DPDK patches and discussions
 help / color / mirror / Atom feed
From: Cindy Lu <lulu@redhat.com>
To: lulu@redhat.com, jasowang@redhat.com, xieyongji@bytedance.com,
	dev@dpdk.org, maxime.coquelin@redhat.com
Subject: [RFC v2 1/2] vduse: add mapping process in vduse create and destroy
Date: Tue, 17 Oct 2023 22:24:02 +0800	[thread overview]
Message-ID: <20231017142403.2995341-2-lulu@redhat.com> (raw)
In-Reply-To: <20231017142403.2995341-1-lulu@redhat.com>

the change in the creation process is add
1> check if we need to do reconnect
2> Use ioctl get the reconnect info (size and max page number) from kernel
3> mapping 1 page for reconnect  status + vq_numbers pages for every vqs,
The change in the destroy process is the add the related unmap process

Signed-off-by: Cindy Lu <lulu@redhat.com>
---
 lib/vhost/vduse.c | 148 ++++++++++++++++++++++++++++++++++++----------
 lib/vhost/vhost.h |  10 ++++
 2 files changed, 127 insertions(+), 31 deletions(-)

diff --git a/lib/vhost/vduse.c b/lib/vhost/vduse.c
index 4f36277d3b..9b7f829a7a 100644
--- a/lib/vhost/vduse.c
+++ b/lib/vhost/vduse.c
@@ -376,6 +376,19 @@ vduse_device_create(const char *path)
 	uint64_t features = VDUSE_NET_SUPPORTED_FEATURES;
 	struct vduse_dev_config *dev_config = NULL;
 	const char *name = path + strlen("/dev/vduse/");
+	char reconnect_dev[PATH_MAX];
+	struct vhost_reconnect_data *log = NULL;
+	struct vduse_reconnect_mmap_info mmap_info;
+	bool reconnect = false;
+
+	ret = snprintf(reconnect_dev, sizeof(reconnect_dev), "%s/%s", "/dev/vduse", name);
+	if (access(reconnect_dev, F_OK) == 0) {
+		reconnect = true;
+		VHOST_LOG_CONFIG(name, INFO, "Device already exists, reconnecting...\n");
+	} else {
+		reconnect = false;
+		VHOST_LOG_CONFIG(name, ERR, "device %s not exists, creating...\n", reconnect_dev);
+	}
 
 	/* If first device, create events dispatcher thread */
 	if (vduse_events_thread == false) {
@@ -407,16 +420,8 @@ vduse_device_create(const char *path)
 	}
 
 	if (ioctl(control_fd, VDUSE_SET_API_VERSION, &ver)) {
-		VHOST_LOG_CONFIG(name, ERR, "Failed to set API version: %" PRIu64 ": %s\n",
-				ver, strerror(errno));
-		ret = -1;
-		goto out_ctrl_close;
-	}
-
-	dev_config = malloc(offsetof(struct vduse_dev_config, config) +
-			sizeof(vnet_config));
-	if (!dev_config) {
-		VHOST_LOG_CONFIG(name, ERR, "Failed to allocate VDUSE config\n");
+		VHOST_LOG_CONFIG(name, ERR, "Failed to set API version: %" PRIu64 ": %s\n", ver,
+				 strerror(errno));
 		ret = -1;
 		goto out_ctrl_close;
 	}
@@ -424,7 +429,7 @@ vduse_device_create(const char *path)
 	ret = rte_vhost_driver_get_queue_num(path, &max_queue_pairs);
 	if (ret < 0) {
 		VHOST_LOG_CONFIG(name, ERR, "Failed to get max queue pairs\n");
-		goto out_free;
+		goto out_ctrl_close;
 	}
 
 	VHOST_LOG_CONFIG(path, INFO, "VDUSE max queue pairs: %u\n", max_queue_pairs);
@@ -435,23 +440,34 @@ vduse_device_create(const char *path)
 	else
 		total_queues += 1; /* Includes ctrl queue */
 
-	vnet_config.max_virtqueue_pairs = max_queue_pairs;
-	memset(dev_config, 0, sizeof(struct vduse_dev_config));
-
-	strncpy(dev_config->name, name, VDUSE_NAME_MAX - 1);
-	dev_config->device_id = VIRTIO_ID_NET;
-	dev_config->vendor_id = 0;
-	dev_config->features = features;
-	dev_config->vq_num = total_queues;
-	dev_config->vq_align = sysconf(_SC_PAGE_SIZE);
-	dev_config->config_size = sizeof(struct virtio_net_config);
-	memcpy(dev_config->config, &vnet_config, sizeof(vnet_config));
+	if (reconnect != true) {
+		dev_config =
+			malloc(offsetof(struct vduse_dev_config, config) + sizeof(vnet_config));
+		if (!dev_config) {
+			VHOST_LOG_CONFIG(name, ERR, "Failed to allocate VDUSE config\n");
+			ret = -1;
+			goto out_ctrl_close;
+		}
 
-	ret = ioctl(control_fd, VDUSE_CREATE_DEV, dev_config);
-	if (ret < 0) {
-		VHOST_LOG_CONFIG(name, ERR, "Failed to create VDUSE device: %s\n",
-				strerror(errno));
-		goto out_free;
+		vnet_config.max_virtqueue_pairs = max_queue_pairs;
+		memset(dev_config, 0, sizeof(struct vduse_dev_config));
+
+		strncpy(dev_config->name, name, VDUSE_NAME_MAX - 1);
+		dev_config->device_id = VIRTIO_ID_NET;
+		dev_config->vendor_id = 0;
+		dev_config->features = features;
+		dev_config->vq_num = total_queues;
+		dev_config->vq_align = sysconf(_SC_PAGE_SIZE);
+		dev_config->config_size = sizeof(struct virtio_net_config);
+		memcpy(dev_config->config, &vnet_config, sizeof(vnet_config));
+
+		ret = ioctl(control_fd, VDUSE_CREATE_DEV, dev_config);
+		free(dev_config);
+		if (ret < 0) {
+			VHOST_LOG_CONFIG(name, ERR, "Failed to create VDUSE device: %s\n",
+					 strerror(errno));
+			goto out_ctrl_close;
+		}
 	}
 
 	dev_fd = open(path, O_RDWR);
@@ -485,10 +501,43 @@ vduse_device_create(const char *path)
 	strncpy(dev->ifname, path, IF_NAME_SZ - 1);
 	dev->vduse_ctrl_fd = control_fd;
 	dev->vduse_dev_fd = dev_fd;
+
+	ret = ioctl(dev_fd, VDUSE_GET_RECONNECT_INFO, &mmap_info);
+	if (ret < 0) {
+		VHOST_LOG_CONFIG(name, ERR, "Failed to get reconnect info VDUSE device: %s\n",
+				 strerror(errno));
+		goto out_dev_close;
+	}
+	dev->mmap_info.size = mmap_info.size;
+	dev->mmap_info.max_index = mmap_info.max_index;
+	log = (struct vhost_reconnect_data *)mmap(NULL, mmap_info.size, PROT_READ | PROT_WRITE,
+						  MAP_SHARED, dev->vduse_dev_fd, 0);
+	if (log == MAP_FAILED) {
+		VHOST_LOG_CONFIG(name, ERR, "Failed to mapping vduse reconnect data\n");
+		goto out_dev_close;
+	}
+
+	dev->log = log;
+
+	if (reconnect == true) {
+		dev->status = dev->log->status;
+		log->version = VHOST_VDUSE_API_VERSION;
+		log->reconnect_time += 1;
+		log->nr_vrings = total_queues;
+	}
+
 	vhost_setup_virtio_net(dev->vid, true, true, true, true);
 
+	if (total_queues > mmap_info.max_index - 1) {
+		VHOST_LOG_CONFIG(name, ERR, "The max vring number %d lager then %d\n", total_queues,
+				 mmap_info.max_index - 1);
+		goto out_dev_close;
+	}
+
 	for (i = 0; i < total_queues; i++) {
 		struct vduse_vq_config vq_cfg = { 0 };
+		struct vhost_reconnect_vring *log_vq;
+		struct vhost_virtqueue *vq;
 
 		ret = alloc_vring_queue(dev, i);
 		if (ret) {
@@ -496,6 +545,22 @@ vduse_device_create(const char *path)
 			goto out_dev_destroy;
 		}
 
+		log_vq = (struct vhost_reconnect_vring *)mmap(NULL, mmap_info.size,
+							      PROT_READ | PROT_WRITE, MAP_SHARED,
+							      dev->vduse_dev_fd,
+							      (i + 1) * mmap_info.size);
+		if (log_vq == MAP_FAILED) {
+			VHOST_LOG_CONFIG(name, ERR, "Failed to mapping vring %d reconnect data\n",
+					 i);
+
+			goto out_dev_destroy;
+		}
+
+		vq = dev->virtqueue[i];
+		vq->log = log_vq;
+		if (reconnect)
+			continue;
+
 		vq_cfg.index = i;
 		vq_cfg.max_size = 1024;
 
@@ -516,7 +581,8 @@ vduse_device_create(const char *path)
 	}
 	fdset_pipe_notify(&vduse.fdset);
 
-	free(dev_config);
+	if (reconnect && dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK)
+		vduse_device_start(dev, true);
 
 	return 0;
 
@@ -526,8 +592,6 @@ vduse_device_create(const char *path)
 	if (dev_fd >= 0)
 		close(dev_fd);
 	ioctl(control_fd, VDUSE_DESTROY_DEV, name);
-out_free:
-	free(dev_config);
 out_ctrl_close:
 	close(control_fd);
 
@@ -553,7 +617,29 @@ vduse_device_destroy(const char *path)
 
 	if (vid == RTE_MAX_VHOST_DEVICE)
 		return -1;
-
+	if (dev->log) {
+		for (uint32_t i = 0; i < dev->log->nr_vrings; i++) {
+			struct vhost_virtqueue *vq;
+
+			vq = dev->virtqueue[i];
+			if (vq->log) {
+				ret = munmap(vq->log, dev->mmap_info.size);
+				if (ret) {
+					VHOST_LOG_CONFIG(name, ERR,
+							 "Failed to unmap device %s vq %d: %s\n",
+							 path, i, strerror(errno));
+					ret = -1;
+				}
+			}
+		}
+		ret = munmap(dev->log, dev->mmap_info.size);
+		if (ret) {
+			VHOST_LOG_CONFIG(name, ERR, "Failed to unmap device %s dev status: %s\n",
+					 path, strerror(errno));
+			ret = -1;
+		}
+	}
+	dev->log = NULL;
 	if (dev->cvq && dev->cvq->kickfd >= 0) {
 		fdset_del(&vduse.fdset, dev->cvq->kickfd);
 		fdset_pipe_notify(&vduse.fdset);
diff --git a/lib/vhost/vhost.h b/lib/vhost/vhost.h
index c8f2a0d43a..1879c6875b 100644
--- a/lib/vhost/vhost.h
+++ b/lib/vhost/vhost.h
@@ -19,6 +19,7 @@
 #include <rte_ether.h>
 #include <rte_malloc.h>
 #include <rte_dmadev.h>
+#include <linux/vduse.h>
 
 #include "rte_vhost.h"
 #include "vdpa_driver.h"
@@ -344,6 +345,8 @@ struct vhost_virtqueue {
 
 	struct vhost_vring_addr ring_addrs;
 	struct virtqueue_stats	stats;
+
+	struct vhost_reconnect_vring *log;
 } __rte_cache_aligned;
 
 /* Virtio device status as per Virtio specification */
@@ -537,6 +540,9 @@ struct virtio_net {
 	struct rte_vhost_user_extern_ops extern_ops;
 
 	struct vhost_backend_ops *backend_ops;
+
+	struct vhost_reconnect_data *log;
+	struct vduse_reconnect_mmap_info mmap_info;
 } __rte_cache_aligned;
 
 static inline void
@@ -582,6 +588,10 @@ vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
 		vq->avail_wrap_counter ^= 1;
 		vq->last_avail_idx -= vq->size;
 	}
+	if (vq->log) {
+		vq->log->last_avail_idx = vq->last_avail_idx;
+		vq->log->avail_wrap_counter = vq->avail_wrap_counter;
+	}
 }
 
 void __vhost_log_cache_write(struct virtio_net *dev,
-- 
2.34.3


  reply	other threads:[~2023-10-17 14:24 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-10-17 14:24 [RFC v2 0/2] vduse: Add support for reconnection Cindy Lu
2023-10-17 14:24 ` Cindy Lu [this message]
2023-10-17 14:24 ` [RFC v2 2/2] vhost: add reconnection support to VDUSE (WIP) Cindy Lu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20231017142403.2995341-2-lulu@redhat.com \
    --to=lulu@redhat.com \
    --cc=dev@dpdk.org \
    --cc=jasowang@redhat.com \
    --cc=maxime.coquelin@redhat.com \
    --cc=xieyongji@bytedance.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).