iokernel: support mlx5

.gitignore

@@ -3,7 +3,7 @@
 *.a
 [._]*.sw[a-p]
 iokerneld
-iokerenld-noht
+iokerneld-noht
 *~
 .cproject
 .project

Makefile

@@ -6,7 +6,9 @@ LD	= gcc
 CC	= gcc
 AR	= ar
 SPARSE	= sparse
-MLX=$(shell lspci | grep 'ConnectX-3' || echo "")
+# uncomment to autodetect MLX5
+# MLX5=$(shell lspci | grep 'ConnectX-5' || echo "")
+MLX4=$(shell lspci | grep 'ConnectX-3' || echo "")
 
 CHECKFLAGS = -D__CHECKER__ -Waddress-space
 
@@ -21,8 +23,12 @@ ifneq ($(TCP_RX_STATS),)
 CFLAGS += -DTCP_RX_STATS
 endif
 
-ifneq ($(MLX),)
-CFLAGS += -DMLX
+ifneq ($(MLX5),)
+CFLAGS += -DMLX5
+else
+ifneq ($(MLX4),)
+CFLAGS += -DMLX4
+endif
 endif
 
 # handy for debugging
@@ -71,9 +77,13 @@ DPDK_LIBS += -lrte_mempool
 DPDK_LIBS += -lrte_mempool_stack
 DPDK_LIBS += -lrte_ring
 # additional libs for running with Mellanox NICs
-ifneq ($(MLX),)
+ifneq ($(MLX5),)
+DPDK_LIBS +=  -lrte_pmd_mlx5 -libverbs -lmlx5 -lmnl
+else
+ifneq ($(MLX4),)
 DPDK_LIBS += -lrte_pmd_mlx4 -libverbs -lmlx4
 endif
+endif
 
 # must be first
 all: libbase.a libdune.a libnet.a libruntime.a iokerneld iokerneld-noht $(test_targets)

dpdk.sh

@@ -9,7 +9,10 @@ git submodule update --recursive
 # Apply driver patches
 patch -p 1 -d dpdk/ < ixgbe_18_11.patch
 
-if lspci | grep -q 'ConnectX-3'; then
+
+if lspci | grep -q 'ConnectX-5'; then
+   patch -p 1 -d dpdk/ < mlx5_18_11.patch
+elif lspci | grep -q 'ConnectX-3'; then
     patch -p 1 -d dpdk/ < mlx4_18_11.patch
     sed -i 's/CONFIG_RTE_LIBRTE_MLX4_PMD=n/CONFIG_RTE_LIBRTE_MLX4_PMD=y/g' dpdk/config/common_base
 fi

iokernel/defs.h

@@ -12,6 +12,7 @@
 #include <iokernel/control.h>
 #include <net/ethernet.h>
 
+#include "mlx.h"
 #include "ref.h"
 
 /* #define STATS 1 */

iokernel/dp_clients.c

@@ -12,10 +12,6 @@
 
 #include "defs.h"
 
-#ifdef MLX
-#include <mlx4_custom.h>
-#endif
-
 #define MAC_TO_PROC_ENTRIES	128
 
 static struct lrpc_chan_out lrpc_data_to_control;
@@ -36,7 +32,7 @@ static void dp_clients_add_client(struct proc *p)
 		log_err("dp_clients: failed to add MAC to hash table in add_client");
 
 #ifdef MLX
-	p->mr = mlx4_manual_reg_mr(dp.port, p->region.base, p->region.len, &p->lkey);
+	p->mr = mlx_reg_mem(dp.port, p->region.base, p->region.len, &p->lkey);
 	if (!p->mr)
 		log_err("dp clients: failed to register memory with MLX nic");
 #endif
@@ -78,7 +74,7 @@ static void dp_clients_remove_client(struct proc *p)
 		log_err("dp_clients: failed to remove MAC from hash table in remove "
 				"client");
 #ifdef MLX
-	mlx4_manual_dereg_mr(p->mr);
+	mlx_dereg_mem(p->mr);
 #endif
 
 	/* TODO: free queued packets/commands? */

iokernel/mlx.h

@@ -0,0 +1,17 @@
+#pragma once
+
+#ifdef MLX4
+
+#include <mlx4_custom.h>
+#define MLX
+#define mlx_reg_mem mlx4_manual_reg_mr
+#define mlx_dereg_mem mlx4_manual_dereg_mr
+
+
+#elif defined(MLX5)
+#include <mlx5_custom.h>
+#define MLX
+#define mlx_reg_mem mlx5_manual_reg_mr
+#define mlx_dereg_mem mlx5_manual_dereg_mr
+
+#endif

mlx5_18_11.patch

@@ -0,0 +1,150 @@
+From 970312d4e39fad07a67683e24b0870efd9d7b3e1 Mon Sep 17 00:00:00 2001
+From: Josh Fried <joshuafried@gmail.com>
+Date: Thu, 31 Jan 2019 12:53:00 -0500
+Subject: [PATCH] MLX5 support for the iokernel
+
+---
+ config/common_base                        |  2 +-
+ drivers/net/mlx5/Makefile                 |  2 ++
+ drivers/net/mlx5/mlx5_custom.h            |  8 ++++++++
+ drivers/net/mlx5/mlx5_mr.c                | 17 +++++++++++++++++
+ drivers/net/mlx5/mlx5_mr.h                |  2 ++
+ drivers/net/mlx5/mlx5_rxtx.h              | 22 +++++++---------------
+ drivers/net/mlx5/rte_pmd_mlx5_version.map |  4 ++++
+ 7 files changed, 41 insertions(+), 16 deletions(-)
+ create mode 100644 drivers/net/mlx5/mlx5_custom.h
+
+diff --git a/config/common_base b/config/common_base
+index d12ae98bc..616b15fed 100644
+--- a/config/common_base
++++ b/config/common_base
+@@ -317,7 +317,7 @@ CONFIG_RTE_LIBRTE_MLX4_DLOPEN_DEPS=n
+ # Compile burst-oriented Mellanox ConnectX-4, ConnectX-5 & Bluefield
+ # (MLX5) PMD
+ #
+-CONFIG_RTE_LIBRTE_MLX5_PMD=n
++CONFIG_RTE_LIBRTE_MLX5_PMD=y
+ CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
+ CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS=n
+
+diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
+index 895cdfeed..a9f25ff5c 100644
+--- a/drivers/net/mlx5/Makefile
++++ b/drivers/net/mlx5/Makefile
+@@ -10,6 +10,8 @@ LIB_GLUE = $(LIB_GLUE_BASE).$(LIB_GLUE_VERSION)
+ LIB_GLUE_BASE = librte_pmd_mlx5_glue.so
+ LIB_GLUE_VERSION = 18.11.0
+
++SYMLINK-$(CONFIG_RTE_LIBRTE_MLX5_PMD)-include += mlx5_custom.h
++
+ # Sources.
+ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5.c
+ ifneq ($(CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS),y)
+diff --git a/drivers/net/mlx5/mlx5_custom.h b/drivers/net/mlx5/mlx5_custom.h
+new file mode 100644
+index 000000000..925721de9
+--- /dev/null
++++ b/drivers/net/mlx5/mlx5_custom.h
+@@ -0,0 +1,8 @@
++
++#ifndef RTE_PMD_MLX5_CUSTOM_H
++#define RTE_PMD_MLX5_CUSTOM_H
++
++void *mlx5_manual_reg_mr(uint8_t port_id, void *addr, size_t length, uint32_t *lkey_out);
++void mlx5_manual_dereg_mr(void *ibv_mr);
++
++#endif /* RTE_PMD_MLX5_CUSTOM_H */
+diff --git a/drivers/net/mlx5/mlx5_mr.c b/drivers/net/mlx5/mlx5_mr.c
+index 442b2d232..00ef98ab6 100644
+--- a/drivers/net/mlx5/mlx5_mr.c
++++ b/drivers/net/mlx5/mlx5_mr.c
+@@ -1368,3 +1368,20 @@ mlx5_mr_release(struct rte_eth_dev *dev)
+ 	/* Free all remaining MRs. */
+ 	mlx5_mr_garbage_collect(dev);
+ }
++
++void *
++mlx5_manual_reg_mr(uint8_t port_id, void *addr, size_t length, uint32_t *lkey_out)
++{
++	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
++	struct priv *priv = dev->data->dev_private;
++	struct ibv_mr *ibv_mr = mlx5_glue->reg_mr(priv->pd, addr, length, IBV_ACCESS_LOCAL_WRITE);
++	if (ibv_mr && lkey_out) *lkey_out = rte_cpu_to_be_32(ibv_mr->lkey);
++
++	return ibv_mr;
++}
++
++void
++mlx5_manual_dereg_mr(void *ibv_mr)
++{
++	mlx5_glue->dereg_mr(ibv_mr);
++}
+diff --git a/drivers/net/mlx5/mlx5_mr.h b/drivers/net/mlx5/mlx5_mr.h
+index a57003fe9..9db10045e 100644
+--- a/drivers/net/mlx5/mlx5_mr.h
++++ b/drivers/net/mlx5/mlx5_mr.h
+@@ -26,6 +26,8 @@
+ #include <rte_rwlock.h>
+ #include <rte_bitmap.h>
+
++#include "mlx5_custom.h"
++
+ /* Memory Region object. */
+ struct mlx5_mr {
+ 	LIST_ENTRY(mlx5_mr) mr; /**< Pointer to the prev/next entry. */
+diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
+index f47d327cf..18612cc2d 100644
+--- a/drivers/net/mlx5/mlx5_rxtx.h
++++ b/drivers/net/mlx5/mlx5_rxtx.h
+@@ -656,6 +656,10 @@ mlx5_rx_addr2mr(struct mlx5_rxq_data *rxq, uintptr_t addr)
+
+ #define mlx5_rx_mb2mr(rxq, mb) mlx5_rx_addr2mr(rxq, (uintptr_t)((mb)->buf_addr))
+
++struct mem_info {
++	uint32_t lkey;
++};
++
+ /**
+  * Query LKey from a packet buffer for Tx. If not found, add the mempool.
+  *
+@@ -668,22 +672,10 @@ mlx5_rx_addr2mr(struct mlx5_rxq_data *rxq, uintptr_t addr)
+  *   Searched LKey on success, UINT32_MAX on no match.
+  */
+ static __rte_always_inline uint32_t
+-mlx5_tx_mb2mr(struct mlx5_txq_data *txq, struct rte_mbuf *mb)
++mlx5_tx_mb2mr(struct mlx5_txq_data *txq __rte_unused, struct rte_mbuf *mb)
+ {
+-	struct mlx5_mr_ctrl *mr_ctrl = &txq->mr_ctrl;
+-	uintptr_t addr = (uintptr_t)mb->buf_addr;
+-	uint32_t lkey;
+-
+-	/* Check generation bit to see if there's any change on existing MRs. */
+-	if (unlikely(*mr_ctrl->dev_gen_ptr != mr_ctrl->cur_gen))
+-		mlx5_mr_flush_local_cache(mr_ctrl);
+-	/* Linear search on MR cache array. */
+-	lkey = mlx5_mr_lookup_cache(mr_ctrl->cache, &mr_ctrl->mru,
+-				    MLX5_MR_CACHE_N, addr);
+-	if (likely(lkey != UINT32_MAX))
+-		return lkey;
+-	/* Take slower bottom-half on miss. */
+-	return mlx5_tx_mb2mr_bh(txq, mb);
++	struct mem_info *m = (struct mem_info *)(((char *) mb) + sizeof(struct rte_mbuf));
++	return m->lkey;
+ }
+
+ /**
+diff --git a/drivers/net/mlx5/rte_pmd_mlx5_version.map b/drivers/net/mlx5/rte_pmd_mlx5_version.map
+index ad607bbed..eb819010c 100644
+--- a/drivers/net/mlx5/rte_pmd_mlx5_version.map
++++ b/drivers/net/mlx5/rte_pmd_mlx5_version.map
+@@ -1,3 +1,7 @@
+ DPDK_2.2 {
+ 	local: *;
++
++	global:
++	mlx5_manual_reg_mr;
++	mlx5_manual_dereg_mr;
+ };
+-- 
+2.17.1
+