net/mlx5: add control of excessive memory pinning by kernel

A new PMD parameter (mr_ext_memseg_en) is added to control extension of
memseg when creating a MR. It is enabled by default.

If enabled, mlx5_mr_create() tries to maximize the range of MR
registration so that the LKey lookup tables on datapath become smaller
and get the best performance. However, it may worsen memory utilization
because registered memory is pinned by kernel driver. Even if a page in
the extended chunk is freed, that doesn't become reusable until the
entire memory is freed and the MR is destroyed.

To make freed pages available immediately, this parameter has to be
turned off but it could drop performance.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index fa9bf73da7af9fad908ab182cc2a95a19418f85b..e5e3d9061ed93335245bd2b69ccfcec5175cb78f 100644 (file)
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -485,6 +485,17 @@ Run-time configuration
 
   Disabled by default.
 
+- ``mr_ext_memseg_en`` parameter [int]
+
+  A nonzero value enables extending memseg when registering DMA memory. If
+  enabled, the number of entries in MR (Memory Region) lookup table on datapath
+  is minimized and it benefits performance. On the other hand, it worsens memory
+  utilization because registered memory is pinned by kernel driver. Even if a
+  page in the extended chunk is freed, that doesn't become reusable until the
+  entire memory is freed.
+
+  Enabled by default.
+
 - ``representor`` parameter [list]
 
   This parameter can be used to instantiate DPDK Ethernet devices from

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 2b7a6d121fd451175d8b01d0fe5c886204019b52..40445056f52b8e389fb07072c7edc466f66f5c87 100644 (file)
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -107,6 +107,9 @@
 /* Activate Netlink support in VF mode. */
 #define MLX5_VF_NL_EN "vf_nl_en"
 
+/* Enable extending memsegs when creating a MR. */
+#define MLX5_MR_EXT_MEMSEG_EN "mr_ext_memseg_en"
+
 /* Select port representors to instantiate. */
 #define MLX5_REPRESENTOR "representor"
 
@@ -732,6 +735,8 @@ mlx5_args_check(const char *key, const char *val, void *opaque)
                config->vf_nl_en = !!tmp;
        } else if (strcmp(MLX5_DV_FLOW_EN, key) == 0) {
                config->dv_flow_en = !!tmp;
+       } else if (strcmp(MLX5_MR_EXT_MEMSEG_EN, key) == 0) {
+               config->mr_ext_memseg_en = !!tmp;
        } else {
                DRV_LOG(WARNING, "%s: unknown parameter", key);
                rte_errno = EINVAL;
@@ -773,6 +778,7 @@ mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs)
                MLX5_L3_VXLAN_EN,
                MLX5_VF_NL_EN,
                MLX5_DV_FLOW_EN,
+               MLX5_MR_EXT_MEMSEG_EN,
                MLX5_REPRESENTOR,
                NULL,
        };
@@ -1853,6 +1859,7 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
                .txqs_vec = MLX5_ARG_UNSET,
                .inline_max_packet_sz = MLX5_ARG_UNSET,
                .vf_nl_en = 1,
+               .mr_ext_memseg_en = 1,
                .mprq = {
                        .enabled = 0, /* Disabled by default. */
                        .stride_num_n = MLX5_MPRQ_STRIDE_NUM_N,

diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 12692505c3f45b676c957299917d71102af72d5f..b3445f198f37c11e00bfdd70786e7ebf9394ef6f 100644 (file)
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -167,6 +167,8 @@ struct mlx5_dev_config {
        unsigned int tx_vec_en:1; /* Tx vector is enabled. */
        unsigned int rx_vec_en:1; /* Rx vector is enabled. */
        unsigned int mpw_hdr_dseg:1; /* Enable DSEGs in the title WQEBB. */
+       unsigned int mr_ext_memseg_en:1;
+       /* Whether memseg should be extended for MR creation. */
        unsigned int l3_vxlan_en:1; /* Enable L3 VXLAN flow creation. */
        unsigned int vf_nl_en:1; /* Enable Netlink requests in VF mode. */
        unsigned int dv_flow_en:1; /* Enable DV flow. */

diff --git a/drivers/net/mlx5/mlx5_mr.c b/drivers/net/mlx5/mlx5_mr.c
index e7f55be6e10bbd2259ac663e3271250771317a25..78d829722e3d1f8cef89599224592ccfd0786243 100644 (file)
--- a/drivers/net/mlx5/mlx5_mr.c
+++ b/drivers/net/mlx5/mlx5_mr.c
@@ -535,6 +535,7 @@ mlx5_mr_create(struct rte_eth_dev *dev, struct mlx5_mr_cache *entry,
               uintptr_t addr)
 {
        struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_dev_config *config = &priv->config;
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
        const struct rte_memseg_list *msl;
        const struct rte_memseg *ms;
@@ -570,14 +571,24 @@ mlx5_mr_create(struct rte_eth_dev *dev, struct mlx5_mr_cache *entry,
         */
        mlx5_mr_garbage_collect(dev);
        /*
-        * Find out a contiguous virtual address chunk in use, to which the
-        * given address belongs, in order to register maximum range. In the
-        * best case where mempools are not dynamically recreated and
+        * If enabled, find out a contiguous virtual address chunk in use, to
+        * which the given address belongs, in order to register maximum range.
+        * In the best case where mempools are not dynamically recreated and
         * '--socket-mem' is specified as an EAL option, it is very likely to
         * have only one MR(LKey) per a socket and per a hugepage-size even
-        * though the system memory is highly fragmented.
+        * though the system memory is highly fragmented. As the whole memory
+        * chunk will be pinned by kernel, it can't be reused unless entire
+        * chunk is freed from EAL.
+        *
+        * If disabled, just register one memseg (page). Then, memory
+        * consumption will be minimized but it may drop performance if there
+        * are many MRs to lookup on the datapath.
         */
-       if (!rte_memseg_contig_walk(mr_find_contig_memsegs_cb, &data)) {
+       if (!config->mr_ext_memseg_en) {
+               data.msl = rte_mem_virt2memseg_list((void *)addr);
+               data.start = RTE_ALIGN_FLOOR(addr, data.msl->page_sz);
+               data.end = data.start + data.msl->page_sz;
+       } else if (!rte_memseg_contig_walk(mr_find_contig_memsegs_cb, &data)) {
                DRV_LOG(WARNING,
                        "port %u unable to find virtually contiguous"
                        " chunk for address (%p)."