[dpdk-dev] [PATCH v7 00/10] vhost: support inflight share memory protocol feature

[dpdk-dev] [PATCH v7 00/10] vhost: support inflight share memory protocol feature

[Jin Yu]

This patches introduces two new messages VHOST_USER_GET_INFLIGHT_FD
and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
buffer between qemu and backend.
Now It can both support split and packed ring. The example code show
how these API work. The test has passed.

How to test the example:
1, Qemu need two patches.
https://patchwork.kernel.org/patch/10766813/
https://patchwork.kernel.org/patch/10861411/(QEMU merged)
it also needs some manual modifications:
we should confirm that before we send get inflight we have already
sent the set features but it seems Qemu didn't do like this. So we
manually revise this, we can add below code in vhost_dev_get_inflight
    int r;

    r = vhost_dev_set_features(dev, dev->log_enabled);
    if (r < 0) {
        return -errno;
    }
before get_inflight_fd.
2, Guest OS version >= 5.0
3, run the example
4, run the qemu with vhost-user-blk-pci.
 eg:
        -chardev socket,id=spdk_vhost_blk0,reconnect=1,path=xxxx\
        -device vhost-user-blk-pci,ring_packed=on,chardev=spdk_vhost_blk0,num-queues=1\
5, run fio in the guest
6, kill the example and run again.
7, the fio in the guest should continue run without errors.

Jin Yu (10):
  vhost: add the inflight description
  vhost: add packed ring
  vhost: add the inflight structure
  vhost: add two new messages to support a shared buffer
  vhost: checkout the resubmit inflight information
  vhost: add the APIs to operate inflight ring
  vhost: add APIs for user getting inflight ring
  vhost: add vring functions packed ring support
  vhost: add an API for judging vq format
  vhost: add vhost-user-blk example which support inflight

 examples/vhost_blk/Makefile            |   67 ++
 examples/vhost_blk/blk.c               |  125 +++
 examples/vhost_blk/blk_spec.h          |   95 ++
 examples/vhost_blk/meson.build         |   20 +
 examples/vhost_blk/vhost_blk.c         | 1313 ++++++++++++++++++++++++
 examples/vhost_blk/vhost_blk.h         |  116 +++
 examples/vhost_blk/vhost_blk_compat.c  |  195 ++++
 lib/librte_vhost/rte_vhost.h           |  250 ++++-
 lib/librte_vhost/rte_vhost_version.map |    9 +
 lib/librte_vhost/vhost.c               |  416 +++++++-
 lib/librte_vhost/vhost.h               |   16 +
 lib/librte_vhost/vhost_user.c          |  470 ++++++++-
 lib/librte_vhost/vhost_user.h          |   12 +-
 13 files changed, 3091 insertions(+), 13 deletions(-)
 create mode 100644 examples/vhost_blk/Makefile
 create mode 100644 examples/vhost_blk/blk.c
 create mode 100644 examples/vhost_blk/blk_spec.h
 create mode 100644 examples/vhost_blk/meson.build
 create mode 100644 examples/vhost_blk/vhost_blk.c
 create mode 100644 examples/vhost_blk/vhost_blk.h
 create mode 100644 examples/vhost_blk/vhost_blk_compat.c

-- 
2.17.2

[dpdk-dev] [PATCH v7 01/10] vhost: add the inflight description

[Jin Yu]

This patch add the inflight message description and
the inflight share fd protocol feature flag.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
v1 - specify the APIs are split-ring only
v2 - fix APIs and judge split or packed
v3 - add rte_vhost_ prefix and fix one issue
v4 - add the packed ring support
v5 - revise get_vring_base func depend on Tiwei's suggestion
v6 - divide patch into small patches
v7 - updated base on Maxime's comments
---
 lib/librte_vhost/rte_vhost.h  | 4 ++++
 lib/librte_vhost/vhost_user.h | 8 ++++++++
 2 files changed, 12 insertions(+)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 0226b3eff..9943575ce 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -71,6 +71,10 @@ extern "C" {
 #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER 11
 #endif
 
+#ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
+#define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
+#endif
+
 /** Indicate whether protocol features negotiation is supported. */
 #ifndef VHOST_USER_F_PROTOCOL_FEATURES
 #define VHOST_USER_F_PROTOCOL_FEATURES	30
diff --git a/lib/librte_vhost/vhost_user.h b/lib/librte_vhost/vhost_user.h
index 2a650fe4b..17a1d7bca 100644
--- a/lib/librte_vhost/vhost_user.h
+++ b/lib/librte_vhost/vhost_user.h
@@ -112,6 +112,13 @@ typedef struct VhostUserVringArea {
 	uint64_t offset;
 } VhostUserVringArea;
 
+typedef struct VhostUserInflight {
+	uint64_t mmap_size;
+	uint64_t mmap_offset;
+	uint16_t num_queues;
+	uint16_t queue_size;
+} VhostUserInflight;
+
 typedef struct VhostUserMsg {
 	union {
 		uint32_t master; /* a VhostUserRequest value */
@@ -134,6 +141,7 @@ typedef struct VhostUserMsg {
 		struct vhost_iotlb_msg iotlb;
 		VhostUserCryptoSessionParam crypto_session;
 		VhostUserVringArea area;
+		VhostUserInflight inflight;
 	} payload;
 	int fds[VHOST_MEMORY_MAX_NREGIONS];
 	int fd_num;
-- 
2.17.2

[dpdk-dev] [PATCH v7 02/10] vhost: add packed ring

[Jin Yu]

This patch add the packed ring in the rte_vhost_vring.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/rte_vhost.h | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 9943575ce..7257f0965 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -103,9 +103,18 @@ struct rte_vhost_memory {
 };
 
 struct rte_vhost_vring {
-	struct vring_desc	*desc;
-	struct vring_avail	*avail;
-	struct vring_used	*used;
+	union {
+		struct vring_desc *desc;
+		struct vring_packed_desc *desc_packed;
+	};
+	union {
+		struct vring_avail *avail;
+		struct vring_packed_desc_event *driver_event;
+	};
+	union {
+		struct vring_used *used;
+		struct vring_packed_desc_event *device_event;
+	};
 	uint64_t		log_guest_addr;
 
 	/** Deprecated, use rte_vhost_vring_call() instead. */
-- 
2.17.2

[dpdk-dev] [PATCH v7 03/10] vhost: add the inflight structure

[Jin Yu]

This patch adds the inflight queue region structure include
the split and packed.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/rte_vhost.h | 43 ++++++++++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 7257f0965..644b3e0d7 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -102,6 +102,49 @@ struct rte_vhost_memory {
 	struct rte_vhost_mem_region regions[];
 };
 
+struct rte_vhost_inflight_desc_split {
+	uint8_t inflight;
+	uint8_t padding[5];
+	uint16_t next;
+	uint64_t counter;
+};
+
+struct rte_vhost_inflight_info_split {
+	uint64_t features;
+	uint16_t version;
+	uint16_t desc_num;
+	uint16_t last_inflight_io;
+	uint16_t used_idx;
+	struct rte_vhost_inflight_desc_split desc[0];
+};
+
+struct rte_vhost_inflight_desc_packed {
+	uint8_t inflight;
+	uint8_t padding;
+	uint16_t next;
+	uint16_t last;
+	uint16_t num;
+	uint64_t counter;
+	uint16_t id;
+	uint16_t flags;
+	uint32_t len;
+	uint64_t addr;
+	};
+
+struct rte_vhost_inflight_info_packed {
+	uint64_t features;
+	uint16_t version;
+	uint16_t desc_num;
+	uint16_t free_head;
+	uint16_t old_free_head;
+	uint16_t used_idx;
+	uint16_t old_used_idx;
+	uint8_t used_wrap_counter;
+	uint8_t old_used_wrap_counter;
+	uint8_t padding[7];
+	struct rte_vhost_inflight_desc_packed desc[0];
+};
+
 struct rte_vhost_vring {
 	union {
 		struct vring_desc *desc;
-- 
2.17.2

[dpdk-dev] [PATCH v7 04/10] vhost: add two new messages to support a shared buffer

[Jin Yu]

This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD
and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
buffer between qemu and backend.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/vhost.h      |   7 +
 lib/librte_vhost/vhost_user.c | 260 +++++++++++++++++++++++++++++++++-
 lib/librte_vhost/vhost_user.h |   4 +-
 3 files changed, 269 insertions(+), 2 deletions(-)

diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index 884befa85..d67ba849a 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -286,6 +286,12 @@ struct guest_page {
 	uint64_t size;
 };
 
+struct inflight_mem_info {
+	int		fd;
+	void		*addr;
+	uint64_t	size;
+};
+
 /**
  * Device structure contains all configuration information relating
  * to the device.
@@ -303,6 +309,7 @@ struct virtio_net {
 	uint32_t		nr_vring;
 	int			dequeue_zero_copy;
 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
+	struct inflight_mem_info *inflight_info;
 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 	char			ifname[IF_NAME_SZ];
 	uint64_t		log_size;
diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
index c9e29ece8..66582a30f 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -37,6 +37,10 @@
 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
 #include <linux/userfaultfd.h>
 #endif
+#ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
+#include <linux/memfd.h>
+#define MEMFD_SUPPORTED
+#endif
 
 #include <rte_common.h>
 #include <rte_malloc.h>
@@ -49,6 +53,15 @@
 #define VIRTIO_MIN_MTU 68
 #define VIRTIO_MAX_MTU 65535
 
+#define INFLIGHT_ALIGNMENT	64
+#define INFLIGHT_VERSION	0xabcd
+#define VIRTQUEUE_MAX_SIZE	1024
+
+#define CLOEXEC		0x0001U
+
+#define ALIGN_DOWN(n, m) ((n) / (m) * (m))
+#define ALIGN_UP(n, m) ALIGN_DOWN((n) + (m) - 1, (m))
+
 static const char *vhost_message_str[VHOST_USER_MAX] = {
 	[VHOST_USER_NONE] = "VHOST_USER_NONE",
 	[VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
@@ -78,6 +91,8 @@ static const char *vhost_message_str[VHOST_USER_MAX] = {
 	[VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
 	[VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
 	[VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
+	[VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
+	[VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
 };
 
 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
@@ -160,6 +175,21 @@ vhost_backend_cleanup(struct virtio_net *dev)
 		dev->log_addr = 0;
 	}
 
+	if (dev->inflight_info->addr) {
+		munmap(dev->inflight_info->addr, dev->inflight_info->size);
+		dev->inflight_info->addr = NULL;
+	}
+
+	if (dev->inflight_info->fd > 0) {
+		close(dev->inflight_info->fd);
+		dev->inflight_info->fd = -1;
+	}
+
+	if (dev->inflight_info) {
+		free(dev->inflight_info);
+		dev->inflight_info = NULL;
+	}
+
 	if (dev->slave_req_fd >= 0) {
 		close(dev->slave_req_fd);
 		dev->slave_req_fd = -1;
@@ -1165,6 +1195,233 @@ virtio_is_ready(struct virtio_net *dev)
 	return 1;
 }
 
+static int
+mem_create(const char *name, unsigned int flags)
+{
+#ifdef MEMFD_SUPPORTED
+	return memfd_create(name, flags);
+#else
+	RTE_LOG(ERR, VHOST_CONFIG,
+		"doesn't support memfd--name:%s and flag:%x\n",
+		name, flags);
+	return -1;
+#endif
+}
+
+static void *
+inflight_mem_alloc(const char *name, size_t size, int *fd)
+{
+	void *ptr;
+	int mfd = -1;
+	char fname[20] = "/tmp/memfd-XXXXXX";
+
+	*fd = -1;
+	mfd = mem_create(name, CLOEXEC);
+	if (mfd != -1) {
+		if (ftruncate(mfd, size) == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"ftruncate fail for alloc inflight buffer\n");
+			close(mfd);
+			return NULL;
+		}
+	} else {
+		mfd = mkstemp(fname);
+		unlink(fname);
+
+		if (mfd == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"mkstemp fail for alloc inflight buffer\n");
+			return NULL;
+		}
+
+		if (ftruncate(mfd, size) == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"ftruncate fail for alloc inflight buffer\n");
+			close(mfd);
+			return NULL;
+		}
+	}
+
+	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
+	if (ptr == MAP_FAILED) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"mmap fail for alloc inflight buffer\n");
+		close(mfd);
+		return NULL;
+	}
+
+	*fd = mfd;
+	return ptr;
+}
+
+static uint32_t
+get_pervq_shm_size_split(uint16_t queue_size)
+{
+	return ALIGN_UP(sizeof(struct rte_vhost_inflight_desc_split) *
+			queue_size + sizeof(uint64_t) + sizeof(uint16_t) * 4,
+			INFLIGHT_ALIGNMENT);
+}
+
+static uint32_t
+get_pervq_shm_size_packed(uint16_t queue_size)
+{
+	return ALIGN_UP(sizeof(struct rte_vhost_inflight_desc_packed) *
+			queue_size + sizeof(uint64_t) + sizeof(uint16_t) * 6 +
+			sizeof(uint8_t) * 9, INFLIGHT_ALIGNMENT);
+}
+
+static int
+vhost_user_get_inflight_fd(struct virtio_net **pdev,
+				     VhostUserMsg *msg,
+				     int main_fd __rte_unused)
+{
+	int fd, i, j;
+	uint64_t pervq_inflight_size, mmap_size;
+	void *addr;
+	uint16_t num_queues, queue_size;
+	struct virtio_net *dev = *pdev;
+	struct rte_vhost_inflight_info_packed *inflight_packed = NULL;
+
+	if (msg->size != sizeof(msg->payload.inflight)) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Invalid get_inflight_fd message size is %d",
+			msg->size);
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	dev->inflight_info = calloc(1, sizeof(struct inflight_mem_info));
+	if (!dev->inflight_info) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Failed to alloc dev inflight area");
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	num_queues = msg->payload.inflight.num_queues;
+	queue_size = msg->payload.inflight.queue_size;
+
+	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
+		msg->payload.inflight.num_queues);
+	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
+		msg->payload.inflight.queue_size);
+
+	if (vq_is_packed(dev))
+		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
+	else
+		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
+
+	mmap_size = num_queues * pervq_inflight_size;
+	addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
+	if (!addr) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Failed to alloc vhost inflight area");
+			msg->payload.inflight.mmap_size = 0;
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+	memset(addr, 0, mmap_size);
+
+	dev->inflight_info->addr = addr;
+	dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
+	dev->inflight_info->fd = msg->fds[0] = fd;
+	msg->payload.inflight.mmap_offset = 0;
+	msg->fd_num = 1;
+
+	if (vq_is_packed(dev)) {
+		for (i = 0; i < num_queues; i++) {
+			inflight_packed =
+				(struct rte_vhost_inflight_info_packed *)addr;
+			inflight_packed->used_wrap_counter = 1;
+			inflight_packed->old_used_wrap_counter = 1;
+			for (j = 0; j < queue_size; j++)
+				inflight_packed->desc[j].next = j + 1;
+			addr = (void *)((char *)addr + pervq_inflight_size);
+		}
+	}
+
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"send inflight mmap_size: %"PRIu64"\n",
+		msg->payload.inflight.mmap_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"send inflight mmap_offset: %"PRIu64"\n",
+		msg->payload.inflight.mmap_offset);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"send inflight fd: %d\n", msg->fds[0]);
+
+	return RTE_VHOST_MSG_RESULT_REPLY;
+}
+
+static int
+vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
+		int main_fd __rte_unused)
+{
+	int fd;
+	uint64_t mmap_size, mmap_offset;
+	uint16_t num_queues, queue_size;
+	uint32_t pervq_inflight_size;
+	void *addr;
+	struct virtio_net *dev = *pdev;
+
+	fd = msg->fds[0];
+	if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Invalid set_inflight_fd message size is %d,fd is %d\n",
+			msg->size, fd);
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	mmap_size = msg->payload.inflight.mmap_size;
+	mmap_offset = msg->payload.inflight.mmap_offset;
+	num_queues = msg->payload.inflight.num_queues;
+	queue_size = msg->payload.inflight.queue_size;
+
+	if (vq_is_packed(dev))
+		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
+	else
+		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
+
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd num_queues: %u\n", num_queues);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd queue_size: %u\n", queue_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd fd: %d\n", fd);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd pervq_inflight_size: %d\n",
+		pervq_inflight_size);
+
+	if (!dev->inflight_info) {
+		dev->inflight_info = calloc(1,
+					    sizeof(struct inflight_mem_info));
+		if (dev->inflight_info == NULL) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"Failed to alloc dev inflight area");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+	}
+
+	if (dev->inflight_info->addr)
+		munmap(dev->inflight_info->addr, dev->inflight_info->size);
+
+	addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
+			fd, mmap_offset);
+	if (addr == MAP_FAILED) {
+		RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	if (dev->inflight_info->fd)
+		close(dev->inflight_info->fd);
+
+	dev->inflight_info->fd = fd;
+	dev->inflight_info->addr = addr;
+	dev->inflight_info->size = mmap_size;
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
 static int
 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
 			int main_fd __rte_unused)
@@ -1762,9 +2019,10 @@ static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
 	[VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
 	[VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
 	[VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
+	[VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
+	[VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
 };
 
-
 /* return bytes# of read on success or negative val on failure. */
 static int
 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
diff --git a/lib/librte_vhost/vhost_user.h b/lib/librte_vhost/vhost_user.h
index 17a1d7bca..6563f7315 100644
--- a/lib/librte_vhost/vhost_user.h
+++ b/lib/librte_vhost/vhost_user.h
@@ -54,7 +54,9 @@ typedef enum VhostUserRequest {
 	VHOST_USER_POSTCOPY_ADVISE = 28,
 	VHOST_USER_POSTCOPY_LISTEN = 29,
 	VHOST_USER_POSTCOPY_END = 30,
-	VHOST_USER_MAX = 31
+	VHOST_USER_GET_INFLIGHT_FD = 31,
+	VHOST_USER_SET_INFLIGHT_FD = 32,
+	VHOST_USER_MAX = 33
 } VhostUserRequest;
 
 typedef enum VhostUserSlaveRequest {
-- 
2.17.2

[dpdk-dev] [PATCH v7 05/10] vhost: checkout the resubmit inflight information

[Jin Yu]

This patch shows how to checkout the inflight ring and construct
the resubmit information also include destroying resubmit info.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/rte_vhost.h  |  19 +++
 lib/librte_vhost/vhost.c      |  29 ++++-
 lib/librte_vhost/vhost.h      |   9 ++
 lib/librte_vhost/vhost_user.c | 212 +++++++++++++++++++++++++++++++++-
 4 files changed, 267 insertions(+), 2 deletions(-)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 644b3e0d7..6ac738730 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -145,6 +145,25 @@ struct rte_vhost_inflight_info_packed {
 	struct rte_vhost_inflight_desc_packed desc[0];
 };
 
+struct rte_vhost_resubmit_desc {
+	uint16_t index;
+	uint64_t counter;
+};
+
+struct rte_vhost_resubmit_info {
+	struct rte_vhost_resubmit_desc *resubmit_list;
+	uint16_t resubmit_num;
+};
+
+struct rte_vhost_ring_inflight {
+	union {
+		struct rte_vhost_inflight_info_split *inflight_split;
+		struct rte_vhost_inflight_info_packed *inflight_packed;
+	};
+
+	struct rte_vhost_resubmit_info *resubmit_inflight;
+};
+
 struct rte_vhost_vring {
 	union {
 		struct vring_desc *desc;
diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 981837b5d..660ac2a07 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -242,6 +242,31 @@ cleanup_vq(struct vhost_virtqueue *vq, int destroy)
 		close(vq->kickfd);
 }
 
+void
+cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+	if (!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
+		return;
+
+	if (vq_is_packed(dev)) {
+		if (vq->inflight_packed)
+			vq->inflight_packed = NULL;
+	} else {
+		if (vq->inflight_split)
+			vq->inflight_split = NULL;
+	}
+
+	if (vq->resubmit_inflight) {
+		if (vq->resubmit_inflight->resubmit_list) {
+			free(vq->resubmit_inflight->resubmit_list);
+			vq->resubmit_inflight->resubmit_list = NULL;
+		}
+		free(vq->resubmit_inflight);
+		vq->resubmit_inflight = NULL;
+	}
+}
+
 /*
  * Unmap any memory, close any file descriptors and
  * free any memory owned by a device.
@@ -253,8 +278,10 @@ cleanup_device(struct virtio_net *dev, int destroy)
 
 	vhost_backend_cleanup(dev);
 
-	for (i = 0; i < dev->nr_vring; i++)
+	for (i = 0; i < dev->nr_vring; i++) {
 		cleanup_vq(dev->virtqueue[i], destroy);
+		cleanup_vq_inflight(dev, dev->virtqueue[i]);
+	}
 }
 
 void
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index d67ba849a..ab95999c4 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -128,6 +128,14 @@ struct vhost_virtqueue {
 	/* Physical address of used ring, for logging */
 	uint64_t		log_guest_addr;
 
+	/* inflight share memory info */
+	union {
+		struct rte_vhost_inflight_info_split *inflight_split;
+		struct rte_vhost_inflight_info_packed *inflight_packed;
+	};
+	struct rte_vhost_resubmit_info *resubmit_inflight;
+	uint64_t		global_counter;
+
 	uint16_t		nr_zmbuf;
 	uint16_t		zmbuf_size;
 	uint16_t		last_zmbuf_idx;
@@ -474,6 +482,7 @@ void vhost_destroy_device(int);
 void vhost_destroy_device_notify(struct virtio_net *dev);
 
 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
+void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 
 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
index 66582a30f..e3a7aee51 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -336,6 +336,7 @@ vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 
 			dev->virtqueue[dev->nr_vring] = NULL;
 			cleanup_vq(vq, 1);
+			cleanup_vq_inflight(dev, vq);
 			free_vq(dev, vq);
 		}
 	}
@@ -1353,11 +1354,12 @@ static int
 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
 		int main_fd __rte_unused)
 {
-	int fd;
+	int fd, i;
 	uint64_t mmap_size, mmap_offset;
 	uint16_t num_queues, queue_size;
 	uint32_t pervq_inflight_size;
 	void *addr;
+	struct vhost_virtqueue *vq;
 	struct virtio_net *dev = *pdev;
 
 	fd = msg->fds[0];
@@ -1419,6 +1421,18 @@ vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
 	dev->inflight_info->addr = addr;
 	dev->inflight_info->size = mmap_size;
 
+	for (i = 0; i < num_queues; i++) {
+		vq = dev->virtqueue[i];
+		if (vq_is_packed(dev)) {
+			vq->inflight_packed = addr;
+			vq->inflight_packed->desc_num = queue_size;
+		} else {
+			vq->inflight_split = addr;
+			vq->inflight_split->desc_num = queue_size;
+		}
+		addr = (void *)((char *)addr + pervq_inflight_size);
+	}
+
 	return RTE_VHOST_MSG_RESULT_OK;
 }
 
@@ -1458,6 +1472,188 @@ static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
 	return RTE_VHOST_MSG_RESULT_OK;
 }
 
+static int
+resubmit_desc_compare(const void *a, const void *b)
+{
+	const struct rte_vhost_resubmit_desc *desc0 =
+		(const struct rte_vhost_resubmit_desc *)a;
+	const struct rte_vhost_resubmit_desc *desc1 =
+		(const struct rte_vhost_resubmit_desc *)b;
+
+	if (desc1->counter > desc0->counter &&
+		(desc1->counter - desc0->counter) < VIRTQUEUE_MAX_SIZE * 2)
+		return 1;
+
+	return -1;
+}
+
+static int
+vhost_check_queue_inflights_split(struct virtio_net *dev,
+	struct vhost_virtqueue *vq)
+{
+	uint16_t i = 0;
+	uint16_t resubmit_num = 0, last_io, num;
+	struct vring_used *used = vq->used;
+	struct rte_vhost_resubmit_info *resubmit = NULL;
+	struct rte_vhost_inflight_info_split *inflight_split;
+
+	if (!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
+		return RTE_VHOST_MSG_RESULT_OK;
+
+	if ((!vq->inflight_split))
+		return RTE_VHOST_MSG_RESULT_ERR;
+
+	if (!vq->inflight_split->version) {
+		vq->inflight_split->version = INFLIGHT_VERSION;
+		return RTE_VHOST_MSG_RESULT_OK;
+	}
+
+	inflight_split = vq->inflight_split;
+	vq->resubmit_inflight = NULL;
+	vq->global_counter = 0;
+	last_io = inflight_split->last_inflight_io;
+
+	if (inflight_split->used_idx != used->idx) {
+		inflight_split->desc[last_io].inflight = 0;
+		rte_compiler_barrier();
+		inflight_split->used_idx = used->idx;
+	}
+
+	for (i = 0; i < inflight_split->desc_num; i++) {
+		if (inflight_split->desc[i].inflight == 1)
+			resubmit_num++;
+	}
+
+	vq->last_avail_idx += resubmit_num;
+
+	if (resubmit_num) {
+		resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
+		if (!resubmit) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for resubmit info.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		resubmit->resubmit_list = calloc(resubmit_num,
+			sizeof(struct rte_vhost_resubmit_desc));
+		if (!resubmit->resubmit_list) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for inflight desc.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		num = 0;
+		for (i = 0; i < vq->inflight_split->desc_num; i++) {
+			if (vq->inflight_split->desc[i].inflight == 1) {
+				resubmit->resubmit_list[num].index = i;
+				resubmit->resubmit_list[num].counter =
+					inflight_split->desc[i].counter;
+				num++;
+			}
+		}
+		resubmit->resubmit_num = num;
+
+		if (resubmit->resubmit_num > 1)
+			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
+				sizeof(struct rte_vhost_resubmit_desc),
+				resubmit_desc_compare);
+
+		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
+		vq->resubmit_inflight = resubmit;
+	}
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
+static int
+vhost_check_queue_inflights_packed(struct virtio_net *dev,
+						 struct vhost_virtqueue *vq)
+{
+	uint16_t i = 0;
+	uint16_t resubmit_num = 0, old_used_idx, num;
+	struct rte_vhost_resubmit_info *resubmit = NULL;
+	struct rte_vhost_inflight_info_packed *inflight_packed;
+
+	if (!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
+		return RTE_VHOST_MSG_RESULT_OK;
+
+	if ((!vq->inflight_packed))
+		return RTE_VHOST_MSG_RESULT_ERR;
+
+	if (!vq->inflight_packed->version) {
+		vq->inflight_packed->version = INFLIGHT_VERSION;
+		return RTE_VHOST_MSG_RESULT_OK;
+	}
+
+	inflight_packed = vq->inflight_packed;
+	vq->resubmit_inflight = NULL;
+	vq->global_counter = 0;
+	old_used_idx = inflight_packed->old_used_idx;
+
+	if (inflight_packed->used_idx != old_used_idx) {
+		if (inflight_packed->desc[old_used_idx].inflight == 0) {
+			inflight_packed->old_used_idx =
+				inflight_packed->used_idx;
+			inflight_packed->old_used_wrap_counter =
+				inflight_packed->used_wrap_counter;
+			inflight_packed->old_free_head =
+				inflight_packed->free_head;
+		} else {
+			inflight_packed->used_idx =
+				inflight_packed->old_used_idx;
+			inflight_packed->used_wrap_counter =
+				inflight_packed->old_used_wrap_counter;
+			inflight_packed->free_head =
+				inflight_packed->old_free_head;
+		}
+	}
+
+	for (i = 0; i < inflight_packed->desc_num; i++) {
+		if (inflight_packed->desc[i].inflight == 1)
+			resubmit_num++;
+	}
+
+	if (resubmit_num) {
+		resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
+		if (resubmit == NULL) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for resubmit info.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		resubmit->resubmit_list = calloc(resubmit_num,
+			sizeof(struct rte_vhost_resubmit_desc));
+		if (resubmit->resubmit_list == NULL) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for resubmit desc.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		num = 0;
+		for (i = 0; i < inflight_packed->desc_num; i++) {
+			if (vq->inflight_packed->desc[i].inflight == 1) {
+				resubmit->resubmit_list[num].index = i;
+				resubmit->resubmit_list[num].counter =
+					inflight_packed->desc[i].counter;
+				num++;
+			}
+		}
+		resubmit->resubmit_num = num;
+
+		if (resubmit->resubmit_num > 1)
+			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
+				sizeof(struct rte_vhost_resubmit_desc),
+				resubmit_desc_compare);
+
+		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
+		vq->resubmit_inflight = resubmit;
+	}
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
 static int
 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
 			int main_fd __rte_unused)
@@ -1499,6 +1695,20 @@ vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
 		close(vq->kickfd);
 	vq->kickfd = file.fd;
 
+	if (vq_is_packed(dev)) {
+		if (vhost_check_queue_inflights_packed(dev, vq)) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"Failed to inflights for vq: %d\n", file.index);
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+	} else {
+		if (vhost_check_queue_inflights_split(dev, vq)) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"Failed to inflights for vq: %d\n", file.index);
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+	}
+
 	return RTE_VHOST_MSG_RESULT_OK;
 }
 
-- 
2.17.2

[dpdk-dev] [PATCH v7 06/10] vhost: add the APIs to operate inflight ring

[Jin Yu]

This patch introduces three APIs to operate the inflight
ring. Three APIs are set, set last and clear. It includes
split and packed ring.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/rte_vhost.h           | 116 +++++++++++
 lib/librte_vhost/rte_vhost_version.map |   6 +
 lib/librte_vhost/vhost.c               | 272 +++++++++++++++++++++++++
 3 files changed, 394 insertions(+)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 6ac738730..f0e86f5e2 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -693,6 +693,122 @@ int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
 int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 			      struct rte_vhost_vring *vring);
 
+/**
+ * Set split inflight descriptor.
+ *
+ * This function save descriptors that has been comsumed in available
+ * ring
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param idx
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t idx);
+
+/**
+ * Set packed inflight descriptor and get corresponding inflight entry
+ *
+ * This function save descriptors that has been comsumed
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param head
+ *  head of descriptors
+ * @param last
+ *  last of descriptors
+ * @param inflight_entry
+ *  corresponding inflight entry
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head, uint16_t last, uint16_t *inflight_entry);
+
+/**
+ * Save the head of list that the last batch of used descriptors.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param idx
+ *  descriptor entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_set_last_inflight_io_split(int vid,
+	uint16_t vring_idx, uint16_t idx);
+
+/**
+ * Update the inflight free_head, used_idx and used_wrap_counter.
+ *
+ * This function will update status first before updating descriptors
+ * to used
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param head
+ *  head of descriptors
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_set_last_inflight_io_packed(int vid,
+	uint16_t vring_idx, uint16_t head);
+
+/**
+ * Clear the split inflight status.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param last_used_idx
+ *  last used idx of used ring
+ * @param idx
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t last_used_idx, uint16_t idx);
+
+/**
+ * Clear the packed inflight status.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param head
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head);
+
 /**
  * Notify the guest that used descriptors have been added to the vring.  This
  * function acts as a memory barrier.
diff --git a/lib/librte_vhost/rte_vhost_version.map b/lib/librte_vhost/rte_vhost_version.map
index 5f1d4a75c..bc70bfaa5 100644
--- a/lib/librte_vhost/rte_vhost_version.map
+++ b/lib/librte_vhost/rte_vhost_version.map
@@ -87,4 +87,10 @@ EXPERIMENTAL {
 	rte_vdpa_relay_vring_used;
 	rte_vhost_extern_callback_register;
 	rte_vhost_driver_set_protocol_features;
+	rte_vhost_set_inflight_desc_split;
+	rte_vhost_set_inflight_desc_packed;
+	rte_vhost_set_last_inflight_io_split;
+	rte_vhost_set_last_inflight_io_packed;
+	rte_vhost_clr_inflight_desc_split;
+	rte_vhost_clr_inflight_desc_packed;
 };
diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 660ac2a07..44ae87d4f 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -783,6 +783,278 @@ rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 	return 0;
 }
 
+int
+rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (unlikely(!vq->inflight_split))
+		return -1;
+
+	if (unlikely(idx >= vq->size))
+		return -1;
+
+	vq->inflight_split->desc[idx].counter = vq->global_counter++;
+	vq->inflight_split->desc[idx].inflight = 1;
+	return 0;
+}
+
+int
+rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head, uint16_t last, uint16_t *inflight_entry)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+	struct rte_vhost_inflight_info_packed *inflight_info;
+	struct vring_packed_desc *desc;
+	uint16_t old_free_head, free_head;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(!vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	inflight_info = vq->inflight_packed;
+	if (unlikely(!inflight_info))
+		return -1;
+
+	if (unlikely(head >= vq->size))
+		return -1;
+
+	desc = vq->desc_packed;
+	old_free_head = inflight_info->old_free_head;
+	if (unlikely(old_free_head >= vq->size))
+		return -1;
+
+	free_head = old_free_head;
+
+	/* init header descriptor */
+	inflight_info->desc[old_free_head].num = 0;
+	inflight_info->desc[old_free_head].counter = vq->global_counter++;
+	inflight_info->desc[old_free_head].inflight = 1;
+
+	/* save desc entry in flight entry */
+	while (head != ((last + 1) % vq->size)) {
+		inflight_info->desc[old_free_head].num++;
+		inflight_info->desc[free_head].addr = desc[head].addr;
+		inflight_info->desc[free_head].len = desc[head].len;
+		inflight_info->desc[free_head].flags = desc[head].flags;
+		inflight_info->desc[free_head].id = desc[head].id;
+
+		inflight_info->desc[old_free_head].last = free_head;
+		free_head = inflight_info->desc[free_head].next;
+		inflight_info->free_head = free_head;
+		head = (head + 1) % vq->size;
+	}
+
+	inflight_info->old_free_head = free_head;
+	*inflight_entry = old_free_head;
+
+	return 0;
+}
+
+int
+rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t last_used_idx, uint16_t idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (unlikely(!vq->inflight_split))
+		return -1;
+
+	if (unlikely(idx >= vq->size))
+		return -1;
+
+	rte_compiler_barrier();
+
+	vq->inflight_split->desc[idx].inflight = 0;
+
+	rte_compiler_barrier();
+
+	vq->inflight_split->used_idx = last_used_idx;
+	return 0;
+}
+
+int
+rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+	struct rte_vhost_inflight_info_packed *inflight_info;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(!vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	inflight_info = vq->inflight_packed;
+	if (unlikely(!inflight_info))
+		return -1;
+
+	if (unlikely(head >= vq->size))
+		return -1;
+
+	rte_compiler_barrier();
+
+	inflight_info->desc[head].inflight = 0;
+
+	rte_compiler_barrier();
+
+	inflight_info->old_free_head = inflight_info->free_head;
+	inflight_info->old_used_idx = inflight_info->used_idx;
+	inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
+
+	return 0;
+}
+
+int
+rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
+	uint16_t idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (unlikely(!vq->inflight_split))
+		return -1;
+
+	vq->inflight_split->last_inflight_io = idx;
+	return 0;
+}
+
+int
+rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
+	uint16_t head)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+	struct rte_vhost_inflight_info_packed *inflight_info;
+	uint16_t last;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(!vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	inflight_info = vq->inflight_packed;
+	if (unlikely(!inflight_info))
+		return -1;
+
+	if (unlikely(head >= vq->size))
+		return -1;
+
+	last = inflight_info->desc[head].last;
+	if (unlikely(last >= vq->size))
+		return -1;
+
+	inflight_info->desc[last].next = inflight_info->free_head;
+	inflight_info->free_head = head;
+	inflight_info->used_idx += inflight_info->desc[head].num;
+	if (inflight_info->used_idx >= inflight_info->desc_num) {
+		inflight_info->used_idx -= inflight_info->desc_num;
+		inflight_info->used_wrap_counter =
+			!inflight_info->used_wrap_counter;
+	}
+
+	return 0;
+}
+
 int
 rte_vhost_vring_call(int vid, uint16_t vring_idx)
 {
-- 
2.17.2

[dpdk-dev] [PATCH v7 07/10] vhost: add APIs for user getting inflight ring

[Jin Yu]

This patch introduces two APIs. one is for getting inflgiht
ring and the other is for getting base.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/rte_vhost.h           | 40 +++++++++++++++++
 lib/librte_vhost/rte_vhost_version.map |  2 +
 lib/librte_vhost/vhost.c               | 59 ++++++++++++++++++++++++++
 3 files changed, 101 insertions(+)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index f0e86f5e2..8b6435d0d 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -693,6 +693,23 @@ int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
 int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 			      struct rte_vhost_vring *vring);
 
+/**
+ * Get guest inflight vring info, including inflight ring and resubmit list.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param vring
+ *  the structure to hold the requested inflight vring info
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
+	struct rte_vhost_ring_inflight *vring);
+
 /**
  * Set split inflight descriptor.
  *
@@ -867,6 +884,29 @@ int __rte_experimental
 rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 		uint16_t *last_avail_idx, uint16_t *last_used_idx);
 
+/**
+ * Get last_avail/last_used of the vhost virtqueue
+ *
+ * This function is designed for the reconnection and it's specific for
+ * the packed ring as we can get the two parameters from the inflight
+ * queueregion
+ *
+ * @param vid
+ *  vhost device ID
+ * @param queue_id
+ *  vhost queue index
+ * @param last_avail_idx
+ *  vhost last_avail_idx to get
+ * @param last_used_idx
+ *  vhost last_used_idx to get
+ * @return
+ *  0 on success, -1 on failure
+ */
+__rte_experimental
+int
+rte_vhost_get_vring_base_from_inflight(int vid,
+	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx);
+
 /**
  * Set last_avail/used_idx of the vhost virtqueue
  *
diff --git a/lib/librte_vhost/rte_vhost_version.map b/lib/librte_vhost/rte_vhost_version.map
index bc70bfaa5..ce517b127 100644
--- a/lib/librte_vhost/rte_vhost_version.map
+++ b/lib/librte_vhost/rte_vhost_version.map
@@ -93,4 +93,6 @@ EXPERIMENTAL {
 	rte_vhost_set_last_inflight_io_packed;
 	rte_vhost_clr_inflight_desc_split;
 	rte_vhost_clr_inflight_desc_packed;
+	rte_vhost_get_vhost_ring_inflight;
+	rte_vhost_get_vring_base_from_inflight;
 };
diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 44ae87d4f..95f99ea51 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -783,6 +783,41 @@ rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 	return 0;
 }
 
+int
+rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
+		struct rte_vhost_ring_inflight *vring)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (vring_idx >= VHOST_MAX_VRING)
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (vq_is_packed(dev)) {
+		if (unlikely(!vq->inflight_packed))
+			return -1;
+
+		vring->inflight_packed = vq->inflight_packed;
+	} else {
+		if (unlikely(!vq->inflight_split))
+			return -1;
+
+		vring->inflight_split = vq->inflight_split;
+	}
+
+	vring->resubmit_inflight = vq->resubmit_inflight;
+
+	return 0;
+}
+
 int
 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
 	uint16_t idx)
@@ -1249,6 +1284,30 @@ int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 	return 0;
 }
 
+int
+rte_vhost_get_vring_base_from_inflight(int vid,
+	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx)
+{
+	struct rte_vhost_inflight_info_packed *inflight_info;
+	struct virtio_net *dev = get_device(vid);
+
+	if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
+		return -1;
+
+	if (!vq_is_packed(dev))
+		return -1;
+
+	inflight_info = dev->virtqueue[queue_id]->inflight_packed;
+	if (!inflight_info)
+		return -1;
+
+	*last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
+				inflight_info->old_used_idx;
+	*last_used_idx = *last_avail_idx;
+
+	return 0;
+}
+
 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
 		uint16_t last_avail_idx, uint16_t last_used_idx)
 {
-- 
2.17.2

[dpdk-dev] [PATCH v7 08/10] vhost: add vring functions packed ring support

[Jin Yu]

This patch add packed ring support in two APIs
so user can get the packed ring`.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/vhost.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 49 insertions(+), 19 deletions(-)

diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 95f99ea51..1b6dcca98 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -771,9 +771,15 @@ rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 	if (!vq)
 		return -1;
 
-	vring->desc  = vq->desc;
-	vring->avail = vq->avail;
-	vring->used  = vq->used;
+	if (vq_is_packed(dev)) {
+		vring->desc_packed = vq->desc_packed;
+		vring->driver_event = vq->driver_event;
+		vring->device_event = vq->device_event;
+	} else {
+		vring->desc = vq->desc;
+		vring->avail = vq->avail;
+		vring->used = vq->used;
+	}
 	vring->log_guest_addr  = vq->log_guest_addr;
 
 	vring->callfd  = vq->callfd;
@@ -1273,13 +1279,51 @@ int rte_vhost_get_log_base(int vid, uint64_t *log_base,
 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 		uint16_t *last_avail_idx, uint16_t *last_used_idx)
 {
+	struct vhost_virtqueue *vq;
 	struct virtio_net *dev = get_device(vid);
 
 	if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
 		return -1;
 
-	*last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
-	*last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
+	vq = dev->virtqueue[queue_id];
+	if (!vq)
+		return -1;
+
+	if (vq_is_packed(dev)) {
+		*last_avail_idx = (vq->avail_wrap_counter << 15) |
+					vq->last_avail_idx;
+		*last_used_idx = (vq->used_wrap_counter << 15) |
+					vq->last_used_idx;
+	} else {
+		*last_avail_idx = vq->last_avail_idx;
+		*last_used_idx = vq->last_used_idx;
+	}
+
+	return 0;
+}
+
+int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
+		uint16_t last_avail_idx, uint16_t last_used_idx)
+{
+	struct vhost_virtqueue *vq;
+	struct virtio_net *dev = get_device(vid);
+
+	if (!dev)
+		return -1;
+
+	vq = dev->virtqueue[queue_id];
+	if (!vq)
+		return -1;
+
+	if (vq_is_packed(dev)) {
+		vq->last_avail_idx = last_avail_idx & 0x7fff;
+		vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
+		vq->last_used_idx = last_used_idx & 0x7fff;
+		vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
+	} else {
+		vq->last_avail_idx = last_avail_idx;
+		vq->last_used_idx = last_used_idx;
+	}
 
 	return 0;
 }
@@ -1308,20 +1352,6 @@ rte_vhost_get_vring_base_from_inflight(int vid,
 	return 0;
 }
 
-int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
-		uint16_t last_avail_idx, uint16_t last_used_idx)
-{
-	struct virtio_net *dev = get_device(vid);
-
-	if (!dev)
-		return -1;
-
-	dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
-	dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
-
-	return 0;
-}
-
 int rte_vhost_extern_callback_register(int vid,
 		struct rte_vhost_user_extern_ops const * const ops, void *ctx)
 {
-- 
2.17.2

[dpdk-dev] [PATCH v7 09/10] vhost: add an API for judging vq format

[Jin Yu]

This patch introduces an API for getting virtqueue format.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: Jin Yu <jin.yu@intel.com>
---
 lib/librte_vhost/rte_vhost.h           | 13 +++++++++++++
 lib/librte_vhost/rte_vhost_version.map |  1 +
 lib/librte_vhost/vhost.c               | 12 ++++++++++++
 3 files changed, 26 insertions(+)

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 8b6435d0d..6d768c550 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -11,6 +11,7 @@
  */
 
 #include <stdint.h>
+#include <stdbool.h>
 #include <sys/eventfd.h>
 
 #include <rte_memory.h>
@@ -678,6 +679,18 @@ uint16_t rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
  */
 int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
 
+/**
+ * Get vq is packed
+ *
+ * @param vid
+ *  vhost device ID
+ * @return
+ *  true on packed, false on split
+ */
+__rte_experimental
+bool
+rte_vhost_vq_is_packed(int vid);
+
 /**
  * Get guest vring info, including the vring address, vring size, etc.
  *
diff --git a/lib/librte_vhost/rte_vhost_version.map b/lib/librte_vhost/rte_vhost_version.map
index ce517b127..0dce0d2bb 100644
--- a/lib/librte_vhost/rte_vhost_version.map
+++ b/lib/librte_vhost/rte_vhost_version.map
@@ -95,4 +95,5 @@ EXPERIMENTAL {
 	rte_vhost_clr_inflight_desc_packed;
 	rte_vhost_get_vhost_ring_inflight;
 	rte_vhost_get_vring_base_from_inflight;
+	rte_vhost_vq_is_packed;
 };
diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 1b6dcca98..372a25375 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -753,6 +753,18 @@ rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 	return 0;
 }
 
+bool
+rte_vhost_vq_is_packed(int vid)
+{
+	struct virtio_net *dev;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	return vq_is_packed(dev);
+}
+
 int
 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 			  struct rte_vhost_vring *vring)
-- 
2.17.2

[dpdk-dev] [PATCH v7 10/10] vhost: add vhost-user-blk example which support inflight

[Jin Yu]

A vhost-user-blk example that support inflight feature. It uses the
new APIs that introduced in the first patch, so It can show how there
APIs work to support inflight feature.

Signed-off-by: Jin Yu <jin.yu@intel.com>
---
V1 - add the case.
V2 - add the rte_vhost prefix.
V3 - add packed ring support
---
 examples/vhost_blk/Makefile           |   67 ++
 examples/vhost_blk/blk.c              |  125 +++
 examples/vhost_blk/blk_spec.h         |   95 ++
 examples/vhost_blk/meson.build        |   20 +
 examples/vhost_blk/vhost_blk.c        | 1313 +++++++++++++++++++++++++
 examples/vhost_blk/vhost_blk.h        |  116 +++
 examples/vhost_blk/vhost_blk_compat.c |  195 ++++
 7 files changed, 1931 insertions(+)
 create mode 100644 examples/vhost_blk/Makefile
 create mode 100644 examples/vhost_blk/blk.c
 create mode 100644 examples/vhost_blk/blk_spec.h
 create mode 100644 examples/vhost_blk/meson.build
 create mode 100644 examples/vhost_blk/vhost_blk.c
 create mode 100644 examples/vhost_blk/vhost_blk.h
 create mode 100644 examples/vhost_blk/vhost_blk_compat.c

diff --git a/examples/vhost_blk/Makefile b/examples/vhost_blk/Makefile
new file mode 100644
index 000000000..52e9befd8
--- /dev/null
+++ b/examples/vhost_blk/Makefile
@@ -0,0 +1,67 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2010-2017 Intel Corporation
+
+# binary name
+APP = vhost-blk
+
+# all source are stored in SRCS-y
+SRCS-y := blk.c vhost_blk.c vhost_blk_compat.c
+
+# Build using pkg-config variables if possible
+$(shell pkg-config --exists libdpdk)
+ifeq ($(.SHELLSTATUS),0)
+
+all: shared
+.PHONY: shared static
+shared: build/$(APP)-shared
+	ln -sf $(APP)-shared build/$(APP)
+static: build/$(APP)-static
+	ln -sf $(APP)-static build/$(APP)
+
+CFLAGS += -D_FILE_OFFSET_BITS=64
+LDFLAGS += -pthread
+
+PC_FILE := $(shell pkg-config --path libdpdk)
+CFLAGS += -O3 $(shell pkg-config --cflags libdpdk)
+LDFLAGS_SHARED = $(shell pkg-config --libs libdpdk)
+LDFLAGS_STATIC = -Wl,-Bstatic $(shell pkg-config --static --libs libdpdk)
+
+build/$(APP)-shared: $(SRCS-y) Makefile $(PC_FILE) | build
+	$(CC) $(CFLAGS) $(SRCS-y) -o $@ $(LDFLAGS) $(LDFLAGS_SHARED)
+
+build/$(APP)-static: $(SRCS-y) Makefile $(PC_FILE) | build
+	$(CC) $(CFLAGS) $(SRCS-y) -o $@ $(LDFLAGS) $(LDFLAGS_STATIC)
+
+build:
+	@mkdir -p $@
+
+.PHONY: clean
+clean:
+	rm -f build/$(APP) build/$(APP)-static build/$(APP)-shared
+	test -d build && rmdir -p build || true
+
+else # Build using legacy build system
+
+ifeq ($(RTE_SDK),)
+$(error "Please define RTE_SDK environment variable")
+endif
+
+# Default target, detect a build directory, by looking for a path with a .config
+RTE_TARGET ?= $(notdir $(abspath $(dir $(firstword $(wildcard $(RTE_SDK)/*/.config)))))
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+ifneq ($(CONFIG_RTE_EXEC_ENV_LINUX),y)
+$(info This application can only operate in a linux environment, \
+please change the definition of the RTE_TARGET environment variable)
+all:
+else
+
+CFLAGS += -D_FILE_OFFSET_BITS=64
+CFLAGS += -O2
+#CFLAGS += $(WERROR_FLAGS)
+
+include $(RTE_SDK)/mk/rte.extapp.mk
+
+endif
+endif
diff --git a/examples/vhost_blk/blk.c b/examples/vhost_blk/blk.c
new file mode 100644
index 000000000..424ed3015
--- /dev/null
+++ b/examples/vhost_blk/blk.c
@@ -0,0 +1,125 @@
+// SPDX-License-Identifier: BSD-3-Clause
+// Copyright(c) 2010-2019 Intel Corporation
+
+/**
+ * This work is largely based on the "vhost-user-blk" implementation by
+ * SPDK(https://github.com/spdk/spdk).
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <assert.h>
+#include <ctype.h>
+#include <string.h>
+#include <stddef.h>
+
+#include <rte_atomic.h>
+#include <rte_cycles.h>
+#include <rte_log.h>
+#include <rte_malloc.h>
+#include <rte_byteorder.h>
+#include <rte_string_fns.h>
+
+#include "vhost_blk.h"
+#include "blk_spec.h"
+
+static void
+vhost_strcpy_pad(void *dst, const char *src, size_t size, int pad)
+{
+	size_t len;
+
+	len = strlen(src);
+	if (len < size) {
+		memcpy(dst, src, len);
+		memset((char *)dst + len, pad, size - len);
+	} else {
+		memcpy(dst, src, size);
+	}
+}
+
+static int
+vhost_bdev_blk_readwrite(struct vhost_block_dev *bdev,
+			  struct vhost_blk_task *task,
+			  uint64_t lba_512, __rte_unused uint32_t xfer_len)
+{
+	uint32_t i;
+	uint64_t offset;
+	uint32_t nbytes = 0;
+
+	offset = lba_512 * 512;
+
+	for (i = 0; i < task->iovs_cnt; i++) {
+		if (task->dxfer_dir == BLK_DIR_TO_DEV)
+			memcpy(bdev->data + offset, task->iovs[i].iov_base,
+			       task->iovs[i].iov_len);
+		else
+			memcpy(task->iovs[i].iov_base, bdev->data + offset,
+			       task->iovs[i].iov_len);
+		offset += task->iovs[i].iov_len;
+		nbytes += task->iovs[i].iov_len;
+	}
+
+	return nbytes;
+}
+
+int
+vhost_bdev_process_blk_commands(struct vhost_block_dev *bdev,
+				 struct vhost_blk_task *task)
+{
+	int used_len;
+
+	if (unlikely(task->data_len > (bdev->blockcnt * bdev->blocklen))) {
+		fprintf(stderr, "read or write beyond capacity\n");
+		return VIRTIO_BLK_S_UNSUPP;
+	}
+
+	switch (task->req->type) {
+	case VIRTIO_BLK_T_IN:
+		if (unlikely(task->data_len == 0 ||
+			(task->data_len & (512 - 1)) != 0)) {
+			fprintf(stderr,
+				"%s - passed IO buffer is not multiple of 512b"
+				"(req_idx = %"PRIu16").\n",
+				task->req->type ? "WRITE" : "READ",
+				task->head_idx);
+			return VIRTIO_BLK_S_UNSUPP;
+		}
+
+		task->dxfer_dir = BLK_DIR_FROM_DEV;
+		vhost_bdev_blk_readwrite(bdev, task,
+					 task->req->sector, task->data_len);
+		break;
+	case VIRTIO_BLK_T_OUT:
+		if (unlikely(task->data_len == 0 ||
+			(task->data_len & (512 - 1)) != 0)) {
+			fprintf(stderr,
+				"%s - passed IO buffer is not multiple of 512b"
+				"(req_idx = %"PRIu16").\n",
+				task->req->type ? "WRITE" : "READ",
+				task->head_idx);
+			return VIRTIO_BLK_S_UNSUPP;
+		}
+
+		if (task->readtype) {
+			fprintf(stderr, "type isn't right\n");
+			return VIRTIO_BLK_S_IOERR;
+		}
+		task->dxfer_dir = BLK_DIR_TO_DEV;
+		vhost_bdev_blk_readwrite(bdev, task,
+					 task->req->sector, task->data_len);
+		break;
+	case VIRTIO_BLK_T_GET_ID:
+		if (!task->iovs_cnt || task->data_len)
+			return VIRTIO_BLK_S_UNSUPP;
+		used_len = min(VIRTIO_BLK_ID_BYTES, task->data_len);
+		vhost_strcpy_pad(task->iovs[0].iov_base,
+				 bdev->product_name, used_len, ' ');
+		break;
+	default:
+		fprintf(stderr, "unsupported cmd\n");
+		return VIRTIO_BLK_S_UNSUPP;
+	}
+
+	return VIRTIO_BLK_S_OK;
+}
diff --git a/examples/vhost_blk/blk_spec.h b/examples/vhost_blk/blk_spec.h
new file mode 100644
index 000000000..5875e2f86
--- /dev/null
+++ b/examples/vhost_blk/blk_spec.h
@@ -0,0 +1,95 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2019 Intel Corporation
+ */
+
+#ifndef _BLK_SPEC_H
+#define _BLK_SPEC_H
+
+#include <stdint.h>
+
+#ifndef VHOST_USER_MEMORY_MAX_NREGIONS
+#define VHOST_USER_MEMORY_MAX_NREGIONS		8
+#endif
+
+#ifndef VHOST_USER_MAX_CONFIG_SIZE
+#define VHOST_USER_MAX_CONFIG_SIZE		256
+#endif
+
+#ifndef VHOST_USER_PROTOCOL_F_CONFIG
+#define VHOST_USER_PROTOCOL_F_CONFIG		9
+#endif
+
+#ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
+#define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD	12
+#endif
+
+#define VIRTIO_BLK_ID_BYTES	20	/* ID string length */
+
+#define VIRTIO_BLK_T_IN			0
+#define VIRTIO_BLK_T_OUT		1
+#define VIRTIO_BLK_T_FLUSH		4
+#define VIRTIO_BLK_T_GET_ID		8
+#define VIRTIO_BLK_T_DISCARD		11
+#define VIRTIO_BLK_T_WRITE_ZEROES	13
+
+#define VIRTIO_BLK_S_OK			0
+#define VIRTIO_BLK_S_IOERR		1
+#define VIRTIO_BLK_S_UNSUPP		2
+
+enum vhost_user_request {
+	VHOST_USER_NONE = 0,
+	VHOST_USER_GET_FEATURES = 1,
+	VHOST_USER_SET_FEATURES = 2,
+	VHOST_USER_SET_OWNER = 3,
+	VHOST_USER_RESET_OWNER = 4,
+	VHOST_USER_SET_MEM_TABLE = 5,
+	VHOST_USER_SET_LOG_BASE = 6,
+	VHOST_USER_SET_LOG_FD = 7,
+	VHOST_USER_SET_VRING_NUM = 8,
+	VHOST_USER_SET_VRING_ADDR = 9,
+	VHOST_USER_SET_VRING_BASE = 10,
+	VHOST_USER_GET_VRING_BASE = 11,
+	VHOST_USER_SET_VRING_KICK = 12,
+	VHOST_USER_SET_VRING_CALL = 13,
+	VHOST_USER_SET_VRING_ERR = 14,
+	VHOST_USER_GET_PROTOCOL_FEATURES = 15,
+	VHOST_USER_SET_PROTOCOL_FEATURES = 16,
+	VHOST_USER_GET_QUEUE_NUM = 17,
+	VHOST_USER_SET_VRING_ENABLE = 18,
+	VHOST_USER_MAX
+};
+
+/** Get/set config msg payload */
+struct vhost_user_config {
+	uint32_t offset;
+	uint32_t size;
+	uint32_t flags;
+	uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
+};
+
+/** Fixed-size vhost_memory struct */
+struct vhost_memory_padded {
+	uint32_t nregions;
+	uint32_t padding;
+	struct vhost_memory_region regions[VHOST_USER_MEMORY_MAX_NREGIONS];
+};
+
+struct vhost_user_msg {
+	enum vhost_user_request request;
+
+#define VHOST_USER_VERSION_MASK     0x3
+#define VHOST_USER_REPLY_MASK       (0x1 << 2)
+	uint32_t flags;
+	uint32_t size; /**< the following payload size */
+	union {
+#define VHOST_USER_VRING_IDX_MASK   0xff
+#define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
+		uint64_t u64;
+		struct vhost_vring_state state;
+		struct vhost_vring_addr addr;
+		struct vhost_memory_padded memory;
+		struct vhost_user_config cfg;
+	} payload;
+} __attribute((packed));
+
+#endif
diff --git a/examples/vhost_blk/meson.build b/examples/vhost_blk/meson.build
new file mode 100644
index 000000000..028aa4f62
--- /dev/null
+++ b/examples/vhost_blk/meson.build
@@ -0,0 +1,20 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2017 Intel Corporation
+
+# meson file, for building this example as part of a main DPDK build.
+#
+# To build this example as a standalone application with an already-installed
+# DPDK instance, use 'make'
+
+if not is_linux
+	build = false
+endif
+
+if not cc.has_header('linux/virtio_blk.h')
+	build = false
+endif
+
+deps += 'vhost'
+sources = files(
+	'blk.c', 'vhost_blk.c', 'vhost_blk_compat.c'
+)
diff --git a/examples/vhost_blk/vhost_blk.c b/examples/vhost_blk/vhost_blk.c
new file mode 100644
index 000000000..e83c47e14
--- /dev/null
+++ b/examples/vhost_blk/vhost_blk.c
@@ -0,0 +1,1313 @@
+// SPDX-License-Identifier: BSD-3-Clause
+// Copyright(c) 2010-2017 Intel Corporation
+
+#include <stdint.h>
+#include <unistd.h>
+#include <stdbool.h>
+#include <signal.h>
+#include <assert.h>
+#include <semaphore.h>
+#include <linux/virtio_blk.h>
+#include <linux/virtio_ring.h>
+
+#include <rte_atomic.h>
+#include <rte_cycles.h>
+#include <rte_log.h>
+#include <rte_malloc.h>
+#include <rte_vhost.h>
+
+#include "vhost_blk.h"
+#include "blk_spec.h"
+
+#define VIRTQ_DESC_F_NEXT 1
+#define VIRTQ_DESC_F_AVAIL (1 << 7)
+#define VIRTQ_DESC_F_USED (1 << 15)
+
+#define MAX_TASK	12
+
+#define VHOST_BLK_FEATURES ((1ULL << VIRTIO_F_RING_PACKED) | \
+			    (1ULL << VIRTIO_F_VERSION_1) |\
+			    (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) | \
+			    (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))
+
+/* Path to folder where character device will be created. Can be set by user. */
+static char dev_pathname[PATH_MAX] = "";
+static sem_t exit_sem;
+
+struct vhost_blk_ctrlr *
+vhost_blk_ctrlr_find(const char *ctrlr_name)
+{
+	/* currently we only support 1 socket file fd */
+	return g_vhost_ctrlr;
+}
+
+static uint64_t gpa_to_vva(int vid, uint64_t gpa, uint64_t *len)
+{
+	char path[PATH_MAX];
+	struct vhost_blk_ctrlr *ctrlr;
+	int ret = 0;
+
+	ret = rte_vhost_get_ifname(vid, path, PATH_MAX);
+	if (ret) {
+		fprintf(stderr, "Cannot get socket name\n");
+		assert(ret != 0);
+	}
+
+	ctrlr = vhost_blk_ctrlr_find(path);
+	if (!ctrlr) {
+		fprintf(stderr, "Controller is not ready\n");
+		assert(ctrlr != NULL);
+	}
+
+	assert(ctrlr->mem != NULL);
+
+	return rte_vhost_va_from_guest_pa(ctrlr->mem, gpa, len);
+}
+
+static struct vring_packed_desc *
+descriptor_get_next_packed(struct rte_vhost_vring *vq,
+			     uint16_t *idx)
+{
+	if (vq->desc_packed[*idx % vq->size].flags & VIRTQ_DESC_F_NEXT) {
+		*idx += 1;
+		return &vq->desc_packed[*idx % vq->size];
+	}
+
+	return NULL;
+}
+
+static bool
+descriptor_has_next_packed(struct vring_packed_desc *cur_desc)
+{
+	return !!(cur_desc->flags & VRING_DESC_F_NEXT);
+}
+
+static bool
+descriptor_is_wr_packed(struct vring_packed_desc *cur_desc)
+{
+	return !!(cur_desc->flags & VRING_DESC_F_WRITE);
+}
+
+static struct rte_vhost_inflight_desc_packed *
+inflight_desc_get_next(struct rte_vhost_inflight_info_packed *inflight_packed,
+			       struct rte_vhost_inflight_desc_packed *cur_desc)
+{
+	if (!!(cur_desc->flags & VIRTQ_DESC_F_NEXT))
+		return &inflight_packed->desc[cur_desc->next];
+
+	return NULL;
+}
+
+static bool
+inflight_desc_has_next(struct rte_vhost_inflight_desc_packed *cur_desc)
+{
+	return !!(cur_desc->flags & VRING_DESC_F_NEXT);
+}
+
+static bool
+inflight_desc_is_wr(struct rte_vhost_inflight_desc_packed *cur_desc)
+{
+	return !!(cur_desc->flags & VRING_DESC_F_WRITE);
+}
+
+static void
+inflight_process_payload_chain_packed(struct inflight_blk_task *task)
+{
+	void *data;
+	uint64_t chunck_len;
+	struct vhost_blk_task *blk_task;
+	struct rte_vhost_inflight_desc_packed *desc;
+
+	blk_task = &task->blk_task;
+	blk_task->iovs_cnt = 0;
+
+	do {
+		desc = task->inflight_desc;
+		chunck_len = desc->len;
+		data = (void *)(uintptr_t)gpa_to_vva(blk_task->bdev->vid,
+						     desc->addr,
+						     &chunck_len);
+		if (!data || chunck_len != desc->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			return;
+		}
+
+		blk_task->iovs[blk_task->iovs_cnt].iov_base = data;
+		blk_task->iovs[blk_task->iovs_cnt].iov_len = desc->len;
+		blk_task->data_len += desc->len;
+		blk_task->iovs_cnt++;
+		task->inflight_desc = inflight_desc_get_next(
+					task->inflight_packed, desc);
+	} while (inflight_desc_has_next(task->inflight_desc));
+
+	chunck_len = task->inflight_desc->len;
+	blk_task->status = (void *)(uintptr_t)gpa_to_vva(
+		blk_task->bdev->vid, task->inflight_desc->addr, &chunck_len);
+	if (!blk_task->status || chunck_len != task->inflight_desc->len)
+		fprintf(stderr, "failed to translate desc address.\n");
+}
+
+static void
+inflight_submit_completion_packed(struct inflight_blk_task *task,
+					      uint32_t q_idx, uint16_t *used_id,
+					      bool *used_wrap_counter)
+{
+	struct vhost_blk_ctrlr *ctrlr;
+	struct rte_vhost_vring *vq;
+	struct vring_packed_desc *desc;
+	uint16_t flags;
+	uint16_t entry_num;
+	int ret;
+
+	ctrlr = vhost_blk_ctrlr_find(dev_pathname);
+	vq = task->blk_task.vq;
+
+	ret = rte_vhost_set_last_inflight_io_packed(ctrlr->bdev->vid, q_idx,
+						    task->blk_task.head_idx);
+	if (ret != 0)
+		fprintf(stderr, "fail to set last inflight io\n");
+
+	desc = &vq->desc_packed[*used_id];
+	desc->id = task->blk_task.buffer_id;
+	rte_compiler_barrier();
+	if (*used_wrap_counter)
+		desc->flags |= VIRTQ_DESC_F_AVAIL | VIRTQ_DESC_F_USED;
+	else
+		desc->flags &= ~(VIRTQ_DESC_F_AVAIL | VIRTQ_DESC_F_USED);
+	rte_compiler_barrier();
+
+	*used_id += task->blk_task.iovs_cnt + 2;
+	if (*used_id >= vq->size) {
+		*used_id -= vq->size;
+		*used_wrap_counter = !(*used_wrap_counter);
+	}
+
+	ret = rte_vhost_clr_inflight_desc_packed(ctrlr->bdev->vid, q_idx,
+						 task->blk_task.head_idx);
+	if (ret != 0)
+		fprintf(stderr, "fail to clear inflight io\n");
+
+	/* Send an interrupt back to the guest VM so that it knows
+	 * a completion is ready to be processed.
+	 */
+	rte_vhost_vring_call(task->blk_task.bdev->vid, q_idx);
+}
+
+static void
+submit_completion_packed(struct vhost_blk_task *task, uint32_t q_idx,
+				  uint16_t *used_id, bool *used_wrap_counter)
+{
+	struct vhost_blk_ctrlr *ctrlr;
+	struct rte_vhost_vring *vq;
+	struct vring_packed_desc *desc;
+	uint16_t entry_num;
+	int ret;
+
+	ctrlr = vhost_blk_ctrlr_find(dev_pathname);
+	vq = task->vq;
+
+	ret = rte_vhost_set_last_inflight_io_packed(ctrlr->bdev->vid, q_idx,
+						    task->inflight_idx);
+	if (ret != 0)
+		fprintf(stderr, "fail to set last inflight io\n");
+
+	desc = &vq->desc_packed[*used_id];
+	desc->id = task->buffer_id;
+	rte_compiler_barrier();
+	if (*used_wrap_counter)
+		desc->flags |= VIRTQ_DESC_F_AVAIL | VIRTQ_DESC_F_USED;
+	else
+		desc->flags &= ~(VIRTQ_DESC_F_AVAIL | VIRTQ_DESC_F_USED);
+	rte_compiler_barrier();
+
+	*used_id += task->iovs_cnt + 2;
+	if (*used_id >= vq->size) {
+		*used_id -= vq->size;
+		*used_wrap_counter = !(*used_wrap_counter);
+	}
+
+	ret = rte_vhost_clr_inflight_desc_packed(ctrlr->bdev->vid, q_idx,
+						 task->inflight_idx);
+	if (ret != 0)
+		fprintf(stderr, "fail to clear inflight io\n");
+
+	/* Send an interrupt back to the guest VM so that it knows
+	 * a completion is ready to be processed.
+	 */
+	rte_vhost_vring_call(task->bdev->vid, q_idx);
+}
+
+static void
+vhost_process_payload_chain_packed(struct vhost_blk_task *task,
+	uint16_t *idx)
+{
+	void *data;
+	uint64_t chunck_len;
+
+	task->iovs_cnt = 0;
+
+	do {
+		chunck_len = task->desc_packed->len;
+		data = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+						     task->desc_packed->addr,
+							 &chunck_len);
+		if (!data || chunck_len != task->desc_packed->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			return;
+		}
+
+		task->iovs[task->iovs_cnt].iov_base = data;
+		task->iovs[task->iovs_cnt].iov_len = task->desc_packed->len;
+		task->data_len += task->desc_packed->len;
+		task->iovs_cnt++;
+		task->desc_packed = descriptor_get_next_packed(task->vq, idx);
+	} while (descriptor_has_next_packed(task->desc_packed));
+
+	task->last_idx = *idx % task->vq->size;
+	chunck_len = task->desc_packed->len;
+	task->status = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+						   task->desc_packed->addr,
+						   &chunck_len);
+	if (!task->status || chunck_len != task->desc_packed->len)
+		fprintf(stderr, "failed to translate desc address.\n");
+}
+
+
+static int
+descriptor_is_available(struct rte_vhost_vring *vring, uint16_t idx,
+					bool avail_wrap_counter)
+{
+	uint16_t flags = vring->desc_packed[idx].flags;
+
+	return ((!!(flags & VIRTQ_DESC_F_AVAIL) == avail_wrap_counter) &&
+		(!!(flags & VIRTQ_DESC_F_USED) != avail_wrap_counter));
+}
+
+static int
+descriptor_is_used(struct rte_vhost_vring *vring, uint16_t idx,
+					bool used_wrap_counter)
+{
+	uint16_t flags = vring->desc_packed[idx].flags;
+
+	return ((!!(flags & VIRTQ_DESC_F_AVAIL) == used_wrap_counter) &&
+		(!!(flags & VIRTQ_DESC_F_USED) == used_wrap_counter));
+}
+
+static void
+process_requestq_packed(struct vhost_blk_ctrlr *ctrlr, uint32_t q_idx)
+{
+	bool avail_wrap_counter, used_wrap_counter;
+	uint16_t avail_idx, used_idx;
+	int ret;
+	uint64_t chunck_len;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_vring *vq;
+	struct vhost_blk_task *task;
+
+	blk_vq = &ctrlr->bdev->queues[q_idx];
+	vq = &blk_vq->vq;
+
+	avail_idx = blk_vq->last_avail_idx;
+	avail_wrap_counter = blk_vq->avail_wrap_counter;
+	used_idx = blk_vq->last_used_idx;
+	used_wrap_counter = blk_vq->used_wrap_counter;
+
+	task = rte_zmalloc(NULL, sizeof(*task), 0);
+	assert(task != NULL);
+	task->vq = vq;
+	task->bdev = ctrlr->bdev;
+
+	while (descriptor_is_available(vq, avail_idx, avail_wrap_counter)) {
+		task->head_idx = avail_idx;
+		task->desc_packed = &task->vq->desc_packed[task->head_idx];
+		task->iovs_cnt = 0;
+		task->data_len = 0;
+		task->req = NULL;
+		task->status = NULL;
+
+		/* does not support indirect descriptors */
+		assert((task->desc_packed->flags & VRING_DESC_F_INDIRECT) == 0);
+
+		chunck_len = task->desc_packed->len;
+		task->req = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+			task->desc_packed->addr, &chunck_len);
+		if (!task->req || chunck_len != task->desc_packed->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			rte_free(task);
+			return;
+		}
+
+		task->desc_packed = descriptor_get_next_packed(task->vq,
+								&avail_idx);
+		assert(task->desc_packed != NULL);
+		if (!descriptor_has_next_packed(task->desc_packed)) {
+			task->dxfer_dir = BLK_DIR_NONE;
+			task->last_idx = avail_idx % vq->size;
+			chunck_len = task->desc_packed->len;
+			task->status = (void *)(uintptr_t)
+					      gpa_to_vva(task->bdev->vid,
+							task->desc_packed->addr,
+							&chunck_len);
+			if (!task->status ||
+				chunck_len != task->desc_packed->len) {
+				fprintf(stderr,
+					"failed to translate desc address.\n");
+				rte_free(task);
+				return;
+			}
+		} else {
+			task->readtype = descriptor_is_wr_packed(
+							task->desc_packed);
+			vhost_process_payload_chain_packed(task, &avail_idx);
+		}
+		task->buffer_id = vq->desc_packed[task->last_idx].id;
+		rte_vhost_set_inflight_desc_packed(ctrlr->bdev->vid, q_idx,
+						   task->head_idx,
+						   task->last_idx,
+						   &task->inflight_idx);
+
+		if (++avail_idx >= vq->size) {
+			avail_idx -= vq->size;
+			avail_wrap_counter = !avail_wrap_counter;
+		}
+		blk_vq->last_avail_idx = avail_idx;
+		blk_vq->avail_wrap_counter = avail_wrap_counter;
+
+		ret = vhost_bdev_process_blk_commands(ctrlr->bdev, task);
+		if (ret) {
+			/* invalid response */
+			*task->status = VIRTIO_BLK_S_IOERR;
+		} else {
+			/* successfully */
+			*task->status = VIRTIO_BLK_S_OK;
+		}
+
+		submit_completion_packed(task, q_idx, &used_idx,
+						&used_wrap_counter);
+		blk_vq->last_used_idx = used_idx;
+		blk_vq->used_wrap_counter = used_wrap_counter;
+	}
+
+	rte_free(task);
+}
+
+static void
+submit_inflight_vq_packed(struct vhost_blk_ctrlr *ctrlr,
+	uint16_t q_idx)
+{
+	bool used_wrap_counter;
+	int req_idx, ret;
+	uint16_t used_idx;
+	uint64_t chunck_len;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_ring_inflight *inflight_vq;
+	struct rte_vhost_resubmit_info *resubmit_info;
+	struct rte_vhost_vring *vq;
+	struct inflight_blk_task *task;
+	struct vhost_blk_task *blk_task;
+	struct rte_vhost_inflight_info_packed *inflight_info;
+
+	blk_vq = &ctrlr->bdev->queues[q_idx];
+	vq = &blk_vq->vq;
+	inflight_vq = &blk_vq->inflight_vq;
+	resubmit_info = inflight_vq->resubmit_inflight;
+	inflight_info = inflight_vq->inflight_packed;
+	used_idx = blk_vq->last_used_idx;
+	used_wrap_counter = blk_vq->used_wrap_counter;
+
+	task = rte_malloc(NULL, sizeof(*task), 0);
+	if (!task) {
+		fprintf(stderr, "fail to allocate memory\n");
+		return;
+	}
+	blk_task = &task->blk_task;
+	blk_task->vq = vq;
+	blk_task->bdev = ctrlr->bdev;
+	task->inflight_packed = inflight_vq->inflight_packed;
+
+	while (resubmit_info->resubmit_num-- > 0) {
+		req_idx = resubmit_info->resubmit_num;
+		blk_task->head_idx =
+			resubmit_info->resubmit_list[req_idx].index;
+		task->inflight_desc =
+			&inflight_info->desc[blk_task->head_idx];
+		task->blk_task.iovs_cnt = 0;
+		task->blk_task.data_len = 0;
+		task->blk_task.req = NULL;
+		task->blk_task.status = NULL;
+
+		/* update the avail idx too
+		 * as it's initial value equals to used idx
+		 */
+		blk_vq->last_avail_idx += task->inflight_desc->num;
+		if (blk_vq->last_avail_idx >= vq->size) {
+			blk_vq->last_avail_idx -= vq->size;
+			blk_vq->avail_wrap_counter =
+				!blk_vq->avail_wrap_counter;
+		}
+
+		/* does not support indirect descriptors */
+		assert(task->inflight_desc != NULL);
+		assert((task->inflight_desc->flags &
+			VRING_DESC_F_INDIRECT) == 0);
+
+		chunck_len = task->inflight_desc->len;
+		blk_task->req = (void *)(uintptr_t)
+				     gpa_to_vva(blk_task->bdev->vid,
+						task->inflight_desc->addr,
+						&chunck_len);
+		if (!blk_task->req ||
+			chunck_len != task->inflight_desc->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			rte_free(task);
+			return;
+		}
+
+		task->inflight_desc = inflight_desc_get_next(
+			task->inflight_packed, task->inflight_desc);
+		assert(task->inflight_desc != NULL);
+		if (!inflight_desc_has_next(task->inflight_desc)) {
+			blk_task->dxfer_dir = BLK_DIR_NONE;
+			chunck_len = task->inflight_desc->len;
+			blk_task->status = (void *)(uintptr_t)
+				gpa_to_vva(blk_task->bdev->vid,
+						task->inflight_desc->addr,
+						&chunck_len);
+			if (!blk_task->status ||
+			    chunck_len != task->inflight_desc->len) {
+				fprintf(stderr,
+					"failed to translate desc address.\n");
+				rte_free(task);
+				return;
+			}
+		} else {
+			blk_task->readtype =
+			inflight_desc_is_wr(task->inflight_desc);
+			inflight_process_payload_chain_packed(task);
+		}
+
+		blk_task->buffer_id = task->inflight_desc->id;
+
+		ret = vhost_bdev_process_blk_commands(ctrlr->bdev, blk_task);
+		if (ret)
+			/* invalid response */
+			*blk_task->status = VIRTIO_BLK_S_IOERR;
+		else
+			/* successfully */
+			*blk_task->status = VIRTIO_BLK_S_OK;
+
+		inflight_submit_completion_packed(task, q_idx, &used_idx,
+						  &used_wrap_counter);
+
+		blk_vq->last_used_idx = used_idx;
+		blk_vq->used_wrap_counter = used_wrap_counter;
+	}
+
+	rte_free(task);
+}
+
+static void
+process_batch_requests_packed(struct vhost_blk_ctrlr *ctrlr,
+	uint16_t q_idx)
+{
+	bool avail_wrap_counter, used_wrap_counter;
+	uint16_t avail_idx, used_idx;
+	int ret, i, j;
+	uint64_t chunck_len;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_vring *vq;
+	struct vhost_blk_task *task[MAX_TASK] = {NULL};
+
+	blk_vq = &ctrlr->bdev->queues[q_idx];
+	vq = &blk_vq->vq;
+
+	avail_idx = blk_vq->last_avail_idx;
+	avail_wrap_counter = blk_vq->avail_wrap_counter;
+	used_idx = blk_vq->last_used_idx;
+	used_wrap_counter = blk_vq->used_wrap_counter;
+
+	for (i = 0; i < MAX_TASK; i++) {
+		task[i] = rte_zmalloc(NULL, sizeof(struct vhost_blk_task), 0);
+		assert(task[i] != NULL);
+		task[i]->vq = vq;
+		task[i]->bdev = ctrlr->bdev;
+	}
+
+	i = 0;
+	j = 0;
+	while (j++ < 1000000 &&
+	       descriptor_is_available(vq, avail_idx, avail_wrap_counter) &&
+	       i < MAX_TASK) {
+		task[i]->head_idx = avail_idx;
+		task[i]->desc_packed = &vq->desc_packed[task[i]->head_idx];
+		task[i]->iovs_cnt = 0;
+		task[i]->data_len = 0;
+		task[i]->req = NULL;
+		task[i]->status = NULL;
+
+		/* does not support indirect descriptors */
+		assert((task[i]->desc_packed->flags &
+			VRING_DESC_F_INDIRECT) == 0);
+
+		chunck_len = task[i]->desc_packed->len;
+		task[i]->req = (void *)(uintptr_t)gpa_to_vva(task[i]->bdev->vid,
+			task[i]->desc_packed->addr, &chunck_len);
+		if (!task[i]->req || chunck_len != task[i]->desc_packed->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			goto free_task;
+		}
+
+		task[i]->desc_packed = descriptor_get_next_packed(task[i]->vq,
+								  &avail_idx);
+		assert(task[i]->desc_packed != NULL);
+		if (!descriptor_has_next_packed(task[i]->desc_packed)) {
+			task[i]->dxfer_dir = BLK_DIR_NONE;
+			task[i]->last_idx = avail_idx % vq->size;
+			chunck_len = task[i]->desc_packed->len;
+			task[i]->status = (void *)(uintptr_t)
+				gpa_to_vva(task[i]->bdev->vid,
+						task[i]->desc_packed->addr,
+						&chunck_len);
+			if (!task[i]->status ||
+				chunck_len != task[i]->desc_packed->len) {
+				fprintf(stderr,
+					"failed to translate desc address.\n");
+				goto free_task;
+			}
+		} else {
+			task[i]->readtype =
+				descriptor_is_wr_packed(task[i]->desc_packed);
+			vhost_process_payload_chain_packed(task[i], &avail_idx);
+		}
+
+		task[i]->buffer_id = vq->desc_packed[task[i]->last_idx].id;
+		rte_vhost_set_inflight_desc_packed(ctrlr->bdev->vid, q_idx,
+						   task[i]->head_idx,
+						   task[i]->last_idx,
+						   &task[i]->inflight_idx);
+
+		if (++avail_idx >= vq->size) {
+			avail_idx -= vq->size;
+			avail_wrap_counter = !avail_wrap_counter;
+		}
+		blk_vq->last_avail_idx = avail_idx;
+		blk_vq->avail_wrap_counter = avail_wrap_counter;
+
+		i++;
+		if (g_should_stop == 1)
+			goto free_task;
+	}
+
+	for (j = 0; j < i; j++) {
+		ret = vhost_bdev_process_blk_commands(ctrlr->bdev, task[j]);
+		if (ret) {
+			/* invalid response */
+			*task[j]->status = VIRTIO_BLK_S_IOERR;
+		} else {
+			/* successfully */
+			*task[j]->status = VIRTIO_BLK_S_OK;
+		}
+
+		submit_completion_packed(task[j], q_idx, &used_idx,
+						&used_wrap_counter);
+		blk_vq->last_used_idx = used_idx;
+		blk_vq->used_wrap_counter = used_wrap_counter;
+
+		if (g_should_stop == 1)
+			goto free_task;
+	}
+
+free_task:
+	for (j = 0; j < MAX_TASK; j++)
+		rte_free(task[j]);
+}
+
+static struct vring_desc *
+descriptor_get_next_split(struct vring_desc *vq_desc,
+				   struct vring_desc *cur_desc)
+{
+	return &vq_desc[cur_desc->next];
+}
+
+static bool
+descriptor_has_next_split(struct vring_desc *cur_desc)
+{
+	return !!(cur_desc->flags & VRING_DESC_F_NEXT);
+}
+
+static bool
+descriptor_is_wr_split(struct vring_desc *cur_desc)
+{
+	return !!(cur_desc->flags & VRING_DESC_F_WRITE);
+}
+
+static void
+vhost_process_payload_chain_split(struct vhost_blk_task *task)
+{
+	void *data;
+	uint64_t chunck_len;
+
+	task->iovs_cnt = 0;
+
+	do {
+		chunck_len = task->desc_split->len;
+		data = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+						     task->desc_split->addr,
+						     &chunck_len);
+		if (!data || chunck_len != task->desc_split->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			return;
+		}
+
+		task->iovs[task->iovs_cnt].iov_base = data;
+		task->iovs[task->iovs_cnt].iov_len = task->desc_split->len;
+		task->data_len += task->desc_split->len;
+		task->iovs_cnt++;
+		task->desc_split =
+		descriptor_get_next_split(task->vq->desc, task->desc_split);
+	} while (descriptor_has_next_split(task->desc_split));
+
+	chunck_len = task->desc_split->len;
+	task->status = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+						     task->desc_split->addr,
+						     &chunck_len);
+	if (!task->status || chunck_len != task->desc_split->len)
+		fprintf(stderr, "failed to translate desc address.\n");
+}
+
+static void
+submit_completion_split(struct vhost_blk_task *task, uint32_t vid,
+	uint32_t q_idx)
+{
+	struct rte_vhost_vring *vq;
+	struct vring_used *used;
+
+	vq = task->vq;
+	used = vq->used;
+
+	rte_vhost_set_last_inflight_io_split(vid, q_idx, task->req_idx);
+
+	/* Fill out the next entry in the "used" ring.  id = the
+	 * index of the descriptor that contained the blk request.
+	 * len = the total amount of data transferred for the blk
+	 * request. We must report the correct len, for variable
+	 * length blk CDBs, where we may return less data than
+	 * allocated by the guest VM.
+	 */
+	used->ring[used->idx & (vq->size - 1)].id = task->req_idx;
+	used->ring[used->idx & (vq->size - 1)].len = task->data_len;
+	rte_compiler_barrier();
+	used->idx++;
+	rte_compiler_barrier();
+
+	rte_vhost_clr_inflight_desc_split(vid, q_idx, used->idx, task->req_idx);
+
+	/* Send an interrupt back to the guest VM so that it knows
+	 * a completion is ready to be processed.
+	 */
+	rte_vhost_vring_call(task->bdev->vid, q_idx);
+}
+
+static void
+submit_inflight_vq_split(struct vhost_blk_ctrlr *ctrlr,
+	uint32_t q_idx)
+{
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_ring_inflight *inflight_vq;
+	struct rte_vhost_resubmit_info *resubmit_inflight;
+	struct rte_vhost_resubmit_desc *resubmit_list;
+	struct vhost_blk_task *task;
+	int req_idx;
+	uint64_t chunck_len;
+	int ret;
+
+	blk_vq = &ctrlr->bdev->queues[q_idx];
+	inflight_vq = &blk_vq->inflight_vq;
+	resubmit_inflight = inflight_vq->resubmit_inflight;
+	resubmit_list = resubmit_inflight->resubmit_list;
+
+	task = rte_zmalloc(NULL, sizeof(*task), 0);
+	assert(task != NULL);
+
+	task->ctrlr = ctrlr;
+	task->bdev = ctrlr->bdev;
+	task->vq = &blk_vq->vq;
+
+	while (resubmit_inflight->resubmit_num-- > 0) {
+		req_idx = resubmit_list[resubmit_inflight->resubmit_num].index;
+		task->req_idx = req_idx;
+		task->desc_split = &task->vq->desc[task->req_idx];
+		task->iovs_cnt = 0;
+		task->data_len = 0;
+		task->req = NULL;
+		task->status = NULL;
+
+		/* does not support indirect descriptors */
+		assert(task->desc_split != NULL);
+		assert((task->desc_split->flags & VRING_DESC_F_INDIRECT) == 0);
+
+		chunck_len = task->desc_split->len;
+		task->req = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+				task->desc_split->addr, &chunck_len);
+		if (!task->req || chunck_len != task->desc_split->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			rte_free(task);
+			return;
+		}
+
+		task->desc_split = descriptor_get_next_split(task->vq->desc,
+							     task->desc_split);
+		if (!descriptor_has_next_split(task->desc_split)) {
+			task->dxfer_dir = BLK_DIR_NONE;
+			chunck_len = task->desc_split->len;
+			task->status = (void *)(uintptr_t)
+				       gpa_to_vva(task->bdev->vid,
+						  task->desc_split->addr,
+						  &chunck_len);
+			if (!task->status ||
+				chunck_len != task->desc_split->len) {
+				fprintf(stderr,
+					"failed to translate desc address.\n");
+				rte_free(task);
+				return;
+			}
+		} else {
+			task->readtype =
+				descriptor_is_wr_split(task->desc_split);
+			vhost_process_payload_chain_split(task);
+		}
+
+		ret = vhost_bdev_process_blk_commands(ctrlr->bdev, task);
+		if (ret) {
+			/* invalid response */
+			*task->status = VIRTIO_BLK_S_IOERR;
+		} else {
+			/* successfully */
+			*task->status = VIRTIO_BLK_S_OK;
+		}
+		submit_completion_split(task, ctrlr->bdev->vid, q_idx);
+	}
+
+	rte_free(task);
+}
+
+static void
+process_requestq_split(struct vhost_blk_ctrlr *ctrlr, uint32_t q_idx)
+{
+	int ret;
+	int req_idx;
+	uint16_t last_idx;
+	uint64_t chunck_len;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_vring *vq;
+	struct vhost_blk_task *task;
+
+	blk_vq = &ctrlr->bdev->queues[q_idx];
+	vq = &blk_vq->vq;
+
+	task = rte_zmalloc(NULL, sizeof(*task), 0);
+	assert(task != NULL);
+	task->ctrlr = ctrlr;
+	task->bdev = ctrlr->bdev;
+	task->vq = vq;
+
+	while (vq->avail->idx != blk_vq->last_avail_idx) {
+		last_idx = blk_vq->last_avail_idx & (vq->size - 1);
+		req_idx = vq->avail->ring[last_idx];
+		task->req_idx = req_idx;
+		task->desc_split = &task->vq->desc[task->req_idx];
+		task->iovs_cnt = 0;
+		task->data_len = 0;
+		task->req = NULL;
+		task->status = NULL;
+
+		rte_vhost_set_inflight_desc_split(ctrlr->bdev->vid, q_idx,
+							task->req_idx);
+
+		/* does not support indirect descriptors */
+		assert((task->desc_split->flags & VRING_DESC_F_INDIRECT) == 0);
+
+		chunck_len = task->desc_split->len;
+		task->req = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
+				task->desc_split->addr, &chunck_len);
+		if (!task->req || chunck_len != task->desc_split->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			rte_free(task);
+			return;
+		}
+
+		task->desc_split = descriptor_get_next_split(task->vq->desc,
+							     task->desc_split);
+		if (!descriptor_has_next_split(task->desc_split)) {
+			task->dxfer_dir = BLK_DIR_NONE;
+			chunck_len = task->desc_split->len;
+			task->status = (void *)(uintptr_t)
+					      gpa_to_vva(task->bdev->vid,
+							 task->desc_split->addr,
+							 &chunck_len);
+			if (!task->status ||
+				chunck_len != task->desc_split->len) {
+				fprintf(stderr,
+					"failed to translate desc address.\n");
+				rte_free(task);
+				return;
+			}
+		} else {
+			task->readtype =
+				descriptor_is_wr_split(task->desc_split);
+			vhost_process_payload_chain_split(task);
+		}
+		blk_vq->last_avail_idx++;
+
+		ret = vhost_bdev_process_blk_commands(ctrlr->bdev, task);
+		if (ret) {
+			/* invalid response */
+			*task->status = VIRTIO_BLK_S_IOERR;
+		} else {
+			/* successfully */
+			*task->status = VIRTIO_BLK_S_OK;
+		}
+
+		submit_completion_split(task, ctrlr->bdev->vid, q_idx);
+	}
+
+	rte_free(task);
+}
+
+static void
+process_batch_requests_split(struct vhost_blk_ctrlr *ctrlr,
+	uint32_t q_idx)
+{
+	int i, j;
+	int ret;
+	int req_idx;
+	uint16_t last_idx;
+	uint64_t chunck_len;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_vring *vq;
+	struct vhost_blk_task *task[MAX_TASK] = {NULL};
+
+	blk_vq = &ctrlr->bdev->queues[q_idx];
+	vq = &blk_vq->vq;
+
+	for (i = 0; i < MAX_TASK; i++) {
+		task[i] = rte_zmalloc(NULL, sizeof(struct vhost_blk_task), 0);
+		assert(task[i] != NULL);
+		task[i]->ctrlr = ctrlr;
+		task[i]->bdev = ctrlr->bdev;
+		task[i]->vq = vq;
+	}
+
+	i = 0;
+	j = 0;
+	while (j++ < 1000000 && vq->avail->idx != blk_vq->last_avail_idx &&
+	       i < MAX_TASK) {
+		last_idx = blk_vq->last_avail_idx & (vq->size - 1);
+		req_idx = vq->avail->ring[last_idx];
+		task[i]->req_idx = req_idx;
+		task[i]->desc_split = &task[i]->vq->desc[task[i]->req_idx];
+		task[i]->iovs_cnt = 0;
+		task[i]->data_len = 0;
+		task[i]->req = NULL;
+		task[i]->status = NULL;
+
+		rte_vhost_set_inflight_desc_split(ctrlr->bdev->vid, q_idx,
+						  task[i]->req_idx);
+
+		/* does not support indirect descriptors */
+		assert((task[i]->desc_split->flags &
+			VRING_DESC_F_INDIRECT) == 0);
+
+		chunck_len = task[i]->desc_split->len;
+		task[i]->req = (void *)(uintptr_t)gpa_to_vva(task[i]->bdev->vid,
+				task[i]->desc_split->addr, &chunck_len);
+		if (!task[i]->req || chunck_len != task[i]->desc_split->len) {
+			fprintf(stderr, "failed to translate desc address.\n");
+			goto free_tasks;
+		}
+
+		task[i]->desc_split = descriptor_get_next_split(
+					task[i]->vq->desc, task[i]->desc_split);
+		if (!descriptor_has_next_split(task[i]->desc_split)) {
+			task[i]->dxfer_dir = BLK_DIR_NONE;
+			chunck_len = task[i]->desc_split->len;
+			task[i]->status = (void *)(uintptr_t)
+				gpa_to_vva(task[i]->bdev->vid,
+						task[i]->desc_split->addr,
+						&chunck_len);
+			if (!task[i]->status ||
+				chunck_len != task[i]->desc_split->len) {
+				fprintf(stderr,
+					"failed to translate desc address.\n");
+				goto free_tasks;
+			}
+		} else {
+			task[i]->readtype =
+				descriptor_is_wr_split(task[i]->desc_split);
+			vhost_process_payload_chain_split(task[i]);
+		}
+		blk_vq->last_avail_idx++;
+		i++;
+
+		if (g_should_stop == 1)
+			goto free_tasks;
+	}
+
+	for (j = 0; j < i; j++) {
+		ret = vhost_bdev_process_blk_commands(ctrlr->bdev, task[j]);
+		if (ret) {
+			/* invalid response */
+			*task[j]->status = VIRTIO_BLK_S_IOERR;
+		} else {
+			/* successfully */
+			*task[j]->status = VIRTIO_BLK_S_OK;
+		}
+
+		submit_completion_split(task[j], ctrlr->bdev->vid, q_idx);
+
+		if (g_should_stop == 1)
+			goto free_tasks;
+	}
+
+free_tasks:
+	for (i = 0; i < MAX_TASK; i++)
+		rte_free(task[i]);
+}
+
+static void *
+ctrlr_worker(void *arg)
+{
+	struct vhost_blk_ctrlr *ctrlr = (struct vhost_blk_ctrlr *)arg;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_ring_inflight *inflight_vq;
+	cpu_set_t cpuset;
+	pthread_t thread;
+	int i, ret;
+
+	fprintf(stdout, "Ctrlr Worker Thread start\n");
+
+	if (ctrlr == NULL || ctrlr->bdev == NULL) {
+		fprintf(stderr,
+			"%s: Error, invalid argument passed to worker thread\n",
+			__func__);
+		exit(0);
+	}
+
+	thread = pthread_self();
+	CPU_ZERO(&cpuset);
+	CPU_SET(0, &cpuset);
+	pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
+
+	for (i = 0; i < NUM_OF_BLK_QUEUES; i++) {
+		blk_vq = &ctrlr->bdev->queues[i];
+		inflight_vq = &blk_vq->inflight_vq;
+		if (inflight_vq->resubmit_inflight != NULL &&
+		    inflight_vq->resubmit_inflight->resubmit_num != 0) {
+			if (ctrlr->packed_ring)
+				submit_inflight_vq_packed(ctrlr, i);
+			else
+				submit_inflight_vq_split(ctrlr, i);
+		}
+	}
+
+	while (!g_should_stop && ctrlr->bdev != NULL) {
+		for (i = 0; i < NUM_OF_BLK_QUEUES; i++) {
+			if (ctrlr->packed_ring)
+				process_requestq_packed(ctrlr, i);
+			else
+				process_requestq_split(ctrlr, i);
+		}
+	}
+
+	g_should_stop = 2;
+	fprintf(stdout, "Ctrlr Worker Thread Exiting\n");
+	sem_post(&exit_sem);
+	return NULL;
+}
+
+static int
+new_device(int vid)
+{
+	char path[PATH_MAX];
+	struct vhost_blk_ctrlr *ctrlr;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_vring *vq;
+	pthread_t tid;
+	int i, ret;
+	uint64_t features;
+
+	ctrlr = vhost_blk_ctrlr_find(path);
+	if (!ctrlr) {
+		fprintf(stderr, "Controller is not ready\n");
+		return -1;
+	}
+
+	if (ctrlr->started)
+		return 0;
+
+	ctrlr->bdev->vid = vid;
+	ctrlr->packed_ring = rte_vhost_vq_is_packed(vid);
+
+	ret = rte_vhost_get_mem_table(vid, &ctrlr->mem);
+	if (ret)
+		fprintf(stderr, "Get Controller memory region failed\n");
+	assert(ctrlr->mem != NULL);
+
+	/* Disable Notifications and init last idx */
+	for (i = 0; i < NUM_OF_BLK_QUEUES; i++) {
+		blk_vq = &ctrlr->bdev->queues[i];
+		vq = &blk_vq->vq;
+
+		ret = rte_vhost_get_vhost_vring(ctrlr->bdev->vid, i, vq);
+		assert(ret == 0);
+
+		ret = rte_vhost_get_vring_base(ctrlr->bdev->vid, i,
+					       &blk_vq->last_avail_idx,
+					       &blk_vq->last_used_idx);
+		assert(ret == 0);
+
+		ret = rte_vhost_get_vhost_ring_inflight(ctrlr->bdev->vid, i,
+							&blk_vq->inflight_vq);
+		assert(ret == 0);
+
+		if (ctrlr->packed_ring) {
+			/* for the reconnection */
+			ret = rte_vhost_get_vring_base_from_inflight(
+				ctrlr->bdev->vid, i,
+				&blk_vq->last_avail_idx,
+				&blk_vq->last_used_idx);
+			assert(ret == 0);
+
+			blk_vq->avail_wrap_counter = blk_vq->last_avail_idx &
+				(1 << 15);
+			blk_vq->last_avail_idx = blk_vq->last_avail_idx &
+				0x7fff;
+			blk_vq->used_wrap_counter = blk_vq->last_used_idx &
+				(1 << 15);
+			blk_vq->last_used_idx = blk_vq->last_used_idx &
+				0x7fff;
+		}
+
+		rte_vhost_enable_guest_notification(vid, i, 0);
+	}
+
+	/* start polling vring */
+	g_should_stop = 0;
+	fprintf(stdout, "New Device %s, Device ID %d\n", path, vid);
+	if (pthread_create(&tid, NULL, &ctrlr_worker, ctrlr) < 0) {
+		fprintf(stderr, "Worker Thread Started Failed\n");
+		return -1;
+	}
+
+	/* device has been started */
+	ctrlr->started = 1;
+	pthread_detach(tid);
+	return 0;
+}
+
+static void
+destroy_device(int vid)
+{
+	char path[PATH_MAX];
+	struct vhost_blk_ctrlr *ctrlr;
+	struct vhost_blk_queue *blk_vq;
+	struct rte_vhost_vring *vq;
+	int i, ret;
+
+	ret = rte_vhost_get_ifname(vid, path, PATH_MAX);
+	if (ret) {
+		fprintf(stderr, "Destroy Ctrlr Failed\n");
+		return;
+	}
+	fprintf(stdout, "Destroy %s Device ID %d\n", path, vid);
+	ctrlr = vhost_blk_ctrlr_find(path);
+	if (!ctrlr) {
+		fprintf(stderr, "Destroy Ctrlr Failed\n");
+		return;
+	}
+
+	if (!ctrlr->started)
+		return;
+
+	g_should_stop = 1;
+	while (g_should_stop != 2)
+		;
+
+	for (i = 0; i < NUM_OF_BLK_QUEUES; i++) {
+		blk_vq = &ctrlr->bdev->queues[i];
+		if (ctrlr->packed_ring) {
+			blk_vq->last_avail_idx |= (blk_vq->avail_wrap_counter <<
+				15);
+			blk_vq->last_used_idx |= (blk_vq->used_wrap_counter <<
+				15);
+		}
+		rte_vhost_set_vring_base(ctrlr->bdev->vid, i,
+					 blk_vq->last_avail_idx,
+					 blk_vq->last_used_idx);
+	}
+
+	free(ctrlr->mem);
+
+	ctrlr->started = 0;
+	sem_wait(&exit_sem);
+}
+
+static int
+new_connection(int vid)
+{
+	/* extend the proper features for block device */
+	vhost_session_install_rte_compat_hooks(vid);
+}
+
+struct vhost_device_ops vhost_blk_device_ops = {
+	.new_device =  new_device,
+	.destroy_device = destroy_device,
+	.new_connection = new_connection,
+};
+
+static struct vhost_block_dev *
+vhost_blk_bdev_construct(const char *bdev_name,
+	const char *bdev_serial, uint32_t blk_size, uint64_t blk_cnt,
+	bool wce_enable)
+{
+	struct vhost_block_dev *bdev;
+
+	bdev = rte_zmalloc(NULL, sizeof(*bdev), RTE_CACHE_LINE_SIZE);
+	if (!bdev)
+		return NULL;
+
+	strncpy(bdev->name, bdev_name, sizeof(bdev->name));
+	strncpy(bdev->product_name, bdev_serial, sizeof(bdev->product_name));
+	bdev->blocklen = blk_size;
+	bdev->blockcnt = blk_cnt;
+	bdev->write_cache = wce_enable;
+
+	fprintf(stdout, "blocklen=%d, blockcnt=%d\n", bdev->blocklen,
+		bdev->blockcnt);
+
+	/* use memory as disk storage space */
+	bdev->data = rte_zmalloc(NULL, blk_cnt * blk_size, 0);
+	if (!bdev->data) {
+		fprintf(stderr, "no enough reserved huge memory for disk\n");
+		free(bdev);
+		return NULL;
+	}
+
+	return bdev;
+}
+
+static struct vhost_blk_ctrlr *
+vhost_blk_ctrlr_construct(const char *ctrlr_name)
+{
+	int ret;
+	struct vhost_blk_ctrlr *ctrlr;
+	char *path;
+	char cwd[PATH_MAX];
+
+	/* always use current directory */
+	path = getcwd(cwd, PATH_MAX);
+	if (!path) {
+		fprintf(stderr, "Cannot get current working directory\n");
+		return NULL;
+	}
+	snprintf(dev_pathname, sizeof(dev_pathname), "%s/%s", path, ctrlr_name);
+
+	if (access(dev_pathname, F_OK) != -1) {
+		if (unlink(dev_pathname) != 0)
+			rte_exit(EXIT_FAILURE, "Cannot remove %s.\n",
+				 dev_pathname);
+	}
+
+	if (rte_vhost_driver_register(dev_pathname, 0) != 0) {
+		fprintf(stderr, "socket %s already exists\n", dev_pathname);
+		return NULL;
+	}
+
+	ret = rte_vhost_driver_set_features(dev_pathname, VHOST_BLK_FEATURES);
+	if (ret != 0) {
+		fprintf(stderr, "Set vhost driver features failed\n");
+		rte_vhost_driver_unregister(dev_pathname);
+		return NULL;
+	}
+
+	/* set proper features */
+	vhost_dev_install_rte_compat_hooks(dev_pathname);
+
+	ctrlr = rte_zmalloc(NULL, sizeof(*ctrlr), RTE_CACHE_LINE_SIZE);
+	if (!ctrlr) {
+		rte_vhost_driver_unregister(dev_pathname);
+		return NULL;
+	}
+
+	/* hardcoded block device information with 128MiB */
+	ctrlr->bdev = vhost_blk_bdev_construct("malloc0", "vhost_blk_malloc0",
+						4096, 32768, 0);
+	if (!ctrlr->bdev) {
+		rte_free(ctrlr);
+		rte_vhost_driver_unregister(dev_pathname);
+		return NULL;
+	}
+
+	rte_vhost_driver_callback_register(dev_pathname,
+					   &vhost_blk_device_ops);
+
+	return ctrlr;
+}
+
+static void
+signal_handler(__rte_unused int signum)
+{
+	struct vhost_blk_ctrlr *ctrlr;
+
+	if (access(dev_pathname, F_OK) == 0)
+		unlink(dev_pathname);
+
+	g_should_stop = 1;
+
+	while (g_should_stop != 2)
+		;
+
+	ctrlr = vhost_blk_ctrlr_find(NULL);
+	if (ctrlr != NULL) {
+		if (ctrlr->bdev != NULL) {
+			rte_free(ctrlr->bdev->data);
+			rte_free(ctrlr->bdev);
+		}
+		rte_free(ctrlr);
+	}
+
+	rte_vhost_driver_unregister(dev_pathname);
+	exit(0);
+}
+
+int main(int argc, char *argv[])
+{
+	int ret;
+
+	signal(SIGINT, signal_handler);
+
+	/* init EAL */
+	ret = rte_eal_init(argc, argv);
+	if (ret < 0)
+		rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
+
+	g_vhost_ctrlr = vhost_blk_ctrlr_construct("vhost.socket");
+	if (g_vhost_ctrlr == NULL) {
+		fprintf(stderr, "Construct vhost blk controller failed\n");
+		return 0;
+	}
+
+	if (sem_init(&exit_sem, 0, 0) < 0) {
+		fprintf(stderr, "Error init exit_sem\n");
+		return -1;
+	}
+
+	rte_vhost_driver_start(dev_pathname);
+
+	/* loop for exit the application */
+	while (1)
+		sleep(1);
+
+	return 0;
+}
+
diff --git a/examples/vhost_blk/vhost_blk.h b/examples/vhost_blk/vhost_blk.h
new file mode 100644
index 000000000..f40d49f46
--- /dev/null
+++ b/examples/vhost_blk/vhost_blk.h
@@ -0,0 +1,116 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2010-2017 Intel Corporation
+ */
+
+#ifndef _VHOST_BLK_H_
+#define _VHOST_BLK_H_
+
+#include <stdio.h>
+#include <sys/uio.h>
+#include <stdint.h>
+#include <linux/virtio_blk.h>
+#include <linux/virtio_ring.h>
+
+#include <rte_vhost.h>
+
+struct vhost_blk_queue {
+	struct rte_vhost_vring vq;
+	struct rte_vhost_ring_inflight inflight_vq;
+	uint16_t last_avail_idx;
+	uint16_t last_used_idx;
+	bool avail_wrap_counter;
+	bool used_wrap_counter;
+};
+
+#define NUM_OF_BLK_QUEUES 1
+
+#ifndef VIRTIO_F_RING_PACKED
+#define VIRTIO_F_RING_PACKED 34
+#endif
+
+#define min(a, b) (((a) < (b)) ? (a) : (b))
+
+struct vhost_block_dev {
+	/** ID for vhost library. */
+	int vid;
+	/** Queues for the block device */
+	struct vhost_blk_queue queues[NUM_OF_BLK_QUEUES];
+	/** Unique name for this block device. */
+	char name[64];
+
+	/** Unique product name for this kind of block device. */
+	char product_name[256];
+
+	/** Size in bytes of a logical block for the backend */
+	uint32_t blocklen;
+
+	/** Number of blocks */
+	uint64_t blockcnt;
+
+	/** write cache enabled, not used at the moment */
+	int write_cache;
+
+	/** use memory as disk storage space */
+	uint8_t *data;
+};
+
+struct vhost_blk_ctrlr {
+	uint8_t started;
+	uint8_t packed_ring;
+	uint8_t need_restart;
+	/** Only support 1 LUN for the example */
+	struct vhost_block_dev *bdev;
+	/** VM memory region */
+	struct rte_vhost_memory *mem;
+} __rte_cache_aligned;
+
+#define VHOST_BLK_MAX_IOVS 128
+
+enum blk_data_dir {
+	BLK_DIR_NONE = 0,
+	BLK_DIR_TO_DEV = 1,
+	BLK_DIR_FROM_DEV = 2,
+};
+
+struct vhost_blk_task {
+	uint8_t readtype;
+	uint8_t req_idx;
+	uint16_t head_idx;
+	uint16_t last_idx;
+	uint16_t inflight_idx;
+	uint16_t buffer_id;
+	uint32_t dxfer_dir;
+	uint32_t data_len;
+	struct virtio_blk_outhdr *req;
+
+	volatile uint8_t *status;
+
+	struct iovec iovs[VHOST_BLK_MAX_IOVS];
+	uint32_t iovs_cnt;
+	struct vring_packed_desc *desc_packed;
+	struct vring_desc *desc_split;
+	struct rte_vhost_vring *vq;
+	struct vhost_block_dev *bdev;
+	struct vhost_blk_ctrlr *ctrlr;
+};
+
+struct inflight_blk_task {
+	struct vhost_blk_task blk_task;
+	struct rte_vhost_inflight_desc_packed *inflight_desc;
+	struct rte_vhost_inflight_info_packed *inflight_packed;
+};
+
+struct vhost_blk_ctrlr *g_vhost_ctrlr;
+struct vhost_device_ops vhost_blk_device_ops;
+int g_should_stop;
+
+int vhost_bdev_process_blk_commands(struct vhost_block_dev *bdev,
+				     struct vhost_blk_task *task);
+
+void vhost_session_install_rte_compat_hooks(uint32_t vid);
+
+void vhost_dev_install_rte_compat_hooks(const char *path);
+
+struct vhost_blk_ctrlr *vhost_blk_ctrlr_find(const char *ctrlr_name);
+
+#endif /* _VHOST_blk_H_ */
diff --git a/examples/vhost_blk/vhost_blk_compat.c b/examples/vhost_blk/vhost_blk_compat.c
new file mode 100644
index 000000000..8035051d2
--- /dev/null
+++ b/examples/vhost_blk/vhost_blk_compat.c
@@ -0,0 +1,195 @@
+// SPDX-License-Identifier: BSD-3-Clause
+// Copyright(c) 2010-2017 Intel Corporation
+
+#ifndef _VHOST_BLK_COMPAT_H_
+#define _VHOST_BLK_COMPAT_H_
+
+#include <sys/uio.h>
+#include <stdint.h>
+#include <linux/virtio_blk.h>
+#include <linux/virtio_ring.h>
+
+#include <rte_vhost.h>
+#include "vhost_blk.h"
+#include "blk_spec.h"
+
+#define VHOST_MAX_VQUEUES	256
+#define SPDK_VHOST_MAX_VQ_SIZE	1024
+
+#define VHOST_USER_GET_CONFIG 24
+#define VHOST_USER_SET_CONFIG 25
+
+static int
+vhost_blk_get_config(struct vhost_block_dev *bdev, uint8_t *config,
+			  uint32_t len)
+{
+	struct virtio_blk_config blkcfg;
+	uint32_t blk_size;
+	uint64_t blkcnt;
+
+	if (bdev == NULL) {
+		/* We can't just return -1 here as this GET_CONFIG message might
+		 * be caused by a QEMU VM reboot. Returning -1 will indicate an
+		 * error to QEMU, who might then decide to terminate itself.
+		 * We don't want that. A simple reboot shouldn't break the
+		 * system.
+		 *
+		 * Presenting a block device with block size 0 and block count 0
+		 * doesn't cause any problems on QEMU side and the virtio-pci
+		 * device is even still available inside the VM, but there will
+		 * be no block device created for it - the kernel drivers will
+		 * silently reject it.
+		 */
+		blk_size = 0;
+		blkcnt = 0;
+	} else {
+		blk_size = bdev->blocklen;
+		blkcnt = bdev->blockcnt;
+	}
+
+	memset(&blkcfg, 0, sizeof(blkcfg));
+	blkcfg.blk_size = blk_size;
+	/* minimum I/O size in blocks */
+	blkcfg.min_io_size = 1;
+	/* expressed in 512 Bytes sectors */
+	blkcfg.capacity = (blkcnt * blk_size) / 512;
+	/* QEMU can overwrite this value when started */
+	blkcfg.num_queues = VHOST_MAX_VQUEUES;
+
+	fprintf(stdout, "block device:blk_size = %d, blkcnt = %d\n", blk_size,
+		blkcnt);
+
+	memcpy(config, &blkcfg, min(len, sizeof(blkcfg)));
+
+	return 0;
+}
+
+static enum rte_vhost_msg_result
+extern_vhost_pre_msg_handler(int vid, void *_msg)
+{
+	char path[PATH_MAX];
+	struct vhost_blk_ctrlr *ctrlr;
+	struct vhost_user_msg *msg = _msg;
+	int ret;
+
+	ret = rte_vhost_get_ifname(vid, path, PATH_MAX);
+	if (ret) {
+		fprintf(stderr, "Cannot get socket name\n");
+		return -1;
+	}
+
+	ctrlr = vhost_blk_ctrlr_find(path);
+	if (!ctrlr) {
+		fprintf(stderr, "Controller is not ready\n");
+		return -1;
+	}
+
+	switch (msg->request) {
+	case VHOST_USER_GET_VRING_BASE:
+		if (!g_should_stop && ctrlr->started)
+			vhost_blk_device_ops.destroy_device(vid);
+		break;
+	case VHOST_USER_SET_VRING_BASE:
+	case VHOST_USER_SET_VRING_ADDR:
+	case VHOST_USER_SET_VRING_NUM:
+	case VHOST_USER_SET_VRING_KICK:
+		if (!g_should_stop && ctrlr->started)
+			vhost_blk_device_ops.destroy_device(vid);
+		break;
+	case VHOST_USER_SET_VRING_CALL:
+	case VHOST_USER_SET_MEM_TABLE:
+		if (!g_should_stop && ctrlr->started) {
+			vhost_blk_device_ops.destroy_device(vid);
+			ctrlr->need_restart = 1;
+		}
+		break;
+	case VHOST_USER_GET_CONFIG: {
+		int rc = 0;
+
+		rc = vhost_blk_get_config(ctrlr->bdev,
+					  msg->payload.cfg.region,
+					  msg->payload.cfg.size);
+		if (rc != 0)
+			msg->size = 0;
+
+		return RTE_VHOST_MSG_RESULT_REPLY;
+	}
+	case VHOST_USER_SET_CONFIG:
+	default:
+		break;
+	}
+
+	return RTE_VHOST_MSG_RESULT_NOT_HANDLED;
+}
+
+static enum rte_vhost_msg_result
+extern_vhost_post_msg_handler(int vid, void *_msg)
+{
+	char path[PATH_MAX];
+	struct vhost_blk_ctrlr *ctrlr;
+	struct vhost_user_msg *msg = _msg;
+	int ret;
+
+	ret = rte_vhost_get_ifname(vid, path, PATH_MAX);
+	if (ret) {
+		fprintf(stderr, "Cannot get socket name\n");
+		return -1;
+	}
+
+	ctrlr = vhost_blk_ctrlr_find(path);
+	if (!ctrlr) {
+		fprintf(stderr, "Controller is not ready\n");
+		return -1;
+	}
+
+	if (ctrlr->need_restart) {
+		vhost_blk_device_ops.new_device(vid);
+		ctrlr->need_restart = 0;
+	}
+
+	switch (msg->request) {
+	case VHOST_USER_SET_FEATURES:
+		break;
+	case VHOST_USER_SET_VRING_KICK:
+		/* vhost-user spec tells us to start polling a queue after
+		 * receiving its SET_VRING_KICK message. Let's do it!
+		 */
+		if (g_should_stop && !ctrlr->started)
+			vhost_blk_device_ops.new_device(vid);
+		break;
+	default:
+		break;
+	}
+
+	return RTE_VHOST_MSG_RESULT_NOT_HANDLED;
+}
+
+struct rte_vhost_user_extern_ops g_extern_vhost_ops = {
+	.pre_msg_handle = extern_vhost_pre_msg_handler,
+	.post_msg_handle = extern_vhost_post_msg_handler,
+};
+
+void
+vhost_session_install_rte_compat_hooks(uint32_t vid)
+{
+	int rc;
+
+	rc = rte_vhost_extern_callback_register(vid, &g_extern_vhost_ops, NULL);
+	if (rc != 0)
+		fprintf(stderr,
+			"rte_vhost_extern_callback_register() failed for vid = %d\n",
+			vid);
+}
+
+void
+vhost_dev_install_rte_compat_hooks(const char *path)
+{
+	uint64_t protocol_features = 0;
+
+	rte_vhost_driver_get_protocol_features(path, &protocol_features);
+	protocol_features |= (1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
+	protocol_features |= (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD);
+	rte_vhost_driver_set_protocol_features(path, protocol_features);
+}
+
+#endif
-- 
2.17.2

Re: [dpdk-dev] [PATCH v7 00/10] vhost: support inflight share memory protocol feature

[Tiwei Bie]

> [PATCH v7 00/10] vhost: support inflight share memory protocol feature

This should be v8. Please increase the version number when
sending a new series.

Thanks,
Tiwei

On Fri, Sep 20, 2019 at 08:00:52PM +0800, Jin Yu wrote:
> This patches introduces two new messages VHOST_USER_GET_INFLIGHT_FD
> and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
> buffer between qemu and backend.
> Now It can both support split and packed ring. The example code show
> how these API work. The test has passed.
> 
> How to test the example:
> 1, Qemu need two patches.
> https://patchwork.kernel.org/patch/10766813/
> https://patchwork.kernel.org/patch/10861411/(QEMU merged)
> it also needs some manual modifications:
> we should confirm that before we send get inflight we have already
> sent the set features but it seems Qemu didn't do like this. So we
> manually revise this, we can add below code in vhost_dev_get_inflight
>     int r;
> 
>     r = vhost_dev_set_features(dev, dev->log_enabled);
>     if (r < 0) {
>         return -errno;
>     }
> before get_inflight_fd.
> 2, Guest OS version >= 5.0
> 3, run the example
> 4, run the qemu with vhost-user-blk-pci.
>  eg:
>         -chardev socket,id=spdk_vhost_blk0,reconnect=1,path=xxxx\
>         -device vhost-user-blk-pci,ring_packed=on,chardev=spdk_vhost_blk0,num-queues=1\
> 5, run fio in the guest
> 6, kill the example and run again.
> 7, the fio in the guest should continue run without errors.
> 
> Jin Yu (10):
>   vhost: add the inflight description
>   vhost: add packed ring
>   vhost: add the inflight structure
>   vhost: add two new messages to support a shared buffer
>   vhost: checkout the resubmit inflight information
>   vhost: add the APIs to operate inflight ring
>   vhost: add APIs for user getting inflight ring
>   vhost: add vring functions packed ring support
>   vhost: add an API for judging vq format
>   vhost: add vhost-user-blk example which support inflight
> 
>  examples/vhost_blk/Makefile            |   67 ++
>  examples/vhost_blk/blk.c               |  125 +++
>  examples/vhost_blk/blk_spec.h          |   95 ++
>  examples/vhost_blk/meson.build         |   20 +
>  examples/vhost_blk/vhost_blk.c         | 1313 ++++++++++++++++++++++++
>  examples/vhost_blk/vhost_blk.h         |  116 +++
>  examples/vhost_blk/vhost_blk_compat.c  |  195 ++++
>  lib/librte_vhost/rte_vhost.h           |  250 ++++-
>  lib/librte_vhost/rte_vhost_version.map |    9 +
>  lib/librte_vhost/vhost.c               |  416 +++++++-
>  lib/librte_vhost/vhost.h               |   16 +
>  lib/librte_vhost/vhost_user.c          |  470 ++++++++-
>  lib/librte_vhost/vhost_user.h          |   12 +-
>  13 files changed, 3091 insertions(+), 13 deletions(-)
>  create mode 100644 examples/vhost_blk/Makefile
>  create mode 100644 examples/vhost_blk/blk.c
>  create mode 100644 examples/vhost_blk/blk_spec.h
>  create mode 100644 examples/vhost_blk/meson.build
>  create mode 100644 examples/vhost_blk/vhost_blk.c
>  create mode 100644 examples/vhost_blk/vhost_blk.h
>  create mode 100644 examples/vhost_blk/vhost_blk_compat.c
> 
> -- 
> 2.17.2
> 

Re: [dpdk-dev] [PATCH v7 10/10] vhost: add vhost-user-blk example which support inflight

[Tiwei Bie]

On Fri, Sep 20, 2019 at 08:01:02PM +0800, Jin Yu wrote:
> A vhost-user-blk example that support inflight feature. It uses the
> new APIs that introduced in the first patch, so It can show how there

s/It/it/
s/there/these/

> APIs work to support inflight feature.
> 
> Signed-off-by: Jin Yu <jin.yu@intel.com>
> ---
> V1 - add the case.
> V2 - add the rte_vhost prefix.
> V3 - add packed ring support
> ---
>  examples/vhost_blk/Makefile           |   67 ++
>  examples/vhost_blk/blk.c              |  125 +++
>  examples/vhost_blk/blk_spec.h         |   95 ++
>  examples/vhost_blk/meson.build        |   20 +
>  examples/vhost_blk/vhost_blk.c        | 1313 +++++++++++++++++++++++++
>  examples/vhost_blk/vhost_blk.h        |  116 +++
>  examples/vhost_blk/vhost_blk_compat.c |  195 ++++
>  7 files changed, 1931 insertions(+)

I met some build issues when trying this example.

examples/vhost_blk/vhost_blk.c: In function ‘descriptor_get_next_packed’:
examples/vhost_blk/vhost_blk.c:71:21: error: invalid use of undefined type ‘struct vring_packed_desc’
  if (vq->desc_packed[*idx % vq->size].flags & VIRTQ_DESC_F_NEXT) {
                     ^
examples/vhost_blk/vhost_blk.c:71:21: error: dereferencing pointer to incomplete type ‘struct vring_packed_desc’
examples/vhost_blk/vhost_blk.c:73:26: error: invalid use of undefined type ‘struct vring_packed_desc’
   return &vq->desc_packed[*idx % vq->size];
                          ^
examples/vhost_blk/vhost_blk.c: In function ‘inflight_submit_completion_packed’:
examples/vhost_blk/vhost_blk.c:165:2: warning: ‘rte_vhost_set_last_inflight_io_packed’ is deprecated: Symbol is not yet part of stable ABI [-Wdeprecated-declarations]
  ret = rte_vhost_set_last_inflight_io_packed(ctrlr->bdev->vid, q_idx,
  ^~~
In file included from examples/vhost_blk/vhost_blk.c:17:0:
x86_64-native-linuxapp-gcc/include/rte_vhost.h:810:1: note: declared here
 rte_vhost_set_last_inflight_io_packed(int vid,
 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                         ^
...

Re: [dpdk-dev] [PATCH v7 04/10] vhost: add two new messages to support a shared buffer

[Tiwei Bie]

On Fri, Sep 20, 2019 at 08:00:56PM +0800, Jin Yu wrote:
> This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD
> and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
> buffer between qemu and backend.
> 
> Signed-off-by: Lin Li <lilin24@baidu.com>
> Signed-off-by: Xun Ni <nixun@baidu.com>
> Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
> Signed-off-by: Jin Yu <jin.yu@intel.com>
> ---
>  lib/librte_vhost/vhost.h      |   7 +
>  lib/librte_vhost/vhost_user.c | 260 +++++++++++++++++++++++++++++++++-
>  lib/librte_vhost/vhost_user.h |   4 +-
>  3 files changed, 269 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
> index 884befa85..d67ba849a 100644
> --- a/lib/librte_vhost/vhost.h
> +++ b/lib/librte_vhost/vhost.h
> @@ -286,6 +286,12 @@ struct guest_page {
>  	uint64_t size;
>  };
>  
> +struct inflight_mem_info {
> +	int		fd;
> +	void		*addr;
> +	uint64_t	size;
> +};
> +
>  /**
>   * Device structure contains all configuration information relating
>   * to the device.
> @@ -303,6 +309,7 @@ struct virtio_net {
>  	uint32_t		nr_vring;
>  	int			dequeue_zero_copy;
>  	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
> +	struct inflight_mem_info *inflight_info;
>  #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
>  	char			ifname[IF_NAME_SZ];
>  	uint64_t		log_size;
> diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
> index c9e29ece8..66582a30f 100644
> --- a/lib/librte_vhost/vhost_user.c
> +++ b/lib/librte_vhost/vhost_user.c
> @@ -37,6 +37,10 @@
>  #ifdef RTE_LIBRTE_VHOST_POSTCOPY
>  #include <linux/userfaultfd.h>
>  #endif
> +#ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
> +#include <linux/memfd.h>
> +#define MEMFD_SUPPORTED
> +#endif
>  
>  #include <rte_common.h>
>  #include <rte_malloc.h>
> @@ -49,6 +53,15 @@
>  #define VIRTIO_MIN_MTU 68
>  #define VIRTIO_MAX_MTU 65535
>  
> +#define INFLIGHT_ALIGNMENT	64
> +#define INFLIGHT_VERSION	0xabcd

Why the version is 0xabcd? Shouldn't it be 1?

https://git.qemu.org/?p=qemu.git;a=blob;f=docs/interop/vhost-user.rst;h=7827b710aa0a223f6f184b229f4b5ced6420b638;hb=HEAD#l551
https://git.qemu.org/?p=qemu.git;a=blob;f=docs/interop/vhost-user.rst;h=7827b710aa0a223f6f184b229f4b5ced6420b638;hb=HEAD#l658

> +#define VIRTQUEUE_MAX_SIZE	1024
> +
> +#define CLOEXEC		0x0001U
> +
> +#define ALIGN_DOWN(n, m) ((n) / (m) * (m))
> +#define ALIGN_UP(n, m) ALIGN_DOWN((n) + (m) - 1, (m))
> +
>  static const char *vhost_message_str[VHOST_USER_MAX] = {
>  	[VHOST_USER_NONE] = "VHOST_USER_NONE",
>  	[VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
> @@ -78,6 +91,8 @@ static const char *vhost_message_str[VHOST_USER_MAX] = {
>  	[VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
>  	[VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
>  	[VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
> +	[VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
> +	[VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
>  };
>  
>  static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
> @@ -160,6 +175,21 @@ vhost_backend_cleanup(struct virtio_net *dev)
>  		dev->log_addr = 0;
>  	}
>  
> +	if (dev->inflight_info->addr) {
> +		munmap(dev->inflight_info->addr, dev->inflight_info->size);
> +		dev->inflight_info->addr = NULL;
> +	}
> +
> +	if (dev->inflight_info->fd > 0) {
> +		close(dev->inflight_info->fd);
> +		dev->inflight_info->fd = -1;
> +	}
> +
> +	if (dev->inflight_info) {
> +		free(dev->inflight_info);
> +		dev->inflight_info = NULL;
> +	}

You should check dev->inflight_info first before dereferencing it.

> +
>  	if (dev->slave_req_fd >= 0) {
>  		close(dev->slave_req_fd);
>  		dev->slave_req_fd = -1;
> @@ -1165,6 +1195,233 @@ virtio_is_ready(struct virtio_net *dev)
>  	return 1;
>  }
>  
> +static int
> +mem_create(const char *name, unsigned int flags)
> +{
> +#ifdef MEMFD_SUPPORTED
> +	return memfd_create(name, flags);
> +#else
> +	RTE_LOG(ERR, VHOST_CONFIG,
> +		"doesn't support memfd--name:%s and flag:%x\n",
> +		name, flags);

You probably don't want to always print an error when
memfd isn't available.

> +	return -1;
> +#endif
> +}
> +
> +static void *
> +inflight_mem_alloc(const char *name, size_t size, int *fd)
> +{
> +	void *ptr;
> +	int mfd = -1;
> +	char fname[20] = "/tmp/memfd-XXXXXX";
> +
> +	*fd = -1;
> +	mfd = mem_create(name, CLOEXEC);
> +	if (mfd != -1) {
> +		if (ftruncate(mfd, size) == -1) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +				"ftruncate fail for alloc inflight buffer\n");
> +			close(mfd);
> +			return NULL;
> +		}
> +	} else {
> +		mfd = mkstemp(fname);
> +		unlink(fname);
> +
> +		if (mfd == -1) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +				"mkstemp fail for alloc inflight buffer\n");
> +			return NULL;
> +		}
> +
> +		if (ftruncate(mfd, size) == -1) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +				"ftruncate fail for alloc inflight buffer\n");
> +			close(mfd);
> +			return NULL;
> +		}
> +	}
> +
> +	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
> +	if (ptr == MAP_FAILED) {
> +		RTE_LOG(ERR, VHOST_CONFIG,
> +			"mmap fail for alloc inflight buffer\n");
> +		close(mfd);
> +		return NULL;
> +	}
> +
> +	*fd = mfd;
> +	return ptr;
> +}
> +
> +static uint32_t
> +get_pervq_shm_size_split(uint16_t queue_size)
> +{
> +	return ALIGN_UP(sizeof(struct rte_vhost_inflight_desc_split) *
> +			queue_size + sizeof(uint64_t) + sizeof(uint16_t) * 4,
> +			INFLIGHT_ALIGNMENT);
> +}
> +
> +static uint32_t
> +get_pervq_shm_size_packed(uint16_t queue_size)
> +{
> +	return ALIGN_UP(sizeof(struct rte_vhost_inflight_desc_packed) *
> +			queue_size + sizeof(uint64_t) + sizeof(uint16_t) * 6 +
> +			sizeof(uint8_t) * 9, INFLIGHT_ALIGNMENT);
> +}
> +
> +static int
> +vhost_user_get_inflight_fd(struct virtio_net **pdev,
> +				     VhostUserMsg *msg,
> +				     int main_fd __rte_unused)
> +{
> +	int fd, i, j;
> +	uint64_t pervq_inflight_size, mmap_size;
> +	void *addr;
> +	uint16_t num_queues, queue_size;
> +	struct virtio_net *dev = *pdev;
> +	struct rte_vhost_inflight_info_packed *inflight_packed = NULL;
> +
> +	if (msg->size != sizeof(msg->payload.inflight)) {
> +		RTE_LOG(ERR, VHOST_CONFIG,
> +			"Invalid get_inflight_fd message size is %d",
> +			msg->size);
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +	}
> +
> +	dev->inflight_info = calloc(1, sizeof(struct inflight_mem_info));

You need to check whether dev->inflight_info has been allocated,
otherwise memory may leak.

> +	if (!dev->inflight_info) {
> +		RTE_LOG(ERR, VHOST_CONFIG,
> +			"Failed to alloc dev inflight area");
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +	}
> +
> +	num_queues = msg->payload.inflight.num_queues;
> +	queue_size = msg->payload.inflight.queue_size;
> +
> +	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
> +		msg->payload.inflight.num_queues);
> +	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
> +		msg->payload.inflight.queue_size);
> +
> +	if (vq_is_packed(dev))
> +		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
> +	else
> +		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
> +
> +	mmap_size = num_queues * pervq_inflight_size;
> +	addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
> +	if (!addr) {
> +		RTE_LOG(ERR, VHOST_CONFIG,
> +			"Failed to alloc vhost inflight area");
> +			msg->payload.inflight.mmap_size = 0;
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +	}
> +	memset(addr, 0, mmap_size);
> +
> +	dev->inflight_info->addr = addr;
> +	dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
> +	dev->inflight_info->fd = msg->fds[0] = fd;
> +	msg->payload.inflight.mmap_offset = 0;
> +	msg->fd_num = 1;
> +
> +	if (vq_is_packed(dev)) {
> +		for (i = 0; i < num_queues; i++) {
> +			inflight_packed =
> +				(struct rte_vhost_inflight_info_packed *)addr;
> +			inflight_packed->used_wrap_counter = 1;
> +			inflight_packed->old_used_wrap_counter = 1;
> +			for (j = 0; j < queue_size; j++)
> +				inflight_packed->desc[j].next = j + 1;
> +			addr = (void *)((char *)addr + pervq_inflight_size);
> +		}
> +	}
> +
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"send inflight mmap_size: %"PRIu64"\n",
> +		msg->payload.inflight.mmap_size);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"send inflight mmap_offset: %"PRIu64"\n",
> +		msg->payload.inflight.mmap_offset);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"send inflight fd: %d\n", msg->fds[0]);
> +
> +	return RTE_VHOST_MSG_RESULT_REPLY;
> +}
> +
> +static int
> +vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
> +		int main_fd __rte_unused)
> +{
> +	int fd;
> +	uint64_t mmap_size, mmap_offset;
> +	uint16_t num_queues, queue_size;
> +	uint32_t pervq_inflight_size;
> +	void *addr;
> +	struct virtio_net *dev = *pdev;
> +
> +	fd = msg->fds[0];
> +	if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
> +		RTE_LOG(ERR, VHOST_CONFIG,
> +			"Invalid set_inflight_fd message size is %d,fd is %d\n",
> +			msg->size, fd);
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +	}
> +
> +	mmap_size = msg->payload.inflight.mmap_size;
> +	mmap_offset = msg->payload.inflight.mmap_offset;
> +	num_queues = msg->payload.inflight.num_queues;
> +	queue_size = msg->payload.inflight.queue_size;
> +
> +	if (vq_is_packed(dev))
> +		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
> +	else
> +		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
> +
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"set_inflight_fd num_queues: %u\n", num_queues);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"set_inflight_fd queue_size: %u\n", queue_size);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"set_inflight_fd fd: %d\n", fd);
> +	RTE_LOG(INFO, VHOST_CONFIG,
> +		"set_inflight_fd pervq_inflight_size: %d\n",
> +		pervq_inflight_size);
> +
> +	if (!dev->inflight_info) {
> +		dev->inflight_info = calloc(1,
> +					    sizeof(struct inflight_mem_info));
> +		if (dev->inflight_info == NULL) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +				"Failed to alloc dev inflight area");
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +	}
> +
> +	if (dev->inflight_info->addr)
> +		munmap(dev->inflight_info->addr, dev->inflight_info->size);

dev->inflight_info->addr won't be zeroed when below mmap() fails.

> +
> +	addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
> +			fd, mmap_offset);
> +	if (addr == MAP_FAILED) {
> +		RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +	}
> +
> +	if (dev->inflight_info->fd)
> +		close(dev->inflight_info->fd);
> +
> +	dev->inflight_info->fd = fd;
> +	dev->inflight_info->addr = addr;
> +	dev->inflight_info->size = mmap_size;
> +
> +	return RTE_VHOST_MSG_RESULT_OK;
> +}
> +
>  static int
>  vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
>  			int main_fd __rte_unused)
> @@ -1762,9 +2019,10 @@ static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
>  	[VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
>  	[VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
>  	[VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
> +	[VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
> +	[VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
>  };
>  
> -
>  /* return bytes# of read on success or negative val on failure. */
>  static int
>  read_vhost_message(int sockfd, struct VhostUserMsg *msg)
> diff --git a/lib/librte_vhost/vhost_user.h b/lib/librte_vhost/vhost_user.h
> index 17a1d7bca..6563f7315 100644
> --- a/lib/librte_vhost/vhost_user.h
> +++ b/lib/librte_vhost/vhost_user.h
> @@ -54,7 +54,9 @@ typedef enum VhostUserRequest {
>  	VHOST_USER_POSTCOPY_ADVISE = 28,
>  	VHOST_USER_POSTCOPY_LISTEN = 29,
>  	VHOST_USER_POSTCOPY_END = 30,
> -	VHOST_USER_MAX = 31
> +	VHOST_USER_GET_INFLIGHT_FD = 31,
> +	VHOST_USER_SET_INFLIGHT_FD = 32,
> +	VHOST_USER_MAX = 33
>  } VhostUserRequest;
>  
>  typedef enum VhostUserSlaveRequest {
> -- 
> 2.17.2
> 

Re: [dpdk-dev] [PATCH v7 05/10] vhost: checkout the resubmit inflight information

[Tiwei Bie]

On Fri, Sep 20, 2019 at 08:00:57PM +0800, Jin Yu wrote:
> @@ -1458,6 +1472,188 @@ static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
>  	return RTE_VHOST_MSG_RESULT_OK;
>  }
>  
> +static int
> +resubmit_desc_compare(const void *a, const void *b)
> +{
> +	const struct rte_vhost_resubmit_desc *desc0 =
> +		(const struct rte_vhost_resubmit_desc *)a;
> +	const struct rte_vhost_resubmit_desc *desc1 =
> +		(const struct rte_vhost_resubmit_desc *)b;
> +
> +	if (desc1->counter > desc0->counter &&
> +		(desc1->counter - desc0->counter) < VIRTQUEUE_MAX_SIZE * 2)

Why is this hardcoded to 1024 * 2 here?

> +		return 1;
> +
> +	return -1;
> +}
> +
> +static int
> +vhost_check_queue_inflights_split(struct virtio_net *dev,
> +	struct vhost_virtqueue *vq)
> +{
> +	uint16_t i = 0;
> +	uint16_t resubmit_num = 0, last_io, num;
> +	struct vring_used *used = vq->used;
> +	struct rte_vhost_resubmit_info *resubmit = NULL;
> +	struct rte_vhost_inflight_info_split *inflight_split;
> +
> +	if (!(dev->protocol_features &
> +		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
> +		return RTE_VHOST_MSG_RESULT_OK;
> +
> +	if ((!vq->inflight_split))
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +
> +	if (!vq->inflight_split->version) {
> +		vq->inflight_split->version = INFLIGHT_VERSION;
> +		return RTE_VHOST_MSG_RESULT_OK;
> +	}
> +
> +	inflight_split = vq->inflight_split;
> +	vq->resubmit_inflight = NULL;

This is invoked in vhost message handler, you need to
check whether it has been allocated first.


> +	vq->global_counter = 0;
> +	last_io = inflight_split->last_inflight_io;
> +
> +	if (inflight_split->used_idx != used->idx) {
> +		inflight_split->desc[last_io].inflight = 0;
> +		rte_compiler_barrier();
> +		inflight_split->used_idx = used->idx;
> +	}
> +
> +	for (i = 0; i < inflight_split->desc_num; i++) {
> +		if (inflight_split->desc[i].inflight == 1)
> +			resubmit_num++;
> +	}
> +
> +	vq->last_avail_idx += resubmit_num;
> +
> +	if (resubmit_num) {
> +		resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
> +		if (!resubmit) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +			 "Failed to allocate memory for resubmit info.\n");
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +
> +		resubmit->resubmit_list = calloc(resubmit_num,
> +			sizeof(struct rte_vhost_resubmit_desc));
> +		if (!resubmit->resubmit_list) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +			 "Failed to allocate memory for inflight desc.\n");
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +
> +		num = 0;
> +		for (i = 0; i < vq->inflight_split->desc_num; i++) {
> +			if (vq->inflight_split->desc[i].inflight == 1) {
> +				resubmit->resubmit_list[num].index = i;
> +				resubmit->resubmit_list[num].counter =
> +					inflight_split->desc[i].counter;
> +				num++;
> +			}
> +		}
> +		resubmit->resubmit_num = num;
> +
> +		if (resubmit->resubmit_num > 1)
> +			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
> +				sizeof(struct rte_vhost_resubmit_desc),
> +				resubmit_desc_compare);
> +
> +		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
> +		vq->resubmit_inflight = resubmit;
> +	}
> +
> +	return RTE_VHOST_MSG_RESULT_OK;
> +}
> +
> +static int
> +vhost_check_queue_inflights_packed(struct virtio_net *dev,
> +						 struct vhost_virtqueue *vq)
> +{
> +	uint16_t i = 0;
> +	uint16_t resubmit_num = 0, old_used_idx, num;
> +	struct rte_vhost_resubmit_info *resubmit = NULL;
> +	struct rte_vhost_inflight_info_packed *inflight_packed;
> +
> +	if (!(dev->protocol_features &
> +		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
> +		return RTE_VHOST_MSG_RESULT_OK;
> +
> +	if ((!vq->inflight_packed))
> +		return RTE_VHOST_MSG_RESULT_ERR;
> +
> +	if (!vq->inflight_packed->version) {
> +		vq->inflight_packed->version = INFLIGHT_VERSION;
> +		return RTE_VHOST_MSG_RESULT_OK;
> +	}
> +
> +	inflight_packed = vq->inflight_packed;
> +	vq->resubmit_inflight = NULL;

Ditto.


> +	vq->global_counter = 0;
> +	old_used_idx = inflight_packed->old_used_idx;
> +
> +	if (inflight_packed->used_idx != old_used_idx) {
> +		if (inflight_packed->desc[old_used_idx].inflight == 0) {
> +			inflight_packed->old_used_idx =
> +				inflight_packed->used_idx;
> +			inflight_packed->old_used_wrap_counter =
> +				inflight_packed->used_wrap_counter;
> +			inflight_packed->old_free_head =
> +				inflight_packed->free_head;
> +		} else {
> +			inflight_packed->used_idx =
> +				inflight_packed->old_used_idx;
> +			inflight_packed->used_wrap_counter =
> +				inflight_packed->old_used_wrap_counter;
> +			inflight_packed->free_head =
> +				inflight_packed->old_free_head;
> +		}
> +	}
> +
> +	for (i = 0; i < inflight_packed->desc_num; i++) {
> +		if (inflight_packed->desc[i].inflight == 1)
> +			resubmit_num++;
> +	}
> +
> +	if (resubmit_num) {
> +		resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
> +		if (resubmit == NULL) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +			 "Failed to allocate memory for resubmit info.\n");
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +
> +		resubmit->resubmit_list = calloc(resubmit_num,
> +			sizeof(struct rte_vhost_resubmit_desc));
> +		if (resubmit->resubmit_list == NULL) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +			 "Failed to allocate memory for resubmit desc.\n");
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +
> +		num = 0;
> +		for (i = 0; i < inflight_packed->desc_num; i++) {
> +			if (vq->inflight_packed->desc[i].inflight == 1) {
> +				resubmit->resubmit_list[num].index = i;
> +				resubmit->resubmit_list[num].counter =
> +					inflight_packed->desc[i].counter;
> +				num++;
> +			}
> +		}
> +		resubmit->resubmit_num = num;
> +
> +		if (resubmit->resubmit_num > 1)
> +			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
> +				sizeof(struct rte_vhost_resubmit_desc),
> +				resubmit_desc_compare);
> +
> +		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
> +		vq->resubmit_inflight = resubmit;
> +	}
> +
> +	return RTE_VHOST_MSG_RESULT_OK;
> +}
> +
>  static int
>  vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
>  			int main_fd __rte_unused)
> @@ -1499,6 +1695,20 @@ vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
>  		close(vq->kickfd);
>  	vq->kickfd = file.fd;
>  
> +	if (vq_is_packed(dev)) {
> +		if (vhost_check_queue_inflights_packed(dev, vq)) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +				"Failed to inflights for vq: %d\n", file.index);
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +	} else {
> +		if (vhost_check_queue_inflights_split(dev, vq)) {
> +			RTE_LOG(ERR, VHOST_CONFIG,
> +				"Failed to inflights for vq: %d\n", file.index);
> +			return RTE_VHOST_MSG_RESULT_ERR;
> +		}
> +	}
> +
>  	return RTE_VHOST_MSG_RESULT_OK;
>  }
>  
> -- 
> 2.17.2
> 

Re: [dpdk-dev] [PATCH v7 00/10] vhost: support inflight share memory protocol feature

[Yu, Jin]

> -----Original Message-----
> From: Bie, Tiwei
> Sent: Wednesday, September 25, 2019 10:26 PM
> To: Yu, Jin <jin.yu@intel.com>
> Cc: dev@dpdk.org; Liu, Changpeng <changpeng.liu@intel.com>;
> maxime.coquelin@redhat.com; Wang, Zhihong <zhihong.wang@intel.com>
> Subject: Re: [PATCH v7 00/10] vhost: support inflight share memory protocol
> feature
> 
> 
> > [PATCH v7 00/10] vhost: support inflight share memory protocol feature
> 
> This should be v8. Please increase the version number when sending a new series.

I thought it's a small fix. I will fix it. Thanks.
> 
> Thanks,
> Tiwei
> 
> On Fri, Sep 20, 2019 at 08:00:52PM +0800, Jin Yu wrote:
> > This patches introduces two new messages VHOST_USER_GET_INFLIGHT_FD
> > and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared buffer
> > between qemu and backend.
> > Now It can both support split and packed ring. The example code show
> > how these API work. The test has passed.
> >
> > How to test the example:
> > 1, Qemu need two patches.
> > https://patchwork.kernel.org/patch/10766813/
> > https://patchwork.kernel.org/patch/10861411/(QEMU merged) it also
> > needs some manual modifications:
> > we should confirm that before we send get inflight we have already
> > sent the set features but it seems Qemu didn't do like this. So we
> > manually revise this, we can add below code in vhost_dev_get_inflight
> >     int r;
> >
> >     r = vhost_dev_set_features(dev, dev->log_enabled);
> >     if (r < 0) {
> >         return -errno;
> >     }
> > before get_inflight_fd.
> > 2, Guest OS version >= 5.0
> > 3, run the example
> > 4, run the qemu with vhost-user-blk-pci.
> >  eg:
> >         -chardev socket,id=spdk_vhost_blk0,reconnect=1,path=xxxx\
> >         -device
> > vhost-user-blk-pci,ring_packed=on,chardev=spdk_vhost_blk0,num-queues=1
> > \
> > 5, run fio in the guest
> > 6, kill the example and run again.
> > 7, the fio in the guest should continue run without errors.
> >
> > Jin Yu (10):
> >   vhost: add the inflight description
> >   vhost: add packed ring
> >   vhost: add the inflight structure
> >   vhost: add two new messages to support a shared buffer
> >   vhost: checkout the resubmit inflight information
> >   vhost: add the APIs to operate inflight ring
> >   vhost: add APIs for user getting inflight ring
> >   vhost: add vring functions packed ring support
> >   vhost: add an API for judging vq format
> >   vhost: add vhost-user-blk example which support inflight
> >
> >  examples/vhost_blk/Makefile            |   67 ++
> >  examples/vhost_blk/blk.c               |  125 +++
> >  examples/vhost_blk/blk_spec.h          |   95 ++
> >  examples/vhost_blk/meson.build         |   20 +
> >  examples/vhost_blk/vhost_blk.c         | 1313 ++++++++++++++++++++++++
> >  examples/vhost_blk/vhost_blk.h         |  116 +++
> >  examples/vhost_blk/vhost_blk_compat.c  |  195 ++++
> >  lib/librte_vhost/rte_vhost.h           |  250 ++++-
> >  lib/librte_vhost/rte_vhost_version.map |    9 +
> >  lib/librte_vhost/vhost.c               |  416 +++++++-
> >  lib/librte_vhost/vhost.h               |   16 +
> >  lib/librte_vhost/vhost_user.c          |  470 ++++++++-
> >  lib/librte_vhost/vhost_user.h          |   12 +-
> >  13 files changed, 3091 insertions(+), 13 deletions(-)  create mode
> > 100644 examples/vhost_blk/Makefile  create mode 100644
> > examples/vhost_blk/blk.c  create mode 100644
> > examples/vhost_blk/blk_spec.h  create mode 100644
> > examples/vhost_blk/meson.build  create mode 100644
> > examples/vhost_blk/vhost_blk.c  create mode 100644
> > examples/vhost_blk/vhost_blk.h  create mode 100644
> > examples/vhost_blk/vhost_blk_compat.c
> >
> > --
> > 2.17.2
> >

Re: [dpdk-dev] [PATCH v7 10/10] vhost: add vhost-user-blk example which support inflight

[Yu, Jin]

> -----Original Message-----
> From: Bie, Tiwei
> Sent: Wednesday, September 25, 2019 10:46 PM
> To: Yu, Jin <jin.yu@intel.com>
> Cc: dev@dpdk.org; Liu, Changpeng <changpeng.liu@intel.com>;
> maxime.coquelin@redhat.com; Wang, Zhihong <zhihong.wang@intel.com>
> Subject: Re: [PATCH v7 10/10] vhost: add vhost-user-blk example which support
> inflight
> 
> On Fri, Sep 20, 2019 at 08:01:02PM +0800, Jin Yu wrote:
> > A vhost-user-blk example that support inflight feature. It uses the
> > new APIs that introduced in the first patch, so It can show how there
> 
> s/It/it/
> s/there/these/

Got it. Thanks.
> 
> > APIs work to support inflight feature.
> >
> > Signed-off-by: Jin Yu <jin.yu@intel.com>
> > ---
> > V1 - add the case.
> > V2 - add the rte_vhost prefix.
> > V3 - add packed ring support
> > ---
> >  examples/vhost_blk/Makefile           |   67 ++
> >  examples/vhost_blk/blk.c              |  125 +++
> >  examples/vhost_blk/blk_spec.h         |   95 ++
> >  examples/vhost_blk/meson.build        |   20 +
> >  examples/vhost_blk/vhost_blk.c        | 1313 +++++++++++++++++++++++++
> >  examples/vhost_blk/vhost_blk.h        |  116 +++
> >  examples/vhost_blk/vhost_blk_compat.c |  195 ++++
> >  7 files changed, 1931 insertions(+)
> 
> I met some build issues when trying this example.
> 
> examples/vhost_blk/vhost_blk.c: In function ‘descriptor_get_next_packed’:
> examples/vhost_blk/vhost_blk.c:71:21: error: invalid use of undefined type
> ‘struct vring_packed_desc’
>   if (vq->desc_packed[*idx % vq->size].flags & VIRTQ_DESC_F_NEXT) {
>                      ^
> examples/vhost_blk/vhost_blk.c:71:21: error: dereferencing pointer to
> incomplete type ‘struct vring_packed_desc’
> examples/vhost_blk/vhost_blk.c:73:26: error: invalid use of undefined type
> ‘struct vring_packed_desc’
>    return &vq->desc_packed[*idx % vq->size];

The new version of virtio_ring.h have defined the vring_packed_desc.
For the compatibility， I will add the definition in the rte_vhost.h.
Thanks.

>                           ^
> examples/vhost_blk/vhost_blk.c: In function
> ‘inflight_submit_completion_packed’:
> examples/vhost_blk/vhost_blk.c:165:2: warning:
> ‘rte_vhost_set_last_inflight_io_packed’ is deprecated: Symbol is not yet part of
> stable ABI [-Wdeprecated-declarations]

Should I ignore this warning？ I'm not sure about this.
The reason is the _rte_experiment keyword?
Thanks.
>   ret = rte_vhost_set_last_inflight_io_packed(ctrlr->bdev->vid, q_idx,
>   ^~~
> In file included from examples/vhost_blk/vhost_blk.c:17:0:
> x86_64-native-linuxapp-gcc/include/rte_vhost.h:810:1: note: declared here
> rte_vhost_set_last_inflight_io_packed(int vid,
> ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>                          ^
> ...

Re: [dpdk-dev] [PATCH v7 10/10] vhost: add vhost-user-blk example which support inflight

[Tiwei Bie]

On Thu, Sep 26, 2019 at 10:29:19PM +0800, Yu, Jin wrote:
> > -----Original Message-----
> > From: Bie, Tiwei
> > Sent: Wednesday, September 25, 2019 10:46 PM
> > To: Yu, Jin <jin.yu@intel.com>
> > Cc: dev@dpdk.org; Liu, Changpeng <changpeng.liu@intel.com>;
> > maxime.coquelin@redhat.com; Wang, Zhihong <zhihong.wang@intel.com>
> > Subject: Re: [PATCH v7 10/10] vhost: add vhost-user-blk example which support
> > inflight
> > 
> > On Fri, Sep 20, 2019 at 08:01:02PM +0800, Jin Yu wrote:
> > > A vhost-user-blk example that support inflight feature. It uses the
> > > new APIs that introduced in the first patch, so It can show how there
> > 
> > s/It/it/
> > s/there/these/
> 
> Got it. Thanks.
> > 
> > > APIs work to support inflight feature.
> > >
> > > Signed-off-by: Jin Yu <jin.yu@intel.com>
> > > ---
> > > V1 - add the case.
> > > V2 - add the rte_vhost prefix.
> > > V3 - add packed ring support
> > > ---
> > >  examples/vhost_blk/Makefile           |   67 ++
> > >  examples/vhost_blk/blk.c              |  125 +++
> > >  examples/vhost_blk/blk_spec.h         |   95 ++
> > >  examples/vhost_blk/meson.build        |   20 +
> > >  examples/vhost_blk/vhost_blk.c        | 1313 +++++++++++++++++++++++++
> > >  examples/vhost_blk/vhost_blk.h        |  116 +++
> > >  examples/vhost_blk/vhost_blk_compat.c |  195 ++++
> > >  7 files changed, 1931 insertions(+)
> > 
> > I met some build issues when trying this example.
> > 
> > examples/vhost_blk/vhost_blk.c: In function ‘descriptor_get_next_packed’:
> > examples/vhost_blk/vhost_blk.c:71:21: error: invalid use of undefined type
> > ‘struct vring_packed_desc’
> >   if (vq->desc_packed[*idx % vq->size].flags & VIRTQ_DESC_F_NEXT) {
> >                      ^
> > examples/vhost_blk/vhost_blk.c:71:21: error: dereferencing pointer to
> > incomplete type ‘struct vring_packed_desc’
> > examples/vhost_blk/vhost_blk.c:73:26: error: invalid use of undefined type
> > ‘struct vring_packed_desc’
> >    return &vq->desc_packed[*idx % vq->size];
> 
> The new version of virtio_ring.h have defined the vring_packed_desc.
> For the compatibility， I will add the definition in the rte_vhost.h.

No. They should be added in this example instead of
the vhost API.

> Thanks.
> 
> >                           ^
> > examples/vhost_blk/vhost_blk.c: In function
> > ‘inflight_submit_completion_packed’:
> > examples/vhost_blk/vhost_blk.c:165:2: warning:
> > ‘rte_vhost_set_last_inflight_io_packed’ is deprecated: Symbol is not yet part of
> > stable ABI [-Wdeprecated-declarations]
> 
> Should I ignore this warning？ I'm not sure about this.
> The reason is the _rte_experiment keyword?

You need something like this:

https://github.com/DPDK/dpdk/blob/bd253daa7717835f88bbc58b09a94d0060380396/examples/vhost/Makefile#L29
https://github.com/DPDK/dpdk/blob/bd253daa7717835f88bbc58b09a94d0060380396/examples/vhost/meson.build#L13


> Thanks.
> >   ret = rte_vhost_set_last_inflight_io_packed(ctrlr->bdev->vid, q_idx,
> >   ^~~
> > In file included from examples/vhost_blk/vhost_blk.c:17:0:
> > x86_64-native-linuxapp-gcc/include/rte_vhost.h:810:1: note: declared here
> > rte_vhost_set_last_inflight_io_packed(int vid,
> > ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >                          ^
> > ...

Re: [dpdk-dev] [PATCH v7 04/10] vhost: add two new messages to support a shared buffer

[Yu, Jin]

> -----Original Message-----
> From: Bie, Tiwei
> Sent: Thursday, September 26, 2019 3:39 PM
> To: Yu, Jin <jin.yu@intel.com>
> Cc: dev@dpdk.org; Liu, Changpeng <changpeng.liu@intel.com>;
> maxime.coquelin@redhat.com; Wang, Zhihong <zhihong.wang@intel.com>; Lin
> Li <lilin24@baidu.com>; Xun Ni <nixun@baidu.com>; Yu Zhang
> <zhangyu31@baidu.com>
> Subject: Re: [PATCH v7 04/10] vhost: add two new messages to support a shared
> buffer
> 
> On Fri, Sep 20, 2019 at 08:00:56PM +0800, Jin Yu wrote:
> > This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD and
> > VHOST_USER_SET_INFLIGHT_FD to support transferring a shared buffer
> > between qemu and backend.
> >
> > Signed-off-by: Lin Li <lilin24@baidu.com>
> > Signed-off-by: Xun Ni <nixun@baidu.com>
> > Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
> > Signed-off-by: Jin Yu <jin.yu@intel.com>
> > ---
> >  lib/librte_vhost/vhost.h      |   7 +
> >  lib/librte_vhost/vhost_user.c | 260 +++++++++++++++++++++++++++++++++-
> >  lib/librte_vhost/vhost_user.h |   4 +-
> >  3 files changed, 269 insertions(+), 2 deletions(-)
> >
> > diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h index
> > 884befa85..d67ba849a 100644
> > --- a/lib/librte_vhost/vhost.h
> > +++ b/lib/librte_vhost/vhost.h
> > @@ -286,6 +286,12 @@ struct guest_page {
> >  	uint64_t size;
> >  };
> >
> > +struct inflight_mem_info {
> > +	int		fd;
> > +	void		*addr;
> > +	uint64_t	size;
> > +};
> > +
> >  /**
> >   * Device structure contains all configuration information relating
> >   * to the device.
> > @@ -303,6 +309,7 @@ struct virtio_net {
> >  	uint32_t		nr_vring;
> >  	int			dequeue_zero_copy;
> >  	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
> > +	struct inflight_mem_info *inflight_info;
> >  #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
> >  	char			ifname[IF_NAME_SZ];
> >  	uint64_t		log_size;
> > diff --git a/lib/librte_vhost/vhost_user.c
> > b/lib/librte_vhost/vhost_user.c index c9e29ece8..66582a30f 100644
> > --- a/lib/librte_vhost/vhost_user.c
> > +++ b/lib/librte_vhost/vhost_user.c
> > @@ -37,6 +37,10 @@
> >  #ifdef RTE_LIBRTE_VHOST_POSTCOPY
> >  #include <linux/userfaultfd.h>
> >  #endif
> > +#ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
> > +#include <linux/memfd.h> #define MEMFD_SUPPORTED #endif
> >
> >  #include <rte_common.h>
> >  #include <rte_malloc.h>
> > @@ -49,6 +53,15 @@
> >  #define VIRTIO_MIN_MTU 68
> >  #define VIRTIO_MAX_MTU 65535
> >
> > +#define INFLIGHT_ALIGNMENT	64
> > +#define INFLIGHT_VERSION	0xabcd
> 
> Why the version is 0xabcd? Shouldn't it be 1?
> 
> https://git.qemu.org/?p=qemu.git;a=blob;f=docs/interop/vhost-
> user.rst;h=7827b710aa0a223f6f184b229f4b5ced6420b638;hb=HEAD#l551
> https://git.qemu.org/?p=qemu.git;a=blob;f=docs/interop/vhost-
> user.rst;h=7827b710aa0a223f6f184b229f4b5ced6420b638;hb=HEAD#l658
Yeah, I didn't fix it. Will fix it in next version.
> 
> > +#define VIRTQUEUE_MAX_SIZE	1024
> > +
> > +#define CLOEXEC		0x0001U
> > +
> > +#define ALIGN_DOWN(n, m) ((n) / (m) * (m)) #define ALIGN_UP(n, m)
> > +ALIGN_DOWN((n) + (m) - 1, (m))
> > +
> >  static const char *vhost_message_str[VHOST_USER_MAX] = {
> >  	[VHOST_USER_NONE] = "VHOST_USER_NONE",
> >  	[VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES", @@
> -78,6
> > +91,8 @@ static const char *vhost_message_str[VHOST_USER_MAX] = {
> >  	[VHOST_USER_POSTCOPY_ADVISE]  =
> "VHOST_USER_POSTCOPY_ADVISE",
> >  	[VHOST_USER_POSTCOPY_LISTEN]  =
> "VHOST_USER_POSTCOPY_LISTEN",
> >  	[VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
> > +	[VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
> > +	[VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
> >  };
> >
> >  static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg); @@
> > -160,6 +175,21 @@ vhost_backend_cleanup(struct virtio_net *dev)
> >  		dev->log_addr = 0;
> >  	}
> >
> > +	if (dev->inflight_info->addr) {
> > +		munmap(dev->inflight_info->addr, dev->inflight_info->size);
> > +		dev->inflight_info->addr = NULL;
> > +	}
> > +
> > +	if (dev->inflight_info->fd > 0) {
> > +		close(dev->inflight_info->fd);
> > +		dev->inflight_info->fd = -1;
> > +	}
> > +
> > +	if (dev->inflight_info) {
> > +		free(dev->inflight_info);
> > +		dev->inflight_info = NULL;
> > +	}
> 
> You should check dev->inflight_info first before dereferencing it.
Got it. Thanks.
> 
> > +
> >  	if (dev->slave_req_fd >= 0) {
> >  		close(dev->slave_req_fd);
> >  		dev->slave_req_fd = -1;
> > @@ -1165,6 +1195,233 @@ virtio_is_ready(struct virtio_net *dev)
> >  	return 1;
> >  }
> >
> > +static int
> > +mem_create(const char *name, unsigned int flags) { #ifdef
> > +MEMFD_SUPPORTED
> > +	return memfd_create(name, flags);
> > +#else
> > +	RTE_LOG(ERR, VHOST_CONFIG,
> > +		"doesn't support memfd--name:%s and flag:%x\n",
> > +		name, flags);
> 
> You probably don't want to always print an error when memfd isn't available.
I saw below error message from ci/Intel-compilation. 
../lib/librte_vhost/vhost_user.c:1202:24: error: unused parameter ‘name’ [-Werror=unused-parameter]
 mem_create(const char *name, unsigned int flags)
                        ^~~~
../lib/librte_vhost/vhost_user.c:1202:43: error: unused parameter ‘flags’ [-Werror=unused-parameter]
 mem_create(const char *name, unsigned int flags)

That's why I add this print.Should I ignore this?
Thanks.

> 
> > +	return -1;
> > +#endif
> > +}
> > +
> > +static void *
> > +inflight_mem_alloc(const char *name, size_t size, int *fd) {
> > +	void *ptr;
> > +	int mfd = -1;
> > +	char fname[20] = "/tmp/memfd-XXXXXX";
> > +
> > +	*fd = -1;
> > +	mfd = mem_create(name, CLOEXEC);
> > +	if (mfd != -1) {
> > +		if (ftruncate(mfd, size) == -1) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +				"ftruncate fail for alloc inflight buffer\n");
> > +			close(mfd);
> > +			return NULL;
> > +		}
> > +	} else {
> > +		mfd = mkstemp(fname);
> > +		unlink(fname);
> > +
> > +		if (mfd == -1) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +				"mkstemp fail for alloc inflight buffer\n");
> > +			return NULL;
> > +		}
> > +
> > +		if (ftruncate(mfd, size) == -1) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +				"ftruncate fail for alloc inflight buffer\n");
> > +			close(mfd);
> > +			return NULL;
> > +		}
> > +	}
> > +
> > +	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
> > +	if (ptr == MAP_FAILED) {
> > +		RTE_LOG(ERR, VHOST_CONFIG,
> > +			"mmap fail for alloc inflight buffer\n");
> > +		close(mfd);
> > +		return NULL;
> > +	}
> > +
> > +	*fd = mfd;
> > +	return ptr;
> > +}
> > +
> > +static uint32_t
> > +get_pervq_shm_size_split(uint16_t queue_size) {
> > +	return ALIGN_UP(sizeof(struct rte_vhost_inflight_desc_split) *
> > +			queue_size + sizeof(uint64_t) + sizeof(uint16_t) * 4,
> > +			INFLIGHT_ALIGNMENT);
> > +}
> > +
> > +static uint32_t
> > +get_pervq_shm_size_packed(uint16_t queue_size) {
> > +	return ALIGN_UP(sizeof(struct rte_vhost_inflight_desc_packed) *
> > +			queue_size + sizeof(uint64_t) + sizeof(uint16_t) * 6 +
> > +			sizeof(uint8_t) * 9, INFLIGHT_ALIGNMENT); }
> > +
> > +static int
> > +vhost_user_get_inflight_fd(struct virtio_net **pdev,
> > +				     VhostUserMsg *msg,
> > +				     int main_fd __rte_unused)
> > +{
> > +	int fd, i, j;
> > +	uint64_t pervq_inflight_size, mmap_size;
> > +	void *addr;
> > +	uint16_t num_queues, queue_size;
> > +	struct virtio_net *dev = *pdev;
> > +	struct rte_vhost_inflight_info_packed *inflight_packed = NULL;
> > +
> > +	if (msg->size != sizeof(msg->payload.inflight)) {
> > +		RTE_LOG(ERR, VHOST_CONFIG,
> > +			"Invalid get_inflight_fd message size is %d",
> > +			msg->size);
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +	}
> > +
> > +	dev->inflight_info = calloc(1, sizeof(struct inflight_mem_info));
> 
> You need to check whether dev->inflight_info has been allocated, otherwise
> memory may leak.
Yeah, Thanks. Will fix it.
> 
> > +	if (!dev->inflight_info) {
> > +		RTE_LOG(ERR, VHOST_CONFIG,
> > +			"Failed to alloc dev inflight area");
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +	}
> > +
> > +	num_queues = msg->payload.inflight.num_queues;
> > +	queue_size = msg->payload.inflight.queue_size;
> > +
> > +	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
> > +		msg->payload.inflight.num_queues);
> > +	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
> > +		msg->payload.inflight.queue_size);
> > +
> > +	if (vq_is_packed(dev))
> > +		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
> > +	else
> > +		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
> > +
> > +	mmap_size = num_queues * pervq_inflight_size;
> > +	addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
> > +	if (!addr) {
> > +		RTE_LOG(ERR, VHOST_CONFIG,
> > +			"Failed to alloc vhost inflight area");
> > +			msg->payload.inflight.mmap_size = 0;
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +	}
> > +	memset(addr, 0, mmap_size);
> > +
> > +	dev->inflight_info->addr = addr;
> > +	dev->inflight_info->size = msg->payload.inflight.mmap_size =
> mmap_size;
> > +	dev->inflight_info->fd = msg->fds[0] = fd;
> > +	msg->payload.inflight.mmap_offset = 0;
> > +	msg->fd_num = 1;
> > +
> > +	if (vq_is_packed(dev)) {
> > +		for (i = 0; i < num_queues; i++) {
> > +			inflight_packed =
> > +				(struct rte_vhost_inflight_info_packed *)addr;
> > +			inflight_packed->used_wrap_counter = 1;
> > +			inflight_packed->old_used_wrap_counter = 1;
> > +			for (j = 0; j < queue_size; j++)
> > +				inflight_packed->desc[j].next = j + 1;
> > +			addr = (void *)((char *)addr + pervq_inflight_size);
> > +		}
> > +	}
> > +
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"send inflight mmap_size: %"PRIu64"\n",
> > +		msg->payload.inflight.mmap_size);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"send inflight mmap_offset: %"PRIu64"\n",
> > +		msg->payload.inflight.mmap_offset);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"send inflight fd: %d\n", msg->fds[0]);
> > +
> > +	return RTE_VHOST_MSG_RESULT_REPLY;
> > +}
> > +
> > +static int
> > +vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
> > +		int main_fd __rte_unused)
> > +{
> > +	int fd;
> > +	uint64_t mmap_size, mmap_offset;
> > +	uint16_t num_queues, queue_size;
> > +	uint32_t pervq_inflight_size;
> > +	void *addr;
> > +	struct virtio_net *dev = *pdev;
> > +
> > +	fd = msg->fds[0];
> > +	if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
> > +		RTE_LOG(ERR, VHOST_CONFIG,
> > +			"Invalid set_inflight_fd message size is %d,fd is %d\n",
> > +			msg->size, fd);
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +	}
> > +
> > +	mmap_size = msg->payload.inflight.mmap_size;
> > +	mmap_offset = msg->payload.inflight.mmap_offset;
> > +	num_queues = msg->payload.inflight.num_queues;
> > +	queue_size = msg->payload.inflight.queue_size;
> > +
> > +	if (vq_is_packed(dev))
> > +		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
> > +	else
> > +		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
> > +
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"set_inflight_fd num_queues: %u\n", num_queues);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"set_inflight_fd queue_size: %u\n", queue_size);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"set_inflight_fd fd: %d\n", fd);
> > +	RTE_LOG(INFO, VHOST_CONFIG,
> > +		"set_inflight_fd pervq_inflight_size: %d\n",
> > +		pervq_inflight_size);
> > +
> > +	if (!dev->inflight_info) {
> > +		dev->inflight_info = calloc(1,
> > +					    sizeof(struct inflight_mem_info));
> > +		if (dev->inflight_info == NULL) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +				"Failed to alloc dev inflight area");
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +	}
> > +
> > +	if (dev->inflight_info->addr)
> > +		munmap(dev->inflight_info->addr, dev->inflight_info->size);
> 
> dev->inflight_info->addr won't be zeroed when below mmap() fails.
I think it should be zero when below mmap() fails as this function would return directly, so
The dev->inflight_info->addr would not been set.
> 
> > +
> > +	addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
> > +			fd, mmap_offset);
> > +	if (addr == MAP_FAILED) {
> > +		RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share
> memory.\n");
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +	}
> > +
> > +	if (dev->inflight_info->fd)
> > +		close(dev->inflight_info->fd);
> > +
> > +	dev->inflight_info->fd = fd;
> > +	dev->inflight_info->addr = addr;
> > +	dev->inflight_info->size = mmap_size;
> > +
> > +	return RTE_VHOST_MSG_RESULT_OK;
> > +}
> > +
> >  static int
> >  vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg
> *msg,
> >  			int main_fd __rte_unused)
> > @@ -1762,9 +2019,10 @@ static vhost_message_handler_t
> vhost_message_handlers[VHOST_USER_MAX] = {
> >  	[VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
> >  	[VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
> >  	[VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
> > +	[VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
> > +	[VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
> >  };
> >
> > -
> >  /* return bytes# of read on success or negative val on failure. */
> > static int  read_vhost_message(int sockfd, struct VhostUserMsg *msg)
> > diff --git a/lib/librte_vhost/vhost_user.h
> > b/lib/librte_vhost/vhost_user.h index 17a1d7bca..6563f7315 100644
> > --- a/lib/librte_vhost/vhost_user.h
> > +++ b/lib/librte_vhost/vhost_user.h
> > @@ -54,7 +54,9 @@ typedef enum VhostUserRequest {
> >  	VHOST_USER_POSTCOPY_ADVISE = 28,
> >  	VHOST_USER_POSTCOPY_LISTEN = 29,
> >  	VHOST_USER_POSTCOPY_END = 30,
> > -	VHOST_USER_MAX = 31
> > +	VHOST_USER_GET_INFLIGHT_FD = 31,
> > +	VHOST_USER_SET_INFLIGHT_FD = 32,
> > +	VHOST_USER_MAX = 33
> >  } VhostUserRequest;
> >
> >  typedef enum VhostUserSlaveRequest {
> > --
> > 2.17.2
> >

Re: [dpdk-dev] [PATCH v7 05/10] vhost: checkout the resubmit inflight information

[Yu, Jin]

> -----Original Message-----
> From: Bie, Tiwei
> Sent: Thursday, September 26, 2019 4:05 PM
> To: Yu, Jin <jin.yu@intel.com>
> Cc: dev@dpdk.org; Liu, Changpeng <changpeng.liu@intel.com>;
> maxime.coquelin@redhat.com; Wang, Zhihong <zhihong.wang@intel.com>; Lin
> Li <lilin24@baidu.com>; Xun Ni <nixun@baidu.com>; Yu Zhang
> <zhangyu31@baidu.com>
> Subject: Re: [PATCH v7 05/10] vhost: checkout the resubmit inflight information
> 
> On Fri, Sep 20, 2019 at 08:00:57PM +0800, Jin Yu wrote:
> > @@ -1458,6 +1472,188 @@ static int vhost_user_set_vring_err(struct
> virtio_net **pdev __rte_unused,
> >  	return RTE_VHOST_MSG_RESULT_OK;
> >  }
> >
> > +static int
> > +resubmit_desc_compare(const void *a, const void *b) {
> > +	const struct rte_vhost_resubmit_desc *desc0 =
> > +		(const struct rte_vhost_resubmit_desc *)a;
> > +	const struct rte_vhost_resubmit_desc *desc1 =
> > +		(const struct rte_vhost_resubmit_desc *)b;
> > +
> > +	if (desc1->counter > desc0->counter &&
> > +		(desc1->counter - desc0->counter) < VIRTQUEUE_MAX_SIZE * 2)
> 
> Why is this hardcoded to 1024 * 2 here?
I will check this. Thanks.
> 
> > +		return 1;
> > +
> > +	return -1;
> > +}
> > +
> > +static int
> > +vhost_check_queue_inflights_split(struct virtio_net *dev,
> > +	struct vhost_virtqueue *vq)
> > +{
> > +	uint16_t i = 0;
> > +	uint16_t resubmit_num = 0, last_io, num;
> > +	struct vring_used *used = vq->used;
> > +	struct rte_vhost_resubmit_info *resubmit = NULL;
> > +	struct rte_vhost_inflight_info_split *inflight_split;
> > +
> > +	if (!(dev->protocol_features &
> > +		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
> > +		return RTE_VHOST_MSG_RESULT_OK;
> > +
> > +	if ((!vq->inflight_split))
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +
> > +	if (!vq->inflight_split->version) {
> > +		vq->inflight_split->version = INFLIGHT_VERSION;
> > +		return RTE_VHOST_MSG_RESULT_OK;
> > +	}
> > +
> > +	inflight_split = vq->inflight_split;
> > +	vq->resubmit_inflight = NULL;
> 
> This is invoked in vhost message handler, you need to check whether it has been
> allocated first.
Got it, Thanks.
> 
> 
> > +	vq->global_counter = 0;
> > +	last_io = inflight_split->last_inflight_io;
> > +
> > +	if (inflight_split->used_idx != used->idx) {
> > +		inflight_split->desc[last_io].inflight = 0;
> > +		rte_compiler_barrier();
> > +		inflight_split->used_idx = used->idx;
> > +	}
> > +
> > +	for (i = 0; i < inflight_split->desc_num; i++) {
> > +		if (inflight_split->desc[i].inflight == 1)
> > +			resubmit_num++;
> > +	}
> > +
> > +	vq->last_avail_idx += resubmit_num;
> > +
> > +	if (resubmit_num) {
> > +		resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
> > +		if (!resubmit) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +			 "Failed to allocate memory for resubmit info.\n");
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +
> > +		resubmit->resubmit_list = calloc(resubmit_num,
> > +			sizeof(struct rte_vhost_resubmit_desc));
> > +		if (!resubmit->resubmit_list) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +			 "Failed to allocate memory for inflight desc.\n");
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +
> > +		num = 0;
> > +		for (i = 0; i < vq->inflight_split->desc_num; i++) {
> > +			if (vq->inflight_split->desc[i].inflight == 1) {
> > +				resubmit->resubmit_list[num].index = i;
> > +				resubmit->resubmit_list[num].counter =
> > +					inflight_split->desc[i].counter;
> > +				num++;
> > +			}
> > +		}
> > +		resubmit->resubmit_num = num;
> > +
> > +		if (resubmit->resubmit_num > 1)
> > +			qsort(resubmit->resubmit_list, resubmit-
> >resubmit_num,
> > +				sizeof(struct rte_vhost_resubmit_desc),
> > +				resubmit_desc_compare);
> > +
> > +		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
> > +		vq->resubmit_inflight = resubmit;
> > +	}
> > +
> > +	return RTE_VHOST_MSG_RESULT_OK;
> > +}
> > +
> > +static int
> > +vhost_check_queue_inflights_packed(struct virtio_net *dev,
> > +						 struct vhost_virtqueue *vq)
> > +{
> > +	uint16_t i = 0;
> > +	uint16_t resubmit_num = 0, old_used_idx, num;
> > +	struct rte_vhost_resubmit_info *resubmit = NULL;
> > +	struct rte_vhost_inflight_info_packed *inflight_packed;
> > +
> > +	if (!(dev->protocol_features &
> > +		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
> > +		return RTE_VHOST_MSG_RESULT_OK;
> > +
> > +	if ((!vq->inflight_packed))
> > +		return RTE_VHOST_MSG_RESULT_ERR;
> > +
> > +	if (!vq->inflight_packed->version) {
> > +		vq->inflight_packed->version = INFLIGHT_VERSION;
> > +		return RTE_VHOST_MSG_RESULT_OK;
> > +	}
> > +
> > +	inflight_packed = vq->inflight_packed;
> > +	vq->resubmit_inflight = NULL;
> 
> Ditto.
> 
> 
> > +	vq->global_counter = 0;
> > +	old_used_idx = inflight_packed->old_used_idx;
> > +
> > +	if (inflight_packed->used_idx != old_used_idx) {
> > +		if (inflight_packed->desc[old_used_idx].inflight == 0) {
> > +			inflight_packed->old_used_idx =
> > +				inflight_packed->used_idx;
> > +			inflight_packed->old_used_wrap_counter =
> > +				inflight_packed->used_wrap_counter;
> > +			inflight_packed->old_free_head =
> > +				inflight_packed->free_head;
> > +		} else {
> > +			inflight_packed->used_idx =
> > +				inflight_packed->old_used_idx;
> > +			inflight_packed->used_wrap_counter =
> > +				inflight_packed->old_used_wrap_counter;
> > +			inflight_packed->free_head =
> > +				inflight_packed->old_free_head;
> > +		}
> > +	}
> > +
> > +	for (i = 0; i < inflight_packed->desc_num; i++) {
> > +		if (inflight_packed->desc[i].inflight == 1)
> > +			resubmit_num++;
> > +	}
> > +
> > +	if (resubmit_num) {
> > +		resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
> > +		if (resubmit == NULL) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +			 "Failed to allocate memory for resubmit info.\n");
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +
> > +		resubmit->resubmit_list = calloc(resubmit_num,
> > +			sizeof(struct rte_vhost_resubmit_desc));
> > +		if (resubmit->resubmit_list == NULL) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +			 "Failed to allocate memory for resubmit desc.\n");
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +
> > +		num = 0;
> > +		for (i = 0; i < inflight_packed->desc_num; i++) {
> > +			if (vq->inflight_packed->desc[i].inflight == 1) {
> > +				resubmit->resubmit_list[num].index = i;
> > +				resubmit->resubmit_list[num].counter =
> > +					inflight_packed->desc[i].counter;
> > +				num++;
> > +			}
> > +		}
> > +		resubmit->resubmit_num = num;
> > +
> > +		if (resubmit->resubmit_num > 1)
> > +			qsort(resubmit->resubmit_list, resubmit-
> >resubmit_num,
> > +				sizeof(struct rte_vhost_resubmit_desc),
> > +				resubmit_desc_compare);
> > +
> > +		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
> > +		vq->resubmit_inflight = resubmit;
> > +	}
> > +
> > +	return RTE_VHOST_MSG_RESULT_OK;
> > +}
> > +
> >  static int
> >  vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg
> *msg,
> >  			int main_fd __rte_unused)
> > @@ -1499,6 +1695,20 @@ vhost_user_set_vring_kick(struct virtio_net
> **pdev, struct VhostUserMsg *msg,
> >  		close(vq->kickfd);
> >  	vq->kickfd = file.fd;
> >
> > +	if (vq_is_packed(dev)) {
> > +		if (vhost_check_queue_inflights_packed(dev, vq)) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +				"Failed to inflights for vq: %d\n", file.index);
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +	} else {
> > +		if (vhost_check_queue_inflights_split(dev, vq)) {
> > +			RTE_LOG(ERR, VHOST_CONFIG,
> > +				"Failed to inflights for vq: %d\n", file.index);
> > +			return RTE_VHOST_MSG_RESULT_ERR;
> > +		}
> > +	}
> > +
> >  	return RTE_VHOST_MSG_RESULT_OK;
> >  }
> >
> > --
> > 2.17.2
> >

Re: [dpdk-dev] [PATCH v7 04/10] vhost: add two new messages to support a shared buffer

[Tiwei Bie]

On Thu, Sep 26, 2019 at 11:06:22PM +0800, Yu, Jin wrote:
> > >
> > > +static int
> > > +mem_create(const char *name, unsigned int flags) { #ifdef
> > > +MEMFD_SUPPORTED
> > > +	return memfd_create(name, flags);
> > > +#else
> > > +	RTE_LOG(ERR, VHOST_CONFIG,
> > > +		"doesn't support memfd--name:%s and flag:%x\n",
> > > +		name, flags);
> > 
> > You probably don't want to always print an error when memfd isn't available.
> I saw below error message from ci/Intel-compilation. 
> ../lib/librte_vhost/vhost_user.c:1202:24: error: unused parameter ‘name’ [-Werror=unused-parameter]
>  mem_create(const char *name, unsigned int flags)
>                         ^~~~
> ../lib/librte_vhost/vhost_user.c:1202:43: error: unused parameter ‘flags’ [-Werror=unused-parameter]
>  mem_create(const char *name, unsigned int flags)
> 
> That's why I add this print.Should I ignore this?

Looks better to do it like this (compile test only):

static void *
inflight_mem_alloc(const char *name, size_t size, int *fd)
{
	char fname[20] = "/tmp/memfd-XXXXXX";
	int mfd = -1;
	void *addr;

#ifdef MEMFD_SUPPORTED
	mfd = memfd_create(name, MFD_CLOEXEC);
#else
	RTE_SET_USED(name);
#endif
	if (mfd == -1) {
		mfd = mkstemp(fname);
		if (mfd == -1) {
			RTE_LOG(ERR, VHOST_CONFIG,
				"failed to create memory file for inflight buffer\n");
			return NULL;
		}
		unlink(fname);
	}

	if (ftruncate(mfd, size) == -1) {
		RTE_LOG(ERR, VHOST_CONFIG,
			"failed to truncate memory file for inflight buffer\n");
		close(mfd);
		return NULL;
	}

	addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
	if (addr == MAP_FAILED) {
		RTE_LOG(ERR, VHOST_CONFIG,
			"failed to map memory file for inflight buffer\n");
		close(mfd);
		return NULL;
	}

	*fd = mfd;
	return addr;
}

And you don't need to define CLOEXEC by yourself.


> Thanks.
> 
> > 
> > > +	return -1;
> > > +#endif
> > > +}
> > > +
> > > +static void *
> > > +inflight_mem_alloc(const char *name, size_t size, int *fd) {
> > > +	void *ptr;
> > > +	int mfd = -1;
> > > +	char fname[20] = "/tmp/memfd-XXXXXX";
> > > +
> > > +	*fd = -1;
> > > +	mfd = mem_create(name, CLOEXEC);
> > > +	if (mfd != -1) {
> > > +		if (ftruncate(mfd, size) == -1) {
> > > +			RTE_LOG(ERR, VHOST_CONFIG,
> > > +				"ftruncate fail for alloc inflight buffer\n");
> > > +			close(mfd);
> > > +			return NULL;
> > > +		}
> > > +	} else {
> > > +		mfd = mkstemp(fname);
> > > +		unlink(fname);
> > > +
> > > +		if (mfd == -1) {
> > > +			RTE_LOG(ERR, VHOST_CONFIG,
> > > +				"mkstemp fail for alloc inflight buffer\n");
> > > +			return NULL;
> > > +		}
> > > +
> > > +		if (ftruncate(mfd, size) == -1) {
> > > +			RTE_LOG(ERR, VHOST_CONFIG,
> > > +				"ftruncate fail for alloc inflight buffer\n");
> > > +			close(mfd);
> > > +			return NULL;
> > > +		}
> > > +	}
> > > +
> > > +	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
> > > +	if (ptr == MAP_FAILED) {
> > > +		RTE_LOG(ERR, VHOST_CONFIG,
> > > +			"mmap fail for alloc inflight buffer\n");
> > > +		close(mfd);
> > > +		return NULL;
> > > +	}
> > > +
> > > +	*fd = mfd;
> > > +	return ptr;
> > > +}