net/iavf: add thread for event callbacks

[ upstream commit cb5c1b9 ]

All callbacks registered for ethdev events are called in
eal-intr-thread, and some of them execute virtchnl commands.
Because interrupts are disabled in the intr thread, no response
will be received for these commands. So all callbacks should
be called in a new context.

When the device is bonded, the bond pmd registers a callback for
the LSC event to execute virtchnl commands to reinitialize the
device, and it would also raise the above issue.

This commit adds a new thread to call all event callbacks.

Fixes: 48de41c ("net/avf: enable link status update")
Fixes: 8410842 ("net/iavf: support asynchronous virtual channel message")

Signed-off-by: Yiding Zhou <yidingx.zhou@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>

drivers/net/iavf/iavf.h

@@ -400,6 +400,8 @@ _atomic_set_async_response_cmd(struct iavf_info *vf, enum virtchnl_ops ops)
 }
 int iavf_check_api_version(struct iavf_adapter *adapter);
 int iavf_get_vf_resource(struct iavf_adapter *adapter);
+void iavf_dev_event_handler_fini(void);
+int iavf_dev_event_handler_init(void);
 void iavf_handle_virtchnl_msg(struct rte_eth_dev *dev);
 int iavf_enable_vlan_strip(struct iavf_adapter *adapter);
 int iavf_disable_vlan_strip(struct iavf_adapter *adapter);

drivers/net/iavf/iavf_ethdev.c

@@ -2561,6 +2561,9 @@ iavf_dev_init(struct rte_eth_dev *eth_dev)
 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.addr,
 			&eth_dev->data->mac_addrs[0]);
 
+	if (iavf_dev_event_handler_init())
+		goto init_vf_err;
+
 	if (vf->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_WB_ON_ITR) {
 		/* register callback func to eal lib */
 		rte_intr_callback_register(pci_dev->intr_handle,
@@ -2715,6 +2718,8 @@ iavf_dev_uninit(struct rte_eth_dev *dev)
 
 	iavf_dev_close(dev);
 
+	iavf_dev_event_handler_fini();
+
 	return 0;
 }
 

drivers/net/iavf/iavf_vchnl.c

@@ -2,6 +2,7 @@
  * Copyright(c) 2017 Intel Corporation
  */
 
+#include <fcntl.h>
 #include <stdio.h>
 #include <errno.h>
 #include <stdint.h>
@@ -11,6 +12,7 @@
 #include <inttypes.h>
 #include <rte_byteorder.h>
 #include <rte_common.h>
+#include <rte_os_shim.h>
 
 #include <rte_debug.h>
 #include <rte_alarm.h>
@@ -27,6 +29,146 @@
 #define MAX_TRY_TIMES 2000
 #define ASQ_DELAY_MS  1
 
+#define MAX_EVENT_PENDING 16
+
+struct iavf_event_element {
+	TAILQ_ENTRY(iavf_event_element) next;
+	struct rte_eth_dev *dev;
+	enum rte_eth_event_type event;
+	void *param;
+	size_t param_alloc_size;
+	uint8_t param_alloc_data[0];
+};
+
+struct iavf_event_handler {
+	uint32_t ndev;
+	pthread_t tid;
+	int fd[2];
+	pthread_mutex_t lock;
+	TAILQ_HEAD(event_list, iavf_event_element) pending;
+};
+
+static struct iavf_event_handler event_handler = {
+	.fd = {-1, -1},
+};
+
+#ifndef TAILQ_FOREACH_SAFE
+#define TAILQ_FOREACH_SAFE(var, head, field, tvar) \
+	for ((var) = TAILQ_FIRST((head)); \
+		(var) && ((tvar) = TAILQ_NEXT((var), field), 1); \
+	(var) = (tvar))
+#endif
+
+static void *
+iavf_dev_event_handle(void *param __rte_unused)
+{
+	struct iavf_event_handler *handler = &event_handler;
+	TAILQ_HEAD(event_list, iavf_event_element) pending;
+
+	while (true) {
+		char unused[MAX_EVENT_PENDING];
+		ssize_t nr = read(handler->fd[0], &unused, sizeof(unused));
+		if (nr <= 0)
+			break;
+
+		TAILQ_INIT(&pending);
+		pthread_mutex_lock(&handler->lock);
+		TAILQ_CONCAT(&pending, &handler->pending, next);
+		pthread_mutex_unlock(&handler->lock);
+
+		struct iavf_event_element *pos, *save_next;
+		TAILQ_FOREACH_SAFE(pos, &pending, next, save_next) {
+			TAILQ_REMOVE(&pending, pos, next);
+			rte_eth_dev_callback_process(pos->dev, pos->event, pos->param);
+			rte_free(pos);
+		}
+	}
+
+	return NULL;
+}
+
+static void
+iavf_dev_event_post(struct rte_eth_dev *dev,
+		enum rte_eth_event_type event,
+		void *param, size_t param_alloc_size)
+{
+	struct iavf_event_handler *handler = &event_handler;
+	char notify_byte;
+	struct iavf_event_element *elem = rte_malloc(NULL, sizeof(*elem) + param_alloc_size, 0);
+	if (!elem)
+		return;
+
+	elem->dev = dev;
+	elem->event = event;
+	elem->param = param;
+	elem->param_alloc_size = param_alloc_size;
+	if (param && param_alloc_size) {
+		rte_memcpy(elem->param_alloc_data, param, param_alloc_size);
+		elem->param = elem->param_alloc_data;
+	}
+
+	pthread_mutex_lock(&handler->lock);
+	TAILQ_INSERT_TAIL(&handler->pending, elem, next);
+	pthread_mutex_unlock(&handler->lock);
+
+	ssize_t nw = write(handler->fd[1], &notify_byte, 1);
+	RTE_SET_USED(nw);
+}
+
+int
+iavf_dev_event_handler_init(void)
+{
+	struct iavf_event_handler *handler = &event_handler;
+
+	if (__atomic_add_fetch(&handler->ndev, 1, __ATOMIC_RELAXED) != 1)
+		return 0;
+#if defined(RTE_EXEC_ENV_IS_WINDOWS) && RTE_EXEC_ENV_IS_WINDOWS != 0
+	int err = _pipe(handler->fd, MAX_EVENT_PENDING, O_BINARY);
+#else
+	int err = pipe(handler->fd);
+#endif
+	if (err != 0) {
+		__atomic_sub_fetch(&handler->ndev, 1, __ATOMIC_RELAXED);
+		return -1;
+	}
+
+	TAILQ_INIT(&handler->pending);
+	pthread_mutex_init(&handler->lock, NULL);
+
+	if (rte_ctrl_thread_create(&handler->tid, "iavf-event-thread",
+				NULL, iavf_dev_event_handle, NULL)) {
+		__atomic_sub_fetch(&handler->ndev, 1, __ATOMIC_RELAXED);
+		return -1;
+	}
+
+	return 0;
+}
+
+void
+iavf_dev_event_handler_fini(void)
+{
+	struct iavf_event_handler *handler = &event_handler;
+
+	if (__atomic_sub_fetch(&handler->ndev, 1, __ATOMIC_RELAXED) != 0)
+		return;
+
+	int unused = pthread_cancel(handler->tid);
+	RTE_SET_USED(unused);
+	close(handler->fd[0]);
+	close(handler->fd[1]);
+	handler->fd[0] = -1;
+	handler->fd[1] = -1;
+
+	pthread_join(handler->tid, NULL);
+	pthread_mutex_destroy(&handler->lock);
+
+	struct iavf_event_element *pos, *save_next;
+	TAILQ_FOREACH_SAFE(pos, &handler->pending, next, save_next) {
+		TAILQ_REMOVE(&handler->pending, pos, next);
+		rte_free(pos);
+	}
+}
+
 static uint32_t
 iavf_convert_link_speed(enum virtchnl_link_speed virt_link_speed)
 {
@@ -278,8 +420,8 @@ iavf_handle_pf_event_msg(struct rte_eth_dev *dev, uint8_t *msg,
 	case VIRTCHNL_EVENT_RESET_IMPENDING:
 		PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_RESET_IMPENDING event");
 		vf->vf_reset = true;
-		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
-					      NULL);
+		iavf_dev_event_post(dev, RTE_ETH_EVENT_INTR_RESET,
+					      NULL, 0);
 		break;
 	case VIRTCHNL_EVENT_LINK_CHANGE:
 		PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_LINK_CHANGE event");
@@ -293,7 +435,7 @@ iavf_handle_pf_event_msg(struct rte_eth_dev *dev, uint8_t *msg,
 			vf->link_speed = iavf_convert_link_speed(speed);
 		}
 		iavf_dev_link_update(dev, 0);
-		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
+		iavf_dev_event_post(dev, RTE_ETH_EVENT_INTR_LSC, NULL, 0);
 		break;
 	case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
 		PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_PF_DRIVER_CLOSE event");
@@ -359,9 +501,8 @@ iavf_handle_virtchnl_msg(struct rte_eth_dev *dev)
 					desc.subtype =
 						RTE_ETH_EVENT_IPSEC_UNKNOWN;
 					desc.metadata = ev->ipsec_event_data;
-					rte_eth_dev_callback_process(dev,
-							RTE_ETH_EVENT_IPSEC,
-							&desc);
+					iavf_dev_event_post(dev, RTE_ETH_EVENT_IPSEC,
+							&desc, sizeof(desc));
 					return;
 				}