqe = {}; struct ib_block_iter biter; + struct mlx5_ksm *cur_ksm; struct mlx5_mtt *cur_mtt; size_t orig_sg_length; - struct mlx5_mtt *mtt; size_t final_size; + void *curr_entry; struct ib_sge sg; + void *entry; u64 offset = 0; int err = 0; - if (WARN_ON(mr->umem->is_odp)) - return -EINVAL; - - mtt = mlx5r_umr_create_xlt( - dev, &sg, ib_umem_num_dma_blocks(mr->umem, 1 << mr->page_shift), - sizeof(*mtt), flags); - if (!mtt) + entry = mlx5r_umr_create_xlt(dev, &sg, + ib_umem_num_dma_blocks(mr->umem, 1 << mr->page_shift), + ent_size, flags); + if (!entry) return -ENOMEM; orig_sg_length = sg.length; - mlx5r_umr_set_update_xlt_ctrl_seg(&wqe.ctrl_seg, flags, &sg); mlx5r_umr_set_update_xlt_mkey_seg(dev, &wqe.mkey_seg, mr, mr->page_shift); + if (dd) { + /* Use the data direct internal kernel PD */ + MLX5_SET(mkc, &wqe.mkey_seg, pd, dev->ddr.pdn); + cur_ksm = entry; + } else { + cur_mtt = entry; + } + mlx5r_umr_set_update_xlt_data_seg(&wqe.data_seg, &sg); - cur_mtt = mtt; + curr_entry = entry; rdma_umem_for_each_dma_block(mr->umem, &biter, BIT(mr->page_shift)) { - if (cur_mtt == (void *)mtt + sg.length) { + if (curr_entry == entry + sg.length) { dma_sync_single_for_device(ddev, sg.addr, sg.length, DMA_TO_DEVICE); @@ -652,23 +716,31 @@ int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) DMA_TO_DEVICE); offset += sg.length; mlx5r_umr_update_offset(&wqe.ctrl_seg, offset); - - cur_mtt = mtt; + if (dd) + cur_ksm = entry; + else + cur_mtt = entry; } - cur_mtt->ptag = - cpu_to_be64(rdma_block_iter_dma_address(&biter) | - MLX5_IB_MTT_PRESENT); - - if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP)) - cur_mtt->ptag = 0; - - cur_mtt++; + if (dd) { + cur_ksm->va = cpu_to_be64(rdma_block_iter_dma_address(&biter)); + cur_ksm->key = cpu_to_be32(dev->ddr.mkey); + cur_ksm++; + curr_entry = cur_ksm; + } else { + cur_mtt->ptag = + cpu_to_be64(rdma_block_iter_dma_address(&biter) | + MLX5_IB_MTT_PRESENT); + if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP)) + cur_mtt->ptag = 0; + cur_mtt++; + curr_entry = cur_mtt; + } } - final_size = (void *)cur_mtt - (void *)mtt; + final_size = curr_entry - entry; sg.length = ALIGN(final_size, MLX5_UMR_FLEX_ALIGNMENT); - memset(cur_mtt, 0, sg.length - final_size); + memset(curr_entry, 0, sg.length - final_size); mlx5r_umr_final_update_xlt(dev, &wqe, mr, &sg, flags); dma_sync_single_for_device(ddev, sg.addr, sg.length, DMA_TO_DEVICE); @@ -676,10 +748,32 @@ int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) err: sg.length = orig_sg_length; - mlx5r_umr_unmap_free_xlt(dev, mtt, &sg); + mlx5r_umr_unmap_free_xlt(dev, entry, &sg); return err; } +int mlx5r_umr_update_data_direct_ksm_pas(struct mlx5_ib_mr *mr, unsigned int flags) +{ + /* No invalidation flow is expected */ + if (WARN_ON(!mr->umem->is_dmabuf) || (flags & MLX5_IB_UPD_XLT_ZAP)) + return -EINVAL; + + return _mlx5r_umr_update_mr_pas(mr, flags, true); +} + +/* + * Send the DMA list to the HW for a normal MR using UMR. + * Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP + * flag may be used. + */ +int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +{ + if (WARN_ON(mr->umem->is_odp)) + return -EINVAL; + + return _mlx5r_umr_update_mr_pas(mr, flags, false); +} + static bool umr_can_use_indirect_mkey(struct mlx5_ib_dev *dev) { return !MLX5_CAP_GEN(dev->mdev, umr_indirect_mkey_disabled); @@ -746,7 +840,17 @@ int mlx5r_umr_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, size_to_map = npages * desc_size; dma_sync_single_for_cpu(ddev, sg.addr, sg.length, DMA_TO_DEVICE); - mlx5_odp_populate_xlt(xlt, idx, npages, mr, flags); + /* + * npages is the maximum number of pages to map, but we + * can't guarantee that all pages are actually mapped. + * + * For example, if page is p2p of type which is not supported + * for mapping, the number of pages mapped will be less than + * requested. + */ + err = mlx5_odp_populate_xlt(xlt, idx, npages, mr, flags); + if (err) + return err; dma_sync_single_for_device(ddev, sg.addr, sg.length, DMA_TO_DEVICE); sg.length = ALIGN(size_to_map, MLX5_UMR_FLEX_ALIGNMENT); diff --git a/drivers/infiniband/hw/mlx5/umr.h b/drivers/infiniband/hw/mlx5/umr.h index 3799bb758e49..4a02c9b5aad8 100644 --- a/drivers/infiniband/hw/mlx5/umr.h +++ b/drivers/infiniband/hw/mlx5/umr.h @@ -16,6 +16,9 @@ int mlx5r_umr_resource_init(struct mlx5_ib_dev *dev); void mlx5r_umr_resource_cleanup(struct mlx5_ib_dev *dev); +int mlx5r_umr_init(struct mlx5_ib_dev *dev); +void mlx5r_umr_cleanup(struct mlx5_ib_dev *dev); + static inline bool mlx5r_umr_can_load_pas(struct mlx5_ib_dev *dev, size_t length) { @@ -92,6 +95,7 @@ int mlx5r_umr_revoke_mr(struct mlx5_ib_mr *mr); int mlx5r_umr_rereg_pd_access(struct mlx5_ib_mr *mr, struct ib_pd *pd, int access_flags); int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags); +int mlx5r_umr_update_data_direct_ksm_pas(struct mlx5_ib_mr *mr, unsigned int flags); int mlx5r_umr_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, int page_shift, int flags); diff --git a/drivers/infiniband/hw/mlx5/wr.c b/drivers/infiniband/hw/mlx5/wr.c index df1d1b0a3ef7..9947feb7fb8a 100644 --- a/drivers/infiniband/hw/mlx5/wr.c +++ b/drivers/infiniband/hw/mlx5/wr.c @@ -78,7 +78,7 @@ static void set_eth_seg(const struct ib_send_wr *wr, struct mlx5_ib_qp *qp, */ copysz = min_t(u64, *cur_edge - (void *)eseg->inline_hdr.start, left); - memcpy(eseg->inline_hdr.start, pdata, copysz); + memcpy(eseg->inline_hdr.data, pdata, copysz); stride = ALIGN(sizeof(struct mlx5_wqe_eth_seg) - sizeof(eseg->inline_hdr.start) + copysz, 16); *size += stride / 16; |