From: Juan Zhou zhoujuan51@h-partners.com
Yixing Liu (2): Update kernel headers libhns: Support congestion control algorithm configuration
kernel-headers/rdma/hns-abi.h | 18 ++++- providers/hns/hns_roce_u.c | 1 + providers/hns/hns_roce_u.h | 6 ++ providers/hns/hns_roce_u_verbs.c | 114 +++++++++++++++++++++++++++++-- providers/hns/hnsdv.h | 24 ++++++- providers/hns/libhns.map | 1 + 6 files changed, 156 insertions(+), 8 deletions(-)
From: Yixing Liu liuyixing1@huawei.com
To commit ?? ("RDMA/hns: Support congestion control algorithm configuration at QP granularity").
Signed-off-by: Yixing Liu liuyixing1@huawei.com Reviewed-by: Yangyang Li liyangyang20@huawei.com --- kernel-headers/rdma/hns-abi.h | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-)
diff --git a/kernel-headers/rdma/hns-abi.h b/kernel-headers/rdma/hns-abi.h index 5988a62..bd19927 100644 --- a/kernel-headers/rdma/hns-abi.h +++ b/kernel-headers/rdma/hns-abi.h @@ -63,6 +63,18 @@ struct hns_roce_ib_create_srq_resp { __u32 reserved; };
+enum hns_roce_create_qp_comp_mask { + HNS_ROCE_CREATE_QP_MASK_CREATE_FLAGS = 1 << 0, + HNS_ROCE_CREATE_QP_MASK_CONGEST_TYPE = 1 << 1, +}; + +enum hns_roce_congest_type_flags { + HNS_ROCE_CREATE_QP_FLAGS_DCQCN = 1 << 0, + HNS_ROCE_CREATE_QP_FLAGS_LDCP = 1 << 1, + HNS_ROCE_CREATE_QP_FLAGS_HC3 = 1 << 2, + HNS_ROCE_CREATE_QP_FLAGS_DIP = 1 << 3, +}; + struct hns_roce_ib_create_qp { __aligned_u64 buf_addr; __aligned_u64 db_addr; @@ -71,6 +83,9 @@ struct hns_roce_ib_create_qp { __u8 sq_no_prefetch; __u8 reserved[5]; __aligned_u64 sdb_addr; + __aligned_u64 comp_mask; + __aligned_u64 create_flags; + __aligned_u64 congest_type_flags; };
enum hns_roce_qp_cap_flags { @@ -123,7 +138,8 @@ struct hns_roce_ib_alloc_ucontext_resp { __u32 config; __u32 max_inline_data; __u8 mac_type; - __u8 rsv1[7]; + __u8 congest_type; + __u8 rsv1[6]; __u32 dca_qps; __u32 dca_mmap_size; __aligned_u64 dca_mmap_key;
From: Yixing Liu liuyixing1@huawei.com
driver inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I6N1G4
---------------------------------------------------------------
Added the use of direct verbs to implement QP-level user-configurable congestion control algorithms. Among them, the user mode driver mainly provides interfaces for users to choose, and the kernel mode driver is responsible for filling the resources of different algorithms and providing the supported algorithm types for user mode.
At the same time, provide a direct verbs interface for users to query the type of congestion control algorithm.
Signed-off-by: Yixing Liu liuyixing1@huawei.com Reviewed-by: Yangyang Li liyangyang20@huawei.com --- providers/hns/hns_roce_u.c | 1 + providers/hns/hns_roce_u.h | 6 ++ providers/hns/hns_roce_u_verbs.c | 107 +++++++++++++++++++++++++++++-- providers/hns/hnsdv.h | 22 +++++++ providers/hns/libhns.map | 1 + 5 files changed, 131 insertions(+), 6 deletions(-)
diff --git a/providers/hns/hns_roce_u.c b/providers/hns/hns_roce_u.c index 3d29838..87f9ed8 100644 --- a/providers/hns/hns_roce_u.c +++ b/providers/hns/hns_roce_u.c @@ -299,6 +299,7 @@ static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev, goto err_free;
hr_dev->mac_type = resp.mac_type; + hr_dev->congest_type = resp.congest_type;
if (!resp.cqe_size) context->cqe_size = HNS_ROCE_CQE_SIZE; diff --git a/providers/hns/hns_roce_u.h b/providers/hns/hns_roce_u.h index 76c7adb..41e9599 100644 --- a/providers/hns/hns_roce_u.h +++ b/providers/hns/hns_roce_u.h @@ -172,6 +172,7 @@ struct hns_roce_device { const struct hns_roce_u_hw *u_hw; int hw_version; uint8_t mac_type; + uint8_t congest_type; };
struct hns_roce_buf { @@ -230,6 +231,11 @@ struct hns_roce_v2_reset_state { uint32_t is_reset; };
+struct hns_roce_cmd_flag { + uint32_t create_flags; + uint32_t congest_type_flags; +}; + struct hns_roce_context { struct verbs_context ibv_ctx; void *uar; diff --git a/providers/hns/hns_roce_u_verbs.c b/providers/hns/hns_roce_u_verbs.c index 282ab74..6c6120c 100644 --- a/providers/hns/hns_roce_u_verbs.c +++ b/providers/hns/hns_roce_u_verbs.c @@ -89,10 +89,10 @@ int hns_roce_u_query_device(struct ibv_context *context, struct ibv_device_attr_ex *attr, size_t attr_size) { struct ib_uverbs_ex_query_device_resp resp; + unsigned int major, minor, sub_minor; size_t resp_size = sizeof(resp); - int ret; uint64_t raw_fw_ver; - unsigned int major, minor, sub_minor; + int ret;
ret = ibv_cmd_query_device_any(context, input, attr, attr_size, &resp, &resp_size); @@ -110,6 +110,27 @@ int hns_roce_u_query_device(struct ibv_context *context, return 0; }
+int hnsdv_query_device(struct ibv_context *context, + struct hnsdv_context *attrs_out) +{ + struct hns_roce_device *hr_dev = to_hr_dev(context->device); + + if (!hr_dev) { + verbs_err(verbs_get_ctx(context), "not a HNS RoCE device!\n"); + return EOPNOTSUPP; + } + + if (!attrs_out) + return EINVAL; + + memset(attrs_out, 0, sizeof(*attrs_out)); + + attrs_out->comp_mask |= HNSDV_CONTEXT_MASK_CONGEST_TYPE; + attrs_out->congest_type = hr_dev->congest_type; + + return 0; +} + int hns_roce_u_query_port(struct ibv_context *context, uint8_t port, struct ibv_port_attr *attr) { @@ -956,6 +977,67 @@ int hns_roce_u_destroy_srq(struct ibv_srq *ibv_srq) return 0; }
+enum { + HNSDV_QP_SUP_COMP_MASK = HNSDV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS | + HNSDV_QP_INIT_ATTR_MASK_QP_CONGEST_TYPE, +}; + +static int check_qp_congest_type(struct hns_roce_context *ctx, + struct hnsdv_qp_init_attr *hns_attr, + struct hns_roce_cmd_flag *cmd_flag) +{ + struct hns_roce_device *hr_dev = to_hr_dev(ctx->ibv_ctx.context.device); + + if (!check_comp_mask(hns_attr->congest_type, hr_dev->congest_type)) { + verbs_err(&ctx->ibv_ctx, "unsupported congest type 0x%x.\n", + hns_attr->congest_type); + return -EOPNOTSUPP; + } + + switch (hns_attr->congest_type) { + case HNSDV_QP_CREATE_ENABLE_DCQCN: + cmd_flag->congest_type_flags |= HNS_ROCE_CREATE_QP_FLAGS_DCQCN; + break; + case HNSDV_QP_CREATE_ENABLE_LDCP: + cmd_flag->congest_type_flags |= HNS_ROCE_CREATE_QP_FLAGS_LDCP; + break; + case HNSDV_QP_CREATE_ENABLE_HC3: + cmd_flag->congest_type_flags |= HNS_ROCE_CREATE_QP_FLAGS_LDCP; + break; + case HNSDV_QP_CREATE_ENABLE_DIP: + cmd_flag->congest_type_flags |= HNS_ROCE_CREATE_QP_FLAGS_DIP; + break; + default: + verbs_err(&ctx->ibv_ctx, + "unsupported congestion control algorithm configuration.\n"); + return -EOPNOTSUPP; + } + + return 0; +} + +static int check_hnsdv_qp_attr(struct hns_roce_context *ctx, + struct hnsdv_qp_init_attr *hns_attr, + struct hns_roce_cmd_flag *cmd_flag) +{ + int ret; + + if (!hns_attr) + return 0; + + if (!check_comp_mask(hns_attr->comp_mask, HNSDV_QP_SUP_COMP_MASK)) { + verbs_err(&ctx->ibv_ctx, "invalid hnsdv comp_mask 0x%x.\n", + hns_attr->comp_mask); + return -EINVAL; + } + + ret = check_qp_congest_type(ctx, hns_attr, cmd_flag); + if (ret) + return ret; + + return 0; +} + enum { CREATE_QP_SUP_COMP_MASK = IBV_QP_INIT_ATTR_PD | IBV_QP_INIT_ATTR_XRCD | IBV_QP_INIT_ATTR_SEND_OPS_FLAGS, @@ -1050,7 +1132,9 @@ static int verify_qp_create_cap(struct hns_roce_context *ctx, }
static int verify_qp_create_attr(struct hns_roce_context *ctx, - struct ibv_qp_init_attr_ex *attr) + struct ibv_qp_init_attr_ex *attr, + struct hnsdv_qp_init_attr *hns_attr, + struct hns_roce_cmd_flag *cmd_flag) { int ret;
@@ -1058,6 +1142,10 @@ static int verify_qp_create_attr(struct hns_roce_context *ctx, if (ret) return ret;
+ ret = check_hnsdv_qp_attr(ctx, hns_attr, cmd_flag); + if (ret) + return ret; + return verify_qp_create_cap(ctx, attr); }
@@ -1452,7 +1540,8 @@ static int hns_roce_store_qp(struct hns_roce_context *ctx, static int qp_exec_create_cmd(struct ibv_qp_init_attr_ex *attr, struct hns_roce_qp *qp, struct hns_roce_context *ctx, - uint64_t *dwqe_mmap_key) + uint64_t *dwqe_mmap_key, + struct hns_roce_cmd_flag *cmd_flag) { struct hns_roce_create_qp_ex_resp resp_ex = {}; struct hns_roce_create_qp_ex cmd_ex = {}; @@ -1464,6 +1553,11 @@ static int qp_exec_create_cmd(struct ibv_qp_init_attr_ex *attr, cmd_ex.log_sq_stride = qp->sq.wqe_shift; cmd_ex.log_sq_bb_count = hr_ilog32(qp->sq.wqe_cnt);
+ if (cmd_flag->congest_type_flags) { + cmd_ex.comp_mask |= HNS_ROCE_CREATE_QP_MASK_CONGEST_TYPE; + cmd_ex.congest_type_flags = cmd_flag->congest_type_flags; + } + ret = ibv_cmd_create_qp_ex2(&ctx->ibv_ctx.context, &qp->verbs_qp, attr, &cmd_ex.ibv_cmd, sizeof(cmd_ex), &resp_ex.ibv_resp, sizeof(resp_ex)); @@ -1543,11 +1637,12 @@ static struct ibv_qp *create_qp(struct ibv_context *ibv_ctx, struct hnsdv_qp_init_attr *hns_attr) { struct hns_roce_context *context = to_hr_ctx(ibv_ctx); + struct hns_roce_cmd_flag cmd_flag = {}; struct hns_roce_qp *qp; uint64_t dwqe_mmap_key; int ret;
- ret = verify_qp_create_attr(context, attr); + ret = verify_qp_create_attr(context, attr, hns_attr, &cmd_flag); if (ret) goto err;
@@ -1567,7 +1662,7 @@ static struct ibv_qp *create_qp(struct ibv_context *ibv_ctx, if (ret) goto err_buf;
- ret = qp_exec_create_cmd(attr, qp, context, &dwqe_mmap_key); + ret = qp_exec_create_cmd(attr, qp, context, &dwqe_mmap_key, &cmd_flag); if (ret) goto err_cmd;
diff --git a/providers/hns/hnsdv.h b/providers/hns/hnsdv.h index cfe1611..e15b428 100644 --- a/providers/hns/hnsdv.h +++ b/providers/hns/hnsdv.h @@ -45,19 +45,41 @@ enum hnsdv_qp_create_flags { HNSDV_QP_CREATE_ENABLE_DCA_MODE = 1 << 0, };
+enum hnsdv_qp_congest_ctrl_type { + HNSDV_QP_CREATE_ENABLE_DCQCN = 1 << 0, + HNSDV_QP_CREATE_ENABLE_LDCP = 1 << 1, + HNSDV_QP_CREATE_ENABLE_HC3 = 1 << 2, + HNSDV_QP_CREATE_ENABLE_DIP = 1 << 3, +}; + enum hnsdv_qp_init_attr_mask { HNSDV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS = 1 << 0, + HNSDV_QP_INIT_ATTR_MASK_QP_CONGEST_TYPE = 1 << 1, };
struct hnsdv_qp_init_attr { uint64_t comp_mask; /* Use enum hnsdv_qp_init_attr_mask */ uint32_t create_flags; /* Use enum hnsdv_qp_create_flags */ + uint8_t congest_type; /* Use enum hnsdv_qp_congest_ctrl_type */ };
struct ibv_qp *hnsdv_create_qp(struct ibv_context *context, struct ibv_qp_init_attr_ex *qp_attr, struct hnsdv_qp_init_attr *hns_qp_attr);
+enum hnsdv_query_context_comp_mask { + HNSDV_CONTEXT_MASK_CONGEST_TYPE = 1 << 0, +}; + +struct hnsdv_context { + uint64_t comp_mask; /* use enum hnsdv_query_context_comp_mask */ + uint64_t flags; + uint8_t congest_type; /* Use enum hnsdv_qp_congest_ctrl_type */ +}; + +int hnsdv_query_device(struct ibv_context *ctx_in, + struct hnsdv_context *attrs_out); + #ifdef __cplusplus } #endif diff --git a/providers/hns/libhns.map b/providers/hns/libhns.map index aed491c..ebf28eb 100644 --- a/providers/hns/libhns.map +++ b/providers/hns/libhns.map @@ -5,5 +5,6 @@ HNS_1.0 { hnsdv_is_supported; hnsdv_open_device; hnsdv_create_qp; + hnsdv_query_device; local: *; };
-
Chengchang Tang