net/mlx5: replace IPC socket with EAL API
authorYongseok Koh <yskoh@mellanox.com>
Mon, 1 Apr 2019 21:12:54 +0000 (14:12 -0700)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 5 Apr 2019 15:45:22 +0000 (17:45 +0200)
Socket API is used for IPC in order for secondary process to acquire
Verb command file descriptor. The FD is used to remap UAR address.
The multi-process APIs (rte_mp) in EAL are newly introduced.
mlx5_socket.c is replaced with mlx5_mp.c, which uses the new APIs.

As it is PMD global infrastructure, only one IPC channel is established.
All the IPC message types may have port_id in the message if there is
need to reference a specific device.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
drivers/net/mlx5/Makefile
drivers/net/mlx5/meson.build
drivers/net/mlx5/mlx5.c
drivers/net/mlx5/mlx5.h
drivers/net/mlx5/mlx5_ethdev.c
drivers/net/mlx5/mlx5_mp.c [new file with mode: 0644]
drivers/net/mlx5/mlx5_socket.c [deleted file]

index c326494..e5587e9 100644 (file)
@@ -34,7 +34,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_flow.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_flow_dv.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_flow_tcf.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_flow_verbs.c
-SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_socket.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_mp.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_nl.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5_devx_cmds.c
 
index e3cb9bc..fe880b9 100644 (file)
@@ -41,7 +41,7 @@ if build
                'mlx5_rxmode.c',
                'mlx5_rxq.c',
                'mlx5_rxtx.c',
-               'mlx5_socket.c',
+               'mlx5_mp.c',
                'mlx5_stats.c',
                'mlx5_trigger.c',
                'mlx5_txq.c',
index 2208cc9..61ef29c 100644 (file)
@@ -325,6 +325,7 @@ mlx5_prepare_shared_data(void)
                        rte_mem_event_callback_register("MLX5_MEM_EVENT_CB",
                                                        mlx5_mr_mem_event_cb,
                                                        NULL);
+                       mlx5_mp_init();
                }
        }
        rte_spinlock_unlock(&mlx5_shared_data_lock);
@@ -454,8 +455,6 @@ mlx5_dev_close(struct rte_eth_dev *dev)
                rte_free(priv->rss_conf.rss_key);
        if (priv->reta_idx != NULL)
                rte_free(priv->reta_idx);
-       if (priv->primary_socket)
-               mlx5_socket_uninit(dev);
        if (priv->config.vf)
                mlx5_nl_mac_addr_flush(dev);
        if (priv->nl_socket_route >= 0)
@@ -970,7 +969,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
                if (err)
                        return NULL;
                /* Receive command fd from primary process */
-               err = mlx5_socket_connect(eth_dev);
+               err = mlx5_mp_req_verbs_cmd_fd(eth_dev);
                if (err < 0)
                        return NULL;
                /* Remap UAR for Tx queues. */
index 7402798..f454adb 100644 (file)
@@ -56,6 +56,24 @@ enum {
        PCI_DEVICE_ID_MELLANOX_CONNECTX6VF = 0x101c,
 };
 
+/* Request types for IPC. */
+enum mlx5_mp_req_type {
+       MLX5_MP_REQ_VERBS_CMD_FD = 1,
+};
+
+/* Pameters for IPC. */
+struct mlx5_mp_param {
+       enum mlx5_mp_req_type type;
+       int port_id;
+       int result;
+};
+
+/** Request timeout for IPC. */
+#define MLX5_MP_REQ_TIMEOUT_SEC 5
+
+/** Key string for IPC. */
+#define MLX5_MP_NAME "net_mlx5_mp"
+
 /** Switch information returned by mlx5_nl_switch_info(). */
 struct mlx5_switch_info {
        uint32_t master:1; /**< Master device. */
@@ -272,9 +290,7 @@ struct mlx5_priv {
        uint32_t link_speed_capa; /* Link speed capabilities. */
        struct mlx5_xstats_ctrl xstats_ctrl; /* Extended stats control. */
        struct mlx5_stats_ctrl stats_ctrl; /* Stats control. */
-       int primary_socket; /* Unix socket for primary process. */
        void *uar_base; /* Reserved address space for UAR mapping */
-       struct rte_intr_handle intr_handle_socket; /* Interrupt handler. */
        struct mlx5_dev_config config; /* Device configuration. */
        struct mlx5_verbs_alloc_ctx verbs_alloc_ctx;
        /* Context for Verbs allocator. */
@@ -432,12 +448,9 @@ int mlx5_ctrl_flow(struct rte_eth_dev *dev,
 int mlx5_flow_create_drop_queue(struct rte_eth_dev *dev);
 void mlx5_flow_delete_drop_queue(struct rte_eth_dev *dev);
 
-/* mlx5_socket.c */
-
-int mlx5_socket_init(struct rte_eth_dev *priv);
-void mlx5_socket_uninit(struct rte_eth_dev *priv);
-void mlx5_socket_handle(struct rte_eth_dev *priv);
-int mlx5_socket_connect(struct rte_eth_dev *priv);
+/* mlx5_mp.c */
+int mlx5_mp_req_verbs_cmd_fd(struct rte_eth_dev *dev);
+void mlx5_mp_init(void);
 
 /* mlx5_nl.c */
 
index 7273bd9..aab8e67 100644 (file)
@@ -1095,20 +1095,6 @@ mlx5_dev_interrupt_handler(void *cb_arg)
        }
 }
 
-/**
- * Handle interrupts from the socket.
- *
- * @param cb_arg
- *   Callback argument.
- */
-static void
-mlx5_dev_handler_socket(void *cb_arg)
-{
-       struct rte_eth_dev *dev = cb_arg;
-
-       mlx5_socket_handle(dev);
-}
-
 /**
  * Uninstall shared asynchronous device events handler.
  * This function is implemeted to support event sharing
@@ -1208,14 +1194,7 @@ exit:
 void
 mlx5_dev_interrupt_handler_uninstall(struct rte_eth_dev *dev)
 {
-       struct mlx5_priv *priv = dev->data->dev_private;
-
        mlx5_dev_shared_handler_uninstall(dev);
-       if (priv->primary_socket)
-               rte_intr_callback_unregister(&priv->intr_handle_socket,
-                                            mlx5_dev_handler_socket, dev);
-       priv->intr_handle_socket.fd = 0;
-       priv->intr_handle_socket.type = RTE_INTR_HANDLE_UNKNOWN;
 }
 
 /**
@@ -1227,20 +1206,7 @@ mlx5_dev_interrupt_handler_uninstall(struct rte_eth_dev *dev)
 void
 mlx5_dev_interrupt_handler_install(struct rte_eth_dev *dev)
 {
-       struct mlx5_priv *priv = dev->data->dev_private;
-       int ret;
-
        mlx5_dev_shared_handler_install(dev);
-       ret = mlx5_socket_init(dev);
-       if (ret)
-               DRV_LOG(ERR, "port %u cannot initialise socket: %s",
-                       dev->data->port_id, strerror(rte_errno));
-       else if (priv->primary_socket) {
-               priv->intr_handle_socket.fd = priv->primary_socket;
-               priv->intr_handle_socket.type = RTE_INTR_HANDLE_EXT;
-               rte_intr_callback_register(&priv->intr_handle_socket,
-                                          mlx5_dev_handler_socket, dev);
-       }
 }
 
 /**
diff --git a/drivers/net/mlx5/mlx5_mp.c b/drivers/net/mlx5/mlx5_mp.c
new file mode 100644 (file)
index 0000000..71a2b66
--- /dev/null
@@ -0,0 +1,139 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2019 6WIND S.A.
+ * Copyright 2019 Mellanox Technologies, Ltd
+ */
+
+#include <assert.h>
+#include <stdio.h>
+#include <time.h>
+
+#include <rte_eal.h>
+#include <rte_ethdev_driver.h>
+#include <rte_string_fns.h>
+
+#include "mlx5.h"
+#include "mlx5_utils.h"
+
+/**
+ * Initialize IPC message.
+ *
+ * @param[in] dev
+ *   Pointer to Ethernet structure.
+ * @param[out] msg
+ *   Pointer to message to fill in.
+ * @param[in] type
+ *   Message type.
+ */
+static inline void
+mp_init_msg(struct rte_eth_dev *dev, struct rte_mp_msg *msg,
+           enum mlx5_mp_req_type type)
+{
+       struct mlx5_mp_param *param = (struct mlx5_mp_param *)msg->param;
+
+       memset(msg, 0, sizeof(*msg));
+       strlcpy(msg->name, MLX5_MP_NAME, sizeof(msg->name));
+       msg->len_param = sizeof(*param);
+       param->type = type;
+       param->port_id = dev->data->port_id;
+}
+
+/**
+ * IPC message handler of primary process.
+ *
+ * @param[in] dev
+ *   Pointer to Ethernet structure.
+ * @param[in] peer
+ *   Pointer to the peer socket path.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mp_primary_handle(const struct rte_mp_msg *mp_msg, const void *peer)
+{
+       struct rte_mp_msg mp_res;
+       struct mlx5_mp_param *res = (struct mlx5_mp_param *)mp_res.param;
+       const struct mlx5_mp_param *param =
+               (const struct mlx5_mp_param *)mp_msg->param;
+       struct rte_eth_dev *dev;
+       struct mlx5_priv *priv;
+       int ret;
+
+       assert(rte_eal_process_type() == RTE_PROC_PRIMARY);
+       if (!rte_eth_dev_is_valid_port(param->port_id)) {
+               rte_errno = ENODEV;
+               DRV_LOG(ERR, "port %u invalid port ID", param->port_id);
+               return -rte_errno;
+       }
+       dev = &rte_eth_devices[param->port_id];
+       priv = dev->data->dev_private;
+       switch (param->type) {
+       case MLX5_MP_REQ_VERBS_CMD_FD:
+               mp_init_msg(dev, &mp_res, param->type);
+               mp_res.num_fds = 1;
+               mp_res.fds[0] = priv->sh->ctx->cmd_fd;
+               res->result = 0;
+               ret = rte_mp_reply(&mp_res, peer);
+               break;
+       default:
+               rte_errno = EINVAL;
+               DRV_LOG(ERR, "port %u invalid mp request type",
+                       dev->data->port_id);
+               return -rte_errno;
+       }
+       return ret;
+}
+
+/**
+ * Request Verbs command file descriptor for mmap to the primary process.
+ *
+ * @param[in] dev
+ *   Pointer to Ethernet structure.
+ *
+ * @return
+ *   fd on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_mp_req_verbs_cmd_fd(struct rte_eth_dev *dev)
+{
+       struct rte_mp_msg mp_req;
+       struct rte_mp_msg *mp_res;
+       struct rte_mp_reply mp_rep;
+       struct mlx5_mp_param *res;
+       struct timespec ts = {.tv_sec = MLX5_MP_REQ_TIMEOUT_SEC, .tv_nsec = 0};
+       int ret;
+
+       assert(rte_eal_process_type() == RTE_PROC_SECONDARY);
+       mp_init_msg(dev, &mp_req, MLX5_MP_REQ_VERBS_CMD_FD);
+       ret = rte_mp_request_sync(&mp_req, &mp_rep, &ts);
+       if (ret) {
+               DRV_LOG(ERR, "port %u request to primary process failed",
+                       dev->data->port_id);
+               return -rte_errno;
+       }
+       assert(mp_rep.nb_received == 1);
+       mp_res = &mp_rep.msgs[0];
+       res = (struct mlx5_mp_param *)mp_res->param;
+       if (res->result) {
+               rte_errno = -res->result;
+               DRV_LOG(ERR,
+                       "port %u failed to get command FD from primary process",
+                       dev->data->port_id);
+               ret = -rte_errno;
+               goto exit;
+       }
+       assert(mp_res->num_fds == 1);
+       ret = mp_res->fds[0];
+       DRV_LOG(DEBUG, "port %u command FD from primary is %d",
+               dev->data->port_id, ret);
+exit:
+       free(mp_rep.msgs);
+       return ret;
+}
+
+void
+mlx5_mp_init(void)
+{
+       if (rte_eal_process_type() == RTE_PROC_PRIMARY)
+               rte_mp_action_register(MLX5_MP_NAME, mp_primary_handle);
+}
diff --git a/drivers/net/mlx5/mlx5_socket.c b/drivers/net/mlx5/mlx5_socket.c
deleted file mode 100644 (file)
index 8fa6430..0000000
+++ /dev/null
@@ -1,306 +0,0 @@
-/* SPDX-License-Identifier: BSD-3-Clause
- * Copyright 2016 6WIND S.A.
- * Copyright 2016 Mellanox Technologies, Ltd
- */
-
-#include <sys/types.h>
-#include <sys/socket.h>
-#include <sys/un.h>
-#include <fcntl.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <sys/stat.h>
-
-#include "mlx5.h"
-#include "mlx5_utils.h"
-
-/**
- * Initialise the socket to communicate with the secondary process
- *
- * @param[in] dev
- *   Pointer to Ethernet device.
- *
- * @return
- *   0 on success, a negative errno value otherwise and rte_errno is set.
- */
-int
-mlx5_socket_init(struct rte_eth_dev *dev)
-{
-       struct mlx5_priv *priv = dev->data->dev_private;
-       struct sockaddr_un sun = {
-               .sun_family = AF_UNIX,
-       };
-       int ret;
-       int flags;
-
-       /*
-        * Close the last socket that was used to communicate
-        * with the secondary process
-        */
-       if (priv->primary_socket)
-               mlx5_socket_uninit(dev);
-       /*
-        * Initialise the socket to communicate with the secondary
-        * process.
-        */
-       ret = socket(AF_UNIX, SOCK_STREAM, 0);
-       if (ret < 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING, "port %u secondary process not supported: %s",
-                       dev->data->port_id, strerror(errno));
-               goto error;
-       }
-       priv->primary_socket = ret;
-       flags = fcntl(priv->primary_socket, F_GETFL, 0);
-       if (flags == -1) {
-               rte_errno = errno;
-               goto error;
-       }
-       ret = fcntl(priv->primary_socket, F_SETFL, flags | O_NONBLOCK);
-       if (ret < 0) {
-               rte_errno = errno;
-               goto error;
-       }
-       snprintf(sun.sun_path, sizeof(sun.sun_path), "/var/tmp/%s_%d",
-                MLX5_DRIVER_NAME, priv->primary_socket);
-       remove(sun.sun_path);
-       ret = bind(priv->primary_socket, (const struct sockaddr *)&sun,
-                  sizeof(sun));
-       if (ret < 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING,
-                       "port %u cannot bind socket, secondary process not"
-                       " supported: %s",
-                       dev->data->port_id, strerror(errno));
-               goto close;
-       }
-       ret = listen(priv->primary_socket, 0);
-       if (ret < 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING, "port %u secondary process not supported: %s",
-                       dev->data->port_id, strerror(errno));
-               goto close;
-       }
-       return 0;
-close:
-       remove(sun.sun_path);
-error:
-       claim_zero(close(priv->primary_socket));
-       priv->primary_socket = 0;
-       return -rte_errno;
-}
-
-/**
- * Un-Initialise the socket to communicate with the secondary process
- *
- * @param[in] dev
- */
-void
-mlx5_socket_uninit(struct rte_eth_dev *dev)
-{
-       struct mlx5_priv *priv = dev->data->dev_private;
-
-       MKSTR(path, "/var/tmp/%s_%d", MLX5_DRIVER_NAME, priv->primary_socket);
-       claim_zero(close(priv->primary_socket));
-       priv->primary_socket = 0;
-       claim_zero(remove(path));
-}
-
-/**
- * Handle socket interrupts.
- *
- * @param dev
- *   Pointer to Ethernet device.
- */
-void
-mlx5_socket_handle(struct rte_eth_dev *dev)
-{
-       struct mlx5_priv *priv = dev->data->dev_private;
-       int conn_sock;
-       int ret = 0;
-       struct cmsghdr *cmsg = NULL;
-       struct ucred *cred = NULL;
-       char buf[CMSG_SPACE(sizeof(struct ucred))] = { 0 };
-       char vbuf[1024] = { 0 };
-       struct iovec io = {
-               .iov_base = vbuf,
-               .iov_len = sizeof(*vbuf),
-       };
-       struct msghdr msg = {
-               .msg_iov = &io,
-               .msg_iovlen = 1,
-               .msg_control = buf,
-               .msg_controllen = sizeof(buf),
-       };
-       int *fd;
-
-       /* Accept the connection from the client. */
-       conn_sock = accept(priv->primary_socket, NULL, NULL);
-       if (conn_sock < 0) {
-               DRV_LOG(WARNING, "port %u connection failed: %s",
-                       dev->data->port_id, strerror(errno));
-               return;
-       }
-       ret = setsockopt(conn_sock, SOL_SOCKET, SO_PASSCRED, &(int){1},
-                                        sizeof(int));
-       if (ret < 0) {
-               ret = errno;
-               DRV_LOG(WARNING, "port %u cannot change socket options: %s",
-                       dev->data->port_id, strerror(rte_errno));
-               goto error;
-       }
-       ret = recvmsg(conn_sock, &msg, MSG_WAITALL);
-       if (ret < 0) {
-               ret = errno;
-               DRV_LOG(WARNING, "port %u received an empty message: %s",
-                       dev->data->port_id, strerror(rte_errno));
-               goto error;
-       }
-       /* Expect to receive credentials only. */
-       cmsg = CMSG_FIRSTHDR(&msg);
-       if (cmsg == NULL) {
-               DRV_LOG(WARNING, "port %u no message", dev->data->port_id);
-               goto error;
-       }
-       if ((cmsg->cmsg_type == SCM_CREDENTIALS) &&
-               (cmsg->cmsg_len >= sizeof(*cred))) {
-               cred = (struct ucred *)CMSG_DATA(cmsg);
-               assert(cred != NULL);
-       }
-       cmsg = CMSG_NXTHDR(&msg, cmsg);
-       if (cmsg != NULL) {
-               DRV_LOG(WARNING, "port %u message wrongly formatted",
-                       dev->data->port_id);
-               goto error;
-       }
-       /* Make sure all the ancillary data was received and valid. */
-       if ((cred == NULL) || (cred->uid != getuid()) ||
-           (cred->gid != getgid())) {
-               DRV_LOG(WARNING, "port %u wrong credentials",
-                       dev->data->port_id);
-               goto error;
-       }
-       /* Set-up the ancillary data. */
-       cmsg = CMSG_FIRSTHDR(&msg);
-       assert(cmsg != NULL);
-       cmsg->cmsg_level = SOL_SOCKET;
-       cmsg->cmsg_type = SCM_RIGHTS;
-       cmsg->cmsg_len = CMSG_LEN(sizeof(priv->sh->ctx->cmd_fd));
-       fd = (int *)CMSG_DATA(cmsg);
-       *fd = priv->sh->ctx->cmd_fd;
-       ret = sendmsg(conn_sock, &msg, 0);
-       if (ret < 0)
-               DRV_LOG(WARNING, "port %u cannot send response",
-                       dev->data->port_id);
-error:
-       close(conn_sock);
-}
-
-/**
- * Connect to the primary process.
- *
- * @param[in] dev
- *   Pointer to Ethernet structure.
- *
- * @return
- *   fd on success, negative errno value otherwise and rte_errno is set.
- */
-int
-mlx5_socket_connect(struct rte_eth_dev *dev)
-{
-       struct mlx5_priv *priv = dev->data->dev_private;
-       struct sockaddr_un sun = {
-               .sun_family = AF_UNIX,
-       };
-       int socket_fd = -1;
-       int *fd = NULL;
-       int ret;
-       struct ucred *cred;
-       char buf[CMSG_SPACE(sizeof(*cred))] = { 0 };
-       char vbuf[1024] = { 0 };
-       struct iovec io = {
-               .iov_base = vbuf,
-               .iov_len = sizeof(*vbuf),
-       };
-       struct msghdr msg = {
-               .msg_control = buf,
-               .msg_controllen = sizeof(buf),
-               .msg_iov = &io,
-               .msg_iovlen = 1,
-       };
-       struct cmsghdr *cmsg;
-
-       ret = socket(AF_UNIX, SOCK_STREAM, 0);
-       if (ret < 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING, "port %u cannot connect to primary",
-                       dev->data->port_id);
-               goto error;
-       }
-       socket_fd = ret;
-       snprintf(sun.sun_path, sizeof(sun.sun_path), "/var/tmp/%s_%d",
-                MLX5_DRIVER_NAME, priv->primary_socket);
-       ret = connect(socket_fd, (const struct sockaddr *)&sun, sizeof(sun));
-       if (ret < 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING, "port %u cannot connect to primary",
-                       dev->data->port_id);
-               goto error;
-       }
-       cmsg = CMSG_FIRSTHDR(&msg);
-       if (cmsg == NULL) {
-               rte_errno = EINVAL;
-               DRV_LOG(DEBUG, "port %u cannot get first message",
-                       dev->data->port_id);
-               goto error;
-       }
-       cmsg->cmsg_level = SOL_SOCKET;
-       cmsg->cmsg_type = SCM_CREDENTIALS;
-       cmsg->cmsg_len = CMSG_LEN(sizeof(*cred));
-       cred = (struct ucred *)CMSG_DATA(cmsg);
-       if (cred == NULL) {
-               rte_errno = EINVAL;
-               DRV_LOG(DEBUG, "port %u no credentials received",
-                       dev->data->port_id);
-               goto error;
-       }
-       cred->pid = getpid();
-       cred->uid = getuid();
-       cred->gid = getgid();
-       ret = sendmsg(socket_fd, &msg, MSG_DONTWAIT);
-       if (ret < 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING,
-                       "port %u cannot send credentials to primary: %s",
-                       dev->data->port_id, strerror(errno));
-               goto error;
-       }
-       ret = recvmsg(socket_fd, &msg, MSG_WAITALL);
-       if (ret <= 0) {
-               rte_errno = errno;
-               DRV_LOG(WARNING, "port %u no message from primary: %s",
-                       dev->data->port_id, strerror(errno));
-               goto error;
-       }
-       cmsg = CMSG_FIRSTHDR(&msg);
-       if (cmsg == NULL) {
-               rte_errno = EINVAL;
-               DRV_LOG(WARNING, "port %u no file descriptor received",
-                       dev->data->port_id);
-               goto error;
-       }
-       fd = (int *)CMSG_DATA(cmsg);
-       if (*fd < 0) {
-               DRV_LOG(WARNING, "port %u no file descriptor received: %s",
-                       dev->data->port_id, strerror(errno));
-               rte_errno = *fd;
-               goto error;
-       }
-       ret = *fd;
-       close(socket_fd);
-       return ret;
-error:
-       if (socket_fd != -1)
-               close(socket_fd);
-       return -rte_errno;
-}