#define MAX_QUEUES 128
#endif
+#define NUM_MBUFS_DEFAULT 0x24000
+
/* the maximum number of external ports supported */
#define MAX_SUP_PORTS 1
#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;
+
struct dma_for_vhost dma_bind[RTE_MAX_VHOST_DEVICE];
int16_t dmas_id[RTE_DMADEV_DEFAULT_MAX];
static int dma_count;
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 = {
#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 inline uint32_t
+get_async_flag_by_socketid(int socketid)
+{
+ return dma_bind[socketid].async_flag;
+}
+
+static inline void
+init_vid2socketid_array(int vid, int socketid)
+{
+ vid2socketid[vid] = socketid;
+}
+
static inline bool
is_dma_configured(int16_t dev_id)
{
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 };
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) {
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;
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:
return retval;
}
+ if (dev_info.max_vmdq_pools == 0) {
+ RTE_LOG(ERR, VHOST_PORT, "Failed to get VMDq info.\n");
+ return -1;
+ }
rxconf = &dev_info.default_rxconf;
txconf = &dev_info.default_txconf;
" --tx-csum [0|1] disable/enable TX checksum offload.\n"
" --tso [0|1] disable/enable TCP segment offload.\n"
" --client register a vhost-user socket as client mode.\n"
- " --dma-type register dma type for your vhost async driver. For example \"ioat\" for now.\n"
- " --dmas register dma channel for specific vhost device.\n",
+ " --dmas register dma channel for specific vhost device.\n"
+ " --total-num-mbufs [0-N] set the number of mbufs to be allocated in mbuf pools, the default value is 147456.\n",
prgname);
}
OPT_BUILTIN_NET_DRIVER_NUM,
#define OPT_DMAS "dmas"
OPT_DMAS_NUM,
+#define OPT_NUM_MBUFS "total-num-mbufs"
+ OPT_NUM_MBUFS_NUM,
};
/*
NULL, OPT_BUILTIN_NET_DRIVER_NUM},
{OPT_DMAS, required_argument,
NULL, OPT_DMAS_NUM},
+ {OPT_NUM_MBUFS, required_argument,
+ NULL, OPT_NUM_MBUFS_NUM},
{NULL, 0, 0, 0},
};
}
break;
+ case OPT_NUM_MBUFS_NUM:
+ ret = parse_num_opt(optarg, INT32_MAX);
+ if (ret == -1) {
+ RTE_LOG(INFO, VHOST_CONFIG,
+ "Invalid argument for total-num-mbufs [0..N]\n");
+ us_vhost_usage(prgname);
+ return -1;
+ }
+
+ if (total_num_mbufs < ret)
+ total_num_mbufs = ret;
+ break;
+
case OPT_CLIENT_NUM:
client_mode = 1;
break;
{
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);
- if (complete_count) {
+ if (complete_count)
free_pkts(p_cpl, complete_count);
- __atomic_sub_fetch(&vdev->pkts_inflight, complete_count, __ATOMIC_SEQ_CST);
- }
}
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);
- __atomic_add_fetch(&vdev->pkts_inflight, ret, __ATOMIC_SEQ_CST);
-
- 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,
__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);
}
}
}
+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)
{
}
}
- 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);
- __atomic_add_fetch(&vdev->pkts_inflight, enqueue_count, __ATOMIC_SEQ_CST);
-
- 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,
__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;
+ int16_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)
{
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) {
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
vdev->vid);
if (dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled) {
- uint16_t n_pkt = 0;
- int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
- struct rte_mbuf *m_cpl[vdev->pkts_inflight];
-
- while (vdev->pkts_inflight) {
- n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, VIRTIO_RXQ,
- m_cpl, vdev->pkts_inflight, dma_id, 0);
- free_pkts(m_cpl, n_pkt);
- __atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST);
- }
-
+ 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 inline 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 inline 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.
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,
}
}
+ 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);
"(%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
if (queue_id != VIRTIO_RXQ)
return 0;
- if (dma_bind[vid].dmas[queue_id].async_enabled) {
- if (!enable) {
- uint16_t n_pkt = 0;
- int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
- struct rte_mbuf *m_cpl[vdev->pkts_inflight];
-
- while (vdev->pkts_inflight) {
- n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, queue_id,
- m_cpl, vdev->pkts_inflight, dma_id, 0);
- free_pkts(m_cpl, n_pkt);
- __atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST);
- }
- }
+ if (dma_bind[vid2socketid[vid]].dmas[queue_id].async_enabled) {
+ if (!enable)
+ vhost_clear_queue_thread_unsafe(vdev, queue_id);
}
return 0;
exit(0);
}
-/*
- * While creating an mbuf pool, one key thing is to figure out how
- * many mbuf entries is enough for our use. FYI, here are some
- * guidelines:
- *
- * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
- *
- * - For each switch core (A CPU core does the packet switch), we need
- * also make some reservation for receiving the packets from virtio
- * Tx queue. How many is enough depends on the usage. It's normally
- * a simple calculation like following:
- *
- * MAX_PKT_BURST * max packet size / mbuf size
- *
- * So, we definitely need allocate more mbufs when TSO is enabled.
- *
- * - Similarly, for each switching core, we should serve @nr_rx_desc
- * mbufs for receiving the packets from physical NIC device.
- *
- * - We also need make sure, for each switch core, we have allocated
- * enough mbufs to fill up the mbuf cache.
- */
-static void
-create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
- uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
-{
- uint32_t nr_mbufs;
- uint32_t nr_mbufs_per_core;
- uint32_t mtu = 1500;
-
- if (mergeable)
- mtu = 9000;
- if (enable_tso)
- mtu = 64 * 1024;
-
- nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST /
- (mbuf_size - RTE_PKTMBUF_HEADROOM);
- nr_mbufs_per_core += nr_rx_desc;
- nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
-
- nr_mbufs = nr_queues * nr_rx_desc;
- nr_mbufs += nr_mbufs_per_core * nr_switch_core;
- nr_mbufs *= nr_port;
-
- mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
- nr_mbuf_cache, 0, mbuf_size,
- rte_socket_id());
- if (mbuf_pool == NULL)
- rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
-}
-
static void
reset_dma(void)
{
* many queues here. We probably should only do allocation for
* those queues we are going to use.
*/
- create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
- MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
+ mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", total_num_mbufs,
+ MBUF_CACHE_SIZE, 0, MBUF_DATA_SIZE,
+ rte_socket_id());
+ if (mbuf_pool == NULL)
+ rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
if (vm2vm_mode == VM2VM_HARDWARE) {
/* Enable VT loop back to let L2 switch to do it. */
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);
git.droids-corp.org / dpdk.git / blobdiff