@@ -169,9 +169,12 @@ demonstrates how to use the async vhost APIs. It's used in combination with dmas
**--dmas**
This parameter is used to specify the assigned DMA device of a vhost device.
Async vhost-user net driver will be used if --dmas is set. For example
---dmas [txd0@00:04.0,txd1@00:04.1] means use DMA channel 00:04.0 for vhost
-device 0 enqueue operation and use DMA channel 00:04.1 for vhost device 1
-enqueue operation.
+--dmas [txd0@00:04.0,txd1@00:04.1,rxd0@00:04.2,rxd1@00:04.3] means use
+DMA channel 00:04.0/00:04.2 for vhost device 0 enqueue/dequeue operation
+and use DMA channel 00:04.1/00:04.3 for vhost device 1 enqueue/dequeue
+operation. The index of the device corresponds to the socket file in order,
+that means vhost device 0 is created through the first socket file, vhost
+device 1 is created through the second socket file, and so on.
Common Issues
-------------
@@ -63,6 +63,9 @@
#define DMA_RING_SIZE 4096
+#define ASYNC_ENQUEUE_VHOST 1
+#define ASYNC_DEQUEUE_VHOST 2
+
/* number of mbufs in all pools - if specified on command-line. */
static int total_num_mbufs = NUM_MBUFS_DEFAULT;
@@ -116,6 +119,8 @@ static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
static char *socket_files;
static int nb_sockets;
+static struct vhost_queue_ops vdev_queue_ops[RTE_MAX_VHOST_DEVICE];
+
/* empty VMDq configuration structure. Filled in programmatically */
static struct rte_eth_conf vmdq_conf_default = {
.rxmode = {
@@ -205,6 +210,18 @@ struct vhost_bufftable *vhost_txbuff[RTE_MAX_LCORE * RTE_MAX_VHOST_DEVICE];
#define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \
/ US_PER_S * BURST_TX_DRAIN_US)
+static int vid2socketid[RTE_MAX_VHOST_DEVICE];
+
+static uint32_t get_async_flag_by_socketid(int socketid)
+{
+ return dma_bind[socketid].async_flag;
+}
+
+static void init_vid2socketid_array(int vid, int socketid)
+{
+ vid2socketid[vid] = socketid;
+}
+
static inline bool
is_dma_configured(int16_t dev_id)
{
@@ -224,7 +241,7 @@ open_dma(const char *value)
char *addrs = input;
char *ptrs[2];
char *start, *end, *substr;
- int64_t vid;
+ int64_t socketid, vring_id;
struct rte_dma_info info;
struct rte_dma_conf dev_config = { .nb_vchans = 1 };
@@ -262,7 +279,9 @@ open_dma(const char *value)
while (i < args_nr) {
char *arg_temp = dma_arg[i];
+ char *txd, *rxd;
uint8_t sub_nr;
+ int async_flag;
sub_nr = rte_strsplit(arg_temp, strlen(arg_temp), ptrs, 2, '@');
if (sub_nr != 2) {
@@ -270,14 +289,23 @@ open_dma(const char *value)
goto out;
}
- start = strstr(ptrs[0], "txd");
- if (start == NULL) {
+ txd = strstr(ptrs[0], "txd");
+ rxd = strstr(ptrs[0], "rxd");
+ if (txd) {
+ start = txd;
+ vring_id = VIRTIO_RXQ;
+ async_flag = ASYNC_ENQUEUE_VHOST;
+ } else if (rxd) {
+ start = rxd;
+ vring_id = VIRTIO_TXQ;
+ async_flag = ASYNC_DEQUEUE_VHOST;
+ } else {
ret = -1;
goto out;
}
start += 3;
- vid = strtol(start, &end, 0);
+ socketid = strtol(start, &end, 0);
if (end == start) {
ret = -1;
goto out;
@@ -338,7 +366,8 @@ open_dma(const char *value)
dmas_id[dma_count++] = dev_id;
done:
- (dma_info + vid)->dmas[VIRTIO_RXQ].dev_id = dev_id;
+ (dma_info + socketid)->dmas[vring_id].dev_id = dev_id;
+ (dma_info + socketid)->async_flag |= async_flag;
i++;
}
out:
@@ -990,7 +1019,7 @@ complete_async_pkts(struct vhost_dev *vdev)
{
struct rte_mbuf *p_cpl[MAX_PKT_BURST];
uint16_t complete_count;
- int16_t dma_id = dma_bind[vdev->vid].dmas[VIRTIO_RXQ].dev_id;
+ int16_t dma_id = dma_bind[vid2socketid[vdev->vid]].dmas[VIRTIO_RXQ].dev_id;
complete_count = rte_vhost_poll_enqueue_completed(vdev->vid,
VIRTIO_RXQ, p_cpl, MAX_PKT_BURST, dma_id, 0);
@@ -1029,22 +1058,7 @@ drain_vhost(struct vhost_dev *vdev)
uint16_t nr_xmit = vhost_txbuff[buff_idx]->len;
struct rte_mbuf **m = vhost_txbuff[buff_idx]->m_table;
- if (builtin_net_driver) {
- ret = vs_enqueue_pkts(vdev, VIRTIO_RXQ, m, nr_xmit);
- } else if (dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled) {
- uint16_t enqueue_fail = 0;
- int16_t dma_id = dma_bind[vdev->vid].dmas[VIRTIO_RXQ].dev_id;
-
- complete_async_pkts(vdev);
- ret = rte_vhost_submit_enqueue_burst(vdev->vid, VIRTIO_RXQ, m, nr_xmit, dma_id, 0);
-
- enqueue_fail = nr_xmit - ret;
- if (enqueue_fail)
- free_pkts(&m[ret], nr_xmit - ret);
- } else {
- ret = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
- m, nr_xmit);
- }
+ ret = vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev, VIRTIO_RXQ, m, nr_xmit);
if (enable_stats) {
__atomic_add_fetch(&vdev->stats.rx_total_atomic, nr_xmit,
@@ -1053,7 +1067,7 @@ drain_vhost(struct vhost_dev *vdev)
__ATOMIC_SEQ_CST);
}
- if (!dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled)
+ if (!dma_bind[vid2socketid[vdev->vid]].dmas[VIRTIO_RXQ].async_enabled)
free_pkts(m, nr_xmit);
}
@@ -1325,6 +1339,32 @@ drain_mbuf_table(struct mbuf_table *tx_q)
}
}
+uint16_t
+async_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mbuf **pkts, uint32_t rx_count)
+{
+ uint16_t enqueue_count;
+ uint16_t enqueue_fail = 0;
+ uint16_t dma_id = dma_bind[vid2socketid[dev->vid]].dmas[VIRTIO_RXQ].dev_id;
+
+ complete_async_pkts(dev);
+ enqueue_count = rte_vhost_submit_enqueue_burst(dev->vid, queue_id,
+ pkts, rx_count, dma_id, 0);
+
+ enqueue_fail = rx_count - enqueue_count;
+ if (enqueue_fail)
+ free_pkts(&pkts[enqueue_count], enqueue_fail);
+
+ return enqueue_count;
+}
+
+uint16_t
+sync_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mbuf **pkts, uint32_t rx_count)
+{
+ return rte_vhost_enqueue_burst(dev->vid, queue_id, pkts, rx_count);
+}
+
static __rte_always_inline void
drain_eth_rx(struct vhost_dev *vdev)
{
@@ -1355,25 +1395,8 @@ drain_eth_rx(struct vhost_dev *vdev)
}
}
- if (builtin_net_driver) {
- enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
- pkts, rx_count);
- } else if (dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled) {
- uint16_t enqueue_fail = 0;
- int16_t dma_id = dma_bind[vdev->vid].dmas[VIRTIO_RXQ].dev_id;
-
- complete_async_pkts(vdev);
- enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid,
- VIRTIO_RXQ, pkts, rx_count, dma_id, 0);
-
- enqueue_fail = rx_count - enqueue_count;
- if (enqueue_fail)
- free_pkts(&pkts[enqueue_count], enqueue_fail);
-
- } else {
- enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
- pkts, rx_count);
- }
+ enqueue_count = vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev,
+ VIRTIO_RXQ, pkts, rx_count);
if (enable_stats) {
__atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count,
@@ -1382,10 +1405,31 @@ drain_eth_rx(struct vhost_dev *vdev)
__ATOMIC_SEQ_CST);
}
- if (!dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled)
+ if (!dma_bind[vid2socketid[vdev->vid]].dmas[VIRTIO_RXQ].async_enabled)
free_pkts(pkts, rx_count);
}
+uint16_t async_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mempool *mbuf_pool,
+ struct rte_mbuf **pkts, uint16_t count)
+{
+ int nr_inflight;
+ uint16_t dequeue_count;
+ uint16_t dma_id = dma_bind[vid2socketid[dev->vid]].dmas[VIRTIO_TXQ].dev_id;
+
+ dequeue_count = rte_vhost_async_try_dequeue_burst(dev->vid, queue_id,
+ mbuf_pool, pkts, count, &nr_inflight, dma_id, 0);
+
+ return dequeue_count;
+}
+
+uint16_t sync_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mempool *mbuf_pool,
+ struct rte_mbuf **pkts, uint16_t count)
+{
+ return rte_vhost_dequeue_burst(dev->vid, queue_id, mbuf_pool, pkts, count);
+}
+
static __rte_always_inline void
drain_virtio_tx(struct vhost_dev *vdev)
{
@@ -1393,13 +1437,8 @@ drain_virtio_tx(struct vhost_dev *vdev)
uint16_t count;
uint16_t i;
- if (builtin_net_driver) {
- count = vs_dequeue_pkts(vdev, VIRTIO_TXQ, mbuf_pool,
- pkts, MAX_PKT_BURST);
- } else {
- count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ,
- mbuf_pool, pkts, MAX_PKT_BURST);
- }
+ count = vdev_queue_ops[vdev->vid].dequeue_pkt_burst(vdev,
+ VIRTIO_TXQ, mbuf_pool, pkts, MAX_PKT_BURST);
/* setup VMDq for the first packet */
if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
@@ -1478,6 +1517,26 @@ switch_worker(void *arg __rte_unused)
return 0;
}
+static void
+vhost_clear_queue_thread_unsafe(struct vhost_dev *vdev, uint16_t queue_id)
+{
+ uint16_t n_pkt = 0;
+ int pkts_inflight;
+
+ int16_t dma_id = dma_bind[vid2socketid[vdev->vid]].dmas[queue_id].dev_id;
+ pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vdev->vid, queue_id);
+
+ struct rte_mbuf *m_cpl[pkts_inflight];
+
+ while (pkts_inflight) {
+ n_pkt = rte_vhost_clear_queue_thread_unsafe(vdev->vid, queue_id, m_cpl,
+ pkts_inflight, dma_id, 0);
+ free_pkts(m_cpl, n_pkt);
+ pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vdev->vid,
+ queue_id);
+ }
+}
+
/*
* Remove a device from the specific data core linked list and from the
* main linked list. Synchronization occurs through the use of the
@@ -1535,27 +1594,79 @@ destroy_device(int vid)
vdev->vid);
if (dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled) {
- uint16_t n_pkt = 0;
- int pkts_inflight;
- int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
- pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid, VIRTIO_RXQ);
- struct rte_mbuf *m_cpl[pkts_inflight];
-
- while (pkts_inflight) {
- n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, VIRTIO_RXQ,
- m_cpl, pkts_inflight, dma_id, 0);
- free_pkts(m_cpl, n_pkt);
- pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid,
- VIRTIO_RXQ);
- }
-
+ vhost_clear_queue_thread_unsafe(vdev, VIRTIO_RXQ);
rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled = false;
}
+ if (dma_bind[vid].dmas[VIRTIO_TXQ].async_enabled) {
+ vhost_clear_queue_thread_unsafe(vdev, VIRTIO_TXQ);
+ rte_vhost_async_channel_unregister(vid, VIRTIO_TXQ);
+ dma_bind[vid].dmas[VIRTIO_TXQ].async_enabled = false;
+ }
+
rte_free(vdev);
}
+static int
+get_socketid_by_vid(int vid)
+{
+ int i;
+ char ifname[PATH_MAX];
+ rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
+
+ for (i = 0; i < nb_sockets; i++) {
+ char *file = socket_files + i * PATH_MAX;
+ if (strcmp(file, ifname) == 0)
+ return i;
+ }
+
+ return -1;
+}
+
+static int
+init_vhost_queue_ops(int vid)
+{
+ if (builtin_net_driver) {
+ vdev_queue_ops[vid].enqueue_pkt_burst = builtin_enqueue_pkts;
+ vdev_queue_ops[vid].dequeue_pkt_burst = builtin_dequeue_pkts;
+ } else {
+ if (dma_bind[vid2socketid[vid]].dmas[VIRTIO_RXQ].async_enabled)
+ vdev_queue_ops[vid].enqueue_pkt_burst = async_enqueue_pkts;
+ else
+ vdev_queue_ops[vid].enqueue_pkt_burst = sync_enqueue_pkts;
+
+ if (dma_bind[vid2socketid[vid]].dmas[VIRTIO_TXQ].async_enabled)
+ vdev_queue_ops[vid].dequeue_pkt_burst = async_dequeue_pkts;
+ else
+ vdev_queue_ops[vid].dequeue_pkt_burst = sync_dequeue_pkts;
+ }
+
+ return 0;
+}
+
+static int
+vhost_async_channel_register(int vid)
+{
+ int rx_ret = 0, tx_ret = 0;
+
+ if (dma_bind[vid2socketid[vid]].dmas[VIRTIO_RXQ].dev_id != INVALID_DMA_ID) {
+ rx_ret = rte_vhost_async_channel_register(vid, VIRTIO_RXQ);
+ if (rx_ret == 0)
+ dma_bind[vid2socketid[vid]].dmas[VIRTIO_RXQ].async_enabled = true;
+ }
+
+ if (dma_bind[vid2socketid[vid]].dmas[VIRTIO_TXQ].dev_id != INVALID_DMA_ID) {
+ tx_ret = rte_vhost_async_channel_register(vid, VIRTIO_TXQ);
+ if (tx_ret == 0)
+ dma_bind[vid2socketid[vid]].dmas[VIRTIO_TXQ].async_enabled = true;
+ }
+
+ return rx_ret | tx_ret;
+}
+
+
+
/*
* A new device is added to a data core. First the device is added to the main linked list
* and then allocated to a specific data core.
@@ -1567,6 +1678,8 @@ new_device(int vid)
uint16_t i;
uint32_t device_num_min = num_devices;
struct vhost_dev *vdev;
+ int ret;
+
vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
if (vdev == NULL) {
RTE_LOG(INFO, VHOST_DATA,
@@ -1589,6 +1702,17 @@ new_device(int vid)
}
}
+ int socketid = get_socketid_by_vid(vid);
+ if (socketid == -1)
+ return -1;
+
+ init_vid2socketid_array(vid, socketid);
+
+ ret = vhost_async_channel_register(vid);
+
+ if (init_vhost_queue_ops(vid) != 0)
+ return -1;
+
if (builtin_net_driver)
vs_vhost_net_setup(vdev);
@@ -1620,16 +1744,7 @@ new_device(int vid)
"(%d) device has been added to data core %d\n",
vid, vdev->coreid);
- if (dma_bind[vid].dmas[VIRTIO_RXQ].dev_id != INVALID_DMA_ID) {
- int ret;
-
- ret = rte_vhost_async_channel_register(vid, VIRTIO_RXQ);
- if (ret == 0)
- dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled = true;
- return ret;
- }
-
- return 0;
+ return ret;
}
static int
@@ -1647,22 +1762,9 @@ vring_state_changed(int vid, uint16_t queue_id, int enable)
if (queue_id != VIRTIO_RXQ)
return 0;
- if (dma_bind[vid].dmas[queue_id].async_enabled) {
- if (!enable) {
- uint16_t n_pkt = 0;
- int pkts_inflight;
- pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid, queue_id);
- int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
- struct rte_mbuf *m_cpl[pkts_inflight];
-
- while (pkts_inflight) {
- n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, queue_id,
- m_cpl, pkts_inflight, dma_id, 0);
- free_pkts(m_cpl, n_pkt);
- pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid,
- queue_id);
- }
- }
+ if (dma_bind[vid2socketid[vid]].dmas[queue_id].async_enabled) {
+ if (!enable)
+ vhost_clear_queue_thread_unsafe(vdev, queue_id);
}
return 0;
@@ -1887,7 +1989,7 @@ main(int argc, char *argv[])
for (i = 0; i < nb_sockets; i++) {
char *file = socket_files + i * PATH_MAX;
- if (dma_count)
+ if (dma_count && get_async_flag_by_socketid(i) != 0)
flags = flags | RTE_VHOST_USER_ASYNC_COPY;
ret = rte_vhost_driver_register(file, flags);
@@ -61,6 +61,19 @@ struct vhost_dev {
struct vhost_queue queues[MAX_QUEUE_PAIRS * 2];
} __rte_cache_aligned;
+typedef uint16_t (*vhost_enqueue_burst_t)(struct vhost_dev *dev,
+ uint16_t queue_id, struct rte_mbuf **pkts,
+ uint32_t count);
+
+typedef uint16_t (*vhost_dequeue_burst_t)(struct vhost_dev *dev,
+ uint16_t queue_id, struct rte_mempool *mbuf_pool,
+ struct rte_mbuf **pkts, uint16_t count);
+
+struct vhost_queue_ops {
+ vhost_enqueue_burst_t enqueue_pkt_burst;
+ vhost_dequeue_burst_t dequeue_pkt_burst;
+};
+
TAILQ_HEAD(vhost_dev_tailq_list, vhost_dev);
@@ -87,6 +100,7 @@ struct dma_info {
struct dma_for_vhost {
struct dma_info dmas[RTE_MAX_QUEUES_PER_PORT * 2];
+ uint32_t async_flag;
};
/* we implement non-extra virtio net features */
@@ -97,7 +111,19 @@ void vs_vhost_net_remove(struct vhost_dev *dev);
uint16_t vs_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
struct rte_mbuf **pkts, uint32_t count);
-uint16_t vs_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
- struct rte_mempool *mbuf_pool,
- struct rte_mbuf **pkts, uint16_t count);
+uint16_t builtin_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mbuf **pkts, uint32_t count);
+uint16_t builtin_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mempool *mbuf_pool,
+ struct rte_mbuf **pkts, uint16_t count);
+uint16_t sync_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mbuf **pkts, uint32_t count);
+uint16_t sync_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mempool *mbuf_pool,
+ struct rte_mbuf **pkts, uint16_t count);
+uint16_t async_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mbuf **pkts, uint32_t count);
+uint16_t async_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mempool *mbuf_pool,
+ struct rte_mbuf **pkts, uint16_t count);
#endif /* _MAIN_H_ */
@@ -238,6 +238,13 @@ vs_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
return count;
}
+uint16_t
+builtin_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mbuf **pkts, uint32_t count)
+{
+ return vs_enqueue_pkts(dev, queue_id, pkts, count);
+}
+
static __rte_always_inline int
dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
struct rte_mbuf *m, uint16_t desc_idx,
@@ -363,7 +370,7 @@ dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
return 0;
}
-uint16_t
+static uint16_t
vs_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
{
@@ -440,3 +447,10 @@ vs_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
return i;
}
+
+uint16_t
+builtin_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+ struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
+{
+ return vs_dequeue_pkts(dev, queue_id, mbuf_pool, pkts, count);
+}