common/mlx5: support DevX indirect mkey creation

Add option to create an indirect mkey by the current
mlx5_devx_cmd_mkey_create command.

Indirect mkey points to set of direct mkeys.
By this way, the HW\SW can reference fragmented memory by one object.

Align the net/mlx5 driver usage in the above command.

Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>

diff --git a/drivers/common/mlx5/mlx5_devx_cmds.c b/drivers/common/mlx5/mlx5_devx_cmds.c
index 3a10ff0..2197705 100644 (file)
--- a/drivers/common/mlx5/mlx5_devx_cmds.c
+++ b/drivers/common/mlx5/mlx5_devx_cmds.c
@@ -142,7 +142,11 @@ struct mlx5_devx_obj *
 mlx5_devx_cmd_mkey_create(struct ibv_context *ctx,
                          struct mlx5_devx_mkey_attr *attr)
 {
-       uint32_t in[MLX5_ST_SZ_DW(create_mkey_in)] = {0};
+       struct mlx5_klm *klm_array = attr->klm_array;
+       int klm_num = attr->klm_num;
+       int in_size_dw = MLX5_ST_SZ_DW(create_mkey_in) +
+                    (klm_num ? RTE_ALIGN(klm_num, 4) : 0) * MLX5_ST_SZ_DW(klm);
+       uint32_t in[in_size_dw];
        uint32_t out[MLX5_ST_SZ_DW(create_mkey_out)] = {0};
        void *mkc;
        struct mlx5_devx_obj *mkey = rte_zmalloc("mkey", sizeof(*mkey), 0);
@@ -153,27 +157,52 @@ mlx5_devx_cmd_mkey_create(struct ibv_context *ctx,
                rte_errno = ENOMEM;
                return NULL;
        }
+       memset(in, 0, in_size_dw * 4);
        pgsize = sysconf(_SC_PAGESIZE);
-       translation_size = (RTE_ALIGN(attr->size, pgsize) * 8) / 16;
        MLX5_SET(create_mkey_in, in, opcode, MLX5_CMD_OP_CREATE_MKEY);
+       mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
+       if (klm_num > 0) {
+               int i;
+               uint8_t *klm = (uint8_t *)MLX5_ADDR_OF(create_mkey_in, in,
+                                                      klm_pas_mtt);
+               translation_size = RTE_ALIGN(klm_num, 4);
+               for (i = 0; i < klm_num; i++) {
+                       MLX5_SET(klm, klm, byte_count, klm_array[i].byte_count);
+                       MLX5_SET(klm, klm, mkey, klm_array[i].mkey);
+                       MLX5_SET64(klm, klm, address, klm_array[i].address);
+                       klm += MLX5_ST_SZ_BYTES(klm);
+               }
+               for (; i < (int)translation_size; i++) {
+                       MLX5_SET(klm, klm, mkey, 0x0);
+                       MLX5_SET64(klm, klm, address, 0x0);
+                       klm += MLX5_ST_SZ_BYTES(klm);
+               }
+               MLX5_SET(mkc, mkc, access_mode_1_0, attr->log_entity_size ?
+                        MLX5_MKC_ACCESS_MODE_KLM_FBS :
+                        MLX5_MKC_ACCESS_MODE_KLM);
+               MLX5_SET(mkc, mkc, log_page_size, attr->log_entity_size);
+       } else {
+               translation_size = (RTE_ALIGN(attr->size, pgsize) * 8) / 16;
+               MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT);
+               MLX5_SET(mkc, mkc, log_page_size, rte_log2_u32(pgsize));
+       }
        MLX5_SET(create_mkey_in, in, translations_octword_actual_size,
                 translation_size);
        MLX5_SET(create_mkey_in, in, mkey_umem_id, attr->umem_id);
-       mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
+       MLX5_SET(create_mkey_in, in, pg_access, attr->pg_access);
        MLX5_SET(mkc, mkc, lw, 0x1);
        MLX5_SET(mkc, mkc, lr, 0x1);
-       MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT);
        MLX5_SET(mkc, mkc, qpn, 0xffffff);
        MLX5_SET(mkc, mkc, pd, attr->pd);
        MLX5_SET(mkc, mkc, mkey_7_0, attr->umem_id & 0xFF);
        MLX5_SET(mkc, mkc, translations_octword_size, translation_size);
        MLX5_SET64(mkc, mkc, start_addr, attr->addr);
        MLX5_SET64(mkc, mkc, len, attr->size);
-       MLX5_SET(mkc, mkc, log_page_size, rte_log2_u32(pgsize));
-       mkey->obj = mlx5_glue->devx_obj_create(ctx, in, sizeof(in), out,
+       mkey->obj = mlx5_glue->devx_obj_create(ctx, in, in_size_dw * 4, out,
                                               sizeof(out));
        if (!mkey->obj) {
-               DRV_LOG(ERR, "Can't create mkey - error %d", errno);
+               DRV_LOG(ERR, "Can't create %sdirect mkey - error %d\n",
+                       klm_num ? "an in" : "a ", errno);
                rte_errno = errno;
                rte_free(mkey);
                return NULL;

diff --git a/drivers/common/mlx5/mlx5_devx_cmds.h b/drivers/common/mlx5/mlx5_devx_cmds.h
index c1c9e99..c76c172 100644 (file)
--- a/drivers/common/mlx5/mlx5_devx_cmds.h
+++ b/drivers/common/mlx5/mlx5_devx_cmds.h
@@ -6,6 +6,7 @@
 #define RTE_PMD_MLX5_DEVX_CMDS_H_
 
 #include "mlx5_glue.h"
+#include "mlx5_prm.h"
 
 
 /* devX creation object */
@@ -14,11 +15,26 @@ struct mlx5_devx_obj {
        int id; /* The object ID. */
 };
 
+/* UMR memory buffer used to define 1 entry in indirect mkey. */
+struct mlx5_klm {
+       uint32_t byte_count;
+       uint32_t mkey;
+       uint64_t address;
+};
+
+/* This is limitation of libibverbs: in length variable type is u16. */
+#define MLX5_DEVX_MAX_KLM_ENTRIES ((UINT16_MAX - \
+               MLX5_ST_SZ_DW(create_mkey_in) * 4) / (MLX5_ST_SZ_DW(klm) * 4))
+
 struct mlx5_devx_mkey_attr {
        uint64_t addr;
        uint64_t size;
        uint32_t umem_id;
        uint32_t pd;
+       uint32_t log_entity_size;
+       uint32_t pg_access:1;
+       struct mlx5_klm *klm_array;
+       int klm_num;
 };
 
 /* HCA qos attributes. */
@@ -216,6 +232,7 @@ struct mlx5_devx_modify_sq_attr {
        uint32_t hairpin_peer_vhca:16;
 };
 
+
 /* mlx5_devx_cmds.c */
 
 struct mlx5_devx_obj *mlx5_devx_cmd_flow_counter_alloc(struct ibv_context *ctx,

diff --git a/drivers/common/mlx5/mlx5_prm.h b/drivers/common/mlx5/mlx5_prm.h
index efd6ad4..db15bb6 100644 (file)
--- a/drivers/common/mlx5/mlx5_prm.h
+++ b/drivers/common/mlx5/mlx5_prm.h
@@ -726,6 +726,8 @@ enum {
 
 enum {
        MLX5_MKC_ACCESS_MODE_MTT   = 0x1,
+       MLX5_MKC_ACCESS_MODE_KLM   = 0x2,
+       MLX5_MKC_ACCESS_MODE_KLM_FBS = 0x3,
 };
 
 /* Flow counters. */
@@ -790,6 +792,16 @@ struct mlx5_ifc_query_flow_counter_in_bits {
        u8         flow_counter_id[0x20];
 };
 
+#define MLX5_MAX_KLM_BYTE_COUNT 0x80000000u
+#define MLX5_MIN_KLM_FIXED_BUFFER_SIZE 0x1000u
+
+
+struct mlx5_ifc_klm_bits {
+       u8         byte_count[0x20];
+       u8         mkey[0x20];
+       u8         address[0x40];
+};
+
 struct mlx5_ifc_mkc_bits {
        u8         reserved_at_0[0x1];
        u8         free[0x1];

diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 1b31602..5610d94 100644 (file)
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -3885,6 +3885,10 @@ flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
        mkey_attr.size = size;
        mkey_attr.umem_id = mem_mng->umem->umem_id;
        mkey_attr.pd = sh->pdn;
+       mkey_attr.log_entity_size = 0;
+       mkey_attr.pg_access = 0;
+       mkey_attr.klm_array = NULL;
+       mkey_attr.klm_num = 0;
        mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
        if (!mem_mng->dm) {
                mlx5_glue->devx_umem_dereg(mem_mng->umem);